[pull] master from tensorflow:master - #8860

Merged

pull[bot] merged 22 commits into

Cache-Cloud:masterfrom

tensorflow:master

Sep 30, 2026

Commits on Sep 25, 2026

Return 0 from the zero branch of the xlogy/xlog1py/xdivy GPU kernels
vishwakt
committed
2a43bef
Update the xlogy/xlog1py/xdivy GPU test baselines for +0
vishwakt
committed
d854010

Commits on Sep 30, 2026

Set intra op threadpool in ynn param when using NanoRt Executable.

SiqiaoWu1993
authored and
tensorflower-gardener
committed
f424878
Automated Code Change

cantonios
authored and
tensorflower-gardener
committed
059acf0
Protect XNNPACK runtime creation and destruction with workspace_mutex_.
tensorflower-gardener
committed
0a3b7c4
[xla:cpu] ir_compiler: remove obsolete comment about AllowFPOpFusion

cota
authored and
tensorflower-gardener
committed
4be465e
Modernize manual status checks, deprecated TF_* macros, and StrCat error builders

toli-y
authored and
tensorflower-gardener
committed
d064165
[XLA] Add PrecisionConfig algorithms for FP8x3 and FP8x4 BF16 dot emulation

SamGinzburg
authored and
tensorflower-gardener
committed
dc599b3
Use HeapSimulator to decide schedule-by-structure reruns

deqiangc
authored and
tensorflower-gardener
committed
e324dfe
[xla:cpu] sink same-block, contract, single-use fmul before fadd/fsub

cota
authored and
tensorflower-gardener
committed
e8fecb2
Previously, `buildMaxAndArgmaxBody` computed `selected_value` using `SelectOp` with the index tie-breaking condition rather than `MaxOp`. While `SelectOp` produces the same value for normal numbers…

pgmoka
authored and
tensorflower-gardener
committed
fdf964b
Fix data race in ThreadSafeStatus::status() by returning absl::Status by value so it is copied while the lock is held.

dmiltr3
authored and
tensorflower-gardener
committed
a599fd9
[XLA][SPMD] Optimize 2-concatenate for multi devices

wsmoses
authored and
tensorflower-gardener
committed
c53bc6e
Add llvm_xz dependency and BUILD definition.

mrguenther
authored and
tensorflower-gardener
committed
51e5cf1
Support broadcast-based grouped-query attention (GQA) during prefill in YNNPACK.

vksnk
authored and
tensorflower-gardener
committed
2091fe0
Automated Code Change
tensorflower-gardener
committed
276976b
Merge pull request #128034 from vishwakt:fix-xfunctor-gpu-kernels
tensorflower-gardener
committed
90d577c
Don't rewrite integer Pow(x, -1) to Reciprocal in Grappler, so integer Pow keeps raising InvalidArgumentError for negative exponents.

dmiltr3
authored and
tensorflower-gardener
committed
0fa5880
Automated Code Change
tensorflower-gardener
committed
da199e6
[Mosaic] Fix tpu.memref_squeeze verification and layout propagation for unusual tilings

tlongeri
authored and
tensorflower-gardener
committed
1154e44
Preserve metadata and frontend attributes in AllGatherDecomposer.

derdrdirk
authored and
tensorflower-gardener
committed
10141aa
Allow concurrent `workflow_dispatch` runs in `postsubmit_benchmark.yml`

Andrew-Dame
authored and
tensorflower-gardener
committed
103b097

Navigation Menu

[pull] master from tensorflow:master - #8860

[pull] master from tensorflow:master#8860
pull[bot] merged 22 commits into
Cache-Cloud:masterfrom
tensorflow:master

Commits on Sep 25, 2026

Return 0 from the zero branch of the xlogy/xlog1py/xdivy GPU kernels

Update the xlogy/xlog1py/xdivy GPU test baselines for +0

Commits on Sep 30, 2026

Set intra op threadpool in ynn param when using NanoRt Executable.

Automated Code Change

Protect XNNPACK runtime creation and destruction with workspace_mutex_.

[xla:cpu] ir_compiler: remove obsolete comment about AllowFPOpFusion

Modernize manual status checks, deprecated TF_* macros, and StrCat error builders

[XLA] Add PrecisionConfig algorithms for FP8x3 and FP8x4 BF16 dot emulation

Use HeapSimulator to decide schedule-by-structure reruns

[xla:cpu] sink same-block, contract, single-use fmul before fadd/fsub

Previously, `buildMaxAndArgmaxBody` computed `selected_value` using `SelectOp` with the index tie-breaking condition rather than `MaxOp`. While `SelectOp` produces the same value for normal numbers…

Fix data race in ThreadSafeStatus::status() by returning absl::Status by value so it is copied while the lock is held.

[XLA][SPMD] Optimize 2-concatenate for multi devices

Add llvm_xz dependency and BUILD definition.

Support broadcast-based grouped-query attention (GQA) during prefill in YNNPACK.

Automated Code Change

Merge pull request #128034 from vishwakt:fix-xfunctor-gpu-kernels

Don't rewrite integer Pow(x, -1) to Reciprocal in Grappler, so integer Pow keeps raising InvalidArgumentError for negative exponents.

Automated Code Change

[Mosaic] Fix tpu.memref_squeeze verification and layout propagation for unusual tilings

Preserve metadata and frontend attributes in AllGatherDecomposer.

Allow concurrent `workflow_dispatch` runs in `postsubmit_benchmark.yml`