hash stringlengths 40 40 | msg stringlengths 1 131k | author stringlengths 1 33 | email stringlengths 0 57 | date int64 1,447B 1,698B |
|---|---|---|---|---|
04f34b50ccf906a29f7261d8f94182bb8fcde624 | Fork Model Analyzer.
PiperOrigin-RevId: 564500466 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,467,297,000 |
af73df530c87a257db908ae6efc6bbf9482a504e | [SPMD] Add sharding propagation logics/tests for shard_like/as. (3/N)
1. Adding bidirectional propagation paths among nodes within the same shard_group
2. Enforce the shardings from the same shard_as group to take the same sharding in the end.
PiperOrigin-RevId: 564504486 | Tongfei Guo | tongfei@google.com | 1,694,468,154,000 |
da7114251a91579b1c8389457a4819d3b24cff61 | [xla] Avoid cloning unused instructions.
Rearrange the code that unroll the last while-loop iteration to avoid cloning
the instruction being pipelined, which is unused in the iteration. This is to
prepare for the next change to support backward pipelining recv-done without
generating the unused recv and recv-done pair... | Bixia Zheng | bixia@google.com | 1,694,470,580,000 |
4551e3efdf9b26648c9b545ba20b74a4606248b6 | #tf-data-service Make sure dynamic sharding works without job_names.
PiperOrigin-RevId: 564516421 | Yang Chen | yangchen@google.com | 1,694,470,720,000 |
68cc4388bc7ea92a0eb24abd28377323da254f79 | #tf.data Log `tf.data.Options` used to ease debugging.
PiperOrigin-RevId: 564518057 | Wilsin Gosti | wilsin@google.com | 1,694,471,064,000 |
f4901034536be6c63ca330cb51091055196e6fdc | Add back TensorflowInterface to libtpu
PiperOrigin-RevId: 564526383 | Haibo Huang | hhb@google.com | 1,694,472,984,000 |
ced9f3b12f5e04e1e8b54df7d57c50faaf3a43fc | Remove c/kernels/ops/* from libtpu
PiperOrigin-RevId: 564526562 | Haibo Huang | hhb@google.com | 1,694,473,025,000 |
12e65cf0e6f2ca2de61a24758def5a753b521bd0 | Add int4 registrations for sharding.
Needed for sharding int4 models.
PiperOrigin-RevId: 564528408 | Antonio Sanchez | cantonios@google.com | 1,694,473,417,000 |
ff5dce3d739e2791b2de0a3b3f49b10bf13cd78c | Merge pull request #61798 from bhbruce:fix-cmake-error
PiperOrigin-RevId: 564528951 | TensorFlower Gardener | gardener@tensorflow.org | 1,694,474,495,000 |
73ee76cc3fc9460ffd705cbbf449b6c26dd72d72 | Merge pull request #61491 from benbarsdell:upstream-SparseSegmentReduceGradV2-part3
PiperOrigin-RevId: 564529533 | TensorFlower Gardener | gardener@tensorflow.org | 1,694,474,872,000 |
e06c3a6f25debbe637fd692be6228815d54f86c1 | [tflite] Add Eager API for activation functions.
Leaky Relu, Elu, Relu, Relu1, Relu0To1, Relu6, Tanh, Sigmoid, and HardSwish
PiperOrigin-RevId: 564537536 | Robert David | lrdx@google.com | 1,694,475,682,000 |
d70aeead5c66030f7a9bcfb6e13735d38898c720 | [xla:gpu2] Remove RuntimeBackend option from gpu2 compiler pipeline
We are not going to pursue direct lowering from LMHLO to API calls, though we might consider "reference implementation API" in the future, but it will go through intermediate "XLA runtime dialect". For now remove dead code.
PiperOrigin-RevId: 5645380... | Eugene Zhulenev | ezhulenev@google.com | 1,694,475,807,000 |
45a8301d717a5b358d44e45d2b9022ff9be10175 | Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId: 564546472 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,477,921,000 |
cf10eb3a7e5a9398e19ea36509dc73519bac0970 | Update ops-related pbtxt files.
PiperOrigin-RevId: 564546615 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,477,956,000 |
a84630670be7a6b4175ce100b6fbdd4183e0646a | [tflite-gpu] Make GPU Async zero-copy.
The user will provide inputs via AHWB. Behind the scenes we bind the AHWB's to SSBO. Both cl & gl inference are supported.
