Delete folder stable-diffusion.cpp with huggingface_hub
Browse files- stable-diffusion.cpp/.dockerignore +0 -6
- stable-diffusion.cpp/.github/workflows/build.yml +0 -201
- stable-diffusion.cpp/.gitignore +0 -5
- stable-diffusion.cpp/.gitmodules +0 -3
- stable-diffusion.cpp/CMakeLists.txt +0 -45
- stable-diffusion.cpp/Dockerfile +0 -17
- stable-diffusion.cpp/LICENSE +0 -21
- stable-diffusion.cpp/README.md +0 -198
- stable-diffusion.cpp/assets/a lovely cat.png +0 -0
- stable-diffusion.cpp/assets/f16.png +0 -0
- stable-diffusion.cpp/assets/f32.png +0 -0
- stable-diffusion.cpp/assets/img2img_output.png +0 -0
- stable-diffusion.cpp/assets/q4_0.png +0 -0
- stable-diffusion.cpp/assets/q4_1.png +0 -0
- stable-diffusion.cpp/assets/q5_0.png +0 -0
- stable-diffusion.cpp/assets/q5_1.png +0 -0
- stable-diffusion.cpp/assets/q8_0.png +0 -0
- stable-diffusion.cpp/examples/CMakeLists.txt +0 -8
- stable-diffusion.cpp/examples/main.cpp +0 -473
- stable-diffusion.cpp/examples/stb_image.h +0 -0
- stable-diffusion.cpp/examples/stb_image_write.h +0 -1741
- stable-diffusion.cpp/models/.gitignore +0 -5
- stable-diffusion.cpp/models/README.md +0 -26
- stable-diffusion.cpp/models/convert.py +0 -385
- stable-diffusion.cpp/models/requirements.txt +0 -4
- stable-diffusion.cpp/models/vocab.json +0 -0
- stable-diffusion.cpp/rng.h +0 -35
- stable-diffusion.cpp/rng_philox.h +0 -125
- stable-diffusion.cpp/stable-diffusion.cpp +0 -0
- stable-diffusion.cpp/stable-diffusion.h +0 -74
stable-diffusion.cpp/.dockerignore
DELETED
@@ -1,6 +0,0 @@
|
|
1 |
-
build*/
|
2 |
-
test/
|
3 |
-
|
4 |
-
.cache/
|
5 |
-
*.swp
|
6 |
-
models/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/.github/workflows/build.yml
DELETED
@@ -1,201 +0,0 @@
|
|
1 |
-
name: CI
|
2 |
-
|
3 |
-
on:
|
4 |
-
workflow_dispatch: # allows manual triggering
|
5 |
-
inputs:
|
6 |
-
create_release:
|
7 |
-
description: 'Create new release'
|
8 |
-
required: true
|
9 |
-
type: boolean
|
10 |
-
push:
|
11 |
-
branches:
|
12 |
-
- master
|
13 |
-
- ci
|
14 |
-
paths: ['.github/workflows/**', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu']
|
15 |
-
pull_request:
|
16 |
-
types: [opened, synchronize, reopened]
|
17 |
-
paths: ['**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu']
|
18 |
-
|
19 |
-
env:
|
20 |
-
BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
|
21 |
-
|
22 |
-
jobs:
|
23 |
-
ubuntu-latest-cmake:
|
24 |
-
runs-on: ubuntu-latest
|
25 |
-
|
26 |
-
steps:
|
27 |
-
- name: Clone
|
28 |
-
id: checkout
|
29 |
-
uses: actions/checkout@v3
|
30 |
-
with:
|
31 |
-
submodules: recursive
|
32 |
-
|
33 |
-
|
34 |
-
- name: Dependencies
|
35 |
-
id: depends
|
36 |
-
run: |
|
37 |
-
sudo apt-get update
|
38 |
-
sudo apt-get install build-essential
|
39 |
-
|
40 |
-
- name: Build
|
41 |
-
id: cmake_build
|
42 |
-
run: |
|
43 |
-
mkdir build
|
44 |
-
cd build
|
45 |
-
cmake ..
|
46 |
-
cmake --build . --config Release
|
47 |
-
|
48 |
-
#- name: Test
|
49 |
-
#id: cmake_test
|
50 |
-
#run: |
|
51 |
-
#cd build
|
52 |
-
#ctest --verbose --timeout 900
|
53 |
-
|
54 |
-
macOS-latest-cmake:
|
55 |
-
runs-on: macos-latest
|
56 |
-
|
57 |
-
steps:
|
58 |
-
- name: Clone
|
59 |
-
id: checkout
|
60 |
-
uses: actions/checkout@v3
|
61 |
-
with:
|
62 |
-
submodules: recursive
|
63 |
-
|
64 |
-
- name: Dependencies
|
65 |
-
id: depends
|
66 |
-
continue-on-error: true
|
67 |
-
run: |
|
68 |
-
brew update
|
69 |
-
|
70 |
-
- name: Build
|
71 |
-
id: cmake_build
|
72 |
-
run: |
|
73 |
-
sysctl -a
|
74 |
-
mkdir build
|
75 |
-
cd build
|
76 |
-
cmake ..
|
77 |
-
cmake --build . --config Release
|
78 |
-
|
79 |
-
#- name: Test
|
80 |
-
#id: cmake_test
|
81 |
-
#run: |
|
82 |
-
#cd build
|
83 |
-
#ctest --verbose --timeout 900
|
84 |
-
|
85 |
-
windows-latest-cmake:
|
86 |
-
runs-on: windows-latest
|
87 |
-
|
88 |
-
strategy:
|
89 |
-
matrix:
|
90 |
-
include:
|
91 |
-
- build: 'noavx'
|
92 |
-
defines: '-DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF'
|
93 |
-
- build: 'avx2'
|
94 |
-
defines: '-DGGML_AVX2=ON'
|
95 |
-
- build: 'avx'
|
96 |
-
defines: '-DGGML_AVX2=OFF'
|
97 |
-
- build: 'avx512'
|
98 |
-
defines: '-DGGML_AVX512=ON'
|
99 |
-
|
100 |
-
steps:
|
101 |
-
- name: Clone
|
102 |
-
id: checkout
|
103 |
-
uses: actions/checkout@v3
|
104 |
-
with:
|
105 |
-
submodules: recursive
|
106 |
-
|
107 |
-
- name: Build
|
108 |
-
id: cmake_build
|
109 |
-
run: |
|
110 |
-
mkdir build
|
111 |
-
cd build
|
112 |
-
cmake .. ${{ matrix.defines }}
|
113 |
-
cmake --build . --config Release
|
114 |
-
|
115 |
-
- name: Check AVX512F support
|
116 |
-
id: check_avx512f
|
117 |
-
if: ${{ matrix.build == 'avx512' }}
|
118 |
-
continue-on-error: true
|
119 |
-
run: |
|
120 |
-
cd build
|
121 |
-
$vcdir = $(vswhere -latest -products * -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath)
|
122 |
-
$msvc = $(join-path $vcdir $('VC\Tools\MSVC\'+$(gc -raw $(join-path $vcdir 'VC\Auxiliary\Build\Microsoft.VCToolsVersion.default.txt')).Trim()))
|
123 |
-
$cl = $(join-path $msvc 'bin\Hostx64\x64\cl.exe')
|
124 |
-
echo 'int main(void){unsigned int a[4];__cpuid(a,7);return !(a[1]&65536);}' >> avx512f.c
|
125 |
-
& $cl /O2 /GS- /kernel avx512f.c /link /nodefaultlib /entry:main
|
126 |
-
.\avx512f.exe && echo "AVX512F: YES" && ( echo HAS_AVX512F=1 >> $env:GITHUB_ENV ) || echo "AVX512F: NO"
|
127 |
-
|
128 |
-
#- name: Test
|
129 |
-
#id: cmake_test
|
130 |
-
#run: |
|
131 |
-
#cd build
|
132 |
-
#ctest -C Release --verbose --timeout 900
|
133 |
-
|
134 |
-
- name: Get commit hash
|
135 |
-
id: commit
|
136 |
-
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
|
137 |
-
uses: pr-mpt/actions-commit-hash@v2
|
138 |
-
|
139 |
-
- name: Pack artifacts
|
140 |
-
id: pack_artifacts
|
141 |
-
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
|
142 |
-
run: |
|
143 |
-
Copy-Item ggml/LICENSE .\build\bin\Release\ggml.txt
|
144 |
-
Copy-Item LICENSE .\build\bin\Release\stable-diffusion.cpp.txt
|
145 |
-
7z a sd-${{ env.BRANCH_NAME }}-${{ steps.commit.outputs.short }}-bin-win-${{ matrix.build }}-x64.zip .\build\bin\Release\*
|
146 |
-
|
147 |
-
- name: Upload artifacts
|
148 |
-
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
|
149 |
-
uses: actions/upload-artifact@v3
|
150 |
-
with:
|
151 |
-
path: |
|
152 |
-
sd-${{ env.BRANCH_NAME }}-${{ steps.commit.outputs.short }}-bin-win-${{ matrix.build }}-x64.zip
|
153 |
-
|
154 |
-
release:
|
155 |
-
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
|
156 |
-
|
157 |
-
runs-on: ubuntu-latest
|
158 |
-
|
159 |
-
needs:
|
160 |
-
- ubuntu-latest-cmake
|
161 |
-
- macOS-latest-cmake
|
162 |
-
- windows-latest-cmake
|
163 |
-
|
164 |
-
steps:
|
165 |
-
- name: Download artifacts
|
166 |
-
id: download-artifact
|
167 |
-
uses: actions/download-artifact@v3
|
168 |
-
|
169 |
-
- name: Get commit hash
|
170 |
-
id: commit
|
171 |
-
uses: pr-mpt/actions-commit-hash@v2
|
172 |
-
|
173 |
-
- name: Create release
|
174 |
-
id: create_release
|
175 |
-
uses: anzz1/action-create-release@v1
|
176 |
-
env:
|
177 |
-
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
178 |
-
with:
|
179 |
-
tag_name: ${{ env.BRANCH_NAME }}-${{ steps.commit.outputs.short }}
|
180 |
-
|
181 |
-
- name: Upload release
|
182 |
-
id: upload_release
|
183 |
-
uses: actions/github-script@v3
|
184 |
-
with:
|
185 |
-
github-token: ${{secrets.GITHUB_TOKEN}}
|
186 |
-
script: |
|
187 |
-
const path = require('path');
|
188 |
-
const fs = require('fs');
|
189 |
-
const release_id = '${{ steps.create_release.outputs.id }}';
|
190 |
-
for (let file of await fs.readdirSync('./artifact')) {
|
191 |
-
if (path.extname(file) === '.zip') {
|
192 |
-
console.log('uploadReleaseAsset', file);
|
193 |
-
await github.repos.uploadReleaseAsset({
|
194 |
-
owner: context.repo.owner,
|
195 |
-
repo: context.repo.repo,
|
196 |
-
release_id: release_id,
|
197 |
-
name: file,
|
198 |
-
data: await fs.readFileSync(`./artifact/${file}`)
|
199 |
-
});
|
200 |
-
}
|
201 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/.gitignore
DELETED
@@ -1,5 +0,0 @@
|
|
1 |
-
build*/
|
2 |
-
test/
|
3 |
-
|
4 |
-
.cache/
|
5 |
-
*.swp
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/.gitmodules
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
[submodule "ggml"]
|
2 |
-
path = ggml
|
3 |
-
url = https://github.com/leejet/ggml.git
|
|
|
|
|
|
|
|
stable-diffusion.cpp/CMakeLists.txt
DELETED
@@ -1,45 +0,0 @@
|
|
1 |
-
cmake_minimum_required(VERSION 3.12)
|
2 |
-
project("stable-diffusion")
|
3 |
-
|
4 |
-
set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
|
5 |
-
|
6 |
-
if (NOT XCODE AND NOT MSVC AND NOT CMAKE_BUILD_TYPE)
|
7 |
-
set(CMAKE_BUILD_TYPE Release CACHE STRING "Build type" FORCE)
|
8 |
-
set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release" "MinSizeRel" "RelWithDebInfo")
|
9 |
-
endif()
|
10 |
-
|
11 |
-
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
|
12 |
-
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
|
13 |
-
|
14 |
-
if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
|
15 |
-
set(SD_STANDALONE ON)
|
16 |
-
else()
|
17 |
-
set(SD_STANDALONE OFF)
|
18 |
-
endif()
|
19 |
-
|
20 |
-
#
|
21 |
-
# Option list
|
22 |
-
#
|
23 |
-
|
24 |
-
# general
|
25 |
-
#option(SD_BUILD_TESTS "sd: build tests" ${SD_STANDALONE})
|
26 |
-
option(SD_BUILD_EXAMPLES "sd: build examples" ${SD_STANDALONE})
|
27 |
-
option(BUILD_SHARED_LIBS "sd: build shared libs" OFF)
|
28 |
-
#option(SD_BUILD_SERVER "sd: build server example" ON)
|
29 |
-
|
30 |
-
|
31 |
-
# deps
|
32 |
-
add_subdirectory(ggml)
|
33 |
-
|
34 |
-
set(SD_LIB stable-diffusion)
|
35 |
-
|
36 |
-
add_library(${SD_LIB} stable-diffusion.h stable-diffusion.cpp)
|
37 |
-
target_link_libraries(${SD_LIB} PUBLIC ggml)
|
38 |
-
target_include_directories(${SD_LIB} PUBLIC .)
|
39 |
-
target_compile_features(${SD_LIB} PUBLIC cxx_std_11)
|
40 |
-
|
41 |
-
|
42 |
-
if (SD_BUILD_EXAMPLES)
|
43 |
-
add_subdirectory(examples)
|
44 |
-
endif()
|
45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/Dockerfile
DELETED
@@ -1,17 +0,0 @@
|
|
1 |
-
ARG UBUNTU_VERSION=22.04
|
2 |
-
|
3 |
-
FROM ubuntu:$UBUNTU_VERSION as build
|
4 |
-
|
5 |
-
RUN apt-get update && apt-get install -y build-essential git cmake
|
6 |
-
|
7 |
-
WORKDIR /sd.cpp
|
8 |
-
|
9 |
-
COPY . .
|
10 |
-
|
11 |
-
RUN mkdir build && cd build && cmake .. && cmake --build . --config Release
|
12 |
-
|
13 |
-
FROM ubuntu:$UBUNTU_VERSION as runtime
|
14 |
-
|
15 |
-
COPY --from=build /sd.cpp/build/bin/sd /sd
|
16 |
-
|
17 |
-
ENTRYPOINT [ "/sd" ]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/LICENSE
DELETED
@@ -1,21 +0,0 @@
|
|
1 |
-
MIT License
|
2 |
-
|
3 |
-
Copyright (c) 2023 leejet
|
4 |
-
|
5 |
-
Permission is hereby granted, free of charge, to any person obtaining a copy
|
6 |
-
of this software and associated documentation files (the "Software"), to deal
|
7 |
-
in the Software without restriction, including without limitation the rights
|
8 |
-
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
9 |
-
copies of the Software, and to permit persons to whom the Software is
|
10 |
-
furnished to do so, subject to the following conditions:
|
11 |
-
|
12 |
-
The above copyright notice and this permission notice shall be included in all
|
13 |
-
copies or substantial portions of the Software.
|
14 |
-
|
15 |
-
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
16 |
-
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
17 |
-
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
18 |
-
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
19 |
-
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
20 |
-
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
21 |
-
SOFTWARE.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/README.md
DELETED
@@ -1,198 +0,0 @@
|
|
1 |
-
<p align="center">
|
2 |
-
<img src="./assets/a%20lovely%20cat.png" width="256x">
|
3 |
-
</p>
|
4 |
-
|
5 |
-
# stable-diffusion.cpp
|
6 |
-
|
7 |
-
Inference of [Stable Diffusion](https://github.com/CompVis/stable-diffusion) in pure C/C++
|
8 |
-
|
9 |
-
## Features
|
10 |
-
|
11 |
-
- Plain C/C++ implementation based on [ggml](https://github.com/ggerganov/ggml), working in the same way as [llama.cpp](https://github.com/ggerganov/llama.cpp)
|
12 |
-
- 16-bit, 32-bit float support
|
13 |
-
- 4-bit, 5-bit and 8-bit integer quantization support
|
14 |
-
- Accelerated memory-efficient CPU inference
|
15 |
-
- Only requires ~2.3GB when using txt2img with fp16 precision to generate a 512x512 image
|
16 |
-
- AVX, AVX2 and AVX512 support for x86 architectures
|
17 |
-
- SD1.x and SD2.x support
|
18 |
-
- Original `txt2img` and `img2img` mode
|
19 |
-
- Negative prompt
|
20 |
-
- [stable-diffusion-webui](https://github.com/AUTOMATIC1111/stable-diffusion-webui) style tokenizer (not all the features, only token weighting for now)
|
21 |
-
- Sampling method
|
22 |
-
- `Euler A`
|
23 |
-
- `Euler`
|
24 |
-
- `Heun`
|
25 |
-
- `DPM2`
|
26 |
-
- `DPM++ 2M`
|
27 |
-
- [`DPM++ 2M v2`](https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/8457)
|
28 |
-
- `DPM++ 2S a`
|
29 |
-
- Cross-platform reproducibility (`--rng cuda`, consistent with the `stable-diffusion-webui GPU RNG`)
|
30 |
-
- Embedds generation parameters into png output as webui-compatible text string
|
31 |
-
- Supported platforms
|
32 |
-
- Linux
|
33 |
-
- Mac OS
|
34 |
-
- Windows
|
35 |
-
- Android (via Termux)
|
36 |
-
|
37 |
-
### TODO
|
38 |
-
|
39 |
-
- [ ] More sampling methods
|
40 |
-
- [ ] GPU support
|
41 |
-
- [ ] Make inference faster
|
42 |
-
- The current implementation of ggml_conv_2d is slow and has high memory usage
|
43 |
-
- [ ] Continuing to reduce memory usage (quantizing the weights of ggml_conv_2d)
|
44 |
-
- [ ] LoRA support
|
45 |
-
- [ ] k-quants support
|
46 |
-
|
47 |
-
## Usage
|
48 |
-
|
49 |
-
### Get the Code
|
50 |
-
|
51 |
-
```
|
52 |
-
git clone --recursive https://github.com/leejet/stable-diffusion.cpp
|
53 |
-
cd stable-diffusion.cpp
|
54 |
-
```
|
55 |
-
|
56 |
-
- If you have already cloned the repository, you can use the following command to update the repository to the latest code.
|
57 |
-
|
58 |
-
```
|
59 |
-
cd stable-diffusion.cpp
|
60 |
-
git pull origin master
|
61 |
-
git submodule init
|
62 |
-
git submodule update
|
63 |
-
```
|
64 |
-
|
65 |
-
### Convert weights
|
66 |
-
|
67 |
-
- download original weights(.ckpt or .safetensors). For example
|
68 |
-
- Stable Diffusion v1.4 from https://huggingface.co/CompVis/stable-diffusion-v-1-4-original
|
69 |
-
- Stable Diffusion v1.5 from https://huggingface.co/runwayml/stable-diffusion-v1-5
|
70 |
-
- Stable Diffuison v2.1 from https://huggingface.co/stabilityai/stable-diffusion-2-1
|
71 |
-
|
72 |
-
```shell
|
73 |
-
curl -L -O https://huggingface.co/CompVis/stable-diffusion-v-1-4-original/resolve/main/sd-v1-4.ckpt
|
74 |
-
# curl -L -O https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors
|
75 |
-
# curl -L -O https://huggingface.co/stabilityai/stable-diffusion-2-1/blob/main/v2-1_768-nonema-pruned.safetensors
|
76 |
-
```
|
77 |
-
|
78 |
-
- convert weights to ggml model format
|
79 |
-
|
80 |
-
```shell
|
81 |
-
cd models
|
82 |
-
pip install -r requirements.txt
|
83 |
-
python convert.py [path to weights] --out_type [output precision]
|
84 |
-
# For example, python convert.py sd-v1-4.ckpt --out_type f16
|
85 |
-
```
|
86 |
-
|
87 |
-
### Quantization
|
88 |
-
|
89 |
-
You can specify the output model format using the --out_type parameter
|
90 |
-
|
91 |
-
- `f16` for 16-bit floating-point
|
92 |
-
- `f32` for 32-bit floating-point
|
93 |
-
- `q8_0` for 8-bit integer quantization
|
94 |
-
- `q5_0` or `q5_1` for 5-bit integer quantization
|
95 |
-
- `q4_0` or `q4_1` for 4-bit integer quantization
|
96 |
-
|
97 |
-
### Build
|
98 |
-
|
99 |
-
#### Build from scratch
|
100 |
-
|
101 |
-
```shell
|
102 |
-
mkdir build
|
103 |
-
cd build
|
104 |
-
cmake ..
|
105 |
-
cmake --build . --config Release
|
106 |
-
```
|
107 |
-
|
108 |
-
##### Using OpenBLAS
|
109 |
-
|
110 |
-
```
|
111 |
-
cmake .. -DGGML_OPENBLAS=ON
|
112 |
-
cmake --build . --config Release
|
113 |
-
```
|
114 |
-
|
115 |
-
### Run
|
116 |
-
|
117 |
-
```
|
118 |
-
usage: ./bin/sd [arguments]
|
119 |
-
|
120 |
-
arguments:
|
121 |
-
-h, --help show this help message and exit
|
122 |
-
-M, --mode [txt2img or img2img] generation mode (default: txt2img)
|
123 |
-
-t, --threads N number of threads to use during computation (default: -1).
|
124 |
-
If threads <= 0, then threads will be set to the number of CPU physical cores
|
125 |
-
-m, --model [MODEL] path to model
|
126 |
-
-i, --init-img [IMAGE] path to the input image, required by img2img
|
127 |
-
-o, --output OUTPUT path to write result image to (default: .\output.png)
|
128 |
-
-p, --prompt [PROMPT] the prompt to render
|
129 |
-
-n, --negative-prompt PROMPT the negative prompt (default: "")
|
130 |
-
--cfg-scale SCALE unconditional guidance scale: (default: 7.0)
|
131 |
-
--strength STRENGTH strength for noising/unnoising (default: 0.75)
|
132 |
-
1.0 corresponds to full destruction of information in init image
|
133 |
-
-H, --height H image height, in pixel space (default: 512)
|
134 |
-
-W, --width W image width, in pixel space (default: 512)
|
135 |
-
--sampling-method {euler, euler_a, heun, dpm++2m, dpm++2mv2}
|
136 |
-
sampling method (default: "euler_a")
|
137 |
-
--steps STEPS number of sample steps (default: 20)
|
138 |
-
--rng {std_default, cuda} RNG (default: cuda)
|
139 |
-
-s SEED, --seed SEED RNG seed (default: 42, use random seed for < 0)
|
140 |
-
-v, --verbose print extra info
|
141 |
-
```
|
142 |
-
|
143 |
-
#### txt2img example
|
144 |
-
|
145 |
-
```
|
146 |
-
./bin/sd -m ../models/sd-v1-4-ggml-model-f16.bin -p "a lovely cat"
|
147 |
-
```
|
148 |
-
|
149 |
-
Using formats of different precisions will yield results of varying quality.
|
150 |
-
|
151 |
-
| f32 | f16 |q8_0 |q5_0 |q5_1 |q4_0 |q4_1 |
|
152 |
-
| ---- |---- |---- |---- |---- |---- |---- |
|
153 |
-
| ![](./assets/f32.png) |![](./assets/f16.png) |![](./assets/q8_0.png) |![](./assets/q5_0.png) |![](./assets/q5_1.png) |![](./assets/q4_0.png) |![](./assets/q4_1.png) |
|
154 |
-
|
155 |
-
#### img2img example
|
156 |
-
|
157 |
-
- `./output.png` is the image generated from the above txt2img pipeline
|
158 |
-
|
159 |
-
|
160 |
-
```
|
161 |
-
./bin/sd --mode img2img -m ../models/sd-v1-4-ggml-model-f16.bin -p "cat with blue eyes" -i ./output.png -o ./img2img_output.png --strength 0.4
|
162 |
-
```
|
163 |
-
|
164 |
-
<p align="center">
|
165 |
-
<img src="./assets/img2img_output.png" width="256x">
|
166 |
-
</p>
|
167 |
-
|
168 |
-
### Docker
|
169 |
-
|
170 |
-
#### Building using Docker
|
171 |
-
|
172 |
-
```shell
|
173 |
-
docker build -t sd .
|
174 |
-
```
|
175 |
-
|
176 |
-
#### Run
|
177 |
-
|
178 |
-
```shell
|
179 |
-
docker run -v /path/to/models:/models -v /path/to/output/:/output sd [args...]
