Skip to content

[webgpu] Fix the incorrect shapes when profiling #26928

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
qjia7 merged 2 commits into from
Jan 14, 2026
Merged

[webgpu] Fix the incorrect shapes when profiling #26928

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
b7d0dc1
[webgpu] Fix the incorrect shapes when profiling
qjia7 Jan 7, 2026
f7cbdb7
use override shape instead of original shape
qjia7 Jan 12, 2026
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
14 changes: 6 additions & 8 deletions onnxruntime/core/providers/webgpu/webgpu_context.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -628,17 +628,15 @@ void WebGpuContext::CollectProfilingData(profiling::Events& events) {

for (size_t i = 0; i < pending_kernels.size(); i++) {
const PendingKernelInfo& pending_kernel_info = pending_kernels[i];
const auto& inputs = pending_kernel_info.inputs;
const auto& outputs = pending_kernel_info.outputs;
const auto& input_shapes = pending_kernel_info.input_shapes;
const auto& output_shapes = pending_kernel_info.output_shapes;

SS(shapes, 128);
for (size_t s = 0; s < inputs.size(); s++) {
const auto& input = inputs[s];
shapes << "inputs[" << s << "] = " << input.override_shape.ToString() << " ";
for (size_t s = 0; s < input_shapes.size(); s++) {
shapes << "inputs[" << s << "] = " << input_shapes[s].ToString() << " ";
}
for (size_t s = 0; s < outputs.size(); s++) {
const auto& output = outputs[s];
shapes << "outputs[" << s << "] = " << output.override_shape.ToString() << " ";
for (size_t s = 0; s < output_shapes.size(); s++) {
shapes << "outputs[" << s << "] = " << output_shapes[s].ToString() << " ";
}

if (gpu_timestamp_offset_ == 0) {
Expand Down
16 changes: 13 additions & 3 deletions onnxruntime/core/providers/webgpu/webgpu_context.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -268,16 +268,26 @@ class WebGpuContext final {
std::string_view cache_key,
const std::vector<ProgramInput>& inputs,
const std::vector<ProgramOutput>& outputs)
: name{absl::StrJoin({kernel_name, kernel_type, program_name}, "&")}, cache_key{cache_key}, inputs{inputs}, outputs{outputs} {}
: name{absl::StrJoin({kernel_name, kernel_type, program_name}, "&")}, cache_key{cache_key} {
// Store shape information instead of tensor pointers to avoid accessing released tensors
input_shapes.reserve(inputs.size());
for (const auto& input : inputs) {
input_shapes.emplace_back(input.use_override_shape ? input.override_shape : input.tensor->Shape());
}
output_shapes.reserve(outputs.size());
for (const auto& output : outputs) {
output_shapes.emplace_back(output.use_override_shape ? output.override_shape : output.tensor->Shape());
}
}

PendingKernelInfo(PendingKernelInfo&&) = default;
PendingKernelInfo& operator=(PendingKernelInfo&&) = default;
ORT_DISALLOW_COPY_AND_ASSIGNMENT(PendingKernelInfo);

std::string name;
std::string cache_key;
std::vector<ProgramInput> inputs;
std::vector<ProgramOutput> outputs;
std::vector<TensorShape> input_shapes;
std::vector<TensorShape> output_shapes;
};

struct PendingQueryInfo {
Expand Down
Loading