hash stringlengths 40 40 | msg stringlengths 1 131k | author stringlengths 1 33 | email stringlengths 0 57 | date int64 1,447B 1,698B |
|---|---|---|---|---|
0d07a2382e2c744aabce0ea42d9965875166935b | Check for variants in ZerosLike or AddN before legalize to custom tensorlist.
PiperOrigin-RevId: 570822448 | Luke Boyer | lukeboyer@google.com | 1,696,457,399,000 |
12d891dafb80118960224ba2762f4070add77e58 | Legalize some MHLO broadcasted And/Or/Xor ops to TF BitwiseAnd/BitwiseOr/BitwiseXor ops directly. tf.BroadcastTo op folder folds constants by default, this would increase the size of models converted from StableHLO because StableHLO requires explicit broadcasting. This change helps reduce model size by removing unneces... | Yishuang Pang | ypang@google.com | 1,696,457,787,000 |
8b8765eb05a9910cfc1accd9e201bf02a6c5fe11 | Integrate LLVM at llvm/llvm-project@173fd67a124d
Updates LLVM usage to match
[173fd67a124d](https://github.com/llvm/llvm-project/commit/173fd67a124d)
PiperOrigin-RevId: 570825329 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,457,956,000 |
f244daaa32bf89a561cd06c7d99b6c07d2e8d858 | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/99f2812e48077157a369a7574b829040ef4d1dad.
PiperOrigin-RevId: 570838674 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,460,628,000 |
1064ef978e2f14418c2c386badb4b1c97a6bad30 | Remove `register_extension_info` from xla.bzl and fix load statement order
PiperOrigin-RevId: 570841231 | David Dunleavy | ddunleavy@google.com | 1,696,461,194,000 |
f875204ca8afeb86293306fd75195de1d8765108 | Temporarily remove profile output and fix target typo
This is the only instance where the bazelrc references our hard-coded CI output paths, and it's not working correctly in some of the new CI layouts we're testing. We're not making use of this flag right now. If needed, we can restore it later.
PiperOrigin-RevId: 5... | Austin Anderson | angerson@google.com | 1,696,461,835,000 |
4d2b786cfc168479a51cbbb739679ebc3c30f550 | Legalize some MHLO broadcasted compare ops to TF Equal/NotEqual op directly. tf.BroadcastTo op folder folds constants by default, this would increase the size of models converted from StableHLO because StableHLO requires explicit broadcasting. This change helps reduce model size by removing unnecessary broadcasts at le... | Yishuang Pang | ypang@google.com | 1,696,462,463,000 |
8fb2d0901a965b5c1485bbac9fb0fd1bc76f42a4 | Transform batch_matmul to fully_connected in tflite if the weight is rank2 constant.
PiperOrigin-RevId: 570848144 | Mason Chang | masonchang@google.com | 1,696,462,820,000 |
1cdc904ee7b14bf31d4dec6cec87ce12d387dd12 | Remove AddGraphExportLoweringPasses as users should migrate to tf_to_dialect_executor
PiperOrigin-RevId: 570857911 | Mason Chang | masonchang@google.com | 1,696,465,426,000 |
6c872d682a16bf140ef047005b8661384501d434 | Remove upper_cost_threshold in TFRT serving
PiperOrigin-RevId: 570864891 | Kuangyuan Chen | chky@google.com | 1,696,467,567,000 |
d8e55c054734c31734a2164a4b35ad971261346f | Implement Zero-point-offset for Convolution
Currently we only support Conv2D and Conv3D with NHWC format. We also assume that RHS zp is 0 and LHS dilation is 1.
PiperOrigin-RevId: 570867408 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,468,277,000 |
cc579007fec72159cd5d457b8f977a28ff169069 | [stream_executor] NFC: Remove stream_executor_internal header from public API
`stream_executor_internal.h` should be used only for implementing platform-specific StreamExecutor backend, and not to get access to implementation details.
