hieroglyphs qingxu98 commited on
Commit
f81fd9e
0 Parent(s):

Duplicate from qingxu98/gpt-academic

Browse files

Co-authored-by: fuqingxu <qingxu98@users.noreply.huggingface.co>

This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +5 -0
  2. .github/ISSUE_TEMPLATE/bug_report.yml +75 -0
  3. .github/ISSUE_TEMPLATE/feature_request.yml +28 -0
  4. .github/workflows/build-with-chatglm.yml +44 -0
  5. .github/workflows/build-with-jittorllms.yml +44 -0
  6. .github/workflows/build-with-latex.yml +44 -0
  7. .github/workflows/build-without-local-llms.yml +44 -0
  8. .gitignore +152 -0
  9. Dockerfile +34 -0
  10. LICENSE +674 -0
  11. README.md +373 -0
  12. app.py +237 -0
  13. check_proxy.py +169 -0
  14. colorful.py +61 -0
  15. config.py +220 -0
  16. core_functional.py +94 -0
  17. crazy_functional.py +439 -0
  18. crazy_functions/Langchain知识库.py +107 -0
  19. crazy_functions/Latex全文润色.py +243 -0
  20. crazy_functions/Latex全文翻译.py +175 -0
  21. crazy_functions/Latex输出PDF结果.py +300 -0
  22. crazy_functions/__init__.py +0 -0
  23. crazy_functions/chatglm微调工具.py +141 -0
  24. crazy_functions/crazy_functions_test.py +231 -0
  25. crazy_functions/crazy_utils.py +753 -0
  26. crazy_functions/latex_fns/latex_actions.py +447 -0
  27. crazy_functions/latex_fns/latex_toolbox.py +459 -0
  28. crazy_functions/latex_utils.py +788 -0
  29. crazy_functions/live_audio/aliyunASR.py +130 -0
  30. crazy_functions/live_audio/audio_io.py +51 -0
  31. crazy_functions/pdf_fns/parse_pdf.py +25 -0
  32. crazy_functions/test_project/cpp/cppipc/buffer.cpp +87 -0
  33. crazy_functions/test_project/cpp/cppipc/ipc.cpp +701 -0
  34. crazy_functions/test_project/cpp/cppipc/policy.h +25 -0
  35. crazy_functions/test_project/cpp/cppipc/pool_alloc.cpp +17 -0
  36. crazy_functions/test_project/cpp/cppipc/prod_cons.h +433 -0
  37. crazy_functions/test_project/cpp/cppipc/queue.h +216 -0
  38. crazy_functions/test_project/cpp/cppipc/shm.cpp +103 -0
  39. crazy_functions/test_project/cpp/cppipc/waiter.h +83 -0
  40. crazy_functions/test_project/cpp/cppipc/来源 +3 -0
  41. crazy_functions/test_project/cpp/libJPG/jpgd.cpp +3276 -0
  42. crazy_functions/test_project/cpp/libJPG/jpgd.h +316 -0
  43. crazy_functions/test_project/cpp/libJPG/jpge.cpp +1049 -0
  44. crazy_functions/test_project/cpp/libJPG/jpge.h +172 -0
  45. crazy_functions/test_project/cpp/libJPG/来源 +3 -0
  46. crazy_functions/test_project/cpp/longcode/jpgd.cpp +3276 -0
  47. crazy_functions/test_project/cpp/longcode/jpge.cpp +1049 -0
  48. crazy_functions/test_project/cpp/longcode/prod_cons.h +433 -0
  49. crazy_functions/test_project/latex/attention/background.tex +58 -0
  50. crazy_functions/test_project/latex/attention/introduction.tex +18 -0
.gitattributes ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ *.h linguist-detectable=false
2
+ *.cpp linguist-detectable=false
3
+ *.tex linguist-detectable=false
4
+ *.cs linguist-detectable=false
5
+ *.tps linguist-detectable=false
.github/ISSUE_TEMPLATE/bug_report.yml ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: Report Bug | 报告BUG
2
+ description: "Report bug"
3
+ title: "[Bug]: "
4
+ labels: []
5
+ body:
6
+ - type: dropdown
7
+ id: download
8
+ attributes:
9
+ label: Installation Method | 安装方法与平台
10
+ options:
11
+ - Please choose | 请选择
12
+ - Pip Install (I ignored requirements.txt)
13
+ - Pip Install (I used latest requirements.txt)
14
+ - Anaconda (I ignored requirements.txt)
15
+ - Anaconda (I used latest requirements.txt)
16
+ - Docker(Windows/Mac)
17
+ - Docker(Linux)
18
+ - Docker-Compose(Windows/Mac)
19
+ - Docker-Compose(Linux)
20
+ - Huggingface
21
+ - Others (Please Describe)
22
+ validations:
23
+ required: true
24
+
25
+ - type: dropdown
26
+ id: version
27
+ attributes:
28
+ label: Version | 版本
29
+ options:
30
+ - Please choose | 请选择
31
+ - Latest | 最新版
32
+ - Others | 非最新版
33
+ validations:
34
+ required: true
35
+
36
+ - type: dropdown
37
+ id: os
38
+ attributes:
39
+ label: OS | 操作系统
40
+ options:
41
+ - Please choose | 请选择
42
+ - Windows
43
+ - Mac
44
+ - Linux
45
+ - Docker
46
+ validations:
47
+ required: true
48
+
49
+ - type: textarea
50
+ id: describe
51
+ attributes:
52
+ label: Describe the bug | 简述
53
+ description: Describe the bug | 简述
54
+ validations:
55
+ required: true
56
+
57
+ - type: textarea
58
+ id: screenshot
59
+ attributes:
60
+ label: Screen Shot | 有帮助的截图
61
+ description: Screen Shot | 有帮助的截图
62
+ validations:
63
+ required: true
64
+
65
+ - type: textarea
66
+ id: traceback
67
+ attributes:
68
+ label: Terminal Traceback & Material to Help Reproduce Bugs | 终端traceback(如有) + 帮助我们复现的测试材料样本(如有)
69
+ description: Terminal Traceback & Material to Help Reproduce Bugs | 终端traceback(如有) + 帮助我们复现的测试材料样本(如有)
70
+
71
+
72
+
73
+
74
+
75
+
.github/ISSUE_TEMPLATE/feature_request.yml ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: Feature Request | 功能请求
2
+ description: "Feature Request"
3
+ title: "[Feature]: "
4
+ labels: []
5
+ body:
6
+ - type: dropdown
7
+ id: download
8
+ attributes:
9
+ label: Class | 类型
10
+ options:
11
+ - Please choose | 请选择
12
+ - 其他
13
+ - 函数插件
14
+ - 大语言模型
15
+ - 程序主体
16
+ validations:
17
+ required: false
18
+
19
+ - type: textarea
20
+ id: traceback
21
+ attributes:
22
+ label: Feature Request | 功能请求
23
+ description: Feature Request | 功能请求
24
+
25
+
26
+
27
+
28
+
.github/workflows/build-with-chatglm.yml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # https://docs.github.com/en/actions/publishing-packages/publishing-docker-images#publishing-images-to-github-packages
2
+ name: Create and publish a Docker image for ChatGLM support
3
+
4
+ on:
5
+ push:
6
+ branches:
7
+ - 'master'
8
+
9
+ env:
10
+ REGISTRY: ghcr.io
11
+ IMAGE_NAME: ${{ github.repository }}_chatglm_moss
12
+
13
+ jobs:
14
+ build-and-push-image:
15
+ runs-on: ubuntu-latest
16
+ permissions:
17
+ contents: read
18
+ packages: write
19
+
20
+ steps:
21
+ - name: Checkout repository
22
+ uses: actions/checkout@v3
23
+
24
+ - name: Log in to the Container registry
25
+ uses: docker/login-action@v2
26
+ with:
27
+ registry: ${{ env.REGISTRY }}
28
+ username: ${{ github.actor }}
29
+ password: ${{ secrets.GITHUB_TOKEN }}
30
+
31
+ - name: Extract metadata (tags, labels) for Docker
32
+ id: meta
33
+ uses: docker/metadata-action@v4
34
+ with:
35
+ images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
36
+
37
+ - name: Build and push Docker image
38
+ uses: docker/build-push-action@v4
39
+ with:
40
+ context: .
41
+ push: true
42
+ file: docs/GithubAction+ChatGLM+Moss
43
+ tags: ${{ steps.meta.outputs.tags }}
44
+ labels: ${{ steps.meta.outputs.labels }}
.github/workflows/build-with-jittorllms.yml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # https://docs.github.com/en/actions/publishing-packages/publishing-docker-images#publishing-images-to-github-packages
2
+ name: Create and publish a Docker image for ChatGLM support
3
+
4
+ on:
5
+ push:
6
+ branches:
7
+ - 'master'
8
+
9
+ env:
10
+ REGISTRY: ghcr.io
11
+ IMAGE_NAME: ${{ github.repository }}_jittorllms
12
+
13
+ jobs:
14
+ build-and-push-image:
15
+ runs-on: ubuntu-latest
16
+ permissions:
17
+ contents: read
18
+ packages: write
19
+
20
+ steps:
21
+ - name: Checkout repository
22
+ uses: actions/checkout@v3
23
+
24
+ - name: Log in to the Container registry
25
+ uses: docker/login-action@v2
26
+ with:
27
+ registry: ${{ env.REGISTRY }}
28
+ username: ${{ github.actor }}
29
+ password: ${{ secrets.GITHUB_TOKEN }}
30
+
31
+ - name: Extract metadata (tags, labels) for Docker
32
+ id: meta
33
+ uses: docker/metadata-action@v4
34
+ with:
35
+ images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
36
+
37
+ - name: Build and push Docker image
38
+ uses: docker/build-push-action@v4
39
+ with:
40
+ context: .
41
+ push: true
42
+ file: docs/GithubAction+JittorLLMs
43
+ tags: ${{ steps.meta.outputs.tags }}
44
+ labels: ${{ steps.meta.outputs.labels }}
.github/workflows/build-with-latex.yml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # https://docs.github.com/en/actions/publishing-packages/publishing-docker-images#publishing-images-to-github-packages
2
+ name: Create and publish a Docker image for Latex support
3
+
4
+ on:
5
+ push:
6
+ branches:
7
+ - 'master'
8
+
9
+ env:
10
+ REGISTRY: ghcr.io
11
+ IMAGE_NAME: ${{ github.repository }}_with_latex
12
+
13
+ jobs:
14
+ build-and-push-image:
15
+ runs-on: ubuntu-latest
16
+ permissions:
17
+ contents: read
18
+ packages: write
19
+
20
+ steps:
21
+ - name: Checkout repository
22
+ uses: actions/checkout@v3
23
+
24
+ - name: Log in to the Container registry
25
+ uses: docker/login-action@v2
26
+ with:
27
+ registry: ${{ env.REGISTRY }}
28
+ username: ${{ github.actor }}
29
+ password: ${{ secrets.GITHUB_TOKEN }}
30
+
31
+ - name: Extract metadata (tags, labels) for Docker
32
+ id: meta
33
+ uses: docker/metadata-action@v4
34
+ with:
35
+ images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
36
+
37
+ - name: Build and push Docker image
38
+ uses: docker/build-push-action@v4
39
+ with:
40
+ context: .
41
+ push: true
42
+ file: docs/GithubAction+NoLocal+Latex
43
+ tags: ${{ steps.meta.outputs.tags }}
44
+ labels: ${{ steps.meta.outputs.labels }}
.github/workflows/build-without-local-llms.yml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # https://docs.github.com/en/actions/publishing-packages/publishing-docker-images#publishing-images-to-github-packages
2
+ name: Create and publish a Docker image
3
+
4
+ on:
5
+ push:
6
+ branches:
7
+ - 'master'
8
+
9
+ env:
10
+ REGISTRY: ghcr.io
11
+ IMAGE_NAME: ${{ github.repository }}_nolocal
12
+
13
+ jobs:
14
+ build-and-push-image:
15
+ runs-on: ubuntu-latest
16
+ permissions:
17
+ contents: read
18
+ packages: write
19
+
20
+ steps:
21
+ - name: Checkout repository
22
+ uses: actions/checkout@v3
23
+
24
+ - name: Log in to the Container registry
25
+ uses: docker/login-action@v2
26
+ with:
27
+ registry: ${{ env.REGISTRY }}
28
+ username: ${{ github.actor }}
29
+ password: ${{ secrets.GITHUB_TOKEN }}
30
+
31
+ - name: Extract metadata (tags, labels) for Docker
32
+ id: meta
33
+ uses: docker/metadata-action@v4
34
+ with:
35
+ images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
36
+
37
+ - name: Build and push Docker image
38
+ uses: docker/build-push-action@v4
39
+ with:
40
+ context: .
41
+ push: true
42
+ file: docs/GithubAction+NoLocal
43
+ tags: ${{ steps.meta.outputs.tags }}
44
+ labels: ${{ steps.meta.outputs.labels }}
.gitignore ADDED
@@ -0,0 +1,152 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Byte-compiled / optimized / DLL files
2
+ __pycache__/
3
+ *.py[cod]
4
+ *$py.class
5
+
6
+ # C extensions
7
+ *.so
8
+
9
+ # Distribution / packaging
10
+ .Python
11
+ build/
12
+ develop-eggs/
13
+ dist/
14
+ downloads/
15
+ eggs/
16
+ .eggs/
17
+ lib/
18
+ lib64/
19
+ parts/
20
+ sdist/
21
+ var/
22
+ wheels/
23
+ pip-wheel-metadata/
24
+ share/python-wheels/
25
+ *.egg-info/
26
+ .installed.cfg
27
+ *.egg
28
+ MANIFEST
29
+
30
+ # PyInstaller
31
+ # Usually these files are written by a python script from a template
32
+ # before PyInstaller builds the exe, so as to inject date/other infos into it.
33
+ *.manifest
34
+ *.spec
35
+ # Installer logs
36
+ pip-log.txt
37
+ pip-delete-this-directory.txt
38
+
39
+ # Unit test / coverage reports
40
+ htmlcov/
41
+ .tox/
42
+ .nox/
43
+ .coverage
44
+ .coverage.*
45
+ .cache
46
+ nosetests.xml
47
+ coverage.xml
48
+ *.cover
49
+ *.py,cover
50
+ .hypothesis/
51
+ .pytest_cache/
52
+
53
+ # Translations
54
+ *.mo
55
+ *.pot
56
+ github
57
+ .github
58
+ TEMP
59
+ TRASH
60
+
61
+ # Django stuff:
62
+ *.log
63
+ local_settings.py
64
+ db.sqlite3
65
+ db.sqlite3-journal
66
+
67
+ # Flask stuff:
68
+ instance/
69
+ .webassets-cache
70
+
71
+ # Scrapy stuff:
72
+ .scrapy
73
+
74
+ # Sphinx documentation
75
+ docs/_build/
76
+
77
+ # PyBuilder
78
+ target/
79
+
80
+ # Jupyter Notebook
81
+ .ipynb_checkpoints
82
+
83
+ # IPython
84
+ profile_default/
85
+ ipython_config.py
86
+
87
+ # pyenv
88
+ .python-version
89
+
90
+ # pipenv
91
+ # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
92
+ # However, in case of collaboration, if having platform-specific dependencies or dependencies
93
+ # having no cross-platform support, pipenv may install dependencies that don't work, or not
94
+ # install all needed dependencies.
95
+ #Pipfile.lock
96
+
97
+ # PEP 582; used by e.g. github.com/David-OConnor/pyflow
98
+ __pypackages__/
99
+
100
+ # Celery stuff
101
+ celerybeat-schedule
102
+ celerybeat.pid
103
+
104
+ # SageMath parsed files
105
+ *.sage.py
106
+
107
+ # Environments
108
+ .env
109
+ .venv
110
+ env/
111
+ venv/
112
+ ENV/
113
+ env.bak/
114
+ venv.bak/
115
+
116
+ # Spyder project settings
117
+ .spyderproject
118
+ .spyproject
119
+
120
+ # Rope project settings
121
+ .ropeproject
122
+
123
+ # mkdocs documentation
124
+ /site
125
+
126
+ # mypy
127
+ .mypy_cache/
128
+ .dmypy.json
129
+ dmypy.json
130
+
131
+ # Pyre type checker
132
+ .pyre/
133
+
134
+ .vscode
135
+ .idea
136
+
137
+ history
138
+ ssr_conf
139
+ config_private.py
140
+ gpt_log
141
+ private.md
142
+ private_upload
143
+ other_llms
144
+ cradle*
145
+ debug*
146
+ private*
147
+ crazy_functions/test_project/pdf_and_word
148
+ crazy_functions/test_samples
149
+ request_llm/jittorllms
150
+ multi-language
151
+ request_llm/moss
152
+ media
Dockerfile ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # 此Dockerfile适用于“无本地模型”的环境构建,如果需要使用chatglm等本地模型或者latex运行依赖,请参考 docker-compose.yml
2
+ # 如何构建: 先修改 `config.py`, 然后 `docker build -t gpt-academic . `
3
+ # 如何运行(Linux下): `docker run --rm -it --net=host gpt-academic `
4
+ # 如何运行(其他操作系统,选择任意一个固定端口50923): `docker run --rm -it -e WEB_PORT=50923 -p 50923:50923 gpt-academic `
5
+ FROM python:3.11
6
+
7
+
8
+ # 非必要步骤,更换pip源
9
+ RUN echo '[global]' > /etc/pip.conf && \
10
+ echo 'index-url = https://mirrors.aliyun.com/pypi/simple/' >> /etc/pip.conf && \
11
+ echo 'trusted-host = mirrors.aliyun.com' >> /etc/pip.conf
12
+
13
+
14
+ # 进入工作路径
15
+ WORKDIR /gpt
16
+
17
+
18
+ # 安装大部分依赖,利用Docker缓存加速以后的构建
19
+ COPY requirements.txt ./
20
+ COPY ./docs/gradio-3.32.2-py3-none-any.whl ./docs/gradio-3.32.2-py3-none-any.whl
21
+ RUN pip3 install -r requirements.txt
22
+
23
+
24
+ # 装载项目文件,安装剩余依赖
25
+ COPY . .
26
+ RUN pip3 install -r requirements.txt
27
+
28
+
29
+ # 非必要步骤,用于预热模块
30
+ RUN python3 -c 'from check_proxy import warm_up_modules; warm_up_modules()'
31
+
32
+
33
+ # 启动
34
+ CMD ["python3", "-u", "main.py"]
LICENSE ADDED
@@ -0,0 +1,674 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ GNU GENERAL PUBLIC LICENSE
2
+ Version 3, 29 June 2007
3
+
4
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
5
+ Everyone is permitted to copy and distribute verbatim copies
6
+ of this license document, but changing it is not allowed.
7
+
8
+ Preamble
9
+
10
+ The GNU General Public License is a free, copyleft license for
11
+ software and other kinds of works.
12
+
13
+ The licenses for most software and other practical works are designed
14
+ to take away your freedom to share and change the works. By contrast,
15
+ the GNU General Public License is intended to guarantee your freedom to
16
+ share and change all versions of a program--to make sure it remains free
17
+ software for all its users. We, the Free Software Foundation, use the
18
+ GNU General Public License for most of our software; it applies also to
19
+ any other work released this way by its authors. You can apply it to
20
+ your programs, too.
21
+
22
+ When we speak of free software, we are referring to freedom, not
23
+ price. Our General Public Licenses are designed to make sure that you
24
+ have the freedom to distribute copies of free software (and charge for
25
+ them if you wish), that you receive source code or can get it if you
26
+ want it, that you can change the software or use pieces of it in new
27
+ free programs, and that you know you can do these things.
28
+
29
+ To protect your rights, we need to prevent others from denying you
30
+ these rights or asking you to surrender the rights. Therefore, you have
31
+ certain responsibilities if you distribute copies of the software, or if
32
+ you modify it: responsibilities to respect the freedom of others.
33
+
34
+ For example, if you distribute copies of such a program, whether
35
+ gratis or for a fee, you must pass on to the recipients the same
36
+ freedoms that you received. You must make sure that they, too, receive
37
+ or can get the source code. And you must show them these terms so they
38
+ know their rights.
39
+
40
+ Developers that use the GNU GPL protect your rights with two steps:
41
+ (1) assert copyright on the software, and (2) offer you this License
42
+ giving you legal permission to copy, distribute and/or modify it.
43
+
44
+ For the developers' and authors' protection, the GPL clearly explains
45
+ that there is no warranty for this free software. For both users' and
46
+ authors' sake, the GPL requires that modified versions be marked as
47
+ changed, so that their problems will not be attributed erroneously to
48
+ authors of previous versions.
49
+
50
+ Some devices are designed to deny users access to install or run
51
+ modified versions of the software inside them, although the manufacturer
52
+ can do so. This is fundamentally incompatible with the aim of
53
+ protecting users' freedom to change the software. The systematic
54
+ pattern of such abuse occurs in the area of products for individuals to
55
+ use, which is precisely where it is most unacceptable. Therefore, we
56
+ have designed this version of the GPL to prohibit the practice for those
57
+ products. If such problems arise substantially in other domains, we
58
+ stand ready to extend this provision to those domains in future versions
59
+ of the GPL, as needed to protect the freedom of users.
60
+
61
+ Finally, every program is threatened constantly by software patents.
62
+ States should not allow patents to restrict development and use of
63
+ software on general-purpose computers, but in those that do, we wish to
64
+ avoid the special danger that patents applied to a free program could
65
+ make it effectively proprietary. To prevent this, the GPL assures that
66
+ patents cannot be used to render the program non-free.
67
+
68
+ The precise terms and conditions for copying, distribution and
69
+ modification follow.
70
+
71
+ TERMS AND CONDITIONS
72
+
73
+ 0. Definitions.
74
+
75
+ "This License" refers to version 3 of the GNU General Public License.
76
+
77
+ "Copyright" also means copyright-like laws that apply to other kinds of
78
+ works, such as semiconductor masks.
79
+
80
+ "The Program" refers to any copyrightable work licensed under this
81
+ License. Each licensee is addressed as "you". "Licensees" and
82
+ "recipients" may be individuals or organizations.
83
+
84
+ To "modify" a work means to copy from or adapt all or part of the work
85
+ in a fashion requiring copyright permission, other than the making of an
86
+ exact copy. The resulting work is called a "modified version" of the
87
+ earlier work or a work "based on" the earlier work.
88
+
89
+ A "covered work" means either the unmodified Program or a work based
90
+ on the Program.
91
+
92
+ To "propagate" a work means to do anything with it that, without
93
+ permission, would make you directly or secondarily liable for
94
+ infringement under applicable copyright law, except executing it on a
95
+ computer or modifying a private copy. Propagation includes copying,
96
+ distribution (with or without modification), making available to the
97
+ public, and in some countries other activities as well.
98
+
99
+ To "convey" a work means any kind of propagation that enables other
100
+ parties to make or receive copies. Mere interaction with a user through
101
+ a computer network, with no transfer of a copy, is not conveying.
102
+
103
+ An interactive user interface displays "Appropriate Legal Notices"
104
+ to the extent that it includes a convenient and prominently visible
105
+ feature that (1) displays an appropriate copyright notice, and (2)
106
+ tells the user that there is no warranty for the work (except to the
107
+ extent that warranties are provided), that licensees may convey the
108
+ work under this License, and how to view a copy of this License. If
109
+ the interface presents a list of user commands or options, such as a
110
+ menu, a prominent item in the list meets this criterion.
111
+
112
+ 1. Source Code.
113
+
114
+ The "source code" for a work means the preferred form of the work
115
+ for making modifications to it. "Object code" means any non-source
116
+ form of a work.
117
+
118
+ A "Standard Interface" means an interface that either is an official
119
+ standard defined by a recognized standards body, or, in the case of
120
+ interfaces specified for a particular programming language, one that
121
+ is widely used among developers working in that language.
122
+
123
+ The "System Libraries" of an executable work include anything, other
124
+ than the work as a whole, that (a) is included in the normal form of
125
+ packaging a Major Component, but which is not part of that Major
126
+ Component, and (b) serves only to enable use of the work with that
127
+ Major Component, or to implement a Standard Interface for which an
128
+ implementation is available to the public in source code form. A
129
+ "Major Component", in this context, means a major essential component
130
+ (kernel, window system, and so on) of the specific operating system
131
+ (if any) on which the executable work runs, or a compiler used to
132
+ produce the work, or an object code interpreter used to run it.
133
+
134
+ The "Corresponding Source" for a work in object code form means all
135
+ the source code needed to generate, install, and (for an executable
136
+ work) run the object code and to modify the work, including scripts to
137
+ control those activities. However, it does not include the work's
138
+ System Libraries, or general-purpose tools or generally available free
139
+ programs which are used unmodified in performing those activities but
140
+ which are not part of the work. For example, Corresponding Source
141
+ includes interface definition files associated with source files for
142
+ the work, and the source code for shared libraries and dynamically
143
+ linked subprograms that the work is specifically designed to require,
144
+ such as by intimate data communication or control flow between those
145
+ subprograms and other parts of the work.
146
+
147
+ The Corresponding Source need not include anything that users
148
+ can regenerate automatically from other parts of the Corresponding
149
+ Source.
150
+
151
+ The Corresponding Source for a work in source code form is that
152
+ same work.
153
+
154
+ 2. Basic Permissions.
155
+
156
+ All rights granted under this License are granted for the term of
157
+ copyright on the Program, and are irrevocable provided the stated
158
+ conditions are met. This License explicitly affirms your unlimited
159
+ permission to run the unmodified Program. The output from running a
160
+ covered work is covered by this License only if the output, given its
161
+ content, constitutes a covered work. This License acknowledges your
162
+ rights of fair use or other equivalent, as provided by copyright law.
163
+
164
+ You may make, run and propagate covered works that you do not
165
+ convey, without conditions so long as your license otherwise remains
166
+ in force. You may convey covered works to others for the sole purpose
167
+ of having them make modifications exclusively for you, or provide you
168
+ with facilities for running those works, provided that you comply with
169
+ the terms of this License in conveying all material for which you do
170
+ not control copyright. Those thus making or running the covered works
171
+ for you must do so exclusively on your behalf, under your direction
172
+ and control, on terms that prohibit them from making any copies of
173
+ your copyrighted material outside their relationship with you.
174
+
175
+ Conveying under any other circumstances is permitted solely under
176
+ the conditions stated below. Sublicensing is not allowed; section 10
177
+ makes it unnecessary.
178
+
179
+ 3. Protecting Users' Legal Rights From Anti-Circumvention Law.
180
+
181
+ No covered work shall be deemed part of an effective technological
182
+ measure under any applicable law fulfilling obligations under article
183
+ 11 of the WIPO copyright treaty adopted on 20 December 1996, or
184
+ similar laws prohibiting or restricting circumvention of such
185
+ measures.
186
+
187
+ When you convey a covered work, you waive any legal power to forbid
188
+ circumvention of technological measures to the extent such circumvention
189
+ is effected by exercising rights under this License with respect to
190
+ the covered work, and you disclaim any intention to limit operation or
191
+ modification of the work as a means of enforcing, against the work's
192
+ users, your or third parties' legal rights to forbid circumvention of
193
+ technological measures.
194
+
195
+ 4. Conveying Verbatim Copies.
196
+
197
+ You may convey verbatim copies of the Program's source code as you
198
+ receive it, in any medium, provided that you conspicuously and
199
+ appropriately publish on each copy an appropriate copyright notice;
200
+ keep intact all notices stating that this License and any
201
+ non-permissive terms added in accord with section 7 apply to the code;
202
+ keep intact all notices of the absence of any warranty; and give all
203
+ recipients a copy of this License along with the Program.
204
+
205
+ You may charge any price or no price for each copy that you convey,
206
+ and you may offer support or warranty protection for a fee.
207
+
208
+ 5. Conveying Modified Source Versions.
209
+
210
+ You may convey a work based on the Program, or the modifications to
211
+ produce it from the Program, in the form of source code under the
212
+ terms of section 4, provided that you also meet all of these conditions:
213
+
214
+ a) The work must carry prominent notices stating that you modified
215
+ it, and giving a relevant date.
216
+
217
+ b) The work must carry prominent notices stating that it is
218
+ released under this License and any conditions added under section
219
+ 7. This requirement modifies the requirement in section 4 to
220
+ "keep intact all notices".
221
+
222
+ c) You must license the entire work, as a whole, under this
223
+ License to anyone who comes into possession of a copy. This
224
+ License will therefore apply, along with any applicable section 7
225
+ additional terms, to the whole of the work, and all its parts,
226
+ regardless of how they are packaged. This License gives no
227
+ permission to license the work in any other way, but it does not
228
+ invalidate such permission if you have separately received it.
229
+
230
+ d) If the work has interactive user interfaces, each must display
231
+ Appropriate Legal Notices; however, if the Program has interactive
232
+ interfaces that do not display Appropriate Legal Notices, your
233
+ work need not make them do so.
234
+
235
+ A compilation of a covered work with other separate and independent
236
+ works, which are not by their nature extensions of the covered work,
237
+ and which are not combined with it such as to form a larger program,
238
+ in or on a volume of a storage or distribution medium, is called an
239
+ "aggregate" if the compilation and its resulting copyright are not
240
+ used to limit the access or legal rights of the compilation's users
241
+ beyond what the individual works permit. Inclusion of a covered work
242
+ in an aggregate does not cause this License to apply to the other
243
+ parts of the aggregate.
244
+
245
+ 6. Conveying Non-Source Forms.
246
+
247
+ You may convey a covered work in object code form under the terms
248
+ of sections 4 and 5, provided that you also convey the
249
+ machine-readable Corresponding Source under the terms of this License,
250
+ in one of these ways:
251
+
252
+ a) Convey the object code in, or embodied in, a physical product
253
+ (including a physical distribution medium), accompanied by the
254
+ Corresponding Source fixed on a durable physical medium
255
+ customarily used for software interchange.
256
+
257
+ b) Convey the object code in, or embodied in, a physical product
258
+ (including a physical distribution medium), accompanied by a
259
+ written offer, valid for at least three years and valid for as
260
+ long as you offer spare parts or customer support for that product
261
+ model, to give anyone who possesses the object code either (1) a
262
+ copy of the Corresponding Source for all the software in the
263
+ product that is covered by this License, on a durable physical
264
+ medium customarily used for software interchange, for a price no
265
+ more than your reasonable cost of physically performing this
266
+ conveying of source, or (2) access to copy the
267
+ Corresponding Source from a network server at no charge.
268
+
269
+ c) Convey individual copies of the object code with a copy of the
270
+ written offer to provide the Corresponding Source. This
271
+ alternative is allowed only occasionally and noncommercially, and
272
+ only if you received the object code with such an offer, in accord
273
+ with subsection 6b.
274
+
275
+ d) Convey the object code by offering access from a designated
276
+ place (gratis or for a charge), and offer equivalent access to the
277
+ Corresponding Source in the same way through the same place at no
278
+ further charge. You need not require recipients to copy the
279
+ Corresponding Source along with the object code. If the place to
280
+ copy the object code is a network server, the Corresponding Source
281
+ may be on a different server (operated by you or a third party)
282
+ that supports equivalent copying facilities, provided you maintain
283
+ clear directions next to the object code saying where to find the
284
+ Corresponding Source. Regardless of what server hosts the
285
+ Corresponding Source, you remain obligated to ensure that it is
286
+ available for as long as needed to satisfy these requirements.
287
+
288
+ e) Convey the object code using peer-to-peer transmission, provided
289
+ you inform other peers where the object code and Corresponding
290
+ Source of the work are being offered to the general public at no
291
+ charge under subsection 6d.
292
+
293
+ A separable portion of the object code, whose source code is excluded
294
+ from the Corresponding Source as a System Library, need not be
295
+ included in conveying the object code work.
296
+
297
+ A "User Product" is either (1) a "consumer product", which means any
298
+ tangible personal property which is normally used for personal, family,
299
+ or household purposes, or (2) anything designed or sold for incorporation
300
+ into a dwelling. In determining whether a product is a consumer product,
301
+ doubtful cases shall be resolved in favor of coverage. For a particular
302
+ product received by a particular user, "normally used" refers to a
303
+ typical or common use of that class of product, regardless of the status
304
+ of the particular user or of the way in which the particular user
305
+ actually uses, or expects or is expected to use, the product. A product
306
+ is a consumer product regardless of whether the product has substantial
307
+ commercial, industrial or non-consumer uses, unless such uses represent
308
+ the only significant mode of use of the product.
309
+
310
+ "Installation Information" for a User Product means any methods,
311
+ procedures, authorization keys, or other information required to install
312
+ and execute modified versions of a covered work in that User Product from
313
+ a modified version of its Corresponding Source. The information must
314
+ suffice to ensure that the continued functioning of the modified object
315
+ code is in no case prevented or interfered with solely because
316
+ modification has been made.
317
+
318
+ If you convey an object code work under this section in, or with, or
319
+ specifically for use in, a User Product, and the conveying occurs as
320
+ part of a transaction in which the right of possession and use of the
321
+ User Product is transferred to the recipient in perpetuity or for a
322
+ fixed term (regardless of how the transaction is characterized), the
323
+ Corresponding Source conveyed under this section must be accompanied
324
+ by the Installation Information. But this requirement does not apply
325
+ if neither you nor any third party retains the ability to install
326
+ modified object code on the User Product (for example, the work has
327
+ been installed in ROM).
328
+
329
+ The requirement to provide Installation Information does not include a
330
+ requirement to continue to provide support service, warranty, or updates
331
+ for a work that has been modified or installed by the recipient, or for
332
+ the User Product in which it has been modified or installed. Access to a
333
+ network may be denied when the modification itself materially and
334
+ adversely affects the operation of the network or violates the rules and
335
+ protocols for communication across the network.
336
+
337
+ Corresponding Source conveyed, and Installation Information provided,
338
+ in accord with this section must be in a format that is publicly
339
+ documented (and with an implementation available to the public in
340
+ source code form), and must require no special password or key for
341
+ unpacking, reading or copying.
342
+
343
+ 7. Additional Terms.
344
+
345
+ "Additional permissions" are terms that supplement the terms of this
346
+ License by making exceptions from one or more of its conditions.
347
+ Additional permissions that are applicable to the entire Program shall
348
+ be treated as though they were included in this License, to the extent
349
+ that they are valid under applicable law. If additional permissions
350
+ apply only to part of the Program, that part may be used separately
351
+ under those permissions, but the entire Program remains governed by
352
+ this License without regard to the additional permissions.
353
+
354
+ When you convey a copy of a covered work, you may at your option
355
+ remove any additional permissions from that copy, or from any part of
356
+ it. (Additional permissions may be written to require their own
357
+ removal in certain cases when you modify the work.) You may place
358
+ additional permissions on material, added by you to a covered work,
359
+ for which you have or can give appropriate copyright permission.
360
+
361
+ Notwithstanding any other provision of this License, for material you
362
+ add to a covered work, you may (if authorized by the copyright holders of
363
+ that material) supplement the terms of this License with terms:
364
+
365
+ a) Disclaiming warranty or limiting liability differently from the
366
+ terms of sections 15 and 16 of this License; or
367
+
368
+ b) Requiring preservation of specified reasonable legal notices or
369
+ author attributions in that material or in the Appropriate Legal
370
+ Notices displayed by works containing it; or
371
+
372
+ c) Prohibiting misrepresentation of the origin of that material, or
373
+ requiring that modified versions of such material be marked in
374
+ reasonable ways as different from the original version; or
375
+
376
+ d) Limiting the use for publicity purposes of names of licensors or
377
+ authors of the material; or
378
+
379
+ e) Declining to grant rights under trademark law for use of some
380
+ trade names, trademarks, or service marks; or
381
+
382
+ f) Requiring indemnification of licensors and authors of that
383
+ material by anyone who conveys the material (or modified versions of
384
+ it) with contractual assumptions of liability to the recipient, for
385
+ any liability that these contractual assumptions directly impose on
386
+ those licensors and authors.
387
+
388
+ All other non-permissive additional terms are considered "further
389
+ restrictions" within the meaning of section 10. If the Program as you
390
+ received it, or any part of it, contains a notice stating that it is
391
+ governed by this License along with a term that is a further
392
+ restriction, you may remove that term. If a license document contains
393
+ a further restriction but permits relicensing or conveying under this
394
+ License, you may add to a covered work material governed by the terms
395
+ of that license document, provided that the further restriction does
396
+ not survive such relicensing or conveying.
397
+
398
+ If you add terms to a covered work in accord with this section, you
399
+ must place, in the relevant source files, a statement of the
400
+ additional terms that apply to those files, or a notice indicating
401
+ where to find the applicable terms.
402
+
403
+ Additional terms, permissive or non-permissive, may be stated in the
404
+ form of a separately written license, or stated as exceptions;
405
+ the above requirements apply either way.
406
+
407
+ 8. Termination.
408
+
409
+ You may not propagate or modify a covered work except as expressly
410
+ provided under this License. Any attempt otherwise to propagate or
411
+ modify it is void, and will automatically terminate your rights under
412
+ this License (including any patent licenses granted under the third
413
+ paragraph of section 11).
414
+
415
+ However, if you cease all violation of this License, then your
416
+ license from a particular copyright holder is reinstated (a)
417
+ provisionally, unless and until the copyright holder explicitly and
418
+ finally terminates your license, and (b) permanently, if the copyright
419
+ holder fails to notify you of the violation by some reasonable means
420
+ prior to 60 days after the cessation.
421
+
422
+ Moreover, your license from a particular copyright holder is
423
+ reinstated permanently if the copyright holder notifies you of the
424
+ violation by some reasonable means, this is the first time you have
425
+ received notice of violation of this License (for any work) from that
426
+ copyright holder, and you cure the violation prior to 30 days after
427
+ your receipt of the notice.
428
+
429
+ Termination of your rights under this section does not terminate the
430
+ licenses of parties who have received copies or rights from you under
431
+ this License. If your rights have been terminated and not permanently
432
+ reinstated, you do not qualify to receive new licenses for the same
433
+ material under section 10.
434
+
435
+ 9. Acceptance Not Required for Having Copies.
436
+
437
+ You are not required to accept this License in order to receive or
438
+ run a copy of the Program. Ancillary propagation of a covered work
439
+ occurring solely as a consequence of using peer-to-peer transmission
440
+ to receive a copy likewise does not require acceptance. However,
441
+ nothing other than this License grants you permission to propagate or
442
+ modify any covered work. These actions infringe copyright if you do
443
+ not accept this License. Therefore, by modifying or propagating a
444
+ covered work, you indicate your acceptance of this License to do so.
445
+
446
+ 10. Automatic Licensing of Downstream Recipients.
447
+
448
+ Each time you convey a covered work, the recipient automatically
449
+ receives a license from the original licensors, to run, modify and
450
+ propagate that work, subject to this License. You are not responsible
451
+ for enforcing compliance by third parties with this License.
452
+
453
+ An "entity transaction" is a transaction transferring control of an
454
+ organization, or substantially all assets of one, or subdividing an
455
+ organization, or merging organizations. If propagation of a covered
456
+ work results from an entity transaction, each party to that
457
+ transaction who receives a copy of the work also receives whatever
458
+ licenses to the work the party's predecessor in interest had or could
459
+ give under the previous paragraph, plus a right to possession of the
460
+ Corresponding Source of the work from the predecessor in interest, if
461
+ the predecessor has it or can get it with reasonable efforts.
462
+
463
+ You may not impose any further restrictions on the exercise of the
464
+ rights granted or affirmed under this License. For example, you may
465
+ not impose a license fee, royalty, or other charge for exercise of
466
+ rights granted under this License, and you may not initiate litigation
467
+ (including a cross-claim or counterclaim in a lawsuit) alleging that
468
+ any patent claim is infringed by making, using, selling, offering for
469
+ sale, or importing the Program or any portion of it.
470
+
471
+ 11. Patents.
472
+
473
+ A "contributor" is a copyright holder who authorizes use under this
474
+ License of the Program or a work on which the Program is based. The
475
+ work thus licensed is called the contributor's "contributor version".
476
+
477
+ A contributor's "essential patent claims" are all patent claims
478
+ owned or controlled by the contributor, whether already acquired or
479
+ hereafter acquired, that would be infringed by some manner, permitted
480
+ by this License, of making, using, or selling its contributor version,
481
+ but do not include claims that would be infringed only as a
482
+ consequence of further modification of the contributor version. For
483
+ purposes of this definition, "control" includes the right to grant
484
+ patent sublicenses in a manner consistent with the requirements of
485
+ this License.
486
+
487
+ Each contributor grants you a non-exclusive, worldwide, royalty-free
488
+ patent license under the contributor's essential patent claims, to
489
+ make, use, sell, offer for sale, import and otherwise run, modify and
490
+ propagate the contents of its contributor version.
491
+
492
+ In the following three paragraphs, a "patent license" is any express
493
+ agreement or commitment, however denominated, not to enforce a patent
494
+ (such as an express permission to practice a patent or covenant not to
495
+ sue for patent infringement). To "grant" such a patent license to a
496
+ party means to make such an agreement or commitment not to enforce a
497
+ patent against the party.
498
+
499
+ If you convey a covered work, knowingly relying on a patent license,
500
+ and the Corresponding Source of the work is not available for anyone
501
+ to copy, free of charge and under the terms of this License, through a
502
+ publicly available network server or other readily accessible means,
503
+ then you must either (1) cause the Corresponding Source to be so
504
+ available, or (2) arrange to deprive yourself of the benefit of the
505
+ patent license for this particular work, or (3) arrange, in a manner
506
+ consistent with the requirements of this License, to extend the patent
507
+ license to downstream recipients. "Knowingly relying" means you have
508
+ actual knowledge that, but for the patent license, your conveying the
509
+ covered work in a country, or your recipient's use of the covered work
510
+ in a country, would infringe one or more identifiable patents in that
511
+ country that you have reason to believe are valid.
512
+
513
+ If, pursuant to or in connection with a single transaction or
514
+ arrangement, you convey, or propagate by procuring conveyance of, a
515
+ covered work, and grant a patent license to some of the parties
516
+ receiving the covered work authorizing them to use, propagate, modify
517
+ or convey a specific copy of the covered work, then the patent license
518
+ you grant is automatically extended to all recipients of the covered
519
+ work and works based on it.
520
+
521
+ A patent license is "discriminatory" if it does not include within
522
+ the scope of its coverage, prohibits the exercise of, or is
523
+ conditioned on the non-exercise of one or more of the rights that are
524
+ specifically granted under this License. You may not convey a covered
525
+ work if you are a party to an arrangement with a third party that is
526
+ in the business of distributing software, under which you make payment
527
+ to the third party based on the extent of your activity of conveying
528
+ the work, and under which the third party grants, to any of the
529
+ parties who would receive the covered work from you, a discriminatory
530
+ patent license (a) in connection with copies of the covered work
531
+ conveyed by you (or copies made from those copies), or (b) primarily
532
+ for and in connection with specific products or compilations that
533
+ contain the covered work, unless you entered into that arrangement,
534
+ or that patent license was granted, prior to 28 March 2007.
535
+
536
+ Nothing in this License shall be construed as excluding or limiting
537
+ any implied license or other defenses to infringement that may
538
+ otherwise be available to you under applicable patent law.
539
+
540
+ 12. No Surrender of Others' Freedom.
541
+
542
+ If conditions are imposed on you (whether by court order, agreement or
543
+ otherwise) that contradict the conditions of this License, they do not
544
+ excuse you from the conditions of this License. If you cannot convey a
545
+ covered work so as to satisfy simultaneously your obligations under this
546
+ License and any other pertinent obligations, then as a consequence you may
547
+ not convey it at all. For example, if you agree to terms that obligate you
548
+ to collect a royalty for further conveying from those to whom you convey
549
+ the Program, the only way you could satisfy both those terms and this
550
+ License would be to refrain entirely from conveying the Program.
551
+
552
+ 13. Use with the GNU Affero General Public License.
553
+
554
+ Notwithstanding any other provision of this License, you have
555
+ permission to link or combine any covered work with a work licensed
556
+ under version 3 of the GNU Affero General Public License into a single
557
+ combined work, and to convey the resulting work. The terms of this
558
+ License will continue to apply to the part which is the covered work,
559
+ but the special requirements of the GNU Affero General Public License,
560
+ section 13, concerning interaction through a network will apply to the
561
+ combination as such.
562
+
563
+ 14. Revised Versions of this License.
564
+
565
+ The Free Software Foundation may publish revised and/or new versions of
566
+ the GNU General Public License from time to time. Such new versions will
567
+ be similar in spirit to the present version, but may differ in detail to
568
+ address new problems or concerns.
569
+
570
+ Each version is given a distinguishing version number. If the
571
+ Program specifies that a certain numbered version of the GNU General
572
+ Public License "or any later version" applies to it, you have the
573
+ option of following the terms and conditions either of that numbered
574
+ version or of any later version published by the Free Software
575
+ Foundation. If the Program does not specify a version number of the
576
+ GNU General Public License, you may choose any version ever published
577
+ by the Free Software Foundation.
578
+
579
+ If the Program specifies that a proxy can decide which future
580
+ versions of the GNU General Public License can be used, that proxy's
581
+ public statement of acceptance of a version permanently authorizes you
582
+ to choose that version for the Program.
583
+
584
+ Later license versions may give you additional or different
585
+ permissions. However, no additional obligations are imposed on any
586
+ author or copyright holder as a result of your choosing to follow a
587
+ later version.
588
+
589
+ 15. Disclaimer of Warranty.
590
+
591
+ THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
592
+ APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
593
+ HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
594
+ OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
595
+ THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
596
+ PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
597
+ IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
598
+ ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
599
+
600
+ 16. Limitation of Liability.
601
+
602
+ IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
603
+ WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
604
+ THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
605
+ GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
606
+ USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
607
+ DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
608
+ PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
609
+ EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
610
+ SUCH DAMAGES.
611
+
612
+ 17. Interpretation of Sections 15 and 16.
613
+
614
+ If the disclaimer of warranty and limitation of liability provided
615
+ above cannot be given local legal effect according to their terms,
616
+ reviewing courts shall apply local law that most closely approximates
617
+ an absolute waiver of all civil liability in connection with the
618
+ Program, unless a warranty or assumption of liability accompanies a
619
+ copy of the Program in return for a fee.
620
+
621
+ END OF TERMS AND CONDITIONS
622
+
623
+ How to Apply These Terms to Your New Programs
624
+
625
+ If you develop a new program, and you want it to be of the greatest
626
+ possible use to the public, the best way to achieve this is to make it
627
+ free software which everyone can redistribute and change under these terms.
628
+
629
+ To do so, attach the following notices to the program. It is safest
630
+ to attach them to the start of each source file to most effectively
631
+ state the exclusion of warranty; and each file should have at least
632
+ the "copyright" line and a pointer to where the full notice is found.
633
+
634
+ <one line to give the program's name and a brief idea of what it does.>
635
+ Copyright (C) <year> <name of author>
636
+
637
+ This program is free software: you can redistribute it and/or modify
638
+ it under the terms of the GNU General Public License as published by
639
+ the Free Software Foundation, either version 3 of the License, or
640
+ (at your option) any later version.
641
+
642
+ This program is distributed in the hope that it will be useful,
643
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
644
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
645
+ GNU General Public License for more details.
646
+
647
+ You should have received a copy of the GNU General Public License
648
+ along with this program. If not, see <https://www.gnu.org/licenses/>.
649
+
650
+ Also add information on how to contact you by electronic and paper mail.
651
+
652
+ If the program does terminal interaction, make it output a short
653
+ notice like this when it starts in an interactive mode:
654
+
655
+ <program> Copyright (C) <year> <name of author>
656
+ This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
657
+ This is free software, and you are welcome to redistribute it
658
+ under certain conditions; type `show c' for details.
659
+
660
+ The hypothetical commands `show w' and `show c' should show the appropriate
661
+ parts of the General Public License. Of course, your program's commands
662
+ might be different; for a GUI interface, you would use an "about box".
663
+
664
+ You should also get your employer (if you work as a programmer) or school,
665
+ if any, to sign a "copyright disclaimer" for the program, if necessary.
666
+ For more information on this, and how to apply and follow the GNU GPL, see
667
+ <https://www.gnu.org/licenses/>.
668
+
669
+ The GNU General Public License does not permit incorporating your program
670
+ into proprietary programs. If your program is a subroutine library, you
671
+ may consider it more useful to permit linking proprietary applications with
672
+ the library. If this is what you want to do, use the GNU Lesser General
673
+ Public License instead of this License. But first, please read
674
+ <https://www.gnu.org/licenses/why-not-lgpl.html>.
README.md ADDED
@@ -0,0 +1,373 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ title: GPT-Academic
3
+ emoji: 😻
4
+ colorFrom: blue
5
+ colorTo: blue
6
+ sdk: gradio
7
+ sdk_version: 3.32.0
8
+ app_file: app.py
9
+ pinned: false
10
+ duplicated_from: qingxu98/gpt-academic
11
+ ---
12
+
13
+ # ChatGPT 学术优化
14
+ > **Note**
15
+ >
16
+ > 2023.7.8: Gradio, Pydantic依赖调整,已修改 `requirements.txt`。请及时**更新代码**,安装依赖时,请严格选择`requirements.txt`中**指定的版本**
17
+ >
18
+ > `pip install -r requirements.txt`
19
+
20
+
21
+ # <div align=center><img src="docs/logo.png" width="40"> GPT 学术优化 (GPT Academic)</div>
22
+
23
+ **如果喜欢这个项目,请给它一个Star;如果您发明了好用的快捷键或函数插件,欢迎发pull requests!**
24
+
25
+ If you like this project, please give it a Star. If you've come up with more useful academic shortcuts or functional plugins, feel free to open an issue or pull request. We also have a README in [English|](docs/README_EN.md)[日本語|](docs/README_JP.md)[한국어|](https://github.com/mldljyh/ko_gpt_academic)[Русский|](docs/README_RS.md)[Français](docs/README_FR.md) translated by this project itself.
26
+ To translate this project to arbitary language with GPT, read and run [`multi_language.py`](multi_language.py) (experimental).
27
+
28
+ > **Note**
29
+ >
30
+ > 1.请注意只有 **高亮(如红色)** 标识的函数插件(按钮)才支持读取文件,部分插件位于插件区的**下拉菜单**中。另外我们以**最高优先级**欢迎和处理任何新插件的PR。
31
+ >
32
+ > 2.本项目中每个文件的功能都在自译解[`self_analysis.md`](https://github.com/binary-husky/gpt_academic/wiki/chatgpt-academic%E9%A1%B9%E7%9B%AE%E8%87%AA%E8%AF%91%E8%A7%A3%E6%8A%A5%E5%91%8A)详细说明。随着版本的迭代,您也可以随时自行点击相关函数插件,调用GPT重新生成项目的自我解析报告。常见问题汇总在[`wiki`](https://github.com/binary-husky/gpt_academic/wiki/%E5%B8%B8%E8%A7%81%E9%97%AE%E9%A2%98)当中。[安装方法](#installation)。
33
+ >
34
+ > 3.本项目兼容并鼓励尝试国产大语言模型ChatGLM和Moss等等。支持多个api-key共存,可在配置文件中填写如`API_KEY="openai-key1,openai-key2,azure-key3,api2d-key4"`。需要临时更换`API_KEY`时,在输入区输入临时的`API_KEY`然后回车键提交后即可生效。
35
+
36
+
37
+
38
+
39
+ <div align="center">
40
+
41
+ 功能(⭐= 近期新增功能) | 描述
42
+ --- | ---
43
+ ⭐[接入新模型](https://github.com/binary-husky/gpt_academic/wiki/%E5%A6%82%E4%BD%95%E5%88%87%E6%8D%A2%E6%A8%A1%E5%9E%8B)! | 百度[千帆](https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Nlks5zkzu)与文心一言, [通义千问](https://modelscope.cn/models/qwen/Qwen-7B-Chat/summary),上海AI-Lab[书生](https://github.com/InternLM/InternLM),讯飞[星火](https://xinghuo.xfyun.cn/),[LLaMa2](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)
44
+ 一键润色 | 支持一键润色、一键查找论文语法错误
45
+ 一键中英互译 | 一键中英互译
46
+ 一键代码解释 | 显示代码、解释代码、生成代码、给代码加注释
47
+ [自定义快捷键](https://www.bilibili.com/video/BV14s4y1E7jN) | 支持自定义快捷键
48
+ 模块化设计 | 支持自定义强大的[函数插件](https://github.com/binary-husky/gpt_academic/tree/master/crazy_functions),插件支持[热更新](https://github.com/binary-husky/gpt_academic/wiki/%E5%87%BD%E6%95%B0%E6%8F%92%E4%BB%B6%E6%8C%87%E5%8D%97)
49
+ [自我程序剖析](https://www.bilibili.com/video/BV1cj411A7VW) | [函数插件] [一键读懂](https://github.com/binary-husky/gpt_academic/wiki/chatgpt-academic%E9%A1%B9%E7%9B%AE%E8%87%AA%E8%AF%91%E8%A7%A3%E6%8A%A5%E5%91%8A)本项目的源代码
50
+ [程序剖析](https://www.bilibili.com/video/BV1cj411A7VW) | [函数插件] 一键可以剖析其他Python/C/C++/Java/Lua/...项目树
51
+ 读论文、[翻译](https://www.bilibili.com/video/BV1KT411x7Wn)论文 | [函数插件] 一键解读latex/pdf论文全文并生成摘要
52
+ Latex全文[翻译](https://www.bilibili.com/video/BV1nk4y1Y7Js/)、[润色](https://www.bilibili.com/video/BV1FT411H7c5/) | [函数插件] 一键翻译或润色latex论文
53
+ 批量注释生成 | [函数插件] 一键批量生成函数注释
54
+ Markdown[中英互译](https://www.bilibili.com/video/BV1yo4y157jV/) | [函数插件] 看到上面5种语言的[README](https://github.com/binary-husky/gpt_academic/blob/master/docs/README_EN.md)了吗?
55
+ chat分析报告生成 | [函数插件] 运行后自动生成总结汇报
56
+ [PDF论文全文翻译功能](https://www.bilibili.com/video/BV1KT411x7Wn) | [函数插件] PDF论文提取题目&摘要+翻译全文(多线程)
57
+ [Arxiv小助手](https://www.bilibili.com/video/BV1LM4y1279X) | [函数插件] 输入arxiv文章url即可一键翻译摘要+下载PDF
58
+ Latex论文一键校对 | [函数插件] 仿Grammarly对Latex文章进行语法、拼写纠错+输出对照PDF
59
+ [谷歌学术统合小助手](https://www.bilibili.com/video/BV19L411U7ia) | [函数插件] 给定任意谷歌学术搜索页面URL,让gpt帮你[写relatedworks](https://www.bilibili.com/video/BV1GP411U7Az/)
60
+ 互联网信息聚合+GPT | [函数插件] 一键[让GPT从互联网获取信息](https://www.bilibili.com/video/BV1om4y127ck)回答问题,让信息永不过时
61
+ ⭐Arxiv论文精细翻译 ([Docker](https://github.com/binary-husky/gpt_academic/pkgs/container/gpt_academic_with_latex)) | [函数插件] 一键[以超高质量翻译arxiv论文](https://www.bilibili.com/video/BV1dz4y1v77A/),目前最好的论文翻译工具
62
+ ⭐[实时语音对话输入](https://github.com/binary-husky/gpt_academic/blob/master/docs/use_audio.md) | [函数插件] 异步[监听音频](https://www.bilibili.com/video/BV1AV4y187Uy/),自动断句,自动寻找回答时机
63
+ 公式/图片/表格显示 | 可以同时显示公式的[tex形式和渲染形式](https://user-images.githubusercontent.com/96192199/230598842-1d7fcddd-815d-40ee-af60-baf488a199df.png),支持公式、代码高亮
64
+ 多线程函数插件支持 | 支持多线调用chatgpt,一键处理[海量文本](https://www.bilibili.com/video/BV1FT411H7c5/)或程序
65
+ 启动暗色[主题](https://github.com/binary-husky/gpt_academic/issues/173) | 在浏览器url后面添加```/?__theme=dark```可以切换dark主题
66
+ [多LLM模型](https://www.bilibili.com/video/BV1wT411p7yf)支持 | 同时被GPT3.5、GPT4、[清华ChatGLM2](https://github.com/THUDM/ChatGLM2-6B)、[复旦MOSS](https://github.com/OpenLMLab/MOSS)同时伺候的感觉一定会很不错吧?
67
+ ⭐ChatGLM2微调模型 | 支持加载ChatGLM2微调模型,提供ChatGLM2微调辅助插件
68
+ 更多LLM模型接入,支持[huggingface部署](https://huggingface.co/spaces/qingxu98/gpt-academic) | 加入Newbing接口(新必应),引入清华[Jittorllms](https://github.com/Jittor/JittorLLMs)支持[LLaMA](https://github.com/facebookresearch/llama)和[盘古α](https://openi.org.cn/pangu/)
69
+ ⭐[虚空终端](https://github.com/binary-husky/void-terminal)pip包 | 脱离GUI,在Python中直接调用本项目的函数插件(开发中)
70
+ 更多新功能展示 (图像生成等) …… | 见本文档结尾处 ……
71
+ </div>
72
+
73
+
74
+ - 新界面(修改`config.py`中的LAYOUT选项即可实现“左右布局”和“上下布局”的切换)
75
+ <div align="center">
76
+ <img src="https://user-images.githubusercontent.com/96192199/230361456-61078362-a966-4eb5-b49e-3c62ef18b860.gif" width="700" >
77
+ </div>
78
+
79
+
80
+ - 所有按钮都通过读取functional.py动态生成,可随意加自定义功能,解放粘贴板
81
+ <div align="center">
82
+ <img src="https://user-images.githubusercontent.com/96192199/231975334-b4788e91-4887-412f-8b43-2b9c5f41d248.gif" width="700" >
83
+ </div>
84
+
85
+ - 润色/纠错
86
+ <div align="center">
87
+ <img src="https://user-images.githubusercontent.com/96192199/231980294-f374bdcb-3309-4560-b424-38ef39f04ebd.gif" width="700" >
88
+ </div>
89
+
90
+ - 如果输出包含公式,会同时以tex形式和渲染形式显示,方便复制和阅读
91
+ <div align="center">
92
+ <img src="https://user-images.githubusercontent.com/96192199/230598842-1d7fcddd-815d-40ee-af60-baf488a199df.png" width="700" >
93
+ </div>
94
+
95
+ - 懒得看项目代码?整个工程直接给chatgpt炫嘴里
96
+ <div align="center">
97
+ <img src="https://user-images.githubusercontent.com/96192199/226935232-6b6a73ce-8900-4aee-93f9-733c7e6fef53.png" width="700" >
98
+ </div>
99
+
100
+ - 多种大语言模型混合调用(ChatGLM + OpenAI-GPT3.5 + [API2D](https://api2d.com/)-GPT4)
101
+ <div align="center">
102
+ <img src="https://user-images.githubusercontent.com/96192199/232537274-deca0563-7aa6-4b5d-94a2-b7c453c47794.png" width="700" >
103
+ </div>
104
+
105
+ # Installation
106
+ ### 安装方法I:直接运行 (Windows, Linux or MacOS)
107
+
108
+ 1. 下载项目
109
+ ```sh
110
+ git clone --depth=1 https://github.com/binary-husky/gpt_academic.git
111
+ cd gpt_academic
112
+ ```
113
+
114
+ 2. 配置API_KEY
115
+
116
+ 在`config.py`中,配置API KEY等设置,[点击查看特殊网络环境设置方法](https://github.com/binary-husky/gpt_academic/issues/1) 。
117
+
118
+ (P.S. 程序运行时会优先检查是否存在名为`config_private.py`的私密配置文件,并用其中的配置覆盖`config.py`的同名配置。因此,如果您能理解我们的配置读取逻辑,我们强烈建议您在`config.py`旁边创建一个名为`config_private.py`的新配置文件,并把`config.py`中的配置转移(复制)到`config_private.py`中。`config_private.py`不受git管控,可以让您的隐私信息更加安全。P.S.项目同样支持通过`环境变量`配置大多数选项,环境变量的书写格式参考`docker-compose`文件。读取优先级: `环境变量` > `config_private.py` > `config.py`)
119
+
120
+
121
+ 3. 安装依赖
122
+ ```sh
123
+ # (选择I: 如熟悉python)(python版本3.9以上,越新越好),备注:使用官方pip源或者阿里pip源,临时换源方法:python -m pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/
124
+ python -m pip install -r requirements.txt
125
+
126
+ # (选择II: 如不熟悉python)使用anaconda,步骤也是类似的 (https://www.bilibili.com/video/BV1rc411W7Dr):
127
+ conda create -n gptac_venv python=3.11 # 创建anaconda环境
128
+ conda activate gptac_venv # 激活anaconda环境
129
+ python -m pip install -r requirements.txt # 这个步骤和pip安装一样的步骤
130
+ ```
131
+
132
+
133
+ <details><summary>如果需要支持清华ChatGLM2/复旦MOSS/RWKV作为后端,请点击展开此处</summary>
134
+ <p>
135
+
136
+ 【可选步骤】如果需要支持清华ChatGLM2/复旦MOSS作为后端,需要额外安装更多依赖(前提条件:熟悉Python + 用过Pytorch + 电脑配置够强):
137
+ ```sh
138
+ # 【可选步骤I】支持清华ChatGLM2。清华ChatGLM备注:如果遇到"Call ChatGLM fail 不能正常加载ChatGLM的参数" 错误,参考如下: 1:以上默认安装的为torch+cpu版,使用cuda需要卸载torch重新安装torch+cuda; 2:如因本机配置不够无法加载模型,可以修改request_llm/bridge_chatglm.py中的模型精度, 将 AutoTokenizer.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True) 都修改为 AutoTokenizer.from_pretrained("THUDM/chatglm-6b-int4", trust_remote_code=True)
139
+ python -m pip install -r request_llm/requirements_chatglm.txt
140
+
141
+ # 【可选步骤II】支持复旦MOSS
142
+ python -m pip install -r request_llm/requirements_moss.txt
143
+ git clone --depth=1 https://github.com/OpenLMLab/MOSS.git request_llm/moss # 注意执行此行代码时,必须处于项目根路径
144
+
145
+ # 【可选步骤III】支持RWKV Runner
146
+ 参考wiki:https://github.com/binary-husky/gpt_academic/wiki/%E9%80%82%E9%85%8DRWKV-Runner
147
+
148
+ # 【可选步骤IV】确保config.py配置文件的AVAIL_LLM_MODELS包含了期望的模型,目前支持的全部模型如下(jittorllms系列目前仅支持docker方案):
149
+ AVAIL_LLM_MODELS = ["gpt-3.5-turbo", "api2d-gpt-3.5-turbo", "gpt-4", "api2d-gpt-4", "chatglm", "newbing", "moss"] # + ["jittorllms_rwkv", "jittorllms_pangualpha", "jittorllms_llama"]
150
+ ```
151
+
152
+ </p>
153
+ </details>
154
+
155
+
156
+
157
+ 4. 运行
158
+ ```sh
159
+ python main.py
160
+ ```
161
+
162
+ ### 安装方法II:使用Docker
163
+
164
+ 1. 仅ChatGPT(推荐大多数人选择,等价于docker-compose方案1)
165
+ [![basic](https://github.com/binary-husky/gpt_academic/actions/workflows/build-without-local-llms.yml/badge.svg?branch=master)](https://github.com/binary-husky/gpt_academic/actions/workflows/build-without-local-llms.yml)
166
+ [![basiclatex](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-latex.yml/badge.svg?branch=master)](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-latex.yml)
167
+ [![basicaudio](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-audio-assistant.yml/badge.svg?branch=master)](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-audio-assistant.yml)
168
+
169
+ ``` sh
170
+ git clone --depth=1 https://github.com/binary-husky/gpt_academic.git # 下载项目
171
+ cd gpt_academic # 进入路径
172
+ nano config.py # 用任意文本编辑器编辑config.py, 配置 “Proxy”, “API_KEY” 以及 “WEB_PORT” (例如50923) 等
173
+ docker build -t gpt-academic . # 安装
174
+
175
+ #(最后一步-Linux操作系统)用`--net=host`更方便快捷
176
+ docker run --rm -it --net=host gpt-academic
177
+ #(最后一步-MacOS/Windows操作系统)只能用-p选项将容器上的端口(例如50923)暴露给主机上的端口
178
+ docker run --rm -it -e WEB_PORT=50923 -p 50923:50923 gpt-academic
179
+ ```
180
+ P.S. 如果需要依赖Latex的插件功能,请见Wiki。另外,您也可以直接使用docker-compose获取Latex功能(修改docker-compose.yml,保留方案4并删除其他方案)。
181
+
182
+ 2. ChatGPT + ChatGLM2 + MOSS + LLAMA2 + 通义千问(需要熟悉[Nvidia Docker](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#installing-on-ubuntu-and-debian)运行时)
183
+ [![chatglm](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-chatglm.yml/badge.svg?branch=master)](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-chatglm.yml)
184
+
185
+ ``` sh
186
+ # 修改docker-compose.yml,保留方案2并删除其他方案。修改docker-compose.yml中方案2的配置,参考其中注释即可
187
+ docker-compose up
188
+ ```
189
+
190
+ 3. ChatGPT + LLAMA + 盘古 + RWKV(需要熟悉[Nvidia Docker](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#installing-on-ubuntu-and-debian)运行时)
191
+ [![jittorllms](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-jittorllms.yml/badge.svg?branch=master)](https://github.com/binary-husky/gpt_academic/actions/workflows/build-with-jittorllms.yml)
192
+
193
+ ``` sh
194
+ # 修改docker-compose.yml,保留方案3并删除其他方案。修改docker-compose.yml中方案3的配置,参考其中注释即可
195
+ docker-compose up
196
+ ```
197
+
198
+
199
+ ### 安装方法III:其他部署姿势
200
+ 1. 一键运行脚本。
201
+ 完全不熟悉python环境的Windows用户可以下载[Release](https://github.com/binary-husky/gpt_academic/releases)中发布的一键运行脚本安装无本地模型的版本。
202
+ 脚本的贡献来源是[oobabooga](https://github.com/oobabooga/one-click-installers)。
203
+
204
+ 2. 使用docker-compose运行。
205
+ 请阅读docker-compose.yml后,按照其中的提示操作即可
206
+
207
+ 3. 如何使用反代URL
208
+ 按照`config.py`中的说明配置API_URL_REDIRECT即可。
209
+
210
+ 4. 微软云AzureAPI
211
+ 按照`config.py`中的说明配置即可(AZURE_ENDPOINT等四个配置)
212
+
213
+ 5. 远程云服务器部署(需要云服务器知识与经验)。
214
+ 请访问[部署wiki-1](https://github.com/binary-husky/gpt_academic/wiki/%E4%BA%91%E6%9C%8D%E5%8A%A1%E5%99%A8%E8%BF%9C%E7%A8%8B%E9%83%A8%E7%BD%B2%E6%8C%87%E5%8D%97)
215
+
216
+ 6. 使用Sealos[一键部署](https://github.com/binary-husky/gpt_academic/issues/993)。
217
+
218
+ 7. 使用WSL2(Windows Subsystem for Linux 子系统)。
219
+ 请访问[部署wiki-2](https://github.com/binary-husky/gpt_academic/wiki/%E4%BD%BF%E7%94%A8WSL2%EF%BC%88Windows-Subsystem-for-Linux-%E5%AD%90%E7%B3%BB%E7%BB%9F%EF%BC%89%E9%83%A8%E7%BD%B2)
220
+
221
+ 8. 如何在二级网址(如`http://localhost/subpath`)下运行。
222
+ 请访问[FastAPI运行说明](docs/WithFastapi.md)
223
+
224
+
225
+ # Advanced Usage
226
+ ### I:自定义新的便捷按钮(学术快捷键)
227
+ 任意文本编辑器打开`core_functional.py`,添加条目如下,然后重启程序即可。(如果按钮已经添加成功并可见,那么前缀、后缀都支持热修改,无需重启程序即可生效。)
228
+ 例如
229
+ ```
230
+ "超级英译中": {
231
+ # 前缀,会被加在你的输入之前。例如,用来描述你的要求,例如翻译、解释代码、润色等等
232
+ "Prefix": "请翻译把下面一段内容成中文,然后用一个markdown表格逐一解释文中出现的专有名词:\n\n",
233
+
234
+ # 后缀,会被加在你的输入之后。例如,配合前缀可以把你的输入内容用引号圈起来。
235
+ "Suffix": "",
236
+ },
237
+ ```
238
+ <div align="center">
239
+ <img src="https://user-images.githubusercontent.com/96192199/226899272-477c2134-ed71-4326-810c-29891fe4a508.png" width="500" >
240
+ </div>
241
+
242
+ ### II:自定义函数插件
243
+
244
+ 编写强大的函数插件来执行任何你想得到的和想不到的任务。
245
+ 本项目的插件编写、调试难度很低,只要您具备一定的python基础知识,就可以仿照我们提供的模板实现自己的插件功能。
246
+ 详情请参考[函数插件指南](https://github.com/binary-husky/gpt_academic/wiki/%E5%87%BD%E6%95%B0%E6%8F%92%E4%BB%B6%E6%8C%87%E5%8D%97)。
247
+
248
+
249
+ # Latest Update
250
+ ### I:新功能动态
251
+
252
+ 1. 对话保存功能。在函数插件区调用 `保存当前的对话` 即可将当前对话保存为可读+可复原的html文件,
253
+ 另外在函数插件区(下拉菜单)调用 `载入对话历史存档` ,即可还原之前的会话。
254
+ Tip:不指定文件直接点击 `载入对话历史存档` 可以查看历史html存档缓存。
255
+ <div align="center">
256
+ <img src="https://user-images.githubusercontent.com/96192199/235222390-24a9acc0-680f-49f5-bc81-2f3161f1e049.png" width="500" >
257
+ </div>
258
+
259
+ 2. ⭐Latex/Arxiv论文翻译功能⭐
260
+ <div align="center">
261
+ <img src="https://github.com/binary-husky/gpt_academic/assets/96192199/002a1a75-ace0-4e6a-94e2-ec1406a746f1" height="250" > ===>
262
+ <img src="https://github.com/binary-husky/gpt_academic/assets/96192199/9fdcc391-f823-464f-9322-f8719677043b" height="250" >
263
+ </div>
264
+
265
+ 3. 生成报告。大部分插件都会在执行结束后,生成工作报告
266
+ <div align="center">
267
+ <img src="https://user-images.githubusercontent.com/96192199/227503770-fe29ce2c-53fd-47b0-b0ff-93805f0c2ff4.png" height="250" >
268
+ <img src="https://user-images.githubusercontent.com/96192199/227504617-7a497bb3-0a2a-4b50-9a8a-95ae60ea7afd.png" height="250" >
269
+ </div>
270
+
271
+ 4. 模块化功能设计,简单的接口却能支持强大的功能
272
+ <div align="center">
273
+ <img src="https://user-images.githubusercontent.com/96192199/229288270-093643c1-0018-487a-81e6-1d7809b6e90f.png" height="400" >
274
+ <img src="https://user-images.githubusercontent.com/96192199/227504931-19955f78-45cd-4d1c-adac-e71e50957915.png" height="400" >
275
+ </div>
276
+
277
+ 5. 译解其他开源项目
278
+ <div align="center">
279
+ <img src="https://user-images.githubusercontent.com/96192199/226935232-6b6a73ce-8900-4aee-93f9-733c7e6fef53.png" height="250" >
280
+ <img src="https://user-images.githubusercontent.com/96192199/226969067-968a27c1-1b9c-486b-8b81-ab2de8d3f88a.png" height="250" >
281
+ </div>
282
+
283
+ 6. 装饰[live2d](https://github.com/fghrsh/live2d_demo)的小功能(默认关闭,需要修改`config.py`)
284
+ <div align="center">
285
+ <img src="https://user-images.githubusercontent.com/96192199/236432361-67739153-73e8-43fe-8111-b61296edabd9.png" width="500" >
286
+ </div>
287
+
288
+ 7. 新增MOSS大语言模型支持
289
+ <div align="center">
290
+ <img src="https://user-images.githubusercontent.com/96192199/236639178-92836f37-13af-4fdd-984d-b4450fe30336.png" width="500" >
291
+ </div>
292
+
293
+ 8. OpenAI图像生成
294
+ <div align="center">
295
+ <img src="https://github.com/binary-husky/gpt_academic/assets/96192199/bc7ab234-ad90-48a0-8d62-f703d9e74665" width="500" >
296
+ </div>
297
+
298
+ 9. OpenAI音频解析与总结
299
+ <div align="center">
300
+ <img src="https://github.com/binary-husky/gpt_academic/assets/96192199/709ccf95-3aee-498a-934a-e1c22d3d5d5b" width="500" >
301
+ </div>
302
+
303
+ 10. Latex全文校对纠错
304
+ <div align="center">
305
+ <img src="https://github.com/binary-husky/gpt_academic/assets/96192199/651ccd98-02c9-4464-91e1-77a6b7d1b033" height="200" > ===>
306
+ <img src="https://github.com/binary-husky/gpt_academic/assets/96192199/476f66d9-7716-4537-b5c1-735372c25adb" height="200">
307
+ </div>
308
+
309
+ 11. 语言、主题切换
310
+ <div align="center">
311
+ <img src="https://github.com/binary-husky/gpt_academic/assets/96192199/b6799499-b6fb-4f0c-9c8e-1b441872f4e8" width="500" >
312
+ </div>
313
+
314
+
315
+ ### II:版本:
316
+ - version 3.5(Todo): 使用自然语言调用本项目的所有函数插件(高优先级)
317
+ - version 3.49: 支持百度千帆平台和文心一言
318
+ - version 3.48: 支持阿里达摩院通义千问,上海AI-Lab书生,讯飞星火
319
+ - version 3.46: 支持完全脱手操作的实时语音对话
320
+ - version 3.45: 支持自定义ChatGLM2微调模型
321
+ - version 3.44: 正式支持Azure,优化界面易用性
322
+ - version 3.4: +arxiv论文翻译、latex论文批改功能
323
+ - version 3.3: +互联网信息综合功能
324
+ - version 3.2: 函数插件支持更多参数接口 (保存对话功能, 解读任意语言代码+同时询问任意的LLM组合)
325
+ - version 3.1: 支持同时问询多个gpt模型!支持api2d,支持多个apikey负载均衡
326
+ - version 3.0: 对chatglm和其他小型llm的支持
327
+ - version 2.6: 重构了插件结构,提高了交互性,加入更多插件
328
+ - version 2.5: 自更新,解决总结大工程源代码时文本过长、token溢出的问题
329
+ - version 2.4: (1)新增PDF全文翻译功能; (2)新增输入区切换位置的功能; (3)新增垂直布局选项; (4)多线程函数插件优化。
330
+ - version 2.3: 增强多线程交互性
331
+ - version 2.2: 函数插件支持热重载
332
+ - version 2.1: 可折叠式布局
333
+ - version 2.0: 引入模块化函数插件
334
+ - version 1.0: 基础功能
335
+
336
+ gpt_academic开发者QQ群-2:610599535
337
+
338
+ - 已知问题
339
+ - 某些浏览器翻译插件干扰此软件前端的运行
340
+ - 官方Gradio目前有很多兼容性Bug,请务必使用`requirement.txt`安装Gradio
341
+
342
+ ### III:主题
343
+ 可以通过修改`THEME`选项(config.py)变更主题
344
+ 1. `Chuanhu-Small-and-Beautiful` [网址](https://github.com/GaiZhenbiao/ChuanhuChatGPT/)
345
+
346
+
347
+ ### IV:参考与学习
348
+
349
+ ```
350
+ 代码中参考了很多其他优秀项目中的设计,顺序不分先后:
351
+
352
+ # 清华ChatGLM2-6B:
353
+ https://github.com/THUDM/ChatGLM2-6B
354
+
355
+ # 清华JittorLLMs:
356
+ https://github.com/Jittor/JittorLLMs
357
+
358
+ # ChatPaper:
359
+ https://github.com/kaixindelele/ChatPaper
360
+
361
+ # Edge-GPT:
362
+ https://github.com/acheong08/EdgeGPT
363
+
364
+ # ChuanhuChatGPT:
365
+ https://github.com/GaiZhenbiao/ChuanhuChatGPT
366
+
367
+ # Oobabooga one-click installer:
368
+ https://github.com/oobabooga/one-click-installers
369
+
370
+ # More:
371
+ https://github.com/gradio-app/gradio
372
+ https://github.com/fghrsh/live2d_demo
373
+ ```
app.py ADDED
@@ -0,0 +1,237 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os; os.environ['no_proxy'] = '*' # 避免代理网络产生意外污染
2
+
3
+ def main():
4
+ import subprocess, sys
5
+ subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'gradio-stable-fork'])
6
+ import gradio as gr
7
+ from request_llm.bridge_all import predict
8
+ from toolbox import format_io, find_free_port, on_file_uploaded, on_report_generated, get_conf, ArgsGeneralWrapper, load_chat_cookies, DummyWith
9
+ # 建议您复制一个config_private.py放自己的秘密, 如API和代理网址, 避免不小心传github被别人看到
10
+ proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT, LAYOUT, AVAIL_LLM_MODELS, AUTO_CLEAR_TXT = \
11
+ get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION', 'CHATBOT_HEIGHT', 'LAYOUT', 'AVAIL_LLM_MODELS', 'AUTO_CLEAR_TXT')
12
+ ENABLE_AUDIO, AUTO_CLEAR_TXT = get_conf('ENABLE_AUDIO', 'AUTO_CLEAR_TXT')
13
+ # 如果WEB_PORT是-1, 则随机选取WEB端口
14
+ PORT = find_free_port() if WEB_PORT <= 0 else WEB_PORT
15
+ if not AUTHENTICATION: AUTHENTICATION = None
16
+
17
+ from check_proxy import get_current_version
18
+ from themes.theme import adjust_theme, advanced_css, theme_declaration
19
+ initial_prompt = "Serve me as a writing and programming assistant."
20
+ title_html = f"<h1 align=\"center\">GPT 学术优化 {get_current_version()}</h1>{theme_declaration}"
21
+ description = """代码开源和更新[地址🚀](https://github.com/binary-husky/chatgpt_academic),感谢热情的[开发者们❤️](https://github.com/binary-husky/chatgpt_academic/graphs/contributors)"""
22
+
23
+ # 问询记录, python 版本建议3.9+(越新越好)
24
+ import logging, uuid
25
+ os.makedirs("gpt_log", exist_ok=True)
26
+ try:logging.basicConfig(filename="gpt_log/chat_secrets.log", level=logging.INFO, encoding="utf-8", format="%(asctime)s %(levelname)-8s %(message)s", datefmt="%Y-%m-%d %H:%M:%S")
27
+ except:logging.basicConfig(filename="gpt_log/chat_secrets.log", level=logging.INFO, format="%(asctime)s %(levelname)-8s %(message)s", datefmt="%Y-%m-%d %H:%M:%S")
28
+ # Disable logging output from the 'httpx' logger
29
+ logging.getLogger("httpx").setLevel(logging.WARNING)
30
+ print("所有问询记录将自动保存在本地目录./gpt_log/chat_secrets.log, 请注意自我隐私保护哦!")
31
+
32
+ # 一些普通功能模块
33
+ from core_functional import get_core_functions
34
+ functional = get_core_functions()
35
+
36
+ # 高级函数插件
37
+ from crazy_functional import get_crazy_functions
38
+ crazy_fns = get_crazy_functions()
39
+
40
+ # 处理markdown文本格式的转变
41
+ gr.Chatbot.postprocess = format_io
42
+
43
+ # 做一些外观色彩上的调整
44
+ set_theme = adjust_theme()
45
+
46
+ # 代理与自动更新
47
+ from check_proxy import check_proxy, auto_update, warm_up_modules
48
+ proxy_info = check_proxy(proxies)
49
+
50
+ gr_L1 = lambda: gr.Row().style()
51
+ gr_L2 = lambda scale, elem_id: gr.Column(scale=scale, elem_id=elem_id)
52
+ if LAYOUT == "TOP-DOWN":
53
+ gr_L1 = lambda: DummyWith()
54
+ gr_L2 = lambda scale, elem_id: gr.Row()
55
+ CHATBOT_HEIGHT /= 2
56
+
57
+ cancel_handles = []
58
+ with gr.Blocks(title="GPT 学术优化", theme=set_theme, analytics_enabled=False, css=advanced_css) as demo:
59
+ gr.HTML(title_html)
60
+ gr.HTML('''<center><a href="https://huggingface.co/spaces/qingxu98/gpt-academic?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>请您打开此页面后务必点击上方的“复制空间”(Duplicate Space)按钮!<font color="#FF00FF">使用时,先在输入框填入API-KEY然后回车。</font><br/>切忌在“复制空间”(Duplicate Space)之前填入API_KEY或进行提问,否则您的API_KEY将极可能被空间所有者攫取!<br/>支持任意数量的OpenAI的密钥和API2D的密钥共存,例如输入"OpenAI密钥1,API2D密钥2",然后提交,即可同时使用两种模型接口。</center>''')
61
+ cookies = gr.State(load_chat_cookies())
62
+ with gr_L1():
63
+ with gr_L2(scale=2, elem_id="gpt-chat"):
64
+ chatbot = gr.Chatbot(label=f"当前模型:{LLM_MODEL}", elem_id="gpt-chatbot")
65
+ if LAYOUT == "TOP-DOWN": chatbot.style(height=CHATBOT_HEIGHT)
66
+ history = gr.State([])
67
+ with gr_L2(scale=1, elem_id="gpt-panel"):
68
+ with gr.Accordion("输入区", open=True, elem_id="input-panel") as area_input_primary:
69
+ with gr.Row():
70
+ txt = gr.Textbox(show_label=False, lines=2, placeholder="输入问题或API密钥,输入多个密钥时,用英文逗号间隔。支持OpenAI密钥和API2D密钥共存。").style(container=False)
71
+ with gr.Row():
72
+ submitBtn = gr.Button("提交", variant="primary")
73
+ with gr.Row():
74
+ resetBtn = gr.Button("重置", variant="secondary"); resetBtn.style(size="sm")
75
+ stopBtn = gr.Button("停止", variant="secondary"); stopBtn.style(size="sm")
76
+ clearBtn = gr.Button("清除", variant="secondary", visible=False); clearBtn.style(size="sm")
77
+ if ENABLE_AUDIO:
78
+ with gr.Row():
79
+ audio_mic = gr.Audio(source="microphone", type="numpy", streaming=True, show_label=False).style(container=False)
80
+ with gr.Row():
81
+ status = gr.Markdown(f"Tip: 按Enter提交, 按Shift+Enter换行。当前模型: {LLM_MODEL} \n {proxy_info}", elem_id="state-panel")
82
+ with gr.Accordion("基础功能区", open=True, elem_id="basic-panel") as area_basic_fn:
83
+ with gr.Row():
84
+ for k in functional:
85
+ if ("Visible" in functional[k]) and (not functional[k]["Visible"]): continue
86
+ variant = functional[k]["Color"] if "Color" in functional[k] else "secondary"
87
+ functional[k]["Button"] = gr.Button(k, variant=variant)
88
+ with gr.Accordion("函数插件区", open=True, elem_id="plugin-panel") as area_crazy_fn:
89
+ with gr.Row():
90
+ gr.Markdown("插件可读取“输入区”文本/路径作为参数(上传文件自动修正路径)")
91
+ with gr.Row():
92
+ for k in crazy_fns:
93
+ if not crazy_fns[k].get("AsButton", True): continue
94
+ variant = crazy_fns[k]["Color"] if "Color" in crazy_fns[k] else "secondary"
95
+ crazy_fns[k]["Button"] = gr.Button(k, variant=variant)
96
+ crazy_fns[k]["Button"].style(size="sm")
97
+ with gr.Row():
98
+ with gr.Accordion("更多函数插件", open=True):
99
+ dropdown_fn_list = [k for k in crazy_fns.keys() if not crazy_fns[k].get("AsButton", True)]
100
+ with gr.Row():
101
+ dropdown = gr.Dropdown(dropdown_fn_list, value=r"打开插件列表", label="", show_label=False).style(container=False)
102
+ with gr.Row():
103
+ plugin_advanced_arg = gr.Textbox(show_label=True, label="高级参数输入区", visible=False,
104
+ placeholder="这里是特殊函数插件的高级参数输入区").style(container=False)
105
+ with gr.Row():
106
+ switchy_bt = gr.Button(r"请先从插件列表中选择", variant="secondary")
107
+ with gr.Row():
108
+ with gr.Accordion("点击展开“文件上传区”。上传本地文件/压缩包供函数插件调用。", open=False) as area_file_up:
109
+ file_upload = gr.Files(label="任何文件, 但推荐上传压缩文件(zip, tar)", file_count="multiple")
110
+ with gr.Accordion("更换模型 & SysPrompt & 交互界面布局", open=(LAYOUT == "TOP-DOWN"), elem_id="interact-panel"):
111
+ system_prompt = gr.Textbox(show_label=True, placeholder=f"System Prompt", label="System prompt", value=initial_prompt)
112
+ top_p = gr.Slider(minimum=-0, maximum=1.0, value=1.0, step=0.01,interactive=True, label="Top-p (nucleus sampling)",)
113
+ temperature = gr.Slider(minimum=-0, maximum=2.0, value=1.0, step=0.01, interactive=True, label="Temperature",)
114
+ max_length_sl = gr.Slider(minimum=256, maximum=8192, value=4096, step=1, interactive=True, label="Local LLM MaxLength",)
115
+ checkboxes = gr.CheckboxGroup(["基础功能区", "函数插件区", "底部输入区", "输入清除键", "插件参数区"], value=["基础功能区", "函数插件区"], label="显示/隐藏功能区")
116
+ md_dropdown = gr.Dropdown(AVAIL_LLM_MODELS, value=LLM_MODEL, label="更换LLM模型/请求源").style(container=False)
117
+
118
+ gr.Markdown(description)
119
+ with gr.Accordion("备选输入区", open=True, visible=False, elem_id="input-panel2") as area_input_secondary:
120
+ with gr.Row():
121
+ txt2 = gr.Textbox(show_label=False, placeholder="Input question here.", label="输入区2").style(container=False)
122
+ with gr.Row():
123
+ submitBtn2 = gr.Button("提交", variant="primary")
124
+ with gr.Row():
125
+ resetBtn2 = gr.Button("重置", variant="secondary"); resetBtn2.style(size="sm")
126
+ stopBtn2 = gr.Button("停止", variant="secondary"); stopBtn2.style(size="sm")
127
+ clearBtn2 = gr.Button("清除", variant="secondary", visible=False); clearBtn2.style(size="sm")
128
+ # 功能区显示开关与功能区的互动
129
+ def fn_area_visibility(a):
130
+ ret = {}
131
+ ret.update({area_basic_fn: gr.update(visible=("基础功能区" in a))})
132
+ ret.update({area_crazy_fn: gr.update(visible=("函数插件区" in a))})
133
+ ret.update({area_input_primary: gr.update(visible=("底部输入区" not in a))})
134
+ ret.update({area_input_secondary: gr.update(visible=("底部输入区" in a))})
135
+ ret.update({clearBtn: gr.update(visible=("输入清除键" in a))})
136
+ ret.update({clearBtn2: gr.update(visible=("输入清除键" in a))})
137
+ ret.update({plugin_advanced_arg: gr.update(visible=("插件参数区" in a))})
138
+ if "底部输入区" in a: ret.update({txt: gr.update(value="")})
139
+ return ret
140
+ checkboxes.select(fn_area_visibility, [checkboxes], [area_basic_fn, area_crazy_fn, area_input_primary, area_input_secondary, txt, txt2, clearBtn, clearBtn2, plugin_advanced_arg] )
141
+ # 整理反复出现的控件句柄组合
142
+ input_combo = [cookies, max_length_sl, md_dropdown, txt, txt2, top_p, temperature, chatbot, history, system_prompt, plugin_advanced_arg]
143
+ output_combo = [cookies, chatbot, history, status]
144
+ predict_args = dict(fn=ArgsGeneralWrapper(predict), inputs=input_combo, outputs=output_combo)
145
+ # 提交按钮、重置按钮
146
+ cancel_handles.append(txt.submit(**predict_args))
147
+ cancel_handles.append(txt2.submit(**predict_args))
148
+ cancel_handles.append(submitBtn.click(**predict_args))
149
+ cancel_handles.append(submitBtn2.click(**predict_args))
150
+ resetBtn.click(lambda: ([], [], "已重置"), None, [chatbot, history, status])
151
+ resetBtn2.click(lambda: ([], [], "已重置"), None, [chatbot, history, status])
152
+ clearBtn.click(lambda: ("",""), None, [txt, txt2])
153
+ clearBtn2.click(lambda: ("",""), None, [txt, txt2])
154
+ if AUTO_CLEAR_TXT:
155
+ submitBtn.click(lambda: ("",""), None, [txt, txt2])
156
+ submitBtn2.click(lambda: ("",""), None, [txt, txt2])
157
+ txt.submit(lambda: ("",""), None, [txt, txt2])
158
+ txt2.submit(lambda: ("",""), None, [txt, txt2])
159
+ # 基础功能区的回调函数注册
160
+ for k in functional:
161
+ if ("Visible" in functional[k]) and (not functional[k]["Visible"]): continue
162
+ click_handle = functional[k]["Button"].click(fn=ArgsGeneralWrapper(predict), inputs=[*input_combo, gr.State(True), gr.State(k)], outputs=output_combo)
163
+ cancel_handles.append(click_handle)
164
+ # 文件上传区,接收文件后与chatbot的互动
165
+ file_upload.upload(on_file_uploaded, [file_upload, chatbot, txt, txt2, checkboxes], [chatbot, txt, txt2])
166
+ # 函数插件-固定按钮区
167
+ for k in crazy_fns:
168
+ if not crazy_fns[k].get("AsButton", True): continue
169
+ click_handle = crazy_fns[k]["Button"].click(ArgsGeneralWrapper(crazy_fns[k]["Function"]), [*input_combo, gr.State(PORT)], output_combo)
170
+ click_handle.then(on_report_generated, [cookies, file_upload, chatbot], [cookies, file_upload, chatbot])
171
+ cancel_handles.append(click_handle)
172
+ # 函数插件-下拉菜单与随变按钮的互动
173
+ def on_dropdown_changed(k):
174
+ variant = crazy_fns[k]["Color"] if "Color" in crazy_fns[k] else "secondary"
175
+ ret = {switchy_bt: gr.update(value=k, variant=variant)}
176
+ if crazy_fns[k].get("AdvancedArgs", False): # 是否唤起高级插件参数区
177
+ ret.update({plugin_advanced_arg: gr.update(visible=True, label=f"插件[{k}]的高级参数说明:" + crazy_fns[k].get("ArgsReminder", [f"没有提供高级参数功能说明"]))})
178
+ else:
179
+ ret.update({plugin_advanced_arg: gr.update(visible=False, label=f"插件[{k}]不需要高级参数。")})
180
+ return ret
181
+ dropdown.select(on_dropdown_changed, [dropdown], [switchy_bt, plugin_advanced_arg] )
182
+ def on_md_dropdown_changed(k):
183
+ return {chatbot: gr.update(label="当前模型:"+k)}
184
+ md_dropdown.select(on_md_dropdown_changed, [md_dropdown], [chatbot] )
185
+ # 随变按钮的回调函数注册
186
+ def route(request: gr.Request, k, *args, **kwargs):
187
+ if k in [r"打开插件列表", r"请先从插件列表中选择"]: return
188
+ yield from ArgsGeneralWrapper(crazy_fns[k]["Function"])(request, *args, **kwargs)
189
+ click_handle = switchy_bt.click(route,[switchy_bt, *input_combo, gr.State(PORT)], output_combo)
190
+ click_handle.then(on_report_generated, [cookies, file_upload, chatbot], [cookies, file_upload, chatbot])
191
+ cancel_handles.append(click_handle)
192
+ # 终止按钮的回调函数注册
193
+ stopBtn.click(fn=None, inputs=None, outputs=None, cancels=cancel_handles)
194
+ stopBtn2.click(fn=None, inputs=None, outputs=None, cancels=cancel_handles)
195
+ if ENABLE_AUDIO:
196
+ from crazy_functions.live_audio.audio_io import RealtimeAudioDistribution
197
+ rad = RealtimeAudioDistribution()
198
+ def deal_audio(audio, cookies):
199
+ rad.feed(cookies['uuid'].hex, audio)
200
+ audio_mic.stream(deal_audio, inputs=[audio_mic, cookies])
201
+
202
+ def init_cookie(cookies, chatbot):
203
+ # 为每一位访问的用户赋予一个独一无二的uuid编码
204
+ cookies.update({'uuid': uuid.uuid4()})
205
+ return cookies
206
+ demo.load(init_cookie, inputs=[cookies, chatbot], outputs=[cookies])
207
+ demo.load(lambda: 0, inputs=None, outputs=None, _js='()=>{ChatBotHeight();}')
208
+
209
+ # gradio的inbrowser触发不太稳定,回滚代码到原始的浏览器打开函数
210
+ def auto_opentab_delay():
211
+ import threading, webbrowser, time
212
+ print(f"如果浏览器没有自动打开,请复制并转到以下URL:")
213
+ print(f"\t(亮色主题): http://localhost:{PORT}")
214
+ print(f"\t(暗色主题): http://localhost:{PORT}/?__theme=dark")
215
+ def open():
216
+ time.sleep(2) # 打开浏览器
217
+ DARK_MODE, = get_conf('DARK_MODE')
218
+ if DARK_MODE: webbrowser.open_new_tab(f"http://localhost:{PORT}/?__theme=dark")
219
+ else: webbrowser.open_new_tab(f"http://localhost:{PORT}")
220
+ threading.Thread(target=open, name="open-browser", daemon=True).start()
221
+ threading.Thread(target=auto_update, name="self-upgrade", daemon=True).start()
222
+ threading.Thread(target=warm_up_modules, name="warm-up", daemon=True).start()
223
+
224
+ auto_opentab_delay()
225
+ demo.queue(concurrency_count=CONCURRENT_COUNT).launch(server_name="0.0.0.0", share=False, favicon_path="docs/logo.png", blocked_paths=["config.py","config_private.py","docker-compose.yml","Dockerfile"])
226
+
227
+ # 如果需要在二级路径下运行
228
+ # CUSTOM_PATH, = get_conf('CUSTOM_PATH')
229
+ # if CUSTOM_PATH != "/":
230
+ # from toolbox import run_gradio_in_subpath
231
+ # run_gradio_in_subpath(demo, auth=AUTHENTICATION, port=PORT, custom_path=CUSTOM_PATH)
232
+ # else:
233
+ # demo.launch(server_name="0.0.0.0", server_port=PORT, auth=AUTHENTICATION, favicon_path="docs/logo.png",
234
+ # blocked_paths=["config.py","config_private.py","docker-compose.yml","Dockerfile"])
235
+
236
+ if __name__ == "__main__":
237
+ main()
check_proxy.py ADDED
@@ -0,0 +1,169 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ def check_proxy(proxies):
3
+ import requests
4
+ proxies_https = proxies['https'] if proxies is not None else '无'
5
+ try:
6
+ response = requests.get("https://ipapi.co/json/", proxies=proxies, timeout=4)
7
+ data = response.json()
8
+ print(f'查询代理的地理位置,返回的结果是{data}')
9
+ if 'country_name' in data:
10
+ country = data['country_name']
11
+ result = f"代理配置 {proxies_https}, 代理所在地:{country}"
12
+ elif 'error' in data:
13
+ alternative = _check_with_backup_source(proxies)
14
+ if alternative is None:
15
+ result = f"代理配置 {proxies_https}, 代理所在地:未知,IP查询频率受限"
16
+ else:
17
+ result = f"代理配置 {proxies_https}, 代理所在地:{alternative}"
18
+ else:
19
+ result = f"代理配置 {proxies_https}, 代理数据解析失败:{data}"
20
+ print(result)
21
+ return result
22
+ except:
23
+ result = f"代理配置 {proxies_https}, 代理所在地查询超时,代理可能无效"
24
+ print(result)
25
+ return result
26
+
27
+ def _check_with_backup_source(proxies):
28
+ import random, string, requests
29
+ random_string = ''.join(random.choices(string.ascii_letters + string.digits, k=32))
30
+ try: return requests.get(f"http://{random_string}.edns.ip-api.com/json", proxies=proxies, timeout=4).json()['dns']['geo']
31
+ except: return None
32
+
33
+ def backup_and_download(current_version, remote_version):
34
+ """
35
+ 一键更新协议:备份和下载
36
+ """
37
+ from toolbox import get_conf
38
+ import shutil
39
+ import os
40
+ import requests
41
+ import zipfile
42
+ os.makedirs(f'./history', exist_ok=True)
43
+ backup_dir = f'./history/backup-{current_version}/'
44
+ new_version_dir = f'./history/new-version-{remote_version}/'
45
+ if os.path.exists(new_version_dir):
46
+ return new_version_dir
47
+ os.makedirs(new_version_dir)
48
+ shutil.copytree('./', backup_dir, ignore=lambda x, y: ['history'])
49
+ proxies, = get_conf('proxies')
50
+ r = requests.get(
51
+ 'https://github.com/binary-husky/chatgpt_academic/archive/refs/heads/master.zip', proxies=proxies, stream=True)
52
+ zip_file_path = backup_dir+'/master.zip'
53
+ with open(zip_file_path, 'wb+') as f:
54
+ f.write(r.content)
55
+ dst_path = new_version_dir
56
+ with zipfile.ZipFile(zip_file_path, "r") as zip_ref:
57
+ for zip_info in zip_ref.infolist():
58
+ dst_file_path = os.path.join(dst_path, zip_info.filename)
59
+ if os.path.exists(dst_file_path):
60
+ os.remove(dst_file_path)
61
+ zip_ref.extract(zip_info, dst_path)
62
+ return new_version_dir
63
+
64
+
65
+ def patch_and_restart(path):
66
+ """
67
+ 一键更新协议:覆盖和重启
68
+ """
69
+ from distutils import dir_util
70
+ import shutil
71
+ import os
72
+ import sys
73
+ import time
74
+ import glob
75
+ from colorful import print亮黄, print亮绿, print亮红
76
+ # if not using config_private, move origin config.py as config_private.py
77
+ if not os.path.exists('config_private.py'):
78
+ print亮黄('由于您没有设置config_private.py私密配置,现将您的现有配置移动至config_private.py以防止配置丢失,',
79
+ '另外您可以随时在history子文件夹下找回旧版的程序。')
80
+ shutil.copyfile('config.py', 'config_private.py')
81
+ path_new_version = glob.glob(path + '/*-master')[0]
82
+ dir_util.copy_tree(path_new_version, './')
83
+ print亮绿('代码已经更新,即将更新pip包依赖……')
84
+ for i in reversed(range(5)): time.sleep(1); print(i)
85
+ try:
86
+ import subprocess
87
+ subprocess.check_call([sys.executable, '-m', 'pip', 'install', '-r', 'requirements.txt'])
88
+ except:
89
+ print亮红('pip包依赖安装出现问题,需要手动安装新增的依赖库 `python -m pip install -r requirements.txt`,然后在用常规的`python main.py`的方式启动。')
90
+ print亮绿('更新完成,您可以随时在history子文件夹下找回旧版的程序,5s之后重启')
91
+ print亮红('假如重启失败,您可能需要手动安装新增的依赖库 `python -m pip install -r requirements.txt`,然后在用常规的`python main.py`的方式启动。')
92
+ print(' ------------------------------ -----------------------------------')
93
+ for i in reversed(range(8)): time.sleep(1); print(i)
94
+ os.execl(sys.executable, sys.executable, *sys.argv)
95
+
96
+
97
+ def get_current_version():
98
+ import json
99
+ try:
100
+ with open('./version', 'r', encoding='utf8') as f:
101
+ current_version = json.loads(f.read())['version']
102
+ except:
103
+ current_version = ""
104
+ return current_version
105
+
106
+
107
+ def auto_update(raise_error=False):
108
+ """
109
+ 一键更新协议:查询版本和用户意见
110
+ """
111
+ try:
112
+ from toolbox import get_conf
113
+ import requests
114
+ import time
115
+ import json
116
+ proxies, = get_conf('proxies')
117
+ response = requests.get(
118
+ "https://raw.githubusercontent.com/binary-husky/chatgpt_academic/master/version", proxies=proxies, timeout=5)
119
+ remote_json_data = json.loads(response.text)
120
+ remote_version = remote_json_data['version']
121
+ if remote_json_data["show_feature"]:
122
+ new_feature = "新功能:" + remote_json_data["new_feature"]
123
+ else:
124
+ new_feature = ""
125
+ with open('./version', 'r', encoding='utf8') as f:
126
+ current_version = f.read()
127
+ current_version = json.loads(current_version)['version']
128
+ if (remote_version - current_version) >= 0.01-1e-5:
129
+ from colorful import print亮黄
130
+ print亮黄(
131
+ f'\n新版本可用。新版本:{remote_version},当前版本:{current_version}。{new_feature}')
132
+ print('(1)Github更新地址:\nhttps://github.com/binary-husky/chatgpt_academic\n')
133
+ user_instruction = input('(2)是否一键更新代码(Y+回车=确认,输入其他/无输入+回车=不更新)?')
134
+ if user_instruction in ['Y', 'y']:
135
+ path = backup_and_download(current_version, remote_version)
136
+ try:
137
+ patch_and_restart(path)
138
+ except:
139
+ msg = '更新失败。'
140
+ if raise_error:
141
+ from toolbox import trimmed_format_exc
142
+ msg += trimmed_format_exc()
143
+ print(msg)
144
+ else:
145
+ print('自动更新程序:已禁用')
146
+ return
147
+ else:
148
+ return
149
+ except:
150
+ msg = '自动更新程序:已禁用。建议排查:代理网络配置。'
151
+ if raise_error:
152
+ from toolbox import trimmed_format_exc
153
+ msg += trimmed_format_exc()
154
+ print(msg)
155
+
156
+ def warm_up_modules():
157
+ print('正在执行一些模块的预热...')
158
+ from request_llm.bridge_all import model_info
159
+ enc = model_info["gpt-3.5-turbo"]['tokenizer']
160
+ enc.encode("模块预热", disallowed_special=())
161
+ enc = model_info["gpt-4"]['tokenizer']
162
+ enc.encode("模块预热", disallowed_special=())
163
+
164
+ if __name__ == '__main__':
165
+ import os
166
+ os.environ['no_proxy'] = '*' # 避免代理网络产生意外污染
167
+ from toolbox import get_conf
168
+ proxies, = get_conf('proxies')
169
+ check_proxy(proxies)
colorful.py ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import platform
2
+ from sys import stdout
3
+
4
+ if platform.system()=="Linux":
5
+ pass
6
+ else:
7
+ from colorama import init
8
+ init()
9
+
10
+ # Do you like the elegance of Chinese characters?
11
+ def print红(*kw,**kargs):
12
+ print("\033[0;31m",*kw,"\033[0m",**kargs)
13
+ def print绿(*kw,**kargs):
14
+ print("\033[0;32m",*kw,"\033[0m",**kargs)
15
+ def print黄(*kw,**kargs):
16
+ print("\033[0;33m",*kw,"\033[0m",**kargs)
17
+ def print蓝(*kw,**kargs):
18
+ print("\033[0;34m",*kw,"\033[0m",**kargs)
19
+ def print紫(*kw,**kargs):
20
+ print("\033[0;35m",*kw,"\033[0m",**kargs)
21
+ def print靛(*kw,**kargs):
22
+ print("\033[0;36m",*kw,"\033[0m",**kargs)
23
+
24
+ def print亮红(*kw,**kargs):
25
+ print("\033[1;31m",*kw,"\033[0m",**kargs)
26
+ def print亮绿(*kw,**kargs):
27
+ print("\033[1;32m",*kw,"\033[0m",**kargs)
28
+ def print亮黄(*kw,**kargs):
29
+ print("\033[1;33m",*kw,"\033[0m",**kargs)
30
+ def print亮蓝(*kw,**kargs):
31
+ print("\033[1;34m",*kw,"\033[0m",**kargs)
32
+ def print亮紫(*kw,**kargs):
33
+ print("\033[1;35m",*kw,"\033[0m",**kargs)
34
+ def print亮靛(*kw,**kargs):
35
+ print("\033[1;36m",*kw,"\033[0m",**kargs)
36
+
37
+ # Do you like the elegance of Chinese characters?
38
+ def sprint红(*kw):
39
+ return "\033[0;31m"+' '.join(kw)+"\033[0m"
40
+ def sprint绿(*kw):
41
+ return "\033[0;32m"+' '.join(kw)+"\033[0m"
42
+ def sprint黄(*kw):
43
+ return "\033[0;33m"+' '.join(kw)+"\033[0m"
44
+ def sprint蓝(*kw):
45
+ return "\033[0;34m"+' '.join(kw)+"\033[0m"
46
+ def sprint紫(*kw):
47
+ return "\033[0;35m"+' '.join(kw)+"\033[0m"
48
+ def sprint靛(*kw):
49
+ return "\033[0;36m"+' '.join(kw)+"\033[0m"
50
+ def sprint亮红(*kw):
51
+ return "\033[1;31m"+' '.join(kw)+"\033[0m"
52
+ def sprint亮绿(*kw):
53
+ return "\033[1;32m"+' '.join(kw)+"\033[0m"
54
+ def sprint亮黄(*kw):
55
+ return "\033[1;33m"+' '.join(kw)+"\033[0m"
56
+ def sprint亮蓝(*kw):
57
+ return "\033[1;34m"+' '.join(kw)+"\033[0m"
58
+ def sprint亮紫(*kw):
59
+ return "\033[1;35m"+' '.join(kw)+"\033[0m"
60
+ def sprint亮靛(*kw):
61
+ return "\033[1;36m"+' '.join(kw)+"\033[0m"
config.py ADDED
@@ -0,0 +1,220 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ 以下所有配置也都支持利用环境变量覆写,环境变量配置格式见docker-compose.yml。
3
+ 读取优先级:环境变量 > config_private.py > config.py
4
+ --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- ---
5
+ All the following configurations also support using environment variables to override,
6
+ and the environment variable configuration format can be seen in docker-compose.yml.
