update_dir3.cuh
6.53 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
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
#ifndef STIM_CUDA_UPDATE_DIR3_H
#define STIM_CUDA_UPDATE_DIR3_H
# include <iostream>
# include <cuda.h>
#include <stim/cuda/cudatools.h>
#include "cpyToshare.cuh"
// this kernel calculates the voting direction for the next iteration based on the angle between the location of this voter and the maximum vote value in its voting area.
template<typename T>
__global__ void update_dir3(T* gpu_dir, T* gpu_grad, T* gpu_vote, T cos_phi, int rx, int ry, int rz, int x, int y, int z){
extern __shared__ float s_vote[];
//calculate x,y,z coordinates for this thread
int xi = blockIdx.x * blockDim.x + threadIdx.x;
//find the grid size along y
int grid_y = y / blockDim.y;
int blockidx_y = blockIdx.y % grid_y;
int yi = blockidx_y * blockDim.y + threadIdx.y;
int zi = blockIdx.y / grid_y;
//compute the global 1D index for this pixel
int i = zi * x * y + yi * x + xi;
if(xi >= x|| yi >= y || zi>= z) return;
// find the starting points for this block along the x and y directions
int bxi = blockIdx.x * blockDim.x;
int byi = blockidx_y * blockDim.y;
//find the starting points and the size of the window, which will be copied to the 2D-shared memory
int bxs = bxi - rx;
int bys = byi - ry;
int xwidth = 2 * rx + blockDim.x;
int ywidth = 2 * ry + blockDim.y;
//compute the coordinations of this pixel in the 2D-shared memory.
int sx_rx = threadIdx.x + rx;
int sy_ry = threadIdx.y + ry;
//find the gradient values along the x, y ,z axis, and the gradient magnitude for the voter
float g_v_x = gpu_grad[i * 3 + 0];
float g_v_y = gpu_grad[i * 3 + 1];
float g_v_z = gpu_grad[i * 3 + 2];
float mag_v = sqrt( g_v_x * g_v_x + g_v_y * g_v_y + g_v_z * g_v_z);
// define a local variable to maximum value of the vote image in the voting area for this voter
float max = 0;
float l_vote = 0;
// define local variables for the x, y, and z coordinations point to the vote direction
float id_x = g_v_x;
float id_y = g_v_y;
float id_z = g_v_z;
int rx_sq = rx * rx;
int ry_sq = ry * ry;
int rz_sq = rz * rz;
for (int z_p = -rz; z_p<=rz; z_p++){
int zi_p = zi + z_p;
if ((zi_p) >=0 && (zi_p) < z){
//call the function to copy one slide of vote date to the 2D-shared memory.
__syncthreads();
cpyG2S2D<float>(s_vote, gpu_vote, bxs, bys, zi + z_p, xwidth, ywidth, threadIdx, blockDim, x, y);
__syncthreads();
float z_sq = z_p * z_p;
float d_z_sq = (z_sq)/rz_sq;
for(int y_p = -ry; y_p <= ry; y_p++){
float y_sq = y_p * y_p;
float yz_sq = y_sq + z_sq;
int yi_p = (yi + y_p) ;
float d_yz_sq = (y_sq)/ry_sq + d_z_sq;
unsigned int s_y1d = (sy_ry + y_p) * xwidth;
for(int x_p = -rx; x_p <= rx; x_p++){
//check if the current pixel is inside of the data-set.
int xi_p = (xi + x_p) ;
if (yi_p >=0 && yi_p < y && xi_p >=0 && xi_p < x){
//calculate the distance between the pixel and the current voter.
float x_sq = x_p * x_p;
float d_pv = sqrt(x_sq + yz_sq);
// calculate the angle between the pixel and the current voter.
float cos_diff = (g_v_x * x_p + g_v_y * y_p + g_v_z * z_p)/(d_pv * mag_v);
if ((((x_sq)/rx_sq + d_yz_sq )<= 1) && (cos_diff >= cos_phi)){
//calculate the 1D index for the current pixel in the 2D-shared memory
unsigned int id_s = s_y1d + (sx_rx + x_p);
l_vote = s_vote[id_s];
// compare the vote value of this pixel with the max value to find the maxima and its index.
