Add torch.backends.cuda.math_sdp.fp32_precision#2841

Closed

anatoliylitv wants to merge 133 commits intorocm7.2_internal_testingfrom

xinyazhang/math_sdp_ieee

Commits on Nov 24, 2025

[dynamo] Remove redundant _nonvar_fields assignments in UserDefinedObjectVariable subclasses (#167801)

hinriksnaer
authored and
pytorchmergebot
committed
Revert "[DTensor] compute shape and offset for arbitrary _StridedShard (#168146)"
pytorchmergebot
committed
[dynamo] Skip optree tests when optree isn't installed (#168931)

jansel
authored and
pytorchmergebot
committed
[dynamo] Fix local test failures for test/dynamo/test_after_aot.py (#168914)

jansel
authored and
pytorchmergebot
committed
[dynamo] Fix local test failures for test_compiler_bisector.py (#168915)

jansel
authored and
pytorchmergebot
committed
[dynamo] Fix more cases of tests leaking config changes (#168924)

jansel
authored and
pytorchmergebot
committed
[dynamo] Isolate test/dynamo/test_aot_compile.py tests in subproc (#168925)

jansel
authored and
pytorchmergebot
committed
[dynamo] Make test custom op naming unique (#168926)

jansel
authored and
pytorchmergebot
committed
Revert "Checking if the input is finite before calculation in lowering of pow func (#167723)"
pytorchmergebot
committed
Revert "Move CUDAEvent to c10 (#158219)"
pytorchmergebot
committed
Fix local_map default partitioner issue (#168396)

soulitzer
authored and
pytorchmergebot
committed
[CI][CUDA][B200][Smoke Test] Add one more B200 smoke test. Change periodic frequency from every 6 hours to every 2 hours (#168990)

nWEIdia
authored and
pytorchmergebot
committed
[AOTI] Skip emit_multi_arch_kernel when CUDA version is lower than 12.8 (#168985)

desertfire
authored and
pytorchmergebot
committed
[CD] [aarch64] unify the build.sh to build for aarch64 wheel (#166044)

tinglvv
authored and
pytorchmergebot
committed
Add human-readable type name comment to TYPE_MATCH guards for debugging clarity (#168272)

manigkrish
authored and
pytorchmergebot
committed
[3.14] Fix nn.Module annotations lookup (#168325)

rtimpe
authored and
pytorchmergebot
committed
[precompile] set strict_autograd_cache=True when precompiling (#168989)

bobrenjc93
authored and
pytorchmergebot
committed
Refactor call_function (#168932)

aorenste
authored and
pytorchmergebot
committed
Revert "[DebugMode] wait before hashing collectives by default (#168119)"
pytorchmergebot
committed
[BE] Delete Pytorch-circleci-labels (#169003)

malfet
authored and
pytorchmergebot
committed
[inductor] fix picking wrong contiguous node (#168371)

shunting314
authored and
pytorchmergebot
committed
[simplefsdp] fix dsv3 estimation (#168199)

ruisizhang123
authored and
pytorchmergebot
committed
[hoo] Invoke subgraph + effect (#167231)

angelayi
authored and
pytorchmergebot
committed
[invoke_subgraph] Don't run the graph twice when autograd enabled (#167245)

angelayi
authored and
pytorchmergebot
committed
[hoo] Fix unlift of effects with invoke_subgraph (#167363)

angelayi
authored and
pytorchmergebot
committed
[3/N] Use context managers (#167788)

cyyever
authored and
pytorchmergebot
committed
inductor: fix failure in test_flex_decoding on <sm80 (#165404)

jsmolic
authored and
pytorchmergebot
committed

Commits on Nov 25, 2025

Fix take_along_dim negative index handling (#146211) (#152161)

KaaustaaubShankar
authored and
pytorchmergebot
committed
Adding check for step size=0 in unfold backward to avoid divide by 0 … (#162720)

mansiag05
authored and
pytorchmergebot
committed
[6/N] Use key in dict for existence checks (#168350)

cyyever
authored and
pytorchmergebot
committed
[precompile] move strict_autograd_cache=True decorator to aot_compile_fullgraph (#169008)

bobrenjc93
authored and
pytorchmergebot
committed
[xpu][fix] Refine memory pool logic when expandable segement enabled (#168956)

guangyey
authored and
pytorchmergebot
committed
Remove useless parent method delegation (#168355)

cyyever
authored and
pytorchmergebot
committed
[Inductor][Quant]Support qconv_pointwise.tensor and qconv2d_pointwise.binary_tensor (#166608)

jiayisunx
authored and
pytorchmergebot
committed
[1/N] Use TYPE_CHECKING (#165852)

cyyever
authored and
pytorchmergebot
committed
[CPU] add onednn context cache for qlinear to improve performance (#168150)