7
+ Configuration reading priority: environment variable > config_private.py > config.py
8
+ """
9
+
10
+ # [step 1]>> API_KEY = "sk-123456789xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx123456789"。极少数情况下,还需要填写组织(格式如org-123456789abcdefghijklmno的),请向下翻,找 API_ORG 设置项
11
+ API_KEY = "此处填API密钥" # 可同时填写多个API-KEY,用英文逗号分割,例如API_KEY = "sk-openaikey1,sk-openaikey2,fkxxxx-api2dkey3,azure-apikey4"
12
+
13
+
14
+ # [step 1]>> API_KEY = "sk-123456789xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx123456789"。极少数情况下,还需要填写组织(格式如org-123456789abcdefghijklmno的),请向下翻,找 API_ORG 设置项
15
+ API_KEY = "此处填API密钥" # 可同时填写多个API-KEY,用英文逗号分割,例如API_KEY = "sk-openaikey1,sk-openaikey2,fkxxxx-api2dkey3,azure-apikey4"
16
+
17
+
18
+ # [step 2]>> 改为True应用代理,如果直接在海外服务器部署,此处不修改;如果使用本地或无地域限制的大模型时,此处也不需要修改
19
+ USE_PROXY = False
20
+ if USE_PROXY:
21
+ """
22
+ 填写格式是 [协议]:// [地址] :[端口],填写之前不要忘记把USE_PROXY改成True,如果直接在海外服务器部署,此处不修改
23
+ <配置教程&视频教程> https://github.com/binary-husky/gpt_academic/issues/1>
24
+ [协议] 常见协议无非socks5h/http; 例如 v2**y 和 ss* 的默认本地协议是socks5h; 而cl**h 的默认本地协议是http
25
+ [地址] 懂的都懂,不懂就填localhost或者127.0.0.1肯定错不了(localhost意思是代理软件安装在本机上)
26
+ [端口] 在代理软件的设置里找。虽然不同的代理软件界面不一样,但端口号都应该在最显眼的位置上
27
+ """
28
+ # 代理网络的地址,打开你的*学*网软件查看代理的协议(socks5h / http)、地址(localhost)和端口(11284)
29
+ proxies = {
30
+ # [协议]:// [地址] :[端口]
31
+ "http": "socks5h://localhost:11284", # 再例如 "http": "http://127.0.0.1:7890",
32
+ "https": "socks5h://localhost:11284", # 再例如 "https": "http://127.0.0.1:7890",
33
+ }
34
+ else:
35
+ proxies = None
36
+
37
+ # ------------------------------------ 以下配置可以优化体验, 但大部分场合下并不需要修改 ------------------------------------
38
+
39
+ # 重新URL重新定向,实现更换API_URL的作用(高危设置! 常规情况下不要修改! 通过修改此设置,您将把您的API-KEY和对话隐私完全暴露给您设定的中间人!)
40
+ # 格式: API_URL_REDIRECT = {"https://api.openai.com/v1/chat/completions": "在这里填写重定向的api.openai.com的URL"}
41
+ # 举例: API_URL_REDIRECT = {"https://api.openai.com/v1/chat/completions": "https://reverse-proxy-url/v1/chat/completions"}
42
+ API_URL_REDIRECT = {}
43
+
44
+
45
+ # 多线程函数插件中,默认允许多少路线程同时访问OpenAI。Free trial users的限制是每分钟3次,Pay-as-you-go users的限制是每分钟3500次
46
+ # 一言以蔽之:免费(5刀)用户填3,OpenAI绑了信用卡的用户可以填 16 或者更高。提高限制请查询:https://platform.openai.com/docs/guides/rate-limits/overview
47
+ DEFAULT_WORKER_NUM = 3
48
+
49
+
50
+ # 对话窗的高度
51
+ CHATBOT_HEIGHT = 1115
52
+
53
+
54
+ # 代码高亮
55
+ CODE_HIGHLIGHT = True
56
+
57
+
58
+ # 窗口布局
59
+ LAYOUT = "LEFT-RIGHT" # "LEFT-RIGHT"(左右布局) # "TOP-DOWN"(上下布局)
60
+ DARK_MODE = True # 暗色模式 / 亮色模式
61
+
62
+
63
+ # 发送请求到OpenAI后,等待多久判定为超时
64
+ TIMEOUT_SECONDS = 30
65
+
66
+
67
+ # 网页的端口, -1代表随机端口
68
+ WEB_PORT = -1
69
+
70
+
71
+ # 如果OpenAI不响应(网络卡顿、代理失败、KEY失效),重试的次数限制
72
+ MAX_RETRY = 2
73
+
74
+ # OpenAI模型选择是(gpt4现在只对申请成功的人开放)
75
+ LLM_MODEL = "gpt-3.5-turbo" # 可选 "chatglm"
76
+ AVAIL_LLM_MODELS = ["gpt-3.5-turbo", "gpt-4", "api2d-gpt-4", "api2d-gpt-3.5-turbo", "spark", "azure-gpt-3.5"]
77
+
78
+ # ChatGLM(2) Finetune Model Path (如果使用ChatGLM2微调模型,需要把"chatglmft"加入AVAIL_LLM_MODELS中)
79
+ ChatGLM_PTUNING_CHECKPOINT = "" # 例如"/home/hmp/ChatGLM2-6B/ptuning/output/6b-pt-128-1e-2/checkpoint-100"
80
+
81
+
82
+ # 本地LLM模型如ChatGLM的执行方式 CPU/GPU
83
+ LOCAL_MODEL_DEVICE = "cpu" # 可选 "cuda"
84
+ LOCAL_MODEL_QUANT = "FP16" # 默认 "FP16" "INT4" 启用量化INT4版本 "INT8" 启用量化INT8版本
85
+
86
+
87
+ # 设置gradio的并行线程数(不需要修改)
88
+ CONCURRENT_COUNT = 100
89
+
90
+
91
+ # 是否在提交时自动清空输入框
92
+ AUTO_CLEAR_TXT = False
93
+
94
+
95
+ # 色彩主体,可选 ["Default", "Chuanhu-Small-and-Beautiful"]
96
+ THEME = "Default"
97
+
98
+
99
+ # 加一���live2d装饰
100
+ ADD_WAIFU = False
101
+
102
+
103
+ # 设置用户名和密码(不需要修改)(相关功能不稳定,与gradio版本和网络都相关,如果本地使用不建议加这个)
104
+ # [("username", "password"), ("username2", "password2"), ...]
105
+ AUTHENTICATION = []
106
+
107
+
108
+ # 如果需要在二级路径下运行(常规情况下,不要修改!!)(需要配合修改main.py才能生效!)
109
+ CUSTOM_PATH = "/"
110
+
111
+
112
+ # 极少数情况下,openai的官方KEY需要伴随组织编码(格式如org-xxxxxxxxxxxxxxxxxxxxxxxx)使用
113
+ API_ORG = ""
114
+
115
+
116
+ # 如果需要使用Slack Claude,使用教程详情见 request_llm/README.md
117
+ SLACK_CLAUDE_BOT_ID = ''
118
+ SLACK_CLAUDE_USER_TOKEN = ''
119
+
120
+
121
+ # 如果需要使用AZURE 详情请见额外文档 docs\use_azure.md
122
+ AZURE_ENDPOINT = "https://你亲手写的api名称.openai.azure.com/"
123
+ AZURE_API_KEY = "填入azure openai api的密钥" # 建议直接在API_KEY处填写,该选项即将被弃用
124
+ AZURE_ENGINE = "填入你亲手写的部署名" # 读 docs\use_azure.md
125
+
126
+
127
+ # 使用Newbing
128
+ NEWBING_STYLE = "creative" # ["creative", "balanced", "precise"]
129
+ NEWBING_COOKIES = """
130
+ put your new bing cookies here
131
+ """
132
+
133
+
134
+ # 阿里云实时语音识别 配置难度较高 仅建议高手用户使用 参考 https://github.com/binary-husky/gpt_academic/blob/master/docs/use_audio.md
135
+ ENABLE_AUDIO = False
136
+ ALIYUN_TOKEN="" # 例如 f37f30e0f9934c34a992f6f64f7eba4f
137
+ ALIYUN_APPKEY="" # 例如 RoPlZrM88DnAFkZK
138
+ ALIYUN_ACCESSKEY="" # (无需填写)
139
+ ALIYUN_SECRET="" # (无需填写)
140
+
141
+
142
+ # 接入讯飞星火大模型 https://console.xfyun.cn/services/iat
143
+ XFYUN_APPID = "00000000"
144
+ XFYUN_API_SECRET = "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"
145
+ XFYUN_API_KEY = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
146
+
147
+
148
+ # Claude API KEY
149
+ ANTHROPIC_API_KEY = ""
150
+
151
+
152
+ # 自定义API KEY格式
153
+ CUSTOM_API_KEY_PATTERN = ""
154
+
155
+
156
+ # HUGGINGFACE的TOKEN,下载LLAMA时起作用 https://huggingface.co/docs/hub/security-tokens
157
+ HUGGINGFACE_ACCESS_TOKEN = "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"
158
+
159
+
160
+ # GROBID服务器地址(填写多个可以均衡负载),用于高质量地读取PDF文档
161
+ # 获取方法:复制以下空间https://huggingface.co/spaces/qingxu98/grobid,设为public,然后GROBID_URL = "https://(你的hf用户名如qingxu98)-(你的填写的空间名如grobid).hf.space"
162
+ GROBID_URLS = [
163
+ "https://qingxu98-grobid.hf.space","https://qingxu98-grobid2.hf.space","https://qingxu98-grobid3.hf.space",
164
+ "https://shaocongma-grobid.hf.space","https://FBR123-grobid.hf.space",
165
+ ]
166
+
167
+
168
+
169
+ """
170
+ 在线大模型配置关联关系示意图
171
+
172
+ ├── "gpt-3.5-turbo" 等openai模型
173
+ │ ├── API_KEY
174
+ │ ├── CUSTOM_API_KEY_PATTERN(不常用)
175
+ │ ├── API_ORG(不常用)
176
+ │ └── API_URL_REDIRECT(不常用)
177
+
178
+ ├── "azure-gpt-3.5" 等azure模型
179
+ │ ├── API_KEY
180
+ │ ├── AZURE_ENDPOINT
181
+ │ ├── AZURE_API_KEY
182
+ │ ├── AZURE_ENGINE
183
+ │ └── API_URL_REDIRECT
184
+
185
+ ├── "spark" 星火认知大模型
186
+ │ ├── XFYUN_APPID
187
+ │ ├── XFYUN_API_SECRET
188
+ │ └── XFYUN_API_KEY
189
+
190
+ ├── "claude-1-100k" 等claude模型
191
+ │ └── ANTHROPIC_API_KEY
192
+
193
+ ├── "stack-claude"
194
+ │ ├── SLACK_CLAUDE_BOT_ID
195
+ │ └── SLACK_CLAUDE_USER_TOKEN
196
+
197
+ ├── "qianfan" 百度千帆大模型库
198
+ │ ├── BAIDU_CLOUD_QIANFAN_MODEL
199
+ │ ├── BAIDU_CLOUD_API_KEY
200
+ │ └── BAIDU_CLOUD_SECRET_KEY
201
+
202
+ ├── "newbing" Newbing接口不再稳定,不推荐使用
203
+ ├── NEWBING_STYLE
204
+ └── NEWBING_COOKIES
205
+
206
+
207
+
208
+ 插件在线服务配置依赖关系示意图
209
+
210
+ ├── 语音功能
211
+ │ ├── ENABLE_AUDIO
212
+ │ ├── ALIYUN_TOKEN
213
+ │ ├── ALIYUN_APPKEY
214
+ │ ├── ALIYUN_ACCESSKEY
215
+ │ └── ALIYUN_SECRET
216
+
217
+ ├── PDF文档精准解析
218
+ │ └── GROBID_URLS
219
+
220
+ """
core_functional.py ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # 'primary' 颜色对应 theme.py 中的 primary_hue
2
+ # 'secondary' 颜色对应 theme.py 中的 neutral_hue
3
+ # 'stop' 颜色对应 theme.py 中的 color_er
4
+ import importlib
5
+ from toolbox import clear_line_break
6
+
7
+
8
+ def get_core_functions():
9
+ return {
10
+ "英语学术润色": {
11
+ # 前缀,会被加在你的输入之前。例如,用来描述你的要求,例如翻译、解释代码、润色等等
12
+ "Prefix": r"Below is a paragraph from an academic paper. Polish the writing to meet the academic style, " +
13
+ r"improve the spelling, grammar, clarity, concision and overall readability. When necessary, rewrite the whole sentence. " +
14
+ r"Furthermore, list all modification and explain the reasons to do so in markdown table." + "\n\n",
15
+ # 后缀,会被加在你的输入之后。例如,配合前缀可以把你的输入内容用引号圈起来
16
+ "Suffix": r"",
17
+ # 按钮颜色 (默认 secondary)
18
+ "Color": r"secondary",
19
+ # 按钮是否可见 (默认 True,即可见)
20
+ "Visible": True,
21
+ # 是否在触发时清除历史 (默认 False,即不处理之前的对话历史)
22
+ "AutoClearHistory": False
23
+ },
24
+ "中文学术润色": {
25
+ "Prefix": r"作为一名中文学术论文写作改进助理,你的任务是改进所提供文本的拼写、语法、清晰、简洁和整体可读性," +
26
+ r"同时分解长句,减少重复,并提供改进建议。请只提供文本的更正版本,避免包括解释。请编辑以下文本" + "\n\n",
27
+ "Suffix": r"",
28
+ },
29
+ "查找语法错误": {
30
+ "Prefix": r"Can you help me ensure that the grammar and the spelling is correct? " +
31
+ r"Do not try to polish the text, if no mistake is found, tell me that this paragraph is good." +
32
+ r"If you find grammar or spelling mistakes, please list mistakes you find in a two-column markdown table, " +
33
+ r"put the original text the first column, " +
34
+ r"put the corrected text in the second column and highlight the key words you fixed.""\n"
35
+ r"Example:""\n"
36
+ r"Paragraph: How is you? Do you knows what is it?""\n"
37
+ r"| Original sentence | Corrected sentence |""\n"
38
+ r"| :--- | :--- |""\n"
39
+ r"| How **is** you? | How **are** you? |""\n"
40
+ r"| Do you **knows** what **is** **it**? | Do you **know** what **it** **is** ? |""\n"
41
+ r"Below is a paragraph from an academic paper. "
42
+ r"You need to report all grammar and spelling mistakes as the example before."
43
+ + "\n\n",
44
+ "Suffix": r"",
45
+ "PreProcess": clear_line_break, # 预处理:清除换行符
46
+ },
47
+ "中译英": {
48
+ "Prefix": r"Please translate following sentence to English:" + "\n\n",
49
+ "Suffix": r"",
50
+ },
51
+ "学术中英互译": {
52
+ "Prefix": r"I want you to act as a scientific English-Chinese translator, " +
53
+ r"I will provide you with some paragraphs in one language " +
54
+ r"and your task is to accurately and academically translate the paragraphs only into the other language. " +
55
+ r"Do not repeat the original provided paragraphs after translation. " +
56
+ r"You should use artificial intelligence tools, " +
57
+ r"such as natural language processing, and rhetorical knowledge " +
58
+ r"and experience about effective writing techniques to reply. " +
59
+ r"I'll give you my paragraphs as follows, tell me what language it is written in, and then translate:" + "\n\n",
60
+ "Suffix": "",
61
+ "Color": "secondary",
62
+ },
63
+ "英译中": {
64
+ "Prefix": r"翻译成地道的中文:" + "\n\n",
65
+ "Suffix": r"",
66
+ },
67
+ "找图片": {
68
+ "Prefix": r"我需要你找一张网络图片。使用Unsplash API(https://source.unsplash.com/960x640/?<英语关键词>)获取图片URL," +
69
+ r"然后请使用Markdown格式封装,并且不要有反斜线,不要用代码块。现在,请按以下描述给我发送图片:" + "\n\n",
70
+ "Suffix": r"",
71
+ "Visible": False,
72
+ },
73
+ "解释代码": {
74
+ "Prefix": r"请解释以下代码:" + "\n```\n",
75
+ "Suffix": "\n```\n",
76
+ },
77
+ "参考文献转Bib": {
78
+ "Prefix": r"Here are some bibliography items, please transform them into bibtex style." +
79
+ r"Note that, reference styles maybe more than one kind, you should transform each item correctly." +
80
+ r"Items need to be transformed:",
81
+ "Suffix": r"",
82
+ }
83
+ }
84
+
85
+
86
+ def handle_core_functionality(additional_fn, inputs, history, chatbot):
87
+ import core_functional
88
+ importlib.reload(core_functional) # 热更新prompt
89
+ core_functional = core_functional.get_core_functions()
90
+ if "PreProcess" in core_functional[additional_fn]: inputs = core_functional[additional_fn]["PreProcess"](inputs) # 获取预处理函数(如果有的话)
91
+ inputs = core_functional[additional_fn]["Prefix"] + inputs + core_functional[additional_fn]["Suffix"]
92
+ if core_functional[additional_fn].get("AutoClearHistory", False):
93
+ history = []
94
+ return inputs, history
crazy_functional.py ADDED
@@ -0,0 +1,439 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import HotReload # HotReload 的意思是热更新,修改函数插件后,不需要重启程序,代码直接生效
2
+
3
+
4
+ def get_crazy_functions():
5
+ ###################### 第一组插件 ###########################
6
+ from crazy_functions.读文章写摘要 import 读文章写摘要
7
+ from crazy_functions.生成函数注释 import 批量生成函数注释
8
+ from crazy_functions.解析项目源代码 import 解析项目本身
9
+ from crazy_functions.解析项目源代码 import 解析一个Python项目
10
+ from crazy_functions.解析项目源代码 import 解析一个C项目的头文件
11
+ from crazy_functions.解析项目源代码 import 解析一个C项目
12
+ from crazy_functions.解析项目源代码 import 解析一个Golang项目
13
+ from crazy_functions.解析项目源代码 import 解析一个Rust项目
14
+ from crazy_functions.解析项目源代码 import 解析一个Java项目
15
+ from crazy_functions.解析项目源代码 import 解析一个前端项目
16
+ from crazy_functions.高级功能函数模板 import 高阶功能模板函数
17
+ from crazy_functions.代码重写为全英文_多线程 import 全项目切换英文
18
+ from crazy_functions.Latex全文润色 import Latex英文润色
19
+ from crazy_functions.询问多个大语言模型 import 同时问询
20
+ from crazy_functions.解析项目源代码 import 解析一个Lua项目
21
+ from crazy_functions.解析项目源代码 import 解析一个CSharp项目
22
+ from crazy_functions.总结word文档 import 总结word文档
23
+ from crazy_functions.解析JupyterNotebook import 解析ipynb文件
24
+ from crazy_functions.对话历史存档 import 对话历史存档
25
+ from crazy_functions.对话历史存档 import 载入对话历史存档
26
+ from crazy_functions.对话历史存档 import 删除所有本地对话历史记录
27
+ from crazy_functions.辅助功能 import 清除缓存
28
+
29
+ from crazy_functions.批量Markdown翻译 import Markdown英译中
30
+ function_plugins = {
31
+ "解析整个Python项目": {
32
+ "Color": "stop", # 按钮颜色
33
+ "Function": HotReload(解析一个Python项目)
34
+ },
35
+ "载入对话历史存档(先上传存档或输入路径)": {
36
+ "Color": "stop",
37
+ "AsButton":False,
38
+ "Function": HotReload(载入对话历史存档)
39
+ },
40
+ "删除所有本地对话历史记录(请谨慎操作)": {
41
+ "AsButton":False,
42
+ "Function": HotReload(删除所有本地对话历史记录)
43
+ },
44
+ "清除所有缓存文件(请谨慎操作)": {
45
+ "Color": "stop",
46
+ "AsButton": False, # 加入下拉菜单中
47
+ "Function": HotReload(清除缓存)
48
+ },
49
+ "解析Jupyter Notebook文件": {
50
+ "Color": "stop",
51
+ "AsButton":False,
52
+ "Function": HotReload(解析ipynb文件),
53
+ "AdvancedArgs": True, # 调用时,唤起高级参数输入区(默认False)
54
+ "ArgsReminder": "若输入0,则不解析notebook中的Markdown块", # 高级参数输入区的显示提示
55
+ },
56
+ "批量总结Word文档": {
57
+ "Color": "stop",
58
+ "Function": HotReload(总结word文档)
59
+ },
60
+ "解析整个C++项目头文件": {
61
+ "Color": "stop", # 按钮颜色
62
+ "AsButton": False, # 加入下拉菜单中
63
+ "Function": HotReload(解析一个C项目的头文件)
64
+ },
65
+ "解析整个C++项目(.cpp/.hpp/.c/.h)": {
66
+ "Color": "stop", # 按钮颜色
67
+ "AsButton": False, # 加入下拉菜单中
68
+ "Function": HotReload(解析一个C项目)
69
+ },
70
+ "解析整个Go项目": {
71
+ "Color": "stop", # 按钮颜色
72
+ "AsButton": False, # 加入下拉菜单中
73
+ "Function": HotReload(解析一个Golang项目)
74
+ },
75
+ "解析整个Rust项目": {
76
+ "Color": "stop", # 按钮颜色
77
+ "AsButton": False, # 加入下拉菜单中
78
+ "Function": HotReload(解析一个Rust项目)
79
+ },
80
+ "解析整个Java项目": {
81
+ "Color": "stop", # 按钮颜色
82
+ "AsButton": False, # 加入下拉菜单中
83
+ "Function": HotReload(解析一个Java项目)
84
+ },
85
+ "解析整个前端项目(js,ts,css等)": {
86
+ "Color": "stop", # 按钮颜色
87
+ "AsButton": False, # 加入下拉菜单中
88
+ "Function": HotReload(解析一个前端项目)
89
+ },
90
+ "解析整个Lua项目": {
91
+ "Color": "stop", # 按钮颜色
92
+ "AsButton": False, # 加入下拉菜单中
93
+ "Function": HotReload(解析一个Lua项目)
94
+ },
95
+ "解析整个CSharp项目": {
96
+ "Color": "stop", # 按钮颜色
97
+ "AsButton": False, # 加入下拉菜单中
98
+ "Function": HotReload(解析一个CSharp项目)
99
+ },
100
+ "读Tex论文写摘要": {
101
+ "Color": "stop", # 按钮颜色
102
+ "Function": HotReload(读文章写摘要)
103
+ },
104
+ "Markdown/Readme英译中": {
105
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
106
+ "Color": "stop",
107
+ "Function": HotReload(Markdown英译中)
108
+ },
109
+ "批量生成函数注释": {
110
+ "Color": "stop", # 按钮颜色
111
+ "AsButton": False, # 加入下拉菜单中
112
+ "Function": HotReload(批量生成函数注释)
113
+ },
114
+ "保存当前的对话": {
115
+ "Function": HotReload(对话历史存档)
116
+ },
117
+ "[多线程Demo] 解析此项目本身(源码自译解)": {
118
+ "AsButton": False, # 加入下拉菜单中
119
+ "Function": HotReload(解析项目本身)
120
+ },
121
+ # "[老旧的Demo] 把本项目源代码切换成全英文": {
122
+ # # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
123
+ # "AsButton": False, # 加入下拉菜单中
124
+ # "Function": HotReload(全项目切换英文)
125
+ # },
126
+ "[插件demo] 历史上的今天": {
127
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
128
+ "Function": HotReload(高阶功能模板函数)
129
+ },
130
+
131
+ }
132
+ ###################### 第二组插件 ###########################
133
+ # [第二组插件]: 经过充分测试
134
+ from crazy_functions.批量总结PDF文档 import 批量总结PDF文档
135
+ # from crazy_functions.批量总结PDF文档pdfminer import 批量总结PDF文档pdfminer
136
+ from crazy_functions.批量翻译PDF文档_多线程 import 批量翻译PDF文档
137
+ from crazy_functions.谷歌检索小助手 import 谷歌检索小助手
138
+ from crazy_functions.理解PDF文档内容 import 理解PDF文档内容标准文件输入
139
+ from crazy_functions.Latex全文润色 import Latex中文润色
140
+ from crazy_functions.Latex全文润色 import Latex英文纠错
141
+ from crazy_functions.Latex全文翻译 import Latex中译英
142
+ from crazy_functions.Latex全文翻译 import Latex英译中
143
+ from crazy_functions.批量Markdown翻译 import Markdown中译英
144
+
145
+ function_plugins.update({
146
+ "批量翻译PDF文档(多线程)": {
147
+ "Color": "stop",
148
+ "AsButton": True, # 加入下拉菜单中
149
+ "Function": HotReload(批量翻译PDF文档)
150
+ },
151
+ "询问多个GPT模型": {
152
+ "Color": "stop", # 按钮颜色
153
+ "Function": HotReload(同时问询)
154
+ },
155
+ "[测试功能] 批量总结PDF文档": {
156
+ "Color": "stop",
157
+ "AsButton": False, # 加入下拉菜单中
158
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
159
+ "Function": HotReload(批量总结PDF文档)
160
+ },
161
+ # "[测试功能] 批量总结PDF文档pdfminer": {
162
+ # "Color": "stop",
163
+ # "AsButton": False, # 加入下拉菜单中
164
+ # "Function": HotReload(批量总结PDF文档pdfminer)
165
+ # },
166
+ "谷歌学术检索助手(输入谷歌学术搜索页url)": {
167
+ "Color": "stop",
168
+ "AsButton": False, # 加入下拉菜单中
169
+ "Function": HotReload(谷歌检索小助手)
170
+ },
171
+ "理解PDF文档内容 (模仿ChatPDF)": {
172
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
173
+ "Color": "stop",
174
+ "AsButton": False, # 加入下拉菜单中
175
+ "Function": HotReload(理解PDF文档内容标准文件输入)
176
+ },
177
+ "英文Latex项目全文润色(输入路径或上传压缩包)": {
178
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
179
+ "Color": "stop",
180
+ "AsButton": False, # 加入下拉菜单中
181
+ "Function": HotReload(Latex英文润色)
182
+ },
183
+ "英文Latex项目全文纠错(输入路径或上传压缩包)": {
184
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
185
+ "Color": "stop",
186
+ "AsButton": False, # 加入下拉菜单中
187
+ "Function": HotReload(Latex英文纠错)
188
+ },
189
+ "中文Latex项目全文润色(输入路径或上传压缩包)": {
190
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
191
+ "Color": "stop",
192
+ "AsButton": False, # 加入下拉菜单中
193
+ "Function": HotReload(Latex中文润色)
194
+ },
195
+ "Latex项目全文中译英(输入路径或上传压缩包)": {
196
+ # HotReload 的意思是热更新,修改函数插件代码后��不需要重启程序,代码直接生效
197
+ "Color": "stop",
198
+ "AsButton": False, # 加入下拉菜单中
199
+ "Function": HotReload(Latex中译英)
200
+ },
201
+ "Latex项目全文英译中(输入路径或上传压缩包)": {
202
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
203
+ "Color": "stop",
204
+ "AsButton": False, # 加入下拉菜单中
205
+ "Function": HotReload(Latex英译中)
206
+ },
207
+ "批量Markdown中译英(输入路径或上传压缩包)": {
208
+ # HotReload 的意思是热更新,修改函数插件代码后,不需要重启程序,代码直接生效
209
+ "Color": "stop",
210
+ "AsButton": False, # 加入下拉菜单中
211
+ "Function": HotReload(Markdown中译英)
212
+ },
213
+
214
+
215
+ })
216
+
217
+ ###################### 第三组插件 ###########################
218
+ # [第三组插件]: 尚未充分测试的函数插件
219
+
220
+ try:
221
+ from crazy_functions.下载arxiv论文翻译摘要 import 下载arxiv论文并翻译摘要
222
+ function_plugins.update({
223
+ "一键下载arxiv论文并翻译摘要(先在input输入编号,如1812.10695)": {
224
+ "Color": "stop",
225
+ "AsButton": False, # 加入下拉菜单中
226
+ "Function": HotReload(下载arxiv论文并翻译摘要)
227
+ }
228
+ })
229
+ except:
230
+ print('Load function plugin failed')
231
+
232
+ try:
233
+ from crazy_functions.联网的ChatGPT import 连接网络回答问题
234
+ function_plugins.update({
235
+ "连接网络回答问题(输入问题后点击该插件,需要访问谷歌)": {
236
+ "Color": "stop",
237
+ "AsButton": False, # 加入下拉菜单中
238
+ "Function": HotReload(连接网络回答问题)
239
+ }
240
+ })
241
+ from crazy_functions.联网的ChatGPT_bing版 import 连接bing搜索回答问题
242
+ function_plugins.update({
243
+ "连接网络回答问题(中文Bing版,输入问题后点击该插件)": {
244
+ "Color": "stop",
245
+ "AsButton": False, # 加入下拉菜单中
246
+ "Function": HotReload(连接bing搜索回答问题)
247
+ }
248
+ })
249
+ except:
250
+ print('Load function plugin failed')
251
+
252
+ try:
253
+ from crazy_functions.解析项目源代码 import 解析任意code项目
254
+ function_plugins.update({
255
+ "解析项目源代码(手动指定和筛选源代码文件类型)": {
256
+ "Color": "stop",
257
+ "AsButton": False,
258
+ "AdvancedArgs": True, # 调用时,唤起高级参数输入区(默认False)
259
+ "ArgsReminder": "输入时用逗号隔开, *代表通配符, 加了^代表不匹配; 不输入代表全部匹配。例如: \"*.c, ^*.cpp, config.toml, ^*.toml\"", # 高级参数输入区的显示提示
260
+ "Function": HotReload(解析任意code项目)
261
+ },
262
+ })
263
+ except:
264
+ print('Load function plugin failed')
265
+
266
+ try:
267
+ from crazy_functions.询问多个大语言模型 import 同时问询_指定模型
268
+ function_plugins.update({
269
+ "询问多个GPT模型(手动指定询问哪些模型)": {
270
+ "Color": "stop",
271
+ "AsButton": False,
272
+ "AdvancedArgs": True, # 调用时,唤起高级参数输入区(默认False)
273
+ "ArgsReminder": "支持任意数量的llm接口,用&符号分隔。例如chatglm&gpt-3.5-turbo&api2d-gpt-4", # 高级参数输入区的显示提示
274
+ "Function": HotReload(同时问询_指定模型)
275
+ },
276
+ })
277
+ except:
278
+ print('Load function plugin failed')
279
+
280
+ try:
281
+ from crazy_functions.图片生成 import 图片生成
282
+ function_plugins.update({
283
+ "图片生成(先切换模型到openai或api2d)": {
284
+ "Color": "stop",
285
+ "AsButton": False,
286
+ "AdvancedArgs": True, # 调用时,唤起高级参数输入区(默认False)
287
+ "ArgsReminder": "在这里输入分辨率, 如256x256(默认)", # 高级参数输入区的显示提示
288
+ "Function": HotReload(图片生成)
289
+ },
290
+ })
291
+ except:
292
+ print('Load function plugin failed')
293
+
294
+ try:
295
+ from crazy_functions.总结音视频 import 总结音视频
296
+ function_plugins.update({
297
+ "批量总结音视频(输入路径或上传压缩包)": {
298
+ "Color": "stop",
299
+ "AsButton": False,
300
+ "AdvancedArgs": True,
301
+ "ArgsReminder": "调用openai api 使用whisper-1模型, 目前支持的格式:mp4, m4a, wav, mpga, mpeg, mp3。此处可以输入解析提示,例如:解析为简体中文(默认)。",
302
+ "Function": HotReload(总结音视频)
303
+ }
304
+ })
305
+ except:
306
+ print('Load function plugin failed')
307
+
308
+ try:
309
+ from crazy_functions.数学动画生成manim import 动画生成
310
+ function_plugins.update({
311
+ "数学动画生成(Manim)": {
312
+ "Color": "stop",
313
+ "AsButton": False,
314
+ "Function": HotReload(动画生成)
315
+ }
316
+ })
317
+ except:
318
+ print('Load function plugin failed')
319
+
320
+ try:
321
+ from crazy_functions.批量Markdown翻译 import Markdown翻译指定语言
322
+ function_plugins.update({
323
+ "Markdown翻译(手动指定语言)": {
324
+ "Color": "stop",
325
+ "AsButton": False,
326
+ "AdvancedArgs": True,
327
+ "ArgsReminder": "请输入要翻译成哪种语言,默认为Chinese。",
328
+ "Function": HotReload(Markdown翻译指定语言)
329
+ }
330
+ })
331
+ except:
332
+ print('Load function plugin failed')
333
+
334
+ try:
335
+ from crazy_functions.Langchain知识库 import 知识库问答
336
+ function_plugins.update({
337
+ "构建知识库(请先上传文件素材)": {
338
+ "Color": "stop",
339
+ "AsButton": False,
340
+ "AdvancedArgs": True,
341
+ "ArgsReminder": "待注入的知识库名称id, 默认为default",
342
+ "Function": HotReload(知识库问答)
343
+ }
344
+ })
345
+ except:
346
+ print('Load function plugin failed')
347
+
348
+ try:
349
+ from crazy_functions.Langchain知识库 import 读取知识库作答
350
+ function_plugins.update({
351
+ "知识库问答": {
352
+ "Color": "stop",
353
+ "AsButton": False,
354
+ "AdvancedArgs": True,
355
+ "ArgsReminder": "待提取的知识库名称id, 默认为default, 您需要首先调用构建知识库",
356
+ "Function": HotReload(读取知识库作答)
357
+ }
358
+ })
359
+ except:
360
+ print('Load function plugin failed')
361
+
362
+ try:
363
+ from crazy_functions.交互功能函数模板 import 交互功能模板函数
364
+ function_plugins.update({
365
+ "交互功能模板函数": {
366
+ "Color": "stop",
367
+ "AsButton": False,
368
+ "Function": HotReload(交互功能模板函数)
369
+ }
370
+ })
371
+ except:
372
+ print('Load function plugin failed')
373
+
374
+ # try:
375
+ # from crazy_functions.chatglm微调工具 import 微调数据集生成
376
+ # function_plugins.update({
377
+ # "黑盒模型学习: 微调数据集生成 (先上传数据集)": {
378
+ # "Color": "stop",
379
+ # "AsButton": False,
380
+ # "AdvancedArgs": True,
381
+ # "ArgsReminder": "针对数据集输入(如 绿帽子*深蓝色衬衫*黑色运动裤)给出指令,例如您可以将以下命令复制到下方: --llm_to_learn=azure-gpt-3.5 --prompt_prefix='根据下面的服装类型提示,想象一个穿着者,对这个人外貌、身处的环境、内心世界、过去经历进行描写。要求:100字以内,用第二人称。' --system_prompt=''",
382
+ # "Function": HotReload(微调数据集生成)
383
+ # }
384
+ # })
385
+ # except:
386
+ # print('Load function plugin failed')
387
+
388
+ try:
389
+ from crazy_functions.Latex输出PDF结果 import Latex英文纠错加PDF对比
390
+ function_plugins.update({
391
+ "Latex英文纠错+高亮修正位置 [需Latex]": {
392
+ "Color": "stop",
393
+ "AsButton": False,
394
+ "AdvancedArgs": True,
395
+ "ArgsReminder": "如果有必要, 请在此处追加更细致的矫错指令(使用英文)。",
396
+ "Function": HotReload(Latex英文纠错加PDF对比)
397
+ }
398
+ })
399
+ from crazy_functions.Latex输出PDF结果 import Latex翻译中文并重新编译PDF
400
+ function_plugins.update({
401
+ "Arixv论文精细翻译(输入arxivID)[需Latex]": {
402
+ "Color": "stop",
403
+ "AsButton": False,
404
+ "AdvancedArgs": True,
405
+ "ArgsReminder":
406
+ "如果有必要, 请在此处给出自定义翻译命令, 解决部分词汇翻译不准确的问题。 "+
407
+ "例如当单词'agent'翻译不准确时, 请尝试把以下指令复制到高级参数区: " + 'If the term "agent" is used in this section, it should be translated to "智能体". ',
408
+ "Function": HotReload(Latex翻译中文并重新编译PDF)
409
+ }
410
+ })
411
+ function_plugins.update({
412
+ "本地Latex论文精细翻译(上传Latex项目)[需Latex]": {
413
+ "Color": "stop",
414
+ "AsButton": False,
415
+ "AdvancedArgs": True,
416
+ "ArgsReminder":
417
+ "如果有必要, 请在此处给出自定义翻译命令, 解决部分词汇翻译不准确的问题。 "+
418
+ "例如当单词'agent'翻译不准确时, 请尝试把以下指令复制到高级参数区: " + 'If the term "agent" is used in this section, it should be translated to "智能体". ',
419
+ "Function": HotReload(Latex翻译中文并重新编译PDF)
420
+ }
421
+ })
422
+ except:
423
+ print('Load function plugin failed')
424
+
425
+ # try:
426
+ # from crazy_functions.虚空终端 import 终端
427
+ # function_plugins.update({
428
+ # "超级终端": {
429
+ # "Color": "stop",
430
+ # "AsButton": False,
431
+ # # "AdvancedArgs": True,
432
+ # # "ArgsReminder": "",
433
+ # "Function": HotReload(终端)
434
+ # }
435
+ # })
436
+ # except:
437
+ # print('Load function plugin failed')
438
+
439
+ return function_plugins
crazy_functions/Langchain知识库.py ADDED
@@ -0,0 +1,107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import CatchException, update_ui, ProxyNetworkActivate
2
+ from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive, get_files_from_everything
3
+
4
+
5
+
6
+ @CatchException
7
+ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
8
+ """
9
+ txt 输入栏用户输入的文本,例如需要翻译的一段话,再例如一个包含了待处理文件的路径
10
+ llm_kwargs gpt模型参数, 如温度和top_p等, 一般原样传递下去就行
11
+ plugin_kwargs 插件模型的参数,暂时没有用武之地
12
+ chatbot 聊天显示框的句柄,用于显示给用户
13
+ history 聊天历史,前情提要
14
+ system_prompt 给gpt的静默提醒
15
+ web_port 当前软件运行的端口号
16
+ """
17
+ history = [] # 清空历史,以免输入溢出
18
+ chatbot.append(("这是什么功能?", "[Local Message] 从一批文件(txt, md, tex)中读取数据构建知识库, 然后进行问答。"))
19
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
20
+
21
+ # resolve deps
22
+ try:
23
+ from zh_langchain import construct_vector_store
24
+ from langchain.embeddings.huggingface import HuggingFaceEmbeddings
25
+ from .crazy_utils import knowledge_archive_interface
26
+ except Exception as e:
27
+ chatbot.append(
28
+ ["依赖不足",
29
+ "导入依赖失败。正在尝试自动安装,请查看终端的输出或耐心等待..."]
30
+ )
31
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
32
+ from .crazy_utils import try_install_deps
33
+ try_install_deps(['zh_langchain==0.2.1', 'pypinyin'])
34
+
35
+ # < --------------------读取参数--------------- >
36
+ if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg")
37
+ kai_id = plugin_kwargs.get("advanced_arg", 'default')
38
+
39
+ # < --------------------读取文件--------------- >
40
+ file_manifest = []
41
+ spl = ["txt", "doc", "docx", "email", "epub", "html", "json", "md", "msg", "pdf", "ppt", "pptx", "rtf"]
42
+ for sp in spl:
43
+ _, file_manifest_tmp, _ = get_files_from_everything(txt, type=f'.{sp}')
44
+ file_manifest += file_manifest_tmp
45
+
46
+ if len(file_manifest) == 0:
47
+ chatbot.append(["没有找到任何可读取文件", "当前支持的格式包括: txt, md, docx, pptx, pdf, json等"])
48
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
49
+ return
50
+
51
+ # < -------------------预热文本向量化模组--------------- >
52
+ chatbot.append(['<br/>'.join(file_manifest), "正在预热文本向量化模组, 如果是第一次运行, 将消耗较长时间下载中文向量化模型..."])
53
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
54
+ print('Checking Text2vec ...')
55
+ from langchain.embeddings.huggingface import HuggingFaceEmbeddings
56
+ with ProxyNetworkActivate(): # 临时地激活代理网络
57
+ HuggingFaceEmbeddings(model_name="GanymedeNil/text2vec-large-chinese")
58
+
59
+ # < -------------------构建知识库--------------- >
60
+ chatbot.append(['<br/>'.join(file_manifest), "正在构建知识库..."])
61
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
62
+ print('Establishing knowledge archive ...')
63
+ with ProxyNetworkActivate(): # 临时地激活代理网络
64
+ kai = knowledge_archive_interface()
65
+ kai.feed_archive(file_manifest=file_manifest, id=kai_id)
66
+ kai_files = kai.get_loaded_file()
67
+ kai_files = '<br/>'.join(kai_files)
68
+ # chatbot.append(['知识库构建成功', "正在将知识库存储至cookie中"])
69
+ # yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
70
+ # chatbot._cookies['langchain_plugin_embedding'] = kai.get_current_archive_id()
71
+ # chatbot._cookies['lock_plugin'] = 'crazy_functions.Langchain知识库->读取知识库作答'
72
+ # chatbot.append(['完成', "“根据知识库作答”函数插件已经接管问答系统, 提问吧! 但注意, 您接下来不能再使用其他插件了,刷新页面即可以退出知识库问答模式。"])
73
+ chatbot.append(['构建完成', f"当前知识库内的有效文件:\n\n---\n\n{kai_files}\n\n---\n\n请切换至“知识库问答”插件进行知识库访问, 或者使用此插件继续上传更多文件。"])
74
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间,我们先及时地做一次界面更新
75
+
76
+ @CatchException
77
+ def 读取知识库作答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port=-1):
78
+ # resolve deps
79
+ try:
80
+ from zh_langchain import construct_vector_store
81
+ from langchain.embeddings.huggingface import HuggingFaceEmbeddings
82
+ from .crazy_utils import knowledge_archive_interface
83
+ except Exception as e:
84
+ chatbot.append(["依赖不足", "导入依赖失败。正在尝试自动安装,请查看终端的��出或耐心等待..."])
85
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
86
+ from .crazy_utils import try_install_deps
87
+ try_install_deps(['zh_langchain==0.2.1'])
88
+
89
+ # < ------------------- --------------- >
90
+ kai = knowledge_archive_interface()
91
+
92
+ if 'langchain_plugin_embedding' in chatbot._cookies:
93
+ resp, prompt = kai.answer_with_archive_by_id(txt, chatbot._cookies['langchain_plugin_embedding'])
94
+ else:
95
+ if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg")
96
+ kai_id = plugin_kwargs.get("advanced_arg", 'default')
97
+ resp, prompt = kai.answer_with_archive_by_id(txt, kai_id)
98
+
99
+ chatbot.append((txt, '[Local Message] ' + prompt))
100
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间,我们先及时地做一次界面更新
101
+ gpt_say = yield from request_gpt_model_in_new_thread_with_ui_alive(
102
+ inputs=prompt, inputs_show_user=txt,
103
+ llm_kwargs=llm_kwargs, chatbot=chatbot, history=[],
104
+ sys_prompt=system_prompt
105
+ )
106
+ history.extend((prompt, gpt_say))
107
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间,我们先及时地做一次界面更新
crazy_functions/Latex全文润色.py ADDED
@@ -0,0 +1,243 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import update_ui, trimmed_format_exc
2
+ from toolbox import CatchException, report_execption, write_results_to_file, zip_folder
3
+
4
+
5
+ class PaperFileGroup():
6
+ def __init__(self):
7
+ self.file_paths = []
8
+ self.file_contents = []
9
+ self.sp_file_contents = []
10
+ self.sp_file_index = []
11
+ self.sp_file_tag = []
12
+
13
+ # count_token
14
+ from request_llm.bridge_all import model_info
15
+ enc = model_info["gpt-3.5-turbo"]['tokenizer']
16
+ def get_token_num(txt): return len(enc.encode(txt, disallowed_special=()))
17
+ self.get_token_num = get_token_num
18
+
19
+ def run_file_split(self, max_token_limit=1900):
20
+ """
21
+ 将长文本分离开来
22
+ """
23
+ for index, file_content in enumerate(self.file_contents):
24
+ if self.get_token_num(file_content) < max_token_limit:
25
+ self.sp_file_contents.append(file_content)
26
+ self.sp_file_index.append(index)
27
+ self.sp_file_tag.append(self.file_paths[index])
28
+ else:
29
+ from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
30
+ segments = breakdown_txt_to_satisfy_token_limit_for_pdf(file_content, self.get_token_num, max_token_limit)
31
+ for j, segment in enumerate(segments):
32
+ self.sp_file_contents.append(segment)
33
+ self.sp_file_index.append(index)
34
+ self.sp_file_tag.append(self.file_paths[index] + f".part-{j}.tex")
35
+
36
+ print('Segmentation: done')
37
+ def merge_result(self):
38
+ self.file_result = ["" for _ in range(len(self.file_paths))]
39
+ for r, k in zip(self.sp_file_result, self.sp_file_index):
40
+ self.file_result[k] += r
41
+
42
+ def write_result(self):
43
+ manifest = []
44
+ for path, res in zip(self.file_paths, self.file_result):
45
+ with open(path + '.polish.tex', 'w', encoding='utf8') as f:
46
+ manifest.append(path + '.polish.tex')
47
+ f.write(res)
48
+ return manifest
49
+
50
+ def zip_result(self):
51
+ import os, time
52
+ folder = os.path.dirname(self.file_paths[0])
53
+ t = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime())
54
+ zip_folder(folder, './gpt_log/', f'{t}-polished.zip')
55
+
56
+
57
+ def 多文件润色(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en', mode='polish'):
58
+ import time, os, re
59
+ from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
60
+
61
+
62
+ # <-------- 读取Latex文件,删除其中的所有注释 ---------->
63
+ pfg = PaperFileGroup()
64
+
65
+ for index, fp in enumerate(file_manifest):
66
+ with open(fp, 'r', encoding='utf-8', errors='replace') as f:
67
+ file_content = f.read()
68
+ # 定义注释的正则表达式
69
+ comment_pattern = r'(?<!\\)%.*'
70
+ # 使用正则表达式查找注释,并替换为空字符串
71
+ clean_tex_content = re.sub(comment_pattern, '', file_content)
72
+ # 记录删除注释后的文本
73
+ pfg.file_paths.append(fp)
74
+ pfg.file_contents.append(clean_tex_content)
75
+
76
+ # <-------- 拆分过长的latex文件 ---------->
77
+ pfg.run_file_split(max_token_limit=1024)
78
+ n_split = len(pfg.sp_file_contents)
79
+
80
+
81
+ # <-------- 多线程润色开始 ---------->
82
+ if language == 'en':
83
+ if mode == 'polish':
84
+ inputs_array = ["Below is a section from an academic paper, polish this section to meet the academic standard, " +
85
+ "improve the grammar, clarity and overall readability, do not modify any latex command such as \section, \cite and equations:" +
86
+ f"\n\n{frag}" for frag in pfg.sp_file_contents]
87
+ else:
88
+ inputs_array = [r"Below is a section from an academic paper, proofread this section." +
89
+ r"Do not modify any latex command such as \section, \cite, \begin, \item and equations. " +
90
+ r"Answer me only with the revised text:" +
91
+ f"\n\n{frag}" for frag in pfg.sp_file_contents]
92
+ inputs_show_user_array = [f"Polish {f}" for f in pfg.sp_file_tag]
93
+ sys_prompt_array = ["You are a professional academic paper writer." for _ in range(n_split)]
94
+ elif language == 'zh':
95
+ if mode == 'polish':
96
+ inputs_array = [f"以下是一篇学术论文中的一段内容,请将此部分润色以满足学术标准,提高语法、清晰度和整体可读性,不要修改任何LaTeX命令,例如\section,\cite和方程式:" +
97
+ f"\n\n{frag}" for frag in pfg.sp_file_contents]
98
+ else:
99
+ inputs_array = [f"以下是一篇学术论文中的一段内容,请对这部分内容进行语法矫正。不要修改任何LaTeX命令,例如\section,\cite和方程式:" +
100
+ f"\n\n{frag}" for frag in pfg.sp_file_contents]
101
+ inputs_show_user_array = [f"润色 {f}" for f in pfg.sp_file_tag]
102
+ sys_prompt_array=["你是一位专业的中文学术论文作家。" for _ in range(n_split)]
103
+
104
+
105
+ gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
106
+ inputs_array=inputs_array,
107
+ inputs_show_user_array=inputs_show_user_array,
108
+ llm_kwargs=llm_kwargs,
109
+ chatbot=chatbot,
110
+ history_array=[[""] for _ in range(n_split)],
111
+ sys_prompt_array=sys_prompt_array,
112
+ # max_workers=5, # 并行任务数量限制,最多同时执行5个,其他的排队等待
113
+ scroller_max_len = 80
114
+ )
115
+
116
+ # <-------- 文本碎片重组为完整的tex文件,整理结果为压缩包 ---------->
117
+ try:
118
+ pfg.sp_file_result = []
119
+ for i_say, gpt_say in zip(gpt_response_collection[0::2], gpt_response_collection[1::2]):
120
+ pfg.sp_file_result.append(gpt_say)
121
+ pfg.merge_result()
122
+ pfg.write_result()
123
+ pfg.zip_result()
124
+ except:
125
+ print(trimmed_format_exc())
126
+
127
+ # <-------- 整理结果,退出 ---------->
128
+ create_report_file_name = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) + f"-chatgpt.polish.md"
129
+ res = write_results_to_file(gpt_response_collection, file_name=create_report_file_name)
130
+ history = gpt_response_collection
131
+ chatbot.append((f"{fp}完成了吗?", res))
132
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
133
+
134
+
135
+ @CatchException
136
+ def Latex英文润色(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
137
+ # 基本信息:功能、贡献者
138
+ chatbot.append([
139
+ "函数插件功能?",
140
+ "对整个Latex项目进行润色。函数插件贡献者: Binary-Husky"])
141
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
142
+
143
+ # 尝试导入依赖,如果缺少依赖,则给出安装建议
144
+ try:
145
+ import tiktoken
146
+ except:
147
+ report_execption(chatbot, history,
148
+ a=f"解析项目: {txt}",
149
+ b=f"导入软件依赖失败。使用该模块需要额外依赖,安装方法```pip install --upgrade tiktoken```。")
150
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
151
+ return
152
+ history = [] # 清空历史,以免输入溢出
153
+ import glob, os
154
+ if os.path.exists(txt):
155
+ project_folder = txt
156
+ else:
157
+ if txt == "": txt = '空空如也的输入栏'
158
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
159
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
160
+ return
161
+ file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
162
+ if len(file_manifest) == 0:
163
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
164
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
165
+ return
166
+ yield from 多文件润色(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en')
167
+
168
+
169
+
170
+
171
+
172
+
173
+ @CatchException
174
+ def Latex中文润色(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
175
+ # 基本信息:功能、贡献者
176
+ chatbot.append([
177
+ "函数插件功能?",
178
+ "对整个Latex项目进行润色。函数插件贡献者: Binary-Husky"])
179
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
180
+
181
+ # 尝试导入依赖,如果缺少依赖,则给出安装建议
182
+ try:
183
+ import tiktoken
184
+ except:
185
+ report_execption(chatbot, history,
186
+ a=f"解析项目: {txt}",
187
+ b=f"导入软件依赖失败。使用该模块需要额外依赖,安装方法```pip install --upgrade tiktoken```。")
188
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
189
+ return
190
+ history = [] # 清空历史,以免输入溢出
191
+ import glob, os
192
+ if os.path.exists(txt):
193
+ project_folder = txt
194
+ else:
195
+ if txt == "": txt = '空空如也的输入栏'
196
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
197
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
198
+ return
199
+ file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
200
+ if len(file_manifest) == 0:
201
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
202
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
203
+ return
204
+ yield from 多文件润色(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='zh')
205
+
206
+
207
+
208
+
209
+ @CatchException
210
+ def Latex英文纠错(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
211
+ # 基本信息:功能、贡献者
212
+ chatbot.append([
213
+ "函数插件功能?",
214
+ "对整个Latex项目进行纠错。函数插件贡献者: Binary-Husky"])
215
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
216
+
217
+ # 尝试导入依赖,如果缺少依赖,则给出安装建议
218
+ try:
219
+ import tiktoken
220
+ except:
221
+ report_execption(chatbot, history,
222
+ a=f"解析项目: {txt}",
223
+ b=f"导入软件依赖失败。使用该模块需要额外依赖,安装方法```pip install --upgrade tiktoken```。")
224
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
225
+ return
226
+ history = [] # 清空历史,以免输入溢出
227
+ import glob, os
228
+ if os.path.exists(txt):
229
+ project_folder = txt
230
+ else:
231
+ if txt == "": txt = '空空如也的输入栏'
232
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
233
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
234
+ return
235
+ file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
236
+ if len(file_manifest) == 0:
237
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
238
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
239
+ return
240
+ yield from 多文件润色(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en', mode='proofread')
241
+
242
+
243
+
crazy_functions/Latex全文翻译.py ADDED
@@ -0,0 +1,175 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import update_ui
2
+ from toolbox import CatchException, report_execption, write_results_to_file
3
+ fast_debug = False
4
+
5
+ class PaperFileGroup():
6
+ def __init__(self):
7
+ self.file_paths = []
8
+ self.file_contents = []
9
+ self.sp_file_contents = []
10
+ self.sp_file_index = []
11
+ self.sp_file_tag = []
12
+
13
+ # count_token
14
+ from request_llm.bridge_all import model_info
15
+ enc = model_info["gpt-3.5-turbo"]['tokenizer']
16
+ def get_token_num(txt): return len(enc.encode(txt, disallowed_special=()))
17
+ self.get_token_num = get_token_num
18
+
19
+ def run_file_split(self, max_token_limit=1900):
20
+ """
21
+ 将长文本分离开来
22
+ """
23
+ for index, file_content in enumerate(self.file_contents):
24
+ if self.get_token_num(file_content) < max_token_limit:
25
+ self.sp_file_contents.append(file_content)
26
+ self.sp_file_index.append(index)
27
+ self.sp_file_tag.append(self.file_paths[index])
28
+ else:
29
+ from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
30
+ segments = breakdown_txt_to_satisfy_token_limit_for_pdf(file_content, self.get_token_num, max_token_limit)
31
+ for j, segment in enumerate(segments):
32
+ self.sp_file_contents.append(segment)
33
+ self.sp_file_index.append(index)
34
+ self.sp_file_tag.append(self.file_paths[index] + f".part-{j}.tex")
35
+
36
+ print('Segmentation: done')
37
+
38
+ def 多文件翻译(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en'):
39
+ import time, os, re
40
+ from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
41
+
42
+ # <-------- 读取Latex文件,删除其中的所有注释 ---------->
43
+ pfg = PaperFileGroup()
44
+
45
+ for index, fp in enumerate(file_manifest):
46
+ with open(fp, 'r', encoding='utf-8', errors='replace') as f:
47
+ file_content = f.read()
48
+ # 定义注释的正则表达式
49
+ comment_pattern = r'(?<!\\)%.*'
50
+ # 使用正则表达式查找注释,并替换为空字符串
51
+ clean_tex_content = re.sub(comment_pattern, '', file_content)
52
+ # 记录删除注释后的文本
53
+ pfg.file_paths.append(fp)
54
+ pfg.file_contents.append(clean_tex_content)
55
+
56
+ # <-------- 拆分过长的latex文件 ---------->
57
+ pfg.run_file_split(max_token_limit=1024)
58
+ n_split = len(pfg.sp_file_contents)
59
+
60
+ # <-------- 抽取摘要 ---------->
61
+ # if language == 'en':
62
+ # abs_extract_inputs = f"Please write an abstract for this paper"
63
+
64
+ # # 单线,获取文章meta信息
65
+ # paper_meta_info = yield from request_gpt_model_in_new_thread_with_ui_alive(
66
+ # inputs=abs_extract_inputs,
67
+ # inputs_show_user=f"正在抽取摘要信息。",
68
+ # llm_kwargs=llm_kwargs,
69
+ # chatbot=chatbot, history=[],
70
+ # sys_prompt="Your job is to collect information from materials。",
71
+ # )
72
+
73
+ # <-------- 多线程润色开始 ---------->
74
+ if language == 'en->zh':
75
+ inputs_array = ["Below is a section from an English academic paper, translate it into Chinese, do not modify any latex command such as \section, \cite and equations:" +
76
+ f"\n\n{frag}" for frag in pfg.sp_file_contents]
77
+ inputs_show_user_array = [f"翻译 {f}" for f in pfg.sp_file_tag]
78
+ sys_prompt_array = ["You are a professional academic paper translator." for _ in range(n_split)]
79
+ elif language == 'zh->en':
80
+ inputs_array = [f"Below is a section from a Chinese academic paper, translate it into English, do not modify any latex command such as \section, \cite and equations:" +
81
+ f"\n\n{frag}" for frag in pfg.sp_file_contents]
82
+ inputs_show_user_array = [f"翻译 {f}" for f in pfg.sp_file_tag]
83
+ sys_prompt_array = ["You are a professional academic paper translator." for _ in range(n_split)]
84
+
85
+ gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
86
+ inputs_array=inputs_array,
87
+ inputs_show_user_array=inputs_show_user_array,
88
+ llm_kwargs=llm_kwargs,
89
+ chatbot=chatbot,
90
+ history_array=[[""] for _ in range(n_split)],
91
+ sys_prompt_array=sys_prompt_array,
92
+ # max_workers=5, # OpenAI所允许的最大并行过载
93
+ scroller_max_len = 80
94
+ )
95
+
96
+ # <-------- 整理结果,退出 ---------->
97
+ create_report_file_name = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) + f"-chatgpt.polish.md"
98
+ res = write_results_to_file(gpt_response_collection, file_name=create_report_file_name)
99
+ history = gpt_response_collection
100
+ chatbot.append((f"{fp}完成了吗?", res))
101
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
102
+
103
+
104
+
105
+
106
+
107
+ @CatchException
108
+ def Latex英译中(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
109
+ # 基本信息:功能、贡献者
110
+ chatbot.append([
111
+ "函数插件功能?",
112
+ "对整个Latex项目进行翻译。函数插件贡献者: Binary-Husky"])
113
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
114
+
115
+ # 尝试导入依赖,如果缺少依赖,则给出安装建议
116
+ try:
117
+ import tiktoken
118
+ except:
119
+ report_execption(chatbot, history,
120
+ a=f"解析项目: {txt}",
121
+ b=f"导入软件依赖失败。使用该模块需要额外依赖,安装方法```pip install --upgrade tiktoken```。")
122
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
123
+ return
124
+ history = [] # 清空历史,以免输入溢出
125
+ import glob, os
126
+ if os.path.exists(txt):
127
+ project_folder = txt
128
+ else:
129
+ if txt == "": txt = '空空如也的输入栏'
130
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
131
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
132
+ return
133
+ file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
134
+ if len(file_manifest) == 0:
135
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
136
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
137
+ return
138
+ yield from 多文件翻译(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='en->zh')
139
+
140
+
141
+
142
+
143
+
144
+ @CatchException
145
+ def Latex中译英(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
146
+ # 基本信息:功能、贡献者
147
+ chatbot.append([
148
+ "函数插件功能?",
149
+ "对整个Latex项目进行翻译。函数插件贡献者: Binary-Husky"])
150
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
151
+
152
+ # 尝试导入依赖,如果缺少依赖,则给出安装建议
153
+ try:
154
+ import tiktoken
155
+ except:
156
+ report_execption(chatbot, history,
157
+ a=f"解析项目: {txt}",
158
+ b=f"导入软件依赖失败。使用该模块需要额外依赖,安装方法```pip install --upgrade tiktoken```。")
159
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
160
+ return
161
+ history = [] # 清空历史,以免输入溢出
162
+ import glob, os
163
+ if os.path.exists(txt):
164
+ project_folder = txt
165
+ else:
166
+ if txt == "": txt = '空空如也的输入栏'
167
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
168
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
169
+ return
170
+ file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
171
+ if len(file_manifest) == 0:
172
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
173
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
174
+ return
175
+ yield from 多文件翻译(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, language='zh->en')
crazy_functions/Latex输出PDF结果.py ADDED
@@ -0,0 +1,300 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import update_ui, trimmed_format_exc, get_conf, objdump, objload, promote_file_to_downloadzone
2
+ from toolbox import CatchException, report_execption, update_ui_lastest_msg, zip_result, gen_time_str
3
+ from functools import partial
4
+ import glob, os, requests, time
5
+ pj = os.path.join
6
+ ARXIV_CACHE_DIR = os.path.expanduser(f"~/arxiv_cache/")
7
+
8
+ # =================================== 工具函数 ===============================================
9
+ 专业词汇声明 = 'If the term "agent" is used in this section, it should be translated to "智能体". '
10
+ def switch_prompt(pfg, mode, more_requirement):
11
+ """
12
+ Generate prompts and system prompts based on the mode for proofreading or translating.
13
+ Args:
14
+ - pfg: Proofreader or Translator instance.
15
+ - mode: A string specifying the mode, either 'proofread' or 'translate_zh'.
16
+
17
+ Returns:
18
+ - inputs_array: A list of strings containing prompts for users to respond to.
19
+ - sys_prompt_array: A list of strings containing prompts for system prompts.
20
+ """
21
+ n_split = len(pfg.sp_file_contents)
22
+ if mode == 'proofread_en':
23
+ inputs_array = [r"Below is a section from an academic paper, proofread this section." +
24
+ r"Do not modify any latex command such as \section, \cite, \begin, \item and equations. " + more_requirement +
25
+ r"Answer me only with the revised text:" +
26
+ f"\n\n{frag}" for frag in pfg.sp_file_contents]
27
+ sys_prompt_array = ["You are a professional academic paper writer." for _ in range(n_split)]
28
+ elif mode == 'translate_zh':
29
+ inputs_array = [r"Below is a section from an English academic paper, translate it into Chinese. " + more_requirement +
30
+ r"Do not modify any latex command such as \section, \cite, \begin, \item and equations. " +
31
+ r"Answer me only with the translated text:" +
32
+ f"\n\n{frag}" for frag in pfg.sp_file_contents]
33
+ sys_prompt_array = ["You are a professional translator." for _ in range(n_split)]
34
+ else:
35
+ assert False, "未知指令"
36
+ return inputs_array, sys_prompt_array
37
+
38
+ def desend_to_extracted_folder_if_exist(project_folder):
39
+ """
40
+ Descend into the extracted folder if it exists, otherwise return the original folder.
41
+
42
+ Args:
43
+ - project_folder: A string specifying the folder path.
44
+
45
+ Returns:
46
+ - A string specifying the path to the extracted folder, or the original folder if there is no extracted folder.
47
+ """
48
+ maybe_dir = [f for f in glob.glob(f'{project_folder}/*') if os.path.isdir(f)]
49
+ if len(maybe_dir) == 0: return project_folder
50
+ if maybe_dir[0].endswith('.extract'): return maybe_dir[0]
51
+ return project_folder
52
+
53
+ def move_project(project_folder, arxiv_id=None):
54
+ """
55
+ Create a new work folder and copy the project folder to it.
56
+
57
+ Args:
58
+ - project_folder: A string specifying the folder path of the project.
59
+
60
+ Returns:
61
+ - A string specifying the path to the new work folder.
62
+ """
63
+ import shutil, time
64
+ time.sleep(2) # avoid time string conflict
65
+ if arxiv_id is not None:
66
+ new_workfolder = pj(ARXIV_CACHE_DIR, arxiv_id, 'workfolder')
67
+ else:
68
+ new_workfolder = f'gpt_log/{gen_time_str()}'
69
+ try:
70
+ shutil.rmtree(new_workfolder)
71
+ except:
72
+ pass
73
+
74
+ # align subfolder if there is a folder wrapper
75
+ items = glob.glob(pj(project_folder,'*'))
76
+ if len(glob.glob(pj(project_folder,'*.tex'))) == 0 and len(items) == 1:
77
+ if os.path.isdir(items[0]): project_folder = items[0]
78
+
79
+ shutil.copytree(src=project_folder, dst=new_workfolder)
80
+ return new_workfolder
81
+
82
+ def arxiv_download(chatbot, history, txt):
83
+ def check_cached_translation_pdf(arxiv_id):
84
+ translation_dir = pj(ARXIV_CACHE_DIR, arxiv_id, 'translation')
85
+ if not os.path.exists(translation_dir):
86
+ os.makedirs(translation_dir)
87
+ target_file = pj(translation_dir, 'translate_zh.pdf')
88
+ if os.path.exists(target_file):
89
+ promote_file_to_downloadzone(target_file, rename_file=None, chatbot=chatbot)
90
+ return target_file
91
+ return False
92
+ def is_float(s):
93
+ try:
94
+ float(s)
95
+ return True
96
+ except ValueError:
97
+ return False
98
+ if ('.' in txt) and ('/' not in txt) and is_float(txt): # is arxiv ID
99
+ txt = 'https://arxiv.org/abs/' + txt.strip()
100
+ if ('.' in txt) and ('/' not in txt) and is_float(txt[:10]): # is arxiv ID
101
+ txt = 'https://arxiv.org/abs/' + txt[:10]
102
+ if not txt.startswith('https://arxiv.org'):
103
+ return txt, None
104
+
105
+ # <-------------- inspect format ------------->
106
+ chatbot.append([f"检测到arxiv文档连接", '尝试下载 ...'])
107
+ yield from update_ui(chatbot=chatbot, history=history)
108
+ time.sleep(1) # 刷新界面
109
+
110
+ url_ = txt # https://arxiv.org/abs/1707.06690
111
+ if not txt.startswith('https://arxiv.org/abs/'):
112
+ msg = f"解析arxiv网址失败, 期望格式例如: https://arxiv.org/abs/1707.06690。实际得到格式: {url_}"
113
+ yield from update_ui_lastest_msg(msg, chatbot=chatbot, history=history) # 刷新界面
114
+ return msg, None
115
+ # <-------------- set format ------------->
116
+ arxiv_id = url_.split('/abs/')[-1]
117
+ if 'v' in arxiv_id: arxiv_id = arxiv_id[:10]
118
+ cached_translation_pdf = check_cached_translation_pdf(arxiv_id)
119
+ if cached_translation_pdf: return cached_translation_pdf, arxiv_id
120
+
121
+ url_tar = url_.replace('/abs/', '/e-print/')
122
+ translation_dir = pj(ARXIV_CACHE_DIR, arxiv_id, 'e-print')
123
+ extract_dst = pj(ARXIV_CACHE_DIR, arxiv_id, 'extract')
124
+ os.makedirs(translation_dir, exist_ok=True)
125
+
126
+ # <-------------- download arxiv source file ------------->
127
+ dst = pj(translation_dir, arxiv_id+'.tar')
128
+ if os.path.exists(dst):
129
+ yield from update_ui_lastest_msg("调用缓存", chatbot=chatbot, history=history) # 刷新界面
130
+ else:
131
+ yield from update_ui_lastest_msg("开始下载", chatbot=chatbot, history=history) # 刷新界面
132
+ proxies, = get_conf('proxies')
133
+ r = requests.get(url_tar, proxies=proxies)
134
+ with open(dst, 'wb+') as f:
135
+ f.write(r.content)
136
+ # <-------------- extract file ------------->
137
+ yield from update_ui_lastest_msg("下载完成", chatbot=chatbot, history=history) # 刷新界面
138
+ from toolbox import extract_archive
139
+ extract_archive(file_path=dst, dest_dir=extract_dst)
140
+ return extract_dst, arxiv_id
141
+ # ========================================= 插件主程序1 =====================================================
142
+
143
+
144
+ @CatchException
145
+ def Latex英文纠错加PDF对比(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
146
+ # <-------------- information about this plugin ------------->
147
+ chatbot.append([ "函数插件功能?",
148
+ "对整个Latex项目进行纠错, 用latex编译为PDF对修正处做高亮。函数插件贡献者: Binary-Husky。注意事项: 目前仅支持GPT3.5/GPT4,其他模型转化效果未知。目前对机器学习类文献转化效果最好,其他类型文献转化效果未知。仅在Windows系统进行了测试,其他操作系统表现未知。"])
149
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
150
+
151
+ # <-------------- more requirements ------------->
152
+ if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg")
153
+ more_req = plugin_kwargs.get("advanced_arg", "")
154
+ _switch_prompt_ = partial(switch_prompt, more_requirement=more_req)
155
+
156
+ # <-------------- check deps ------------->
157
+ try:
158
+ import glob, os, time, subprocess
159
+ subprocess.Popen(['pdflatex', '-version'])
160
+ from .latex_fns.latex_actions import Latex精细分解与转化, 编译Latex
161
+ except Exception as e:
162
+ chatbot.append([ f"解析项目: {txt}",
163
+ f"尝试执行Latex指令失败。Latex没有安装, 或者不在环境变量PATH中。安装方法https://tug.org/texlive/。报错信息\n\n```\n\n{trimmed_format_exc()}\n\n```\n\n"])
164
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
165
+ return
166
+
167
+
168
+ # <-------------- clear history and read input ------------->
169
+ history = []
170
+ if os.path.exists(txt):
171
+ project_folder = txt
172
+ else:
173
+ if txt == "": txt = '空空如也的输入栏'
174
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
175
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
176
+ return
177
+ file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
178
+ if len(file_manifest) == 0:
179
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
180
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
181
+ return
182
+
183
+
184
+ # <-------------- if is a zip/tar file ------------->
185
+ project_folder = desend_to_extracted_folder_if_exist(project_folder)
186
+
187
+
188
+ # <-------------- move latex project away from temp folder ------------->
189
+ project_folder = move_project(project_folder, arxiv_id=None)
190
+
191
+
192
+ # <-------------- if merge_translate_zh is already generated, skip gpt req ------------->
193
+ if not os.path.exists(project_folder + '/merge_proofread_en.tex'):
194
+ yield from Latex精细分解与转化(file_manifest, project_folder, llm_kwargs, plugin_kwargs,
195
+ chatbot, history, system_prompt, mode='proofread_en', switch_prompt=_switch_prompt_)
196
+
197
+
198
+ # <-------------- compile PDF ------------->
199
+ success = yield from 编译Latex(chatbot, history, main_file_original='merge', main_file_modified='merge_proofread_en',
200
+ work_folder_original=project_folder, work_folder_modified=project_folder, work_folder=project_folder)
201
+
202
+
203
+ # <-------------- zip PDF ------------->
204
+ zip_res = zip_result(project_folder)
205
+ if success:
206
+ chatbot.append((f"成功啦", '请查收结果(压缩包)...'))
207
+ yield from update_ui(chatbot=chatbot, history=history); time.sleep(1) # 刷新界面
208
+ promote_file_to_downloadzone(file=zip_res, chatbot=chatbot)
209
+ else:
210
+ chatbot.append((f"失败了", '虽然PDF生成失败了, 但请查收结果(压缩包), 内含已经翻译的Tex文档, 也是可读的, 您可以到Github Issue区, 用该压缩包+对话历史存档进行反馈 ...'))
211
+ yield from update_ui(chatbot=chatbot, history=history); time.sleep(1) # 刷新界面
212
+ promote_file_to_downloadzone(file=zip_res, chatbot=chatbot)
213
+
214
+ # <-------------- we are done ------------->
215
+ return success
216
+
217
+
218
+ # ========================================= 插件主程序2 =====================================================
219
+
220
+ @CatchException
221
+ def Latex翻译中文并重新编译PDF(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
222
+ # <-------------- information about this plugin ------------->
223
+ chatbot.append([
224
+ "函数插件功能?",
225
+ "对整个Latex项目进行翻译, 生成中文PDF。函数插件贡献者: Binary-Husky。注意事项: 此插件Windows支持最佳,Linux下必须使用Docker安装,详见项目主README.md。目前仅支持GPT3.5/GPT4,其他模型转化效果未知。目前对机器学习类文献转化效果最好,其他类型文献转化效果未知。"])
226
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
227
+
228
+ # <-------------- more requirements ------------->
229
+ if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg")
230
+ more_req = plugin_kwargs.get("advanced_arg", "")
231
+ _switch_prompt_ = partial(switch_prompt, more_requirement=more_req)
232
+
233
+ # <-------------- check deps ------------->
234
+ try:
235
+ import glob, os, time, subprocess
236
+ subprocess.Popen(['pdflatex', '-version'])
237
+ from .latex_fns.latex_actions import Latex精细分解与转化, 编译Latex
238
+ except Exception as e:
239
+ chatbot.append([ f"解析项目: {txt}",
240
+ f"尝试执行Latex指令失败。Latex没有安装, 或者不在环境变量PATH中。安装方法https://tug.org/texlive/。报错信息\n\n```\n\n{trimmed_format_exc()}\n\n```\n\n"])
241
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
242
+ return
243
+
244
+
245
+ # <-------------- clear history and read input ------------->
246
+ history = []
247
+ txt, arxiv_id = yield from arxiv_download(chatbot, history, txt)
248
+ if txt.endswith('.pdf'):
249
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"发现已经存在翻译好的PDF文档")
250
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
251
+ return
252
+
253
+
254
+ if os.path.exists(txt):
255
+ project_folder = txt
256
+ else:
257
+ if txt == "": txt = '空空如也的输入栏'
258
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
259
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
260
+ return
261
+
262
+ file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)]
263
+ if len(file_manifest) == 0:
264
+ report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
265
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
266
+ return
267
+
268
+
269
+ # <-------------- if is a zip/tar file ------------->
270
+ project_folder = desend_to_extracted_folder_if_exist(project_folder)
271
+
272
+
273
+ # <-------------- move latex project away from temp folder ------------->
274
+ project_folder = move_project(project_folder, arxiv_id)
275
+
276
+
277
+ # <-------------- if merge_translate_zh is already generated, skip gpt req ------------->
278
+ if not os.path.exists(project_folder + '/merge_translate_zh.tex'):
279
+ yield from Latex精细分解与转化(file_manifest, project_folder, llm_kwargs, plugin_kwargs,
280
+ chatbot, history, system_prompt, mode='translate_zh', switch_prompt=_switch_prompt_)
281
+
282
+
283
+ # <-------------- compile PDF ------------->
284
+ success = yield from 编译Latex(chatbot, history, main_file_original='merge', main_file_modified='merge_translate_zh', mode='translate_zh',
285
+ work_folder_original=project_folder, work_folder_modified=project_folder, work_folder=project_folder)
286
+
287
+ # <-------------- zip PDF ------------->
288
+ zip_res = zip_result(project_folder)
289
+ if success:
290
+ chatbot.append((f"成功啦", '请查收结果(压缩包)...'))
291
+ yield from update_ui(chatbot=chatbot, history=history); time.sleep(1) # 刷新界面
292
+ promote_file_to_downloadzone(file=zip_res, chatbot=chatbot)
293
+ else:
294
+ chatbot.append((f"失败了", '虽然PDF生成失败了, 但请查收结果(压缩包), 内含已经翻译的Tex文档, 也是可读的, 您可以��Github Issue区, 用该压缩包+对话历史存档进行反馈 ...'))
295
+ yield from update_ui(chatbot=chatbot, history=history); time.sleep(1) # 刷新界面
296
+ promote_file_to_downloadzone(file=zip_res, chatbot=chatbot)
297
+
298
+
299
+ # <-------------- we are done ------------->
300
+ return success
crazy_functions/__init__.py ADDED
File without changes
crazy_functions/chatglm微调工具.py ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import CatchException, update_ui, promote_file_to_downloadzone
2
+ from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
3
+ import datetime, json
4
+
5
+ def fetch_items(list_of_items, batch_size):
6
+ for i in range(0, len(list_of_items), batch_size):
7
+ yield list_of_items[i:i + batch_size]
8
+
9
+ def string_to_options(arguments):
10
+ import argparse
11
+ import shlex
12
+
13
+ # Create an argparse.ArgumentParser instance
14
+ parser = argparse.ArgumentParser()
15
+
16
+ # Add command-line arguments
17
+ parser.add_argument("--llm_to_learn", type=str, help="LLM model to learn", default="gpt-3.5-turbo")
18
+ parser.add_argument("--prompt_prefix", type=str, help="Prompt prefix", default='')
19
+ parser.add_argument("--system_prompt", type=str, help="System prompt", default='')
20
+ parser.add_argument("--batch", type=int, help="System prompt", default=50)
21
+ parser.add_argument("--pre_seq_len", type=int, help="pre_seq_len", default=50)
22
+ parser.add_argument("--learning_rate", type=float, help="learning_rate", default=2e-2)
23
+ parser.add_argument("--num_gpus", type=int, help="num_gpus", default=1)
24
+ parser.add_argument("--json_dataset", type=str, help="json_dataset", default="")
25
+ parser.add_argument("--ptuning_directory", type=str, help="ptuning_directory", default="")
26
+
27
+
28
+
29
+ # Parse the arguments
30
+ args = parser.parse_args(shlex.split(arguments))
31
+
32
+ return args
33
+
34
+ @CatchException
35
+ def 微调数据集生成(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
36
+ """
37
+ txt 输入栏用户输入的文本,例如需要翻译的一段话,再例如一个包含了待处理文件的路径
38
+ llm_kwargs gpt模型参数,如温度和top_p等,一般原样传递下去就行
39
+ plugin_kwargs 插件模型的参数
40
+ chatbot 聊天显示框的句柄,用于显示给用户
41
+ history 聊天历史,前情提要
42
+ system_prompt 给gpt的静默提醒
43
+ web_port 当前软件运行的端口号
44
+ """
45
+ history = [] # 清空历史,以免输入溢出
46
+ chatbot.append(("这是什么功能?", "[Local Message] 微调数据集生成"))
47
+ if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg")
48
+ args = plugin_kwargs.get("advanced_arg", None)
49
+ if args is None:
50
+ chatbot.append(("没给定指令", "退出"))
51
+ yield from update_ui(chatbot=chatbot, history=history); return
52
+ else:
53
+ arguments = string_to_options(arguments=args)
54
+
55
+ dat = []
56
+ with open(txt, 'r', encoding='utf8') as f:
57
+ for line in f.readlines():
58
+ json_dat = json.loads(line)
59
+ dat.append(json_dat["content"])
60
+
61
+ llm_kwargs['llm_model'] = arguments.llm_to_learn
62
+ for batch in fetch_items(dat, arguments.batch):
63
+ res = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
64
+ inputs_array=[f"{arguments.prompt_prefix}\n\n{b}" for b in (batch)],
65
+ inputs_show_user_array=[f"Show Nothing" for _ in (batch)],
66
+ llm_kwargs=llm_kwargs,
67
+ chatbot=chatbot,
68
+ history_array=[[] for _ in (batch)],
69
+ sys_prompt_array=[arguments.system_prompt for _ in (batch)],
70
+ max_workers=10 # OpenAI所允许的最大并行过载
71
+ )
72
+
73
+ with open(txt+'.generated.json', 'a+', encoding='utf8') as f:
74
+ for b, r in zip(batch, res[1::2]):
75
+ f.write(json.dumps({"content":b, "summary":r}, ensure_ascii=False)+'\n')
76
+
77
+ promote_file_to_downloadzone(txt+'.generated.json', rename_file='generated.json', chatbot=chatbot)
78
+ return
79
+
80
+
81
+
82
+ @CatchException
83
+ def 启动微调(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
84
+ """
85
+ txt 输入栏用户输入的文本,例如需要翻译的一段话,再例如一个包含了待处理文件的路径
86
+ llm_kwargs gpt模型参数,如温度和top_p等,一般原样传递下去就行
87
+ plugin_kwargs 插件模型的参数
88
+ chatbot 聊天显示框的句柄,用于显示给用户
89
+ history 聊天历史,前情提要
90
+ system_prompt 给gpt的静默提醒
91
+ web_port 当前软件运行的端口号
92
+ """
93
+ import subprocess
94
+ history = [] # 清空历史,以免输入溢出
95
+ chatbot.append(("这是什么功能?", "[Local Message] 微调数据集生成"))
96
+ if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg")
97
+ args = plugin_kwargs.get("advanced_arg", None)
98
+ if args is None:
99
+ chatbot.append(("没给定指令", "退出"))
100
+ yield from update_ui(chatbot=chatbot, history=history); return
101
+ else:
102
+ arguments = string_to_options(arguments=args)
103
+
104
+
105
+
106
+ pre_seq_len = arguments.pre_seq_len # 128
107
+ learning_rate = arguments.learning_rate # 2e-2
108
+ num_gpus = arguments.num_gpus # 1
109
+ json_dataset = arguments.json_dataset # 't_code.json'
110
+ ptuning_directory = arguments.ptuning_directory # '/home/hmp/ChatGLM2-6B/ptuning'
111
+
112
+ command = f"torchrun --standalone --nnodes=1 --nproc-per-node={num_gpus} main.py \
113
+ --do_train \
114
+ --train_file AdvertiseGen/{json_dataset} \
115
+ --validation_file AdvertiseGen/{json_dataset} \
116
+ --preprocessing_num_workers 20 \
117
+ --prompt_column content \
118
+ --response_column summary \
119
+ --overwrite_cache \
120
+ --model_name_or_path THUDM/chatglm2-6b \
121
+ --output_dir output/clothgen-chatglm2-6b-pt-{pre_seq_len}-{learning_rate} \
122
+ --overwrite_output_dir \
123
+ --max_source_length 256 \
124
+ --max_target_length 256 \
125
+ --per_device_train_batch_size 1 \
126
+ --per_device_eval_batch_size 1 \
127
+ --gradient_accumulation_steps 16 \
128
+ --predict_with_generate \
129
+ --max_steps 100 \
130
+ --logging_steps 10 \
131
+ --save_steps 20 \
132
+ --learning_rate {learning_rate} \
133
+ --pre_seq_len {pre_seq_len} \
134
+ --quantization_bit 4"
135
+
136
+ process = subprocess.Popen(command, shell=True, cwd=ptuning_directory)
137
+ try:
138
+ process.communicate(timeout=3600*24)
139
+ except subprocess.TimeoutExpired:
140
+ process.kill()
141
+ return
crazy_functions/crazy_functions_test.py ADDED
@@ -0,0 +1,231 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ 这是什么?
3
+ 这个文件用于函数插件的单元测试
4
+ 运行方法 python crazy_functions/crazy_functions_test.py
5
+ """
6
+
7
+ # ==============================================================================================================================
8
+
9
+ def validate_path():
10
+ import os, sys
11
+ dir_name = os.path.dirname(__file__)
12
+ root_dir_assume = os.path.abspath(os.path.dirname(__file__) + '/..')
13
+ os.chdir(root_dir_assume)
14
+ sys.path.append(root_dir_assume)
15
+ validate_path() # validate path so you can run from base directory
16
+
17
+ # ==============================================================================================================================
18
+
19
+ from colorful import *
20
+ from toolbox import get_conf, ChatBotWithCookies
21
+ import contextlib
22
+ import os
23
+ import sys
24
+ from functools import wraps
25
+ proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT, LAYOUT, API_KEY = \
26
+ get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION', 'CHATBOT_HEIGHT', 'LAYOUT', 'API_KEY')
27
+
28
+ llm_kwargs = {
29
+ 'api_key': API_KEY,
30
+ 'llm_model': LLM_MODEL,
31
+ 'top_p':1.0,
32
+ 'max_length': None,
33
+ 'temperature':1.0,
34
+ }
35
+ plugin_kwargs = { }
36
+ chatbot = ChatBotWithCookies(llm_kwargs)
37
+ history = []
38
+ system_prompt = "Serve me as a writing and programming assistant."
39
+ web_port = 1024
40
+
41
+ # ==============================================================================================================================
42
+
43
+ def silence_stdout(func):
44
+ @wraps(func)
45
+ def wrapper(*args, **kwargs):
46
+ _original_stdout = sys.stdout
47
+ sys.stdout = open(os.devnull, 'w')
48
+ for q in func(*args, **kwargs):
49
+ sys.stdout = _original_stdout
50
+ yield q
51
+ sys.stdout = open(os.devnull, 'w')
52
+ sys.stdout.close()
53
+ sys.stdout = _original_stdout
54
+ return wrapper
55
+
56
+ class CLI_Printer():
57
+ def __init__(self) -> None:
58
+ self.pre_buf = ""
59
+
60
+ def print(self, buf):
61
+ bufp = ""
62
+ for index, chat in enumerate(buf):
63
+ a, b = chat
64
+ bufp += sprint亮靛('[Me]:' + a) + '\n'
65
+ bufp += '[GPT]:' + b
66
+ if index < len(buf)-1:
67
+ bufp += '\n'
68
+
69
+ if self.pre_buf!="" and bufp.startswith(self.pre_buf):
70
+ print(bufp[len(self.pre_buf):], end='')
71
+ else:
72
+ print('\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n'+bufp, end='')
73
+ self.pre_buf = bufp
74
+ return
75
+
76
+ cli_printer = CLI_Printer()
77
+ # ==============================================================================================================================
78
+ def test_解析一个Python项目():
79
+ from crazy_functions.解析项目源代码 import 解析一个Python项目
80
+ txt = "crazy_functions/test_project/python/dqn"
81
+ for cookies, cb, hist, msg in 解析一个Python项目(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
82
+ print(cb)
83
+
84
+ def test_解析一个Cpp项目():
85
+ from crazy_functions.解析项目源代码 import 解析一个C项目
86
+ txt = "crazy_functions/test_project/cpp/cppipc"
87
+ for cookies, cb, hist, msg in 解析一个C项目(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
88
+ print(cb)
89
+
90
+ def test_Latex英文润色():
91
+ from crazy_functions.Latex全文润色 import Latex英文润色
92
+ txt = "crazy_functions/test_project/latex/attention"
93
+ for cookies, cb, hist, msg in Latex英文润色(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
94
+ print(cb)
95
+
96
+ def test_Markdown中译英():
97
+ from crazy_functions.批量Markdown翻译 import Markdown中译英
98
+ txt = "README.md"
99
+ for cookies, cb, hist, msg in Markdown中译英(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
100
+ print(cb)
101
+
102
+ def test_批量翻译PDF文档():
103
+ from crazy_functions.批量翻译PDF文档_多线程 import 批量翻译PDF文档
104
+ txt = "crazy_functions/test_project/pdf_and_word"
105
+ for cookies, cb, hist, msg in 批量翻译PDF文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
106
+ print(cb)
107
+
108
+ def test_谷歌检索小助手():
109
+ from crazy_functions.谷歌检索小助手 import 谷歌检索小助手
110
+ txt = "https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=auto+reinforcement+learning&btnG="
111
+ for cookies, cb, hist, msg in 谷歌检索小助手(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
112
+ print(cb)
113
+
114
+ def test_总结word文档():
115
+ from crazy_functions.总结word文档 import 总结word文档
116
+ txt = "crazy_functions/test_project/pdf_and_word"
117
+ for cookies, cb, hist, msg in 总结word文档(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
118
+ print(cb)
119
+
120
+ def test_下载arxiv论文并翻译摘要():
121
+ from crazy_functions.下载arxiv论文翻译摘要 import 下载arxiv论文并翻译摘要
122
+ txt = "1812.10695"
123
+ for cookies, cb, hist, msg in 下载arxiv论文并翻译摘要(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
124
+ print(cb)
125
+
126
+ def test_联网回答问题():
127
+ from crazy_functions.联网的ChatGPT import 连接网络回答问题
128
+ # txt = "谁是应急食品?"
129
+ # >> '根据以上搜索结果可以得知,应急食品是“原神”游戏中的角色派蒙的外号。'
130
+ # txt = "道路千万条,安全第一条。后面两句是?"
131
+ # >> '行车不规范,亲人两行泪。'
132
+ # txt = "You should have gone for the head. What does that mean?"
133
+ # >> The phrase "You should have gone for the head" is a quote from the Marvel movies, Avengers: Infinity War and Avengers: Endgame. It was spoken by the character Thanos in Infinity War and by Thor in Endgame.
134
+ txt = "AutoGPT是什么?"
135
+ for cookies, cb, hist, msg in 连接网络回答问题(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
136
+ print("当前问答:", cb[-1][-1].replace("\n"," "))
137
+ for i, it in enumerate(cb): print亮蓝(it[0]); print亮黄(it[1])
138
+
139
+ def test_解析ipynb文件():
140
+ from crazy_functions.解析JupyterNotebook import 解析ipynb文件
141
+ txt = "crazy_functions/test_samples"
142
+ for cookies, cb, hist, msg in 解析ipynb文件(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
143
+ print(cb)
144
+
145
+
146
+ def test_数学动画生成manim():
147
+ from crazy_functions.数学动画生成manim import 动画生成
148
+ txt = "A ball split into 2, and then split into 4, and finally split into 8."
149
+ for cookies, cb, hist, msg in 动画生成(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
150
+ print(cb)
151
+
152
+
153
+
154
+ def test_Markdown多语言():
155
+ from crazy_functions.批量Markdown翻译 import Markdown翻译指定语言
156
+ txt = "README.md"
157
+ history = []
158
+ for lang in ["English", "French", "Japanese", "Korean", "Russian", "Italian", "German", "Portuguese", "Arabic"]:
159
+ plugin_kwargs = {"advanced_arg": lang}
160
+ for cookies, cb, hist, msg in Markdown翻译指定语言(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
161
+ print(cb)
162
+
163
+ def test_Langchain知识库():
164
+ from crazy_functions.Langchain知识库 import 知识库问答
165
+ txt = "./"
166
+ chatbot = ChatBotWithCookies(llm_kwargs)
167
+ for cookies, cb, hist, msg in silence_stdout(知识库问答)(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
168
+ cli_printer.print(cb) # print(cb)
169
+
170
+ chatbot = ChatBotWithCookies(cookies)
171
+ from crazy_functions.Langchain知识库 import 读取知识库作答
172
+ txt = "What is the installation method?"
173
+ for cookies, cb, hist, msg in silence_stdout(读取知识库作答)(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
174
+ cli_printer.print(cb) # print(cb)
175
+
176
+ def test_Langchain知识库读取():
177
+ from crazy_functions.Langchain知识库 import 读取知识库作答
178
+ txt = "远程云服务器部署?"
179
+ for cookies, cb, hist, msg in silence_stdout(读取知识库作答)(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
180
+ cli_printer.print(cb) # print(cb)
181
+
182
+ def test_Latex():
183
+ from crazy_functions.Latex输出PDF结果 import Latex英文纠错加PDF对比, Latex翻译中文并重新编译PDF
184
+
185
+ # txt = r"https://arxiv.org/abs/1706.03762"
186
+ # txt = r"https://arxiv.org/abs/1902.03185"
187
+ # txt = r"https://arxiv.org/abs/2305.18290"
188
+ # txt = r"https://arxiv.org/abs/2305.17608"
189
+ # txt = r"https://arxiv.org/abs/2211.16068" # ACE
190
+ # txt = r"C:\Users\x\arxiv_cache\2211.16068\workfolder" # ACE
191
+ # txt = r"https://arxiv.org/abs/2002.09253"
192
+ # txt = r"https://arxiv.org/abs/2306.07831"
193
+ # txt = r"https://arxiv.org/abs/2212.10156"
194
+ # txt = r"https://arxiv.org/abs/2211.11559"
195
+ # txt = r"https://arxiv.org/abs/2303.08774"
196
+ txt = r"https://arxiv.org/abs/2303.12712"
197
+ # txt = r"C:\Users\fuqingxu\arxiv_cache\2303.12712\workfolder"
198
+
199
+
200
+ for cookies, cb, hist, msg in (Latex翻译中文并重新编译PDF)(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
201
+ cli_printer.print(cb) # print(cb)
202
+
203
+
204
+
205
+ # txt = "2302.02948.tar"
206
+ # print(txt)
207
+ # main_tex, work_folder = Latex预处理(txt)
208
+ # print('main tex:', main_tex)
209
+ # res = 编译Latex(main_tex, work_folder)
210
+ # # for cookies, cb, hist, msg in silence_stdout(编译Latex)(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port):
211
+ # cli_printer.print(cb) # print(cb)
212
+
213
+
214
+
215
+ # test_解析一个Python项目()
216
+ # test_Latex英文润色()
217
+ # test_Markdown中译英()
218
+ # test_批量翻译PDF文档()
219
+ # test_谷歌检索小助手()
220
+ # test_总结word文档()
221
+ # test_下载arxiv论文并翻译摘要()
222
+ # test_解析一个Cpp项目()
223
+ # test_联网回答问题()
224
+ # test_解析ipynb文件()
225
+ # test_数学动画生成manim()
226
+ # test_Langchain知识库()
227
+ # test_Langchain知识库读取()
228
+ if __name__ == "__main__":
229
+ test_Latex()
230
+ input("程序完成,回车退出。")
231
+ print("退出。")
crazy_functions/crazy_utils.py ADDED
@@ -0,0 +1,753 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import update_ui, get_conf, trimmed_format_exc
2
+ import threading
3
+
4
+ def input_clipping(inputs, history, max_token_limit):
5
+ import numpy as np
6
+ from request_llm.bridge_all import model_info
7
+ enc = model_info["gpt-3.5-turbo"]['tokenizer']
8
+ def get_token_num(txt): return len(enc.encode(txt, disallowed_special=()))
9
+
10
+ mode = 'input-and-history'
11
+ # 当 输入部分的token占比 小于 全文的一半时,只裁剪历史
12
+ input_token_num = get_token_num(inputs)
13
+ if input_token_num < max_token_limit//2:
14
+ mode = 'only-history'
15
+ max_token_limit = max_token_limit - input_token_num
16
+
17
+ everything = [inputs] if mode == 'input-and-history' else ['']
18
+ everything.extend(history)
19
+ n_token = get_token_num('\n'.join(everything))
20
+ everything_token = [get_token_num(e) for e in everything]
21
+ delta = max(everything_token) // 16 # 截断时的颗粒度
22
+
23
+ while n_token > max_token_limit:
24
+ where = np.argmax(everything_token)
25
+ encoded = enc.encode(everything[where], disallowed_special=())
26
+ clipped_encoded = encoded[:len(encoded)-delta]
27
+ everything[where] = enc.decode(clipped_encoded)[:-1] # -1 to remove the may-be illegal char
28
+ everything_token[where] = get_token_num(everything[where])
29
+ n_token = get_token_num('\n'.join(everything))
30
+
31
+ if mode == 'input-and-history':
32
+ inputs = everything[0]
33
+ else:
34
+ pass
35
+ history = everything[1:]
36
+ return inputs, history
37
+
38
+ def request_gpt_model_in_new_thread_with_ui_alive(
39
+ inputs, inputs_show_user, llm_kwargs,
40
+ chatbot, history, sys_prompt, refresh_interval=0.2,
41
+ handle_token_exceed=True,
42
+ retry_times_at_unknown_error=2,
43
+ ):
44
+ """
45
+ Request GPT model,请求GPT模型同时维持用户界面活跃。
46
+
47
+ 输入参数 Args (以_array结尾的输入变量都是列表,列表长度为子任务的数量,执行时,会把列表拆解,放到每个子线程中分别执行):
48
+ inputs (string): List of inputs (输入)
49
+ inputs_show_user (string): List of inputs to show user(展现在报告中的输入,借助此参数,在汇总报告中隐藏啰嗦的真实输入,增强报告的可读性)
50
+ top_p (float): Top p value for sampling from model distribution (GPT参数,浮点数)
51
+ temperature (float): Temperature value for sampling from model distribution(GPT参数,浮点数)
52
+ chatbot: chatbot inputs and outputs (用户界面对话窗口句柄,用于数据流可视化)
53
+ history (list): List of chat history (历史,对话历史列表)
54
+ sys_prompt (string): List of system prompts (系统输入,列表,用于输入给GPT的前提提示,比如你是翻译官怎样怎样)
55
+ refresh_interval (float, optional): Refresh interval for UI (default: 0.2) (刷新时间间隔频率,建议低于1,不可高于3,仅仅服务于视觉效果)
56
+ handle_token_exceed:是否自动处理token溢出的情况,如果选择自动处理,则会在溢出时暴力截断,默认开启
57
+ retry_times_at_unknown_error:失败时的重试次数
58
+
59
+ 输出 Returns:
60
+ future: 输出,GPT返回的结果
61
+ """
62
+ import time
63
+ from concurrent.futures import ThreadPoolExecutor
64
+ from request_llm.bridge_all import predict_no_ui_long_connection
65
+ # 用户反馈
66
+ chatbot.append([inputs_show_user, ""])
67
+ yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
68
+ executor = ThreadPoolExecutor(max_workers=16)
69
+ mutable = ["", time.time(), ""]
70
+ def _req_gpt(inputs, history, sys_prompt):
71
+ retry_op = retry_times_at_unknown_error
72
+ exceeded_cnt = 0
73
+ while True:
74
+ # watchdog error
75
+ if len(mutable) >= 2 and (time.time()-mutable[1]) > 5:
76
+ raise RuntimeError("检测到程序终止。")
77
+ try:
78
+ # 【第一种情况】:顺利完成
79
+ result = predict_no_ui_long_connection(
80
+ inputs=inputs, llm_kwargs=llm_kwargs,
81
+ history=history, sys_prompt=sys_prompt, observe_window=mutable)
82
+ return result
83
+ except ConnectionAbortedError as token_exceeded_error:
84
+ # 【第二种情况】:Token溢出
85
+ if handle_token_exceed:
86
+ exceeded_cnt += 1
87
+ # 【选择处理】 尝试计算比例,尽可能多地保留文本
88
+ from toolbox import get_reduce_token_percent
89
+ p_ratio, n_exceed = get_reduce_token_percent(str(token_exceeded_error))
90
+ MAX_TOKEN = 4096
91
+ EXCEED_ALLO = 512 + 512 * exceeded_cnt
92
+ inputs, history = input_clipping(inputs, history, max_token_limit=MAX_TOKEN-EXCEED_ALLO)
93
+ mutable[0] += f'[Local Message] 警告,文本过长将进行截断,Token溢出数:{n_exceed}。\n\n'
94
+ continue # 返回重试
95
+ else:
96
+ # 【选择放弃】
97
+ tb_str = '```\n' + trimmed_format_exc() + '```'
98
+ mutable[0] += f"[Local Message] 警告,在执行过程中遭遇问题, Traceback:\n\n{tb_str}\n\n"
99
+ return mutable[0] # 放弃
100
+ except:
101
+ # 【第三种情况】:其他错误:重试几次
102
+ tb_str = '```\n' + trimmed_format_exc() + '```'
103
+ print(tb_str)
104
+ mutable[0] += f"[Local Message] 警告,在执行过程中遭遇问题, Traceback:\n\n{tb_str}\n\n"
105
+ if retry_op > 0:
106
+ retry_op -= 1
107
+ mutable[0] += f"[Local Message] 重试中,请稍等 {retry_times_at_unknown_error-retry_op}/{retry_times_at_unknown_error}:\n\n"
108
+ if ("Rate limit reached" in tb_str) or ("Too Many Requests" in tb_str):
109
+ time.sleep(30)
110
+ time.sleep(5)
111
+ continue # 返回重试
112
+ else:
113
+ time.sleep(5)
114
+ return mutable[0] # 放弃
115
+
116
+ # 提交任务
117
+ future = executor.submit(_req_gpt, inputs, history, sys_prompt)
118
+ while True:
119
+ # yield一次以刷新前端页面
120
+ time.sleep(refresh_interval)
121
+ # “喂狗”(看门狗)
122
+ mutable[1] = time.time()
123
+ if future.done():
124
+ break
125
+ chatbot[-1] = [chatbot[-1][0], mutable[0]]
126
+ yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
127
+
128
+ final_result = future.result()
129
+ chatbot[-1] = [chatbot[-1][0], final_result]
130
+ yield from update_ui(chatbot=chatbot, history=[]) # 如果最后成功了,则删除报错信息
131
+ return final_result
132
+
133
+ def can_multi_process(llm):
134
+ if llm.startswith('gpt-'): return True
135
+ if llm.startswith('api2d-'): return True
136
+ if llm.startswith('azure-'): return True
137
+ return False
138
+
139
+ def request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
140
+ inputs_array, inputs_show_user_array, llm_kwargs,
141
+ chatbot, history_array, sys_prompt_array,
142
+ refresh_interval=0.2, max_workers=-1, scroller_max_len=30,
143
+ handle_token_exceed=True, show_user_at_complete=False,
144
+ retry_times_at_unknown_error=2,
145
+ ):
146
+ """
147
+ Request GPT model using multiple threads with UI and high efficiency
148
+ 请求GPT模型的[多线程]版。
149
+ 具备以下功能:
150
+ 实时在UI上反馈远程数据流
151
+ 使用线程池,可调节线程池的大小避免openai的流量限制错误
152
+ 处理中途中止的情况
153
+ 网络等出问题时,会把traceback和已经接收的数据转入输出
154
+
155
+ 输入参数 Args (以_array结尾的输入变量都是列表,列表长度为子任务的数量,执行时,会把列表拆解,放到每个子线程中分别执行):
156
+ inputs_array (list): List of inputs (每个子任务的输入)
157
+ inputs_show_user_array (list): List of inputs to show user(每个子任务展现在报告中的输入,借助此参数,在汇总报告中隐藏啰嗦的真实输入,增强报告的可读性)
158
+ llm_kwargs: llm_kwargs参数
159
+ chatbot: chatbot (用户界面对话窗口句柄,用于数据流可视化)
160
+ history_array (list): List of chat history (历史对话输入,双层列表,第一层列表是子任务分解,第二层列表是对话历史)
161
+ sys_prompt_array (list): List of system prompts (系统输入,列表,用于输入给GPT的前提提示,比如你是翻译官怎样怎样)
162
+ refresh_interval (float, optional): Refresh interval for UI (default: 0.2) (刷新时间间隔频率,建议低于1,不可高于3,仅仅服务于视觉效果)
163
+ max_workers (int, optional): Maximum number of threads (default: see config.py) (最大线程数,如果子任务非常多,需要用此选项防止高频地请求openai导致错误)
164
+ scroller_max_len (int, optional): Maximum length for scroller (default: 30)(数据流的显示最后收到的多少个字符,仅仅服务于视觉效果)
165
+ handle_token_exceed (bool, optional): (是否在输入过长时,自动缩减文本)
166
+ handle_token_exceed:是否自动处理token溢出的情况,如果选择自动处理,则会在溢出时暴力截断,默认开启
167
+ show_user_at_complete (bool, optional): (在结束时,把完整输入-输出结果显示在聊天框)
168
+ retry_times_at_unknown_error:子任务失败时的重试次数
169
+
170
+ 输出 Returns:
171
+ list: List of GPT model responses (每个子任务的输出汇总,如果某个子任务出错,response中会携带traceback报错信息,方便调试和定位问题。)
172
+ """
173
+ import time, random
174
+ from concurrent.futures import ThreadPoolExecutor
175
+ from request_llm.bridge_all import predict_no_ui_long_connection
176
+ assert len(inputs_array) == len(history_array)
177
+ assert len(inputs_array) == len(sys_prompt_array)
178
+ if max_workers == -1: # 读取配置文件
179
+ try: max_workers, = get_conf('DEFAULT_WORKER_NUM')
180
+ except: max_workers = 8
181
+ if max_workers <= 0: max_workers = 3
182
+ # 屏蔽掉 chatglm的多线程,可能会导致严重卡顿
183
+ if not can_multi_process(llm_kwargs['llm_model']):
184
+ max_workers = 1
185
+
186
+ executor = ThreadPoolExecutor(max_workers=max_workers)
187
+ n_frag = len(inputs_array)
188
+ # 用户反馈
189
+ chatbot.append(["请开始多线程操作。", ""])
190
+ yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
191
+ # 跨线程传递
192
+ mutable = [["", time.time(), "等待中"] for _ in range(n_frag)]
193
+
194
+ # 子线程任务
195
+ def _req_gpt(index, inputs, history, sys_prompt):
196
+ gpt_say = ""
197
+ retry_op = retry_times_at_unknown_error
198
+ exceeded_cnt = 0
199
+ mutable[index][2] = "执行中"
200
+ while True:
201
+ # watchdog error
202
+ if len(mutable[index]) >= 2 and (time.time()-mutable[index][1]) > 5:
203
+ raise RuntimeError("检测到程序终止。")
204
+ try:
205
+ # 【第一种情况】:顺利完成
206
+ # time.sleep(10); raise RuntimeError("测试")
207
+ gpt_say = predict_no_ui_long_connection(
208
+ inputs=inputs, llm_kwargs=llm_kwargs, history=history,
209
+ sys_prompt=sys_prompt, observe_window=mutable[index], console_slience=True
210
+ )
211
+ mutable[index][2] = "已成功"
212
+ return gpt_say
213
+ except ConnectionAbortedError as token_exceeded_error:
214
+ # 【第二种情况】:Token溢出,
215
+ if handle_token_exceed:
216
+ exceeded_cnt += 1
217
+ # 【选择处理】 尝试计算比例,尽可能多地保留文本
218
+ from toolbox import get_reduce_token_percent
219
+ p_ratio, n_exceed = get_reduce_token_percent(str(token_exceeded_error))
220
+ MAX_TOKEN = 4096
221
+ EXCEED_ALLO = 512 + 512 * exceeded_cnt
222
+ inputs, history = input_clipping(inputs, history, max_token_limit=MAX_TOKEN-EXCEED_ALLO)
223
+ gpt_say += f'[Local Message] 警告,文本过长将进行截断,Token溢出数:{n_exceed}。\n\n'
224
+ mutable[index][2] = f"截断重试"
225
+ continue # 返回重试
226
+ else:
227
+ # 【选择放弃】
228
+ tb_str = '```\n' + trimmed_format_exc() + '```'
229
+ gpt_say += f"[Local Message] 警告,线程{index}在执行过程中遭遇问题, Traceback:\n\n{tb_str}\n\n"
230
+ if len(mutable[index][0]) > 0: gpt_say += "此线程失败前收到的回答:\n\n" + mutable[index][0]
231
+ mutable[index][2] = "输入过长已放弃"
232
+ return gpt_say # 放弃
233
+ except:
234
+ # 【第三种情况】:其他错误
235
+ tb_str = '```\n' + trimmed_format_exc() + '```'
236
+ print(tb_str)
237
+ gpt_say += f"[Local Message] 警告,线程{index}在执行过程中遭遇问题, Traceback:\n\n{tb_str}\n\n"
238
+ if len(mutable[index][0]) > 0: gpt_say += "此线程失败前收到的回答:\n\n" + mutable[index][0]
239
+ if retry_op > 0:
240
+ retry_op -= 1
241
+ wait = random.randint(5, 20)
242
+ if ("Rate limit reached" in tb_str) or ("Too Many Requests" in tb_str):
243
+ wait = wait * 3
244
+ fail_info = "OpenAI绑定信用卡可解除频率限制 "
245
+ else:
246
+ fail_info = ""
247
+ # 也许等待十几秒后,情况会好转
248
+ for i in range(wait):
249
+ mutable[index][2] = f"{fail_info}等待重试 {wait-i}"; time.sleep(1)
250
+ # 开始重试
251
+ mutable[index][2] = f"重试中 {retry_times_at_unknown_error-retry_op}/{retry_times_at_unknown_error}"
252
+ continue # 返回重试
253
+ else:
254
+ mutable[index][2] = "已失败"
255
+ wait = 5
256
+ time.sleep(5)
257
+ return gpt_say # 放弃
258
+
259
+ # 异步任务开始
260
+ futures = [executor.submit(_req_gpt, index, inputs, history, sys_prompt) for index, inputs, history, sys_prompt in zip(
261
+ range(len(inputs_array)), inputs_array, history_array, sys_prompt_array)]
262
+ cnt = 0
263
+ while True:
264
+ # yield一次以刷新前端页面
265
+ time.sleep(refresh_interval)
266
+ cnt += 1
267
+ worker_done = [h.done() for h in futures]
268
+ # 更好的UI视觉效果
269
+ observe_win = []
270
+ # 每个线程都要“喂狗”(看门狗)
271
+ for thread_index, _ in enumerate(worker_done):
272
+ mutable[thread_index][1] = time.time()
273
+ # 在前端打印些好玩的东西
274
+ for thread_index, _ in enumerate(worker_done):
275
+ print_something_really_funny = "[ ...`"+mutable[thread_index][0][-scroller_max_len:].\
276
+ replace('\n', '').replace('```', '...').replace(
277
+ ' ', '.').replace('<br/>', '.....').replace('$', '.')+"`... ]"
278
+ observe_win.append(print_something_really_funny)
279
+ # 在前端打印些好玩的东西
280
+ stat_str = ''.join([f'`{mutable[thread_index][2]}`: {obs}\n\n'
281
+ if not done else f'`{mutable[thread_index][2]}`\n\n'
282
+ for thread_index, done, obs in zip(range(len(worker_done)), worker_done, observe_win)])
283
+ # 在前端打印些好玩的东西
284
+ chatbot[-1] = [chatbot[-1][0], f'多线程操作已经开始,完成情况: \n\n{stat_str}' + ''.join(['.']*(cnt % 10+1))]
285
+ yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
286
+ if all(worker_done):
287
+ executor.shutdown()
288
+ break
289
+
290
+ # 异步任务结束
291
+ gpt_response_collection = []
292
+ for inputs_show_user, f in zip(inputs_show_user_array, futures):
293
+ gpt_res = f.result()
294
+ gpt_response_collection.extend([inputs_show_user, gpt_res])
295
+
296
+ # 是否在结束时,在界面上显示结果
297
+ if show_user_at_complete:
298
+ for inputs_show_user, f in zip(inputs_show_user_array, futures):
299
+ gpt_res = f.result()
300
+ chatbot.append([inputs_show_user, gpt_res])
301
+ yield from update_ui(chatbot=chatbot, history=[]) # 刷新界面
302
+ time.sleep(0.3)
303
+ return gpt_response_collection
304
+
305
+
306
+ def breakdown_txt_to_satisfy_token_limit(txt, get_token_fn, limit):
307
+ def cut(txt_tocut, must_break_at_empty_line): # 递归
308
+ if get_token_fn(txt_tocut) <= limit:
309
+ return [txt_tocut]
310
+ else:
311
+ lines = txt_tocut.split('\n')
312
+ estimated_line_cut = limit / get_token_fn(txt_tocut) * len(lines)
313
+ estimated_line_cut = int(estimated_line_cut)
314
+ for cnt in reversed(range(estimated_line_cut)):
315
+ if must_break_at_empty_line:
316
+ if lines[cnt] != "":
317
+ continue
318
+ print(cnt)
319
+ prev = "\n".join(lines[:cnt])
320
+ post = "\n".join(lines[cnt:])
321
+ if get_token_fn(prev) < limit:
322
+ break
323
+ if cnt == 0:
324
+ raise RuntimeError("存在一行极长的文本!")
