Skip to content
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

[WebNN EP] Update WebNN normalization ops #18817

Merged
merged 6 commits into from
Jan 9, 2024
Merged

[WebNN EP] Update WebNN normalization ops #18817

Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
22c19ad
[WebNN EP] Update WebNN normalization ops
Dec 6, 2023
d89cecc
Resolve comment for Layernorm axis
Dec 19, 2023
d0e0d3a
Support 3D input for instanceNormalization
Dec 26, 2023
9566ca9
Support instanceNorm with input rank > 4
Jan 4, 2024
4912fc6
Resolve comments
Jan 4, 2024
5d0e4e0
Resolve comment
Jan 8, 2024
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
31 changes: 14 additions & 17 deletions onnxruntime/core/providers/webnn/builders/impl/normalization_op_builder.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,29 +93,26 @@ Status NormalizationOpBuilder::AddToModelBuilderImpl(ModelBuilder& model_builder
output = model_builder.GetBuilder().call<emscripten::val>("layerNormalization", input, options);
} else if (op_type == "InstanceNormalization") {
// WebNN spec only supports 4D input for instanceNormalization.
// Supports 3D input by prepanding 1 size dimension.
// Supports 3D input by prepending 1 size dimension.
// For models with dimensions greater than 4, they will be reshaped into 4D.
if (input_shape.size() != 4) {
constexpr size_t webnn_shape_rank = 4;
if (input_shape.size() != webnn_shape_rank) {
std::vector<uint32_t> new_shape;
fdwr marked this conversation as resolved.
Show resolved Hide resolved
new_shape.reserve(std::max(input_shape.size(), size_t(4)));
std::transform(input_shape.begin(), input_shape.end(),
std::back_inserter(new_shape),
[](int64_t dim) -> uint32_t { return SafeInt<uint32_t>(dim); });
if (model_builder.GetPreferredLayout() == DataLayout::NHWC) {
if (input_shape.size() == 3) {
new_shape.insert(new_shape.begin() + 2, 1);
} else {
uint32_t sum = std::accumulate(new_shape.begin() + 2, new_shape.end() - 1, 1, std::multiplies<uint32_t>());
new_shape.erase(new_shape.begin() + 2, new_shape.end() - 1);
new_shape.insert(new_shape.begin() + 2, sum);
}

size_t insertion_offset = (model_builder.GetPreferredLayout() == DataLayout::NHWC) ? 2 : 3;
ptrdiff_t excess_rank = new_shape.size() - webnn_shape_rank;
auto insertion_point = new_shape.begin() + insertion_offset;
if (input_shape.size() < 4) {
zesongw marked this conversation as resolved.
Show resolved Hide resolved
new_shape.insert(insertion_point, -excess_rank, 1);
} else {
if (input_shape.size() == 3) {
new_shape.emplace_back(1);
} else {
uint32_t sum = std::accumulate(new_shape.begin() + 3, new_shape.end(), 1, std::multiplies<uint32_t>());
new_shape.erase(new_shape.begin() + 3, new_shape.end());
new_shape.insert(new_shape.begin() + 3, sum);
}
uint32_t sum = std::accumulate(
insertion_point, insertion_point + excess_rank + 1, 1, std::multiplies<uint32_t>());
new_shape.erase(insertion_point, insertion_point + excess_rank);
*insertion_point = sum;
}
input = model_builder.GetBuilder().call<emscripten::val>("reshape", input, emscripten::val::array(new_shape));
}
Expand Down
Loading