Skip to content
Navigation Menu
{{ message }}
-
Notifications
You must be signed in to change notification settings - Fork 84
Expand file tree
/
Copy pathmultinode_testcases.cpp
More file actions
371 lines (287 loc) · 17 KB
/
Copy pathmultinode_testcases.cpp
File metadata and controls
371 lines (287 loc) · 17 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
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
/*
* SPDX-FileCopyrightText: Copyright (c) 2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <cuda.h>
#include "testcase.h"
#include "memcpy.h"
#include "common.h"
#include "output.h"
#ifdef MULTINODE
#include <mpi.h>
#include "multinode_memcpy.h"
// DtoD Read test - copy from dst to src (backwards) using src contxt
void MultinodeDeviceToDeviceReadCE::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(worldSize, worldSize, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorCE(), new NodeHelperMulti(), PREFER_DST_CONTEXT);
auto pairs = generateTestPairs(worldSize, targetNumPairs);
for (const auto& pair : pairs) {
int srcDeviceId = pair.first;
int peerDeviceId = pair.second;
MultinodeDeviceBufferUnicast srcNode(size, srcDeviceId);
MultinodeDeviceBufferUnicast peerNode(size, peerDeviceId);
// swap src and peer nodes, but use srcNodes (the copy's destination) context
bandwidthValues.value(srcDeviceId, peerDeviceId) = memcpyInstance.doMemcpy(peerNode, srcNode);
}
output->addTestcaseResults(bandwidthValues, "memcpy CE GPU(row) -> GPU(column) bandwidth (GB/s)");
}
// DtoD Write test - copy from src to dst using src context
void MultinodeDeviceToDeviceWriteCE::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(worldSize, worldSize, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorCE(), new NodeHelperMulti());
auto pairs = generateTestPairs(worldSize, targetNumPairs);
for (const auto& pair : pairs) {
int srcDeviceId = pair.first;
int peerDeviceId = pair.second;
MultinodeDeviceBufferUnicast srcNode(size, srcDeviceId);
MultinodeDeviceBufferUnicast peerNode(size, peerDeviceId);
bandwidthValues.value(srcDeviceId, peerDeviceId) = memcpyInstance.doMemcpy(srcNode, peerNode);
}
output->addTestcaseResults(bandwidthValues, "memcpy CE GPU(row) <- GPU(column) bandwidth (GB/s)");
}
// DtoD Bidir Read test - copy from dst to src (backwards) using src contxt
void MultinodeDeviceToDeviceBidirReadCE::run(unsigned long long size, unsigned long long loopCount) {
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorCE(), new NodeHelperMulti(), PREFER_DST_CONTEXT, MemcpyOperation::VECTOR_BW);
PeerValueMatrix<double> bandwidthValuesRead1(worldSize, worldSize, key + "_read1");
PeerValueMatrix<double> bandwidthValuesRead2(worldSize, worldSize, key + "_read2");
PeerValueMatrix<double> bandwidthValuesTotal(worldSize, worldSize, key + "_total");
auto pairs = generateTestPairs(worldSize, targetNumPairs);
for (const auto& pair : pairs) {
int srcDeviceId = pair.first;
int peerDeviceId = pair.second;
// Double the size of the interference copy to ensure it interferes correctly
MultinodeDeviceBufferUnicast src1(size, srcDeviceId), src2(size, srcDeviceId);
MultinodeDeviceBufferUnicast peer1(size, peerDeviceId), peer2(size, peerDeviceId);
// swap src and peer nodes, but use srcNodes (the copy's destination) context
std::vector<const MemcpyBuffer*> srcNodes = {&peer1, &src2};
std::vector<const MemcpyBuffer*> peerNodes = {&src1, &peer2};
auto results = memcpyInstance.doMemcpyVector(srcNodes, peerNodes);
bandwidthValuesRead1.value(srcDeviceId, peerDeviceId) = results[0];
bandwidthValuesRead2.value(srcDeviceId, peerDeviceId) = results[1];
bandwidthValuesTotal.value(srcDeviceId, peerDeviceId) = results[0] + results[1];
}
output->addTestcaseResults(bandwidthValuesRead1, "memcpy CE CPU(row) <-> GPU(column) Read1 bandwidth (GB/s)");