325
+ # print(len(post))
326
+ # 列表递归接龙
327
+ result = [prev]
328
+ result.extend(cut(post, must_break_at_empty_line))
329
+ return result
330
+ try:
331
+ return cut(txt, must_break_at_empty_line=True)
332
+ except RuntimeError:
333
+ return cut(txt, must_break_at_empty_line=False)
334
+
335
+
336
+ def force_breakdown(txt, limit, get_token_fn):
337
+ """
338
+ 当无法用标点、空行分割时,我们用最暴力的方法切割
339
+ """
340
+ for i in reversed(range(len(txt))):
341
+ if get_token_fn(txt[:i]) < limit:
342
+ return txt[:i], txt[i:]
343
+ return "Tiktoken未知错误", "Tiktoken未知错误"
344
+
345
+ def breakdown_txt_to_satisfy_token_limit_for_pdf(txt, get_token_fn, limit):
346
+ # 递归
347
+ def cut(txt_tocut, must_break_at_empty_line, break_anyway=False):
348
+ if get_token_fn(txt_tocut) <= limit:
349
+ return [txt_tocut]
350
+ else:
351
+ lines = txt_tocut.split('\n')
352
+ estimated_line_cut = limit / get_token_fn(txt_tocut) * len(lines)
353
+ estimated_line_cut = int(estimated_line_cut)
354
+ cnt = 0
355
+ for cnt in reversed(range(estimated_line_cut)):
356
+ if must_break_at_empty_line:
357
+ if lines[cnt] != "":
358
+ continue
359
+ prev = "\n".join(lines[:cnt])
360
+ post = "\n".join(lines[cnt:])
361
+ if get_token_fn(prev) < limit:
362
+ break
363
+ if cnt == 0:
364
+ if break_anyway:
365
+ prev, post = force_breakdown(txt_tocut, limit, get_token_fn)
366
+ else:
367
+ raise RuntimeError(f"存在一行极长的文本!{txt_tocut}")
368
+ # print(len(post))
369
+ # 列表递归接龙
370
+ result = [prev]
371
+ result.extend(cut(post, must_break_at_empty_line, break_anyway=break_anyway))
372
+ return result
373
+ try:
374
+ # 第1次尝试,将双空行(\n\n)作为切分点
375
+ return cut(txt, must_break_at_empty_line=True)
376
+ except RuntimeError:
377
+ try:
378
+ # 第2次尝试,将单空行(\n)作为切分点
379
+ return cut(txt, must_break_at_empty_line=False)
380
+ except RuntimeError:
381
+ try:
382
+ # 第3次尝试,将英文句号(.)作为切分点
383
+ res = cut(txt.replace('.', '。\n'), must_break_at_empty_line=False) # 这个中文的句号是故意的,作为一个标识而存在
384
+ return [r.replace('。\n', '.') for r in res]
385
+ except RuntimeError as e:
386
+ try:
387
+ # 第4次尝试,将中文句号(。)作为切分点
388
+ res = cut(txt.replace('。', '。。\n'), must_break_at_empty_line=False)
389
+ return [r.replace('。。\n', '。') for r in res]
390
+ except RuntimeError as e:
391
+ # 第5次尝试,没办法了,随便切一下敷衍吧
392
+ return cut(txt, must_break_at_empty_line=False, break_anyway=True)
393
+
394
+
395
+
396
+ def read_and_clean_pdf_text(fp):
397
+ """
398
+ 这个函数用于分割pdf,用了很多trick,逻辑较乱,效果奇好
399
+
400
+ **输入参数说明**
401
+ - `fp`:需要读取和清理文本的pdf文件路径
402
+
403
+ **输出参数说明**
404
+ - `meta_txt`:清理后的文本内容字符串
405
+ - `page_one_meta`:第一页清理后的文本内容列表
406
+
407
+ **函数功能**
408
+ 读取pdf文件并清理其中的文本内容,清理规则包括:
409
+ - 提取所有块元的文本信息,并合并为一个字符串
410
+ - 去除短块(字符数小于100)并替换为回车符
411
+ - 清理多余的空行
412
+ - 合并小写字母开头的段落块并替换为空格
413
+ - 清除重复的换行
414
+ - 将每个换行符替换为两个换行符,使每个段落之间有两个换行符分隔
415
+ """
416
+ import fitz, copy
417
+ import re
418
+ import numpy as np
419
+ from colorful import print亮黄, print亮绿
420
+ fc = 0 # Index 0 文本
421
+ fs = 1 # Index 1 字体
422
+ fb = 2 # Index 2 框框
423
+ REMOVE_FOOT_NOTE = True # 是否丢弃掉 不是正文的内容 (比正文字体小,如参考文献、脚注、图注等)
424
+ REMOVE_FOOT_FFSIZE_PERCENT = 0.95 # 小于正文的?时,判定为不是正文(有些文章的正文部分字体大小不是100%统一的,有肉眼不可见的小变化)
425
+ def primary_ffsize(l):
426
+ """
427
+ 提取文本块主字体
428
+ """
429
+ fsize_statiscs = {}
430
+ for wtf in l['spans']:
431
+ if wtf['size'] not in fsize_statiscs: fsize_statiscs[wtf['size']] = 0
432
+ fsize_statiscs[wtf['size']] += len(wtf['text'])
433
+ return max(fsize_statiscs, key=fsize_statiscs.get)
434
+
435
+ def ffsize_same(a,b):
436
+ """
437
+ 提取字体大小是否近似相等
438
+ """
439
+ return abs((a-b)/max(a,b)) < 0.02
440
+
441
+ with fitz.open(fp) as doc:
442
+ meta_txt = []
443
+ meta_font = []
444
+
445
+ meta_line = []
446
+ meta_span = []
447
+ ############################## <第 1 步,搜集初始信息> ##################################
448
+ for index, page in enumerate(doc):
449
+ # file_content += page.get_text()
450
+ text_areas = page.get_text("dict") # 获取页面上的文本信息
451
+ for t in text_areas['blocks']:
452
+ if 'lines' in t:
453
+ pf = 998
454
+ for l in t['lines']:
455
+ txt_line = "".join([wtf['text'] for wtf in l['spans']])
456
+ if len(txt_line) == 0: continue
457
+ pf = primary_ffsize(l)
458
+ meta_line.append([txt_line, pf, l['bbox'], l])
459
+ for wtf in l['spans']: # for l in t['lines']:
460
+ meta_span.append([wtf['text'], wtf['size'], len(wtf['text'])])
461
+ # meta_line.append(["NEW_BLOCK", pf])
462
+ # 块元提取 for each word segment with in line for each line cross-line words for each block
463
+ meta_txt.extend([" ".join(["".join([wtf['text'] for wtf in l['spans']]) for l in t['lines']]).replace(
464
+ '- ', '') for t in text_areas['blocks'] if 'lines' in t])
465
+ meta_font.extend([np.mean([np.mean([wtf['size'] for wtf in l['spans']])
466
+ for l in t['lines']]) for t in text_areas['blocks'] if 'lines' in t])
467
+ if index == 0:
468
+ page_one_meta = [" ".join(["".join([wtf['text'] for wtf in l['spans']]) for l in t['lines']]).replace(
469
+ '- ', '') for t in text_areas['blocks'] if 'lines' in t]
470
+
471
+ ############################## <第 2 步,获取正文主字体> ##################################
472
+ fsize_statiscs = {}
473
+ for span in meta_span:
474
+ if span[1] not in fsize_statiscs: fsize_statiscs[span[1]] = 0
475
+ fsize_statiscs[span[1]] += span[2]
476
+ main_fsize = max(fsize_statiscs, key=fsize_statiscs.get)
477
+ if REMOVE_FOOT_NOTE:
478
+ give_up_fize_threshold = main_fsize * REMOVE_FOOT_FFSIZE_PERCENT
479
+
480
+ ############################## <第 3 步,切分和重新整合> ##################################
481
+ mega_sec = []
482
+ sec = []
483
+ for index, line in enumerate(meta_line):
484
+ if index == 0:
485
+ sec.append(line[fc])
486
+ continue
487
+ if REMOVE_FOOT_NOTE:
488
+ if meta_line[index][fs] <= give_up_fize_threshold:
489
+ continue
490
+ if ffsize_same(meta_line[index][fs], meta_line[index-1][fs]):
491
+ # 尝试识别段落
492
+ if meta_line[index][fc].endswith('.') and\
493
+ (meta_line[index-1][fc] != 'NEW_BLOCK') and \
494
+ (meta_line[index][fb][2] - meta_line[index][fb][0]) < (meta_line[index-1][fb][2] - meta_line[index-1][fb][0]) * 0.7:
495
+ sec[-1] += line[fc]
496
+ sec[-1] += "\n\n"
497
+ else:
498
+ sec[-1] += " "
499
+ sec[-1] += line[fc]
500
+ else:
501
+ if (index+1 < len(meta_line)) and \
502
+ meta_line[index][fs] > main_fsize:
503
+ # 单行 + 字体大
504
+ mega_sec.append(copy.deepcopy(sec))
505
+ sec = []
506
+ sec.append("# " + line[fc])
507
+ else:
508
+ # 尝试识别section
509
+ if meta_line[index-1][fs] > meta_line[index][fs]:
510
+ sec.append("\n" + line[fc])
511
+ else:
512
+ sec.append(line[fc])
513
+ mega_sec.append(copy.deepcopy(sec))
514
+
515
+ finals = []
516
+ for ms in mega_sec:
517
+ final = " ".join(ms)
518
+ final = final.replace('- ', ' ')
519
+ finals.append(final)
520
+ meta_txt = finals
521
+
522
+ ############################## <第 4 步,乱七八糟的后处理> ##################################
523
+ def 把字符太少的块清除为回车(meta_txt):
524
+ for index, block_txt in enumerate(meta_txt):
525
+ if len(block_txt) < 100:
526
+ meta_txt[index] = '\n'
527
+ return meta_txt
528
+ meta_txt = 把字符太少的块清除为回车(meta_txt)
529
+
530
+ def 清理多余的空行(meta_txt):
531
+ for index in reversed(range(1, len(meta_txt))):
532
+ if meta_txt[index] == '\n' and meta_txt[index-1] == '\n':
533
+ meta_txt.pop(index)
534
+ return meta_txt
535
+ meta_txt = 清理多余的空行(meta_txt)
536
+
537
+ def 合并小写开头的段落块(meta_txt):
538
+ def starts_with_lowercase_word(s):
539
+ pattern = r"^[a-z]+"
540
+ match = re.match(pattern, s)
541
+ if match:
542
+ return True
543
+ else:
544
+ return False
545
+ for _ in range(100):
546
+ for index, block_txt in enumerate(meta_txt):
547
+ if starts_with_lowercase_word(block_txt):
548
+ if meta_txt[index-1] != '\n':
549
+ meta_txt[index-1] += ' '
550
+ else:
551
+ meta_txt[index-1] = ''
552
+ meta_txt[index-1] += meta_txt[index]
553
+ meta_txt[index] = '\n'
554
+ return meta_txt
555
+ meta_txt = 合并小写开头的段落块(meta_txt)
556
+ meta_txt = 清理多余的空行(meta_txt)
557
+
558
+ meta_txt = '\n'.join(meta_txt)
559
+ # 清除重复的换行
560
+ for _ in range(5):
561
+ meta_txt = meta_txt.replace('\n\n', '\n')
562
+
563
+ # 换行 -> 双换行
564
+ meta_txt = meta_txt.replace('\n', '\n\n')
565
+
566
+ ############################## <第 5 步,展示分割效果> ##################################
567
+ # for f in finals:
568
+ # print亮黄(f)
569
+ # print亮绿('***************************')
570
+
571
+ return meta_txt, page_one_meta
572
+
573
+
574
+ def get_files_from_everything(txt, type): # type='.md'
575
+ """
576
+ 这个函数是用来获取指定目录下所有指定类型(如.md)的文件,并且对于网络上的文件,也可以获取它。
577
+ 下面是对每个参数和返回值的说明:
578
+ 参数
579
+ - txt: 路径或网址,表示要搜索的文件或者文件夹路径或网络上的文件。
580
+ - type: 字符串,表示要搜索的文件类型。默认是.md。
581
+ 返回值
582
+ - success: 布尔值,表示函数是否成功执行。
583
+ - file_manifest: 文件路径列表,里面包含以指定类型为后缀名的所有文件的绝对路径。
584
+ - project_folder: 字符串,表示文件所在的文件夹路径。如果是网络上的文件,就是临时文件夹的路径。
585
+ 该函数详细注释已添加,请确认是否满足您的需要。
586
+ """
587
+ import glob, os
588
+
589
+ success = True
590
+ if txt.startswith('http'):
591
+ # 网络的远程文件
592
+ import requests
593
+ from toolbox import get_conf
594
+ proxies, = get_conf('proxies')
595
+ r = requests.get(txt, proxies=proxies)
596
+ with open('./gpt_log/temp'+type, 'wb+') as f: f.write(r.content)
597
+ project_folder = './gpt_log/'
598
+ file_manifest = ['./gpt_log/temp'+type]
599
+ elif txt.endswith(type):
600
+ # 直接给定文件
601
+ file_manifest = [txt]
602
+ project_folder = os.path.dirname(txt)
603
+ elif os.path.exists(txt):
604
+ # 本地路径,递归搜索
605
+ project_folder = txt
606
+ file_manifest = [f for f in glob.glob(f'{project_folder}/**/*'+type, recursive=True)]
607
+ if len(file_manifest) == 0:
608
+ success = False
609
+ else:
610
+ project_folder = None
611
+ file_manifest = []
612
+ success = False
613
+
614
+ return success, file_manifest, project_folder
615
+
616
+
617
+
618
+
619
+ def Singleton(cls):
620
+ _instance = {}
621
+
622
+ def _singleton(*args, **kargs):
623
+ if cls not in _instance:
624
+ _instance[cls] = cls(*args, **kargs)
625
+ return _instance[cls]
626
+
627
+ return _singleton
628
+
629
+
630
+ @Singleton
631
+ class knowledge_archive_interface():
632
+ def __init__(self) -> None:
633
+ self.threadLock = threading.Lock()
634
+ self.current_id = ""
635
+ self.kai_path = None
636
+ self.qa_handle = None
637
+ self.text2vec_large_chinese = None
638
+
639
+ def get_chinese_text2vec(self):
640
+ if self.text2vec_large_chinese is None:
641
+ # < -------------------预热文本向量化模组--------------- >
642
+ from toolbox import ProxyNetworkActivate
643
+ print('Checking Text2vec ...')
644
+ from langchain.embeddings.huggingface import HuggingFaceEmbeddings
645
+ with ProxyNetworkActivate(): # 临时地激活代理网络
646
+ self.text2vec_large_chinese = HuggingFaceEmbeddings(model_name="GanymedeNil/text2vec-large-chinese")
647
+
648
+ return self.text2vec_large_chinese
649
+
650
+
651
+ def feed_archive(self, file_manifest, id="default"):
652
+ self.threadLock.acquire()
653
+ # import uuid
654
+ self.current_id = id
655
+ from zh_langchain import construct_vector_store
656
+ self.qa_handle, self.kai_path = construct_vector_store(
657
+ vs_id=self.current_id,
658
+ files=file_manifest,
659
+ sentence_size=100,
660
+ history=[],
661
+ one_conent="",
662
+ one_content_segmentation="",
663
+ text2vec = self.get_chinese_text2vec(),
664
+ )
665
+ self.threadLock.release()
666
+
667
+ def get_current_archive_id(self):
668
+ return self.current_id
669
+
670
+ def get_loaded_file(self):
671
+ return self.qa_handle.get_loaded_file()
672
+
673
+ def answer_with_archive_by_id(self, txt, id):
674
+ self.threadLock.acquire()
675
+ if not self.current_id == id:
676
+ self.current_id = id
677
+ from zh_langchain import construct_vector_store
678
+ self.qa_handle, self.kai_path = construct_vector_store(
679
+ vs_id=self.current_id,
680
+ files=[],
681
+ sentence_size=100,
682
+ history=[],
683
+ one_conent="",
684
+ one_content_segmentation="",
685
+ text2vec = self.get_chinese_text2vec(),
686
+ )
687
+ VECTOR_SEARCH_SCORE_THRESHOLD = 0
688
+ VECTOR_SEARCH_TOP_K = 4
689
+ CHUNK_SIZE = 512
690
+ resp, prompt = self.qa_handle.get_knowledge_based_conent_test(
691
+ query = txt,
692
+ vs_path = self.kai_path,
693
+ score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD,
694
+ vector_search_top_k=VECTOR_SEARCH_TOP_K,
695
+ chunk_conent=True,
696
+ chunk_size=CHUNK_SIZE,
697
+ text2vec = self.get_chinese_text2vec(),
698
+ )
699
+ self.threadLock.release()
700
+ return resp, prompt
701
+
702
+ def try_install_deps(deps):
703
+ for dep in deps:
704
+ import subprocess, sys
705
+ subprocess.check_call([sys.executable, '-m', 'pip', 'install', '--user', dep])
706
+
707
+
708
+ class construct_html():
709
+ def __init__(self) -> None:
710
+ self.css = """
711
+ .row {
712
+ display: flex;
713
+ flex-wrap: wrap;
714
+ }
715
+
716
+ .column {
717
+ flex: 1;
718
+ padding: 10px;
719
+ }
720
+
721
+ .table-header {
722
+ font-weight: bold;
723
+ border-bottom: 1px solid black;
724
+ }
725
+
726
+ .table-row {
727
+ border-bottom: 1px solid lightgray;
728
+ }
729
+
730
+ .table-cell {
731
+ padding: 5px;
732
+ }
733
+ """
734
+ self.html_string = f'<!DOCTYPE html><head><meta charset="utf-8"><title>翻译结果</title><style>{self.css}</style></head>'
735
+
736
+
737
+ def add_row(self, a, b):
738
+ tmp = """
739
+ <div class="row table-row">
740
+ <div class="column table-cell">REPLACE_A</div>
741
+ <div class="column table-cell">REPLACE_B</div>
742
+ </div>
743
+ """
744
+ from toolbox import markdown_convertion
745
+ tmp = tmp.replace('REPLACE_A', markdown_convertion(a))
746
+ tmp = tmp.replace('REPLACE_B', markdown_convertion(b))
747
+ self.html_string += tmp
748
+
749
+
750
+ def save_file(self, file_name):
751
+ with open(f'./gpt_log/{file_name}', 'w', encoding='utf8') as f:
752
+ f.write(self.html_string.encode('utf-8', 'ignore').decode())
753
+
crazy_functions/latex_fns/latex_actions.py ADDED
@@ -0,0 +1,447 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import update_ui, update_ui_lastest_msg # 刷新Gradio前端界面
2
+ from toolbox import zip_folder, objdump, objload, promote_file_to_downloadzone
3
+ from .latex_toolbox import PRESERVE, TRANSFORM
4
+ from .latex_toolbox import set_forbidden_text, set_forbidden_text_begin_end, set_forbidden_text_careful_brace
5
+ from .latex_toolbox import reverse_forbidden_text_careful_brace, reverse_forbidden_text, convert_to_linklist, post_process
6
+ from .latex_toolbox import fix_content, find_main_tex_file, merge_tex_files, compile_latex_with_timeout
7
+
8
+ import os, shutil
9
+ import re
10
+ import numpy as np
11
+
12
+ pj = os.path.join
13
+
14
+
15
+ def split_subprocess(txt, project_folder, return_dict, opts):
16
+ """
17
+ break down latex file to a linked list,
18
+ each node use a preserve flag to indicate whether it should
19
+ be proccessed by GPT.
20
+ """
21
+ text = txt
22
+ mask = np.zeros(len(txt), dtype=np.uint8) + TRANSFORM
23
+
24
+ # 吸收title与作者以上的部分
25
+ text, mask = set_forbidden_text(text, mask, r"^(.*?)\\maketitle", re.DOTALL)
26
+ text, mask = set_forbidden_text(text, mask, r"^(.*?)\\begin{document}", re.DOTALL)
27
+ # 吸收iffalse注释
28
+ text, mask = set_forbidden_text(text, mask, r"\\iffalse(.*?)\\fi", re.DOTALL)
29
+ # 吸收在42行以内的begin-end组合
30
+ text, mask = set_forbidden_text_begin_end(text, mask, r"\\begin\{([a-z\*]*)\}(.*?)\\end\{\1\}", re.DOTALL, limit_n_lines=42)
31
+ # 吸收匿名公式
32
+ text, mask = set_forbidden_text(text, mask, [ r"\$\$([^$]+)\$\$", r"\\\[.*?\\\]" ], re.DOTALL)
33
+ # 吸收其他杂项
34
+ text, mask = set_forbidden_text(text, mask, [ r"\\section\{(.*?)\}", r"\\section\*\{(.*?)\}", r"\\subsection\{(.*?)\}", r"\\subsubsection\{(.*?)\}" ])
35
+ text, mask = set_forbidden_text(text, mask, [ r"\\bibliography\{(.*?)\}", r"\\bibliographystyle\{(.*?)\}" ])
36
+ text, mask = set_forbidden_text(text, mask, r"\\begin\{thebibliography\}.*?\\end\{thebibliography\}", re.DOTALL)
37
+ text, mask = set_forbidden_text(text, mask, r"\\begin\{lstlisting\}(.*?)\\end\{lstlisting\}", re.DOTALL)
38
+ text, mask = set_forbidden_text(text, mask, r"\\begin\{wraptable\}(.*?)\\end\{wraptable\}", re.DOTALL)
39
+ text, mask = set_forbidden_text(text, mask, r"\\begin\{algorithm\}(.*?)\\end\{algorithm\}", re.DOTALL)
40
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{wrapfigure\}(.*?)\\end\{wrapfigure\}", r"\\begin\{wrapfigure\*\}(.*?)\\end\{wrapfigure\*\}"], re.DOTALL)
41
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{figure\}(.*?)\\end\{figure\}", r"\\begin\{figure\*\}(.*?)\\end\{figure\*\}"], re.DOTALL)
42
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{multline\}(.*?)\\end\{multline\}", r"\\begin\{multline\*\}(.*?)\\end\{multline\*\}"], re.DOTALL)
43
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{table\}(.*?)\\end\{table\}", r"\\begin\{table\*\}(.*?)\\end\{table\*\}"], re.DOTALL)
44
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{minipage\}(.*?)\\end\{minipage\}", r"\\begin\{minipage\*\}(.*?)\\end\{minipage\*\}"], re.DOTALL)
45
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{align\*\}(.*?)\\end\{align\*\}", r"\\begin\{align\}(.*?)\\end\{align\}"], re.DOTALL)
46
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{equation\}(.*?)\\end\{equation\}", r"\\begin\{equation\*\}(.*?)\\end\{equation\*\}"], re.DOTALL)
47
+ text, mask = set_forbidden_text(text, mask, [r"\\includepdf\[(.*?)\]\{(.*?)\}", r"\\clearpage", r"\\newpage", r"\\appendix", r"\\tableofcontents", r"\\include\{(.*?)\}"])
48
+ text, mask = set_forbidden_text(text, mask, [r"\\vspace\{(.*?)\}", r"\\hspace\{(.*?)\}", r"\\label\{(.*?)\}", r"\\begin\{(.*?)\}", r"\\end\{(.*?)\}", r"\\item "])
49
+ text, mask = set_forbidden_text_careful_brace(text, mask, r"\\hl\{(.*?)\}", re.DOTALL)
50
+ # reverse 操作必须放在最后
51
+ text, mask = reverse_forbidden_text_careful_brace(text, mask, r"\\caption\{(.*?)\}", re.DOTALL, forbid_wrapper=True)
52
+ text, mask = reverse_forbidden_text_careful_brace(text, mask, r"\\abstract\{(.*?)\}", re.DOTALL, forbid_wrapper=True)
53
+ text, mask = reverse_forbidden_text(text, mask, r"\\begin\{abstract\}(.*?)\\end\{abstract\}", re.DOTALL, forbid_wrapper=True)
54
+ root = convert_to_linklist(text, mask)
55
+
56
+ # 最后一步处理,增强稳健性
57
+ root = post_process(root)
58
+
59
+ # 输出html调试文件,用红色标注处保留区(PRESERVE),用黑色标注转换区(TRANSFORM)
60
+ with open(pj(project_folder, 'debug_log.html'), 'w', encoding='utf8') as f:
61
+ segment_parts_for_gpt = []
62
+ nodes = []
63
+ node = root
64
+ while True:
65
+ nodes.append(node)
66
+ show_html = node.string.replace('\n','<br/>')
67
+ if not node.preserve:
68
+ segment_parts_for_gpt.append(node.string)
69
+ f.write(f'<p style="color:black;">#{node.range}{show_html}#</p>')
70
+ else:
71
+ f.write(f'<p style="color:red;">{show_html}</p>')
72
+ node = node.next
73
+ if node is None: break
74
+
75
+ for n in nodes: n.next = None # break
76
+ return_dict['nodes'] = nodes
77
+ return_dict['segment_parts_for_gpt'] = segment_parts_for_gpt
78
+ return return_dict
79
+
80
+ class LatexPaperSplit():
81
+ """
82
+ break down latex file to a linked list,
83
+ each node use a preserve flag to indicate whether it should
84
+ be proccessed by GPT.
85
+ """
86
+ def __init__(self) -> None:
87
+ self.nodes = None
88
+ self.msg = "*{\\scriptsize\\textbf{警告:该PDF由GPT-Academic开源项目调用大语言模型+Latex翻译插件一键生成," + \
89
+ "版权归原文作者所有。翻译内容可靠性无保障,请仔细鉴别并以原文为准。" + \
90
+ "项目Github地址 \\url{https://github.com/binary-husky/gpt_academic/}。"
91
+ # 请您不要删除或修改这行警告,除非您是论文的原作者(如果您是论文原作者,欢迎加REAME中的QQ联系开发者)
92
+ self.msg_declare = "为了防止大语言模型的意外谬误产生扩散影响,禁止移除或修改此警告。}}\\\\"
93
+
94
+
95
+ def merge_result(self, arr, mode, msg, buggy_lines=[], buggy_line_surgery_n_lines=10):
96
+ """
97
+ Merge the result after the GPT process completed
98
+ """
99
+ result_string = ""
100
+ node_cnt = 0
101
+ line_cnt = 0
102
+
103
+ for node in self.nodes:
104
+ if node.preserve:
105
+ line_cnt += node.string.count('\n')
106
+ result_string += node.string
107
+ else:
108
+ translated_txt = fix_content(arr[node_cnt], node.string)
109
+ begin_line = line_cnt
110
+ end_line = line_cnt + translated_txt.count('\n')
111
+
112
+ # reverse translation if any error
113
+ if any([begin_line-buggy_line_surgery_n_lines <= b_line <= end_line+buggy_line_surgery_n_lines for b_line in buggy_lines]):
114
+ translated_txt = node.string
115
+
116
+ result_string += translated_txt
117
+ node_cnt += 1
118
+ line_cnt += translated_txt.count('\n')
119
+
120
+ if mode == 'translate_zh':
121
+ pattern = re.compile(r'\\begin\{abstract\}.*\n')
122
+ match = pattern.search(result_string)
123
+ if not match:
124
+ # match \abstract{xxxx}
125
+ pattern_compile = re.compile(r"\\abstract\{(.*?)\}", flags=re.DOTALL)
126
+ match = pattern_compile.search(result_string)
127
+ position = match.regs[1][0]
128
+ else:
129
+ # match \begin{abstract}xxxx\end{abstract}
130
+ position = match.end()
131
+ result_string = result_string[:position] + self.msg + msg + self.msg_declare + result_string[position:]
132
+ return result_string
133
+
134
+
135
+ def split(self, txt, project_folder, opts):
136
+ """
137
+ break down latex file to a linked list,
138
+ each node use a preserve flag to indicate whether it should
139
+ be proccessed by GPT.
140
+ P.S. use multiprocessing to avoid timeout error
141
+ """
142
+ import multiprocessing
143
+ manager = multiprocessing.Manager()
144
+ return_dict = manager.dict()
145
+ p = multiprocessing.Process(
146
+ target=split_subprocess,
147
+ args=(txt, project_folder, return_dict, opts))
148
+ p.start()
149
+ p.join()
150
+ p.close()
151
+ self.nodes = return_dict['nodes']
152
+ self.sp = return_dict['segment_parts_for_gpt']
153
+ return self.sp
154
+
155
+
156
+ class LatexPaperFileGroup():
157
+ """
158
+ use tokenizer to break down text according to max_token_limit
159
+ """
160
+ def __init__(self):
161
+ self.file_paths = []
162
+ self.file_contents = []
163
+ self.sp_file_contents = []
164
+ self.sp_file_index = []
165
+ self.sp_file_tag = []
166
+
167
+ # count_token
168
+ from request_llm.bridge_all import model_info
169
+ enc = model_info["gpt-3.5-turbo"]['tokenizer']
170
+ def get_token_num(txt): return len(enc.encode(txt, disallowed_special=()))
171
+ self.get_token_num = get_token_num
172
+
173
+ def run_file_split(self, max_token_limit=1900):
174
+ """
175
+ use tokenizer to break down text according to max_token_limit
176
+ """
177
+ for index, file_content in enumerate(self.file_contents):
178
+ if self.get_token_num(file_content) < max_token_limit:
179
+ self.sp_file_contents.append(file_content)
180
+ self.sp_file_index.append(index)
181
+ self.sp_file_tag.append(self.file_paths[index])
182
+ else:
183
+ from ..crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
184
+ segments = breakdown_txt_to_satisfy_token_limit_for_pdf(file_content, self.get_token_num, max_token_limit)
185
+ for j, segment in enumerate(segments):
186
+ self.sp_file_contents.append(segment)
187
+ self.sp_file_index.append(index)
188
+ self.sp_file_tag.append(self.file_paths[index] + f".part-{j}.tex")
189
+ print('Segmentation: done')
190
+
191
+ def merge_result(self):
192
+ self.file_result = ["" for _ in range(len(self.file_paths))]
193
+ for r, k in zip(self.sp_file_result, self.sp_file_index):
194
+ self.file_result[k] += r
195
+
196
+ def write_result(self):
197
+ manifest = []
198
+ for path, res in zip(self.file_paths, self.file_result):
199
+ with open(path + '.polish.tex', 'w', encoding='utf8') as f:
200
+ manifest.append(path + '.polish.tex')
201
+ f.write(res)
202
+ return manifest
203
+
204
+
205
+ def Latex精细分解与转化(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, mode='proofread', switch_prompt=None, opts=[]):
206
+ import time, os, re
207
+ from ..crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
208
+ from .latex_actions import LatexPaperFileGroup, LatexPaperSplit
209
+
210
+ # <-------- 寻找主tex文件 ---------->
211
+ maintex = find_main_tex_file(file_manifest, mode)
212
+ chatbot.append((f"定位主Latex文件", f'[Local Message] 分析结果:该项目的Latex主文件是{maintex}, 如果分析错误, 请立即终止程序, 删除或修改歧义文件, 然后重试。主程序即将开始, 请稍候。'))
213
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
214
+ time.sleep(3)
215
+
216
+ # <-------- 读取Latex文件, 将多文件tex工程融合为一个巨型tex ---------->
217
+ main_tex_basename = os.path.basename(maintex)
218
+ assert main_tex_basename.endswith('.tex')
219
+ main_tex_basename_bare = main_tex_basename[:-4]
220
+ may_exist_bbl = pj(project_folder, f'{main_tex_basename_bare}.bbl')
221
+ if os.path.exists(may_exist_bbl):
222
+ shutil.copyfile(may_exist_bbl, pj(project_folder, f'merge.bbl'))
223
+ shutil.copyfile(may_exist_bbl, pj(project_folder, f'merge_{mode}.bbl'))
224
+ shutil.copyfile(may_exist_bbl, pj(project_folder, f'merge_diff.bbl'))
225
+
226
+ with open(maintex, 'r', encoding='utf-8', errors='replace') as f:
227
+ content = f.read()
228
+ merged_content = merge_tex_files(project_folder, content, mode)
229
+
230
+ with open(project_folder + '/merge.tex', 'w', encoding='utf-8', errors='replace') as f:
231
+ f.write(merged_content)
232
+
233
+ # <-------- 精细切分latex文件 ---------->
234
+ chatbot.append((f"Latex文件融合完成", f'[Local Message] 正在精细切分latex文件,这需要一段时间计算,文档越长耗时越长,请耐心等待。'))
235
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
236
+ lps = LatexPaperSplit()
237
+ res = lps.split(merged_content, project_folder, opts) # 消耗时间的函数
238
+
239
+ # <-------- 拆分过长的latex片段 ---------->
240
+ pfg = LatexPaperFileGroup()
241
+ for index, r in enumerate(res):
242
+ pfg.file_paths.append('segment-' + str(index))
243
+ pfg.file_contents.append(r)
244
+
245
+ pfg.run_file_split(max_token_limit=1024)
246
+ n_split = len(pfg.sp_file_contents)
247
+
248
+ # <-------- 根据需要切换prompt ---------->
249
+ inputs_array, sys_prompt_array = switch_prompt(pfg, mode)
250
+ inputs_show_user_array = [f"{mode} {f}" for f in pfg.sp_file_tag]
251
+
252
+ if os.path.exists(pj(project_folder,'temp.pkl')):
253
+
254
+ # <-------- 【仅调试】如果存在调试缓存文件,则跳过GPT请求环节 ---------->
255
+ pfg = objload(file=pj(project_folder,'temp.pkl'))
256
+
257
+ else:
258
+ # <-------- gpt 多线程请求 ---------->
259
+ gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
260
+ inputs_array=inputs_array,
261
+ inputs_show_user_array=inputs_show_user_array,
262
+ llm_kwargs=llm_kwargs,
263
+ chatbot=chatbot,
264
+ history_array=[[""] for _ in range(n_split)],
265
+ sys_prompt_array=sys_prompt_array,
266
+ # max_workers=5, # 并行任务数量限制, 最多同时执行5个, 其他的排队等待
267
+ scroller_max_len = 40
268
+ )
269
+
270
+ # <-------- 文本碎片重组为完整的tex片段 ---------->
271
+ pfg.sp_file_result = []
272
+ for i_say, gpt_say, orig_content in zip(gpt_response_collection[0::2], gpt_response_collection[1::2], pfg.sp_file_contents):
273
+ pfg.sp_file_result.append(gpt_say)
274
+ pfg.merge_result()
275
+
276
+ # <-------- 临时存储用于调试 ---------->
277
+ pfg.get_token_num = None
278
+ objdump(pfg, file=pj(project_folder,'temp.pkl'))
279
+
280
+ write_html(pfg.sp_file_contents, pfg.sp_file_result, chatbot=chatbot, project_folder=project_folder)
281
+
282
+ # <-------- 写出文件 ---------->
283
+ msg = f"当前大语言模型: {llm_kwargs['llm_model']},当前语言模型温度设定: {llm_kwargs['temperature']}。"
284
+ final_tex = lps.merge_result(pfg.file_result, mode, msg)
285
+ objdump((lps, pfg.file_result, mode, msg), file=pj(project_folder,'merge_result.pkl'))
286
+
287
+ with open(project_folder + f'/merge_{mode}.tex', 'w', encoding='utf-8', errors='replace') as f:
288
+ if mode != 'translate_zh' or "binary" in final_tex: f.write(final_tex)
289
+
290
+
291
+ # <-------- 整理结果, 退出 ---------->
292
+ chatbot.append((f"完成了吗?", 'GPT结果已输出, 即将编译PDF'))
293
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
294
+
295
+ # <-------- 返回 ---------->
296
+ return project_folder + f'/merge_{mode}.tex'
297
+
298
+
299
+ def remove_buggy_lines(file_path, log_path, tex_name, tex_name_pure, n_fix, work_folder_modified, fixed_line=[]):
300
+ try:
301
+ with open(log_path, 'r', encoding='utf-8', errors='replace') as f:
302
+ log = f.read()
303
+ import re
304
+ buggy_lines = re.findall(tex_name+':([0-9]{1,5}):', log)
305
+ buggy_lines = [int(l) for l in buggy_lines]
306
+ buggy_lines = sorted(buggy_lines)
307
+ buggy_line = buggy_lines[0]-1
308
+ print("reversing tex line that has errors", buggy_line)
309
+
310
+ # 重组,逆转出错的段落
311
+ if buggy_line not in fixed_line:
312
+ fixed_line.append(buggy_line)
313
+
314
+ lps, file_result, mode, msg = objload(file=pj(work_folder_modified,'merge_result.pkl'))
315
+ final_tex = lps.merge_result(file_result, mode, msg, buggy_lines=fixed_line, buggy_line_surgery_n_lines=5*n_fix)
316
+
317
+ with open(pj(work_folder_modified, f"{tex_name_pure}_fix_{n_fix}.tex"), 'w', encoding='utf-8', errors='replace') as f:
318
+ f.write(final_tex)
319
+
320
+ return True, f"{tex_name_pure}_fix_{n_fix}", buggy_lines
321
+ except:
322
+ print("Fatal error occurred, but we cannot identify error, please download zip, read latex log, and compile manually.")
323
+ return False, -1, [-1]
324
+
325
+
326
+ def 编译Latex(chatbot, history, main_file_original, main_file_modified, work_folder_original, work_folder_modified, work_folder, mode='default'):
327
+ import os, time
328
+ n_fix = 1
329
+ fixed_line = []
330
+ max_try = 32
331
+ chatbot.append([f"正在编译PDF文档", f'编译已经开始。当前工作路径为{work_folder},如果程序停顿5分钟以上,请直接去该路径下取回翻译结果,或者重启之后再度尝试 ...']); yield from update_ui(chatbot=chatbot, history=history)
332
+ chatbot.append([f"正在编译PDF文档", '...']); yield from update_ui(chatbot=chatbot, history=history); time.sleep(1); chatbot[-1] = list(chatbot[-1]) # 刷新界面
333
+ yield from update_ui_lastest_msg('编译已经开始...', chatbot, history) # 刷新Gradio前端界面
334
+
335
+ while True:
336
+ import os
337
+ may_exist_bbl = pj(work_folder_modified, f'merge.bbl')
338
+ target_bbl = pj(work_folder_modified, f'{main_file_modified}.bbl')
339
+ if os.path.exists(may_exist_bbl) and not os.path.exists(target_bbl):
340
+ shutil.copyfile(may_exist_bbl, target_bbl)
341
+
342
+ # https://stackoverflow.com/questions/738755/dont-make-me-manually-abort-a-latex-compile-when-theres-an-error
343
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 编译原始PDF ...', chatbot, history) # 刷新Gradio前端界面
344
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_original}.tex', work_folder_original)
345
+
346
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 编译转化后的PDF ...', chatbot, history) # 刷新Gradio前端界面
347
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_modified}.tex', work_folder_modified)
348
+
349
+ if ok and os.path.exists(pj(work_folder_modified, f'{main_file_modified}.pdf')):
350
+ # 只有第二步成功,才能继续下面的步骤
351
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 编译BibTex ...', chatbot, history) # 刷新Gradio前端界面
352
+ if not os.path.exists(pj(work_folder_original, f'{main_file_original}.bbl')):
353
+ ok = compile_latex_with_timeout(f'bibtex {main_file_original}.aux', work_folder_original)
354
+ if not os.path.exists(pj(work_folder_modified, f'{main_file_modified}.bbl')):
355
+ ok = compile_latex_with_timeout(f'bibtex {main_file_modified}.aux', work_folder_modified)
356
+
357
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 编译文献交叉引用 ...', chatbot, history) # 刷新Gradio前端界面
358
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_original}.tex', work_folder_original)
359
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_modified}.tex', work_folder_modified)
360
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_original}.tex', work_folder_original)
361
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_modified}.tex', work_folder_modified)
362
+
363
+ if mode!='translate_zh':
364
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 使用latexdiff生成论文转化前后对比 ...', chatbot, history) # 刷新Gradio前端界面
365
+ print( f'latexdiff --encoding=utf8 --append-safecmd=subfile {work_folder_original}/{main_file_original}.tex {work_folder_modified}/{main_file_modified}.tex --flatten > {work_folder}/merge_diff.tex')
366
+ ok = compile_latex_with_timeout(f'latexdiff --encoding=utf8 --append-safecmd=subfile {work_folder_original}/{main_file_original}.tex {work_folder_modified}/{main_file_modified}.tex --flatten > {work_folder}/merge_diff.tex')
367
+
368
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 正在编译对比PDF ...', chatbot, history) # 刷新Gradio前端界面
369
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error merge_diff.tex', work_folder)
370
+ ok = compile_latex_with_timeout(f'bibtex merge_diff.aux', work_folder)
371
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error merge_diff.tex', work_folder)
372
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error merge_diff.tex', work_folder)
373
+
374
+ # <---------- 检查结果 ----------->
375
+ results_ = ""
376
+ original_pdf_success = os.path.exists(pj(work_folder_original, f'{main_file_original}.pdf'))
377
+ modified_pdf_success = os.path.exists(pj(work_folder_modified, f'{main_file_modified}.pdf'))
378
+ diff_pdf_success = os.path.exists(pj(work_folder, f'merge_diff.pdf'))
379
+ results_ += f"原始PDF编译是否成功: {original_pdf_success};"
380
+ results_ += f"转化PDF编译是否成功: {modified_pdf_success};"
381
+ results_ += f"对比PDF编译是否成功: {diff_pdf_success};"
382
+ yield from update_ui_lastest_msg(f'第{n_fix}编译结束:<br/>{results_}...', chatbot, history) # 刷新Gradio前端界面
383
+
384
+ if diff_pdf_success:
385
+ result_pdf = pj(work_folder_modified, f'merge_diff.pdf') # get pdf path
386
+ promote_file_to_downloadzone(result_pdf, rename_file=None, chatbot=chatbot) # promote file to web UI
387
+ if modified_pdf_success:
388
+ yield from update_ui_lastest_msg(f'转化PDF编译已经成功, 即将退出 ...', chatbot, history) # 刷新Gradio前端界面
389
+ result_pdf = pj(work_folder_modified, f'{main_file_modified}.pdf') # get pdf path
390
+ origin_pdf = pj(work_folder_original, f'{main_file_original}.pdf') # get pdf path
391
+ if os.path.exists(pj(work_folder, '..', 'translation')):
392
+ shutil.copyfile(result_pdf, pj(work_folder, '..', 'translation', 'translate_zh.pdf'))
393
+ promote_file_to_downloadzone(result_pdf, rename_file=None, chatbot=chatbot) # promote file to web UI
394
+ # 将两个PDF拼接
395
+ if original_pdf_success:
396
+ try:
397
+ from .latex_toolbox import merge_pdfs
398
+ concat_pdf = pj(work_folder_modified, f'comparison.pdf')
399
+ merge_pdfs(origin_pdf, result_pdf, concat_pdf)
400
+ promote_file_to_downloadzone(concat_pdf, rename_file=None, chatbot=chatbot) # promote file to web UI
401
+ except Exception as e:
402
+ pass
403
+ return True # 成功啦
404
+ else:
405
+ if n_fix>=max_try: break
406
+ n_fix += 1
407
+ can_retry, main_file_modified, buggy_lines = remove_buggy_lines(
408
+ file_path=pj(work_folder_modified, f'{main_file_modified}.tex'),
409
+ log_path=pj(work_folder_modified, f'{main_file_modified}.log'),
410
+ tex_name=f'{main_file_modified}.tex',
411
+ tex_name_pure=f'{main_file_modified}',
412
+ n_fix=n_fix,
413
+ work_folder_modified=work_folder_modified,
414
+ fixed_line=fixed_line
415
+ )
416
+ yield from update_ui_lastest_msg(f'由于最为关键的转化PDF编译失败, 将根据报错信息修正tex源文件并重试, 当前报错的latex代码处于第{buggy_lines}行 ...', chatbot, history) # 刷新Gradio前端界面
417
+ if not can_retry: break
418
+
419
+ return False # 失败啦
420
+
421
+
422
+ def write_html(sp_file_contents, sp_file_result, chatbot, project_folder):
423
+ # write html
424
+ try:
425
+ import shutil
426
+ from ..crazy_utils import construct_html
427
+ from toolbox import gen_time_str
428
+ ch = construct_html()
429
+ orig = ""
430
+ trans = ""
431
+ final = []
432
+ for c,r in zip(sp_file_contents, sp_file_result):
433
+ final.append(c)
434
+ final.append(r)
435
+ for i, k in enumerate(final):
436
+ if i%2==0:
437
+ orig = k
438
+ if i%2==1:
439
+ trans = k
440
+ ch.add_row(a=orig, b=trans)
441
+ create_report_file_name = f"{gen_time_str()}.trans.html"
442
+ ch.save_file(create_report_file_name)
443
+ shutil.copyfile(pj('./gpt_log/', create_report_file_name), pj(project_folder, create_report_file_name))
444
+ promote_file_to_downloadzone(file=f'./gpt_log/{create_report_file_name}', chatbot=chatbot)
445
+ except:
446
+ from toolbox import trimmed_format_exc
447
+ print('writing html result failed:', trimmed_format_exc())
crazy_functions/latex_fns/latex_toolbox.py ADDED
@@ -0,0 +1,459 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os, shutil
2
+ import re
3
+ import numpy as np
4
+ PRESERVE = 0
5
+ TRANSFORM = 1
6
+
7
+ pj = os.path.join
8
+
9
+ class LinkedListNode():
10
+ """
11
+ Linked List Node
12
+ """
13
+ def __init__(self, string, preserve=True) -> None:
14
+ self.string = string
15
+ self.preserve = preserve
16
+ self.next = None
17
+ self.range = None
18
+ # self.begin_line = 0
19
+ # self.begin_char = 0
20
+
21
+ def convert_to_linklist(text, mask):
22
+ root = LinkedListNode("", preserve=True)
23
+ current_node = root
24
+ for c, m, i in zip(text, mask, range(len(text))):
25
+ if (m==PRESERVE and current_node.preserve) \
26
+ or (m==TRANSFORM and not current_node.preserve):
27
+ # add
28
+ current_node.string += c
29
+ else:
30
+ current_node.next = LinkedListNode(c, preserve=(m==PRESERVE))
31
+ current_node = current_node.next
32
+ return root
33
+
34
+ def post_process(root):
35
+ # 修复括号
36
+ node = root
37
+ while True:
38
+ string = node.string
39
+ if node.preserve:
40
+ node = node.next
41
+ if node is None: break
42
+ continue
43
+ def break_check(string):
44
+ str_stack = [""] # (lv, index)
45
+ for i, c in enumerate(string):
46
+ if c == '{':
47
+ str_stack.append('{')
48
+ elif c == '}':
49
+ if len(str_stack) == 1:
50
+ print('stack fix')
51
+ return i
52
+ str_stack.pop(-1)
53
+ else:
54
+ str_stack[-1] += c
55
+ return -1
56
+ bp = break_check(string)
57
+
58
+ if bp == -1:
59
+ pass
60
+ elif bp == 0:
61
+ node.string = string[:1]
62
+ q = LinkedListNode(string[1:], False)
63
+ q.next = node.next
64
+ node.next = q
65
+ else:
66
+ node.string = string[:bp]
67
+ q = LinkedListNode(string[bp:], False)
68
+ q.next = node.next
69
+ node.next = q
70
+
71
+ node = node.next
72
+ if node is None: break
73
+
74
+ # 屏蔽空行和太短的句子
75
+ node = root
76
+ while True:
77
+ if len(node.string.strip('\n').strip(''))==0: node.preserve = True
78
+ if len(node.string.strip('\n').strip(''))<42: node.preserve = True
79
+ node = node.next
80
+ if node is None: break
81
+ node = root
82
+ while True:
83
+ if node.next and node.preserve and node.next.preserve:
84
+ node.string += node.next.string
85
+ node.next = node.next.next
86
+ node = node.next
87
+ if node is None: break
88
+
89
+ # 将前后断行符脱离
90
+ node = root
91
+ prev_node = None
92
+ while True:
93
+ if not node.preserve:
94
+ lstriped_ = node.string.lstrip().lstrip('\n')
95
+ if (prev_node is not None) and (prev_node.preserve) and (len(lstriped_)!=len(node.string)):
96
+ prev_node.string += node.string[:-len(lstriped_)]
97
+ node.string = lstriped_
98
+ rstriped_ = node.string.rstrip().rstrip('\n')
99
+ if (node.next is not None) and (node.next.preserve) and (len(rstriped_)!=len(node.string)):
100
+ node.next.string = node.string[len(rstriped_):] + node.next.string
101
+ node.string = rstriped_
102
+ # =====
103
+ prev_node = node
104
+ node = node.next
105
+ if node is None: break
106
+
107
+ # 标注节点的行数范围
108
+ node = root
109
+ n_line = 0
110
+ expansion = 2
111
+ while True:
112
+ n_l = node.string.count('\n')
113
+ node.range = [n_line-expansion, n_line+n_l+expansion] # 失败时,扭转的范围
114
+ n_line = n_line+n_l
115
+ node = node.next
116
+ if node is None: break
117
+ return root
118
+
119
+
120
+ """
121
+ =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
122
+ Latex segmentation with a binary mask (PRESERVE=0, TRANSFORM=1)
123
+ =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
124
+ """
125
+
126
+
127
+ def set_forbidden_text(text, mask, pattern, flags=0):
128
+ """
129
+ Add a preserve text area in this paper
130
+ e.g. with pattern = r"\\begin\{algorithm\}(.*?)\\end\{algorithm\}"
131
+ you can mask out (mask = PRESERVE so that text become untouchable for GPT)
132
+ everything between "\begin{equation}" and "\end{equation}"
133
+ """
134
+ if isinstance(pattern, list): pattern = '|'.join(pattern)
135
+ pattern_compile = re.compile(pattern, flags)
136
+ for res in pattern_compile.finditer(text):
137
+ mask[res.span()[0]:res.span()[1]] = PRESERVE
138
+ return text, mask
139
+
140
+ def reverse_forbidden_text(text, mask, pattern, flags=0, forbid_wrapper=True):
141
+ """
142
+ Move area out of preserve area (make text editable for GPT)
143
+ count the number of the braces so as to catch compelete text area.
144
+ e.g.
145
+ \begin{abstract} blablablablablabla. \end{abstract}
146
+ """
147
+ if isinstance(pattern, list): pattern = '|'.join(pattern)
148
+ pattern_compile = re.compile(pattern, flags)
149
+ for res in pattern_compile.finditer(text):
150
+ if not forbid_wrapper:
151
+ mask[res.span()[0]:res.span()[1]] = TRANSFORM
152
+ else:
153
+ mask[res.regs[0][0]: res.regs[1][0]] = PRESERVE # '\\begin{abstract}'
154
+ mask[res.regs[1][0]: res.regs[1][1]] = TRANSFORM # abstract
155
+ mask[res.regs[1][1]: res.regs[0][1]] = PRESERVE # abstract
156
+ return text, mask
157
+
158
+ def set_forbidden_text_careful_brace(text, mask, pattern, flags=0):
159
+ """
160
+ Add a preserve text area in this paper (text become untouchable for GPT).
161
+ count the number of the braces so as to catch compelete text area.
162
+ e.g.
163
+ \caption{blablablablabla\texbf{blablabla}blablabla.}
164
+ """
165
+ pattern_compile = re.compile(pattern, flags)
166
+ for res in pattern_compile.finditer(text):
167
+ brace_level = -1
168
+ p = begin = end = res.regs[0][0]
169
+ for _ in range(1024*16):
170
+ if text[p] == '}' and brace_level == 0: break
171
+ elif text[p] == '}': brace_level -= 1
172
+ elif text[p] == '{': brace_level += 1
173
+ p += 1
174
+ end = p+1
175
+ mask[begin:end] = PRESERVE
176
+ return text, mask
177
+
178
+ def reverse_forbidden_text_careful_brace(text, mask, pattern, flags=0, forbid_wrapper=True):
179
+ """
180
+ Move area out of preserve area (make text editable for GPT)
181
+ count the number of the braces so as to catch compelete text area.
182
+ e.g.
183
+ \caption{blablablablabla\texbf{blablabla}blablabla.}
184
+ """
185
+ pattern_compile = re.compile(pattern, flags)
186
+ for res in pattern_compile.finditer(text):
187
+ brace_level = 0
188
+ p = begin = end = res.regs[1][0]
189
+ for _ in range(1024*16):
190
+ if text[p] == '}' and brace_level == 0: break
191
+ elif text[p] == '}': brace_level -= 1
192
+ elif text[p] == '{': brace_level += 1
193
+ p += 1
194
+ end = p
195
+ mask[begin:end] = TRANSFORM
196
+ if forbid_wrapper:
197
+ mask[res.regs[0][0]:begin] = PRESERVE
198
+ mask[end:res.regs[0][1]] = PRESERVE
199
+ return text, mask
200
+
201
+ def set_forbidden_text_begin_end(text, mask, pattern, flags=0, limit_n_lines=42):
202
+ """
203
+ Find all \begin{} ... \end{} text block that with less than limit_n_lines lines.
204
+ Add it to preserve area
205
+ """
206
+ pattern_compile = re.compile(pattern, flags)
207
+ def search_with_line_limit(text, mask):
208
+ for res in pattern_compile.finditer(text):
209
+ cmd = res.group(1) # begin{what}
210
+ this = res.group(2) # content between begin and end
211
+ this_mask = mask[res.regs[2][0]:res.regs[2][1]]
212
+ white_list = ['document', 'abstract', 'lemma', 'definition', 'sproof',
213
+ 'em', 'emph', 'textit', 'textbf', 'itemize', 'enumerate']
214
+ if (cmd in white_list) or this.count('\n') >= limit_n_lines: # use a magical number 42
215
+ this, this_mask = search_with_line_limit(this, this_mask)
216
+ mask[res.regs[2][0]:res.regs[2][1]] = this_mask
217
+ else:
218
+ mask[res.regs[0][0]:res.regs[0][1]] = PRESERVE
219
+ return text, mask
220
+ return search_with_line_limit(text, mask)
221
+
222
+
223
+
224
+ """
225
+ =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
226
+ Latex Merge File
227
+ =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
228
+ """
229
+
230
+ def find_main_tex_file(file_manifest, mode):
231
+ """
232
+ 在多Tex文档中,寻找主文件,必须包含documentclass,返回找到的第一个。
233
+ P.S. 但愿没人把latex模板放在里面传进来 (6.25 加入判定latex模板的代码)
234
+ """
235
+ canidates = []
236
+ for texf in file_manifest:
237
+ if os.path.basename(texf).startswith('merge'):
238
+ continue
239
+ with open(texf, 'r', encoding='utf8', errors='ignore') as f:
240
+ file_content = f.read()
241
+ if r'\documentclass' in file_content:
242
+ canidates.append(texf)
243
+ else:
244
+ continue
245
+
246
+ if len(canidates) == 0:
247
+ raise RuntimeError('无法找到一个主Tex文件(包含documentclass关键字)')
248
+ elif len(canidates) == 1:
249
+ return canidates[0]
250
+ else: # if len(canidates) >= 2 通过一些Latex模板中常见(但通常不会出现在正文)的单词,对不同latex源文件扣分,取评分最高者返回
251
+ canidates_score = []
252
+ # 给出一些判定模板文档的词作为扣分项
253
+ unexpected_words = ['\LaTeX', 'manuscript', 'Guidelines', 'font', 'citations', 'rejected', 'blind review', 'reviewers']
254
+ expected_words = ['\input', '\ref', '\cite']
255
+ for texf in canidates:
256
+ canidates_score.append(0)
257
+ with open(texf, 'r', encoding='utf8', errors='ignore') as f:
258
+ file_content = f.read()
259
+ for uw in unexpected_words:
260
+ if uw in file_content:
261
+ canidates_score[-1] -= 1
262
+ for uw in expected_words:
263
+ if uw in file_content:
264
+ canidates_score[-1] += 1
265
+ select = np.argmax(canidates_score) # 取评分最高者返回
266
+ return canidates[select]
267
+
268
+ def rm_comments(main_file):
269
+ new_file_remove_comment_lines = []
270
+ for l in main_file.splitlines():
271
+ # 删除整行���空注释
272
+ if l.lstrip().startswith("%"):
273
+ pass
274
+ else:
275
+ new_file_remove_comment_lines.append(l)
276
+ main_file = '\n'.join(new_file_remove_comment_lines)
277
+ # main_file = re.sub(r"\\include{(.*?)}", r"\\input{\1}", main_file) # 将 \include 命令转换为 \input 命令
278
+ main_file = re.sub(r'(?<!\\)%.*', '', main_file) # 使用正则表达式查找半行注释, 并替换为空字符串
279
+ return main_file
280
+
281
+ def find_tex_file_ignore_case(fp):
282
+ dir_name = os.path.dirname(fp)
283
+ base_name = os.path.basename(fp)
284
+ # 如果输入的文件路径是正确的
285
+ if os.path.exists(pj(dir_name, base_name)): return pj(dir_name, base_name)
286
+ # 如果不正确,试着加上.tex后缀试试
287
+ if not base_name.endswith('.tex'): base_name+='.tex'
288
+ if os.path.exists(pj(dir_name, base_name)): return pj(dir_name, base_name)
289
+ # 如果还找不到,解除大小写限制,再试一次
290
+ import glob
291
+ for f in glob.glob(dir_name+'/*.tex'):
292
+ base_name_s = os.path.basename(fp)
293
+ if base_name_s.lower() == base_name.lower(): return f
294
+ return None
295
+
296
+ def merge_tex_files_(project_foler, main_file, mode):
297
+ """
298
+ Merge Tex project recrusively
299
+ """
300
+ main_file = rm_comments(main_file)
301
+ for s in reversed([q for q in re.finditer(r"\\input\{(.*?)\}", main_file, re.M)]):
302
+ f = s.group(1)
303
+ fp = os.path.join(project_foler, f)
304
+ fp = find_tex_file_ignore_case(fp)
305
+ if fp:
306
+ with open(fp, 'r', encoding='utf-8', errors='replace') as fx: c = fx.read()
307
+ else:
308
+ raise RuntimeError(f'找不到{fp},Tex源文件缺失!')
309
+ c = merge_tex_files_(project_foler, c, mode)
310
+ main_file = main_file[:s.span()[0]] + c + main_file[s.span()[1]:]
311
+ return main_file
312
+
313
+ def merge_tex_files(project_foler, main_file, mode):
314
+ """
315
+ Merge Tex project recrusively
316
+ P.S. 顺便把CTEX塞进去以支持中文
317
+ P.S. 顺便把Latex的注释去除
318
+ """
319
+ main_file = merge_tex_files_(project_foler, main_file, mode)
320
+ main_file = rm_comments(main_file)
321
+
322
+ if mode == 'translate_zh':
323
+ # find paper documentclass
324
+ pattern = re.compile(r'\\documentclass.*\n')
325
+ match = pattern.search(main_file)
326
+ assert match is not None, "Cannot find documentclass statement!"
327
+ position = match.end()
328
+ add_ctex = '\\usepackage{ctex}\n'
329
+ add_url = '\\usepackage{url}\n' if '{url}' not in main_file else ''
330
+ main_file = main_file[:position] + add_ctex + add_url + main_file[position:]
331
+ # fontset=windows
332
+ import platform
333
+ main_file = re.sub(r"\\documentclass\[(.*?)\]{(.*?)}", r"\\documentclass[\1,fontset=windows,UTF8]{\2}",main_file)
334
+ main_file = re.sub(r"\\documentclass{(.*?)}", r"\\documentclass[fontset=windows,UTF8]{\1}",main_file)
335
+ # find paper abstract
336
+ pattern_opt1 = re.compile(r'\\begin\{abstract\}.*\n')
337
+ pattern_opt2 = re.compile(r"\\abstract\{(.*?)\}", flags=re.DOTALL)
338
+ match_opt1 = pattern_opt1.search(main_file)
339
+ match_opt2 = pattern_opt2.search(main_file)
340
+ assert (match_opt1 is not None) or (match_opt2 is not None), "Cannot find paper abstract section!"
341
+ return main_file
342
+
343
+
344
+ """
345
+ =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
346
+ Post process
347
+ =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
348
+ """
349
+ def mod_inbraket(match):
350
+ """
351
+ 为啥chatgpt会把cite里面的逗号换成中文逗号呀
352
+ """
353
+ # get the matched string
354
+ cmd = match.group(1)
355
+ str_to_modify = match.group(2)
356
+ # modify the matched string
357
+ str_to_modify = str_to_modify.replace(':', ':') # 前面是中文冒号,后面是英文冒号
358
+ str_to_modify = str_to_modify.replace(',', ',') # 前面是中文逗号,后面是英文逗号
359
+ # str_to_modify = 'BOOM'
360
+ return "\\" + cmd + "{" + str_to_modify + "}"
361
+
362
+ def fix_content(final_tex, node_string):
363
+ """
364
+ Fix common GPT errors to increase success rate
365
+ """
366
+ final_tex = re.sub(r"(?<!\\)%", "\\%", final_tex)
367
+ final_tex = re.sub(r"\\([a-z]{2,10})\ \{", r"\\\1{", string=final_tex)
368
+ final_tex = re.sub(r"\\\ ([a-z]{2,10})\{", r"\\\1{", string=final_tex)
369
+ final_tex = re.sub(r"\\([a-z]{2,10})\{([^\}]*?)\}", mod_inbraket, string=final_tex)
370
+
371
+ if "Traceback" in final_tex and "[Local Message]" in final_tex:
372
+ final_tex = node_string # 出问题了,还原原文
373
+ if node_string.count('\\begin') != final_tex.count('\\begin'):
374
+ final_tex = node_string # 出问题了,还原原文
375
+ if node_string.count('\_') > 0 and node_string.count('\_') > final_tex.count('\_'):
376
+ # walk and replace any _ without \
377
+ final_tex = re.sub(r"(?<!\\)_", "\\_", final_tex)
378
+
379
+ def compute_brace_level(string):
380
+ # this function count the number of { and }
381
+ brace_level = 0
382
+ for c in string:
383
+ if c == "{": brace_level += 1
384
+ elif c == "}": brace_level -= 1
385
+ return brace_level
386
+ def join_most(tex_t, tex_o):
387
+ # this function join translated string and original string when something goes wrong
388
+ p_t = 0
389
+ p_o = 0
390
+ def find_next(string, chars, begin):
391
+ p = begin
392
+ while p < len(string):
393
+ if string[p] in chars: return p, string[p]
394
+ p += 1
395
+ return None, None
396
+ while True:
397
+ res1, char = find_next(tex_o, ['{','}'], p_o)
398
+ if res1 is None: break
399
+ res2, char = find_next(tex_t, [char], p_t)
400
+ if res2 is None: break
401
+ p_o = res1 + 1
402
+ p_t = res2 + 1
403
+ return tex_t[:p_t] + tex_o[p_o:]
404
+
405
+ if compute_brace_level(final_tex) != compute_brace_level(node_string):
406
+ # 出问题了,还原部分原文,保证括号正确
407
+ final_tex = join_most(final_tex, node_string)
408
+ return final_tex
409
+
410
+ def compile_latex_with_timeout(command, cwd, timeout=60):
411
+ import subprocess
412
+ process = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, cwd=cwd)
413
+ try:
414
+ stdout, stderr = process.communicate(timeout=timeout)
415
+ except subprocess.TimeoutExpired:
416
+ process.kill()
417
+ stdout, stderr = process.communicate()
418
+ print("Process timed out!")
419
+ return False
420
+ return True
421
+
422
+
423
+
424
+ def merge_pdfs(pdf1_path, pdf2_path, output_path):
425
+ import PyPDF2
426
+ Percent = 0.8
427
+ # Open the first PDF file
428
+ with open(pdf1_path, 'rb') as pdf1_file:
429
+ pdf1_reader = PyPDF2.PdfFileReader(pdf1_file)
430
+ # Open the second PDF file
431
+ with open(pdf2_path, 'rb') as pdf2_file:
432
+ pdf2_reader = PyPDF2.PdfFileReader(pdf2_file)
433
+ # Create a new PDF file to store the merged pages
434
+ output_writer = PyPDF2.PdfFileWriter()
435
+ # Determine the number of pages in each PDF file
436
+ num_pages = max(pdf1_reader.numPages, pdf2_reader.numPages)
437
+ # Merge the pages from the two PDF files
438
+ for page_num in range(num_pages):
439
+ # Add the page from the first PDF file
440
+ if page_num < pdf1_reader.numPages:
441
+ page1 = pdf1_reader.getPage(page_num)
442
+ else:
443
+ page1 = PyPDF2.PageObject.createBlankPage(pdf1_reader)
444
+ # Add the page from the second PDF file
445
+ if page_num < pdf2_reader.numPages:
446
+ page2 = pdf2_reader.getPage(page_num)
447
+ else:
448
+ page2 = PyPDF2.PageObject.createBlankPage(pdf1_reader)
449
+ # Create a new empty page with double width
450
+ new_page = PyPDF2.PageObject.createBlankPage(
451
+ width = int(int(page1.mediaBox.getWidth()) + int(page2.mediaBox.getWidth()) * Percent),
452
+ height = max(page1.mediaBox.getHeight(), page2.mediaBox.getHeight())
453
+ )
454
+ new_page.mergeTranslatedPage(page1, 0, 0)
455
+ new_page.mergeTranslatedPage(page2, int(int(page1.mediaBox.getWidth())-int(page2.mediaBox.getWidth())* (1-Percent)), 0)
456
+ output_writer.addPage(new_page)
457
+ # Save the merged PDF file
458
+ with open(output_path, 'wb') as output_file:
459
+ output_writer.write(output_file)
crazy_functions/latex_utils.py ADDED
@@ -0,0 +1,788 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from toolbox import update_ui, update_ui_lastest_msg # 刷新Gradio前端界面
2
+ from toolbox import zip_folder, objdump, objload, promote_file_to_downloadzone
3
+ import os, shutil
4
+ import re
5
+ import numpy as np
6
+ pj = os.path.join
7
+
8
+ """
9
+ ========================================================================
10
+ Part One
11
+ Latex segmentation with a binary mask (PRESERVE=0, TRANSFORM=1)
12
+ ========================================================================
13
+ """
14
+ PRESERVE = 0
15
+ TRANSFORM = 1
16
+
17
+ def set_forbidden_text(text, mask, pattern, flags=0):
18
+ """
19
+ Add a preserve text area in this paper
20
+ e.g. with pattern = r"\\begin\{algorithm\}(.*?)\\end\{algorithm\}"
21
+ you can mask out (mask = PRESERVE so that text become untouchable for GPT)
22
+ everything between "\begin{equation}" and "\end{equation}"
23
+ """
24
+ if isinstance(pattern, list): pattern = '|'.join(pattern)
25
+ pattern_compile = re.compile(pattern, flags)
26
+ for res in pattern_compile.finditer(text):
27
+ mask[res.span()[0]:res.span()[1]] = PRESERVE
28
+ return text, mask
29
+
30
+ def reverse_forbidden_text(text, mask, pattern, flags=0, forbid_wrapper=True):
31
+ """
32
+ Move area out of preserve area (make text editable for GPT)
33
+ count the number of the braces so as to catch compelete text area.
34
+ e.g.
35
+ \begin{abstract} blablablablablabla. \end{abstract}
36
+ """
37
+ if isinstance(pattern, list): pattern = '|'.join(pattern)
38
+ pattern_compile = re.compile(pattern, flags)
39
+ for res in pattern_compile.finditer(text):
40
+ if not forbid_wrapper:
41
+ mask[res.span()[0]:res.span()[1]] = TRANSFORM
42
+ else:
43
+ mask[res.regs[0][0]: res.regs[1][0]] = PRESERVE # '\\begin{abstract}'
44
+ mask[res.regs[1][0]: res.regs[1][1]] = TRANSFORM # abstract
45
+ mask[res.regs[1][1]: res.regs[0][1]] = PRESERVE # abstract
46
+ return text, mask
47
+
48
+ def set_forbidden_text_careful_brace(text, mask, pattern, flags=0):
49
+ """
50
+ Add a preserve text area in this paper (text become untouchable for GPT).
51
+ count the number of the braces so as to catch compelete text area.
52
+ e.g.
53
+ \caption{blablablablabla\texbf{blablabla}blablabla.}
54
+ """
55
+ pattern_compile = re.compile(pattern, flags)
56
+ for res in pattern_compile.finditer(text):
57
+ brace_level = -1
58
+ p = begin = end = res.regs[0][0]
59
+ for _ in range(1024*16):
60
+ if text[p] == '}' and brace_level == 0: break
61
+ elif text[p] == '}': brace_level -= 1
62
+ elif text[p] == '{': brace_level += 1
63
+ p += 1
64
+ end = p+1
65
+ mask[begin:end] = PRESERVE
66
+ return text, mask
67
+
68
+ def reverse_forbidden_text_careful_brace(text, mask, pattern, flags=0, forbid_wrapper=True):
69
+ """
70
+ Move area out of preserve area (make text editable for GPT)
71
+ count the number of the braces so as to catch compelete text area.
72
+ e.g.
73
+ \caption{blablablablabla\texbf{blablabla}blablabla.}
74
+ """
75
+ pattern_compile = re.compile(pattern, flags)
76
+ for res in pattern_compile.finditer(text):
77
+ brace_level = 0
78
+ p = begin = end = res.regs[1][0]
79
+ for _ in range(1024*16):
80
+ if text[p] == '}' and brace_level == 0: break
81
+ elif text[p] == '}': brace_level -= 1
82
+ elif text[p] == '{': brace_level += 1
83
+ p += 1
84
+ end = p
85
+ mask[begin:end] = TRANSFORM
86
+ if forbid_wrapper:
87
+ mask[res.regs[0][0]:begin] = PRESERVE
88
+ mask[end:res.regs[0][1]] = PRESERVE
89
+ return text, mask
90
+
91
+ def set_forbidden_text_begin_end(text, mask, pattern, flags=0, limit_n_lines=42):
92
+ """
93
+ Find all \begin{} ... \end{} text block that with less than limit_n_lines lines.
94
+ Add it to preserve area
95
+ """
96
+ pattern_compile = re.compile(pattern, flags)
97
+ def search_with_line_limit(text, mask):
98
+ for res in pattern_compile.finditer(text):
99
+ cmd = res.group(1) # begin{what}
100
+ this = res.group(2) # content between begin and end
101
+ this_mask = mask[res.regs[2][0]:res.regs[2][1]]
102
+ white_list = ['document', 'abstract', 'lemma', 'definition', 'sproof',
103
+ 'em', 'emph', 'textit', 'textbf', 'itemize', 'enumerate']
104
+ if (cmd in white_list) or this.count('\n') >= limit_n_lines: # use a magical number 42
105
+ this, this_mask = search_with_line_limit(this, this_mask)
106
+ mask[res.regs[2][0]:res.regs[2][1]] = this_mask
107
+ else:
108
+ mask[res.regs[0][0]:res.regs[0][1]] = PRESERVE
109
+ return text, mask
110
+ return search_with_line_limit(text, mask)
111
+
112
+ class LinkedListNode():
113
+ """
114
+ Linked List Node
115
+ """
116
+ def __init__(self, string, preserve=True) -> None:
117
+ self.string = string
118
+ self.preserve = preserve
119
+ self.next = None
120
+ # self.begin_line = 0
121
+ # self.begin_char = 0
122
+
123
+ def convert_to_linklist(text, mask):
124
+ root = LinkedListNode("", preserve=True)
125
+ current_node = root
126
+ for c, m, i in zip(text, mask, range(len(text))):
127
+ if (m==PRESERVE and current_node.preserve) \
128
+ or (m==TRANSFORM and not current_node.preserve):
129
+ # add
130
+ current_node.string += c
131
+ else:
132
+ current_node.next = LinkedListNode(c, preserve=(m==PRESERVE))
133
+ current_node = current_node.next
134
+ return root
135
+ """
136
+ ========================================================================
137
+ Latex Merge File
138
+ ========================================================================
139
+ """
140
+
141
+ def 寻找Latex主文件(file_manifest, mode):
142
+ """
143
+ 在多Tex文档中,寻找主文件,必须包含documentclass,返回找到的第一个。
144
+ P.S. 但愿没人把latex模板放在里面传进来 (6.25 加入判定latex模板的代码)
145
+ """
146
+ canidates = []
147
+ for texf in file_manifest:
148
+ if os.path.basename(texf).startswith('merge'):
149
+ continue
150
+ with open(texf, 'r', encoding='utf8') as f:
151
+ file_content = f.read()
152
+ if r'\documentclass' in file_content:
153
+ canidates.append(texf)
154
+ else:
155
+ continue
156
+
157
+ if len(canidates) == 0:
158
+ raise RuntimeError('无法找到一个主Tex文件(包含documentclass关键字)')
159
+ elif len(canidates) == 1:
160
+ return canidates[0]
161
+ else: # if len(canidates) >= 2 通过一些Latex模板中常见(但通常不会出现在正文)的单词,对不同latex源文件扣分,取评分最高者返回
162
+ canidates_score = []
163
+ # 给出一些判定模板文档的词作为扣分项
164
+ unexpected_words = ['\LaTeX', 'manuscript', 'Guidelines', 'font', 'citations', 'rejected', 'blind review', 'reviewers']
165
+ expected_words = ['\input', '\ref', '\cite']
166
+ for texf in canidates:
167
+ canidates_score.append(0)
168
+ with open(texf, 'r', encoding='utf8') as f:
169
+ file_content = f.read()
170
+ for uw in unexpected_words:
171
+ if uw in file_content:
172
+ canidates_score[-1] -= 1
173
+ for uw in expected_words:
174
+ if uw in file_content:
175
+ canidates_score[-1] += 1
176
+ select = np.argmax(canidates_score) # 取评分最高者返回
177
+ return canidates[select]
178
+
179
+ def rm_comments(main_file):
180
+ new_file_remove_comment_lines = []
181
+ for l in main_file.splitlines():
182
+ # 删除整行的空注释
183
+ if l.lstrip().startswith("%"):
184
+ pass
185
+ else:
186
+ new_file_remove_comment_lines.append(l)
187
+ main_file = '\n'.join(new_file_remove_comment_lines)
188
+ # main_file = re.sub(r"\\include{(.*?)}", r"\\input{\1}", main_file) # 将 \include 命令转换为 \input 命令
189
+ main_file = re.sub(r'(?<!\\)%.*', '', main_file) # 使用正则表达式查找半行注释, 并替换为空字符串
190
+ return main_file
191
+
192
+ def merge_tex_files_(project_foler, main_file, mode):
193
+ """
194
+ Merge Tex project recrusively
195
+ """
196
+ main_file = rm_comments(main_file)
197
+ for s in reversed([q for q in re.finditer(r"\\input\{(.*?)\}", main_file, re.M)]):
198
+ f = s.group(1)
199
+ fp = os.path.join(project_foler, f)
200
+ if os.path.exists(fp):
201
+ # e.g., \input{srcs/07_appendix.tex}
202
+ with open(fp, 'r', encoding='utf-8', errors='replace') as fx:
203
+ c = fx.read()
204
+ else:
205
+ # e.g., \input{srcs/07_appendix}
206
+ with open(fp+'.tex', 'r', encoding='utf-8', errors='replace') as fx:
207
+ c = fx.read()
208
+ c = merge_tex_files_(project_foler, c, mode)
209
+ main_file = main_file[:s.span()[0]] + c + main_file[s.span()[1]:]
210
+ return main_file
211
+
212
+ def merge_tex_files(project_foler, main_file, mode):
213
+ """
214
+ Merge Tex project recrusively
215
+ P.S. 顺便把CTEX塞进去以支持中文
216
+ P.S. 顺便把Latex的注释去除
217
+ """
218
+ main_file = merge_tex_files_(project_foler, main_file, mode)
219
+ main_file = rm_comments(main_file)
220
+
221
+ if mode == 'translate_zh':
222
+ # find paper documentclass
223
+ pattern = re.compile(r'\\documentclass.*\n')
224
+ match = pattern.search(main_file)
225
+ assert match is not None, "Cannot find documentclass statement!"
226
+ position = match.end()
227
+ add_ctex = '\\usepackage{ctex}\n'
228
+ add_url = '\\usepackage{url}\n' if '{url}' not in main_file else ''
229
+ main_file = main_file[:position] + add_ctex + add_url + main_file[position:]
230
+ # fontset=windows
231
+ import platform
232
+ main_file = re.sub(r"\\documentclass\[(.*?)\]{(.*?)}", r"\\documentclass[\1,fontset=windows,UTF8]{\2}",main_file)
233
+ main_file = re.sub(r"\\documentclass{(.*?)}", r"\\documentclass[fontset=windows,UTF8]{\1}",main_file)
234
+ # find paper abstract
235
+ pattern_opt1 = re.compile(r'\\begin\{abstract\}.*\n')
236
+ pattern_opt2 = re.compile(r"\\abstract\{(.*?)\}", flags=re.DOTALL)
237
+ match_opt1 = pattern_opt1.search(main_file)
238
+ match_opt2 = pattern_opt2.search(main_file)
239
+ assert (match_opt1 is not None) or (match_opt2 is not None), "Cannot find paper abstract section!"
240
+ return main_file
241
+
242
+
243
+
244
+ """
245
+ ========================================================================
246
+ Post process
247
+ ========================================================================
248
+ """
249
+ def mod_inbraket(match):
250
+ """
251
+ 为啥chatgpt会把cite里面的逗号换成中文逗号呀
252
+ """
253
+ # get the matched string
254
+ cmd = match.group(1)
255
+ str_to_modify = match.group(2)
256
+ # modify the matched string
257
+ str_to_modify = str_to_modify.replace(':', ':') # 前面是中文冒号,后面是英文冒号
258
+ str_to_modify = str_to_modify.replace(',', ',') # 前面是中文逗号,后面是英文逗号
259
+ # str_to_modify = 'BOOM'
260
+ return "\\" + cmd + "{" + str_to_modify + "}"
261
+
262
+ def fix_content(final_tex, node_string):
263
+ """
264
+ Fix common GPT errors to increase success rate
265
+ """
266
+ final_tex = re.sub(r"(?<!\\)%", "\\%", final_tex)
267
+ final_tex = re.sub(r"\\([a-z]{2,10})\ \{", r"\\\1{", string=final_tex)
268
+ final_tex = re.sub(r"\\\ ([a-z]{2,10})\{", r"\\\1{", string=final_tex)
269
+ final_tex = re.sub(r"\\([a-z]{2,10})\{([^\}]*?)\}", mod_inbraket, string=final_tex)
270
+
271
+ if "Traceback" in final_tex and "[Local Message]" in final_tex:
272
+ final_tex = node_string # 出问题了,还原原文
273
+ if node_string.count('\\begin') != final_tex.count('\\begin'):
274
+ final_tex = node_string # 出问题了,还原原文
275
+ if node_string.count('\_') > 0 and node_string.count('\_') > final_tex.count('\_'):
276
+ # walk and replace any _ without \
277
+ final_tex = re.sub(r"(?<!\\)_", "\\_", final_tex)
278
+
279
+ def compute_brace_level(string):
280
+ # this function count the number of { and }
281
+ brace_level = 0
282
+ for c in string:
283
+ if c == "{": brace_level += 1
284
+ elif c == "}": brace_level -= 1
285
+ return brace_level
286
+ def join_most(tex_t, tex_o):
287
+ # this function join translated string and original string when something goes wrong
288
+ p_t = 0
289
+ p_o = 0
290
+ def find_next(string, chars, begin):
291
+ p = begin
292
+ while p < len(string):
293
+ if string[p] in chars: return p, string[p]
294
+ p += 1
295
+ return None, None
296
+ while True:
297
+ res1, char = find_next(tex_o, ['{','}'], p_o)
298
+ if res1 is None: break
299
+ res2, char = find_next(tex_t, [char], p_t)
300
+ if res2 is None: break
301
+ p_o = res1 + 1
302
+ p_t = res2 + 1
303
+ return tex_t[:p_t] + tex_o[p_o:]
304
+
305
+ if compute_brace_level(final_tex) != compute_brace_level(node_string):
306
+ # 出问题了,还原部分原文,保证括号正确
307
+ final_tex = join_most(final_tex, node_string)
308
+ return final_tex
309
+
310
+ def split_subprocess(txt, project_folder, return_dict, opts):
311
+ """
312
+ break down latex file to a linked list,
313
+ each node use a preserve flag to indicate whether it should
314
+ be proccessed by GPT.
315
+ """
316
+ text = txt
317
+ mask = np.zeros(len(txt), dtype=np.uint8) + TRANSFORM
318
+
319
+ # 吸收title与作者以上的部分
320
+ text, mask = set_forbidden_text(text, mask, r"(.*?)\\maketitle", re.DOTALL)
321
+ # 吸收iffalse注释
322
+ text, mask = set_forbidden_text(text, mask, r"\\iffalse(.*?)\\fi", re.DOTALL)
323
+ # 吸收在42行以内的begin-end组合
324
+ text, mask = set_forbidden_text_begin_end(text, mask, r"\\begin\{([a-z\*]*)\}(.*?)\\end\{\1\}", re.DOTALL, limit_n_lines=42)
325
+ # 吸收匿名公式
326
+ text, mask = set_forbidden_text(text, mask, [ r"\$\$(.*?)\$\$", r"\\\[.*?\\\]" ], re.DOTALL)
327
+ # 吸收其他杂项
328
+ text, mask = set_forbidden_text(text, mask, [ r"\\section\{(.*?)\}", r"\\section\*\{(.*?)\}", r"\\subsection\{(.*?)\}", r"\\subsubsection\{(.*?)\}" ])
329
+ text, mask = set_forbidden_text(text, mask, [ r"\\bibliography\{(.*?)\}", r"\\bibliographystyle\{(.*?)\}" ])
330
+ text, mask = set_forbidden_text(text, mask, r"\\begin\{thebibliography\}.*?\\end\{thebibliography\}", re.DOTALL)
331
+ text, mask = set_forbidden_text(text, mask, r"\\begin\{lstlisting\}(.*?)\\end\{lstlisting\}", re.DOTALL)
332
+ text, mask = set_forbidden_text(text, mask, r"\\begin\{wraptable\}(.*?)\\end\{wraptable\}", re.DOTALL)
333
+ text, mask = set_forbidden_text(text, mask, r"\\begin\{algorithm\}(.*?)\\end\{algorithm\}", re.DOTALL)
334
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{wrapfigure\}(.*?)\\end\{wrapfigure\}", r"\\begin\{wrapfigure\*\}(.*?)\\end\{wrapfigure\*\}"], re.DOTALL)
335
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{figure\}(.*?)\\end\{figure\}", r"\\begin\{figure\*\}(.*?)\\end\{figure\*\}"], re.DOTALL)
336
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{multline\}(.*?)\\end\{multline\}", r"\\begin\{multline\*\}(.*?)\\end\{multline\*\}"], re.DOTALL)
337
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{table\}(.*?)\\end\{table\}", r"\\begin\{table\*\}(.*?)\\end\{table\*\}"], re.DOTALL)
338
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{minipage\}(.*?)\\end\{minipage\}", r"\\begin\{minipage\*\}(.*?)\\end\{minipage\*\}"], re.DOTALL)
339
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{align\*\}(.*?)\\end\{align\*\}", r"\\begin\{align\}(.*?)\\end\{align\}"], re.DOTALL)
340
+ text, mask = set_forbidden_text(text, mask, [r"\\begin\{equation\}(.*?)\\end\{equation\}", r"\\begin\{equation\*\}(.*?)\\end\{equation\*\}"], re.DOTALL)
341
+ text, mask = set_forbidden_text(text, mask, [r"\\includepdf\[(.*?)\]\{(.*?)\}", r"\\clearpage", r"\\newpage", r"\\appendix", r"\\tableofcontents", r"\\include\{(.*?)\}"])
342
+ text, mask = set_forbidden_text(text, mask, [r"\\vspace\{(.*?)\}", r"\\hspace\{(.*?)\}", r"\\label\{(.*?)\}", r"\\begin\{(.*?)\}", r"\\end\{(.*?)\}", r"\\item "])
343
+ text, mask = set_forbidden_text_careful_brace(text, mask, r"\\hl\{(.*?)\}", re.DOTALL)
344
+ # reverse 操作必须放在最后
345
+ text, mask = reverse_forbidden_text_careful_brace(text, mask, r"\\caption\{(.*?)\}", re.DOTALL, forbid_wrapper=True)
346
+ text, mask = reverse_forbidden_text_careful_brace(text, mask, r"\\abstract\{(.*?)\}", re.DOTALL, forbid_wrapper=True)
347
+ text, mask = reverse_forbidden_text(text, mask, r"\\begin\{abstract\}(.*?)\\end\{abstract\}", re.DOTALL, forbid_wrapper=True)
348
+ root = convert_to_linklist(text, mask)
349
+
350
+ # 修复括号
351
+ node = root
352
+ while True:
353
+ string = node.string
354
+ if node.preserve:
355
+ node = node.next
356
+ if node is None: break
357
+ continue
358
+ def break_check(string):
359
+ str_stack = [""] # (lv, index)
360
+ for i, c in enumerate(string):
361
+ if c == '{':
362
+ str_stack.append('{')
363
+ elif c == '}':
364
+ if len(str_stack) == 1:
365
+ print('stack fix')
366
+ return i
367
+ str_stack.pop(-1)
368
+ else:
369
+ str_stack[-1] += c
370
+ return -1
371
+ bp = break_check(string)
372
+
373
+ if bp == -1:
374
+ pass
375
+ elif bp == 0:
376
+ node.string = string[:1]
377
+ q = LinkedListNode(string[1:], False)
378
+ q.next = node.next
379
+ node.next = q
380
+ else:
381
+ node.string = string[:bp]
382
+ q = LinkedListNode(string[bp:], False)
383
+ q.next = node.next
384
+ node.next = q
385
+
386
+ node = node.next
387
+ if node is None: break
388
+
389
+ # 屏蔽空行和太短的句子
390
+ node = root
391
+ while True:
392
+ if len(node.string.strip('\n').strip(''))==0: node.preserve = True
393
+ if len(node.string.strip('\n').strip(''))<42: node.preserve = True
394
+ node = node.next
395
+ if node is None: break
396
+ node = root
397
+ while True:
398
+ if node.next and node.preserve and node.next.preserve:
399
+ node.string += node.next.string
400
+ node.next = node.next.next
401
+ node = node.next
402
+ if node is None: break
403
+
404
+ # 将前后断行符脱离
405
+ node = root
406
+ prev_node = None
407
+ while True:
408
+ if not node.preserve:
409
+ lstriped_ = node.string.lstrip().lstrip('\n')
410
+ if (prev_node is not None) and (prev_node.preserve) and (len(lstriped_)!=len(node.string)):
411
+ prev_node.string += node.string[:-len(lstriped_)]
412
+ node.string = lstriped_
413
+ rstriped_ = node.string.rstrip().rstrip('\n')
414
+ if (node.next is not None) and (node.next.preserve) and (len(rstriped_)!=len(node.string)):
415
+ node.next.string = node.string[len(rstriped_):] + node.next.string
416
+ node.string = rstriped_
417
+ # =====
418
+ prev_node = node
419
+ node = node.next
420
+ if node is None: break
421
+ # 输出html调试文件,用红色标注处保留区(PRESERVE),用黑色标注转换区(TRANSFORM)
422
+ with open(pj(project_folder, 'debug_log.html'), 'w', encoding='utf8') as f:
423
+ segment_parts_for_gpt = []
424
+ nodes = []
425
+ node = root
426
+ while True:
427
+ nodes.append(node)
428
+ show_html = node.string.replace('\n','<br/>')
429
+ if not node.preserve:
430
+ segment_parts_for_gpt.append(node.string)
431
+ f.write(f'<p style="color:black;">#{show_html}#</p>')
432
+ else:
433
+ f.write(f'<p style="color:red;">{show_html}</p>')
434
+ node = node.next
435
+ if node is None: break
436
+
437
+ for n in nodes: n.next = None # break
438
+ return_dict['nodes'] = nodes
439
+ return_dict['segment_parts_for_gpt'] = segment_parts_for_gpt
440
+ return return_dict
441
+
442
+
443
+
444
+ class LatexPaperSplit():
445
+ """
446
+ break down latex file to a linked list,
447
+ each node use a preserve flag to indicate whether it should
448
+ be proccessed by GPT.
449
+ """
450
+ def __init__(self) -> None:
451
+ self.nodes = None
452
+ self.msg = "*{\\scriptsize\\textbf{警告:该PDF由GPT-Academic开源项目调用大语言模型+Latex翻译插件一键生成," + \
453
+ "版权归原文作者所有。翻译内容可靠性无保障,请仔细鉴别并以原文为准。" + \
454
+ "项目Github地址 \\url{https://github.com/binary-husky/gpt_academic/}。"
455
+ # 请您不要删除或修改这行警告,除非您是论文的原作者(如果您是论文原作者,欢迎加REAME中的QQ联系开发者)
456
+ self.msg_declare = "为了防止大语言模型的意外谬误产生扩散影响,禁止移除或修改此警告。}}\\\\"
457
+
458
+ def merge_result(self, arr, mode, msg):
459
+ """
460
+ Merge the result after the GPT process completed
461
+ """
462
+ result_string = ""
463
+ p = 0
464
+ for node in self.nodes:
465
+ if node.preserve:
466
+ result_string += node.string
467
+ else:
468
+ result_string += fix_content(arr[p], node.string)
469
+ p += 1
470
+ if mode == 'translate_zh':
471
+ pattern = re.compile(r'\\begin\{abstract\}.*\n')
472
+ match = pattern.search(result_string)
473
+ if not match:
474
+ # match \abstract{xxxx}
475
+ pattern_compile = re.compile(r"\\abstract\{(.*?)\}", flags=re.DOTALL)
476
+ match = pattern_compile.search(result_string)
477
+ position = match.regs[1][0]
478
+ else:
479
+ # match \begin{abstract}xxxx\end{abstract}
480
+ position = match.end()
481
+ result_string = result_string[:position] + self.msg + msg + self.msg_declare + result_string[position:]
482
+ return result_string
483
+
484
+ def split(self, txt, project_folder, opts):
485
+ """
486
+ break down latex file to a linked list,
487
+ each node use a preserve flag to indicate whether it should
488
+ be proccessed by GPT.
489
+ P.S. use multiprocessing to avoid timeout error
490
+ """
491
+ import multiprocessing
492
+ manager = multiprocessing.Manager()
493
+ return_dict = manager.dict()
494
+ p = multiprocessing.Process(
495
+ target=split_subprocess,
496
+ args=(txt, project_folder, return_dict, opts))
497
+ p.start()
498
+ p.join()
499
+ p.close()
500
+ self.nodes = return_dict['nodes']
501
+ self.sp = return_dict['segment_parts_for_gpt']
502
+ return self.sp
503
+
504
+
505
+
506
+ class LatexPaperFileGroup():
507
+ """
508
+ use tokenizer to break down text according to max_token_limit
509
+ """
510
+ def __init__(self):
511
+ self.file_paths = []
512
+ self.file_contents = []
513
+ self.sp_file_contents = []
514
+ self.sp_file_index = []
515
+ self.sp_file_tag = []
516
+
517
+ # count_token
518
+ from request_llm.bridge_all import model_info
519
+ enc = model_info["gpt-3.5-turbo"]['tokenizer']
520
+ def get_token_num(txt): return len(enc.encode(txt, disallowed_special=()))
521
+ self.get_token_num = get_token_num
522
+
523
+ def run_file_split(self, max_token_limit=1900):
524
+ """
525
+ use tokenizer to break down text according to max_token_limit
526
+ """
527
+ for index, file_content in enumerate(self.file_contents):
528
+ if self.get_token_num(file_content) < max_token_limit:
529
+ self.sp_file_contents.append(file_content)
530
+ self.sp_file_index.append(index)
531
+ self.sp_file_tag.append(self.file_paths[index])
532
+ else:
533
+ from .crazy_utils import breakdown_txt_to_satisfy_token_limit_for_pdf
534
+ segments = breakdown_txt_to_satisfy_token_limit_for_pdf(file_content, self.get_token_num, max_token_limit)
535
+ for j, segment in enumerate(segments):
536
+ self.sp_file_contents.append(segment)
537
+ self.sp_file_index.append(index)
538
+ self.sp_file_tag.append(self.file_paths[index] + f".part-{j}.tex")
539
+ print('Segmentation: done')
540
+
541
+ def merge_result(self):
542
+ self.file_result = ["" for _ in range(len(self.file_paths))]
543
+ for r, k in zip(self.sp_file_result, self.sp_file_index):
544
+ self.file_result[k] += r
545
+
546
+ def write_result(self):
547
+ manifest = []
548
+ for path, res in zip(self.file_paths, self.file_result):
549
+ with open(path + '.polish.tex', 'w', encoding='utf8') as f:
550
+ manifest.append(path + '.polish.tex')
551
+ f.write(res)
552
+ return manifest
553
+
554
+ def write_html(sp_file_contents, sp_file_result, chatbot, project_folder):
555
+
556
+ # write html
557
+ try:
558
+ import shutil
559
+ from .crazy_utils import construct_html
560
+ from toolbox import gen_time_str
561
+ ch = construct_html()
562
+ orig = ""
563
+ trans = ""
564
+ final = []
565
+ for c,r in zip(sp_file_contents, sp_file_result):
566
+ final.append(c)
567
+ final.append(r)
568
+ for i, k in enumerate(final):
569
+ if i%2==0:
570
+ orig = k
571
+ if i%2==1:
572
+ trans = k
573
+ ch.add_row(a=orig, b=trans)
574
+ create_report_file_name = f"{gen_time_str()}.trans.html"
575
+ ch.save_file(create_report_file_name)
576
+ shutil.copyfile(pj('./gpt_log/', create_report_file_name), pj(project_folder, create_report_file_name))
577
+ promote_file_to_downloadzone(file=f'./gpt_log/{create_report_file_name}', chatbot=chatbot)
578
+ except:
579
+ from toolbox import trimmed_format_exc
580
+ print('writing html result failed:', trimmed_format_exc())
581
+
582
+ def Latex精细分解与转化(file_manifest, project_folder, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, mode='proofread', switch_prompt=None, opts=[]):
583
+ import time, os, re
584
+ from .crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency
585
+ from .latex_utils import LatexPaperFileGroup, merge_tex_files, LatexPaperSplit, 寻找Latex主文件
586
+
587
+ # <-------- 寻找主tex文件 ---------->
588
+ maintex = 寻找Latex主文件(file_manifest, mode)
589
+ chatbot.append((f"定位主Latex文件", f'[Local Message] 分析结果:该项目的Latex主文件是{maintex}, 如果分析错误, 请立即终止程序, 删除或修改歧义文件, 然后重试。主程序即将开始, 请稍候。'))
590
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
591
+ time.sleep(3)
592
+
593
+ # <-------- 读取Latex文件, 将多文件tex工程融合为一个巨型tex ---------->
594
+ main_tex_basename = os.path.basename(maintex)
595
+ assert main_tex_basename.endswith('.tex')
596
+ main_tex_basename_bare = main_tex_basename[:-4]
597
+ may_exist_bbl = pj(project_folder, f'{main_tex_basename_bare}.bbl')
598
+ if os.path.exists(may_exist_bbl):
599
+ shutil.copyfile(may_exist_bbl, pj(project_folder, f'merge.bbl'))
600
+ shutil.copyfile(may_exist_bbl, pj(project_folder, f'merge_{mode}.bbl'))
601
+ shutil.copyfile(may_exist_bbl, pj(project_folder, f'merge_diff.bbl'))
602
+
603
+ with open(maintex, 'r', encoding='utf-8', errors='replace') as f:
604
+ content = f.read()
605
+ merged_content = merge_tex_files(project_folder, content, mode)
606
+
607
+ with open(project_folder + '/merge.tex', 'w', encoding='utf-8', errors='replace') as f:
608
+ f.write(merged_content)
609
+
610
+ # <-------- 精细切分latex文件 ---------->
611
+ chatbot.append((f"Latex文件融合完成", f'[Local Message] 正在精细切分latex文件,这需要一段时间计算,文档越长耗时越长,请耐心等待。'))
612
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
613
+ lps = LatexPaperSplit()
614
+ res = lps.split(merged_content, project_folder, opts) # 消耗时间的函数
615
+
616
+ # <-------- 拆分过长的latex片段 ---------->
617
+ pfg = LatexPaperFileGroup()
618
+ for index, r in enumerate(res):
619
+ pfg.file_paths.append('segment-' + str(index))
620
+ pfg.file_contents.append(r)
621
+
622
+ pfg.run_file_split(max_token_limit=1024)
623
+ n_split = len(pfg.sp_file_contents)
624
+
625
+ # <-------- 根据需要切换prompt ---------->
626
+ inputs_array, sys_prompt_array = switch_prompt(pfg, mode)
627
+ inputs_show_user_array = [f"{mode} {f}" for f in pfg.sp_file_tag]
628
+
629
+ if os.path.exists(pj(project_folder,'temp.pkl')):
630
+
631
+ # <-------- 【仅调试】如果存在调试缓存文件,则跳过GPT请求环节 ---------->
632
+ pfg = objload(file=pj(project_folder,'temp.pkl'))
633
+
634
+ else:
635
+ # <-------- gpt 多线程请求 ---------->
636
+ gpt_response_collection = yield from request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency(
637
+ inputs_array=inputs_array,
638
+ inputs_show_user_array=inputs_show_user_array,
639
+ llm_kwargs=llm_kwargs,
640
+ chatbot=chatbot,
641
+ history_array=[[""] for _ in range(n_split)],
642
+ sys_prompt_array=sys_prompt_array,
643
+ # max_workers=5, # 并行任务数量限制, 最多同时执行5个, 其他的排队等待
644
+ scroller_max_len = 40
645
+ )
646
+
647
+ # <-------- 文本碎片重组为完整的tex片段 ---------->
648
+ pfg.sp_file_result = []
649
+ for i_say, gpt_say, orig_content in zip(gpt_response_collection[0::2], gpt_response_collection[1::2], pfg.sp_file_contents):
650
+ pfg.sp_file_result.append(gpt_say)
651
+ pfg.merge_result()
652
+
653
+ # <-------- 临时存储用于调试 ---------->
654
+ pfg.get_token_num = None
655
+ objdump(pfg, file=pj(project_folder,'temp.pkl'))
656
+
657
+ write_html(pfg.sp_file_contents, pfg.sp_file_result, chatbot=chatbot, project_folder=project_folder)
658
+
659
+ # <-------- 写出文件 ---------->
660
+ msg = f"当前大语言模型: {llm_kwargs['llm_model']},当前语言模型温度设定: {llm_kwargs['temperature']}。"
661
+ final_tex = lps.merge_result(pfg.file_result, mode, msg)
662
+ with open(project_folder + f'/merge_{mode}.tex', 'w', encoding='utf-8', errors='replace') as f:
663
+ if mode != 'translate_zh' or "binary" in final_tex: f.write(final_tex)
664
+
665
+
666
+ # <-------- 整理结果, 退出 ---------->
667
+ chatbot.append((f"完成了吗?", 'GPT结果已输出, 正在编译PDF'))
668
+ yield from update_ui(chatbot=chatbot, history=history) # 刷新界面
669
+
670
+ # <-------- 返回 ---------->
671
+ return project_folder + f'/merge_{mode}.tex'
672
+
673
+
674
+
675
+ def remove_buggy_lines(file_path, log_path, tex_name, tex_name_pure, n_fix, work_folder_modified):
676
+ try:
677
+ with open(log_path, 'r', encoding='utf-8', errors='replace') as f:
678
+ log = f.read()
679
+ with open(file_path, 'r', encoding='utf-8', errors='replace') as f:
680
+ file_lines = f.readlines()
681
+ import re
682
+ buggy_lines = re.findall(tex_name+':([0-9]{1,5}):', log)
683
+ buggy_lines = [int(l) for l in buggy_lines]
684
+ buggy_lines = sorted(buggy_lines)
685
+ print("removing lines that has errors", buggy_lines)
686
+ file_lines.pop(buggy_lines[0]-1)
687
+ with open(pj(work_folder_modified, f"{tex_name_pure}_fix_{n_fix}.tex"), 'w', encoding='utf-8', errors='replace') as f:
688
+ f.writelines(file_lines)
689
+ return True, f"{tex_name_pure}_fix_{n_fix}", buggy_lines
690
+ except:
691
+ print("Fatal error occurred, but we cannot identify error, please download zip, read latex log, and compile manually.")
692
+ return False, -1, [-1]
693
+
694
+ def compile_latex_with_timeout(command, cwd, timeout=60):
695
+ import subprocess
696
+ process = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, cwd=cwd)
697
+ try:
698
+ stdout, stderr = process.communicate(timeout=timeout)
699
+ except subprocess.TimeoutExpired:
700
+ process.kill()
701
+ stdout, stderr = process.communicate()
702
+ print("Process timed out!")
