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 20240513 #2543

Merged
merged 1,259 commits into from
May 20, 2024
Merged

Develop upstream sync 20240513 #2543

merged 1,259 commits into from
May 20, 2024

Conversation

i-chaochen
Copy link

@i-chaochen i-chaochen commented May 13, 2024 •
edited
Loading

  1. update rocm_solvers.cc [ROCm] fixed rocm_solvers build tensorflow/tensorflow#68249
  2. two CPU unit tests due to python version e0aef0f
  3. ClipByValue on DT_INT64 https://github.com/ROCm/frameworks-internal/issues/8064

tensorflower-gardener and others added 30 commits May 8, 2024 11:22
@tensorflower-gardener
Reverts 604adec
4382439 
PiperOrigin-RevId: 631863025
@tensorflower-gardener
Update ops-related pbtxt files.
7a49df9 
PiperOrigin-RevId: 631864481
@deqiangc @tensorflower-gardener
Fix asan error due to additional error logging
eb93fb5 
PiperOrigin-RevId: 631870090
@LukeBoyer @tensorflower-gardener
Add pattern match and composite outlining for GELU.
9589a74 
PiperOrigin-RevId: 631871121
@tensorflower-gardener
Make sure all TfLiteOperator's fields are initialized at construction
3926a51 
Field inplace_operator wasn't being initialized, most likely an omission.

PiperOrigin-RevId: 631875486
@klucke @tensorflower-gardener
Fix "moving a temporary object prevents copy elision [-Werror,-Wpessi…
bc89f9b 
…mizing-move]" error when compiling gpu_fused_mha_test with clang.

PiperOrigin-RevId: 631876424
@d0k @tensorflower-gardener
Integrate LLVM at llvm/llvm-project@3a8316216807
8190442 
Updates LLVM usage to match
[3a8316216807](llvm/llvm-project@3a8316216807)

PiperOrigin-RevId: 631878634
@yueshengys @tensorflower-gardener
Some preparation changes for enabling async dispatch on JAX CPU.
f991995 
To prevent too much parallelism for non-parallel computations, we add a enqueue event to make sure next computation won't be enqueued until last one is done per user thread.

In `~PyArray_Storage()`, we now release the Python GIL then destroy the underlying buffer to prevent deadlock caused by interactions between argument donations and host callbacks on CPU backend.

PiperOrigin-RevId: 631887633
@tensorflower-gardener
Internal change, only affects BUILD files.
c148954 
PiperOrigin-RevId: 631894583
@rocketas @tensorflower-gardener
Fix stale comment on CreateSplitIntoIslandPerOpPass.
d7fcb9d 
PiperOrigin-RevId: 631897585
@tensorflower-gardener
Update ops-related pbtxt files.
064e8eb 
PiperOrigin-RevId: 631903263
@hawkinsp @tensorflower-gardener
[PJRT] Remove method alias PjRtClient::GetFullTopologyForCompilation.
e494302 
PiperOrigin-RevId: 631915115
@vamsimanchala @tensorflower-gardener
Reverts 4382439
e8960ea 
PiperOrigin-RevId: 631919121
@tensorflower-gardener
Add missing absl random bit_gen_ref target
f5bd515 
PiperOrigin-RevId: 631920259
@Tongfei-Guo @tensorflower-gardener
[XLA:SPMD] Use the default way to call once and shard barrier custom-…
69b50ff 
…call registration instead of always linking it.

PiperOrigin-RevId: 631928144
@jimlinntu @tensorflower-gardener
#tf-data Restores the seed_generator_ and three seeds in global_shu…
b0def2a 
…ffle_dataset_op.cc to ensure the iterator can generate the same result after restoring

PiperOrigin-RevId: 631932883
@ezhulenev @tensorflower-gardener
[xla] NFC: Add tests for async custom call buffer assignment
3be6dbf 
PiperOrigin-RevId: 631935435
@klucke @tensorflower-gardener
Unify StreamExecutorInterface::GetSymbol and StreamExecutor::GetUntyp…
f551e4a 
…edSymbol.

PiperOrigin-RevId: 631936160
@tensorflower-gardener
Update ops-related pbtxt files.
12fd812 
PiperOrigin-RevId: 631940651
@hertschuh @tensorflower-gardener
Configure autograph to not convert tf_keras code.
79ae121 
Fixes keras-team/tf-keras#777

PiperOrigin-RevId: 631941146
@tensorflower-gardener
Extends the TFLite Benchmarker to support multiple signatures.
50ef93f 
PiperOrigin-RevId: 631947318
@wilsingosti @tensorflower-gardener
No public description
ff862d8 
PiperOrigin-RevId: 631952018
@pschuh @tensorflower-gardener
For nested full-to-shard we must also exclude SPMDShardToFullShape
c656774 
because the hlo->sharding() and operand->sharding() will have conflicting
values for IsManualSubgroup().