|
180 |
-
# For example
|
181 |
-
# docker run -v ./models:/models -v ./build:/output sd -m /models/sd-v1-4-ggml-model-f16.bin -p "a lovely cat" -v -o /output/output.png
|
182 |
-
```
|
183 |
-
|
184 |
-
## Memory/Disk Requirements
|
185 |
-
|
186 |
-
| precision | f32 | f16 |q8_0 |q5_0 |q5_1 |q4_0 |q4_1 |
|
187 |
-
| ---- | ---- |---- |---- |---- |---- |---- |---- |
|
188 |
-
| **Disk** | 2.7G | 2.0G | 1.7G | 1.6G | 1.6G | 1.5G | 1.5G |
|
189 |
-
| **Memory**(txt2img - 512 x 512) | ~2.8G | ~2.3G | ~2.1G | ~2.0G | ~2.0G | ~2.0G | ~2.0G |
|
190 |
-
|
191 |
-
|
192 |
-
## References
|
193 |
-
|
194 |
-
- [ggml](https://github.com/ggerganov/ggml)
|
195 |
-
- [stable-diffusion](https://github.com/CompVis/stable-diffusion)
|
196 |
-
- [stable-diffusion-stability-ai](https://github.com/Stability-AI/stablediffusion)
|
197 |
-
- [stable-diffusion-webui](https://github.com/AUTOMATIC1111/stable-diffusion-webui)
|
198 |
-
- [k-diffusion](https://github.com/crowsonkb/k-diffusion)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/assets/a lovely cat.png
DELETED
Binary file (695 kB)
|
|
stable-diffusion.cpp/assets/f16.png
DELETED
Binary file (695 kB)
|
|
stable-diffusion.cpp/assets/f32.png
DELETED
Binary file (695 kB)
|
|
stable-diffusion.cpp/assets/img2img_output.png
DELETED
Binary file (601 kB)
|
|
stable-diffusion.cpp/assets/q4_0.png
DELETED
Binary file (665 kB)
|
|
stable-diffusion.cpp/assets/q4_1.png
DELETED
Binary file (681 kB)
|
|
stable-diffusion.cpp/assets/q5_0.png
DELETED
Binary file (682 kB)
|
|
stable-diffusion.cpp/assets/q5_1.png
DELETED
Binary file (674 kB)
|
|
stable-diffusion.cpp/assets/q8_0.png
DELETED
Binary file (674 kB)
|
|
stable-diffusion.cpp/examples/CMakeLists.txt
DELETED
@@ -1,8 +0,0 @@
|
|
1 |
-
# TODO: move into its own subdirectoy
|
2 |
-
# TODO: make stb libs a target (maybe common)
|
3 |
-
set(SD_TARGET sd)
|
4 |
-
|
5 |
-
add_executable(${SD_TARGET} main.cpp stb_image.h stb_image_write.h)
|
6 |
-
install(TARGETS ${SD_TARGET} RUNTIME)
|
7 |
-
target_link_libraries(${SD_TARGET} PRIVATE stable-diffusion ${CMAKE_THREAD_LIBS_INIT})
|
8 |
-
target_compile_features(${SD_TARGET} PUBLIC cxx_std_11)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/examples/main.cpp
DELETED
@@ -1,473 +0,0 @@
|
|
1 |
-
#include <stdio.h>
|
2 |
-
#include <ctime>
|
3 |
-
#include <fstream>
|
4 |
-
#include <iostream>
|
5 |
-
#include <random>
|
6 |
-
#include <string>
|
7 |
-
#include <thread>
|
8 |
-
#include <unordered_set>
|
9 |
-
|
10 |
-
#include "stable-diffusion.h"
|
11 |
-
|
12 |
-
#define STB_IMAGE_IMPLEMENTATION
|
13 |
-
#include "stb_image.h"
|
14 |
-
|
15 |
-
#define STB_IMAGE_WRITE_IMPLEMENTATION
|
16 |
-
#define STB_IMAGE_WRITE_STATIC
|
17 |
-
#include "stb_image_write.h"
|
18 |
-
|
19 |
-
#if defined(__APPLE__) && defined(__MACH__)
|
20 |
-
#include <sys/sysctl.h>
|
21 |
-
#include <sys/types.h>
|
22 |
-
#endif
|
23 |
-
|
24 |
-
#if !defined(_WIN32)
|
25 |
-
#include <sys/ioctl.h>
|
26 |
-
#include <unistd.h>
|
27 |
-
#endif
|
28 |
-
|
29 |
-
#define TXT2IMG "txt2img"
|
30 |
-
#define IMG2IMG "img2img"
|
31 |
-
|
32 |
-
// get_num_physical_cores is copy from
|
33 |
-
// https://github.com/ggerganov/llama.cpp/blob/master/examples/common.cpp
|
34 |
-
// LICENSE: https://github.com/ggerganov/llama.cpp/blob/master/LICENSE
|
35 |
-
int32_t get_num_physical_cores() {
|
36 |
-
#ifdef __linux__
|
37 |
-
// enumerate the set of thread siblings, num entries is num cores
|
38 |
-
std::unordered_set<std::string> siblings;
|
39 |
-
for (uint32_t cpu = 0; cpu < UINT32_MAX; ++cpu) {
|
40 |
-
std::ifstream thread_siblings("/sys/devices/system/cpu" + std::to_string(cpu) + "/topology/thread_siblings");
|
41 |
-
if (!thread_siblings.is_open()) {
|
42 |
-
break; // no more cpus
|
43 |
-
}
|
44 |
-
std::string line;
|
45 |
-
if (std::getline(thread_siblings, line)) {
|
46 |
-
siblings.insert(line);
|
47 |
-
}
|
48 |
-
}
|
49 |
-
if (siblings.size() > 0) {
|
50 |
-
return static_cast<int32_t>(siblings.size());
|
51 |
-
}
|
52 |
-
#elif defined(__APPLE__) && defined(__MACH__)
|
53 |
-
int32_t num_physical_cores;
|
54 |
-
size_t len = sizeof(num_physical_cores);
|
55 |
-
int result = sysctlbyname("hw.perflevel0.physicalcpu", &num_physical_cores, &len, NULL, 0);
|
56 |
-
if (result == 0) {
|
57 |
-
return num_physical_cores;
|
58 |
-
}
|
59 |
-
result = sysctlbyname("hw.physicalcpu", &num_physical_cores, &len, NULL, 0);
|
60 |
-
if (result == 0) {
|
61 |
-
return num_physical_cores;
|
62 |
-
}
|
63 |
-
#elif defined(_WIN32)
|
64 |
-
// TODO: Implement
|
65 |
-
#endif
|
66 |
-
unsigned int n_threads = std::thread::hardware_concurrency();
|
67 |
-
return n_threads > 0 ? (n_threads <= 4 ? n_threads : n_threads / 2) : 4;
|
68 |
-
}
|
69 |
-
|
70 |
-
const char* rng_type_to_str[] = {
|
71 |
-
"std_default",
|
72 |
-
"cuda",
|
73 |
-
};
|
74 |
-
|
75 |
-
// Names of the sampler method, same order as enum SampleMethod in stable-diffusion.h
|
76 |
-
const char* sample_method_str[] = {
|
77 |
-
"euler_a",
|
78 |
-
"euler",
|
79 |
-
"heun",
|
80 |
-
"dpm2",
|
81 |
-
"dpm++2s_a",
|
82 |
-
"dpm++2m",
|
83 |
-
"dpm++2mv2"};
|
84 |
-
|
85 |
-
// Names of the sigma schedule overrides, same order as Schedule in stable-diffusion.h
|
86 |
-
const char* schedule_str[] = {
|
87 |
-
"default",
|
88 |
-
"discrete",
|
89 |
-
"karras"};
|
90 |
-
|
91 |
-
struct Option {
|
92 |
-
int n_threads = -1;
|
93 |
-
std::string mode = TXT2IMG;
|
94 |
-
std::string model_path;
|
95 |
-
std::string output_path = "output.png";
|
96 |
-
std::string init_img;
|
97 |
-
std::string prompt;
|
98 |
-
std::string negative_prompt;
|
99 |
-
float cfg_scale = 7.0f;
|
100 |
-
int w = 512;
|
101 |
-
int h = 512;
|
102 |
-
SampleMethod sample_method = EULER_A;
|
103 |
-
Schedule schedule = DEFAULT;
|
104 |
-
int sample_steps = 20;
|
105 |
-
float strength = 0.75f;
|
106 |
-
RNGType rng_type = CUDA_RNG;
|
107 |
-
int64_t seed = 42;
|
108 |
-
bool verbose = false;
|
109 |
-
|
110 |
-
void print() {
|
111 |
-
printf("Option: \n");
|
112 |
-
printf(" n_threads: %d\n", n_threads);
|
113 |
-
printf(" mode: %s\n", mode.c_str());
|
114 |
-
printf(" model_path: %s\n", model_path.c_str());
|
115 |
-
printf(" output_path: %s\n", output_path.c_str());
|
116 |
-
printf(" init_img: %s\n", init_img.c_str());
|
117 |
-
printf(" prompt: %s\n", prompt.c_str());
|
118 |
-
printf(" negative_prompt: %s\n", negative_prompt.c_str());
|
119 |
-
printf(" cfg_scale: %.2f\n", cfg_scale);
|
120 |
-
printf(" width: %d\n", w);
|
121 |
-
printf(" height: %d\n", h);
|
122 |
-
printf(" sample_method: %s\n", sample_method_str[sample_method]);
|
123 |
-
printf(" schedule: %s\n", schedule_str[schedule]);
|
124 |
-
printf(" sample_steps: %d\n", sample_steps);
|
125 |
-
printf(" strength: %.2f\n", strength);
|
126 |
-
printf(" rng: %s\n", rng_type_to_str[rng_type]);
|
127 |
-
printf(" seed: %ld\n", seed);
|
128 |
-
}
|
129 |
-
};
|
130 |
-
|
131 |
-
void print_usage(int argc, const char* argv[]) {
|
132 |
-
printf("usage: %s [arguments]\n", argv[0]);
|
133 |
-
printf("\n");
|
134 |
-
printf("arguments:\n");
|
135 |
-
printf(" -h, --help show this help message and exit\n");
|
136 |
-
printf(" -M, --mode [txt2img or img2img] generation mode (default: txt2img)\n");
|
137 |
-
printf(" -t, --threads N number of threads to use during computation (default: -1).\n");
|
138 |
-
printf(" If threads <= 0, then threads will be set to the number of CPU physical cores\n");
|
139 |
-
printf(" -m, --model [MODEL] path to model\n");
|
140 |
-
printf(" -i, --init-img [IMAGE] path to the input image, required by img2img\n");
|
141 |
-
printf(" -o, --output OUTPUT path to write result image to (default: .\\output.png)\n");
|
142 |
-
printf(" -p, --prompt [PROMPT] the prompt to render\n");
|
143 |
-
printf(" -n, --negative-prompt PROMPT the negative prompt (default: \"\")\n");
|
144 |
-
printf(" --cfg-scale SCALE unconditional guidance scale: (default: 7.0)\n");
|
145 |
-
printf(" --strength STRENGTH strength for noising/unnoising (default: 0.75)\n");
|
146 |
-
printf(" 1.0 corresponds to full destruction of information in init image\n");
|
147 |
-
printf(" -H, --height H image height, in pixel space (default: 512)\n");
|
148 |
-
printf(" -W, --width W image width, in pixel space (default: 512)\n");
|
149 |
-
printf(" --sampling-method {euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2}\n");
|
150 |
-
printf(" sampling method (default: \"euler_a\")\n");
|
151 |
-
printf(" --steps STEPS number of sample steps (default: 20)\n");
|
152 |
-
printf(" --rng {std_default, cuda} RNG (default: cuda)\n");
|
153 |
-
printf(" -s SEED, --seed SEED RNG seed (default: 42, use random seed for < 0)\n");
|
154 |
-
printf(" --schedule {discrete, karras} Denoiser sigma schedule (default: discrete)\n");
|
155 |
-
printf(" -v, --verbose print extra info\n");
|
156 |
-
}
|
157 |
-
|
158 |
-
void parse_args(int argc, const char* argv[], Option* opt) {
|
159 |
-
bool invalid_arg = false;
|
160 |
-
|
161 |
-
for (int i = 1; i < argc; i++) {
|
162 |
-
std::string arg = argv[i];
|
163 |
-
|
164 |
-
if (arg == "-t" || arg == "--threads") {
|
165 |
-
if (++i >= argc) {
|
166 |
-
invalid_arg = true;
|
167 |
-
break;
|
168 |
-
}
|
169 |
-
opt->n_threads = std::stoi(argv[i]);
|
170 |
-
} else if (arg == "-M" || arg == "--mode") {
|
171 |
-
if (++i >= argc) {
|
172 |
-
invalid_arg = true;
|
173 |
-
break;
|
174 |
-
}
|
175 |
-
opt->mode = argv[i];
|
176 |
-
|
177 |
-
} else if (arg == "-m" || arg == "--model") {
|
178 |
-
if (++i >= argc) {
|
179 |
-
invalid_arg = true;
|
180 |
-
break;
|
181 |
-
}
|
182 |
-
opt->model_path = argv[i];
|
183 |
-
} else if (arg == "-i" || arg == "--init-img") {
|
184 |
-
if (++i >= argc) {
|
185 |
-
invalid_arg = true;
|
186 |
-
break;
|
187 |
-
}
|
188 |
-
opt->init_img = argv[i];
|
189 |
-
} else if (arg == "-o" || arg == "--output") {
|
190 |
-
if (++i >= argc) {
|
191 |
-
invalid_arg = true;
|
192 |
-
break;
|
193 |
-
}
|
194 |
-
opt->output_path = argv[i];
|
195 |
-
} else if (arg == "-p" || arg == "--prompt") {
|
196 |
-
if (++i >= argc) {
|
197 |
-
invalid_arg = true;
|
198 |
-
break;
|
199 |
-
}
|
200 |
-
opt->prompt = argv[i];
|
201 |
-
} else if (arg == "-n" || arg == "--negative-prompt") {
|
202 |
-
if (++i >= argc) {
|
203 |
-
invalid_arg = true;
|
204 |
-
break;
|
205 |
-
}
|
206 |
-
opt->negative_prompt = argv[i];
|
207 |
-
} else if (arg == "--cfg-scale") {
|
208 |
-
if (++i >= argc) {
|
209 |
-
invalid_arg = true;
|
210 |
-
break;
|
211 |
-
}
|
212 |
-
opt->cfg_scale = std::stof(argv[i]);
|
213 |
-
} else if (arg == "--strength") {
|
214 |
-
if (++i >= argc) {
|
215 |
-
invalid_arg = true;
|
216 |
-
break;
|
217 |
-
}
|
218 |
-
opt->strength = std::stof(argv[i]);
|
219 |
-
} else if (arg == "-H" || arg == "--height") {
|
220 |
-
if (++i >= argc) {
|
221 |
-
invalid_arg = true;
|
222 |
-
break;
|
223 |
-
}
|
224 |
-
opt->h = std::stoi(argv[i]);
|
225 |
-
} else if (arg == "-W" || arg == "--width") {
|
226 |
-
if (++i >= argc) {
|
227 |
-
invalid_arg = true;
|
228 |
-
break;
|
229 |
-
}
|
230 |
-
opt->w = std::stoi(argv[i]);
|
231 |
-
} else if (arg == "--steps") {
|
232 |
-
if (++i >= argc) {
|
233 |
-
invalid_arg = true;
|
234 |
-
break;
|
235 |
-
}
|
236 |
-
opt->sample_steps = std::stoi(argv[i]);
|
237 |
-
} else if (arg == "--rng") {
|
238 |
-
if (++i >= argc) {
|
239 |
-
invalid_arg = true;
|
240 |
-
break;
|
241 |
-
}
|
242 |
-
std::string rng_type_str = argv[i];
|
243 |
-
if (rng_type_str == "std_default") {
|
244 |
-
opt->rng_type = STD_DEFAULT_RNG;
|
245 |
-
} else if (rng_type_str == "cuda") {
|
246 |
-
opt->rng_type = CUDA_RNG;
|
247 |
-
} else {
|
248 |
-
invalid_arg = true;
|
249 |
-
break;
|
250 |
-
}
|
251 |
-
} else if (arg == "--schedule") {
|
252 |
-
if (++i >= argc) {
|
253 |
-
invalid_arg = true;
|
254 |
-
break;
|
255 |
-
}
|
256 |
-
const char* schedule_selected = argv[i];
|
257 |
-
int schedule_found = -1;
|
258 |
-
for (int d = 0; d < N_SCHEDULES; d++) {
|
259 |
-
if (!strcmp(schedule_selected, schedule_str[d])) {
|
260 |
-
schedule_found = d;
|
261 |
-
}
|
262 |
-
}
|
263 |
-
if (schedule_found == -1) {
|
264 |
-
invalid_arg = true;
|
265 |
-
break;
|
266 |
-
}
|
267 |
-
opt->schedule = (Schedule)schedule_found;
|
268 |
-
} else if (arg == "-s" || arg == "--seed") {
|
269 |
-
if (++i >= argc) {
|
270 |
-
invalid_arg = true;
|
271 |
-
break;
|
272 |
-
}
|
273 |
-
opt->seed = std::stoll(argv[i]);
|
274 |
-
} else if (arg == "--sampling-method") {
|
275 |
-
if (++i >= argc) {
|
276 |
-
invalid_arg = true;
|
277 |
-
break;
|
278 |
-
}
|
279 |
-
const char* sample_method_selected = argv[i];
|
280 |
-
int sample_method_found = -1;
|
281 |
-
for (int m = 0; m < N_SAMPLE_METHODS; m++) {
|
282 |
-
if (!strcmp(sample_method_selected, sample_method_str[m])) {
|
283 |
-
sample_method_found = m;
|
284 |
-
}
|
285 |
-
}
|
286 |
-
if (sample_method_found == -1) {
|
287 |
-
invalid_arg = true;
|
288 |
-
break;
|
289 |
-
}
|
290 |
-
opt->sample_method = (SampleMethod)sample_method_found;
|
291 |
-
} else if (arg == "-h" || arg == "--help") {
|
292 |
-
print_usage(argc, argv);
|
293 |
-
exit(0);
|
294 |
-
} else if (arg == "-v" || arg == "--verbose") {
|
295 |
-
opt->verbose = true;
|
296 |
-
} else {
|
297 |
-
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
|
298 |
-
print_usage(argc, argv);
|
299 |
-
exit(1);
|
300 |
-
}
|
301 |
-
if (invalid_arg) {
|
302 |
-
fprintf(stderr, "error: invalid parameter for argument: %s\n", arg.c_str());
|
303 |
-
print_usage(argc, argv);
|
304 |
-
exit(1);
|
305 |
-
}
|
306 |
-
}
|
307 |
-
|
308 |
-
if (opt->n_threads <= 0) {
|
309 |
-
opt->n_threads = get_num_physical_cores();
|
310 |
-
}
|
311 |
-
|
312 |
-
if (opt->mode != TXT2IMG && opt->mode != IMG2IMG) {
|
313 |
-
fprintf(stderr, "error: invalid mode %s, must be one of ['%s', '%s']\n",
|
314 |
-
opt->mode.c_str(), TXT2IMG, IMG2IMG);
|
315 |
-
exit(1);
|
316 |
-
}
|
317 |
-
|
318 |
-
if (opt->prompt.length() == 0) {
|
319 |
-
fprintf(stderr, "error: the following arguments are required: prompt\n");
|
320 |
-
print_usage(argc, argv);
|
321 |
-
exit(1);
|
322 |
-
}
|
323 |
-
|
324 |
-
if (opt->model_path.length() == 0) {
|
325 |
-
fprintf(stderr, "error: the following arguments are required: model_path\n");
|
326 |
-
print_usage(argc, argv);
|
327 |
-
exit(1);
|
328 |
-
}
|
329 |
-
|
330 |
-
if (opt->mode == IMG2IMG && opt->init_img.length() == 0) {
|
331 |
-
fprintf(stderr, "error: when using the img2img mode, the following arguments are required: init-img\n");
|
332 |
-
print_usage(argc, argv);
|
333 |
-
exit(1);
|
334 |
-
}
|
335 |
-
|
336 |
-
if (opt->output_path.length() == 0) {
|
337 |
-
fprintf(stderr, "error: the following arguments are required: output_path\n");
|
338 |
-
print_usage(argc, argv);
|
339 |
-
exit(1);
|
340 |
-
}
|
341 |
-
|
342 |
-
if (opt->w <= 0 || opt->w % 64 != 0) {
|
343 |
-
fprintf(stderr, "error: the width must be a multiple of 64\n");
|
344 |
-
exit(1);
|
345 |
-
}
|
346 |
-
|
347 |
-
if (opt->h <= 0 || opt->h % 64 != 0) {
|
348 |
-
fprintf(stderr, "error: the height must be a multiple of 64\n");
|
349 |
-
exit(1);
|
350 |
-
}
|
351 |
-
|
352 |
-
if (opt->sample_steps <= 0) {
|
353 |
-
fprintf(stderr, "error: the sample_steps must be greater than 0\n");
|
354 |
-
exit(1);
|
355 |
-
}
|
356 |
-
|
357 |
-
if (opt->strength < 0.f || opt->strength > 1.f) {
|
358 |
-
fprintf(stderr, "error: can only work with strength in [0.0, 1.0]\n");
|
359 |
-
exit(1);
|
360 |
-
}
|
361 |
-
|
362 |
-
if (opt->seed < 0) {
|
363 |
-
srand((int)time(NULL));
|
364 |
-
opt->seed = rand();
|
365 |
-
}
|
366 |
-
}
|
367 |
-
|
368 |
-
std::string basename(const std::string& path) {
|
369 |
-
size_t pos = path.find_last_of('/');
|
370 |
-
if (pos != std::string::npos) {
|
371 |
-
return path.substr(pos + 1);
|
372 |
-
}
|
373 |
-
pos = path.find_last_of('\\');
|
374 |
-
if (pos != std::string::npos) {
|
375 |
-
return path.substr(pos + 1);
|
376 |
-
}
|
377 |
-
return path;
|
378 |
-
}
|
379 |
-
|
380 |
-
int main(int argc, const char* argv[]) {
|
381 |
-
Option opt;
|
382 |
-
parse_args(argc, argv, &opt);
|
383 |
-
|
384 |
-
if (opt.verbose) {
|
385 |
-
opt.print();
|
386 |
-
printf("%s", sd_get_system_info().c_str());
|
387 |
-
set_sd_log_level(SDLogLevel::DEBUG);
|
388 |
-
}
|
389 |
-
|
390 |
-
bool vae_decode_only = true;
|
391 |
-
std::vector<uint8_t> init_img;
|
392 |
-
if (opt.mode == IMG2IMG) {
|
393 |
-
vae_decode_only = false;
|
394 |
-
|
395 |
-
int c = 0;
|
396 |
-
unsigned char* img_data = stbi_load(opt.init_img.c_str(), &opt.w, &opt.h, &c, 3);
|
397 |
-
if (img_data == NULL) {
|
398 |
-
fprintf(stderr, "load image from '%s' failed\n", opt.init_img.c_str());
|
399 |
-
return 1;
|
400 |
-
}
|
401 |
-
if (c != 3) {
|
402 |
-
fprintf(stderr, "input image must be a 3 channels RGB image, but got %d channels\n", c);
|
403 |
-
free(img_data);
|
404 |
-
return 1;
|
405 |
-
}
|
406 |
-
if (opt.w <= 0 || opt.w % 64 != 0) {
|
407 |
-
fprintf(stderr, "error: the width of image must be a multiple of 64\n");
|
408 |
-
free(img_data);
|
409 |
-
return 1;
|
410 |
-
}
|
411 |
-
if (opt.h <= 0 || opt.h % 64 != 0) {
|
412 |
-
fprintf(stderr, "error: the height of image must be a multiple of 64\n");
|
413 |
-
free(img_data);
|
414 |
-
return 1;
|
415 |
-
}
|
416 |
-
init_img.assign(img_data, img_data + (opt.w * opt.h * c));
|
417 |
-
}
|
418 |
-
|
419 |
-
StableDiffusion sd(opt.n_threads, vae_decode_only, true, opt.rng_type);
|
420 |
-
if (!sd.load_from_file(opt.model_path, opt.schedule)) {
|
421 |
-
return 1;
|
422 |
-
}
|
423 |
-
|
424 |
-
std::vector<uint8_t> img;
|
425 |
-
if (opt.mode == TXT2IMG) {
|
426 |
-
img = sd.txt2img(opt.prompt,
|
427 |
-
opt.negative_prompt,
|
428 |
-
opt.cfg_scale,
|
429 |
-
opt.w,
|
430 |
-
opt.h,
|
431 |
-
opt.sample_method,
|
432 |
-
opt.sample_steps,
|
433 |
-
opt.seed);
|
434 |
-
} else {
|
435 |
-
img = sd.img2img(init_img,
|
436 |
-
opt.prompt,
|
437 |
-
opt.negative_prompt,
|
438 |
-
opt.cfg_scale,
|
439 |
-
opt.w,
|
440 |
-
opt.h,
|
441 |
-
opt.sample_method,
|
442 |
-
opt.sample_steps,
|
443 |
-
opt.strength,
|
444 |
-
opt.seed);
|
445 |
-
}
|
446 |
-
|
447 |
-
if (img.size() == 0) {
|
448 |
-
fprintf(stderr, "generate failed\n");
|
449 |
-
return 1;
|
450 |
-
}
|
451 |
-
|
452 |
-
std::string parameter_string = opt.prompt + "\n";
|
453 |
-
if (opt.negative_prompt.size() != 0) {
|
454 |
-
parameter_string += "Negative prompt: " + opt.negative_prompt + "\n";
|
455 |
-
}
|
456 |
-
parameter_string += "Steps: " + std::to_string(opt.sample_steps) + ", ";
|
457 |
-
parameter_string += "CFG scale: " + std::to_string(opt.cfg_scale) + ", ";
|
458 |
-
parameter_string += "Seed: " + std::to_string(opt.seed) + ", ";
|
459 |
-
parameter_string += "Size: " + std::to_string(opt.w) + "x" + std::to_string(opt.h) + ", ";
|
460 |
-
parameter_string += "Model: " + basename(opt.model_path) + ", ";
|
461 |
-
parameter_string += "RNG: " + std::string(rng_type_to_str[opt.rng_type]) + ", ";
|
462 |
-
parameter_string += "Sampler: " + std::string(sample_method_str[opt.sample_method]);
|
463 |
-
if (opt.schedule == KARRAS) {
|
464 |
-
parameter_string += " karras";
|
465 |
-
}
|
466 |
-
parameter_string += ", ";
|
467 |
-
parameter_string += "Version: stable-diffusion.cpp";
|
468 |
-
|
469 |
-
stbi_write_png(opt.output_path.c_str(), opt.w, opt.h, 3, img.data(), 0, parameter_string.c_str());
|
470 |
-
printf("save result image to '%s'\n", opt.output_path.c_str());
|
471 |
-
|
472 |
-
return 0;
|
473 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/examples/stb_image.h
DELETED
The diff for this file is too large to render.
