Skip to content

Make the data types of output and input consistenst for batchnorm #84410

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
CaoE wants to merge 2 commits into from
Closed

Make the data types of output and input consistenst for batchnorm #84410

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2a1128d
make the data types of output and input consistenst for batchnorm
CaoE Sep 1, 2022
68f02d2
Apply suggestions from code review
malfet Oct 12, 2022
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: 1 addition & 2 deletions aten/src/ATen/native/Normalization.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,8 +142,7 @@ std::tuple<Tensor,Tensor,Tensor> batch_norm_cpu_transform_input_template(
.check_all_same_dtype(false)
.promote_inputs_to_common_dtype(false)
.build();

cpu_kernel(iter, [=](scalar_t input, param_t mean, param_t invstd, param_t weight, param_t bias) {
cpu_kernel(iter, [=](scalar_t input, param_t mean, param_t invstd, param_t weight, param_t bias) -> scalar_t {
return ((input - mean) * invstd) * weight + bias;
});
return std::make_tuple(output, save_mean, save_invstd);
Expand Down
9 changes: 9 additions & 0 deletions test/test_nn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8843,6 +8843,15 @@ def test_batchnorm_non_contig_cpu(self):
self.assertTrue(ref_out.is_contiguous())
self.assertEqual(out, ref_out)

input_bf = torch.arange(24, dtype=torch.bfloat16).reshape(1, 3, 2, 4)
input_bf = input_bf.permute(0, 2, 1, 3)
input_f = input_bf.float()
bn_mix = torch.nn.BatchNorm2d(2).float().eval()
ref_bn_f = deepcopy(bn_mix)
out_bf = bn_mix(input_bf)
ref_out_bf = ref_bn_f(input_f)
self.assertEqual(ref_out_bf, out_bf.float(), atol=0.05, rtol=0.05)

@unittest.skipIf(not TEST_CUDA, "CUDA unavailable")
@unittest.skipIf(not TEST_CUDNN, "needs cudnn")
def test_batchnorm_cudnn_nhwc(self):
Expand Down