gradient.cuh
2.47 KB
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
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
#ifndef STIM_CUDA_GRADIENT_H
#define STIM_CUDA_GRADIENT_H
#include <iostream>
#include <cuda.h>
#include <stim/cuda/cudatools.h>
namespace stim{
namespace cuda{
template<typename T>
__global__ void gradient_2d(T* out, T* in, size_t x, size_t y){
// calculate the 2D coordinates for this current thread.
size_t xi = blockIdx.x * blockDim.x + threadIdx.x;
size_t yi = blockIdx.y * blockDim.y + threadIdx.y;
// convert 2D coordinates to 1D
size_t i = yi * x + xi;
//return if the pixel is outside of the image
if(xi >= x || yi >= y) return;
//calculate indices for the forward difference
size_t i_xp = yi * x + (xi + 1);
size_t i_yp = (yi + 1) * x + xi;
//calculate indices for the backward difference
size_t i_xn = yi * x + (xi - 1);
size_t i_yn = (yi - 1) * x + xi;
//use forward differences if a coordinate is zero
if(xi == 0)
out[i * 2 + 0] = in[i_xp] - in[i];
else if (xi == x - 1)
out[i * 2 + 0] = in[i] - in[i_xn];
else
out[i * 2 + 0] = (in[i_xp] - in[i_xn]) / 2;
if(yi == 0)
out[i * 2 + 1] = in[i_yp] - in[i];
else if(yi == y-1)
out[i * 2 + 1] = in[i] - in[i_yn];
else
out[i * 2 + 1] = (in[i_yp] - in[i_yn]) / 2;
}
template<typename T>
void gpu_gradient_2d(T* gpuGrad, T* gpuI, size_t x, size_t y){
//get the maximum number of threads per block for the CUDA device
unsigned int max_threads = stim::maxThreadsPerBlock();
dim3 threads(max_threads, 1);
dim3 blocks((unsigned int)(x/threads.x) + 1 , (unsigned int)y);
//call the GPU kernel to determine the gradient
gradient_2d<T> <<< blocks, threads >>>(gpuGrad, gpuI, x, y);
}
template<typename T>
void cpu_gradient_2d(T* out, T* in, unsigned int x, unsigned int y){
//get the number of pixels in the image
unsigned int pixels = x * y;
unsigned int bytes = pixels * sizeof(T);
//allocate space on the GPU for the input image
T* gpuIn;
HANDLE_ERROR(cudaMalloc(&gpuIn, bytes));
//copy the image data to the GPU
HANDLE_ERROR(cudaMemcpy(gpuIn, in, bytes, cudaMemcpyHostToDevice));
//allocate space on the GPU for the output gradient image
T* gpuOut;
cudaMalloc(&gpuOut, bytes * 2); //the output image will have two channels (x, y)
//call the GPU version of this function
gpu_gradient_2d(gpuOut, gpuIn, x, y);
//copy the results to the CPU
cudaMemcpy(out, gpuOut, bytes * 2, cudaMemcpyDeviceToHost);
//free allocated memory
cudaFree(gpuOut);
cudaFree(gpuIn);
}
}
}
#endif