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
|
/* ************************************************************************
* Copyright (C) 2016-2022 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell cop-
* ies of the Software, and to permit persons to whom the Software is furnished
* to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IM-
* PLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
* FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
* COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
* IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNE-
* CTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
* ************************************************************************ */
/*
README: Test multiple OpenMP threads
The main thread creates NUM_THREADS handles/streams
Each OpenMP thread calls a rocblas routine asscociated with one handle, one stream
The main thread finally destroy all handles/streams
An alternate valid way is each thread creates its own handle/stream and destroy it
locally.
But in the second way, the handles/streams can not persist across multiple parallel
regions.
In this example, we have two parallel regions
It is NOT recommended that multiple thread share the same rocblas handle.
Yet, it is safe that multiple thread shared the same stream.
If users do not create streams explicitely like what I am doing here,
all rocblas routine take the NULL (0) stream.
*/
#include "rocblas.hpp"
#include "utility.hpp"
#include <cstdio>
#include <cstdlib>
#include <hip/hip_runtime.h>
#include <iostream>
#ifdef _OPENMP
#include <omp.h>
#endif
#include <vector>
#define NUM_THREADS 4
/* ============================================================================================ */
int main()
{
constexpr rocblas_int N = 102400;
float alpha = 10.0;
omp_set_num_threads(NUM_THREADS);
int thread_id;
printf("%d OpenMP threads performing rocblas_scal \n", NUM_THREADS);
// Naming: dX is in GPU (device) memory. hK is in CPU (host) memory, plz follow this practice
std::vector<float> hx(N * NUM_THREADS);
#if 0
std::vector<float> hz(N * NUM_THREADS);
#endif
float *dx, *dy;
double gpu_time_used;
rocblas_handle handles[NUM_THREADS];
hipStream_t streams[NUM_THREADS];
// Create handle/stream have overhead
for(rocblas_int i = 0; i < NUM_THREADS; i++)
{
rocblas_create_handle(&handles[i]);
hipStreamCreate(&streams[i]);
}
// allocate memory on device
hipMalloc(&dx, N * NUM_THREADS * sizeof(float));
hipMalloc(&dy, N * NUM_THREADS * sizeof(float));
// Initial Data on CPU
srand(1);
rocblas_init<float>(hx, 1, N * NUM_THREADS, 1);
#if 0
// copy vector is easy in STL; hz = hx: save a copy in hz which will be output of CPU BLAS
hz = hx;
#endif
hipMemcpy(dx, hx.data(), sizeof(float) * N * NUM_THREADS, hipMemcpyHostToDevice);
printf("N rocblas(us) \n");
gpu_time_used = get_time_us_sync_device(); // in microseconds
// 1st parallel rocblas routine call : scal x
// spawn openmp threads
#pragma omp parallel private(thread_id)
{
thread_id = omp_get_thread_num(); // thread_id from 0,...,NUM_THREADS-1
// associate each handle with a stream
rocblas_set_stream(handles[thread_id], streams[thread_id]);
/* =====================================================================
ROCBLAS template interface
=================================================================== */
rocblas_scal<float>(handles[thread_id], N, &alpha, dx + thread_id * N, 1);
// Blocks until all stream has completed all operations.
hipStreamSynchronize(streams[thread_id]);
}
// 2nd parallel rocblas routine call : copy x to y
// spawn openmp threads
#pragma omp parallel private(thread_id)
{
thread_id = omp_get_thread_num(); // thread_id from 0,...,NUM_THREADS-1
// associate each handle with a stream
rocblas_set_stream(handles[thread_id], streams[thread_id]);
/* =====================================================================
ROCBLAS template interface
=================================================================== */
rocblas_copy<float>(handles[thread_id], N, dx + thread_id * N, 1, dy + thread_id * N, 1);
// Blocks until all stream has completed all operations.
hipStreamSynchronize(streams[thread_id]);
}
gpu_time_used = get_time_us_sync_device() - gpu_time_used;
// copy output from device to CPU
hipMemcpy(hx.data(), dy, sizeof(float) * N * NUM_THREADS, hipMemcpyDeviceToHost);
#if 0
//verify rocblas_scal result
for(rocblas_int i=0;i<N*NUM_THREADS;i++){
if(hz[i] * alpha != hx[i]){
printf("error in element %d: CPU=%f, GPU=%f ", i, hz[i]*alpha, hx[i]);
break;
}
}
#endif
printf("%d %8.2f \n", (int)N * NUM_THREADS, gpu_time_used);
hipFree(dx);
hipFree(dy);
// Destroy handle/streams
for(rocblas_int i = 0; i < NUM_THREADS; i++)
{
rocblas_destroy_handle(handles[i]);
hipStreamDestroy(streams[i]);
}
return 0;
}
|