PiperOrigin-RevId: 570901850 | Eugene Zhulenev | ezhulenev@google.com | 1,696,481,821,000 |
319af2b3894ab06a0c9d53dc11d7f97fd7b82756 | [stream_executor] Add performance benchmarks for CUDA command buffers
----------------------------------------------------------------------
Benchmark Time CPU Iterations
----------------------------------------------------------------------
BM_CreateCommandBuffer/8 16... | Eugene Zhulenev | ezhulenev@google.com | 1,696,483,319,000 |
9d2607201ed37fef0b3d23249a2f757a9047a680 | NFC: replace uses of TF_DISALLOW_COPY_AND_ASSIGN/SE_DISALLOW_COPY_AND_ASSIGN macros with implementation.
PiperOrigin-RevId: 570913788 | Christian Sigg | csigg@google.com | 1,696,486,041,000 |
3e459a9503de26b24e0d6ee3ed694850da9596df | Merge branch 'master' into master | Guakocius | 98476579+Guakocius@users.noreply.github.com | 1,696,489,792,000 |
7d08b5ac00dc33494cd812e395a0616cc94ce3fa | Allow reduction users in multi-output fusions with buffer aliasing (FusionCanShareBufferHint)
PiperOrigin-RevId: 570942664 | Sergey Kozub | sergeykozub@google.com | 1,696,495,959,000 |
2c98406b22e9de15c0512529c98240d5b3d72619 | compat: Update forward compatibility horizon to 2023-10-05
PiperOrigin-RevId: 570944412 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,496,530,000 |
0a255348ac18dfdbb2ddf76a35337fc4f3dc740f | Update GraphDef version to 1640.
PiperOrigin-RevId: 570944422 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,496,532,000 |
8db9f5603146faf4cfbafce2c6d1a0104285fee9 | PR #6000: [ROCm] Fixing XLA tests build brake
Imported from GitHub PR https://github.com/openxla/xla/pull/6000
Recently there was a PR which broke our build process: https://github.com/openxla/xla/commit/933e2adb0e8c0c97db0c1f6793a7a59dc5ba6059
by adding CUDA-specific functions without any macro guards.
Additionally... | pemeliya | 141146080+pemeliya@users.noreply.github.com | 1,696,497,295,000 |
1ad489c1a5686703c695ad4bf2aeeae6087b3a7f | [Linaro:ARM_CI] Use portserver during testing
The use of portpicker is not thread safe by default so start
portserver before testing so that each invocation of portpicker
will defer to the single instance of portserver and thus avoid
the non thread safe behaviour that leads to duplicate ports being
attempted to be use... | Andrew Goodbody | andrew.goodbody@linaro.org | 1,696,498,772,000 |
5136c05062e83b77117ff6a4d622b80382008365 | Internal Code Change
PiperOrigin-RevId: 570961358 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,501,624,000 |
465e399fda2ee9b6620a3c342bd989ac74ddace1 | Extract out Bridge build targets out of tensorflow_passes in preparation to move
them out to internal directory.
PiperOrigin-RevId: 570995527 | Mason Chang | masonchang@google.com | 1,696,511,986,000 |
a458909b010418f88275edfaef819954dbd24956 | Display ResultStore links to Bazel RBE invocations.
PiperOrigin-RevId: 571004660 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,514,499,000 |
f9ae0356fff15ed2123cbbc2ca02854f942b506a | Extract EnableDetailedLogging out into its own helper method
PiperOrigin-RevId: 571008489 | Mason Chang | masonchang@google.com | 1,696,515,591,000 |
9f72699c17a9c790917004a2b60988e0f414ca5b | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,010,000 |
f4238c862f65ff727e39640d8bb5e0d515d3cd44 | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,018,000 |
3c857b2629c847693036ce6531eae32289078c48 | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,026,000 |
11a37825005cec22e1325027942df4f76b51695f | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,036,000 |
2ef1f7d6f3717068bfbb1e55e7348cfe044ae9d8 | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,050,000 |
ca5f639b2f6025e2e1c2cae7083c7275a10e82e8 | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,063,000 |
898a98cedd410e98ec1e5d33f6f70624089edb0f | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,076,000 |
eb31a86028aaedc4581fee938985c0eefe11de32 | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,092,000 |
85a8227f521a75672a0b83cad09d2a388813c32e | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,105,000 |
3cbdfb2eafa68269b8fd2231ea4e57fd93a53c9f | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,117,000 |
a61513037ab64d80d14444efaea79a7247864ab7 | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,130,000 |
ca1380122fa5de4d85acb27378ae16a0ce1773fb | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,141,000 |
d91e5641a7542e4ae7a5a2c4e24acdbb3fd3f0cd | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,150,000 |
20eebe47ec93134d08ef10e5eabaf3ef6a6294cc | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,158,000 |
eb524e64cddec8f6e3750afdff96711d8351f213 | Apply reviewer suggested change | Mihai Maruseac | mihai.maruseac@gmail.com | 1,696,518,167,000 |
5251b89936b7c52ba1c0f10a6ea9950d557d0c31 | Rollback of [XLA:GPU] Allow reduction users in multi-output fusions with buffer aliasing (FusionCanShareBufferHint)
PiperOrigin-RevId: 571017871 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,518,094,000 |
a9c481cd7ce8c3fe7b2f01160cfdbde89925ead9 | Disable int4 tests when XLA is used.
