场景说明
执行两个输入张量a，b以及参数结构体M，执行两个矩阵的乘法运算，将计算结果存储于输出张量output。例如：
| a: [1,2,
3,4]
b: [5,6,
7,8]
output: [19,22,
43,50]
|
代码示例
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 | #include <cmath>
#include <random>
#include <cstdint>
#include <iostream>
#include "ktfop.h"
int main()
{
using namespace ktfop;
MatMulParams<float> params;
params.order = CblasRowMajor; //矩阵存储顺序(行优先or列优先)
params.transA = CblasNoTrans; //A矩阵是否转置
params.transB = CblasNoTrans; //B矩阵是否转置
params.m = 2; //A矩阵的行数
params.n = 2; //B矩阵的列数
params.k = 2; //A矩阵的列数或B矩阵的行数
params.alpha = 1.0f; //alpha参数
params.lda = 2; //A矩阵存储步长
params.ldb = 2; //B矩阵存储步长
params.beta = 0.0f; //beta参数
params.ldc = 2; //C矩阵存储步长
float a[4] = {1, 2, 3, 4}; //初始化参与计算的矩阵
float b[4] = {5, 6, 7, 8};
float c[4] = {0, 0, 0, 0}; //保存计算结果的矩阵
int ret = -1;
ret = Matmul(a, b, c, params); //调用Matmul算子，结果存储于c数组
std::cout << "c: [";
for (int i = 0; i < 4; ++i) {
std::cout << c[i];
if (i < 3) {
std::cout << ", ";
}
}
std::cout << "]" << std::endl;
ret = Matmul(static_cast<float *>(nullptr), b, c, params); //输入空指针，打印日志"ERROR Parameter verification failed for the MatMul Op."
ret = Matmul(a, b, static_cast<float *>(nullptr), params); //输入空指针，打印日志"ERROR Parameter verification failed for the MatMul Op."
}
|