703
+ return False
704
+ return True
705
+
706
+ def 编译Latex(chatbot, history, main_file_original, main_file_modified, work_folder_original, work_folder_modified, work_folder, mode='default'):
707
+ import os, time
708
+ current_dir = os.getcwd()
709
+ n_fix = 1
710
+ max_try = 32
711
+ chatbot.append([f"正在编译PDF文档", f'编译已经开始。当前工作路径为{work_folder},如果程序停顿5分钟以上,请直接去该路径下取回翻译结果,或者重启之后再度尝试 ...']); yield from update_ui(chatbot=chatbot, history=history)
712
+ chatbot.append([f"正在编译PDF文档", '...']); yield from update_ui(chatbot=chatbot, history=history); time.sleep(1); chatbot[-1] = list(chatbot[-1]) # 刷新界面
713
+ yield from update_ui_lastest_msg('编译已经开始...', chatbot, history) # 刷新Gradio前端界面
714
+
715
+ while True:
716
+ import os
717
+
718
+ # https://stackoverflow.com/questions/738755/dont-make-me-manually-abort-a-latex-compile-when-theres-an-error
719
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 编译原始PDF ...', chatbot, history) # 刷新Gradio前端界面
720
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_original}.tex', work_folder_original)
721
+
722
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 编译转化后的PDF ...', chatbot, history) # 刷新Gradio前端界面
723
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_modified}.tex', work_folder_modified)
724
+
725
+ if ok and os.path.exists(pj(work_folder_modified, f'{main_file_modified}.pdf')):
726
+ # 只有第二步成功,才能继续下面的步骤
727
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 编译BibTex ...', chatbot, history) # 刷新Gradio前端界面
728
+ if not os.path.exists(pj(work_folder_original, f'{main_file_original}.bbl')):
729
+ ok = compile_latex_with_timeout(f'bibtex {main_file_original}.aux', work_folder_original)
730
+ if not os.path.exists(pj(work_folder_modified, f'{main_file_modified}.bbl')):
731
+ ok = compile_latex_with_timeout(f'bibtex {main_file_modified}.aux', work_folder_modified)
732
+
733
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 编译文献交叉引用 ...', chatbot, history) # 刷新Gradio前端界面
734
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_original}.tex', work_folder_original)
735
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_modified}.tex', work_folder_modified)
736
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_original}.tex', work_folder_original)
737
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error {main_file_modified}.tex', work_folder_modified)
738
+
739
+ if mode!='translate_zh':
740
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 使用latexdiff生成论文转化前后对比 ...', chatbot, history) # 刷新Gradio前端界面
741
+ print( f'latexdiff --encoding=utf8 --append-safecmd=subfile {work_folder_original}/{main_file_original}.tex {work_folder_modified}/{main_file_modified}.tex --flatten > {work_folder}/merge_diff.tex')
742
+ ok = compile_latex_with_timeout(f'latexdiff --encoding=utf8 --append-safecmd=subfile {work_folder_original}/{main_file_original}.tex {work_folder_modified}/{main_file_modified}.tex --flatten > {work_folder}/merge_diff.tex')
743
+
744
+ yield from update_ui_lastest_msg(f'尝试第 {n_fix}/{max_try} 次编译, 正在编译对比PDF ...', chatbot, history) # 刷新Gradio前端界面
745
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error merge_diff.tex', work_folder)
746
+ ok = compile_latex_with_timeout(f'bibtex merge_diff.aux', work_folder)
747
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error merge_diff.tex', work_folder)
748
+ ok = compile_latex_with_timeout(f'pdflatex -interaction=batchmode -file-line-error merge_diff.tex', work_folder)
749
+
750
+
751
+ # <---------- 检查结果 ----------->
752
+ results_ = ""
753
+ original_pdf_success = os.path.exists(pj(work_folder_original, f'{main_file_original}.pdf'))
754
+ modified_pdf_success = os.path.exists(pj(work_folder_modified, f'{main_file_modified}.pdf'))
755
+ diff_pdf_success = os.path.exists(pj(work_folder, f'merge_diff.pdf'))
756
+ results_ += f"原始PDF编译是否成功: {original_pdf_success};"
757
+ results_ += f"转化PDF编译是否成功: {modified_pdf_success};"
758
+ results_ += f"对比PDF编译是否成功: {diff_pdf_success};"
759
+ yield from update_ui_lastest_msg(f'第{n_fix}编译结束:<br/>{results_}...', chatbot, history) # 刷新Gradio前端界面
760
+
761
+ if diff_pdf_success:
762
+ result_pdf = pj(work_folder_modified, f'merge_diff.pdf') # get pdf path
763
+ promote_file_to_downloadzone(result_pdf, rename_file=None, chatbot=chatbot) # promote file to web UI
764
+ if modified_pdf_success:
765
+ yield from update_ui_lastest_msg(f'转化PDF编译已经成功, 即将退出 ...', chatbot, history) # 刷新Gradio前端界面
766
+ result_pdf = pj(work_folder_modified, f'{main_file_modified}.pdf') # get pdf path
767
+ if os.path.exists(pj(work_folder, '..', 'translation')):
768
+ shutil.copyfile(result_pdf, pj(work_folder, '..', 'translation', 'translate_zh.pdf'))
769
+ promote_file_to_downloadzone(result_pdf, rename_file=None, chatbot=chatbot) # promote file to web UI
770
+ return True # 成功啦
771
+ else:
772
+ if n_fix>=max_try: break
773
+ n_fix += 1
774
+ can_retry, main_file_modified, buggy_lines = remove_buggy_lines(
775
+ file_path=pj(work_folder_modified, f'{main_file_modified}.tex'),
776
+ log_path=pj(work_folder_modified, f'{main_file_modified}.log'),
777
+ tex_name=f'{main_file_modified}.tex',
778
+ tex_name_pure=f'{main_file_modified}',
779
+ n_fix=n_fix,
780
+ work_folder_modified=work_folder_modified,
781
+ )
782
+ yield from update_ui_lastest_msg(f'由于最为关键的转化PDF编译失败, 将根据报错信息修正tex源文件并重试, 当前报错的latex代码处于第{buggy_lines}行 ...', chatbot, history) # 刷新Gradio前端界面
783
+ if not can_retry: break
784
+
785
+ return False # 失败啦
786
+
787
+
788
+
crazy_functions/live_audio/aliyunASR.py ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import time, threading, json
2
+
3
+
4
+ class AliyunASR():
5
+
6
+ def test_on_sentence_begin(self, message, *args):
7
+ # print("test_on_sentence_begin:{}".format(message))
8
+ pass
9
+
10
+ def test_on_sentence_end(self, message, *args):
11
+ # print("test_on_sentence_end:{}".format(message))
12
+ message = json.loads(message)
13
+ self.parsed_sentence = message['payload']['result']
14
+ self.event_on_entence_end.set()
15
+ print(self.parsed_sentence)
16
+
17
+ def test_on_start(self, message, *args):
18
+ # print("test_on_start:{}".format(message))
19
+ pass
20
+
21
+ def test_on_error(self, message, *args):
22
+ print("on_error args=>{}".format(args))
23
+ pass
24
+
25
+ def test_on_close(self, *args):
26
+ self.aliyun_service_ok = False
27
+ pass
28
+
29
+ def test_on_result_chg(self, message, *args):
30
+ # print("test_on_chg:{}".format(message))
31
+ message = json.loads(message)
32
+ self.parsed_text = message['payload']['result']
33
+ self.event_on_result_chg.set()
34
+
35
+ def test_on_completed(self, message, *args):
36
+ # print("on_completed:args=>{} message=>{}".format(args, message))
37
+ pass
38
+
39
+
40
+ def audio_convertion_thread(self, uuid):
41
+ # 在一个异步线程中采集音频
42
+ import nls # pip install git+https://github.com/aliyun/alibabacloud-nls-python-sdk.git
43
+ import tempfile
44
+ from scipy import io
45
+ from toolbox import get_conf
46
+ from .audio_io import change_sample_rate
47
+ from .audio_io import RealtimeAudioDistribution
48
+ NEW_SAMPLERATE = 16000
49
+ rad = RealtimeAudioDistribution()
50
+ rad.clean_up()
51
+ temp_folder = tempfile.gettempdir()
52
+ TOKEN, APPKEY = get_conf('ALIYUN_TOKEN', 'ALIYUN_APPKEY')
53
+ if len(TOKEN) == 0:
54
+ TOKEN = self.get_token()
55
+ self.aliyun_service_ok = True
56
+ URL="wss://nls-gateway.aliyuncs.com/ws/v1"
57
+ sr = nls.NlsSpeechTranscriber(
58
+ url=URL,
59
+ token=TOKEN,
60
+ appkey=APPKEY,
61
+ on_sentence_begin=self.test_on_sentence_begin,
62
+ on_sentence_end=self.test_on_sentence_end,
63
+ on_start=self.test_on_start,
64
+ on_result_changed=self.test_on_result_chg,
65
+ on_completed=self.test_on_completed,
66
+ on_error=self.test_on_error,
67
+ on_close=self.test_on_close,
68
+ callback_args=[uuid.hex]
69
+ )
70
+
71
+ r = sr.start(aformat="pcm",
72
+ enable_intermediate_result=True,
73
+ enable_punctuation_prediction=True,
74
+ enable_inverse_text_normalization=True)
75
+
76
+ while not self.stop:
77
+ # time.sleep(self.capture_interval)
78
+ audio = rad.read(uuid.hex)
79
+ if audio is not None:
80
+ # convert to pcm file
81
+ temp_file = f'{temp_folder}/{uuid.hex}.pcm' #
82
+ dsdata = change_sample_rate(audio, rad.rate, NEW_SAMPLERATE) # 48000 --> 16000
83
+ io.wavfile.write(temp_file, NEW_SAMPLERATE, dsdata)
84
+ # read pcm binary
85
+ with open(temp_file, "rb") as f: data = f.read()
86
+ # print('audio len:', len(audio), '\t ds len:', len(dsdata), '\t need n send:', len(data)//640)
87
+ slices = zip(*(iter(data),) * 640) # 640个字节为一组
88
+ for i in slices: sr.send_audio(bytes(i))
89
+ else:
90
+ time.sleep(0.1)
91
+
92
+ if not self.aliyun_service_ok:
93
+ self.stop = True
94
+ self.stop_msg = 'Aliyun音频服务异常,请检查ALIYUN_TOKEN和ALIYUN_APPKEY是否过期。'
95
+ r = sr.stop()
96
+
97
+ def get_token(self):
98
+ from toolbox import get_conf
99
+ import json
100
+ from aliyunsdkcore.request import CommonRequest
101
+ from aliyunsdkcore.client import AcsClient
102
+ AccessKey_ID, AccessKey_secret = get_conf('ALIYUN_ACCESSKEY', 'ALIYUN_SECRET')
103
+
104
+ # 创建AcsClient实例
105
+ client = AcsClient(
106
+ AccessKey_ID,
107
+ AccessKey_secret,
108
+ "cn-shanghai"
109
+ )
110
+
111
+ # 创建request,并设置参数。
112
+ request = CommonRequest()
113
+ request.set_method('POST')
114
+ request.set_domain('nls-meta.cn-shanghai.aliyuncs.com')
115
+ request.set_version('2019-02-28')
116
+ request.set_action_name('CreateToken')
117
+
118
+ try:
119
+ response = client.do_action_with_exception(request)
120
+ print(response)
121
+ jss = json.loads(response)
122
+ if 'Token' in jss and 'Id' in jss['Token']:
123
+ token = jss['Token']['Id']
124
+ expireTime = jss['Token']['ExpireTime']
125
+ print("token = " + token)
126
+ print("expireTime = " + str(expireTime))
127
+ except Exception as e:
128
+ print(e)
129
+
130
+ return token
crazy_functions/live_audio/audio_io.py ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import numpy as np
2
+ from scipy import interpolate
3
+
4
+ def Singleton(cls):
5
+ _instance = {}
6
+
7
+ def _singleton(*args, **kargs):
8
+ if cls not in _instance:
9
+ _instance[cls] = cls(*args, **kargs)
10
+ return _instance[cls]
11
+
12
+ return _singleton
13
+
14
+
15
+ @Singleton
16
+ class RealtimeAudioDistribution():
17
+ def __init__(self) -> None:
18
+ self.data = {}
19
+ self.max_len = 1024*1024
20
+ self.rate = 48000 # 只读,每秒采样数量
21
+
22
+ def clean_up(self):
23
+ self.data = {}
24
+
25
+ def feed(self, uuid, audio):
26
+ self.rate, audio_ = audio
27
+ # print('feed', len(audio_), audio_[-25:])
28
+ if uuid not in self.data:
29
+ self.data[uuid] = audio_
30
+ else:
31
+ new_arr = np.concatenate((self.data[uuid], audio_))
32
+ if len(new_arr) > self.max_len: new_arr = new_arr[-self.max_len:]
33
+ self.data[uuid] = new_arr
34
+
35
+ def read(self, uuid):
36
+ if uuid in self.data:
37
+ res = self.data.pop(uuid)
38
+ print('\r read-', len(res), '-', max(res), end='', flush=True)
39
+ else:
40
+ res = None
41
+ return res
42
+
43
+ def change_sample_rate(audio, old_sr, new_sr):
44
+ duration = audio.shape[0] / old_sr
45
+
46
+ time_old = np.linspace(0, duration, audio.shape[0])
47
+ time_new = np.linspace(0, duration, int(audio.shape[0] * new_sr / old_sr))
48
+
49
+ interpolator = interpolate.interp1d(time_old, audio.T)
50
+ new_audio = interpolator(time_new).T
51
+ return new_audio.astype(np.int16)
crazy_functions/pdf_fns/parse_pdf.py ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import requests
2
+ import random
3
+ from functools import lru_cache
4
+ class GROBID_OFFLINE_EXCEPTION(Exception): pass
5
+
6
+ def get_avail_grobid_url():
7
+ from toolbox import get_conf
8
+ GROBID_URLS, = get_conf('GROBID_URLS')
9
+ if len(GROBID_URLS) == 0: return None
10
+ try:
11
+ _grobid_url = random.choice(GROBID_URLS) # 随机负载均衡
12
+ if _grobid_url.endswith('/'): _grobid_url = _grobid_url.rstrip('/')
13
+ res = requests.get(_grobid_url+'/api/isalive')
14
+ if res.text=='true': return _grobid_url
15
+ else: return None
16
+ except:
17
+ return None
18
+
19
+ @lru_cache(maxsize=32)
20
+ def parse_pdf(pdf_path, grobid_url):
21
+ import scipdf # pip install scipdf_parser
22
+ if grobid_url.endswith('/'): grobid_url = grobid_url.rstrip('/')
23
+ article_dict = scipdf.parse_pdf_to_dict(pdf_path, grobid_url=grobid_url)
24
+ return article_dict
25
+
crazy_functions/test_project/cpp/cppipc/buffer.cpp ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #include "libipc/buffer.h"
2
+ #include "libipc/utility/pimpl.h"
3
+
4
+ #include <cstring>
5
+
6
+ namespace ipc {
7
+
8
+ bool operator==(buffer const & b1, buffer const & b2) {
9
+ return (b1.size() == b2.size()) && (std::memcmp(b1.data(), b2.data(), b1.size()) == 0);
10
+ }
11
+
12
+ bool operator!=(buffer const & b1, buffer const & b2) {
13
+ return !(b1 == b2);
14
+ }
15
+
16
+ class buffer::buffer_ : public pimpl<buffer_> {
17
+ public:
18
+ void* p_;
19
+ std::size_t s_;
20
+ void* a_;
21
+ buffer::destructor_t d_;
22
+
23
+ buffer_(void* p, std::size_t s, buffer::destructor_t d, void* a)
24
+ : p_(p), s_(s), a_(a), d_(d) {
25
+ }
26
+
27
+ ~buffer_() {
28
+ if (d_ == nullptr) return;
29
+ d_((a_ == nullptr) ? p_ : a_, s_);
30
+ }
31
+ };
32
+
33
+ buffer::buffer()
34
+ : buffer(nullptr, 0, nullptr, nullptr) {
35
+ }
36
+
37
+ buffer::buffer(void* p, std::size_t s, destructor_t d)
38
+ : p_(p_->make(p, s, d, nullptr)) {
39
+ }
40
+
41
+ buffer::buffer(void* p, std::size_t s, destructor_t d, void* additional)
42
+ : p_(p_->make(p, s, d, additional)) {
43
+ }
44
+
45
+ buffer::buffer(void* p, std::size_t s)
46
+ : buffer(p, s, nullptr) {
47
+ }
48
+
49
+ buffer::buffer(char const & c)
50
+ : buffer(const_cast<char*>(&c), 1) {
51
+ }
52
+
53
+ buffer::buffer(buffer&& rhs)
54
+ : buffer() {
55
+ swap(rhs);
56
+ }
57
+
58
+ buffer::~buffer() {
59
+ p_->clear();
60
+ }
61
+
62
+ void buffer::swap(buffer& rhs) {
63
+ std::swap(p_, rhs.p_);
64
+ }
65
+
66
+ buffer& buffer::operator=(buffer rhs) {
67
+ swap(rhs);
68
+ return *this;
69
+ }
70
+
71
+ bool buffer::empty() const noexcept {
72
+ return (impl(p_)->p_ == nullptr) || (impl(p_)->s_ == 0);
73
+ }
74
+
75
+ void* buffer::data() noexcept {
76
+ return impl(p_)->p_;
77
+ }
78
+
79
+ void const * buffer::data() const noexcept {
80
+ return impl(p_)->p_;
81
+ }
82
+
83
+ std::size_t buffer::size() const noexcept {
84
+ return impl(p_)->s_;
85
+ }
86
+
87
+ } // namespace ipc
crazy_functions/test_project/cpp/cppipc/ipc.cpp ADDED
@@ -0,0 +1,701 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ #include <type_traits>
3
+ #include <cstring>
4
+ #include <algorithm>
5
+ #include <utility> // std::pair, std::move, std::forward
6
+ #include <atomic>
7
+ #include <type_traits> // aligned_storage_t
8
+ #include <string>
9
+ #include <vector>
10
+ #include <array>
11
+ #include <cassert>
12
+
13
+ #include "libipc/ipc.h"
14
+ #include "libipc/def.h"
15
+ #include "libipc/shm.h"
16
+ #include "libipc/pool_alloc.h"
17
+ #include "libipc/queue.h"
18
+ #include "libipc/policy.h"
19
+ #include "libipc/rw_lock.h"
20
+ #include "libipc/waiter.h"
21
+
22
+ #include "libipc/utility/log.h"
23
+ #include "libipc/utility/id_pool.h"
24
+ #include "libipc/utility/scope_guard.h"
25
+ #include "libipc/utility/utility.h"
26
+
27
+ #include "libipc/memory/resource.h"
28
+ #include "libipc/platform/detail.h"
29
+ #include "libipc/circ/elem_array.h"
30
+
31
+ namespace {
32
+
33
+ using msg_id_t = std::uint32_t;
34
+ using acc_t = std::atomic<msg_id_t>;
35
+
36
+ template <std::size_t DataSize, std::size_t AlignSize>
37
+ struct msg_t;
38
+
39
+ template <std::size_t AlignSize>
40
+ struct msg_t<0, AlignSize> {
41
+ msg_id_t cc_id_;
42
+ msg_id_t id_;
43
+ std::int32_t remain_;
44
+ bool storage_;
45
+ };
46
+
47
+ template <std::size_t DataSize, std::size_t AlignSize>
48
+ struct msg_t : msg_t<0, AlignSize> {
49
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
50
+
51
+ msg_t() = default;
52
+ msg_t(msg_id_t cc_id, msg_id_t id, std::int32_t remain, void const * data, std::size_t size)
53
+ : msg_t<0, AlignSize> {cc_id, id, remain, (data == nullptr) || (size == 0)} {
54
+ if (this->storage_) {
55
+ if (data != nullptr) {
56
+ // copy storage-id
57
+ *reinterpret_cast<ipc::storage_id_t*>(&data_) =
58
+ *static_cast<ipc::storage_id_t const *>(data);
59
+ }
60
+ }
61
+ else std::memcpy(&data_, data, size);
62
+ }
63
+ };
64
+
65
+ template <typename T>
66
+ ipc::buff_t make_cache(T& data, std::size_t size) {
67
+ auto ptr = ipc::mem::alloc(size);
68
+ std::memcpy(ptr, &data, (ipc::detail::min)(sizeof(data), size));
69
+ return { ptr, size, ipc::mem::free };
70
+ }
71
+
72
+ struct cache_t {
73
+ std::size_t fill_;
74
+ ipc::buff_t buff_;
75
+
76
+ cache_t(std::size_t f, ipc::buff_t && b)
77
+ : fill_(f), buff_(std::move(b))
78
+ {}
79
+
80
+ void append(void const * data, std::size_t size) {
81
+ if (fill_ >= buff_.size() || data == nullptr || size == 0) return;
82
+ auto new_fill = (ipc::detail::min)(fill_ + size, buff_.size());
83
+ std::memcpy(static_cast<ipc::byte_t*>(buff_.data()) + fill_, data, new_fill - fill_);
84
+ fill_ = new_fill;
85
+ }
86
+ };
87
+
88
+ auto cc_acc() {
89
+ static ipc::shm::handle acc_h("__CA_CONN__", sizeof(acc_t));
90
+ return static_cast<acc_t*>(acc_h.get());
91
+ }
92
+
93
+ IPC_CONSTEXPR_ std::size_t align_chunk_size(std::size_t size) noexcept {
94
+ return (((size - 1) / ipc::large_msg_align) + 1) * ipc::large_msg_align;
95
+ }
96
+
97
+ IPC_CONSTEXPR_ std::size_t calc_chunk_size(std::size_t size) noexcept {
98
+ return ipc::make_align(alignof(std::max_align_t), align_chunk_size(
99
+ ipc::make_align(alignof(std::max_align_t), sizeof(std::atomic<ipc::circ::cc_t>)) + size));
100
+ }
101
+
102
+ struct chunk_t {
103
+ std::atomic<ipc::circ::cc_t> &conns() noexcept {
104
+ return *reinterpret_cast<std::atomic<ipc::circ::cc_t> *>(this);
105
+ }
106
+
107
+ void *data() noexcept {
108
+ return reinterpret_cast<ipc::byte_t *>(this)
109
+ + ipc::make_align(alignof(std::max_align_t), sizeof(std::atomic<ipc::circ::cc_t>));
110
+ }
111
+ };
112
+
113
+ struct chunk_info_t {
114
+ ipc::id_pool<> pool_;
115
+ ipc::spin_lock lock_;
116
+
117
+ IPC_CONSTEXPR_ static std::size_t chunks_mem_size(std::size_t chunk_size) noexcept {
118
+ return ipc::id_pool<>::max_count * chunk_size;
119
+ }
120
+
121
+ ipc::byte_t *chunks_mem() noexcept {
122
+ return reinterpret_cast<ipc::byte_t *>(this + 1);
123
+ }
124
+
125
+ chunk_t *at(std::size_t chunk_size, ipc::storage_id_t id) noexcept {
126
+ if (id < 0) return nullptr;
127
+ return reinterpret_cast<chunk_t *>(chunks_mem() + (chunk_size * id));
128
+ }
129
+ };
130
+
131
+ auto& chunk_storages() {
132
+ class chunk_handle_t {
133
+ ipc::shm::handle handle_;
134
+
135
+ public:
136
+ chunk_info_t *get_info(std::size_t chunk_size) {
137
+ if (!handle_.valid() &&
138
+ !handle_.acquire( ("__CHUNK_INFO__" + ipc::to_string(chunk_size)).c_str(),
139
+ sizeof(chunk_info_t) + chunk_info_t::chunks_mem_size(chunk_size) )) {
140
+ ipc::error("[chunk_storages] chunk_shm.id_info_.acquire failed: chunk_size = %zd\n", chunk_size);
141
+ return nullptr;
142
+ }
143
+ auto info = static_cast<chunk_info_t*>(handle_.get());
144
+ if (info == nullptr) {
145
+ ipc::error("[chunk_storages] chunk_shm.id_info_.get failed: chunk_size = %zd\n", chunk_size);
146
+ return nullptr;
147
+ }
148
+ return info;
149
+ }
150
+ };
151
+ static ipc::map<std::size_t, chunk_handle_t> chunk_hs;
152
+ return chunk_hs;
153
+ }
154
+
155
+ chunk_info_t *chunk_storage_info(std::size_t chunk_size) {
156
+ auto &storages = chunk_storages();
157
+ std::decay_t<decltype(storages)>::iterator it;
158
+ {
159
+ static ipc::rw_lock lock;
160
+ IPC_UNUSED_ std::shared_lock<ipc::rw_lock> guard {lock};
161
+ if ((it = storages.find(chunk_size)) == storages.end()) {
162
+ using chunk_handle_t = std::decay_t<decltype(storages)>::value_type::second_type;
163
+ guard.unlock();
164
+ IPC_UNUSED_ std::lock_guard<ipc::rw_lock> guard {lock};
165
+ it = storages.emplace(chunk_size, chunk_handle_t{}).first;
166
+ }
167
+ }
168
+ return it->second.get_info(chunk_size);
169
+ }
170
+
171
+ std::pair<ipc::storage_id_t, void*> acquire_storage(std::size_t size, ipc::circ::cc_t conns) {
172
+ std::size_t chunk_size = calc_chunk_size(size);
173
+ auto info = chunk_storage_info(chunk_size);
174
+ if (info == nullptr) return {};
175
+
176
+ info->lock_.lock();
177
+ info->pool_.prepare();
178
+ // got an unique id
179
+ auto id = info->pool_.acquire();
180
+ info->lock_.unlock();
181
+
182
+ auto chunk = info->at(chunk_size, id);
183
+ if (chunk == nullptr) return {};
184
+ chunk->conns().store(conns, std::memory_order_relaxed);
185
+ return { id, chunk->data() };
186
+ }
187
+
188
+ void *find_storage(ipc::storage_id_t id, std::size_t size) {
189
+ if (id < 0) {
190
+ ipc::error("[find_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
191
+ return nullptr;
192
+ }
193
+ std::size_t chunk_size = calc_chunk_size(size);
194
+ auto info = chunk_storage_info(chunk_size);
195
+ if (info == nullptr) return nullptr;
196
+ return info->at(chunk_size, id)->data();
197
+ }
198
+
199
+ void release_storage(ipc::storage_id_t id, std::size_t size) {
200
+ if (id < 0) {
201
+ ipc::error("[release_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
202
+ return;
203
+ }
204
+ std::size_t chunk_size = calc_chunk_size(size);
205
+ auto info = chunk_storage_info(chunk_size);
206
+ if (info == nullptr) return;
207
+ info->lock_.lock();
208
+ info->pool_.release(id);
209
+ info->lock_.unlock();
210
+ }
211
+
212
+ template <ipc::relat Rp, ipc::relat Rc>
213
+ bool sub_rc(ipc::wr<Rp, Rc, ipc::trans::unicast>,
214
+ std::atomic<ipc::circ::cc_t> &/*conns*/, ipc::circ::cc_t /*curr_conns*/, ipc::circ::cc_t /*conn_id*/) noexcept {
215
+ return true;
216
+ }
217
+
218
+ template <ipc::relat Rp, ipc::relat Rc>
219
+ bool sub_rc(ipc::wr<Rp, Rc, ipc::trans::broadcast>,
220
+ std::atomic<ipc::circ::cc_t> &conns, ipc::circ::cc_t curr_conns, ipc::circ::cc_t conn_id) noexcept {
221
+ auto last_conns = curr_conns & ~conn_id;
222
+ for (unsigned k = 0;;) {
223
+ auto chunk_conns = conns.load(std::memory_order_acquire);
224
+ if (conns.compare_exchange_weak(chunk_conns, chunk_conns & last_conns, std::memory_order_release)) {
225
+ return (chunk_conns & last_conns) == 0;
226
+ }
227
+ ipc::yield(k);
228
+ }
229
+ }
230
+
231
+ template <typename Flag>
232
+ void recycle_storage(ipc::storage_id_t id, std::size_t size, ipc::circ::cc_t curr_conns, ipc::circ::cc_t conn_id) {
233
+ if (id < 0) {
234
+ ipc::error("[recycle_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
235
+ return;
236
+ }
237
+ std::size_t chunk_size = calc_chunk_size(size);
238
+ auto info = chunk_storage_info(chunk_size);
239
+ if (info == nullptr) return;
240
+
241
+ auto chunk = info->at(chunk_size, id);
242
+ if (chunk == nullptr) return;
243
+
244
+ if (!sub_rc(Flag{}, chunk->conns(), curr_conns, conn_id)) {
245
+ return;
246
+ }
247
+ info->lock_.lock();
248
+ info->pool_.release(id);
249
+ info->lock_.unlock();
250
+ }
251
+
252
+ template <typename MsgT>
253
+ bool clear_message(void* p) {
254
+ auto msg = static_cast<MsgT*>(p);
255
+ if (msg->storage_) {
256
+ std::int32_t r_size = static_cast<std::int32_t>(ipc::data_length) + msg->remain_;
257
+ if (r_size <= 0) {
258
+ ipc::error("[clear_message] invalid msg size: %d\n", (int)r_size);
259
+ return true;
260
+ }
261
+ release_storage(
262
+ *reinterpret_cast<ipc::storage_id_t*>(&msg->data_),
263
+ static_cast<std::size_t>(r_size));
264
+ }
265
+ return true;
266
+ }
267
+
268
+ struct conn_info_head {
269
+
270
+ ipc::string name_;
271
+ msg_id_t cc_id_; // connection-info id
272
+ ipc::detail::waiter cc_waiter_, wt_waiter_, rd_waiter_;
273
+ ipc::shm::handle acc_h_;
274
+
275
+ conn_info_head(char const * name)
276
+ : name_ {name}
277
+ , cc_id_ {(cc_acc() == nullptr) ? 0 : cc_acc()->fetch_add(1, std::memory_order_relaxed)}
278
+ , cc_waiter_{("__CC_CONN__" + name_).c_str()}
279
+ , wt_waiter_{("__WT_CONN__" + name_).c_str()}
280
+ , rd_waiter_{("__RD_CONN__" + name_).c_str()}
281
+ , acc_h_ {("__AC_CONN__" + name_).c_str(), sizeof(acc_t)} {
282
+ }
283
+
284
+ void quit_waiting() {
285
+ cc_waiter_.quit_waiting();
286
+ wt_waiter_.quit_waiting();
287
+ rd_waiter_.quit_waiting();
288
+ }
289
+
290
+ auto acc() {
291
+ return static_cast<acc_t*>(acc_h_.get());
292
+ }
293
+
294
+ auto& recv_cache() {
295
+ thread_local ipc::unordered_map<msg_id_t, cache_t> tls;
296
+ return tls;
297
+ }
298
+ };
299
+
300
+ template <typename W, typename F>
301
+ bool wait_for(W& waiter, F&& pred, std::uint64_t tm) {
302
+ if (tm == 0) return !pred();
303
+ for (unsigned k = 0; pred();) {
304
+ bool ret = true;
305
+ ipc::sleep(k, [&k, &ret, &waiter, &pred, tm] {
306
+ ret = waiter.wait_if(std::forward<F>(pred), tm);
307
+ k = 0;
308
+ });
309
+ if (!ret) return false; // timeout or fail
310
+ if (k == 0) break; // k has been reset
311
+ }
312
+ return true;
313
+ }
314
+
315
+ template <typename Policy,
316
+ std::size_t DataSize = ipc::data_length,
317
+ std::size_t AlignSize = (ipc::detail::min)(DataSize, alignof(std::max_align_t))>
318
+ struct queue_generator {
319
+
320
+ using queue_t = ipc::queue<msg_t<DataSize, AlignSize>, Policy>;
321
+
322
+ struct conn_info_t : conn_info_head {
323
+ queue_t que_;
324
+
325
+ conn_info_t(char const * name)
326
+ : conn_info_head{name}
327
+ , que_{("__QU_CONN__" +
328
+ ipc::to_string(DataSize) + "__" +
329
+ ipc::to_string(AlignSize) + "__" + name).c_str()} {
330
+ }
331
+
332
+ void disconnect_receiver() {
333
+ bool dis = que_.disconnect();
334
+ this->quit_waiting();
335
+ if (dis) {
336
+ this->recv_cache().clear();
337
+ }
338
+ }
339
+ };
340
+ };
341
+
342
+ template <typename Policy>
343
+ struct detail_impl {
344
+
345
+ using policy_t = Policy;
346
+ using flag_t = typename policy_t::flag_t;
347
+ using queue_t = typename queue_generator<policy_t>::queue_t;
348
+ using conn_info_t = typename queue_generator<policy_t>::conn_info_t;
349
+
350
+ constexpr static conn_info_t* info_of(ipc::handle_t h) noexcept {
351
+ return static_cast<conn_info_t*>(h);
352
+ }
353
+
354
+ constexpr static queue_t* queue_of(ipc::handle_t h) noexcept {
355
+ return (info_of(h) == nullptr) ? nullptr : &(info_of(h)->que_);
356
+ }
357
+
358
+ /* API implementations */
359
+
360
+ static void disconnect(ipc::handle_t h) {
361
+ auto que = queue_of(h);
362
+ if (que == nullptr) {
363
+ return;
364
+ }
365
+ que->shut_sending();
366
+ assert(info_of(h) != nullptr);
367
+ info_of(h)->disconnect_receiver();
368
+ }
369
+
370
+ static bool reconnect(ipc::handle_t * ph, bool start_to_recv) {
371
+ assert(ph != nullptr);
372
+ assert(*ph != nullptr);
373
+ auto que = queue_of(*ph);
374
+ if (que == nullptr) {
375
+ return false;
376
+ }
377
+ if (start_to_recv) {
378
+ que->shut_sending();
379
+ if (que->connect()) { // wouldn't connect twice
380
+ info_of(*ph)->cc_waiter_.broadcast();
381
+ return true;
382
+ }
383
+ return false;
384
+ }
385
+ // start_to_recv == false
386
+ if (que->connected()) {
387
+ info_of(*ph)->disconnect_receiver();
388
+ }
389
+ return que->ready_sending();
390
+ }
391
+
392
+ static bool connect(ipc::handle_t * ph, char const * name, bool start_to_recv) {
393
+ assert(ph != nullptr);
394
+ if (*ph == nullptr) {
395
+ *ph = ipc::mem::alloc<conn_info_t>(name);
396
+ }
397
+ return reconnect(ph, start_to_recv);
398
+ }
399
+
400
+ static void destroy(ipc::handle_t h) {
401
+ disconnect(h);
402
+ ipc::mem::free(info_of(h));
403
+ }
404
+
405
+ static std::size_t recv_count(ipc::handle_t h) noexcept {
406
+ auto que = queue_of(h);
407
+ if (que == nullptr) {
408
+ return ipc::invalid_value;
409
+ }
410
+ return que->conn_count();
411
+ }
412
+
413
+ static bool wait_for_recv(ipc::handle_t h, std::size_t r_count, std::uint64_t tm) {
414
+ auto que = queue_of(h);
415
+ if (que == nullptr) {
416
+ return false;
417
+ }
418
+ return wait_for(info_of(h)->cc_waiter_, [que, r_count] {
419
+ return que->conn_count() < r_count;
420
+ }, tm);
421
+ }
422
+
423
+ template <typename F>
424
+ static bool send(F&& gen_push, ipc::handle_t h, void const * data, std::size_t size) {
425
+ if (data == nullptr || size == 0) {
426
+ ipc::error("fail: send(%p, %zd)\n", data, size);
427
+ return false;
428
+ }
429
+ auto que = queue_of(h);
430
+ if (que == nullptr) {
431
+ ipc::error("fail: send, queue_of(h) == nullptr\n");
432
+ return false;
433
+ }
434
+ if (que->elems() == nullptr) {
435
+ ipc::error("fail: send, queue_of(h)->elems() == nullptr\n");
436
+ return false;
437
+ }
438
+ if (!que->ready_sending()) {
439
+ ipc::error("fail: send, que->ready_sending() == false\n");
440
+ return false;
441
+ }
442
+ ipc::circ::cc_t conns = que->elems()->connections(std::memory_order_relaxed);
443
+ if (conns == 0) {
444
+ ipc::error("fail: send, there is no receiver on this connection.\n");
445
+ return false;
446
+ }
447
+ // calc a new message id
448
+ auto acc = info_of(h)->acc();
449
+ if (acc == nullptr) {
450
+ ipc::error("fail: send, info_of(h)->acc() == nullptr\n");
451
+ return false;
452
+ }
453
+ auto msg_id = acc->fetch_add(1, std::memory_order_relaxed);
454
+ auto try_push = std::forward<F>(gen_push)(info_of(h), que, msg_id);
455
+ if (size > ipc::large_msg_limit) {
456
+ auto dat = acquire_storage(size, conns);
457
+ void * buf = dat.second;
458
+ if (buf != nullptr) {
459
+ std::memcpy(buf, data, size);
460
+ return try_push(static_cast<std::int32_t>(size) -
461
+ static_cast<std::int32_t>(ipc::data_length), &(dat.first), 0);
462
+ }
463
+ // try using message fragment
464
+ //ipc::log("fail: shm::handle for big message. msg_id: %zd, size: %zd\n", msg_id, size);
465
+ }
466
+ // push message fragment
467
+ std::int32_t offset = 0;
468
+ for (std::int32_t i = 0; i < static_cast<std::int32_t>(size / ipc::data_length); ++i, offset += ipc::data_length) {
469
+ if (!try_push(static_cast<std::int32_t>(size) - offset - static_cast<std::int32_t>(ipc::data_length),
470
+ static_cast<ipc::byte_t const *>(data) + offset, ipc::data_length)) {
471
+ return false;
472
+ }
473
+ }
474
+ // if remain > 0, this is the last message fragment
475
+ std::int32_t remain = static_cast<std::int32_t>(size) - offset;
476
+ if (remain > 0) {
477
+ if (!try_push(remain - static_cast<std::int32_t>(ipc::data_length),
478
+ static_cast<ipc::byte_t const *>(data) + offset,
479
+ static_cast<std::size_t>(remain))) {
480
+ return false;
481
+ }
482
+ }
483
+ return true;
484
+ }
485
+
486
+ static bool send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
487
+ return send([tm](auto info, auto que, auto msg_id) {
488
+ return [tm, info, que, msg_id](std::int32_t remain, void const * data, std::size_t size) {
489
+ if (!wait_for(info->wt_waiter_, [&] {
490
+ return !que->push(
491
+ [](void*) { return true; },
492
+ info->cc_id_, msg_id, remain, data, size);
493
+ }, tm)) {
494
+ ipc::log("force_push: msg_id = %zd, remain = %d, size = %zd\n", msg_id, remain, size);
495
+ if (!que->force_push(
496
+ clear_message<typename queue_t::value_t>,
497
+ info->cc_id_, msg_id, remain, data, size)) {
498
+ return false;
499
+ }
500
+ }
501
+ info->rd_waiter_.broadcast();
502
+ return true;
503
+ };
504
+ }, h, data, size);
505
+ }
506
+
507
+ static bool try_send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
508
+ return send([tm](auto info, auto que, auto msg_id) {
509
+ return [tm, info, que, msg_id](std::int32_t remain, void const * data, std::size_t size) {
510
+ if (!wait_for(info->wt_waiter_, [&] {
511
+ return !que->push(
512
+ [](void*) { return true; },
513
+ info->cc_id_, msg_id, remain, data, size);
514
+ }, tm)) {
515
+ return false;
516
+ }
517
+ info->rd_waiter_.broadcast();
518
+ return true;
519
+ };
520
+ }, h, data, size);
521
+ }
522
+
523
+ static ipc::buff_t recv(ipc::handle_t h, std::uint64_t tm) {
524
+ auto que = queue_of(h);
525
+ if (que == nullptr) {
526
+ ipc::error("fail: recv, queue_of(h) == nullptr\n");
527
+ return {};
528
+ }
529
+ if (!que->connected()) {
530
+ // hasn't connected yet, just return.
531
+ return {};
532
+ }
533
+ auto& rc = info_of(h)->recv_cache();
534
+ for (;;) {
535
+ // pop a new message
536
+ typename queue_t::value_t msg;
537
+ if (!wait_for(info_of(h)->rd_waiter_, [que, &msg] {
538
+ return !que->pop(msg);
539
+ }, tm)) {
540
+ // pop failed, just return.
541
+ return {};
542
+ }
543
+ info_of(h)->wt_waiter_.broadcast();
544
+ if ((info_of(h)->acc() != nullptr) && (msg.cc_id_ == info_of(h)->cc_id_)) {
545
+ continue; // ignore message to self
546
+ }
547
+ // msg.remain_ may minus & abs(msg.remain_) < data_length
548
+ std::int32_t r_size = static_cast<std::int32_t>(ipc::data_length) + msg.remain_;
549
+ if (r_size <= 0) {
550
+ ipc::error("fail: recv, r_size = %d\n", (int)r_size);
551
+ return {};
552
+ }
553
+ std::size_t msg_size = static_cast<std::size_t>(r_size);
554
+ // large message
555
+ if (msg.storage_) {
556
+ ipc::storage_id_t buf_id = *reinterpret_cast<ipc::storage_id_t*>(&msg.data_);
557
+ void* buf = find_storage(buf_id, msg_size);
558
+ if (buf != nullptr) {
559
+ struct recycle_t {
560
+ ipc::storage_id_t storage_id;
561
+ ipc::circ::cc_t curr_conns;
562
+ ipc::circ::cc_t conn_id;
563
+ } *r_info = ipc::mem::alloc<recycle_t>(recycle_t{
564
+ buf_id, que->elems()->connections(std::memory_order_relaxed), que->connected_id()
565
+ });
566
+ if (r_info == nullptr) {
567
+ ipc::log("fail: ipc::mem::alloc<recycle_t>.\n");
568
+ return ipc::buff_t{buf, msg_size}; // no recycle
569
+ } else {
570
+ return ipc::buff_t{buf, msg_size, [](void* p_info, std::size_t size) {
571
+ auto r_info = static_cast<recycle_t *>(p_info);
572
+ IPC_UNUSED_ auto finally = ipc::guard([r_info] {
573
+ ipc::mem::free(r_info);
574
+ });
575
+ recycle_storage<flag_t>(r_info->storage_id, size, r_info->curr_conns, r_info->conn_id);
576
+ }, r_info};
577
+ }
578
+ } else {
579
+ ipc::log("fail: shm::handle for large message. msg_id: %zd, buf_id: %zd, size: %zd\n", msg.id_, buf_id, msg_size);
580
+ continue;
581
+ }
582
+ }
583
+ // find cache with msg.id_
584
+ auto cac_it = rc.find(msg.id_);
585
+ if (cac_it == rc.end()) {
586
+ if (msg_size <= ipc::data_length) {
587
+ return make_cache(msg.data_, msg_size);
588
+ }
589
+ // gc
590
+ if (rc.size() > 1024) {
591
+ std::vector<msg_id_t> need_del;
592
+ for (auto const & pair : rc) {
593
+ auto cmp = std::minmax(msg.id_, pair.first);
594
+ if (cmp.second - cmp.first > 8192) {
595
+ need_del.push_back(pair.first);
596
+ }
597
+ }
598
+ for (auto id : need_del) rc.erase(id);
599
+ }
600
+ // cache the first message fragment
601
+ rc.emplace(msg.id_, cache_t { ipc::data_length, make_cache(msg.data_, msg_size) });
602
+ }
603
+ // has cached before this message
604
+ else {
605
+ auto& cac = cac_it->second;
606
+ // this is the last message fragment
607
+ if (msg.remain_ <= 0) {
608
+ cac.append(&(msg.data_), msg_size);
609
+ // finish this message, erase it from cache
610
+ auto buff = std::move(cac.buff_);
611
+ rc.erase(cac_it);
612
+ return buff;
613
+ }
614
+ // there are remain datas after this message
615
+ cac.append(&(msg.data_), ipc::data_length);
616
+ }
617
+ }
618
+ }
619
+
620
+ static ipc::buff_t try_recv(ipc::handle_t h) {
621
+ return recv(h, 0);
622
+ }
623
+
624
+ }; // detail_impl<Policy>
625
+
626
+ template <typename Flag>
627
+ using policy_t = ipc::policy::choose<ipc::circ::elem_array, Flag>;
628
+
629
+ } // internal-linkage
630
+
631
+ namespace ipc {
632
+
633
+ template <typename Flag>
634
+ ipc::handle_t chan_impl<Flag>::inited() {
635
+ ipc::detail::waiter::init();
636
+ return nullptr;
637
+ }
638
+
639
+ template <typename Flag>
640
+ bool chan_impl<Flag>::connect(ipc::handle_t * ph, char const * name, unsigned mode) {
641
+ return detail_impl<policy_t<Flag>>::connect(ph, name, mode & receiver);
642
+ }
643
+
644
+ template <typename Flag>
645
+ bool chan_impl<Flag>::reconnect(ipc::handle_t * ph, unsigned mode) {
646
+ return detail_impl<policy_t<Flag>>::reconnect(ph, mode & receiver);
647
+ }
648
+
649
+ template <typename Flag>
650
+ void chan_impl<Flag>::disconnect(ipc::handle_t h) {
651
+ detail_impl<policy_t<Flag>>::disconnect(h);
652
+ }
653
+
654
+ template <typename Flag>
655
+ void chan_impl<Flag>::destroy(ipc::handle_t h) {
656
+ detail_impl<policy_t<Flag>>::destroy(h);
657
+ }
658
+
659
+ template <typename Flag>
660
+ char const * chan_impl<Flag>::name(ipc::handle_t h) {
661
+ auto info = detail_impl<policy_t<Flag>>::info_of(h);
662
+ return (info == nullptr) ? nullptr : info->name_.c_str();
663
+ }
664
+
665
+ template <typename Flag>
666
+ std::size_t chan_impl<Flag>::recv_count(ipc::handle_t h) {
667
+ return detail_impl<policy_t<Flag>>::recv_count(h);
668
+ }
669
+
670
+ template <typename Flag>
671
+ bool chan_impl<Flag>::wait_for_recv(ipc::handle_t h, std::size_t r_count, std::uint64_t tm) {
672
+ return detail_impl<policy_t<Flag>>::wait_for_recv(h, r_count, tm);
673
+ }
674
+
675
+ template <typename Flag>
676
+ bool chan_impl<Flag>::send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
677
+ return detail_impl<policy_t<Flag>>::send(h, data, size, tm);
678
+ }
679
+
680
+ template <typename Flag>
681
+ buff_t chan_impl<Flag>::recv(ipc::handle_t h, std::uint64_t tm) {
682
+ return detail_impl<policy_t<Flag>>::recv(h, tm);
683
+ }
684
+
685
+ template <typename Flag>
686
+ bool chan_impl<Flag>::try_send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
687
+ return detail_impl<policy_t<Flag>>::try_send(h, data, size, tm);
688
+ }
689
+
690
+ template <typename Flag>
691
+ buff_t chan_impl<Flag>::try_recv(ipc::handle_t h) {
692
+ return detail_impl<policy_t<Flag>>::try_recv(h);
693
+ }
694
+
695
+ template struct chan_impl<ipc::wr<relat::single, relat::single, trans::unicast >>;
696
+ // template struct chan_impl<ipc::wr<relat::single, relat::multi , trans::unicast >>; // TBD
697
+ // template struct chan_impl<ipc::wr<relat::multi , relat::multi , trans::unicast >>; // TBD
698
+ template struct chan_impl<ipc::wr<relat::single, relat::multi , trans::broadcast>>;
699
+ template struct chan_impl<ipc::wr<relat::multi , relat::multi , trans::broadcast>>;
700
+
701
+ } // namespace ipc
crazy_functions/test_project/cpp/cppipc/policy.h ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #include <type_traits>
4
+
5
+ #include "libipc/def.h"
6
+ #include "libipc/prod_cons.h"
7
+
8
+ #include "libipc/circ/elem_array.h"
9
+
10
+ namespace ipc {
11
+ namespace policy {
12
+
13
+ template <template <typename, std::size_t...> class Elems, typename Flag>
14
+ struct choose;
15
+
16
+ template <typename Flag>
17
+ struct choose<circ::elem_array, Flag> {
18
+ using flag_t = Flag;
19
+
20
+ template <std::size_t DataSize, std::size_t AlignSize>
21
+ using elems_t = circ::elem_array<ipc::prod_cons_impl<flag_t>, DataSize, AlignSize>;
22
+ };
23
+
24
+ } // namespace policy
25
+ } // namespace ipc
crazy_functions/test_project/cpp/cppipc/pool_alloc.cpp ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #include "libipc/pool_alloc.h"
2
+
3
+ #include "libipc/memory/resource.h"
4
+
5
+ namespace ipc {
6
+ namespace mem {
7
+
8
+ void* pool_alloc::alloc(std::size_t size) {
9
+ return async_pool_alloc::alloc(size);
10
+ }
11
+
12
+ void pool_alloc::free(void* p, std::size_t size) {
13
+ async_pool_alloc::free(p, size);
14
+ }
15
+
16
+ } // namespace mem
17
+ } // namespace ipc
crazy_functions/test_project/cpp/cppipc/prod_cons.h ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #include <atomic>
4
+ #include <utility>
5
+ #include <cstring>
6
+ #include <type_traits>
7
+ #include <cstdint>
8
+
9
+ #include "libipc/def.h"
10
+
11
+ #include "libipc/platform/detail.h"
12
+ #include "libipc/circ/elem_def.h"
13
+ #include "libipc/utility/log.h"
14
+ #include "libipc/utility/utility.h"
15
+
16
+ namespace ipc {
17
+
18
+ ////////////////////////////////////////////////////////////////
19
+ /// producer-consumer implementation
20
+ ////////////////////////////////////////////////////////////////
21
+
22
+ template <typename Flag>
23
+ struct prod_cons_impl;
24
+
25
+ template <>
26
+ struct prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
27
+
28
+ template <std::size_t DataSize, std::size_t AlignSize>
29
+ struct elem_t {
30
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
31
+ };
32
+
33
+ alignas(cache_line_size) std::atomic<circ::u2_t> rd_; // read index
34
+ alignas(cache_line_size) std::atomic<circ::u2_t> wt_; // write index
35
+
36
+ constexpr circ::u2_t cursor() const noexcept {
37
+ return 0;
38
+ }
39
+
40
+ template <typename W, typename F, typename E>
41
+ bool push(W* /*wrapper*/, F&& f, E* elems) {
42
+ auto cur_wt = circ::index_of(wt_.load(std::memory_order_relaxed));
43
+ if (cur_wt == circ::index_of(rd_.load(std::memory_order_acquire) - 1)) {
44
+ return false; // full
45
+ }
46
+ std::forward<F>(f)(&(elems[cur_wt].data_));
47
+ wt_.fetch_add(1, std::memory_order_release);
48
+ return true;
49
+ }
50
+
51
+ /**
52
+ * In single-single-unicast, 'force_push' means 'no reader' or 'the only one reader is dead'.
53
+ * So we could just disconnect all connections of receiver, and return false.
54
+ */
55
+ template <typename W, typename F, typename E>
56
+ bool force_push(W* wrapper, F&&, E*) {
57
+ wrapper->elems()->disconnect_receiver(~static_cast<circ::cc_t>(0u));
58
+ return false;
59
+ }
60
+
61
+ template <typename W, typename F, typename R, typename E>
62
+ bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E* elems) {
63
+ auto cur_rd = circ::index_of(rd_.load(std::memory_order_relaxed));
64
+ if (cur_rd == circ::index_of(wt_.load(std::memory_order_acquire))) {
65
+ return false; // empty
66
+ }
67
+ std::forward<F>(f)(&(elems[cur_rd].data_));
68
+ std::forward<R>(out)(true);
69
+ rd_.fetch_add(1, std::memory_order_release);
70
+ return true;
71
+ }
72
+ };
73
+
74
+ template <>
75
+ struct prod_cons_impl<wr<relat::single, relat::multi , trans::unicast>>
76
+ : prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
77
+
78
+ template <typename W, typename F, typename E>
79
+ bool force_push(W* wrapper, F&&, E*) {
80
+ wrapper->elems()->disconnect_receiver(1);
81
+ return false;
82
+ }
83
+
84
+ template <typename W, typename F, typename R,
85
+ template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
86
+ bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
87
+ byte_t buff[DS];
88
+ for (unsigned k = 0;;) {
89
+ auto cur_rd = rd_.load(std::memory_order_relaxed);
90
+ if (circ::index_of(cur_rd) ==
91
+ circ::index_of(wt_.load(std::memory_order_acquire))) {
92
+ return false; // empty
93
+ }
94
+ std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
95
+ if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
96
+ std::forward<F>(f)(buff);
97
+ std::forward<R>(out)(true);
98
+ return true;
99
+ }
100
+ ipc::yield(k);
101
+ }
102
+ }
103
+ };
104
+
105
+ template <>
106
+ struct prod_cons_impl<wr<relat::multi , relat::multi, trans::unicast>>
107
+ : prod_cons_impl<wr<relat::single, relat::multi, trans::unicast>> {
108
+
109
+ using flag_t = std::uint64_t;
110
+
111
+ template <std::size_t DataSize, std::size_t AlignSize>
112
+ struct elem_t {
113
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
114
+ std::atomic<flag_t> f_ct_ { 0 }; // commit flag
115
+ };
116
+
117
+ alignas(cache_line_size) std::atomic<circ::u2_t> ct_; // commit index
118
+
119
+ template <typename W, typename F, typename E>
120
+ bool push(W* /*wrapper*/, F&& f, E* elems) {
121
+ circ::u2_t cur_ct, nxt_ct;
122
+ for (unsigned k = 0;;) {
123
+ cur_ct = ct_.load(std::memory_order_relaxed);
124
+ if (circ::index_of(nxt_ct = cur_ct + 1) ==
125
+ circ::index_of(rd_.load(std::memory_order_acquire))) {
126
+ return false; // full
127
+ }
128
+ if (ct_.compare_exchange_weak(cur_ct, nxt_ct, std::memory_order_acq_rel)) {
129
+ break;
130
+ }
131
+ ipc::yield(k);
132
+ }
133
+ auto* el = elems + circ::index_of(cur_ct);
134
+ std::forward<F>(f)(&(el->data_));
135
+ // set flag & try update wt
136
+ el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
137
+ while (1) {
138
+ auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
139
+ if (cur_ct != wt_.load(std::memory_order_relaxed)) {
140
+ return true;
141
+ }
142
+ if ((~cac_ct) != cur_ct) {
143
+ return true;
144
+ }
145
+ if (!el->f_ct_.compare_exchange_strong(cac_ct, 0, std::memory_order_relaxed)) {
146
+ return true;
147
+ }
148
+ wt_.store(nxt_ct, std::memory_order_release);
149
+ cur_ct = nxt_ct;
150
+ nxt_ct = cur_ct + 1;
151
+ el = elems + circ::index_of(cur_ct);
152
+ }
153
+ return true;
154
+ }
155
+
156
+ template <typename W, typename F, typename E>
157
+ bool force_push(W* wrapper, F&&, E*) {
158
+ wrapper->elems()->disconnect_receiver(1);
159
+ return false;
160
+ }
161
+
162
+ template <typename W, typename F, typename R,
163
+ template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
164
+ bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
165
+ byte_t buff[DS];
166
+ for (unsigned k = 0;;) {
167
+ auto cur_rd = rd_.load(std::memory_order_relaxed);
168
+ auto cur_wt = wt_.load(std::memory_order_acquire);
169
+ auto id_rd = circ::index_of(cur_rd);
170
+ auto id_wt = circ::index_of(cur_wt);
171
+ if (id_rd == id_wt) {
172
+ auto* el = elems + id_wt;
173
+ auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
174
+ if ((~cac_ct) != cur_wt) {
175
+ return false; // empty
176
+ }
177
+ if (el->f_ct_.compare_exchange_weak(cac_ct, 0, std::memory_order_relaxed)) {
178
+ wt_.store(cur_wt + 1, std::memory_order_release);
179
+ }
180
+ k = 0;
181
+ }
182
+ else {
183
+ std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
184
+ if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
185
+ std::forward<F>(f)(buff);
186
+ std::forward<R>(out)(true);
187
+ return true;
188
+ }
189
+ ipc::yield(k);
190
+ }
191
+ }
192
+ }
193
+ };
194
+
195
+ template <>
196
+ struct prod_cons_impl<wr<relat::single, relat::multi, trans::broadcast>> {
197
+
198
+ using rc_t = std::uint64_t;
199
+
200
+ enum : rc_t {
201
+ ep_mask = 0x00000000ffffffffull,
202
+ ep_incr = 0x0000000100000000ull
203
+ };
204
+
205
+ template <std::size_t DataSize, std::size_t AlignSize>
206
+ struct elem_t {
207
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
208
+ std::atomic<rc_t> rc_ { 0 }; // read-counter
209
+ };
210
+
211
+ alignas(cache_line_size) std::atomic<circ::u2_t> wt_; // write index
212
+ alignas(cache_line_size) rc_t epoch_ { 0 }; // only one writer
213
+
214
+ circ::u2_t cursor() const noexcept {
215
+ return wt_.load(std::memory_order_acquire);
216
+ }
217
+
218
+ template <typename W, typename F, typename E>
219
+ bool push(W* wrapper, F&& f, E* elems) {
220
+ E* el;
221
+ for (unsigned k = 0;;) {
222
+ circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
223
+ if (cc == 0) return false; // no reader
224
+ el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
225
+ // check all consumers have finished reading this element
226
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
227
+ circ::cc_t rem_cc = cur_rc & ep_mask;
228
+ if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch_)) {
229
+ return false; // has not finished yet
230
+ }
231
+ // consider rem_cc to be 0 here
232
+ if (el->rc_.compare_exchange_weak(
233
+ cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
234
+ break;
235
+ }
236
+ ipc::yield(k);
237
+ }
238
+ std::forward<F>(f)(&(el->data_));
239
+ wt_.fetch_add(1, std::memory_order_release);
240
+ return true;
241
+ }
242
+
243
+ template <typename W, typename F, typename E>
244
+ bool force_push(W* wrapper, F&& f, E* elems) {
245
+ E* el;
246
+ epoch_ += ep_incr;
247
+ for (unsigned k = 0;;) {
248
+ circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
249
+ if (cc == 0) return false; // no reader
250
+ el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
251
+ // check all consumers have finished reading this element
252
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
253
+ circ::cc_t rem_cc = cur_rc & ep_mask;
254
+ if (cc & rem_cc) {
255
+ ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
256
+ cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
257
+ if (cc == 0) return false; // no reader
258
+ }
259
+ // just compare & exchange
260
+ if (el->rc_.compare_exchange_weak(
261
+ cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
262
+ break;
263
+ }
264
+ ipc::yield(k);
265
+ }
266
+ std::forward<F>(f)(&(el->data_));
267
+ wt_.fetch_add(1, std::memory_order_release);
268
+ return true;
269
+ }
270
+
271
+ template <typename W, typename F, typename R, typename E>
272
+ bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E* elems) {
273
+ if (cur == cursor()) return false; // acquire
274
+ auto* el = elems + circ::index_of(cur++);
275
+ std::forward<F>(f)(&(el->data_));
276
+ for (unsigned k = 0;;) {
277
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
278
+ if ((cur_rc & ep_mask) == 0) {
279
+ std::forward<R>(out)(true);
280
+ return true;
281
+ }
282
+ auto nxt_rc = cur_rc & ~static_cast<rc_t>(wrapper->connected_id());
283
+ if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
284
+ std::forward<R>(out)((nxt_rc & ep_mask) == 0);
285
+ return true;
286
+ }
287
+ ipc::yield(k);
288
+ }
289
+ }
290
+ };
291
+
292
+ template <>
293
+ struct prod_cons_impl<wr<relat::multi, relat::multi, trans::broadcast>> {
294
+
295
+ using rc_t = std::uint64_t;
296
+ using flag_t = std::uint64_t;
297
+
298
+ enum : rc_t {
299
+ rc_mask = 0x00000000ffffffffull,
300
+ ep_mask = 0x00ffffffffffffffull,
301
+ ep_incr = 0x0100000000000000ull,
302
+ ic_mask = 0xff000000ffffffffull,
303
+ ic_incr = 0x0000000100000000ull
304
+ };
305
+
306
+ template <std::size_t DataSize, std::size_t AlignSize>
307
+ struct elem_t {
308
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
309
+ std::atomic<rc_t > rc_ { 0 }; // read-counter
310
+ std::atomic<flag_t> f_ct_ { 0 }; // commit flag
311
+ };
312
+
313
+ alignas(cache_line_size) std::atomic<circ::u2_t> ct_; // commit index
314
+ alignas(cache_line_size) std::atomic<rc_t> epoch_ { 0 };
315
+
316
+ circ::u2_t cursor() const noexcept {
317
+ return ct_.load(std::memory_order_acquire);
318
+ }
319
+
320
+ constexpr static rc_t inc_rc(rc_t rc) noexcept {
321
+ return (rc & ic_mask) | ((rc + ic_incr) & ~ic_mask);
322
+ }
323
+
324
+ constexpr static rc_t inc_mask(rc_t rc) noexcept {
325
+ return inc_rc(rc) & ~rc_mask;
326
+ }
327
+
328
+ template <typename W, typename F, typename E>
329
+ bool push(W* wrapper, F&& f, E* elems) {
330
+ E* el;
331
+ circ::u2_t cur_ct;
332
+ rc_t epoch = epoch_.load(std::memory_order_acquire);
333
+ for (unsigned k = 0;;) {
334
+ circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
335
+ if (cc == 0) return false; // no reader
336
+ el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
337
+ // check all consumers have finished reading this element
338
+ auto cur_rc = el->rc_.load(std::memory_order_relaxed);
339
+ circ::cc_t rem_cc = cur_rc & rc_mask;
340
+ if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch)) {
341
+ return false; // has not finished yet
342
+ }
343
+ else if (!rem_cc) {
344
+ auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
345
+ if ((cur_fl != cur_ct) && cur_fl) {
346
+ return false; // full
347
+ }
348
+ }
349
+ // consider rem_cc to be 0 here
350
+ if (el->rc_.compare_exchange_weak(
351
+ cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed) &&
352
+ epoch_.compare_exchange_weak(epoch, epoch, std::memory_order_acq_rel)) {
353
+ break;
354
+ }
355
+ ipc::yield(k);
356
+ }
357
+ // only one thread/process would touch here at one time
358
+ ct_.store(cur_ct + 1, std::memory_order_release);
359
+ std::forward<F>(f)(&(el->data_));
360
+ // set flag & try update wt
361
+ el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
362
+ return true;
363
+ }
364
+
365
+ template <typename W, typename F, typename E>
366
+ bool force_push(W* wrapper, F&& f, E* elems) {
367
+ E* el;
368
+ circ::u2_t cur_ct;
369
+ rc_t epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
370
+ for (unsigned k = 0;;) {
371
+ circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
372
+ if (cc == 0) return false; // no reader
373
+ el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
374
+ // check all consumers have finished reading this element
375
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
376
+ circ::cc_t rem_cc = cur_rc & rc_mask;
377
+ if (cc & rem_cc) {
378
+ ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
379
+ cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
380
+ if (cc == 0) return false; // no reader
381
+ }
382
+ // just compare & exchange
383
+ if (el->rc_.compare_exchange_weak(
384
+ cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed)) {
385
+ if (epoch == epoch_.load(std::memory_order_acquire)) {
386
+ break;
387
+ }
388
+ else if (push(wrapper, std::forward<F>(f), elems)) {
389
+ return true;
390
+ }
391
+ epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
392
+ }
393
+ ipc::yield(k);
394
+ }
395
+ // only one thread/process would touch here at one time
396
+ ct_.store(cur_ct + 1, std::memory_order_release);
397
+ std::forward<F>(f)(&(el->data_));
398
+ // set flag & try update wt
399
+ el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
400
+ return true;
401
+ }
402
+
403
+ template <typename W, typename F, typename R, typename E, std::size_t N>
404
+ bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E(& elems)[N]) {
405
+ auto* el = elems + circ::index_of(cur);
406
+ auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
407
+ if (cur_fl != ~static_cast<flag_t>(cur)) {
408
+ return false; // empty
409
+ }
410
+ ++cur;
411
+ std::forward<F>(f)(&(el->data_));
412
+ for (unsigned k = 0;;) {
413
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
414
+ if ((cur_rc & rc_mask) == 0) {
415
+ std::forward<R>(out)(true);
416
+ el->f_ct_.store(cur + N - 1, std::memory_order_release);
417
+ return true;
418
+ }
419
+ auto nxt_rc = inc_rc(cur_rc) & ~static_cast<rc_t>(wrapper->connected_id());
420
+ bool last_one = false;
421
+ if ((last_one = (nxt_rc & rc_mask) == 0)) {
422
+ el->f_ct_.store(cur + N - 1, std::memory_order_release);
423
+ }
424
+ if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
425
+ std::forward<R>(out)(last_one);
426
+ return true;
427
+ }
428
+ ipc::yield(k);
429
+ }
430
+ }
431
+ };
432
+
433
+ } // namespace ipc
crazy_functions/test_project/cpp/cppipc/queue.h ADDED
@@ -0,0 +1,216 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #include <type_traits>
4
+ #include <new>
5
+ #include <utility> // [[since C++14]]: std::exchange
6
+ #include <algorithm>
7
+ #include <atomic>
8
+ #include <tuple>
9
+ #include <thread>
10
+ #include <chrono>
11
+ #include <string>
12
+ #include <cassert> // assert
13
+
14
+ #include "libipc/def.h"
15
+ #include "libipc/shm.h"
16
+ #include "libipc/rw_lock.h"
17
+
18
+ #include "libipc/utility/log.h"
19
+ #include "libipc/platform/detail.h"
20
+ #include "libipc/circ/elem_def.h"
21
+
22
+ namespace ipc {
23
+ namespace detail {
24
+
25
+ class queue_conn {
26
+ protected:
27
+ circ::cc_t connected_ = 0;
28
+ shm::handle elems_h_;
29
+
30
+ template <typename Elems>
31
+ Elems* open(char const * name) {
32
+ if (name == nullptr || name[0] == '\0') {
33
+ ipc::error("fail open waiter: name is empty!\n");
34
+ return nullptr;
35
+ }
36
+ if (!elems_h_.acquire(name, sizeof(Elems))) {
37
+ return nullptr;
38
+ }
39
+ auto elems = static_cast<Elems*>(elems_h_.get());
40
+ if (elems == nullptr) {
41
+ ipc::error("fail acquire elems: %s\n", name);
42
+ return nullptr;
43
+ }
44
+ elems->init();
45
+ return elems;
46
+ }
47
+
48
+ void close() {
49
+ elems_h_.release();
50
+ }
51
+
52
+ public:
53
+ queue_conn() = default;
54
+ queue_conn(const queue_conn&) = delete;
55
+ queue_conn& operator=(const queue_conn&) = delete;
56
+
57
+ bool connected() const noexcept {
58
+ return connected_ != 0;
59
+ }
60
+
61
+ circ::cc_t connected_id() const noexcept {
62
+ return connected_;
63
+ }
64
+
65
+ template <typename Elems>
66
+ auto connect(Elems* elems) noexcept
67
+ /*needs 'optional' here*/
68
+ -> std::tuple<bool, bool, decltype(std::declval<Elems>().cursor())> {
69
+ if (elems == nullptr) return {};
70
+ // if it's already connected, just return
71
+ if (connected()) return {connected(), false, 0};
72
+ connected_ = elems->connect_receiver();
73
+ return {connected(), true, elems->cursor()};
74
+ }
75
+
76
+ template <typename Elems>
77
+ bool disconnect(Elems* elems) noexcept {
78
+ if (elems == nullptr) return false;
79
+ // if it's already disconnected, just return false
80
+ if (!connected()) return false;
81
+ elems->disconnect_receiver(std::exchange(connected_, 0));
82
+ return true;
83
+ }
84
+ };
85
+
86
+ template <typename Elems>
87
+ class queue_base : public queue_conn {
88
+ using base_t = queue_conn;
89
+
90
+ public:
91
+ using elems_t = Elems;
92
+ using policy_t = typename elems_t::policy_t;
93
+
94
+ protected:
95
+ elems_t * elems_ = nullptr;
96
+ decltype(std::declval<elems_t>().cursor()) cursor_ = 0;
97
+ bool sender_flag_ = false;
98
+
99
+ public:
100
+ using base_t::base_t;
101
+
102
+ queue_base() = default;
103
+
104
+ explicit queue_base(char const * name)
105
+ : queue_base{} {
106
+ elems_ = open<elems_t>(name);
107
+ }
108
+
109
+ explicit queue_base(elems_t * elems) noexcept
110
+ : queue_base{} {
111
+ assert(elems != nullptr);
112
+ elems_ = elems;
113
+ }
114
+
115
+ /* not virtual */ ~queue_base() {
116
+ base_t::close();
117
+ }
118
+
119
+ elems_t * elems() noexcept { return elems_; }
120
+ elems_t const * elems() const noexcept { return elems_; }
121
+
122
+ bool ready_sending() noexcept {
123
+ if (elems_ == nullptr) return false;
124
+ return sender_flag_ || (sender_flag_ = elems_->connect_sender());
125
+ }
126
+
127
+ void shut_sending() noexcept {
128
+ if (elems_ == nullptr) return;
129
+ if (!sender_flag_) return;
130
+ elems_->disconnect_sender();
131
+ }
132
+
133
+ bool connect() noexcept {
134
+ auto tp = base_t::connect(elems_);
135
+ if (std::get<0>(tp) && std::get<1>(tp)) {
136
+ cursor_ = std::get<2>(tp);
137
+ return true;
138
+ }
139
+ return std::get<0>(tp);
140
+ }
141
+
142
+ bool disconnect() noexcept {
143
+ return base_t::disconnect(elems_);
144
+ }
145
+
146
+ std::size_t conn_count() const noexcept {
147
+ return (elems_ == nullptr) ? static_cast<std::size_t>(invalid_value) : elems_->conn_count();
148
+ }
149
+
150
+ bool valid() const noexcept {
151
+ return elems_ != nullptr;
152
+ }
153
+
154
+ bool empty() const noexcept {
155
+ return !valid() || (cursor_ == elems_->cursor());
156
+ }
157
+
158
+ template <typename T, typename F, typename... P>
159
+ bool push(F&& prep, P&&... params) {
160
+ if (elems_ == nullptr) return false;
161
+ return elems_->push(this, [&](void* p) {
162
+ if (prep(p)) ::new (p) T(std::forward<P>(params)...);
163
+ });
164
+ }
165
+
166
+ template <typename T, typename F, typename... P>
167
+ bool force_push(F&& prep, P&&... params) {
168
+ if (elems_ == nullptr) return false;
169
+ return elems_->force_push(this, [&](void* p) {
170
+ if (prep(p)) ::new (p) T(std::forward<P>(params)...);
171
+ });
172
+ }
173
+
174
+ template <typename T, typename F>
175
+ bool pop(T& item, F&& out) {
176
+ if (elems_ == nullptr) {
177
+ return false;
178
+ }
179
+ return elems_->pop(this, &(this->cursor_), [&item](void* p) {
180
+ ::new (&item) T(std::move(*static_cast<T*>(p)));
181
+ }, std::forward<F>(out));
182
+ }
183
+ };
184
+
185
+ } // namespace detail
186
+
187
+ template <typename T, typename Policy>
188
+ class queue final : public detail::queue_base<typename Policy::template elems_t<sizeof(T), alignof(T)>> {
189
+ using base_t = detail::queue_base<typename Policy::template elems_t<sizeof(T), alignof(T)>>;
190
+
191
+ public:
192
+ using value_t = T;
193
+
194
+ using base_t::base_t;
195
+
196
+ template <typename... P>
197
+ bool push(P&&... params) {
198
+ return base_t::template push<T>(std::forward<P>(params)...);
199
+ }
200
+
201
+ template <typename... P>
202
+ bool force_push(P&&... params) {
203
+ return base_t::template force_push<T>(std::forward<P>(params)...);
204
+ }
205
+
206
+ bool pop(T& item) {
207
+ return base_t::pop(item, [](bool) {});
208
+ }
209
+
210
+ template <typename F>
211
+ bool pop(T& item, F&& out) {
212
+ return base_t::pop(item, std::forward<F>(out));
213
+ }
214
+ };
215
+
216
+ } // namespace ipc
crazy_functions/test_project/cpp/cppipc/shm.cpp ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ #include <string>
3
+ #include <utility>
4
+
5
+ #include "libipc/shm.h"
6
+
7
+ #include "libipc/utility/pimpl.h"
8
+ #include "libipc/memory/resource.h"
9
+
10
+ namespace ipc {
11
+ namespace shm {
12
+
13
+ class handle::handle_ : public pimpl<handle_> {
14
+ public:
15
+ shm::id_t id_ = nullptr;
16
+ void* m_ = nullptr;
17
+
18
+ ipc::string n_;
19
+ std::size_t s_ = 0;
20
+ };
21
+
22
+ handle::handle()
23
+ : p_(p_->make()) {
24
+ }
25
+
26
+ handle::handle(char const * name, std::size_t size, unsigned mode)
27
+ : handle() {
28
+ acquire(name, size, mode);
29
+ }
30
+
31
+ handle::handle(handle&& rhs)
32
+ : handle() {
33
+ swap(rhs);
34
+ }
35
+
36
+ handle::~handle() {
37
+ release();
38
+ p_->clear();
39
+ }
40
+
41
+ void handle::swap(handle& rhs) {
42
+ std::swap(p_, rhs.p_);
43
+ }
44
+
45
+ handle& handle::operator=(handle rhs) {
46
+ swap(rhs);
47
+ return *this;
48
+ }
49
+
50
+ bool handle::valid() const noexcept {
51
+ return impl(p_)->m_ != nullptr;
52
+ }
53
+
54
+ std::size_t handle::size() const noexcept {
55
+ return impl(p_)->s_;
56
+ }
57
+
58
+ char const * handle::name() const noexcept {
59
+ return impl(p_)->n_.c_str();
60
+ }
61
+
62
+ std::int32_t handle::ref() const noexcept {
63
+ return shm::get_ref(impl(p_)->id_);
64
+ }
65
+
66
+ void handle::sub_ref() noexcept {
67
+ shm::sub_ref(impl(p_)->id_);
68
+ }
69
+
70
+ bool handle::acquire(char const * name, std::size_t size, unsigned mode) {
71
+ release();
72
+ impl(p_)->id_ = shm::acquire((impl(p_)->n_ = name).c_str(), size, mode);
73
+ impl(p_)->m_ = shm::get_mem(impl(p_)->id_, &(impl(p_)->s_));
74
+ return valid();
75
+ }
76
+
77
+ std::int32_t handle::release() {
78
+ if (impl(p_)->id_ == nullptr) return -1;
79
+ return shm::release(detach());
80
+ }
81
+
82
+ void* handle::get() const {
83
+ return impl(p_)->m_;
84
+ }
85
+
86
+ void handle::attach(id_t id) {
87
+ if (id == nullptr) return;
88
+ release();
89
+ impl(p_)->id_ = id;
90
+ impl(p_)->m_ = shm::get_mem(impl(p_)->id_, &(impl(p_)->s_));
91
+ }
92
+
93
+ id_t handle::detach() {
94
+ auto old = impl(p_)->id_;
95
+ impl(p_)->id_ = nullptr;
96
+ impl(p_)->m_ = nullptr;
97
+ impl(p_)->s_ = 0;
98
+ impl(p_)->n_.clear();
99
+ return old;
100
+ }
101
+
102
+ } // namespace shm
103
+ } // namespace ipc
crazy_functions/test_project/cpp/cppipc/waiter.h ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #include <utility>
4
+ #include <string>
5
+ #include <mutex>
6
+ #include <atomic>
7
+
8
+ #include "libipc/def.h"
9
+ #include "libipc/mutex.h"
10
+ #include "libipc/condition.h"
11
+ #include "libipc/platform/detail.h"
12
+
13
+ namespace ipc {
14
+ namespace detail {
15
+
16
+ class waiter {
17
+ ipc::sync::condition cond_;
18
+ ipc::sync::mutex lock_;
19
+ std::atomic<bool> quit_ {false};
20
+
21
+ public:
22
+ static void init();
23
+
24
+ waiter() = default;
25
+ waiter(char const *name) {
26
+ open(name);
27
+ }
28
+
29
+ ~waiter() {
30
+ close();
31
+ }
32
+
33
+ bool valid() const noexcept {
34
+ return cond_.valid() && lock_.valid();
35
+ }
36
+
37
+ bool open(char const *name) noexcept {
38
+ quit_.store(false, std::memory_order_relaxed);
39
+ if (!cond_.open((std::string{"_waiter_cond_"} + name).c_str())) {
40
+ return false;
41
+ }
42
+ if (!lock_.open((std::string{"_waiter_lock_"} + name).c_str())) {
43
+ cond_.close();
44
+ return false;
45
+ }
46
+ return valid();
47
+ }
48
+
49
+ void close() noexcept {
50
+ cond_.close();
51
+ lock_.close();
52
+ }
53
+
54
+ template <typename F>
55
+ bool wait_if(F &&pred, std::uint64_t tm = ipc::invalid_value) noexcept {
56
+ IPC_UNUSED_ std::lock_guard<ipc::sync::mutex> guard {lock_};
57
+ while ([this, &pred] {
58
+ return !quit_.load(std::memory_order_relaxed)
59
+ && std::forward<F>(pred)();
60
+ }()) {
61
+ if (!cond_.wait(lock_, tm)) return false;
62
+ }
63
+ return true;
64
+ }
65
+
66
+ bool notify() noexcept {
67
+ std::lock_guard<ipc::sync::mutex>{lock_}; // barrier
68
+ return cond_.notify(lock_);
69
+ }
70
+
71
+ bool broadcast() noexcept {
72
+ std::lock_guard<ipc::sync::mutex>{lock_}; // barrier
73
+ return cond_.broadcast(lock_);
74
+ }
75
+
76
+ bool quit_waiting() {
77
+ quit_.store(true, std::memory_order_release);
78
+ return broadcast();
79
+ }
80
+ };
81
+
82
+ } // namespace detail
83
+ } // namespace ipc
crazy_functions/test_project/cpp/cppipc/来源 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ https://github.com/mutouyun/cpp-ipc
2
+
3
+ A high-performance inter-process communication library using shared memory on Linux/Windows.
crazy_functions/test_project/cpp/libJPG/jpgd.cpp ADDED
@@ -0,0 +1,3276 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ // jpgd.cpp - C++ class for JPEG decompression.
2
+ // Public domain, Rich Geldreich <richgel99@gmail.com>
3
+ // Last updated Apr. 16, 2011
4
+ // Alex Evans: Linear memory allocator (taken from jpge.h).
5
+ //
6
+ // Supports progressive and baseline sequential JPEG image files, and the most common chroma subsampling factors: Y, H1V1, H2V1, H1V2, and H2V2.
7
+ //
8
+ // Chroma upsampling quality: H2V2 is upsampled in the frequency domain, H2V1 and H1V2 are upsampled using point sampling.
9
+ // Chroma upsampling reference: "Fast Scheme for Image Size Change in the Compressed Domain"
10
+ // http://vision.ai.uiuc.edu/~dugad/research/dct/index.html
11
+
12
+ #include "jpgd.h"
13
+ #include <string.h>
14
+
15
+ #include <assert.h>
16
+ // BEGIN EPIC MOD
17
+ #define JPGD_ASSERT(x) { assert(x); CA_ASSUME(x); } (void)0
18
+ // END EPIC MOD
19
+
20
+ #ifdef _MSC_VER
21
+ #pragma warning (disable : 4611) // warning C4611: interaction between '_setjmp' and C++ object destruction is non-portable
22
+ #endif
23
+
24
+ // Set to 1 to enable freq. domain chroma upsampling on images using H2V2 subsampling (0=faster nearest neighbor sampling).
25
+ // This is slower, but results in higher quality on images with highly saturated colors.
26
+ #define JPGD_SUPPORT_FREQ_DOMAIN_UPSAMPLING 1
27
+
28
+ #define JPGD_TRUE (1)
29
+ #define JPGD_FALSE (0)
30
+
31
+ #define JPGD_MAX(a,b) (((a)>(b)) ? (a) : (b))
32
+ #define JPGD_MIN(a,b) (((a)<(b)) ? (a) : (b))
33
+
34
+ namespace jpgd {
35
+
36
+ static inline void *jpgd_malloc(size_t nSize) { return FMemory::Malloc(nSize); }
37
+ static inline void jpgd_free(void *p) { FMemory::Free(p); }
38
+
39
+ // BEGIN EPIC MOD
40
+ //@UE3 - use UE3 BGRA encoding instead of assuming RGBA
41
+ // stolen from IImageWrapper.h
42
+ enum ERGBFormatJPG
43
+ {
44
+ Invalid = -1,
45
+ RGBA = 0,
46
+ BGRA = 1,
47
+ Gray = 2,
48
+ };
49
+ static ERGBFormatJPG jpg_format;
50
+ // END EPIC MOD
51
+
52
+ // DCT coefficients are stored in this sequence.
53
+ static int g_ZAG[64] = { 0,1,8,16,9,2,3,10,17,24,32,25,18,11,4,5,12,19,26,33,40,48,41,34,27,20,13,6,7,14,21,28,35,42,49,56,57,50,43,36,29,22,15,23,30,37,44,51,58,59,52,45,38,31,39,46,53,60,61,54,47,55,62,63 };
54
+
55
+ enum JPEG_MARKER
56
+ {
57
+ M_SOF0 = 0xC0, M_SOF1 = 0xC1, M_SOF2 = 0xC2, M_SOF3 = 0xC3, M_SOF5 = 0xC5, M_SOF6 = 0xC6, M_SOF7 = 0xC7, M_JPG = 0xC8,
58
+ M_SOF9 = 0xC9, M_SOF10 = 0xCA, M_SOF11 = 0xCB, M_SOF13 = 0xCD, M_SOF14 = 0xCE, M_SOF15 = 0xCF, M_DHT = 0xC4, M_DAC = 0xCC,
59
+ M_RST0 = 0xD0, M_RST1 = 0xD1, M_RST2 = 0xD2, M_RST3 = 0xD3, M_RST4 = 0xD4, M_RST5 = 0xD5, M_RST6 = 0xD6, M_RST7 = 0xD7,
60
+ M_SOI = 0xD8, M_EOI = 0xD9, M_SOS = 0xDA, M_DQT = 0xDB, M_DNL = 0xDC, M_DRI = 0xDD, M_DHP = 0xDE, M_EXP = 0xDF,
61
+ M_APP0 = 0xE0, M_APP15 = 0xEF, M_JPG0 = 0xF0, M_JPG13 = 0xFD, M_COM = 0xFE, M_TEM = 0x01, M_ERROR = 0x100, RST0 = 0xD0
62
+ };
63
+
64
+ enum JPEG_SUBSAMPLING { JPGD_GRAYSCALE = 0, JPGD_YH1V1, JPGD_YH2V1, JPGD_YH1V2, JPGD_YH2V2 };
65
+
66
+ #define CONST_BITS 13
67
+ #define PASS1_BITS 2
68
+ #define SCALEDONE ((int32)1)
69
+
70
+ #define FIX_0_298631336 ((int32)2446) /* FIX(0.298631336) */
71
+ #define FIX_0_390180644 ((int32)3196) /* FIX(0.390180644) */
72
+ #define FIX_0_541196100 ((int32)4433) /* FIX(0.541196100) */
73
+ #define FIX_0_765366865 ((int32)6270) /* FIX(0.765366865) */
74
+ #define FIX_0_899976223 ((int32)7373) /* FIX(0.899976223) */
75
+ #define FIX_1_175875602 ((int32)9633) /* FIX(1.175875602) */
76
+ #define FIX_1_501321110 ((int32)12299) /* FIX(1.501321110) */
77
+ #define FIX_1_847759065 ((int32)15137) /* FIX(1.847759065) */
78
+ #define FIX_1_961570560 ((int32)16069) /* FIX(1.961570560) */
79
+ #define FIX_2_053119869 ((int32)16819) /* FIX(2.053119869) */
80
+ #define FIX_2_562915447 ((int32)20995) /* FIX(2.562915447) */
81
+ #define FIX_3_072711026 ((int32)25172) /* FIX(3.072711026) */
82
+
83
+ #define DESCALE(x,n) (((x) + (SCALEDONE << ((n)-1))) >> (n))
84
+ #define DESCALE_ZEROSHIFT(x,n) (((x) + (128 << (n)) + (SCALEDONE << ((n)-1))) >> (n))
85
+
86
+ #define MULTIPLY(var, cnst) ((var) * (cnst))
87
+
88
+ #define CLAMP(i) ((static_cast<uint>(i) > 255) ? (((~i) >> 31) & 0xFF) : (i))
89
+
90
+ // Compiler creates a fast path 1D IDCT for X non-zero columns
91
+ template <int NONZERO_COLS>
92
+ struct Row
93
+ {
94
+ static void idct(int* pTemp, const jpgd_block_t* pSrc)
95
+ {
96
+ // ACCESS_COL() will be optimized at compile time to either an array access, or 0.
97
+ #define ACCESS_COL(x) (((x) < NONZERO_COLS) ? (int)pSrc[x] : 0)
98
+
99
+ const int z2 = ACCESS_COL(2), z3 = ACCESS_COL(6);
100
+
101
+ const int z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
102
+ const int tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
103
+ const int tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
104
+
105
+ const int tmp0 = (ACCESS_COL(0) + ACCESS_COL(4)) << CONST_BITS;
106
+ const int tmp1 = (ACCESS_COL(0) - ACCESS_COL(4)) << CONST_BITS;
107
+
108
+ const int tmp10 = tmp0 + tmp3, tmp13 = tmp0 - tmp3, tmp11 = tmp1 + tmp2, tmp12 = tmp1 - tmp2;
109
+
110
+ const int atmp0 = ACCESS_COL(7), atmp1 = ACCESS_COL(5), atmp2 = ACCESS_COL(3), atmp3 = ACCESS_COL(1);
111
+
112
+ const int bz1 = atmp0 + atmp3, bz2 = atmp1 + atmp2, bz3 = atmp0 + atmp2, bz4 = atmp1 + atmp3;
113
+ const int bz5 = MULTIPLY(bz3 + bz4, FIX_1_175875602);
114
+
115
+ const int az1 = MULTIPLY(bz1, - FIX_0_899976223);
116
+ const int az2 = MULTIPLY(bz2, - FIX_2_562915447);
117
+ const int az3 = MULTIPLY(bz3, - FIX_1_961570560) + bz5;
118
+ const int az4 = MULTIPLY(bz4, - FIX_0_390180644) + bz5;
119
+
120
+ const int btmp0 = MULTIPLY(atmp0, FIX_0_298631336) + az1 + az3;
121
+ const int btmp1 = MULTIPLY(atmp1, FIX_2_053119869) + az2 + az4;
122
+ const int btmp2 = MULTIPLY(atmp2, FIX_3_072711026) + az2 + az3;
123
+ const int btmp3 = MULTIPLY(atmp3, FIX_1_501321110) + az1 + az4;
124
+
125
+ pTemp[0] = DESCALE(tmp10 + btmp3, CONST_BITS-PASS1_BITS);
126
+ pTemp[7] = DESCALE(tmp10 - btmp3, CONST_BITS-PASS1_BITS);
127
+ pTemp[1] = DESCALE(tmp11 + btmp2, CONST_BITS-PASS1_BITS);
128
+ pTemp[6] = DESCALE(tmp11 - btmp2, CONST_BITS-PASS1_BITS);
129
+ pTemp[2] = DESCALE(tmp12 + btmp1, CONST_BITS-PASS1_BITS);
130
+ pTemp[5] = DESCALE(tmp12 - btmp1, CONST_BITS-PASS1_BITS);
131
+ pTemp[3] = DESCALE(tmp13 + btmp0, CONST_BITS-PASS1_BITS);
132
+ pTemp[4] = DESCALE(tmp13 - btmp0, CONST_BITS-PASS1_BITS);
133
+ }
134
+ };
135
+
136
+ template <>
137
+ struct Row<0>
138
+ {
139
+ static void idct(int* pTemp, const jpgd_block_t* pSrc)
140
+ {
141
+ #ifdef _MSC_VER
142
+ pTemp; pSrc;
143
+ #endif
144
+ }
145
+ };
146
+
147
+ template <>
148
+ struct Row<1>
149
+ {
150
+ static void idct(int* pTemp, const jpgd_block_t* pSrc)
151
+ {
152
+ const int dcval = (pSrc[0] << PASS1_BITS);
153
+
154
+ pTemp[0] = dcval;
155
+ pTemp[1] = dcval;
156
+ pTemp[2] = dcval;
157
+ pTemp[3] = dcval;
158
+ pTemp[4] = dcval;
159
+ pTemp[5] = dcval;
160
+ pTemp[6] = dcval;
161
+ pTemp[7] = dcval;
162
+ }
163
+ };
164
+
165
+ // Compiler creates a fast path 1D IDCT for X non-zero rows
166
+ template <int NONZERO_ROWS>
167
+ struct Col
168
+ {
169
+ static void idct(uint8* pDst_ptr, const int* pTemp)
170
+ {
171
+ // ACCESS_ROW() will be optimized at compile time to either an array access, or 0.
172
+ #define ACCESS_ROW(x) (((x) < NONZERO_ROWS) ? pTemp[x * 8] : 0)
173
+
174
+ const int z2 = ACCESS_ROW(2);
175
+ const int z3 = ACCESS_ROW(6);
176
+
177
+ const int z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
178
+ const int tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
179
+ const int tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
180
+
181
+ const int tmp0 = (ACCESS_ROW(0) + ACCESS_ROW(4)) << CONST_BITS;
182
+ const int tmp1 = (ACCESS_ROW(0) - ACCESS_ROW(4)) << CONST_BITS;
183
+
184
+ const int tmp10 = tmp0 + tmp3, tmp13 = tmp0 - tmp3, tmp11 = tmp1 + tmp2, tmp12 = tmp1 - tmp2;
185
+
186
+ const int atmp0 = ACCESS_ROW(7), atmp1 = ACCESS_ROW(5), atmp2 = ACCESS_ROW(3), atmp3 = ACCESS_ROW(1);
187
+
188
+ const int bz1 = atmp0 + atmp3, bz2 = atmp1 + atmp2, bz3 = atmp0 + atmp2, bz4 = atmp1 + atmp3;
189
+ const int bz5 = MULTIPLY(bz3 + bz4, FIX_1_175875602);
190
+
191
+ const int az1 = MULTIPLY(bz1, - FIX_0_899976223);
192
+ const int az2 = MULTIPLY(bz2, - FIX_2_562915447);
193
+ const int az3 = MULTIPLY(bz3, - FIX_1_961570560) + bz5;
194
+ const int az4 = MULTIPLY(bz4, - FIX_0_390180644) + bz5;
195
+
196
+ const int btmp0 = MULTIPLY(atmp0, FIX_0_298631336) + az1 + az3;
197
+ const int btmp1 = MULTIPLY(atmp1, FIX_2_053119869) + az2 + az4;
198
+ const int btmp2 = MULTIPLY(atmp2, FIX_3_072711026) + az2 + az3;
199
+ const int btmp3 = MULTIPLY(atmp3, FIX_1_501321110) + az1 + az4;
200
+
201
+ int i = DESCALE_ZEROSHIFT(tmp10 + btmp3, CONST_BITS+PASS1_BITS+3);
202
+ pDst_ptr[8*0] = (uint8)CLAMP(i);
203
+
204
+ i = DESCALE_ZEROSHIFT(tmp10 - btmp3, CONST_BITS+PASS1_BITS+3);
205
+ pDst_ptr[8*7] = (uint8)CLAMP(i);
206
+
207
+ i = DESCALE_ZEROSHIFT(tmp11 + btmp2, CONST_BITS+PASS1_BITS+3);
208
+ pDst_ptr[8*1] = (uint8)CLAMP(i);
209
+
210
+ i = DESCALE_ZEROSHIFT(tmp11 - btmp2, CONST_BITS+PASS1_BITS+3);
211
+ pDst_ptr[8*6] = (uint8)CLAMP(i);
212
+
213
+ i = DESCALE_ZEROSHIFT(tmp12 + btmp1, CONST_BITS+PASS1_BITS+3);
214
+ pDst_ptr[8*2] = (uint8)CLAMP(i);
215
+
216
+ i = DESCALE_ZEROSHIFT(tmp12 - btmp1, CONST_BITS+PASS1_BITS+3);
217
+ pDst_ptr[8*5] = (uint8)CLAMP(i);
218
+
219
+ i = DESCALE_ZEROSHIFT(tmp13 + btmp0, CONST_BITS+PASS1_BITS+3);
220
+ pDst_ptr[8*3] = (uint8)CLAMP(i);
221
+
222
+ i = DESCALE_ZEROSHIFT(tmp13 - btmp0, CONST_BITS+PASS1_BITS+3);
223
+ pDst_ptr[8*4] = (uint8)CLAMP(i);
224
+ }
225
+ };
226
+
227
+ template <>
228
+ struct Col<1>
229
+ {
230
+ static void idct(uint8* pDst_ptr, const int* pTemp)
231
+ {
232
+ int dcval = DESCALE_ZEROSHIFT(pTemp[0], PASS1_BITS+3);
233
+ const uint8 dcval_clamped = (uint8)CLAMP(dcval);
234
+ pDst_ptr[0*8] = dcval_clamped;
235
+ pDst_ptr[1*8] = dcval_clamped;
236
+ pDst_ptr[2*8] = dcval_clamped;
237
+ pDst_ptr[3*8] = dcval_clamped;
238
+ pDst_ptr[4*8] = dcval_clamped;
239
+ pDst_ptr[5*8] = dcval_clamped;
240
+ pDst_ptr[6*8] = dcval_clamped;
241
+ pDst_ptr[7*8] = dcval_clamped;
242
+ }
243
+ };
244
+
245
+ static const uint8 s_idct_row_table[] =
246
+ {
247
+ 1,0,0,0,0,0,0,0, 2,0,0,0,0,0,0,0, 2,1,0,0,0,0,0,0, 2,1,1,0,0,0,0,0, 2,2,1,0,0,0,0,0, 3,2,1,0,0,0,0,0, 4,2,1,0,0,0,0,0, 4,3,1,0,0,0,0,0,
248
+ 4,3,2,0,0,0,0,0, 4,3,2,1,0,0,0,0, 4,3,2,1,1,0,0,0, 4,3,2,2,1,0,0,0, 4,3,3,2,1,0,0,0, 4,4,3,2,1,0,0,0, 5,4,3,2,1,0,0,0, 6,4,3,2,1,0,0,0,
249
+ 6,5,3,2,1,0,0,0, 6,5,4,2,1,0,0,0, 6,5,4,3,1,0,0,0, 6,5,4,3,2,0,0,0, 6,5,4,3,2,1,0,0, 6,5,4,3,2,1,1,0, 6,5,4,3,2,2,1,0, 6,5,4,3,3,2,1,0,
250
+ 6,5,4,4,3,2,1,0, 6,5,5,4,3,2,1,0, 6,6,5,4,3,2,1,0, 7,6,5,4,3,2,1,0, 8,6,5,4,3,2,1,0, 8,7,5,4,3,2,1,0, 8,7,6,4,3,2,1,0, 8,7,6,5,3,2,1,0,
251
+ 8,7,6,5,4,2,1,0, 8,7,6,5,4,3,1,0, 8,7,6,5,4,3,2,0, 8,7,6,5,4,3,2,1, 8,7,6,5,4,3,2,2, 8,7,6,5,4,3,3,2, 8,7,6,5,4,4,3,2, 8,7,6,5,5,4,3,2,
252
+ 8,7,6,6,5,4,3,2, 8,7,7,6,5,4,3,2, 8,8,7,6,5,4,3,2, 8,8,8,6,5,4,3,2, 8,8,8,7,5,4,3,2, 8,8,8,7,6,4,3,2, 8,8,8,7,6,5,3,2, 8,8,8,7,6,5,4,2,
253
+ 8,8,8,7,6,5,4,3, 8,8,8,7,6,5,4,4, 8,8,8,7,6,5,5,4, 8,8,8,7,6,6,5,4, 8,8,8,7,7,6,5,4, 8,8,8,8,7,6,5,4, 8,8,8,8,8,6,5,4, 8,8,8,8,8,7,5,4,
254
+ 8,8,8,8,8,7,6,4, 8,8,8,8,8,7,6,5, 8,8,8,8,8,7,6,6, 8,8,8,8,8,7,7,6, 8,8,8,8,8,8,7,6, 8,8,8,8,8,8,8,6, 8,8,8,8,8,8,8,7, 8,8,8,8,8,8,8,8,
255
+ };
256
+
257
+ static const uint8 s_idct_col_table[] = { 1, 1, 2, 3, 3, 3, 3, 3, 3, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8 };
258
+
259
+ void idct(const jpgd_block_t* pSrc_ptr, uint8* pDst_ptr, int block_max_zag)
260
+ {
261
+ JPGD_ASSERT(block_max_zag >= 1);
262
+ JPGD_ASSERT(block_max_zag <= 64);
263
+
264
+ if (block_max_zag == 1)
265
+ {
266
+ int k = ((pSrc_ptr[0] + 4) >> 3) + 128;
267
+ k = CLAMP(k);
268
+ k = k | (k<<8);
269
+ k = k | (k<<16);
270
+
271
+ for (int i = 8; i > 0; i--)
272
+ {
273
+ *(int*)&pDst_ptr[0] = k;
274
+ *(int*)&pDst_ptr[4] = k;
275
+ pDst_ptr += 8;
276
+ }
277
+ return;
278
+ }
279
+
280
+ int temp[64];
281
+
282
+ const jpgd_block_t* pSrc = pSrc_ptr;
283
+ int* pTemp = temp;
284
+
285
+ const uint8* pRow_tab = &s_idct_row_table[(block_max_zag - 1) * 8];
286
+ int i;
287
+ for (i = 8; i > 0; i--, pRow_tab++)
288
+ {
289
+ switch (*pRow_tab)
290
+ {
291
+ case 0: Row<0>::idct(pTemp, pSrc); break;
292
+ case 1: Row<1>::idct(pTemp, pSrc); break;
293
+ case 2: Row<2>::idct(pTemp, pSrc); break;
294
+ case 3: Row<3>::idct(pTemp, pSrc); break;
295
+ case 4: Row<4>::idct(pTemp, pSrc); break;
296
+ case 5: Row<5>::idct(pTemp, pSrc); break;
297
+ case 6: Row<6>::idct(pTemp, pSrc); break;
298
+ case 7: Row<7>::idct(pTemp, pSrc); break;
299
+ case 8: Row<8>::idct(pTemp, pSrc); break;
300
+ }
301
+
302
+ pSrc += 8;
303
+ pTemp += 8;
304
+ }
305
+
306
+ pTemp = temp;
307
+
308
+ const int nonzero_rows = s_idct_col_table[block_max_zag - 1];
309
+ for (i = 8; i > 0; i--)
310
+ {
311
+ switch (nonzero_rows)
312
+ {
313
+ case 1: Col<1>::idct(pDst_ptr, pTemp); break;
314
+ case 2: Col<2>::idct(pDst_ptr, pTemp); break;
315
+ case 3: Col<3>::idct(pDst_ptr, pTemp); break;
316
+ case 4: Col<4>::idct(pDst_ptr, pTemp); break;
317
+ case 5: Col<5>::idct(pDst_ptr, pTemp); break;
318
+ case 6: Col<6>::idct(pDst_ptr, pTemp); break;
319
+ case 7: Col<7>::idct(pDst_ptr, pTemp); break;
320
+ case 8: Col<8>::idct(pDst_ptr, pTemp); break;
321
+ }
322
+
323
+ pTemp++;
324
+ pDst_ptr++;
325
+ }
326
+ }
327
+
328
+ void idct_4x4(const jpgd_block_t* pSrc_ptr, uint8* pDst_ptr)
329
+ {
330
+ int temp[64];
331
+ int* pTemp = temp;
332
+ const jpgd_block_t* pSrc = pSrc_ptr;
333
+
334
+ for (int i = 4; i > 0; i--)
335
+ {
336
+ Row<4>::idct(pTemp, pSrc);
337
+ pSrc += 8;
338
+ pTemp += 8;
339
+ }
340
+
341
+ pTemp = temp;
342
+ for (int i = 8; i > 0; i--)
343
+ {
344
+ Col<4>::idct(pDst_ptr, pTemp);
345
+ pTemp++;
346
+ pDst_ptr++;
347
+ }
348
+ }
349
+
350
+ // Retrieve one character from the input stream.
351
+ inline uint jpeg_decoder::get_char()
352
+ {
353
+ // Any bytes remaining in buffer?
354
+ if (!m_in_buf_left)
355
+ {
356
+ // Try to get more bytes.
357
+ prep_in_buffer();
358
+ // Still nothing to get?
359
+ if (!m_in_buf_left)
360
+ {
361
+ // Pad the end of the stream with 0xFF 0xD9 (EOI marker)
362
+ int t = m_tem_flag;
363
+ m_tem_flag ^= 1;
364
+ if (t)
365
+ return 0xD9;
366
+ else
367
+ return 0xFF;
368
+ }
369
+ }
370
+
371
+ uint c = *m_pIn_buf_ofs++;
372
+ m_in_buf_left--;
373
+
374
+ return c;
375
+ }
376
+
377
+ // Same as previous method, except can indicate if the character is a pad character or not.
378
+ inline uint jpeg_decoder::get_char(bool *pPadding_flag)
379
+ {
380
+ if (!m_in_buf_left)
381
+ {
382
+ prep_in_buffer();
383
+ if (!m_in_buf_left)
384
+ {
385
+ *pPadding_flag = true;
386
+ int t = m_tem_flag;
387
+ m_tem_flag ^= 1;
388
+ if (t)
389
+ return 0xD9;
390
+ else
391
+ return 0xFF;
392
+ }
393
+ }
394
+
395
+ *pPadding_flag = false;
396
+
397
+ uint c = *m_pIn_buf_ofs++;
398
+ m_in_buf_left--;
399
+
400
+ return c;
401
+ }
402
+
403
+ // Inserts a previously retrieved character back into the input buffer.
404
+ inline void jpeg_decoder::stuff_char(uint8 q)
405
+ {
406
+ *(--m_pIn_buf_ofs) = q;
407
+ m_in_buf_left++;
408
+ }
409
+
410
+ // Retrieves one character from the input stream, but does not read past markers. Will continue to return 0xFF when a marker is encountered.
411
+ inline uint8 jpeg_decoder::get_octet()
412
+ {
413
+ bool padding_flag;
414
+ int c = get_char(&padding_flag);
415
+
416
+ if (c == 0xFF)
417
+ {
418
+ if (padding_flag)
419
+ return 0xFF;
420
+
421
+ c = get_char(&padding_flag);
422
+ if (padding_flag)
423
+ {
424
+ stuff_char(0xFF);
425
+ return 0xFF;
426
+ }
427
+
428
+ if (c == 0x00)
429
+ return 0xFF;
430
+ else
431
+ {
432
+ stuff_char(static_cast<uint8>(c));
433
+ stuff_char(0xFF);
434
+ return 0xFF;
435
+ }
436
+ }
437
+
438
+ return static_cast<uint8>(c);
439
+ }
440
+
441
+ // Retrieves a variable number of bits from the input stream. Does not recognize markers.
442
+ inline uint jpeg_decoder::get_bits(int num_bits)
443
+ {
444
+ if (!num_bits)
445
+ return 0;
446
+
447
+ uint i = m_bit_buf >> (32 - num_bits);
448
+
449
+ if ((m_bits_left -= num_bits) <= 0)
450
+ {
451
+ m_bit_buf <<= (num_bits += m_bits_left);
452
+
453
+ uint c1 = get_char();
454
+ uint c2 = get_char();
455
+ m_bit_buf = (m_bit_buf & 0xFFFF0000) | (c1 << 8) | c2;
456
+
457
+ m_bit_buf <<= -m_bits_left;
458
+
459
+ m_bits_left += 16;
460
+
461
+ JPGD_ASSERT(m_bits_left >= 0);
462
+ }
463
+ else
464
+ m_bit_buf <<= num_bits;
465
+
466
+ return i;
467
+ }
468
+
469
+ // Retrieves a variable number of bits from the input stream. Markers will not be read into the input bit buffer. Instead, an infinite number of all 1's will be returned when a marker is encountered.
470
+ inline uint jpeg_decoder::get_bits_no_markers(int num_bits)
471
+ {
472
+ if (!num_bits)
473
+ return 0;
474
+
475
+ uint i = m_bit_buf >> (32 - num_bits);
476
+
477
+ if ((m_bits_left -= num_bits) <= 0)
478
+ {
479
+ m_bit_buf <<= (num_bits += m_bits_left);
480
+
481
+ if ((m_in_buf_left < 2) || (m_pIn_buf_ofs[0] == 0xFF) || (m_pIn_buf_ofs[1] == 0xFF))
482
+ {
483
+ uint c1 = get_octet();
484
+ uint c2 = get_octet();
485
+ m_bit_buf |= (c1 << 8) | c2;
486
+ }
487
+ else
488
+ {
489
+ m_bit_buf |= ((uint)m_pIn_buf_ofs[0] << 8) | m_pIn_buf_ofs[1];
490
+ m_in_buf_left -= 2;
491
+ m_pIn_buf_ofs += 2;
492
+ }
493
+
494
+ m_bit_buf <<= -m_bits_left;
495
+
496
+ m_bits_left += 16;
497
+
498
+ JPGD_ASSERT(m_bits_left >= 0);
499
+ }
500
+ else
501
+ m_bit_buf <<= num_bits;
502
+
503
+ return i;
504
+ }
505
+
506
+ // Decodes a Huffman encoded symbol.
507
+ inline int jpeg_decoder::huff_decode(huff_tables *pH)
508
+ {
509
+ int symbol;
510
+
511
+ // Check first 8-bits: do we have a complete symbol?
512
+ if ((symbol = pH->look_up[m_bit_buf >> 24]) < 0)
513
+ {
514
+ // Decode more bits, use a tree traversal to find symbol.
515
+ int ofs = 23;
516
+ do
517
+ {
518
+ symbol = pH->tree[-(int)(symbol + ((m_bit_buf >> ofs) & 1))];
519
+ ofs--;
520
+ } while (symbol < 0);
521
+
522
+ get_bits_no_markers(8 + (23 - ofs));
523
+ }
524
+ else
525
+ get_bits_no_markers(pH->code_size[symbol]);
526
+
527
+ return symbol;
528
+ }
529
+
530
+ // Decodes a Huffman encoded symbol.
531
+ inline int jpeg_decoder::huff_decode(huff_tables *pH, int& extra_bits)
532
+ {
533
+ int symbol;
534
+
535
+ // Check first 8-bits: do we have a complete symbol?
536
+ if ((symbol = pH->look_up2[m_bit_buf >> 24]) < 0)
537
+ {
538
+ // Use a tree traversal to find symbol.
539
+ int ofs = 23;
540
+ do
541
+ {
542
+ symbol = pH->tree[-(int)(symbol + ((m_bit_buf >> ofs) & 1))];
543
+ ofs--;
544
+ } while (symbol < 0);
545
+
546
+ get_bits_no_markers(8 + (23 - ofs));
547
+
548
+ extra_bits = get_bits_no_markers(symbol & 0xF);
549
+ }
550
+ else
551
+ {
552
+ JPGD_ASSERT(((symbol >> 8) & 31) == pH->code_size[symbol & 255] + ((symbol & 0x8000) ? (symbol & 15) : 0));
553
+
554
+ if (symbol & 0x8000)
555
+ {
556
+ get_bits_no_markers((symbol >> 8) & 31);
557
+ extra_bits = symbol >> 16;
558
+ }
559
+ else
560
+ {
561
+ int code_size = (symbol >> 8) & 31;
562
+ int num_extra_bits = symbol & 0xF;
563
+ int bits = code_size + num_extra_bits;
564
+ if (bits <= (m_bits_left + 16))
565
+ extra_bits = get_bits_no_markers(bits) & ((1 << num_extra_bits) - 1);
566
+ else
567
+ {
568
+ get_bits_no_markers(code_size);
569
+ extra_bits = get_bits_no_markers(num_extra_bits);
570
+ }
571
+ }
572
+
573
+ symbol &= 0xFF;
574
+ }
575
+
576
+ return symbol;
577
+ }
578
+
579
+ // Tables and macro used to fully decode the DPCM differences.
580
+ static const int s_extend_test[16] = { 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 };
581
+ static const int s_extend_offset[16] = { 0, -1, -3, -7, -15, -31, -63, -127, -255, -511, -1023, -2047, -4095, -8191, -16383, -32767 };
582
+ static const int s_extend_mask[] = { 0, (1<<0), (1<<1), (1<<2), (1<<3), (1<<4), (1<<5), (1<<6), (1<<7), (1<<8), (1<<9), (1<<10), (1<<11), (1<<12), (1<<13), (1<<14), (1<<15), (1<<16) };
583
+ #define HUFF_EXTEND(x,s) ((x) < s_extend_test[s] ? (x) + s_extend_offset[s] : (x))
584
+
585
+ // Clamps a value between 0-255.