XLA does not yet support int4. I'll revert this change once it does support it. I explicitly caused XLA to raise an error in https://github.com/openxla/xla/commit/42d69767d33faba12bfdf6799ed2b0313754b4bb, which caused these TF tests to fail.
PiperOrigin-RevId: 571029725 | Reed Wanderman-Milne | reedwm@google.com | 1,696,520,591,000 |
5a46ad34b8addb3a79227fbe264c3a1cce016ffc | Merge pull request #61988 from Guakocius:master
PiperOrigin-RevId: 571039580 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,523,234,000 |
5118a877a31b33d3a13503ab62c3188daa32b40b | Initial import of guesslang from https://github.com/yoeo/guesslang.
PiperOrigin-RevId: 571046450 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,524,117,000 |
3524b23ae1ed391f910a64c6211f9c81e9566f9e | Check in generated pyi files for some py_extension targets.
PiperOrigin-RevId: 571051476 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,525,080,000 |
4e78f725375c81d700177e273cf7c0b22784c0b6 | Merge pull request #61878 from Intel-tensorflow:mraunak/win_oneDNN
PiperOrigin-RevId: 571063279 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,527,610,000 |
223fca869563405963ce99e92b342ec03191f3fe | Check in generated pyi files for some py_extension targets.
PiperOrigin-RevId: 571067137 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,527,978,000 |
f5c62a620b26c2a1a252b0ab7192eb2aa71aa80c | Add multi-device SPMD test target.
PiperOrigin-RevId: 571068377 | Justin Szaday | twelve@google.com | 1,696,528,227,000 |
aa94dd6d0a58097ca81dfe60f0802f24e25797dc | Integrate StableHLO at openxla/stablehlo@46a25064
PiperOrigin-RevId: 571068693 | Gunhyun Park | gunhyun@google.com | 1,696,528,290,000 |
8541a32b305099bea39795d66a034adae1fbd790 | [XLA:SPMD] Fix wrong broadcast shape in the Gather trivial sliced dimension partitioning.
PiperOrigin-RevId: 571073338 | Tongfei Guo | tongfei@google.com | 1,696,529,170,000 |
8a5e6bee8253e287cf200dae1de3ec1ea00ab4f5 | Merge pull request #61913 from kobrineli:fix-nullderef
PiperOrigin-RevId: 571079178 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,530,492,000 |
a089c4b2777dcd2f88f55229b920102f10264427 | Fix TF_NCCL_VERSION detection.
The current macro substitution wasn't working and is resolving to %{nccl_version} rather than a version string. Instead, add code to create a NCCL config header (nccl_config.h), and use that to detect the .so version we should be opening.
This change should only affect NCCL when obtaine... | Peter Hawkins | phawkins@google.com | 1,696,530,688,000 |
030d66b6298038c37ba47cf6c1c465ea9c8d0d15 | Fix bats output directory for newer scripts
PiperOrigin-RevId: 571082308 | Austin Anderson | angerson@google.com | 1,696,530,762,000 |
4e68afb135286a1c6335949504a1dee50085e2b3 | Make test::TensorEq public so that people don't have to keep reimplementing it
PiperOrigin-RevId: 571096079 | Victor Akabutu | victorakabutu@google.com | 1,696,533,672,000 |
c02f018d21a5c6ae22c48f07f65757cd2aea6f61 | #tf-data-service Retry on `DEADLINE_EXCEEDED`s.
