-
Notifications
You must be signed in to change notification settings - Fork 0
/
HW4_cuda.cu
266 lines (222 loc) · 7.19 KB
/
HW4_cuda.cu
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
#include <stdio.h>
#include <stdlib.h>
const int INF = 1000000000;
int V = 20010;
void input(char *inFileName);
void output(char *outFileName);
void block_FW(int B);
int ceil(int a, int b);
int n, m; // Number of vertices, edges
int* host_ptr = NULL;
size_t pitch;
// for device
int* device_ptr = NULL;
__global__ void gpu_phase1(int* dist, int B, int Round, int block_start_x, int block_start_y, int n, size_t pitch);
__global__ void gpu_phase2(int* dist, int B, int Round, int block_start_x, int block_start_y, int n, int pos, size_t pitch);
__global__ void gpu_phase3(int* dist, int B, int Round, int block_start_x, int block_start_y, int n, size_t pitch);
int main(int argc, char* argv[])
{
input(argv[1]);
int B = atoi(argv[3]);
// allocate memory for device
cudaMallocPitch(&device_ptr, &pitch, V*sizeof(int), V);
cudaMemcpy2D(device_ptr, pitch, host_ptr, V*sizeof(int), V*sizeof(int), V, cudaMemcpyHostToDevice);
block_FW(B);
cudaMemcpy2D(host_ptr, V*sizeof(int), device_ptr, pitch, V*sizeof(int), V, cudaMemcpyDeviceToHost);
cudaFree(device_ptr);
output(argv[2]);
free(host_ptr);
return 0;
}
void input(char *inFileName)
{
FILE *infile = fopen(inFileName, "r");
fscanf(infile, "%d %d", &n, &m);
// Malloc host memory
V = n + 10;
host_ptr = (int*)malloc((size_t)V * V * sizeof(int));
for (int i = 0; i < n; ++i) {
for (int j = 0; j < n; ++j) {
if (i == j) host_ptr[i*V+j] = 0;
else host_ptr[i*V+j] = INF;
}
}
while (--m >= 0) {
int a, b, v;
fscanf(infile, "%d %d %d", &a, &b, &v);
host_ptr[a*V+b] = v;
}
fclose(infile);
}
void output(char *outFileName)
{
FILE *outfile = fopen(outFileName, "w");
for (int i = 0; i < n; ++i) {
for (int j = 0; j < n; ++j) {
if (host_ptr[i*V+j] >= INF)
host_ptr[i*V+j] = INF;
}
fwrite(&host_ptr[i*V], sizeof(int), n, outfile);
}
fclose(outfile);
}
int ceil(int a, int b)
{
return (a + b -1)/b;
}
void block_FW(int B)
{
int round = ceil(n, B);
dim3 blocks = {1, 1};
dim3 threads = {(unsigned int)B, (unsigned int)B};
printf("B: %d, Round: %d\n", B, round);
for (unsigned int r = 0; r < round; ++r) {
if (r % 10 == 0)
printf("%d %d\n", r, round);
/* Phase 1*/
blocks = {1, 1};
gpu_phase1<<<blocks, threads, B*B*1*sizeof(int)>>>(device_ptr, B, r, r, r, n, pitch/sizeof(int));
/* Phase 2*/
if (r > 0) {
// left
blocks = {1, r};
gpu_phase2<<<blocks, threads, B*B*2*sizeof(int)>>>(device_ptr, B, r, r, 0, n, 1, pitch/sizeof(int));
// up
blocks = {r, 1};
gpu_phase2<<<blocks, threads, B*B*2*sizeof(int)>>>(device_ptr, B, r, 0, r, n, 0, pitch/sizeof(int));
}
if (r < round - 1) {
// right
blocks = {1, round - r -1};
gpu_phase2<<<blocks, threads, B*B*2*sizeof(int)>>>(device_ptr, B, r, r, r +1, n, 1, pitch/sizeof(int));
// down
blocks = {round - r -1, 1};
gpu_phase2<<<blocks, threads, B*B*2*sizeof(int)>>>(device_ptr, B, r, r +1, r, n, 0, pitch/sizeof(int));
}
/* Phase 3*/
if (r == 0) {
// down right
blocks = {round - r -1, round - r -1};
gpu_phase3<<<blocks, threads, B*B*3*sizeof(int)>>>(device_ptr, B, r, r +1, r +1, n, pitch/sizeof(int));
}
else if (r == round - 1) {
// upper left
blocks = {r, r};
gpu_phase3<<<blocks, threads, B*B*3*sizeof(int)>>>(device_ptr, B, r, 0, 0, n, pitch/sizeof(int));
}
else {
// down right
blocks = {round - r -1, round - r -1};
gpu_phase3<<<blocks, threads, B*B*3*sizeof(int)>>>(device_ptr, B, r, r +1, r +1, n, pitch/sizeof(int));
// upper left
