[Pytorch] Disable use of qnnpack with ceil_mode avgpool (#79028)
Summary:
In pooling ops qnnpack does not handle ceil_mode. This is protected in maxpool
but not in avgpool. This diff adds that.
Test Plan: CI
Differential Revision: D36975661
Pull Request resolved: https://github.com/pytorch/pytorch/pull/79028
Approved by: https://github.com/dzdang
diff --git a/aten/src/ATen/native/quantized/cpu/AveragePool2d.cpp b/aten/src/ATen/native/quantized/cpu/AveragePool2d.cpp
index 45e1b2a..264707c 100644
--- a/aten/src/ATen/native/quantized/cpu/AveragePool2d.cpp
+++ b/aten/src/ATen/native/quantized/cpu/AveragePool2d.cpp
@@ -361,7 +361,7 @@
Tensor output;
#ifdef USE_PYTORCH_QNNPACK
if (at::globalContext().qEngine() == at::QEngine::QNNPACK &&
- input.scalar_type() == kQUInt8) {
+ input.scalar_type() == kQUInt8 && !ceil_mode) {
return at::native::qnnp_avgpool_helper::qnnpack_avg_pool2d(
input,
kernel_size,
