mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-08-06 14:36:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

video(DIS): OpenCL workaround for AMDGPU

Browse Source
This commit is contained in:
Alexander Alekhin 2019-05-03 18:57:05 +00:00
parent 0e1dd63f5e
commit 04caf0549d
1 changed files with 6 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
modules/video/src/opencl/dis_flow.cl
View File

@ -316,16 +316,18 @@ float3 processPatchMeanNorm(const __global uchar *I0_ptr, const __global uchar *
float sum_diff = 0.0, sum_diff_sq = 0.0; float sum_diff = 0.0, sum_diff_sq = 0.0;
float sum_I0x_mul = 0.0, sum_I0y_mul = 0.0; float sum_I0x_mul = 0.0, sum_I0y_mul = 0.0;
int n = patch_sz * patch_sz; int n = patch_sz * patch_sz;
uchar8 I1_vec1, I1_vec2; uchar8 I1_vec1;
uchar I1_val1, I1_val2; uchar8 I1_vec2 = vload8(0, I1_ptr);
uchar I1_val1;
uchar I1_val2 = I1_ptr[patch_sz];
for (int i = 0; i < 8; i++) for (int i = 0; i < 8; i++)
{ {
uchar8 I0_vec = vload8(0, I0_ptr + i * I0_stride); uchar8 I0_vec = vload8(0, I0_ptr + i * I0_stride);
I1_vec1 = (i == 0) ? vload8(0, I1_ptr + i * I1_stride) : I1_vec2; I1_vec1 = I1_vec2;
I1_vec2 = vload8(0, I1_ptr + (i + 1) * I1_stride); I1_vec2 = vload8(0, I1_ptr + (i + 1) * I1_stride);
I1_val1 = (i == 0) ? I1_ptr[i * I1_stride + patch_sz] : I1_val2; I1_val1 = I1_val2;
I1_val2 = I1_ptr[(i + 1) * I1_stride + patch_sz]; I1_val2 = I1_ptr[(i + 1) * I1_stride + patch_sz];
float8 vec = w00 * convert_float8(I1_vec1) + w01 * convert_float8((uchar8)(I1_vec1.s123, I1_vec1.s4567, I1_val1)) + float8 vec = w00 * convert_float8(I1_vec1) + w01 * convert_float8((uchar8)(I1_vec1.s123, I1_vec1.s4567, I1_val1)) +