See raw diff
|
|
stable-diffusion.cpp/examples/stb_image_write.h
DELETED
@@ -1,1741 +0,0 @@
|
|
1 |
-
/* stb_image_write - v1.16 - public domain - http://nothings.org/stb
|
2 |
-
writes out PNG/BMP/TGA/JPEG/HDR images to C stdio - Sean Barrett 2010-2015
|
3 |
-
no warranty implied; use at your own risk
|
4 |
-
|
5 |
-
Before #including,
|
6 |
-
|
7 |
-
#define STB_IMAGE_WRITE_IMPLEMENTATION
|
8 |
-
|
9 |
-
in the file that you want to have the implementation.
|
10 |
-
|
11 |
-
Will probably not work correctly with strict-aliasing optimizations.
|
12 |
-
|
13 |
-
ABOUT:
|
14 |
-
|
15 |
-
This header file is a library for writing images to C stdio or a callback.
|
16 |
-
|
17 |
-
The PNG output is not optimal; it is 20-50% larger than the file
|
18 |
-
written by a decent optimizing implementation; though providing a custom
|
19 |
-
zlib compress function (see STBIW_ZLIB_COMPRESS) can mitigate that.
|
20 |
-
This library is designed for source code compactness and simplicity,
|
21 |
-
not optimal image file size or run-time performance.
|
22 |
-
|
23 |
-
BUILDING:
|
24 |
-
|
25 |
-
You can #define STBIW_ASSERT(x) before the #include to avoid using assert.h.
|
26 |
-
You can #define STBIW_MALLOC(), STBIW_REALLOC(), and STBIW_FREE() to replace
|
27 |
-
malloc,realloc,free.
|
28 |
-
You can #define STBIW_MEMMOVE() to replace memmove()
|
29 |
-
You can #define STBIW_ZLIB_COMPRESS to use a custom zlib-style compress function
|
30 |
-
for PNG compression (instead of the builtin one), it must have the following signature:
|
31 |
-
unsigned char * my_compress(unsigned char *data, int data_len, int *out_len, int quality);
|
32 |
-
The returned data will be freed with STBIW_FREE() (free() by default),
|
33 |
-
so it must be heap allocated with STBIW_MALLOC() (malloc() by default),
|
34 |
-
|
35 |
-
UNICODE:
|
36 |
-
|
37 |
-
If compiling for Windows and you wish to use Unicode filenames, compile
|
38 |
-
with
|
39 |
-
#define STBIW_WINDOWS_UTF8
|
40 |
-
and pass utf8-encoded filenames. Call stbiw_convert_wchar_to_utf8 to convert
|
41 |
-
Windows wchar_t filenames to utf8.
|
42 |
-
|
43 |
-
USAGE:
|
44 |
-
|
45 |
-
There are five functions, one for each image file format:
|
46 |
-
|
47 |
-
int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes);
|
48 |
-
int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data);
|
49 |
-
int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data);
|
50 |
-
int stbi_write_jpg(char const *filename, int w, int h, int comp, const void *data, int quality);
|
51 |
-
int stbi_write_hdr(char const *filename, int w, int h, int comp, const float *data);
|
52 |
-
|
53 |
-
void stbi_flip_vertically_on_write(int flag); // flag is non-zero to flip data vertically
|
54 |
-
|
55 |
-
There are also five equivalent functions that use an arbitrary write function. You are
|
56 |
-
expected to open/close your file-equivalent before and after calling these:
|
57 |
-
|
58 |
-
int stbi_write_png_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data, int stride_in_bytes);
|
59 |
-
int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data);
|
60 |
-
int stbi_write_tga_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data);
|
61 |
-
int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const float *data);
|
62 |
-
int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality);
|
63 |
-
|
64 |
-
where the callback is:
|
65 |
-
void stbi_write_func(void *context, void *data, int size);
|
66 |
-
|
67 |
-
You can configure it with these global variables:
|
68 |
-
int stbi_write_tga_with_rle; // defaults to true; set to 0 to disable RLE
|
69 |
-
int stbi_write_png_compression_level; // defaults to 8; set to higher for more compression
|
70 |
-
int stbi_write_force_png_filter; // defaults to -1; set to 0..5 to force a filter mode
|
71 |
-
|
72 |
-
|
73 |
-
You can define STBI_WRITE_NO_STDIO to disable the file variant of these
|
74 |
-
functions, so the library will not use stdio.h at all. However, this will
|
75 |
-
also disable HDR writing, because it requires stdio for formatted output.
|
76 |
-
|
77 |
-
Each function returns 0 on failure and non-0 on success.
|
78 |
-
|
79 |
-
The functions create an image file defined by the parameters. The image
|
80 |
-
is a rectangle of pixels stored from left-to-right, top-to-bottom.
|
81 |
-
Each pixel contains 'comp' channels of data stored interleaved with 8-bits
|
82 |
-
per channel, in the following order: 1=Y, 2=YA, 3=RGB, 4=RGBA. (Y is
|
83 |
-
monochrome color.) The rectangle is 'w' pixels wide and 'h' pixels tall.
|
84 |
-
The *data pointer points to the first byte of the top-left-most pixel.
|
85 |
-
For PNG, "stride_in_bytes" is the distance in bytes from the first byte of
|
86 |
-
a row of pixels to the first byte of the next row of pixels.
|
87 |
-
|
88 |
-
PNG creates output files with the same number of components as the input.
|
89 |
-
The BMP format expands Y to RGB in the file format and does not
|
90 |
-
output alpha.
|
91 |
-
|
92 |
-
PNG supports writing rectangles of data even when the bytes storing rows of
|
93 |
-
data are not consecutive in memory (e.g. sub-rectangles of a larger image),
|
94 |
-
by supplying the stride between the beginning of adjacent rows. The other
|
95 |
-
formats do not. (Thus you cannot write a native-format BMP through the BMP
|
96 |
-
writer, both because it is in BGR order and because it may have padding
|
97 |
-
at the end of the line.)
|
98 |
-
|
99 |
-
PNG allows you to set the deflate compression level by setting the global
|
100 |
-
variable 'stbi_write_png_compression_level' (it defaults to 8).
|
101 |
-
|
102 |
-
HDR expects linear float data. Since the format is always 32-bit rgb(e)
|
103 |
-
data, alpha (if provided) is discarded, and for monochrome data it is
|
104 |
-
replicated across all three channels.
|
105 |
-
|
106 |
-
TGA supports RLE or non-RLE compressed data. To use non-RLE-compressed
|
107 |
-
data, set the global variable 'stbi_write_tga_with_rle' to 0.
|
108 |
-
|
109 |
-
JPEG does ignore alpha channels in input data; quality is between 1 and 100.
|
110 |
-
Higher quality looks better but results in a bigger image.
|
111 |
-
JPEG baseline (no JPEG progressive).
|
112 |
-
|
113 |
-
CREDITS:
|
114 |
-
|
115 |
-
|
116 |
-
Sean Barrett - PNG/BMP/TGA
|
117 |
-
Baldur Karlsson - HDR
|
118 |
-
Jean-Sebastien Guay - TGA monochrome
|
119 |
-
Tim Kelsey - misc enhancements
|
120 |
-
Alan Hickman - TGA RLE
|
121 |
-
Emmanuel Julien - initial file IO callback implementation
|
122 |
-
Jon Olick - original jo_jpeg.cpp code
|
123 |
-
Daniel Gibson - integrate JPEG, allow external zlib
|
124 |
-
Aarni Koskela - allow choosing PNG filter
|
125 |
-
|
126 |
-
bugfixes:
|
127 |
-
github:Chribba
|
128 |
-
Guillaume Chereau
|
129 |
-
github:jry2
|
130 |
-
github:romigrou
|
131 |
-
Sergio Gonzalez
|
132 |
-
Jonas Karlsson
|
133 |
-
Filip Wasil
|
134 |
-
Thatcher Ulrich
|
135 |
-
github:poppolopoppo
|
136 |
-
Patrick Boettcher
|
137 |
-
github:xeekworx
|
138 |
-
Cap Petschulat
|
139 |
-
Simon Rodriguez
|
140 |
-
Ivan Tikhonov
|
141 |
-
github:ignotion
|
142 |
-
Adam Schackart
|
143 |
-
Andrew Kensler
|
144 |
-
|
145 |
-
LICENSE
|
146 |
-
|
147 |
-
See end of file for license information.
|
148 |
-
|
149 |
-
*/
|
150 |
-
|
151 |
-
#ifndef INCLUDE_STB_IMAGE_WRITE_H
|
152 |
-
#define INCLUDE_STB_IMAGE_WRITE_H
|
153 |
-
|
154 |
-
#include <stdlib.h>
|
155 |
-
|
156 |
-
// if STB_IMAGE_WRITE_STATIC causes problems, try defining STBIWDEF to 'inline' or 'static inline'
|
157 |
-
#ifndef STBIWDEF
|
158 |
-
#ifdef STB_IMAGE_WRITE_STATIC
|
159 |
-
#define STBIWDEF static
|
160 |
-
#else
|
161 |
-
#ifdef __cplusplus
|
162 |
-
#define STBIWDEF extern "C"
|
163 |
-
#else
|
164 |
-
#define STBIWDEF extern
|
165 |
-
#endif
|
166 |
-
#endif
|
167 |
-
#endif
|
168 |
-
|
169 |
-
#ifndef STB_IMAGE_WRITE_STATIC // C++ forbids static forward declarations
|
170 |
-
STBIWDEF int stbi_write_tga_with_rle;
|
171 |
-
STBIWDEF int stbi_write_png_compression_level;
|
172 |
-
STBIWDEF int stbi_write_force_png_filter;
|
173 |
-
#endif
|
174 |
-
|
175 |
-
#ifndef STBI_WRITE_NO_STDIO
|
176 |
-
STBIWDEF int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes, const char* parameters = NULL);
|
177 |
-
STBIWDEF int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data);
|
178 |
-
STBIWDEF int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data);
|
179 |
-
STBIWDEF int stbi_write_hdr(char const *filename, int w, int h, int comp, const float *data);
|
180 |
-
STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void *data, int quality);
|
181 |
-
|
182 |
-
#ifdef STBIW_WINDOWS_UTF8
|
183 |
-
STBIWDEF int stbiw_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input);
|
184 |
-
#endif
|
185 |
-
#endif
|
186 |
-
|
187 |
-
typedef void stbi_write_func(void *context, void *data, int size);
|
188 |
-
|
189 |
-
STBIWDEF int stbi_write_png_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data, int stride_in_bytes);
|
190 |
-
STBIWDEF int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data);
|
191 |
-
STBIWDEF int stbi_write_tga_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data);
|
192 |
-
STBIWDEF int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const float *data);
|
193 |
-
STBIWDEF int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality);
|
194 |
-
|
195 |
-
STBIWDEF void stbi_flip_vertically_on_write(int flip_boolean);
|
196 |
-
|
197 |
-
#endif//INCLUDE_STB_IMAGE_WRITE_H
|
198 |
-
|
199 |
-
#ifdef STB_IMAGE_WRITE_IMPLEMENTATION
|
200 |
-
|
201 |
-
#ifdef _WIN32
|
202 |
-
#ifndef _CRT_SECURE_NO_WARNINGS
|
203 |
-
#define _CRT_SECURE_NO_WARNINGS
|
204 |
-
#endif
|
205 |
-
#ifndef _CRT_NONSTDC_NO_DEPRECATE
|
206 |
-
#define _CRT_NONSTDC_NO_DEPRECATE
|
207 |
-
#endif
|
208 |
-
#endif
|
209 |
-
|
210 |
-
#ifndef STBI_WRITE_NO_STDIO
|
211 |
-
#include <stdio.h>
|
212 |
-
#endif // STBI_WRITE_NO_STDIO
|
213 |
-
|
214 |
-
#include <stdarg.h>
|
215 |
-
#include <stdlib.h>
|
216 |
-
#include <string.h>
|
217 |
-
#include <math.h>
|
218 |
-
|
219 |
-
#if defined(STBIW_MALLOC) && defined(STBIW_FREE) && (defined(STBIW_REALLOC) || defined(STBIW_REALLOC_SIZED))
|
220 |
-
// ok
|
221 |
-
#elif !defined(STBIW_MALLOC) && !defined(STBIW_FREE) && !defined(STBIW_REALLOC) && !defined(STBIW_REALLOC_SIZED)
|
222 |
-
// ok
|
223 |
-
#else
|
224 |
-
#error "Must define all or none of STBIW_MALLOC, STBIW_FREE, and STBIW_REALLOC (or STBIW_REALLOC_SIZED)."
|
225 |
-
#endif
|
226 |
-
|
227 |
-
#ifndef STBIW_MALLOC
|
228 |
-
#define STBIW_MALLOC(sz) malloc(sz)
|
229 |
-
#define STBIW_REALLOC(p,newsz) realloc(p,newsz)
|
230 |
-
#define STBIW_FREE(p) free(p)
|
231 |
-
#endif
|
232 |
-
|
233 |
-
#ifndef STBIW_REALLOC_SIZED
|
234 |
-
#define STBIW_REALLOC_SIZED(p,oldsz,newsz) STBIW_REALLOC(p,newsz)
|
235 |
-
#endif
|
236 |
-
|
237 |
-
|
238 |
-
#ifndef STBIW_MEMMOVE
|
239 |
-
#define STBIW_MEMMOVE(a,b,sz) memmove(a,b,sz)
|
240 |
-
#endif
|
241 |
-
|
242 |
-
|
243 |
-
#ifndef STBIW_ASSERT
|
244 |
-
#include <assert.h>
|
245 |
-
#define STBIW_ASSERT(x) assert(x)
|
246 |
-
#endif
|
247 |
-
|
248 |
-
#define STBIW_UCHAR(x) (unsigned char) ((x) & 0xff)
|
249 |
-
|
250 |
-
#ifdef STB_IMAGE_WRITE_STATIC
|
251 |
-
static int stbi_write_png_compression_level = 8;
|
252 |
-
static int stbi_write_tga_with_rle = 1;
|
253 |
-
static int stbi_write_force_png_filter = -1;
|
254 |
-
#else
|
255 |
-
int stbi_write_png_compression_level = 8;
|
256 |
-
int stbi_write_tga_with_rle = 1;
|
257 |
-
int stbi_write_force_png_filter = -1;
|
258 |
-
#endif
|
259 |
-
|
260 |
-
static int stbi__flip_vertically_on_write = 0;
|
261 |
-
|
262 |
-
STBIWDEF void stbi_flip_vertically_on_write(int flag)
|
263 |
-
{
|
264 |
-
stbi__flip_vertically_on_write = flag;
|
265 |
-
}
|
266 |
-
|
267 |
-
typedef struct
|
268 |
-
{
|
269 |
-
stbi_write_func *func;
|
270 |
-
void *context;
|
271 |
-
unsigned char buffer[64];
|
272 |
-
int buf_used;
|
273 |
-
} stbi__write_context;
|
274 |
-
|
275 |
-
// initialize a callback-based context
|
276 |
-
static void stbi__start_write_callbacks(stbi__write_context *s, stbi_write_func *c, void *context)
|
277 |
-
{
|
278 |
-
s->func = c;
|
279 |
-
s->context = context;
|
280 |
-
}
|
281 |
-
|
282 |
-
#ifndef STBI_WRITE_NO_STDIO
|
283 |
-
|
284 |
-
static void stbi__stdio_write(void *context, void *data, int size)
|
285 |
-
{
|
286 |
-
fwrite(data,1,size,(FILE*) context);
|
287 |
-
}
|
288 |
-
|
289 |
-
#if defined(_WIN32) && defined(STBIW_WINDOWS_UTF8)
|
290 |
-
#ifdef __cplusplus
|
291 |
-
#define STBIW_EXTERN extern "C"
|
292 |
-
#else
|
293 |
-
#define STBIW_EXTERN extern
|
294 |
-
#endif
|
295 |
-
STBIW_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, const char *str, int cbmb, wchar_t *widestr, int cchwide);
|
296 |
-
STBIW_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, const wchar_t *widestr, int cchwide, char *str, int cbmb, const char *defchar, int *used_default);
|
297 |
-
|
298 |
-
STBIWDEF int stbiw_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input)
|
299 |
-
{
|
300 |
-
return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int) bufferlen, NULL, NULL);
|
301 |
-
}
|
302 |
-
#endif
|
303 |
-
|
304 |
-
static FILE *stbiw__fopen(char const *filename, char const *mode)
|
305 |
-
{
|
306 |
-
FILE *f;
|
307 |
-
#if defined(_WIN32) && defined(STBIW_WINDOWS_UTF8)
|
308 |
-
wchar_t wMode[64];
|
309 |
-
wchar_t wFilename[1024];
|
310 |
-
if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename)/sizeof(*wFilename)))
|
311 |
-
return 0;
|
312 |
-
|
313 |
-
if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode)/sizeof(*wMode)))
|
314 |
-
return 0;
|
315 |
-
|
316 |
-
#if defined(_MSC_VER) && _MSC_VER >= 1400
|
317 |
-
if (0 != _wfopen_s(&f, wFilename, wMode))
|
318 |
-
f = 0;
|
319 |
-
#else
|
320 |
-
f = _wfopen(wFilename, wMode);
|
321 |
-
#endif
|
322 |
-
|
323 |
-
#elif defined(_MSC_VER) && _MSC_VER >= 1400
|
324 |
-
if (0 != fopen_s(&f, filename, mode))
|
325 |
-
f=0;
|
326 |
-
#else
|
327 |
-
f = fopen(filename, mode);
|
328 |
-
#endif
|
329 |
-
return f;
|
330 |
-
}
|
331 |
-
|
332 |
-
static int stbi__start_write_file(stbi__write_context *s, const char *filename)
|
333 |
-
{
|
334 |
-
FILE *f = stbiw__fopen(filename, "wb");
|
335 |
-
stbi__start_write_callbacks(s, stbi__stdio_write, (void *) f);
|
336 |
-
return f != NULL;
|
337 |
-
}
|
338 |
-
|
339 |
-
static void stbi__end_write_file(stbi__write_context *s)
|
340 |
-
{
|
341 |
-
fclose((FILE *)s->context);
|
342 |
-
}
|
343 |
-
|
344 |
-
#endif // !STBI_WRITE_NO_STDIO
|
345 |
-
|
346 |
-
typedef unsigned int stbiw_uint32;
|
347 |
-
typedef int stb_image_write_test[sizeof(stbiw_uint32)==4 ? 1 : -1];
|
348 |
-
|
349 |
-
static void stbiw__writefv(stbi__write_context *s, const char *fmt, va_list v)
|
350 |
-
{
|
351 |
-
while (*fmt) {
|
352 |
-
switch (*fmt++) {
|
353 |
-
case ' ': break;
|
354 |
-
case '1': { unsigned char x = STBIW_UCHAR(va_arg(v, int));
|
355 |
-
s->func(s->context,&x,1);
|
356 |
-
break; }
|
357 |
-
case '2': { int x = va_arg(v,int);
|
358 |
-
unsigned char b[2];
|
359 |
-
b[0] = STBIW_UCHAR(x);
|
360 |
-
b[1] = STBIW_UCHAR(x>>8);
|
361 |
-
s->func(s->context,b,2);
|
362 |
-
break; }
|
363 |
-
case '4': { stbiw_uint32 x = va_arg(v,int);
|
364 |
-
unsigned char b[4];
|
365 |
-
b[0]=STBIW_UCHAR(x);
|
366 |
-
b[1]=STBIW_UCHAR(x>>8);
|
367 |
-
b[2]=STBIW_UCHAR(x>>16);
|
368 |
-
b[3]=STBIW_UCHAR(x>>24);
|
369 |
-
s->func(s->context,b,4);
|
370 |
-
break; }
|
371 |
-
default:
|
372 |
-
STBIW_ASSERT(0);
|
373 |
-
return;
|
374 |
-
}
|
375 |
-
}
|
376 |
-
}
|
377 |
-
|
378 |
-
static void stbiw__writef(stbi__write_context *s, const char *fmt, ...)
|
379 |
-
{
|
380 |
-
va_list v;
|
381 |
-
va_start(v, fmt);
|
382 |
-
stbiw__writefv(s, fmt, v);
|
383 |
-
va_end(v);
|
384 |
-
}
|
385 |
-
|
386 |
-
static void stbiw__write_flush(stbi__write_context *s)
|
387 |
-
{
|
388 |
-
if (s->buf_used) {
|
389 |
-
s->func(s->context, &s->buffer, s->buf_used);
|
390 |
-
s->buf_used = 0;
|
391 |
-
}
|
392 |
-
}
|
393 |
-
|
394 |
-
static void stbiw__putc(stbi__write_context *s, unsigned char c)
|
395 |
-
{
|
396 |
-
s->func(s->context, &c, 1);
|
397 |
-
}
|
398 |
-
|
399 |
-
static void stbiw__write1(stbi__write_context *s, unsigned char a)
|
400 |
-
{
|
401 |
-
if ((size_t)s->buf_used + 1 > sizeof(s->buffer))
|
402 |
-
stbiw__write_flush(s);
|
403 |
-
s->buffer[s->buf_used++] = a;
|
404 |
-
}
|
405 |
-
|
406 |
-
static void stbiw__write3(stbi__write_context *s, unsigned char a, unsigned char b, unsigned char c)
|
407 |
-
{
|
408 |
-
int n;
|
409 |
-
if ((size_t)s->buf_used + 3 > sizeof(s->buffer))
|
410 |
-
stbiw__write_flush(s);
|
411 |
-
n = s->buf_used;
|
412 |
-
s->buf_used = n+3;
|
413 |
-
s->buffer[n+0] = a;
|
414 |
-
s->buffer[n+1] = b;
|
415 |
-
s->buffer[n+2] = c;
|
416 |
-
}
|
417 |
-
|
418 |
-
static void stbiw__write_pixel(stbi__write_context *s, int rgb_dir, int comp, int write_alpha, int expand_mono, unsigned char *d)
|
419 |
-
{
|
420 |
-
unsigned char bg[3] = { 255, 0, 255}, px[3];
|
421 |
-
int k;
|
422 |
-
|
423 |
-
if (write_alpha < 0)
|
424 |
-
stbiw__write1(s, d[comp - 1]);
|
425 |
-
|
426 |
-
switch (comp) {
|
427 |
-
case 2: // 2 pixels = mono + alpha, alpha is written separately, so same as 1-channel case
|
428 |
-
case 1:
|
429 |
-
if (expand_mono)
|
430 |
-
stbiw__write3(s, d[0], d[0], d[0]); // monochrome bmp
|
431 |
-
else
|
432 |
-
stbiw__write1(s, d[0]); // monochrome TGA
|
433 |
-
break;
|
434 |
-
case 4:
|
435 |
-
if (!write_alpha) {
|
436 |
-
// composite against pink background
|
437 |
-
for (k = 0; k < 3; ++k)
|
438 |
-
px[k] = bg[k] + ((d[k] - bg[k]) * d[3]) / 255;
|
439 |
-
stbiw__write3(s, px[1 - rgb_dir], px[1], px[1 + rgb_dir]);
|
440 |
-
break;
|
441 |
-
}
|
442 |
-
/* FALLTHROUGH */
|
443 |
-
case 3:
|
444 |
-
stbiw__write3(s, d[1 - rgb_dir], d[1], d[1 + rgb_dir]);
|
445 |
-
break;
|
446 |
-
}
|
447 |
-
if (write_alpha > 0)
|
448 |
-
stbiw__write1(s, d[comp - 1]);
|
449 |
-
}
|
450 |
-
|
451 |
-
static void stbiw__write_pixels(stbi__write_context *s, int rgb_dir, int vdir, int x, int y, int comp, void *data, int write_alpha, int scanline_pad, int expand_mono)
|
452 |
-
{
|
453 |
-
stbiw_uint32 zero = 0;
|
454 |
-
int i,j, j_end;
|
455 |
-
|
456 |
-
if (y <= 0)
|
457 |
-
return;
|
458 |
-
|
459 |
-
if (stbi__flip_vertically_on_write)
|
460 |
-
vdir *= -1;
|
461 |
-
|
462 |
-
if (vdir < 0) {
|
463 |
-
j_end = -1; j = y-1;
|
464 |
-
} else {
|
465 |
-
j_end = y; j = 0;
|
466 |
-
}
|
467 |
-
|
468 |
-
for (; j != j_end; j += vdir) {
|
469 |
-
for (i=0; i < x; ++i) {
|
470 |
-
unsigned char *d = (unsigned char *) data + (j*x+i)*comp;
|
471 |
-
stbiw__write_pixel(s, rgb_dir, comp, write_alpha, expand_mono, d);
|
472 |
-
}
|
473 |
-
stbiw__write_flush(s);
|
474 |
-
s->func(s->context, &zero, scanline_pad);
|
475 |
-
}
|
476 |
-
}
|
477 |
-
|
478 |
-
static int stbiw__outfile(stbi__write_context *s, int rgb_dir, int vdir, int x, int y, int comp, int expand_mono, void *data, int alpha, int pad, const char *fmt, ...)
