-
Notifications
You must be signed in to change notification settings - Fork 5
/
fiddlelink.cu
239 lines (175 loc) · 7.55 KB
/
fiddlelink.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
#include<assert.h>
#include<stdlib.h>
#include<stdio.h>
#include<time.h>
#include<cuda.h>
#include "fiddlelink.h"
void *peer_access(void *addr){
int *tx1 = (int*)addr;
int *rx1 = (int*)(addr + sizeof(int));
int tx = *tx1;
int rx = *rx1;
printf("tx is %d, rx is %d\n",tx,rx);
cudaSetDevice(tx);
cudaDeviceEnablePeerAccess(rx,0);
}
void pair_stream(int rx, int tx, void* dst, void* src, double size, int type){
if(type == 3){
if(dst == src)
{
printf("stream src and dst addtress should be different.\n");
exit(0);
}
else{
int split = sizeof(char)*8;
long int chunk = size/split;
cudaStream_t stream[2];
cudaStreamCreate(&stream[0]);
cudaStreamCreate(&stream[1]);
cudaDeviceDisablePeerAccess(tx);
cudaDeviceDisablePeerAccess(rx);
cudaMemcpyAsync(dst,src,chunk,cudaMemcpyDeviceToDevice,stream[0]);
cudaDeviceEnablePeerAccess(tx,0);
cudaDeviceEnablePeerAccess(rx,0);
cudaMemcpyAsync(dst,src,chunk,cudaMemcpyDeviceToDevice,stream[1]);
long int address_src = (long long unsigned)src;
long int address_dst = (long long unsigned)dst;
for(int i = 0; i<split;i++)
{
long int address_src1 = address_src+i*chunk;
long int address_dst1 = address_dst+i*chunk;
void* src1 = (void*)address_src1;
void* dst1 = (void*)address_dst1;
long int address_src2 = address_src + (i+1)*chunk;
long int address_dst2 = address_dst + (i+1)*chunk;
void* src2 = (void*)address_src2;
void* dst2 = (void*)address_dst2;
long int address_src3 = address_src + (i+2)*chunk;
long int address_dst3 = address_dst + (i+2)*chunk;
void* src3 = (void*)address_src3;
void* dst3 = (void*)address_dst3;
if(cudaMemcpyAsync(dst1,src1,chunk,cudaMemcpyDeviceToDevice,stream[1])==cudaSuccess && cudaMemcpyAsync(dst1,src1,chunk,cudaMemcpyDeviceToDevice,stream[0])==cudaSuccess )
{
cudaDeviceEnablePeerAccess(tx,0);
cudaMemcpyAsync(dst2,src2,chunk,cudaMemcpyDeviceToDevice,stream[1]);
cudaDeviceDisablePeerAccess(tx);
cudaMemcpyAsync(dst3,src3,chunk,cudaMemcpyDeviceToDevice,stream[0]);
i = i + 1;
}
else if (cudaMemcpyAsync(dst1,src1,chunk,cudaMemcpyDeviceToDevice,stream[0])==cudaSuccess && cudaMemcpyAsync(dst1,src1,chunk,cudaMemcpyDeviceToDevice,stream[1])!=cudaSuccess)
{
cudaDeviceDisablePeerAccess(tx);
cudaMemcpyAsync(dst2,src2,chunk,cudaMemcpyDeviceToDevice,stream[0]);
}
else if (cudaMemcpyAsync(dst1,src1,chunk,cudaMemcpyDeviceToDevice,stream[0])!=cudaSuccess && cudaMemcpyAsync(dst1,src1,chunk,cudaMemcpyDeviceToDevice,stream[1])==cudaSuccess)
{
cudaDeviceEnablePeerAccess(tx,0);
cudaMemcpyAsync(dst2,src2,chunk,cudaMemcpyDeviceToDevice,stream[1]);
}
}
cudaStreamDestroy(stream[0]);
cudaStreamDestroy(stream[1]);
}
}
else if(type == 4){
cudaStream_t stream[2];
cudaStreamCreate (&stream[0]);
cudaStreamCreate (&stream[1]);
long int stream0_size = size*8/16;
long int stream1_size = size*6/16;
long int stream2_size = size*2/16;
long int address_src = (long long unsigned)src;
long int address_src1 = address_src+stream0_size;
long int address_dst = (long long unsigned)dst;
