cv::transpose

modules/core/src/matrix.cpp

@@ -3002,8 +3002,8 @@ static bool ocl_transpose( InputArray _src, OutputArray _dst )
         k.args(ocl::KernelArg::ReadOnly(src),
                ocl::KernelArg::WriteOnlyNoSize(dst));
 
-    size_t localsize[3]  = { TILE_DIM, BLOCK_ROWS, 1 };
-    size_t globalsize[3] = { src.cols, inplace ? src.rows : divUp(src.rows, TILE_DIM) * BLOCK_ROWS, 1 };
+    size_t localsize[2]  = { TILE_DIM, BLOCK_ROWS };
+    size_t globalsize[2] = { src.cols, inplace ? src.rows : divUp(src.rows, TILE_DIM) * BLOCK_ROWS };
 
     return k.run(2, globalsize, localsize, false);
 }

modules/core/src/opencl/transpose.cl

@@ -53,7 +53,7 @@
 #define TSIZE ((int)sizeof(T1)*3)
 #endif
 
-#define LDS_STEP      TILE_DIM
+#define LDS_STEP      (TILE_DIM + 1)
 
 __kernel void transpose(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols,
                         __global uchar * dstptr, int dst_step, int dst_offset)
@@ -90,6 +90,7 @@ __kernel void transpose(__global const uchar * srcptr, int src_step, int src_off
     {
         int index_src = mad24(y, src_step, mad24(x, TSIZE, src_offset));
 
+        #pragma unroll
         for (int i = 0; i < TILE_DIM; i += BLOCK_ROWS)
             if (y + i < src_rows)
             {
@@ -103,6 +104,7 @@ __kernel void transpose(__global const uchar * srcptr, int src_step, int src_off
     {
         int index_dst = mad24(y_index, dst_step, mad24(x_index, TSIZE, dst_offset));
 
+        #pragma unroll
         for (int i = 0; i < TILE_DIM; i += BLOCK_ROWS)
             if ((y_index + i) < src_cols)
             {