output->addTestcaseResults(bandwidthValuesRead2, "memcpy CE CPU(row) <-> GPU(column) Read2 bandwidth (GB/s)");
output->addTestcaseResults(bandwidthValuesTotal, "memcpy CE CPU(row) <-> GPU(column) Total bandwidth (GB/s)");
}
// DtoD Bidir Write test - copy from src to dst using src context
void MultinodeDeviceToDeviceBidirWriteCE::run(unsigned long long size, unsigned long long loopCount) {
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorCE(), new NodeHelperMulti(), PREFER_SRC_CONTEXT, MemcpyOperation::VECTOR_BW);
PeerValueMatrix<double> bandwidthValuesWrite1(worldSize, worldSize, key + "_write1");
PeerValueMatrix<double> bandwidthValuesWrite2(worldSize, worldSize, key + "_write2");
PeerValueMatrix<double> bandwidthValuesTotal(worldSize, worldSize, key + "_total");
auto pairs = generateTestPairs(worldSize, targetNumPairs);
for (const auto& pair : pairs) {
int srcDeviceId = pair.first;
int peerDeviceId = pair.second;
// Double the size of the interference copy to ensure it interferes correctly
MultinodeDeviceBufferUnicast src1(size, srcDeviceId), src2(size, srcDeviceId);
MultinodeDeviceBufferUnicast peer1(size, peerDeviceId), peer2(size, peerDeviceId);
std::vector<const MemcpyBuffer*> srcNodes = {&src1, &peer2};
std::vector<const MemcpyBuffer*> peerNodes = {&peer1, &src2};
auto results = memcpyInstance.doMemcpyVector(srcNodes, peerNodes);
bandwidthValuesWrite1.value(srcDeviceId, peerDeviceId) = results[0];
bandwidthValuesWrite2.value(srcDeviceId, peerDeviceId) = results[1];
bandwidthValuesTotal.value(srcDeviceId, peerDeviceId) = results[0] + results[1];
}
output->addTestcaseResults(bandwidthValuesWrite1, "memcpy CE CPU(row) <-> GPU(column) Read1 bandwidth (GB/s)");
output->addTestcaseResults(bandwidthValuesWrite2, "memcpy CE CPU(row) <-> GPU(column) Read2 bandwidth (GB/s)");
output->addTestcaseResults(bandwidthValuesTotal, "memcpy CE CPU(row) <-> GPU(column) Total bandwidth (GB/s)");
}
// DtoD Read test - copy from dst to src (backwards) using src contxt
void MultinodeDeviceToDeviceReadSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(worldSize, worldSize, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorSM(), new NodeHelperMulti(), PREFER_DST_CONTEXT);
auto pairs = generateTestPairs(worldSize, targetNumPairs);
for (const auto& pair : pairs) {
int srcDeviceId = pair.first;
int peerDeviceId = pair.second;
MultinodeDeviceBufferUnicast srcNode(size, srcDeviceId);
MultinodeDeviceBufferUnicast peerNode(size, peerDeviceId);
// swap src and peer nodes, but use srcNodes (the copy's destination) context
bandwidthValues.value(srcDeviceId, peerDeviceId) = memcpyInstance.doMemcpy(peerNode, srcNode);
}
output->addTestcaseResults(bandwidthValues, "memcpy CE GPU(row) -> GPU(column) bandwidth (GB/s)");
}
// DtoD Write test - copy from src to dst using src context
void MultinodeDeviceToDeviceWriteSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(worldSize, worldSize, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorSM(), new NodeHelperMulti());
auto pairs = generateTestPairs(worldSize, targetNumPairs);
for (const auto& pair : pairs) {
int srcDeviceId = pair.first;
int peerDeviceId = pair.second;
MultinodeDeviceBufferUnicast srcNode(size, srcDeviceId);
MultinodeDeviceBufferUnicast peerNode(size, peerDeviceId);
bandwidthValues.value(srcDeviceId, peerDeviceId) = memcpyInstance.doMemcpy(srcNode, peerNode);
}
output->addTestcaseResults(bandwidthValues, "memcpy SM GPU(row) <- GPU(column) bandwidth (GB/s)");
}
// DtoD Bidir Read test - copy from dst to src (backwards) using src contxt
void MultinodeDeviceToDeviceBidirReadSM::run(unsigned long long size, unsigned long long loopCount) {
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorSM(), new NodeHelperMulti(), PREFER_DST_CONTEXT, MemcpyOperation::VECTOR_BW);