if (l_vote>max) {
max = l_vote;
id_x = x_p;
id_y = y_p;
id_z = z_p;
}
}
}
}
}
}
}
float m_id = sqrt (id_x*id_x + id_y*id_y + id_z*id_z);
gpu_dir[i * 3 + 0] = mag_v * (id_x/m_id);
gpu_dir[i * 3 + 1] = mag_v * (id_y/m_id);
gpu_dir[i * 3 + 2] = mag_v * (id_z/m_id);
}
// this kernel updates the gradient direction by the calculated voting direction.
template<typename T>
__global__ void update_grad3(T* gpu_grad, T* gpu_dir, int x, int y, int z){
//calculate x,y,z coordinates for this thread
int xi = blockIdx.x * blockDim.x + threadIdx.x;
//find the grid size along y
int grid_y = y / blockDim.y;
int blockidx_y = blockIdx.y % grid_y;
int yi = blockidx_y * blockDim.y + threadIdx.y;
int zi = blockIdx.y / grid_y;
int i = zi * x * y + yi * x + xi;
if(xi >= x || yi >= y || zi >= z) return;
//update the gradient image with the new direction direction
gpu_grad[i * 3 + 0] = gpu_dir [i * 3 + 0];
gpu_grad[i * 3 + 1] = gpu_dir [i * 3 + 1];
gpu_grad[i * 3 + 2] = gpu_dir [i * 3 + 2];
}
template<typename T>
void gpu_update_dir3(T* gpu_grad, T* gpu_vote, T phi, T cos_phi, unsigned int r[], unsigned int x, unsigned int y, unsigned int z){
unsigned int max_threads = stim::maxThreadsPerBlock();
dim3 threads(sqrt (max_threads),sqrt (max_threads));
dim3 blocks(x / threads.x + 1, (y / threads.y + 1) * z);
unsigned int shared_bytes = (threads.x + 2*r[0])*(threads.y + 2*r[1])*sizeof(T);
// allocate space on the GPU for the updated vote direction
T* gpu_dir;
cudaMalloc(&gpu_dir, x * y * z * sizeof(T) * 3);
//call the kernel to calculate the new voting direction
update_dir3 <<< blocks, threads, shared_bytes >>>(gpu_dir, gpu_grad, gpu_vote, cos_phi, r[0], r[1], r[2], x , y, z);
//call the kernel to update the gradient direction
update_grad3 <<< blocks, threads >>>(gpu_grad, gpu_dir, x , y, z);
//free allocated memory
cudaFree(gpu_dir);
}
template<typename T>
void cpu_update_dir3(T* cpu_grad, T* cpu_vote, T cos_phi, unsigned int r[], unsigned int x, unsigned int y, unsigned int z){
//calculate the number of bytes in the array
unsigned int bytes = x * y * z * sizeof(T);
//allocate space on the GPU for the Vote data
T* gpu_vote;
cudaMalloc(&gpu_vote, bytes);
//copy the input vote data to the GPU
cudaMemcpy(gpu_vote, cpu_vote, bytes, cudaMemcpyHostToDevice);
//allocate space on the GPU for the Gradient data
T* gpu_grad;
cudaMalloc(&gpu_grad, bytes*3);
//copy the Gradient data to the GPU
cudaMemcpy(gpu_grad, cpu_grad, bytes*3, cudaMemcpyHostToDevice);
//call the GPU version of the update direction function
gpu_update_dir3<T>(gpu_grad, gpu_vote, cos_phi, r, x , y, z);
//copy the new gradient image back to the CPU
cudaMemcpy(cpu_grad, gpu_grad, bytes*3, cudaMemcpyDeviceToHost) ;
//free allocated memory
cudaFree(gpu_vote);
cudaFree(gpu_grad);
}
#endif