Intel MKL库常用模块的C语言实现

其他 2020-05-01 11:19:05 阅读次数: 0

文章目录

说明1：Intel MKL官方参考手册以及官方参考例程（位于安装目录下examples文件夹中）是最好的参考资料，本文程序大多根据官方例程改写，均可正常运行，我用的编译环境是VS2015+Intel MKL库2019；

说明2：由于本人主要是使用Intel MKL库做通信系统仿真，因此文章函数大多是复数单精度浮点形式，实数域函数参考官方手册对参数类型进行相应修改即可，也可以参考以下文章：MKL学习——基本操作C++实现，该文章主要是针对实数域的操作。

1. 随机数生成

1.1 等概率的01分布 viRngUniform

/* 生成信源（等概率的01序列）—— 等概率的伯努利分布 */

#include <stdio.h>
#include <math.h>

#include "mkl.h"
#include "errcheck.inc"

#define SEED    111
#define BRNG    VSL_BRNG_MCG31
#define METHOD  VSL_RNG_METHOD_UNIFORM_STD
#define N       1000    //随机数的数量
#define NN      100     

int main()
{
	int r[N];
	VSLStreamStatePtr stream;
	int i, errcode;
	int a = 0, b = 2;

	/***** Initialize *****/
	errcode = vslNewStream(&stream, BRNG, SEED);
	CheckVslError(errcode);   // 检测随机数生成过程是否出错

	/***** Call RNG *****/
	errcode = viRngUniform(METHOD, stream, N, r, a, b);
	CheckVslError(errcode);

	/***** Printing results *****/
	printf("Results (first 100 of 1000):\n");
	printf("---------------------------\n");
        for (i = 0; i<NN; i++) {
            printf("r[%d]=%d\n", i, r[i]);
        }

	/***** Deinitialize *****/
	errcode = vslDeleteStream(&stream);
	CheckVslError(errcode);

	return 0;
}

1.2 （复）高斯分布 vsRngGaussian

/* 生成独立同分布的准静态瑞利衰落信道（还需要将元素储存到矩阵中） */

#include <stdio.h>
#include <math.h>

#include "mkl.h"
#include "errcheck.inc"

#define SEED    777
#define BRNG    VSL_BRNG_MCG31
#define METHOD  VSL_RNG_METHOD_GAUSSIAN_ICDF
#define N       1000
#define NN      10

int main()
{
	float real[N];
	float imag[N];
	MKL_Complex8 H[1000];

	VSLStreamStatePtr stream;
	int i, errcode1,errcode2;
	float a = 0.0, sigma = 1.0;

	/***** Initialize *****/
	vslNewStream(&stream, BRNG, SEED);
	
	/***** Call RNG *****/
	errcode1 = vsRngGaussian(METHOD, stream, N, real, a, sigma);
	CheckVslError(errcode1);

	errcode2 = vsRngGaussian(METHOD, stream, N, imag, a, sigma);
	CheckVslError(errcode2);

	for (i = 0; i<NN; i++) {
		H[i].real = real[i]/(float)sqrt(2);
		H[i].imag = imag[i]/(float)sqrt(2);
	}

	/***** Printing results *****/
	printf("Results (first 10 of 1000):\n");
	printf("---------------------------\n");
	for (i = 0; i<NN; i++) {
		printf("real[%d]=%.4f\n", i, real[i]);
	}

	printf("---------------------------\n");
	for (i = 0; i<NN; i++) {
		printf("imag[%d]=%.4f\n", i, imag[i]);
	}

	printf("---------------------------\n");
	for (i = 0; i<NN; i++) {
		printf("H[%d]=%.4f+(%.4fi)\n", i, H[i].real, H[i].imag);
	}

	/***** Deinitialize *****/
	vslDeleteStream(&stream);

	return 0;
}

2. 向量-向量运算

2.1 向量数乘与加法 cblas_caxpby

/* 向量数乘与加法 y = alpha*x+beta*y */

#include <stdio.h>
#include <stdlib.h>
#include "mkl.h"

int main( )
{
	MKL_INT         n, incx, incy;
	MKL_Complex8    alpha, beta;
	MKL_Complex8   *x, *y;
	MKL_INT         len_x, len_y;