586
+ inline uint8 jpeg_decoder::clamp(int i)
587
+ {
588
+ if (static_cast<uint>(i) > 255)
589
+ i = (((~i) >> 31) & 0xFF);
590
+
591
+ return static_cast<uint8>(i);
592
+ }
593
+
594
+ namespace DCT_Upsample
595
+ {
596
+ struct Matrix44
597
+ {
598
+ typedef int Element_Type;
599
+ enum { NUM_ROWS = 4, NUM_COLS = 4 };
600
+
601
+ Element_Type v[NUM_ROWS][NUM_COLS];
602
+
603
+ inline int rows() const { return NUM_ROWS; }
604
+ inline int cols() const { return NUM_COLS; }
605
+
606
+ inline const Element_Type & at(int r, int c) const { return v[r][c]; }
607
+ inline Element_Type & at(int r, int c) { return v[r][c]; }
608
+
609
+ inline Matrix44() { }
610
+
611
+ inline Matrix44& operator += (const Matrix44& a)
612
+ {
613
+ for (int r = 0; r < NUM_ROWS; r++)
614
+ {
615
+ at(r, 0) += a.at(r, 0);
616
+ at(r, 1) += a.at(r, 1);
617
+ at(r, 2) += a.at(r, 2);
618
+ at(r, 3) += a.at(r, 3);
619
+ }
620
+ return *this;
621
+ }
622
+
623
+ inline Matrix44& operator -= (const Matrix44& a)
624
+ {
625
+ for (int r = 0; r < NUM_ROWS; r++)
626
+ {
627
+ at(r, 0) -= a.at(r, 0);
628
+ at(r, 1) -= a.at(r, 1);
629
+ at(r, 2) -= a.at(r, 2);
630
+ at(r, 3) -= a.at(r, 3);
631
+ }
632
+ return *this;
633
+ }
634
+
635
+ friend inline Matrix44 operator + (const Matrix44& a, const Matrix44& b)
636
+ {
637
+ Matrix44 ret;
638
+ for (int r = 0; r < NUM_ROWS; r++)
639
+ {
640
+ ret.at(r, 0) = a.at(r, 0) + b.at(r, 0);
641
+ ret.at(r, 1) = a.at(r, 1) + b.at(r, 1);
642
+ ret.at(r, 2) = a.at(r, 2) + b.at(r, 2);
643
+ ret.at(r, 3) = a.at(r, 3) + b.at(r, 3);
644
+ }
645
+ return ret;
646
+ }
647
+
648
+ friend inline Matrix44 operator - (const Matrix44& a, const Matrix44& b)
649
+ {
650
+ Matrix44 ret;
651
+ for (int r = 0; r < NUM_ROWS; r++)
652
+ {
653
+ ret.at(r, 0) = a.at(r, 0) - b.at(r, 0);
654
+ ret.at(r, 1) = a.at(r, 1) - b.at(r, 1);
655
+ ret.at(r, 2) = a.at(r, 2) - b.at(r, 2);
656
+ ret.at(r, 3) = a.at(r, 3) - b.at(r, 3);
657
+ }
658
+ return ret;
659
+ }
660
+
661
+ static inline void add_and_store(jpgd_block_t* pDst, const Matrix44& a, const Matrix44& b)
662
+ {
663
+ for (int r = 0; r < 4; r++)
664
+ {
665
+ pDst[0*8 + r] = static_cast<jpgd_block_t>(a.at(r, 0) + b.at(r, 0));
666
+ pDst[1*8 + r] = static_cast<jpgd_block_t>(a.at(r, 1) + b.at(r, 1));
667
+ pDst[2*8 + r] = static_cast<jpgd_block_t>(a.at(r, 2) + b.at(r, 2));
668
+ pDst[3*8 + r] = static_cast<jpgd_block_t>(a.at(r, 3) + b.at(r, 3));
669
+ }
670
+ }
671
+
672
+ static inline void sub_and_store(jpgd_block_t* pDst, const Matrix44& a, const Matrix44& b)
673
+ {
674
+ for (int r = 0; r < 4; r++)
675
+ {
676
+ pDst[0*8 + r] = static_cast<jpgd_block_t>(a.at(r, 0) - b.at(r, 0));
677
+ pDst[1*8 + r] = static_cast<jpgd_block_t>(a.at(r, 1) - b.at(r, 1));
678
+ pDst[2*8 + r] = static_cast<jpgd_block_t>(a.at(r, 2) - b.at(r, 2));
679
+ pDst[3*8 + r] = static_cast<jpgd_block_t>(a.at(r, 3) - b.at(r, 3));
680
+ }
681
+ }
682
+ };
683
+
684
+ const int FRACT_BITS = 10;
685
+ const int SCALE = 1 << FRACT_BITS;
686
+
687
+ typedef int Temp_Type;
688
+ #define D(i) (((i) + (SCALE >> 1)) >> FRACT_BITS)
689
+ #define F(i) ((int)((i) * SCALE + .5f))
690
+
691
+ // Any decent C++ compiler will optimize this at compile time to a 0, or an array access.
692
+ #define AT(c, r) ((((c)>=NUM_COLS)||((r)>=NUM_ROWS)) ? 0 : pSrc[(c)+(r)*8])
693
+
694
+ // NUM_ROWS/NUM_COLS = # of non-zero rows/cols in input matrix
695
+ template<int NUM_ROWS, int NUM_COLS>
696
+ struct P_Q
697
+ {
698
+ static void calc(Matrix44& P, Matrix44& Q, const jpgd_block_t* pSrc)
699
+ {
700
+ // 4x8 = 4x8 times 8x8, matrix 0 is constant
701
+ const Temp_Type X000 = AT(0, 0);
702
+ const Temp_Type X001 = AT(0, 1);
703
+ const Temp_Type X002 = AT(0, 2);
704
+ const Temp_Type X003 = AT(0, 3);
705
+ const Temp_Type X004 = AT(0, 4);
706
+ const Temp_Type X005 = AT(0, 5);
707
+ const Temp_Type X006 = AT(0, 6);
708
+ const Temp_Type X007 = AT(0, 7);
709
+ const Temp_Type X010 = D(F(0.415735f) * AT(1, 0) + F(0.791065f) * AT(3, 0) + F(-0.352443f) * AT(5, 0) + F(0.277785f) * AT(7, 0));
710
+ const Temp_Type X011 = D(F(0.415735f) * AT(1, 1) + F(0.791065f) * AT(3, 1) + F(-0.352443f) * AT(5, 1) + F(0.277785f) * AT(7, 1));
711
+ const Temp_Type X012 = D(F(0.415735f) * AT(1, 2) + F(0.791065f) * AT(3, 2) + F(-0.352443f) * AT(5, 2) + F(0.277785f) * AT(7, 2));
712
+ const Temp_Type X013 = D(F(0.415735f) * AT(1, 3) + F(0.791065f) * AT(3, 3) + F(-0.352443f) * AT(5, 3) + F(0.277785f) * AT(7, 3));
713
+ const Temp_Type X014 = D(F(0.415735f) * AT(1, 4) + F(0.791065f) * AT(3, 4) + F(-0.352443f) * AT(5, 4) + F(0.277785f) * AT(7, 4));
714
+ const Temp_Type X015 = D(F(0.415735f) * AT(1, 5) + F(0.791065f) * AT(3, 5) + F(-0.352443f) * AT(5, 5) + F(0.277785f) * AT(7, 5));
715
+ const Temp_Type X016 = D(F(0.415735f) * AT(1, 6) + F(0.791065f) * AT(3, 6) + F(-0.352443f) * AT(5, 6) + F(0.277785f) * AT(7, 6));
716
+ const Temp_Type X017 = D(F(0.415735f) * AT(1, 7) + F(0.791065f) * AT(3, 7) + F(-0.352443f) * AT(5, 7) + F(0.277785f) * AT(7, 7));
717
+ const Temp_Type X020 = AT(4, 0);
718
+ const Temp_Type X021 = AT(4, 1);
719
+ const Temp_Type X022 = AT(4, 2);
720
+ const Temp_Type X023 = AT(4, 3);
721
+ const Temp_Type X024 = AT(4, 4);
722
+ const Temp_Type X025 = AT(4, 5);
723
+ const Temp_Type X026 = AT(4, 6);
724
+ const Temp_Type X027 = AT(4, 7);
725
+ const Temp_Type X030 = D(F(0.022887f) * AT(1, 0) + F(-0.097545f) * AT(3, 0) + F(0.490393f) * AT(5, 0) + F(0.865723f) * AT(7, 0));
726
+ const Temp_Type X031 = D(F(0.022887f) * AT(1, 1) + F(-0.097545f) * AT(3, 1) + F(0.490393f) * AT(5, 1) + F(0.865723f) * AT(7, 1));
727
+ const Temp_Type X032 = D(F(0.022887f) * AT(1, 2) + F(-0.097545f) * AT(3, 2) + F(0.490393f) * AT(5, 2) + F(0.865723f) * AT(7, 2));
728
+ const Temp_Type X033 = D(F(0.022887f) * AT(1, 3) + F(-0.097545f) * AT(3, 3) + F(0.490393f) * AT(5, 3) + F(0.865723f) * AT(7, 3));
729
+ const Temp_Type X034 = D(F(0.022887f) * AT(1, 4) + F(-0.097545f) * AT(3, 4) + F(0.490393f) * AT(5, 4) + F(0.865723f) * AT(7, 4));
730
+ const Temp_Type X035 = D(F(0.022887f) * AT(1, 5) + F(-0.097545f) * AT(3, 5) + F(0.490393f) * AT(5, 5) + F(0.865723f) * AT(7, 5));
731
+ const Temp_Type X036 = D(F(0.022887f) * AT(1, 6) + F(-0.097545f) * AT(3, 6) + F(0.490393f) * AT(5, 6) + F(0.865723f) * AT(7, 6));
732
+ const Temp_Type X037 = D(F(0.022887f) * AT(1, 7) + F(-0.097545f) * AT(3, 7) + F(0.490393f) * AT(5, 7) + F(0.865723f) * AT(7, 7));
733
+
734
+ // 4x4 = 4x8 times 8x4, matrix 1 is constant
735
+ P.at(0, 0) = X000;
736
+ P.at(0, 1) = D(X001 * F(0.415735f) + X003 * F(0.791065f) + X005 * F(-0.352443f) + X007 * F(0.277785f));
737
+ P.at(0, 2) = X004;
738
+ P.at(0, 3) = D(X001 * F(0.022887f) + X003 * F(-0.097545f) + X005 * F(0.490393f) + X007 * F(0.865723f));
739
+ P.at(1, 0) = X010;
740
+ P.at(1, 1) = D(X011 * F(0.415735f) + X013 * F(0.791065f) + X015 * F(-0.352443f) + X017 * F(0.277785f));
741
+ P.at(1, 2) = X014;
742
+ P.at(1, 3) = D(X011 * F(0.022887f) + X013 * F(-0.097545f) + X015 * F(0.490393f) + X017 * F(0.865723f));
743
+ P.at(2, 0) = X020;
744
+ P.at(2, 1) = D(X021 * F(0.415735f) + X023 * F(0.791065f) + X025 * F(-0.352443f) + X027 * F(0.277785f));
745
+ P.at(2, 2) = X024;
746
+ P.at(2, 3) = D(X021 * F(0.022887f) + X023 * F(-0.097545f) + X025 * F(0.490393f) + X027 * F(0.865723f));
747
+ P.at(3, 0) = X030;
748
+ P.at(3, 1) = D(X031 * F(0.415735f) + X033 * F(0.791065f) + X035 * F(-0.352443f) + X037 * F(0.277785f));
749
+ P.at(3, 2) = X034;
750
+ P.at(3, 3) = D(X031 * F(0.022887f) + X033 * F(-0.097545f) + X035 * F(0.490393f) + X037 * F(0.865723f));
751
+ // 40 muls 24 adds
752
+
753
+ // 4x4 = 4x8 times 8x4, matrix 1 is constant
754
+ Q.at(0, 0) = D(X001 * F(0.906127f) + X003 * F(-0.318190f) + X005 * F(0.212608f) + X007 * F(-0.180240f));
755
+ Q.at(0, 1) = X002;
756
+ Q.at(0, 2) = D(X001 * F(-0.074658f) + X003 * F(0.513280f) + X005 * F(0.768178f) + X007 * F(-0.375330f));
757
+ Q.at(0, 3) = X006;
758
+ Q.at(1, 0) = D(X011 * F(0.906127f) + X013 * F(-0.318190f) + X015 * F(0.212608f) + X017 * F(-0.180240f));
759
+ Q.at(1, 1) = X012;
760
+ Q.at(1, 2) = D(X011 * F(-0.074658f) + X013 * F(0.513280f) + X015 * F(0.768178f) + X017 * F(-0.375330f));
761
+ Q.at(1, 3) = X016;
762
+ Q.at(2, 0) = D(X021 * F(0.906127f) + X023 * F(-0.318190f) + X025 * F(0.212608f) + X027 * F(-0.180240f));
763
+ Q.at(2, 1) = X022;
764
+ Q.at(2, 2) = D(X021 * F(-0.074658f) + X023 * F(0.513280f) + X025 * F(0.768178f) + X027 * F(-0.375330f));
765
+ Q.at(2, 3) = X026;
766
+ Q.at(3, 0) = D(X031 * F(0.906127f) + X033 * F(-0.318190f) + X035 * F(0.212608f) + X037 * F(-0.180240f));
767
+ Q.at(3, 1) = X032;
768
+ Q.at(3, 2) = D(X031 * F(-0.074658f) + X033 * F(0.513280f) + X035 * F(0.768178f) + X037 * F(-0.375330f));
769
+ Q.at(3, 3) = X036;
770
+ // 40 muls 24 adds
771
+ }
772
+ };
773
+
774
+ template<int NUM_ROWS, int NUM_COLS>
775
+ struct R_S
776
+ {
777
+ static void calc(Matrix44& R, Matrix44& S, const jpgd_block_t* pSrc)
778
+ {
779
+ // 4x8 = 4x8 times 8x8, matrix 0 is constant
780
+ const Temp_Type X100 = D(F(0.906127f) * AT(1, 0) + F(-0.318190f) * AT(3, 0) + F(0.212608f) * AT(5, 0) + F(-0.180240f) * AT(7, 0));
781
+ const Temp_Type X101 = D(F(0.906127f) * AT(1, 1) + F(-0.318190f) * AT(3, 1) + F(0.212608f) * AT(5, 1) + F(-0.180240f) * AT(7, 1));
782
+ const Temp_Type X102 = D(F(0.906127f) * AT(1, 2) + F(-0.318190f) * AT(3, 2) + F(0.212608f) * AT(5, 2) + F(-0.180240f) * AT(7, 2));
783
+ const Temp_Type X103 = D(F(0.906127f) * AT(1, 3) + F(-0.318190f) * AT(3, 3) + F(0.212608f) * AT(5, 3) + F(-0.180240f) * AT(7, 3));
784
+ const Temp_Type X104 = D(F(0.906127f) * AT(1, 4) + F(-0.318190f) * AT(3, 4) + F(0.212608f) * AT(5, 4) + F(-0.180240f) * AT(7, 4));
785
+ const Temp_Type X105 = D(F(0.906127f) * AT(1, 5) + F(-0.318190f) * AT(3, 5) + F(0.212608f) * AT(5, 5) + F(-0.180240f) * AT(7, 5));
786
+ const Temp_Type X106 = D(F(0.906127f) * AT(1, 6) + F(-0.318190f) * AT(3, 6) + F(0.212608f) * AT(5, 6) + F(-0.180240f) * AT(7, 6));
787
+ const Temp_Type X107 = D(F(0.906127f) * AT(1, 7) + F(-0.318190f) * AT(3, 7) + F(0.212608f) * AT(5, 7) + F(-0.180240f) * AT(7, 7));
788
+ const Temp_Type X110 = AT(2, 0);
789
+ const Temp_Type X111 = AT(2, 1);
790
+ const Temp_Type X112 = AT(2, 2);
791
+ const Temp_Type X113 = AT(2, 3);
792
+ const Temp_Type X114 = AT(2, 4);
793
+ const Temp_Type X115 = AT(2, 5);
794
+ const Temp_Type X116 = AT(2, 6);
795
+ const Temp_Type X117 = AT(2, 7);
796
+ const Temp_Type X120 = D(F(-0.074658f) * AT(1, 0) + F(0.513280f) * AT(3, 0) + F(0.768178f) * AT(5, 0) + F(-0.375330f) * AT(7, 0));
797
+ const Temp_Type X121 = D(F(-0.074658f) * AT(1, 1) + F(0.513280f) * AT(3, 1) + F(0.768178f) * AT(5, 1) + F(-0.375330f) * AT(7, 1));
798
+ const Temp_Type X122 = D(F(-0.074658f) * AT(1, 2) + F(0.513280f) * AT(3, 2) + F(0.768178f) * AT(5, 2) + F(-0.375330f) * AT(7, 2));
799
+ const Temp_Type X123 = D(F(-0.074658f) * AT(1, 3) + F(0.513280f) * AT(3, 3) + F(0.768178f) * AT(5, 3) + F(-0.375330f) * AT(7, 3));
800
+ const Temp_Type X124 = D(F(-0.074658f) * AT(1, 4) + F(0.513280f) * AT(3, 4) + F(0.768178f) * AT(5, 4) + F(-0.375330f) * AT(7, 4));
801
+ const Temp_Type X125 = D(F(-0.074658f) * AT(1, 5) + F(0.513280f) * AT(3, 5) + F(0.768178f) * AT(5, 5) + F(-0.375330f) * AT(7, 5));
802
+ const Temp_Type X126 = D(F(-0.074658f) * AT(1, 6) + F(0.513280f) * AT(3, 6) + F(0.768178f) * AT(5, 6) + F(-0.375330f) * AT(7, 6));
803
+ const Temp_Type X127 = D(F(-0.074658f) * AT(1, 7) + F(0.513280f) * AT(3, 7) + F(0.768178f) * AT(5, 7) + F(-0.375330f) * AT(7, 7));
804
+ const Temp_Type X130 = AT(6, 0);
805
+ const Temp_Type X131 = AT(6, 1);
806
+ const Temp_Type X132 = AT(6, 2);
807
+ const Temp_Type X133 = AT(6, 3);
808
+ const Temp_Type X134 = AT(6, 4);
809
+ const Temp_Type X135 = AT(6, 5);
810
+ const Temp_Type X136 = AT(6, 6);
811
+ const Temp_Type X137 = AT(6, 7);
812
+ // 80 muls 48 adds
813
+
814
+ // 4x4 = 4x8 times 8x4, matrix 1 is constant
815
+ R.at(0, 0) = X100;
816
+ R.at(0, 1) = D(X101 * F(0.415735f) + X103 * F(0.791065f) + X105 * F(-0.352443f) + X107 * F(0.277785f));
817
+ R.at(0, 2) = X104;
818
+ R.at(0, 3) = D(X101 * F(0.022887f) + X103 * F(-0.097545f) + X105 * F(0.490393f) + X107 * F(0.865723f));
819
+ R.at(1, 0) = X110;
820
+ R.at(1, 1) = D(X111 * F(0.415735f) + X113 * F(0.791065f) + X115 * F(-0.352443f) + X117 * F(0.277785f));
821
+ R.at(1, 2) = X114;
822
+ R.at(1, 3) = D(X111 * F(0.022887f) + X113 * F(-0.097545f) + X115 * F(0.490393f) + X117 * F(0.865723f));
823
+ R.at(2, 0) = X120;
824
+ R.at(2, 1) = D(X121 * F(0.415735f) + X123 * F(0.791065f) + X125 * F(-0.352443f) + X127 * F(0.277785f));
825
+ R.at(2, 2) = X124;
826
+ R.at(2, 3) = D(X121 * F(0.022887f) + X123 * F(-0.097545f) + X125 * F(0.490393f) + X127 * F(0.865723f));
827
+ R.at(3, 0) = X130;
828
+ R.at(3, 1) = D(X131 * F(0.415735f) + X133 * F(0.791065f) + X135 * F(-0.352443f) + X137 * F(0.277785f));
829
+ R.at(3, 2) = X134;
830
+ R.at(3, 3) = D(X131 * F(0.022887f) + X133 * F(-0.097545f) + X135 * F(0.490393f) + X137 * F(0.865723f));
831
+ // 40 muls 24 adds
832
+ // 4x4 = 4x8 times 8x4, matrix 1 is constant
833
+ S.at(0, 0) = D(X101 * F(0.906127f) + X103 * F(-0.318190f) + X105 * F(0.212608f) + X107 * F(-0.180240f));
834
+ S.at(0, 1) = X102;
835
+ S.at(0, 2) = D(X101 * F(-0.074658f) + X103 * F(0.513280f) + X105 * F(0.768178f) + X107 * F(-0.375330f));
836
+ S.at(0, 3) = X106;
837
+ S.at(1, 0) = D(X111 * F(0.906127f) + X113 * F(-0.318190f) + X115 * F(0.212608f) + X117 * F(-0.180240f));
838
+ S.at(1, 1) = X112;
839
+ S.at(1, 2) = D(X111 * F(-0.074658f) + X113 * F(0.513280f) + X115 * F(0.768178f) + X117 * F(-0.375330f));
840
+ S.at(1, 3) = X116;
841
+ S.at(2, 0) = D(X121 * F(0.906127f) + X123 * F(-0.318190f) + X125 * F(0.212608f) + X127 * F(-0.180240f));
842
+ S.at(2, 1) = X122;
843
+ S.at(2, 2) = D(X121 * F(-0.074658f) + X123 * F(0.513280f) + X125 * F(0.768178f) + X127 * F(-0.375330f));
844
+ S.at(2, 3) = X126;
845
+ S.at(3, 0) = D(X131 * F(0.906127f) + X133 * F(-0.318190f) + X135 * F(0.212608f) + X137 * F(-0.180240f));
846
+ S.at(3, 1) = X132;
847
+ S.at(3, 2) = D(X131 * F(-0.074658f) + X133 * F(0.513280f) + X135 * F(0.768178f) + X137 * F(-0.375330f));
848
+ S.at(3, 3) = X136;
849
+ // 40 muls 24 adds
850
+ }
851
+ };
852
+ } // end namespace DCT_Upsample
853
+
854
+ // Unconditionally frees all allocated m_blocks.
855
+ void jpeg_decoder::free_all_blocks()
856
+ {
857
+ m_pStream = NULL;
858
+ for (mem_block *b = m_pMem_blocks; b; )
859
+ {
860
+ mem_block *n = b->m_pNext;
861
+ jpgd_free(b);
862
+ b = n;
863
+ }
864
+ m_pMem_blocks = NULL;
865
+ }
866
+
867
+ // This method handles all errors.
868
+ // It could easily be changed to use C++ exceptions.
869
+ void jpeg_decoder::stop_decoding(jpgd_status status)
870
+ {
871
+ m_error_code = status;
872
+ free_all_blocks();
873
+ longjmp(m_jmp_state, status);
874
+
875
+ // we shouldn't get here as longjmp shouldn't return, but we put it here to make it explicit
876
+ // that this function doesn't return, otherwise we get this error:
877
+ //
878
+ // error : function declared 'noreturn' should not return
879
+ exit(1);
880
+ }
881
+
882
+ void *jpeg_decoder::alloc(size_t nSize, bool zero)
883
+ {
884
+ nSize = (JPGD_MAX(nSize, 1) + 3) & ~3;
885
+ char *rv = NULL;
886
+ for (mem_block *b = m_pMem_blocks; b; b = b->m_pNext)
887
+ {
888
+ if ((b->m_used_count + nSize) <= b->m_size)
889
+ {
890
+ rv = b->m_data + b->m_used_count;
891
+ b->m_used_count += nSize;
892
+ break;
893
+ }
894
+ }
895
+ if (!rv)
896
+ {
897
+ int capacity = JPGD_MAX(32768 - 256, (nSize + 2047) & ~2047);
898
+ mem_block *b = (mem_block*)jpgd_malloc(sizeof(mem_block) + capacity);
899
+ if (!b) stop_decoding(JPGD_NOTENOUGHMEM);
900
+ b->m_pNext = m_pMem_blocks; m_pMem_blocks = b;
901
+ b->m_used_count = nSize;
902
+ b->m_size = capacity;
903
+ rv = b->m_data;
904
+ }
905
+ if (zero) memset(rv, 0, nSize);
906
+ return rv;
907
+ }
908
+
909
+ void jpeg_decoder::word_clear(void *p, uint16 c, uint n)
910
+ {
911
+ uint8 *pD = (uint8*)p;
912
+ const uint8 l = c & 0xFF, h = (c >> 8) & 0xFF;
913
+ while (n)
914
+ {
915
+ pD[0] = l; pD[1] = h; pD += 2;
916
+ n--;
917
+ }
918
+ }
919
+
920
+ // Refill the input buffer.
921
+ // This method will sit in a loop until (A) the buffer is full or (B)
922
+ // the stream's read() method reports and end of file condition.
923
+ void jpeg_decoder::prep_in_buffer()
924
+ {
925
+ m_in_buf_left = 0;
926
+ m_pIn_buf_ofs = m_in_buf;
927
+
928
+ if (m_eof_flag)
929
+ return;
930
+
931
+ do
932
+ {
933
+ int bytes_read = m_pStream->read(m_in_buf + m_in_buf_left, JPGD_IN_BUF_SIZE - m_in_buf_left, &m_eof_flag);
934
+ if (bytes_read == -1)
935
+ stop_decoding(JPGD_STREAM_READ);
936
+
937
+ m_in_buf_left += bytes_read;
938
+ } while ((m_in_buf_left < JPGD_IN_BUF_SIZE) && (!m_eof_flag));
939
+
940
+ m_total_bytes_read += m_in_buf_left;
941
+
942
+ // Pad the end of the block with M_EOI (prevents the decompressor from going off the rails if the stream is invalid).
943
+ // (This dates way back to when this decompressor was written in C/asm, and the all-asm Huffman decoder did some fancy things to increase perf.)
944
+ word_clear(m_pIn_buf_ofs + m_in_buf_left, 0xD9FF, 64);
945
+ }
946
+
947
+ // Read a Huffman code table.
948
+ void jpeg_decoder::read_dht_marker()
949
+ {
950
+ int i, index, count;
951
+ uint8 huff_num[17];
952
+ uint8 huff_val[256];
953
+
954
+ uint num_left = get_bits(16);
955
+
956
+ if (num_left < 2)
957
+ stop_decoding(JPGD_BAD_DHT_MARKER);
958
+
959
+ num_left -= 2;
960
+
961
+ while (num_left)
962
+ {
963
+ index = get_bits(8);
964
+
965
+ huff_num[0] = 0;
966
+
967
+ count = 0;
968
+
969
+ for (i = 1; i <= 16; i++)
970
+ {
971
+ huff_num[i] = static_cast<uint8>(get_bits(8));
972
+ count += huff_num[i];
973
+ }
974
+
975
+ if (count > 255)
976
+ stop_decoding(JPGD_BAD_DHT_COUNTS);
977
+
978
+ for (i = 0; i < count; i++)
979
+ huff_val[i] = static_cast<uint8>(get_bits(8));
980
+
981
+ i = 1 + 16 + count;
982
+
983
+ if (num_left < (uint)i)
984
+ stop_decoding(JPGD_BAD_DHT_MARKER);
985
+
986
+ num_left -= i;
987
+
988
+ if ((index & 0x10) > 0x10)
989
+ stop_decoding(JPGD_BAD_DHT_INDEX);
990
+
991
+ index = (index & 0x0F) + ((index & 0x10) >> 4) * (JPGD_MAX_HUFF_TABLES >> 1);
992
+
993
+ if (index >= JPGD_MAX_HUFF_TABLES)
994
+ stop_decoding(JPGD_BAD_DHT_INDEX);
995
+
996
+ if (!m_huff_num[index])
997
+ m_huff_num[index] = (uint8 *)alloc(17);
998
+
999
+ if (!m_huff_val[index])
1000
+ m_huff_val[index] = (uint8 *)alloc(256);
1001
+
1002
+ m_huff_ac[index] = (index & 0x10) != 0;
1003
+ memcpy(m_huff_num[index], huff_num, 17);
1004
+ memcpy(m_huff_val[index], huff_val, 256);
1005
+ }
1006
+ }
1007
+
1008
+ // Read a quantization table.
1009
+ void jpeg_decoder::read_dqt_marker()
1010
+ {
1011
+ int n, i, prec;
1012
+ uint num_left;
1013
+ uint temp;
1014
+
1015
+ num_left = get_bits(16);
1016
+
1017
+ if (num_left < 2)
1018
+ stop_decoding(JPGD_BAD_DQT_MARKER);
1019
+
1020
+ num_left -= 2;
1021
+
1022
+ while (num_left)
1023
+ {
1024
+ n = get_bits(8);
1025
+ prec = n >> 4;
1026
+ n &= 0x0F;
1027
+
1028
+ if (n >= JPGD_MAX_QUANT_TABLES)
1029
+ stop_decoding(JPGD_BAD_DQT_TABLE);
1030
+
1031
+ if (!m_quant[n])
1032
+ m_quant[n] = (jpgd_quant_t *)alloc(64 * sizeof(jpgd_quant_t));
1033
+
1034
+ // read quantization entries, in zag order
1035
+ for (i = 0; i < 64; i++)
1036
+ {
1037
+ temp = get_bits(8);
1038
+
1039
+ if (prec)
1040
+ temp = (temp << 8) + get_bits(8);
1041
+
1042
+ m_quant[n][i] = static_cast<jpgd_quant_t>(temp);
1043
+ }
1044
+
1045
+ i = 64 + 1;
1046
+
1047
+ if (prec)
1048
+ i += 64;
1049
+
1050
+ if (num_left < (uint)i)
1051
+ stop_decoding(JPGD_BAD_DQT_LENGTH);
1052
+
1053
+ num_left -= i;
1054
+ }
1055
+ }
1056
+
1057
+ // Read the start of frame (SOF) marker.
1058
+ void jpeg_decoder::read_sof_marker()
1059
+ {
1060
+ int i;
1061
+ uint num_left;
1062
+
1063
+ num_left = get_bits(16);
1064
+
1065
+ if (get_bits(8) != 8) /* precision: sorry, only 8-bit precision is supported right now */
1066
+ stop_decoding(JPGD_BAD_PRECISION);
1067
+
1068
+ m_image_y_size = get_bits(16);
1069
+
1070
+ if ((m_image_y_size < 1) || (m_image_y_size > JPGD_MAX_HEIGHT))
1071
+ stop_decoding(JPGD_BAD_HEIGHT);
1072
+
1073
+ m_image_x_size = get_bits(16);
1074
+
1075
+ if ((m_image_x_size < 1) || (m_image_x_size > JPGD_MAX_WIDTH))
1076
+ stop_decoding(JPGD_BAD_WIDTH);
1077
+
1078
+ m_comps_in_frame = get_bits(8);
1079
+
1080
+ if (m_comps_in_frame > JPGD_MAX_COMPONENTS)
1081
+ stop_decoding(JPGD_TOO_MANY_COMPONENTS);
1082
+
1083
+ if (num_left != (uint)(m_comps_in_frame * 3 + 8))
1084
+ stop_decoding(JPGD_BAD_SOF_LENGTH);
1085
+
1086
+ for (i = 0; i < m_comps_in_frame; i++)
1087
+ {
1088
+ m_comp_ident[i] = get_bits(8);
1089
+ m_comp_h_samp[i] = get_bits(4);
1090
+ m_comp_v_samp[i] = get_bits(4);
1091
+ m_comp_quant[i] = get_bits(8);
1092
+ }
1093
+ }
1094
+
1095
+ // Used to skip unrecognized markers.
1096
+ void jpeg_decoder::skip_variable_marker()
1097
+ {
1098
+ uint num_left;
1099
+
1100
+ num_left = get_bits(16);
1101
+
1102
+ if (num_left < 2)
1103
+ stop_decoding(JPGD_BAD_VARIABLE_MARKER);
1104
+
1105
+ num_left -= 2;
1106
+
1107
+ while (num_left)
1108
+ {
1109
+ get_bits(8);
1110
+ num_left--;
1111
+ }
1112
+ }
1113
+
1114
+ // Read a define restart interval (DRI) marker.
1115
+ void jpeg_decoder::read_dri_marker()
1116
+ {
1117
+ if (get_bits(16) != 4)
1118
+ stop_decoding(JPGD_BAD_DRI_LENGTH);
1119
+
1120
+ m_restart_interval = get_bits(16);
1121
+ }
1122
+
1123
+ // Read a start of scan (SOS) marker.
1124
+ void jpeg_decoder::read_sos_marker()
1125
+ {
1126
+ uint num_left;
1127
+ int i, ci, n, c, cc;
1128
+
1129
+ num_left = get_bits(16);
1130
+
1131
+ n = get_bits(8);
1132
+
1133
+ m_comps_in_scan = n;
1134
+
1135
+ num_left -= 3;
1136
+
1137
+ if ( (num_left != (uint)(n * 2 + 3)) || (n < 1) || (n > JPGD_MAX_COMPS_IN_SCAN) )
1138
+ stop_decoding(JPGD_BAD_SOS_LENGTH);
1139
+
1140
+ for (i = 0; i < n; i++)
1141
+ {
1142
+ cc = get_bits(8);
1143
+ c = get_bits(8);
1144
+ num_left -= 2;
1145
+
1146
+ for (ci = 0; ci < m_comps_in_frame; ci++)
1147
+ if (cc == m_comp_ident[ci])
1148
+ break;
1149
+
1150
+ if (ci >= m_comps_in_frame)
1151
+ stop_decoding(JPGD_BAD_SOS_COMP_ID);
1152
+
1153
+ m_comp_list[i] = ci;
1154
+ m_comp_dc_tab[ci] = (c >> 4) & 15;
1155
+ m_comp_ac_tab[ci] = (c & 15) + (JPGD_MAX_HUFF_TABLES >> 1);
1156
+ }
1157
+
1158
+ m_spectral_start = get_bits(8);
1159
+ m_spectral_end = get_bits(8);
1160
+ m_successive_high = get_bits(4);
1161
+ m_successive_low = get_bits(4);
1162
+
1163
+ if (!m_progressive_flag)
1164
+ {
1165
+ m_spectral_start = 0;
1166
+ m_spectral_end = 63;
1167
+ }
1168
+
1169
+ num_left -= 3;
1170
+
1171
+ while (num_left) /* read past whatever is num_left */
1172
+ {
1173
+ get_bits(8);
1174
+ num_left--;
1175
+ }
1176
+ }
1177
+
1178
+ // Finds the next marker.
1179
+ int jpeg_decoder::next_marker()
1180
+ {
1181
+ uint c, bytes;
1182
+
1183
+ bytes = 0;
1184
+
1185
+ do
1186
+ {
1187
+ do
1188
+ {
1189
+ bytes++;
1190
+ c = get_bits(8);
1191
+ } while (c != 0xFF);
1192
+
1193
+ do
1194
+ {
1195
+ c = get_bits(8);
1196
+ } while (c == 0xFF);
1197
+
1198
+ } while (c == 0);
1199
+
1200
+ // If bytes > 0 here, there where extra bytes before the marker (not good).
1201
+
1202
+ return c;
1203
+ }
1204
+
1205
+ // Process markers. Returns when an SOFx, SOI, EOI, or SOS marker is
1206
+ // encountered.
1207
+ int jpeg_decoder::process_markers()
1208
+ {
1209
+ int c;
1210
+
1211
+ for ( ; ; )
1212
+ {
1213
+ c = next_marker();
1214
+
1215
+ switch (c)
1216
+ {
1217
+ case M_SOF0:
1218
+ case M_SOF1:
1219
+ case M_SOF2:
1220
+ case M_SOF3:
1221
+ case M_SOF5:
1222
+ case M_SOF6:
1223
+ case M_SOF7:
1224
+ // case M_JPG:
1225
+ case M_SOF9:
1226
+ case M_SOF10:
1227
+ case M_SOF11:
1228
+ case M_SOF13:
1229
+ case M_SOF14:
1230
+ case M_SOF15:
1231
+ case M_SOI:
1232
+ case M_EOI:
1233
+ case M_SOS:
1234
+ {
1235
+ return c;
1236
+ }
1237
+ case M_DHT:
1238
+ {
1239
+ read_dht_marker();
1240
+ break;
1241
+ }
1242
+ // No arithmitic support - dumb patents!
1243
+ case M_DAC:
1244
+ {
1245
+ stop_decoding(JPGD_NO_ARITHMITIC_SUPPORT);
1246
+ break;
1247
+ }
1248
+ case M_DQT:
1249
+ {
1250
+ read_dqt_marker();
1251
+ break;
1252
+ }
1253
+ case M_DRI:
1254
+ {
1255
+ read_dri_marker();
1256
+ break;
1257
+ }
1258
+ //case M_APP0: /* no need to read the JFIF marker */
1259
+
1260
+ case M_JPG:
1261
+ case M_RST0: /* no parameters */
1262
+ case M_RST1:
1263
+ case M_RST2:
1264
+ case M_RST3:
1265
+ case M_RST4:
1266
+ case M_RST5:
1267
+ case M_RST6:
1268
+ case M_RST7:
1269
+ case M_TEM:
1270
+ {
1271
+ stop_decoding(JPGD_UNEXPECTED_MARKER);
1272
+ break;
1273
+ }
1274
+ default: /* must be DNL, DHP, EXP, APPn, JPGn, COM, or RESn or APP0 */
1275
+ {
1276
+ skip_variable_marker();
1277
+ break;
1278
+ }
1279
+ }
1280
+ }
1281
+ }
1282
+
1283
+ // Finds the start of image (SOI) marker.
1284
+ // This code is rather defensive: it only checks the first 512 bytes to avoid
1285
+ // false positives.
1286
+ void jpeg_decoder::locate_soi_marker()
1287
+ {
1288
+ uint lastchar, thischar;
1289
+ uint bytesleft;
1290
+
1291
+ lastchar = get_bits(8);
1292
+
1293
+ thischar = get_bits(8);
1294
+
1295
+ /* ok if it's a normal JPEG file without a special header */
1296
+
1297
+ if ((lastchar == 0xFF) && (thischar == M_SOI))
1298
+ return;
1299
+
1300
+ bytesleft = 4096; //512;
1301
+
1302
+ for ( ; ; )
1303
+ {
1304
+ if (--bytesleft == 0)
1305
+ stop_decoding(JPGD_NOT_JPEG);
1306
+
1307
+ lastchar = thischar;
1308
+
1309
+ thischar = get_bits(8);
1310
+
1311
+ if (lastchar == 0xFF)
1312
+ {
1313
+ if (thischar == M_SOI)
1314
+ break;
1315
+ else if (thischar == M_EOI) // get_bits will keep returning M_EOI if we read past the end
1316
+ stop_decoding(JPGD_NOT_JPEG);
1317
+ }
1318
+ }
1319
+
1320
+ // Check the next character after marker: if it's not 0xFF, it can't be the start of the next marker, so the file is bad.
1321
+ thischar = (m_bit_buf >> 24) & 0xFF;
1322
+
1323
+ if (thischar != 0xFF)
1324
+ stop_decoding(JPGD_NOT_JPEG);
1325
+ }
1326
+
1327
+ // Find a start of frame (SOF) marker.
1328
+ void jpeg_decoder::locate_sof_marker()
1329
+ {
1330
+ locate_soi_marker();
1331
+
1332
+ int c = process_markers();
1333
+
1334
+ switch (c)
1335
+ {
1336
+ case M_SOF2:
1337
+ m_progressive_flag = JPGD_TRUE;
1338
+ case M_SOF0: /* baseline DCT */
1339
+ case M_SOF1: /* extended sequential DCT */
1340
+ {
1341
+ read_sof_marker();
1342
+ break;
1343
+ }
1344
+ case M_SOF9: /* Arithmitic coding */
1345
+ {
1346
+ stop_decoding(JPGD_NO_ARITHMITIC_SUPPORT);
1347
+ break;
1348
+ }
1349
+ default:
1350
+ {
1351
+ stop_decoding(JPGD_UNSUPPORTED_MARKER);
1352
+ break;
1353
+ }
1354
+ }
1355
+ }
1356
+
1357
+ // Find a start of scan (SOS) marker.
1358
+ int jpeg_decoder::locate_sos_marker()
1359
+ {
1360
+ int c;
1361
+
1362
+ c = process_markers();
1363
+
1364
+ if (c == M_EOI)
1365
+ return JPGD_FALSE;
1366
+ else if (c != M_SOS)
1367
+ stop_decoding(JPGD_UNEXPECTED_MARKER);
1368
+
1369
+ read_sos_marker();
1370
+
1371
+ return JPGD_TRUE;
1372
+ }
1373
+
1374
+ // Reset everything to default/uninitialized state.
1375
+ void jpeg_decoder::init(jpeg_decoder_stream *pStream)
1376
+ {
1377
+ m_pMem_blocks = NULL;
1378
+ m_error_code = JPGD_SUCCESS;
1379
+ m_ready_flag = false;
1380
+ m_image_x_size = m_image_y_size = 0;
1381
+ m_pStream = pStream;
1382
+ m_progressive_flag = JPGD_FALSE;
1383
+
1384
+ memset(m_huff_ac, 0, sizeof(m_huff_ac));
1385
+ memset(m_huff_num, 0, sizeof(m_huff_num));
1386
+ memset(m_huff_val, 0, sizeof(m_huff_val));
1387
+ memset(m_quant, 0, sizeof(m_quant));
1388
+
1389
+ m_scan_type = 0;
1390
+ m_comps_in_frame = 0;
1391
+
1392
+ memset(m_comp_h_samp, 0, sizeof(m_comp_h_samp));
1393
+ memset(m_comp_v_samp, 0, sizeof(m_comp_v_samp));
1394
+ memset(m_comp_quant, 0, sizeof(m_comp_quant));
1395
+ memset(m_comp_ident, 0, sizeof(m_comp_ident));
1396
+ memset(m_comp_h_blocks, 0, sizeof(m_comp_h_blocks));
1397
+ memset(m_comp_v_blocks, 0, sizeof(m_comp_v_blocks));
1398
+
1399
+ m_comps_in_scan = 0;
1400
+ memset(m_comp_list, 0, sizeof(m_comp_list));
1401
+ memset(m_comp_dc_tab, 0, sizeof(m_comp_dc_tab));
1402
+ memset(m_comp_ac_tab, 0, sizeof(m_comp_ac_tab));
1403
+
1404
+ m_spectral_start = 0;
1405
+ m_spectral_end = 0;
1406
+ m_successive_low = 0;
1407
+ m_successive_high = 0;
1408
+ m_max_mcu_x_size = 0;
1409
+ m_max_mcu_y_size = 0;
1410
+ m_blocks_per_mcu = 0;
1411
+ m_max_blocks_per_row = 0;
1412
+ m_mcus_per_row = 0;
1413
+ m_mcus_per_col = 0;
1414
+ m_expanded_blocks_per_component = 0;
1415
+ m_expanded_blocks_per_mcu = 0;
1416
+ m_expanded_blocks_per_row = 0;
1417
+ m_freq_domain_chroma_upsample = false;
1418
+
1419
+ memset(m_mcu_org, 0, sizeof(m_mcu_org));
1420
+
1421
+ m_total_lines_left = 0;
1422
+ m_mcu_lines_left = 0;
1423
+ m_real_dest_bytes_per_scan_line = 0;
1424
+ m_dest_bytes_per_scan_line = 0;
1425
+ m_dest_bytes_per_pixel = 0;
1426
+
1427
+ memset(m_pHuff_tabs, 0, sizeof(m_pHuff_tabs));
1428
+
1429
+ memset(m_dc_coeffs, 0, sizeof(m_dc_coeffs));
1430
+ memset(m_ac_coeffs, 0, sizeof(m_ac_coeffs));
1431
+ memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
1432
+
1433
+ m_eob_run = 0;
1434
+
1435
+ memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
1436
+
1437
+ m_pIn_buf_ofs = m_in_buf;
1438
+ m_in_buf_left = 0;
1439
+ m_eof_flag = false;
1440
+ m_tem_flag = 0;
1441
+
1442
+ memset(m_in_buf_pad_start, 0, sizeof(m_in_buf_pad_start));
1443
+ memset(m_in_buf, 0, sizeof(m_in_buf));
1444
+ memset(m_in_buf_pad_end, 0, sizeof(m_in_buf_pad_end));
1445
+
1446
+ m_restart_interval = 0;
1447
+ m_restarts_left = 0;
1448
+ m_next_restart_num = 0;
1449
+
1450
+ m_max_mcus_per_row = 0;
1451
+ m_max_blocks_per_mcu = 0;
1452
+ m_max_mcus_per_col = 0;
1453
+
1454
+ memset(m_last_dc_val, 0, sizeof(m_last_dc_val));
1455
+ m_pMCU_coefficients = NULL;
1456
+ m_pSample_buf = NULL;
1457
+
1458
+ m_total_bytes_read = 0;
1459
+
1460
+ m_pScan_line_0 = NULL;
1461
+ m_pScan_line_1 = NULL;
1462
+
1463
+ // Ready the input buffer.
1464
+ prep_in_buffer();
1465
+
1466
+ // Prime the bit buffer.
1467
+ m_bits_left = 16;
1468
+ m_bit_buf = 0;
1469
+
1470
+ get_bits(16);
1471
+ get_bits(16);
1472
+
1473
+ for (int i = 0; i < JPGD_MAX_BLOCKS_PER_MCU; i++)
1474
+ m_mcu_block_max_zag[i] = 64;
1475
+ }
1476
+
1477
+ #define SCALEBITS 16
1478
+ #define ONE_HALF ((int) 1 << (SCALEBITS-1))
1479
+ #define FIX(x) ((int) ((x) * (1L<<SCALEBITS) + 0.5f))
1480
+
1481
+ // Create a few tables that allow us to quickly convert YCbCr to RGB.
1482
+ void jpeg_decoder::create_look_ups()
1483
+ {
1484
+ for (int i = 0; i <= 255; i++)
1485
+ {
1486
+ int k = i - 128;
1487
+ m_crr[i] = ( FIX(1.40200f) * k + ONE_HALF) >> SCALEBITS;
1488
+ m_cbb[i] = ( FIX(1.77200f) * k + ONE_HALF) >> SCALEBITS;
1489
+ m_crg[i] = (-FIX(0.71414f)) * k;
1490
+ m_cbg[i] = (-FIX(0.34414f)) * k + ONE_HALF;
1491
+ }
1492
+ }
1493
+
1494
+ // This method throws back into the stream any bytes that where read
1495
+ // into the bit buffer during initial marker scanning.
1496
+ void jpeg_decoder::fix_in_buffer()
1497
+ {
1498
+ // In case any 0xFF's where pulled into the buffer during marker scanning.
1499
+ JPGD_ASSERT((m_bits_left & 7) == 0);
1500
+
1501
+ if (m_bits_left == 16)
1502
+ stuff_char( (uint8)(m_bit_buf & 0xFF));
1503
+
1504
+ if (m_bits_left >= 8)
1505
+ stuff_char( (uint8)((m_bit_buf >> 8) & 0xFF));
1506
+
1507
+ stuff_char((uint8)((m_bit_buf >> 16) & 0xFF));
1508
+ stuff_char((uint8)((m_bit_buf >> 24) & 0xFF));
1509
+
1510
+ m_bits_left = 16;
1511
+ get_bits_no_markers(16);
1512
+ get_bits_no_markers(16);
1513
+ }
1514
+
1515
+ void jpeg_decoder::transform_mcu(int mcu_row)
1516
+ {
1517
+ jpgd_block_t* pSrc_ptr = m_pMCU_coefficients;
1518
+ uint8* pDst_ptr = m_pSample_buf + mcu_row * m_blocks_per_mcu * 64;
1519
+
1520
+ for (int mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++)
1521
+ {
1522
+ idct(pSrc_ptr, pDst_ptr, m_mcu_block_max_zag[mcu_block]);
1523
+ pSrc_ptr += 64;
1524
+ pDst_ptr += 64;
1525
+ }
1526
+ }
1527
+
1528
+ static const uint8 s_max_rc[64] =
1529
+ {
1530
+ 17, 18, 34, 50, 50, 51, 52, 52, 52, 68, 84, 84, 84, 84, 85, 86, 86, 86, 86, 86,
1531
+ 102, 118, 118, 118, 118, 118, 118, 119, 120, 120, 120, 120, 120, 120, 120, 136,
1532
+ 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136,
1533
+ 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136
1534
+ };
1535
+
1536
+ void jpeg_decoder::transform_mcu_expand(int mcu_row)
1537
+ {
1538
+ jpgd_block_t* pSrc_ptr = m_pMCU_coefficients;
1539
+ uint8* pDst_ptr = m_pSample_buf + mcu_row * m_expanded_blocks_per_mcu * 64;
1540
+
1541
+ // Y IDCT
1542
+ int mcu_block;
1543
+ for (mcu_block = 0; mcu_block < m_expanded_blocks_per_component; mcu_block++)
1544
+ {
1545
+ idct(pSrc_ptr, pDst_ptr, m_mcu_block_max_zag[mcu_block]);
1546
+ pSrc_ptr += 64;
1547
+ pDst_ptr += 64;
1548
+ }
1549
+
1550
+ // Chroma IDCT, with upsampling
1551
+ jpgd_block_t temp_block[64];
1552
+
1553
+ for (int i = 0; i < 2; i++)
1554
+ {
1555
+ DCT_Upsample::Matrix44 P, Q, R, S;
1556
+
1557
+ JPGD_ASSERT(m_mcu_block_max_zag[mcu_block] >= 1);
1558
+ JPGD_ASSERT(m_mcu_block_max_zag[mcu_block] <= 64);
1559
+
1560
+ switch (s_max_rc[m_mcu_block_max_zag[mcu_block++] - 1])
1561
+ {
1562
+ case 1*16+1:
1563
+ DCT_Upsample::P_Q<1, 1>::calc(P, Q, pSrc_ptr);
1564
+ DCT_Upsample::R_S<1, 1>::calc(R, S, pSrc_ptr);
1565
+ break;
1566
+ case 1*16+2:
1567
+ DCT_Upsample::P_Q<1, 2>::calc(P, Q, pSrc_ptr);
1568
+ DCT_Upsample::R_S<1, 2>::calc(R, S, pSrc_ptr);
1569
+ break;
1570
+ case 2*16+2:
1571
+ DCT_Upsample::P_Q<2, 2>::calc(P, Q, pSrc_ptr);
1572
+ DCT_Upsample::R_S<2, 2>::calc(R, S, pSrc_ptr);
1573
+ break;
1574
+ case 3*16+2:
1575
+ DCT_Upsample::P_Q<3, 2>::calc(P, Q, pSrc_ptr);
1576
+ DCT_Upsample::R_S<3, 2>::calc(R, S, pSrc_ptr);
1577
+ break;
1578
+ case 3*16+3:
1579
+ DCT_Upsample::P_Q<3, 3>::calc(P, Q, pSrc_ptr);
1580
+ DCT_Upsample::R_S<3, 3>::calc(R, S, pSrc_ptr);
1581
+ break;
1582
+ case 3*16+4:
1583
+ DCT_Upsample::P_Q<3, 4>::calc(P, Q, pSrc_ptr);
1584
+ DCT_Upsample::R_S<3, 4>::calc(R, S, pSrc_ptr);
1585
+ break;
1586
+ case 4*16+4:
1587
+ DCT_Upsample::P_Q<4, 4>::calc(P, Q, pSrc_ptr);
1588
+ DCT_Upsample::R_S<4, 4>::calc(R, S, pSrc_ptr);
1589
+ break;
1590
+ case 5*16+4:
1591
+ DCT_Upsample::P_Q<5, 4>::calc(P, Q, pSrc_ptr);
1592
+ DCT_Upsample::R_S<5, 4>::calc(R, S, pSrc_ptr);
1593
+ break;
1594
+ case 5*16+5:
1595
+ DCT_Upsample::P_Q<5, 5>::calc(P, Q, pSrc_ptr);
1596
+ DCT_Upsample::R_S<5, 5>::calc(R, S, pSrc_ptr);
1597
+ break;
1598
+ case 5*16+6:
1599
+ DCT_Upsample::P_Q<5, 6>::calc(P, Q, pSrc_ptr);
1600
+ DCT_Upsample::R_S<5, 6>::calc(R, S, pSrc_ptr);
1601
+ break;
1602
+ case 6*16+6:
1603
+ DCT_Upsample::P_Q<6, 6>::calc(P, Q, pSrc_ptr);
1604
+ DCT_Upsample::R_S<6, 6>::calc(R, S, pSrc_ptr);
1605
+ break;
1606
+ case 7*16+6:
1607
+ DCT_Upsample::P_Q<7, 6>::calc(P, Q, pSrc_ptr);
1608
+ DCT_Upsample::R_S<7, 6>::calc(R, S, pSrc_ptr);
1609
+ break;
1610
+ case 7*16+7:
1611
+ DCT_Upsample::P_Q<7, 7>::calc(P, Q, pSrc_ptr);
1612
+ DCT_Upsample::R_S<7, 7>::calc(R, S, pSrc_ptr);
1613
+ break;
1614
+ case 7*16+8:
1615
+ DCT_Upsample::P_Q<7, 8>::calc(P, Q, pSrc_ptr);
1616
+ DCT_Upsample::R_S<7, 8>::calc(R, S, pSrc_ptr);
1617
+ break;
1618
+ case 8*16+8:
1619
+ DCT_Upsample::P_Q<8, 8>::calc(P, Q, pSrc_ptr);
1620
+ DCT_Upsample::R_S<8, 8>::calc(R, S, pSrc_ptr);
1621
+ break;
1622
+ default:
1623
+ JPGD_ASSERT(false);
1624
+ }
1625
+
1626
+ DCT_Upsample::Matrix44 a(P + Q); P -= Q;
1627
+ DCT_Upsample::Matrix44& b = P;
1628
+ DCT_Upsample::Matrix44 c(R + S); R -= S;
1629
+ DCT_Upsample::Matrix44& d = R;
1630
+
1631
+ DCT_Upsample::Matrix44::add_and_store(temp_block, a, c);
1632
+ idct_4x4(temp_block, pDst_ptr);
1633
+ pDst_ptr += 64;
1634
+
1635
+ DCT_Upsample::Matrix44::sub_and_store(temp_block, a, c);
1636
+ idct_4x4(temp_block, pDst_ptr);
1637
+ pDst_ptr += 64;
1638
+
1639
+ DCT_Upsample::Matrix44::add_and_store(temp_block, b, d);
1640
+ idct_4x4(temp_block, pDst_ptr);
1641
+ pDst_ptr += 64;
1642
+
1643
+ DCT_Upsample::Matrix44::sub_and_store(temp_block, b, d);
1644
+ idct_4x4(temp_block, pDst_ptr);
1645
+ pDst_ptr += 64;
1646
+
1647
+ pSrc_ptr += 64;
1648
+ }
1649
+ }
1650
+
1651
+ // Loads and dequantizes the next row of (already decoded) coefficients.
1652
+ // Progressive images only.
1653
+ void jpeg_decoder::load_next_row()
1654
+ {
1655
+ int i;
1656
+ jpgd_block_t *p;
1657
+ jpgd_quant_t *q;
1658
+ int mcu_row, mcu_block, row_block = 0;
1659
+ int component_num, component_id;
1660
+ int block_x_mcu[JPGD_MAX_COMPONENTS];
1661
+
1662
+ memset(block_x_mcu, 0, JPGD_MAX_COMPONENTS * sizeof(int));
1663
+
1664
+ for (mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++)
1665
+ {
1666
+ int block_x_mcu_ofs = 0, block_y_mcu_ofs = 0;
1667
+
1668
+ for (mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++)
1669
+ {
1670
+ component_id = m_mcu_org[mcu_block];
1671
+ q = m_quant[m_comp_quant[component_id]];
1672
+
1673
+ p = m_pMCU_coefficients + 64 * mcu_block;
1674
+
1675
+ jpgd_block_t* pAC = coeff_buf_getp(m_ac_coeffs[component_id], block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
1676
+ jpgd_block_t* pDC = coeff_buf_getp(m_dc_coeffs[component_id], block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
1677
+ p[0] = pDC[0];
1678
+ memcpy(&p[1], &pAC[1], 63 * sizeof(jpgd_block_t));
1679
+
1680
+ for (i = 63; i > 0; i--)
1681
+ if (p[g_ZAG[i]])
1682
+ break;
1683
+
1684
+ m_mcu_block_max_zag[mcu_block] = i + 1;
1685
+
1686
+ for ( ; i >= 0; i--)
1687
+ if (p[g_ZAG[i]])
1688
+ p[g_ZAG[i]] = static_cast<jpgd_block_t>(p[g_ZAG[i]] * q[i]);
1689
+
1690
+ row_block++;
1691
+
1692
+ if (m_comps_in_scan == 1)
1693
+ block_x_mcu[component_id]++;
1694
+ else
1695
+ {
1696
+ if (++block_x_mcu_ofs == m_comp_h_samp[component_id])
1697
+ {
1698
+ block_x_mcu_ofs = 0;
1699
+
1700
+ if (++block_y_mcu_ofs == m_comp_v_samp[component_id])
1701
+ {
1702
+ block_y_mcu_ofs = 0;
1703
+
1704
+ block_x_mcu[component_id] += m_comp_h_samp[component_id];
1705
+ }
1706
+ }
1707
+ }
1708
+ }
1709
+
1710
+ if (m_freq_domain_chroma_upsample)
1711
+ transform_mcu_expand(mcu_row);
1712
+ else
1713
+ transform_mcu(mcu_row);
1714
+ }
1715
+
1716
+ if (m_comps_in_scan == 1)
1717
+ m_block_y_mcu[m_comp_list[0]]++;
1718
+ else
1719
+ {
1720
+ for (component_num = 0; component_num < m_comps_in_scan; component_num++)
1721
+ {
1722
+ component_id = m_comp_list[component_num];
1723
+
1724
+ m_block_y_mcu[component_id] += m_comp_v_samp[component_id];
1725
+ }
1726
+ }
1727
+ }
1728
+
1729
+ // Restart interval processing.
1730
+ void jpeg_decoder::process_restart()
1731
+ {
1732
+ int i;
1733
+ int c = 0;
1734
+
1735
+ // Align to a byte boundry
1736
+ // FIXME: Is this really necessary? get_bits_no_markers() never reads in markers!
1737
+ //get_bits_no_markers(m_bits_left & 7);
1738
+
1739
+ // Let's scan a little bit to find the marker, but not _too_ far.
1740
+ // 1536 is a "fudge factor" that determines how much to scan.
1741
+ for (i = 1536; i > 0; i--)
1742
+ if (get_char() == 0xFF)
1743
+ break;
1744
+
1745
+ if (i == 0)
1746
+ stop_decoding(JPGD_BAD_RESTART_MARKER);
1747
+
1748
+ for ( ; i > 0; i--)
1749
+ if ((c = get_char()) != 0xFF)
1750
+ break;
1751
+
1752
+ if (i == 0)
1753
+ stop_decoding(JPGD_BAD_RESTART_MARKER);
1754
+
1755
+ // Is it the expected marker? If not, something bad happened.
1756
+ if (c != (m_next_restart_num + M_RST0))
1757
+ stop_decoding(JPGD_BAD_RESTART_MARKER);
1758
+
1759
+ // Reset each component's DC prediction values.
1760
+ memset(&m_last_dc_val, 0, m_comps_in_frame * sizeof(uint));
1761
+
1762
+ m_eob_run = 0;
1763
+
1764
+ m_restarts_left = m_restart_interval;
1765
+
1766
+ m_next_restart_num = (m_next_restart_num + 1) & 7;
1767
+
1768
+ // Get the bit buffer going again...
1769
+
1770
+ m_bits_left = 16;
1771
+ get_bits_no_markers(16);
1772
+ get_bits_no_markers(16);
1773
+ }
1774
+
1775
+ static inline int dequantize_ac(int c, int q) { c *= q; return c; }
1776
+
1777
+ // Decodes and dequantizes the next row of coefficients.
1778
+ void jpeg_decoder::decode_next_row()
1779
+ {
1780
+ int row_block = 0;
1781
+
1782
+ for (int mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++)
1783
+ {
1784
+ if ((m_restart_interval) && (m_restarts_left == 0))
1785
+ process_restart();
1786
+
1787
+ jpgd_block_t* p = m_pMCU_coefficients;
1788
+ for (int mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++, p += 64)
1789
+ {
1790
+ int component_id = m_mcu_org[mcu_block];
1791
+ jpgd_quant_t* q = m_quant[m_comp_quant[component_id]];
1792
+
1793
+ int r, s;
1794
+ s = huff_decode(m_pHuff_tabs[m_comp_dc_tab[component_id]], r);
1795
+ s = HUFF_EXTEND(r, s);
1796
+
1797
+ m_last_dc_val[component_id] = (s += m_last_dc_val[component_id]);
1798
+
1799
+ p[0] = static_cast<jpgd_block_t>(s * q[0]);
1800
+
1801
+ int prev_num_set = m_mcu_block_max_zag[mcu_block];
1802
+
1803
+ huff_tables *pH = m_pHuff_tabs[m_comp_ac_tab[component_id]];
1804
+
1805
+ int k;
1806
+ for (k = 1; k < 64; k++)
1807
+ {
1808
+ int extra_bits;
1809
+ s = huff_decode(pH, extra_bits);
1810
+
1811
+ r = s >> 4;
1812
+ s &= 15;
1813
+
1814
+ if (s)
1815
+ {
1816
+ if (r)
1817
+ {
1818
+ if ((k + r) > 63)
1819
+ stop_decoding(JPGD_DECODE_ERROR);
1820
+
1821
+ if (k < prev_num_set)
1822
+ {
1823
+ int n = JPGD_MIN(r, prev_num_set - k);
1824
+ int kt = k;
1825
+ while (n--)
1826
+ p[g_ZAG[kt++]] = 0;
1827
+ }
1828
+
1829
+ k += r;
1830
+ }
1831
+
1832
+ s = HUFF_EXTEND(extra_bits, s);
1833
+
1834
+ JPGD_ASSERT(k < 64);
1835
+
1836
+ p[g_ZAG[k]] = static_cast<jpgd_block_t>(dequantize_ac(s, q[k])); //s * q[k];
1837
+ }
1838
+ else
1839
+ {
1840
+ if (r == 15)
1841
+ {
1842
+ if ((k + 16) > 64)
1843
+ stop_decoding(JPGD_DECODE_ERROR);
1844
+
1845
+ if (k < prev_num_set)
1846
+ {
1847
+ int n = JPGD_MIN(16, prev_num_set - k);
1848
+ int kt = k;
1849
+ while (n--)
1850
+ {
1851
+ JPGD_ASSERT(kt <= 63);
1852
+ p[g_ZAG[kt++]] = 0;
1853
+ }
1854
+ }
1855
+
1856
+ k += 16 - 1; // - 1 because the loop counter is k
1857
+ // BEGIN EPIC MOD
1858
+ JPGD_ASSERT(k < 64 && p[g_ZAG[k]] == 0);
1859
+ // END EPIC MOD
1860
+ }
1861
+ else
1862
+ break;
1863
+ }
1864
+ }
1865
+
1866
+ if (k < prev_num_set)
1867
+ {
1868
+ int kt = k;
1869
+ while (kt < prev_num_set)
1870
+ p[g_ZAG[kt++]] = 0;
1871
+ }
1872
+
1873
+ m_mcu_block_max_zag[mcu_block] = k;
1874
+
1875
+ row_block++;
1876
+ }
1877
+
1878
+ if (m_freq_domain_chroma_upsample)
1879
+ transform_mcu_expand(mcu_row);
1880
+ else
1881
+ transform_mcu(mcu_row);
1882
+
1883
+ m_restarts_left--;
1884
+ }
1885
+ }
1886
+
1887
+ // YCbCr H1V1 (1x1:1:1, 3 m_blocks per MCU) to RGB
1888
+ void jpeg_decoder::H1V1Convert()
1889
+ {
1890
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
1891
+ uint8 *d = m_pScan_line_0;
1892
+ uint8 *s = m_pSample_buf + row * 8;
1893
+
1894
+ for (int i = m_max_mcus_per_row; i > 0; i--)
1895
+ {
1896
+ for (int j = 0; j < 8; j++)
1897
+ {
1898
+ int y = s[j];
1899
+ int cb = s[64+j];
1900
+ int cr = s[128+j];
1901
+
1902
+ if (jpg_format == ERGBFormatJPG::BGRA)
1903
+ {
1904
+ d[0] = clamp(y + m_cbb[cb]);
1905
+ d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
1906
+ d[2] = clamp(y + m_crr[cr]);
1907
+ d[3] = 255;
1908
+ }
1909
+ else
1910
+ {
1911
+ d[0] = clamp(y + m_crr[cr]);
1912
+ d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
1913
+ d[2] = clamp(y + m_cbb[cb]);
1914
+ d[3] = 255;
1915
+ }
1916
+ d += 4;
1917
+ }
1918
+
1919
+ s += 64*3;
1920
+ }
1921
+ }
1922
+
1923
+ // YCbCr H2V1 (2x1:1:1, 4 m_blocks per MCU) to RGB
1924
+ void jpeg_decoder::H2V1Convert()
1925
+ {
1926
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
1927
+ uint8 *d0 = m_pScan_line_0;
1928
+ uint8 *y = m_pSample_buf + row * 8;
1929
+ uint8 *c = m_pSample_buf + 2*64 + row * 8;
1930
+
1931
+ for (int i = m_max_mcus_per_row; i > 0; i--)
1932
+ {
1933
+ for (int l = 0; l < 2; l++)
1934
+ {
1935
+ for (int j = 0; j < 4; j++)
1936
+ {
1937
+ int cb = c[0];
1938
+ int cr = c[64];
1939
+
1940
+ int rc = m_crr[cr];
1941
+ int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
1942
+ int bc = m_cbb[cb];
1943
+
1944
+ int yy = y[j<<1];
1945
+ if (jpg_format == ERGBFormatJPG::BGRA)
1946
+ {
1947
+ d0[0] = clamp(yy+bc);
1948
+ d0[1] = clamp(yy+gc);
1949
+ d0[2] = clamp(yy+rc);
1950
+ d0[3] = 255;
1951
+ yy = y[(j<<1)+1];
1952
+ d0[4] = clamp(yy+bc);
1953
+ d0[5] = clamp(yy+gc);
1954
+ d0[6] = clamp(yy+rc);
1955
+ d0[7] = 255;
1956
+ }
1957
+ else
1958
+ {
1959
+ d0[0] = clamp(yy+rc);
1960
+ d0[1] = clamp(yy+gc);
1961
+ d0[2] = clamp(yy+bc);
1962
+ d0[3] = 255;
1963
+ yy = y[(j<<1)+1];
1964
+ d0[4] = clamp(yy+rc);
1965
+ d0[5] = clamp(yy+gc);
1966
+ d0[6] = clamp(yy+bc);
1967
+ d0[7] = 255;
1968
+ }
1969
+
1970
+ d0 += 8;
1971
+
1972
+ c++;
1973
+ }
1974
+ y += 64;
1975
+ }
1976
+
1977
+ y += 64*4 - 64*2;
1978
+ c += 64*4 - 8;
1979
+ }
1980
+ }
1981
+
1982
+ // YCbCr H2V1 (1x2:1:1, 4 m_blocks per MCU) to RGB
1983
+ void jpeg_decoder::H1V2Convert()
1984
+ {
1985
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
1986
+ uint8 *d0 = m_pScan_line_0;
1987
+ uint8 *d1 = m_pScan_line_1;
1988
+ uint8 *y;
1989
+ uint8 *c;
1990
+
1991
+ if (row < 8)
1992
+ y = m_pSample_buf + row * 8;
1993
+ else
1994
+ y = m_pSample_buf + 64*1 + (row & 7) * 8;
1995
+
1996
+ c = m_pSample_buf + 64*2 + (row >> 1) * 8;
1997
+
1998
+ for (int i = m_max_mcus_per_row; i > 0; i--)
1999
+ {
2000
+ for (int j = 0; j < 8; j++)
2001
+ {
2002
+ int cb = c[0+j];
2003
+ int cr = c[64+j];
2004
+
2005
+ int rc = m_crr[cr];
2006
+ int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
2007
+ int bc = m_cbb[cb];
2008
+
2009
+ int yy = y[j];
2010
+ if (jpg_format == ERGBFormatJPG::BGRA)
2011
+ {
2012
+ d0[0] = clamp(yy+bc);
2013
+ d0[1] = clamp(yy+gc);
2014
+ d0[2] = clamp(yy+rc);
2015
+ d0[3] = 255;
2016
+ yy = y[8+j];
2017
+ d1[0] = clamp(yy+bc);
2018
+ d1[1] = clamp(yy+gc);
2019
+ d1[2] = clamp(yy+rc);
2020
+ d1[3] = 255;
2021
+ }
2022
+ else
2023
+ {
2024
+ d0[0] = clamp(yy+rc);
2025
+ d0[1] = clamp(yy+gc);
2026
+ d0[2] = clamp(yy+bc);
2027
+ d0[3] = 255;
2028
+ yy = y[8+j];
2029
+ d1[0] = clamp(yy+rc);
2030
+ d1[1] = clamp(yy+gc);
2031
+ d1[2] = clamp(yy+bc);
2032
+ d1[3] = 255;
2033
+ }
2034
+
2035
+ d0 += 4;
2036
+ d1 += 4;
2037
+ }
2038
+
2039
+ y += 64*4;
2040
+ c += 64*4;
2041
+ }
2042
+ }
2043
+
2044
+ // YCbCr H2V2 (2x2:1:1, 6 m_blocks per MCU) to RGB
2045
+ void jpeg_decoder::H2V2Convert()
2046
+ {
2047
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
2048
+ uint8 *d0 = m_pScan_line_0;
2049
+ uint8 *d1 = m_pScan_line_1;
2050
+ uint8 *y;
2051
+ uint8 *c;
2052
+
2053
+ if (row < 8)
2054
+ y = m_pSample_buf + row * 8;
2055
+ else
2056
+ y = m_pSample_buf + 64*2 + (row & 7) * 8;
2057
+
2058
+ c = m_pSample_buf + 64*4 + (row >> 1) * 8;
2059
+
2060
+ for (int i = m_max_mcus_per_row; i > 0; i--)
2061
+ {
2062
+ for (int l = 0; l < 2; l++)
2063
+ {
2064
+ for (int j = 0; j < 8; j += 2)
2065
+ {
2066
+ int cb = c[0];
2067
+ int cr = c[64];
2068
+
2069
+ int rc = m_crr[cr];
2070
+ int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
2071
+ int bc = m_cbb[cb];
2072
+
2073
+ int yy = y[j];
2074
+ if (jpg_format == ERGBFormatJPG::BGRA)
2075
+ {
2076
+ d0[0] = clamp(yy+bc);
2077
+ d0[1] = clamp(yy+gc);
2078
+ d0[2] = clamp(yy+rc);
2079
+ d0[3] = 255;
2080
+ yy = y[j+1];
2081
+ d0[4] = clamp(yy+bc);
2082
+ d0[5] = clamp(yy+gc);
2083
+ d0[6] = clamp(yy+rc);
2084
+ d0[7] = 255;
2085
+ yy = y[j+8];
2086
+ d1[0] = clamp(yy+bc);
2087
+ d1[1] = clamp(yy+gc);
2088
+ d1[2] = clamp(yy+rc);
2089
+ d1[3] = 255;
2090
+ yy = y[j+8+1];
2091
+ d1[4] = clamp(yy+bc);
2092
+ d1[5] = clamp(yy+gc);
2093
+ d1[6] = clamp(yy+rc);
2094
+ d1[7] = 255;
2095
+ }
2096
+ else
2097
+ {
2098
+ d0[0] = clamp(yy+rc);
2099
+ d0[1] = clamp(yy+gc);
2100
+ d0[2] = clamp(yy+bc);
2101
+ d0[3] = 255;
2102
+ yy = y[j+1];
2103
+ d0[4] = clamp(yy+rc);
2104
+ d0[5] = clamp(yy+gc);
2105
+ d0[6] = clamp(yy+bc);
2106
+ d0[7] = 255;
2107
+ yy = y[j+8];
2108
+ d1[0] = clamp(yy+rc);
2109
+ d1[1] = clamp(yy+gc);
2110
+ d1[2] = clamp(yy+bc);
2111
+ d1[3] = 255;
2112
+ yy = y[j+8+1];
2113
+ d1[4] = clamp(yy+rc);
2114
+ d1[5] = clamp(yy+gc);
2115
+ d1[6] = clamp(yy+bc);
2116
+ d1[7] = 255;
2117
+ }
2118
+
2119
+ d0 += 8;
2120
+ d1 += 8;
2121
+
2122
+ c++;
2123
+ }
2124
+ y += 64;
2125
+ }
2126
+
2127
+ y += 64*6 - 64*2;
2128
+ c += 64*6 - 8;
2129
+ }
2130
+ }
2131
+
2132
+ // Y (1 block per MCU) to 8-bit grayscale
2133
+ void jpeg_decoder::gray_convert()
2134
+ {
2135
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
2136
+ uint8 *d = m_pScan_line_0;
2137
+ uint8 *s = m_pSample_buf + row * 8;
2138
+
2139
+ for (int i = m_max_mcus_per_row; i > 0; i--)
2140
+ {
2141
+ *(uint *)d = *(uint *)s;
2142
+ *(uint *)(&d[4]) = *(uint *)(&s[4]);
2143
+
2144
+ s += 64;
2145
+ d += 8;
2146
+ }
2147
+ }
2148
+
2149
+ void jpeg_decoder::expanded_convert()
2150
+ {
2151
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
2152
+
2153
+ uint8* Py = m_pSample_buf + (row / 8) * 64 * m_comp_h_samp[0] + (row & 7) * 8;
2154
+
2155
+ uint8* d = m_pScan_line_0;
2156
+
2157
+ for (int i = m_max_mcus_per_row; i > 0; i--)
2158
+ {
2159
+ for (int k = 0; k < m_max_mcu_x_size; k += 8)
2160
+ {
2161
+ const int Y_ofs = k * 8;
2162
+ const int Cb_ofs = Y_ofs + 64 * m_expanded_blocks_per_component;
2163
+ const int Cr_ofs = Y_ofs + 64 * m_expanded_blocks_per_component * 2;
2164
+ for (int j = 0; j < 8; j++)
2165
+ {
2166
+ int y = Py[Y_ofs + j];
2167
+ int cb = Py[Cb_ofs + j];
2168
+ int cr = Py[Cr_ofs + j];
2169
+
2170
+ if (jpg_format == ERGBFormatJPG::BGRA)
2171
+ {
2172
+ d[0] = clamp(y + m_cbb[cb]);
2173
+ d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
2174
+ d[2] = clamp(y + m_crr[cr]);
2175
+ d[3] = 255;
2176
+ }
2177
+ else
2178
+ {
2179
+ d[0] = clamp(y + m_crr[cr]);
2180
+ d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
2181
+ d[2] = clamp(y + m_cbb[cb]);
2182
+ d[3] = 255;
2183
+ }
2184
+
2185
+ d += 4;
2186
+ }
2187
+ }
2188
+
2189
+ Py += 64 * m_expanded_blocks_per_mcu;
2190
+ }
2191
+ }
2192
+
2193
+ // Find end of image (EOI) marker, so we can return to the user the exact size of the input stream.
2194
+ void jpeg_decoder::find_eoi()
2195
+ {
2196
+ if (!m_progressive_flag)
2197
+ {
2198
+ // Attempt to read the EOI marker.
2199
+ //get_bits_no_markers(m_bits_left & 7);
2200
+
2201
+ // Prime the bit buffer
2202
+ m_bits_left = 16;
2203
+ get_bits(16);
2204
+ get_bits(16);
2205
+
2206
+ // The next marker _should_ be EOI
2207
+ process_markers();
2208
+ }
2209
+
2210
+ m_total_bytes_read -= m_in_buf_left;
2211
+ }
2212
+
2213
+ int jpeg_decoder::decode(const void** pScan_line, uint* pScan_line_len)
2214
+ {
2215
+ if ((m_error_code) || (!m_ready_flag))
2216
+ return JPGD_FAILED;
2217
+
2218
+ if (m_total_lines_left == 0)
2219
+ return JPGD_DONE;
2220
+
2221
+ if (m_mcu_lines_left == 0)
2222
+ {
2223
+ if (setjmp(m_jmp_state))
2224
+ return JPGD_FAILED;
2225
+
2226
+ if (m_progressive_flag)
2227
+ load_next_row();
2228
+ else
2229
+ decode_next_row();
2230
+
2231
+ // Find the EOI marker if that was the last row.
2232
+ if (m_total_lines_left <= m_max_mcu_y_size)
2233
+ find_eoi();
2234
+
2235
+ m_mcu_lines_left = m_max_mcu_y_size;
2236
+ }
2237
+
2238
+ if (m_freq_domain_chroma_upsample)
2239
+ {
2240
+ expanded_convert();
2241
+ *pScan_line = m_pScan_line_0;
2242
+ }
2243
+ else
2244
+ {
2245
+ switch (m_scan_type)
2246
+ {
2247
+ case JPGD_YH2V2:
2248
+ {
2249
+ if ((m_mcu_lines_left & 1) == 0)
2250
+ {
2251
+ H2V2Convert();
2252
+ *pScan_line = m_pScan_line_0;
2253
+ }
2254
+ else
2255
+ *pScan_line = m_pScan_line_1;
2256
+
2257
+ break;
2258
+ }
2259
+ case JPGD_YH2V1:
2260
+ {
2261
+ H2V1Convert();
2262
+ *pScan_line = m_pScan_line_0;
2263
+ break;
2264
+ }
2265
+ case JPGD_YH1V2:
2266
+ {
2267
+ if ((m_mcu_lines_left & 1) == 0)
2268
+ {
2269
+ H1V2Convert();
2270
+ *pScan_line = m_pScan_line_0;
2271
+ }
2272
+ else
2273
+ *pScan_line = m_pScan_line_1;
2274
+
2275
+ break;
2276
+ }
2277
+ case JPGD_YH1V1:
2278
+ {
2279
+ H1V1Convert();
2280
+ *pScan_line = m_pScan_line_0;
2281
+ break;
2282
+ }
2283
+ case JPGD_GRAYSCALE:
2284
+ {
2285
+ gray_convert();
2286
+ *pScan_line = m_pScan_line_0;
2287
+
2288
+ break;
2289
+ }
2290
+ }
2291
+ }
2292
+
2293
+ *pScan_line_len = m_real_dest_bytes_per_scan_line;
2294
+
2295
+ m_mcu_lines_left--;
2296
+ m_total_lines_left--;
2297
+
2298
+ return JPGD_SUCCESS;
2299
+ }
2300
+
2301
+ // Creates the tables needed for efficient Huffman decoding.
2302
+ void jpeg_decoder::make_huff_table(int index, huff_tables *pH)
2303
+ {
2304
+ int p, i, l, si;
2305
+ uint8 huffsize[257];
2306
+ uint huffcode[257];
2307
+ uint code;
2308
+ uint subtree;
2309
+ int code_size;
2310
+ int lastp;
2311
+ int nextfreeentry;
2312
+ int currententry;
2313
+
2314
+ pH->ac_table = m_huff_ac[index] != 0;
2315
+
2316
+ p = 0;
2317
+
2318
+ for (l = 1; l <= 16; l++)
2319
+ {
2320
+ for (i = 1; i <= m_huff_num[index][l]; i++)
2321
+ huffsize[p++] = static_cast<uint8>(l);
2322
+ }
2323
+
2324
+ huffsize[p] = 0;
2325
+
2326
+ lastp = p;
2327
+
2328
+ code = 0;
2329
+ si = huffsize[0];
2330
+ p = 0;
2331
+
2332
+ while (huffsize[p])
2333
+ {
2334
+ while (huffsize[p] == si)
2335
+ {
2336
+ huffcode[p++] = code;
2337
+ code++;
2338
+ }
2339
+
2340
+ code <<= 1;
2341
+ si++;
2342
+ }
2343
+
2344
+ memset(pH->look_up, 0, sizeof(pH->look_up));
2345
+ memset(pH->look_up2, 0, sizeof(pH->look_up2));
2346
+ memset(pH->tree, 0, sizeof(pH->tree));
2347
+ memset(pH->code_size, 0, sizeof(pH->code_size));
2348
+
2349
+ nextfreeentry = -1;
2350
+
2351
+ p = 0;
2352
+
2353
+ while (p < lastp)
2354
+ {
2355
+ i = m_huff_val[index][p];
2356
+ code = huffcode[p];
2357
+ code_size = huffsize[p];
2358
+
2359
+ pH->code_size[i] = static_cast<uint8>(code_size);
2360
+
2361
+ if (code_size <= 8)
2362
+ {
2363
+ code <<= (8 - code_size);
2364
+
2365
+ for (l = 1 << (8 - code_size); l > 0; l--)
2366
+ {
2367
+ JPGD_ASSERT(i < 256);
2368
+
2369
+ pH->look_up[code] = i;
2370
+
2371
+ bool has_extrabits = false;
2372
+ int extra_bits = 0;
2373
+ int num_extra_bits = i & 15;
2374
+
2375
+ int bits_to_fetch = code_size;
2376
+ if (num_extra_bits)
2377
+ {
2378
+ int total_codesize = code_size + num_extra_bits;
2379
+ if (total_codesize <= 8)
2380
+ {
2381
+ has_extrabits = true;
2382
+ extra_bits = ((1 << num_extra_bits) - 1) & (code >> (8 - total_codesize));
2383
+ JPGD_ASSERT(extra_bits <= 0x7FFF);
2384
+ bits_to_fetch += num_extra_bits;
2385
+ }
2386
+ }
2387
+
2388
+ if (!has_extrabits)
2389
+ pH->look_up2[code] = i | (bits_to_fetch << 8);
2390
+ else
2391
+ pH->look_up2[code] = i | 0x8000 | (extra_bits << 16) | (bits_to_fetch << 8);
2392
+
2393
+ code++;
2394
+ }
2395
+ }
2396
+ else
2397
+ {
2398
+ subtree = (code >> (code_size - 8)) & 0xFF;
2399
+
2400
+ currententry = pH->look_up[subtree];
2401
+
2402
+ if (currententry == 0)
2403
+ {
2404
+ pH->look_up[subtree] = currententry = nextfreeentry;
2405
+ pH->look_up2[subtree] = currententry = nextfreeentry;
2406
+
2407
+ nextfreeentry -= 2;
2408
+ }
2409
+
2410
+ code <<= (16 - (code_size - 8));
2411
+
2412
+ for (l = code_size; l > 9; l--)
2413
+ {
2414
+ if ((code & 0x8000) == 0)
2415
+ currententry--;
2416
+
2417
+ if (pH->tree[-currententry - 1] == 0)
2418
+ {
2419
+ pH->tree[-currententry - 1] = nextfreeentry;
2420
+
2421
+ currententry = nextfreeentry;
2422
+
2423
+ nextfreeentry -= 2;
2424
+ }
2425
+ else
2426
+ currententry = pH->tree[-currententry - 1];
2427
+
2428
+ code <<= 1;
2429
+ }
2430
+
2431
+ if ((code & 0x8000) == 0)
2432
+ currententry--;
2433
+
2434
+ pH->tree[-currententry - 1] = i;
2435
+ }
2436
+
2437
+ p++;
2438
+ }
2439
+ }
2440
+
2441
+ // Verifies the quantization tables needed for this scan are available.
2442
+ void jpeg_decoder::check_quant_tables()
2443
+ {
2444
+ for (int i = 0; i < m_comps_in_scan; i++)
2445
+ if (m_quant[m_comp_quant[m_comp_list[i]]] == NULL)
2446
+ stop_decoding(JPGD_UNDEFINED_QUANT_TABLE);
2447
+ }
2448
+
2449
+ // Verifies that all the Huffman tables needed for this scan are available.
2450
+ void jpeg_decoder::check_huff_tables()
2451
+ {
2452
+ for (int i = 0; i < m_comps_in_scan; i++)
2453
+ {
2454
+ if ((m_spectral_start == 0) && (m_huff_num[m_comp_dc_tab[m_comp_list[i]]] == NULL))
2455
+ stop_decoding(JPGD_UNDEFINED_HUFF_TABLE);
2456
+
2457
+ if ((m_spectral_end > 0) && (m_huff_num[m_comp_ac_tab[m_comp_list[i]]] == NULL))
2458
+ stop_decoding(JPGD_UNDEFINED_HUFF_TABLE);
2459
+ }
2460
+
2461
+ for (int i = 0; i < JPGD_MAX_HUFF_TABLES; i++)
2462
+ if (m_huff_num[i])
2463
+ {
2464
+ if (!m_pHuff_tabs[i])
2465
+ m_pHuff_tabs[i] = (huff_tables *)alloc(sizeof(huff_tables));
2466
+
2467
+ make_huff_table(i, m_pHuff_tabs[i]);
2468
+ }
2469
+ }
2470
+
2471
+ // Determines the component order inside each MCU.
2472
+ // Also calcs how many MCU's are on each row, etc.
2473
+ void jpeg_decoder::calc_mcu_block_order()
2474
+ {
2475
+ int component_num, component_id;
2476
+ int max_h_samp = 0, max_v_samp = 0;
2477
+
2478
+ for (component_id = 0; component_id < m_comps_in_frame; component_id++)
2479
+ {
2480
+ if (m_comp_h_samp[component_id] > max_h_samp)
2481
+ max_h_samp = m_comp_h_samp[component_id];
2482
+
2483
+ if (m_comp_v_samp[component_id] > max_v_samp)
2484
+ max_v_samp = m_comp_v_samp[component_id];
2485
+ }
2486
+
2487
+ for (component_id = 0; component_id < m_comps_in_frame; component_id++)
2488
+ {
2489
+ m_comp_h_blocks[component_id] = ((((m_image_x_size * m_comp_h_samp[component_id]) + (max_h_samp - 1)) / max_h_samp) + 7) / 8;
2490
+ m_comp_v_blocks[component_id] = ((((m_image_y_size * m_comp_v_samp[component_id]) + (max_v_samp - 1)) / max_v_samp) + 7) / 8;
2491
+ }
2492
+
2493
+ if (m_comps_in_scan == 1)
2494
+ {
2495
+ m_mcus_per_row = m_comp_h_blocks[m_comp_list[0]];
2496
+ m_mcus_per_col = m_comp_v_blocks[m_comp_list[0]];
2497
+ }
2498
+ else
2499
+ {
2500
+ m_mcus_per_row = (((m_image_x_size + 7) / 8) + (max_h_samp - 1)) / max_h_samp;
2501
+ m_mcus_per_col = (((m_image_y_size + 7) / 8) + (max_v_samp - 1)) / max_v_samp;
2502
+ }
2503
+
2504
+ if (m_comps_in_scan == 1)
2505
+ {
2506
+ m_mcu_org[0] = m_comp_list[0];
2507
+
2508
+ m_blocks_per_mcu = 1;
2509
+ }
2510
+ else
2511
+ {
2512
+ m_blocks_per_mcu = 0;
2513
+
2514
+ for (component_num = 0; component_num < m_comps_in_scan; component_num++)
2515
+ {
2516
+ int num_blocks;
2517
+
2518
+ component_id = m_comp_list[component_num];
2519
+
2520
+ num_blocks = m_comp_h_samp[component_id] * m_comp_v_samp[component_id];
2521
+
2522
+ while (num_blocks--)
2523
+ m_mcu_org[m_blocks_per_mcu++] = component_id;
2524
+ }
2525
+ }
2526
+ }
2527
+
2528
+ // Starts a new scan.
2529
+ int jpeg_decoder::init_scan()
2530
+ {
2531
+ if (!locate_sos_marker())
2532
+ return JPGD_FALSE;
2533
+
2534
+ calc_mcu_block_order();
2535
+
2536
+ check_huff_tables();
2537
+
2538
+ check_quant_tables();
2539
+
2540
+ memset(m_last_dc_val, 0, m_comps_in_frame * sizeof(uint));
2541
+
2542
+ m_eob_run = 0;
2543
+
2544
+ if (m_restart_interval)
2545
+ {
2546
+ m_restarts_left = m_restart_interval;
2547
+ m_next_restart_num = 0;
2548
+ }
2549
+
2550
+ fix_in_buffer();
2551
+
2552
+ return JPGD_TRUE;
2553
+ }
2554
+
2555
+ // Starts a frame. Determines if the number of components or sampling factors
2556
+ // are supported.
2557
+ void jpeg_decoder::init_frame()
2558
+ {
2559
+ int i;
2560
+
2561
+ if (m_comps_in_frame == 1)
2562
+ {
2563
+ if ((m_comp_h_samp[0] != 1) || (m_comp_v_samp[0] != 1))
2564
+ stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
2565
+
2566
+ m_scan_type = JPGD_GRAYSCALE;
2567
+ m_max_blocks_per_mcu = 1;
2568
+ m_max_mcu_x_size = 8;
2569
+ m_max_mcu_y_size = 8;
2570
+ }
2571
+ else if (m_comps_in_frame == 3)
2572
+ {
2573
+ if ( ((m_comp_h_samp[1] != 1) || (m_comp_v_samp[1] != 1)) ||
2574
+ ((m_comp_h_samp[2] != 1) || (m_comp_v_samp[2] != 1)) )
2575
+ stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
2576
+
2577
+ if ((m_comp_h_samp[0] == 1) && (m_comp_v_samp[0] == 1))
2578
+ {
2579
+ m_scan_type = JPGD_YH1V1;
2580
+
2581
+ m_max_blocks_per_mcu = 3;
2582
+ m_max_mcu_x_size = 8;
2583
+ m_max_mcu_y_size = 8;
2584
+ }
2585
+ else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 1))
2586
+ {
2587
+ m_scan_type = JPGD_YH2V1;
2588
+ m_max_blocks_per_mcu = 4;
2589
+ m_max_mcu_x_size = 16;
2590
+ m_max_mcu_y_size = 8;
2591
+ }
2592
+ else if ((m_comp_h_samp[0] == 1) && (m_comp_v_samp[0] == 2))
2593
+ {
2594
+ m_scan_type = JPGD_YH1V2;
2595
+ m_max_blocks_per_mcu = 4;
2596
+ m_max_mcu_x_size = 8;
2597
+ m_max_mcu_y_size = 16;
2598
+ }
2599
+ else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 2))
2600
+ {
2601
+ m_scan_type = JPGD_YH2V2;
2602
+ m_max_blocks_per_mcu = 6;
2603
+ m_max_mcu_x_size = 16;
2604
+ m_max_mcu_y_size = 16;
2605
+ }
2606
+ else
2607
+ stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
2608
+ }
2609
+ else
2610
+ stop_decoding(JPGD_UNSUPPORTED_COLORSPACE);
2611
+
2612
+ m_max_mcus_per_row = (m_image_x_size + (m_max_mcu_x_size - 1)) / m_max_mcu_x_size;
2613
+ m_max_mcus_per_col = (m_image_y_size + (m_max_mcu_y_size - 1)) / m_max_mcu_y_size;
2614
+
2615
+ // These values are for the *destination* pixels: after conversion.
2616
+ if (m_scan_type == JPGD_GRAYSCALE)
2617
+ m_dest_bytes_per_pixel = 1;
2618
+ else
2619
+ m_dest_bytes_per_pixel = 4;
2620
+
2621
+ m_dest_bytes_per_scan_line = ((m_image_x_size + 15) & 0xFFF0) * m_dest_bytes_per_pixel;
2622
+
2623
+ m_real_dest_bytes_per_scan_line = (m_image_x_size * m_dest_bytes_per_pixel);
2624
+
2625
+ // Initialize two scan line buffers.
2626
+ m_pScan_line_0 = (uint8 *)alloc(m_dest_bytes_per_scan_line, true);
2627
+ if ((m_scan_type == JPGD_YH1V2) || (m_scan_type == JPGD_YH2V2))
2628
+ m_pScan_line_1 = (uint8 *)alloc(m_dest_bytes_per_scan_line, true);
2629
+
2630
+ m_max_blocks_per_row = m_max_mcus_per_row * m_max_blocks_per_mcu;
2631
+
2632
+ // Should never happen
2633
+ if (m_max_blocks_per_row > JPGD_MAX_BLOCKS_PER_ROW)
2634
+ stop_decoding(JPGD_ASSERTION_ERROR);
2635
+
2636
+ // Allocate the coefficient buffer, enough for one MCU
2637
+ m_pMCU_coefficients = (jpgd_block_t*)alloc(m_max_blocks_per_mcu * 64 * sizeof(jpgd_block_t));
2638
+
2639
+ for (i = 0; i < m_max_blocks_per_mcu; i++)
2640
+ m_mcu_block_max_zag[i] = 64;
2641
+
2642
+ m_expanded_blocks_per_component = m_comp_h_samp[0] * m_comp_v_samp[0];
2643
+ m_expanded_blocks_per_mcu = m_expanded_blocks_per_component * m_comps_in_frame;
2644
+ m_expanded_blocks_per_row = m_max_mcus_per_row * m_expanded_blocks_per_mcu;
2645
+ // Freq. domain chroma upsampling is only supported for H2V2 subsampling factor.
2646
+ // BEGIN EPIC MOD
2647
+ #if JPGD_SUPPORT_FREQ_DOMAIN_UPSAMPLING
2648
+ m_freq_domain_chroma_upsample = (m_expanded_blocks_per_mcu == 4*3);
2649
+ #else
2650
+ m_freq_domain_chroma_upsample = 0;
2651
+ #endif
2652
+ // END EPIC MOD
2653
+
2654
+ if (m_freq_domain_chroma_upsample)
2655
+ m_pSample_buf = (uint8 *)alloc(m_expanded_blocks_per_row * 64);
2656
+ else
2657
+ m_pSample_buf = (uint8 *)alloc(m_max_blocks_per_row * 64);
2658
+
2659
+ m_total_lines_left = m_image_y_size;
2660
+
2661
+ m_mcu_lines_left = 0;
2662
+
2663
+ create_look_ups();
2664
+ }
2665
+
2666
+ // The coeff_buf series of methods originally stored the coefficients
2667
+ // into a "virtual" file which was located in EMS, XMS, or a disk file. A cache
2668
+ // was used to make this process more efficient. Now, we can store the entire
2669
+ // thing in RAM.
2670
+ jpeg_decoder::coeff_buf* jpeg_decoder::coeff_buf_open(int block_num_x, int block_num_y, int block_len_x, int block_len_y)
2671
+ {
2672
+ coeff_buf* cb = (coeff_buf*)alloc(sizeof(coeff_buf));
2673
+
2674
+ cb->block_num_x = block_num_x;
2675
+ cb->block_num_y = block_num_y;
2676
+ cb->block_len_x = block_len_x;
2677
+ cb->block_len_y = block_len_y;
2678
+ cb->block_size = (block_len_x * block_len_y) * sizeof(jpgd_block_t);
2679
+ cb->pData = (uint8 *)alloc(cb->block_size * block_num_x * block_num_y, true);
2680
+ return cb;
2681
+ }
2682
+
2683
+ inline jpgd_block_t *jpeg_decoder::coeff_buf_getp(coeff_buf *cb, int block_x, int block_y)
2684
+ {
2685
+ JPGD_ASSERT((block_x < cb->block_num_x) && (block_y < cb->block_num_y));
2686
+ return (jpgd_block_t *)(cb->pData + block_x * cb->block_size + block_y * (cb->block_size * cb->block_num_x));
2687
+ }
2688
+
2689
+ // The following methods decode the various types of m_blocks encountered
2690
+ // in progressively encoded images.
2691
+ void jpeg_decoder::decode_block_dc_first(jpeg_decoder *pD, int component_id, int block_x, int block_y)
2692
+ {
2693
+ int s, r;
2694
+ jpgd_block_t *p = pD->coeff_buf_getp(pD->m_dc_coeffs[component_id], block_x, block_y);
2695
+
2696
+ if ((s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_dc_tab[component_id]])) != 0)
2697
+ {
2698
+ r = pD->get_bits_no_markers(s);
2699
+ s = HUFF_EXTEND(r, s);
2700
+ }
2701
+
2702
+ pD->m_last_dc_val[component_id] = (s += pD->m_last_dc_val[component_id]);
2703
+
2704
+ p[0] = static_cast<jpgd_block_t>(s << pD->m_successive_low);
2705
+ }
2706
+
2707
+ void jpeg_decoder::decode_block_dc_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y)
2708
+ {
2709
+ if (pD->get_bits_no_markers(1))
2710
+ {
2711
+ jpgd_block_t *p = pD->coeff_buf_getp(pD->m_dc_coeffs[component_id], block_x, block_y);
2712
+
2713
+ p[0] |= (1 << pD->m_successive_low);
2714
+ }
2715
+ }
2716
+
2717
+ void jpeg_decoder::decode_block_ac_first(jpeg_decoder *pD, int component_id, int block_x, int block_y)
2718
+ {
2719
+ int k, s, r;
2720
+
2721
+ if (pD->m_eob_run)
2722
+ {
2723
+ pD->m_eob_run--;
2724
+ return;
2725
+ }
2726
+
2727
+ jpgd_block_t *p = pD->coeff_buf_getp(pD->m_ac_coeffs[component_id], block_x, block_y);
2728
+
2729
+ for (k = pD->m_spectral_start; k <= pD->m_spectral_end; k++)
2730
+ {
2731
+ s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_ac_tab[component_id]]);
2732
+
2733
+ r = s >> 4;
2734
+ s &= 15;
2735
+
2736
+ if (s)
2737
+ {
2738
+ if ((k += r) > 63)
2739
+ pD->stop_decoding(JPGD_DECODE_ERROR);
2740
+
2741
+ r = pD->get_bits_no_markers(s);
2742
+ s = HUFF_EXTEND(r, s);
2743
+
2744
+ p[g_ZAG[k]] = static_cast<jpgd_block_t>(s << pD->m_successive_low);
2745
+ }
2746
+ else
2747
+ {
2748
+ if (r == 15)
2749
+ {
2750
+ if ((k += 15) > 63)
2751
+ pD->stop_decoding(JPGD_DECODE_ERROR);
2752
+ }
2753
+ else
2754
+ {
2755
+ pD->m_eob_run = 1 << r;
2756
+
2757
+ if (r)
2758
+ pD->m_eob_run += pD->get_bits_no_markers(r);
2759
+
2760
+ pD->m_eob_run--;
2761
+
2762
+ break;
2763
+ }
2764
+ }
2765
+ }
2766
+ }
2767
+
2768
+ void jpeg_decoder::decode_block_ac_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y)
2769
+ {
2770
+ int s, k, r;
2771
+ int p1 = 1 << pD->m_successive_low;
2772
+ int m1 = (-1) << pD->m_successive_low;
2773
+ jpgd_block_t *p = pD->coeff_buf_getp(pD->m_ac_coeffs[component_id], block_x, block_y);
2774
+
2775
+ k = pD->m_spectral_start;
2776
+
2777
+ if (pD->m_eob_run == 0)
2778
+ {
2779
+ for ( ; k <= pD->m_spectral_end; k++)
2780
+ {
2781
+ s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_ac_tab[component_id]]);
2782
+
2783
+ r = s >> 4;
2784
+ s &= 15;
2785
+
2786
+ if (s)
2787
+ {
2788
+ if (s != 1)
2789
+ pD->stop_decoding(JPGD_DECODE_ERROR);
2790
+
2791
+ if (pD->get_bits_no_markers(1))
2792
+ s = p1;
2793
+ else
2794
+ s = m1;
2795
+ }
2796
+ else
2797
+ {
2798
+ if (r != 15)
2799
+ {
2800
+ pD->m_eob_run = 1 << r;
2801
+
2802
+ if (r)
2803
+ pD->m_eob_run += pD->get_bits_no_markers(r);
2804
+
2805
+ break;
2806
+ }
2807
+ }
2808
+
2809
+ do
2810
+ {
2811
+ // BEGIN EPIC MOD
2812
+ JPGD_ASSERT(k < 64);
2813
+ // END EPIC MOD
2814
+
2815
+ jpgd_block_t *this_coef = p + g_ZAG[k];
2816
+
2817
+ if (*this_coef != 0)
2818
+ {
2819
+ if (pD->get_bits_no_markers(1))
2820
+ {
2821
+ if ((*this_coef & p1) == 0)
2822
+ {
2823
+ if (*this_coef >= 0)
2824
+ *this_coef = static_cast<jpgd_block_t>(*this_coef + p1);
2825
+ else
2826
+ *this_coef = static_cast<jpgd_block_t>(*this_coef + m1);
2827
+ }
2828
+ }
2829
+ }
2830
+ else
2831
+ {
2832
+ if (--r < 0)
2833
+ break;
2834
+ }
2835
+
2836
+ k++;
2837
+
2838
+ } while (k <= pD->m_spectral_end);
2839
+
2840
+ if ((s) && (k < 64))
2841
+ {
2842
+ p[g_ZAG[k]] = static_cast<jpgd_block_t>(s);
2843
+ }
2844
+ }
2845
+ }
2846
+
2847
+ if (pD->m_eob_run > 0)
2848
+ {
2849
+ for ( ; k <= pD->m_spectral_end; k++)
2850
+ {
2851
+ // BEGIN EPIC MOD
2852
+ JPGD_ASSERT(k < 64);
2853
+ // END EPIC MOD
2854
+
2855
+ jpgd_block_t *this_coef = p + g_ZAG[k];
2856
+
2857
+ if (*this_coef != 0)
2858
+ {
2859
+ if (pD->get_bits_no_markers(1))
2860
+ {
2861
+ if ((*this_coef & p1) == 0)
2862
+ {
2863
+ if (*this_coef >= 0)
2864
+ *this_coef = static_cast<jpgd_block_t>(*this_coef + p1);
2865
+ else
2866
+ *this_coef = static_cast<jpgd_block_t>(*this_coef + m1);
2867
+ }
2868
+ }
2869
+ }
2870
+ }
2871
+
2872
+ pD->m_eob_run--;
2873
+ }
2874
+ }
2875
+
2876
+ // Decode a scan in a progressively encoded image.
2877
+ void jpeg_decoder::decode_scan(pDecode_block_func decode_block_func)
2878
+ {
2879
+ int mcu_row, mcu_col, mcu_block;
2880
+ int block_x_mcu[JPGD_MAX_COMPONENTS], m_block_y_mcu[JPGD_MAX_COMPONENTS];
2881
+
2882
+ memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
2883
+
2884
+ for (mcu_col = 0; mcu_col < m_mcus_per_col; mcu_col++)
2885
+ {
2886
+ int component_num, component_id;
2887
+
2888
+ memset(block_x_mcu, 0, sizeof(block_x_mcu));
2889
+
2890
+ for (mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++)
2891
+ {
2892
+ int block_x_mcu_ofs = 0, block_y_mcu_ofs = 0;
2893
+
2894
+ if ((m_restart_interval) && (m_restarts_left == 0))
2895
+ process_restart();
2896
+
2897
+ for (mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++)
2898
+ {
2899
+ component_id = m_mcu_org[mcu_block];
2900
+
2901
+ decode_block_func(this, component_id, block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
2902
+
2903
+ if (m_comps_in_scan == 1)
2904
+ block_x_mcu[component_id]++;
2905
+ else
2906
+ {
2907
+ if (++block_x_mcu_ofs == m_comp_h_samp[component_id])
2908
+ {
2909
+ block_x_mcu_ofs = 0;
2910
+
2911
+ if (++block_y_mcu_ofs == m_comp_v_samp[component_id])
2912
+ {
2913
+ block_y_mcu_ofs = 0;
2914
+ block_x_mcu[component_id] += m_comp_h_samp[component_id];
2915
+ }
2916
+ }
2917
+ }
2918
+ }
2919
+
2920
+ m_restarts_left--;
2921
+ }
2922
+
2923
+ if (m_comps_in_scan == 1)
2924
+ m_block_y_mcu[m_comp_list[0]]++;
2925
+ else
2926
+ {
2927
+ for (component_num = 0; component_num < m_comps_in_scan; component_num++)
2928
+ {
2929
+ component_id = m_comp_list[component_num];
2930
+ m_block_y_mcu[component_id] += m_comp_v_samp[component_id];
2931
+ }
2932
+ }
2933
+ }
2934
+ }
2935
+
2936
+ // Decode a progressively encoded image.
2937
+ void jpeg_decoder::init_progressive()
2938
+ {
2939
+ int i;
2940
+
2941
+ if (m_comps_in_frame == 4)
2942
+ stop_decoding(JPGD_UNSUPPORTED_COLORSPACE);
2943
+
2944
+ // Allocate the coefficient buffers.
