Skip to content

port cuda specific case with hook #2184

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

port cuda specific case with hook #2184

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
df7ebb6
port cuda specific case with hook
daisyden Oct 16, 2025
6ab6446
replace hardcoded xpu with device variable
daisyden Oct 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 34 additions & 0 deletions test/xpu/test_nn_xpu.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@
from torch.testing._internal.common_device_type import (
dtypes,
instantiate_device_type_tests,
largeTensorTest,
)
from torch.testing._internal.common_dtype import get_all_math_dtypes, integral_types
from torch.testing._internal.common_utils import (
Expand Down Expand Up @@ -3786,6 +3787,39 @@ def test_cross_entropy_loss_2d_out_of_bounds_class_index(self):
)


@dtypes(torch.float, torch.half)
@largeTensorTest("20GB")
@largeTensorTest("64GB", "cpu")
def _test_warp_softmax_64bit_indexing(self, device, dtype):
def run_test(*shape):
x = torch.randn(shape, device="xpu", dtype=torch.float16, requires_grad=True)
Copy link

Copilot AI Oct 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The test receives a device argument (from the instantiated device-type tests and the second @largeTensorTest decorator specifying 'cpu'), but the tensor is hard-coded with device='xpu'. This prevents the CPU-annotated variant from actually exercising a CPU path and can cause mismatched expectations. Use the passed-in device variable for allocation: x = torch.randn(shape, device=device, dtype=torch.float16, requires_grad=True).

Suggested change
x = torch.randn(shape, device="xpu", dtype=torch.float16, requires_grad=True)
x = torch.randn(shape, device=device, dtype=torch.float16, requires_grad=True)

Copilot uses AI. Check for mistakes.
y = F.log_softmax(x, dim=-1, dtype=dtype)
y.backward(y)
with torch.no_grad():
xx = x.cpu().requires_grad_()
yy = F.log_softmax(xx.float(), dim=-1).to(dtype)
yy.backward(yy)
# workaround to reduce memory usage vs. self.assertEqual, see #84944
rtol, atol = torch.testing._comparison.get_tolerances(
dtype, rtol=None, atol=None
)
self.assertTrue(torch.allclose(y.cpu(), yy, rtol=rtol, atol=atol))
# x is half
rtol, _ = torch.testing._comparison.get_tolerances(
torch.half, rtol=None, atol=None
)
self.assertTrue(torch.allclose(x.grad.cpu(), xx.grad, rtol=rtol, atol=1e-3))

run_test(
1100000000, 2
) # Illegal memory access https://github.com/pytorch/pytorch/issues/52715
run_test(
2200000000, 1
) # invalid configuration argument https://github.com/pytorch/pytorch/issues/52716


TestNNDeviceType.test_warp_softmax_64bit_indexing = _test_warp_softmax_64bit_indexing

TestNNDeviceType.test_cross_entropy_loss_2d_out_of_bounds_class_index = (
_test_cross_entropy_loss_2d_out_of_bounds_class_index
)
Expand Down
Loading