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
|
/* ************************************************************************
* Copyright (C) 2016-2024 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.
*
* ************************************************************************ */
#include "client_utility.hpp"
#include <hip/hip_runtime.h>
#include <rocblas/rocblas.h>
#define DIM1 1023
#define DIM2 1024
#define DIM3 1025
template <typename T>
void mat_mat_mult(T alpha,
T beta,
int M,
int N,
int K,
const T* A,
int As1,
int As2,
const T* B,
int Bs1,
int Bs2,
T* C,
int Cs1,
int Cs2)
{
for(int i1 = 0; i1 < M; i1++)
{
for(int i2 = 0; i2 < N; i2++)
{
T t = 0.0;
for(int i3 = 0; i3 < K; i3++)
{
t += A[i1 * As1 + i3 * As2] * B[i3 * Bs1 + i2 * Bs2];
}
C[i1 * Cs1 + i2 * Cs2] = beta * C[i1 * Cs1 + i2 * Cs2] + alpha * t;
}
}
}
int main()
{
rocblas_operation transa = rocblas_operation_none, transb = rocblas_operation_transpose;
float alpha = 1.1, beta = 0.9;
rocblas_int m = DIM1, n = DIM2, k = DIM3;
rocblas_int lda, ldb, ldc, size_a, size_b, size_c;
int a_stride_1, a_stride_2, b_stride_1, b_stride_2;
rocblas_cout << "sgemm example" << std::endl;
if(transa == rocblas_operation_none)
{
lda = m;
size_a = k * lda;
a_stride_1 = 1;
a_stride_2 = lda;
rocblas_cout << "N";
}
else
{
lda = k;
size_a = m * lda;
a_stride_1 = lda;
a_stride_2 = 1;
rocblas_cout << "T";
}
if(transb == rocblas_operation_none)
{
ldb = k;
size_b = n * ldb;
b_stride_1 = 1;
b_stride_2 = ldb;
rocblas_cout << "N: ";
}
else
{
ldb = n;
size_b = k * ldb;
b_stride_1 = ldb;
b_stride_2 = 1;
rocblas_cout << "T: ";
}
ldc = m;
size_c = n * ldc;
// Naming: da is in GPU (device) memory. ha is in CPU (host) memory
std::vector<float> ha(size_a);
std::vector<float> hb(size_b);
std::vector<float> hc(size_c);
std::vector<float> hc_gold(size_c);
// initial data on host
srand(1);
for(int i = 0; i < size_a; ++i)
{
ha[i] = rand() % 17;
}
for(int i = 0; i < size_b; ++i)
{
hb[i] = rand() % 17;
}
for(int i = 0; i < size_c; ++i)
{
hc[i] = rand() % 17;
}
hc_gold = hc;
// allocate memory on device
float *da, *db, *dc;
CHECK_HIP_ERROR(hipMalloc(&da, size_a * sizeof(float)));
CHECK_HIP_ERROR(hipMalloc(&db, size_b * sizeof(float)));
CHECK_HIP_ERROR(hipMalloc(&dc, size_c * sizeof(float)));
// copy matrices from host to device
CHECK_HIP_ERROR(hipMemcpy(da, ha.data(), sizeof(float) * size_a, hipMemcpyHostToDevice));
CHECK_HIP_ERROR(hipMemcpy(db, hb.data(), sizeof(float) * size_b, hipMemcpyHostToDevice));
CHECK_HIP_ERROR(hipMemcpy(dc, hc.data(), sizeof(float) * size_c, hipMemcpyHostToDevice));
rocblas_handle handle;
CHECK_ROCBLAS_ERROR(rocblas_create_handle(&handle));
CHECK_ROCBLAS_ERROR(
rocblas_sgemm(handle, transa, transb, m, n, k, &alpha, da, lda, db, ldb, &beta, dc, ldc));
// copy output from device to CPU
CHECK_HIP_ERROR(hipMemcpy(hc.data(), dc, sizeof(float) * size_c, hipMemcpyDeviceToHost));
rocblas_cout << "m, n, k, lda, ldb, ldc = " << m << ", " << n << ", " << k << ", " << lda
<< ", " << ldb << ", " << ldc << std::endl;
float max_relative_error = std::numeric_limits<float>::min();
// calculate golden or correct result
mat_mat_mult<float>(alpha,
beta,
m,
n,
k,
ha.data(),
a_stride_1,
a_stride_2,
hb.data(),
b_stride_1,
b_stride_2,
hc_gold.data(),
1,
ldc);
for(int i = 0; i < size_c; i++)
{
float relative_error = (hc_gold[i] - hc[i]) / hc_gold[i];
relative_error = relative_error > 0 ? relative_error : -relative_error;
max_relative_error
= relative_error < max_relative_error ? max_relative_error : relative_error;
}
float eps = std::numeric_limits<float>::epsilon();
float tolerance = 10;
if(max_relative_error != max_relative_error || max_relative_error > eps * tolerance)
{
rocblas_cout << "FAIL: max_relative_error = " << max_relative_error << std::endl;
}
else
{
rocblas_cout << "PASS: max_relative_error = " << max_relative_error << std::endl;
}
CHECK_HIP_ERROR(hipFree(da));
CHECK_HIP_ERROR(hipFree(db));
CHECK_HIP_ERROR(hipFree(dc));
CHECK_ROCBLAS_ERROR(rocblas_destroy_handle(handle));
return EXIT_SUCCESS;
}
|