2945
+ for (i = 0; i < m_comps_in_frame; i++)
2946
+ {
2947
+ m_dc_coeffs[i] = coeff_buf_open(m_max_mcus_per_row * m_comp_h_samp[i], m_max_mcus_per_col * m_comp_v_samp[i], 1, 1);
2948
+ m_ac_coeffs[i] = coeff_buf_open(m_max_mcus_per_row * m_comp_h_samp[i], m_max_mcus_per_col * m_comp_v_samp[i], 8, 8);
2949
+ }
2950
+
2951
+ for ( ; ; )
2952
+ {
2953
+ int dc_only_scan, refinement_scan;
2954
+ pDecode_block_func decode_block_func;
2955
+
2956
+ if (!init_scan())
2957
+ break;
2958
+
2959
+ dc_only_scan = (m_spectral_start == 0);
2960
+ refinement_scan = (m_successive_high != 0);
2961
+
2962
+ if ((m_spectral_start > m_spectral_end) || (m_spectral_end > 63))
2963
+ stop_decoding(JPGD_BAD_SOS_SPECTRAL);
2964
+
2965
+ if (dc_only_scan)
2966
+ {
2967
+ if (m_spectral_end)
2968
+ stop_decoding(JPGD_BAD_SOS_SPECTRAL);
2969
+ }
2970
+ else if (m_comps_in_scan != 1) /* AC scans can only contain one component */
2971
+ stop_decoding(JPGD_BAD_SOS_SPECTRAL);
2972
+
2973
+ if ((refinement_scan) && (m_successive_low != m_successive_high - 1))
2974
+ stop_decoding(JPGD_BAD_SOS_SUCCESSIVE);
2975
+
2976
+ if (dc_only_scan)
2977
+ {
2978
+ if (refinement_scan)
2979
+ decode_block_func = decode_block_dc_refine;
2980
+ else
2981
+ decode_block_func = decode_block_dc_first;
2982
+ }
2983
+ else
2984
+ {
2985
+ if (refinement_scan)
2986
+ decode_block_func = decode_block_ac_refine;
2987
+ else
2988
+ decode_block_func = decode_block_ac_first;
2989
+ }
2990
+
2991
+ decode_scan(decode_block_func);
2992
+
2993
+ m_bits_left = 16;
2994
+ get_bits(16);
2995
+ get_bits(16);
2996
+ }
2997
+
2998
+ m_comps_in_scan = m_comps_in_frame;
2999
+
3000
+ for (i = 0; i < m_comps_in_frame; i++)
3001
+ m_comp_list[i] = i;
3002
+
3003
+ calc_mcu_block_order();
3004
+ }
3005
+
3006
+ void jpeg_decoder::init_sequential()
3007
+ {
3008
+ if (!init_scan())
3009
+ stop_decoding(JPGD_UNEXPECTED_MARKER);
3010
+ }
3011
+
3012
+ void jpeg_decoder::decode_start()
3013
+ {
3014
+ init_frame();
3015
+
3016
+ if (m_progressive_flag)
3017
+ init_progressive();
3018
+ else
3019
+ init_sequential();
3020
+ }
3021
+
3022
+ void jpeg_decoder::decode_init(jpeg_decoder_stream *pStream)
3023
+ {
3024
+ init(pStream);
3025
+ locate_sof_marker();
3026
+ }
3027
+
3028
+ jpeg_decoder::jpeg_decoder(jpeg_decoder_stream *pStream)
3029
+ {
3030
+ if (setjmp(m_jmp_state))
3031
+ return;
3032
+ decode_init(pStream);
3033
+ }
3034
+
3035
+ int jpeg_decoder::begin_decoding()
3036
+ {
3037
+ if (m_ready_flag)
3038
+ return JPGD_SUCCESS;
3039
+
3040
+ if (m_error_code)
3041
+ return JPGD_FAILED;
3042
+
3043
+ if (setjmp(m_jmp_state))
3044
+ return JPGD_FAILED;
3045
+
3046
+ decode_start();
3047
+
3048
+ m_ready_flag = true;
3049
+
3050
+ return JPGD_SUCCESS;
3051
+ }
3052
+
3053
+ jpeg_decoder::~jpeg_decoder()
3054
+ {
3055
+ free_all_blocks();
3056
+ }
3057
+
3058
+ jpeg_decoder_file_stream::jpeg_decoder_file_stream()
3059
+ {
3060
+ m_pFile = NULL;
3061
+ m_eof_flag = false;
3062
+ m_error_flag = false;
3063
+ }
3064
+
3065
+ void jpeg_decoder_file_stream::close()
3066
+ {
3067
+ if (m_pFile)
3068
+ {
3069
+ fclose(m_pFile);
3070
+ m_pFile = NULL;
3071
+ }
3072
+
3073
+ m_eof_flag = false;
3074
+ m_error_flag = false;
3075
+ }
3076
+
3077
+ jpeg_decoder_file_stream::~jpeg_decoder_file_stream()
3078
+ {
3079
+ close();
3080
+ }
3081
+
3082
+ bool jpeg_decoder_file_stream::open(const char *Pfilename)
3083
+ {
3084
+ close();
3085
+
3086
+ m_eof_flag = false;
3087
+ m_error_flag = false;
3088
+
3089
+ #if defined(_MSC_VER)
3090
+ m_pFile = NULL;
3091
+ fopen_s(&m_pFile, Pfilename, "rb");
3092
+ #else
3093
+ m_pFile = fopen(Pfilename, "rb");
3094
+ #endif
3095
+ return m_pFile != NULL;
3096
+ }
3097
+
3098
+ int jpeg_decoder_file_stream::read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag)
3099
+ {
3100
+ if (!m_pFile)
3101
+ return -1;
3102
+
3103
+ if (m_eof_flag)
3104
+ {
3105
+ *pEOF_flag = true;
3106
+ return 0;
3107
+ }
3108
+
3109
+ if (m_error_flag)
3110
+ return -1;
3111
+
3112
+ int bytes_read = static_cast<int>(fread(pBuf, 1, max_bytes_to_read, m_pFile));
3113
+ if (bytes_read < max_bytes_to_read)
3114
+ {
3115
+ if (ferror(m_pFile))
3116
+ {
3117
+ m_error_flag = true;
3118
+ return -1;
3119
+ }
3120
+
3121
+ m_eof_flag = true;
3122
+ *pEOF_flag = true;
3123
+ }
3124
+
3125
+ return bytes_read;
3126
+ }
3127
+
3128
+ bool jpeg_decoder_mem_stream::open(const uint8 *pSrc_data, uint size)
3129
+ {
3130
+ close();
3131
+ m_pSrc_data = pSrc_data;
3132
+ m_ofs = 0;
3133
+ m_size = size;
3134
+ return true;
3135
+ }
3136
+
3137
+ int jpeg_decoder_mem_stream::read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag)
3138
+ {
3139
+ *pEOF_flag = false;
3140
+
3141
+ if (!m_pSrc_data)
3142
+ return -1;
3143
+
3144
+ uint bytes_remaining = m_size - m_ofs;
3145
+ if ((uint)max_bytes_to_read > bytes_remaining)
3146
+ {
3147
+ max_bytes_to_read = bytes_remaining;
3148
+ *pEOF_flag = true;
3149
+ }
3150
+
3151
+ memcpy(pBuf, m_pSrc_data + m_ofs, max_bytes_to_read);
3152
+ m_ofs += max_bytes_to_read;
3153
+
3154
+ return max_bytes_to_read;
3155
+ }
3156
+
3157
+ unsigned char *decompress_jpeg_image_from_stream(jpeg_decoder_stream *pStream, int *width, int *height, int *actual_comps, int req_comps)
3158
+ {
3159
+ if (!actual_comps)
3160
+ return NULL;
3161
+ *actual_comps = 0;
3162
+
3163
+ if ((!pStream) || (!width) || (!height) || (!req_comps))
3164
+ return NULL;
3165
+
3166
+ if ((req_comps != 1) && (req_comps != 3) && (req_comps != 4))
3167
+ return NULL;
3168
+
3169
+ jpeg_decoder decoder(pStream);
3170
+ if (decoder.get_error_code() != JPGD_SUCCESS)
3171
+ return NULL;
3172
+
3173
+ const int image_width = decoder.get_width(), image_height = decoder.get_height();
3174
+ *width = image_width;
3175
+ *height = image_height;
3176
+ *actual_comps = decoder.get_num_components();
3177
+
3178
+ if (decoder.begin_decoding() != JPGD_SUCCESS)
3179
+ return NULL;
3180
+
3181
+ const int dst_bpl = image_width * req_comps;
3182
+
3183
+ uint8 *pImage_data = (uint8*)jpgd_malloc(dst_bpl * image_height);
3184
+ if (!pImage_data)
3185
+ return NULL;
3186
+
3187
+ for (int y = 0; y < image_height; y++)
3188
+ {
3189
+ const uint8* pScan_line = 0;
3190
+ uint scan_line_len;
3191
+ if (decoder.decode((const void**)&pScan_line, &scan_line_len) != JPGD_SUCCESS)
3192
+ {
3193
+ jpgd_free(pImage_data);
3194
+ return NULL;
3195
+ }
3196
+
3197
+ uint8 *pDst = pImage_data + y * dst_bpl;
3198
+
3199
+ if (((req_comps == 4) && (decoder.get_num_components() == 3)) ||
3200
+ ((req_comps == 1) && (decoder.get_num_components() == 1)))
3201
+ {
3202
+ memcpy(pDst, pScan_line, dst_bpl);
3203
+ }
3204
+ else if (decoder.get_num_components() == 1)
3205
+ {
3206
+ if (req_comps == 3)
3207
+ {
3208
+ for (int x = 0; x < image_width; x++)
3209
+ {
3210
+ uint8 luma = pScan_line[x];
3211
+ pDst[0] = luma;
3212
+ pDst[1] = luma;
3213
+ pDst[2] = luma;
3214
+ pDst += 3;
3215
+ }
3216
+ }
3217
+ else
3218
+ {
3219
+ for (int x = 0; x < image_width; x++)
3220
+ {
3221
+ uint8 luma = pScan_line[x];
3222
+ pDst[0] = luma;
3223
+ pDst[1] = luma;
3224
+ pDst[2] = luma;
3225
+ pDst[3] = 255;
3226
+ pDst += 4;
3227
+ }
3228
+ }
3229
+ }
3230
+ else if (decoder.get_num_components() == 3)
3231
+ {
3232
+ if (req_comps == 1)
3233
+ {
3234
+ const int YR = 19595, YG = 38470, YB = 7471;
3235
+ for (int x = 0; x < image_width; x++)
3236
+ {
3237
+ int r = pScan_line[x*4+0];
3238
+ int g = pScan_line[x*4+1];
3239
+ int b = pScan_line[x*4+2];
3240
+ *pDst++ = static_cast<uint8>((r * YR + g * YG + b * YB + 32768) >> 16);
3241
+ }
3242
+ }
3243
+ else
3244
+ {
3245
+ for (int x = 0; x < image_width; x++)
3246
+ {
3247
+ pDst[0] = pScan_line[x*4+0];
3248
+ pDst[1] = pScan_line[x*4+1];
3249
+ pDst[2] = pScan_line[x*4+2];
3250
+ pDst += 3;
3251
+ }
3252
+ }
3253
+ }
3254
+ }
3255
+
3256
+ return pImage_data;
3257
+ }
3258
+
3259
+ // BEGIN EPIC MOD
3260
+ unsigned char *decompress_jpeg_image_from_memory(const unsigned char *pSrc_data, int src_data_size, int *width, int *height, int *actual_comps, int req_comps, int format)
3261
+ {
3262
+ jpg_format = (ERGBFormatJPG)format;
3263
+ // EMD EPIC MOD
3264
+ jpgd::jpeg_decoder_mem_stream mem_stream(pSrc_data, src_data_size);
3265
+ return decompress_jpeg_image_from_stream(&mem_stream, width, height, actual_comps, req_comps);
3266
+ }
3267
+
3268
+ unsigned char *decompress_jpeg_image_from_file(const char *pSrc_filename, int *width, int *height, int *actual_comps, int req_comps)
3269
+ {
3270
+ jpgd::jpeg_decoder_file_stream file_stream;
3271
+ if (!file_stream.open(pSrc_filename))
3272
+ return NULL;
3273
+ return decompress_jpeg_image_from_stream(&file_stream, width, height, actual_comps, req_comps);
3274
+ }
3275
+
3276
+ } // namespace jpgd
crazy_functions/test_project/cpp/libJPG/jpgd.h ADDED
@@ -0,0 +1,316 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ // jpgd.h - C++ class for JPEG decompression.
2
+ // Public domain, Rich Geldreich <richgel99@gmail.com>
3
+ #ifndef JPEG_DECODER_H
4
+ #define JPEG_DECODER_H
5
+
6
+ #include <stdlib.h>
7
+ #include <stdio.h>
8
+ #include <setjmp.h>
9
+
10
+ namespace jpgd
11
+ {
12
+ typedef unsigned char uint8;
13
+ typedef signed short int16;
14
+ typedef unsigned short uint16;
15
+ typedef unsigned int uint;
16
+ typedef signed int int32;
17
+
18
+ // Loads a JPEG image from a memory buffer or a file.
19
+ // req_comps can be 1 (grayscale), 3 (RGB), or 4 (RGBA).
20
+ // On return, width/height will be set to the image's dimensions, and actual_comps will be set to the either 1 (grayscale) or 3 (RGB).
21
+ // Notes: For more control over where and how the source data is read, see the decompress_jpeg_image_from_stream() function below, or call the jpeg_decoder class directly.
22
+ // Requesting a 8 or 32bpp image is currently a little faster than 24bpp because the jpeg_decoder class itself currently always unpacks to either 8 or 32bpp.
23
+ // BEGIN EPIC MOD
24
+ //unsigned char *decompress_jpeg_image_from_memory(const unsigned char *pSrc_data, int src_data_size, int *width, int *height, int *actual_comps, int req_comps);
25
+ unsigned char *decompress_jpeg_image_from_memory(const unsigned char *pSrc_data, int src_data_size, int *width, int *height, int *actual_comps, int req_comps, int format);
26
+ // END EPIC MOD
27
+ unsigned char *decompress_jpeg_image_from_file(const char *pSrc_filename, int *width, int *height, int *actual_comps, int req_comps);
28
+
29
+ // Success/failure error codes.
30
+ enum jpgd_status
31
+ {
32
+ JPGD_SUCCESS = 0, JPGD_FAILED = -1, JPGD_DONE = 1,
33
+ JPGD_BAD_DHT_COUNTS = -256, JPGD_BAD_DHT_INDEX, JPGD_BAD_DHT_MARKER, JPGD_BAD_DQT_MARKER, JPGD_BAD_DQT_TABLE,
34
+ JPGD_BAD_PRECISION, JPGD_BAD_HEIGHT, JPGD_BAD_WIDTH, JPGD_TOO_MANY_COMPONENTS,
35
+ JPGD_BAD_SOF_LENGTH, JPGD_BAD_VARIABLE_MARKER, JPGD_BAD_DRI_LENGTH, JPGD_BAD_SOS_LENGTH,
36
+ JPGD_BAD_SOS_COMP_ID, JPGD_W_EXTRA_BYTES_BEFORE_MARKER, JPGD_NO_ARITHMITIC_SUPPORT, JPGD_UNEXPECTED_MARKER,
37
+ JPGD_NOT_JPEG, JPGD_UNSUPPORTED_MARKER, JPGD_BAD_DQT_LENGTH, JPGD_TOO_MANY_BLOCKS,
38
+ JPGD_UNDEFINED_QUANT_TABLE, JPGD_UNDEFINED_HUFF_TABLE, JPGD_NOT_SINGLE_SCAN, JPGD_UNSUPPORTED_COLORSPACE,
39
+ JPGD_UNSUPPORTED_SAMP_FACTORS, JPGD_DECODE_ERROR, JPGD_BAD_RESTART_MARKER, JPGD_ASSERTION_ERROR,
40
+ JPGD_BAD_SOS_SPECTRAL, JPGD_BAD_SOS_SUCCESSIVE, JPGD_STREAM_READ, JPGD_NOTENOUGHMEM
41
+ };
42
+
43
+ // Input stream interface.
44
+ // Derive from this class to read input data from sources other than files or memory. Set m_eof_flag to true when no more data is available.
45
+ // The decoder is rather greedy: it will keep on calling this method until its internal input buffer is full, or until the EOF flag is set.
46
+ // It the input stream contains data after the JPEG stream's EOI (end of image) marker it will probably be pulled into the internal buffer.
47
+ // Call the get_total_bytes_read() method to determine the actual size of the JPEG stream after successful decoding.
48
+ class jpeg_decoder_stream
49
+ {
50
+ public:
51
+ jpeg_decoder_stream() { }
52
+ virtual ~jpeg_decoder_stream() { }
53
+
54
+ // The read() method is called when the internal input buffer is empty.
55
+ // Parameters:
56
+ // pBuf - input buffer
57
+ // max_bytes_to_read - maximum bytes that can be written to pBuf
58
+ // pEOF_flag - set this to true if at end of stream (no more bytes remaining)
59
+ // Returns -1 on error, otherwise return the number of bytes actually written to the buffer (which may be 0).
60
+ // Notes: This method will be called in a loop until you set *pEOF_flag to true or the internal buffer is full.
61
+ virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag) = 0;
62
+ };
63
+
64
+ // stdio FILE stream class.
65
+ class jpeg_decoder_file_stream : public jpeg_decoder_stream
66
+ {
67
+ jpeg_decoder_file_stream(const jpeg_decoder_file_stream &);
68
+ jpeg_decoder_file_stream &operator =(const jpeg_decoder_file_stream &);
69
+
70
+ FILE *m_pFile;
71
+ bool m_eof_flag, m_error_flag;
72
+
73
+ public:
74
+ jpeg_decoder_file_stream();
75
+ virtual ~jpeg_decoder_file_stream();
76
+
77
+ bool open(const char *Pfilename);
78
+ void close();
79
+
80
+ virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag);
81
+ };
82
+
83
+ // Memory stream class.
84
+ class jpeg_decoder_mem_stream : public jpeg_decoder_stream
85
+ {
86
+ const uint8 *m_pSrc_data;
87
+ uint m_ofs, m_size;
88
+
89
+ public:
90
+ jpeg_decoder_mem_stream() : m_pSrc_data(NULL), m_ofs(0), m_size(0) { }
91
+ jpeg_decoder_mem_stream(const uint8 *pSrc_data, uint size) : m_pSrc_data(pSrc_data), m_ofs(0), m_size(size) { }
92
+
93
+ virtual ~jpeg_decoder_mem_stream() { }
94
+
95
+ bool open(const uint8 *pSrc_data, uint size);
96
+ void close() { m_pSrc_data = NULL; m_ofs = 0; m_size = 0; }
97
+
98
+ virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag);
99
+ };
100
+
101
+ // Loads JPEG file from a jpeg_decoder_stream.
102
+ unsigned char *decompress_jpeg_image_from_stream(jpeg_decoder_stream *pStream, int *width, int *height, int *actual_comps, int req_comps);
103
+
104
+ enum
105
+ {
106
+ JPGD_IN_BUF_SIZE = 8192, JPGD_MAX_BLOCKS_PER_MCU = 10, JPGD_MAX_HUFF_TABLES = 8, JPGD_MAX_QUANT_TABLES = 4,
107
+ JPGD_MAX_COMPONENTS = 4, JPGD_MAX_COMPS_IN_SCAN = 4, JPGD_MAX_BLOCKS_PER_ROW = 8192, JPGD_MAX_HEIGHT = 16384, JPGD_MAX_WIDTH = 16384
108
+ };
109
+
110
+ typedef int16 jpgd_quant_t;
111
+ typedef int16 jpgd_block_t;
112
+
113
+ class jpeg_decoder
114
+ {
115
+ public:
116
+ // Call get_error_code() after constructing to determine if the stream is valid or not. You may call the get_width(), get_height(), etc.
117
+ // methods after the constructor is called. You may then either destruct the object, or begin decoding the image by calling begin_decoding(), then decode() on each scanline.
118
+ jpeg_decoder(jpeg_decoder_stream *pStream);
119
+
120
+ ~jpeg_decoder();
121
+
122
+ // Call this method after constructing the object to begin decompression.
123
+ // If JPGD_SUCCESS is returned you may then call decode() on each scanline.
124
+ int begin_decoding();
125
+
126
+ // Returns the next scan line.
127
+ // For grayscale images, pScan_line will point to a buffer containing 8-bit pixels (get_bytes_per_pixel() will return 1).
128
+ // Otherwise, it will always point to a buffer containing 32-bit RGBA pixels (A will always be 255, and get_bytes_per_pixel() will return 4).
129
+ // Returns JPGD_SUCCESS if a scan line has been returned.
130
+ // Returns JPGD_DONE if all scan lines have been returned.
131
+ // Returns JPGD_FAILED if an error occurred. Call get_error_code() for a more info.
132
+ int decode(const void** pScan_line, uint* pScan_line_len);
133
+
134
+ inline jpgd_status get_error_code() const { return m_error_code; }
135
+
136
+ inline int get_width() const { return m_image_x_size; }
137
+ inline int get_height() const { return m_image_y_size; }
138
+
139
+ inline int get_num_components() const { return m_comps_in_frame; }
140
+
141
+ inline int get_bytes_per_pixel() const { return m_dest_bytes_per_pixel; }
142
+ inline int get_bytes_per_scan_line() const { return m_image_x_size * get_bytes_per_pixel(); }
143
+
144
+ // Returns the total number of bytes actually consumed by the decoder (which should equal the actual size of the JPEG file).
145
+ inline int get_total_bytes_read() const { return m_total_bytes_read; }
146
+
147
+ private:
148
+ jpeg_decoder(const jpeg_decoder &);
149
+ jpeg_decoder &operator =(const jpeg_decoder &);
150
+
151
+ typedef void (*pDecode_block_func)(jpeg_decoder *, int, int, int);
152
+
153
+ struct huff_tables
154
+ {
155
+ bool ac_table;
156
+ uint look_up[256];
157
+ uint look_up2[256];
158
+ uint8 code_size[256];
159
+ uint tree[512];
160
+ };
161
+
162
+ struct coeff_buf
163
+ {
164
+ uint8 *pData;
165
+ int block_num_x, block_num_y;
166
+ int block_len_x, block_len_y;
167
+ int block_size;
168
+ };
169
+
170
+ struct mem_block
171
+ {
172
+ mem_block *m_pNext;
173
+ size_t m_used_count;
174
+ size_t m_size;
175
+ char m_data[1];
176
+ };
177
+
178
+ jmp_buf m_jmp_state;
179
+ mem_block *m_pMem_blocks;
180
+ int m_image_x_size;
181
+ int m_image_y_size;
182
+ jpeg_decoder_stream *m_pStream;
183
+ int m_progressive_flag;
184
+ uint8 m_huff_ac[JPGD_MAX_HUFF_TABLES];
185
+ uint8* m_huff_num[JPGD_MAX_HUFF_TABLES]; // pointer to number of Huffman codes per bit size
186
+ uint8* m_huff_val[JPGD_MAX_HUFF_TABLES]; // pointer to Huffman codes per bit size
187
+ jpgd_quant_t* m_quant[JPGD_MAX_QUANT_TABLES]; // pointer to quantization tables
188
+ int m_scan_type; // Gray, Yh1v1, Yh1v2, Yh2v1, Yh2v2 (CMYK111, CMYK4114 no longer supported)
189
+ int m_comps_in_frame; // # of components in frame
190
+ int m_comp_h_samp[JPGD_MAX_COMPONENTS]; // component's horizontal sampling factor
191
+ int m_comp_v_samp[JPGD_MAX_COMPONENTS]; // component's vertical sampling factor
192
+ int m_comp_quant[JPGD_MAX_COMPONENTS]; // component's quantization table selector
193
+ int m_comp_ident[JPGD_MAX_COMPONENTS]; // component's ID
194
+ int m_comp_h_blocks[JPGD_MAX_COMPONENTS];
195
+ int m_comp_v_blocks[JPGD_MAX_COMPONENTS];
196
+ int m_comps_in_scan; // # of components in scan
197
+ int m_comp_list[JPGD_MAX_COMPS_IN_SCAN]; // components in this scan
198
+ int m_comp_dc_tab[JPGD_MAX_COMPONENTS]; // component's DC Huffman coding table selector
199
+ int m_comp_ac_tab[JPGD_MAX_COMPONENTS]; // component's AC Huffman coding table selector
200
+ int m_spectral_start; // spectral selection start
201
+ int m_spectral_end; // spectral selection end
202
+ int m_successive_low; // successive approximation low
203
+ int m_successive_high; // successive approximation high
204
+ int m_max_mcu_x_size; // MCU's max. X size in pixels
205
+ int m_max_mcu_y_size; // MCU's max. Y size in pixels
206
+ int m_blocks_per_mcu;
207
+ int m_max_blocks_per_row;
208
+ int m_mcus_per_row, m_mcus_per_col;
209
+ int m_mcu_org[JPGD_MAX_BLOCKS_PER_MCU];
210
+ int m_total_lines_left; // total # lines left in image
211
+ int m_mcu_lines_left; // total # lines left in this MCU
212
+ int m_real_dest_bytes_per_scan_line;
213
+ int m_dest_bytes_per_scan_line; // rounded up
214
+ int m_dest_bytes_per_pixel; // 4 (RGB) or 1 (Y)
215
+ huff_tables* m_pHuff_tabs[JPGD_MAX_HUFF_TABLES];
216
+ coeff_buf* m_dc_coeffs[JPGD_MAX_COMPONENTS];
217
+ coeff_buf* m_ac_coeffs[JPGD_MAX_COMPONENTS];
218
+ int m_eob_run;
219
+ int m_block_y_mcu[JPGD_MAX_COMPONENTS];
220
+ uint8* m_pIn_buf_ofs;
221
+ int m_in_buf_left;
222
+ int m_tem_flag;
223
+ bool m_eof_flag;
224
+ uint8 m_in_buf_pad_start[128];
225
+ uint8 m_in_buf[JPGD_IN_BUF_SIZE + 128];
226
+ uint8 m_in_buf_pad_end[128];
227
+ int m_bits_left;
228
+ uint m_bit_buf;
229
+ int m_restart_interval;
230
+ int m_restarts_left;
231
+ int m_next_restart_num;
232
+ int m_max_mcus_per_row;
233
+ int m_max_blocks_per_mcu;
234
+ int m_expanded_blocks_per_mcu;
235
+ int m_expanded_blocks_per_row;
236
+ int m_expanded_blocks_per_component;
237
+ bool m_freq_domain_chroma_upsample;
238
+ int m_max_mcus_per_col;
239
+ uint m_last_dc_val[JPGD_MAX_COMPONENTS];
240
+ jpgd_block_t* m_pMCU_coefficients;
241
+ int m_mcu_block_max_zag[JPGD_MAX_BLOCKS_PER_MCU];
242
+ uint8* m_pSample_buf;
243
+ int m_crr[256];
244
+ int m_cbb[256];
245
+ int m_crg[256];
246
+ int m_cbg[256];
247
+ uint8* m_pScan_line_0;
248
+ uint8* m_pScan_line_1;
249
+ jpgd_status m_error_code;
250
+ bool m_ready_flag;
251
+ int m_total_bytes_read;
252
+
253
+ void free_all_blocks();
254
+ // BEGIN EPIC MOD
255
+ UE_NORETURN void stop_decoding(jpgd_status status);
256
+ // END EPIC MOD
257
+ void *alloc(size_t n, bool zero = false);
258
+ void word_clear(void *p, uint16 c, uint n);
259
+ void prep_in_buffer();
260
+ void read_dht_marker();
261
+ void read_dqt_marker();
262
+ void read_sof_marker();
263
+ void skip_variable_marker();
264
+ void read_dri_marker();
265
+ void read_sos_marker();
266
+ int next_marker();
267
+ int process_markers();
268
+ void locate_soi_marker();
269
+ void locate_sof_marker();
270
+ int locate_sos_marker();
271
+ void init(jpeg_decoder_stream * pStream);
272
+ void create_look_ups();
273
+ void fix_in_buffer();
274
+ void transform_mcu(int mcu_row);
275
+ void transform_mcu_expand(int mcu_row);
276
+ coeff_buf* coeff_buf_open(int block_num_x, int block_num_y, int block_len_x, int block_len_y);
277
+ inline jpgd_block_t *coeff_buf_getp(coeff_buf *cb, int block_x, int block_y);
278
+ void load_next_row();
279
+ void decode_next_row();
280
+ void make_huff_table(int index, huff_tables *pH);
281
+ void check_quant_tables();
282
+ void check_huff_tables();
283
+ void calc_mcu_block_order();
284
+ int init_scan();
285
+ void init_frame();
286
+ void process_restart();
287
+ void decode_scan(pDecode_block_func decode_block_func);
288
+ void init_progressive();
289
+ void init_sequential();
290
+ void decode_start();
291
+ void decode_init(jpeg_decoder_stream * pStream);
292
+ void H2V2Convert();
293
+ void H2V1Convert();
294
+ void H1V2Convert();
295
+ void H1V1Convert();
296
+ void gray_convert();
297
+ void expanded_convert();
298
+ void find_eoi();
299
+ inline uint get_char();
300
+ inline uint get_char(bool *pPadding_flag);
301
+ inline void stuff_char(uint8 q);
302
+ inline uint8 get_octet();
303
+ inline uint get_bits(int num_bits);
304
+ inline uint get_bits_no_markers(int numbits);
305
+ inline int huff_decode(huff_tables *pH);
306
+ inline int huff_decode(huff_tables *pH, int& extrabits);
307
+ static inline uint8 clamp(int i);
308
+ static void decode_block_dc_first(jpeg_decoder *pD, int component_id, int block_x, int block_y);
309
+ static void decode_block_dc_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y);
310
+ static void decode_block_ac_first(jpeg_decoder *pD, int component_id, int block_x, int block_y);
311
+ static void decode_block_ac_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y);
312
+ };
313
+
314
+ } // namespace jpgd
315
+
316
+ #endif // JPEG_DECODER_H
crazy_functions/test_project/cpp/libJPG/jpge.cpp ADDED
@@ -0,0 +1,1049 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ // jpge.cpp - C++ class for JPEG compression.
2
+ // Public domain, Rich Geldreich <richgel99@gmail.com>
3
+ // v1.01, Dec. 18, 2010 - Initial release
4
+ // v1.02, Apr. 6, 2011 - Removed 2x2 ordered dither in H2V1 chroma subsampling method load_block_16_8_8(). (The rounding factor was 2, when it should have been 1. Either way, it wasn't helping.)
5
+ // v1.03, Apr. 16, 2011 - Added support for optimized Huffman code tables, optimized dynamic memory allocation down to only 1 alloc.
6
+ // Also from Alex Evans: Added RGBA support, linear memory allocator (no longer needed in v1.03).
7
+ // v1.04, May. 19, 2012: Forgot to set m_pFile ptr to NULL in cfile_stream::close(). Thanks to Owen Kaluza for reporting this bug.
8
+ // Code tweaks to fix VS2008 static code analysis warnings (all looked harmless).
9
+ // Code review revealed method load_block_16_8_8() (used for the non-default H2V1 sampling mode to downsample chroma) somehow didn't get the rounding factor fix from v1.02.
10
+
11
+ #include "jpge.h"
12
+
13
+ #include <stdlib.h>
14
+ #include <string.h>
15
+ #if PLATFORM_WINDOWS
16
+ #include <malloc.h>
17
+ #endif
18
+
19
+ #define JPGE_MAX(a,b) (((a)>(b))?(a):(b))
20
+ #define JPGE_MIN(a,b) (((a)<(b))?(a):(b))
21
+
22
+ namespace jpge {
23
+
24
+ static inline void *jpge_malloc(size_t nSize) { return FMemory::Malloc(nSize); }
25
+ static inline void jpge_free(void *p) { FMemory::Free(p);; }
26
+
27
+ // Various JPEG enums and tables.
28
+ enum { M_SOF0 = 0xC0, M_DHT = 0xC4, M_SOI = 0xD8, M_EOI = 0xD9, M_SOS = 0xDA, M_DQT = 0xDB, M_APP0 = 0xE0 };
29
+ enum { DC_LUM_CODES = 12, AC_LUM_CODES = 256, DC_CHROMA_CODES = 12, AC_CHROMA_CODES = 256, MAX_HUFF_SYMBOLS = 257, MAX_HUFF_CODESIZE = 32 };
30
+
31
+ static uint8 s_zag[64] = { 0,1,8,16,9,2,3,10,17,24,32,25,18,11,4,5,12,19,26,33,40,48,41,34,27,20,13,6,7,14,21,28,35,42,49,56,57,50,43,36,29,22,15,23,30,37,44,51,58,59,52,45,38,31,39,46,53,60,61,54,47,55,62,63 };
32
+ static int16 s_std_lum_quant[64] = { 16,11,12,14,12,10,16,14,13,14,18,17,16,19,24,40,26,24,22,22,24,49,35,37,29,40,58,51,61,60,57,51,56,55,64,72,92,78,64,68,87,69,55,56,80,109,81,87,95,98,103,104,103,62,77,113,121,112,100,120,92,101,103,99 };
33
+ static int16 s_std_croma_quant[64] = { 17,18,18,24,21,24,47,26,26,47,99,66,56,66,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99 };
34
+ static uint8 s_dc_lum_bits[17] = { 0,0,1,5,1,1,1,1,1,1,0,0,0,0,0,0,0 };
35
+ static uint8 s_dc_lum_val[DC_LUM_CODES] = { 0,1,2,3,4,5,6,7,8,9,10,11 };
36
+ static uint8 s_ac_lum_bits[17] = { 0,0,2,1,3,3,2,4,3,5,5,4,4,0,0,1,0x7d };
37
+ static uint8 s_ac_lum_val[AC_LUM_CODES] =
38
+ {
39
+ 0x01,0x02,0x03,0x00,0x04,0x11,0x05,0x12,0x21,0x31,0x41,0x06,0x13,0x51,0x61,0x07,0x22,0x71,0x14,0x32,0x81,0x91,0xa1,0x08,0x23,0x42,0xb1,0xc1,0x15,0x52,0xd1,0xf0,
40
+ 0x24,0x33,0x62,0x72,0x82,0x09,0x0a,0x16,0x17,0x18,0x19,0x1a,0x25,0x26,0x27,0x28,0x29,0x2a,0x34,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,0x49,
41
+ 0x4a,0x53,0x54,0x55,0x56,0x57,0x58,0x59,0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x83,0x84,0x85,0x86,0x87,0x88,0x89,
42
+ 0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,0xb5,0xb6,0xb7,0xb8,0xb9,0xba,0xc2,0xc3,0xc4,0xc5,
43
+ 0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,0xe1,0xe2,0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf1,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,
44
+ 0xf9,0xfa
45
+ };
46
+ static uint8 s_dc_chroma_bits[17] = { 0,0,3,1,1,1,1,1,1,1,1,1,0,0,0,0,0 };
47
+ static uint8 s_dc_chroma_val[DC_CHROMA_CODES] = { 0,1,2,3,4,5,6,7,8,9,10,11 };
48
+ static uint8 s_ac_chroma_bits[17] = { 0,0,2,1,2,4,4,3,4,7,5,4,4,0,1,2,0x77 };
49
+ static uint8 s_ac_chroma_val[AC_CHROMA_CODES] =
50
+ {
51
+ 0x00,0x01,0x02,0x03,0x11,0x04,0x05,0x21,0x31,0x06,0x12,0x41,0x51,0x07,0x61,0x71,0x13,0x22,0x32,0x81,0x08,0x14,0x42,0x91,0xa1,0xb1,0xc1,0x09,0x23,0x33,0x52,0xf0,
52
+ 0x15,0x62,0x72,0xd1,0x0a,0x16,0x24,0x34,0xe1,0x25,0xf1,0x17,0x18,0x19,0x1a,0x26,0x27,0x28,0x29,0x2a,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,
53
+ 0x49,0x4a,0x53,0x54,0x55,0x56,0x57,0x58,0x59,0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x82,0x83,0x84,0x85,0x86,0x87,
54
+ 0x88,0x89,0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,0xb5,0xb6,0xb7,0xb8,0xb9,0xba,0xc2,0xc3,
55
+ 0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,0xe2,0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,
56
+ 0xf9,0xfa
57
+ };
58
+
59
+ // Low-level helper functions.
60
+ template <class T> inline void clear_obj(T &obj) { memset(&obj, 0, sizeof(obj)); }
61
+
62
+ const int YR = 19595, YG = 38470, YB = 7471, CB_R = -11059, CB_G = -21709, CB_B = 32768, CR_R = 32768, CR_G = -27439, CR_B = -5329;
63
+ static inline uint8 clamp(int i) { if (static_cast<uint>(i) > 255U) { if (i < 0) i = 0; else if (i > 255) i = 255; } return static_cast<uint8>(i); }
64
+
65
+ static void RGB_to_YCC(uint8* pDst, const uint8 *pSrc, int num_pixels)
66
+ {
67
+ for ( ; num_pixels; pDst += 3, pSrc += 3, num_pixels--)
68
+ {
69
+ const int r = pSrc[0], g = pSrc[1], b = pSrc[2];
70
+ pDst[0] = static_cast<uint8>((r * YR + g * YG + b * YB + 32768) >> 16);
71
+ pDst[1] = clamp(128 + ((r * CB_R + g * CB_G + b * CB_B + 32768) >> 16));
72
+ pDst[2] = clamp(128 + ((r * CR_R + g * CR_G + b * CR_B + 32768) >> 16));
73
+ }
74
+ }
75
+
76
+ static void RGB_to_Y(uint8* pDst, const uint8 *pSrc, int num_pixels)
77
+ {
78
+ for ( ; num_pixels; pDst++, pSrc += 3, num_pixels--)
79
+ pDst[0] = static_cast<uint8>((pSrc[0] * YR + pSrc[1] * YG + pSrc[2] * YB + 32768) >> 16);
80
+ }
81
+
82
+ static void RGBA_to_YCC(uint8* pDst, const uint8 *pSrc, int num_pixels)
83
+ {
84
+ for ( ; num_pixels; pDst += 3, pSrc += 4, num_pixels--)
85
+ {
86
+ const int r = pSrc[0], g = pSrc[1], b = pSrc[2];
87
+ pDst[0] = static_cast<uint8>((r * YR + g * YG + b * YB + 32768) >> 16);
88
+ pDst[1] = clamp(128 + ((r * CB_R + g * CB_G + b * CB_B + 32768) >> 16));
89
+ pDst[2] = clamp(128 + ((r * CR_R + g * CR_G + b * CR_B + 32768) >> 16));
90
+ }
91
+ }
92
+
93
+ static void RGBA_to_Y(uint8* pDst, const uint8 *pSrc, int num_pixels)
94
+ {
95
+ for ( ; num_pixels; pDst++, pSrc += 4, num_pixels--)
96
+ pDst[0] = static_cast<uint8>((pSrc[0] * YR + pSrc[1] * YG + pSrc[2] * YB + 32768) >> 16);
97
+ }
98
+
99
+ static void Y_to_YCC(uint8* pDst, const uint8* pSrc, int num_pixels)
100
+ {
101
+ for( ; num_pixels; pDst += 3, pSrc++, num_pixels--) { pDst[0] = pSrc[0]; pDst[1] = 128; pDst[2] = 128; }
102
+ }
103
+
104
+ // Forward DCT - DCT derived from jfdctint.
105
+ #define CONST_BITS 13
106
+ #define ROW_BITS 2
107
+ #define DCT_DESCALE(x, n) (((x) + (((int32)1) << ((n) - 1))) >> (n))
108
+ #define DCT_MUL(var, c) (static_cast<int16>(var) * static_cast<int32>(c))
109
+ #define DCT1D(s0, s1, s2, s3, s4, s5, s6, s7) \
110
+ int32 t0 = s0 + s7, t7 = s0 - s7, t1 = s1 + s6, t6 = s1 - s6, t2 = s2 + s5, t5 = s2 - s5, t3 = s3 + s4, t4 = s3 - s4; \
111
+ int32 t10 = t0 + t3, t13 = t0 - t3, t11 = t1 + t2, t12 = t1 - t2; \
112
+ int32 u1 = DCT_MUL(t12 + t13, 4433); \
113
+ s2 = u1 + DCT_MUL(t13, 6270); \
114
+ s6 = u1 + DCT_MUL(t12, -15137); \
115
+ u1 = t4 + t7; \
116
+ int32 u2 = t5 + t6, u3 = t4 + t6, u4 = t5 + t7; \
117
+ int32 z5 = DCT_MUL(u3 + u4, 9633); \
118
+ t4 = DCT_MUL(t4, 2446); t5 = DCT_MUL(t5, 16819); \
119
+ t6 = DCT_MUL(t6, 25172); t7 = DCT_MUL(t7, 12299); \
120
+ u1 = DCT_MUL(u1, -7373); u2 = DCT_MUL(u2, -20995); \
121
+ u3 = DCT_MUL(u3, -16069); u4 = DCT_MUL(u4, -3196); \
122
+ u3 += z5; u4 += z5; \
123
+ s0 = t10 + t11; s1 = t7 + u1 + u4; s3 = t6 + u2 + u3; s4 = t10 - t11; s5 = t5 + u2 + u4; s7 = t4 + u1 + u3;
124
+
125
+ static void DCT2D(int32 *p)
126
+ {
127
+ int32 c, *q = p;
128
+ for (c = 7; c >= 0; c--, q += 8)
129
+ {
130
+ int32 s0 = q[0], s1 = q[1], s2 = q[2], s3 = q[3], s4 = q[4], s5 = q[5], s6 = q[6], s7 = q[7];
131
+ DCT1D(s0, s1, s2, s3, s4, s5, s6, s7);
132
+ q[0] = s0 << ROW_BITS; q[1] = DCT_DESCALE(s1, CONST_BITS-ROW_BITS); q[2] = DCT_DESCALE(s2, CONST_BITS-ROW_BITS); q[3] = DCT_DESCALE(s3, CONST_BITS-ROW_BITS);
133
+ q[4] = s4 << ROW_BITS; q[5] = DCT_DESCALE(s5, CONST_BITS-ROW_BITS); q[6] = DCT_DESCALE(s6, CONST_BITS-ROW_BITS); q[7] = DCT_DESCALE(s7, CONST_BITS-ROW_BITS);
134
+ }
135
+ for (q = p, c = 7; c >= 0; c--, q++)
136
+ {
137
+ int32 s0 = q[0*8], s1 = q[1*8], s2 = q[2*8], s3 = q[3*8], s4 = q[4*8], s5 = q[5*8], s6 = q[6*8], s7 = q[7*8];
138
+ DCT1D(s0, s1, s2, s3, s4, s5, s6, s7);
139
+ q[0*8] = DCT_DESCALE(s0, ROW_BITS+3); q[1*8] = DCT_DESCALE(s1, CONST_BITS+ROW_BITS+3); q[2*8] = DCT_DESCALE(s2, CONST_BITS+ROW_BITS+3); q[3*8] = DCT_DESCALE(s3, CONST_BITS+ROW_BITS+3);
140
+ q[4*8] = DCT_DESCALE(s4, ROW_BITS+3); q[5*8] = DCT_DESCALE(s5, CONST_BITS+ROW_BITS+3); q[6*8] = DCT_DESCALE(s6, CONST_BITS+ROW_BITS+3); q[7*8] = DCT_DESCALE(s7, CONST_BITS+ROW_BITS+3);
141
+ }
142
+ }
143
+
144
+ struct sym_freq { uint m_key, m_sym_index; };
145
+
146
+ // Radix sorts sym_freq[] array by 32-bit key m_key. Returns ptr to sorted values.
147
+ static inline sym_freq* radix_sort_syms(uint num_syms, sym_freq* pSyms0, sym_freq* pSyms1)
148
+ {
149
+ const uint cMaxPasses = 4;
150
+ uint32 hist[256 * cMaxPasses]; clear_obj(hist);
151
+ for (uint i = 0; i < num_syms; i++) { uint freq = pSyms0[i].m_key; hist[freq & 0xFF]++; hist[256 + ((freq >> 8) & 0xFF)]++; hist[256*2 + ((freq >> 16) & 0xFF)]++; hist[256*3 + ((freq >> 24) & 0xFF)]++; }
152
+ sym_freq* pCur_syms = pSyms0, *pNew_syms = pSyms1;
153
+ uint total_passes = cMaxPasses; while ((total_passes > 1) && (num_syms == hist[(total_passes - 1) * 256])) total_passes--;
154
+ for (uint pass_shift = 0, pass = 0; pass < total_passes; pass++, pass_shift += 8)
155
+ {
156
+ const uint32* pHist = &hist[pass << 8];
157
+ uint offsets[256], cur_ofs = 0;
158
+ for (uint i = 0; i < 256; i++) { offsets[i] = cur_ofs; cur_ofs += pHist[i]; }
159
+ for (uint i = 0; i < num_syms; i++)
160
+ pNew_syms[offsets[(pCur_syms[i].m_key >> pass_shift) & 0xFF]++] = pCur_syms[i];
161
+ sym_freq* t = pCur_syms; pCur_syms = pNew_syms; pNew_syms = t;
162
+ }
163
+ return pCur_syms;
164
+ }
165
+
166
+ // calculate_minimum_redundancy() originally written by: Alistair Moffat, alistair@cs.mu.oz.au, Jyrki Katajainen, jyrki@diku.dk, November 1996.
167
+ static void calculate_minimum_redundancy(sym_freq *A, int n)
168
+ {
169
+ int root, leaf, next, avbl, used, dpth;
170
+ if (n==0) return; else if (n==1) { A[0].m_key = 1; return; }
171
+ A[0].m_key += A[1].m_key; root = 0; leaf = 2;
172
+ for (next=1; next < n-1; next++)
173
+ {
174
+ if (leaf>=n || A[root].m_key<A[leaf].m_key) { A[next].m_key = A[root].m_key; A[root++].m_key = next; } else A[next].m_key = A[leaf++].m_key;
175
+ if (leaf>=n || (root<next && A[root].m_key<A[leaf].m_key)) { A[next].m_key += A[root].m_key; A[root++].m_key = next; } else A[next].m_key += A[leaf++].m_key;
176
+ }
177
+ A[n-2].m_key = 0;
178
+ for (next=n-3; next>=0; next--) A[next].m_key = A[A[next].m_key].m_key+1;
179
+ avbl = 1; used = dpth = 0; root = n-2; next = n-1;
180
+ while (avbl>0)
181
+ {
182
+ while (root>=0 && (int)A[root].m_key==dpth) { used++; root--; }
183
+ while (avbl>used) { A[next--].m_key = dpth; avbl--; }
184
+ avbl = 2*used; dpth++; used = 0;
185
+ }
186
+ }
187
+
188
+ // Limits canonical Huffman code table's max code size to max_code_size.
189
+ static void huffman_enforce_max_code_size(int *pNum_codes, int code_list_len, int max_code_size)
190
+ {
191
+ if (code_list_len <= 1) return;
192
+
193
+ for (int i = max_code_size + 1; i <= MAX_HUFF_CODESIZE; i++) pNum_codes[max_code_size] += pNum_codes[i];
194
+
195
+ uint32 total = 0;
196
+ for (int i = max_code_size; i > 0; i--)
197
+ total += (((uint32)pNum_codes[i]) << (max_code_size - i));
198
+
199
+ while (total != (1UL << max_code_size))
200
+ {
201
+ pNum_codes[max_code_size]--;
202
+ for (int i = max_code_size - 1; i > 0; i--)
203
+ {
204
+ if (pNum_codes[i]) { pNum_codes[i]--; pNum_codes[i + 1] += 2; break; }
205
+ }
206
+ total--;
207
+ }
208
+ }
209
+
210
+ // Generates an optimized offman table.
211
+ void jpeg_encoder::optimize_huffman_table(int table_num, int table_len)
212
+ {
213
+ sym_freq syms0[MAX_HUFF_SYMBOLS], syms1[MAX_HUFF_SYMBOLS];
214
+ syms0[0].m_key = 1; syms0[0].m_sym_index = 0; // dummy symbol, assures that no valid code contains all 1's
215
+ int num_used_syms = 1;
216
+ const uint32 *pSym_count = &m_huff_count[table_num][0];
217
+ for (int i = 0; i < table_len; i++)
218
+ if (pSym_count[i]) { syms0[num_used_syms].m_key = pSym_count[i]; syms0[num_used_syms++].m_sym_index = i + 1; }
219
+ sym_freq* pSyms = radix_sort_syms(num_used_syms, syms0, syms1);
220
+ calculate_minimum_redundancy(pSyms, num_used_syms);
221
+
222
+ // Count the # of symbols of each code size.
223
+ int num_codes[1 + MAX_HUFF_CODESIZE]; clear_obj(num_codes);
224
+ for (int i = 0; i < num_used_syms; i++)
225
+ num_codes[pSyms[i].m_key]++;
226
+
227
+ const uint JPGE_CODE_SIZE_LIMIT = 16; // the maximum possible size of a JPEG Huffman code (valid range is [9,16] - 9 vs. 8 because of the dummy symbol)
228
+ huffman_enforce_max_code_size(num_codes, num_used_syms, JPGE_CODE_SIZE_LIMIT);
229
+
230
+ // Compute m_huff_bits array, which contains the # of symbols per code size.
231
+ clear_obj(m_huff_bits[table_num]);
232
+ for (int i = 1; i <= (int)JPGE_CODE_SIZE_LIMIT; i++)
233
+ m_huff_bits[table_num][i] = static_cast<uint8>(num_codes[i]);
234
+
235
+ // Remove the dummy symbol added above, which must be in largest bucket.
236
+ for (int i = JPGE_CODE_SIZE_LIMIT; i >= 1; i--)
237
+ {
238
+ if (m_huff_bits[table_num][i]) { m_huff_bits[table_num][i]--; break; }
239
+ }
240
+
241
+ // Compute the m_huff_val array, which contains the symbol indices sorted by code size (smallest to largest).
242
+ for (int i = num_used_syms - 1; i >= 1; i--)
243
+ m_huff_val[table_num][num_used_syms - 1 - i] = static_cast<uint8>(pSyms[i].m_sym_index - 1);
244
+ }
245
+
246
+ // JPEG marker generation.
247
+ void jpeg_encoder::emit_byte(uint8 i)
248
+ {
249
+ m_all_stream_writes_succeeded = m_all_stream_writes_succeeded && m_pStream->put_obj(i);
250
+ }
251
+
252
+ void jpeg_encoder::emit_word(uint i)
253
+ {
254
+ emit_byte(uint8(i >> 8)); emit_byte(uint8(i & 0xFF));
255
+ }
256
+
257
+ void jpeg_encoder::emit_marker(int marker)
258
+ {
259
+ emit_byte(uint8(0xFF)); emit_byte(uint8(marker));
260
+ }
261
+
262
+ // Emit JFIF marker
263
+ void jpeg_encoder::emit_jfif_app0()
264
+ {
265
+ emit_marker(M_APP0);
266
+ emit_word(2 + 4 + 1 + 2 + 1 + 2 + 2 + 1 + 1);
267
+ emit_byte(0x4A); emit_byte(0x46); emit_byte(0x49); emit_byte(0x46); /* Identifier: ASCII "JFIF" */
268
+ emit_byte(0);
269
+ emit_byte(1); /* Major version */
270
+ emit_byte(1); /* Minor version */
271
+ emit_byte(0); /* Density unit */
272
+ emit_word(1);
273
+ emit_word(1);
274
+ emit_byte(0); /* No thumbnail image */
275
+ emit_byte(0);
276
+ }
277
+
278
+ // Emit quantization tables
279
+ void jpeg_encoder::emit_dqt()
280
+ {
281
+ for (int i = 0; i < ((m_num_components == 3) ? 2 : 1); i++)
282
+ {
283
+ emit_marker(M_DQT);
284
+ emit_word(64 + 1 + 2);
285
+ emit_byte(static_cast<uint8>(i));
286
+ for (int j = 0; j < 64; j++)
287
+ emit_byte(static_cast<uint8>(m_quantization_tables[i][j]));
288
+ }
289
+ }
290
+
291
+ // Emit start of frame marker
292
+ void jpeg_encoder::emit_sof()
293
+ {
294
+ emit_marker(M_SOF0); /* baseline */
295
+ emit_word(3 * m_num_components + 2 + 5 + 1);
296
+ emit_byte(8); /* precision */
297
+ emit_word(m_image_y);
298
+ emit_word(m_image_x);
299
+ emit_byte(m_num_components);
300
+ for (int i = 0; i < m_num_components; i++)
301
+ {
302
+ emit_byte(static_cast<uint8>(i + 1)); /* component ID */
303
+ emit_byte((m_comp_h_samp[i] << 4) + m_comp_v_samp[i]); /* h and v sampling */
304
+ emit_byte(i > 0); /* quant. table num */
305
+ }
306
+ }
307
+
308
+ // Emit Huffman table.
309
+ void jpeg_encoder::emit_dht(uint8 *bits, uint8 *val, int index, bool ac_flag)
310
+ {
311
+ emit_marker(M_DHT);
312
+
313
+ int length = 0;
314
+ for (int i = 1; i <= 16; i++)
315
+ length += bits[i];
316
+
317
+ emit_word(length + 2 + 1 + 16);
318
+ emit_byte(static_cast<uint8>(index + (ac_flag << 4)));
319
+
320
+ for (int i = 1; i <= 16; i++)
321
+ emit_byte(bits[i]);
322
+
323
+ for (int i = 0; i < length; i++)
324
+ emit_byte(val[i]);
325
+ }
326
+
327
+ // Emit all Huffman tables.
328
+ void jpeg_encoder::emit_dhts()
329
+ {
330
+ emit_dht(m_huff_bits[0+0], m_huff_val[0+0], 0, false);
331
+ emit_dht(m_huff_bits[2+0], m_huff_val[2+0], 0, true);
332
+ if (m_num_components == 3)
333
+ {
334
+ emit_dht(m_huff_bits[0+1], m_huff_val[0+1], 1, false);
335
+ emit_dht(m_huff_bits[2+1], m_huff_val[2+1], 1, true);
336
+ }
337
+ }
338
+
339
+ // emit start of scan
340
+ void jpeg_encoder::emit_sos()
341
+ {
342
+ emit_marker(M_SOS);
343
+ emit_word(2 * m_num_components + 2 + 1 + 3);
344
+ emit_byte(m_num_components);
345
+ for (int i = 0; i < m_num_components; i++)
346
+ {
347
+ emit_byte(static_cast<uint8>(i + 1));
348
+ if (i == 0)
349
+ emit_byte((0 << 4) + 0);
350
+ else
351
+ emit_byte((1 << 4) + 1);
352
+ }
353
+ emit_byte(0); /* spectral selection */
354
+ emit_byte(63);
355
+ emit_byte(0);
356
+ }
357
+
358
+ // Emit all markers at beginning of image file.
359
+ void jpeg_encoder::emit_markers()
360
+ {
361
+ emit_marker(M_SOI);
362
+ emit_jfif_app0();
363
+ emit_dqt();
364
+ emit_sof();
365
+ emit_dhts();
366
+ emit_sos();
367
+ }
368
+
369
+ // Compute the actual canonical Huffman codes/code sizes given the JPEG huff bits and val arrays.
370
+ void jpeg_encoder::compute_huffman_table(uint *codes, uint8 *code_sizes, uint8 *bits, uint8 *val)
371
+ {
372
+ int i, l, last_p, si;
373
+ uint8 huff_size[257];
374
+ uint huff_code[257];
375
+ uint code;
376
+
377
+ int p = 0;
378
+ for (l = 1; l <= 16; l++)
379
+ for (i = 1; i <= bits[l]; i++)
380
+ huff_size[p++] = (char)l;
381
+
382
+ huff_size[p] = 0; last_p = p; // write sentinel
383
+
384
+ code = 0; si = huff_size[0]; p = 0;
385
+
386
+ while (huff_size[p])
387
+ {
388
+ while (huff_size[p] == si)
389
+ huff_code[p++] = code++;
390
+ code <<= 1;
391
+ si++;
392
+ }
393
+
394
+ memset(codes, 0, sizeof(codes[0])*256);
395
+ memset(code_sizes, 0, sizeof(code_sizes[0])*256);
396
+ for (p = 0; p < last_p; p++)
397
+ {
398
+ codes[val[p]] = huff_code[p];
399
+ code_sizes[val[p]] = huff_size[p];
400
+ }
401
+ }
402
+
403
+ // Quantization table generation.
404
+ void jpeg_encoder::compute_quant_table(int32 *pDst, int16 *pSrc)
405
+ {
406
+ int32 q;
407
+ if (m_params.m_quality < 50)
408
+ q = 5000 / m_params.m_quality;
409
+ else
410
+ q = 200 - m_params.m_quality * 2;
411
+ for (int i = 0; i < 64; i++)
412
+ {
413
+ int32 j = *pSrc++; j = (j * q + 50L) / 100L;
414
+ *pDst++ = JPGE_MIN(JPGE_MAX(j, 1), 255);
415
+ }
416
+ }
417
+
418
+ // Higher-level methods.
419
+ void jpeg_encoder::first_pass_init()
420
+ {
421
+ m_bit_buffer = 0; m_bits_in = 0;
422
+ memset(m_last_dc_val, 0, 3 * sizeof(m_last_dc_val[0]));
423
+ m_mcu_y_ofs = 0;
424
+ m_pass_num = 1;
425
+ }
426
+
427
+ bool jpeg_encoder::second_pass_init()
428
+ {
429
+ compute_huffman_table(&m_huff_codes[0+0][0], &m_huff_code_sizes[0+0][0], m_huff_bits[0+0], m_huff_val[0+0]);
430
+ compute_huffman_table(&m_huff_codes[2+0][0], &m_huff_code_sizes[2+0][0], m_huff_bits[2+0], m_huff_val[2+0]);
431
+ if (m_num_components > 1)
432
+ {
433
+ compute_huffman_table(&m_huff_codes[0+1][0], &m_huff_code_sizes[0+1][0], m_huff_bits[0+1], m_huff_val[0+1]);
434
+ compute_huffman_table(&m_huff_codes[2+1][0], &m_huff_code_sizes[2+1][0], m_huff_bits[2+1], m_huff_val[2+1]);
435
+ }
436
+ first_pass_init();
437
+ emit_markers();
438
+ m_pass_num = 2;
439
+ return true;
440
+ }
441
+
442
+ bool jpeg_encoder::jpg_open(int p_x_res, int p_y_res, int src_channels)
443
+ {
444
+ m_num_components = 3;
445
+ switch (m_params.m_subsampling)
446
+ {
447
+ case Y_ONLY:
448
+ {
449
+ m_num_components = 1;
450
+ m_comp_h_samp[0] = 1; m_comp_v_samp[0] = 1;
451
+ m_mcu_x = 8; m_mcu_y = 8;
452
+ break;
453
+ }
454
+ case H1V1:
455
+ {
456
+ m_comp_h_samp[0] = 1; m_comp_v_samp[0] = 1;
457
+ m_comp_h_samp[1] = 1; m_comp_v_samp[1] = 1;
458
+ m_comp_h_samp[2] = 1; m_comp_v_samp[2] = 1;
459
+ m_mcu_x = 8; m_mcu_y = 8;
460
+ break;
461
+ }
462
+ case H2V1:
463
+ {
464
+ m_comp_h_samp[0] = 2; m_comp_v_samp[0] = 1;
465
+ m_comp_h_samp[1] = 1; m_comp_v_samp[1] = 1;
466
+ m_comp_h_samp[2] = 1; m_comp_v_samp[2] = 1;
467
+ m_mcu_x = 16; m_mcu_y = 8;
468
+ break;
469
+ }
470
+ case H2V2:
471
+ {
472
+ m_comp_h_samp[0] = 2; m_comp_v_samp[0] = 2;
473
+ m_comp_h_samp[1] = 1; m_comp_v_samp[1] = 1;
474
+ m_comp_h_samp[2] = 1; m_comp_v_samp[2] = 1;
475
+ m_mcu_x = 16; m_mcu_y = 16;
476
+ }
477
+ }
478
+
479
+ m_image_x = p_x_res; m_image_y = p_y_res;
480
+ m_image_bpp = src_channels;
481
+ m_image_bpl = m_image_x * src_channels;
482
+ m_image_x_mcu = (m_image_x + m_mcu_x - 1) & (~(m_mcu_x - 1));
483
+ m_image_y_mcu = (m_image_y + m_mcu_y - 1) & (~(m_mcu_y - 1));
484
+ m_image_bpl_xlt = m_image_x * m_num_components;
485
+ m_image_bpl_mcu = m_image_x_mcu * m_num_components;
486
+ m_mcus_per_row = m_image_x_mcu / m_mcu_x;
487
+
488
+ if ((m_mcu_lines[0] = static_cast<uint8*>(jpge_malloc(m_image_bpl_mcu * m_mcu_y))) == NULL) return false;
489
+ for (int i = 1; i < m_mcu_y; i++)
490
+ m_mcu_lines[i] = m_mcu_lines[i-1] + m_image_bpl_mcu;
491
+
492
+ compute_quant_table(m_quantization_tables[0], s_std_lum_quant);
493
+ compute_quant_table(m_quantization_tables[1], m_params.m_no_chroma_discrim_flag ? s_std_lum_quant : s_std_croma_quant);
494
+
495
+ m_out_buf_left = JPGE_OUT_BUF_SIZE;
496
+ m_pOut_buf = m_out_buf;
497
+
498
+ if (m_params.m_two_pass_flag)
499
+ {
500
+ clear_obj(m_huff_count);
501
+ first_pass_init();
502
+ }
503
+ else
504
+ {
505
+ memcpy(m_huff_bits[0+0], s_dc_lum_bits, 17); memcpy(m_huff_val [0+0], s_dc_lum_val, DC_LUM_CODES);
506
+ memcpy(m_huff_bits[2+0], s_ac_lum_bits, 17); memcpy(m_huff_val [2+0], s_ac_lum_val, AC_LUM_CODES);
507
+ memcpy(m_huff_bits[0+1], s_dc_chroma_bits, 17); memcpy(m_huff_val [0+1], s_dc_chroma_val, DC_CHROMA_CODES);
508
+ memcpy(m_huff_bits[2+1], s_ac_chroma_bits, 17); memcpy(m_huff_val [2+1], s_ac_chroma_val, AC_CHROMA_CODES);
509
+ if (!second_pass_init()) return false; // in effect, skip over the first pass
510
+ }
511
+ return m_all_stream_writes_succeeded;
512
+ }
513
+
514
+ void jpeg_encoder::load_block_8_8_grey(int x)
515
+ {
516
+ uint8 *pSrc;
517
+ sample_array_t *pDst = m_sample_array;
518
+ x <<= 3;
519
+ for (int i = 0; i < 8; i++, pDst += 8)
520
+ {
521
+ pSrc = m_mcu_lines[i] + x;
522
+ pDst[0] = pSrc[0] - 128; pDst[1] = pSrc[1] - 128; pDst[2] = pSrc[2] - 128; pDst[3] = pSrc[3] - 128;
523
+ pDst[4] = pSrc[4] - 128; pDst[5] = pSrc[5] - 128; pDst[6] = pSrc[6] - 128; pDst[7] = pSrc[7] - 128;
524
+ }
525
+ }
526
+
527
+ void jpeg_encoder::load_block_8_8(int x, int y, int c)
528
+ {
529
+ uint8 *pSrc;
530
+ sample_array_t *pDst = m_sample_array;
531
+ x = (x * (8 * 3)) + c;
532
+ y <<= 3;
533
+ for (int i = 0; i < 8; i++, pDst += 8)
534
+ {
535
+ pSrc = m_mcu_lines[y + i] + x;
536
+ pDst[0] = pSrc[0 * 3] - 128; pDst[1] = pSrc[1 * 3] - 128; pDst[2] = pSrc[2 * 3] - 128; pDst[3] = pSrc[3 * 3] - 128;
537
+ pDst[4] = pSrc[4 * 3] - 128; pDst[5] = pSrc[5 * 3] - 128; pDst[6] = pSrc[6 * 3] - 128; pDst[7] = pSrc[7 * 3] - 128;
538
+ }
539
+ }
540
+
541
+ void jpeg_encoder::load_block_16_8(int x, int c)
542
+ {
543
+ uint8 *pSrc1, *pSrc2;
544
+ sample_array_t *pDst = m_sample_array;
545
+ x = (x * (16 * 3)) + c;
546
+ int a = 0, b = 2;
547
+ for (int i = 0; i < 16; i += 2, pDst += 8)
548
+ {
549
+ pSrc1 = m_mcu_lines[i + 0] + x;
550
+ pSrc2 = m_mcu_lines[i + 1] + x;
551
+ pDst[0] = ((pSrc1[ 0 * 3] + pSrc1[ 1 * 3] + pSrc2[ 0 * 3] + pSrc2[ 1 * 3] + a) >> 2) - 128; pDst[1] = ((pSrc1[ 2 * 3] + pSrc1[ 3 * 3] + pSrc2[ 2 * 3] + pSrc2[ 3 * 3] + b) >> 2) - 128;
552
+ pDst[2] = ((pSrc1[ 4 * 3] + pSrc1[ 5 * 3] + pSrc2[ 4 * 3] + pSrc2[ 5 * 3] + a) >> 2) - 128; pDst[3] = ((pSrc1[ 6 * 3] + pSrc1[ 7 * 3] + pSrc2[ 6 * 3] + pSrc2[ 7 * 3] + b) >> 2) - 128;
553
+ pDst[4] = ((pSrc1[ 8 * 3] + pSrc1[ 9 * 3] + pSrc2[ 8 * 3] + pSrc2[ 9 * 3] + a) >> 2) - 128; pDst[5] = ((pSrc1[10 * 3] + pSrc1[11 * 3] + pSrc2[10 * 3] + pSrc2[11 * 3] + b) >> 2) - 128;
554
+ pDst[6] = ((pSrc1[12 * 3] + pSrc1[13 * 3] + pSrc2[12 * 3] + pSrc2[13 * 3] + a) >> 2) - 128; pDst[7] = ((pSrc1[14 * 3] + pSrc1[15 * 3] + pSrc2[14 * 3] + pSrc2[15 * 3] + b) >> 2) - 128;
555
+ int temp = a; a = b; b = temp;
556
+ }
557
+ }
558
+
559
+ void jpeg_encoder::load_block_16_8_8(int x, int c)
560
+ {
561
+ uint8 *pSrc1;
562
+ sample_array_t *pDst = m_sample_array;
563
+ x = (x * (16 * 3)) + c;
564
+ for (int i = 0; i < 8; i++, pDst += 8)
565
+ {
566
+ pSrc1 = m_mcu_lines[i + 0] + x;
567
+ pDst[0] = ((pSrc1[ 0 * 3] + pSrc1[ 1 * 3]) >> 1) - 128; pDst[1] = ((pSrc1[ 2 * 3] + pSrc1[ 3 * 3]) >> 1) - 128;
568
+ pDst[2] = ((pSrc1[ 4 * 3] + pSrc1[ 5 * 3]) >> 1) - 128; pDst[3] = ((pSrc1[ 6 * 3] + pSrc1[ 7 * 3]) >> 1) - 128;
569
+ pDst[4] = ((pSrc1[ 8 * 3] + pSrc1[ 9 * 3]) >> 1) - 128; pDst[5] = ((pSrc1[10 * 3] + pSrc1[11 * 3]) >> 1) - 128;
570
+ pDst[6] = ((pSrc1[12 * 3] + pSrc1[13 * 3]) >> 1) - 128; pDst[7] = ((pSrc1[14 * 3] + pSrc1[15 * 3]) >> 1) - 128;
571
+ }
572
+ }
573
+
574
+ void jpeg_encoder::load_quantized_coefficients(int component_num)
575
+ {
576
+ int32 *q = m_quantization_tables[component_num > 0];
577
+ int16 *pDst = m_coefficient_array;
578
+ for (int i = 0; i < 64; i++)
579
+ {
580
+ sample_array_t j = m_sample_array[s_zag[i]];
581
+ if (j < 0)
582
+ {
583
+ if ((j = -j + (*q >> 1)) < *q)
584
+ *pDst++ = 0;
585
+ else
586
+ *pDst++ = static_cast<int16>(-(j / *q));
587
+ }
588
+ else
589
+ {
590
+ if ((j = j + (*q >> 1)) < *q)
591
+ *pDst++ = 0;
592
+ else
593
+ *pDst++ = static_cast<int16>((j / *q));
594
+ }
595
+ q++;
596
+ }
597
+ }
598
+
599
+ void jpeg_encoder::flush_output_buffer()
600
+ {
601
+ if (m_out_buf_left != JPGE_OUT_BUF_SIZE)
602
+ m_all_stream_writes_succeeded = m_all_stream_writes_succeeded && m_pStream->put_buf(m_out_buf, JPGE_OUT_BUF_SIZE - m_out_buf_left);
603
+ m_pOut_buf = m_out_buf;
604
+ m_out_buf_left = JPGE_OUT_BUF_SIZE;
605
+ }
606
+
607
+ void jpeg_encoder::put_bits(uint bits, uint len)
608
+ {
609
+ m_bit_buffer |= ((uint32)bits << (24 - (m_bits_in += len)));
610
+ while (m_bits_in >= 8)
611
+ {
612
+ uint8 c;
613
+ #define JPGE_PUT_BYTE(c) { *m_pOut_buf++ = (c); if (--m_out_buf_left == 0) flush_output_buffer(); }
614
+ JPGE_PUT_BYTE(c = (uint8)((m_bit_buffer >> 16) & 0xFF));
615
+ if (c == 0xFF) JPGE_PUT_BYTE(0);
616
+ m_bit_buffer <<= 8;
617
+ m_bits_in -= 8;
618
+ }
619
+ }
620
+
621
+ void jpeg_encoder::code_coefficients_pass_one(int component_num)
622
+ {
623
+ if (component_num >= 3) return; // just to shut up static analysis
624
+ int i, run_len, nbits, temp1;
625
+ int16 *src = m_coefficient_array;
626
+ uint32 *dc_count = component_num ? m_huff_count[0 + 1] : m_huff_count[0 + 0], *ac_count = component_num ? m_huff_count[2 + 1] : m_huff_count[2 + 0];
627
+
628
+ temp1 = src[0] - m_last_dc_val[component_num];
629
+ m_last_dc_val[component_num] = src[0];
630
+ if (temp1 < 0) temp1 = -temp1;
631
+
632
+ nbits = 0;
633
+ while (temp1)
634
+ {
635
+ nbits++; temp1 >>= 1;
636
+ }
637
+
638
+ dc_count[nbits]++;
639
+ for (run_len = 0, i = 1; i < 64; i++)
640
+ {
641
+ if ((temp1 = m_coefficient_array[i]) == 0)
642
+ run_len++;
643
+ else
644
+ {
645
+ while (run_len >= 16)
646
+ {
647
+ ac_count[0xF0]++;
648
+ run_len -= 16;
649
+ }
650
+ if (temp1 < 0) temp1 = -temp1;
651
+ nbits = 1;
652
+ while (temp1 >>= 1) nbits++;
653
+ ac_count[(run_len << 4) + nbits]++;
654
+ run_len = 0;
655
+ }
656
+ }
657
+ if (run_len) ac_count[0]++;
658
+ }
659
+
660
+ void jpeg_encoder::code_coefficients_pass_two(int component_num)
661
+ {
662
+ int i, j, run_len, nbits, temp1, temp2;
663
+ int16 *pSrc = m_coefficient_array;
664
+ uint *codes[2];
665
+ uint8 *code_sizes[2];
666
+
667
+ if (component_num == 0)
668
+ {
669
+ codes[0] = m_huff_codes[0 + 0]; codes[1] = m_huff_codes[2 + 0];
670
+ code_sizes[0] = m_huff_code_sizes[0 + 0]; code_sizes[1] = m_huff_code_sizes[2 + 0];
671
+ }
672
+ else
673
+ {
674
+ codes[0] = m_huff_codes[0 + 1]; codes[1] = m_huff_codes[2 + 1];
675
+ code_sizes[0] = m_huff_code_sizes[0 + 1]; code_sizes[1] = m_huff_code_sizes[2 + 1];
676
+ }
677
+
678
+ temp1 = temp2 = pSrc[0] - m_last_dc_val[component_num];
679
+ m_last_dc_val[component_num] = pSrc[0];
680
+
681
+ if (temp1 < 0)
682
+ {
683
+ temp1 = -temp1; temp2--;
684
+ }
685
+
686
+ nbits = 0;
687
+ while (temp1)
688
+ {
689
+ nbits++; temp1 >>= 1;
690
+ }
691
+
692
+ put_bits(codes[0][nbits], code_sizes[0][nbits]);
693
+ if (nbits) put_bits(temp2 & ((1 << nbits) - 1), nbits);
694
+
695
+ for (run_len = 0, i = 1; i < 64; i++)
696
+ {
697
+ if ((temp1 = m_coefficient_array[i]) == 0)
698
+ run_len++;
699
+ else
700
+ {
701
+ while (run_len >= 16)
702
+ {
703
+ put_bits(codes[1][0xF0], code_sizes[1][0xF0]);
704
+ run_len -= 16;
705
+ }
706
+ if ((temp2 = temp1) < 0)
707
+ {
708
+ temp1 = -temp1;
709
+ temp2--;
710
+ }
711
+ nbits = 1;
712
+ while (temp1 >>= 1)
713
+ nbits++;
714
+ j = (run_len << 4) + nbits;
715
+ put_bits(codes[1][j], code_sizes[1][j]);
716
+ put_bits(temp2 & ((1 << nbits) - 1), nbits);
717
+ run_len = 0;
718
+ }
719
+ }
720
+ if (run_len)
721
+ put_bits(codes[1][0], code_sizes[1][0]);
722
+ }
723
+
724
+ void jpeg_encoder::code_block(int component_num)
725
+ {
726
+ DCT2D(m_sample_array);
727
+ load_quantized_coefficients(component_num);
728
+ if (m_pass_num == 1)
729
+ code_coefficients_pass_one(component_num);
730
+ else
731
+ code_coefficients_pass_two(component_num);
732
+ }
733
+
734
+ void jpeg_encoder::process_mcu_row()
735
+ {
736
+ if (m_num_components == 1)
737
+ {
738
+ for (int i = 0; i < m_mcus_per_row; i++)
739
+ {
740
+ load_block_8_8_grey(i); code_block(0);
741
+ }
742
+ }
743
+ else if ((m_comp_h_samp[0] == 1) && (m_comp_v_samp[0] == 1))
744
+ {
745
+ for (int i = 0; i < m_mcus_per_row; i++)
746
+ {
747
+ load_block_8_8(i, 0, 0); code_block(0); load_block_8_8(i, 0, 1); code_block(1); load_block_8_8(i, 0, 2); code_block(2);
748
+ }
749
+ }
750
+ else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 1))
751
+ {
752
+ for (int i = 0; i < m_mcus_per_row; i++)
753
+ {
754
+ load_block_8_8(i * 2 + 0, 0, 0); code_block(0); load_block_8_8(i * 2 + 1, 0, 0); code_block(0);
755
+ load_block_16_8_8(i, 1); code_block(1); load_block_16_8_8(i, 2); code_block(2);
756
+ }
757
+ }
758
+ else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 2))
759
+ {
760
+ for (int i = 0; i < m_mcus_per_row; i++)
761
+ {
762
+ load_block_8_8(i * 2 + 0, 0, 0); code_block(0); load_block_8_8(i * 2 + 1, 0, 0); code_block(0);
763
+ load_block_8_8(i * 2 + 0, 1, 0); code_block(0); load_block_8_8(i * 2 + 1, 1, 0); code_block(0);
764
+ load_block_16_8(i, 1); code_block(1); load_block_16_8(i, 2); code_block(2);
765
+ }
766
+ }
767
+ }
768
+
769
+ bool jpeg_encoder::terminate_pass_one()
770
+ {
771
+ optimize_huffman_table(0+0, DC_LUM_CODES); optimize_huffman_table(2+0, AC_LUM_CODES);
772
+ if (m_num_components > 1)
773
+ {
774
+ optimize_huffman_table(0+1, DC_CHROMA_CODES); optimize_huffman_table(2+1, AC_CHROMA_CODES);
775
+ }
776
+ return second_pass_init();
777
+ }
778
+
779
+ bool jpeg_encoder::terminate_pass_two()
780
+ {
781
+ put_bits(0x7F, 7);
782
+ flush_output_buffer();
783
+ emit_marker(M_EOI);
784
+ m_pass_num++; // purposely bump up m_pass_num, for debugging
785
+ return true;
786
+ }
787
+
788
+ bool jpeg_encoder::process_end_of_image()
789
+ {
790
+ if (m_mcu_y_ofs)
791
+ {
792
+ if (m_mcu_y_ofs < 16) // check here just to shut up static analysis
793
+ {
794
+ for (int i = m_mcu_y_ofs; i < m_mcu_y; i++)
795
+ memcpy(m_mcu_lines[i], m_mcu_lines[m_mcu_y_ofs - 1], m_image_bpl_mcu);
796
+ }
797
+
798
+ process_mcu_row();
799
+ }
800
+
801
+ if (m_pass_num == 1)
802
+ return terminate_pass_one();
803
+ else
804
+ return terminate_pass_two();
805
+ }
806
+
807
+ void jpeg_encoder::load_mcu(const void *pSrc)
808
+ {
809
+ const uint8* Psrc = reinterpret_cast<const uint8*>(pSrc);
810
+
811
+ uint8* pDst = m_mcu_lines[m_mcu_y_ofs]; // OK to write up to m_image_bpl_xlt bytes to pDst
812
+
813
+ if (m_num_components == 1)
814
+ {
815
+ if (m_image_bpp == 4)
816
+ RGBA_to_Y(pDst, Psrc, m_image_x);
817
+ else if (m_image_bpp == 3)
818
+ RGB_to_Y(pDst, Psrc, m_image_x);
819
+ else
820
+ memcpy(pDst, Psrc, m_image_x);
821
+ }
822
+ else
823
+ {
824
+ if (m_image_bpp == 4)
825
+ RGBA_to_YCC(pDst, Psrc, m_image_x);
826
+ else if (m_image_bpp == 3)
827
+ RGB_to_YCC(pDst, Psrc, m_image_x);
828
+ else
829
+ Y_to_YCC(pDst, Psrc, m_image_x);
830
+ }
831
+
832
+ // Possibly duplicate pixels at end of scanline if not a multiple of 8 or 16
833
+ if (m_num_components == 1)
834
+ memset(m_mcu_lines[m_mcu_y_ofs] + m_image_bpl_xlt, pDst[m_image_bpl_xlt - 1], m_image_x_mcu - m_image_x);
835
+ else
836
+ {
837
+ const uint8 y = pDst[m_image_bpl_xlt - 3 + 0], cb = pDst[m_image_bpl_xlt - 3 + 1], cr = pDst[m_image_bpl_xlt - 3 + 2];
838
+ uint8 *q = m_mcu_lines[m_mcu_y_ofs] + m_image_bpl_xlt;
839
+ for (int i = m_image_x; i < m_image_x_mcu; i++)
840
+ {
841
+ *q++ = y; *q++ = cb; *q++ = cr;
842
+ }
843
+ }
844
+
845
+ if (++m_mcu_y_ofs == m_mcu_y)
846
+ {
847
+ process_mcu_row();
848
+ m_mcu_y_ofs = 0;
849
+ }
850
+ }
851
+
852
+ void jpeg_encoder::clear()
853
+ {
854
+ m_mcu_lines[0] = NULL;
855
+ m_pass_num = 0;
856
+ m_all_stream_writes_succeeded = true;
857
+ }
858
+
859
+ jpeg_encoder::jpeg_encoder()
860
+ {
861
+ clear();
862
+ }
863
+
864
+ jpeg_encoder::~jpeg_encoder()
865
+ {
866
+ deinit();
867
+ }
868
+
869
+ bool jpeg_encoder::init(output_stream *pStream, int64_t width, int64_t height, int64_t src_channels, const params &comp_params)
870
+ {
871
+ deinit();
872
+ if (((!pStream) || (width < 1) || (height < 1)) || ((src_channels != 1) && (src_channels != 3) && (src_channels != 4)) || (!comp_params.check_valid())) return false;
873
+ m_pStream = pStream;
874
+ m_params = comp_params;
875
+ return jpg_open(width, height, src_channels);
876
+ }
877
+
878
+ void jpeg_encoder::deinit()
879
+ {
880
+ jpge_free(m_mcu_lines[0]);
881
+ clear();
882
+ }
883
+
884
+ bool jpeg_encoder::process_scanline(const void* pScanline)
885
+ {
886
+ if ((m_pass_num < 1) || (m_pass_num > 2)) return false;
887
+ if (m_all_stream_writes_succeeded)
888
+ {
889
+ if (!pScanline)
890
+ {
891
+ if (!process_end_of_image()) return false;
892
+ }
893
+ else
894
+ {
895
+ load_mcu(pScanline);
896
+ }
897
+ }
898
+ return m_all_stream_writes_succeeded;
899
+ }
900
+
901
+ // Higher level wrappers/examples (optional).
902
+ #include <stdio.h>
903
+
904
+ class cfile_stream : public output_stream
905
+ {
906
+ cfile_stream(const cfile_stream &);
907
+ cfile_stream &operator= (const cfile_stream &);
908
+
909
+ FILE* m_pFile;
910
+ bool m_bStatus;
911
+
912
+ public:
913
+ cfile_stream() : m_pFile(NULL), m_bStatus(false) { }
914
+
915
+ virtual ~cfile_stream()
916
+ {
917
+ close();
918
+ }
919
+
920
+ bool open(const char *pFilename)
921
+ {
922
+ close();
923
+ #if defined(_MSC_VER)
924
+ if (fopen_s(&m_pFile, pFilename, "wb") != 0)
925
+ {
926
+ return false;
927
+ }
928
+ #else
929
+ m_pFile = fopen(pFilename, "wb");
930
+ #endif
931
+ m_bStatus = (m_pFile != NULL);
932
+ return m_bStatus;
933
+ }
934
+
935
+ bool close()
936
+ {
937
+ if (m_pFile)
938
+ {
939
+ if (fclose(m_pFile) == EOF)
940
+ {
941
+ m_bStatus = false;
942
+ }
943
+ m_pFile = NULL;
944
+ }
945
+ return m_bStatus;
946
+ }
947
+
948
+ virtual bool put_buf(const void* pBuf, int64_t len)
949
+ {
950
+ m_bStatus = m_bStatus && (fwrite(pBuf, len, 1, m_pFile) == 1);
951
+ return m_bStatus;
952
+ }
953
+
954
+ uint get_size() const
955
+ {
956
+ return m_pFile ? ftell(m_pFile) : 0;
957
+ }
958
+ };
959
+
960
+ // Writes JPEG image to file.
961
+ bool compress_image_to_jpeg_file(const char *pFilename, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params)
962
+ {
963
+ cfile_stream dst_stream;
964
+ if (!dst_stream.open(pFilename))
965
+ return false;
966
+
967
+ jpge::jpeg_encoder dst_image;
968
+ if (!dst_image.init(&dst_stream, width, height, num_channels, comp_params))
969
+ return false;
970
+
971
+ for (uint pass_index = 0; pass_index < dst_image.get_total_passes(); pass_index++)
972
+ {
973
+ for (int64_t i = 0; i < height; i++)
974
+ {
975
+ // i, width, and num_channels are all 64bit
976
+ const uint8* pBuf = pImage_data + i * width * num_channels;
977
+ if (!dst_image.process_scanline(pBuf))
978
+ return false;
979
+ }
980
+ if (!dst_image.process_scanline(NULL))
981
+ return false;
982
+ }
983
+
984
+ dst_image.deinit();
985
+
986
+ return dst_stream.close();
987
+ }
988
+
989
+ class memory_stream : public output_stream
990
+ {
991
+ memory_stream(const memory_stream &);
992
+ memory_stream &operator= (const memory_stream &);
993
+
994
+ uint8 *m_pBuf;
995
+ uint64_t m_buf_size, m_buf_ofs;
996
+
997
+ public:
998
+ memory_stream(void *pBuf, uint64_t buf_size) : m_pBuf(static_cast<uint8*>(pBuf)), m_buf_size(buf_size), m_buf_ofs(0) { }
999
+
1000
+ virtual ~memory_stream() { }
1001
+
1002
+ virtual bool put_buf(const void* pBuf, int64_t len)
1003
+ {
1004
+ uint64_t buf_remaining = m_buf_size - m_buf_ofs;
1005
+ if ((uint64_t)len > buf_remaining)
1006
+ return false;
1007
+ memcpy(m_pBuf + m_buf_ofs, pBuf, len);
1008
+ m_buf_ofs += len;
1009
+ return true;
1010
+ }
1011
+
1012
+ uint64_t get_size() const
1013
+ {
1014
+ return m_buf_ofs;
1015
+ }
1016
+ };
1017
+
1018
+ bool compress_image_to_jpeg_file_in_memory(void *pDstBuf, int64_t &buf_size, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params)
1019
+ {
1020
+ if ((!pDstBuf) || (!buf_size))
1021
+ return false;
1022
+
1023
+ memory_stream dst_stream(pDstBuf, buf_size);
1024
+
1025
+ buf_size = 0;
1026
+
1027
+ jpge::jpeg_encoder dst_image;
1028
+ if (!dst_image.init(&dst_stream, width, height, num_channels, comp_params))
1029
+ return false;
1030
+
1031
+ for (uint pass_index = 0; pass_index < dst_image.get_total_passes(); pass_index++)
1032
+ {
1033
+ for (int64_t i = 0; i < height; i++)
1034
+ {
1035
+ const uint8* pScanline = pImage_data + i * width * num_channels;
1036
+ if (!dst_image.process_scanline(pScanline))
1037
+ return false;
1038
+ }
1039
+ if (!dst_image.process_scanline(NULL))
1040
+ return false;
1041
+ }
1042
+
1043
+ dst_image.deinit();
1044
+
1045
+ buf_size = dst_stream.get_size();
1046
+ return true;
1047
+ }
1048
+
1049
+ } // namespace jpge
crazy_functions/test_project/cpp/libJPG/jpge.h ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ // jpge.h - C++ class for JPEG compression.
3
+ // Public domain, Rich Geldreich <richgel99@gmail.com>
4
+ // Alex Evans: Added RGBA support, linear memory allocator.
5
+ #ifndef JPEG_ENCODER_H
6
+ #define JPEG_ENCODER_H
7
+
8
+ #include <stdint.h>
9
+
10
+ namespace jpge
11
+ {
12
+ typedef unsigned char uint8;
13
+ typedef signed short int16;
14
+ typedef signed int int32;
15
+ typedef unsigned short uint16;
16
+ typedef unsigned int uint32;
17
+ typedef unsigned int uint;
18
+
19
+ // JPEG chroma subsampling factors. Y_ONLY (grayscale images) and H2V2 (color images) are the most common.
20
+ enum subsampling_t { Y_ONLY = 0, H1V1 = 1, H2V1 = 2, H2V2 = 3 };
21
+
22
+ // JPEG compression parameters structure.
23
+ struct params
24
+ {
25
+ inline params() : m_quality(85), m_subsampling(H2V2), m_no_chroma_discrim_flag(false), m_two_pass_flag(false) { }
26
+
27
+ inline bool check_valid() const
28
+ {
29
+ if ((m_quality < 1) || (m_quality > 100)) return false;
30
+ if ((uint)m_subsampling > (uint)H2V2) return false;
31
+ return true;
32
+ }
33
+
34
+ // Quality: 1-100, higher is better. Typical values are around 50-95.
35
+ int m_quality;
36
+
37
+ // m_subsampling:
38
+ // 0 = Y (grayscale) only
39
+ // 1 = YCbCr, no subsampling (H1V1, YCbCr 1x1x1, 3 blocks per MCU)
40
+ // 2 = YCbCr, H2V1 subsampling (YCbCr 2x1x1, 4 blocks per MCU)
41
+ // 3 = YCbCr, H2V2 subsampling (YCbCr 4x1x1, 6 blocks per MCU-- very common)
42
+ subsampling_t m_subsampling;
43
+
44
+ // Disables CbCr discrimination - only intended for testing.
45
+ // If true, the Y quantization table is also used for the CbCr channels.
46
+ bool m_no_chroma_discrim_flag;
47
+
48
+ bool m_two_pass_flag;
49
+ };
50
+
51
+ // Writes JPEG image to a file.
52
+ // num_channels must be 1 (Y) or 3 (RGB), image pitch must be width*num_channels.
53
+ bool compress_image_to_jpeg_file(const char *pFilename, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params = params());
54
+
55
+ // Writes JPEG image to memory buffer.
56
+ // On entry, buf_size is the size of the output buffer pointed at by pBuf, which should be at least ~1024 bytes.
57
+ // If return value is true, buf_size will be set to the size of the compressed data.
58
+ bool compress_image_to_jpeg_file_in_memory(void *pBuf, int64_t &buf_size, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params = params());
59
+
60
+ // Output stream abstract class - used by the jpeg_encoder class to write to the output stream.
61
+ // put_buf() is generally called with len==JPGE_OUT_BUF_SIZE bytes, but for headers it'll be called with smaller amounts.
62
+ class output_stream
63
+ {
64
+ public:
65
+ virtual ~output_stream() { };
66
+ virtual bool put_buf(const void* Pbuf, int64_t len) = 0;
67
+ template<class T> inline bool put_obj(const T& obj) { return put_buf(&obj, sizeof(T)); }
68
+ };
69
+
70
+ // Lower level jpeg_encoder class - useful if more control is needed than the above helper functions.
71
+ class jpeg_encoder
72
+ {
73
+ public:
74
+ jpeg_encoder();
75
+ ~jpeg_encoder();
76
+
77
+ // Initializes the compressor.
78
+ // pStream: The stream object to use for writing compressed data.
79
+ // params - Compression parameters structure, defined above.
80
+ // width, height - Image dimensions.
81
+ // channels - May be 1, or 3. 1 indicates grayscale, 3 indicates RGB source data.
82
+ // Returns false on out of memory or if a stream write fails.
83
+ bool init(output_stream *pStream, int64_t width, int64_t height, int64_t src_channels, const params &comp_params = params());
84
+
85
+ const params &get_params() const { return m_params; }
86
+
87
+ // Deinitializes the compressor, freeing any allocated memory. May be called at any time.
88
+ void deinit();
89
+
90
+ uint get_total_passes() const { return m_params.m_two_pass_flag ? 2 : 1; }
91
+ inline uint get_cur_pass() { return m_pass_num; }
92
+
93
+ // Call this method with each source scanline.
94
+ // width * src_channels bytes per scanline is expected (RGB or Y format).
95
+ // You must call with NULL after all scanlines are processed to finish compression.
96
+ // Returns false on out of memory or if a stream write fails.
97
+ bool process_scanline(const void* pScanline);
98
+
99
+ private:
100
+ jpeg_encoder(const jpeg_encoder &);
101
+ jpeg_encoder &operator =(const jpeg_encoder &);
102
+
103
+ typedef int32 sample_array_t;
104
+
105
+ output_stream *m_pStream;
106
+ params m_params;
107
+ uint8 m_num_components;
108
+ uint8 m_comp_h_samp[3], m_comp_v_samp[3];
109
+ int m_image_x, m_image_y, m_image_bpp, m_image_bpl;
110
+ int m_image_x_mcu, m_image_y_mcu;
111
+ int m_image_bpl_xlt, m_image_bpl_mcu;
112
+ int m_mcus_per_row;
113
+ int m_mcu_x, m_mcu_y;
114
+ uint8 *m_mcu_lines[16];
115
+ uint8 m_mcu_y_ofs;
116
+ sample_array_t m_sample_array[64];
117
+ int16 m_coefficient_array[64];
118
+ int32 m_quantization_tables[2][64];
119
+ uint m_huff_codes[4][256];
120
+ uint8 m_huff_code_sizes[4][256];
121
+ uint8 m_huff_bits[4][17];
122
+ uint8 m_huff_val[4][256];
123
+ uint32 m_huff_count[4][256];
124
+ int m_last_dc_val[3];
125
+ enum { JPGE_OUT_BUF_SIZE = 2048 };
126
+ uint8 m_out_buf[JPGE_OUT_BUF_SIZE];
127
+ uint8 *m_pOut_buf;
128
+ uint m_out_buf_left;
129
+ uint32 m_bit_buffer;
130
+ uint m_bits_in;
131
+ uint8 m_pass_num;
132
+ bool m_all_stream_writes_succeeded;
133
+
134
+ void optimize_huffman_table(int table_num, int table_len);
135
+ void emit_byte(uint8 i);
136
+ void emit_word(uint i);
137
+ void emit_marker(int marker);
138
+ void emit_jfif_app0();
139
+ void emit_dqt();
140
+ void emit_sof();
141
+ void emit_dht(uint8 *bits, uint8 *val, int index, bool ac_flag);
142
+ void emit_dhts();
143
+ void emit_sos();
144
+ void emit_markers();
145
+ void compute_huffman_table(uint *codes, uint8 *code_sizes, uint8 *bits, uint8 *val);
146
+ void compute_quant_table(int32 *dst, int16 *src);
147
+ void adjust_quant_table(int32 *dst, int32 *src);
148
+ void first_pass_init();
149
+ bool second_pass_init();
150
+ bool jpg_open(int p_x_res, int p_y_res, int src_channels);
151
+ void load_block_8_8_grey(int x);
152
+ void load_block_8_8(int x, int y, int c);
153
+ void load_block_16_8(int x, int c);
154
+ void load_block_16_8_8(int x, int c);
155
+ void load_quantized_coefficients(int component_num);
156
+ void flush_output_buffer();
157
+ void put_bits(uint bits, uint len);
158
+ void code_coefficients_pass_one(int component_num);
159
+ void code_coefficients_pass_two(int component_num);
160
+ void code_block(int component_num);
161
+ void process_mcu_row();
162
+ bool terminate_pass_one();
163
+ bool terminate_pass_two();
164
+ bool process_end_of_image();
165
+ void load_mcu(const void* src);
166
+ void clear();
167
+ void init();
168
+ };
169
+
170
+ } // namespace jpge
171
+
172
+ #endif // JPEG_ENCODER
crazy_functions/test_project/cpp/libJPG/来源 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ jpge.h - C++ class for JPEG compression.
2
+ Public domain, Rich Geldreich <richgel99@gmail.com>
3
+ Alex Evans: Added RGBA support, linear memory allocator.
crazy_functions/test_project/cpp/longcode/jpgd.cpp ADDED
@@ -0,0 +1,3276 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ // jpgd.cpp - C++ class for JPEG decompression.
2
+ // Public domain, Rich Geldreich <richgel99@gmail.com>
3
+ // Last updated Apr. 16, 2011
4
+ // Alex Evans: Linear memory allocator (taken from jpge.h).
5
+ //
6
+ // Supports progressive and baseline sequential JPEG image files, and the most common chroma subsampling factors: Y, H1V1, H2V1, H1V2, and H2V2.
7
+ //
8
+ // Chroma upsampling quality: H2V2 is upsampled in the frequency domain, H2V1 and H1V2 are upsampled using point sampling.
9
+ // Chroma upsampling reference: "Fast Scheme for Image Size Change in the Compressed Domain"
10
+ // http://vision.ai.uiuc.edu/~dugad/research/dct/index.html
11
+
12
+ #include "jpgd.h"
13
+ #include <string.h>
14
+
15
+ #include <assert.h>
16
+ // BEGIN EPIC MOD
17
+ #define JPGD_ASSERT(x) { assert(x); CA_ASSUME(x); } (void)0
18
+ // END EPIC MOD
19
+
20
+ #ifdef _MSC_VER
21
+ #pragma warning (disable : 4611) // warning C4611: interaction between '_setjmp' and C++ object destruction is non-portable
22
+ #endif
23
+
24
+ // Set to 1 to enable freq. domain chroma upsampling on images using H2V2 subsampling (0=faster nearest neighbor sampling).
25
+ // This is slower, but results in higher quality on images with highly saturated colors.
26
+ #define JPGD_SUPPORT_FREQ_DOMAIN_UPSAMPLING 1
27
+
28
+ #define JPGD_TRUE (1)
29
+ #define JPGD_FALSE (0)
30
+
31
+ #define JPGD_MAX(a,b) (((a)>(b)) ? (a) : (b))
32
+ #define JPGD_MIN(a,b) (((a)<(b)) ? (a) : (b))
33
+
34
+ namespace jpgd {
35
+
36
+ static inline void *jpgd_malloc(size_t nSize) { return FMemory::Malloc(nSize); }
37
+ static inline void jpgd_free(void *p) { FMemory::Free(p); }
38
+
39
+ // BEGIN EPIC MOD
40
+ //@UE3 - use UE3 BGRA encoding instead of assuming RGBA
41
+ // stolen from IImageWrapper.h
42
+ enum ERGBFormatJPG
43
+ {
44
+ Invalid = -1,
45
+ RGBA = 0,
46
+ BGRA = 1,
47
+ Gray = 2,
48
+ };
49
+ static ERGBFormatJPG jpg_format;
50
+ // END EPIC MOD
51
+
52
+ // DCT coefficients are stored in this sequence.
53
+ static int g_ZAG[64] = { 0,1,8,16,9,2,3,10,17,24,32,25,18,11,4,5,12,19,26,33,40,48,41,34,27,20,13,6,7,14,21,28,35,42,49,56,57,50,43,36,29,22,15,23,30,37,44,51,58,59,52,45,38,31,39,46,53,60,61,54,47,55,62,63 };
54
+
55
+ enum JPEG_MARKER
56
+ {
57
+ M_SOF0 = 0xC0, M_SOF1 = 0xC1, M_SOF2 = 0xC2, M_SOF3 = 0xC3, M_SOF5 = 0xC5, M_SOF6 = 0xC6, M_SOF7 = 0xC7, M_JPG = 0xC8,
58
+ M_SOF9 = 0xC9, M_SOF10 = 0xCA, M_SOF11 = 0xCB, M_SOF13 = 0xCD, M_SOF14 = 0xCE, M_SOF15 = 0xCF, M_DHT = 0xC4, M_DAC = 0xCC,
59
+ M_RST0 = 0xD0, M_RST1 = 0xD1, M_RST2 = 0xD2, M_RST3 = 0xD3, M_RST4 = 0xD4, M_RST5 = 0xD5, M_RST6 = 0xD6, M_RST7 = 0xD7,
60
+ M_SOI = 0xD8, M_EOI = 0xD9, M_SOS = 0xDA, M_DQT = 0xDB, M_DNL = 0xDC, M_DRI = 0xDD, M_DHP = 0xDE, M_EXP = 0xDF,
61
+ M_APP0 = 0xE0, M_APP15 = 0xEF, M_JPG0 = 0xF0, M_JPG13 = 0xFD, M_COM = 0xFE, M_TEM = 0x01, M_ERROR = 0x100, RST0 = 0xD0
62
+ };
63
+
64
+ enum JPEG_SUBSAMPLING { JPGD_GRAYSCALE = 0, JPGD_YH1V1, JPGD_YH2V1, JPGD_YH1V2, JPGD_YH2V2 };
65
+
66
+ #define CONST_BITS 13
67
+ #define PASS1_BITS 2
68
+ #define SCALEDONE ((int32)1)
69
+
70
+ #define FIX_0_298631336 ((int32)2446) /* FIX(0.298631336) */
71
+ #define FIX_0_390180644 ((int32)3196) /* FIX(0.390180644) */
72
+ #define FIX_0_541196100 ((int32)4433) /* FIX(0.541196100) */
73
+ #define FIX_0_765366865 ((int32)6270) /* FIX(0.765366865) */
74
+ #define FIX_0_899976223 ((int32)7373) /* FIX(0.899976223) */
75
+ #define FIX_1_175875602 ((int32)9633) /* FIX(1.175875602) */
76
+ #define FIX_1_501321110 ((int32)12299) /* FIX(1.501321110) */
77
+ #define FIX_1_847759065 ((int32)15137) /* FIX(1.847759065) */
78
+ #define FIX_1_961570560 ((int32)16069) /* FIX(1.961570560) */
79
+ #define FIX_2_053119869 ((int32)16819) /* FIX(2.053119869) */
80
+ #define FIX_2_562915447 ((int32)20995) /* FIX(2.562915447) */
81
+ #define FIX_3_072711026 ((int32)25172) /* FIX(3.072711026) */
82
+
83
+ #define DESCALE(x,n) (((x) + (SCALEDONE << ((n)-1))) >> (n))
84
+ #define DESCALE_ZEROSHIFT(x,n) (((x) + (128 << (n)) + (SCALEDONE << ((n)-1))) >> (n))
85
+
86
+ #define MULTIPLY(var, cnst) ((var) * (cnst))
87
+
88
+ #define CLAMP(i) ((static_cast<uint>(i) > 255) ? (((~i) >> 31) & 0xFF) : (i))
89
+
90
+ // Compiler creates a fast path 1D IDCT for X non-zero columns
91
+ template <int NONZERO_COLS>
92
+ struct Row
93
+ {
94
+ static void idct(int* pTemp, const jpgd_block_t* pSrc)
95
+ {
96
+ // ACCESS_COL() will be optimized at compile time to either an array access, or 0.
97
+ #define ACCESS_COL(x) (((x) < NONZERO_COLS) ? (int)pSrc[x] : 0)
98
+
99
+ const int z2 = ACCESS_COL(2), z3 = ACCESS_COL(6);
100
+
101
+ const int z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
102
+ const int tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
103
+ const int tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
104
+
105
+ const int tmp0 = (ACCESS_COL(0) + ACCESS_COL(4)) << CONST_BITS;
106
+ const int tmp1 = (ACCESS_COL(0) - ACCESS_COL(4)) << CONST_BITS;
107
+
108
+ const int tmp10 = tmp0 + tmp3, tmp13 = tmp0 - tmp3, tmp11 = tmp1 + tmp2, tmp12 = tmp1 - tmp2;
109
+
110
+ const int atmp0 = ACCESS_COL(7), atmp1 = ACCESS_COL(5), atmp2 = ACCESS_COL(3), atmp3 = ACCESS_COL(1);
111
+
112
+ const int bz1 = atmp0 + atmp3, bz2 = atmp1 + atmp2, bz3 = atmp0 + atmp2, bz4 = atmp1 + atmp3;
113
+ const int bz5 = MULTIPLY(bz3 + bz4, FIX_1_175875602);
114
+
115
+ const int az1 = MULTIPLY(bz1, - FIX_0_899976223);
116
+ const int az2 = MULTIPLY(bz2, - FIX_2_562915447);
117
+ const int az3 = MULTIPLY(bz3, - FIX_1_961570560) + bz5;
118
+ const int az4 = MULTIPLY(bz4, - FIX_0_390180644) + bz5;
119
+
120
+ const int btmp0 = MULTIPLY(atmp0, FIX_0_298631336) + az1 + az3;
121
+ const int btmp1 = MULTIPLY(atmp1, FIX_2_053119869) + az2 + az4;
122
+ const int btmp2 = MULTIPLY(atmp2, FIX_3_072711026) + az2 + az3;
123
+ const int btmp3 = MULTIPLY(atmp3, FIX_1_501321110) + az1 + az4;
124
+
125
+ pTemp[0] = DESCALE(tmp10 + btmp3, CONST_BITS-PASS1_BITS);
126
+ pTemp[7] = DESCALE(tmp10 - btmp3, CONST_BITS-PASS1_BITS);
127
+ pTemp[1] = DESCALE(tmp11 + btmp2, CONST_BITS-PASS1_BITS);
128
+ pTemp[6] = DESCALE(tmp11 - btmp2, CONST_BITS-PASS1_BITS);
129
+ pTemp[2] = DESCALE(tmp12 + btmp1, CONST_BITS-PASS1_BITS);
130
+ pTemp[5] = DESCALE(tmp12 - btmp1, CONST_BITS-PASS1_BITS);
131
+ pTemp[3] = DESCALE(tmp13 + btmp0, CONST_BITS-PASS1_BITS);
132
+ pTemp[4] = DESCALE(tmp13 - btmp0, CONST_BITS-PASS1_BITS);
133
+ }
134
+ };
135
+
136
+ template <>
137
+ struct Row<0>
138
+ {
139
+ static void idct(int* pTemp, const jpgd_block_t* pSrc)
140
+ {
141
+ #ifdef _MSC_VER
142
+ pTemp; pSrc;
143
+ #endif
144
+ }
145
+ };
146
+
147
+ template <>
148
+ struct Row<1>
149
+ {
150
+ static void idct(int* pTemp, const jpgd_block_t* pSrc)
151
+ {
152
+ const int dcval = (pSrc[0] << PASS1_BITS);
153
+
154
+ pTemp[0] = dcval;
155
+ pTemp[1] = dcval;
156
+ pTemp[2] = dcval;
157
+ pTemp[3] = dcval;
158
+ pTemp[4] = dcval;
159
+ pTemp[5] = dcval;
160
+ pTemp[6] = dcval;
161
+ pTemp[7] = dcval;
162
+ }
163
+ };
164
+
165
+ // Compiler creates a fast path 1D IDCT for X non-zero rows
166
+ template <int NONZERO_ROWS>
167
+ struct Col
168
+ {
169
+ static void idct(uint8* pDst_ptr, const int* pTemp)
170
+ {
171
+ // ACCESS_ROW() will be optimized at compile time to either an array access, or 0.
172
+ #define ACCESS_ROW(x) (((x) < NONZERO_ROWS) ? pTemp[x * 8] : 0)
173
+
174
+ const int z2 = ACCESS_ROW(2);
175
+ const int z3 = ACCESS_ROW(6);
176
+
177
+ const int z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
178
+ const int tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
179
+ const int tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
180
+
181
+ const int tmp0 = (ACCESS_ROW(0) + ACCESS_ROW(4)) << CONST_BITS;
182
+ const int tmp1 = (ACCESS_ROW(0) - ACCESS_ROW(4)) << CONST_BITS;
183
+
184
+ const int tmp10 = tmp0 + tmp3, tmp13 = tmp0 - tmp3, tmp11 = tmp1 + tmp2, tmp12 = tmp1 - tmp2;
185
+
186
+ const int atmp0 = ACCESS_ROW(7), atmp1 = ACCESS_ROW(5), atmp2 = ACCESS_ROW(3), atmp3 = ACCESS_ROW(1);
187
+
188
+ const int bz1 = atmp0 + atmp3, bz2 = atmp1 + atmp2, bz3 = atmp0 + atmp2, bz4 = atmp1 + atmp3;
189
+ const int bz5 = MULTIPLY(bz3 + bz4, FIX_1_175875602);
190
+
191
+ const int az1 = MULTIPLY(bz1, - FIX_0_899976223);
192
+ const int az2 = MULTIPLY(bz2, - FIX_2_562915447);
193
+ const int az3 = MULTIPLY(bz3, - FIX_1_961570560) + bz5;
194
+ const int az4 = MULTIPLY(bz4, - FIX_0_390180644) + bz5;
195
+
196
+ const int btmp0 = MULTIPLY(atmp0, FIX_0_298631336) + az1 + az3;
197
+ const int btmp1 = MULTIPLY(atmp1, FIX_2_053119869) + az2 + az4;
198
+ const int btmp2 = MULTIPLY(atmp2, FIX_3_072711026) + az2 + az3;
199
+ const int btmp3 = MULTIPLY(atmp3, FIX_1_501321110) + az1 + az4;
200
+
201
+ int i = DESCALE_ZEROSHIFT(tmp10 + btmp3, CONST_BITS+PASS1_BITS+3);
202
+ pDst_ptr[8*0] = (uint8)CLAMP(i);
203
+
204
+ i = DESCALE_ZEROSHIFT(tmp10 - btmp3, CONST_BITS+PASS1_BITS+3);
205
+ pDst_ptr[8*7] = (uint8)CLAMP(i);
206
+
207
+ i = DESCALE_ZEROSHIFT(tmp11 + btmp2, CONST_BITS+PASS1_BITS+3);
208
+ pDst_ptr[8*1] = (uint8)CLAMP(i);
209
+
210
+ i = DESCALE_ZEROSHIFT(tmp11 - btmp2, CONST_BITS+PASS1_BITS+3);
211
+ pDst_ptr[8*6] = (uint8)CLAMP(i);
212
+
213
+ i = DESCALE_ZEROSHIFT(tmp12 + btmp1, CONST_BITS+PASS1_BITS+3);
214
+ pDst_ptr[8*2] = (uint8)CLAMP(i);
215
+
216
+ i = DESCALE_ZEROSHIFT(tmp12 - btmp1, CONST_BITS+PASS1_BITS+3);
217
+ pDst_ptr[8*5] = (uint8)CLAMP(i);
218
+
219
+ i = DESCALE_ZEROSHIFT(tmp13 + btmp0, CONST_BITS+PASS1_BITS+3);
220
+ pDst_ptr[8*3] = (uint8)CLAMP(i);
221
+
222
+ i = DESCALE_ZEROSHIFT(tmp13 - btmp0, CONST_BITS+PASS1_BITS+3);
223
+ pDst_ptr[8*4] = (uint8)CLAMP(i);
224
+ }
225
+ };
226
+
227
+ template <>
228
+ struct Col<1>
229
+ {
230
+ static void idct(uint8* pDst_ptr, const int* pTemp)
231
+ {
232
+ int dcval = DESCALE_ZEROSHIFT(pTemp[0], PASS1_BITS+3);
233
+ const uint8 dcval_clamped = (uint8)CLAMP(dcval);
234
+ pDst_ptr[0*8] = dcval_clamped;
235
+ pDst_ptr[1*8] = dcval_clamped;
236
+ pDst_ptr[2*8] = dcval_clamped;
237
+ pDst_ptr[3*8] = dcval_clamped;
238
+ pDst_ptr[4*8] = dcval_clamped;
239
+ pDst_ptr[5*8] = dcval_clamped;
240
+ pDst_ptr[6*8] = dcval_clamped;
241
+ pDst_ptr[7*8] = dcval_clamped;
242
+ }
243
+ };
244
+
245
+ static const uint8 s_idct_row_table[] =
246
+ {
247
+ 1,0,0,0,0,0,0,0, 2,0,0,0,0,0,0,0, 2,1,0,0,0,0,0,0, 2,1,1,0,0,0,0,0, 2,2,1,0,0,0,0,0, 3,2,1,0,0,0,0,0, 4,2,1,0,0,0,0,0, 4,3,1,0,0,0,0,0,
248
+ 4,3,2,0,0,0,0,0, 4,3,2,1,0,0,0,0, 4,3,2,1,1,0,0,0, 4,3,2,2,1,0,0,0, 4,3,3,2,1,0,0,0, 4,4,3,2,1,0,0,0, 5,4,3,2,1,0,0,0, 6,4,3,2,1,0,0,0,
249
+ 6,5,3,2,1,0,0,0, 6,5,4,2,1,0,0,0, 6,5,4,3,1,0,0,0, 6,5,4,3,2,0,0,0, 6,5,4,3,2,1,0,0, 6,5,4,3,2,1,1,0, 6,5,4,3,2,2,1,0, 6,5,4,3,3,2,1,0,
250
+ 6,5,4,4,3,2,1,0, 6,5,5,4,3,2,1,0, 6,6,5,4,3,2,1,0, 7,6,5,4,3,2,1,0, 8,6,5,4,3,2,1,0, 8,7,5,4,3,2,1,0, 8,7,6,4,3,2,1,0, 8,7,6,5,3,2,1,0,
251
+ 8,7,6,5,4,2,1,0, 8,7,6,5,4,3,1,0, 8,7,6,5,4,3,2,0, 8,7,6,5,4,3,2,1, 8,7,6,5,4,3,2,2, 8,7,6,5,4,3,3,2, 8,7,6,5,4,4,3,2, 8,7,6,5,5,4,3,2,
252
+ 8,7,6,6,5,4,3,2, 8,7,7,6,5,4,3,2, 8,8,7,6,5,4,3,2, 8,8,8,6,5,4,3,2, 8,8,8,7,5,4,3,2, 8,8,8,7,6,4,3,2, 8,8,8,7,6,5,3,2, 8,8,8,7,6,5,4,2,
253
+ 8,8,8,7,6,5,4,3, 8,8,8,7,6,5,4,4, 8,8,8,7,6,5,5,4, 8,8,8,7,6,6,5,4, 8,8,8,7,7,6,5,4, 8,8,8,8,7,6,5,4, 8,8,8,8,8,6,5,4, 8,8,8,8,8,7,5,4,
254
+ 8,8,8,8,8,7,6,4, 8,8,8,8,8,7,6,5, 8,8,8,8,8,7,6,6, 8,8,8,8,8,7,7,6, 8,8,8,8,8,8,7,6, 8,8,8,8,8,8,8,6, 8,8,8,8,8,8,8,7, 8,8,8,8,8,8,8,8,
255
+ };
256
+
257
+ static const uint8 s_idct_col_table[] = { 1, 1, 2, 3, 3, 3, 3, 3, 3, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8 };
258
+
259
+ void idct(const jpgd_block_t* pSrc_ptr, uint8* pDst_ptr, int block_max_zag)
260
+ {
261
+ JPGD_ASSERT(block_max_zag >= 1);
262
+ JPGD_ASSERT(block_max_zag <= 64);
263
+
264
+ if (block_max_zag == 1)
265
+ {
266
+ int k = ((pSrc_ptr[0] + 4) >> 3) + 128;
267
+ k = CLAMP(k);
268
+ k = k | (k<<8);
269
+ k = k | (k<<16);
270
+
271
+ for (int i = 8; i > 0; i--)
272
+ {
273
+ *(int*)&pDst_ptr[0] = k;
274
+ *(int*)&pDst_ptr[4] = k;
275
+ pDst_ptr += 8;
276
+ }
277
+ return;
278
+ }
279
+
280
+ int temp[64];
281
+
282
+ const jpgd_block_t* pSrc = pSrc_ptr;
283
+ int* pTemp = temp;
284
+
285
+ const uint8* pRow_tab = &s_idct_row_table[(block_max_zag - 1) * 8];
286
+ int i;
287
+ for (i = 8; i > 0; i--, pRow_tab++)
288
+ {
289
+ switch (*pRow_tab)
290
+ {
291
+ case 0: Row<0>::idct(pTemp, pSrc); break;
292
+ case 1: Row<1>::idct(pTemp, pSrc); break;
293
+ case 2: Row<2>::idct(pTemp, pSrc); break;
294
+ case 3: Row<3>::idct(pTemp, pSrc); break;
295
+ case 4: Row<4>::idct(pTemp, pSrc); break;
296
+ case 5: Row<5>::idct(pTemp, pSrc); break;
297
+ case 6: Row<6>::idct(pTemp, pSrc); break;
298
+ case 7: Row<7>::idct(pTemp, pSrc); break;
299
+ case 8: Row<8>::idct(pTemp, pSrc); break;
300
+ }
301
+
302
+ pSrc += 8;
303
+ pTemp += 8;
304
+ }
305
+
306
+ pTemp = temp;
307
+
308
+ const int nonzero_rows = s_idct_col_table[block_max_zag - 1];
309
+ for (i = 8; i > 0; i--)
310
+ {
311
+ switch (nonzero_rows)
312
+ {
313
+ case 1: Col<1>::idct(pDst_ptr, pTemp); break;
314
+ case 2: Col<2>::idct(pDst_ptr, pTemp); break;
315
+ case 3: Col<3>::idct(pDst_ptr, pTemp); break;
316
+ case 4: Col<4>::idct(pDst_ptr, pTemp); break;
317
+ case 5: Col<5>::idct(pDst_ptr, pTemp); break;
318
+ case 6: Col<6>::idct(pDst_ptr, pTemp); break;
319
+ case 7: Col<7>::idct(pDst_ptr, pTemp); break;
320
+ case 8: Col<8>::idct(pDst_ptr, pTemp); break;
321
+ }
322
+
323
+ pTemp++;
324
+ pDst_ptr++;
325
+ }
326
+ }
327
+
328
+ void idct_4x4(const jpgd_block_t* pSrc_ptr, uint8* pDst_ptr)
329
+ {
330
+ int temp[64];
331
+ int* pTemp = temp;
332
+ const jpgd_block_t* pSrc = pSrc_ptr;
333
+
334
+ for (int i = 4; i > 0; i--)
335
+ {
336
+ Row<4>::idct(pTemp, pSrc);
337
+ pSrc += 8;
338
+ pTemp += 8;
339
+ }
340
+
341
+ pTemp = temp;
342
+ for (int i = 8; i > 0; i--)
343
+ {
344
+ Col<4>::idct(pDst_ptr, pTemp);
345
+ pTemp++;
346
+ pDst_ptr++;
347
+ }
348
+ }
349
+
350
+ // Retrieve one character from the input stream.
