Skip to content

Fix a GroupNorm cuda bug when input does not require_grad #44863

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

Closed
xwang233 wants to merge 4 commits into from
Closed

Fix a GroupNorm cuda bug when input does not require_grad #44863

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
9278e24
fix
xwang233 Sep 17, 2020
98c02e4
Merge remote-tracking branch 'upstream/viable/strict' into group-norm…
xwang233 Sep 17, 2020
b06c3b8
update tolerance
xwang233 Sep 17, 2020
d8e061e
remove const
xwang233 Sep 17, 2020
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
3 changes: 2 additions & 1 deletion aten/src/ATen/native/cuda/group_norm_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -489,7 +489,7 @@ void GroupNormBackwardKernelImplInternal(
ComputeInternalGradientsCUDAKernel<T>
<<<N * C, cuda_utils::kCUDABlockReduceNumThreads, 0, cuda_stream>>>(
HxW, dY_data, X_data, ds_data, db_data);
if (dX != nullptr) {
if (dX_data != nullptr) {
Tensor c1 = at::empty({N, C}, X.options().dtype(kAccType));
Tensor c2 = at::empty({N, G}, X.options().dtype(kAccType));
Tensor c3 = at::empty({N, G}, X.options().dtype(kAccType));
Expand Down Expand Up @@ -525,6 +525,7 @@ void GroupNormBackwardKernelImplInternal(
<<<N * C, kCUDANumThreads, 0, cuda_stream>>>(
C, HxW, G, dY_data, X_data, c1_data, c2_data, c3_data, dX_data);
}
AT_CUDA_CHECK(cudaGetLastError());
}
if (dgamma->defined() || dbeta->defined()) {
T* dgamma_data = dgamma->defined() ? dgamma->data_ptr<T>() : nullptr;
Expand Down
19 changes: 14 additions & 5 deletions test/test_nn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9199,8 +9199,10 @@ def _test_GroupNorm_general(self, device, dtype=torch.float):
(2, 6, 4, 2, 2): 3,
(1, 256, 1, 1): 32,
}
for shape, g in good_shape_g.items():
for shape_g, grad in product(good_shape_g.items(), [True, False]):
shape, g = shape_g
x = torch.empty(*shape, device=device, dtype=dtype).uniform_(0, 10)
x.requires_grad_(grad)
b = shape[0]
c = shape[1]

Expand All @@ -9212,8 +9214,13 @@ def _test_GroupNorm_general(self, device, dtype=torch.float):
out_reshaped = output.view(b, g, -1)
mean = out_reshaped.mean(-1)
var = out_reshaped.var(-1, unbiased=False)
self.assertEqual(torch.abs(mean).mean(), 0, atol=1e-5, rtol=0)
self.assertEqual(torch.abs(var).mean(), 1, atol=1e-5, rtol=0)
# TODO: fix numerical issue. See #44863
self.assertEqual(torch.abs(mean).mean(), 0, atol=1e-3, rtol=1e-3)
self.assertEqual(torch.abs(var).mean(), 1, atol=1e-3, rtol=1e-3)

output.backward(torch.randn_like(output))
if output.is_cuda:
torch.cuda.synchronize()

# test that GN applies weight and bias correctly
scale = torch.empty(c, device=device, dtype=dtype).uniform_(0.2, 2)
Expand All @@ -9226,8 +9233,9 @@ def _test_GroupNorm_general(self, device, dtype=torch.float):
out_normed_reshaped = out_normed.view(b, g, -1)
mean = out_normed_reshaped.mean(-1)
var = out_normed_reshaped.var(-1, unbiased=False)
self.assertEqual(torch.abs(mean).mean(), 0, atol=1e-5, rtol=0)
self.assertEqual(torch.abs(var).mean(), 1, atol=1e-5, rtol=0)
# TODO: fix numerical issue. See #44863
self.assertEqual(torch.abs(mean).mean(), 0, atol=1e-3, rtol=1e-3)
self.assertEqual(torch.abs(var).mean(), 1, atol=1e-3, rtol=1e-3)

bad_shape_g = {
(1, 2, 3, 4): 3,
Expand Down Expand Up @@ -9568,6 +9576,7 @@ def test_LayerNorm_general(self, device):
if self.device_type == 'cuda':
self._test_LayerNorm_cuda_half(device)

@onlyOnCPUAndCUDA
def test_GroupNorm_general(self, device):
self._test_GroupNorm_general(device)

Expand Down