OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[EZ][MPS] Improve distribution error checking (#166425)

Browse Source 
Essentially not allow ops on self-overlapping outputs, by adding
`at::assert_no_internal_overlap(self);` check that already used in CPU
and CUDA builds, see
895795f07c/aten/src/ATen/native/DistributionTemplates.h (L366)

This fixes `test_error_inputs_bernoulli_mps`

Should be landed ahead of https://github.com/pytorch/pytorch/pull/165267
Pull Request resolved: https://github.com/pytorch/pytorch/pull/166425
Approved by: https://github.com/Skylion007, https://github.com/seemethere
This commit is contained in:
Nikita Shulga 2025-10-28 10:44:31 -07:00 committed by PyTorch MergeBot
parent 687c15c0b3
commit 1abfa5f70b
2 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
aten/src/ATen/native/mps/operations/Distributions.mm
View File

@ -57,6 +57,7 @@ Tensor& random_mps_impl(Tensor& self,
if (self.numel() == 0) { if (self.numel() == 0) {
return self; return self;
} }
at::assert_no_internal_overlap(self);
// MPS random is broken for 5D+ tensors, see https://github.com/pytorch/pytorch/issues/147624 // MPS random is broken for 5D+ tensors, see https://github.com/pytorch/pytorch/issues/147624
const auto need_reshape = self.ndimension() > 4; const auto need_reshape = self.ndimension() > 4;
auto mps_gen = get_generator_or_default<MPSGeneratorImpl>(gen, at::mps::detail::getDefaultMPSGenerator()); auto mps_gen = get_generator_or_default<MPSGeneratorImpl>(gen, at::mps::detail::getDefaultMPSGenerator());

1
torch/testing/_internal/common_mps.py
View File

@ -812,7 +812,6 @@ if torch.backends.mps.is_available():
"__rmod__", "__rmod__",
"__rsub__", "__rsub__",
"__rpow__", "__rpow__",
"bernoulli",
"clamp_max", "clamp_max",
"clamp_min", "clamp_min",
"masked_scatter", "masked_scatter",