351
+ inline uint jpeg_decoder::get_char()
352
+ {
353
+ // Any bytes remaining in buffer?
354
+ if (!m_in_buf_left)
355
+ {
356
+ // Try to get more bytes.
357
+ prep_in_buffer();
358
+ // Still nothing to get?
359
+ if (!m_in_buf_left)
360
+ {
361
+ // Pad the end of the stream with 0xFF 0xD9 (EOI marker)
362
+ int t = m_tem_flag;
363
+ m_tem_flag ^= 1;
364
+ if (t)
365
+ return 0xD9;
366
+ else
367
+ return 0xFF;
368
+ }
369
+ }
370
+
371
+ uint c = *m_pIn_buf_ofs++;
372
+ m_in_buf_left--;
373
+
374
+ return c;
375
+ }
376
+
377
+ // Same as previous method, except can indicate if the character is a pad character or not.
378
+ inline uint jpeg_decoder::get_char(bool *pPadding_flag)
379
+ {
380
+ if (!m_in_buf_left)
381
+ {
382
+ prep_in_buffer();
383
+ if (!m_in_buf_left)
384
+ {
385
+ *pPadding_flag = true;
386
+ int t = m_tem_flag;
387
+ m_tem_flag ^= 1;
388
+ if (t)
389
+ return 0xD9;
390
+ else
391
+ return 0xFF;
392
+ }
393
+ }
394
+
395
+ *pPadding_flag = false;
396
+
397
+ uint c = *m_pIn_buf_ofs++;
398
+ m_in_buf_left--;
399
+
400
+ return c;
401
+ }
402
+
403
+ // Inserts a previously retrieved character back into the input buffer.
404
+ inline void jpeg_decoder::stuff_char(uint8 q)
405
+ {
406
+ *(--m_pIn_buf_ofs) = q;
407
+ m_in_buf_left++;
408
+ }
409
+
410
+ // Retrieves one character from the input stream, but does not read past markers. Will continue to return 0xFF when a marker is encountered.
411
+ inline uint8 jpeg_decoder::get_octet()
412
+ {
413
+ bool padding_flag;
414
+ int c = get_char(&padding_flag);
415
+
416
+ if (c == 0xFF)
417
+ {
418
+ if (padding_flag)
419
+ return 0xFF;
420
+
421
+ c = get_char(&padding_flag);
422
+ if (padding_flag)
423
+ {
424
+ stuff_char(0xFF);
425
+ return 0xFF;
426
+ }
427
+
428
+ if (c == 0x00)
429
+ return 0xFF;
430
+ else
431
+ {
432
+ stuff_char(static_cast<uint8>(c));
433
+ stuff_char(0xFF);
434
+ return 0xFF;
435
+ }
436
+ }
437
+
438
+ return static_cast<uint8>(c);
439
+ }
440
+
441
+ // Retrieves a variable number of bits from the input stream. Does not recognize markers.
442
+ inline uint jpeg_decoder::get_bits(int num_bits)
443
+ {
444
+ if (!num_bits)
445
+ return 0;
446
+
447
+ uint i = m_bit_buf >> (32 - num_bits);
448
+
449
+ if ((m_bits_left -= num_bits) <= 0)
450
+ {
451
+ m_bit_buf <<= (num_bits += m_bits_left);
452
+
453
+ uint c1 = get_char();
454
+ uint c2 = get_char();
455
+ m_bit_buf = (m_bit_buf & 0xFFFF0000) | (c1 << 8) | c2;
456
+
457
+ m_bit_buf <<= -m_bits_left;
458
+
459
+ m_bits_left += 16;
460
+
461
+ JPGD_ASSERT(m_bits_left >= 0);
462
+ }
463
+ else
464
+ m_bit_buf <<= num_bits;
465
+
466
+ return i;
467
+ }
468
+
469
+ // Retrieves a variable number of bits from the input stream. Markers will not be read into the input bit buffer. Instead, an infinite number of all 1's will be returned when a marker is encountered.
470
+ inline uint jpeg_decoder::get_bits_no_markers(int num_bits)
471
+ {
472
+ if (!num_bits)
473
+ return 0;
474
+
475
+ uint i = m_bit_buf >> (32 - num_bits);
476
+
477
+ if ((m_bits_left -= num_bits) <= 0)
478
+ {
479
+ m_bit_buf <<= (num_bits += m_bits_left);
480
+
481
+ if ((m_in_buf_left < 2) || (m_pIn_buf_ofs[0] == 0xFF) || (m_pIn_buf_ofs[1] == 0xFF))
482
+ {
483
+ uint c1 = get_octet();
484
+ uint c2 = get_octet();
485
+ m_bit_buf |= (c1 << 8) | c2;
486
+ }
487
+ else
488
+ {
489
+ m_bit_buf |= ((uint)m_pIn_buf_ofs[0] << 8) | m_pIn_buf_ofs[1];
490
+ m_in_buf_left -= 2;
491
+ m_pIn_buf_ofs += 2;
492
+ }
493
+
494
+ m_bit_buf <<= -m_bits_left;
495
+
496
+ m_bits_left += 16;
497
+
498
+ JPGD_ASSERT(m_bits_left >= 0);
499
+ }
500
+ else
501
+ m_bit_buf <<= num_bits;
502
+
503
+ return i;
504
+ }
505
+
506
+ // Decodes a Huffman encoded symbol.
507
+ inline int jpeg_decoder::huff_decode(huff_tables *pH)
508
+ {
509
+ int symbol;
510
+
511
+ // Check first 8-bits: do we have a complete symbol?
512
+ if ((symbol = pH->look_up[m_bit_buf >> 24]) < 0)
513
+ {
514
+ // Decode more bits, use a tree traversal to find symbol.
515
+ int ofs = 23;
516
+ do
517
+ {
518
+ symbol = pH->tree[-(int)(symbol + ((m_bit_buf >> ofs) & 1))];
519
+ ofs--;
520
+ } while (symbol < 0);
521
+
522
+ get_bits_no_markers(8 + (23 - ofs));
523
+ }
524
+ else
525
+ get_bits_no_markers(pH->code_size[symbol]);
526
+
527
+ return symbol;
528
+ }
529
+
530
+ // Decodes a Huffman encoded symbol.
531
+ inline int jpeg_decoder::huff_decode(huff_tables *pH, int& extra_bits)
532
+ {
533
+ int symbol;
534
+
535
+ // Check first 8-bits: do we have a complete symbol?
536
+ if ((symbol = pH->look_up2[m_bit_buf >> 24]) < 0)
537
+ {
538
+ // Use a tree traversal to find symbol.
539
+ int ofs = 23;
540
+ do
541
+ {
542
+ symbol = pH->tree[-(int)(symbol + ((m_bit_buf >> ofs) & 1))];
543
+ ofs--;
544
+ } while (symbol < 0);
545
+
546
+ get_bits_no_markers(8 + (23 - ofs));
547
+
548
+ extra_bits = get_bits_no_markers(symbol & 0xF);
549
+ }
550
+ else
551
+ {
552
+ JPGD_ASSERT(((symbol >> 8) & 31) == pH->code_size[symbol & 255] + ((symbol & 0x8000) ? (symbol & 15) : 0));
553
+
554
+ if (symbol & 0x8000)
555
+ {
556
+ get_bits_no_markers((symbol >> 8) & 31);
557
+ extra_bits = symbol >> 16;
558
+ }
559
+ else
560
+ {
561
+ int code_size = (symbol >> 8) & 31;
562
+ int num_extra_bits = symbol & 0xF;
563
+ int bits = code_size + num_extra_bits;
564
+ if (bits <= (m_bits_left + 16))
565
+ extra_bits = get_bits_no_markers(bits) & ((1 << num_extra_bits) - 1);
566
+ else
567
+ {
568
+ get_bits_no_markers(code_size);
569
+ extra_bits = get_bits_no_markers(num_extra_bits);
570
+ }
571
+ }
572
+
573
+ symbol &= 0xFF;
574
+ }
575
+
576
+ return symbol;
577
+ }
578
+
579
+ // Tables and macro used to fully decode the DPCM differences.
580
+ static const int s_extend_test[16] = { 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 };
581
+ static const int s_extend_offset[16] = { 0, -1, -3, -7, -15, -31, -63, -127, -255, -511, -1023, -2047, -4095, -8191, -16383, -32767 };
582
+ static const int s_extend_mask[] = { 0, (1<<0), (1<<1), (1<<2), (1<<3), (1<<4), (1<<5), (1<<6), (1<<7), (1<<8), (1<<9), (1<<10), (1<<11), (1<<12), (1<<13), (1<<14), (1<<15), (1<<16) };
583
+ #define HUFF_EXTEND(x,s) ((x) < s_extend_test[s] ? (x) + s_extend_offset[s] : (x))
584
+
585
+ // Clamps a value between 0-255.
586
+ inline uint8 jpeg_decoder::clamp(int i)
587
+ {
588
+ if (static_cast<uint>(i) > 255)
589
+ i = (((~i) >> 31) & 0xFF);
590
+
591
+ return static_cast<uint8>(i);
592
+ }
593
+
594
+ namespace DCT_Upsample
595
+ {
596
+ struct Matrix44
597
+ {
598
+ typedef int Element_Type;
599
+ enum { NUM_ROWS = 4, NUM_COLS = 4 };
600
+
601
+ Element_Type v[NUM_ROWS][NUM_COLS];
602
+
603
+ inline int rows() const { return NUM_ROWS; }
604
+ inline int cols() const { return NUM_COLS; }
605
+
606
+ inline const Element_Type & at(int r, int c) const { return v[r][c]; }
607
+ inline Element_Type & at(int r, int c) { return v[r][c]; }
608
+
609
+ inline Matrix44() { }
610
+
611
+ inline Matrix44& operator += (const Matrix44& a)
612
+ {
613
+ for (int r = 0; r < NUM_ROWS; r++)
614
+ {
615
+ at(r, 0) += a.at(r, 0);
616
+ at(r, 1) += a.at(r, 1);
617
+ at(r, 2) += a.at(r, 2);
618
+ at(r, 3) += a.at(r, 3);
619
+ }
620
+ return *this;
621
+ }
622
+
623
+ inline Matrix44& operator -= (const Matrix44& a)
624
+ {
625
+ for (int r = 0; r < NUM_ROWS; r++)
626
+ {
627
+ at(r, 0) -= a.at(r, 0);
628
+ at(r, 1) -= a.at(r, 1);
629
+ at(r, 2) -= a.at(r, 2);
630
+ at(r, 3) -= a.at(r, 3);
631
+ }
632
+ return *this;
633
+ }
634
+
635
+ friend inline Matrix44 operator + (const Matrix44& a, const Matrix44& b)
636
+ {
637
+ Matrix44 ret;
638
+ for (int r = 0; r < NUM_ROWS; r++)
639
+ {
640
+ ret.at(r, 0) = a.at(r, 0) + b.at(r, 0);
641
+ ret.at(r, 1) = a.at(r, 1) + b.at(r, 1);
642
+ ret.at(r, 2) = a.at(r, 2) + b.at(r, 2);
643
+ ret.at(r, 3) = a.at(r, 3) + b.at(r, 3);
644
+ }
645
+ return ret;
646
+ }
647
+
648
+ friend inline Matrix44 operator - (const Matrix44& a, const Matrix44& b)
649
+ {
650
+ Matrix44 ret;
651
+ for (int r = 0; r < NUM_ROWS; r++)
652
+ {
653
+ ret.at(r, 0) = a.at(r, 0) - b.at(r, 0);
654
+ ret.at(r, 1) = a.at(r, 1) - b.at(r, 1);
655
+ ret.at(r, 2) = a.at(r, 2) - b.at(r, 2);
656
+ ret.at(r, 3) = a.at(r, 3) - b.at(r, 3);
657
+ }
658
+ return ret;
659
+ }
660
+
661
+ static inline void add_and_store(jpgd_block_t* pDst, const Matrix44& a, const Matrix44& b)
662
+ {
663
+ for (int r = 0; r < 4; r++)
664
+ {
665
+ pDst[0*8 + r] = static_cast<jpgd_block_t>(a.at(r, 0) + b.at(r, 0));
666
+ pDst[1*8 + r] = static_cast<jpgd_block_t>(a.at(r, 1) + b.at(r, 1));
667
+ pDst[2*8 + r] = static_cast<jpgd_block_t>(a.at(r, 2) + b.at(r, 2));
668
+ pDst[3*8 + r] = static_cast<jpgd_block_t>(a.at(r, 3) + b.at(r, 3));
669
+ }
670
+ }
671
+
672
+ static inline void sub_and_store(jpgd_block_t* pDst, const Matrix44& a, const Matrix44& b)
673
+ {
674
+ for (int r = 0; r < 4; r++)
675
+ {
676
+ pDst[0*8 + r] = static_cast<jpgd_block_t>(a.at(r, 0) - b.at(r, 0));
677
+ pDst[1*8 + r] = static_cast<jpgd_block_t>(a.at(r, 1) - b.at(r, 1));
678
+ pDst[2*8 + r] = static_cast<jpgd_block_t>(a.at(r, 2) - b.at(r, 2));
679
+ pDst[3*8 + r] = static_cast<jpgd_block_t>(a.at(r, 3) - b.at(r, 3));
680
+ }
681
+ }
682
+ };
683
+
684
+ const int FRACT_BITS = 10;
685
+ const int SCALE = 1 << FRACT_BITS;
686
+
687
+ typedef int Temp_Type;
688
+ #define D(i) (((i) + (SCALE >> 1)) >> FRACT_BITS)
689
+ #define F(i) ((int)((i) * SCALE + .5f))
690
+
691
+ // Any decent C++ compiler will optimize this at compile time to a 0, or an array access.
692
+ #define AT(c, r) ((((c)>=NUM_COLS)||((r)>=NUM_ROWS)) ? 0 : pSrc[(c)+(r)*8])
693
+
694
+ // NUM_ROWS/NUM_COLS = # of non-zero rows/cols in input matrix
695
+ template<int NUM_ROWS, int NUM_COLS>
696
+ struct P_Q
697
+ {
698
+ static void calc(Matrix44& P, Matrix44& Q, const jpgd_block_t* pSrc)
699
+ {
700
+ // 4x8 = 4x8 times 8x8, matrix 0 is constant
701
+ const Temp_Type X000 = AT(0, 0);
702
+ const Temp_Type X001 = AT(0, 1);
703
+ const Temp_Type X002 = AT(0, 2);
704
+ const Temp_Type X003 = AT(0, 3);
705
+ const Temp_Type X004 = AT(0, 4);
706
+ const Temp_Type X005 = AT(0, 5);
707
+ const Temp_Type X006 = AT(0, 6);
708
+ const Temp_Type X007 = AT(0, 7);
709
+ const Temp_Type X010 = D(F(0.415735f) * AT(1, 0) + F(0.791065f) * AT(3, 0) + F(-0.352443f) * AT(5, 0) + F(0.277785f) * AT(7, 0));
710
+ const Temp_Type X011 = D(F(0.415735f) * AT(1, 1) + F(0.791065f) * AT(3, 1) + F(-0.352443f) * AT(5, 1) + F(0.277785f) * AT(7, 1));
711
+ const Temp_Type X012 = D(F(0.415735f) * AT(1, 2) + F(0.791065f) * AT(3, 2) + F(-0.352443f) * AT(5, 2) + F(0.277785f) * AT(7, 2));
712
+ const Temp_Type X013 = D(F(0.415735f) * AT(1, 3) + F(0.791065f) * AT(3, 3) + F(-0.352443f) * AT(5, 3) + F(0.277785f) * AT(7, 3));
713
+ const Temp_Type X014 = D(F(0.415735f) * AT(1, 4) + F(0.791065f) * AT(3, 4) + F(-0.352443f) * AT(5, 4) + F(0.277785f) * AT(7, 4));
714
+ const Temp_Type X015 = D(F(0.415735f) * AT(1, 5) + F(0.791065f) * AT(3, 5) + F(-0.352443f) * AT(5, 5) + F(0.277785f) * AT(7, 5));
715
+ const Temp_Type X016 = D(F(0.415735f) * AT(1, 6) + F(0.791065f) * AT(3, 6) + F(-0.352443f) * AT(5, 6) + F(0.277785f) * AT(7, 6));
716
+ const Temp_Type X017 = D(F(0.415735f) * AT(1, 7) + F(0.791065f) * AT(3, 7) + F(-0.352443f) * AT(5, 7) + F(0.277785f) * AT(7, 7));
717
+ const Temp_Type X020 = AT(4, 0);
718
+ const Temp_Type X021 = AT(4, 1);
719
+ const Temp_Type X022 = AT(4, 2);
720
+ const Temp_Type X023 = AT(4, 3);
721
+ const Temp_Type X024 = AT(4, 4);
722
+ const Temp_Type X025 = AT(4, 5);
723
+ const Temp_Type X026 = AT(4, 6);
724
+ const Temp_Type X027 = AT(4, 7);
725
+ const Temp_Type X030 = D(F(0.022887f) * AT(1, 0) + F(-0.097545f) * AT(3, 0) + F(0.490393f) * AT(5, 0) + F(0.865723f) * AT(7, 0));
726
+ const Temp_Type X031 = D(F(0.022887f) * AT(1, 1) + F(-0.097545f) * AT(3, 1) + F(0.490393f) * AT(5, 1) + F(0.865723f) * AT(7, 1));
727
+ const Temp_Type X032 = D(F(0.022887f) * AT(1, 2) + F(-0.097545f) * AT(3, 2) + F(0.490393f) * AT(5, 2) + F(0.865723f) * AT(7, 2));
728
+ const Temp_Type X033 = D(F(0.022887f) * AT(1, 3) + F(-0.097545f) * AT(3, 3) + F(0.490393f) * AT(5, 3) + F(0.865723f) * AT(7, 3));
729
+ const Temp_Type X034 = D(F(0.022887f) * AT(1, 4) + F(-0.097545f) * AT(3, 4) + F(0.490393f) * AT(5, 4) + F(0.865723f) * AT(7, 4));
730
+ const Temp_Type X035 = D(F(0.022887f) * AT(1, 5) + F(-0.097545f) * AT(3, 5) + F(0.490393f) * AT(5, 5) + F(0.865723f) * AT(7, 5));
731
+ const Temp_Type X036 = D(F(0.022887f) * AT(1, 6) + F(-0.097545f) * AT(3, 6) + F(0.490393f) * AT(5, 6) + F(0.865723f) * AT(7, 6));
732
+ const Temp_Type X037 = D(F(0.022887f) * AT(1, 7) + F(-0.097545f) * AT(3, 7) + F(0.490393f) * AT(5, 7) + F(0.865723f) * AT(7, 7));
733
+
734
+ // 4x4 = 4x8 times 8x4, matrix 1 is constant
735
+ P.at(0, 0) = X000;
736
+ P.at(0, 1) = D(X001 * F(0.415735f) + X003 * F(0.791065f) + X005 * F(-0.352443f) + X007 * F(0.277785f));
737
+ P.at(0, 2) = X004;
738
+ P.at(0, 3) = D(X001 * F(0.022887f) + X003 * F(-0.097545f) + X005 * F(0.490393f) + X007 * F(0.865723f));
739
+ P.at(1, 0) = X010;
740
+ P.at(1, 1) = D(X011 * F(0.415735f) + X013 * F(0.791065f) + X015 * F(-0.352443f) + X017 * F(0.277785f));
741
+ P.at(1, 2) = X014;
742
+ P.at(1, 3) = D(X011 * F(0.022887f) + X013 * F(-0.097545f) + X015 * F(0.490393f) + X017 * F(0.865723f));
743
+ P.at(2, 0) = X020;
744
+ P.at(2, 1) = D(X021 * F(0.415735f) + X023 * F(0.791065f) + X025 * F(-0.352443f) + X027 * F(0.277785f));
745
+ P.at(2, 2) = X024;
746
+ P.at(2, 3) = D(X021 * F(0.022887f) + X023 * F(-0.097545f) + X025 * F(0.490393f) + X027 * F(0.865723f));
747
+ P.at(3, 0) = X030;
748
+ P.at(3, 1) = D(X031 * F(0.415735f) + X033 * F(0.791065f) + X035 * F(-0.352443f) + X037 * F(0.277785f));
749
+ P.at(3, 2) = X034;
750
+ P.at(3, 3) = D(X031 * F(0.022887f) + X033 * F(-0.097545f) + X035 * F(0.490393f) + X037 * F(0.865723f));
751
+ // 40 muls 24 adds
752
+
753
+ // 4x4 = 4x8 times 8x4, matrix 1 is constant
754
+ Q.at(0, 0) = D(X001 * F(0.906127f) + X003 * F(-0.318190f) + X005 * F(0.212608f) + X007 * F(-0.180240f));
755
+ Q.at(0, 1) = X002;
756
+ Q.at(0, 2) = D(X001 * F(-0.074658f) + X003 * F(0.513280f) + X005 * F(0.768178f) + X007 * F(-0.375330f));
757
+ Q.at(0, 3) = X006;
758
+ Q.at(1, 0) = D(X011 * F(0.906127f) + X013 * F(-0.318190f) + X015 * F(0.212608f) + X017 * F(-0.180240f));
759
+ Q.at(1, 1) = X012;
760
+ Q.at(1, 2) = D(X011 * F(-0.074658f) + X013 * F(0.513280f) + X015 * F(0.768178f) + X017 * F(-0.375330f));
761
+ Q.at(1, 3) = X016;
762
+ Q.at(2, 0) = D(X021 * F(0.906127f) + X023 * F(-0.318190f) + X025 * F(0.212608f) + X027 * F(-0.180240f));
763
+ Q.at(2, 1) = X022;
764
+ Q.at(2, 2) = D(X021 * F(-0.074658f) + X023 * F(0.513280f) + X025 * F(0.768178f) + X027 * F(-0.375330f));
765
+ Q.at(2, 3) = X026;
766
+ Q.at(3, 0) = D(X031 * F(0.906127f) + X033 * F(-0.318190f) + X035 * F(0.212608f) + X037 * F(-0.180240f));
767
+ Q.at(3, 1) = X032;
768
+ Q.at(3, 2) = D(X031 * F(-0.074658f) + X033 * F(0.513280f) + X035 * F(0.768178f) + X037 * F(-0.375330f));
769
+ Q.at(3, 3) = X036;
770
+ // 40 muls 24 adds
771
+ }
772
+ };
773
+
774
+ template<int NUM_ROWS, int NUM_COLS>
775
+ struct R_S
776
+ {
777
+ static void calc(Matrix44& R, Matrix44& S, const jpgd_block_t* pSrc)
778
+ {
779
+ // 4x8 = 4x8 times 8x8, matrix 0 is constant
780
+ const Temp_Type X100 = D(F(0.906127f) * AT(1, 0) + F(-0.318190f) * AT(3, 0) + F(0.212608f) * AT(5, 0) + F(-0.180240f) * AT(7, 0));
781
+ const Temp_Type X101 = D(F(0.906127f) * AT(1, 1) + F(-0.318190f) * AT(3, 1) + F(0.212608f) * AT(5, 1) + F(-0.180240f) * AT(7, 1));
782
+ const Temp_Type X102 = D(F(0.906127f) * AT(1, 2) + F(-0.318190f) * AT(3, 2) + F(0.212608f) * AT(5, 2) + F(-0.180240f) * AT(7, 2));
783
+ const Temp_Type X103 = D(F(0.906127f) * AT(1, 3) + F(-0.318190f) * AT(3, 3) + F(0.212608f) * AT(5, 3) + F(-0.180240f) * AT(7, 3));
784
+ const Temp_Type X104 = D(F(0.906127f) * AT(1, 4) + F(-0.318190f) * AT(3, 4) + F(0.212608f) * AT(5, 4) + F(-0.180240f) * AT(7, 4));
785
+ const Temp_Type X105 = D(F(0.906127f) * AT(1, 5) + F(-0.318190f) * AT(3, 5) + F(0.212608f) * AT(5, 5) + F(-0.180240f) * AT(7, 5));
786
+ const Temp_Type X106 = D(F(0.906127f) * AT(1, 6) + F(-0.318190f) * AT(3, 6) + F(0.212608f) * AT(5, 6) + F(-0.180240f) * AT(7, 6));
787
+ const Temp_Type X107 = D(F(0.906127f) * AT(1, 7) + F(-0.318190f) * AT(3, 7) + F(0.212608f) * AT(5, 7) + F(-0.180240f) * AT(7, 7));
788
+ const Temp_Type X110 = AT(2, 0);
789
+ const Temp_Type X111 = AT(2, 1);
790
+ const Temp_Type X112 = AT(2, 2);
791
+ const Temp_Type X113 = AT(2, 3);
792
+ const Temp_Type X114 = AT(2, 4);
793
+ const Temp_Type X115 = AT(2, 5);
794
+ const Temp_Type X116 = AT(2, 6);
795
+ const Temp_Type X117 = AT(2, 7);
796
+ const Temp_Type X120 = D(F(-0.074658f) * AT(1, 0) + F(0.513280f) * AT(3, 0) + F(0.768178f) * AT(5, 0) + F(-0.375330f) * AT(7, 0));
797
+ const Temp_Type X121 = D(F(-0.074658f) * AT(1, 1) + F(0.513280f) * AT(3, 1) + F(0.768178f) * AT(5, 1) + F(-0.375330f) * AT(7, 1));
798
+ const Temp_Type X122 = D(F(-0.074658f) * AT(1, 2) + F(0.513280f) * AT(3, 2) + F(0.768178f) * AT(5, 2) + F(-0.375330f) * AT(7, 2));
799
+ const Temp_Type X123 = D(F(-0.074658f) * AT(1, 3) + F(0.513280f) * AT(3, 3) + F(0.768178f) * AT(5, 3) + F(-0.375330f) * AT(7, 3));
800
+ const Temp_Type X124 = D(F(-0.074658f) * AT(1, 4) + F(0.513280f) * AT(3, 4) + F(0.768178f) * AT(5, 4) + F(-0.375330f) * AT(7, 4));
801
+ const Temp_Type X125 = D(F(-0.074658f) * AT(1, 5) + F(0.513280f) * AT(3, 5) + F(0.768178f) * AT(5, 5) + F(-0.375330f) * AT(7, 5));
802
+ const Temp_Type X126 = D(F(-0.074658f) * AT(1, 6) + F(0.513280f) * AT(3, 6) + F(0.768178f) * AT(5, 6) + F(-0.375330f) * AT(7, 6));
803
+ const Temp_Type X127 = D(F(-0.074658f) * AT(1, 7) + F(0.513280f) * AT(3, 7) + F(0.768178f) * AT(5, 7) + F(-0.375330f) * AT(7, 7));
804
+ const Temp_Type X130 = AT(6, 0);
805
+ const Temp_Type X131 = AT(6, 1);
806
+ const Temp_Type X132 = AT(6, 2);
807
+ const Temp_Type X133 = AT(6, 3);
808
+ const Temp_Type X134 = AT(6, 4);
809
+ const Temp_Type X135 = AT(6, 5);
810
+ const Temp_Type X136 = AT(6, 6);
811
+ const Temp_Type X137 = AT(6, 7);
812
+ // 80 muls 48 adds
813
+
814
+ // 4x4 = 4x8 times 8x4, matrix 1 is constant
815
+ R.at(0, 0) = X100;
816
+ R.at(0, 1) = D(X101 * F(0.415735f) + X103 * F(0.791065f) + X105 * F(-0.352443f) + X107 * F(0.277785f));
817
+ R.at(0, 2) = X104;
818
+ R.at(0, 3) = D(X101 * F(0.022887f) + X103 * F(-0.097545f) + X105 * F(0.490393f) + X107 * F(0.865723f));
819
+ R.at(1, 0) = X110;
820
+ R.at(1, 1) = D(X111 * F(0.415735f) + X113 * F(0.791065f) + X115 * F(-0.352443f) + X117 * F(0.277785f));
821
+ R.at(1, 2) = X114;
822
+ R.at(1, 3) = D(X111 * F(0.022887f) + X113 * F(-0.097545f) + X115 * F(0.490393f) + X117 * F(0.865723f));
823
+ R.at(2, 0) = X120;
824
+ R.at(2, 1) = D(X121 * F(0.415735f) + X123 * F(0.791065f) + X125 * F(-0.352443f) + X127 * F(0.277785f));
825
+ R.at(2, 2) = X124;
826
+ R.at(2, 3) = D(X121 * F(0.022887f) + X123 * F(-0.097545f) + X125 * F(0.490393f) + X127 * F(0.865723f));
827
+ R.at(3, 0) = X130;
828
+ R.at(3, 1) = D(X131 * F(0.415735f) + X133 * F(0.791065f) + X135 * F(-0.352443f) + X137 * F(0.277785f));
829
+ R.at(3, 2) = X134;
830
+ R.at(3, 3) = D(X131 * F(0.022887f) + X133 * F(-0.097545f) + X135 * F(0.490393f) + X137 * F(0.865723f));
831
+ // 40 muls 24 adds
832
+ // 4x4 = 4x8 times 8x4, matrix 1 is constant
833
+ S.at(0, 0) = D(X101 * F(0.906127f) + X103 * F(-0.318190f) + X105 * F(0.212608f) + X107 * F(-0.180240f));
834
+ S.at(0, 1) = X102;
835
+ S.at(0, 2) = D(X101 * F(-0.074658f) + X103 * F(0.513280f) + X105 * F(0.768178f) + X107 * F(-0.375330f));
836
+ S.at(0, 3) = X106;
837
+ S.at(1, 0) = D(X111 * F(0.906127f) + X113 * F(-0.318190f) + X115 * F(0.212608f) + X117 * F(-0.180240f));
838
+ S.at(1, 1) = X112;
839
+ S.at(1, 2) = D(X111 * F(-0.074658f) + X113 * F(0.513280f) + X115 * F(0.768178f) + X117 * F(-0.375330f));
840
+ S.at(1, 3) = X116;
841
+ S.at(2, 0) = D(X121 * F(0.906127f) + X123 * F(-0.318190f) + X125 * F(0.212608f) + X127 * F(-0.180240f));
842
+ S.at(2, 1) = X122;
843
+ S.at(2, 2) = D(X121 * F(-0.074658f) + X123 * F(0.513280f) + X125 * F(0.768178f) + X127 * F(-0.375330f));
844
+ S.at(2, 3) = X126;
845
+ S.at(3, 0) = D(X131 * F(0.906127f) + X133 * F(-0.318190f) + X135 * F(0.212608f) + X137 * F(-0.180240f));
846
+ S.at(3, 1) = X132;
847
+ S.at(3, 2) = D(X131 * F(-0.074658f) + X133 * F(0.513280f) + X135 * F(0.768178f) + X137 * F(-0.375330f));
848
+ S.at(3, 3) = X136;
849
+ // 40 muls 24 adds
850
+ }
851
+ };
852
+ } // end namespace DCT_Upsample
853
+
854
+ // Unconditionally frees all allocated m_blocks.
855
+ void jpeg_decoder::free_all_blocks()
856
+ {
857
+ m_pStream = NULL;
858
+ for (mem_block *b = m_pMem_blocks; b; )
859
+ {
860
+ mem_block *n = b->m_pNext;
861
+ jpgd_free(b);
862
+ b = n;
863
+ }
864
+ m_pMem_blocks = NULL;
865
+ }
866
+
867
+ // This method handles all errors.
868
+ // It could easily be changed to use C++ exceptions.
869
+ void jpeg_decoder::stop_decoding(jpgd_status status)
870
+ {
871
+ m_error_code = status;
872
+ free_all_blocks();
873
+ longjmp(m_jmp_state, status);
874
+
875
+ // we shouldn't get here as longjmp shouldn't return, but we put it here to make it explicit
876
+ // that this function doesn't return, otherwise we get this error:
877
+ //
878
+ // error : function declared 'noreturn' should not return
879
+ exit(1);
880
+ }
881
+
882
+ void *jpeg_decoder::alloc(size_t nSize, bool zero)
883
+ {
884
+ nSize = (JPGD_MAX(nSize, 1) + 3) & ~3;
885
+ char *rv = NULL;
886
+ for (mem_block *b = m_pMem_blocks; b; b = b->m_pNext)
887
+ {
888
+ if ((b->m_used_count + nSize) <= b->m_size)
889
+ {
890
+ rv = b->m_data + b->m_used_count;
891
+ b->m_used_count += nSize;
892
+ break;
893
+ }
894
+ }
895
+ if (!rv)
896
+ {
897
+ int capacity = JPGD_MAX(32768 - 256, (nSize + 2047) & ~2047);
898
+ mem_block *b = (mem_block*)jpgd_malloc(sizeof(mem_block) + capacity);
899
+ if (!b) stop_decoding(JPGD_NOTENOUGHMEM);
900
+ b->m_pNext = m_pMem_blocks; m_pMem_blocks = b;
901
+ b->m_used_count = nSize;
902
+ b->m_size = capacity;
903
+ rv = b->m_data;
904
+ }
905
+ if (zero) memset(rv, 0, nSize);
906
+ return rv;
907
+ }
908
+
909
+ void jpeg_decoder::word_clear(void *p, uint16 c, uint n)
910
+ {
911
+ uint8 *pD = (uint8*)p;
912
+ const uint8 l = c & 0xFF, h = (c >> 8) & 0xFF;
913
+ while (n)
914
+ {
915
+ pD[0] = l; pD[1] = h; pD += 2;
916
+ n--;
917
+ }
918
+ }
919
+
920
+ // Refill the input buffer.
921
+ // This method will sit in a loop until (A) the buffer is full or (B)
922
+ // the stream's read() method reports and end of file condition.
923
+ void jpeg_decoder::prep_in_buffer()
924
+ {
925
+ m_in_buf_left = 0;
926
+ m_pIn_buf_ofs = m_in_buf;
927
+
928
+ if (m_eof_flag)
929
+ return;
930
+
931
+ do
932
+ {
933
+ int bytes_read = m_pStream->read(m_in_buf + m_in_buf_left, JPGD_IN_BUF_SIZE - m_in_buf_left, &m_eof_flag);
934
+ if (bytes_read == -1)
935
+ stop_decoding(JPGD_STREAM_READ);
936
+
937
+ m_in_buf_left += bytes_read;
938
+ } while ((m_in_buf_left < JPGD_IN_BUF_SIZE) && (!m_eof_flag));
939
+
940
+ m_total_bytes_read += m_in_buf_left;
941
+
942
+ // Pad the end of the block with M_EOI (prevents the decompressor from going off the rails if the stream is invalid).
943
+ // (This dates way back to when this decompressor was written in C/asm, and the all-asm Huffman decoder did some fancy things to increase perf.)
944
+ word_clear(m_pIn_buf_ofs + m_in_buf_left, 0xD9FF, 64);
945
+ }
946
+
947
+ // Read a Huffman code table.
948
+ void jpeg_decoder::read_dht_marker()
949
+ {
950
+ int i, index, count;
951
+ uint8 huff_num[17];
952
+ uint8 huff_val[256];
953
+
954
+ uint num_left = get_bits(16);
955
+
956
+ if (num_left < 2)
957
+ stop_decoding(JPGD_BAD_DHT_MARKER);
958
+
959
+ num_left -= 2;
960
+
961
+ while (num_left)
962
+ {
963
+ index = get_bits(8);
964
+
965
+ huff_num[0] = 0;
966
+
967
+ count = 0;
968
+
969
+ for (i = 1; i <= 16; i++)
970
+ {
971
+ huff_num[i] = static_cast<uint8>(get_bits(8));
972
+ count += huff_num[i];
973
+ }
974
+
975
+ if (count > 255)
976
+ stop_decoding(JPGD_BAD_DHT_COUNTS);
977
+
978
+ for (i = 0; i < count; i++)
979
+ huff_val[i] = static_cast<uint8>(get_bits(8));
980
+
981
+ i = 1 + 16 + count;
982
+
983
+ if (num_left < (uint)i)
984
+ stop_decoding(JPGD_BAD_DHT_MARKER);
985
+
986
+ num_left -= i;
987
+
988
+ if ((index & 0x10) > 0x10)
989
+ stop_decoding(JPGD_BAD_DHT_INDEX);
990
+
991
+ index = (index & 0x0F) + ((index & 0x10) >> 4) * (JPGD_MAX_HUFF_TABLES >> 1);
992
+
993
+ if (index >= JPGD_MAX_HUFF_TABLES)
994
+ stop_decoding(JPGD_BAD_DHT_INDEX);
995
+
996
+ if (!m_huff_num[index])
997
+ m_huff_num[index] = (uint8 *)alloc(17);
998
+
999
+ if (!m_huff_val[index])
1000
+ m_huff_val[index] = (uint8 *)alloc(256);
1001
+
1002
+ m_huff_ac[index] = (index & 0x10) != 0;
1003
+ memcpy(m_huff_num[index], huff_num, 17);
1004
+ memcpy(m_huff_val[index], huff_val, 256);
1005
+ }
1006
+ }
1007
+
1008
+ // Read a quantization table.
1009
+ void jpeg_decoder::read_dqt_marker()
1010
+ {
1011
+ int n, i, prec;
1012
+ uint num_left;
1013
+ uint temp;
1014
+
1015
+ num_left = get_bits(16);
1016
+
1017
+ if (num_left < 2)
1018
+ stop_decoding(JPGD_BAD_DQT_MARKER);
1019
+
1020
+ num_left -= 2;
1021
+
1022
+ while (num_left)
1023
+ {
1024
+ n = get_bits(8);
1025
+ prec = n >> 4;
1026
+ n &= 0x0F;
1027
+
1028
+ if (n >= JPGD_MAX_QUANT_TABLES)
1029
+ stop_decoding(JPGD_BAD_DQT_TABLE);
1030
+
1031
+ if (!m_quant[n])
1032
+ m_quant[n] = (jpgd_quant_t *)alloc(64 * sizeof(jpgd_quant_t));
1033
+
1034
+ // read quantization entries, in zag order
1035
+ for (i = 0; i < 64; i++)
1036
+ {
1037
+ temp = get_bits(8);
1038
+
1039
+ if (prec)
1040
+ temp = (temp << 8) + get_bits(8);
1041
+
1042
+ m_quant[n][i] = static_cast<jpgd_quant_t>(temp);
1043
+ }
1044
+
1045
+ i = 64 + 1;
1046
+
1047
+ if (prec)
1048
+ i += 64;
1049
+
1050
+ if (num_left < (uint)i)
1051
+ stop_decoding(JPGD_BAD_DQT_LENGTH);
1052
+
1053
+ num_left -= i;
1054
+ }
1055
+ }
1056
+
1057
+ // Read the start of frame (SOF) marker.
1058
+ void jpeg_decoder::read_sof_marker()
1059
+ {
1060
+ int i;
1061
+ uint num_left;
1062
+
1063
+ num_left = get_bits(16);
1064
+
1065
+ if (get_bits(8) != 8) /* precision: sorry, only 8-bit precision is supported right now */
1066
+ stop_decoding(JPGD_BAD_PRECISION);
1067
+
1068
+ m_image_y_size = get_bits(16);
1069
+
1070
+ if ((m_image_y_size < 1) || (m_image_y_size > JPGD_MAX_HEIGHT))
1071
+ stop_decoding(JPGD_BAD_HEIGHT);
1072
+
1073
+ m_image_x_size = get_bits(16);
1074
+
1075
+ if ((m_image_x_size < 1) || (m_image_x_size > JPGD_MAX_WIDTH))
1076
+ stop_decoding(JPGD_BAD_WIDTH);
1077
+
1078
+ m_comps_in_frame = get_bits(8);
1079
+
1080
+ if (m_comps_in_frame > JPGD_MAX_COMPONENTS)
1081
+ stop_decoding(JPGD_TOO_MANY_COMPONENTS);
1082
+
1083
+ if (num_left != (uint)(m_comps_in_frame * 3 + 8))
1084
+ stop_decoding(JPGD_BAD_SOF_LENGTH);
1085
+
1086
+ for (i = 0; i < m_comps_in_frame; i++)
1087
+ {
1088
+ m_comp_ident[i] = get_bits(8);
1089
+ m_comp_h_samp[i] = get_bits(4);
1090
+ m_comp_v_samp[i] = get_bits(4);
1091
+ m_comp_quant[i] = get_bits(8);
1092
+ }
1093
+ }
1094
+
1095
+ // Used to skip unrecognized markers.
1096
+ void jpeg_decoder::skip_variable_marker()
1097
+ {
1098
+ uint num_left;
1099
+
1100
+ num_left = get_bits(16);
1101
+
1102
+ if (num_left < 2)
1103
+ stop_decoding(JPGD_BAD_VARIABLE_MARKER);
1104
+
1105
+ num_left -= 2;
1106
+
1107
+ while (num_left)
1108
+ {
1109
+ get_bits(8);
1110
+ num_left--;
1111
+ }
1112
+ }
1113
+
1114
+ // Read a define restart interval (DRI) marker.
1115
+ void jpeg_decoder::read_dri_marker()
1116
+ {
1117
+ if (get_bits(16) != 4)
1118
+ stop_decoding(JPGD_BAD_DRI_LENGTH);
1119
+
1120
+ m_restart_interval = get_bits(16);
1121
+ }
1122
+
1123
+ // Read a start of scan (SOS) marker.
1124
+ void jpeg_decoder::read_sos_marker()
1125
+ {
1126
+ uint num_left;
1127
+ int i, ci, n, c, cc;
1128
+
1129
+ num_left = get_bits(16);
1130
+
1131
+ n = get_bits(8);
1132
+
1133
+ m_comps_in_scan = n;
1134
+
1135
+ num_left -= 3;
1136
+
1137
+ if ( (num_left != (uint)(n * 2 + 3)) || (n < 1) || (n > JPGD_MAX_COMPS_IN_SCAN) )
1138
+ stop_decoding(JPGD_BAD_SOS_LENGTH);
1139
+
1140
+ for (i = 0; i < n; i++)
1141
+ {
1142
+ cc = get_bits(8);
1143
+ c = get_bits(8);
1144
+ num_left -= 2;
1145
+
1146
+ for (ci = 0; ci < m_comps_in_frame; ci++)
1147
+ if (cc == m_comp_ident[ci])
1148
+ break;
1149
+
1150
+ if (ci >= m_comps_in_frame)
1151
+ stop_decoding(JPGD_BAD_SOS_COMP_ID);
1152
+
1153
+ m_comp_list[i] = ci;
1154
+ m_comp_dc_tab[ci] = (c >> 4) & 15;
1155
+ m_comp_ac_tab[ci] = (c & 15) + (JPGD_MAX_HUFF_TABLES >> 1);
1156
+ }
1157
+
1158
+ m_spectral_start = get_bits(8);
1159
+ m_spectral_end = get_bits(8);
1160
+ m_successive_high = get_bits(4);
1161
+ m_successive_low = get_bits(4);
1162
+
1163
+ if (!m_progressive_flag)
1164
+ {
1165
+ m_spectral_start = 0;
1166
+ m_spectral_end = 63;
1167
+ }
1168
+
1169
+ num_left -= 3;
1170
+
1171
+ while (num_left) /* read past whatever is num_left */
1172
+ {
1173
+ get_bits(8);
1174
+ num_left--;
1175
+ }
1176
+ }
1177
+
1178
+ // Finds the next marker.
1179
+ int jpeg_decoder::next_marker()
1180
+ {
1181
+ uint c, bytes;
1182
+
1183
+ bytes = 0;
1184
+
1185
+ do
1186
+ {
1187
+ do
1188
+ {
1189
+ bytes++;
1190
+ c = get_bits(8);
1191
+ } while (c != 0xFF);
1192
+
1193
+ do
1194
+ {
1195
+ c = get_bits(8);
1196
+ } while (c == 0xFF);
1197
+
1198
+ } while (c == 0);
1199
+
1200
+ // If bytes > 0 here, there where extra bytes before the marker (not good).
1201
+
1202
+ return c;
1203
+ }
1204
+
1205
+ // Process markers. Returns when an SOFx, SOI, EOI, or SOS marker is
1206
+ // encountered.
1207
+ int jpeg_decoder::process_markers()
1208
+ {
1209
+ int c;
1210
+
1211
+ for ( ; ; )
1212
+ {
1213
+ c = next_marker();
1214
+
1215
+ switch (c)
1216
+ {
1217
+ case M_SOF0:
1218
+ case M_SOF1:
1219
+ case M_SOF2:
1220
+ case M_SOF3:
1221
+ case M_SOF5:
1222
+ case M_SOF6:
1223
+ case M_SOF7:
1224
+ // case M_JPG:
1225
+ case M_SOF9:
1226
+ case M_SOF10:
1227
+ case M_SOF11:
1228
+ case M_SOF13:
1229
+ case M_SOF14:
1230
+ case M_SOF15:
1231
+ case M_SOI:
1232
+ case M_EOI:
1233
+ case M_SOS:
1234
+ {
1235
+ return c;
1236
+ }
1237
+ case M_DHT:
1238
+ {
1239
+ read_dht_marker();
1240
+ break;
1241
+ }
1242
+ // No arithmitic support - dumb patents!
1243
+ case M_DAC:
1244
+ {
1245
+ stop_decoding(JPGD_NO_ARITHMITIC_SUPPORT);
1246
+ break;
1247
+ }
1248
+ case M_DQT:
1249
+ {
1250
+ read_dqt_marker();
1251
+ break;
1252
+ }
1253
+ case M_DRI:
1254
+ {
1255
+ read_dri_marker();
1256
+ break;
1257
+ }
1258
+ //case M_APP0: /* no need to read the JFIF marker */
1259
+
1260
+ case M_JPG:
1261
+ case M_RST0: /* no parameters */
1262
+ case M_RST1:
1263
+ case M_RST2:
1264
+ case M_RST3:
1265
+ case M_RST4:
1266
+ case M_RST5:
1267
+ case M_RST6:
1268
+ case M_RST7:
1269
+ case M_TEM:
1270
+ {
1271
+ stop_decoding(JPGD_UNEXPECTED_MARKER);
1272
+ break;
1273
+ }
1274
+ default: /* must be DNL, DHP, EXP, APPn, JPGn, COM, or RESn or APP0 */
1275
+ {
1276
+ skip_variable_marker();
1277
+ break;
1278
+ }
1279
+ }
1280
+ }
1281
+ }
1282
+
1283
+ // Finds the start of image (SOI) marker.
1284
+ // This code is rather defensive: it only checks the first 512 bytes to avoid
1285
+ // false positives.
1286
+ void jpeg_decoder::locate_soi_marker()
1287
+ {
1288
+ uint lastchar, thischar;
1289
+ uint bytesleft;
1290
+
1291
+ lastchar = get_bits(8);
1292
+
1293
+ thischar = get_bits(8);
1294
+
1295
+ /* ok if it's a normal JPEG file without a special header */
1296
+
1297
+ if ((lastchar == 0xFF) && (thischar == M_SOI))
1298
+ return;
1299
+
1300
+ bytesleft = 4096; //512;
1301
+
1302
+ for ( ; ; )
1303
+ {
1304
+ if (--bytesleft == 0)
1305
+ stop_decoding(JPGD_NOT_JPEG);
1306
+
1307
+ lastchar = thischar;
1308
+
1309
+ thischar = get_bits(8);
1310
+
1311
+ if (lastchar == 0xFF)
1312
+ {
1313
+ if (thischar == M_SOI)
1314
+ break;
1315
+ else if (thischar == M_EOI) // get_bits will keep returning M_EOI if we read past the end
1316
+ stop_decoding(JPGD_NOT_JPEG);
1317
+ }
1318
+ }
1319
+
1320
+ // Check the next character after marker: if it's not 0xFF, it can't be the start of the next marker, so the file is bad.
1321
+ thischar = (m_bit_buf >> 24) & 0xFF;
1322
+
1323
+ if (thischar != 0xFF)
1324
+ stop_decoding(JPGD_NOT_JPEG);
1325
+ }
1326
+
1327
+ // Find a start of frame (SOF) marker.
1328
+ void jpeg_decoder::locate_sof_marker()
1329
+ {
1330
+ locate_soi_marker();
1331
+
1332
+ int c = process_markers();
1333
+
1334
+ switch (c)
1335
+ {
1336
+ case M_SOF2:
1337
+ m_progressive_flag = JPGD_TRUE;
1338
+ case M_SOF0: /* baseline DCT */
1339
+ case M_SOF1: /* extended sequential DCT */
1340
+ {
1341
+ read_sof_marker();
1342
+ break;
1343
+ }
1344
+ case M_SOF9: /* Arithmitic coding */
1345
+ {
1346
+ stop_decoding(JPGD_NO_ARITHMITIC_SUPPORT);
1347
+ break;
1348
+ }
1349
+ default:
1350
+ {
1351
+ stop_decoding(JPGD_UNSUPPORTED_MARKER);
1352
+ break;
1353
+ }
1354
+ }
1355
+ }
1356
+
1357
+ // Find a start of scan (SOS) marker.
1358
+ int jpeg_decoder::locate_sos_marker()
1359
+ {
1360
+ int c;
1361
+
1362
+ c = process_markers();
1363
+
1364
+ if (c == M_EOI)
1365
+ return JPGD_FALSE;
1366
+ else if (c != M_SOS)
1367
+ stop_decoding(JPGD_UNEXPECTED_MARKER);
1368
+
1369
+ read_sos_marker();
1370
+
1371
+ return JPGD_TRUE;
1372
+ }
1373
+
1374
+ // Reset everything to default/uninitialized state.
1375
+ void jpeg_decoder::init(jpeg_decoder_stream *pStream)
1376
+ {
1377
+ m_pMem_blocks = NULL;
1378
+ m_error_code = JPGD_SUCCESS;
1379
+ m_ready_flag = false;
1380
+ m_image_x_size = m_image_y_size = 0;
1381
+ m_pStream = pStream;
1382
+ m_progressive_flag = JPGD_FALSE;
1383
+
1384
+ memset(m_huff_ac, 0, sizeof(m_huff_ac));
1385
+ memset(m_huff_num, 0, sizeof(m_huff_num));
1386
+ memset(m_huff_val, 0, sizeof(m_huff_val));
1387
+ memset(m_quant, 0, sizeof(m_quant));
1388
+
1389
+ m_scan_type = 0;
1390
+ m_comps_in_frame = 0;
1391
+
1392
+ memset(m_comp_h_samp, 0, sizeof(m_comp_h_samp));
1393
+ memset(m_comp_v_samp, 0, sizeof(m_comp_v_samp));
1394
+ memset(m_comp_quant, 0, sizeof(m_comp_quant));
1395
+ memset(m_comp_ident, 0, sizeof(m_comp_ident));
1396
+ memset(m_comp_h_blocks, 0, sizeof(m_comp_h_blocks));
1397
+ memset(m_comp_v_blocks, 0, sizeof(m_comp_v_blocks));
1398
+
1399
+ m_comps_in_scan = 0;
1400
+ memset(m_comp_list, 0, sizeof(m_comp_list));
1401
+ memset(m_comp_dc_tab, 0, sizeof(m_comp_dc_tab));
1402
+ memset(m_comp_ac_tab, 0, sizeof(m_comp_ac_tab));
1403
+
1404
+ m_spectral_start = 0;
1405
+ m_spectral_end = 0;
1406
+ m_successive_low = 0;
1407
+ m_successive_high = 0;
1408
+ m_max_mcu_x_size = 0;
1409
+ m_max_mcu_y_size = 0;
1410
+ m_blocks_per_mcu = 0;
1411
+ m_max_blocks_per_row = 0;
1412
+ m_mcus_per_row = 0;
1413
+ m_mcus_per_col = 0;
1414
+ m_expanded_blocks_per_component = 0;
1415
+ m_expanded_blocks_per_mcu = 0;
1416
+ m_expanded_blocks_per_row = 0;
1417
+ m_freq_domain_chroma_upsample = false;
1418
+
1419
+ memset(m_mcu_org, 0, sizeof(m_mcu_org));
1420
+
1421
+ m_total_lines_left = 0;
1422
+ m_mcu_lines_left = 0;
1423
+ m_real_dest_bytes_per_scan_line = 0;
1424
+ m_dest_bytes_per_scan_line = 0;
1425
+ m_dest_bytes_per_pixel = 0;
1426
+
1427
+ memset(m_pHuff_tabs, 0, sizeof(m_pHuff_tabs));
1428
+
1429
+ memset(m_dc_coeffs, 0, sizeof(m_dc_coeffs));
1430
+ memset(m_ac_coeffs, 0, sizeof(m_ac_coeffs));
1431
+ memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
1432
+
1433
+ m_eob_run = 0;
1434
+
1435
+ memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
1436
+
1437
+ m_pIn_buf_ofs = m_in_buf;
1438
+ m_in_buf_left = 0;
1439
+ m_eof_flag = false;
1440
+ m_tem_flag = 0;
1441
+
1442
+ memset(m_in_buf_pad_start, 0, sizeof(m_in_buf_pad_start));
1443
+ memset(m_in_buf, 0, sizeof(m_in_buf));
1444
+ memset(m_in_buf_pad_end, 0, sizeof(m_in_buf_pad_end));
1445
+
1446
+ m_restart_interval = 0;
1447
+ m_restarts_left = 0;
1448
+ m_next_restart_num = 0;
1449
+
1450
+ m_max_mcus_per_row = 0;
1451
+ m_max_blocks_per_mcu = 0;
1452
+ m_max_mcus_per_col = 0;
1453
+
1454
+ memset(m_last_dc_val, 0, sizeof(m_last_dc_val));
1455
+ m_pMCU_coefficients = NULL;
1456
+ m_pSample_buf = NULL;
1457
+
1458
+ m_total_bytes_read = 0;
1459
+
1460
+ m_pScan_line_0 = NULL;
1461
+ m_pScan_line_1 = NULL;
1462
+
1463
+ // Ready the input buffer.
1464
+ prep_in_buffer();
1465
+
1466
+ // Prime the bit buffer.
1467
+ m_bits_left = 16;
1468
+ m_bit_buf = 0;
1469
+
1470
+ get_bits(16);
1471
+ get_bits(16);
1472
+
1473
+ for (int i = 0; i < JPGD_MAX_BLOCKS_PER_MCU; i++)
1474
+ m_mcu_block_max_zag[i] = 64;
1475
+ }
1476
+
1477
+ #define SCALEBITS 16
1478
+ #define ONE_HALF ((int) 1 << (SCALEBITS-1))
1479
+ #define FIX(x) ((int) ((x) * (1L<<SCALEBITS) + 0.5f))
1480
+
1481
+ // Create a few tables that allow us to quickly convert YCbCr to RGB.
1482
+ void jpeg_decoder::create_look_ups()
1483
+ {
1484
+ for (int i = 0; i <= 255; i++)
1485
+ {
1486
+ int k = i - 128;
1487
+ m_crr[i] = ( FIX(1.40200f) * k + ONE_HALF) >> SCALEBITS;
1488
+ m_cbb[i] = ( FIX(1.77200f) * k + ONE_HALF) >> SCALEBITS;
1489
+ m_crg[i] = (-FIX(0.71414f)) * k;
1490
+ m_cbg[i] = (-FIX(0.34414f)) * k + ONE_HALF;
1491
+ }
1492
+ }
1493
+
1494
+ // This method throws back into the stream any bytes that where read
1495
+ // into the bit buffer during initial marker scanning.
1496
+ void jpeg_decoder::fix_in_buffer()
1497
+ {
1498
+ // In case any 0xFF's where pulled into the buffer during marker scanning.
1499
+ JPGD_ASSERT((m_bits_left & 7) == 0);
1500
+
1501
+ if (m_bits_left == 16)
1502
+ stuff_char( (uint8)(m_bit_buf & 0xFF));
1503
+
1504
+ if (m_bits_left >= 8)
1505
+ stuff_char( (uint8)((m_bit_buf >> 8) & 0xFF));
1506
+
1507
+ stuff_char((uint8)((m_bit_buf >> 16) & 0xFF));
1508
+ stuff_char((uint8)((m_bit_buf >> 24) & 0xFF));
1509
+
1510
+ m_bits_left = 16;
1511
+ get_bits_no_markers(16);
1512
+ get_bits_no_markers(16);
1513
+ }
1514
+
1515
+ void jpeg_decoder::transform_mcu(int mcu_row)
1516
+ {
1517
+ jpgd_block_t* pSrc_ptr = m_pMCU_coefficients;
1518
+ uint8* pDst_ptr = m_pSample_buf + mcu_row * m_blocks_per_mcu * 64;
1519
+
1520
+ for (int mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++)
1521
+ {
1522
+ idct(pSrc_ptr, pDst_ptr, m_mcu_block_max_zag[mcu_block]);
1523
+ pSrc_ptr += 64;
1524
+ pDst_ptr += 64;
1525
+ }
1526
+ }
1527
+
1528
+ static const uint8 s_max_rc[64] =
1529
+ {
1530
+ 17, 18, 34, 50, 50, 51, 52, 52, 52, 68, 84, 84, 84, 84, 85, 86, 86, 86, 86, 86,
1531
+ 102, 118, 118, 118, 118, 118, 118, 119, 120, 120, 120, 120, 120, 120, 120, 136,
1532
+ 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136,
1533
+ 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136
1534
+ };
1535
+
1536
+ void jpeg_decoder::transform_mcu_expand(int mcu_row)
1537
+ {
1538
+ jpgd_block_t* pSrc_ptr = m_pMCU_coefficients;
1539
+ uint8* pDst_ptr = m_pSample_buf + mcu_row * m_expanded_blocks_per_mcu * 64;
1540
+
1541
+ // Y IDCT
1542
+ int mcu_block;
1543
+ for (mcu_block = 0; mcu_block < m_expanded_blocks_per_component; mcu_block++)
1544
+ {
1545
+ idct(pSrc_ptr, pDst_ptr, m_mcu_block_max_zag[mcu_block]);
1546
+ pSrc_ptr += 64;
1547
+ pDst_ptr += 64;
1548
+ }
1549
+
1550
+ // Chroma IDCT, with upsampling
1551
+ jpgd_block_t temp_block[64];
1552
+
1553
+ for (int i = 0; i < 2; i++)
1554
+ {
1555
+ DCT_Upsample::Matrix44 P, Q, R, S;
1556
+
1557
+ JPGD_ASSERT(m_mcu_block_max_zag[mcu_block] >= 1);
1558
+ JPGD_ASSERT(m_mcu_block_max_zag[mcu_block] <= 64);
1559
+
1560
+ switch (s_max_rc[m_mcu_block_max_zag[mcu_block++] - 1])
1561
+ {
1562
+ case 1*16+1:
1563
+ DCT_Upsample::P_Q<1, 1>::calc(P, Q, pSrc_ptr);
1564
+ DCT_Upsample::R_S<1, 1>::calc(R, S, pSrc_ptr);
1565
+ break;
1566
+ case 1*16+2:
1567
+ DCT_Upsample::P_Q<1, 2>::calc(P, Q, pSrc_ptr);
1568
+ DCT_Upsample::R_S<1, 2>::calc(R, S, pSrc_ptr);
1569
+ break;
1570
+ case 2*16+2:
1571
+ DCT_Upsample::P_Q<2, 2>::calc(P, Q, pSrc_ptr);
1572
+ DCT_Upsample::R_S<2, 2>::calc(R, S, pSrc_ptr);
1573
+ break;
1574
+ case 3*16+2:
1575
+ DCT_Upsample::P_Q<3, 2>::calc(P, Q, pSrc_ptr);
1576
+ DCT_Upsample::R_S<3, 2>::calc(R, S, pSrc_ptr);
1577
+ break;
1578
+ case 3*16+3:
1579
+ DCT_Upsample::P_Q<3, 3>::calc(P, Q, pSrc_ptr);
1580
+ DCT_Upsample::R_S<3, 3>::calc(R, S, pSrc_ptr);
1581
+ break;
1582
+ case 3*16+4:
1583
+ DCT_Upsample::P_Q<3, 4>::calc(P, Q, pSrc_ptr);
1584
+ DCT_Upsample::R_S<3, 4>::calc(R, S, pSrc_ptr);
1585
+ break;
1586
+ case 4*16+4:
1587
+ DCT_Upsample::P_Q<4, 4>::calc(P, Q, pSrc_ptr);
1588
+ DCT_Upsample::R_S<4, 4>::calc(R, S, pSrc_ptr);
1589
+ break;
1590
+ case 5*16+4:
1591
+ DCT_Upsample::P_Q<5, 4>::calc(P, Q, pSrc_ptr);
1592
+ DCT_Upsample::R_S<5, 4>::calc(R, S, pSrc_ptr);
1593
+ break;
1594
+ case 5*16+5:
1595
+ DCT_Upsample::P_Q<5, 5>::calc(P, Q, pSrc_ptr);
1596
+ DCT_Upsample::R_S<5, 5>::calc(R, S, pSrc_ptr);
1597
+ break;
1598
+ case 5*16+6:
1599
+ DCT_Upsample::P_Q<5, 6>::calc(P, Q, pSrc_ptr);
1600
+ DCT_Upsample::R_S<5, 6>::calc(R, S, pSrc_ptr);
1601
+ break;
1602
+ case 6*16+6:
1603
+ DCT_Upsample::P_Q<6, 6>::calc(P, Q, pSrc_ptr);
1604
+ DCT_Upsample::R_S<6, 6>::calc(R, S, pSrc_ptr);
1605
+ break;
1606
+ case 7*16+6:
1607
+ DCT_Upsample::P_Q<7, 6>::calc(P, Q, pSrc_ptr);
1608
+ DCT_Upsample::R_S<7, 6>::calc(R, S, pSrc_ptr);
1609
+ break;
1610
+ case 7*16+7:
1611
+ DCT_Upsample::P_Q<7, 7>::calc(P, Q, pSrc_ptr);
1612
+ DCT_Upsample::R_S<7, 7>::calc(R, S, pSrc_ptr);
1613
+ break;
1614
+ case 7*16+8:
1615
+ DCT_Upsample::P_Q<7, 8>::calc(P, Q, pSrc_ptr);
1616
+ DCT_Upsample::R_S<7, 8>::calc(R, S, pSrc_ptr);
1617
+ break;
1618
+ case 8*16+8:
1619
+ DCT_Upsample::P_Q<8, 8>::calc(P, Q, pSrc_ptr);
1620
+ DCT_Upsample::R_S<8, 8>::calc(R, S, pSrc_ptr);
1621
+ break;
1622
+ default:
1623
+ JPGD_ASSERT(false);
1624
+ }
1625
+
1626
+ DCT_Upsample::Matrix44 a(P + Q); P -= Q;
1627
+ DCT_Upsample::Matrix44& b = P;
1628
+ DCT_Upsample::Matrix44 c(R + S); R -= S;
1629
+ DCT_Upsample::Matrix44& d = R;
1630
+
1631
+ DCT_Upsample::Matrix44::add_and_store(temp_block, a, c);
1632
+ idct_4x4(temp_block, pDst_ptr);
1633
+ pDst_ptr += 64;
1634
+
1635
+ DCT_Upsample::Matrix44::sub_and_store(temp_block, a, c);
1636
+ idct_4x4(temp_block, pDst_ptr);
1637
+ pDst_ptr += 64;
1638
+
1639
+ DCT_Upsample::Matrix44::add_and_store(temp_block, b, d);
1640
+ idct_4x4(temp_block, pDst_ptr);
1641
+ pDst_ptr += 64;
1642
+
1643
+ DCT_Upsample::Matrix44::sub_and_store(temp_block, b, d);
1644
+ idct_4x4(temp_block, pDst_ptr);
1645
+ pDst_ptr += 64;
1646
+
1647
+ pSrc_ptr += 64;
1648
+ }
1649
+ }
1650
+
1651
+ // Loads and dequantizes the next row of (already decoded) coefficients.
1652
+ // Progressive images only.
1653
+ void jpeg_decoder::load_next_row()
1654
+ {
1655
+ int i;
1656
+ jpgd_block_t *p;
1657
+ jpgd_quant_t *q;
1658
+ int mcu_row, mcu_block, row_block = 0;
1659
+ int component_num, component_id;
1660
+ int block_x_mcu[JPGD_MAX_COMPONENTS];
1661
+
1662
+ memset(block_x_mcu, 0, JPGD_MAX_COMPONENTS * sizeof(int));
1663
+
1664
+ for (mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++)
1665
+ {
1666
+ int block_x_mcu_ofs = 0, block_y_mcu_ofs = 0;
1667
+
1668
+ for (mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++)
1669
+ {
1670
+ component_id = m_mcu_org[mcu_block];
1671
+ q = m_quant[m_comp_quant[component_id]];
1672
+
1673
+ p = m_pMCU_coefficients + 64 * mcu_block;
1674
+
1675
+ jpgd_block_t* pAC = coeff_buf_getp(m_ac_coeffs[component_id], block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
1676
+ jpgd_block_t* pDC = coeff_buf_getp(m_dc_coeffs[component_id], block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
1677
+ p[0] = pDC[0];
1678
+ memcpy(&p[1], &pAC[1], 63 * sizeof(jpgd_block_t));
1679
+
1680
+ for (i = 63; i > 0; i--)
1681
+ if (p[g_ZAG[i]])
1682
+ break;
1683
+
1684
+ m_mcu_block_max_zag[mcu_block] = i + 1;
1685
+
1686
+ for ( ; i >= 0; i--)
1687
+ if (p[g_ZAG[i]])
1688
+ p[g_ZAG[i]] = static_cast<jpgd_block_t>(p[g_ZAG[i]] * q[i]);
1689
+
1690
+ row_block++;
1691
+
1692
+ if (m_comps_in_scan == 1)
1693
+ block_x_mcu[component_id]++;
1694
+ else
1695
+ {
1696
+ if (++block_x_mcu_ofs == m_comp_h_samp[component_id])
1697
+ {
1698
+ block_x_mcu_ofs = 0;
1699
+
1700
+ if (++block_y_mcu_ofs == m_comp_v_samp[component_id])
1701
+ {
1702
+ block_y_mcu_ofs = 0;
1703
+
1704
+ block_x_mcu[component_id] += m_comp_h_samp[component_id];
1705
+ }
1706
+ }
1707
+ }
1708
+ }
1709
+
1710
+ if (m_freq_domain_chroma_upsample)
1711
+ transform_mcu_expand(mcu_row);
1712
+ else
1713
+ transform_mcu(mcu_row);
1714
+ }
1715
+
1716
+ if (m_comps_in_scan == 1)
1717
+ m_block_y_mcu[m_comp_list[0]]++;
1718
+ else
1719
+ {
1720
+ for (component_num = 0; component_num < m_comps_in_scan; component_num++)
1721
+ {
1722
+ component_id = m_comp_list[component_num];
1723
+
1724
+ m_block_y_mcu[component_id] += m_comp_v_samp[component_id];
1725
+ }
1726
+ }
1727
+ }
1728
+
1729
+ // Restart interval processing.
1730
+ void jpeg_decoder::process_restart()
1731
+ {
1732
+ int i;
1733
+ int c = 0;
1734
+
1735
+ // Align to a byte boundry
1736
+ // FIXME: Is this really necessary? get_bits_no_markers() never reads in markers!
1737
+ //get_bits_no_markers(m_bits_left & 7);
1738
+
1739
+ // Let's scan a little bit to find the marker, but not _too_ far.
1740
+ // 1536 is a "fudge factor" that determines how much to scan.
1741
+ for (i = 1536; i > 0; i--)
1742
+ if (get_char() == 0xFF)
1743
+ break;
1744
+
1745
+ if (i == 0)
1746
+ stop_decoding(JPGD_BAD_RESTART_MARKER);
1747
+
1748
+ for ( ; i > 0; i--)
1749
+ if ((c = get_char()) != 0xFF)
1750
+ break;
1751
+
1752
+ if (i == 0)
1753
+ stop_decoding(JPGD_BAD_RESTART_MARKER);
1754
+
1755
+ // Is it the expected marker? If not, something bad happened.
1756
+ if (c != (m_next_restart_num + M_RST0))
1757
+ stop_decoding(JPGD_BAD_RESTART_MARKER);
1758
+
1759
+ // Reset each component's DC prediction values.
1760
+ memset(&m_last_dc_val, 0, m_comps_in_frame * sizeof(uint));
1761
+
1762
+ m_eob_run = 0;
1763
+
1764
+ m_restarts_left = m_restart_interval;
1765
+
1766
+ m_next_restart_num = (m_next_restart_num + 1) & 7;
1767
+
1768
+ // Get the bit buffer going again...
1769
+
1770
+ m_bits_left = 16;
1771
+ get_bits_no_markers(16);
1772
+ get_bits_no_markers(16);
1773
+ }
1774
+
1775
+ static inline int dequantize_ac(int c, int q) { c *= q; return c; }
1776
+
1777
+ // Decodes and dequantizes the next row of coefficients.
1778
+ void jpeg_decoder::decode_next_row()
1779
+ {
1780
+ int row_block = 0;
1781
+
1782
+ for (int mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++)
1783
+ {
1784
+ if ((m_restart_interval) && (m_restarts_left == 0))
1785
+ process_restart();
1786
+
1787
+ jpgd_block_t* p = m_pMCU_coefficients;
1788
+ for (int mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++, p += 64)
1789
+ {
1790
+ int component_id = m_mcu_org[mcu_block];
1791
+ jpgd_quant_t* q = m_quant[m_comp_quant[component_id]];
1792
+
1793
+ int r, s;
1794
+ s = huff_decode(m_pHuff_tabs[m_comp_dc_tab[component_id]], r);
1795
+ s = HUFF_EXTEND(r, s);
1796
+
1797
+ m_last_dc_val[component_id] = (s += m_last_dc_val[component_id]);
1798
+
1799
+ p[0] = static_cast<jpgd_block_t>(s * q[0]);
1800
+
1801
+ int prev_num_set = m_mcu_block_max_zag[mcu_block];
1802
+
1803
+ huff_tables *pH = m_pHuff_tabs[m_comp_ac_tab[component_id]];
1804
+
1805
+ int k;
1806
+ for (k = 1; k < 64; k++)
1807
+ {
1808
+ int extra_bits;
1809
+ s = huff_decode(pH, extra_bits);
1810
+
1811
+ r = s >> 4;
1812
+ s &= 15;
1813
+
1814
+ if (s)
1815
+ {
1816
+ if (r)
1817
+ {
1818
+ if ((k + r) > 63)
1819
+ stop_decoding(JPGD_DECODE_ERROR);
1820
+
1821
+ if (k < prev_num_set)
1822
+ {
1823
+ int n = JPGD_MIN(r, prev_num_set - k);
1824
+ int kt = k;
1825
+ while (n--)
1826
+ p[g_ZAG[kt++]] = 0;
1827
+ }
1828
+
1829
+ k += r;
1830
+ }
1831
+
1832
+ s = HUFF_EXTEND(extra_bits, s);
1833
+
1834
+ JPGD_ASSERT(k < 64);
1835
+
1836
+ p[g_ZAG[k]] = static_cast<jpgd_block_t>(dequantize_ac(s, q[k])); //s * q[k];
1837
+ }
1838
+ else
1839
+ {
1840
+ if (r == 15)
1841
+ {
1842
+ if ((k + 16) > 64)
1843
+ stop_decoding(JPGD_DECODE_ERROR);
1844
+
1845
+ if (k < prev_num_set)
1846
+ {
1847
+ int n = JPGD_MIN(16, prev_num_set - k);
1848
+ int kt = k;
1849
+ while (n--)
1850
+ {
1851
+ JPGD_ASSERT(kt <= 63);
1852
+ p[g_ZAG[kt++]] = 0;
1853
+ }
1854
+ }
1855
+
1856
+ k += 16 - 1; // - 1 because the loop counter is k
1857
+ // BEGIN EPIC MOD
1858
+ JPGD_ASSERT(k < 64 && p[g_ZAG[k]] == 0);
1859
+ // END EPIC MOD
1860
+ }
1861
+ else
1862
+ break;
1863
+ }
1864
+ }
1865
+
1866
+ if (k < prev_num_set)
1867
+ {
1868
+ int kt = k;
1869
+ while (kt < prev_num_set)
1870
+ p[g_ZAG[kt++]] = 0;
1871
+ }
1872
+
1873
+ m_mcu_block_max_zag[mcu_block] = k;
1874
+
1875
+ row_block++;
1876
+ }
1877
+
1878
+ if (m_freq_domain_chroma_upsample)
1879
+ transform_mcu_expand(mcu_row);
1880
+ else
1881
+ transform_mcu(mcu_row);
1882
+
1883
+ m_restarts_left--;
1884
+ }
1885
+ }
1886
+
1887
+ // YCbCr H1V1 (1x1:1:1, 3 m_blocks per MCU) to RGB
1888
+ void jpeg_decoder::H1V1Convert()
1889
+ {
1890
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
1891
+ uint8 *d = m_pScan_line_0;
1892
+ uint8 *s = m_pSample_buf + row * 8;
1893
+
1894
+ for (int i = m_max_mcus_per_row; i > 0; i--)
1895
+ {
1896
+ for (int j = 0; j < 8; j++)
1897
+ {
1898
+ int y = s[j];
1899
+ int cb = s[64+j];
1900
+ int cr = s[128+j];
1901
+
1902
+ if (jpg_format == ERGBFormatJPG::BGRA)
1903
+ {
1904
+ d[0] = clamp(y + m_cbb[cb]);
1905
+ d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
1906
+ d[2] = clamp(y + m_crr[cr]);
1907
+ d[3] = 255;
1908
+ }
1909
+ else
1910
+ {
1911
+ d[0] = clamp(y + m_crr[cr]);
1912
+ d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
1913
+ d[2] = clamp(y + m_cbb[cb]);
1914
+ d[3] = 255;
1915
+ }
1916
+ d += 4;
1917
+ }
1918
+
1919
+ s += 64*3;
1920
+ }
1921
+ }
1922
+
1923
+ // YCbCr H2V1 (2x1:1:1, 4 m_blocks per MCU) to RGB
1924
+ void jpeg_decoder::H2V1Convert()
1925
+ {
1926
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
1927
+ uint8 *d0 = m_pScan_line_0;
1928
+ uint8 *y = m_pSample_buf + row * 8;
1929
+ uint8 *c = m_pSample_buf + 2*64 + row * 8;
1930
+
1931
+ for (int i = m_max_mcus_per_row; i > 0; i--)
1932
+ {
1933
+ for (int l = 0; l < 2; l++)
1934
+ {
1935
+ for (int j = 0; j < 4; j++)
1936
+ {
1937
+ int cb = c[0];
1938
+ int cr = c[64];
1939
+
1940
+ int rc = m_crr[cr];
1941
+ int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
1942
+ int bc = m_cbb[cb];
1943
+
1944
+ int yy = y[j<<1];
1945
+ if (jpg_format == ERGBFormatJPG::BGRA)
1946
+ {
1947
+ d0[0] = clamp(yy+bc);
1948
+ d0[1] = clamp(yy+gc);
1949
+ d0[2] = clamp(yy+rc);
1950
+ d0[3] = 255;
1951
+ yy = y[(j<<1)+1];
1952
+ d0[4] = clamp(yy+bc);
1953
+ d0[5] = clamp(yy+gc);
1954
+ d0[6] = clamp(yy+rc);
1955
+ d0[7] = 255;
1956
+ }
1957
+ else
1958
+ {
1959
+ d0[0] = clamp(yy+rc);
1960
+ d0[1] = clamp(yy+gc);
1961
+ d0[2] = clamp(yy+bc);
1962
+ d0[3] = 255;
1963
+ yy = y[(j<<1)+1];
1964
+ d0[4] = clamp(yy+rc);
1965
+ d0[5] = clamp(yy+gc);
1966
+ d0[6] = clamp(yy+bc);
1967
+ d0[7] = 255;
1968
+ }
1969
+
1970
+ d0 += 8;
1971
+
1972
+ c++;
1973
+ }
1974
+ y += 64;
1975
+ }
1976
+
1977
+ y += 64*4 - 64*2;
1978
+ c += 64*4 - 8;
1979
+ }
1980
+ }
1981
+
1982
+ // YCbCr H2V1 (1x2:1:1, 4 m_blocks per MCU) to RGB
1983
+ void jpeg_decoder::H1V2Convert()
1984
+ {
1985
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
1986
+ uint8 *d0 = m_pScan_line_0;
1987
+ uint8 *d1 = m_pScan_line_1;
1988
+ uint8 *y;
1989
+ uint8 *c;
1990
+
1991
+ if (row < 8)
1992
+ y = m_pSample_buf + row * 8;
1993
+ else
1994
+ y = m_pSample_buf + 64*1 + (row & 7) * 8;
1995
+
1996
+ c = m_pSample_buf + 64*2 + (row >> 1) * 8;
1997
+
1998
+ for (int i = m_max_mcus_per_row; i > 0; i--)
1999
+ {
2000
+ for (int j = 0; j < 8; j++)
2001
+ {
2002
+ int cb = c[0+j];
2003
+ int cr = c[64+j];
2004
+
2005
+ int rc = m_crr[cr];
2006
+ int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
2007
+ int bc = m_cbb[cb];
2008
+
2009
+ int yy = y[j];
2010
+ if (jpg_format == ERGBFormatJPG::BGRA)
2011
+ {
2012
+ d0[0] = clamp(yy+bc);
2013
+ d0[1] = clamp(yy+gc);
2014
+ d0[2] = clamp(yy+rc);
2015
+ d0[3] = 255;
2016
+ yy = y[8+j];
2017
+ d1[0] = clamp(yy+bc);
2018
+ d1[1] = clamp(yy+gc);
2019
+ d1[2] = clamp(yy+rc);
2020
+ d1[3] = 255;
2021
+ }
2022
+ else
2023
+ {
2024
+ d0[0] = clamp(yy+rc);
2025
+ d0[1] = clamp(yy+gc);
2026
+ d0[2] = clamp(yy+bc);
2027
+ d0[3] = 255;
2028
+ yy = y[8+j];
2029
+ d1[0] = clamp(yy+rc);
2030
+ d1[1] = clamp(yy+gc);
2031
+ d1[2] = clamp(yy+bc);
2032
+ d1[3] = 255;
2033
+ }
2034
+
2035
+ d0 += 4;
2036
+ d1 += 4;
2037
+ }
2038
+
2039
+ y += 64*4;
2040
+ c += 64*4;
2041
+ }
2042
+ }
2043
+
2044
+ // YCbCr H2V2 (2x2:1:1, 6 m_blocks per MCU) to RGB
2045
+ void jpeg_decoder::H2V2Convert()
2046
+ {
2047
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
2048
+ uint8 *d0 = m_pScan_line_0;
2049
+ uint8 *d1 = m_pScan_line_1;
2050
+ uint8 *y;
2051
+ uint8 *c;
2052
+
2053
+ if (row < 8)
2054
+ y = m_pSample_buf + row * 8;
2055
+ else
2056
+ y = m_pSample_buf + 64*2 + (row & 7) * 8;
2057
+
2058
+ c = m_pSample_buf + 64*4 + (row >> 1) * 8;
2059
+
2060
+ for (int i = m_max_mcus_per_row; i > 0; i--)
2061
+ {
2062
+ for (int l = 0; l < 2; l++)
2063
+ {
2064
+ for (int j = 0; j < 8; j += 2)
2065
+ {
2066
+ int cb = c[0];
2067
+ int cr = c[64];
2068
+
2069
+ int rc = m_crr[cr];
2070
+ int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
2071
+ int bc = m_cbb[cb];
2072
+
2073
+ int yy = y[j];
2074
+ if (jpg_format == ERGBFormatJPG::BGRA)
2075
+ {
2076
+ d0[0] = clamp(yy+bc);
2077
+ d0[1] = clamp(yy+gc);
2078
+ d0[2] = clamp(yy+rc);
2079
+ d0[3] = 255;
2080
+ yy = y[j+1];
2081
+ d0[4] = clamp(yy+bc);
2082
+ d0[5] = clamp(yy+gc);
2083
+ d0[6] = clamp(yy+rc);
2084
+ d0[7] = 255;
2085
+ yy = y[j+8];
2086
+ d1[0] = clamp(yy+bc);
2087
+ d1[1] = clamp(yy+gc);
2088
+ d1[2] = clamp(yy+rc);
2089
+ d1[3] = 255;
2090
+ yy = y[j+8+1];
2091
+ d1[4] = clamp(yy+bc);
2092
+ d1[5] = clamp(yy+gc);
2093
+ d1[6] = clamp(yy+rc);
2094
+ d1[7] = 255;
2095
+ }
2096
+ else
2097
+ {
2098
+ d0[0] = clamp(yy+rc);
2099
+ d0[1] = clamp(yy+gc);
2100
+ d0[2] = clamp(yy+bc);
2101
+ d0[3] = 255;
2102
+ yy = y[j+1];
2103
+ d0[4] = clamp(yy+rc);
2104
+ d0[5] = clamp(yy+gc);
2105
+ d0[6] = clamp(yy+bc);
2106
+ d0[7] = 255;
2107
+ yy = y[j+8];
2108
+ d1[0] = clamp(yy+rc);
2109
+ d1[1] = clamp(yy+gc);
2110
+ d1[2] = clamp(yy+bc);
2111
+ d1[3] = 255;
2112
+ yy = y[j+8+1];
2113
+ d1[4] = clamp(yy+rc);
2114
+ d1[5] = clamp(yy+gc);
2115
+ d1[6] = clamp(yy+bc);
2116
+ d1[7] = 255;
2117
+ }
2118
+
2119
+ d0 += 8;
2120
+ d1 += 8;
2121
+
2122
+ c++;
2123
+ }
2124
+ y += 64;
2125
+ }
2126
+
2127
+ y += 64*6 - 64*2;
2128
+ c += 64*6 - 8;
2129
+ }
2130
+ }
2131
+
2132
+ // Y (1 block per MCU) to 8-bit grayscale
2133
+ void jpeg_decoder::gray_convert()
2134
+ {
2135
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
2136
+ uint8 *d = m_pScan_line_0;
2137
+ uint8 *s = m_pSample_buf + row * 8;
2138
+
2139
+ for (int i = m_max_mcus_per_row; i > 0; i--)
2140
+ {
2141
+ *(uint *)d = *(uint *)s;
2142
+ *(uint *)(&d[4]) = *(uint *)(&s[4]);
2143
+
2144
+ s += 64;
2145
+ d += 8;
2146
+ }
2147
+ }
2148
+
2149
+ void jpeg_decoder::expanded_convert()
2150
+ {
2151
+ int row = m_max_mcu_y_size - m_mcu_lines_left;
2152
+
2153
+ uint8* Py = m_pSample_buf + (row / 8) * 64 * m_comp_h_samp[0] + (row & 7) * 8;
2154
+
2155
+ uint8* d = m_pScan_line_0;
2156
+
2157
+ for (int i = m_max_mcus_per_row; i > 0; i--)
2158
+ {
2159
+ for (int k = 0; k < m_max_mcu_x_size; k += 8)
2160
+ {
2161
+ const int Y_ofs = k * 8;
2162
+ const int Cb_ofs = Y_ofs + 64 * m_expanded_blocks_per_component;
2163
+ const int Cr_ofs = Y_ofs + 64 * m_expanded_blocks_per_component * 2;
2164
+ for (int j = 0; j < 8; j++)
2165
+ {
2166
+ int y = Py[Y_ofs + j];
2167
+ int cb = Py[Cb_ofs + j];
2168
+ int cr = Py[Cr_ofs + j];
2169
+
2170
+ if (jpg_format == ERGBFormatJPG::BGRA)
2171
+ {
2172
+ d[0] = clamp(y + m_cbb[cb]);
2173
+ d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
2174
+ d[2] = clamp(y + m_crr[cr]);
2175
+ d[3] = 255;
2176
+ }
2177
+ else
2178
+ {
2179
+ d[0] = clamp(y + m_crr[cr]);
2180
+ d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
2181
+ d[2] = clamp(y + m_cbb[cb]);
2182
+ d[3] = 255;
2183
+ }
2184
+
2185
+ d += 4;
2186
+ }
2187
+ }
2188
+
2189
+ Py += 64 * m_expanded_blocks_per_mcu;
2190
+ }
2191
+ }
2192
+
2193
+ // Find end of image (EOI) marker, so we can return to the user the exact size of the input stream.
2194
+ void jpeg_decoder::find_eoi()
2195
+ {
2196
+ if (!m_progressive_flag)
2197
+ {
2198
+ // Attempt to read the EOI marker.
2199
+ //get_bits_no_markers(m_bits_left & 7);
2200
+
2201
+ // Prime the bit buffer
2202
+ m_bits_left = 16;
2203
+ get_bits(16);
2204
+ get_bits(16);
2205
+
2206
+ // The next marker _should_ be EOI
2207
+ process_markers();
2208
+ }
2209
+
2210
+ m_total_bytes_read -= m_in_buf_left;
2211
+ }
2212
+
2213
+ int jpeg_decoder::decode(const void** pScan_line, uint* pScan_line_len)
2214
+ {
2215
+ if ((m_error_code) || (!m_ready_flag))
2216
+ return JPGD_FAILED;
2217
+
2218
+ if (m_total_lines_left == 0)
2219
+ return JPGD_DONE;
2220
+
2221
+ if (m_mcu_lines_left == 0)
2222
+ {
2223
+ if (setjmp(m_jmp_state))
2224
+ return JPGD_FAILED;
2225
+
2226
+ if (m_progressive_flag)
2227
+ load_next_row();
2228
+ else
2229
+ decode_next_row();
2230
+
2231
+ // Find the EOI marker if that was the last row.
2232
+ if (m_total_lines_left <= m_max_mcu_y_size)
2233
+ find_eoi();
2234
+
2235
+ m_mcu_lines_left = m_max_mcu_y_size;
2236
+ }
2237
+
2238
+ if (m_freq_domain_chroma_upsample)
2239
+ {
2240
+ expanded_convert();
2241
+ *pScan_line = m_pScan_line_0;
2242
+ }
2243
+ else
2244
+ {
2245
+ switch (m_scan_type)
2246
+ {
2247
+ case JPGD_YH2V2:
2248
+ {
2249
+ if ((m_mcu_lines_left & 1) == 0)
2250
+ {
2251
+ H2V2Convert();
2252
+ *pScan_line = m_pScan_line_0;
2253
+ }
2254
+ else
2255
+ *pScan_line = m_pScan_line_1;
2256
+
2257
+ break;
2258
+ }
2259
+ case JPGD_YH2V1:
2260
+ {
2261
+ H2V1Convert();
2262
+ *pScan_line = m_pScan_line_0;
2263
+ break;
2264
+ }
2265
+ case JPGD_YH1V2:
2266
+ {
2267
+ if ((m_mcu_lines_left & 1) == 0)
2268
+ {
2269
+ H1V2Convert();
2270
+ *pScan_line = m_pScan_line_0;
2271
+ }
2272
+ else
2273
+ *pScan_line = m_pScan_line_1;
2274
+
2275
+ break;
2276
+ }
2277
+ case JPGD_YH1V1:
2278
+ {
2279
+ H1V1Convert();
2280
+ *pScan_line = m_pScan_line_0;
2281
+ break;
2282
+ }
2283
+ case JPGD_GRAYSCALE:
2284
+ {
2285
+ gray_convert();
2286
+ *pScan_line = m_pScan_line_0;
2287
+
2288
+ break;
2289
+ }
2290
+ }
2291
+ }
2292
+
2293
+ *pScan_line_len = m_real_dest_bytes_per_scan_line;
2294
+
2295
+ m_mcu_lines_left--;
2296
+ m_total_lines_left--;
2297
+
2298
+ return JPGD_SUCCESS;
2299
+ }
2300
+
2301
+ // Creates the tables needed for efficient Huffman decoding.
2302
+ void jpeg_decoder::make_huff_table(int index, huff_tables *pH)
2303
+ {
2304
+ int p, i, l, si;
2305
+ uint8 huffsize[257];
2306
+ uint huffcode[257];
2307
+ uint code;
2308
+ uint subtree;
2309
+ int code_size;
2310
+ int lastp;
2311
+ int nextfreeentry;
2312
+ int currententry;
2313
+
2314
+ pH->ac_table = m_huff_ac[index] != 0;
2315
+
2316
+ p = 0;
2317
+
2318
+ for (l = 1; l <= 16; l++)
2319
+ {
2320
+ for (i = 1; i <= m_huff_num[index][l]; i++)
2321
+ huffsize[p++] = static_cast<uint8>(l);
2322
+ }
2323
+
2324
+ huffsize[p] = 0;
2325
+
2326
+ lastp = p;
2327
+
2328
+ code = 0;
2329
+ si = huffsize[0];
2330
+ p = 0;
2331
+
2332
+ while (huffsize[p])
2333
+ {
2334
+ while (huffsize[p] == si)
2335
+ {
2336
+ huffcode[p++] = code;
2337
+ code++;
2338
+ }
2339
+
2340
+ code <<= 1;
2341
+ si++;
2342
+ }
2343
+
2344
+ memset(pH->look_up, 0, sizeof(pH->look_up));
2345
+ memset(pH->look_up2, 0, sizeof(pH->look_up2));
2346
+ memset(pH->tree, 0, sizeof(pH->tree));
2347
+ memset(pH->code_size, 0, sizeof(pH->code_size));
2348
+
2349
+ nextfreeentry = -1;
2350
+
2351
+ p = 0;
2352
+
2353
+ while (p < lastp)
2354
+ {
2355
+ i = m_huff_val[index][p];
2356
+ code = huffcode[p];
2357
+ code_size = huffsize[p];
2358
+
2359
+ pH->code_size[i] = static_cast<uint8>(code_size);
2360
+
2361
+ if (code_size <= 8)
2362
+ {
2363
+ code <<= (8 - code_size);
2364
+
2365
+ for (l = 1 << (8 - code_size); l > 0; l--)
2366
+ {
2367
+ JPGD_ASSERT(i < 256);
2368
+
2369
+ pH->look_up[code] = i;
2370
+
2371
+ bool has_extrabits = false;
2372
+ int extra_bits = 0;
2373
+ int num_extra_bits = i & 15;
2374
+
2375
+ int bits_to_fetch = code_size;
2376
+ if (num_extra_bits)
2377
+ {
2378
+ int total_codesize = code_size + num_extra_bits;
2379
+ if (total_codesize <= 8)
2380
+ {
2381
+ has_extrabits = true;
2382
+ extra_bits = ((1 << num_extra_bits) - 1) & (code >> (8 - total_codesize));
2383
+ JPGD_ASSERT(extra_bits <= 0x7FFF);
2384
+ bits_to_fetch += num_extra_bits;
2385
+ }
2386
+ }
2387
+
2388
+ if (!has_extrabits)
2389
+ pH->look_up2[code] = i | (bits_to_fetch << 8);
2390
+ else
2391
+ pH->look_up2[code] = i | 0x8000 | (extra_bits << 16) | (bits_to_fetch << 8);
2392
+
2393
+ code++;
2394
+ }
2395
+ }
2396
+ else
2397
+ {
2398
+ subtree = (code >> (code_size - 8)) & 0xFF;
2399
+
2400
+ currententry = pH->look_up[subtree];
2401
+
2402
+ if (currententry == 0)
2403
+ {
2404
+ pH->look_up[subtree] = currententry = nextfreeentry;
2405
+ pH->look_up2[subtree] = currententry = nextfreeentry;
2406
+
2407
+ nextfreeentry -= 2;
2408
+ }
2409
+
2410
+ code <<= (16 - (code_size - 8));
2411
+
2412
+ for (l = code_size; l > 9; l--)
2413
+ {
2414
+ if ((code & 0x8000) == 0)
2415
+ currententry--;
2416
+
2417
+ if (pH->tree[-currententry - 1] == 0)
2418
+ {
2419
+ pH->tree[-currententry - 1] = nextfreeentry;
2420
+
2421
+ currententry = nextfreeentry;
2422
+
2423
+ nextfreeentry -= 2;
2424
+ }
2425
+ else
2426
+ currententry = pH->tree[-currententry - 1];
2427
+
2428
+ code <<= 1;
2429
+ }
2430
+
2431
+ if ((code & 0x8000) == 0)
2432
+ currententry--;
2433
+
2434
+ pH->tree[-currententry - 1] = i;
2435
+ }
2436
+
2437
+ p++;
2438
+ }
2439
+ }
2440
+
2441
+ // Verifies the quantization tables needed for this scan are available.
2442
+ void jpeg_decoder::check_quant_tables()
2443
+ {
2444
+ for (int i = 0; i < m_comps_in_scan; i++)
2445
+ if (m_quant[m_comp_quant[m_comp_list[i]]] == NULL)
2446
+ stop_decoding(JPGD_UNDEFINED_QUANT_TABLE);
2447
+ }
2448
+
2449
+ // Verifies that all the Huffman tables needed for this scan are available.
2450
+ void jpeg_decoder::check_huff_tables()
2451
+ {
2452
+ for (int i = 0; i < m_comps_in_scan; i++)
2453
+ {
2454
+ if ((m_spectral_start == 0) && (m_huff_num[m_comp_dc_tab[m_comp_list[i]]] == NULL))
2455
+ stop_decoding(JPGD_UNDEFINED_HUFF_TABLE);
2456
+
2457
+ if ((m_spectral_end > 0) && (m_huff_num[m_comp_ac_tab[m_comp_list[i]]] == NULL))
2458
+ stop_decoding(JPGD_UNDEFINED_HUFF_TABLE);
2459
+ }
2460
+
2461
+ for (int i = 0; i < JPGD_MAX_HUFF_TABLES; i++)
2462
+ if (m_huff_num[i])
2463
+ {
2464
+ if (!m_pHuff_tabs[i])
2465
+ m_pHuff_tabs[i] = (huff_tables *)alloc(sizeof(huff_tables));
2466
+
2467
+ make_huff_table(i, m_pHuff_tabs[i]);
2468
+ }
2469
+ }
2470
+
2471
+ // Determines the component order inside each MCU.
2472
+ // Also calcs how many MCU's are on each row, etc.
2473
+ void jpeg_decoder::calc_mcu_block_order()
2474
+ {
2475
+ int component_num, component_id;
2476
+ int max_h_samp = 0, max_v_samp = 0;
2477
+
2478
+ for (component_id = 0; component_id < m_comps_in_frame; component_id++)
2479
+ {
2480
+ if (m_comp_h_samp[component_id] > max_h_samp)
2481
+ max_h_samp = m_comp_h_samp[component_id];
2482
+
2483
+ if (m_comp_v_samp[component_id] > max_v_samp)
2484
+ max_v_samp = m_comp_v_samp[component_id];
2485
+ }
2486
+
2487
+ for (component_id = 0; component_id < m_comps_in_frame; component_id++)
2488
+ {
2489
+ m_comp_h_blocks[component_id] = ((((m_image_x_size * m_comp_h_samp[component_id]) + (max_h_samp - 1)) / max_h_samp) + 7) / 8;
2490
+ m_comp_v_blocks[component_id] = ((((m_image_y_size * m_comp_v_samp[component_id]) + (max_v_samp - 1)) / max_v_samp) + 7) / 8;
2491
+ }
2492
+
2493
+ if (m_comps_in_scan == 1)
2494
+ {
2495
+ m_mcus_per_row = m_comp_h_blocks[m_comp_list[0]];
2496
+ m_mcus_per_col = m_comp_v_blocks[m_comp_list[0]];
2497
+ }
2498
+ else
2499
+ {
2500
+ m_mcus_per_row = (((m_image_x_size + 7) / 8) + (max_h_samp - 1)) / max_h_samp;
2501
+ m_mcus_per_col = (((m_image_y_size + 7) / 8) + (max_v_samp - 1)) / max_v_samp;
2502
+ }
2503
+
2504
+ if (m_comps_in_scan == 1)
2505
+ {
2506
+ m_mcu_org[0] = m_comp_list[0];
2507
+
2508
+ m_blocks_per_mcu = 1;
2509
+ }
2510
+ else
2511
+ {
2512
+ m_blocks_per_mcu = 0;
2513
+
2514
+ for (component_num = 0; component_num < m_comps_in_scan; component_num++)
2515
+ {
2516
+ int num_blocks;
2517
+
2518
+ component_id = m_comp_list[component_num];
2519
+
2520
+ num_blocks = m_comp_h_samp[component_id] * m_comp_v_samp[component_id];
2521
+
2522
+ while (num_blocks--)
2523
+ m_mcu_org[m_blocks_per_mcu++] = component_id;
2524
+ }
2525
+ }
2526
+ }
2527
+
2528
+ // Starts a new scan.
2529
+ int jpeg_decoder::init_scan()
2530
+ {
2531
+ if (!locate_sos_marker())
2532
+ return JPGD_FALSE;
2533
+
2534
+ calc_mcu_block_order();
2535
+
2536
+ check_huff_tables();
2537
+
2538
+ check_quant_tables();
2539
+
2540
+ memset(m_last_dc_val, 0, m_comps_in_frame * sizeof(uint));
2541
+
2542
+ m_eob_run = 0;
2543
+
2544
+ if (m_restart_interval)
2545
+ {
2546
+ m_restarts_left = m_restart_interval;
2547
+ m_next_restart_num = 0;
2548
+ }
2549
+
2550
+ fix_in_buffer();
2551
+
2552
+ return JPGD_TRUE;
2553
+ }
2554
+
2555
+ // Starts a frame. Determines if the number of components or sampling factors
2556
+ // are supported.
2557
+ void jpeg_decoder::init_frame()
2558
+ {
2559
+ int i;
2560
+
2561
+ if (m_comps_in_frame == 1)
2562
+ {
2563
+ if ((m_comp_h_samp[0] != 1) || (m_comp_v_samp[0] != 1))
2564
+ stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
2565
+
2566
+ m_scan_type = JPGD_GRAYSCALE;
2567
+ m_max_blocks_per_mcu = 1;
2568
+ m_max_mcu_x_size = 8;
2569
+ m_max_mcu_y_size = 8;
2570
+ }
2571
+ else if (m_comps_in_frame == 3)
2572
+ {
2573
+ if ( ((m_comp_h_samp[1] != 1) || (m_comp_v_samp[1] != 1)) ||
2574
+ ((m_comp_h_samp[2] != 1) || (m_comp_v_samp[2] != 1)) )
2575
+ stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
2576
+
2577
+ if ((m_comp_h_samp[0] == 1) && (m_comp_v_samp[0] == 1))
2578
+ {
2579
+ m_scan_type = JPGD_YH1V1;
2580
+
2581
+ m_max_blocks_per_mcu = 3;
2582
+ m_max_mcu_x_size = 8;
2583
+ m_max_mcu_y_size = 8;
2584
+ }
2585
+ else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 1))
2586
+ {
2587
+ m_scan_type = JPGD_YH2V1;
2588
+ m_max_blocks_per_mcu = 4;
2589
+ m_max_mcu_x_size = 16;
2590
+ m_max_mcu_y_size = 8;
2591
+ }
2592
+ else if ((m_comp_h_samp[0] == 1) && (m_comp_v_samp[0] == 2))
2593
+ {
2594
+ m_scan_type = JPGD_YH1V2;
2595
+ m_max_blocks_per_mcu = 4;
2596
+ m_max_mcu_x_size = 8;
2597
+ m_max_mcu_y_size = 16;
2598
+ }
2599
+ else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 2))
2600
+ {
2601
+ m_scan_type = JPGD_YH2V2;
2602
+ m_max_blocks_per_mcu = 6;
2603
+ m_max_mcu_x_size = 16;
2604
+ m_max_mcu_y_size = 16;
2605
+ }
2606
+ else
2607
+ stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
2608
+ }
2609
+ else
2610
+ stop_decoding(JPGD_UNSUPPORTED_COLORSPACE);
2611
+
2612
+ m_max_mcus_per_row = (m_image_x_size + (m_max_mcu_x_size - 1)) / m_max_mcu_x_size;
2613
+ m_max_mcus_per_col = (m_image_y_size + (m_max_mcu_y_size - 1)) / m_max_mcu_y_size;
2614
+
2615
+ // These values are for the *destination* pixels: after conversion.
2616
+ if (m_scan_type == JPGD_GRAYSCALE)
2617
+ m_dest_bytes_per_pixel = 1;
2618
+ else
2619
+ m_dest_bytes_per_pixel = 4;
2620
+
2621
+ m_dest_bytes_per_scan_line = ((m_image_x_size + 15) & 0xFFF0) * m_dest_bytes_per_pixel;
2622
+
2623
+ m_real_dest_bytes_per_scan_line = (m_image_x_size * m_dest_bytes_per_pixel);
2624
+
2625
+ // Initialize two scan line buffers.
2626
+ m_pScan_line_0 = (uint8 *)alloc(m_dest_bytes_per_scan_line, true);
2627
+ if ((m_scan_type == JPGD_YH1V2) || (m_scan_type == JPGD_YH2V2))
2628
+ m_pScan_line_1 = (uint8 *)alloc(m_dest_bytes_per_scan_line, true);
2629
+
2630
+ m_max_blocks_per_row = m_max_mcus_per_row * m_max_blocks_per_mcu;
2631
+
2632
+ // Should never happen
2633
+ if (m_max_blocks_per_row > JPGD_MAX_BLOCKS_PER_ROW)
2634
+ stop_decoding(JPGD_ASSERTION_ERROR);
2635
+
2636
+ // Allocate the coefficient buffer, enough for one MCU
2637
+ m_pMCU_coefficients = (jpgd_block_t*)alloc(m_max_blocks_per_mcu * 64 * sizeof(jpgd_block_t));
2638
+
2639
+ for (i = 0; i < m_max_blocks_per_mcu; i++)
2640
+ m_mcu_block_max_zag[i] = 64;
2641
+
2642
+ m_expanded_blocks_per_component = m_comp_h_samp[0] * m_comp_v_samp[0];
2643
+ m_expanded_blocks_per_mcu = m_expanded_blocks_per_component * m_comps_in_frame;
2644
+ m_expanded_blocks_per_row = m_max_mcus_per_row * m_expanded_blocks_per_mcu;
2645
+ // Freq. domain chroma upsampling is only supported for H2V2 subsampling factor.