PiperOrigin-RevId: 564547199 | Grant Jensen | grantjensen@google.com | 1,694,478,093,000 |
8d47eef8e4a9bb8f0f8f9c8e09359419b6ae2ebe | [xla:gpu] Initialize cuBLAS support before running executable and before capturing gpu graphs
Add the custom call xla.gpu.init_cuBLAS to the beginning of the entry function of the module, when the module contains a gemm.
PiperOrigin-RevId: 564550744 | Anlun Xu | anlunx@google.com | 1,694,479,019,000 |
b1bf09989871721a23f936617a97bb60b3d27a58 | Merge pull request #61768 from boomanaiden154:missing-header-files
PiperOrigin-RevId: 564551916 | TensorFlower Gardener | gardener@tensorflow.org | 1,694,479,808,000 |
8142bcdeceaa1275ff5815d36745b7cdc5243814 | [tosa] Support zero scale in tfl-to-tosa
Sometimes scales are zero in the all zero case. This is obviously
silly but it triggers a failure in the pipeline. Make scale the
minimum scale value as whatevs.
PiperOrigin-RevId: 564556649 | Robert Suderman | suderman@google.com | 1,694,480,671,000 |
6e3904f92757adf4f6dd2d67162fad56468d9a6a | [tflite] Implement Eager API for quantized multiply
PiperOrigin-RevId: 564559805 | Grant Jensen | grantjensen@google.com | 1,694,481,667,000 |
ace5d01ebb5ae6781ed4614c8650202c53684935 | Remove all the scatter re-write patterns from mhlo -> TF.
These patterns are used only by TFLite which now has direct support for stablehlo.scatter.
PiperOrigin-RevId: 564569917 | Majid Dadashi | majiddadashi@google.com | 1,694,485,473,000 |
97a1fa35da876a8c5e328b448babc5627efd6e4a | Adding tf visibility to targets
PiperOrigin-RevId: 564577405 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,488,024,000 |
643e43b101c336611f6adc3da91dbff6ad11aa5f | Fix the checks that check if memory kinds between pjrt_buffer and user sharding is equal or not.
PiperOrigin-RevId: 564578779 | Yash Katariya | yashkatariya@google.com | 1,694,488,373,000 |
4e9db184297de547884b4b3c1d1422658ec37a6e | #tf-data Output of a model node can be deleted by a thread while some of its input nodes are in the process of being deleted. If such a case happens, the `output_` of these input nodes becomes a dangling pointer. This CL adds a check for such a scenario.
PiperOrigin-RevId: 564583894 | Wilsin Gosti | wilsin@google.com | 1,694,490,357,000 |
6324e50357a8288408620a5e9cc1b3022880ef41 | #tf-data The default value of `warm_start` should be `None` instead of `False`. Setting it to `False` as the default value may break the `tf.data.Options` propagation when users set it to `True` somewhere in the input pipeline.
PiperOrigin-RevId: 564585601 | Wilsin Gosti | wilsin@google.com | 1,694,491,035,000 |
2abff427ec6b6f341e92b4fb3b93e201b52f1b96 | PR #5498: Remove unnecessary padding for vector bias for non-fp8 cublasLt Matmul rewrite
Imported from GitHub PR https://github.com/openxla/xla/pull/5498
Remove unnecessary padding for vector bias for non-fp8 cublasLt Matmul rewrite.
Copybara import of the project:
--
fb25577449faf13ef424a827b4896a6f7e2c0f53 by wens... | wenscarl | shuw@nvidia.com | 1,694,491,145,000 |
c1479d758675ccc2de15f045332f14e696dea42a | consistent lowercase testname syntax in examples/adding_an_op tests | redwrasse | mail@redwrasse.io | 1,694,491,894,000 |
0af4b75f09ae8a9c663f72a6f28a3b4492ce5537 | Merge pull request #61748 from tensorflow:sachinprasadhs-patch-2
PiperOrigin-RevId: 564589017 | TensorFlower Gardener | gardener@tensorflow.org | 1,694,492,847,000 |
d66aeda6ccce56c994116cac49378eb049959a38 | Update release Linux configs in the root bazelrc
The following changes are included:
- Starting from TF2.13v, TensorFlow uses Clang as compiler for Linux. Thus, we update the toolchain in release_cpu_linux and release_gpu_linux configs.
