1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
|
__global__ void transpose1(float *dst, float *src, int width, int height, int step) {
__shared__ float data[BLOCK_SIZE_2D][BLOCK_SIZE_2D];
int dx = blockIdx.x * blockDim.x;
int dy = blockIdx.y * blockDim.y;
int x = threadIdx.x + dx;
int y = threadIdx.y + dy;
data[threadIdx.x][threadIdx.y] =
src[x + step + y*width] - src[x + y*width];
__syncthreads();
x = threadIdx.x + dy;
y = threadIdx.y + dx;
dst[x + y*height] = data[threadIdx.y][threadIdx.x];
}
__global__ void transpose2(float *lsum, float *denom, float *src1, float *src2, int width, int height, int step) {
__shared__ float data1[BLOCK_SIZE_2D][BLOCK_SIZE_2D];
__shared__ float data2[BLOCK_SIZE_2D][BLOCK_SIZE_2D];
int dx = blockIdx.x * blockDim.x;
int dy = blockIdx.y * blockDim.y;
int x = threadIdx.x + dx;
int y = threadIdx.y + dy;
data1[threadIdx.x][threadIdx.y] =
src1[x + step + y*width] - src1[x + y*width];
data2[threadIdx.x][threadIdx.y] =
src2[x + step + y*width] - src2[x + y*width];
__syncthreads();
x = threadIdx.x + dy;
y = threadIdx.y + dx;
float val1 = data1[threadIdx.y][threadIdx.x];
float val2 = data2[threadIdx.y][threadIdx.x];
float coef = step * step;
if (x < height) {
lsum[x + y*height] = val1;
denom[x + y*height] = sqrtf(fmaxf((val2 - val1*val1/coef)/(coef - 1) ,0));
}
}
|