[quant][mobile] Return for conv with empty batch (#37779)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37779
We should just return empty output
Test Plan: Imported from OSS
Differential Revision: D21385789
fbshipit-source-id: 4b42f5aaebabfa3f329ed74356bddb33daad98d5
diff --git a/aten/src/ATen/native/quantized/cpu/qnnpack/src/conv-run.cc b/aten/src/ATen/native/quantized/cpu/qnnpack/src/conv-run.cc
index 4b986e2..0e64e6b 100644
--- a/aten/src/ATen/native/quantized/cpu/qnnpack/src/conv-run.cc
+++ b/aten/src/ATen/native/quantized/cpu/qnnpack/src/conv-run.cc
@@ -293,6 +293,11 @@
const size_t dilation_height = conv_p.dilation[1];
const size_t groups = conv_p.groups;
+ if (batch_size == 0) {
+ // If no batches, return
+ return pytorch_qnnp_status_success;
+ }
+
const float convolution_scale =
input_scale * conv_p.kernel_scale / output_scale;
if (convolution_scale >= 1.0f) {