blocks = {r, r};
gpu_phase3<<<blocks, threads, B*B*3*sizeof(int)>>>(device_ptr, B, r, 0, 0, n, pitch/sizeof(int));
// upper right
blocks = {r, round -r -1};
gpu_phase3<<<blocks, threads, B*B*3*sizeof(int)>>>(device_ptr, B, r, 0, r +1, n, pitch/sizeof(int));
// down left
blocks = {round - r -1, r};
gpu_phase3<<<blocks, threads, B*B*3*sizeof(int)>>>(device_ptr, B, r, r +1, 0, n, pitch/sizeof(int));
}
}
}
extern __shared__ int shared_mem[];
__global__ void gpu_phase1(int* dist, int B, int Round, int block_start_x, int block_start_y, int n, size_t pitch)
{
int V = pitch;
int tid = threadIdx.y * B + threadIdx.x;
int i = (block_start_x + blockIdx.x) * B + threadIdx.x;
int j = (block_start_y + blockIdx.y) * B + threadIdx.y;
// need self block - (b_i, b_j)
shared_mem[tid] = dist[j*V+i];
__syncthreads();
for (int k = Round * B; k < (Round +1) * B && k < n; ++k) {
if (i < n && j < n) {
//===== change new posision by: =====//
// new_i = origin_i - B * B_i
// new_j = origin_j - B * B_j
//===================================//
int k_new = k - B * Round;
int i_new = i - B * Round;
int j_new = j - B * Round;
int tmp = shared_mem[k_new*B+i_new] + shared_mem[j_new*B+k_new];
if (tmp < shared_mem[tid]) {
shared_mem[tid] = tmp;
}
}
__syncthreads();
}
dist[j*V+i] = shared_mem[tid];
}
extern __shared__ int shared_mem[];
__global__ void gpu_phase2(int* dist, int B, int Round, int block_start_x, int block_start_y, int n, int pos, size_t pitch)
{
int V = pitch;
int tid = threadIdx.y * B + threadIdx.x;
int b_i = block_start_x + blockIdx.x;
int b_j = block_start_y + blockIdx.y;
int i = b_i * B + threadIdx.x;
int j = b_j * B + threadIdx.y;
// need self block - (b_i, b_j) & pivot block - (Round, Round)
shared_mem[tid+B*B] = dist[j*V+i];
shared_mem[tid] = dist[(Round*B+threadIdx.y)*V+(Round*B+threadIdx.x)];
__syncthreads();
for (int k = Round * B; k < (Round +1) * B && k < n; ++k) {
if (i < n && j < n) {
// up, down
if (pos == 1) {
int k_new_1 = k - B * Round;
int i_new_1 = i - B * Round;
int k_new_2 = k - B * b_i;
int j_new_2 = j - B * b_j;
int tmp = shared_mem[k_new_1*B+i_new_1] + shared_mem[j_new_2*B+k_new_2+B*B];
if (tmp < shared_mem[tid+B*B]) {
shared_mem[tid+B*B] = tmp;
}
}
// left, right
else {
int k_new_1 = k - B * Round;
int j_new_1 = j - B * Round;
int i_new_2 = i - B * b_i;
int k_new_2 = k - B * b_j;
int tmp = shared_mem[k_new_2*B+i_new_2+B*B] + shared_mem[j_new_1*B+k_new_1];
if (tmp < shared_mem[tid+B*B]) {
shared_mem[tid+B*B] = tmp;
}
}
}
__syncthreads();
}
dist[j*V+i] = shared_mem[tid+B*B];
}
__global__ void gpu_phase3(int* dist, int B, int Round, int block_start_x, int block_start_y, int n, size_t pitch)
{
int V = pitch;
int tid = threadIdx.y * B + threadIdx.x;
int b_i = block_start_x + blockIdx.x;
int b_j = block_start_y + blockIdx.y;
int i = b_i * B + threadIdx.x;
int j = b_j * B + threadIdx.y;
// need self block - (b_i, b_j) & row / column block
shared_mem[tid] = dist[j*V+i];
shared_mem[tid+B*B] = dist[(b_j*B+threadIdx.y)*V+(Round*B+threadIdx.x)]; // left, right
shared_mem[tid+B*B*2] = dist[(Round*B+threadIdx.y)*V+(b_i*B+threadIdx.x)]; // up , down
__syncthreads();
for (int k = Round * B; k < (Round +1) * B && k < n; ++k) {
if (i < n && j < n) {
// left, right
int i_new_1 = i - B * b_i;
int k_new_1 = k - B * Round;
// up, down
int k_new_2 = k - B * Round;
int j_new_2 = j - B * b_j;
int tmp = shared_mem[k_new_1*B+i_new_1+B*B*2] + shared_mem[j_new_2*B+k_new_2+B*B];
if (tmp < shared_mem[tid]) {
shared_mem[tid] = tmp;
}
}
__syncthreads();
}
dist[j*V+i] = shared_mem[tid];
}