	/***************** 参数初始化 *****************/
	n = 5;
	incx = 1;
	incy = 1;
	alpha.imag = 0;
	alpha.real = 2;
	beta.imag = 0;
	beta.real = 1;

	len_x = 1 + (n - 1)* incx;
	len_y = 1 + (n - 1)* incy;
	x = (MKL_Complex8 *)mkl_calloc(len_x, sizeof(MKL_Complex8), 64);
	y = (MKL_Complex8 *)mkl_calloc(len_y, sizeof(MKL_Complex8), 64);

	/******************* 给复数向量x,y赋值 *********************/
	for (int i = 0; i < n; i++) {
		x[i].real = (float)i;
		x[i].imag = (float)(i + 1);
		y[i].real = (float)(i + 2);
		y[i].imag = (float)(i + 3);
	}

	/********** Call CBLAS_CAXPBY subroutine ( C Interface )**********/
	cblas_caxpby(n, &alpha, x, incx, &beta, y, incy);

	/********************** Printing results *******************/
	printf("y = ax + by\n");
	for (int i = 0; i < n; i++) {
		printf("----------------------------------------\n");
		printf("y[%d] = %.0f+%.0fi\n", i, y[i].real, y[i].imag);
	}
	
	mkl_free(x);
	mkl_free(y);

	return 0;
}

2.2 向量内积 cblas_cdotu_sub

/* 向量点乘（内积） res = \sigma i=0 to n (x_i*y_i) */

#include <stdio.h>
#include <stdlib.h>
#include "mkl.h"

int main()
{
	MKL_INT        n, incx, incy;
	MKL_Complex8  *x, *y;
	MKL_Complex8   res;
	MKL_INT        len_x, len_y;

	/***************** 参数初始化 *****************/
	n = 5;
	incx = 1;
	incy = 1;

	len_x = 1 + (n - 1)*incx;
	len_y = 1 + (n - 1)*incy;
	x = (MKL_Complex8 *)mkl_calloc(len_x, sizeof(MKL_Complex8), 64);
	y = (MKL_Complex8 *)mkl_calloc(len_y, sizeof(MKL_Complex8), 64);
	
	/******************* 给复数向量x,y赋值 *********************/
	for (int i = 0; i < n; i++) {
		x[i].real = (float)i;
		x[i].imag = (float)(i + 1);
		y[i].real = (float)(i + 2);
		y[i].imag = (float)(i + 3);
	}

	/*      Call CBLAS_CDOTU_SUB subroutine ( C Interface )        */
	cblas_cdotu_sub(n, x, incx, y, incy, &res);

	/********************** Printing results *******************/
	printf("%.0f+%.0fi\n", res.real, res.imag);

	mkl_free(x);
	mkl_free(y);

	return 0;
}

2.3 向量范数 cblas_scnrm2

/* 向量范数 res = ||x|| */

#include <stdio.h>
#include <stdlib.h>
#include "mkl.h"

int main()
{
	MKL_INT        n, incx;
	float          res;
	MKL_Complex8  *x;
	MKL_INT        len_x;

	/***************** 参数初始化 *****************/
	n = 5;
	incx = 1;

	len_x = 1 + (n - 1)*(incx);
	x = (MKL_Complex8 *)mkl_calloc(len_x, sizeof(MKL_Complex8), 64);

	/******************* 给复数向量x赋值 *********************/
	for (int i = 0; i < n; i++) {
		x[i].real = (float)i;
		x[i].imag = (float)(i + 1);
	}

	/************* Call CBLAS_SCNRM2 subroutine ( C Interface ) *********/
	res = cblas_scnrm2(n, x, incx);

