From 5a14aed5da2dcdaedb0c0f0c170a1de89c5f2c54 Mon Sep 17 00:00:00 2001
From: Dmitry Nikolaev <139769634+dnikolaev-amd@users.noreply.github.com>
Date: Fri, 25 Jul 2025 02:21:17 +0200
Subject: [PATCH] [rocm7.0_internal_testing] remove extra transposes in NHWC
 convolutions on MIOpen (#2405)

remove aten::contiguous for NHWC convolutions

Tests:
-
nn/test_convolution.py::TestConvolutionNNDeviceTypeCUDA::test_conv_cudnn_nhwc_cuda_float32
-
nn/test_convolution.py::TestConvolutionNNDeviceTypeCUDA::test_conv_cudnn_nhwc_cuda_float16

Before:
<img width="1255" height="228" alt="image"
src="https://github.com/user-attachments/assets/b125ccab-00c2-4d3a-a341-4583e51d8d57"
/>

After:
<img width="874" height="153" alt="image"
src="https://github.com/user-attachments/assets/ec200754-3622-488e-8762-bff1c2d22818"
/>


Fixes SWDEV-526887
---
 aten/src/ATen/native/miopen/Conv_miopen.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/aten/src/ATen/native/miopen/Conv_miopen.cpp b/aten/src/ATen/native/miopen/Conv_miopen.cpp
index f4e67e4fc307a..f9ac375c205ec 100644
--- a/aten/src/ATen/native/miopen/Conv_miopen.cpp
+++ b/aten/src/ATen/native/miopen/Conv_miopen.cpp
@@ -1200,7 +1200,7 @@ std::tuple<at::Tensor,at::Tensor,at::Tensor> miopen_convolution_transpose_backwa
     IntArrayRef padding, IntArrayRef output_padding, IntArrayRef stride, IntArrayRef dilation, int64_t groups,
     bool benchmark, bool deterministic, std::array<bool,3> output_mask) {
 
-  Tensor grad_output = grad_output_t.contiguous();
+  Tensor grad_output = grad_output_t.contiguous(input.suggest_memory_format());
 
   Tensor grad_input, grad_weight, grad_bias;
   if (output_mask[0]) {
@@ -1452,7 +1452,7 @@ std::tuple<at::Tensor,at::Tensor,at::Tensor> miopen_depthwise_convolution_backwa
     IntArrayRef padding, IntArrayRef stride, IntArrayRef dilation, int64_t groups,
     bool benchmark, bool deterministic, std::array<bool,3> output_mask) {
 
-  Tensor grad_output = grad_output_t.contiguous();
+  Tensor grad_output = grad_output_t.contiguous(input.suggest_memory_format());
 
   Tensor grad_input, grad_weight, grad_bias;
   if (output_mask[0]) {