Skip to content

Fix for AveragePool performance regression by guarding ceil_mode clamp #27331

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
tengli-alaska wants to merge 4 commits into
base: main
Choose a base branch
from
+24 −9
Open

Fix for AveragePool performance regression by guarding ceil_mode clamp #27331

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
10a383b
Guard AveragePool ceil_mode clamp to fix perf regression #27190
tengli-alaska Feb 12, 2026
2e66c83
Removed unrelated vscode settings change
tengli-alaska Feb 12, 2026
07a9512
Removed unrelated vscode settings
tengli-alaska Feb 12, 2026
401f384
Restore .vscode/settings.json
tengli-alaska Feb 14, 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
36 changes: 0 additions & 36 deletions .vscode/settings.json
View file
Open in desktop

This file was deleted.

6 changes: 3 additions & 3 deletions onnxruntime/core/providers/cpu/nn/pool.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -289,7 +289,7 @@ Status AveragePoolV19<T>::Compute(OpKernelContext* context) const {

RunLoop<AveragePool1DTask<T>>(tp, onnxruntime::narrow<size_t>(total_channels),
{X_data, Y_data, x_step, y_step, dilation_h, pooled_height, stride_h(),
height, kernel_shape, pads, pool_attrs_.count_include_pad, p_});
height, kernel_shape, pads, pool_attrs_.count_include_pad, p_, pool_attrs_.ceil_mode});
break;
}

Expand All @@ -301,7 +301,7 @@ Status AveragePoolV19<T>::Compute(OpKernelContext* context) const {
RunLoop<AveragePool2DTask<T>>(
tp, onnxruntime::narrow<size_t>(total_channels),
{X_data, Y_data, x_step, y_step, dilation_h, dilation_w, pooled_height, pooled_width, stride_h(),
stride_w(), height, width, kernel_shape, pads, pool_attrs_.count_include_pad, p_});
stride_w(), height, width, kernel_shape, pads, pool_attrs_.count_include_pad, p_, pool_attrs_.ceil_mode});
break;
}
case 3: {
Expand All @@ -314,7 +314,7 @@ Status AveragePoolV19<T>::Compute(OpKernelContext* context) const {
{X_data, Y_data, x_step, y_step,
dilation_h, dilation_w, dilation_d, pooled_height, pooled_width,
pooled_depth, stride_h(), stride_w(), stride_d(), height,
width, depth, kernel_shape, pads, pool_attrs_.count_include_pad, p_});
width, depth, kernel_shape, pads, pool_attrs_.count_include_pad, p_, pool_attrs_.ceil_mode});
break;
}
default:
Expand Down
27 changes: 21 additions & 6 deletions onnxruntime/core/providers/cpu/nn/pool_functors.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -390,6 +390,7 @@
gsl::span<const int64_t> pads;
bool count_include_pad;
int64_t p;
int64_t ceil_mode;
TensorOpCost Cost() {
double loop_count = static_cast<double>(pooled_height * kernel_shape[0]);
return TensorOpCost{loop_count, loop_count, loop_count};
Expand All @@ -406,7 +407,9 @@
for (int64_t ph = 0; ph < pooled_height; ++ph) {
int64_t hstart = ph * stride_h - pads[0];
int64_t hend = hstart + kernel_shape[0] * dilation_h;
hend = std::min(hend, height + pads[1]);
if (ceil_mode) {
hend = std::min(hend, height + pads[1]);
}
y_d[ph] = 0;
int total_elements = 0;
for (int64_t h = hstart; h < hend; h += dilation_h) {
Expand Down Expand Up @@ -444,6 +447,7 @@
gsl::span<const int64_t> pads;
bool count_include_pad;
int64_t p;
int64_t ceil_mode;

TensorOpCost Cost() {
double loop_count = static_cast<double>(pooled_height * pooled_width * kernel_shape[0] * kernel_shape[1]);
Expand All @@ -462,11 +466,15 @@
for (int64_t ph = 0; ph < pooled_height; ++ph) {
int64_t hstart = ph * stride_h - pads[0];
int64_t hend = hstart + kernel_shape[0] * dilation_h;
hend = std::min(hend, height + pads[1]);
if (ceil_mode) {
hend = std::min(hend, height + pads[1]);
}
for (int64_t pw = 0; pw < pooled_width; ++pw) {
int64_t wstart = pw * stride_w - pads[1];
int64_t wend = wstart + kernel_shape[1] * dilation_w;
wend = std::min(wend, width + pads[3]);
if (ceil_mode) {
wend = std::min(wend, width + pads[3]);
}
const int64_t pool_index = ph * pooled_width + pw;
y_d[pool_index] = 0;
int total_elements = 0;
Expand Down Expand Up @@ -515,6 +523,7 @@
gsl::span<const int64_t> pads;
bool count_include_pad;
int64_t p;
int64_t ceil_mode;

void operator()(std::ptrdiff_t begin, std::ptrdiff_t end) const {
for (std::ptrdiff_t c = begin; c < end; ++c) {
Expand All @@ -535,15 +544,21 @@
for (int64_t ph = 0; ph < pooled_height; ++ph) {
int64_t hstart = ph * stride_h - pads[0];
int64_t hend = hstart + kernel_shape[0] * dilation_h;
hend = std::min(hend, height + pads[1]);
if (ceil_mode) {
hend = std::min(hend, height + pads[1]);
}
for (int64_t pw = 0; pw < pooled_width; ++pw) {
int64_t wstart = pw * stride_w - pads[1];
int64_t wend = wstart + kernel_shape[1] * dilation_w;
wend = std::min(wend, width + pads[3]);
if (ceil_mode) {
wend = std::min(wend, width + pads[3]);
}
for (int64_t pd = 0; pd < pooled_depth; ++pd) {
int64_t dstart = pd * stride_d - pads[2];
int64_t dend = dstart + kernel_shape[2] * dilation_d;
dend = std::min(dend, depth + pads[5]);
if (ceil_mode) {
dend = std::min(dend, depth + pads[5]);

Check warning on line 560 in onnxruntime/core/providers/cpu/nn/pool_functors.h

View workflow job for this annotation

GitHub Actions / Optional Lint C++ 

[cpplint] reported by reviewdog 🐶
 Add #include <algorithm> for min  [build/include_what_you_use] [4]

Raw Output:
onnxruntime/core/providers/cpu/nn/pool_functors.h:560:  Add #include <algorithm> for min  [build/include_what_you_use] [4]
}
const int64_t pool_index = ph * pooled_width * pooled_depth + pw * pooled_depth + pd;
y_d[pool_index] = 0;
int total_elements = 0;
Expand Down
Loading