	/************************* Print output data ********************/
	printf("||x|| = %f\n", res);

	mkl_free(x);
	return 0;
}

3. 矩阵-向量运算

3.1 矩阵存储方案

/* 以“实数矩阵-向量乘法”为例说明矩阵储存方案 y = alpha*A*x + beta*y
 以下代码以行优先 CblasRowMajor 为例 */

#include <stdio.h>
#include <stdlib.h>
#include "mkl.h"

int main()
{
	MKL_INT         m, n, lda, incx, incy;
	float           alpha, beta;
	float          *a, *x, *y;
	CBLAS_LAYOUT    layout = CblasRowMajor;  // 可变为CblasColMajor
	CBLAS_TRANSPOSE trans = CblasNoTrans;
	MKL_INT         nx, ny, len_x, len_y;

	m = 2;
	n = 5;
	incx = incy = 1;
	alpha = 1;	
	beta = 0;

	a = (float *)mkl_calloc(m*n, sizeof(float), 64);
	if (trans == CblasNoTrans) {
		nx = n;
		ny = m;
	}
	else {
		nx = m;
		ny = n;
	}
	if (layout == CblasRowMajor)
		lda = n;
	else
		lda = m;

	len_x = 1 + (nx - 1)*(incx);
	len_y = 1 + (ny - 1)*(incy);
	x = (float *)mkl_calloc(len_x, sizeof(float), 64);
	y = (float *)mkl_calloc(len_y, sizeof(float), 64);

	/*       Print input data          */
	printf("矩阵为\n");
	for (int i = 0; i < m*n; i++) {
		if (i%nx == 0 && i != 0)
			printf("\n");
		a[i] = (float)i;
		printf("%2.0f", a[i]);
	}
	printf("\n");

	printf("向量为\n");
	for (int i = 0; i < len_x; i++)
	{
		x[i] =(float)(i + 1);
		printf("%2.0f", x[i]);
		printf("\n");
	}
	
	/*      Call CBLAS_SGEMV subroutine ( C Interface )         */
	cblas_sgemv(layout, trans, m, n, alpha, a, lda, x, incx, beta, y, incy);

	/*       Print output data           */
	printf("矩阵-向量乘法结果\n");
	for (int i = 0; i < len_y; i++) {
		printf("%2.0f ", y[i]);
		printf("\n");
	}

	mkl_free(a);
	mkl_free(x);
	mkl_free(y);

	return 0;
}

上述过程中，矩阵a由数组 [ 0 1 2 3 4 5 6 7 8 9 ] \begin{bmatrix}0&1&2&3&4 & 5&6&7&8&9 \end{bmatrix} 对其赋值，具体储存过程如下：

行优先CblasRowMajor：初始化时，将数组逐行存入矩阵，可表示如下：
[ 0 1 2 3 4 5 6 7 8 9 ] [ 1 2 3 4 5 ] = [ 40 115 ] \begin{bmatrix}0&1&2&3&4 \\ 5&6&7&8&9 \end{bmatrix} \begin{bmatrix}1 \\ 2 \\3 \\4 \\5 \end{bmatrix}= \begin{bmatrix} 40 \\ 115 \end{bmatrix}
列优先CblasColMajor：行优先：初始化时，将数组逐列存入矩阵，可表示如下：
[ 0 2 4 6 8 1 3 5 7 9 ] [ 1 2 3 4 5 ] = [ 80 95 ] \begin{bmatrix}0&2&4&6&8 \\ 1&3&5&7&9 \end{bmatrix} \begin{bmatrix}1 \\ 2 \\3 \\4 \\5 \end{bmatrix}= \begin{bmatrix} 80 \\ 95 \end{bmatrix}
需注意，不管是行优先还是列优先，只要矩阵不转置，矩阵维度都不会发生变化。

3.2 矩阵-向量乘法 cblas_cgemv

trans参数控制矩阵A以什么形式参与运算，三个取值含义如下：
CblasNoTrans：A；CblasTrans：A的转置；CblasConjTrans：A的共轭转置

/* 复数矩阵-向量乘法 y = alpha*A*x + beta*y */
#include <stdio.h>
#include <stdlib.h>
#include "mkl.h"

int main()
{
	MKL_INT         m, n, lda, incx, incy;
	MKL_Complex8    alpha, beta;
	MKL_Complex8   *a, *x, *y;
	CBLAS_LAYOUT    layout = CblasRowMajor;
	CBLAS_TRANSPOSE trans = CblasNoTrans;
	MKL_INT         nx, ny, len_x, len_y;