- Preserve the old linux build options in the unsupported_cpu_linux and unsupporte... | Kanglan Tang | kanglan@google.com | 1,694,501,658,000 |
4fb43b3b4a827d47d77be01c35c10c38151f8c18 | [XLA:GPU] Remove notap tag from test target.
PiperOrigin-RevId: 564648984 | Benjamin Chetioui | bchetioui@google.com | 1,694,509,307,000 |
89890cb211bda000b41706a54d018513067a4f07 | compat: Update forward compatibility horizon to 2023-09-12
PiperOrigin-RevId: 564649507 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,509,429,000 |
8a3b5773a5b0ccf7c43440af8ada634a07f62d96 | Update GraphDef version to 1617.
PiperOrigin-RevId: 564649508 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,509,429,000 |
ce448e242b2576213ad8bae4f57a8fb84a9ba540 | Integrate LLVM at llvm/llvm-project@5bab59de4463
Updates LLVM usage to match
[5bab59de4463](https://github.com/llvm/llvm-project/commit/5bab59de4463)
PiperOrigin-RevId: 564649900 | Benjamin Kramer | kramerb@google.com | 1,694,509,521,000 |
b24b7acd3d4809ffb4c705b81a0aa84b57db07b6 | NFC: Triton MatMul: Unify duplicated code for RHS/LHS.
Also reduce the number of arguments to most lambdas.
PiperOrigin-RevId: 564654639 | Johannes Reifferscheid | jreiffers@google.com | 1,694,510,888,000 |
4aa2ee612ddc526191ff90b6ea71cc8a1b5f8a06 | Merge pull request #61258 from tensorflow:pjpratik-patch-7
PiperOrigin-RevId: 564685838 | TensorFlower Gardener | gardener@tensorflow.org | 1,694,520,966,000 |
2c8798d24c76423c848e20038da9e18e253b6c92 | NFC: Extract LaunchDimension computation from Triton codegen.
PiperOrigin-RevId: 564686943 | Johannes Reifferscheid | jreiffers@google.com | 1,694,521,004,000 |
9c775b068b64911e3302e3186e6b24cde8e9743d | Use /usertools/cpu.bazelrc for TSL presubmit build
Without this additional bazelrc file TSL will be built with the system compiler which happens to be GCC.
I also had to disable a warning that was raised by clang. It puzzles me a bit that this is not needed for the Tensorflow build which definitely uses Clang.
Piper... | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,521,369,000 |
315ad6f737a9f344b160783d86da4b68751146d6 | Fill refcounts with zeros so that arena memory may be properly shared.
PiperOrigin-RevId: 564693001 | Alan Kelly | alankelly@google.com | 1,694,522,996,000 |
280122a477f3ad268501adaff8ac0a87ac06b1ca | Add TFLite `Dilate` op to MLIR converter.
PiperOrigin-RevId: 564693541 | Quentin Khan | qkhan@google.com | 1,694,523,154,000 |
a34e961be3ef6b769b0a37dac26832f0ed1992d2 | Remove references to llvm::any_isa
It's going away in https://github.com/llvm/llvm-project/commit/5378117327317ac08eb25e52519f482af7571445
PiperOrigin-RevId: 564711879 | Benjamin Kramer | kramerb@google.com | 1,694,527,939,000 |
8ea61a0e0b54df83d75502f576446c58d6c3c812 | [XLA:Python] Remove the use_tfrt flag from make_cpu_client().
use_tfrt=True has been the default for over a year, and the flag currently does nothing.
PiperOrigin-RevId: 564712316 | Peter Hawkins | phawkins@google.com | 1,694,528,066,000 |
b92b789a8ef69e6385d0814b188684270776d5b2 | [StepSecurity] ci: Harden GitHub Actions
Signed-off-by: Joyce Brum <joycebrum@google.com> | Joyce Brum | joycebrum@google.com | 1,694,530,169,000 |
7702906580701bd322899f8b47ef8114fc8b20e1 | A header file name is case corrected to be buildable in Linux environment.
PiperOrigin-RevId: 564719695 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,530,177,000 |
ac641ea578d790b1cd2ba90047afc17906b10b61 | Access tflite tensors to xnnpack map directly instead of using sparse xnnpack_tensors
PiperOrigin-RevId: 564721615 | Alan Kelly | alankelly@google.com | 1,694,530,675,000 |
d4e9abafd70a60323a1b3a7737d545db8aac7e5b | #tf-data Remove an unnecessary mutex lock when removing a model node.
