PartMemCopy.cu
11.2 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
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
#include "cuda_kernels.h"
#include <algorithm>
typedef unsigned char uchar;
typedef unsigned int uint32;
typedef int int32;
#define MAX_SNAPSHOT_WIDTH 320
#define MAX_SNAPSHOT_HEIGHT 320
namespace cuda_common
{
__global__ void kernel_memcopy(unsigned char* d_srcRGB, int src_width, int src_height,
unsigned char* d_dstRGB, int left, int top, int right, int bottom)
{
const int dst_x = blockIdx.x * blockDim.x + threadIdx.x;
const int dst_y = blockIdx.y * blockDim.y + threadIdx.y;
const int dst_width = right - left;
const int dst_height = bottom - top;
if (dst_x < dst_width && dst_y < dst_height)
{
int src_x = left + dst_x;
int src_y = top + dst_y;
//bgr...bgr...bgr...
d_dstRGB[(dst_y*dst_width + dst_x) * 3] = (unsigned char)d_srcRGB[(src_y*src_width + src_x) * 3];
d_dstRGB[(dst_y*dst_width + dst_x)
* 3 + 1] = (unsigned char)d_srcRGB[(src_y*src_width + src_x) * 3 + 1];
d_dstRGB[(dst_y*dst_width + dst_x) * 3 + 2] = (unsigned char)d_srcRGB[(src_y*src_width + src_x) * 3 + 2];
//bbb...ggg...rrr...
//d_dstRGB[(dst_y*dst_width) + dst_x] = (unsigned char)d_srcRGB[(src_y*src_width) + src_x];
//d_dstRGB[(dst_width*dst_height) + (dst_y*dst_width) + dst_x] = (unsigned char)d_srcRGB[(src_width*src_height) + (src_y*src_width) + src_x];
//d_dstRGB[(2 * dst_width*dst_height) + (dst_y*dst_width) + dst_x] = (unsigned char)d_srcRGB[(2 * src_width*src_height) + (src_y*src_width) + src_x];
/* memcpy(d_dstRGB + (dst_y*src_width) + dst_x, d_srcRGB + (src_y*src_width) + src_x, sizeof(float));
memcpy(d_dstRGB + (src_width*src_height) + (dst_y*src_width) + dst_x, d_srcRGB + (src_width*src_height) + (src_y*src_width) + src_x, sizeof(float));
memcpy(d_dstRGB + (2 * src_width*src_height) + (dst_y*src_width) + dst_x, d_srcRGB + (2 * src_width*src_height) + (src_y*src_width) + src_x, sizeof(float));*/
}
}
cudaError_t PartMemCopy(unsigned char* d_srcRGB, int src_width, int src_height, unsigned char* d_dstRGB, int left, int top, int right, int bottom)
{
dim3 block(32, 16, 1);
dim3 grid(((right - left) + (block.x - 1)) / block.x, ((bottom - top) + (block.y - 1)) / block.y, 1);
kernel_memcopy << < grid, block >> > (d_srcRGB, src_width, src_height, d_dstRGB, left, top, right, bottom);
cudaError_t cudaStatus = cudaGetLastError();
if (cudaStatus != cudaSuccess) {
fprintf(stderr, "Part 50 kernel_memcopy launch failed: %s\n", cudaGetErrorString(cudaStatus));
return cudaStatus;
}
cudaStatus = cudaDeviceSynchronize();
if (cudaStatus != cudaSuccess) {
fprintf(stderr, "cudaDeviceSynchronize returned error code %d after launching kernel_bilinear!\n", cudaStatus);
return cudaStatus;
}
return cudaStatus;
}
// __global__ void kernel_memcopy_mean_variance(float* d_srcRGB, int src_width, int src_height,
// unsigned char* vd_dstRGB, int count, int * vleft, int* vtop, int* vright, int * vbottom, float submeanb,float submeang, float submeanr, float varianceb,float varianceg, float variancer)
// {
// const int dst_x = blockIdx.x * blockDim.x + threadIdx.x;
// const int dst_y = blockIdx.y * blockDim.y + threadIdx.y;
// for (int i=0;i<count;i++)
// {
// const int left = vleft[i];
// const int right = vright[i];
// const int top = vtop[i];
// const int bottom = vbottom[i];
//
// const int dst_width = right - left;
// const int dst_height = bottom - top;
//
//
// unsigned char * d_dstRGB = vd_dstRGB + i * ;
//
// if (dst_x < dst_width && dst_y < dst_height)
// {