PeerValueMatrix<double> bandwidthValuesRead1(worldSize, worldSize, key + "_read1");
PeerValueMatrix<double> bandwidthValuesRead2(worldSize, worldSize, key + "_read2");
PeerValueMatrix<double> bandwidthValuesTotal(worldSize, worldSize, key + "_total");
auto pairs = generateTestPairs(worldSize, targetNumPairs);
for (const auto& pair : pairs) {
int srcDeviceId = pair.first;
int peerDeviceId = pair.second;
MultinodeDeviceBufferUnicast src1(size, srcDeviceId), src2(size, srcDeviceId);
MultinodeDeviceBufferUnicast peer1(size, peerDeviceId), peer2(size, peerDeviceId);
// swap src and peer nodes, but use srcNodes (the copy's destination) context
std::vector<const MemcpyBuffer*> srcNodes = {&peer1, &src2};
std::vector<const MemcpyBuffer*> peerNodes = {&src1, &peer2};
auto results = memcpyInstance.doMemcpyVector(srcNodes, peerNodes);
bandwidthValuesRead1.value(srcDeviceId, peerDeviceId) = results[0];
bandwidthValuesRead2.value(srcDeviceId, peerDeviceId) = results[1];
bandwidthValuesTotal.value(srcDeviceId, peerDeviceId) = results[0] + results[1];
}
output->addTestcaseResults(bandwidthValuesRead1, "memcpy SM CPU(row) <-> GPU(column) Read1 bandwidth (GB/s)");
output->addTestcaseResults(bandwidthValuesRead2, "memcpy SM CPU(row) <-> GPU(column) Read2 bandwidth (GB/s)");
output->addTestcaseResults(bandwidthValuesTotal, "memcpy SM CPU(row) <-> GPU(column) Total bandwidth (GB/s)");
}
// DtoD Bidir Write test - copy from src to dst using src context
void MultinodeDeviceToDeviceBidirWriteSM::run(unsigned long long size, unsigned long long loopCount) {
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorSM(), new NodeHelperMulti(), PREFER_SRC_CONTEXT, MemcpyOperation::VECTOR_BW);
PeerValueMatrix<double> bandwidthValuesWrite1(worldSize, worldSize, key + "_write1");
PeerValueMatrix<double> bandwidthValuesWrite2(worldSize, worldSize, key + "_write2");
PeerValueMatrix<double> bandwidthValuesTotal(worldSize, worldSize, key + "_total");
auto pairs = generateTestPairs(worldSize, targetNumPairs);
for (const auto& pair : pairs) {
int srcDeviceId = pair.first;
int peerDeviceId = pair.second;
MultinodeDeviceBufferUnicast src1(size, srcDeviceId), src2(size, srcDeviceId);
MultinodeDeviceBufferUnicast peer1(size, peerDeviceId), peer2(size, peerDeviceId);
std::vector<const MemcpyBuffer*> srcNodes = {&src1, &peer2};
std::vector<const MemcpyBuffer*> peerNodes = {&peer1, &src2};
auto results = memcpyInstance.doMemcpyVector(srcNodes, peerNodes);
bandwidthValuesWrite1.value(srcDeviceId, peerDeviceId) = results[0];
bandwidthValuesWrite2.value(srcDeviceId, peerDeviceId) = results[1];
bandwidthValuesTotal.value(srcDeviceId, peerDeviceId) = results[0] + results[1];
}
output->addTestcaseResults(bandwidthValuesWrite1, "memcpy SM CPU(row) <-> GPU(column) Write1 bandwidth (GB/s)");
output->addTestcaseResults(bandwidthValuesWrite2, "memcpy SM CPU(row) <-> GPU(column) Write2 bandwidth (GB/s)");
output->addTestcaseResults(bandwidthValuesTotal, "memcpy SM CPU(row) <-> GPU(column) Total bandwidth (GB/s)");
}
void MultinodeAllToOneWriteSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, worldSize, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorSM(), new NodeHelperMulti(), PREFER_SRC_CONTEXT, MemcpyOperation::SUM_BW);
for (int dstDeviceId = 0; dstDeviceId < worldSize; dstDeviceId++) {
std::vector<const MemcpyBuffer*> srcNodes;
std::vector<const MemcpyBuffer*> dstNodes;
for (int srcDeviceId = 0; srcDeviceId < worldSize; srcDeviceId++) {
if (dstDeviceId == srcDeviceId) {
continue;
}
srcNodes.push_back(new MultinodeDeviceBufferLocal(size, srcDeviceId));
dstNodes.push_back(new MultinodeDeviceBufferUnicast(size, dstDeviceId));
}
bandwidthValues.value(0, dstDeviceId) = memcpyInstance.doMemcpy(srcNodes, dstNodes);
for (auto node : dstNodes) {
delete node;
}
