forked from pytorch/FBGEMM
-
Notifications
You must be signed in to change notification settings - Fork 0
/
I64Benchmark.cc
136 lines (123 loc) · 3.1 KB
/
I64Benchmark.cc
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
/*
* Copyright (c) Facebook, Inc. and its affiliates.
* All rights reserved.
* This source code is licensed under the BSD-style license found in the
* LICENSE file in the root directory of this source tree.
*/
#include <array>
#include <iostream>
#include <limits>
#include <random>
#include "./BenchUtils.h"
#include "fbgemm/FbgemmI64.h"
#include "fbgemm/Utils.h"
#include "src/RefImplementations.h"
using namespace std;
using namespace fbgemm;
int main() {
vector<char> llc(128 * 1024 * 1024);
// clang-format off
const vector<array<int, 3>> shapes = {
{1024, 1024, 1024}
};
// clang-format on
for (const auto s : shapes) {
const int m = s[0];
const int n = s[1];
const int k = s[2];
cout << "m " << m << " n " << n << " k " << k << endl;
aligned_vector<int64_t> A(m * k);
aligned_vector<int64_t> B(k * n);
aligned_vector<int64_t> C(m * n);
aligned_vector<int64_t> C_ref = C;
randFill(
A, numeric_limits<int64_t>::lowest(), numeric_limits<int64_t>::max());
randFill(
B, numeric_limits<int64_t>::lowest(), numeric_limits<int64_t>::max());
cblas_gemm_i64_i64acc_ref(
matrix_op_t::NoTranspose,
matrix_op_t::NoTranspose,
m,
n,
k,
A.data(),
k,
B.data(),
n,
false, /* accumulation*/
C_ref.data(),
n);
constexpr int NWARMUP = 4;
constexpr int NITER = 16;
double ttot = measureWithWarmup(
[&]() {
cblas_gemm_i64_i64acc(
matrix_op_t::NoTranspose,
matrix_op_t::NoTranspose,
m,
n,
k,
A.data(),
k,
B.data(),
n,
false, /* accumulation*/
C.data(),
n);
},
NWARMUP,
NITER,
[&]() { llc_flush(llc); });
const double ops = 2.0 * m * n * k;
cout << "Gops/s = " << ops / ttot / 1e9 << endl;
compare_buffers(C_ref.data(), C.data(), m, n, n, 5);
cblas_gemm_i64_i64acc_ref(
matrix_op_t::NoTranspose,
matrix_op_t::NoTranspose,
m,
n,
k,
A.data(),
k,
B.data(),
n,
true, /* accumulation*/
C_ref.data(),
n);
aligned_vector<int64_t> C_acc = C;
cblas_gemm_i64_i64acc(
matrix_op_t::NoTranspose,
matrix_op_t::NoTranspose,
m,
n,
k,
A.data(),
k,
B.data(),
n,
true, /* accumulation*/
C_acc.data(),
n);
ttot = measureWithWarmup(
[&]() {
cblas_gemm_i64_i64acc(
matrix_op_t::NoTranspose,
matrix_op_t::NoTranspose,
m,
n,
k,
A.data(),
k,
B.data(),
n,
true, /* accumulation*/
C.data(),
n);
},
NWARMUP,
NITER,
[&]() { llc_flush(llc); });
cout << "Gops/s = " << ops / ttot / 1e9 << endl;
compare_buffers(C_ref.data(), C_acc.data(), m, n, n, 5);
}
}