Skip to content

Check dim order in the optimized layer_norm as the portable one does #21866

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
SuryanshSS1011 wants to merge 1 commit into
base: main
Choose a base branch
from
+36 −0
Open

Check dim order in the optimized layer_norm as the portable one does #21866

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
10 changes: 10 additions & 0 deletions kernels/optimized/cpu/op_native_layer_norm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -149,6 +149,16 @@ std::tuple<Tensor&, Tensor&, Tensor&> opt_native_layer_norm_out(
InvalidArgument,
ret_val);

// Only support default dim order for now.
ET_KERNEL_CHECK(
ctx, tensor_is_default_dim_order(input), InvalidArgument, ret_val);

ET_KERNEL_CHECK(
ctx,
tensors_have_same_dim_order(input, out, mean_out, rstd_out),
InvalidArgument,
ret_val);

Tensor::SizesType mean_rstd_sizes[kTensorDimensionLimit];
size_t mean_rstd_ndim = 0;
get_layer_norm_out_target_size(
Expand Down
26 changes: 26 additions & 0 deletions kernels/test/op_native_layer_norm_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -452,3 +452,29 @@ TEST_F(OpNativeLayerNormTest, DynamicShapeUnbound) {
test_dynamic_shape(
{1, 1}, torch::executor::TensorShapeDynamism::DYNAMIC_UNBOUND);
}

TEST_F(OpNativeLayerNormTest, NonDefaultDimOrderDies) {
TensorFactory<ScalarType::Float> tf;

// mean and rstd share the input's rank with the normalized dims set to 1.
// All four are channels-last so the same-dim-order check passes and only the
// default dim order check can reject.
Tensor input = tf.channels_last_like(
tf.make({1, 3, 2, 2}, {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}));
Tensor out0 = tf.zeros_channels_last({1, 3, 2, 2});
Tensor out1 = tf.zeros_channels_last({1, 3, 2, 1});
Tensor out2 = tf.zeros_channels_last({1, 3, 2, 1});
const std::vector<int64_t> normalized_shape = {2};

ET_EXPECT_KERNEL_FAILURE(
context_,
op_native_layer_norm_out(
input,
IntArrayRef(normalized_shape.data(), normalized_shape.size()),
exec_aten::optional<Tensor>(),
exec_aten::optional<Tensor>(),
1e-5,
out0,
out1,
out2));
}
Loading