forked from NVIDIA/nvbandwidth
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtestcases_sm.cpp
More file actions
222 lines (173 loc) · 9.77 KB
/
testcases_sm.cpp
File metadata and controls
222 lines (173 loc) · 9.77 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
/*
* 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 "testcase.h"
#include "kernels.cuh"
#include "memcpy.h"
void HostToDeviceSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount);
for (int deviceId = 0; deviceId < deviceCount; deviceId++) {
HostNode hostNode(size, deviceId);
DeviceNode deviceNode(size, deviceId);
bandwidthValues.value(0, deviceId) = memcpyInstance.doMemcpy(hostNode, deviceNode);
}
std::cout << "memcpy SM CPU(row) -> GPU(column) bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
void DeviceToHostSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount);
for (int deviceId = 0; deviceId < deviceCount; deviceId++) {
HostNode hostNode(size, deviceId);
DeviceNode deviceNode(size, deviceId);
bandwidthValues.value(0, deviceId) = memcpyInstance.doMemcpy(deviceNode, hostNode);
}
std::cout << "memcpy SM CPU(row) <- GPU(column) bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
// DtoD Read test - copy from dst to src (backwards) using src contxt
void DeviceToDeviceReadSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(deviceCount, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount, MemcpyOperation::PREFER_DST_CONTEXT);
for (int srcDeviceId = 0; srcDeviceId < deviceCount; srcDeviceId++) {
for (int peerDeviceId = 0; peerDeviceId < deviceCount; peerDeviceId++) {
if (peerDeviceId == srcDeviceId) {
continue;
}
DeviceNode srcNode(size, srcDeviceId);
DeviceNode peerNode(size, peerDeviceId);
if (!srcNode.enablePeerAcess(peerNode)) {
continue;
}
// swap src and peer nodes, but use srcNodes (the copy's destination) context
bandwidthValues.value(srcDeviceId, peerDeviceId) = memcpyInstance.doMemcpy(peerNode, srcNode);
}
}
std::cout << "memcpy CE GPU(row) -> GPU(column) bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
// DtoD Write test - copy from src to dst using src context
void DeviceToDeviceWriteSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(deviceCount, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount);
for (int srcDeviceId = 0; srcDeviceId < deviceCount; srcDeviceId++) {
for (int peerDeviceId = 0; peerDeviceId < deviceCount; peerDeviceId++) {
if (peerDeviceId == srcDeviceId) {
continue;
}
DeviceNode srcNode(size, srcDeviceId);
DeviceNode peerNode(size, peerDeviceId);
if (!srcNode.enablePeerAcess(peerNode)) {
continue;
}
bandwidthValues.value(srcDeviceId, peerDeviceId) = memcpyInstance.doMemcpy(srcNode, peerNode);
}
}
std::cout << "memcpy SM GPU(row) <- GPU(column) bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
// DtoD Bidir Read test - copy from dst to src (backwards) using src contxt
void DeviceToDeviceBidirReadSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(deviceCount, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount, MemcpyOperation::PREFER_DST_CONTEXT);
for (int srcDeviceId = 0; srcDeviceId < deviceCount; srcDeviceId++) {
for (int peerDeviceId = 0; peerDeviceId < deviceCount; peerDeviceId++) {
if (peerDeviceId == srcDeviceId) {
continue;
}
DeviceNode src1(size, srcDeviceId), src2(size, srcDeviceId);
DeviceNode peer1(size, peerDeviceId), peer2(size, peerDeviceId);
if (!src1.enablePeerAcess(peer1)) {
continue;
}
// swap src and peer nodes, but use srcNodes (the copy's destination) context
std::vector<const MemcpyNode*> srcNodes = {&peer1, &src2};
std::vector<const MemcpyNode*> peerNodes = {&src1, &peer2};
bandwidthValues.value(srcDeviceId, peerDeviceId) = memcpyInstance.doMemcpy(srcNodes, peerNodes);
}
}
std::cout << "memcpy SM GPU(row) -> GPU(column) bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
// DtoD Bidir Write test - copy from src to dst using src context
void DeviceToDeviceBidirWriteSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(deviceCount, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount);
for (int srcDeviceId = 0; srcDeviceId < deviceCount; srcDeviceId++) {
for (int peerDeviceId = 0; peerDeviceId < deviceCount; peerDeviceId++) {
if (peerDeviceId == srcDeviceId) {
continue;
}
DeviceNode src1(size, srcDeviceId), src2(size, srcDeviceId);
DeviceNode peer1(size, peerDeviceId), peer2(size, peerDeviceId);
if (!src1.enablePeerAcess(peer1)) {
continue;
}
std::vector<const MemcpyNode*> srcNodes = {&src1, &peer2};
std::vector<const MemcpyNode*> peerNodes = {&peer1, &src2};
bandwidthValues.value(srcDeviceId, peerDeviceId) = memcpyInstance.doMemcpy(srcNodes, peerNodes);
}
}
std::cout << "memcpy SM GPU(row) <- GPU(column) bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
void AllToHostSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount, MemcpyOperation::PREFER_SRC_CONTEXT, MemcpyOperation::USE_FIRST_BW);
allHostHelper(size, memcpyInstance, bandwidthValues, false);
std::cout << "memcpy SM CPU(row) <- GPU(column) bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
void HostToAllSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount, MemcpyOperation::PREFER_SRC_CONTEXT, MemcpyOperation::USE_FIRST_BW);
allHostHelper(size, memcpyInstance, bandwidthValues, true);
std::cout << "memcpy SM CPU(row) -> GPU(column) bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
// Write test - copy from src to dst using src context
void AllToOneWriteSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount, MemcpyOperation::PREFER_SRC_CONTEXT, MemcpyOperation::TOTAL_BW);
allToOneHelper(size, memcpyInstance, bandwidthValues, false);
std::cout << "memcpy SM All Gpus -> GPU(column) total bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
// Read test - copy from dst to src (backwards) using src contxt
void AllToOneReadSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount, MemcpyOperation::PREFER_DST_CONTEXT, MemcpyOperation::TOTAL_BW);
allToOneHelper(size, memcpyInstance, bandwidthValues, true);
std::cout << "memcpy SM All GPUs <- GPU(column) total bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
// Write test - copy from src to dst using src context
void OneToAllWriteSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount, MemcpyOperation::PREFER_SRC_CONTEXT, MemcpyOperation::TOTAL_BW);
oneToAllHelper(size, memcpyInstance, bandwidthValues, false);
std::cout << "memcpy SM GPU(column) -> All GPUs total bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}
// Read test - copy from dst to src (backwards) using src contxt
void OneToAllReadSM::run(unsigned long long size, unsigned long long loopCount) {
PeerValueMatrix<double> bandwidthValues(1, deviceCount, key);
MemcpyOperationSM memcpyInstance(loopCount, MemcpyOperation::PREFER_DST_CONTEXT, MemcpyOperation::TOTAL_BW);
oneToAllHelper(size, memcpyInstance, bandwidthValues, true);
std::cout << "memcpy SM GPU(column) <- All GPUs total bandwidth (GB/s)" << std::endl;
std::cout << std::fixed << std::setprecision(2) << bandwidthValues << std::endl;
}