| 1 |
/* |
| 2 |
* Copyright (c) Meta Platforms, Inc. and affiliates. |
| 3 |
* |
| 4 |
* This source code is licensed under the MIT license found in the |
| 5 |
* LICENSE file in the root directory of this source tree. |
| 6 |
*/ |
| 7 |
|
| 8 |
#include <cstdio> |
| 9 |
#include <cstdlib> |
| 10 |
#include <random> |
| 11 |
|
| 12 |
#undef FINTEGER |
| 13 |
#define FINTEGER long |
| 14 |
|
| 15 |
extern "C" { |
| 16 |
|
| 17 |
/* declare BLAS functions, see http://www.netlib.org/clapack/cblas/ */ |
| 18 |
|
| 19 |
int sgemm_( |
| 20 |
const char* transa, |
| 21 |
const char* transb, |
| 22 |
FINTEGER* m, |
| 23 |
FINTEGER* n, |
| 24 |
FINTEGER* k, |
| 25 |
const float* alpha, |
| 26 |
const float* a, |
| 27 |
FINTEGER* lda, |
| 28 |
const float* b, |
| 29 |
FINTEGER* ldb, |
| 30 |
float* beta, |
| 31 |
float* c, |
| 32 |
FINTEGER* ldc); |
| 33 |
|
| 34 |
/* Lapack functions, see http://www.netlib.org/clapack/old/single/sgeqrf.c */ |
| 35 |
|
| 36 |
int sgeqrf_( |
| 37 |
FINTEGER* m, |
| 38 |
FINTEGER* n, |
| 39 |
float* a, |
| 40 |
FINTEGER* lda, |
| 41 |
float* tau, |
| 42 |
float* work, |
| 43 |
FINTEGER* lwork, |
| 44 |
FINTEGER* info); |
| 45 |
} |
| 46 |
|
| 47 |
float* new_random_vec(int size) { |
| 48 |
float* x = new float[size]; |
| 49 |
std::mt19937 rng; |
| 50 |
std::uniform_real_distribution<> distrib; |
| 51 |
for (int i = 0; i < size; i++) |
| 52 |
x[i] = distrib(rng); |
| 53 |
return x; |
| 54 |
} |
| 55 |
|
| 56 |
int main() { |
| 57 |
FINTEGER m = 10, n = 20, k = 30; |
| 58 |
float *a = new_random_vec(m * k), *b = new_random_vec(n * k), |
| 59 |
*c = new float[n * m]; |
| 60 |
float one = 1.0, zero = 0.0; |
| 61 |
|
| 62 |
printf("BLAS test\n"); |
| 63 |
|
| 64 |
sgemm_("Not transposed", |
| 65 |
"Not transposed", |
| 66 |
&m, |
| 67 |
&n, |
| 68 |
&k, |
| 69 |
&one, |
| 70 |
a, |
| 71 |
&m, |
| 72 |
b, |
| 73 |
&k, |
| 74 |
&zero, |
| 75 |
c, |
| 76 |
&m); |
| 77 |
|
| 78 |
printf("errors=\n"); |
| 79 |
|
| 80 |
for (int i = 0; i < m; i++) { |
| 81 |
for (int j = 0; j < n; j++) { |
| 82 |
float accu = 0; |
| 83 |
for (int l = 0; l < k; l++) |
| 84 |
accu += a[i + l * m] * b[l + j * k]; |
| 85 |
printf("%6.3f ", accu - c[i + j * m]); |
| 86 |
} |
| 87 |
printf("\n"); |
| 88 |
} |
| 89 |
|
| 90 |
long info = 0x64bL << 32; |
| 91 |
long mi = 0x64bL << 32 | m; |
| 92 |
float* tau = new float[m]; |
| 93 |
FINTEGER lwork = -1; |
| 94 |
|
| 95 |
float work1; |
| 96 |
|
| 97 |
printf("Intentional Lapack error (appears only for 64-bit INTEGER):\n"); |
| 98 |
sgeqrf_(&mi, &n, c, &m, tau, &work1, &lwork, (FINTEGER*)&info); |
| 99 |
|
| 100 |
// sgeqrf_ (&m, &n, c, &zeroi, tau, &work1, &lwork, (FINTEGER*)&info); |
| 101 |
printf("info=%016lx\n", info); |
| 102 |
|
| 103 |
if (info >> 32 == 0x64b) { |
| 104 |
printf("Lapack uses 32-bit integers\n"); |
| 105 |
} else { |
| 106 |
printf("Lapack uses 64-bit integers\n"); |
| 107 |
} |
| 108 |
|
| 109 |
return 0; |
| 110 |
} |