Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Develop upstream sync 230201 #1987

Merged
merged 545 commits into from
Feb 3, 2023
Merged

Develop upstream sync 230201 #1987

merged 545 commits into from
Feb 3, 2023

Conversation

i-chaochen
Copy link

@i-chaochen i-chaochen commented Feb 1, 2023 •
edited
Loading

Disable unit tests:

  1. multi_client_test_nccl_local_2gpus Failed UT: multi_client_test_nccl_local_2gpus #1980

  2. device_tracer_test_gpu https://github.com/ROCmSoftwarePlatform/frameworks-internal/issues/3175

  3. to fix estimator_export error, I put estimator_export back to tf_export tensorflow@3837d0f

  4. cudnn_determinstic_ops_test needs to check. https://github.com/ROCmSoftwarePlatform/frameworks-internal/issues/3563

5. tensor_or_memref.h is failed to build at cpu test

Also, roc_blas.cc needs to be upstreaming.
https://github.com/ROCmSoftwarePlatform/frameworks-internal/issues/3547

tensorflower-gardener and others added 30 commits January 26, 2023 11:59
@tensorflower-gardener
Fix XLA compilation error due to shape information loss during sort
b5018b6 
Before this change, XLA compilation fails when calling `tf.image.non_max_suppression_padded` with undertemined batch size for input boxes.

The reason behind this failure is because:
1. When calling this op with parameter `sorted_input=False`, `_sort_scores_and_boxes` will be called.
2. In `_sort_scores_and_boxes`, if input boxes has dynamic batch size, during XLA compilation time, `tf.shape(boxes)[1]` will be unknown because of this reshape operation.
3. In the while loop body (`supression_loop_body`), `num_tiles` cannot be determined because `tf.shape(boxes)[1]` is unkonwn during compilation.
4. The op `math_ops.range(num_tiles)` failed to be compiled because `num_tiles` is unknown during compilation time.

This change will fix this issue by adding `num_boxes` to the reshape op.

PiperOrigin-RevId: 504902678
@skye @tensorflower-gardener
[JAX] Allow JAX_USE_PJRT_C_API_ON_TPU to be 0
1869d54 
Prior to this change, it only accepted 1, false, or true.

PiperOrigin-RevId: 504902827
@lhutton1
[tosa] Legalize tfl.atan2 to tosa.table operation
4b48a56 
In a similar fashion to tfl.sin and tlf.cos legalization,
a tosa.table provides an atan approximation. Additional logic
is then used to determine the correct quadrant of the atan2
function.

Signed-off-by: Luke Hutton <[email protected]>
Change-Id: Iae1384009d825d01e5cf48ad7c3ff8fba77114cf
@lhutton1
address comments
cbe054a 
* remove unnecessary type checks
* add note about numerical behaviour of std::atan2
* improve error message for expected inputs
* undo change updating copyright year

Change-Id: Iea8339da437a5ff3e6fe065c715c2c97e696fdbb
@tensorflower-gardener
[xla:runtime] Use custom class for ExecutionReference
2b08df9 
Currently users have to provide dummy deleter if they want to declare ExecutionReference variable. This change solves that issue by declaring ExecutionReference as a class, which wraps std::unique_ptr<ExecutionContext, void(*)(ExecutionContext *)>.

Ex.
struct Test {
  ExecutionReference exec_ref;
};

PiperOrigin-RevId: 504904730
@fionalang @tensorflower-gardener
Update ops references from framework/type_spec.py to `framework/typ…
02438f1 
…e_spec_registry.py`.

PiperOrigin-RevId: 504905987
@w-xinyi @tensorflower-gardener
Update the metric.
f707809 
PiperOrigin-RevId: 504906696
@JunyoungLim @tensorflower-gardener
Factor out get_default_ops and make get_ops_from_nodedef a public met…
4b46509 
…hod in TF selective_registration_header_lib.

