From 8baefc7ae8ecca99e72e9a1f766824666bce7508 Mon Sep 17 00:00:00 2001
From: Tailing Yuan <yuantailing@gmail.com>
Date: Thu, 30 May 2024 02:01:28 +0800
Subject: [PATCH] Fixing copySharedMem in transpose.cu

---
 Samples/6_Performance/transpose/transpose.cu | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Samples/6_Performance/transpose/transpose.cu b/Samples/6_Performance/transpose/transpose.cu
index a8b6fe8d..8a034aa1 100644
--- a/Samples/6_Performance/transpose/transpose.cu
+++ b/Samples/6_Performance/transpose/transpose.cu
@@ -103,7 +103,7 @@ __global__ void copySharedMem(float *odata, float *idata, int width,
 
   for (int i = 0; i < TILE_DIM; i += BLOCK_ROWS) {
     if (xIndex < width && yIndex < height) {
-      tile[threadIdx.y][threadIdx.x] = idata[index];
+      tile[threadIdx.y + i][threadIdx.x] = idata[index + i * width];
     }
   }
 
@@ -111,7 +111,7 @@ __global__ void copySharedMem(float *odata, float *idata, int width,
 
   for (int i = 0; i < TILE_DIM; i += BLOCK_ROWS) {
     if (xIndex < height && yIndex < width) {
-      odata[index] = tile[threadIdx.y][threadIdx.x];
+      odata[index + i * width] = tile[threadIdx.y + i][threadIdx.x];
     }
   }
 }