[xpu][fix] Fix UT test_fuse_mix_order_reductions_combo_kernels (#170297)

Fixes #170296

Pull Request resolved: #170297
Approved by: https://github.com/EikanWang, https://github.com/jansel

[dynamo][DebugMode] make ModTracker a no-op in compiled regions (#170124

)

ModTracker causes a graph break in compiled regions: #169995, so this makes it a no-op by introducing the `torch._dynamo.eval_frame._is_in_compiled_region()` check.

The next PR in the stack makes nn.Module tracking work for DebugMode by introducing an interpreter.

Pull Request resolved: #170124
Approved by: https://github.com/tugsbayasgalan

[ROCm] Enable group gemm on gfx90a (#169356)

Fix concurrency race condition for group gemm and enable group gemm support on
gfx90a architecture.

Test command:
PYTORCH_TEST_WITH_ROCM=1 pytest test/test_matmul_cuda.py -v -k "test_grouped_gemm_2d_2d or test_grouped_gemm_2d_3d or or test_grouped_gemm_3d_3d or test_grouped_gemm_3d_2d"

Pull Request resolved: #169356
Approved by: https://github.com/slayton58, https://github.com/jeffdaily

Co-authored-by: Jeff Daily <[email protected]>

Fix: torch.view_as_complex() does not work on memory layout produced …

…by torch.contiguous() after transpose (#169780)

Fixes #150050, by ignoring the stride divisibility requirement for singleton dimensions (since stride is irrelevant for singleton dimensions).

Pull Request resolved: #169780
Approved by: https://github.com/soulitzer

[codemod][lowrisk] Remove unused exception parameter from caffe2 (#17…

…0325)

Summary:
`-Wunused-exception-parameter` has identified an unused exception parameter. This diff removes it.

This:
```
try {
    ...
} catch (exception& e) {
    // no use of e
}
```
should instead be written as
```
} catch (exception&) {
```

If the code compiles, this is safe to land.

Test Plan: Sandcastle

Reviewed By: dmm-fb

Differential Revision: D89071569

Pull Request resolved: #170325
Approved by: https://github.com/malfet

[18/N] Use Python 3.10 typing (#170280)

This PR uses Python 3.10 typing to some files.

Pull Request resolved: #170280
Approved by: https://github.com/Lucaskabela

[CI] Update update_expected.py to skip cuda-13 results (#170348)

Summary: Since cuda-13 runs skip more tests, we should only use cuda-12 runs to update the expected result files.

Pull Request resolved: #170348
Approved by: https://github.com/eellison

[Release 2.11] Version Bump (#170346)

Same as #162526
Compatibility matrix will be updated separately

Pull Request resolved: #170346
Approved by: https://github.com/huydhn

[ROCm] enable fastSpecializedAtomicAdd for gfx950 (#170330)

Use standard HIP headers for unsafeAtomicAdd. Cannot remove copy/paste of unsafeAtomicAdd as "preview" implementation until ROCm 6.2 can be fully deprecated.

Re-land of #167661

Co-author: @jeffdaily

Pull Request resolved: #170330
Approved by: https://github.com/jeffdaily

Don't call str when in redistribute hotpath (#170366)

Signed-off-by: Edward Z. Yang <[email protected]>
Pull Request resolved: #170366
Approved by: https://github.com/wconstab