// int src_x = left + dst_x;
// int src_y = top + dst_y;
//
// d_dstRGB[(dst_y*dst_width) + dst_x] = (d_srcRGB[(src_y*src_width) + src_x] - submeanb)*varianceb;
// d_dstRGB[(dst_width*dst_height) + (dst_y*dst_width) + dst_x] = (d_srcRGB[(src_width*src_height) + (src_y*src_width) + src_x] -submeang)*varianceg;
// d_dstRGB[(2 * dst_width*dst_height) + (dst_y*dst_width) + dst_x] = (d_srcRGB[(2 * src_width*src_height) + (src_y*src_width) + src_x] - submeanr) * variancer;
//
// }
// }
// }
__global__ void PartCopy_ResizeImgBilinearBGR_Mean_Variance_CUDAKernel(
unsigned char * d_srcRGB, int srcimg_width, int srcimg_height,
int* vleft, int* vtop, int* vright, int * vbottom,
unsigned char** vd_dstRGB, int count, int *dst_width, int *dst_height,
float submeanb, float submeang, float submeanr,
float varianceb, float varianceg, float variancer)
{
int i = blockIdx.z;
//for (int i = 0; i<count; i++)
{
const int left = vleft[i];
const int right = vright[i];
const int top = vtop[i];
const int bottom = vbottom[i];
const int cur_dst_width = dst_width[i];
const int cur_dst_height = dst_height[i];
unsigned char* d_dstRGB = vd_dstRGB[i];
const int src_width = right - left;
const int src_height = bottom - top;
const int x = blockIdx.x * blockDim.x + threadIdx.x;// + left;
const int y = blockIdx.y * blockDim.y + threadIdx.y;//+ top;
const int dst_x = blockIdx.x * blockDim.x + threadIdx.x;
const int dst_y = blockIdx.y * blockDim.y + threadIdx.y;
/*if (dst_x == 0 && dst_y == 0)
printf("%d %d %d %d %d\n", i, vleft[i], vright[i], cur_dst_width, cur_dst_height);*/
unsigned char * src_img = d_srcRGB;
unsigned char * dst_img = d_dstRGB;
if (dst_x < cur_dst_width && dst_y < cur_dst_height)
{
float fx = (x + 0.5)*src_width / (float)cur_dst_width - 0.5 + left;
float fy = (y + 0.5)*src_height / (float)cur_dst_height - 0.5 + top;
int ax = floor(fx);
int ay = floor(fy);
if (ax < 0)
{
ax = 0;
}
if (ax > srcimg_width - 2)
{
ax = srcimg_width - 2;
}
if (ay < 0) {
ay = 0;
}
if (ay > srcimg_height - 2)
{
ay = srcimg_height - 2;
}
int A = ax + ay*srcimg_width;
int B = ax + ay*srcimg_width + 1;
int C = ax + ay*srcimg_width + srcimg_width;
int D = ax + ay*srcimg_width + srcimg_width + 1;
float w1, w2, w3, w4;
w1 = fx - ax;
w2 = 1 - w1;
w3 = fy - ay;
w4 = 1 - w3;
float blue = src_img[A * 3] * w2*w4 + src_img[B * 3] * w1*w4 + src_img[C * 3] * w2*w3 + src_img[D * 3] * w1*w3;
float green = src_img[A * 3 + 1] * w2*w4 + src_img[B * 3 + 1] * w1*w4
+ src_img[C * 3 + 1] * w2*w3 + src_img[D * 3 + 1] * w1*w3;
float red = src_img[A * 3 + 2] * w2*w4 + src_img[B * 3 + 2] * w1*w4
+ src_img[C * 3 + 2] * w2*w3 + src_img[D * 3 + 2] * w1*w3;
/*dst_img[(dst_y * dst_width + dst_x) * 3] = (unsigned char)(blue - submeanb)*varianceb;
dst_img[(dst_y * dst_width + dst_x) * 3 + 1] =(unsigned char) (green - submeang)*varianceg;
dst_img[(dst_y * dst_width + dst_x) * 3 + 2] = (unsigned char) (red - submeanr)*variancer;*/
if (blue < 0)
blue = 0;
else if (blue > 255)
blue = 255;
if (green < 0)
green = 0;
else if (green > 255)
green = 255;
if (red < 0)
red = 0;
else if (red > 255)
red = 255;
dst_img[(dst_y * cur_dst_width + dst_x) * 3] = (unsigned char)blue;
dst_img[(dst_y * cur_dst_width + dst_x) * 3 + 1] = (unsigned char)green;
dst_img[(dst_y * cur_dst_width + dst_x) * 3 + 2] = (unsigned char)red;
/*if (src_img[(dst_y * dst_width + dst_x) * 3] < 0)
src_img[(dst_y * dst_width + dst_x) * 3] = 0;
else if (src_img[(dst_y * dst_width + dst_x) * 3] > 255)
src_img[(dst_y * dst_width + dst_x) * 3] = 255;