PiperOrigin-RevId: 564739804 | Wilsin Gosti | wilsin@google.com | 1,694,534,955,000 |
56803aae482fe077b6fb94e2f4ba96d9ab885de7 | set permissions on workflows
Signed-off-by: Joyce Brum <joycebrum@google.com> | Joyce Brum | joycebrum@google.com | 1,694,536,829,000 |
9f729bcaa23bb337e4005ffd220bd2b27e9cfe33 | set update-nightly.yml permissions
Signed-off-by: Joyce <joycebrum@google.com> | Joyce | joycebrum@google.com | 1,694,537,156,000 |
e3d9d1e00f20e31300f6789c793cad36f76d6598 | Eliminate never called function and unused data members.
PiperOrigin-RevId: 564748979 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,536,936,000 |
fd8872f62f9ee606bb40048ac013774eb69c2875 | [NFC] Actually use the fusion-wrapper pass pipeline
PiperOrigin-RevId: 564749546 | Rahul Joshi | jurahul@google.com | 1,694,537,044,000 |
27beade16250ac0124fd58f50a291342eb50cc34 | PR #5512: Fix link errors in build tests
Imported from GitHub PR https://github.com/openxla/xla/pull/5512
Some build tests were failing due to unresolved symbols so add in their dependencies
This is a redo of https://github.com/tensorflow/tensorflow/pull/61489 which was not merged in time before the code was moved.
... | Andrew Goodbody | andrew.goodbody@linaro.org | 1,694,538,764,000 |
fdef6b6d27631493bb1749f8df11242a4772a8c9 | [XLA:GPU] Add missing dependency on :cuda_gpu_executor from gpu_cudamallocasync_allocator.
This was causing a missing symbol error for certain build targets.
PiperOrigin-RevId: 564766163 | Peter Hawkins | phawkins@google.com | 1,694,540,268,000 |
f8ac04c7200bef6ec1d44fafaf4e28cfb51da9a8 | Don't set singleprint metric if fingerprint isn't present.
PiperOrigin-RevId: 564767887 | Adam Cogdell | adamcogdell@google.com | 1,694,540,586,000 |
062cc8df53b23e4bc7e7e3035d3ce7e1ec361eb4 | Support `ones_like` under `vectorized_map`.
PiperOrigin-RevId: 564789520 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,544,675,000 |
591698c24249d51d0d6a74d4ee9d9e272a2f7314 | Add a transform for broadcast to broadcast_in_dim.
At the moment, StableHLO does not have the spec and the interpreter for `broadcast`. `broadcast` is a special case of `broadcast_in_dim`, and this pass makes `broadcast -> broadcast_in_dim` conversion.
PiperOrigin-RevId: 564796578 | Gunhyun Park | gunhyun@google.com | 1,694,546,157,000 |
b3517259668c52b1968a56a83ee389600417833b | Add int4 support for sliced save/restore.
PiperOrigin-RevId: 564799226 | Antonio Sanchez | cantonios@google.com | 1,694,546,815,000 |
cbecb89379b590df5bcf1fb1b5d5fdfa7bc89ab9 | [tflite] Temporarily add back mhlo.scatter to TF patterns
This broke some tests outside of tflite and darwinn. Temporarily adding back these patterns until the test issues are resolved.
PiperOrigin-RevId: 564800351 | Majid Dadashi | majiddadashi@google.com | 1,694,547,089,000 |
693b0ca69b666919148ab07764afbfdbc52b2333 | [tfcompile] Remove redundant rounding in benchmark timer
Remove `uint64` -> `double` -> `uint64` conversions from `NowMicros()` and their users.
PiperOrigin-RevId: 564802323 | Diego Caballero | diegocaballero@google.com | 1,694,547,549,000 |
3519fadda6b592ea3a9024c970e1d29fbf017c8d | Add an option to override to the size of GPU system
PiperOrigin-RevId: 564804399 | Guanxin Qiao | guanxinq@google.com | 1,694,548,006,000 |
1d7dcd4246d46d372bf68b3a4f83173c7c3d24cd | [XLA:GPU] Adds support for softmax fusions of bf16 data type.