2646
+ // BEGIN EPIC MOD
2647
+ #if JPGD_SUPPORT_FREQ_DOMAIN_UPSAMPLING
2648
+ m_freq_domain_chroma_upsample = (m_expanded_blocks_per_mcu == 4*3);
2649
+ #else
2650
+ m_freq_domain_chroma_upsample = 0;
2651
+ #endif
2652
+ // END EPIC MOD
2653
+
2654
+ if (m_freq_domain_chroma_upsample)
2655
+ m_pSample_buf = (uint8 *)alloc(m_expanded_blocks_per_row * 64);
2656
+ else
2657
+ m_pSample_buf = (uint8 *)alloc(m_max_blocks_per_row * 64);
2658
+
2659
+ m_total_lines_left = m_image_y_size;
2660
+
2661
+ m_mcu_lines_left = 0;
2662
+
2663
+ create_look_ups();
2664
+ }
2665
+
2666
+ // The coeff_buf series of methods originally stored the coefficients
2667
+ // into a "virtual" file which was located in EMS, XMS, or a disk file. A cache
2668
+ // was used to make this process more efficient. Now, we can store the entire
2669
+ // thing in RAM.
2670
+ jpeg_decoder::coeff_buf* jpeg_decoder::coeff_buf_open(int block_num_x, int block_num_y, int block_len_x, int block_len_y)
2671
+ {
2672
+ coeff_buf* cb = (coeff_buf*)alloc(sizeof(coeff_buf));
2673
+
2674
+ cb->block_num_x = block_num_x;
2675
+ cb->block_num_y = block_num_y;
2676
+ cb->block_len_x = block_len_x;
2677
+ cb->block_len_y = block_len_y;
2678
+ cb->block_size = (block_len_x * block_len_y) * sizeof(jpgd_block_t);
2679
+ cb->pData = (uint8 *)alloc(cb->block_size * block_num_x * block_num_y, true);
2680
+ return cb;
2681
+ }
2682
+
2683
+ inline jpgd_block_t *jpeg_decoder::coeff_buf_getp(coeff_buf *cb, int block_x, int block_y)
2684
+ {
2685
+ JPGD_ASSERT((block_x < cb->block_num_x) && (block_y < cb->block_num_y));
2686
+ return (jpgd_block_t *)(cb->pData + block_x * cb->block_size + block_y * (cb->block_size * cb->block_num_x));
2687
+ }
2688
+
2689
+ // The following methods decode the various types of m_blocks encountered
2690
+ // in progressively encoded images.
2691
+ void jpeg_decoder::decode_block_dc_first(jpeg_decoder *pD, int component_id, int block_x, int block_y)
2692
+ {
2693
+ int s, r;
2694
+ jpgd_block_t *p = pD->coeff_buf_getp(pD->m_dc_coeffs[component_id], block_x, block_y);
2695
+
2696
+ if ((s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_dc_tab[component_id]])) != 0)
2697
+ {
2698
+ r = pD->get_bits_no_markers(s);
2699
+ s = HUFF_EXTEND(r, s);
2700
+ }
2701
+
2702
+ pD->m_last_dc_val[component_id] = (s += pD->m_last_dc_val[component_id]);
2703
+
2704
+ p[0] = static_cast<jpgd_block_t>(s << pD->m_successive_low);
2705
+ }
2706
+
2707
+ void jpeg_decoder::decode_block_dc_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y)
2708
+ {
2709
+ if (pD->get_bits_no_markers(1))
2710
+ {
2711
+ jpgd_block_t *p = pD->coeff_buf_getp(pD->m_dc_coeffs[component_id], block_x, block_y);
2712
+
2713
+ p[0] |= (1 << pD->m_successive_low);
2714
+ }
2715
+ }
2716
+
2717
+ void jpeg_decoder::decode_block_ac_first(jpeg_decoder *pD, int component_id, int block_x, int block_y)
2718
+ {
2719
+ int k, s, r;
2720
+
2721
+ if (pD->m_eob_run)
2722
+ {
2723
+ pD->m_eob_run--;
2724
+ return;
2725
+ }
2726
+
2727
+ jpgd_block_t *p = pD->coeff_buf_getp(pD->m_ac_coeffs[component_id], block_x, block_y);
2728
+
2729
+ for (k = pD->m_spectral_start; k <= pD->m_spectral_end; k++)
2730
+ {
2731
+ s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_ac_tab[component_id]]);
2732
+
2733
+ r = s >> 4;
2734
+ s &= 15;
2735
+
2736
+ if (s)
2737
+ {
2738
+ if ((k += r) > 63)
2739
+ pD->stop_decoding(JPGD_DECODE_ERROR);
2740
+
2741
+ r = pD->get_bits_no_markers(s);
2742
+ s = HUFF_EXTEND(r, s);
2743
+
2744
+ p[g_ZAG[k]] = static_cast<jpgd_block_t>(s << pD->m_successive_low);
2745
+ }
2746
+ else
2747
+ {
2748
+ if (r == 15)
2749
+ {
2750
+ if ((k += 15) > 63)
2751
+ pD->stop_decoding(JPGD_DECODE_ERROR);
2752
+ }
2753
+ else
2754
+ {
2755
+ pD->m_eob_run = 1 << r;
2756
+
2757
+ if (r)
2758
+ pD->m_eob_run += pD->get_bits_no_markers(r);
2759
+
2760
+ pD->m_eob_run--;
2761
+
2762
+ break;
2763
+ }
2764
+ }
2765
+ }
2766
+ }
2767
+
2768
+ void jpeg_decoder::decode_block_ac_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y)
2769
+ {
2770
+ int s, k, r;
2771
+ int p1 = 1 << pD->m_successive_low;
2772
+ int m1 = (-1) << pD->m_successive_low;
2773
+ jpgd_block_t *p = pD->coeff_buf_getp(pD->m_ac_coeffs[component_id], block_x, block_y);
2774
+
2775
+ k = pD->m_spectral_start;
2776
+
2777
+ if (pD->m_eob_run == 0)
2778
+ {
2779
+ for ( ; k <= pD->m_spectral_end; k++)
2780
+ {
2781
+ s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_ac_tab[component_id]]);
2782
+
2783
+ r = s >> 4;
2784
+ s &= 15;
2785
+
2786
+ if (s)
2787
+ {
2788
+ if (s != 1)
2789
+ pD->stop_decoding(JPGD_DECODE_ERROR);
2790
+
2791
+ if (pD->get_bits_no_markers(1))
2792
+ s = p1;
2793
+ else
2794
+ s = m1;
2795
+ }
2796
+ else
2797
+ {
2798
+ if (r != 15)
2799
+ {
2800
+ pD->m_eob_run = 1 << r;
2801
+
2802
+ if (r)
2803
+ pD->m_eob_run += pD->get_bits_no_markers(r);
2804
+
2805
+ break;
2806
+ }
2807
+ }
2808
+
2809
+ do
2810
+ {
2811
+ // BEGIN EPIC MOD
2812
+ JPGD_ASSERT(k < 64);
2813
+ // END EPIC MOD
2814
+
2815
+ jpgd_block_t *this_coef = p + g_ZAG[k];
2816
+
2817
+ if (*this_coef != 0)
2818
+ {
2819
+ if (pD->get_bits_no_markers(1))
2820
+ {
2821
+ if ((*this_coef & p1) == 0)
2822
+ {
2823
+ if (*this_coef >= 0)
2824
+ *this_coef = static_cast<jpgd_block_t>(*this_coef + p1);
2825
+ else
2826
+ *this_coef = static_cast<jpgd_block_t>(*this_coef + m1);
2827
+ }
2828
+ }
2829
+ }
2830
+ else
2831
+ {
2832
+ if (--r < 0)
2833
+ break;
2834
+ }
2835
+
2836
+ k++;
2837
+
2838
+ } while (k <= pD->m_spectral_end);
2839
+
2840
+ if ((s) && (k < 64))
2841
+ {
2842
+ p[g_ZAG[k]] = static_cast<jpgd_block_t>(s);
2843
+ }
2844
+ }
2845
+ }
2846
+
2847
+ if (pD->m_eob_run > 0)
2848
+ {
2849
+ for ( ; k <= pD->m_spectral_end; k++)
2850
+ {
2851
+ // BEGIN EPIC MOD
2852
+ JPGD_ASSERT(k < 64);
2853
+ // END EPIC MOD
2854
+
2855
+ jpgd_block_t *this_coef = p + g_ZAG[k];
2856
+
2857
+ if (*this_coef != 0)
2858
+ {
2859
+ if (pD->get_bits_no_markers(1))
2860
+ {
2861
+ if ((*this_coef & p1) == 0)
2862
+ {
2863
+ if (*this_coef >= 0)
2864
+ *this_coef = static_cast<jpgd_block_t>(*this_coef + p1);
2865
+ else
2866
+ *this_coef = static_cast<jpgd_block_t>(*this_coef + m1);
2867
+ }
2868
+ }
2869
+ }
2870
+ }
2871
+
2872
+ pD->m_eob_run--;
2873
+ }
2874
+ }
2875
+
2876
+ // Decode a scan in a progressively encoded image.
2877
+ void jpeg_decoder::decode_scan(pDecode_block_func decode_block_func)
2878
+ {
2879
+ int mcu_row, mcu_col, mcu_block;
2880
+ int block_x_mcu[JPGD_MAX_COMPONENTS], m_block_y_mcu[JPGD_MAX_COMPONENTS];
2881
+
2882
+ memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
2883
+
2884
+ for (mcu_col = 0; mcu_col < m_mcus_per_col; mcu_col++)
2885
+ {
2886
+ int component_num, component_id;
2887
+
2888
+ memset(block_x_mcu, 0, sizeof(block_x_mcu));
2889
+
2890
+ for (mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++)
2891
+ {
2892
+ int block_x_mcu_ofs = 0, block_y_mcu_ofs = 0;
2893
+
2894
+ if ((m_restart_interval) && (m_restarts_left == 0))
2895
+ process_restart();
2896
+
2897
+ for (mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++)
2898
+ {
2899
+ component_id = m_mcu_org[mcu_block];
2900
+
2901
+ decode_block_func(this, component_id, block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
2902
+
2903
+ if (m_comps_in_scan == 1)
2904
+ block_x_mcu[component_id]++;
2905
+ else
2906
+ {
2907
+ if (++block_x_mcu_ofs == m_comp_h_samp[component_id])
2908
+ {
2909
+ block_x_mcu_ofs = 0;
2910
+
2911
+ if (++block_y_mcu_ofs == m_comp_v_samp[component_id])
2912
+ {
2913
+ block_y_mcu_ofs = 0;
2914
+ block_x_mcu[component_id] += m_comp_h_samp[component_id];
2915
+ }
2916
+ }
2917
+ }
2918
+ }
2919
+
2920
+ m_restarts_left--;
2921
+ }
2922
+
2923
+ if (m_comps_in_scan == 1)
2924
+ m_block_y_mcu[m_comp_list[0]]++;
2925
+ else
2926
+ {
2927
+ for (component_num = 0; component_num < m_comps_in_scan; component_num++)
2928
+ {
2929
+ component_id = m_comp_list[component_num];
2930
+ m_block_y_mcu[component_id] += m_comp_v_samp[component_id];
2931
+ }
2932
+ }
2933
+ }
2934
+ }
2935
+
2936
+ // Decode a progressively encoded image.
2937
+ void jpeg_decoder::init_progressive()
2938
+ {
2939
+ int i;
2940
+
2941
+ if (m_comps_in_frame == 4)
2942
+ stop_decoding(JPGD_UNSUPPORTED_COLORSPACE);
2943
+
2944
+ // Allocate the coefficient buffers.
2945
+ for (i = 0; i < m_comps_in_frame; i++)
2946
+ {
2947
+ m_dc_coeffs[i] = coeff_buf_open(m_max_mcus_per_row * m_comp_h_samp[i], m_max_mcus_per_col * m_comp_v_samp[i], 1, 1);
2948
+ m_ac_coeffs[i] = coeff_buf_open(m_max_mcus_per_row * m_comp_h_samp[i], m_max_mcus_per_col * m_comp_v_samp[i], 8, 8);
2949
+ }
2950
+
2951
+ for ( ; ; )
2952
+ {
2953
+ int dc_only_scan, refinement_scan;
2954
+ pDecode_block_func decode_block_func;
2955
+
2956
+ if (!init_scan())
2957
+ break;
2958
+
2959
+ dc_only_scan = (m_spectral_start == 0);
2960
+ refinement_scan = (m_successive_high != 0);
2961
+
2962
+ if ((m_spectral_start > m_spectral_end) || (m_spectral_end > 63))
2963
+ stop_decoding(JPGD_BAD_SOS_SPECTRAL);
2964
+
2965
+ if (dc_only_scan)
2966
+ {
2967
+ if (m_spectral_end)
2968
+ stop_decoding(JPGD_BAD_SOS_SPECTRAL);
2969
+ }
2970
+ else if (m_comps_in_scan != 1) /* AC scans can only contain one component */
2971
+ stop_decoding(JPGD_BAD_SOS_SPECTRAL);
2972
+
2973
+ if ((refinement_scan) && (m_successive_low != m_successive_high - 1))
2974
+ stop_decoding(JPGD_BAD_SOS_SUCCESSIVE);
2975
+
2976
+ if (dc_only_scan)
2977
+ {
2978
+ if (refinement_scan)
2979
+ decode_block_func = decode_block_dc_refine;
2980
+ else
2981
+ decode_block_func = decode_block_dc_first;
2982
+ }
2983
+ else
2984
+ {
2985
+ if (refinement_scan)
2986
+ decode_block_func = decode_block_ac_refine;
2987
+ else
2988
+ decode_block_func = decode_block_ac_first;
2989
+ }
2990
+
2991
+ decode_scan(decode_block_func);
2992
+
2993
+ m_bits_left = 16;
2994
+ get_bits(16);
2995
+ get_bits(16);
2996
+ }
2997
+
2998
+ m_comps_in_scan = m_comps_in_frame;
2999
+
3000
+ for (i = 0; i < m_comps_in_frame; i++)
3001
+ m_comp_list[i] = i;
3002
+
3003
+ calc_mcu_block_order();
3004
+ }
3005
+
3006
+ void jpeg_decoder::init_sequential()
3007
+ {
3008
+ if (!init_scan())
3009
+ stop_decoding(JPGD_UNEXPECTED_MARKER);
3010
+ }
3011
+
3012
+ void jpeg_decoder::decode_start()
3013
+ {
3014
+ init_frame();
3015
+
3016
+ if (m_progressive_flag)
3017
+ init_progressive();
3018
+ else
3019
+ init_sequential();
3020
+ }
3021
+
3022
+ void jpeg_decoder::decode_init(jpeg_decoder_stream *pStream)
3023
+ {
3024
+ init(pStream);
3025
+ locate_sof_marker();
3026
+ }
3027
+
3028
+ jpeg_decoder::jpeg_decoder(jpeg_decoder_stream *pStream)
3029
+ {
3030
+ if (setjmp(m_jmp_state))
3031
+ return;
3032
+ decode_init(pStream);
3033
+ }
3034
+
3035
+ int jpeg_decoder::begin_decoding()
3036
+ {
3037
+ if (m_ready_flag)
3038
+ return JPGD_SUCCESS;
3039
+
3040
+ if (m_error_code)
3041
+ return JPGD_FAILED;
3042
+
3043
+ if (setjmp(m_jmp_state))
3044
+ return JPGD_FAILED;
3045
+
3046
+ decode_start();
3047
+
3048
+ m_ready_flag = true;
3049
+
3050
+ return JPGD_SUCCESS;
3051
+ }
3052
+
3053
+ jpeg_decoder::~jpeg_decoder()
3054
+ {
3055
+ free_all_blocks();
3056
+ }
3057
+
3058
+ jpeg_decoder_file_stream::jpeg_decoder_file_stream()
3059
+ {
3060
+ m_pFile = NULL;
3061
+ m_eof_flag = false;
3062
+ m_error_flag = false;
3063
+ }
3064
+
3065
+ void jpeg_decoder_file_stream::close()
3066
+ {
3067
+ if (m_pFile)
3068
+ {
3069
+ fclose(m_pFile);
3070
+ m_pFile = NULL;
3071
+ }
3072
+
3073
+ m_eof_flag = false;
3074
+ m_error_flag = false;
3075
+ }
3076
+
3077
+ jpeg_decoder_file_stream::~jpeg_decoder_file_stream()
3078
+ {
3079
+ close();
3080
+ }
3081
+
3082
+ bool jpeg_decoder_file_stream::open(const char *Pfilename)
3083
+ {
3084
+ close();
3085
+
3086
+ m_eof_flag = false;
3087
+ m_error_flag = false;
3088
+
3089
+ #if defined(_MSC_VER)
3090
+ m_pFile = NULL;
3091
+ fopen_s(&m_pFile, Pfilename, "rb");
3092
+ #else
3093
+ m_pFile = fopen(Pfilename, "rb");
3094
+ #endif
3095
+ return m_pFile != NULL;
3096
+ }
3097
+
3098
+ int jpeg_decoder_file_stream::read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag)
3099
+ {
3100
+ if (!m_pFile)
3101
+ return -1;
3102
+
3103
+ if (m_eof_flag)
3104
+ {
3105
+ *pEOF_flag = true;
3106
+ return 0;
3107
+ }
3108
+
3109
+ if (m_error_flag)
3110
+ return -1;
3111
+
3112
+ int bytes_read = static_cast<int>(fread(pBuf, 1, max_bytes_to_read, m_pFile));
3113
+ if (bytes_read < max_bytes_to_read)
3114
+ {
3115
+ if (ferror(m_pFile))
3116
+ {
3117
+ m_error_flag = true;
3118
+ return -1;
3119
+ }
3120
+
3121
+ m_eof_flag = true;
3122
+ *pEOF_flag = true;
3123
+ }
3124
+
3125
+ return bytes_read;
3126
+ }
3127
+
3128
+ bool jpeg_decoder_mem_stream::open(const uint8 *pSrc_data, uint size)
3129
+ {
3130
+ close();
3131
+ m_pSrc_data = pSrc_data;
3132
+ m_ofs = 0;
3133
+ m_size = size;
3134
+ return true;
3135
+ }
3136
+
3137
+ int jpeg_decoder_mem_stream::read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag)
3138
+ {
3139
+ *pEOF_flag = false;
3140
+
3141
+ if (!m_pSrc_data)
3142
+ return -1;
3143
+
3144
+ uint bytes_remaining = m_size - m_ofs;
3145
+ if ((uint)max_bytes_to_read > bytes_remaining)
3146
+ {
3147
+ max_bytes_to_read = bytes_remaining;
3148
+ *pEOF_flag = true;
3149
+ }
3150
+
3151
+ memcpy(pBuf, m_pSrc_data + m_ofs, max_bytes_to_read);
3152
+ m_ofs += max_bytes_to_read;
3153
+
3154
+ return max_bytes_to_read;
3155
+ }
3156
+
3157
+ unsigned char *decompress_jpeg_image_from_stream(jpeg_decoder_stream *pStream, int *width, int *height, int *actual_comps, int req_comps)
3158
+ {
3159
+ if (!actual_comps)
3160
+ return NULL;
3161
+ *actual_comps = 0;
3162
+
3163
+ if ((!pStream) || (!width) || (!height) || (!req_comps))
3164
+ return NULL;
3165
+
3166
+ if ((req_comps != 1) && (req_comps != 3) && (req_comps != 4))
3167
+ return NULL;
3168
+
3169
+ jpeg_decoder decoder(pStream);
3170
+ if (decoder.get_error_code() != JPGD_SUCCESS)
3171
+ return NULL;
3172
+
3173
+ const int image_width = decoder.get_width(), image_height = decoder.get_height();
3174
+ *width = image_width;
3175
+ *height = image_height;
3176
+ *actual_comps = decoder.get_num_components();
3177
+
3178
+ if (decoder.begin_decoding() != JPGD_SUCCESS)
3179
+ return NULL;
3180
+
3181
+ const int dst_bpl = image_width * req_comps;
3182
+
3183
+ uint8 *pImage_data = (uint8*)jpgd_malloc(dst_bpl * image_height);
3184
+ if (!pImage_data)
3185
+ return NULL;
3186
+
3187
+ for (int y = 0; y < image_height; y++)
3188
+ {
3189
+ const uint8* pScan_line = 0;
3190
+ uint scan_line_len;
3191
+ if (decoder.decode((const void**)&pScan_line, &scan_line_len) != JPGD_SUCCESS)
3192
+ {
3193
+ jpgd_free(pImage_data);
3194
+ return NULL;
3195
+ }
3196
+
3197
+ uint8 *pDst = pImage_data + y * dst_bpl;
3198
+
3199
+ if (((req_comps == 4) && (decoder.get_num_components() == 3)) ||
3200
+ ((req_comps == 1) && (decoder.get_num_components() == 1)))
3201
+ {
3202
+ memcpy(pDst, pScan_line, dst_bpl);
3203
+ }
3204
+ else if (decoder.get_num_components() == 1)
3205
+ {
3206
+ if (req_comps == 3)
3207
+ {
3208
+ for (int x = 0; x < image_width; x++)
3209
+ {
3210
+ uint8 luma = pScan_line[x];
3211
+ pDst[0] = luma;
3212
+ pDst[1] = luma;
3213
+ pDst[2] = luma;
3214
+ pDst += 3;
3215
+ }
3216
+ }
3217
+ else
3218
+ {
3219
+ for (int x = 0; x < image_width; x++)
3220
+ {
3221
+ uint8 luma = pScan_line[x];
3222
+ pDst[0] = luma;
3223
+ pDst[1] = luma;
3224
+ pDst[2] = luma;
3225
+ pDst[3] = 255;
3226
+ pDst += 4;
3227
+ }
3228
+ }
3229
+ }
3230
+ else if (decoder.get_num_components() == 3)
3231
+ {
3232
+ if (req_comps == 1)
3233
+ {
3234
+ const int YR = 19595, YG = 38470, YB = 7471;
3235
+ for (int x = 0; x < image_width; x++)
3236
+ {
3237
+ int r = pScan_line[x*4+0];
3238
+ int g = pScan_line[x*4+1];
3239
+ int b = pScan_line[x*4+2];
3240
+ *pDst++ = static_cast<uint8>((r * YR + g * YG + b * YB + 32768) >> 16);
3241
+ }
3242
+ }
3243
+ else
3244
+ {
3245
+ for (int x = 0; x < image_width; x++)
3246
+ {
3247
+ pDst[0] = pScan_line[x*4+0];
3248
+ pDst[1] = pScan_line[x*4+1];
3249
+ pDst[2] = pScan_line[x*4+2];
3250
+ pDst += 3;
3251
+ }
3252
+ }
3253
+ }
3254
+ }
3255
+
3256
+ return pImage_data;
3257
+ }
3258
+
3259
+ // BEGIN EPIC MOD
3260
+ unsigned char *decompress_jpeg_image_from_memory(const unsigned char *pSrc_data, int src_data_size, int *width, int *height, int *actual_comps, int req_comps, int format)
3261
+ {
3262
+ jpg_format = (ERGBFormatJPG)format;
3263
+ // EMD EPIC MOD
3264
+ jpgd::jpeg_decoder_mem_stream mem_stream(pSrc_data, src_data_size);
3265
+ return decompress_jpeg_image_from_stream(&mem_stream, width, height, actual_comps, req_comps);
3266
+ }
3267
+
3268
+ unsigned char *decompress_jpeg_image_from_file(const char *pSrc_filename, int *width, int *height, int *actual_comps, int req_comps)
3269
+ {
3270
+ jpgd::jpeg_decoder_file_stream file_stream;
3271
+ if (!file_stream.open(pSrc_filename))
3272
+ return NULL;
3273
+ return decompress_jpeg_image_from_stream(&file_stream, width, height, actual_comps, req_comps);
3274
+ }
3275
+
3276
+ } // namespace jpgd
crazy_functions/test_project/cpp/longcode/jpge.cpp ADDED
@@ -0,0 +1,1049 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ // jpge.cpp - C++ class for JPEG compression.
2
+ // Public domain, Rich Geldreich <richgel99@gmail.com>
3
+ // v1.01, Dec. 18, 2010 - Initial release
4
+ // v1.02, Apr. 6, 2011 - Removed 2x2 ordered dither in H2V1 chroma subsampling method load_block_16_8_8(). (The rounding factor was 2, when it should have been 1. Either way, it wasn't helping.)
5
+ // v1.03, Apr. 16, 2011 - Added support for optimized Huffman code tables, optimized dynamic memory allocation down to only 1 alloc.
6
+ // Also from Alex Evans: Added RGBA support, linear memory allocator (no longer needed in v1.03).
7
+ // v1.04, May. 19, 2012: Forgot to set m_pFile ptr to NULL in cfile_stream::close(). Thanks to Owen Kaluza for reporting this bug.
8
+ // Code tweaks to fix VS2008 static code analysis warnings (all looked harmless).
9
+ // Code review revealed method load_block_16_8_8() (used for the non-default H2V1 sampling mode to downsample chroma) somehow didn't get the rounding factor fix from v1.02.
10
+
11
+ #include "jpge.h"
12
+
13
+ #include <stdlib.h>
14
+ #include <string.h>
15
+ #if PLATFORM_WINDOWS
16
+ #include <malloc.h>
17
+ #endif
18
+
19
+ #define JPGE_MAX(a,b) (((a)>(b))?(a):(b))
20
+ #define JPGE_MIN(a,b) (((a)<(b))?(a):(b))
21
+
22
+ namespace jpge {
23
+
24
+ static inline void *jpge_malloc(size_t nSize) { return FMemory::Malloc(nSize); }
25
+ static inline void jpge_free(void *p) { FMemory::Free(p);; }
26
+
27
+ // Various JPEG enums and tables.
28
+ enum { M_SOF0 = 0xC0, M_DHT = 0xC4, M_SOI = 0xD8, M_EOI = 0xD9, M_SOS = 0xDA, M_DQT = 0xDB, M_APP0 = 0xE0 };
29
+ enum { DC_LUM_CODES = 12, AC_LUM_CODES = 256, DC_CHROMA_CODES = 12, AC_CHROMA_CODES = 256, MAX_HUFF_SYMBOLS = 257, MAX_HUFF_CODESIZE = 32 };
30
+
31
+ static uint8 s_zag[64] = { 0,1,8,16,9,2,3,10,17,24,32,25,18,11,4,5,12,19,26,33,40,48,41,34,27,20,13,6,7,14,21,28,35,42,49,56,57,50,43,36,29,22,15,23,30,37,44,51,58,59,52,45,38,31,39,46,53,60,61,54,47,55,62,63 };
32
+ static int16 s_std_lum_quant[64] = { 16,11,12,14,12,10,16,14,13,14,18,17,16,19,24,40,26,24,22,22,24,49,35,37,29,40,58,51,61,60,57,51,56,55,64,72,92,78,64,68,87,69,55,56,80,109,81,87,95,98,103,104,103,62,77,113,121,112,100,120,92,101,103,99 };
33
+ static int16 s_std_croma_quant[64] = { 17,18,18,24,21,24,47,26,26,47,99,66,56,66,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99 };
34
+ static uint8 s_dc_lum_bits[17] = { 0,0,1,5,1,1,1,1,1,1,0,0,0,0,0,0,0 };
35
+ static uint8 s_dc_lum_val[DC_LUM_CODES] = { 0,1,2,3,4,5,6,7,8,9,10,11 };
36
+ static uint8 s_ac_lum_bits[17] = { 0,0,2,1,3,3,2,4,3,5,5,4,4,0,0,1,0x7d };
37
+ static uint8 s_ac_lum_val[AC_LUM_CODES] =
38
+ {
39
+ 0x01,0x02,0x03,0x00,0x04,0x11,0x05,0x12,0x21,0x31,0x41,0x06,0x13,0x51,0x61,0x07,0x22,0x71,0x14,0x32,0x81,0x91,0xa1,0x08,0x23,0x42,0xb1,0xc1,0x15,0x52,0xd1,0xf0,
40
+ 0x24,0x33,0x62,0x72,0x82,0x09,0x0a,0x16,0x17,0x18,0x19,0x1a,0x25,0x26,0x27,0x28,0x29,0x2a,0x34,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,0x49,
41
+ 0x4a,0x53,0x54,0x55,0x56,0x57,0x58,0x59,0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x83,0x84,0x85,0x86,0x87,0x88,0x89,
42
+ 0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,0xb5,0xb6,0xb7,0xb8,0xb9,0xba,0xc2,0xc3,0xc4,0xc5,
43
+ 0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,0xe1,0xe2,0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf1,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,
44
+ 0xf9,0xfa
45
+ };
46
+ static uint8 s_dc_chroma_bits[17] = { 0,0,3,1,1,1,1,1,1,1,1,1,0,0,0,0,0 };
47
+ static uint8 s_dc_chroma_val[DC_CHROMA_CODES] = { 0,1,2,3,4,5,6,7,8,9,10,11 };
48
+ static uint8 s_ac_chroma_bits[17] = { 0,0,2,1,2,4,4,3,4,7,5,4,4,0,1,2,0x77 };
49
+ static uint8 s_ac_chroma_val[AC_CHROMA_CODES] =
50
+ {
51
+ 0x00,0x01,0x02,0x03,0x11,0x04,0x05,0x21,0x31,0x06,0x12,0x41,0x51,0x07,0x61,0x71,0x13,0x22,0x32,0x81,0x08,0x14,0x42,0x91,0xa1,0xb1,0xc1,0x09,0x23,0x33,0x52,0xf0,
52
+ 0x15,0x62,0x72,0xd1,0x0a,0x16,0x24,0x34,0xe1,0x25,0xf1,0x17,0x18,0x19,0x1a,0x26,0x27,0x28,0x29,0x2a,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,
53
+ 0x49,0x4a,0x53,0x54,0x55,0x56,0x57,0x58,0x59,0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x82,0x83,0x84,0x85,0x86,0x87,
54
+ 0x88,0x89,0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,0xb5,0xb6,0xb7,0xb8,0xb9,0xba,0xc2,0xc3,
55
+ 0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,0xe2,0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,
56
+ 0xf9,0xfa
57
+ };
58
+
59
+ // Low-level helper functions.
60
+ template <class T> inline void clear_obj(T &obj) { memset(&obj, 0, sizeof(obj)); }
61
+
62
+ const int YR = 19595, YG = 38470, YB = 7471, CB_R = -11059, CB_G = -21709, CB_B = 32768, CR_R = 32768, CR_G = -27439, CR_B = -5329;
63
+ static inline uint8 clamp(int i) { if (static_cast<uint>(i) > 255U) { if (i < 0) i = 0; else if (i > 255) i = 255; } return static_cast<uint8>(i); }
64
+
65
+ static void RGB_to_YCC(uint8* pDst, const uint8 *pSrc, int num_pixels)
66
+ {
67
+ for ( ; num_pixels; pDst += 3, pSrc += 3, num_pixels--)
68
+ {
69
+ const int r = pSrc[0], g = pSrc[1], b = pSrc[2];
70
+ pDst[0] = static_cast<uint8>((r * YR + g * YG + b * YB + 32768) >> 16);
71
+ pDst[1] = clamp(128 + ((r * CB_R + g * CB_G + b * CB_B + 32768) >> 16));
72
+ pDst[2] = clamp(128 + ((r * CR_R + g * CR_G + b * CR_B + 32768) >> 16));
73
+ }
74
+ }
75
+
76
+ static void RGB_to_Y(uint8* pDst, const uint8 *pSrc, int num_pixels)
77
+ {
78
+ for ( ; num_pixels; pDst++, pSrc += 3, num_pixels--)
79
+ pDst[0] = static_cast<uint8>((pSrc[0] * YR + pSrc[1] * YG + pSrc[2] * YB + 32768) >> 16);
80
+ }
81
+
82
+ static void RGBA_to_YCC(uint8* pDst, const uint8 *pSrc, int num_pixels)
83
+ {
84
+ for ( ; num_pixels; pDst += 3, pSrc += 4, num_pixels--)
85
+ {
86
+ const int r = pSrc[0], g = pSrc[1], b = pSrc[2];
87
+ pDst[0] = static_cast<uint8>((r * YR + g * YG + b * YB + 32768) >> 16);
88
+ pDst[1] = clamp(128 + ((r * CB_R + g * CB_G + b * CB_B + 32768) >> 16));
89
+ pDst[2] = clamp(128 + ((r * CR_R + g * CR_G + b * CR_B + 32768) >> 16));
90
+ }
91
+ }
92
+
93
+ static void RGBA_to_Y(uint8* pDst, const uint8 *pSrc, int num_pixels)
94
+ {
95
+ for ( ; num_pixels; pDst++, pSrc += 4, num_pixels--)
96
+ pDst[0] = static_cast<uint8>((pSrc[0] * YR + pSrc[1] * YG + pSrc[2] * YB + 32768) >> 16);
97
+ }
98
+
99
+ static void Y_to_YCC(uint8* pDst, const uint8* pSrc, int num_pixels)
100
+ {
101
+ for( ; num_pixels; pDst += 3, pSrc++, num_pixels--) { pDst[0] = pSrc[0]; pDst[1] = 128; pDst[2] = 128; }
102
+ }
103
+
104
+ // Forward DCT - DCT derived from jfdctint.
105
+ #define CONST_BITS 13
106
+ #define ROW_BITS 2
107
+ #define DCT_DESCALE(x, n) (((x) + (((int32)1) << ((n) - 1))) >> (n))
108
+ #define DCT_MUL(var, c) (static_cast<int16>(var) * static_cast<int32>(c))
109
+ #define DCT1D(s0, s1, s2, s3, s4, s5, s6, s7) \
110
+ int32 t0 = s0 + s7, t7 = s0 - s7, t1 = s1 + s6, t6 = s1 - s6, t2 = s2 + s5, t5 = s2 - s5, t3 = s3 + s4, t4 = s3 - s4; \
111
+ int32 t10 = t0 + t3, t13 = t0 - t3, t11 = t1 + t2, t12 = t1 - t2; \
112
+ int32 u1 = DCT_MUL(t12 + t13, 4433); \
113
+ s2 = u1 + DCT_MUL(t13, 6270); \
114
+ s6 = u1 + DCT_MUL(t12, -15137); \
115
+ u1 = t4 + t7; \
116
+ int32 u2 = t5 + t6, u3 = t4 + t6, u4 = t5 + t7; \
117
+ int32 z5 = DCT_MUL(u3 + u4, 9633); \
118
+ t4 = DCT_MUL(t4, 2446); t5 = DCT_MUL(t5, 16819); \
119
+ t6 = DCT_MUL(t6, 25172); t7 = DCT_MUL(t7, 12299); \
120
+ u1 = DCT_MUL(u1, -7373); u2 = DCT_MUL(u2, -20995); \
121
+ u3 = DCT_MUL(u3, -16069); u4 = DCT_MUL(u4, -3196); \
122
+ u3 += z5; u4 += z5; \
123
+ s0 = t10 + t11; s1 = t7 + u1 + u4; s3 = t6 + u2 + u3; s4 = t10 - t11; s5 = t5 + u2 + u4; s7 = t4 + u1 + u3;
124
+
125
+ static void DCT2D(int32 *p)
126
+ {
127
+ int32 c, *q = p;
128
+ for (c = 7; c >= 0; c--, q += 8)
129
+ {
130
+ int32 s0 = q[0], s1 = q[1], s2 = q[2], s3 = q[3], s4 = q[4], s5 = q[5], s6 = q[6], s7 = q[7];
131
+ DCT1D(s0, s1, s2, s3, s4, s5, s6, s7);
132
+ q[0] = s0 << ROW_BITS; q[1] = DCT_DESCALE(s1, CONST_BITS-ROW_BITS); q[2] = DCT_DESCALE(s2, CONST_BITS-ROW_BITS); q[3] = DCT_DESCALE(s3, CONST_BITS-ROW_BITS);
133
+ q[4] = s4 << ROW_BITS; q[5] = DCT_DESCALE(s5, CONST_BITS-ROW_BITS); q[6] = DCT_DESCALE(s6, CONST_BITS-ROW_BITS); q[7] = DCT_DESCALE(s7, CONST_BITS-ROW_BITS);
134
+ }
135
+ for (q = p, c = 7; c >= 0; c--, q++)
136
+ {
137
+ int32 s0 = q[0*8], s1 = q[1*8], s2 = q[2*8], s3 = q[3*8], s4 = q[4*8], s5 = q[5*8], s6 = q[6*8], s7 = q[7*8];
138
+ DCT1D(s0, s1, s2, s3, s4, s5, s6, s7);
139
+ q[0*8] = DCT_DESCALE(s0, ROW_BITS+3); q[1*8] = DCT_DESCALE(s1, CONST_BITS+ROW_BITS+3); q[2*8] = DCT_DESCALE(s2, CONST_BITS+ROW_BITS+3); q[3*8] = DCT_DESCALE(s3, CONST_BITS+ROW_BITS+3);
140
+ q[4*8] = DCT_DESCALE(s4, ROW_BITS+3); q[5*8] = DCT_DESCALE(s5, CONST_BITS+ROW_BITS+3); q[6*8] = DCT_DESCALE(s6, CONST_BITS+ROW_BITS+3); q[7*8] = DCT_DESCALE(s7, CONST_BITS+ROW_BITS+3);
141
+ }
142
+ }
143
+
144
+ struct sym_freq { uint m_key, m_sym_index; };
145
+
146
+ // Radix sorts sym_freq[] array by 32-bit key m_key. Returns ptr to sorted values.
147
+ static inline sym_freq* radix_sort_syms(uint num_syms, sym_freq* pSyms0, sym_freq* pSyms1)
148
+ {
149
+ const uint cMaxPasses = 4;
150
+ uint32 hist[256 * cMaxPasses]; clear_obj(hist);
151
+ for (uint i = 0; i < num_syms; i++) { uint freq = pSyms0[i].m_key; hist[freq & 0xFF]++; hist[256 + ((freq >> 8) & 0xFF)]++; hist[256*2 + ((freq >> 16) & 0xFF)]++; hist[256*3 + ((freq >> 24) & 0xFF)]++; }
152
+ sym_freq* pCur_syms = pSyms0, *pNew_syms = pSyms1;
153
+ uint total_passes = cMaxPasses; while ((total_passes > 1) && (num_syms == hist[(total_passes - 1) * 256])) total_passes--;
154
+ for (uint pass_shift = 0, pass = 0; pass < total_passes; pass++, pass_shift += 8)
155
+ {
156
+ const uint32* pHist = &hist[pass << 8];
157
+ uint offsets[256], cur_ofs = 0;
158
+ for (uint i = 0; i < 256; i++) { offsets[i] = cur_ofs; cur_ofs += pHist[i]; }
159
+ for (uint i = 0; i < num_syms; i++)
160
+ pNew_syms[offsets[(pCur_syms[i].m_key >> pass_shift) & 0xFF]++] = pCur_syms[i];
161
+ sym_freq* t = pCur_syms; pCur_syms = pNew_syms; pNew_syms = t;
162
+ }
163
+ return pCur_syms;
164
+ }
165
+
166
+ // calculate_minimum_redundancy() originally written by: Alistair Moffat, alistair@cs.mu.oz.au, Jyrki Katajainen, jyrki@diku.dk, November 1996.
167
+ static void calculate_minimum_redundancy(sym_freq *A, int n)
168
+ {
169
+ int root, leaf, next, avbl, used, dpth;
170
+ if (n==0) return; else if (n==1) { A[0].m_key = 1; return; }
171
+ A[0].m_key += A[1].m_key; root = 0; leaf = 2;
172
+ for (next=1; next < n-1; next++)
173
+ {
174
+ if (leaf>=n || A[root].m_key<A[leaf].m_key) { A[next].m_key = A[root].m_key; A[root++].m_key = next; } else A[next].m_key = A[leaf++].m_key;
175
+ if (leaf>=n || (root<next && A[root].m_key<A[leaf].m_key)) { A[next].m_key += A[root].m_key; A[root++].m_key = next; } else A[next].m_key += A[leaf++].m_key;
176
+ }
177
+ A[n-2].m_key = 0;
178
+ for (next=n-3; next>=0; next--) A[next].m_key = A[A[next].m_key].m_key+1;
179
+ avbl = 1; used = dpth = 0; root = n-2; next = n-1;
180
+ while (avbl>0)
181
+ {
182
+ while (root>=0 && (int)A[root].m_key==dpth) { used++; root--; }
183
+ while (avbl>used) { A[next--].m_key = dpth; avbl--; }
184
+ avbl = 2*used; dpth++; used = 0;
185
+ }
186
+ }
187
+
188
+ // Limits canonical Huffman code table's max code size to max_code_size.
189
+ static void huffman_enforce_max_code_size(int *pNum_codes, int code_list_len, int max_code_size)
190
+ {
191
+ if (code_list_len <= 1) return;
192
+
193
+ for (int i = max_code_size + 1; i <= MAX_HUFF_CODESIZE; i++) pNum_codes[max_code_size] += pNum_codes[i];
194
+
195
+ uint32 total = 0;
196
+ for (int i = max_code_size; i > 0; i--)
197
+ total += (((uint32)pNum_codes[i]) << (max_code_size - i));
198
+
199
+ while (total != (1UL << max_code_size))
200
+ {
201
+ pNum_codes[max_code_size]--;
202
+ for (int i = max_code_size - 1; i > 0; i--)
203
+ {
204
+ if (pNum_codes[i]) { pNum_codes[i]--; pNum_codes[i + 1] += 2; break; }
205
+ }
206
+ total--;
207
+ }
208
+ }
209
+
210
+ // Generates an optimized offman table.
211
+ void jpeg_encoder::optimize_huffman_table(int table_num, int table_len)
212
+ {
213
+ sym_freq syms0[MAX_HUFF_SYMBOLS], syms1[MAX_HUFF_SYMBOLS];
214
+ syms0[0].m_key = 1; syms0[0].m_sym_index = 0; // dummy symbol, assures that no valid code contains all 1's
215
+ int num_used_syms = 1;
216
+ const uint32 *pSym_count = &m_huff_count[table_num][0];
217
+ for (int i = 0; i < table_len; i++)
218
+ if (pSym_count[i]) { syms0[num_used_syms].m_key = pSym_count[i]; syms0[num_used_syms++].m_sym_index = i + 1; }
219
+ sym_freq* pSyms = radix_sort_syms(num_used_syms, syms0, syms1);
220
+ calculate_minimum_redundancy(pSyms, num_used_syms);
221
+
222
+ // Count the # of symbols of each code size.
223
+ int num_codes[1 + MAX_HUFF_CODESIZE]; clear_obj(num_codes);
224
+ for (int i = 0; i < num_used_syms; i++)
225
+ num_codes[pSyms[i].m_key]++;
226
+
227
+ const uint JPGE_CODE_SIZE_LIMIT = 16; // the maximum possible size of a JPEG Huffman code (valid range is [9,16] - 9 vs. 8 because of the dummy symbol)
228
+ huffman_enforce_max_code_size(num_codes, num_used_syms, JPGE_CODE_SIZE_LIMIT);
229
+
230
+ // Compute m_huff_bits array, which contains the # of symbols per code size.
231
+ clear_obj(m_huff_bits[table_num]);
232
+ for (int i = 1; i <= (int)JPGE_CODE_SIZE_LIMIT; i++)
233
+ m_huff_bits[table_num][i] = static_cast<uint8>(num_codes[i]);
234
+
235
+ // Remove the dummy symbol added above, which must be in largest bucket.
236
+ for (int i = JPGE_CODE_SIZE_LIMIT; i >= 1; i--)
237
+ {
238
+ if (m_huff_bits[table_num][i]) { m_huff_bits[table_num][i]--; break; }
239
+ }
240
+
241
+ // Compute the m_huff_val array, which contains the symbol indices sorted by code size (smallest to largest).
242
+ for (int i = num_used_syms - 1; i >= 1; i--)
243
+ m_huff_val[table_num][num_used_syms - 1 - i] = static_cast<uint8>(pSyms[i].m_sym_index - 1);
244
+ }
245
+
246
+ // JPEG marker generation.
247
+ void jpeg_encoder::emit_byte(uint8 i)
248
+ {
249
+ m_all_stream_writes_succeeded = m_all_stream_writes_succeeded && m_pStream->put_obj(i);
250
+ }
251
+
252
+ void jpeg_encoder::emit_word(uint i)
253
+ {
254
+ emit_byte(uint8(i >> 8)); emit_byte(uint8(i & 0xFF));
255
+ }
256
+
257
+ void jpeg_encoder::emit_marker(int marker)
258
+ {
259
+ emit_byte(uint8(0xFF)); emit_byte(uint8(marker));
260
+ }
261
+
262
+ // Emit JFIF marker
263
+ void jpeg_encoder::emit_jfif_app0()
264
+ {
265
+ emit_marker(M_APP0);
266
+ emit_word(2 + 4 + 1 + 2 + 1 + 2 + 2 + 1 + 1);
267
+ emit_byte(0x4A); emit_byte(0x46); emit_byte(0x49); emit_byte(0x46); /* Identifier: ASCII "JFIF" */
268
+ emit_byte(0);
269
+ emit_byte(1); /* Major version */
270
+ emit_byte(1); /* Minor version */
271
+ emit_byte(0); /* Density unit */
272
+ emit_word(1);
273
+ emit_word(1);
274
+ emit_byte(0); /* No thumbnail image */
275
+ emit_byte(0);
276
+ }
277
+
278
+ // Emit quantization tables
279
+ void jpeg_encoder::emit_dqt()
280
+ {
281
+ for (int i = 0; i < ((m_num_components == 3) ? 2 : 1); i++)
282
+ {
283
+ emit_marker(M_DQT);
284
+ emit_word(64 + 1 + 2);
285
+ emit_byte(static_cast<uint8>(i));
286
+ for (int j = 0; j < 64; j++)
287
+ emit_byte(static_cast<uint8>(m_quantization_tables[i][j]));
288
+ }
289
+ }
290
+
291
+ // Emit start of frame marker
292
+ void jpeg_encoder::emit_sof()
293
+ {
294
+ emit_marker(M_SOF0); /* baseline */
295
+ emit_word(3 * m_num_components + 2 + 5 + 1);
296
+ emit_byte(8); /* precision */
297
+ emit_word(m_image_y);
298
+ emit_word(m_image_x);
299
+ emit_byte(m_num_components);
300
+ for (int i = 0; i < m_num_components; i++)
301
+ {
302
+ emit_byte(static_cast<uint8>(i + 1)); /* component ID */
303
+ emit_byte((m_comp_h_samp[i] << 4) + m_comp_v_samp[i]); /* h and v sampling */
304
+ emit_byte(i > 0); /* quant. table num */
305
+ }
306
+ }
307
+
308
+ // Emit Huffman table.
309
+ void jpeg_encoder::emit_dht(uint8 *bits, uint8 *val, int index, bool ac_flag)
310
+ {
311
+ emit_marker(M_DHT);
312
+
313
+ int length = 0;
314
+ for (int i = 1; i <= 16; i++)
315
+ length += bits[i];
316
+
317
+ emit_word(length + 2 + 1 + 16);
318
+ emit_byte(static_cast<uint8>(index + (ac_flag << 4)));
319
+
320
+ for (int i = 1; i <= 16; i++)
321
+ emit_byte(bits[i]);
322
+
323
+ for (int i = 0; i < length; i++)
324
+ emit_byte(val[i]);
325
+ }
326
+
327
+ // Emit all Huffman tables.
328
+ void jpeg_encoder::emit_dhts()
329
+ {
330
+ emit_dht(m_huff_bits[0+0], m_huff_val[0+0], 0, false);
331
+ emit_dht(m_huff_bits[2+0], m_huff_val[2+0], 0, true);
332
+ if (m_num_components == 3)
333
+ {
334
+ emit_dht(m_huff_bits[0+1], m_huff_val[0+1], 1, false);
335
+ emit_dht(m_huff_bits[2+1], m_huff_val[2+1], 1, true);
336
+ }
337
+ }
338
+
339
+ // emit start of scan
340
+ void jpeg_encoder::emit_sos()
341
+ {
342
+ emit_marker(M_SOS);
343
+ emit_word(2 * m_num_components + 2 + 1 + 3);
344
+ emit_byte(m_num_components);
345
+ for (int i = 0; i < m_num_components; i++)
346
+ {
347
+ emit_byte(static_cast<uint8>(i + 1));
348
+ if (i == 0)
349
+ emit_byte((0 << 4) + 0);
350
+ else
351
+ emit_byte((1 << 4) + 1);
352
+ }
353
+ emit_byte(0); /* spectral selection */
354
+ emit_byte(63);
355
+ emit_byte(0);
356
+ }
357
+
358
+ // Emit all markers at beginning of image file.
359
+ void jpeg_encoder::emit_markers()
360
+ {
361
+ emit_marker(M_SOI);
362
+ emit_jfif_app0();
363
+ emit_dqt();
364
+ emit_sof();
365
+ emit_dhts();
366
+ emit_sos();
367
+ }
368
+
369
+ // Compute the actual canonical Huffman codes/code sizes given the JPEG huff bits and val arrays.
370
+ void jpeg_encoder::compute_huffman_table(uint *codes, uint8 *code_sizes, uint8 *bits, uint8 *val)
371
+ {
372
+ int i, l, last_p, si;
373
+ uint8 huff_size[257];
374
+ uint huff_code[257];
375
+ uint code;
376
+
377
+ int p = 0;
378
+ for (l = 1; l <= 16; l++)
379
+ for (i = 1; i <= bits[l]; i++)
380
+ huff_size[p++] = (char)l;
381
+
382
+ huff_size[p] = 0; last_p = p; // write sentinel
383
+
384
+ code = 0; si = huff_size[0]; p = 0;
385
+
386
+ while (huff_size[p])
387
+ {
388
+ while (huff_size[p] == si)
389
+ huff_code[p++] = code++;
390
+ code <<= 1;
391
+ si++;
392
+ }
393
+
394
+ memset(codes, 0, sizeof(codes[0])*256);
395
+ memset(code_sizes, 0, sizeof(code_sizes[0])*256);
396
+ for (p = 0; p < last_p; p++)
397
+ {
398
+ codes[val[p]] = huff_code[p];
399
+ code_sizes[val[p]] = huff_size[p];
400
+ }
401
+ }
402
+
403
+ // Quantization table generation.
404
+ void jpeg_encoder::compute_quant_table(int32 *pDst, int16 *pSrc)
405
+ {
406
+ int32 q;
407
+ if (m_params.m_quality < 50)
408
+ q = 5000 / m_params.m_quality;
409
+ else
410
+ q = 200 - m_params.m_quality * 2;
411
+ for (int i = 0; i < 64; i++)
412
+ {
413
+ int32 j = *pSrc++; j = (j * q + 50L) / 100L;
414
+ *pDst++ = JPGE_MIN(JPGE_MAX(j, 1), 255);
415
+ }
416
+ }
417
+
418
+ // Higher-level methods.
419
+ void jpeg_encoder::first_pass_init()
420
+ {
421
+ m_bit_buffer = 0; m_bits_in = 0;
422
+ memset(m_last_dc_val, 0, 3 * sizeof(m_last_dc_val[0]));
423
+ m_mcu_y_ofs = 0;
424
+ m_pass_num = 1;
425
+ }
426
+
427
+ bool jpeg_encoder::second_pass_init()
428
+ {
429
+ compute_huffman_table(&m_huff_codes[0+0][0], &m_huff_code_sizes[0+0][0], m_huff_bits[0+0], m_huff_val[0+0]);
430
+ compute_huffman_table(&m_huff_codes[2+0][0], &m_huff_code_sizes[2+0][0], m_huff_bits[2+0], m_huff_val[2+0]);
431
+ if (m_num_components > 1)
432
+ {
433
+ compute_huffman_table(&m_huff_codes[0+1][0], &m_huff_code_sizes[0+1][0], m_huff_bits[0+1], m_huff_val[0+1]);
434
+ compute_huffman_table(&m_huff_codes[2+1][0], &m_huff_code_sizes[2+1][0], m_huff_bits[2+1], m_huff_val[2+1]);
435
+ }
436
+ first_pass_init();
437
+ emit_markers();
438
+ m_pass_num = 2;
439
+ return true;
440
+ }
441
+
442
+ bool jpeg_encoder::jpg_open(int p_x_res, int p_y_res, int src_channels)
443
+ {
444
+ m_num_components = 3;
445
+ switch (m_params.m_subsampling)
446
+ {
447
+ case Y_ONLY:
448
+ {
449
+ m_num_components = 1;
450
+ m_comp_h_samp[0] = 1; m_comp_v_samp[0] = 1;
451
+ m_mcu_x = 8; m_mcu_y = 8;
452
+ break;
453
+ }
454
+ case H1V1:
455
+ {
456
+ m_comp_h_samp[0] = 1; m_comp_v_samp[0] = 1;
457
+ m_comp_h_samp[1] = 1; m_comp_v_samp[1] = 1;
458
+ m_comp_h_samp[2] = 1; m_comp_v_samp[2] = 1;
459
+ m_mcu_x = 8; m_mcu_y = 8;
460
+ break;
461
+ }
462
+ case H2V1:
463
+ {
464
+ m_comp_h_samp[0] = 2; m_comp_v_samp[0] = 1;
465
+ m_comp_h_samp[1] = 1; m_comp_v_samp[1] = 1;
466
+ m_comp_h_samp[2] = 1; m_comp_v_samp[2] = 1;
467
+ m_mcu_x = 16; m_mcu_y = 8;
468
+ break;
469
+ }
470
+ case H2V2:
471
+ {
472
+ m_comp_h_samp[0] = 2; m_comp_v_samp[0] = 2;
473
+ m_comp_h_samp[1] = 1; m_comp_v_samp[1] = 1;
474
+ m_comp_h_samp[2] = 1; m_comp_v_samp[2] = 1;
475
+ m_mcu_x = 16; m_mcu_y = 16;
476
+ }
477
+ }
478
+
479
+ m_image_x = p_x_res; m_image_y = p_y_res;
480
+ m_image_bpp = src_channels;
481
+ m_image_bpl = m_image_x * src_channels;
482
+ m_image_x_mcu = (m_image_x + m_mcu_x - 1) & (~(m_mcu_x - 1));
483
+ m_image_y_mcu = (m_image_y + m_mcu_y - 1) & (~(m_mcu_y - 1));
484
+ m_image_bpl_xlt = m_image_x * m_num_components;
485
+ m_image_bpl_mcu = m_image_x_mcu * m_num_components;
486
+ m_mcus_per_row = m_image_x_mcu / m_mcu_x;
487
+
488
+ if ((m_mcu_lines[0] = static_cast<uint8*>(jpge_malloc(m_image_bpl_mcu * m_mcu_y))) == NULL) return false;
489
+ for (int i = 1; i < m_mcu_y; i++)
490
+ m_mcu_lines[i] = m_mcu_lines[i-1] + m_image_bpl_mcu;
491
+
492
+ compute_quant_table(m_quantization_tables[0], s_std_lum_quant);
493
+ compute_quant_table(m_quantization_tables[1], m_params.m_no_chroma_discrim_flag ? s_std_lum_quant : s_std_croma_quant);
494
+
495
+ m_out_buf_left = JPGE_OUT_BUF_SIZE;
496
+ m_pOut_buf = m_out_buf;
497
+
498
+ if (m_params.m_two_pass_flag)
499
+ {
500
+ clear_obj(m_huff_count);
501
+ first_pass_init();
502
+ }
503
+ else
504
+ {
505
+ memcpy(m_huff_bits[0+0], s_dc_lum_bits, 17); memcpy(m_huff_val [0+0], s_dc_lum_val, DC_LUM_CODES);
506
+ memcpy(m_huff_bits[2+0], s_ac_lum_bits, 17); memcpy(m_huff_val [2+0], s_ac_lum_val, AC_LUM_CODES);
507
+ memcpy(m_huff_bits[0+1], s_dc_chroma_bits, 17); memcpy(m_huff_val [0+1], s_dc_chroma_val, DC_CHROMA_CODES);
508
+ memcpy(m_huff_bits[2+1], s_ac_chroma_bits, 17); memcpy(m_huff_val [2+1], s_ac_chroma_val, AC_CHROMA_CODES);
509
+ if (!second_pass_init()) return false; // in effect, skip over the first pass
510
+ }
511
+ return m_all_stream_writes_succeeded;
512
+ }
513
+
514
+ void jpeg_encoder::load_block_8_8_grey(int x)
515
+ {
516
+ uint8 *pSrc;
517
+ sample_array_t *pDst = m_sample_array;
518
+ x <<= 3;
519
+ for (int i = 0; i < 8; i++, pDst += 8)
520
+ {
521
+ pSrc = m_mcu_lines[i] + x;
522
+ pDst[0] = pSrc[0] - 128; pDst[1] = pSrc[1] - 128; pDst[2] = pSrc[2] - 128; pDst[3] = pSrc[3] - 128;
523
+ pDst[4] = pSrc[4] - 128; pDst[5] = pSrc[5] - 128; pDst[6] = pSrc[6] - 128; pDst[7] = pSrc[7] - 128;
524
+ }
525
+ }
526
+
527
+ void jpeg_encoder::load_block_8_8(int x, int y, int c)
528
+ {
529
+ uint8 *pSrc;
530
+ sample_array_t *pDst = m_sample_array;
531
+ x = (x * (8 * 3)) + c;
532
+ y <<= 3;
533
+ for (int i = 0; i < 8; i++, pDst += 8)
534
+ {
535
+ pSrc = m_mcu_lines[y + i] + x;
536
+ pDst[0] = pSrc[0 * 3] - 128; pDst[1] = pSrc[1 * 3] - 128; pDst[2] = pSrc[2 * 3] - 128; pDst[3] = pSrc[3 * 3] - 128;
537
+ pDst[4] = pSrc[4 * 3] - 128; pDst[5] = pSrc[5 * 3] - 128; pDst[6] = pSrc[6 * 3] - 128; pDst[7] = pSrc[7 * 3] - 128;
538
+ }
539
+ }
540
+
541
+ void jpeg_encoder::load_block_16_8(int x, int c)
542
+ {
543
+ uint8 *pSrc1, *pSrc2;
544
+ sample_array_t *pDst = m_sample_array;
545
+ x = (x * (16 * 3)) + c;
546
+ int a = 0, b = 2;
547
+ for (int i = 0; i < 16; i += 2, pDst += 8)
548
+ {
549
+ pSrc1 = m_mcu_lines[i + 0] + x;
550
+ pSrc2 = m_mcu_lines[i + 1] + x;
551
+ pDst[0] = ((pSrc1[ 0 * 3] + pSrc1[ 1 * 3] + pSrc2[ 0 * 3] + pSrc2[ 1 * 3] + a) >> 2) - 128; pDst[1] = ((pSrc1[ 2 * 3] + pSrc1[ 3 * 3] + pSrc2[ 2 * 3] + pSrc2[ 3 * 3] + b) >> 2) - 128;
552
+ pDst[2] = ((pSrc1[ 4 * 3] + pSrc1[ 5 * 3] + pSrc2[ 4 * 3] + pSrc2[ 5 * 3] + a) >> 2) - 128; pDst[3] = ((pSrc1[ 6 * 3] + pSrc1[ 7 * 3] + pSrc2[ 6 * 3] + pSrc2[ 7 * 3] + b) >> 2) - 128;
553
+ pDst[4] = ((pSrc1[ 8 * 3] + pSrc1[ 9 * 3] + pSrc2[ 8 * 3] + pSrc2[ 9 * 3] + a) >> 2) - 128; pDst[5] = ((pSrc1[10 * 3] + pSrc1[11 * 3] + pSrc2[10 * 3] + pSrc2[11 * 3] + b) >> 2) - 128;
554
+ pDst[6] = ((pSrc1[12 * 3] + pSrc1[13 * 3] + pSrc2[12 * 3] + pSrc2[13 * 3] + a) >> 2) - 128; pDst[7] = ((pSrc1[14 * 3] + pSrc1[15 * 3] + pSrc2[14 * 3] + pSrc2[15 * 3] + b) >> 2) - 128;
555
+ int temp = a; a = b; b = temp;
556
+ }
557
+ }
558
+
559
+ void jpeg_encoder::load_block_16_8_8(int x, int c)
560
+ {
561
+ uint8 *pSrc1;
562
+ sample_array_t *pDst = m_sample_array;
563
+ x = (x * (16 * 3)) + c;
564
+ for (int i = 0; i < 8; i++, pDst += 8)
565
+ {
566
+ pSrc1 = m_mcu_lines[i + 0] + x;
567
+ pDst[0] = ((pSrc1[ 0 * 3] + pSrc1[ 1 * 3]) >> 1) - 128; pDst[1] = ((pSrc1[ 2 * 3] + pSrc1[ 3 * 3]) >> 1) - 128;
568
+ pDst[2] = ((pSrc1[ 4 * 3] + pSrc1[ 5 * 3]) >> 1) - 128; pDst[3] = ((pSrc1[ 6 * 3] + pSrc1[ 7 * 3]) >> 1) - 128;
569
+ pDst[4] = ((pSrc1[ 8 * 3] + pSrc1[ 9 * 3]) >> 1) - 128; pDst[5] = ((pSrc1[10 * 3] + pSrc1[11 * 3]) >> 1) - 128;
570
+ pDst[6] = ((pSrc1[12 * 3] + pSrc1[13 * 3]) >> 1) - 128; pDst[7] = ((pSrc1[14 * 3] + pSrc1[15 * 3]) >> 1) - 128;
571
+ }
572
+ }
573
+
574
+ void jpeg_encoder::load_quantized_coefficients(int component_num)
575
+ {
576
+ int32 *q = m_quantization_tables[component_num > 0];
577
+ int16 *pDst = m_coefficient_array;
578
+ for (int i = 0; i < 64; i++)
579
+ {
580
+ sample_array_t j = m_sample_array[s_zag[i]];
581
+ if (j < 0)
582
+ {
583
+ if ((j = -j + (*q >> 1)) < *q)
584
+ *pDst++ = 0;
585
+ else
586
+ *pDst++ = static_cast<int16>(-(j / *q));
587
+ }
588
+ else
589
+ {
590
+ if ((j = j + (*q >> 1)) < *q)
591
+ *pDst++ = 0;
592
+ else
593
+ *pDst++ = static_cast<int16>((j / *q));
594
+ }
595
+ q++;
596
+ }
597
+ }
598
+
599
+ void jpeg_encoder::flush_output_buffer()
600
+ {
601
+ if (m_out_buf_left != JPGE_OUT_BUF_SIZE)
602
+ m_all_stream_writes_succeeded = m_all_stream_writes_succeeded && m_pStream->put_buf(m_out_buf, JPGE_OUT_BUF_SIZE - m_out_buf_left);
603
+ m_pOut_buf = m_out_buf;
604
+ m_out_buf_left = JPGE_OUT_BUF_SIZE;
605
+ }
606
+
607
+ void jpeg_encoder::put_bits(uint bits, uint len)
608
+ {
609
+ m_bit_buffer |= ((uint32)bits << (24 - (m_bits_in += len)));
610
+ while (m_bits_in >= 8)
611
+ {
612
+ uint8 c;
613
+ #define JPGE_PUT_BYTE(c) { *m_pOut_buf++ = (c); if (--m_out_buf_left == 0) flush_output_buffer(); }
614
+ JPGE_PUT_BYTE(c = (uint8)((m_bit_buffer >> 16) & 0xFF));
615
+ if (c == 0xFF) JPGE_PUT_BYTE(0);
616
+ m_bit_buffer <<= 8;
617
+ m_bits_in -= 8;
618
+ }
619
+ }
620
+
621
+ void jpeg_encoder::code_coefficients_pass_one(int component_num)
622
+ {
623
+ if (component_num >= 3) return; // just to shut up static analysis
624
+ int i, run_len, nbits, temp1;
625
+ int16 *src = m_coefficient_array;
626
+ uint32 *dc_count = component_num ? m_huff_count[0 + 1] : m_huff_count[0 + 0], *ac_count = component_num ? m_huff_count[2 + 1] : m_huff_count[2 + 0];
627
+
628
+ temp1 = src[0] - m_last_dc_val[component_num];
629
+ m_last_dc_val[component_num] = src[0];
630
+ if (temp1 < 0) temp1 = -temp1;
631
+
632
+ nbits = 0;
633
+ while (temp1)
634
+ {
635
+ nbits++; temp1 >>= 1;
636
+ }
637
+
638
+ dc_count[nbits]++;
639
+ for (run_len = 0, i = 1; i < 64; i++)
640
+ {
641
+ if ((temp1 = m_coefficient_array[i]) == 0)
642
+ run_len++;
643
+ else
644
+ {
645
+ while (run_len >= 16)
646
+ {
647
+ ac_count[0xF0]++;
648
+ run_len -= 16;
649
+ }
650
+ if (temp1 < 0) temp1 = -temp1;
651
+ nbits = 1;
652
+ while (temp1 >>= 1) nbits++;
653
+ ac_count[(run_len << 4) + nbits]++;
654
+ run_len = 0;
655
+ }
656
+ }
657
+ if (run_len) ac_count[0]++;
658
+ }
659
+
660
+ void jpeg_encoder::code_coefficients_pass_two(int component_num)
661
+ {
662
+ int i, j, run_len, nbits, temp1, temp2;
663
+ int16 *pSrc = m_coefficient_array;
664
+ uint *codes[2];
665
+ uint8 *code_sizes[2];
666
+
667
+ if (component_num == 0)
668
+ {
669
+ codes[0] = m_huff_codes[0 + 0]; codes[1] = m_huff_codes[2 + 0];
670
+ code_sizes[0] = m_huff_code_sizes[0 + 0]; code_sizes[1] = m_huff_code_sizes[2 + 0];
671
+ }
672
+ else
673
+ {
674
+ codes[0] = m_huff_codes[0 + 1]; codes[1] = m_huff_codes[2 + 1];
675
+ code_sizes[0] = m_huff_code_sizes[0 + 1]; code_sizes[1] = m_huff_code_sizes[2 + 1];
676
+ }
677
+
678
+ temp1 = temp2 = pSrc[0] - m_last_dc_val[component_num];
679
+ m_last_dc_val[component_num] = pSrc[0];
680
+
681
+ if (temp1 < 0)
682
+ {
683
+ temp1 = -temp1; temp2--;
684
+ }
685
+
686
+ nbits = 0;
687
+ while (temp1)
688
+ {
689
+ nbits++; temp1 >>= 1;
690
+ }
691
+
692
+ put_bits(codes[0][nbits], code_sizes[0][nbits]);
693
+ if (nbits) put_bits(temp2 & ((1 << nbits) - 1), nbits);
694
+
695
+ for (run_len = 0, i = 1; i < 64; i++)
696
+ {
697
+ if ((temp1 = m_coefficient_array[i]) == 0)
698
+ run_len++;
699
+ else
700
+ {
701
+ while (run_len >= 16)
702
+ {
703
+ put_bits(codes[1][0xF0], code_sizes[1][0xF0]);
704
+ run_len -= 16;
705
+ }
706
+ if ((temp2 = temp1) < 0)
707
+ {
708
+ temp1 = -temp1;
709
+ temp2--;
710
+ }
711
+ nbits = 1;
712
+ while (temp1 >>= 1)
713
+ nbits++;
714
+ j = (run_len << 4) + nbits;
715
+ put_bits(codes[1][j], code_sizes[1][j]);
716
+ put_bits(temp2 & ((1 << nbits) - 1), nbits);
717
+ run_len = 0;
718
+ }
719
+ }
720
+ if (run_len)
721
+ put_bits(codes[1][0], code_sizes[1][0]);
722
+ }
723
+
724
+ void jpeg_encoder::code_block(int component_num)
725
+ {
726
+ DCT2D(m_sample_array);
727
+ load_quantized_coefficients(component_num);
728
+ if (m_pass_num == 1)
729
+ code_coefficients_pass_one(component_num);
730
+ else
731
+ code_coefficients_pass_two(component_num);
732
+ }
733
+
734
+ void jpeg_encoder::process_mcu_row()
735
+ {
736
+ if (m_num_components == 1)
737
+ {
738
+ for (int i = 0; i < m_mcus_per_row; i++)
739
+ {
740
+ load_block_8_8_grey(i); code_block(0);
741
+ }
742
+ }
743
+ else if ((m_comp_h_samp[0] == 1) && (m_comp_v_samp[0] == 1))
744
+ {
745
+ for (int i = 0; i < m_mcus_per_row; i++)
746
+ {
747
+ load_block_8_8(i, 0, 0); code_block(0); load_block_8_8(i, 0, 1); code_block(1); load_block_8_8(i, 0, 2); code_block(2);
748
+ }
749
+ }
750
+ else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 1))
751
+ {
752
+ for (int i = 0; i < m_mcus_per_row; i++)
753
+ {
754
+ load_block_8_8(i * 2 + 0, 0, 0); code_block(0); load_block_8_8(i * 2 + 1, 0, 0); code_block(0);
755
+ load_block_16_8_8(i, 1); code_block(1); load_block_16_8_8(i, 2); code_block(2);
756
+ }
757
+ }
758
+ else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 2))
759
+ {
760
+ for (int i = 0; i < m_mcus_per_row; i++)
761
+ {
762
+ load_block_8_8(i * 2 + 0, 0, 0); code_block(0); load_block_8_8(i * 2 + 1, 0, 0); code_block(0);
763
+ load_block_8_8(i * 2 + 0, 1, 0); code_block(0); load_block_8_8(i * 2 + 1, 1, 0); code_block(0);
764
+ load_block_16_8(i, 1); code_block(1); load_block_16_8(i, 2); code_block(2);
765
+ }
766
+ }
767
+ }
768
+
769
+ bool jpeg_encoder::terminate_pass_one()
770
+ {
771
+ optimize_huffman_table(0+0, DC_LUM_CODES); optimize_huffman_table(2+0, AC_LUM_CODES);
772
+ if (m_num_components > 1)
773
+ {
774
+ optimize_huffman_table(0+1, DC_CHROMA_CODES); optimize_huffman_table(2+1, AC_CHROMA_CODES);
775
+ }
776
+ return second_pass_init();
777
+ }
778
+
779
+ bool jpeg_encoder::terminate_pass_two()
780
+ {
781
+ put_bits(0x7F, 7);
782
+ flush_output_buffer();
783
+ emit_marker(M_EOI);
784
+ m_pass_num++; // purposely bump up m_pass_num, for debugging
785
+ return true;
786
+ }
787
+
788
+ bool jpeg_encoder::process_end_of_image()
789
+ {
790
+ if (m_mcu_y_ofs)
791
+ {
792
+ if (m_mcu_y_ofs < 16) // check here just to shut up static analysis
793
+ {
794
+ for (int i = m_mcu_y_ofs; i < m_mcu_y; i++)
795
+ memcpy(m_mcu_lines[i], m_mcu_lines[m_mcu_y_ofs - 1], m_image_bpl_mcu);
796
+ }
797
+
798
+ process_mcu_row();
799
+ }
800
+
801
+ if (m_pass_num == 1)
802
+ return terminate_pass_one();
803
+ else
804
+ return terminate_pass_two();
805
+ }
806
+
807
+ void jpeg_encoder::load_mcu(const void *pSrc)
808
+ {
809
+ const uint8* Psrc = reinterpret_cast<const uint8*>(pSrc);
810
+
811
+ uint8* pDst = m_mcu_lines[m_mcu_y_ofs]; // OK to write up to m_image_bpl_xlt bytes to pDst
812
+
813
+ if (m_num_components == 1)
814
+ {
815
+ if (m_image_bpp == 4)
816
+ RGBA_to_Y(pDst, Psrc, m_image_x);
817
+ else if (m_image_bpp == 3)
818
+ RGB_to_Y(pDst, Psrc, m_image_x);
819
+ else
820
+ memcpy(pDst, Psrc, m_image_x);
821
+ }
822
+ else
823
+ {
824
+ if (m_image_bpp == 4)
825
+ RGBA_to_YCC(pDst, Psrc, m_image_x);
826
+ else if (m_image_bpp == 3)
827
+ RGB_to_YCC(pDst, Psrc, m_image_x);
828
+ else
829
+ Y_to_YCC(pDst, Psrc, m_image_x);
830
+ }
831
+
832
+ // Possibly duplicate pixels at end of scanline if not a multiple of 8 or 16
833
+ if (m_num_components == 1)
834
+ memset(m_mcu_lines[m_mcu_y_ofs] + m_image_bpl_xlt, pDst[m_image_bpl_xlt - 1], m_image_x_mcu - m_image_x);
835
+ else
836
+ {
837
+ const uint8 y = pDst[m_image_bpl_xlt - 3 + 0], cb = pDst[m_image_bpl_xlt - 3 + 1], cr = pDst[m_image_bpl_xlt - 3 + 2];
838
+ uint8 *q = m_mcu_lines[m_mcu_y_ofs] + m_image_bpl_xlt;
839
+ for (int i = m_image_x; i < m_image_x_mcu; i++)
840
+ {
841
+ *q++ = y; *q++ = cb; *q++ = cr;
842
+ }
843
+ }
844
+
845
+ if (++m_mcu_y_ofs == m_mcu_y)
846
+ {
847
+ process_mcu_row();
848
+ m_mcu_y_ofs = 0;
849
+ }
850
+ }
851
+
852
+ void jpeg_encoder::clear()
853
+ {
854
+ m_mcu_lines[0] = NULL;
855
+ m_pass_num = 0;
856
+ m_all_stream_writes_succeeded = true;
857
+ }
858
+
859
+ jpeg_encoder::jpeg_encoder()
860
+ {
861
+ clear();
862
+ }
863
+
864
+ jpeg_encoder::~jpeg_encoder()
865
+ {
866
+ deinit();
867
+ }
868
+
869
+ bool jpeg_encoder::init(output_stream *pStream, int64_t width, int64_t height, int64_t src_channels, const params &comp_params)
870
+ {
871
+ deinit();
872
+ if (((!pStream) || (width < 1) || (height < 1)) || ((src_channels != 1) && (src_channels != 3) && (src_channels != 4)) || (!comp_params.check_valid())) return false;
873
+ m_pStream = pStream;
874
+ m_params = comp_params;
875
+ return jpg_open(width, height, src_channels);
876
+ }
877
+
878
+ void jpeg_encoder::deinit()
879
+ {
880
+ jpge_free(m_mcu_lines[0]);
881
+ clear();
882
+ }
883
+
884
+ bool jpeg_encoder::process_scanline(const void* pScanline)
885
+ {
886
+ if ((m_pass_num < 1) || (m_pass_num > 2)) return false;
887
+ if (m_all_stream_writes_succeeded)
888
+ {
889
+ if (!pScanline)
890
+ {
891
+ if (!process_end_of_image()) return false;
892
+ }
893
+ else
894
+ {
895
+ load_mcu(pScanline);
896
+ }
897
+ }
898
+ return m_all_stream_writes_succeeded;
899
+ }
900
+
901
+ // Higher level wrappers/examples (optional).
902
+ #include <stdio.h>
903
+
904
+ class cfile_stream : public output_stream
905
+ {
906
+ cfile_stream(const cfile_stream &);
907
+ cfile_stream &operator= (const cfile_stream &);
908
+
909
+ FILE* m_pFile;
910
+ bool m_bStatus;
911
+
912
+ public:
913
+ cfile_stream() : m_pFile(NULL), m_bStatus(false) { }
914
+
915
+ virtual ~cfile_stream()
916
+ {
917
+ close();
918
+ }
919
+
920
+ bool open(const char *pFilename)
921
+ {
922
+ close();
923
+ #if defined(_MSC_VER)
924
+ if (fopen_s(&m_pFile, pFilename, "wb") != 0)
925
+ {
926
+ return false;
927
+ }
928
+ #else
929
+ m_pFile = fopen(pFilename, "wb");
930
+ #endif
931
+ m_bStatus = (m_pFile != NULL);
932
+ return m_bStatus;
933
+ }
934
+
935
+ bool close()
936
+ {
937
+ if (m_pFile)
938
+ {
939
+ if (fclose(m_pFile) == EOF)
940
+ {
941
+ m_bStatus = false;
942
+ }
943
+ m_pFile = NULL;
944
+ }
945
+ return m_bStatus;
946
+ }
947
+
948
+ virtual bool put_buf(const void* pBuf, int64_t len)
949
+ {
950
+ m_bStatus = m_bStatus && (fwrite(pBuf, len, 1, m_pFile) == 1);
951
+ return m_bStatus;
952
+ }
953
+
954
+ uint get_size() const
955
+ {
956
+ return m_pFile ? ftell(m_pFile) : 0;
957
+ }
958
+ };
959
+
960
+ // Writes JPEG image to file.
961
+ bool compress_image_to_jpeg_file(const char *pFilename, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params)
962
+ {
963
+ cfile_stream dst_stream;
964
+ if (!dst_stream.open(pFilename))
965
+ return false;
966
+
967
+ jpge::jpeg_encoder dst_image;
968
+ if (!dst_image.init(&dst_stream, width, height, num_channels, comp_params))
969
+ return false;
970
+
971
+ for (uint pass_index = 0; pass_index < dst_image.get_total_passes(); pass_index++)
972
+ {
973
+ for (int64_t i = 0; i < height; i++)
974
+ {
975
+ // i, width, and num_channels are all 64bit
976
+ const uint8* pBuf = pImage_data + i * width * num_channels;
977
+ if (!dst_image.process_scanline(pBuf))
978
+ return false;
979
+ }
980
+ if (!dst_image.process_scanline(NULL))
981
+ return false;
982
+ }
983
+
984
+ dst_image.deinit();
985
+
986
+ return dst_stream.close();
987
+ }
988
+
989
+ class memory_stream : public output_stream
990
+ {
991
+ memory_stream(const memory_stream &);
992
+ memory_stream &operator= (const memory_stream &);
993
+
994
+ uint8 *m_pBuf;
995
+ uint64_t m_buf_size, m_buf_ofs;
996
+
997
+ public:
998
+ memory_stream(void *pBuf, uint64_t buf_size) : m_pBuf(static_cast<uint8*>(pBuf)), m_buf_size(buf_size), m_buf_ofs(0) { }
999
+
1000
+ virtual ~memory_stream() { }
1001
+
1002
+ virtual bool put_buf(const void* pBuf, int64_t len)
1003
+ {
1004
+ uint64_t buf_remaining = m_buf_size - m_buf_ofs;
1005
+ if ((uint64_t)len > buf_remaining)
1006
+ return false;
1007
+ memcpy(m_pBuf + m_buf_ofs, pBuf, len);
1008
+ m_buf_ofs += len;
1009
+ return true;
1010
+ }
1011
+
1012
+ uint64_t get_size() const
1013
+ {
1014
+ return m_buf_ofs;
1015
+ }
1016
+ };
1017
+
1018
+ bool compress_image_to_jpeg_file_in_memory(void *pDstBuf, int64_t &buf_size, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params)
1019
+ {
1020
+ if ((!pDstBuf) || (!buf_size))
1021
+ return false;
1022
+
1023
+ memory_stream dst_stream(pDstBuf, buf_size);
1024
+
1025
+ buf_size = 0;
1026
+
1027
+ jpge::jpeg_encoder dst_image;
1028
+ if (!dst_image.init(&dst_stream, width, height, num_channels, comp_params))
1029
+ return false;
1030
+
1031
+ for (uint pass_index = 0; pass_index < dst_image.get_total_passes(); pass_index++)
1032
+ {
1033
+ for (int64_t i = 0; i < height; i++)
1034
+ {
1035
+ const uint8* pScanline = pImage_data + i * width * num_channels;
1036
+ if (!dst_image.process_scanline(pScanline))
1037
+ return false;
1038
+ }
1039
+ if (!dst_image.process_scanline(NULL))
1040
+ return false;
1041
+ }
1042
+
1043
+ dst_image.deinit();
1044
+
1045
+ buf_size = dst_stream.get_size();
1046
+ return true;
1047
+ }
1048
+
1049
+ } // namespace jpge
crazy_functions/test_project/cpp/longcode/prod_cons.h ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #include <atomic>
4
+ #include <utility>
5
+ #include <cstring>
6
+ #include <type_traits>
7
+ #include <cstdint>
8
+
9
+ #include "libipc/def.h"
10
+
11
+ #include "libipc/platform/detail.h"
12
+ #include "libipc/circ/elem_def.h"
13
+ #include "libipc/utility/log.h"
14
+ #include "libipc/utility/utility.h"
15
+
16
+ namespace ipc {
17
+
18
+ ////////////////////////////////////////////////////////////////
19
+ /// producer-consumer implementation
20
+ ////////////////////////////////////////////////////////////////
21
+
22
+ template <typename Flag>
23
+ struct prod_cons_impl;
24
+
25
+ template <>
26
+ struct prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
27
+
28
+ template <std::size_t DataSize, std::size_t AlignSize>
29
+ struct elem_t {
30
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
31
+ };
32
+
33
+ alignas(cache_line_size) std::atomic<circ::u2_t> rd_; // read index
34
+ alignas(cache_line_size) std::atomic<circ::u2_t> wt_; // write index
35
+
36
+ constexpr circ::u2_t cursor() const noexcept {
37
+ return 0;
38
+ }
39
+
40
+ template <typename W, typename F, typename E>
41
+ bool push(W* /*wrapper*/, F&& f, E* elems) {
42
+ auto cur_wt = circ::index_of(wt_.load(std::memory_order_relaxed));
43
+ if (cur_wt == circ::index_of(rd_.load(std::memory_order_acquire) - 1)) {
44
+ return false; // full
45
+ }
46
+ std::forward<F>(f)(&(elems[cur_wt].data_));
47
+ wt_.fetch_add(1, std::memory_order_release);
48
+ return true;
49
+ }
50
+
51
+ /**
52
+ * In single-single-unicast, 'force_push' means 'no reader' or 'the only one reader is dead'.
53
+ * So we could just disconnect all connections of receiver, and return false.
54
+ */
55
+ template <typename W, typename F, typename E>
56
+ bool force_push(W* wrapper, F&&, E*) {
57
+ wrapper->elems()->disconnect_receiver(~static_cast<circ::cc_t>(0u));
58
+ return false;
59
+ }
60
+
61
+ template <typename W, typename F, typename R, typename E>
62
+ bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E* elems) {
63
+ auto cur_rd = circ::index_of(rd_.load(std::memory_order_relaxed));
64
+ if (cur_rd == circ::index_of(wt_.load(std::memory_order_acquire))) {
65
+ return false; // empty
66
+ }
67
+ std::forward<F>(f)(&(elems[cur_rd].data_));
68
+ std::forward<R>(out)(true);
69
+ rd_.fetch_add(1, std::memory_order_release);
70
+ return true;
71
+ }
72
+ };
73
+
74
+ template <>
75
+ struct prod_cons_impl<wr<relat::single, relat::multi , trans::unicast>>
76
+ : prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
77
+
78
+ template <typename W, typename F, typename E>
79
+ bool force_push(W* wrapper, F&&, E*) {
80
+ wrapper->elems()->disconnect_receiver(1);
81
+ return false;
82
+ }
83
+
84
+ template <typename W, typename F, typename R,
85
+ template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
86
+ bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
87
+ byte_t buff[DS];
88
+ for (unsigned k = 0;;) {
89
+ auto cur_rd = rd_.load(std::memory_order_relaxed);
90
+ if (circ::index_of(cur_rd) ==
91
+ circ::index_of(wt_.load(std::memory_order_acquire))) {
92
+ return false; // empty
93
+ }
94
+ std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
95
+ if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
96
+ std::forward<F>(f)(buff);
97
+ std::forward<R>(out)(true);
98
+ return true;
99
+ }
100
+ ipc::yield(k);
101
+ }
102
+ }
103
+ };
104
+
105
+ template <>
106
+ struct prod_cons_impl<wr<relat::multi , relat::multi, trans::unicast>>
107
+ : prod_cons_impl<wr<relat::single, relat::multi, trans::unicast>> {
108
+
109
+ using flag_t = std::uint64_t;
110
+
111
+ template <std::size_t DataSize, std::size_t AlignSize>
112
+ struct elem_t {
113
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
114
+ std::atomic<flag_t> f_ct_ { 0 }; // commit flag
115
+ };
116
+
117
+ alignas(cache_line_size) std::atomic<circ::u2_t> ct_; // commit index
118
+
119
+ template <typename W, typename F, typename E>
120
+ bool push(W* /*wrapper*/, F&& f, E* elems) {
121
+ circ::u2_t cur_ct, nxt_ct;
122
+ for (unsigned k = 0;;) {
123
+ cur_ct = ct_.load(std::memory_order_relaxed);
124
+ if (circ::index_of(nxt_ct = cur_ct + 1) ==
125
+ circ::index_of(rd_.load(std::memory_order_acquire))) {
126
+ return false; // full
127
+ }
128
+ if (ct_.compare_exchange_weak(cur_ct, nxt_ct, std::memory_order_acq_rel)) {
129
+ break;
130
+ }
131
+ ipc::yield(k);
132
+ }
133
+ auto* el = elems + circ::index_of(cur_ct);
134
+ std::forward<F>(f)(&(el->data_));
135
+ // set flag & try update wt
136
+ el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
137
+ while (1) {
138
+ auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
139
+ if (cur_ct != wt_.load(std::memory_order_relaxed)) {
140
+ return true;
141
+ }
142
+ if ((~cac_ct) != cur_ct) {
143
+ return true;
144
+ }
145
+ if (!el->f_ct_.compare_exchange_strong(cac_ct, 0, std::memory_order_relaxed)) {
146
+ return true;
147
+ }
148
+ wt_.store(nxt_ct, std::memory_order_release);
149
+ cur_ct = nxt_ct;
150
+ nxt_ct = cur_ct + 1;
151
+ el = elems + circ::index_of(cur_ct);
152
+ }
153
+ return true;
154
+ }
155
+
156
+ template <typename W, typename F, typename E>
157
+ bool force_push(W* wrapper, F&&, E*) {
158
+ wrapper->elems()->disconnect_receiver(1);
159
+ return false;
160
+ }
161
+
162
+ template <typename W, typename F, typename R,
163
+ template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
164
+ bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
165
+ byte_t buff[DS];
166
+ for (unsigned k = 0;;) {
167
+ auto cur_rd = rd_.load(std::memory_order_relaxed);
168
+ auto cur_wt = wt_.load(std::memory_order_acquire);
169
+ auto id_rd = circ::index_of(cur_rd);
170
+ auto id_wt = circ::index_of(cur_wt);
171
+ if (id_rd == id_wt) {
172
+ auto* el = elems + id_wt;
173
+ auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
174
+ if ((~cac_ct) != cur_wt) {
175
+ return false; // empty
176
+ }
177
+ if (el->f_ct_.compare_exchange_weak(cac_ct, 0, std::memory_order_relaxed)) {
178
+ wt_.store(cur_wt + 1, std::memory_order_release);
179
+ }
180
+ k = 0;
181
+ }
182
+ else {
183
+ std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
184
+ if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
185
+ std::forward<F>(f)(buff);
186
+ std::forward<R>(out)(true);
187
+ return true;
188
+ }
189
+ ipc::yield(k);
190
+ }
191
+ }
192
+ }
193
+ };
194
+
195
+ template <>
196
+ struct prod_cons_impl<wr<relat::single, relat::multi, trans::broadcast>> {
197
+
198
+ using rc_t = std::uint64_t;
199
+
200
+ enum : rc_t {
201
+ ep_mask = 0x00000000ffffffffull,
202
+ ep_incr = 0x0000000100000000ull
203
+ };
204
+
205
+ template <std::size_t DataSize, std::size_t AlignSize>
206
+ struct elem_t {
207
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
208
+ std::atomic<rc_t> rc_ { 0 }; // read-counter
209
+ };
210
+
211
+ alignas(cache_line_size) std::atomic<circ::u2_t> wt_; // write index
212
+ alignas(cache_line_size) rc_t epoch_ { 0 }; // only one writer
213
+
214
+ circ::u2_t cursor() const noexcept {
215
+ return wt_.load(std::memory_order_acquire);
216
+ }
217
+
218
+ template <typename W, typename F, typename E>
219
+ bool push(W* wrapper, F&& f, E* elems) {
220
+ E* el;
221
+ for (unsigned k = 0;;) {
222
+ circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
223
+ if (cc == 0) return false; // no reader
224
+ el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
225
+ // check all consumers have finished reading this element
226
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
227
+ circ::cc_t rem_cc = cur_rc & ep_mask;
228
+ if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch_)) {
229
+ return false; // has not finished yet
230
+ }
231
+ // consider rem_cc to be 0 here
232
+ if (el->rc_.compare_exchange_weak(
233
+ cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
234
+ break;
235
+ }
236
+ ipc::yield(k);
237
+ }
238
+ std::forward<F>(f)(&(el->data_));
239
+ wt_.fetch_add(1, std::memory_order_release);
240
+ return true;
241
+ }
242
+
243
+ template <typename W, typename F, typename E>
244
+ bool force_push(W* wrapper, F&& f, E* elems) {
245
+ E* el;
246
+ epoch_ += ep_incr;
247
+ for (unsigned k = 0;;) {
248
+ circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
249
+ if (cc == 0) return false; // no reader
250
+ el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
251
+ // check all consumers have finished reading this element
252
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
253
+ circ::cc_t rem_cc = cur_rc & ep_mask;
254
+ if (cc & rem_cc) {
255
+ ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
256
+ cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
257
+ if (cc == 0) return false; // no reader
258
+ }
259
+ // just compare & exchange
260
+ if (el->rc_.compare_exchange_weak(
261
+ cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
262
+ break;
263
+ }
264
+ ipc::yield(k);
265
+ }
266
+ std::forward<F>(f)(&(el->data_));
267
+ wt_.fetch_add(1, std::memory_order_release);
268
+ return true;
269
+ }
270
+
271
+ template <typename W, typename F, typename R, typename E>
272
+ bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E* elems) {
273
+ if (cur == cursor()) return false; // acquire
274
+ auto* el = elems + circ::index_of(cur++);
275
+ std::forward<F>(f)(&(el->data_));
276
+ for (unsigned k = 0;;) {
277
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
278
+ if ((cur_rc & ep_mask) == 0) {
279
+ std::forward<R>(out)(true);
280
+ return true;
281
+ }
282
+ auto nxt_rc = cur_rc & ~static_cast<rc_t>(wrapper->connected_id());
283
+ if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
284
+ std::forward<R>(out)((nxt_rc & ep_mask) == 0);
285
+ return true;
286
+ }
287
+ ipc::yield(k);
288
+ }
289
+ }
290
+ };
291
+
292
+ template <>
293
+ struct prod_cons_impl<wr<relat::multi, relat::multi, trans::broadcast>> {
294
+
295
+ using rc_t = std::uint64_t;
296
+ using flag_t = std::uint64_t;
297
+
298
+ enum : rc_t {
299
+ rc_mask = 0x00000000ffffffffull,
300
+ ep_mask = 0x00ffffffffffffffull,
301
+ ep_incr = 0x0100000000000000ull,
302
+ ic_mask = 0xff000000ffffffffull,
303
+ ic_incr = 0x0000000100000000ull
304
+ };
305
+
306
+ template <std::size_t DataSize, std::size_t AlignSize>
307
+ struct elem_t {
308
+ std::aligned_storage_t<DataSize, AlignSize> data_ {};
309
+ std::atomic<rc_t > rc_ { 0 }; // read-counter
310
+ std::atomic<flag_t> f_ct_ { 0 }; // commit flag
311
+ };
312
+
313
+ alignas(cache_line_size) std::atomic<circ::u2_t> ct_; // commit index
314
+ alignas(cache_line_size) std::atomic<rc_t> epoch_ { 0 };
315
+
316
+ circ::u2_t cursor() const noexcept {
317
+ return ct_.load(std::memory_order_acquire);
318
+ }
319
+
320
+ constexpr static rc_t inc_rc(rc_t rc) noexcept {
321
+ return (rc & ic_mask) | ((rc + ic_incr) & ~ic_mask);
322
+ }
323
+
324
+ constexpr static rc_t inc_mask(rc_t rc) noexcept {
325
+ return inc_rc(rc) & ~rc_mask;
326
+ }
327
+
328
+ template <typename W, typename F, typename E>
329
+ bool push(W* wrapper, F&& f, E* elems) {
330
+ E* el;
331
+ circ::u2_t cur_ct;
332
+ rc_t epoch = epoch_.load(std::memory_order_acquire);
333
+ for (unsigned k = 0;;) {
334
+ circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
335
+ if (cc == 0) return false; // no reader
336
+ el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
337
+ // check all consumers have finished reading this element
338
+ auto cur_rc = el->rc_.load(std::memory_order_relaxed);
339
+ circ::cc_t rem_cc = cur_rc & rc_mask;
340
+ if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch)) {
341
+ return false; // has not finished yet
342
+ }
343
+ else if (!rem_cc) {
344
+ auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
345
+ if ((cur_fl != cur_ct) && cur_fl) {
346
+ return false; // full
347
+ }
348
+ }
349
+ // consider rem_cc to be 0 here
350
+ if (el->rc_.compare_exchange_weak(
351
+ cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed) &&
352
+ epoch_.compare_exchange_weak(epoch, epoch, std::memory_order_acq_rel)) {
353
+ break;
354
+ }
355
+ ipc::yield(k);
356
+ }
357
+ // only one thread/process would touch here at one time
358
+ ct_.store(cur_ct + 1, std::memory_order_release);
359
+ std::forward<F>(f)(&(el->data_));
360
+ // set flag & try update wt
361
+ el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
362
+ return true;
363
+ }
364
+
365
+ template <typename W, typename F, typename E>
366
+ bool force_push(W* wrapper, F&& f, E* elems) {
367
+ E* el;
368
+ circ::u2_t cur_ct;
369
+ rc_t epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
370
+ for (unsigned k = 0;;) {
371
+ circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
372
+ if (cc == 0) return false; // no reader
373
+ el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
374
+ // check all consumers have finished reading this element
375
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
376
+ circ::cc_t rem_cc = cur_rc & rc_mask;
377
+ if (cc & rem_cc) {
378
+ ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
379
+ cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
380
+ if (cc == 0) return false; // no reader
381
+ }
382
+ // just compare & exchange
383
+ if (el->rc_.compare_exchange_weak(
384
+ cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed)) {
385
+ if (epoch == epoch_.load(std::memory_order_acquire)) {
386
+ break;
387
+ }
388
+ else if (push(wrapper, std::forward<F>(f), elems)) {
389
+ return true;
390
+ }
391
+ epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
392
+ }
393
+ ipc::yield(k);
394
+ }
395
+ // only one thread/process would touch here at one time
396
+ ct_.store(cur_ct + 1, std::memory_order_release);
397
+ std::forward<F>(f)(&(el->data_));
398
+ // set flag & try update wt
399
+ el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
400
+ return true;
401
+ }
402
+
403
+ template <typename W, typename F, typename R, typename E, std::size_t N>
404
+ bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E(& elems)[N]) {
405
+ auto* el = elems + circ::index_of(cur);
406
+ auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
407
+ if (cur_fl != ~static_cast<flag_t>(cur)) {
408
+ return false; // empty
409
+ }
410
+ ++cur;
411
+ std::forward<F>(f)(&(el->data_));
412
+ for (unsigned k = 0;;) {
413
+ auto cur_rc = el->rc_.load(std::memory_order_acquire);
414
+ if ((cur_rc & rc_mask) == 0) {
415
+ std::forward<R>(out)(true);
416
+ el->f_ct_.store(cur + N - 1, std::memory_order_release);
417
+ return true;
418
+ }
419
+ auto nxt_rc = inc_rc(cur_rc) & ~static_cast<rc_t>(wrapper->connected_id());
420
+ bool last_one = false;
421
+ if ((last_one = (nxt_rc & rc_mask) == 0)) {
422
+ el->f_ct_.store(cur + N - 1, std::memory_order_release);
423
+ }
424
+ if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
425
+ std::forward<R>(out)(last_one);
426
+ return true;
427
+ }
428
+ ipc::yield(k);
429
+ }
430
+ }
431
+ };
432
+
433
+ } // namespace ipc
crazy_functions/test_project/latex/attention/background.tex ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ The goal of reducing sequential computation also forms the foundation of the Extended Neural GPU \citep{extendedngpu}, ByteNet \citep{NalBytenet2017} and ConvS2S \citep{JonasFaceNet2017}, all of which use convolutional neural networks as basic building block, computing hidden representations in parallel for all input and output positions. In these models, the number of operations required to relate signals from two arbitrary input or output positions grows in the distance between positions, linearly for ConvS2S and logarithmically for ByteNet. This makes it more difficult to learn dependencies between distant positions \citep{hochreiter2001gradient}. In the Transformer this is reduced to a constant number of operations, albeit at the cost of reduced effective resolution due to averaging attention-weighted positions, an effect we counteract with Multi-Head Attention as described in section~\ref{sec:attention}.
2
+
3
+ Self-attention, sometimes called intra-attention is an attention mechanism relating different positions of a single sequence in order to compute a representation of the sequence. Self-attention has been used successfully in a variety of tasks including reading comprehension, abstractive summarization, textual entailment and learning task-independent sentence representations \citep{cheng2016long, decomposableAttnModel, paulus2017deep, lin2017structured}.
4
+
5
+ End-to-end memory networks are based on a recurrent attention mechanism instead of sequence-aligned recurrence and have been shown to perform well on simple-language question answering and language modeling tasks \citep{sukhbaatar2015}.
6
+
7
+ To the best of our knowledge, however, the Transformer is the first transduction model relying entirely on self-attention to compute representations of its input and output without using sequence-aligned RNNs or convolution.
8
+ In the following sections, we will describe the Transformer, motivate self-attention and discuss its advantages over models such as \citep{neural_gpu, NalBytenet2017} and \citep{JonasFaceNet2017}.
9
+
10
+
11
+ %\citep{JonasFaceNet2017} report new SOTA on machine translation for English-to-German (EnDe), Enlish-to-French (EnFr) and English-to-Romanian language pairs.
12
+
13
+ %For example,! in MT, we must draw information from both input and previous output words to translate an output word accurately. An attention layer \citep{bahdanau2014neural} can connect a very large number of positions at low computation cost, making it an essential ingredient in competitive recurrent models for machine translation.
14
+
15
+ %A natural question to ask then is, "Could we replace recurrence with attention?". \marginpar{Don't know if it's the most natural question to ask given the previous statements. Also, need to say that the complexity table summarizes these statements} Such a model would be blessed with the computational efficiency of attention and the power of cross-positional communication. In this work, show that pure attention models work remarkably well for MT, achieving new SOTA results on EnDe and EnFr, and can be trained in under $2$ days on xyz architecture.
16
+
17
+ %After the seminal models introduced in \citep{sutskever14, bahdanau2014neural, cho2014learning}, recurrent models have become the dominant solution for both sequence modeling and sequence-to-sequence transduction. Many efforts such as \citep{wu2016google,luong2015effective,jozefowicz2016exploring} have pushed the boundaries of machine translation (MT) and language modeling with recurrent endoder-decoder and recurrent language models. Recent effort \citep{shazeer2017outrageously} has successfully combined the power of conditional computation with sequence models to train very large models for MT, pushing SOTA at lower computational cost.
18
+
19
+ %Recurrent models compute a vector of hidden states $h_t$, for each time step $t$ of computation. $h_t$ is a function of both the input at time $t$ and the previous hidden state $h_t$. This dependence on the previous hidden state precludes processing all timesteps at once, instead requiring long sequences of sequential operations. In practice, this results in greatly reduced computational efficiency, as on modern computing hardware, a single operation on a large batch is much faster than a large number of operations on small batches. The problem gets worse at longer sequence lengths. Although sequential computation is not a severe bottleneck at inference time, as autoregressively generating each output requires all previous outputs, the inability to compute scores at all output positions at once hinders us from rapidly training our models over large datasets. Although impressive work such as \citep{Kuchaiev2017Factorization} is able to significantly accelerate the training of LSTMs with factorization tricks, we are still bound by the linear dependence on sequence length.
20
+
21
+ %If the model could compute hidden states at each time step using only the inputs and outputs, it would be liberated from the dependence on results from previous time steps during training. This line of thought is the foundation of recent efforts such as the Markovian neural GPU \citep{neural_gpu}, ByteNet \citep{NalBytenet2017} and ConvS2S \citep{JonasFaceNet2017}, all of which use convolutional neural networks as a building block to compute hidden representations simultaneously for all timesteps, resulting in $O(1)$ sequential time complexity. \citep{JonasFaceNet2017} report new SOTA on machine translation for English-to-German (EnDe), Enlish-to-French (EnFr) and English-to-Romanian language pairs.
22
+
23
+ %A crucial component for accurate sequence prediction is modeling cross-positional communication. For example, in MT, we must draw information from both input and previous output words to translate an output word accurately. An attention layer \citep{bahdanau2014neural} can connect a very large number of positions at a low computation cost, also $O(1)$ sequential time complexity, making it an essential ingredient in recurrent encoder-decoder architectures for MT. A natural question to ask then is, "Could we replace recurrence with attention?". \marginpar{Don't know if it's the most natural question to ask given the previous statements. Also, need to say that the complexity table summarizes these statements} Such a model would be blessed with the computational efficiency of attention and the power of cross-positional communication. In this work, show that pure attention models work remarkably well for MT, achieving new SOTA results on EnDe and EnFr, and can be trained in under $2$ days on xyz architecture.
24
+
25
+
26
+
27
+ %Note: Facebook model is no better than RNNs in this regard, since it requires a number of layers proportional to the distance you want to communicate. Bytenet is more promising, since it requires a logarithmnic number of layers (does bytenet have SOTA results)?
28
+
29
+ %Note: An attention layer can connect a very large number of positions at a low computation cost in O(1) sequential operations. This is why encoder-decoder attention has been so successful in seq-to-seq models so far. It is only natural, then, to also use attention to connect the timesteps of the same sequence.
30
+
31
+ %Note: I wouldn't say that long sequences are not a problem during inference. It would be great if we could infer with no long sequences. We could just say later on that, while our training graph is constant-depth, our model still requires sequential operations in the decoder part during inference due to the autoregressive nature of the model.
32
+
33
+ %\begin{table}[h!]
34
+ %\caption{Attention models are quite efficient for cross-positional communications when sequence length is smaller than channel depth. $n$ represents the sequence length and $d$ represents the channel depth.}
35
+ %\label{tab:op_complexities}
36
+ %\begin{center}
37
+ %\vspace{-5pt}
38
+ %\scalebox{0.75}{
39
+
40
+ %\begin{tabular}{l|c|c|c}
41
+ %\hline \hline
42
+ %Layer Type & Receptive & Complexity & Sequential \\
43
+ % & Field & & Operations \\
44
+ %\hline
45
+ %Pointwise Feed-Forward & $1$ & $O(n \cdot d^2)$ & $O(1)$ \\
46
+ %\hline
47
+ %Recurrent & $n$ & $O(n \cdot d^2)$ & $O(n)$ \\
48
+ %\hline
49
+ %Convolutional & $r$ & $O(r \cdot n \cdot d^2)$ & $O(1)$ \\
50
+ %\hline
51
+ %Convolutional (separable) & $r$ & $O(r \cdot n \cdot d + n %\cdot d^2)$ & $O(1)$ \\
52
+ %\hline
53
+ %Attention & $r$ & $O(r \cdot n \cdot d)$ & $O(1)$ \\
54
+ %\hline \hline
55
+ %\end{tabular}
56
+ %}
57
+ %\end{center}
58
+ %\end{table}
crazy_functions/test_project/latex/attention/introduction.tex ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Recurrent neural networks, long short-term memory \citep{hochreiter1997} and gated recurrent \citep{gruEval14} neural networks in particular, have been firmly established as state of the art approaches in sequence modeling and transduction problems such as language modeling and machine translation \citep{sutskever14, bahdanau2014neural, cho2014learning}. Numerous efforts have since continued to push the boundaries of recurrent language models and encoder-decoder architectures \citep{wu2016google,luong2015effective,jozefowicz2016exploring}.
2
+
3
+ Recurrent models typically factor computation along the symbol positions of the input and output sequences. Aligning the positions to steps in computation time, they generate a sequence of hidden states $h_t$, as a function of the previous hidden state $h_{t-1}$ and the input for position $t$. This inherently sequential nature precludes parallelization within training examples, which becomes critical at longer sequence lengths, as memory constraints limit batching across examples.
4
+ %\marginpar{not sure if the memory constraints are understandable here}
5
+ Recent work has achieved significant improvements in computational efficiency through factorization tricks \citep{Kuchaiev2017Factorization} and conditional computation \citep{shazeer2017outrageously}, while also improving model performance in case of the latter. The fundamental constraint of sequential computation, however, remains.
6
+
7
+ %\marginpar{@all: there is work on analyzing what attention really does in seq2seq models, couldn't find it right away}
8
+
9
+ Attention mechanisms have become an integral part of compelling sequence modeling and transduction models in various tasks, allowing modeling of dependencies without regard to their distance in the input or output sequences \citep{bahdanau2014neural, structuredAttentionNetworks}. In all but a few cases \citep{decomposableAttnModel}, however, such attention mechanisms are used in conjunction with a recurrent network.
10
+
11
+ %\marginpar{not sure if "cross-positional communication" is understandable without explanation}
12
+ %\marginpar{insert exact training times and stats for the model that reaches sota earliest, maybe even a single GPU model?}
13
+
14
+ In this work we propose the Transformer, a model architecture eschewing recurrence and instead relying entirely on an attention mechanism to draw global dependencies between input and output. The Transformer allows for significantly more parallelization and can reach a new state of the art in translation quality after being trained for as little as twelve hours on eight P100 GPUs.
15
+ %\marginpar{you removed the constant number of repetitions part. I wrote it because I wanted to make it clear that the model does not only perform attention once, while it's also not recurrent. I thought that might be important to get across early.}
16
+
17
+ % Just a standard paragraph with citations, rewrite.
18
+ %After the seminal papers of \citep{sutskever14}, \citep{bahdanau2014neural}, and \citep{cho2014learning}, recurrent models have become the dominant solution for both sequence modeling and sequence-to-sequence transduction. Many efforts such as \citep{wu2016google,luong2015effective,jozefowicz2016exploring} have pushed the boundaries of machine translation and language modeling with recurrent sequence models. Recent effort \citep{shazeer2017outrageously} has combined the power of conditional computation with sequence models to train very large models for machine translation, pushing SOTA at lower computational cost. Recurrent models compute a vector of hidden states $h_t$, for each time step $t$ of computation. $h_t$ is a function of both the input at time $t$ and the previous hidden state $h_t$. This dependence on the previous hidden state encumbers recurrnet models to process multiple inputs at once, and their time complexity is a linear function of the length of the input and output, both during training and inference. [What I want to say here is that although this is fine during decoding, at training time, we are given both input and output and this linear nature does not allow the RNN to process all inputs and outputs simultaneously and haven't been used on datasets that are the of the scale of the web. What's the largest dataset we have ? . Talk about Nividia and possibly other's effors to speed up things, and possibly other efforts that alleviate this, but are still limited by it's comptuational nature]. Rest of the intro: What if you could construct the state based on the actual inputs and outputs, then you could construct them all at once. This has been the foundation of many promising recent efforts, bytenet,facenet (Also talk about quasi rnn here). Now we talk about attention!! Along with cell architectures such as long short-term meory (LSTM) \citep{hochreiter1997}, and gated recurrent units (GRUs) \citep{cho2014learning}, attention has emerged as an essential ingredient in successful sequence models, in particular for machine translation. In recent years, many, if not all, state-of-the-art (SOTA) results in machine translation have been achieved with attention-based sequence models \citep{wu2016google,luong2015effective,jozefowicz2016exploring}. Talk about the neon work on how it played with attention to do self attention! Then talk about what we do.