|
479 |
-
{
|
480 |
-
if (y < 0 || x < 0) {
|
481 |
-
return 0;
|
482 |
-
} else {
|
483 |
-
va_list v;
|
484 |
-
va_start(v, fmt);
|
485 |
-
stbiw__writefv(s, fmt, v);
|
486 |
-
va_end(v);
|
487 |
-
stbiw__write_pixels(s,rgb_dir,vdir,x,y,comp,data,alpha,pad, expand_mono);
|
488 |
-
return 1;
|
489 |
-
}
|
490 |
-
}
|
491 |
-
|
492 |
-
static int stbi_write_bmp_core(stbi__write_context *s, int x, int y, int comp, const void *data)
|
493 |
-
{
|
494 |
-
if (comp != 4) {
|
495 |
-
// write RGB bitmap
|
496 |
-
int pad = (-x*3) & 3;
|
497 |
-
return stbiw__outfile(s,-1,-1,x,y,comp,1,(void *) data,0,pad,
|
498 |
-
"11 4 22 4" "4 44 22 444444",
|
499 |
-
'B', 'M', 14+40+(x*3+pad)*y, 0,0, 14+40, // file header
|
500 |
-
40, x,y, 1,24, 0,0,0,0,0,0); // bitmap header
|
501 |
-
} else {
|
502 |
-
// RGBA bitmaps need a v4 header
|
503 |
-
// use BI_BITFIELDS mode with 32bpp and alpha mask
|
504 |
-
// (straight BI_RGB with alpha mask doesn't work in most readers)
|
505 |
-
return stbiw__outfile(s,-1,-1,x,y,comp,1,(void *)data,1,0,
|
506 |
-
"11 4 22 4" "4 44 22 444444 4444 4 444 444 444 444",
|
507 |
-
'B', 'M', 14+108+x*y*4, 0, 0, 14+108, // file header
|
508 |
-
108, x,y, 1,32, 3,0,0,0,0,0, 0xff0000,0xff00,0xff,0xff000000u, 0, 0,0,0, 0,0,0, 0,0,0, 0,0,0); // bitmap V4 header
|
509 |
-
}
|
510 |
-
}
|
511 |
-
|
512 |
-
STBIWDEF int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data)
|
513 |
-
{
|
514 |
-
stbi__write_context s = { 0 };
|
515 |
-
stbi__start_write_callbacks(&s, func, context);
|
516 |
-
return stbi_write_bmp_core(&s, x, y, comp, data);
|
517 |
-
}
|
518 |
-
|
519 |
-
#ifndef STBI_WRITE_NO_STDIO
|
520 |
-
STBIWDEF int stbi_write_bmp(char const *filename, int x, int y, int comp, const void *data)
|
521 |
-
{
|
522 |
-
stbi__write_context s = { 0 };
|
523 |
-
if (stbi__start_write_file(&s,filename)) {
|
524 |
-
int r = stbi_write_bmp_core(&s, x, y, comp, data);
|
525 |
-
stbi__end_write_file(&s);
|
526 |
-
return r;
|
527 |
-
} else
|
528 |
-
return 0;
|
529 |
-
}
|
530 |
-
#endif //!STBI_WRITE_NO_STDIO
|
531 |
-
|
532 |
-
static int stbi_write_tga_core(stbi__write_context *s, int x, int y, int comp, void *data)
|
533 |
-
{
|
534 |
-
int has_alpha = (comp == 2 || comp == 4);
|
535 |
-
int colorbytes = has_alpha ? comp-1 : comp;
|
536 |
-
int format = colorbytes < 2 ? 3 : 2; // 3 color channels (RGB/RGBA) = 2, 1 color channel (Y/YA) = 3
|
537 |
-
|
538 |
-
if (y < 0 || x < 0)
|
539 |
-
return 0;
|
540 |
-
|
541 |
-
if (!stbi_write_tga_with_rle) {
|
542 |
-
return stbiw__outfile(s, -1, -1, x, y, comp, 0, (void *) data, has_alpha, 0,
|
543 |
-
"111 221 2222 11", 0, 0, format, 0, 0, 0, 0, 0, x, y, (colorbytes + has_alpha) * 8, has_alpha * 8);
|
544 |
-
} else {
|
545 |
-
int i,j,k;
|
546 |
-
int jend, jdir;
|
547 |
-
|
548 |
-
stbiw__writef(s, "111 221 2222 11", 0,0,format+8, 0,0,0, 0,0,x,y, (colorbytes + has_alpha) * 8, has_alpha * 8);
|
549 |
-
|
550 |
-
if (stbi__flip_vertically_on_write) {
|
551 |
-
j = 0;
|
552 |
-
jend = y;
|
553 |
-
jdir = 1;
|
554 |
-
} else {
|
555 |
-
j = y-1;
|
556 |
-
jend = -1;
|
557 |
-
jdir = -1;
|
558 |
-
}
|
559 |
-
for (; j != jend; j += jdir) {
|
560 |
-
unsigned char *row = (unsigned char *) data + j * x * comp;
|
561 |
-
int len;
|
562 |
-
|
563 |
-
for (i = 0; i < x; i += len) {
|
564 |
-
unsigned char *begin = row + i * comp;
|
565 |
-
int diff = 1;
|
566 |
-
len = 1;
|
567 |
-
|
568 |
-
if (i < x - 1) {
|
569 |
-
++len;
|
570 |
-
diff = memcmp(begin, row + (i + 1) * comp, comp);
|
571 |
-
if (diff) {
|
572 |
-
const unsigned char *prev = begin;
|
573 |
-
for (k = i + 2; k < x && len < 128; ++k) {
|
574 |
-
if (memcmp(prev, row + k * comp, comp)) {
|
575 |
-
prev += comp;
|
576 |
-
++len;
|
577 |
-
} else {
|
578 |
-
--len;
|
579 |
-
break;
|
580 |
-
}
|
581 |
-
}
|
582 |
-
} else {
|
583 |
-
for (k = i + 2; k < x && len < 128; ++k) {
|
584 |
-
if (!memcmp(begin, row + k * comp, comp)) {
|
585 |
-
++len;
|
586 |
-
} else {
|
587 |
-
break;
|
588 |
-
}
|
589 |
-
}
|
590 |
-
}
|
591 |
-
}
|
592 |
-
|
593 |
-
if (diff) {
|
594 |
-
unsigned char header = STBIW_UCHAR(len - 1);
|
595 |
-
stbiw__write1(s, header);
|
596 |
-
for (k = 0; k < len; ++k) {
|
597 |
-
stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin + k * comp);
|
598 |
-
}
|
599 |
-
} else {
|
600 |
-
unsigned char header = STBIW_UCHAR(len - 129);
|
601 |
-
stbiw__write1(s, header);
|
602 |
-
stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin);
|
603 |
-
}
|
604 |
-
}
|
605 |
-
}
|
606 |
-
stbiw__write_flush(s);
|
607 |
-
}
|
608 |
-
return 1;
|
609 |
-
}
|
610 |
-
|
611 |
-
STBIWDEF int stbi_write_tga_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data)
|
612 |
-
{
|
613 |
-
stbi__write_context s = { 0 };
|
614 |
-
stbi__start_write_callbacks(&s, func, context);
|
615 |
-
return stbi_write_tga_core(&s, x, y, comp, (void *) data);
|
616 |
-
}
|
617 |
-
|
618 |
-
#ifndef STBI_WRITE_NO_STDIO
|
619 |
-
STBIWDEF int stbi_write_tga(char const *filename, int x, int y, int comp, const void *data)
|
620 |
-
{
|
621 |
-
stbi__write_context s = { 0 };
|
622 |
-
if (stbi__start_write_file(&s,filename)) {
|
623 |
-
int r = stbi_write_tga_core(&s, x, y, comp, (void *) data);
|
624 |
-
stbi__end_write_file(&s);
|
625 |
-
return r;
|
626 |
-
} else
|
627 |
-
return 0;
|
628 |
-
}
|
629 |
-
#endif
|
630 |
-
|
631 |
-
// *************************************************************************************************
|
632 |
-
// Radiance RGBE HDR writer
|
633 |
-
// by Baldur Karlsson
|
634 |
-
|
635 |
-
#define stbiw__max(a, b) ((a) > (b) ? (a) : (b))
|
636 |
-
|
637 |
-
#ifndef STBI_WRITE_NO_STDIO
|
638 |
-
|
639 |
-
static void stbiw__linear_to_rgbe(unsigned char *rgbe, float *linear)
|
640 |
-
{
|
641 |
-
int exponent;
|
642 |
-
float maxcomp = stbiw__max(linear[0], stbiw__max(linear[1], linear[2]));
|
643 |
-
|
644 |
-
if (maxcomp < 1e-32f) {
|
645 |
-
rgbe[0] = rgbe[1] = rgbe[2] = rgbe[3] = 0;
|
646 |
-
} else {
|
647 |
-
float normalize = (float) frexp(maxcomp, &exponent) * 256.0f/maxcomp;
|
648 |
-
|
649 |
-
rgbe[0] = (unsigned char)(linear[0] * normalize);
|
650 |
-
rgbe[1] = (unsigned char)(linear[1] * normalize);
|
651 |
-
rgbe[2] = (unsigned char)(linear[2] * normalize);
|
652 |
-
rgbe[3] = (unsigned char)(exponent + 128);
|
653 |
-
}
|
654 |
-
}
|
655 |
-
|
656 |
-
static void stbiw__write_run_data(stbi__write_context *s, int length, unsigned char databyte)
|
657 |
-
{
|
658 |
-
unsigned char lengthbyte = STBIW_UCHAR(length+128);
|
659 |
-
STBIW_ASSERT(length+128 <= 255);
|
660 |
-
s->func(s->context, &lengthbyte, 1);
|
661 |
-
s->func(s->context, &databyte, 1);
|
662 |
-
}
|
663 |
-
|
664 |
-
static void stbiw__write_dump_data(stbi__write_context *s, int length, unsigned char *data)
|
665 |
-
{
|
666 |
-
unsigned char lengthbyte = STBIW_UCHAR(length);
|
667 |
-
STBIW_ASSERT(length <= 128); // inconsistent with spec but consistent with official code
|
668 |
-
s->func(s->context, &lengthbyte, 1);
|
669 |
-
s->func(s->context, data, length);
|
670 |
-
}
|
671 |
-
|
672 |
-
static void stbiw__write_hdr_scanline(stbi__write_context *s, int width, int ncomp, unsigned char *scratch, float *scanline)
|
673 |
-
{
|
674 |
-
unsigned char scanlineheader[4] = { 2, 2, 0, 0 };
|
675 |
-
unsigned char rgbe[4];
|
676 |
-
float linear[3];
|
677 |
-
int x;
|
678 |
-
|
679 |
-
scanlineheader[2] = (width&0xff00)>>8;
|
680 |
-
scanlineheader[3] = (width&0x00ff);
|
681 |
-
|
682 |
-
/* skip RLE for images too small or large */
|
683 |
-
if (width < 8 || width >= 32768) {
|
684 |
-
for (x=0; x < width; x++) {
|
685 |
-
switch (ncomp) {
|
686 |
-
case 4: /* fallthrough */
|
687 |
-
case 3: linear[2] = scanline[x*ncomp + 2];
|
688 |
-
linear[1] = scanline[x*ncomp + 1];
|
689 |
-
linear[0] = scanline[x*ncomp + 0];
|
690 |
-
break;
|
691 |
-
default:
|
692 |
-
linear[0] = linear[1] = linear[2] = scanline[x*ncomp + 0];
|
693 |
-
break;
|
694 |
-
}
|
695 |
-
stbiw__linear_to_rgbe(rgbe, linear);
|
696 |
-
s->func(s->context, rgbe, 4);
|
697 |
-
}
|
698 |
-
} else {
|
699 |
-
int c,r;
|
700 |
-
/* encode into scratch buffer */
|
701 |
-
for (x=0; x < width; x++) {
|
702 |
-
switch(ncomp) {
|
703 |
-
case 4: /* fallthrough */
|
704 |
-
case 3: linear[2] = scanline[x*ncomp + 2];
|
705 |
-
linear[1] = scanline[x*ncomp + 1];
|
706 |
-
linear[0] = scanline[x*ncomp + 0];
|
707 |
-
break;
|
708 |
-
default:
|
709 |
-
linear[0] = linear[1] = linear[2] = scanline[x*ncomp + 0];
|
710 |
-
break;
|
711 |
-
}
|
712 |
-
stbiw__linear_to_rgbe(rgbe, linear);
|
713 |
-
scratch[x + width*0] = rgbe[0];
|
714 |
-
scratch[x + width*1] = rgbe[1];
|
715 |
-
scratch[x + width*2] = rgbe[2];
|
716 |
-
scratch[x + width*3] = rgbe[3];
|
717 |
-
}
|
718 |
-
|
719 |
-
s->func(s->context, scanlineheader, 4);
|
720 |
-
|
721 |
-
/* RLE each component separately */
|
722 |
-
for (c=0; c < 4; c++) {
|
723 |
-
unsigned char *comp = &scratch[width*c];
|
724 |
-
|
725 |
-
x = 0;
|
726 |
-
while (x < width) {
|
727 |
-
// find first run
|
728 |
-
r = x;
|
729 |
-
while (r+2 < width) {
|
730 |
-
if (comp[r] == comp[r+1] && comp[r] == comp[r+2])
|
731 |
-
break;
|
732 |
-
++r;
|
733 |
-
}
|
734 |
-
if (r+2 >= width)
|
735 |
-
r = width;
|
736 |
-
// dump up to first run
|
737 |
-
while (x < r) {
|
738 |
-
int len = r-x;
|
739 |
-
if (len > 128) len = 128;
|
740 |
-
stbiw__write_dump_data(s, len, &comp[x]);
|
741 |
-
x += len;
|
742 |
-
}
|
743 |
-
// if there's a run, output it
|
744 |
-
if (r+2 < width) { // same test as what we break out of in search loop, so only true if we break'd
|
745 |
-
// find next byte after run
|
746 |
-
while (r < width && comp[r] == comp[x])
|
747 |
-
++r;
|
748 |
-
// output run up to r
|
749 |
-
while (x < r) {
|
750 |
-
int len = r-x;
|
751 |
-
if (len > 127) len = 127;
|
752 |
-
stbiw__write_run_data(s, len, comp[x]);
|
753 |
-
x += len;
|
754 |
-
}
|
755 |
-
}
|
756 |
-
}
|
757 |
-
}
|
758 |
-
}
|
759 |
-
}
|
760 |
-
|
761 |
-
static int stbi_write_hdr_core(stbi__write_context *s, int x, int y, int comp, float *data)
|
762 |
-
{
|
763 |
-
if (y <= 0 || x <= 0 || data == NULL)
|
764 |
-
return 0;
|
765 |
-
else {
|
766 |
-
// Each component is stored separately. Allocate scratch space for full output scanline.
|
767 |
-
unsigned char *scratch = (unsigned char *) STBIW_MALLOC(x*4);
|
768 |
-
int i, len;
|
769 |
-
char buffer[128];
|
770 |
-
char header[] = "#?RADIANCE\n# Written by stb_image_write.h\nFORMAT=32-bit_rle_rgbe\n";
|
771 |
-
s->func(s->context, header, sizeof(header)-1);
|
772 |
-
|
773 |
-
#ifdef __STDC_LIB_EXT1__
|
774 |
-
len = sprintf_s(buffer, sizeof(buffer), "EXPOSURE= 1.0000000000000\n\n-Y %d +X %d\n", y, x);
|
775 |
-
#else
|
776 |
-
len = sprintf(buffer, "EXPOSURE= 1.0000000000000\n\n-Y %d +X %d\n", y, x);
|
777 |
-
#endif
|
778 |
-
s->func(s->context, buffer, len);
|
779 |
-
|
780 |
-
for(i=0; i < y; i++)
|
781 |
-
stbiw__write_hdr_scanline(s, x, comp, scratch, data + comp*x*(stbi__flip_vertically_on_write ? y-1-i : i));
|
782 |
-
STBIW_FREE(scratch);
|
783 |
-
return 1;
|
784 |
-
}
|
785 |
-
}
|
786 |
-
|
787 |
-
STBIWDEF int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const float *data)
|
788 |
-
{
|
789 |
-
stbi__write_context s = { 0 };
|
790 |
-
stbi__start_write_callbacks(&s, func, context);
|
791 |
-
return stbi_write_hdr_core(&s, x, y, comp, (float *) data);
|
792 |
-
}
|
793 |
-
|
794 |
-
STBIWDEF int stbi_write_hdr(char const *filename, int x, int y, int comp, const float *data)
|
795 |
-
{
|
796 |
-
stbi__write_context s = { 0 };
|
797 |
-
if (stbi__start_write_file(&s,filename)) {
|
798 |
-
int r = stbi_write_hdr_core(&s, x, y, comp, (float *) data);
|
799 |
-
stbi__end_write_file(&s);
|
800 |
-
return r;
|
801 |
-
} else
|
802 |
-
return 0;
|
803 |
-
}
|
804 |
-
#endif // STBI_WRITE_NO_STDIO
|
805 |
-
|
806 |
-
|
807 |
-
//////////////////////////////////////////////////////////////////////////////
|
808 |
-
//
|
809 |
-
// PNG writer
|
810 |
-
//
|
811 |
-
|
812 |
-
#ifndef STBIW_ZLIB_COMPRESS
|
813 |
-
// stretchy buffer; stbiw__sbpush() == vector<>::push_back() -- stbiw__sbcount() == vector<>::size()
|
814 |
-
#define stbiw__sbraw(a) ((int *) (void *) (a) - 2)
|
815 |
-
#define stbiw__sbm(a) stbiw__sbraw(a)[0]
|
816 |
-
#define stbiw__sbn(a) stbiw__sbraw(a)[1]
|
817 |
-
|
818 |
-
#define stbiw__sbneedgrow(a,n) ((a)==0 || stbiw__sbn(a)+n >= stbiw__sbm(a))
|
819 |
-
#define stbiw__sbmaybegrow(a,n) (stbiw__sbneedgrow(a,(n)) ? stbiw__sbgrow(a,n) : 0)
|
820 |
-
#define stbiw__sbgrow(a,n) stbiw__sbgrowf((void **) &(a), (n), sizeof(*(a)))
|
821 |
-
|
822 |
-
#define stbiw__sbpush(a, v) (stbiw__sbmaybegrow(a,1), (a)[stbiw__sbn(a)++] = (v))
|
823 |
-
#define stbiw__sbcount(a) ((a) ? stbiw__sbn(a) : 0)
|
824 |
-
#define stbiw__sbfree(a) ((a) ? STBIW_FREE(stbiw__sbraw(a)),0 : 0)
|
825 |
-
|
826 |
-
static void *stbiw__sbgrowf(void **arr, int increment, int itemsize)
|
827 |
-
{
|
828 |
-
int m = *arr ? 2*stbiw__sbm(*arr)+increment : increment+1;
|
829 |
-
void *p = STBIW_REALLOC_SIZED(*arr ? stbiw__sbraw(*arr) : 0, *arr ? (stbiw__sbm(*arr)*itemsize + sizeof(int)*2) : 0, itemsize * m + sizeof(int)*2);
|
830 |
-
STBIW_ASSERT(p);
|
831 |
-
if (p) {
|
832 |
-
if (!*arr) ((int *) p)[1] = 0;
|
833 |
-
*arr = (void *) ((int *) p + 2);
|
834 |
-
stbiw__sbm(*arr) = m;
|
835 |
-
}
|
836 |
-
return *arr;
|
837 |
-
}
|
838 |
-
|
839 |
-
static unsigned char *stbiw__zlib_flushf(unsigned char *data, unsigned int *bitbuffer, int *bitcount)
|
840 |
-
{
|
841 |
-
while (*bitcount >= 8) {
|
842 |
-
stbiw__sbpush(data, STBIW_UCHAR(*bitbuffer));
|
843 |
-
*bitbuffer >>= 8;
|
844 |
-
*bitcount -= 8;
|
845 |
-
}
|
846 |
-
return data;
|
847 |
-
}
|
848 |
-
|
849 |
-
static int stbiw__zlib_bitrev(int code, int codebits)
|
850 |
-
{
|
851 |
-
int res=0;
|
852 |
-
while (codebits--) {
|
853 |
-
res = (res << 1) | (code & 1);
|
854 |
-
code >>= 1;
|
855 |
-
}
|
856 |
-
return res;
|
857 |
-
}
|
858 |
-
|
859 |
-
static unsigned int stbiw__zlib_countm(unsigned char *a, unsigned char *b, int limit)
|
860 |
-
{
|
861 |
-
int i;
|
862 |
-
for (i=0; i < limit && i < 258; ++i)
|
863 |
-
if (a[i] != b[i]) break;
|
864 |
-
return i;
|
865 |
-
}
|
866 |
-
|
867 |
-
static unsigned int stbiw__zhash(unsigned char *data)
|
868 |
-
{
|
869 |
-
stbiw_uint32 hash = data[0] + (data[1] << 8) + (data[2] << 16);
|
870 |
-
hash ^= hash << 3;
|
871 |
-
hash += hash >> 5;
|
872 |
-
hash ^= hash << 4;
|
873 |
-
hash += hash >> 17;
|
874 |
-
hash ^= hash << 25;
|
875 |
-
hash += hash >> 6;
|
876 |
-
return hash;
|
877 |
-
}
|
878 |
-
|
879 |
-
#define stbiw__zlib_flush() (out = stbiw__zlib_flushf(out, &bitbuf, &bitcount))
|
880 |
-
#define stbiw__zlib_add(code,codebits) \
|
881 |
-
(bitbuf |= (code) << bitcount, bitcount += (codebits), stbiw__zlib_flush())
|