PiperOrigin-RevId: 504910088
@tensorflower-gardener
For dynamic shapes, need to read metadata to get literal's real size.
40763f2 
PiperOrigin-RevId: 504910311
@tensorflower-gardener
Removed code related to PipelineSubMeshes.
6285a27 
PiperOrigin-RevId: 504910444
@tensorflower-gardener
[mhlo] Reuse 4 shape functions from StableHLO
27e0323 
Move verifyRecvOp and verifyInfeedOp with TokenType
openxla/stablehlo#852

Move inferFftOp and verifyRngOp with Enums
openxla/stablehlo#853

PiperOrigin-RevId: 504910725
@fionalang @tensorflower-gardener
Update references in python/data/util from framework/type_spec.py…
a342ad9 
… to `framework/type_spec_registry.py`.

PiperOrigin-RevId: 504911860
@rdzhabarov @tensorflower-gardener
[TFG] Fix index out of bound issue in TFG shape inference pass.
62efa48 
Ensure that there is at least a single non control input to the Identity/IdentityN.

PiperOrigin-RevId: 504913650
@fionalang @tensorflower-gardener
Update references in python/ops/linalg from `framework/type_spec.py…
024892d 
…` to `framework/type_spec_registry.py`.

PiperOrigin-RevId: 504917585
@faizan-m @tensorflower-gardener
Migrate remaining tf.function validation to FunctionType
b7d4db7 
PiperOrigin-RevId: 504920230
@grantjensen @tensorflower-gardener
Update selective_build_scripts with NDK r21e
29bb19a 
PiperOrigin-RevId: 504922596
@cheshire @tensorflower-gardener
[XLA:GPU] Layout normalization for dynamic-update-slice
f481e61 
PiperOrigin-RevId: 504923081
@tensorflower-gardener
Preserve HLO names when removing infeed and outfeed
f67e32b 
PiperOrigin-RevId: 504928473
@anlunx @tensorflower-gardener
[XLA:GPU] Reuse GetBestAlgorithm in runtime autotuning
dfc491d 
Export GetBestAlgorithm fom GemmAlgorithmPicker so that runtime autotuning use it as a subroutine.

PiperOrigin-RevId: 504928946
@panzhufeng @tensorflower-gardener
Add check for "ksize" argument in MaxPoolWithArgmax
bec8b93 
PiperOrigin-RevId: 504930976
@tensorflower-gardener
Reduce the number of shape inference reruns for nested function calls.
dab702f 
The shape refiner reruns shape inference for nested function calls every time a materialized argument is requested. Before this change the constant folding mechanism used in the refiner stopped traversing the subgraph right after attending the first unresolved argument. After this change it will continue traversal to cover all unresolved arguments.

This change reduces preprocessing time of some of our models by 75%.

PiperOrigin-RevId: 504931297
@tensorflower-gardener
Migrate remaining tf.function validation to FunctionType
88ec865 
PiperOrigin-RevId: 504935198
@jurahul @tensorflower-gardener
[NFC] Use correct thunk kind for NcclCollectivePermuteStartThunk
a8ae7be 
PiperOrigin-RevId: 504937807
@wilsingosti @tensorflower-gardener
#tf-data A stop gap fix to add available memory check in prefetch aut…
9f4acbb 
…otuner.

PiperOrigin-RevId: 504939086
@wilsingosti @tensorflower-gardener
#tf-data Restart the stage_based_autotune experiment at 5%
a887b9f 
PiperOrigin-RevId: 504939291
@BrianWieder @tensorflower-gardener
Move MatMul flop calculation methods to flops_registry.py.
8326cb7 
This breaks a cycle between "python/framework/graph_util_impl.py" and "python/framework/convert_to_constants.py".