for (auto node : srcNodes) {
delete node;
}
}
output->addTestcaseResults(bandwidthValues, "memcpy SM All Gpus -> GPU(column) total bandwidth (GB/s)");
}
void MultinodeAllFromOneReadSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, worldSize, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorSM(), new NodeHelperMulti(), PREFER_DST_CONTEXT, MemcpyOperation::SUM_BW);
for (int srcDeviceId = 0; srcDeviceId < worldSize; srcDeviceId++) {
std::vector<const MemcpyBuffer*> srcNodes;
std::vector<const MemcpyBuffer*> dstNodes;
for (int dstDeviceId = 0; dstDeviceId < worldSize; dstDeviceId++) {
if (dstDeviceId == srcDeviceId) {
continue;
}
srcNodes.push_back(new MultinodeDeviceBufferUnicast(size, srcDeviceId));
dstNodes.push_back(new MultinodeDeviceBufferLocal(size, dstDeviceId));
}
bandwidthValues.value(0, srcDeviceId) = memcpyInstance.doMemcpy(srcNodes, dstNodes);
for (auto node : dstNodes) {
delete node;
}
for (auto node : srcNodes) {
delete node;
}
}
output->addTestcaseResults(bandwidthValues, "memcpy SM All Gpus <- GPU(column) total bandwidth (GB/s)");
}
void MultinodeBroadcastOneToAllSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, worldSize, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorMulticastWrite(), new NodeHelperMulti(), PREFER_DST_CONTEXT, MemcpyOperation::SUM_BW);
for (int dstDeviceId = 0; dstDeviceId < worldSize; dstDeviceId++) {
std::vector<const MemcpyBuffer*> srcNodes;
std::vector<const MemcpyBuffer*> dstNodes;
srcNodes.push_back(new MultinodeDeviceBufferLocal(size, dstDeviceId));
dstNodes.push_back(new MultinodeDeviceBufferMulticast(size, dstDeviceId));
bandwidthValues.value(0, dstDeviceId) = memcpyInstance.doMemcpy(srcNodes, dstNodes);
for (auto node : dstNodes) {
delete node;
}
for (auto node : srcNodes) {
delete node;
}
}
output->addTestcaseResults(bandwidthValues, "multicast SM GPU(column) -> All Gpus total bandwidth (GB/s)");
}
void MultinodeBroadcastAllToAllSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, 1, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorMulticastWrite(), new NodeHelperMulti(), PREFER_DST_CONTEXT, MemcpyOperation::SUM_BW);
std::vector<const MemcpyBuffer*> srcNodes;
std::vector<const MemcpyBuffer*> dstNodes;
for (int dstDeviceId = 0; dstDeviceId < worldSize; dstDeviceId++) {
srcNodes.push_back(new MultinodeDeviceBufferLocal(size, dstDeviceId));
dstNodes.push_back(new MultinodeDeviceBufferMulticast(size, dstDeviceId));
}
bandwidthValues.value(0, 0) = memcpyInstance.doMemcpy(srcNodes, dstNodes);
for (auto node : dstNodes) {
delete node;
}
for (auto node : srcNodes) {
delete node;
}
output->addTestcaseResults(bandwidthValues, "multicast SM All -> All Gpus total bandwidth (GB/s)");
}
void MultinodeBisectWriteCE::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(worldSize, 1, key);
MemcpyOperation memcpyInstance(loopCount, new MemcpyInitiatorCE(), new NodeHelperMulti(), PREFER_SRC_CONTEXT, MemcpyOperation::VECTOR_BW);
std::vector<std::string> rowLabels;
std::vector<const MemcpyBuffer*> srcNodes, dstNodes;
for (int i = 0; i < worldSize; i++) {
int peer = (i + worldSize / 2) % worldSize;
srcNodes.push_back(new MultinodeDeviceBufferUnicast(size, i));
dstNodes.push_back(new MultinodeDeviceBufferUnicast(size, peer));
std::stringstream s;
s << getPaddedProcessId(i) << "->" << getPaddedProcessId(peer);
rowLabels.push_back(s.str());
}
auto results = memcpyInstance.doMemcpyVector(dstNodes, srcNodes);
for (int i = 0; i < results.size(); i++) {
bandwidthValues.value(i, 0) = results[i];
}
bandwidthValues.setRowLabels(rowLabels);
for (auto node : dstNodes) {
delete node;
}
for (auto node : srcNodes) {
delete node;
}
output->addTestcaseResults(bandwidthValues, "Bisect benchmarking, simultaneous write CE BW");
}
#endif
You can’t perform that action at this time.