	/***************** 参数初始化 *****************/
	m = 2;
	n = 5;
	incx = incy = 1;
	alpha.real = 1;
	alpha.imag = 0;
	beta.real = 0;
	beta.imag = 0;

	a = (MKL_Complex8 *)mkl_calloc(m*n, sizeof(MKL_Complex8), 64);

	if (trans == CblasNoTrans) {
		nx = n;
		ny = m;
	} else {
		nx = m;
		ny = n;
	}
	if (layout == CblasRowMajor)
		lda = n;
	else
		lda = m;

	len_x = 1 + (nx - 1)*(incx);
	len_y = 1 + (ny - 1)*(incy);
	x = (MKL_Complex8 *)mkl_calloc(len_x, sizeof(MKL_Complex8), 64);
	y = (MKL_Complex8 *)mkl_calloc(len_y, sizeof(MKL_Complex8), 64);

	/************** 矩阵与向量赋值 *******************/
	for (int i = 0; i < m*n; i++) {
		a[i].real = (float)(i + 1);
		a[i].imag = (float)i;
	}
	for (int i = 0; i < len_x; i++) {
		x[i].real = (float)i;
		x[i].imag = (float)(i + 1);
	}

	/******* Call CBLAS_CGEMV subroutine ( C Interface )*******/
	cblas_cgemv(layout, trans, m, n, &alpha, a, lda, x, incx, &beta, y, incy);

	/***************** Print output data *****************/
	for (int i = 0; i < len_y; i++) {
		printf("%f+%fi\n", y[i].real, y[i].imag);
	}

	mkl_free(a);
	mkl_free(x);
	mkl_free(y);

	return 0;
}

4. 矩阵-矩阵运算

4.1 矩阵乘法 cblas_cgemm

/* 矩阵乘法 C = alpha A*B + beta*C */

#include <stdio.h>
#include <stdlib.h>
#include "mkl.h"

int main()
{
	MKL_INT         m, n, k;
	MKL_INT         lda, ldb, ldc;
	MKL_Complex8    alpha, beta;
	MKL_Complex8   *a, *b, *c;
	CBLAS_LAYOUT    layout = CblasRowMajor;
	CBLAS_TRANSPOSE transA = CblasNoTrans;
	CBLAS_TRANSPOSE transB = CblasNoTrans;
	MKL_INT         ma, na, mb, nb;

	/***************** 参数初始化 *****************/
	m = 2;
	n = 5;
	k = 3;
	alpha.real = 1;
	alpha.imag = 0;
	beta.real = beta.imag = 0;

	if (transA == CblasNoTrans) {
		ma = m;
		na = k;
	} else {
		ma = k;
		na = m;
	}
	if (transB == CblasNoTrans) {
		mb = k;
		nb = n;
	} else {
		mb = n;
		nb = k;
	}

	a = (MKL_Complex8 *)mkl_calloc(ma*na, sizeof(MKL_Complex8), 64);
	b = (MKL_Complex8 *)mkl_calloc(mb*nb, sizeof(MKL_Complex8), 64);
	c = (MKL_Complex8 *)mkl_calloc(ma*nb, sizeof(MKL_Complex8), 64);

	/************** 矩阵与向量赋值 *******************/
	for (int i = 0; i < ma*na; i++) {
		a[i].real = (float)(i + 1);
		a[i].imag = (float)i;
	}
	for (int i = 0; i < mb*nb; i++) {
		b[i].real = (float)i;
		b[i].imag = (float)(i + 1);
	}
	
	if (layout == CblasRowMajor) {
		lda = na;
		ldb = nb;
		ldc = nb;
	} else {
		lda = ma;
		ldb = mb;
		ldc = ma;
	}

	/*      Call CGEMM subroutine ( C Interface )                  */
	cblas_cgemm(layout, transA, transB, m, n, k, &alpha, a, lda, b, ldb,
		&beta, c, ldc);

