From 1df6f6af8e0f23116e4a4a7d4859cd55da9e10ba Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Thu, 23 Oct 2025 12:54:13 -0500
Subject: [PATCH] Rearrange pointers to fix the reinterpret_cast issue (#3077)

[ROCm/composable_kernel commit: 6df69abeef7e1668fa09ec38b9124d00db4ac114]
---
 test/mx_mfma_op/mx_mfma_op.hpp | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/test/mx_mfma_op/mx_mfma_op.hpp b/test/mx_mfma_op/mx_mfma_op.hpp
index b2e615b9d8..47b4419379 100644
--- a/test/mx_mfma_op/mx_mfma_op.hpp
+++ b/test/mx_mfma_op/mx_mfma_op.hpp
@@ -789,12 +789,12 @@ struct store_C_col_major<CType, CFragT, 32, 32>
             CScalarFragT chunks[vectorSize(CFragT{}) / VW];
         } fragC{cFrag}; // Initialize with input fragment
 
-        *(reinterpret_cast<CScalarFragT*>(output + startOffset))                = fragC.chunks[0];
-        *(reinterpret_cast<CScalarFragT*>(output + startOffset + kMajorOffset)) = fragC.chunks[1];
-        *(reinterpret_cast<CScalarFragT*>(output + startOffset + 2 * kMajorOffset)) =
-            fragC.chunks[2];
-        *(reinterpret_cast<CScalarFragT*>(output + startOffset + 3 * kMajorOffset)) =
-            fragC.chunks[3];
+        CScalarFragT* fragPtr;
+        for(uint32_t idx = 0; idx < vectorSize(CFragT{}) / VW; ++idx)
+        {
+            fragPtr  = reinterpret_cast<CScalarFragT*>(output + startOffset + idx * kMajorOffset);
+            *fragPtr = fragC.chunks[idx];
+        }
     }
 };