Skip to content

[WebNN] Support AveragePool with count_include_pad == 1 #24465

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

Merged
merged 1 commit into from
Apr 20, 2025
Merged

[WebNN] Support AveragePool with count_include_pad == 1 #24465

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
2 changes: 1 addition & 1 deletion js/web/docs/webnn-operators.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ platforms. Check the [WebNN status](https://webmachinelearning.github.io/webnn-s
| And | ai.onnx(7+) | logicalAnd | |
| ArgMax | ai.onnx(7-10, 11, 12, 13+) | argMax | |
| ArgMin | ai.onnx(7-10, 11, 12, 13+) | argMin | |
| AveragePool | ai.onnx(7-9, 10, 11, 12-18, 19+) | averagePool2d | Only supports 4-D input, 2-D 'kernel_shape', 'count_include_pad' value is 0 |
| AveragePool | ai.onnx(7-9, 10, 11, 12-18, 19+) | averagePool2d | Only supports 4-D input, 2-D 'kernel_shape' |
| BatchNormalization | ai.onnx(7-8, 9-13, 14, 15+) | batchNormalization | Only supports 'training_mode' value is 0, one output |
| Cast | ai.onnx(7-8, 9-12, 13-18, 19-20, 21+) | cast | |
| Ceil | ai.onnx(7-12, 13+) | ceil | |
Expand Down
34 changes: 22 additions & 12 deletions onnxruntime/core/providers/webnn/builders/impl/pool_op_builder.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,16 +60,17 @@ Status PoolOpBuilder::AddToModelBuilderImpl(ModelBuilder& model_builder,
emscripten::val options = emscripten::val::object();
options.set("label", node.Name());
NodeAttrHelper helper(node);
const bool is_nhwc = model_builder.GetPreferredLayout() == DataLayout::NHWC;

const auto kernel_shape = helper.Get("kernel_shape", std::vector<int32_t>{0, 0});
const auto onnx_kernel_shape = helper.Get("kernel_shape", std::vector<int32_t>{0, 0});
if (!is_global_pooling) {
options.set("windowDimensions", emscripten::val::array(kernel_shape));
options.set("windowDimensions", emscripten::val::array(onnx_kernel_shape));
}
const auto strides = helper.Get("strides", std::vector<int32_t>{1, 1});
options.set("strides", emscripten::val::array(strides));
const auto dilations = helper.Get("dilations", std::vector<int32_t>{1, 1});
options.set("dilations", emscripten::val::array(dilations));
if (model_builder.GetPreferredLayout() == DataLayout::NHWC) {
if (is_nhwc) {
options.set("layout", emscripten::val("nhwc"));
} else {
options.set("layout", emscripten::val("nchw"));
Expand All @@ -78,7 +79,6 @@ Status PoolOpBuilder::AddToModelBuilderImpl(ModelBuilder& model_builder,
// Add Padding.
// Usually using autopadding is more efficient than using explicit padding.
// Try to see if we can map explicit padding to auto padding.
const auto onnx_kernel_shape = helper.Get("kernel_shape", std::vector<int64_t>{0, 0});
const auto onnx_strides = helper.Get("strides", std::vector<int64_t>{1, 1});
const auto onnx_pads = helper.Get("pads", std::vector<int64_t>{0, 0, 0, 0});
auto pads = helper.Get("pads", std::vector<uint32_t>{0, 0, 0, 0});
Expand All @@ -93,7 +93,7 @@ Status PoolOpBuilder::AddToModelBuilderImpl(ModelBuilder& model_builder,
helper.Get("dilations", std::vector<int64_t>{1, 1}),
auto_pad_type,
pads_out,
model_builder.GetPreferredLayout() == DataLayout::NCHW));
!is_nhwc));
pads = GetNarrowedIntfromInt64<uint32_t>(pads_out);
}
// Permute the ONNX's pads, which is [beginning_height, beginning_width, ending_height, ending_width],
Expand All @@ -105,6 +105,23 @@ Status PoolOpBuilder::AddToModelBuilderImpl(ModelBuilder& model_builder,
options.set("roundingType", ceil_mode == 0 ? emscripten::val("floor")
: emscripten::val("ceil"));

// WebNN doesn't support AveragePool with count_include_pad == 1, emulate it by pad + averagePool2d.
if (op_type == "AveragePool" && helper.Get("count_include_pad", 0) == 1) {
std::vector<uint32_t> beginning_padding{0, 0, pads[0], pads[1]};
std::vector<uint32_t> ending_padding{0, 0, pads[2], pads[3]};
// Unset padding option, because we will use pad op instead.
options.set("padding", emscripten::val::array(std::vector<uint32_t>{0, 0, 0, 0}));
if (is_nhwc) {
beginning_padding = {0, pads[0], pads[1], 0};
ending_padding = {0, pads[2], pads[3], 0};
}

emscripten::val pad_options = emscripten::val::object();
pad_options.set("label", node.Name() + "_pad");
input = model_builder.GetBuilder().call<emscripten::val>("pad", input, emscripten::val::array(beginning_padding),
emscripten::val::array(ending_padding), pad_options);
}

emscripten::val output = model_builder.GetBuilder().call<emscripten::val>(webnn_op_name.c_str(), input, options);
model_builder.AddOperand(node.OutputDefs()[0]->Name(), std::move(output));
return Status::OK();
Expand Down Expand Up @@ -138,13 +155,6 @@ bool PoolOpBuilder::IsOpSupportedImpl(const GraphViewer&,
}
}

if (op_type == "AveragePool") {
if (helper.Get("count_include_pad", 0) != 0) {
LOGS(logger, VERBOSE) << "AveragePool only supports count_include_pad == 0";
return false;
}
}

if (op_type == "MaxPool") {
if (helper.Get("storage_order", 0) == 1) {
LOGS(logger, VERBOSE) << "MaxPool storage_order == 1 is not supported";
Expand Down
Loading