if (src_img[(dst_y * dst_width + dst_x) * 3 + 1] < 0)
src_img[(dst_y * dst_width + dst_x) * 3 + 1] = 0;
else if (src_img[(dst_y * dst_width + dst_x) * 3 + 1] > 255)
src_img[(dst_y * dst_width + dst_x) * 3 + 1] = 255;
if (src_img[(dst_y * dst_width + dst_x) * 3 + 2] < 0)
src_img[(dst_y * dst_width + dst_x) * 3 + 2] = 0;
else if (src_img[(dst_y * dst_width + dst_x) * 3 + 2] > 255)
src_img[(dst_y * dst_width + dst_x) * 3 + 2] = 255;
dst_img[(dst_y * dst_width + dst_x) * 3] = (unsigned char)src_img[(dst_y * dst_width + dst_x) * 3];
dst_img[(dst_y * dst_width + dst_x) * 3 + 1] = (unsigned char)src_img[(dst_y * dst_width + dst_x) * 3 + 1];
dst_img[(dst_y * dst_width + dst_x) * 3 + 2] = (unsigned char)src_img[(dst_y * dst_width + dst_x) * 3 + 2];*/
}
}
}
cudaError_t PartMemResizeBatch(unsigned char* d_srcRGB, int src_width, int src_height, unsigned char** d_dstRGB, int count, int* left, int* top, int* right, int* bottom, int *dst_w, int *dst_h, float submeanb, float submeang, float submeanr,
float varianceb, float varianceg, float variancer)
{
/* cudaEvent_t start, stop;
float time;
cudaEventCreate(&start);
cudaEventCreate(&stop);
cudaEventRecord(start, 0);*/
dim3 block(32, 16, 1);
dim3 grid((*std::max_element(dst_w, dst_w+ count) + (block.x - 1)) / block.x, (*std::max_element(dst_h, dst_h + count) + (block.y - 1)) / block.y, count);
int * gpu_left;
cudaMalloc(&gpu_left, 1000 * sizeof(int));
cudaMemcpy(gpu_left, left, count * sizeof(int), cudaMemcpyHostToDevice);
int * gpu_right;
cudaMalloc(&gpu_right, 1000 * sizeof(int));
cudaMemcpy(gpu_right, right, count * sizeof(int), cudaMemcpyHostToDevice);
int * gpu_top;
cudaMalloc(&gpu_top, 1000 * sizeof(int));
cudaMemcpy(gpu_top, top, count * sizeof(int), cudaMemcpyHostToDevice);
int * gpu_bottom;
cudaMalloc(&gpu_bottom, 1000 * sizeof(int));
cudaMemcpy(gpu_bottom, bottom, count * sizeof(int), cudaMemcpyHostToDevice);
int * gpu_dst_w;
cudaMalloc(&gpu_dst_w, 1000 * sizeof(int));
cudaMemcpy(gpu_dst_w, dst_w, count * sizeof(int), cudaMemcpyHostToDevice);
int * gpu_dst_h;
cudaMalloc(&gpu_dst_h, 1000 * sizeof(int));
cudaMemcpy(gpu_dst_h, dst_h, count * sizeof(int), cudaMemcpyHostToDevice);
unsigned char** gpu_dst_rgb;
cudaMalloc(&gpu_dst_rgb, 1000 * sizeof(unsigned char*));
cudaMemcpy(gpu_dst_rgb, d_dstRGB, count * sizeof(unsigned char*), cudaMemcpyHostToDevice);
//cudaMemcpy(cpu_personfloat, d_srcRGB, 112*224*2*sizeof(float), cudaMemcpyDeviceToHost);
// for(int i=0;i<100;i++)
// {
// printf("the score is %f\t",cpu_personfloat[i]);
// }
PartCopy_ResizeImgBilinearBGR_Mean_Variance_CUDAKernel << < grid, block >> > (
d_srcRGB, src_width, src_height,
gpu_left, gpu_top, gpu_right, gpu_bottom,
gpu_dst_rgb, count, gpu_dst_w, gpu_dst_h,
submeanb, submeang, submeanr,
varianceb, varianceg, variancer);
cudaFree(gpu_top);
cudaFree(gpu_bottom);
cudaFree(gpu_left);
cudaFree(gpu_right);
cudaFree(gpu_dst_w);
cudaFree(gpu_dst_h);
cudaFree(gpu_dst_rgb);
cudaError_t cudaStatus = cudaGetLastError();
if (cudaStatus != cudaSuccess) {
fprintf(stderr, "Part 270 kernel_memcopy launch failed: %s\n", cudaGetErrorString(cudaStatus));
return cudaStatus;
}
cudaStatus = cudaDeviceSynchronize();
if (cudaStatus != cudaSuccess) {
fprintf(stderr, "cudaDeviceSynchronize returned error code %d after launching kernel_bilinear!\n", cudaStatus);
return cudaStatus;
}
/*cudaEventRecord(stop, 0);
cudaEventSynchronize(stop);
cudaEventElapsedTime(&time, start, stop);
cudaEventDestroy(start);
cudaEventDestroy(stop);
printf("ºËº¯ÊýÏûºÄʱ¼ä:%f\n", time);*/
return cudaStatus;
}
}