882 |
-
#define stbiw__zlib_huffa(b,c) stbiw__zlib_add(stbiw__zlib_bitrev(b,c),c)
|
883 |
-
// default huffman tables
|
884 |
-
#define stbiw__zlib_huff1(n) stbiw__zlib_huffa(0x30 + (n), 8)
|
885 |
-
#define stbiw__zlib_huff2(n) stbiw__zlib_huffa(0x190 + (n)-144, 9)
|
886 |
-
#define stbiw__zlib_huff3(n) stbiw__zlib_huffa(0 + (n)-256,7)
|
887 |
-
#define stbiw__zlib_huff4(n) stbiw__zlib_huffa(0xc0 + (n)-280,8)
|
888 |
-
#define stbiw__zlib_huff(n) ((n) <= 143 ? stbiw__zlib_huff1(n) : (n) <= 255 ? stbiw__zlib_huff2(n) : (n) <= 279 ? stbiw__zlib_huff3(n) : stbiw__zlib_huff4(n))
|
889 |
-
#define stbiw__zlib_huffb(n) ((n) <= 143 ? stbiw__zlib_huff1(n) : stbiw__zlib_huff2(n))
|
890 |
-
|
891 |
-
#define stbiw__ZHASH 16384
|
892 |
-
|
893 |
-
#endif // STBIW_ZLIB_COMPRESS
|
894 |
-
|
895 |
-
STBIWDEF unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_len, int quality)
|
896 |
-
{
|
897 |
-
#ifdef STBIW_ZLIB_COMPRESS
|
898 |
-
// user provided a zlib compress implementation, use that
|
899 |
-
return STBIW_ZLIB_COMPRESS(data, data_len, out_len, quality);
|
900 |
-
#else // use builtin
|
901 |
-
static unsigned short lengthc[] = { 3,4,5,6,7,8,9,10,11,13,15,17,19,23,27,31,35,43,51,59,67,83,99,115,131,163,195,227,258, 259 };
|
902 |
-
static unsigned char lengtheb[]= { 0,0,0,0,0,0,0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0 };
|
903 |
-
static unsigned short distc[] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577, 32768 };
|
904 |
-
static unsigned char disteb[] = { 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13 };
|
905 |
-
unsigned int bitbuf=0;
|
906 |
-
int i,j, bitcount=0;
|
907 |
-
unsigned char *out = NULL;
|
908 |
-
unsigned char ***hash_table = (unsigned char***) STBIW_MALLOC(stbiw__ZHASH * sizeof(unsigned char**));
|
909 |
-
if (hash_table == NULL)
|
910 |
-
return NULL;
|
911 |
-
if (quality < 5) quality = 5;
|
912 |
-
|
913 |
-
stbiw__sbpush(out, 0x78); // DEFLATE 32K window
|
914 |
-
stbiw__sbpush(out, 0x5e); // FLEVEL = 1
|
915 |
-
stbiw__zlib_add(1,1); // BFINAL = 1
|
916 |
-
stbiw__zlib_add(1,2); // BTYPE = 1 -- fixed huffman
|
917 |
-
|
918 |
-
for (i=0; i < stbiw__ZHASH; ++i)
|
919 |
-
hash_table[i] = NULL;
|
920 |
-
|
921 |
-
i=0;
|
922 |
-
while (i < data_len-3) {
|
923 |
-
// hash next 3 bytes of data to be compressed
|
924 |
-
int h = stbiw__zhash(data+i)&(stbiw__ZHASH-1), best=3;
|
925 |
-
unsigned char *bestloc = 0;
|
926 |
-
unsigned char **hlist = hash_table[h];
|
927 |
-
int n = stbiw__sbcount(hlist);
|
928 |
-
for (j=0; j < n; ++j) {
|
929 |
-
if (hlist[j]-data > i-32768) { // if entry lies within window
|
930 |
-
int d = stbiw__zlib_countm(hlist[j], data+i, data_len-i);
|
931 |
-
if (d >= best) { best=d; bestloc=hlist[j]; }
|
932 |
-
}
|
933 |
-
}
|
934 |
-
// when hash table entry is too long, delete half the entries
|
935 |
-
if (hash_table[h] && stbiw__sbn(hash_table[h]) == 2*quality) {
|
936 |
-
STBIW_MEMMOVE(hash_table[h], hash_table[h]+quality, sizeof(hash_table[h][0])*quality);
|
937 |
-
stbiw__sbn(hash_table[h]) = quality;
|
938 |
-
}
|
939 |
-
stbiw__sbpush(hash_table[h],data+i);
|
940 |
-
|
941 |
-
if (bestloc) {
|
942 |
-
// "lazy matching" - check match at *next* byte, and if it's better, do cur byte as literal
|
943 |
-
h = stbiw__zhash(data+i+1)&(stbiw__ZHASH-1);
|
944 |
-
hlist = hash_table[h];
|
945 |
-
n = stbiw__sbcount(hlist);
|
946 |
-
for (j=0; j < n; ++j) {
|
947 |
-
if (hlist[j]-data > i-32767) {
|
948 |
-
int e = stbiw__zlib_countm(hlist[j], data+i+1, data_len-i-1);
|
949 |
-
if (e > best) { // if next match is better, bail on current match
|
950 |
-
bestloc = NULL;
|
951 |
-
break;
|
952 |
-
}
|
953 |
-
}
|
954 |
-
}
|
955 |
-
}
|
956 |
-
|
957 |
-
if (bestloc) {
|
958 |
-
int d = (int) (data+i - bestloc); // distance back
|
959 |
-
STBIW_ASSERT(d <= 32767 && best <= 258);
|
960 |
-
for (j=0; best > lengthc[j+1]-1; ++j);
|
961 |
-
stbiw__zlib_huff(j+257);
|
962 |
-
if (lengtheb[j]) stbiw__zlib_add(best - lengthc[j], lengtheb[j]);
|
963 |
-
for (j=0; d > distc[j+1]-1; ++j);
|
964 |
-
stbiw__zlib_add(stbiw__zlib_bitrev(j,5),5);
|
965 |
-
if (disteb[j]) stbiw__zlib_add(d - distc[j], disteb[j]);
|
966 |
-
i += best;
|
967 |
-
} else {
|
968 |
-
stbiw__zlib_huffb(data[i]);
|
969 |
-
++i;
|
970 |
-
}
|
971 |
-
}
|
972 |
-
// write out final bytes
|
973 |
-
for (;i < data_len; ++i)
|
974 |
-
stbiw__zlib_huffb(data[i]);
|
975 |
-
stbiw__zlib_huff(256); // end of block
|
976 |
-
// pad with 0 bits to byte boundary
|
977 |
-
while (bitcount)
|
978 |
-
stbiw__zlib_add(0,1);
|
979 |
-
|
980 |
-
for (i=0; i < stbiw__ZHASH; ++i)
|
981 |
-
(void) stbiw__sbfree(hash_table[i]);
|
982 |
-
STBIW_FREE(hash_table);
|
983 |
-
|
984 |
-
// store uncompressed instead if compression was worse
|
985 |
-
if (stbiw__sbn(out) > data_len + 2 + ((data_len+32766)/32767)*5) {
|
986 |
-
stbiw__sbn(out) = 2; // truncate to DEFLATE 32K window and FLEVEL = 1
|
987 |
-
for (j = 0; j < data_len;) {
|
988 |
-
int blocklen = data_len - j;
|
989 |
-
if (blocklen > 32767) blocklen = 32767;
|
990 |
-
stbiw__sbpush(out, data_len - j == blocklen); // BFINAL = ?, BTYPE = 0 -- no compression
|
991 |
-
stbiw__sbpush(out, STBIW_UCHAR(blocklen)); // LEN
|
992 |
-
stbiw__sbpush(out, STBIW_UCHAR(blocklen >> 8));
|
993 |
-
stbiw__sbpush(out, STBIW_UCHAR(~blocklen)); // NLEN
|
994 |
-
stbiw__sbpush(out, STBIW_UCHAR(~blocklen >> 8));
|
995 |
-
memcpy(out+stbiw__sbn(out), data+j, blocklen);
|
996 |
-
stbiw__sbn(out) += blocklen;
|
997 |
-
j += blocklen;
|
998 |
-
}
|
999 |
-
}
|
1000 |
-
|
1001 |
-
{
|
1002 |
-
// compute adler32 on input
|
1003 |
-
unsigned int s1=1, s2=0;
|
1004 |
-
int blocklen = (int) (data_len % 5552);
|
1005 |
-
j=0;
|
1006 |
-
while (j < data_len) {
|
1007 |
-
for (i=0; i < blocklen; ++i) { s1 += data[j+i]; s2 += s1; }
|
1008 |
-
s1 %= 65521; s2 %= 65521;
|
1009 |
-
j += blocklen;
|
1010 |
-
blocklen = 5552;
|
1011 |
-
}
|
1012 |
-
stbiw__sbpush(out, STBIW_UCHAR(s2 >> 8));
|
1013 |
-
stbiw__sbpush(out, STBIW_UCHAR(s2));
|
1014 |
-
stbiw__sbpush(out, STBIW_UCHAR(s1 >> 8));
|
1015 |
-
stbiw__sbpush(out, STBIW_UCHAR(s1));
|
1016 |
-
}
|
1017 |
-
*out_len = stbiw__sbn(out);
|
1018 |
-
// make returned pointer freeable
|
1019 |
-
STBIW_MEMMOVE(stbiw__sbraw(out), out, *out_len);
|
1020 |
-
return (unsigned char *) stbiw__sbraw(out);
|
1021 |
-
#endif // STBIW_ZLIB_COMPRESS
|
1022 |
-
}
|
1023 |
-
|
1024 |
-
static unsigned int stbiw__crc32(unsigned char *buffer, int len)
|
1025 |
-
{
|
1026 |
-
#ifdef STBIW_CRC32
|
1027 |
-
return STBIW_CRC32(buffer, len);
|
1028 |
-
#else
|
1029 |
-
static unsigned int crc_table[256] =
|
1030 |
-
{
|
1031 |
-
0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA, 0x076DC419, 0x706AF48F, 0xE963A535, 0x9E6495A3,
|
1032 |
-
0x0eDB8832, 0x79DCB8A4, 0xE0D5E91E, 0x97D2D988, 0x09B64C2B, 0x7EB17CBD, 0xE7B82D07, 0x90BF1D91,
|
1033 |
-
0x1DB71064, 0x6AB020F2, 0xF3B97148, 0x84BE41DE, 0x1ADAD47D, 0x6DDDE4EB, 0xF4D4B551, 0x83D385C7,
|
1034 |
-
0x136C9856, 0x646BA8C0, 0xFD62F97A, 0x8A65C9EC, 0x14015C4F, 0x63066CD9, 0xFA0F3D63, 0x8D080DF5,
|
1035 |
-
0x3B6E20C8, 0x4C69105E, 0xD56041E4, 0xA2677172, 0x3C03E4D1, 0x4B04D447, 0xD20D85FD, 0xA50AB56B,
|
1036 |
-
0x35B5A8FA, 0x42B2986C, 0xDBBBC9D6, 0xACBCF940, 0x32D86CE3, 0x45DF5C75, 0xDCD60DCF, 0xABD13D59,
|
1037 |
-
0x26D930AC, 0x51DE003A, 0xC8D75180, 0xBFD06116, 0x21B4F4B5, 0x56B3C423, 0xCFBA9599, 0xB8BDA50F,
|
1038 |
-
0x2802B89E, 0x5F058808, 0xC60CD9B2, 0xB10BE924, 0x2F6F7C87, 0x58684C11, 0xC1611DAB, 0xB6662D3D,
|
1039 |
-
0x76DC4190, 0x01DB7106, 0x98D220BC, 0xEFD5102A, 0x71B18589, 0x06B6B51F, 0x9FBFE4A5, 0xE8B8D433,
|
1040 |
-
0x7807C9A2, 0x0F00F934, 0x9609A88E, 0xE10E9818, 0x7F6A0DBB, 0x086D3D2D, 0x91646C97, 0xE6635C01,
|
1041 |
-
0x6B6B51F4, 0x1C6C6162, 0x856530D8, 0xF262004E, 0x6C0695ED, 0x1B01A57B, 0x8208F4C1, 0xF50FC457,
|
1042 |
-
0x65B0D9C6, 0x12B7E950, 0x8BBEB8EA, 0xFCB9887C, 0x62DD1DDF, 0x15DA2D49, 0x8CD37CF3, 0xFBD44C65,
|
1043 |
-
0x4DB26158, 0x3AB551CE, 0xA3BC0074, 0xD4BB30E2, 0x4ADFA541, 0x3DD895D7, 0xA4D1C46D, 0xD3D6F4FB,
|
1044 |
-
0x4369E96A, 0x346ED9FC, 0xAD678846, 0xDA60B8D0, 0x44042D73, 0x33031DE5, 0xAA0A4C5F, 0xDD0D7CC9,
|
1045 |
-
0x5005713C, 0x270241AA, 0xBE0B1010, 0xC90C2086, 0x5768B525, 0x206F85B3, 0xB966D409, 0xCE61E49F,
|
1046 |
-
0x5EDEF90E, 0x29D9C998, 0xB0D09822, 0xC7D7A8B4, 0x59B33D17, 0x2EB40D81, 0xB7BD5C3B, 0xC0BA6CAD,
|
1047 |
-
0xEDB88320, 0x9ABFB3B6, 0x03B6E20C, 0x74B1D29A, 0xEAD54739, 0x9DD277AF, 0x04DB2615, 0x73DC1683,
|
1048 |
-
0xE3630B12, 0x94643B84, 0x0D6D6A3E, 0x7A6A5AA8, 0xE40ECF0B, 0x9309FF9D, 0x0A00AE27, 0x7D079EB1,
|
1049 |
-
0xF00F9344, 0x8708A3D2, 0x1E01F268, 0x6906C2FE, 0xF762575D, 0x806567CB, 0x196C3671, 0x6E6B06E7,
|
1050 |
-
0xFED41B76, 0x89D32BE0, 0x10DA7A5A, 0x67DD4ACC, 0xF9B9DF6F, 0x8EBEEFF9, 0x17B7BE43, 0x60B08ED5,
|
1051 |
-
0xD6D6A3E8, 0xA1D1937E, 0x38D8C2C4, 0x4FDFF252, 0xD1BB67F1, 0xA6BC5767, 0x3FB506DD, 0x48B2364B,
|
1052 |
-
0xD80D2BDA, 0xAF0A1B4C, 0x36034AF6, 0x41047A60, 0xDF60EFC3, 0xA867DF55, 0x316E8EEF, 0x4669BE79,
|
1053 |
-
0xCB61B38C, 0xBC66831A, 0x256FD2A0, 0x5268E236, 0xCC0C7795, 0xBB0B4703, 0x220216B9, 0x5505262F,
|
1054 |
-
0xC5BA3BBE, 0xB2BD0B28, 0x2BB45A92, 0x5CB36A04, 0xC2D7FFA7, 0xB5D0CF31, 0x2CD99E8B, 0x5BDEAE1D,
|
1055 |
-
0x9B64C2B0, 0xEC63F226, 0x756AA39C, 0x026D930A, 0x9C0906A9, 0xEB0E363F, 0x72076785, 0x05005713,
|
1056 |
-
0x95BF4A82, 0xE2B87A14, 0x7BB12BAE, 0x0CB61B38, 0x92D28E9B, 0xE5D5BE0D, 0x7CDCEFB7, 0x0BDBDF21,
|
1057 |
-
0x86D3D2D4, 0xF1D4E242, 0x68DDB3F8, 0x1FDA836E, 0x81BE16CD, 0xF6B9265B, 0x6FB077E1, 0x18B74777,
|
1058 |
-
0x88085AE6, 0xFF0F6A70, 0x66063BCA, 0x11010B5C, 0x8F659EFF, 0xF862AE69, 0x616BFFD3, 0x166CCF45,
|
1059 |
-
0xA00AE278, 0xD70DD2EE, 0x4E048354, 0x3903B3C2, 0xA7672661, 0xD06016F7, 0x4969474D, 0x3E6E77DB,
|
1060 |
-
0xAED16A4A, 0xD9D65ADC, 0x40DF0B66, 0x37D83BF0, 0xA9BCAE53, 0xDEBB9EC5, 0x47B2CF7F, 0x30B5FFE9,
|
1061 |
-
0xBDBDF21C, 0xCABAC28A, 0x53B39330, 0x24B4A3A6, 0xBAD03605, 0xCDD70693, 0x54DE5729, 0x23D967BF,
|
1062 |
-
0xB3667A2E, 0xC4614AB8, 0x5D681B02, 0x2A6F2B94, 0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D
|
1063 |
-
};
|
1064 |
-
|
1065 |
-
unsigned int crc = ~0u;
|
1066 |
-
int i;
|
1067 |
-
for (i=0; i < len; ++i)
|
1068 |
-
crc = (crc >> 8) ^ crc_table[buffer[i] ^ (crc & 0xff)];
|
1069 |
-
return ~crc;
|
1070 |
-
#endif
|
1071 |
-
}
|
1072 |
-
|
1073 |
-
#define stbiw__wpng4(o,a,b,c,d) ((o)[0]=STBIW_UCHAR(a),(o)[1]=STBIW_UCHAR(b),(o)[2]=STBIW_UCHAR(c),(o)[3]=STBIW_UCHAR(d),(o)+=4)
|
1074 |
-
#define stbiw__wp32(data,v) stbiw__wpng4(data, (v)>>24,(v)>>16,(v)>>8,(v));
|
1075 |
-
#define stbiw__wptag(data,s) stbiw__wpng4(data, s[0],s[1],s[2],s[3])
|
1076 |
-
|
1077 |
-
static void stbiw__wpcrc(unsigned char **data, int len)
|
1078 |
-
{
|
1079 |
-
unsigned int crc = stbiw__crc32(*data - len - 4, len+4);
|
1080 |
-
stbiw__wp32(*data, crc);
|
1081 |
-
}
|
1082 |
-
|
1083 |
-
static unsigned char stbiw__paeth(int a, int b, int c)
|
1084 |
-
{
|
1085 |
-
int p = a + b - c, pa = abs(p-a), pb = abs(p-b), pc = abs(p-c);
|
1086 |
-
if (pa <= pb && pa <= pc) return STBIW_UCHAR(a);
|
1087 |
-
if (pb <= pc) return STBIW_UCHAR(b);
|
1088 |
-
return STBIW_UCHAR(c);
|
1089 |
-
}
|
1090 |
-
|
1091 |
-
// @OPTIMIZE: provide an option that always forces left-predict or paeth predict
|
1092 |
-
static void stbiw__encode_png_line(unsigned char *pixels, int stride_bytes, int width, int height, int y, int n, int filter_type, signed char *line_buffer)
|
1093 |
-
{
|
1094 |
-
static int mapping[] = { 0,1,2,3,4 };
|
1095 |
-
static int firstmap[] = { 0,1,0,5,6 };
|
1096 |
-
int *mymap = (y != 0) ? mapping : firstmap;
|
1097 |
-
int i;
|
1098 |
-
int type = mymap[filter_type];
|
1099 |
-
unsigned char *z = pixels + stride_bytes * (stbi__flip_vertically_on_write ? height-1-y : y);
|
1100 |
-
int signed_stride = stbi__flip_vertically_on_write ? -stride_bytes : stride_bytes;
|
1101 |
-
|
1102 |
-
if (type==0) {
|
1103 |
-
memcpy(line_buffer, z, width*n);
|
1104 |
-
return;
|
1105 |
-
}
|
1106 |
-
|
1107 |
-
// first loop isn't optimized since it's just one pixel
|
1108 |
-
for (i = 0; i < n; ++i) {
|
1109 |
-
switch (type) {
|
1110 |
-
case 1: line_buffer[i] = z[i]; break;
|
1111 |
-
case 2: line_buffer[i] = z[i] - z[i-signed_stride]; break;
|
1112 |
-
case 3: line_buffer[i] = z[i] - (z[i-signed_stride]>>1); break;
|
1113 |
-
case 4: line_buffer[i] = (signed char) (z[i] - stbiw__paeth(0,z[i-signed_stride],0)); break;
|
1114 |
-
case 5: line_buffer[i] = z[i]; break;
|
1115 |
-
case 6: line_buffer[i] = z[i]; break;
|
1116 |
-
}
|
1117 |
-
}
|
1118 |
-
switch (type) {
|
1119 |
-
case 1: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - z[i-n]; break;
|
1120 |
-
case 2: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - z[i-signed_stride]; break;
|
1121 |
-
case 3: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - ((z[i-n] + z[i-signed_stride])>>1); break;
|
1122 |
-
case 4: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - stbiw__paeth(z[i-n], z[i-signed_stride], z[i-signed_stride-n]); break;
|
1123 |
-
case 5: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - (z[i-n]>>1); break;
|
1124 |
-
case 6: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - stbiw__paeth(z[i-n], 0,0); break;
|
1125 |
-
}
|
1126 |
-
}
|
1127 |
-
|
1128 |
-
STBIWDEF unsigned char *stbi_write_png_to_mem(const unsigned char *pixels, int stride_bytes, int x, int y, int n, int *out_len, const char* parameters)
|
1129 |
-
{
|
1130 |
-
int force_filter = stbi_write_force_png_filter;
|
1131 |
-
int param_length = 0;
|
1132 |
-
int ctype[5] = { -1, 0, 4, 2, 6 };
|
1133 |
-
unsigned char sig[8] = { 137,80,78,71,13,10,26,10 };
|
1134 |
-
unsigned char *out,*o, *filt, *zlib;
|
1135 |
-
signed char *line_buffer;
|
1136 |
-
int j,zlen;
|
1137 |
-
|
1138 |
-
if (stride_bytes == 0)
|
1139 |
-
stride_bytes = x * n;
|
1140 |
-
|
1141 |
-
if (force_filter >= 5) {
|
1142 |
-
force_filter = -1;
|
1143 |
-
}
|
1144 |
-
|
1145 |
-
filt = (unsigned char *) STBIW_MALLOC((x*n+1) * y); if (!filt) return 0;
|
1146 |
-
line_buffer = (signed char *) STBIW_MALLOC(x * n); if (!line_buffer) { STBIW_FREE(filt); return 0; }
|
1147 |
-
for (j=0; j < y; ++j) {
|
1148 |
-
int filter_type;
|
1149 |
-
if (force_filter > -1) {
|
1150 |
-
filter_type = force_filter;
|
1151 |
-
stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, force_filter, line_buffer);
|
1152 |
-
} else { // Estimate the best filter by running through all of them:
|
1153 |
-
int best_filter = 0, best_filter_val = 0x7fffffff, est, i;
|
1154 |
-
for (filter_type = 0; filter_type < 5; filter_type++) {
|
1155 |
-
stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, filter_type, line_buffer);
|
1156 |
-
|
1157 |
-
// Estimate the entropy of the line using this filter; the less, the better.