Xia-Weiwen
authored and
pytorchmergebot
committed

Commits on Nov 28, 2025

Remove cudaProfilerInitialize (#168918)

cyyever
authored and
pytorchmergebot
committed
[2/N] Remove unused header inclusion (#165831)

cyyever
authored and
pytorchmergebot
committed
Remove unused inplace loop in test_conv2d_clamp of test_jit_llga_fuser.py‎ (#166691)

cyyever
authored and
pytorchmergebot
committed
[Quant][CPU] fix fp8 qconv (#167611)

jiayisunx
authored and
pytorchmergebot
committed
[dynamo] Support functools.partial as dict key (#169016)

anijain2305
authored and
pytorchmergebot
committed
Remove unused thrust inclusion (#169051)

cyyever
authored and
pytorchmergebot
committed
[CUDA][BugFix] fix truncated error messages (#168942)

lingebeng
authored and
pytorchmergebot
committed
[8/N] Use Python 3.10 typing (#168334)

cyyever
authored and
pytorchmergebot
committed
Fix an unsafe indexing in fx exception handling (#169140)

gmagogsfm
authored and
pytorchmergebot
committed
Enable SIM118 (#167399)

cyyever
authored and
pytorchmergebot
committed
Unset LD_LIBRARY_PATH for binary wheel checks (#168349)

robert-hardwick
authored and
pytorchmergebot
committed
[Code Clean] Better error handling in aten/src/ATen/native/* and aten/src/ATen/mkl/Exceptions.h (#165290)

zhudada0120
authored and
pytorchmergebot
committed
[ROCm][CI] Use hash for download-artifact action instead of tag (#169217)

jithunnair-amd
authored and
pytorchmergebot
committed
[cuDNN] Leak `BenchmarkCache` to avoid teardown segfault, correct compile-time cuDNN version to 9.10.2 in CI (#169153)

eqy
authored and
pytorchmergebot
committed
Revert "Fix local_map default partitioner issue (#168396)"
pytorchmergebot
committed

Commits on Nov 29, 2025

[pytree][dynamo] make `tree_*` functions accept both Python and C++ `PyTreeSpec` (#152624)

XuehaiPan
authored and
pytorchmergebot
committed
[BE][4/5] fix typos in aten/ (aten/src/ATen/native/) (#157553)

XuehaiPan
authored and
pytorchmergebot
committed
[BE][5/5] fix typos in aten/ (aten/src/ATen/) (#157554)

XuehaiPan
authored and
pytorchmergebot
committed
[BE][1/6] fix typos in test/ (#157635)

XuehaiPan
authored and
pytorchmergebot
committed
[9/N] Use Python 3.10 typing (#167806)

cyyever
authored and
pytorchmergebot
committed
[dynamo, 3.14] Ensure `typing.Union` is correctly traced in Dynamo. (#169084)

guilhermeleobas
authored and
pytorchmergebot
committed
[dynamo] Skip Dynamo wrapped tests on cpython tests (#169233)

anijain2305
authored and
pytorchmergebot
committed
[Durin] Bump Kineto Submodule to latest (#169098)

firoz1905
authored and
pytorchmergebot
committed
[openreg] Expand autocast test coverage for custom device (#169029)

authored and
pytorchmergebot
committed

Commits on Nov 30, 2025

Commits on Dec 1, 2025

[3.14] Skip broken numpy test (#169030)

rtimpe
authored and
pytorchmergebot
committed
[3.14] Fix dynamo error on np.broadcast_shapes (#168888)

rtimpe
authored and
pytorchmergebot
committed
Avoid std::tie and returning value constructions in qconv_unpack.cpp (#169207)

cyyever
authored and
pytorchmergebot
committed
Remove unused TVM code in dynamo (#169247)

cyyever
authored and
pytorchmergebot
committed
[xpu][feature] enable Sycl CPP extension on Windows (#162579)

tbohutyn
authored and
pytorchmergebot
committed
[Accelerator] Add Accelerator Capabilities API (#165631)

authored and
pytorchmergebot
committed
add device generalization support for distributed tests (#165067)

harikodali
authored and
pytorchmergebot
committed
[user-streams] Handle the record_stream problem (#168061)

mlazos
authored and
pytorchmergebot
committed
Revert "[dynamo][dicts] Decentralize and Improve key hash implementation for Dict variable tracker (#169204)"
pytorchmergebot
committed
Revert "[Accelerator] Add Accelerator Capabilities API (#165631)"
pytorchmergebot
committed
[inductor] Add option to transpose tensor descriptors on load / store (#165541)

kundaMwiza
authored and
pytorchmergebot
committed
Remove the CUPTI CMake check for kineto (#161370)

cyyever
authored and
pytorchmergebot
committed
[xpu][fix] Support xpu custom raw_alloc/delete in caching allocator (#168957)

guangyey
authored and
pytorchmergebot
committed
Replace vscode recommendation for type checker (#169021)

lolpack
authored and
pytorchmergebot
committed
[reland][inductor] fix the decision of inner reduction (#168391)

shunting314
authored and
pytorchmergebot
committed
bump transformer pin to 4.57.3 (#169226)

BoyuanFeng
authored and
pytorchmergebot
committed
[mergebot] Set header for workflow when calling Dr.CI (#169037)

huydhn
authored and
pytorchmergebot
committed
bucketing compile time improve (#168122)

eellison
authored and
pytorchmergebot
committed
Track overlap per-pg (#169019)

eellison
authored and
pytorchmergebot
committed
[BE] Add missing method docstrings for pytorch quantization classes (#165199)

jiannanWang
authored and
pytorchmergebot
committed
[BE][MPS] Add out-of-bounds checks for embedding_bag (#168930)

malfet
authored and
pytorchmergebot
committed
Revert "Avoid std::tie and returning value constructions in qconv_unpack.cpp (#169207)"
pytorchmergebot
committed

Commits on Dec 2, 2025