A `DEADLINE_EXCEEDED` may indicate a worker having been preempted or downscaled.
PiperOrigin-RevId: 571098292 | Matt Callanan | mpcallanan@google.com | 1,696,534,229,000 |
310d20d49d847aa1a0111101dd7a73d5c3ef85e0 | TPU Builds are failing due to incorrect function call
PiperOrigin-RevId: 571101826 | Chandra Devarakonda | chandrasekhard@google.com | 1,696,535,071,000 |
27b08db38197553bd5f01463397043327b62a669 | Legalize TensorListLength to tflite custom + end2end tests.
PiperOrigin-RevId: 571103368 | Luke Boyer | lukeboyer@google.com | 1,696,535,439,000 |
6fcee23a99553064bd448729a84887ffbc80c82c | Make TFG graph transformation passes visible to the inference converter V2
PiperOrigin-RevId: 571104776 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,535,753,000 |
05185eceddf38175e3c394c2977497d5e6a5222b | Reuse TensorListGetItem to implement TensorListPopBack
PiperOrigin-RevId: 571115374 | Luke Boyer | lukeboyer@google.com | 1,696,538,018,000 |
1216839db9011c7c30fb4c137eec84e01a38542d | Make `BufferInfo` constructor a `constexpr`.
PiperOrigin-RevId: 571120368 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,539,088,000 |
9219692e7399fcb6ff7dbfa0b05af17b97228a3c | #tf-data Turn up `"file_locality"` (v1) experiment to 10% job level.
PiperOrigin-RevId: 571124258 | Matt Callanan | mpcallanan@google.com | 1,696,539,872,000 |
068bee993d59252a46c504d17162486218f3bc82 | [XLA] Implement buffer isolation
PiperOrigin-RevId: 571128305 | Berkin Ilbeyi | berkin@google.com | 1,696,540,732,000 |
fae6dd87b5f8f2058fa33c4523d5717b4caa7a41 | [JAX] Keep CPU host callbacks alive via IFRT, rather than by attaching them to the Python object.
We need to keep callback objects alive as long as any running executables are alive. It is possible to discard the Python data structures for an executable before the runtime has finished running that executable, which ca... | Peter Hawkins | phawkins@google.com | 1,696,543,589,000 |
bb9b59d09300e959fe9f2005fd15de42820c2009 | Split client target into single-source targets.
PiperOrigin-RevId: 571148425 | Juan Martinez Castellanos | juanantoniomc@google.com | 1,696,545,284,000 |
02971e0ad44fb36575fe974be94661f965eb5fed | Update post_training_integer_quant.ipynb
Remove unused labels from the run_tflite_model() method. | Jerome MASSOT | j_massot@yahoo.fr | 1,696,546,472,000 |
4238d1dba2cc4f32e8671533e57ed436ca567d30 | Remove AddGraphExportLoweringPasses as users should migrate to tf_to_dialect_executor
PiperOrigin-RevId: 571153277 | Mason Chang | masonchang@google.com | 1,696,546,457,000 |
972eafd4f7047dc1e7146038d53b9de25e65856f | Export TF Dialect To Executor V2 as a pass pipeline
PiperOrigin-RevId: 571154754 | Mason Chang | masonchang@google.com | 1,696,546,825,000 |
fe98f78d25939c1e665585161a6f9fffb627d08d | Propagate stack traces when creating a GraphExecutionState. Constructing a FunctionLibraryDefinition was previously throwing away stack traces. Extract them from the GraphDef, and store them in the FunctionLibraryDefinition.
PiperOrigin-RevId: 571155259 | Alan Liu | liualan@google.com | 1,696,546,895,000 |
085e4e6cce6f295e46ae4e9f0904703a65d29603 | Fix BISECT and ANY scripts to accept relative paths
Specifying paths is now more consistent across TFCI and TF_* parameters, and all are now "ci/official/thing".