PiperOrigin-RevId: 564807274 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,548,631,000 |
c2f381105de794cf8227b0b79bc93d776e41ba57 | Add CopyToMemorySpace to the PjRtBuffer API. This CL does not implement any instance of the method, but adds the ability to do so in followup CLs.
PiperOrigin-RevId: 564807735 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,548,729,000 |
4b8f8a33ac5568b089d0a656b621f999a710fef2 | Fix NCCL UB issue
This is fixing a UB issue which occurs with newer version of Clang (17+).
The fix is also upstreamed through https://github.com/NVIDIA/nccl/pull/916.
In addition I'm changing the handling of `enqueue.cc` which needs to be compiled
in cuda mode under clang. The previous solution with just passing in ... | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,549,423,000 |
6758034f06481549f37f2411e9aef1bce3dca4ea | #tf-data Turn down `"file_locality_v2"` experiment.
PiperOrigin-RevId: 564813085 | Matt Callanan | mpcallanan@google.com | 1,694,549,898,000 |
c332f6baea26c9994a404dc26a19b3c2735c782a | [XLA:GPU] Shard topk_kernel.cu.cc.
I've seen this file take over 5 minutes to build. Shard it by type.
PiperOrigin-RevId: 564820851 | Peter Hawkins | phawkins@google.com | 1,694,551,468,000 |
d9ae77cb3c1bc03c88d92913de6bf10b9ff9c7ed | Use /usertools/cpu.bazelrc for TSL presubmit build
Without this additional bazelrc file TSL will be built with the system compiler which happens to be GCC.
I also had to disable a warning that was raised by clang. It puzzles me a bit that this is not needed for the Tensorflow build which definitely uses Clang.
Piper... | Fiona Lang | flang@google.com | 1,694,551,782,000 |
66e19b9f72fb03671c31f57200579a1103766ffd | Enable Split op for int4.
PiperOrigin-RevId: 564825677 | Antonio Sanchez | cantonios@google.com | 1,694,552,324,000 |
0f406be2901a9a608e2de8df1669d4029c6681cb | Fix typo in error message
PiperOrigin-RevId: 564840225 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,555,409,000 |
dd5cf20e5785f4ff61d9841b4d0ab23a160b7ea8 | Integrate LLVM at llvm/llvm-project@c1796be93fe5
Updates LLVM usage to match
[c1796be93fe5](https://github.com/llvm/llvm-project/commit/c1796be93fe5)
PiperOrigin-RevId: 564842806 | Benjamin Kramer | kramerb@google.com | 1,694,555,948,000 |
c952951fbeaef9f297a2fdb5ef275a73e734b037 | _python_memory_checker_helper.pyi update to adjust to upstream pybind11 changes.
PiperOrigin-RevId: 564845580 | Ralf W. Grosse-Kunstleve | rwgk@google.com | 1,694,556,549,000 |
3b4eda4fed711ffd763847bdce9f6673b737324f | Remove check for input constant before running tf2xla op from MLIR
This check can be removed since tf2xla can run ops with non-const input even if CompileTimeConstant attribute is set with the help of valueinference.
PiperOrigin-RevId: 564851049 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,557,818,000 |
e61dffa5be78f317148d0172239a1e90dfd790b3 | [pjrt] Rewrite the transpose kernels, removing our dependence on Eigen
For an 8x8 uint32_t transpose, we had:
- 4x `vinsertf128 ymm, ymm, xmm`
- 4x `vperm2f128`.
These are very expensive instructions because they cross the 128-bit lane boundary.
Now, we have `8x vinsertf128` but crucially, the inserted operand now c... | David Majnemer | majnemer@google.com | 1,694,559,861,000 |
d6edddebcda46c48c91baf6f56c7d127236b4c1c | #tf-data Set up noop experiments.
PiperOrigin-RevId: 564864652 | Matt Callanan | mpcallanan@google.com | 1,694,560,750,000 |
e371aa906c1f7fe1694ae6a1d0b83caebba80831 | Describe reference counting behavior for EagerTensor/Handle conversion
PiperOrigin-RevId: 564871149 | Yu Feng | feyu@google.com | 1,694,562,209,000 |
282193e8582f3cf4f102205db156f0163a29a08a | Add error logging instrumentation in phase one TF2XLA bridge.
This is for internal error logging only.
PiperOrigin-RevId: 564878979 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,564,137,000 |
cffae6d8b0421fbc594bd3be15dbd49ddba9c152 | Change visibility for tsl/lib/{histogram/monitoring}.
