mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-01-18 22:44:02 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

dnn(ocl4dnn): drop weights_buf

Browse Source 
- avoid memory access violation during "prefetch" stage
This commit is contained in:
Alexander Alekhin 2018-09-30 20:22:39 +00:00
parent 1059735bfb
commit 0f031b6680
1 changed files with 3 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
modules/dnn/src/opencl/conv_layer_spatial.cl
View File

@ -280,15 +280,6 @@ convolve_simd(
in_addr += INPUT_PITCH;
Dtype weight_buf[WEIGHT_PREF];
int w_idx=0;
for (int i = 0; i < WEIGHT_PREF; i++)
{
weight_buf[i] = weights[weight_addr];
weight_addr += SIMD_SIZE;
}
#define BLOCK_IN(n, c) intel_sub_group_shuffle(in_buf[n], (c))
int kr = 0; // kr = Kernel Row
@ -297,20 +288,18 @@ convolve_simd(
int kc = 0; // kc = Kernel Column
LOOP(KERNEL_WIDTH, kc,
{
Dtype weight_value = weights[weight_addr];
weight_addr += SIMD_SIZE;
for (int br=0; br < OUT_BLOCK_HEIGHT; br++)
{
for(int bc=0; bc < OUT_BLOCK_WIDTH; bc++)
{
Dtype input = BLOCK_IN((br * STRIDE_Y + kr * DILATION_Y), bc * STRIDE_X + kc * DILATION_X);
out[br * OUT_BLOCK_WIDTH + bc] = mad(weight_buf[w_idx % WEIGHT_PREF], input, out[br * OUT_BLOCK_WIDTH + bc]);
out[br * OUT_BLOCK_WIDTH + bc] = mad(weight_value, input, out[br * OUT_BLOCK_WIDTH + bc]);
}
}
weight_buf[w_idx % WEIGHT_PREF] = weights[weight_addr];
weight_addr += SIMD_SIZE;
++w_idx;
});
});
weight_addr -= WEIGHT_PREF * SIMD_SIZE;
}
fm = fm % ALIGNED_NUM_FILTERS;