PiperOrigin-RevId: 571162798 | Austin Anderson | angerson@google.com | 1,696,548,583,000 |
36c59c42bda0cbe771718ffef3a1301743c74d4d | Document cause of segfault.
PiperOrigin-RevId: 571168287 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,549,560,000 |
5a90e2753802c09b0ac22c94ad078d5ac64b844d | Add fields to the CollectiveExecutor trace metadata.
PiperOrigin-RevId: 571168796 | Derek Murray | mrry@google.com | 1,696,549,679,000 |
936be439aceb33d918886fbd63c3020be2d3728c | ExtractTPUCopyWithDynamicShape uses device inspection utility to find TPU 0 and host 0 in the non-replicated case.
PiperOrigin-RevId: 571169160 | Michael Delorimier | mdel@google.com | 1,696,549,732,000 |
8dbd30a76ad2ab8663027f0eef222b51da4e6435 | Integrate LLVM at llvm/llvm-project@73f8ec9edb52
Updates LLVM usage to match
[73f8ec9edb52](https://github.com/llvm/llvm-project/commit/73f8ec9edb52)
PiperOrigin-RevId: 571169396 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,549,772,000 |
c15fc78b86672e7461424ec4fceefbc8b148008d | #tf-data-service Retry on `DEADLINE_EXCEEDED`s.
A `DEADLINE_EXCEEDED` may indicate a worker having been preempted or downscaled.
PiperOrigin-RevId: 571169676 | Matt Callanan | mpcallanan@google.com | 1,696,549,819,000 |
de714987796e53b217a3991c25d8f344897474a9 | [PJRT:CPU] Add CreateErrorBuffer implementation.
PiperOrigin-RevId: 571186318 | Ce Zheng | zce@google.com | 1,696,554,228,000 |
df36243ed4994967040f11ddfeb819ac81f7cf37 | If an instruction has a specified user sharding, and there is only a single
strategy, do not remove the strategy even if invalid. Given that the user has
specified this sharding strategy, we should respect it.
PiperOrigin-RevId: 571190693 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,555,454,000 |
63bc8d2fd33f24699cc957e7b4e16996568736a2 | Merge pull request #62052 from elfringham:use_portserver
PiperOrigin-RevId: 571192481 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,556,526,000 |
c70b5bf2c7264bd09a17ffa3ed4e01d5b9fc97cd | [xla:gpu] Move kParameter instructions to the beginning of the schedule before constructing command buffers
https://github.com/openxla/xla/issues/5756
PiperOrigin-RevId: 571194180 | Anlun Xu | anlunx@google.com | 1,696,557,004,000 |
0b6fdce7dccdc91d48c44cb2048902bb3d29dfec | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/7e9cb2e1ab904f02c14e0741acdf33fcea42b6b4.
PiperOrigin-RevId: 571200931 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,560,033,000 |
c7577ffe2d1c6c0aa5ce30c0b6500b6fe7fd71fc | Merge pull request #59975 from tensorflow:tiruk007-patch-5
PiperOrigin-RevId: 571244649 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,575,784,000 |
3ae2164f2eebf20eb35d9cc530b57d584857c86c | Integrate LLVM at llvm/llvm-project@58678d3bcf32
Updates LLVM usage to match
[58678d3bcf32](https://github.com/llvm/llvm-project/commit/58678d3bcf32)
PiperOrigin-RevId: 571246081 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,575,807,000 |
7d0df464de03eb3ca4c509705ebb7fabaee9c75f | Internal Code Change
PiperOrigin-RevId: 571252798 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,577,712,000 |
750de399c21dc053b998207c18fc5858ebdfa2c1 | Internal Code Change
PiperOrigin-RevId: 571253248 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,577,855,000 |
5790bf22fe1102387c6a3b4cd3302bf0e0cd63c3 | Wait for async results in the multihost HLO runner.
Without this, benchmarking is not possible, as there's no synchronization after running the executable.
PiperOrigin-RevId: 571253810 | Sergey Kozub | sergeykozub@google.com | 1,696,578,024,000 |
92275b51192d0bd185dd8463e3cdbda83cbbd939 | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/c4a4bac3bc8a880886261a6f0a4076641dc5d3ab.