|
1158 |
-
est = 0;
|
1159 |
-
for (i = 0; i < x*n; ++i) {
|
1160 |
-
est += abs((signed char) line_buffer[i]);
|
1161 |
-
}
|
1162 |
-
if (est < best_filter_val) {
|
1163 |
-
best_filter_val = est;
|
1164 |
-
best_filter = filter_type;
|
1165 |
-
}
|
1166 |
-
}
|
1167 |
-
if (filter_type != best_filter) { // If the last iteration already got us the best filter, don't redo it
|
1168 |
-
stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, best_filter, line_buffer);
|
1169 |
-
filter_type = best_filter;
|
1170 |
-
}
|
1171 |
-
}
|
1172 |
-
// when we get here, filter_type contains the filter type, and line_buffer contains the data
|
1173 |
-
filt[j*(x*n+1)] = (unsigned char) filter_type;
|
1174 |
-
STBIW_MEMMOVE(filt+j*(x*n+1)+1, line_buffer, x*n);
|
1175 |
-
}
|
1176 |
-
STBIW_FREE(line_buffer);
|
1177 |
-
zlib = stbi_zlib_compress(filt, y*( x*n+1), &zlen, stbi_write_png_compression_level);
|
1178 |
-
STBIW_FREE(filt);
|
1179 |
-
if (!zlib) return 0;
|
1180 |
-
|
1181 |
-
if(parameters != NULL) {
|
1182 |
-
param_length = strlen(parameters);
|
1183 |
-
param_length += strlen("parameters") + 1; // For the name and the null-byte
|
1184 |
-
}
|
1185 |
-
|
1186 |
-
// each tag requires 12 bytes of overhead
|
1187 |
-
out = (unsigned char *) STBIW_MALLOC(8 + 12+13 + 12+zlen + 12 + ((parameters)?(param_length+12):0));
|
1188 |
-
if (!out) return 0;
|
1189 |
-
*out_len = 8 + 12+13 + 12+zlen + 12 + ((parameters)?(param_length+12):0);
|
1190 |
-
|
1191 |
-
o=out;
|
1192 |
-
STBIW_MEMMOVE(o,sig,8); o+= 8;
|
1193 |
-
stbiw__wp32(o, 13); // header length
|
1194 |
-
stbiw__wptag(o, "IHDR");
|
1195 |
-
stbiw__wp32(o, x);
|
1196 |
-
stbiw__wp32(o, y);
|
1197 |
-
*o++ = 8;
|
1198 |
-
*o++ = STBIW_UCHAR(ctype[n]);
|
1199 |
-
*o++ = 0;
|
1200 |
-
*o++ = 0;
|
1201 |
-
*o++ = 0;
|
1202 |
-
stbiw__wpcrc(&o,13);
|
1203 |
-
|
1204 |
-
if(parameters != NULL) {
|
1205 |
-
stbiw__wp32(o, param_length);
|
1206 |
-
stbiw__wptag(o, "tEXt");
|
1207 |
-
STBIW_MEMMOVE(o, "parameters", strlen("parameters"));
|
1208 |
-
o+=strlen("parameters");
|
1209 |
-
*o++ = 0; // Null pyte separator
|
1210 |
-
STBIW_MEMMOVE(o, parameters, strlen(parameters));
|
1211 |
-
o+=strlen(parameters);
|
1212 |
-
stbiw__wpcrc(&o, param_length);
|
1213 |
-
}
|
1214 |
-
|
1215 |
-
stbiw__wp32(o, zlen);
|
1216 |
-
stbiw__wptag(o, "IDAT");
|
1217 |
-
STBIW_MEMMOVE(o, zlib, zlen);
|
1218 |
-
o += zlen;
|
1219 |
-
STBIW_FREE(zlib);
|
1220 |
-
stbiw__wpcrc(&o, zlen);
|
1221 |
-
|
1222 |
-
stbiw__wp32(o,0);
|
1223 |
-
stbiw__wptag(o, "IEND");
|
1224 |
-
stbiw__wpcrc(&o,0);
|
1225 |
-
|
1226 |
-
STBIW_ASSERT(o == out + *out_len);
|
1227 |
-
|
1228 |
-
return out;
|
1229 |
-
}
|
1230 |
-
|
1231 |
-
#ifndef STBI_WRITE_NO_STDIO
|
1232 |
-
STBIWDEF int stbi_write_png(char const *filename, int x, int y, int comp, const void *data, int stride_bytes, const char* parameters)
|
1233 |
-
{
|
1234 |
-
FILE *f;
|
1235 |
-
int len;
|
1236 |
-
unsigned char *png = stbi_write_png_to_mem((const unsigned char *) data, stride_bytes, x, y, comp, &len, parameters);
|
1237 |
-
if (png == NULL) return 0;
|
1238 |
-
|
1239 |
-
f = stbiw__fopen(filename, "wb");
|
1240 |
-
if (!f) { STBIW_FREE(png); return 0; }
|
1241 |
-
fwrite(png, 1, len, f);
|
1242 |
-
fclose(f);
|
1243 |
-
STBIW_FREE(png);
|
1244 |
-
return 1;
|
1245 |
-
}
|
1246 |
-
#endif
|
1247 |
-
|
1248 |
-
STBIWDEF int stbi_write_png_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int stride_bytes)
|
1249 |
-
{
|
1250 |
-
int len;
|
1251 |
-
unsigned char *png = stbi_write_png_to_mem((const unsigned char *) data, stride_bytes, x, y, comp, &len, NULL);
|
1252 |
-
if (png == NULL) return 0;
|
1253 |
-
func(context, png, len);
|
1254 |
-
STBIW_FREE(png);
|
1255 |
-
return 1;
|
1256 |
-
}
|
1257 |
-
|
1258 |
-
|
1259 |
-
/* ***************************************************************************
|
1260 |
-
*
|
1261 |
-
* JPEG writer
|
1262 |
-
*
|
1263 |
-
* This is based on Jon Olick's jo_jpeg.cpp:
|
1264 |
-
* public domain Simple, Minimalistic JPEG writer - http://www.jonolick.com/code.html
|
1265 |
-
*/
|
1266 |
-
|
1267 |
-
static const unsigned char stbiw__jpg_ZigZag[] = { 0,1,5,6,14,15,27,28,2,4,7,13,16,26,29,42,3,8,12,17,25,30,41,43,9,11,18,
|
1268 |
-
24,31,40,44,53,10,19,23,32,39,45,52,54,20,22,33,38,46,51,55,60,21,34,37,47,50,56,59,61,35,36,48,49,57,58,62,63 };
|
1269 |
-
|
1270 |
-
static void stbiw__jpg_writeBits(stbi__write_context *s, int *bitBufP, int *bitCntP, const unsigned short *bs) {
|
1271 |
-
int bitBuf = *bitBufP, bitCnt = *bitCntP;
|
1272 |
-
bitCnt += bs[1];
|
1273 |
-
bitBuf |= bs[0] << (24 - bitCnt);
|
1274 |
-
while(bitCnt >= 8) {
|
1275 |
-
unsigned char c = (bitBuf >> 16) & 255;
|
1276 |
-
stbiw__putc(s, c);
|
1277 |
-
if(c == 255) {
|
1278 |
-
stbiw__putc(s, 0);
|
1279 |
-
}
|
1280 |
-
bitBuf <<= 8;
|
1281 |
-
bitCnt -= 8;
|
1282 |
-
}
|
1283 |
-
*bitBufP = bitBuf;
|
1284 |
-
*bitCntP = bitCnt;
|
1285 |
-
}
|
1286 |
-
|
1287 |
-
static void stbiw__jpg_DCT(float *d0p, float *d1p, float *d2p, float *d3p, float *d4p, float *d5p, float *d6p, float *d7p) {
|
1288 |
-
float d0 = *d0p, d1 = *d1p, d2 = *d2p, d3 = *d3p, d4 = *d4p, d5 = *d5p, d6 = *d6p, d7 = *d7p;
|
1289 |
-
float z1, z2, z3, z4, z5, z11, z13;
|
1290 |
-
|
1291 |
-
float tmp0 = d0 + d7;
|
1292 |
-
float tmp7 = d0 - d7;
|
1293 |
-
float tmp1 = d1 + d6;
|
1294 |
-
float tmp6 = d1 - d6;
|
1295 |
-
float tmp2 = d2 + d5;
|
1296 |
-
float tmp5 = d2 - d5;
|
1297 |
-
float tmp3 = d3 + d4;
|
1298 |
-
float tmp4 = d3 - d4;
|
1299 |
-
|
1300 |
-
// Even part
|
1301 |
-
float tmp10 = tmp0 + tmp3; // phase 2
|
1302 |
-
float tmp13 = tmp0 - tmp3;
|
1303 |
-
float tmp11 = tmp1 + tmp2;
|
1304 |
-
float tmp12 = tmp1 - tmp2;
|
1305 |
-
|
1306 |
-
d0 = tmp10 + tmp11; // phase 3
|
1307 |
-
d4 = tmp10 - tmp11;
|
1308 |
-
|
1309 |
-
z1 = (tmp12 + tmp13) * 0.707106781f; // c4
|
1310 |
-
d2 = tmp13 + z1; // phase 5
|
1311 |
-
d6 = tmp13 - z1;
|
1312 |
-
|
1313 |
-
// Odd part
|
1314 |
-
tmp10 = tmp4 + tmp5; // phase 2
|
1315 |
-
tmp11 = tmp5 + tmp6;
|
1316 |
-
tmp12 = tmp6 + tmp7;
|
1317 |
-
|
1318 |
-
// The rotator is modified from fig 4-8 to avoid extra negations.
|
1319 |
-
z5 = (tmp10 - tmp12) * 0.382683433f; // c6
|
1320 |
-
z2 = tmp10 * 0.541196100f + z5; // c2-c6
|
1321 |
-
z4 = tmp12 * 1.306562965f + z5; // c2+c6
|
1322 |
-
z3 = tmp11 * 0.707106781f; // c4
|
1323 |
-
|
1324 |
-
z11 = tmp7 + z3; // phase 5
|
1325 |
-
z13 = tmp7 - z3;
|
1326 |
-
|
1327 |
-
*d5p = z13 + z2; // phase 6
|
1328 |
-
*d3p = z13 - z2;
|
1329 |
-
*d1p = z11 + z4;
|
1330 |
-
*d7p = z11 - z4;
|
1331 |
-
|
1332 |
-
*d0p = d0; *d2p = d2; *d4p = d4; *d6p = d6;
|
1333 |
-
}
|
1334 |
-
|
1335 |
-
static void stbiw__jpg_calcBits(int val, unsigned short bits[2]) {
|
1336 |
-
int tmp1 = val < 0 ? -val : val;
|
1337 |
-
val = val < 0 ? val-1 : val;
|
1338 |
-
bits[1] = 1;
|
1339 |
-
while(tmp1 >>= 1) {
|
1340 |
-
++bits[1];
|
1341 |
-
}
|
1342 |
-
bits[0] = val & ((1<<bits[1])-1);
|
1343 |
-
}
|
1344 |
-
|
1345 |
-
static int stbiw__jpg_processDU(stbi__write_context *s, int *bitBuf, int *bitCnt, float *CDU, int du_stride, float *fdtbl, int DC, const unsigned short HTDC[256][2], const unsigned short HTAC[256][2]) {
|
1346 |
-
const unsigned short EOB[2] = { HTAC[0x00][0], HTAC[0x00][1] };
|
1347 |
-
const unsigned short M16zeroes[2] = { HTAC[0xF0][0], HTAC[0xF0][1] };
|
1348 |
-
int dataOff, i, j, n, diff, end0pos, x, y;
|
1349 |
-
int DU[64];
|
1350 |
-
|
1351 |
-
// DCT rows
|
1352 |
-
for(dataOff=0, n=du_stride*8; dataOff<n; dataOff+=du_stride) {
|
1353 |
-
stbiw__jpg_DCT(&CDU[dataOff], &CDU[dataOff+1], &CDU[dataOff+2], &CDU[dataOff+3], &CDU[dataOff+4], &CDU[dataOff+5], &CDU[dataOff+6], &CDU[dataOff+7]);
|
1354 |
-
}
|
1355 |
-
// DCT columns
|
1356 |
-
for(dataOff=0; dataOff<8; ++dataOff) {
|
1357 |
-
stbiw__jpg_DCT(&CDU[dataOff], &CDU[dataOff+du_stride], &CDU[dataOff+du_stride*2], &CDU[dataOff+du_stride*3], &CDU[dataOff+du_stride*4],
|
1358 |
-
&CDU[dataOff+du_stride*5], &CDU[dataOff+du_stride*6], &CDU[dataOff+du_stride*7]);
|
1359 |
-
}
|
1360 |
-
// Quantize/descale/zigzag the coefficients
|
1361 |
-
for(y = 0, j=0; y < 8; ++y) {
|
1362 |
-
for(x = 0; x < 8; ++x,++j) {
|
1363 |
-
float v;
|
1364 |
-
i = y*du_stride+x;
|
1365 |
-
v = CDU[i]*fdtbl[j];
|
1366 |
-
// DU[stbiw__jpg_ZigZag[j]] = (int)(v < 0 ? ceilf(v - 0.5f) : floorf(v + 0.5f));
|
1367 |
-
// ceilf() and floorf() are C99, not C89, but I /think/ they're not needed here anyway?
|
1368 |
-
DU[stbiw__jpg_ZigZag[j]] = (int)(v < 0 ? v - 0.5f : v + 0.5f);
|
1369 |
-
}
|
1370 |
-
}
|
1371 |
-
|
1372 |
-
// Encode DC
|
1373 |
-
diff = DU[0] - DC;
|
1374 |
-
if (diff == 0) {
|
1375 |
-
stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTDC[0]);
|
1376 |
-
} else {
|
1377 |
-
unsigned short bits[2];
|
1378 |
-
stbiw__jpg_calcBits(diff, bits);
|
1379 |
-
stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTDC[bits[1]]);
|
1380 |
-
stbiw__jpg_writeBits(s, bitBuf, bitCnt, bits);
|
1381 |
-
}
|
1382 |
-
// Encode ACs
|
1383 |
-
end0pos = 63;
|
1384 |
-
for(; (end0pos>0)&&(DU[end0pos]==0); --end0pos) {
|
1385 |
-
}
|
1386 |
-
// end0pos = first element in reverse order !=0
|
1387 |
-
if(end0pos == 0) {
|
1388 |
-
stbiw__jpg_writeBits(s, bitBuf, bitCnt, EOB);
|
1389 |
-
return DU[0];
|
1390 |
-
}
|
1391 |
-
for(i = 1; i <= end0pos; ++i) {
|
1392 |
-
int startpos = i;
|
1393 |
-
int nrzeroes;
|
1394 |
-
unsigned short bits[2];
|
1395 |
-
for (; DU[i]==0 && i<=end0pos; ++i) {
|
1396 |
-
}
|
1397 |
-
nrzeroes = i-startpos;
|
1398 |
-
if ( nrzeroes >= 16 ) {
|
1399 |
-
int lng = nrzeroes>>4;
|
1400 |
-
int nrmarker;
|
1401 |
-
for (nrmarker=1; nrmarker <= lng; ++nrmarker)
|
1402 |
-
stbiw__jpg_writeBits(s, bitBuf, bitCnt, M16zeroes);
|
1403 |
-
nrzeroes &= 15;
|
1404 |
-
}
|
1405 |
-
stbiw__jpg_calcBits(DU[i], bits);
|
1406 |
-
stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTAC[(nrzeroes<<4)+bits[1]]);
|
1407 |
-
stbiw__jpg_writeBits(s, bitBuf, bitCnt, bits);
|
1408 |
-
}
|
1409 |
-
if(end0pos != 63) {
|
1410 |
-
stbiw__jpg_writeBits(s, bitBuf, bitCnt, EOB);
|
1411 |
-
}
|
1412 |
-
return DU[0];
|
1413 |
-
}
|
1414 |
-
|
1415 |
-
static int stbi_write_jpg_core(stbi__write_context *s, int width, int height, int comp, const void* data, int quality) {
|
1416 |
-
// Constants that don't pollute global namespace
|
1417 |
-
static const unsigned char std_dc_luminance_nrcodes[] = {0,0,1,5,1,1,1,1,1,1,0,0,0,0,0,0,0};
|
1418 |
-
static const unsigned char std_dc_luminance_values[] = {0,1,2,3,4,5,6,7,8,9,10,11};
|
1419 |
-
static const unsigned char std_ac_luminance_nrcodes[] = {0,0,2,1,3,3,2,4,3,5,5,4,4,0,0,1,0x7d};
|
1420 |
-
static const unsigned char std_ac_luminance_values[] = {
|
1421 |
-
0x01,0x02,0x03,0x00,0x04,0x11,0x05,0x12,0x21,0x31,0x41,0x06,0x13,0x51,0x61,0x07,0x22,0x71,0x14,0x32,0x81,0x91,0xa1,0x08,
|
1422 |
-
0x23,0x42,0xb1,0xc1,0x15,0x52,0xd1,0xf0,0x24,0x33,0x62,0x72,0x82,0x09,0x0a,0x16,0x17,0x18,0x19,0x1a,0x25,0x26,0x27,0x28,
|
1423 |
-
0x29,0x2a,0x34,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4a,0x53,0x54,0x55,0x56,0x57,0x58,0x59,
|
1424 |
-
0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x83,0x84,0x85,0x86,0x87,0x88,0x89,
|
1425 |
-
0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,0xb5,0xb6,
|
1426 |
-
0xb7,0xb8,0xb9,0xba,0xc2,0xc3,0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,0xe1,0xe2,
|
1427 |
-
0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf1,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,0xf9,0xfa
|
1428 |
-
};
|
1429 |
-
static const unsigned char std_dc_chrominance_nrcodes[] = {0,0,3,1,1,1,1,1,1,1,1,1,0,0,0,0,0};
|
1430 |
-
static const unsigned char std_dc_chrominance_values[] = {0,1,2,3,4,5,6,7,8,9,10,11};
|
1431 |
-
static const unsigned char std_ac_chrominance_nrcodes[] = {0,0,2,1,2,4,4,3,4,7,5,4,4,0,1,2,0x77};
|
1432 |
-
static const unsigned char std_ac_chrominance_values[] = {
|
1433 |
-
0x00,0x01,0x02,0x03,0x11,0x04,0x05,0x21,0x31,0x06,0x12,0x41,0x51,0x07,0x61,0x71,0x13,0x22,0x32,0x81,0x08,0x14,0x42,0x91,
|
1434 |
-
0xa1,0xb1,0xc1,0x09,0x23,0x33,0x52,0xf0,0x15,0x62,0x72,0xd1,0x0a,0x16,0x24,0x34,0xe1,0x25,0xf1,0x17,0x18,0x19,0x1a,0x26,
|
1435 |
-
0x27,0x28,0x29,0x2a,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4a,0x53,0x54,0x55,0x56,0x57,0x58,
|
1436 |
-
0x59,0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x82,0x83,0x84,0x85,0x86,0x87,
|
1437 |
-
0x88,0x89,0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,
|
1438 |
-
0xb5,0xb6,0xb7,0xb8,0xb9,0xba,0xc2,0xc3,0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,
|
1439 |
-
0xe2,0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,0xf9,0xfa
|
1440 |
-
};
|
1441 |
-
// Huffman tables
|
1442 |
-
static const unsigned short YDC_HT[256][2] = { {0,2},{2,3},{3,3},{4,3},{5,3},{6,3},{14,4},{30,5},{62,6},{126,7},{254,8},{510,9}};
|
1443 |
-
static const unsigned short UVDC_HT[256][2] = { {0,2},{1,2},{2,2},{6,3},{14,4},{30,5},{62,6},{126,7},{254,8},{510,9},{1022,10},{2046,11}};
|
1444 |
-
static const unsigned short YAC_HT[256][2] = {
|
1445 |
-
{10,4},{0,2},{1,2},{4,3},{11,4},{26,5},{120,7},{248,8},{1014,10},{65410,16},{65411,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1446 |
-
{12,4},{27,5},{121,7},{502,9},{2038,11},{65412,16},{65413,16},{65414,16},{65415,16},{65416,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1447 |
-
{28,5},{249,8},{1015,10},{4084,12},{65417,16},{65418,16},{65419,16},{65420,16},{65421,16},{65422,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1448 |
-
{58,6},{503,9},{4085,12},{65423,16},{65424,16},{65425,16},{65426,16},{65427,16},{65428,16},{65429,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1449 |
-
{59,6},{1016,10},{65430,16},{65431,16},{65432,16},{65433,16},{65434,16},{65435,16},{65436,16},{65437,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1450 |
-
{122,7},{2039,11},{65438,16},{65439,16},{65440,16},{65441,16},{65442,16},{65443,16},{65444,16},{65445,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1451 |
-
{123,7},{4086,12},{65446,16},{65447,16},{65448,16},{65449,16},{65450,16},{65451,16},{65452,16},{65453,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1452 |
-
{250,8},{4087,12},{65454,16},{65455,16},{65456,16},{65457,16},{65458,16},{65459,16},{65460,16},{65461,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1453 |
-
{504,9},{32704,15},{65462,16},{65463,16},{65464,16},{65465,16},{65466,16},{65467,16},{65468,16},{65469,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1454 |
-
{505,9},{65470,16},{65471,16},{65472,16},{65473,16},{65474,16},{65475,16},{65476,16},{65477,16},{65478,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1455 |
-
{506,9},{65479,16},{65480,16},{65481,16},{65482,16},{65483,16},{65484,16},{65485,16},{65486,16},{65487,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1456 |
-
{1017,10},{65488,16},{65489,16},{65490,16},{65491,16},{65492,16},{65493,16},{65494,16},{65495,16},{65496,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1457 |
-
{1018,10},{65497,16},{65498,16},{65499,16},{65500,16},{65501,16},{65502,16},{65503,16},{65504,16},{65505,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1458 |
-
{2040,11},{65506,16},{65507,16},{65508,16},{65509,16},{65510,16},{65511,16},{65512,16},{65513,16},{65514,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1459 |
-
{65515,16},{65516,16},{65517,16},{65518,16},{65519,16},{65520,16},{65521,16},{65522,16},{65523,16},{65524,16},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1460 |
-
{2041,11},{65525,16},{65526,16},{65527,16},{65528,16},{65529,16},{65530,16},{65531,16},{65532,16},{65533,16},{65534,16},{0,0},{0,0},{0,0},{0,0},{0,0}
|
1461 |
-
};
|
1462 |
-
static const unsigned short UVAC_HT[256][2] = {
|
1463 |
-
{0,2},{1,2},{4,3},{10,4},{24,5},{25,5},{56,6},{120,7},{500,9},{1014,10},{4084,12},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1464 |
-
{11,4},{57,6},{246,8},{501,9},{2038,11},{4085,12},{65416,16},{65417,16},{65418,16},{65419,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1465 |
-
{26,5},{247,8},{1015,10},{4086,12},{32706,15},{65420,16},{65421,16},{65422,16},{65423,16},{65424,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1466 |
-
{27,5},{248,8},{1016,10},{4087,12},{65425,16},{65426,16},{65427,16},{65428,16},{65429,16},{65430,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1467 |
-
{58,6},{502,9},{65431,16},{65432,16},{65433,16},{65434,16},{65435,16},{65436,16},{65437,16},{65438,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1468 |
-
{59,6},{1017,10},{65439,16},{65440,16},{65441,16},{65442,16},{65443,16},{65444,16},{65445,16},{65446,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1469 |
-
{121,7},{2039,11},{65447,16},{65448,16},{65449,16},{65450,16},{65451,16},{65452,16},{65453,16},{65454,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1470 |
-
{122,7},{2040,11},{65455,16},{65456,16},{65457,16},{65458,16},{65459,16},{65460,16},{65461,16},{65462,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1471 |
-
{249,8},{65463,16},{65464,16},{65465,16},{65466,16},{65467,16},{65468,16},{65469,16},{65470,16},{65471,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1472 |
-
{503,9},{65472,16},{65473,16},{65474,16},{65475,16},{65476,16},{65477,16},{65478,16},{65479,16},{65480,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1473 |
-
{504,9},{65481,16},{65482,16},{65483,16},{65484,16},{65485,16},{65486,16},{65487,16},{65488,16},{65489,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1474 |
-
{505,9},{65490,16},{65491,16},{65492,16},{65493,16},{65494,16},{65495,16},{65496,16},{65497,16},{65498,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1475 |
-
{506,9},{65499,16},{65500,16},{65501,16},{65502,16},{65503,16},{65504,16},{65505,16},{65506,16},{65507,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1476 |
-
{2041,11},{65508,16},{65509,16},{65510,16},{65511,16},{65512,16},{65513,16},{65514,16},{65515,16},{65516,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1477 |
-
{16352,14},{65517,16},{65518,16},{65519,16},{65520,16},{65521,16},{65522,16},{65523,16},{65524,16},{65525,16},{0,0},{0,0},{0,0},{0,0},{0,0},
|
1478 |
-
{1018,10},{32707,15},{65526,16},{65527,16},{65528,16},{65529,16},{65530,16},{65531,16},{65532,16},{65533,16},{65534,16},{0,0},{0,0},{0,0},{0,0},{0,0}
|
1479 |
-
};
|
1480 |
-
static const int YQT[] = {16,11,10,16,24,40,51,61,12,12,14,19,26,58,60,55,14,13,16,24,40,57,69,56,14,17,22,29,51,87,80,62,18,22,
|
1481 |
-
37,56,68,109,103,77,24,35,55,64,81,104,113,92,49,64,78,87,103,121,120,101,72,92,95,98,112,100,103,99};
|
1482 |
-
static const int UVQT[] = {17,18,24,47,99,99,99,99,18,21,26,66,99,99,99,99,24,26,56,99,99,99,99,99,47,66,99,99,99,99,99,99,
|
1483 |
-
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};
|
1484 |
-
static const float aasf[] = { 1.0f * 2.828427125f, 1.387039845f * 2.828427125f, 1.306562965f * 2.828427125f, 1.175875602f * 2.828427125f,
