Skip to content

webgpu: use naive reduction #28174

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
guschmue merged 5 commits into from
Jun 16, 2026
+15 −9
Merged

webgpu: use naive reduction #28174

Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
275b24e
webgpu: use naive reduction
Apr 22, 2026
aadeda2
Keep to use shared ReduceMean when axes are innermost
Apr 24, 2026
2fcb28f
Address jiajia's comment
Apr 30, 2026
af2b4c9
Merge branch 'main' into use-naive-reduction
Jun 2, 2026
fded7f3
Merge branch 'main' into use-naive-reduction
Jun 5, 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
25 changes: 16 additions & 9 deletions onnxruntime/core/providers/webgpu/reduction/reduction_ops.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -372,7 +372,22 @@ Status ReduceKernel<allow_multi_axes>::ComputeInternal(ComputeContext& context)
return Status::OK();
}

bool use_naive_reduction = name_ == "ArgMin" || name_ == "ArgMax" || (reduce_size < 32 && output_size > 1024) || is_input_empty || input_tensor->Shape().NumDimensions() == 0;
// Prefer the naive ReduceMean path when the shared path would pay extra overhead (transposing non-innermost reduce axes).
Comment thread
guschmue marked this conversation as resolved.
Outdated
constexpr size_t kReduceMeanNaiveMaxReduceSize = 128;
constexpr size_t kReduceMeanNaiveMinOutputSize = 20000;
bool are_axes_innermost = true;
size_t axes_rank = input_axes.size();
for (size_t i = 0; i < input_axes.size() && are_axes_innermost; ++i) {
if (input_axes[axes_rank - 1 - i] != rank - 1 - i) {
are_axes_innermost = false;
break;
}
}
bool use_naive_reduction = name_ == "ArgMin" || name_ == "ArgMax" || (reduce_size < 32 && output_size > 1024) ||
(name_ == "ReduceMean" && !are_axes_innermost &&
reduce_size <= kReduceMeanNaiveMaxReduceSize &&
output_size > kReduceMeanNaiveMinOutputSize) ||
is_input_empty || input_tensor->Shape().NumDimensions() == 0;

if (use_naive_reduction) {
ReduceNaiveProgram program(name_, reduce_op_type, keepdims_, noop_with_empty_axes_, input_axes, is_input_empty);
Expand All @@ -395,14 +410,6 @@ Status ReduceKernel<allow_multi_axes>::ComputeInternal(ComputeContext& context)

return context.RunProgram(program);
} else {
bool are_axes_innermost = true;
size_t axes_rank = input_axes.size();
for (size_t i = 0; i < input_axes.size() && are_axes_innermost; ++i) {
if (input_axes[axes_rank - 1 - i] != rank - 1 - i) {
are_axes_innermost = false;
break;
}
}
Tensor input_transpose;
if (!are_axes_innermost) {
InlinedVector<size_t> perm;
Expand Down
Loading