PiperOrigin-RevId: 564881319 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,564,831,000 |
278f871658329ce838cb0aaf3bfc5f2b348b8de2 | Update version of rules_pkg
PiperOrigin-RevId: 564886962 | Jake Harmon | jakeharmon@google.com | 1,694,566,665,000 |
9256cec6f47be4acffb4c76b11969bfa7c1c8285 | Instrument metrics to track the pjrt compilation status.
Create metrics:
1) '/pjrt/compiler/is_compiling_computation' to record if pjrt compiler is compiling computations.
2) '/pjrt/compiler/is_compiling_module' to record if pjrt compiler is compiling modules.
PiperOrigin-RevId: 564891869 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,568,347,000 |
12fddfb626eab5b6693bbb1ff6e042d50627faad | Use buffer ids to provide an explicitly stable sort.
PiperOrigin-RevId: 564893903 | Russell Power | power@google.com | 1,694,569,077,000 |
3d1802023778a164d35c79536990b35b701e8018 | Fix forward for GCC
GCC is missing some intrinsics, expand them by hand.
PiperOrigin-RevId: 564899448 | David Majnemer | majnemer@google.com | 1,694,571,377,000 |
626e5f223da09ff32259c44ddafc1edee2458621 | Add a property for the number of embedding devices per chip.
PiperOrigin-RevId: 564909053 | Ziyin Huang | ziyinh@google.com | 1,694,574,840,000 |
93dee7392bb05dca346562075082b51f1a1daf1b | Merge pull request #61842 from redwrasse:redwrasse/test-name-syntax-misc
PiperOrigin-RevId: 564914384 | TensorFlower Gardener | gardener@tensorflow.org | 1,694,577,317,000 |
9377294a54fea06e26c9a0ea8b1b55c43be5e73c | Prepare for TensorRT 8.6
The upcoming CUDA-12 upgrades requires TensorRT 8.6 and this version
has a new set of headers which requires an update to the bazel configure script.
I also had to change `find_cuda_config.py` because previously it has only
been reporting the major version of TensorRT back to the configure sc... | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,581,778,000 |
e29f5612938fa336f92194397c8b905b3d5e59b0 | compat: Update forward compatibility horizon to 2023-09-13
PiperOrigin-RevId: 564978209 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,595,721,000 |
0824c23f155d25b7dba4bae9b17b3a0537d347e4 | Update GraphDef version to 1618.
PiperOrigin-RevId: 564978233 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,595,724,000 |
55bae7da65a8830c1f09dd47f3bcd8679e53d2f1 | Remove shared memory from LaunchDimensions.
LaunchDimensions are being decoupled from codegen. Shared memory requirements
are only known after codegen.
Currently, this feature is only used for Triton fusions. All other
fusions that use shared memory allocate it within the kernel.
PiperOrigin-RevId: 564981685 | Johannes Reifferscheid | jreiffers@google.com | 1,694,596,673,000 |
b710b1e09ff92d44f03d447cb73ed671f6763a5e | [XLA:GPU][NFC] Refactor Triton GEMM rewriter.
Add a comparison operator for fragments, improve encapsulation, fix comments.
PiperOrigin-RevId: 564990063 | Ilia Sergachev | sergachev@google.com | 1,694,599,101,000 |
af731a2dba19517a868e4208e30f2420d6b4023a | [XLA:GPU] Fix inconsistent check for reduction unroll factor
From the original code at commit 03d304b, we are only supposed to scale the reduction
tile size of dimX by the unroll factor for column reductions, so the check when creating
`ReductionCodegenInfo` is only valid for column reductions.
PiperOrigin-RevId: 5649... | Son Tuan Vu | vuson@google.com | 1,694,599,525,000 |
25ac48b7c871c81fb50ed551332a0e434e13ce23 | Update the documentation for the TF Lite delegate plug-in interface to clarify
that the corresponding FlatBuffer C API type can also be used instead of
the FlatBuffer C++ API type for the TFLiteSettings FlatBuffer parameter.
PiperOrigin-RevId: 565008668 | Fergus Henderson | fergus@google.com | 1,694,604,395,000 |
1803461c464ebbfaf04f30bed78f1b455171b6f0 | NFC: Extract launch dimensions computation from TritonWrapper.