PiperOrigin-RevId: 504939493
@tensorflower-gardener
Merge pull request tensorflow#59436 from awsaf49:to_ordinal
df98914 
PiperOrigin-RevId: 504940089
@tensorflower-gardener
Change DTensor to pass raw Module to ParallelExecutor
c7aa8db 
PiperOrigin-RevId: 504940378
@lhutton1
address comments
3986b07 
* Cleanup of error messages.
* Cleanup casting inputs/outputs.
* Use CHECK-DAG instead of CHECK for constants.
* Spelling

Change-Id: Ibf01ac5b711944bb2efbd7184820a304cbae8501
Signed-off-by: Luke Hutton <[email protected]>
@yangustc07 @tensorflower-gardener
#tf-data-service Add thread annotations.
91e8d71 
PiperOrigin-RevId: 504949702
tyb0807 and others added 19 commits January 31, 2023 22:23
@tyb0807 @tensorflower-gardener
[xla:cpu] Add debug info to XLA CPU pipeline
12eef94 
This adds a pass that provides some debug info with which basic line number info can be generated.

Adapted from Flang's AddDebugFoundationPass.

PiperOrigin-RevId: 506213461
@tensorflower-gardener
update fuzztest dependency
78034c6 
PiperOrigin-RevId: 506217195
@ddunl @tensorflower-gardener
Remove references to stream_executor/lib
9edf0d9 
PiperOrigin-RevId: 506225078
@tensorflower-gardener
Merge pull request tensorflow#58763 from shawnwang18:upstream/xla_hor…
722fbb8 
…izontal_loop_fusion

PiperOrigin-RevId: 5062358
@tensorflower-gardener
[XLA:GPU] Handle device buffers more safely in run_hlo_module
e8b3ac7 
This fixes double-free errors or memory leaks for example when the running of the HLO is unsuccessful.

The old code-path is also left there, as a lot of our code depends on the ability to run the same HLO multiple times without reallocating the input buffers.

PiperOrigin-RevId: 506238363
@tensorflower-gardener
compat: Update forward compatibility horizon to 2023-02-01
98c1863 
PiperOrigin-RevId: 506239134
@tensorflower-gardener
Update GraphDef version to 1394.
e784dac 
PiperOrigin-RevId: 506239156
@tensorflower-gardener
Fix a typo in the documentation in preemption_watcher.py
d38f8f0 
PiperOrigin-RevId: 506240202
@tensorflower-gardener
Rollback of PR tensorflow#58763
6802154 
PiperOrigin-RevId: 506243978
@pifon2a @tensorflower-gardener
[GmlSt] Group tiling passes for cpu, gpu and triton.
b753419 
PiperOrigin-RevId: 506244287
@tensorflower-gardener
Propagate quantize_params in prepare_pass
f206f16 
PiperOrigin-RevId: 506252805
@i-chaochen
iinnit merge
f0b3a19
@i-chaochen
resolve all conflicts
ead926e
@i-chaochen
enable rocm/hip for multi_client_test
57d4b8d
@i-chaochen
disable nccl 2gpus tests
997bcd9
@i-chaochen
disable device_tracer_test_gpu as it's flaky
5a98ecf
@i-chaochen
disable tensor_or_memref_test and interpreter_value_test
3570021
@i-chaochen
add rocsolvoler_warpper.h env.h
48cd8ae
@i-chaochen
keeep estimator_export in tf_export
6fc8030
@i-chaochen
Copy link
Author

retest Ubuntu-GPU-single please
retest Ubuntu-GPU-multi please
retest Ubuntu-CPU please
retest Ubuntu-sanity please

@jayfurmanek
Copy link

retest Ubuntu-sanity please

@i-chaochen
Copy link
Author

retest Ubuntu-GPU-single please

@i-chaochen i-chaochen mentioned this pull request Feb 3, 2023
Closed
@i-chaochen i-chaochen merged commit a61751c into develop-upstream Feb 3, 2023
@i-chaochen i-chaochen mentioned this pull request Mar 15, 2023
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@jayfurmanek jayfurmanek jayfurmanek approved these changes

@rahulbatra85 rahulbatra85 Awaiting requested review from rahulbatra85

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.