PiperOrigin-RevId: 631952777
@rocketas @tensorflower-gardener
Remove mlir roundtrip.
1a6d038 
PiperOrigin-RevId: 631959980
@tensorflower-gardener
Update flatbuffers to 24.3.25
c17d64d 
PiperOrigin-RevId: 631967585
@ezhulenev @tensorflower-gardener
[xla] NFC: Add tests for async custom call host offloader memory spac…
0f73d4c 
…e propagation

PiperOrigin-RevId: 631967931
@tensorflower-gardener
Introduce std::complex<T> as a new DataType
2ee6025 
PiperOrigin-RevId: 631968244
@tensorflower-gardener
Update ops-related pbtxt files.
66abacb 
PiperOrigin-RevId: 631973755
@anshumang @tensorflower-gardener
Fixes heap use after free in passing a python list() as an `absl:…
ed12cfa 
…:Span` by passing as a `std::vector` instead

PiperOrigin-RevId: 631977962
@tensorflower-gardener
Extend internal Span with more member functions
b337e59 
PiperOrigin-RevId: 631979198
tensorflower-gardener and others added 11 commits May 12, 2024 23:23
@tensorflower-gardener
Update ops-related pbtxt files.
27222f1 
PiperOrigin-RevId: 633100257
@jreiffers @tensorflower-gardener
Support nested tuples.
4f9ebb3 
I thought I could get away with not doing this, but I finally
gave up.

PiperOrigin-RevId: 633120834
@jreiffers @tensorflower-gardener
Add a pattern for removing results of index_switch ops without users.
591095e 
This just matches on index switches with only unused results. Ideally
this would become a canonicalization pattern instead.

index_switch will be used in the reduction group lowering.

PiperOrigin-RevId: 633121827
@tensorflower-gardener
Update ops-related pbtxt files.
4fd667c 
PiperOrigin-RevId: 633125572
@olupton @tensorflower-gardener
PR tensorflow#12253: Do not dump autotuner PTX to stdout when --xla_d…
acc7b7f 
…ump_hlo_as_* are set

Imported from GitHub PR openxla/xla#12253

Without this change then when `--xla_dump_hlo_as_proto=true --xla_dump_to=/some/real/path` is set, the autotuner dumps PTX to stdout.
Copybara import of the project:

--
9149682143373bd17d4dc57635ba34b146e3b358 by Olli Lupton <[email protected]>:

Do not dump autotuning PTX to stdout when --xla_dump_hlo_as_* are set

Merging this change closes tensorflow#12253

PiperOrigin-RevId: 633126913
@akuegel @tensorflower-gardener
Use atomic add for BF16 types on cuda devices which support it.
d3a235f 
This should make Scatter with BF16 type faster.

PiperOrigin-RevId: 633131419
@i-chaochen
init merge
49e847a
@i-chaochen
resolve merge conflict
4ac7a50
@i-chaochen
fixed rocm_solvers due to upstream rocm_executor refactoring
7f4f96d
@zoranjovanovic-ns @i-chaochen
Fixed issue with semver test.
e0aef0f
@i-chaochen
skip ClipByValue on DT_INT64 due to mismatch
be1cc56
@i-chaochen
Copy link
Author

retest Ubuntu-GPU-single please
retest Ubuntu-GPU-multi please

@i-chaochen i-chaochen force-pushed the develop-upstream-sync-20240513 branch from 965d9e6 to bbdb15b Compare May 18, 2024 22:29
@i-chaochen i-chaochen force-pushed the develop-upstream-sync-20240513 branch from bbdb15b to 27d0e2e Compare May 18, 2024 23:20
i-chaochen
i-chaochen commented May 19, 2024
View reviewed changes
third_party/xla/xla/service/gpu/tests/gemm_rewrite_test.cc
bool IsRocm() {
return std::holds_alternative<se::RocmComputeCapability>(Capability());
}

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is unnessary function and we need to remove this in next weekly-sync

@i-chaochen
Copy link
Author

retest Ubuntu-GPU-multi please

2 similar comments
@i-chaochen
Copy link
Author

retest Ubuntu-GPU-multi please

@i-chaochen
Copy link
Author

retest Ubuntu-GPU-multi please

@i-chaochen i-chaochen merged commit f802114 into develop-upstream May 20, 2024
10 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@hsharsha hsharsha hsharsha approved these changes

@draganmladjenovic draganmladjenovic Awaiting requested review from draganmladjenovic

@jayfurmanek jayfurmanek Awaiting requested review from jayfurmanek

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

Successfully merging this pull request may close these issues.