This is another prefactoring to make Triton fusions compatible with
FusionInterface and partially fused HLO. For the former, we need
the LaunchDimension computation to be a separate function. For the
latter, we change the launch dimension function signatur... | Johannes Reifferscheid | jreiffers@google.com | 1,694,606,555,000 |
af5aa0e54e1e0578dcb54db652631df6e94b0095 | [XLA:GPU][NFC] Cosmetic changes to gemm_rewriter_triton.cc
Unify iterator names to *_it;
Use auto* instead of auto if the type is a pointer type.
PiperOrigin-RevId: 565016627 | Tamás Danyluk | tdanyluk@google.com | 1,694,606,826,000 |
4ae35f9f35f38fc12b65b077a7f25eab1137e391 | [NFC] [XLA:GPU] Split up Triton emitter into multiple functions.
PiperOrigin-RevId: 565024582 | George Karpenkov | cheshire@google.com | 1,694,609,472,000 |
410d4576f600b005ede6e6bcdf6b448c8d8e14f4 | [XLA:GPU] Add more comments to gpu_performance_model.
PiperOrigin-RevId: 565025664 | Oleg Shyshkov | shyshkov@google.com | 1,694,609,841,000 |
0ad14a0bf82dab7d18524f2eecb8120276db1e0a | use different agent label for release testing and ci/nightlys | cjflan | 89868659+cjflan@users.noreply.github.com | 1,694,614,941,000 |
44400f85dc5dae2c365e3319be589dc92ee03538 | Change Triton not to depend on :ConversionPasses.
This change was merged upstream in https://github.com/openai/triton/pull/2068 but hasn't made it to the OpenXLA fork yet.
Improves build time in OSS by not building a number of MLIR dialects/passes that are not needed.
PiperOrigin-RevId: 565058458 | Peter Hawkins | phawkins@google.com | 1,694,619,057,000 |
c3cc395eb4c110fa7218ee83ed2eadd083703074 | Use /usertools/cpu.bazelrc for TSL presubmit build
Without this additional bazelrc file TSL will be built with the system compiler which happens to be GCC.
PiperOrigin-RevId: 565060297 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,619,500,000 |
539673ead2b66a9c2dce3fb90e3767efda5deef5 | Switch to using new API generation. | Marc Fisher II | fisherii@google.com | 1,694,190,461,000 |
74d7ec3be99ff5bfbc97ecfb2ec9852ef9e99363 | Integrate LLVM at llvm/llvm-project@8ebe1d1cc1e4
Updates LLVM usage to match
[8ebe1d1cc1e4](https://github.com/llvm/llvm-project/commit/8ebe1d1cc1e4)
PiperOrigin-RevId: 565069541 | Benjamin Kramer | kramerb@google.com | 1,694,621,540,000 |
65bd69126ab25c6ba4b915f0137c0f709fb9e8de | [xla] Change the collective-permute-decomposer to not chain the Send and Recv
instructions through control dependence.
This is because the generated HLO program is correct even without the control
dependence chaining. The purpose of the control dependence chaining is to
support a scheduler, such as the latency hiding ... | Bixia Zheng | bixia@google.com | 1,694,622,526,000 |
f28e73538a5e803a130aba85229d6d3f3035a58d | [XLA] Add support to CollectivePipeliner to sink collectives.
Small collectives might be better off when sinked and there are other potnential use cases
Also fix a bug, where we were accepting reuse of the data that we were storing and changing the tests using that pattern to match the fix.
PiperOrigin-RevId: 5650807... | Marcello Maggioni | maggioni@google.com | 1,694,624,077,000 |
d11423a4e9701166a046ccbc6972bf4535543123 | [mhlo] Remove unused HloLegalizeToLhlo pass.
PiperOrigin-RevId: 565083641 | Oleg Shyshkov | shyshkov@google.com | 1,694,624,652,000 |
be0d1151a6a3352206a435c1c14afa2b1f4c5dc4 | [XLA] Rework dot() sharding propagation to lookahead instructions sharding to choose a sharding for dot() that agrees with the users if possible.
PiperOrigin-RevId: 565086052 | Marcello Maggioni | maggioni@google.com | 1,694,625,114,000 |
26aa3c84c689389b1adac088ac67168f3356cd21 | Excludes building pjrt_c_api_gpu_plugin.so (GPU only target) on MAC.
PiperOrigin-RevId: 565092976 | Jieying Luo | jieying@google.com | 1,694,626,407,000 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.