|
1485 |
-
1.0f * 2.828427125f, 0.785694958f * 2.828427125f, 0.541196100f * 2.828427125f, 0.275899379f * 2.828427125f };
|
1486 |
-
|
1487 |
-
int row, col, i, k, subsample;
|
1488 |
-
float fdtbl_Y[64], fdtbl_UV[64];
|
1489 |
-
unsigned char YTable[64], UVTable[64];
|
1490 |
-
|
1491 |
-
if(!data || !width || !height || comp > 4 || comp < 1) {
|
1492 |
-
return 0;
|
1493 |
-
}
|
1494 |
-
|
1495 |
-
quality = quality ? quality : 90;
|
1496 |
-
subsample = quality <= 90 ? 1 : 0;
|
1497 |
-
quality = quality < 1 ? 1 : quality > 100 ? 100 : quality;
|
1498 |
-
quality = quality < 50 ? 5000 / quality : 200 - quality * 2;
|
1499 |
-
|
1500 |
-
for(i = 0; i < 64; ++i) {
|
1501 |
-
int uvti, yti = (YQT[i]*quality+50)/100;
|
1502 |
-
YTable[stbiw__jpg_ZigZag[i]] = (unsigned char) (yti < 1 ? 1 : yti > 255 ? 255 : yti);
|
1503 |
-
uvti = (UVQT[i]*quality+50)/100;
|
1504 |
-
UVTable[stbiw__jpg_ZigZag[i]] = (unsigned char) (uvti < 1 ? 1 : uvti > 255 ? 255 : uvti);
|
1505 |
-
}
|
1506 |
-
|
1507 |
-
for(row = 0, k = 0; row < 8; ++row) {
|
1508 |
-
for(col = 0; col < 8; ++col, ++k) {
|
1509 |
-
fdtbl_Y[k] = 1 / (YTable [stbiw__jpg_ZigZag[k]] * aasf[row] * aasf[col]);
|
1510 |
-
fdtbl_UV[k] = 1 / (UVTable[stbiw__jpg_ZigZag[k]] * aasf[row] * aasf[col]);
|
1511 |
-
}
|
1512 |
-
}
|
1513 |
-
|
1514 |
-
// Write Headers
|
1515 |
-
{
|
1516 |
-
static const unsigned char head0[] = { 0xFF,0xD8,0xFF,0xE0,0,0x10,'J','F','I','F',0,1,1,0,0,1,0,1,0,0,0xFF,0xDB,0,0x84,0 };
|
1517 |
-
static const unsigned char head2[] = { 0xFF,0xDA,0,0xC,3,1,0,2,0x11,3,0x11,0,0x3F,0 };
|
1518 |
-
const unsigned char head1[] = { 0xFF,0xC0,0,0x11,8,(unsigned char)(height>>8),STBIW_UCHAR(height),(unsigned char)(width>>8),STBIW_UCHAR(width),
|
1519 |
-
3,1,(unsigned char)(subsample?0x22:0x11),0,2,0x11,1,3,0x11,1,0xFF,0xC4,0x01,0xA2,0 };
|
1520 |
-
s->func(s->context, (void*)head0, sizeof(head0));
|
1521 |
-
s->func(s->context, (void*)YTable, sizeof(YTable));
|
1522 |
-
stbiw__putc(s, 1);
|
1523 |
-
s->func(s->context, UVTable, sizeof(UVTable));
|
1524 |
-
s->func(s->context, (void*)head1, sizeof(head1));
|
1525 |
-
s->func(s->context, (void*)(std_dc_luminance_nrcodes+1), sizeof(std_dc_luminance_nrcodes)-1);
|
1526 |
-
s->func(s->context, (void*)std_dc_luminance_values, sizeof(std_dc_luminance_values));
|
1527 |
-
stbiw__putc(s, 0x10); // HTYACinfo
|
1528 |
-
s->func(s->context, (void*)(std_ac_luminance_nrcodes+1), sizeof(std_ac_luminance_nrcodes)-1);
|
1529 |
-
s->func(s->context, (void*)std_ac_luminance_values, sizeof(std_ac_luminance_values));
|
1530 |
-
stbiw__putc(s, 1); // HTUDCinfo
|
1531 |
-
s->func(s->context, (void*)(std_dc_chrominance_nrcodes+1), sizeof(std_dc_chrominance_nrcodes)-1);
|
1532 |
-
s->func(s->context, (void*)std_dc_chrominance_values, sizeof(std_dc_chrominance_values));
|
1533 |
-
stbiw__putc(s, 0x11); // HTUACinfo
|
1534 |
-
s->func(s->context, (void*)(std_ac_chrominance_nrcodes+1), sizeof(std_ac_chrominance_nrcodes)-1);
|
1535 |
-
s->func(s->context, (void*)std_ac_chrominance_values, sizeof(std_ac_chrominance_values));
|
1536 |
-
s->func(s->context, (void*)head2, sizeof(head2));
|
1537 |
-
}
|
1538 |
-
|
1539 |
-
// Encode 8x8 macroblocks
|
1540 |
-
{
|
1541 |
-
static const unsigned short fillBits[] = {0x7F, 7};
|
1542 |
-
int DCY=0, DCU=0, DCV=0;
|
1543 |
-
int bitBuf=0, bitCnt=0;
|
1544 |
-
// comp == 2 is grey+alpha (alpha is ignored)
|
1545 |
-
int ofsG = comp > 2 ? 1 : 0, ofsB = comp > 2 ? 2 : 0;
|
1546 |
-
const unsigned char *dataR = (const unsigned char *)data;
|
1547 |
-
const unsigned char *dataG = dataR + ofsG;
|
1548 |
-
const unsigned char *dataB = dataR + ofsB;
|
1549 |
-
int x, y, pos;
|
1550 |
-
if(subsample) {
|
1551 |
-
for(y = 0; y < height; y += 16) {
|
1552 |
-
for(x = 0; x < width; x += 16) {
|
1553 |
-
float Y[256], U[256], V[256];
|
1554 |
-
for(row = y, pos = 0; row < y+16; ++row) {
|
1555 |
-
// row >= height => use last input row
|
1556 |
-
int clamped_row = (row < height) ? row : height - 1;
|
1557 |
-
int base_p = (stbi__flip_vertically_on_write ? (height-1-clamped_row) : clamped_row)*width*comp;
|
1558 |
-
for(col = x; col < x+16; ++col, ++pos) {
|
1559 |
-
// if col >= width => use pixel from last input column
|
1560 |
-
int p = base_p + ((col < width) ? col : (width-1))*comp;
|
1561 |
-
float r = dataR[p], g = dataG[p], b = dataB[p];
|
1562 |
-
Y[pos]= +0.29900f*r + 0.58700f*g + 0.11400f*b - 128;
|
1563 |
-
U[pos]= -0.16874f*r - 0.33126f*g + 0.50000f*b;
|
1564 |
-
V[pos]= +0.50000f*r - 0.41869f*g - 0.08131f*b;
|
1565 |
-
}
|
1566 |
-
}
|
1567 |
-
DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+0, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
|
1568 |
-
DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+8, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
|
1569 |
-
DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+128, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
|
1570 |
-
DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+136, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
|
1571 |
-
|
1572 |
-
// subsample U,V
|
1573 |
-
{
|
1574 |
-
float subU[64], subV[64];
|
1575 |
-
int yy, xx;
|
1576 |
-
for(yy = 0, pos = 0; yy < 8; ++yy) {
|
1577 |
-
for(xx = 0; xx < 8; ++xx, ++pos) {
|
1578 |
-
int j = yy*32+xx*2;
|
1579 |
-
subU[pos] = (U[j+0] + U[j+1] + U[j+16] + U[j+17]) * 0.25f;
|
1580 |
-
subV[pos] = (V[j+0] + V[j+1] + V[j+16] + V[j+17]) * 0.25f;
|
1581 |
-
}
|
1582 |
-
}
|
1583 |
-
DCU = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, subU, 8, fdtbl_UV, DCU, UVDC_HT, UVAC_HT);
|
1584 |
-
DCV = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, subV, 8, fdtbl_UV, DCV, UVDC_HT, UVAC_HT);
|
1585 |
-
}
|
1586 |
-
}
|
1587 |
-
}
|
1588 |
-
} else {
|
1589 |
-
for(y = 0; y < height; y += 8) {
|
1590 |
-
for(x = 0; x < width; x += 8) {
|
1591 |
-
float Y[64], U[64], V[64];
|
1592 |
-
for(row = y, pos = 0; row < y+8; ++row) {
|
1593 |
-
// row >= height => use last input row
|
1594 |
-
int clamped_row = (row < height) ? row : height - 1;
|
1595 |
-
int base_p = (stbi__flip_vertically_on_write ? (height-1-clamped_row) : clamped_row)*width*comp;
|
1596 |
-
for(col = x; col < x+8; ++col, ++pos) {
|
1597 |
-
// if col >= width => use pixel from last input column
|
1598 |
-
int p = base_p + ((col < width) ? col : (width-1))*comp;
|
1599 |
-
float r = dataR[p], g = dataG[p], b = dataB[p];
|
1600 |
-
Y[pos]= +0.29900f*r + 0.58700f*g + 0.11400f*b - 128;
|
1601 |
-
U[pos]= -0.16874f*r - 0.33126f*g + 0.50000f*b;
|
1602 |
-
V[pos]= +0.50000f*r - 0.41869f*g - 0.08131f*b;
|
1603 |
-
}
|
1604 |
-
}
|
1605 |
-
|
1606 |
-
DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y, 8, fdtbl_Y, DCY, YDC_HT, YAC_HT);
|
1607 |
-
DCU = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, U, 8, fdtbl_UV, DCU, UVDC_HT, UVAC_HT);
|
1608 |
-
DCV = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, V, 8, fdtbl_UV, DCV, UVDC_HT, UVAC_HT);
|
1609 |
-
}
|
1610 |
-
}
|
1611 |
-
}
|
1612 |
-
|
1613 |
-
// Do the bit alignment of the EOI marker
|
1614 |
-
stbiw__jpg_writeBits(s, &bitBuf, &bitCnt, fillBits);
|
1615 |
-
}
|
1616 |
-
|
1617 |
-
// EOI
|
1618 |
-
stbiw__putc(s, 0xFF);
|
1619 |
-
stbiw__putc(s, 0xD9);
|
1620 |
-
|
1621 |
-
return 1;
|
1622 |
-
}
|
1623 |
-
|
1624 |
-
STBIWDEF int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality)
|
1625 |
-
{
|
1626 |
-
stbi__write_context s = { 0 };
|
1627 |
-
stbi__start_write_callbacks(&s, func, context);
|
1628 |
-
return stbi_write_jpg_core(&s, x, y, comp, (void *) data, quality);
|
1629 |
-
}
|
1630 |
-
|
1631 |
-
|
1632 |
-
#ifndef STBI_WRITE_NO_STDIO
|
1633 |
-
STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void *data, int quality)
|
1634 |
-
{
|
1635 |
-
stbi__write_context s = { 0 };
|
1636 |
-
if (stbi__start_write_file(&s,filename)) {
|
1637 |
-
int r = stbi_write_jpg_core(&s, x, y, comp, data, quality);
|
1638 |
-
stbi__end_write_file(&s);
|
1639 |
-
return r;
|
1640 |
-
} else
|
1641 |
-
return 0;
|
1642 |
-
}
|
1643 |
-
#endif
|
1644 |
-
|
1645 |
-
#endif // STB_IMAGE_WRITE_IMPLEMENTATION
|
1646 |
-
|
1647 |
-
/* Revision history
|
1648 |
-
1.16 (2021-07-11)
|
1649 |
-
make Deflate code emit uncompressed blocks when it would otherwise expand
|
1650 |
-
support writing BMPs with alpha channel
|
1651 |
-
1.15 (2020-07-13) unknown
|
1652 |
-
1.14 (2020-02-02) updated JPEG writer to downsample chroma channels
|
1653 |
-
1.13
|
1654 |
-
1.12
|
1655 |
-
1.11 (2019-08-11)
|
1656 |
-
|
1657 |
-
1.10 (2019-02-07)
|
1658 |
-
support utf8 filenames in Windows; fix warnings and platform ifdefs
|
1659 |
-
1.09 (2018-02-11)
|
1660 |
-
fix typo in zlib quality API, improve STB_I_W_STATIC in C++
|
1661 |
-
1.08 (2018-01-29)
|
1662 |
-
add stbi__flip_vertically_on_write, external zlib, zlib quality, choose PNG filter
|
1663 |
-
1.07 (2017-07-24)
|
1664 |
-
doc fix
|
1665 |
-
1.06 (2017-07-23)
|
1666 |
-
writing JPEG (using Jon Olick's code)
|
1667 |
-
1.05 ???
|
1668 |
-
1.04 (2017-03-03)
|
1669 |
-
monochrome BMP expansion
|
1670 |
-
1.03 ???
|
1671 |
-
1.02 (2016-04-02)
|
1672 |
-
avoid allocating large structures on the stack
|
1673 |
-
1.01 (2016-01-16)
|
1674 |
-
STBIW_REALLOC_SIZED: support allocators with no realloc support
|
1675 |
-
avoid race-condition in crc initialization
|
1676 |
-
minor compile issues
|
1677 |
-
1.00 (2015-09-14)
|
1678 |
-
installable file IO function
|
1679 |
-
0.99 (2015-09-13)
|
1680 |
-
warning fixes; TGA rle support
|
1681 |
-
0.98 (2015-04-08)
|
1682 |
-
added STBIW_MALLOC, STBIW_ASSERT etc
|
1683 |
-
0.97 (2015-01-18)
|
1684 |
-
fixed HDR asserts, rewrote HDR rle logic
|
1685 |
-
0.96 (2015-01-17)
|
1686 |
-
add HDR output
|
1687 |
-
fix monochrome BMP
|
1688 |
-
0.95 (2014-08-17)
|
1689 |
-
add monochrome TGA output
|
1690 |
-
0.94 (2014-05-31)
|
1691 |
-
rename private functions to avoid conflicts with stb_image.h
|
1692 |
-
0.93 (2014-05-27)
|
1693 |
-
warning fixes
|
1694 |
-
0.92 (2010-08-01)
|
1695 |
-
casts to unsigned char to fix warnings
|
1696 |
-
0.91 (2010-07-17)
|
1697 |
-
first public release
|
1698 |
-
0.90 first internal release
|
1699 |
-
*/
|
1700 |
-
|
1701 |
-
/*
|
1702 |
-
------------------------------------------------------------------------------
|
1703 |
-
This software is available under 2 licenses -- choose whichever you prefer.
|
1704 |
-
------------------------------------------------------------------------------
|
1705 |
-
ALTERNATIVE A - MIT License
|
1706 |
-
Copyright (c) 2017 Sean Barrett
|
1707 |
-
Permission is hereby granted, free of charge, to any person obtaining a copy of
|
1708 |
-
this software and associated documentation files (the "Software"), to deal in
|
1709 |
-
the Software without restriction, including without limitation the rights to
|
1710 |
-
use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
|
1711 |
-
of the Software, and to permit persons to whom the Software is furnished to do
|
1712 |
-
so, subject to the following conditions:
|
1713 |
-
The above copyright notice and this permission notice shall be included in all
|
1714 |
-
copies or substantial portions of the Software.
|
1715 |
-
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
1716 |
-
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
1717 |
-
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
1718 |
-
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
1719 |
-
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
1720 |
-
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
1721 |
-
SOFTWARE.
|
1722 |
-
------------------------------------------------------------------------------
|
1723 |
-
ALTERNATIVE B - Public Domain (www.unlicense.org)
|
1724 |
-
This is free and unencumbered software released into the public domain.
|
1725 |
-
Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
|
1726 |
-
software, either in source code form or as a compiled binary, for any purpose,
|
1727 |
-
commercial or non-commercial, and by any means.
|
1728 |
-
In jurisdictions that recognize copyright laws, the author or authors of this
|
1729 |
-
software dedicate any and all copyright interest in the software to the public
|
1730 |
-
domain. We make this dedication for the benefit of the public at large and to
|
1731 |
-
the detriment of our heirs and successors. We intend this dedication to be an
|
1732 |
-
overt act of relinquishment in perpetuity of all present and future rights to
|
1733 |
-
this software under copyright law.
|
1734 |
-
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
1735 |
-
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
1736 |
-
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
1737 |
-
AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
|
1738 |
-
ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
|
1739 |
-
WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
1740 |
-
------------------------------------------------------------------------------
|
1741 |
-
*/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/models/.gitignore
DELETED
@@ -1,5 +0,0 @@
|
|
1 |
-
*.bin
|
2 |
-
*.ckpt
|
3 |
-
*.safetensor
|
4 |
-
*.safetensors
|
5 |
-
*.log
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/models/README.md
DELETED
@@ -1,26 +0,0 @@
|
|
1 |
-
# Model Convert Script
|
2 |
-
|
3 |
-
## Requirements
|
4 |
-
|
5 |
-
- vocab.json, from https://huggingface.co/openai/clip-vit-large-patch14/raw/main/vocab.json
|
6 |
-
|
7 |
-
|
8 |
-
```shell
|
9 |
-
pip install -r requirements.txt
|
10 |
-
```
|
11 |
-
|
12 |
-
## Usage
|
13 |
-
```
|
14 |
-
usage: convert.py [-h] [--out_type {f32,f16,q4_0,q4_1,q5_0,q5_1,q8_0}] [--out_file OUT_FILE] model_path
|
15 |
-
|
16 |
-
Convert Stable Diffuison model to GGML compatible file format
|
17 |
-
|
18 |
-
positional arguments:
|
19 |
-
model_path model file path (*.pth, *.pt, *.ckpt, *.safetensors)
|
20 |
-
|
21 |
-
options:
|
22 |
-
-h, --help show this help message and exit
|
23 |
-
--out_type {f32,f16,q4_0,q4_1,q5_0,q5_1,q8_0}
|
24 |
-
output format (default: based on input)
|
25 |
-
--out_file OUT_FILE path to write to; default: based on input and current working directory
|
26 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/models/convert.py
DELETED
@@ -1,385 +0,0 @@
|
|
1 |
-
import struct
|
2 |
-
import json
|
3 |
-
import os
|
4 |
-
|
5 |
-
import numpy as np
|
6 |
-
import torch
|
7 |
-
import safetensors.torch
|
8 |
-
|
9 |
-
this_file_dir = os.path.dirname(__file__)
|
10 |
-
vocab_dir = this_file_dir
|
11 |
-
|
12 |
-
SD1 = 0
|
13 |
-
SD2 = 1
|
14 |
-
|
15 |
-
ggml_ftype_str_to_int = {
|
16 |
-
"f32": 0,
|
17 |
-
"f16": 1,
|
18 |
-
"q4_0": 2,
|
19 |
-
"q4_1": 3,
|
20 |
-
"q5_0": 8,
|
21 |
-
"q5_1": 9,
|
22 |
-
"q8_0": 7
|
23 |
-
}
|
24 |
-
|
25 |
-
ggml_ttype_str_to_int = {
|
26 |
-
"f32": 0,
|
27 |
-
"f16": 1,
|
28 |
-
"q4_0": 2,
|
29 |
-
"q4_1": 3,
|
30 |
-
"q5_0": 6,
|
31 |
-
"q5_1": 7,
|
32 |
-
"q8_0": 8
|
33 |
-
}
|
34 |
-
|
35 |
-
QK4_0 = 32
|
36 |
-
def quantize_q4_0(x):
|
37 |
-
assert x.shape[-1] % QK4_0 == 0 and x.shape[-1] > QK4_0
|
38 |
-
x = x.reshape(-1, QK4_0)
|
39 |
-
max = np.take_along_axis(x, np.argmax(np.abs(x), axis=-1)[:, np.newaxis], axis=-1)
|
40 |
-
d = max / -8
|
41 |
-
qs = ((x / d) + 8).round().clip(min=0, max=15).astype(np.int8)
|
42 |
-
half = QK4_0 // 2
|
43 |
-
qs = qs[:, :half] | (qs[:, half:] << 4)
|
44 |
-
d = d.astype(np.float16).view(np.int8)
|
45 |
-
y = np.concatenate((d, qs), axis=-1)
|
46 |
-
return y
|
47 |
-
|
48 |
-
QK4_1 = 32
|
49 |
-
def quantize_q4_1(x):
|
50 |
-
assert x.shape[-1] % QK4_1 == 0 and x.shape[-1] > QK4_1
|
51 |
-
x = x.reshape(-1, QK4_1)
|
52 |
-
min = np.min(x, axis=-1, keepdims=True)
|
53 |
-
max = np.max(x, axis=-1, keepdims=True)
|
54 |
-
d = (max - min) / ((1 << 4) - 1)
|
55 |
-
qs = ((x - min) / d).round().clip(min=0, max=15).astype(np.int8)
|
56 |
-
half = QK4_1 // 2
|
57 |
-
qs = qs[:, :half] | (qs[:, half:] << 4)
|
58 |
-
d = d.astype(np.float16).view(np.int8)
|
59 |
-
m = min.astype(np.float16).view(np.int8)
|
60 |
-
y = np.concatenate((d, m, qs), axis=-1)
|
61 |
-
return y
|
62 |
-
|
63 |
-
QK5_0 = 32
|
64 |
-
def quantize_q5_0(x):
|
65 |
-
assert x.shape[-1] % QK5_0 == 0 and x.shape[-1] > QK5_0
|
66 |
-
x = x.reshape(-1, QK5_0)
|
67 |
-
max = np.take_along_axis(x, np.argmax(np.abs(x), axis=-1)[:, np.newaxis], axis=-1)
|
68 |
-
d = max / -16
|
69 |
-
xi = ((x / d) + 16).round().clip(min=0, max=31).astype(np.int8)
|
70 |
-
half = QK5_0 // 2
|
71 |
-
qs = (xi[:, :half] & 0x0F) | (xi[:, half:] << 4)
|
72 |
-
qh = np.zeros(qs.shape[:-1], dtype=np.int32)
|
73 |
-
for i in range(QK5_0):
|
74 |
-
qh |= ((xi[:, i] & 0x10) >> 4).astype(np.int32) << i
|
75 |
-
d = d.astype(np.float16).view(np.int8)
|
76 |
-
qh = qh[..., np.newaxis].view(np.int8)
|
77 |
-
y = np.concatenate((d, qh, qs), axis=-1)
|
78 |
-
return y
|
79 |
-
|
80 |
-
QK5_1 = 32
|
81 |
-
def quantize_q5_1(x):
|
82 |
-
assert x.shape[-1] % QK5_1 == 0 and x.shape[-1] > QK5_1
|
83 |
-
x = x.reshape(-1, QK5_1)
|
84 |
-
min = np.min(x, axis=-1, keepdims=True)
|
85 |
-
max = np.max(x, axis=-1, keepdims=True)
|
86 |
-
d = (max - min) / ((1 << 5) - 1)
|
87 |
-
xi = ((x - min) / d).round().clip(min=0, max=31).astype(np.int8)
|
88 |
-
half = QK5_1//2
|
89 |
-
qs = (xi[:, :half] & 0x0F) | (xi[:, half:] << 4)
|
90 |
-
qh = np.zeros(xi.shape[:-1], dtype=np.int32)
|
91 |
-
for i in range(QK5_1):
|
92 |
-
qh |= ((xi[:, i] & 0x10) >> 4).astype(np.int32) << i
|
93 |
-
d = d.astype(np.float16).view(np.int8)
|
94 |
-
m = min.astype(np.float16).view(np.int8)
|
95 |
-
qh = qh[..., np.newaxis].view(np.int8)
|
96 |
-
ndarray = np.concatenate((d, m, qh, qs), axis=-1)
|
97 |
-
return ndarray
|
98 |
-
|
99 |
-
QK8_0 = 32
|
100 |
-
def quantize_q8_0(x):
|
101 |
-
assert x.shape[-1] % QK8_0 == 0 and x.shape[-1] > QK8_0
|
102 |
-
x = x.reshape(-1, QK8_0)
|
103 |
-
amax = np.max(np.abs(x), axis=-1, keepdims=True)
|
104 |
-
d = amax / ((1 << 7) - 1)
|
105 |
-
qs = (x / d).round().clip(min=-128, max=127).astype(np.int8)
|
106 |
-
d = d.astype(np.float16).view(np.int8)
|
107 |
-
y = np.concatenate((d, qs), axis=-1)
|
108 |
-
return y
|
109 |
-
|
110 |
-
# copy from https://github.com/openai/CLIP/blob/main/clip/simple_tokenizer.py#L16
|
111 |
-
def bytes_to_unicode():
|
112 |
-
"""
|
113 |
-
Returns list of utf-8 byte and a corresponding list of unicode strings.
|
114 |
-
The reversible bpe codes work on unicode strings.
|
115 |
-
This means you need a large # of unicode characters in your vocab if you want to avoid UNKs.
|
116 |
-
When you're at something like a 10B token dataset you end up needing around 5K for decent coverage.
|
117 |
-
This is a significant percentage of your normal, say, 32K bpe vocab.
|
118 |
-
To avoid that, we want lookup tables between utf-8 bytes and unicode strings.
|
119 |
-
And avoids mapping to whitespace/control characters the bpe code barfs on.
