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

[inductor] Fallback on complex64 kernels (#98155)

Browse Source 
Later PRs in this stack fixe graph breaks in GoogleFnet which triggers errors from inductor trying to compile torch.complex64, this fixes that.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/98155
Approved by: https://github.com/anijain2305, https://github.com/ngimel
This commit is contained in:
Jason Ansel 2023-04-02 13:40:08 -07:00 committed by PyTorch MergeBot
parent bc9dd969e1
commit 71d850a100
2 changed files with 3 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
test/test_torch.py
View File

@ -2999,6 +2999,7 @@ else:
self.assertEqual(dst, src.neg().conj_physical(), exact_dtype=False)
# FIXME: move to data movement test suite
@skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/98175")
@onlyNativeDeviceTypes
@dtypes(torch.int64, torch.float32, torch.complex64)
def test_copy_transpose_math_view(self, device, dtype):

2
torch/_inductor/lowering.py
View File

@ -1041,6 +1041,8 @@ def fallback_handler(kernel, add_to_fallback_set=True):
def unsupported_output_tensor(t: torch._subclasses.FakeTensor):
if t.dtype in (torch.complex32, torch.complex64, torch.complex128):
return True
return t.is_cpu and config.disable_cpp_codegen