Commit Graph

132403 Commits

Author SHA1 Message Date
Sadeed pv
f0d522b51e
Update SECURITY.md 2022-07-26 10:42:38 +04:00
Sadeed pv
8f90da7975
Update SECURITY.md 2022-07-26 10:41:20 +04:00
Sadeed pv
d38cb2f6a7
Update SECURITY.md 2022-07-26 10:39:12 +04:00
Sadeed pv
fb568f9e29
Update SECURITY.md 2022-07-26 10:37:36 +04:00
Sadeed pv
94e3230e76
Update ISSUES.md 2022-07-26 10:32:43 +04:00
unknown
2c443c6b49 Fix markdown files in security.md and issues.md 2022-07-23 18:11:51 +04:00
A. Unique TensorFlower
d42edf7010 Update GraphDef version to 1201.
PiperOrigin-RevId: 462788603
2022-07-23 02:09:42 -07:00
A. Unique TensorFlower
3726ed139f compat: Update forward compatibility horizon to 2022-07-23
PiperOrigin-RevId: 462788097
2022-07-23 02:05:40 -07:00
TensorFlower Gardener
21122d1ed2 Merge pull request #54379 from Intel-tensorflow:fix_fused_op_test_final
PiperOrigin-RevId: 462744435
2022-07-22 18:33:28 -07:00
Dinghua Li
fe1127e45d Fix an Android build error due to missing std::is_invocable_r in Android NDK.
PiperOrigin-RevId: 462741365
2022-07-22 18:05:15 -07:00
Yuqi Li
1f3c8ebfc7 Remove MobileBERT from Text Embedder Task API doc.
PiperOrigin-RevId: 462732563
2022-07-22 17:00:33 -07:00
TensorFlower Gardener
87641fbfca Merge pull request #56824 from Intel-tensorflow:aimran/oneDNN_upgrade
PiperOrigin-RevId: 462720011
2022-07-22 15:48:19 -07:00
Rahul Joshi
b599708f57 [NFC] Fix description of HLO Recv instruction.
PiperOrigin-RevId: 462718611
2022-07-22 15:40:44 -07:00
A. Unique TensorFlower
d840e52f46 Update TFRT dependency to use revision
f993ddbc75.

PiperOrigin-RevId: 462715093
2022-07-22 15:29:45 -07:00
A. Unique TensorFlower
5bac37eadc Update ops-related pbtxt files.
PiperOrigin-RevId: 462714708
2022-07-22 15:26:17 -07:00
HanBin Yoon
ac3188356c [xla::jitrt] Encode conv dimensions and backend config as XLA structs to custom call.
PiperOrigin-RevId: 462712356
2022-07-22 15:09:12 -07:00
HanBin Yoon
4b2264426a [xla::jitrt] Encode gemm (dot) dimensions (MHLO attr) as XLA struct to custom call.
PiperOrigin-RevId: 462705780
2022-07-22 14:39:06 -07:00
A. Unique TensorFlower
99e7631c7f Allow Mediapipe Tasks to depend on tensorflow/lite/*
PiperOrigin-RevId: 462704636
2022-07-22 14:32:58 -07:00
A. Unique TensorFlower
b427480be6 Update TFRT dependency to use revision
158fa0c499.

PiperOrigin-RevId: 462697231
2022-07-22 14:02:39 -07:00
A. Unique TensorFlower
08fc24e470 Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId: 462695304
2022-07-22 13:58:44 -07:00
A. Unique TensorFlower
af97dcc6a6 Update TFRT dependency to use revision
bddb1b9208.

PiperOrigin-RevId: 462693254
2022-07-22 13:55:19 -07:00
TensorFlower Gardener
731017f85b Merge pull request #56844 from sstamenova:mhlodialectdeps
PiperOrigin-RevId: 462692939
2022-07-22 13:50:48 -07:00
A. Unique TensorFlower
a7a57ed3b1 Add Mha and Tmlp to Cuda AllowList
PiperOrigin-RevId: 462692410
2022-07-22 13:43:05 -07:00
Yang Chen
76aeaa7b99 #tf-data-service register_dataset accepts a dataset_id.
If provided, tf.data service will use the provided ID for the dataset.
If the dataset ID already exists, no new dataset will be registered.
This is useful if multiple training jobs need to use the same dataset
for training. In this case, users should call `register_dataset` with
the same `dataset_id`.

PiperOrigin-RevId: 462690057
2022-07-22 13:26:35 -07:00
George Karpenkov
2347ee9015 [XLA:GPU] [NFC] Minor reduction emitter refactoring
PiperOrigin-RevId: 462689166
2022-07-22 13:21:47 -07:00
A. Unique TensorFlower
129925397b Instrument ExecutorState::Process with activity watcher.
PiperOrigin-RevId: 462685644
2022-07-22 13:12:56 -07:00
Augie Fackler
2e0502651c Integrate LLVM at llvm/llvm-project@9f6ba4be26
Updates LLVM usage to match
[9f6ba4be2685](https://github.com/llvm/llvm-project/commit/9f6ba4be2685)

PiperOrigin-RevId: 462685176
2022-07-22 13:07:03 -07:00
Antonio Sanchez
32d7bd3def Fix MaxPool crash on GPU for invalid filter size.
If the filter size exceeds the input size by one for `VALID` padding,
return an empty tensor. This is consistent with XLA.