|
120 |
-
"""
|
121 |
-
bs = list(range(ord("!"), ord("~")+1))+list(range(ord("¡"), ord("¬")+1))+list(range(ord("®"), ord("ÿ")+1))
|
122 |
-
cs = bs[:]
|
123 |
-
n = 0
|
124 |
-
for b in range(2**8):
|
125 |
-
if b not in bs:
|
126 |
-
bs.append(b)
|
127 |
-
cs.append(2**8+n)
|
128 |
-
n += 1
|
129 |
-
cs = [chr(n) for n in cs]
|
130 |
-
return dict(zip(bs, cs))
|
131 |
-
|
132 |
-
def load_model_from_file(model_path):
|
133 |
-
print("loading model from {}".format(model_path))
|
134 |
-
if model_path.lower().endswith(".safetensors"):
|
135 |
-
pl_sd = safetensors.torch.load_file(model_path, device="cpu")
|
136 |
-
else:
|
137 |
-
pl_sd = torch.load(model_path, map_location="cpu")
|
138 |
-
state_dict = pl_sd["state_dict"] if "state_dict" in pl_sd else pl_sd
|
139 |
-
print("loading model from {} completed".format(model_path))
|
140 |
-
return state_dict
|
141 |
-
|
142 |
-
def get_alpha_comprod(linear_start=0.00085, linear_end=0.0120, timesteps=1000):
|
143 |
-
betas = torch.linspace(linear_start ** 0.5, linear_end ** 0.5, timesteps, dtype=torch.float32) ** 2
|
144 |
-
alphas = 1. - betas
|
145 |
-
alphas_cumprod = np.cumprod(alphas.numpy(), axis=0)
|
146 |
-
return torch.tensor(alphas_cumprod)
|
147 |
-
|
148 |
-
unused_tensors = [
|
149 |
-
"betas",
|
150 |
-
"alphas_cumprod_prev",
|
151 |
-
"sqrt_alphas_cumprod",
|
152 |
-
"sqrt_one_minus_alphas_cumprod",
|
153 |
-
"log_one_minus_alphas_cumprod",
|
154 |
-
"sqrt_recip_alphas_cumprod",
|
155 |
-
"sqrt_recipm1_alphas_cumprod",
|
156 |
-
"posterior_variance",
|
157 |
-
"posterior_log_variance_clipped",
|
158 |
-
"posterior_mean_coef1",
|
159 |
-
"posterior_mean_coef2",
|
160 |
-
"cond_stage_model.transformer.text_model.embeddings.position_ids",
|
161 |
-
"cond_stage_model.model.logit_scale",
|
162 |
-
"cond_stage_model.model.text_projection",
|
163 |
-
"model_ema.decay",
|
164 |
-
"model_ema.num_updates",
|
165 |
-
"control_model",
|
166 |
-
"lora_te_text_model",
|
167 |
-
"embedding_manager"
|
168 |
-
]
|
169 |
-
|
170 |
-
|
171 |
-
def preprocess(state_dict):
|
172 |
-
alphas_cumprod = state_dict.get("alphas_cumprod")
|
173 |
-
if alphas_cumprod != None:
|
174 |
-
# print((np.abs(get_alpha_comprod().numpy() - alphas_cumprod.numpy()) < 0.000001).all())
|
175 |
-
pass
|
176 |
-
else:
|
177 |
-
print("no alphas_cumprod in file, generate new one")
|
178 |
-
alphas_cumprod = get_alpha_comprod()
|
179 |
-
state_dict["alphas_cumprod"] = alphas_cumprod
|
180 |
-
|
181 |
-
new_state_dict = {}
|
182 |
-
for name, w in state_dict.items():
|
183 |
-
# ignore unused tensors
|
184 |
-
if not isinstance(w, torch.Tensor):
|
185 |
-
continue
|
186 |
-
skip = False
|
187 |
-
for unused_tensor in unused_tensors:
|
188 |
-
if name.startswith(unused_tensor):
|
189 |
-
skip = True
|
190 |
-
break
|
191 |
-
if skip:
|
192 |
-
continue
|
193 |
-
|
194 |
-
# # convert BF16 to FP16
|
195 |
-
if w.dtype == torch.bfloat16:
|
196 |
-
w = w.to(torch.float16)
|
197 |
-
|
198 |
-
# convert open_clip to hf CLIPTextModel (for SD2.x)
|
199 |
-
open_clip_to_hf_clip_model = {
|
200 |
-
"cond_stage_model.model.ln_final.bias": "cond_stage_model.transformer.text_model.final_layer_norm.bias",
|
201 |
-
"cond_stage_model.model.ln_final.weight": "cond_stage_model.transformer.text_model.final_layer_norm.weight",
|
202 |
-
"cond_stage_model.model.positional_embedding": "cond_stage_model.transformer.text_model.embeddings.position_embedding.weight",
|
203 |
-
"cond_stage_model.model.token_embedding.weight": "cond_stage_model.transformer.text_model.embeddings.token_embedding.weight",
|
204 |
-
"first_stage_model.decoder.mid.attn_1.to_k.bias": "first_stage_model.decoder.mid.attn_1.k.bias",
|
205 |
-
"first_stage_model.decoder.mid.attn_1.to_k.weight": "first_stage_model.decoder.mid.attn_1.k.weight",
|
206 |
-
"first_stage_model.decoder.mid.attn_1.to_out.0.bias": "first_stage_model.decoder.mid.attn_1.proj_out.bias",
|
207 |
-
"first_stage_model.decoder.mid.attn_1.to_out.0.weight": "first_stage_model.decoder.mid.attn_1.proj_out.weight",
|
208 |
-
"first_stage_model.decoder.mid.attn_1.to_q.bias": "first_stage_model.decoder.mid.attn_1.q.bias",
|
209 |
-
"first_stage_model.decoder.mid.attn_1.to_q.weight": "first_stage_model.decoder.mid.attn_1.q.weight",
|
210 |
-
"first_stage_model.decoder.mid.attn_1.to_v.bias": "first_stage_model.decoder.mid.attn_1.v.bias",
|
211 |
-
"first_stage_model.decoder.mid.attn_1.to_v.weight": "first_stage_model.decoder.mid.attn_1.v.weight",
|
212 |
-
}
|
213 |
-
open_clip_to_hk_clip_resblock = {
|
214 |
-
"attn.out_proj.bias": "self_attn.out_proj.bias",
|
215 |
-
"attn.out_proj.weight": "self_attn.out_proj.weight",
|
216 |
-
"ln_1.bias": "layer_norm1.bias",
|
217 |
-
"ln_1.weight": "layer_norm1.weight",
|
218 |
-
"ln_2.bias": "layer_norm2.bias",
|
219 |
-
"ln_2.weight": "layer_norm2.weight",
|
220 |
-
"mlp.c_fc.bias": "mlp.fc1.bias",
|
221 |
-
"mlp.c_fc.weight": "mlp.fc1.weight",
|
222 |
-
"mlp.c_proj.bias": "mlp.fc2.bias",
|
223 |
-
"mlp.c_proj.weight": "mlp.fc2.weight",
|
224 |
-
}
|
225 |
-
open_clip_resblock_prefix = "cond_stage_model.model.transformer.resblocks."
|
226 |
-
hf_clip_resblock_prefix = "cond_stage_model.transformer.text_model.encoder.layers."
|
227 |
-
if name in open_clip_to_hf_clip_model:
|
228 |
-
new_name = open_clip_to_hf_clip_model[name]
|
229 |
-
print(f"preprocess {name} => {new_name}")
|
230 |
-
name = new_name
|
231 |
-
if name.startswith(open_clip_resblock_prefix):
|
232 |
-
remain = name[len(open_clip_resblock_prefix):]
|
233 |
-
idx = remain.split(".")[0]
|
234 |
-
suffix = remain[len(idx)+1:]
|
235 |
-
if suffix == "attn.in_proj_weight":
|
236 |
-
w_q, w_k, w_v = w.chunk(3)
|
237 |
-
for new_suffix, new_w in zip(["self_attn.q_proj.weight", "self_attn.k_proj.weight", "self_attn.v_proj.weight"], [w_q, w_k, w_v]):
|
238 |
-
new_name = hf_clip_resblock_prefix + idx + "." + new_suffix
|
239 |
-
new_state_dict[new_name] = new_w
|
240 |
-
print(f"preprocess {name}{w.size()} => {new_name}{new_w.size()}")
|
241 |
-
elif suffix == "attn.in_proj_bias":
|
242 |
-
w_q, w_k, w_v = w.chunk(3)
|
243 |
-
for new_suffix, new_w in zip(["self_attn.q_proj.bias", "self_attn.k_proj.bias", "self_attn.v_proj.bias"], [w_q, w_k, w_v]):
|
244 |
-
new_name = hf_clip_resblock_prefix + idx + "." + new_suffix
|
245 |
-
new_state_dict[new_name] = new_w
|
246 |
-
print(f"preprocess {name}{w.size()} => {new_name}{new_w.size()}")
|
247 |
-
else:
|
248 |
-
new_suffix = open_clip_to_hk_clip_resblock[suffix]
|
249 |
-
new_name = hf_clip_resblock_prefix + idx + "." + new_suffix
|
250 |
-
new_state_dict[new_name] = w
|
251 |
-
print(f"preprocess {name} => {new_name}")
|
252 |
-
continue
|
253 |
-
|
254 |
-
# convert unet transformer linear to conv2d 1x1
|
255 |
-
if name.startswith("model.diffusion_model.") and (name.endswith("proj_in.weight") or name.endswith("proj_out.weight")):
|
256 |
-
if len(w.shape) == 2:
|
257 |
-
new_w = w.unsqueeze(2).unsqueeze(3)
|
258 |
-
new_state_dict[name] = new_w
|
259 |
-
print(f"preprocess {name} {w.size()} => {name} {new_w.size()}")
|
260 |
-
continue
|
261 |
-
|
262 |
-
# convert vae attn block linear to conv2d 1x1
|
263 |
-
if name.startswith("first_stage_model.") and "attn_1" in name:
|
264 |
-
if len(w.shape) == 2:
|
265 |
-
new_w = w.unsqueeze(2).unsqueeze(3)
|
266 |
-
new_state_dict[name] = new_w
|
267 |
-
print(f"preprocess {name} {w.size()} => {name} {new_w.size()}")
|
268 |
-
continue
|
269 |
-
|
270 |
-
new_state_dict[name] = w
|
271 |
-
return new_state_dict
|
272 |
-
|
273 |
-
def convert(model_path, out_type = None, out_file=None):
|
274 |
-
# load model
|
275 |
-
with open(os.path.join(vocab_dir, "vocab.json"), encoding="utf-8") as f:
|
276 |
-
clip_vocab = json.load(f)
|
277 |
-
|
278 |
-
state_dict = load_model_from_file(model_path)
|
279 |
-
model_type = SD1
|
280 |
-
if "cond_stage_model.model.token_embedding.weight" in state_dict.keys():
|
281 |
-
model_type = SD2
|
282 |
-
print("Stable diffuison 2.x")
|
283 |
-
else:
|
284 |
-
print("Stable diffuison 1.x")
|
285 |
-
state_dict = preprocess(state_dict)
|
286 |
-
|
287 |
-
# output option
|
288 |
-
if out_type == None:
|
289 |
-
weight = state_dict["model.diffusion_model.input_blocks.0.0.weight"].numpy()
|
290 |
-
if weight.dtype == np.float32:
|
291 |
-
out_type = "f32"
|
292 |
-
elif weight.dtype == np.float16:
|
293 |
-
out_type = "f16"
|
294 |
-
elif weight.dtype == np.float64:
|
295 |
-
out_type = "f32"
|
296 |
-
else:
|
297 |
-
raise Exception("unsupported weight type %s" % weight.dtype)
|
298 |
-
if out_file == None:
|
299 |
-
out_file = os.path.splitext(os.path.basename(model_path))[0] + f"-ggml-model-{out_type}.bin"
|
300 |
-
out_file = os.path.join(os.getcwd(), out_file)
|
301 |
-
print(f"Saving GGML compatible file to {out_file}")
|
302 |
-
|
303 |
-
# convert and save
|
304 |
-
with open(out_file, "wb") as file:
|
305 |
-
# magic: ggml in hex
|
306 |
-
file.write(struct.pack("i", 0x67676D6C))
|
307 |
-
# model & file type
|
308 |
-
ftype = (model_type << 16) | ggml_ftype_str_to_int[out_type]
|
309 |
-
file.write(struct.pack("i", ftype))
|
310 |
-
|
311 |
-
# vocab
|
312 |
-
byte_encoder = bytes_to_unicode()
|
313 |
-
byte_decoder = {v: k for k, v in byte_encoder.items()}
|
314 |
-
file.write(struct.pack("i", len(clip_vocab)))
|
315 |
-
for key in clip_vocab:
|
316 |
-
text = bytearray([byte_decoder[c] for c in key])
|
317 |
-
file.write(struct.pack("i", len(text)))
|
318 |
-
file.write(text)
|
319 |
-
|
320 |
-
# weights
|
321 |
-
for name in state_dict.keys():
|
322 |
-
if not isinstance(state_dict[name], torch.Tensor):
|
323 |
-
continue
|
324 |
-
skip = False
|
325 |
-
for unused_tensor in unused_tensors:
|
326 |
-
if name.startswith(unused_tensor):
|
327 |
-
skip = True
|
328 |
-
break
|
329 |
-
if skip:
|
330 |
-
continue
|
331 |
-
if name in unused_tensors:
|
332 |
-
continue
|
333 |
-
data = state_dict[name].numpy()
|
334 |
-
|
335 |
-
n_dims = len(data.shape)
|
336 |
-
shape = data.shape
|
337 |
-
old_type = data.dtype
|
338 |
-
|
339 |
-
ttype = "f32"
|
340 |
-
if n_dims == 4:
|
341 |
-
data = data.astype(np.float16)
|
342 |
-
ttype = "f16"
|
343 |
-
elif n_dims == 2 and name[-7:] == ".weight":
|
344 |
-
if out_type == "f32":
|
345 |
-
data = data.astype(np.float32)
|
346 |
-
elif out_type == "f16":
|
347 |
-
data = data.astype(np.float16)
|
348 |
-
elif out_type == "q4_0":
|
349 |
-
data = quantize_q4_0(data)
|
350 |
-
elif out_type == "q4_1":
|
351 |
-
data = quantize_q4_1(data)
|
352 |
-
elif out_type == "q5_0":
|
353 |
-
data = quantize_q5_0(data)
|
354 |
-
elif out_type == "q5_1":
|
355 |
-
data = quantize_q5_1(data)
|
356 |
-
elif out_type == "q8_0":
|
357 |
-
data = quantize_q8_0(data)
|
358 |
-
else:
|
359 |
-
raise Exception("invalid out_type {}".format(out_type))
|
360 |
-
ttype = out_type
|
361 |
-
else:
|
362 |
-
data = data.astype(np.float32)
|
363 |
-
ttype = "f32"
|
364 |
-
|
365 |
-
print("Processing tensor: {} with shape {}, {} -> {}".format(name, data.shape, old_type, ttype))
|
366 |
-
|
367 |
-
# header
|
368 |
-
name_bytes = name.encode("utf-8")
|
369 |
-
file.write(struct.pack("iii", n_dims, len(name_bytes), ggml_ttype_str_to_int[ttype]))
|
370 |
-
for i in range(n_dims):
|
371 |
-
file.write(struct.pack("i", shape[n_dims - 1 - i]))
|
372 |
-
file.write(name_bytes)
|
373 |
-
# data
|
374 |
-
data.tofile(file)
|
375 |
-
print("Convert done")
|
376 |
-
print(f"Saved GGML compatible file to {out_file}")
|
377 |
-
|
378 |
-
if __name__ == "__main__":
|
379 |
-
import argparse
|
380 |
-
parser = argparse.ArgumentParser(description="Convert Stable Diffuison model to GGML compatible file format")
|
381 |
-
parser.add_argument("--out_type", choices=["f32", "f16", "q4_0", "q4_1", "q5_0", "q5_1", "q8_0"], help="output format (default: based on input)")
|
382 |
-
parser.add_argument("--out_file", help="path to write to; default: based on input and current working directory")
|
383 |
-
parser.add_argument("model_path", help="model file path (*.pth, *.pt, *.ckpt, *.safetensors)")
|
384 |
-
args = parser.parse_args()
|
385 |
-
convert(args.model_path, args.out_type, args.out_file)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/models/requirements.txt
DELETED
@@ -1,4 +0,0 @@
|
|
1 |
-
numpy
|
2 |
-
torch
|
3 |
-
safetensors
|
4 |
-
pytorch_lightning
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/models/vocab.json
DELETED
The diff for this file is too large to render.
See raw diff
|
|
stable-diffusion.cpp/rng.h
DELETED
@@ -1,35 +0,0 @@
|
|
1 |
-
#ifndef __RNG_H__
|
2 |
-
#define __RNG_H__
|
3 |
-
|
4 |
-
#include <random>
|
5 |
-
#include <vector>
|
6 |
-
|
7 |
-
class RNG {
|
8 |
-
public:
|
9 |
-
virtual void manual_seed(uint64_t seed) = 0;
|
10 |
-
virtual std::vector<float> randn(uint32_t n) = 0;
|
11 |
-
};
|
12 |
-
|
13 |
-
class STDDefaultRNG : public RNG {
|
14 |
-
private:
|
15 |
-
std::default_random_engine generator;
|
16 |
-
|
17 |
-
public:
|
18 |
-
void manual_seed(uint64_t seed) {
|
19 |
-
generator.seed(seed);
|
20 |
-
}
|
21 |
-
|
22 |
-
std::vector<float> randn(uint32_t n) {
|
23 |
-
std::vector<float> result;
|
24 |
-
float mean = 0.0;
|
25 |
-
float stddev = 1.0;
|
26 |
-
std::normal_distribution<float> distribution(mean, stddev);
|
27 |
-
for (int i = 0; i < n; i++) {
|
28 |
-
float random_number = distribution(generator);
|
29 |
-
result.push_back(random_number);
|
30 |
-
}
|
31 |
-
return result;
|
32 |
-
}
|
33 |
-
};
|
34 |
-
|
35 |
-
#endif // __RNG_H__
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/rng_philox.h
DELETED
@@ -1,125 +0,0 @@
|
|
1 |
-
#ifndef __RNG_PHILOX_H__
|
2 |
-
#define __RNG_PHILOX_H__
|
3 |
-
|
4 |
-
#include <cmath>
|
5 |
-
#include <vector>
|
6 |
-
|
7 |
-
#include "rng.h"
|
8 |
-
|
9 |
-
// RNG imitiating torch cuda randn on CPU.
|
10 |
-
// Port from: https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/5ef669de080814067961f28357256e8fe27544f4/modules/rng_philox.py
|
11 |
-
class PhiloxRNG : public RNG {
|
12 |
-
private:
|
13 |
-
uint64_t seed;
|
14 |
-
uint32_t offset;
|
15 |
-
|
16 |
-
private:
|
17 |
-
std::vector<uint32_t> philox_m = {0xD2511F53, 0xCD9E8D57};
|
18 |
-
std::vector<uint32_t> philox_w = {0x9E3779B9, 0xBB67AE85};
|
19 |
-
float two_pow32_inv = 2.3283064e-10;
|
20 |
-
float two_pow32_inv_2pi = 2.3283064e-10 * 6.2831855;
|
21 |
-
|
22 |
-
std::vector<uint32_t> uint32(uint64_t x) {
|
23 |
-
std::vector<uint32_t> result(2);
|
24 |
-
result[0] = static_cast<uint32_t>(x & 0xFFFFFFFF);
|
25 |
-
result[1] = static_cast<uint32_t>(x >> 32);
|
26 |
-
return result;
|
27 |
-
}
|
28 |
-
|
29 |
-
std::vector<std::vector<uint32_t>> uint32(const std::vector<uint64_t>& x) {
|
30 |
-
int N = x.size();
|
31 |
-
std::vector<std::vector<uint32_t>> result(2, std::vector<uint32_t>(N));
|
32 |
-
|
33 |
-
for (int i = 0; i < N; ++i) {
|
34 |
-
result[0][i] = static_cast<uint32_t>(x[i] & 0xFFFFFFFF);
|
35 |
-
result[1][i] = static_cast<uint32_t>(x[i] >> 32);
|
36 |
-
}
|
37 |
-
|
38 |
-
return result;
|
39 |
-
}
|
40 |
-
|
41 |
-
// A single round of the Philox 4x32 random number generator.
|
42 |
-
void philox4_round(std::vector<std::vector<uint32_t>>& counter,
|
43 |
-
const std::vector<std::vector<uint32_t>>& key) {
|
44 |
-
uint32_t N = counter[0].size();
|
45 |
-
for (uint32_t i = 0; i < N; i++) {
|
46 |
-
std::vector<uint32_t> v1 = uint32(static_cast<uint64_t>(counter[0][i]) * static_cast<uint64_t>(philox_m[0]));
|
47 |
-
std::vector<uint32_t> v2 = uint32(static_cast<uint64_t>(counter[2][i]) * static_cast<uint64_t>(philox_m[1]));
|
48 |
-
|
49 |
-
counter[0][i] = v2[1] ^ counter[1][i] ^ key[0][i];
|
50 |
-
counter[1][i] = v2[0];
|
51 |
-
counter[2][i] = v1[1] ^ counter[3][i] ^ key[1][i];
|
52 |
-
counter[3][i] = v1[0];
|
53 |
-
}
|
54 |
-
}
|
55 |
-
|
56 |
-
// Generates 32-bit random numbers using the Philox 4x32 random number generator.
|
57 |
-
// Parameters:
|
58 |
-
// counter : A 4xN array of 32-bit integers representing the counter values (offset into generation).
|
59 |
-
// key : A 2xN array of 32-bit integers representing the key values (seed).
|
60 |
-
// rounds : The number of rounds to perform.
|
61 |
-
// Returns:
|
62 |
-
// std::vector<std::vector<uint32_t>>: A 4xN array of 32-bit integers containing the generated random numbers.
|
63 |
-
std::vector<std::vector<uint32_t>> philox4_32(std::vector<std::vector<uint32_t>>& counter,
|
64 |
-
std::vector<std::vector<uint32_t>>& key,
|
65 |
-
int rounds = 10) {
|
66 |
-
uint32_t N = counter[0].size();
|
67 |
-
for (int i = 0; i < rounds - 1; ++i) {
|
68 |
-
philox4_round(counter, key);
|
69 |
-
|
70 |
-
for (uint32_t j = 0; j < N; ++j) {
|
71 |
-
key[0][j] += philox_w[0];
|
72 |
-
key[1][j] += philox_w[1];
|
73 |
-
}
|
74 |
-
}
|
75 |
-
|
76 |
-
philox4_round(counter, key);
|
77 |
-
return counter;
|
78 |
-
}
|
79 |
-
|
80 |
-
float box_muller(float x, float y) {
|
81 |
-
float u = x * two_pow32_inv + two_pow32_inv / 2;
|
82 |
-
float v = y * two_pow32_inv_2pi + two_pow32_inv_2pi / 2;
|
83 |
-
|
84 |
-
float s = sqrt(-2.0 * log(u));
|
85 |
-
|
86 |
-
float r1 = s * sin(v);
|
87 |
-
return r1;
|
88 |
-
}
|
89 |
-
|
90 |
-
public:
|
91 |
-
PhiloxRNG(uint64_t seed = 0) {
|
92 |
-
this->seed = seed;
|
93 |
-
this->offset = 0;
|
94 |
-
}
|
95 |
-
|
96 |
-
void manual_seed(uint64_t seed) {
|
97 |
-
this->seed = seed;
|
98 |
-
this->offset = 0;
|
99 |
-
}
|
100 |
-
|
101 |
-
std::vector<float> randn(uint32_t n) {
|
102 |
-
std::vector<std::vector<uint32_t>> counter(4, std::vector<uint32_t>(n, 0));
|
103 |
-
for (uint32_t i = 0; i < n; i++) {
|
104 |
-
counter[0][i] = this->offset;
|
105 |
-
}
|
106 |
-
|
107 |
-
for (uint32_t i = 0; i < n; i++) {
|
108 |
-
counter[2][i] = i;
|
109 |
-
}
|
110 |
-
this->offset += 1;
|
111 |
-
|
112 |
-
std::vector<uint64_t> key(n, this->seed);
|
113 |
-
std::vector<std::vector<uint32_t>> key_uint32 = uint32(key);
|
114 |
-
|
115 |
-
std::vector<std::vector<uint32_t>> g = philox4_32(counter, key_uint32);
|
116 |
-
|
117 |
-
std::vector<float> result;
|
118 |
-
for (int i = 0; i < n; ++i) {
|
119 |
-
result.push_back(box_muller(g[0][i], g[1][i]));
|
120 |
-
}
|
121 |
-
return result;
|
122 |
-
}
|
123 |
-
};
|
124 |
-
|
125 |
-
#endif // __RNG_PHILOX_H__
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
stable-diffusion.cpp/stable-diffusion.cpp
DELETED
The diff for this file is too large to render.
See raw diff
|
|
stable-diffusion.cpp/stable-diffusion.h
DELETED
@@ -1,74 +0,0 @@
|
|
1 |
-
#ifndef __STABLE_DIFFUSION_H__
|
2 |
-
#define __STABLE_DIFFUSION_H__
|
3 |
-
|
4 |
-
#include <memory>
|
5 |
-
#include <vector>
|
6 |
-
|
7 |
-
enum SDLogLevel {
|
8 |
-
DEBUG,
|
9 |
-
INFO,
|
10 |
-
WARN,
|
11 |
-
ERROR
|
12 |
-
};
|
13 |
-
|
14 |
-
enum RNGType {
|
15 |
-
STD_DEFAULT_RNG,
|
16 |
-
CUDA_RNG
|
17 |
-
};
|
18 |
-
|
19 |
-
enum SampleMethod {
|
20 |
-
EULER_A,
|
21 |
-
EULER,
|
22 |
-
HEUN,
|
23 |
-
DPM2,
|
24 |
-
DPMPP2S_A,
|
25 |
-
DPMPP2M,
|
26 |
-
DPMPP2Mv2,
|
27 |
-
N_SAMPLE_METHODS
|
28 |
-
};
|
29 |
-
|
30 |
-
enum Schedule {
|
31 |
-
DEFAULT,
|
32 |
-
DISCRETE,
|
33 |
-
KARRAS,
|
34 |
-
N_SCHEDULES
|
35 |
-
};
|
36 |
-
|
37 |
-
class StableDiffusionGGML;
|
38 |
-
|
39 |
-
class StableDiffusion {
|
40 |
-
private:
|
41 |
-
std::shared_ptr<StableDiffusionGGML> sd;
|
42 |
-
|
43 |
-
public:
|
44 |
-
StableDiffusion(int n_threads = -1,
|
45 |
-
bool vae_decode_only = false,
|
46 |
-
bool free_params_immediately = false,
|
47 |
-
RNGType rng_type = STD_DEFAULT_RNG);
|
48 |
-
bool load_from_file(const std::string& file_path, Schedule d = DEFAULT);
|
49 |
-
std::vector<uint8_t> txt2img(
|
50 |
-
const std::string& prompt,
|
51 |
-
const std::string& negative_prompt,
|
52 |
-
float cfg_scale,
|
53 |
-
int width,
|
54 |
-
int height,
|
55 |
-
SampleMethod sample_method,
|
56 |
-
int sample_steps,
|
57 |
-
int64_t seed);
|
58 |
-
std::vector<uint8_t> img2img(
|
59 |
-
const std::vector<uint8_t>& init_img,
|
60 |
-
const std::string& prompt,
|
61 |
-
const std::string& negative_prompt,
|
62 |
-
float cfg_scale,
|
63 |
-
int width,
|
64 |
-
int height,
|
65 |
-
SampleMethod sample_method,
|
66 |
-
int sample_steps,
|
67 |
-
float strength,
|
68 |
-
int64_t seed);
|
69 |
-
};
|
70 |
-
|
71 |
-
void set_sd_log_level(SDLogLevel level);
|
72 |
-
std::string sd_get_system_info();
|
73 |
-
|
74 |
-
#endif // __STABLE_DIFFUSION_H__
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|