PiperOrigin-RevId: 571254197 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,578,125,000 |
dfe6d8b9e86e8ed81a22a4662c9850d771e636d8 | compat: Update forward compatibility horizon to 2023-10-06
PiperOrigin-RevId: 571271064 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,582,933,000 |
82471c7278d506ff2c47181d8cfbf8ae82b797b6 | Update GraphDef version to 1641.
PiperOrigin-RevId: 571271073 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,582,934,000 |
b33f31f93c1a9d9f25b4052f4318354994a97ec5 | Internal Code Change
PiperOrigin-RevId: 571279747 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,585,239,000 |
f1488515ca8f7673c882d305ab76897e35c5ade3 | [XLA:GPU] Remove a duplicate (costly) LogAndVerify call from EmitTritonFusion
There is already a LogAndVerify at the end of TritonWrapper.
PiperOrigin-RevId: 571279912 | Tamás Danyluk | tdanyluk@google.com | 1,696,585,303,000 |
43426b083bb48e3d0696344c97f406de628cef0e | hipblas-lt changes for tensorflow | Pavel Emeliyanenko | pavel.emeliyanenko@amd.com | 1,695,309,897,000 |
446e8e2cc39b73ecc8268dc5eaa019f68619c829 | fixes according to review + one missing file | Pavel Emeliyanenko | pavel.emeliyanenko@amd.com | 1,695,820,483,000 |
a398716cf0a31857a6c69ed1f825aa6de591862a | fixing linker errors | Pavel Emeliyanenko | pavel.emeliyanenko@amd.com | 1,696,236,875,000 |
70ba8a948e474faaeb572ba2c158fd1268e89652 | rebased and fixed BUILD file | Pavel Emeliyanenko | pavel.emeliyanenko@amd.com | 1,696,589,696,000 |
09722c78aa4a9601c95dcaec3d13ed72b9a362e3 | Trim the unnecessary device_name layer.
PiperOrigin-RevId: 571301108 | Shiqing Yan | shiqing@google.com | 1,696,592,161,000 |
7da0102e16a6847262037d72b7e13f6df9a455fc | GpuPerformanceModel: use HLO fusion analysis to get more accurate thread counts.
Note: this probably isn't great for compile time. We'll deal with it later.
PiperOrigin-RevId: 571312897 | Johannes Reifferscheid | jreiffers@google.com | 1,696,596,372,000 |
df0a4f8c5b4c482015044f0c911413f4055407a1 | Adds support for 3D device mesh shapes (i.e., having three dimensions with size greater than one).
PiperOrigin-RevId: 571321597 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,599,078,000 |
165910603af9a073d07031c18937a9c26ca83f14 | [XLA:GPU] Copy module to new context faster
Serialize to Bitcode instead of text.
This way of (de)serialization is about 1,73x faster.
PiperOrigin-RevId: 571328596 | Tamás Danyluk | tdanyluk@google.com | 1,696,601,350,000 |
ba63f639eb85ce090b51499dab5a45e098f08b5e | Propagate shape into functions by inferred type when input type is ranked
PiperOrigin-RevId: 571332268 | Chen Li | chencli@google.com | 1,696,602,539,000 |
3353e238d4335c6cbee7eea60b5a70a553a43aa0 | Integrate LLVM at llvm/llvm-project@ff5e2babcb46
Updates LLVM usage to match
[ff5e2babcb46](https://github.com/llvm/llvm-project/commit/ff5e2babcb46)
PiperOrigin-RevId: 571338203 | Krasimir Georgiev | krasimir@google.com | 1,696,604,331,000 |
9c8c0a097f1ea1e73e80cd24cd52cc59aecd24fc | [XLA:GPU] [NFC] Remove dead code in GPU compiler
Remove unused boilerplate.
PiperOrigin-RevId: 571340881 | George Karpenkov | cheshire@google.com | 1,696,604,987,000 |
3f1d26404ace031b6c98c1e502347a0ac439fb59 | Make optimized graph cache writes atomic
PiperOrigin-RevId: 571347004 | Philip Pham | phillypham@google.com | 1,696,606,799,000 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.