PiperOrigin-RevId: 462684864
2022-07-22 13:03:45 -07:00
A. Unique TensorFlower
86a827c7b4 Automated visibility attribute cleanup.
PiperOrigin-RevId: 462683823
2022-07-22 12:59:42 -07:00
TensorFlower Gardener
8aa8d0ad1e Merge pull request #56841 from elfringham:cross_trainer_fix
PiperOrigin-RevId: 462683641
2022-07-22 12:56:23 -07:00
Dinghua Li
8bd711813c Instrument ExecutorState::Process with activity watcher.
PiperOrigin-RevId: 462682671
2022-07-22 12:50:25 -07:00
Robert Suderman
e725f4a8c4 Update tfl.gather lowering to tosa.gather for dynamism
Dynamic shapes can be supported by tosa.gather so long
as tosa.reshape is never passed more than a single
dynamic shape. Update the lowering and tests to expand
dynamism.

PiperOrigin-RevId: 462682345
2022-07-22 12:46:32 -07:00
A. Unique TensorFlower
28e801b08e Update TFRT dependency to use revision
6c1f130ded.

PiperOrigin-RevId: 462680527
2022-07-22 12:38:26 -07:00
Chris Jones
dae21b7b86 [XLA:GPU] Fix for fusing vector bias then matrix bias adds into cublasLt matmul.
PiperOrigin-RevId: 462667320
2022-07-22 11:37:22 -07:00
A. Unique TensorFlower
f9ea427340 Automated visibility attribute cleanup.
PiperOrigin-RevId: 462649060
2022-07-22 10:17:16 -07:00
A. Unique TensorFlower
a65715658a [mlir][chlo] Lower chlo.asin operator for complex element type.
See 6a2e3dcd5f/tensorflow/compiler/xla/client/lib/math.cc (L1168-L1171) for implementation in XLA (which uses the same sequence of operations for both complex and non-complex type).

PiperOrigin-RevId: 462644260
2022-07-22 09:58:08 -07:00
A. Unique TensorFlower
a1f5585ac9 Update TFRT dependency to use revision
01e5021627.

PiperOrigin-RevId: 462642799
2022-07-22 09:50:26 -07:00
Kuangyuan Chen
177a072d37 Implement error handling in send/recv based host callbacks.
PiperOrigin-RevId: 462641478
2022-07-22 09:45:16 -07:00
Benjamin Kramer
579c67bd60 [xla:cpu] Try to pick the correct ABI for fp16 conversion runtime functions
The x86_64 psABI defines fp16 as passed in floating point registers instead of
general purpose registers. This means using uint16_t stopped working when LLVM
recently adopted that ABI rule.

The proper fix for this is using _Float16 instead but compiler support is still
spotty. GCC has supported it for a couple of versions, Clang as of now only in
trunk and other compilers don't have it at all.

To fix this we pick _Float16 when it's available on new GCC and Clang, and fall
back to float and careful casting if we don't have it.

PiperOrigin-RevId: 462640054
2022-07-22 09:34:35 -07:00
Augie Fackler
9c9c87120f Integrate LLVM at llvm/llvm-project@41958f76d8
Updates LLVM usage to match
[41958f76d8a2](https://github.com/llvm/llvm-project/commit/41958f76d8a2)

PiperOrigin-RevId: 462638952
2022-07-22 09:30:48 -07:00
TensorFlower Gardener
cf791c2672 Merge pull request #56788 from ROCmSoftwarePlatform:fix-upstream-rocm-unit-tests
PiperOrigin-RevId: 462631309
2022-07-22 08:49:27 -07:00
George Karpenkov
ed24612418 [TF2XLA] Reland: fix layout for light outside compilation custom call
Fix: make a test more resilient.
PiperOrigin-RevId: 462613935
2022-07-22 07:13:06 -07:00
A. Unique TensorFlower
e2009cbe95 Update TFRT dependency to use revision
28bd500e25.

PiperOrigin-RevId: 462598614
2022-07-22 05:25:34 -07:00
Benjamin Kramer
103dd9dcfe Integrate LLVM at llvm/llvm-project@35b80c448b
Updates LLVM usage to match
[35b80c448bbc](https://github.com/llvm/llvm-project/commit/35b80c448bbc)

PiperOrigin-RevId: 462591852
2022-07-22 04:33:31 -07:00
A. Unique TensorFlower
b7a835fc76 compat: Update forward compatibility horizon to 2022-07-22
PiperOrigin-RevId: 462571546
2022-07-22 02:07:56 -07:00
A. Unique TensorFlower
5382e652b7 Update GraphDef version to 1200.
PiperOrigin-RevId: 462571532
2022-07-22 02:04:32 -07:00
A. Unique TensorFlower
e3493018f5 Write the measured costs to disk from SharedContext and add unit tests.
PiperOrigin-RevId: 462547558
2022-07-21 23:27:55 -07:00
Hanhan Wang
e0a0325527 Fixes for cmake BUILD
PiperOrigin-RevId: 462545892
2022-07-21 23:14:38 -07:00
A. Unique TensorFlower
f3cf67ac57 Add IsScalar / IsVector (rank) checks to input min/max tensors for FakeQuantWithMinMaxVarsPerChannelGradientOp and FakeQuantWithMinMaxVarsGradientOp.
PiperOrigin-RevId: 462542629
2022-07-21 22:52:41 -07:00
TensorFlower Gardener
7f64135668 Merge pull request #56828 from Intel-tensorflow:bhavanis/pooling_ops_3d_test_fix
PiperOrigin-RevId: 462542402
2022-07-21 22:47:48 -07:00