long int address_dst1 = address_dst+stream0_size;
void* src1 = (void*)address_src1;
void* dst1 = (void*)address_dst1;
long int address_src2 = address_src1+stream1_size;
long int address_dst2 = address_dst1+stream1_size;
void* src2 = (void*)address_src2;
void* dst2 = (void*)address_dst2;
cudaMemcpyAsync(dst,src,stream0_size,cudaMemcpyDeviceToDevice,stream[0]);
cudaDeviceEnablePeerAccess(tx,0);
cudaMemcpyAsync(dst1,src1,stream1_size,cudaMemcpyDeviceToDevice,stream[1]);
if (cudaMemcpyAsync(dst1,src1,stream1_size,cudaMemcpyDeviceToDevice,stream[1])==cudaSuccess)
{
cudaDeviceEnablePeerAccess(tx,0);
cudaMemcpyAsync(dst2,src2,stream2_size,cudaMemcpyDeviceToDevice,stream[1]);
}
/*
else if(cudaMemcpyAsync(dst1,src1,stream1_size,cudaMemcpyDeviceToDevice,stream[1])==cudaSuccess && cudaMemcpyAsync(dst,src,stream0_size,cudaMemcpyDeviceToDevice,stream[0])==cudaSuccess)
{
long int address_src3 = address_src2+stream2_size*2/3;
long int address_dst3 = address_dst2+stream2_size*2/3;
void* src3 = (void*)address_src3;
void* dst3 = (void*)address_dst3;
cudaMemcpyAsync(dst2,src2,stream2_size,cudaMemcpyDeviceToDevice,stream[1]);
cudaDeviceDisablePeerAccess(tx);
cudaMemcpyAsync(dst3,src3,stream2_size/3,cudaMemcpyDeviceToDevice,stream[0]);
}*/
else if (cudaMemcpyAsync(dst,src,stream0_size,cudaMemcpyDeviceToDevice,stream[0])==cudaSuccess && cudaMemcpyAsync(dst1,src1,stream1_size,cudaMemcpyDeviceToDevice,stream[1])!=cudaSuccess)
{
cudaDeviceDisablePeerAccess(tx);
cudaMemcpyAsync(dst2,src2,stream2_size,cudaMemcpyDeviceToDevice,stream[0]);
}
}
else if(type == 2){
if(dst == src)
{
printf("stream src and dst addtress should be different.\n");
exit(0);
}
else{
long int stream0_size = size*11/16;
long int stream1_size = size*5/16;
long int address_src = (long long unsigned)src;
long int address_src1 = address_src+stream0_size;
long int address_dst = (long long unsigned)dst;
long int address_dst1 = address_dst+stream0_size;
void* src1 = (void*)address_src1;
void* dst1 = (void*)address_dst1;
cudaSetDevice(tx);
cudaStream_t stream[2];
cudaStreamCreate (&stream[0]);
cudaStreamCreate (&stream[1]);
// cudaDeviceDisablePeerAccess(rx);
// cudaMemcpyAsync(dst1,src1,stream1_size,cudaMemcpyDeviceToDevice,stream[1]);
cudaDeviceEnablePeerAccess(rx,0);
cudaMemcpyAsync(dst,src,stream0_size,cudaMemcpyDeviceToDevice,stream[0]);
cudaDeviceDisablePeerAccess(rx);
cudaMemcpyAsync(dst1,src1,stream1_size,cudaMemcpyDeviceToDevice,stream[1]);
cudaStreamDestroy(stream[0]);
cudaStreamDestroy(stream[1]);
}
}
else if(type == 1){
if(dst == src)
{
printf("stream src and dst address should be different.\n");
exit(0);
}
else{
cudaSetDevice(tx);
cudaDeviceEnablePeerAccess(rx,0);
cudaStream_t stream[1];
cudaStreamCreate (&stream[0]);
cudaMemcpyAsync(dst,src,size,cudaMemcpyDeviceToDevice,stream[0]);
cudaStreamDestroy(stream[0]);
}
}
else if(type == 0){
if(dst == src)
{
printf("stream src and dst address should be different.\n");
exit(0);
}
else{
cudaSetDevice(tx);
// cudaDeviceDisablePeerAccess(rx);
cudaStream_t stream[1];
cudaStreamCreate(&stream[0]);
cudaMemcpyAsync(dst,src,size,cudaMemcpyDeviceToDevice,stream[0]);
cudaStreamDestroy(stream[0]);
}
}
}