	/*       Print output data                                     */
	for (int i = 0; i < ma*nb; i++) {
		printf("%f+%fi\n", c[i].real, c[i].imag);
	}

	mkl_free(a);
	mkl_free(b);
	mkl_free(c);

	return 0;
}

4.2 矩阵求逆 LAPACKE_cgetrf+LAPACKE_cgetri

/* 矩阵求逆 */

#include <stdio.h>
#include <stdlib.h>
#include "mkl.h"

#define LAYOUT    LAPACK_ROW_MAJOR
#define N          3

int main()
{
	int layout = LAYOUT;
	int m = N;
	int n = N;
	int info1, info2, i;
	int lda = n;
	int ipiv[3];
	MKL_Complex8 *a;
	a = (MKL_Complex8 *)mkl_calloc(m*n, sizeof(MKL_Complex8), 64);

	/************* 矩阵赋值 ***************/
	for (i=0; i<m*n; i++) {
		a[i].real = (float)0;
		a[i].imag = (float)0;
	}
	a[0].real = a[5].real = 1;
	a[1].real = a[4].real = 2;
	a[2].real = a[3].real = 3;
	a[6].imag = 1;
	a[7].imag = 3;
	a[8].imag = 2;

    /* LU分解 + 求逆 */
	info1 = LAPACKE_cgetrf(layout, m, n, a, lda, ipiv);
	info2 = LAPACKE_cgetri(layout, m, a, lda, ipiv);
    
    /* print results*/
	printf("%d\n", info2);
	for (i = 0; i<N*N; i++) {
		printf("%f+(%f)i ", a[i].real, a[i].imag);
	}
	printf("\n");

	mkl_free(a);

	return 0;
}

4.3 矩阵的QR分解 LAPACKE_cgeqrf+LAPACKE_cungqr

#include <stdio.h>
#include <stdlib.h>
#include <math.h>

#include "mkl.h"

int main()
{
	int nR = 4;
	int nT = 3;
	int iseed[4] = { 1,2,3,5 };
	MKL_Complex8* H = NULL;
	H = (MKL_Complex8 *)mkl_calloc((nR + nT)*nT, sizeof(MKL_Complex8), 64);
	LAPACKE_clarnv(3, iseed, (nR + nT)*nT, H);  // 复高斯分布CN(0,2)

	printf("------- Before QR decomposition -------\n");
	for (int i = 0; i < (nR + nT)*nT; i++) {
		if (i%nT == 0)
			printf("\n");
		printf("%f+(%f)i ", H[i].real, H[i].imag);
	}
	printf("\n");

	MKL_Complex8 tau[3];   // 注意此处数组大小应随nT变化
	int info1 = LAPACKE_cgeqrf(LAPACK_ROW_MAJOR, nR + nT, nT, H, nT, tau);		// QR分解
	int info2 = LAPACKE_cungqr(LAPACK_ROW_MAJOR, nR + nT, nT, nT, H, nT, tau);	// 仅输出Q矩阵的前nT列

	printf("------- After QR decomposition -------\n");
	for (int i = 0; i < (nR + nT)*nT; i++) {
		if (i%nT == 0)
			printf("\n");
		printf("%f+(%f)i ", H[i].real, H[i].imag);
	}
	printf("\n");

	mkl_free(H);

	return 0;
}

5. 常用的子函数模块

5.1 计时函数

double start = dsecnd();
....
double stop  = dsecnd();
printf("Elapsed time = %f seconds\n", stop - start);
// 首次调用时初始化会花费一些时间，如果想要更精确的结果，可在正式计时开始前先调用一次

5.2 Print matrix

void print_matrix(MKL_LAYOUT layout, MKL_INT m, MKL_INT n, MKL_Complex8* a, MKL_INT lda) {
	MKL_INT i, j;
	for (i = 0; i < m; i++) {
		for (j = 0; j < n; j++) {
			if (layout == MKL_COL_MAJOR) {
				printf(" (%6.3f,%6.3f)", a[i + j * lda].real, a[i + j * lda].imag);
			} else {
				printf(" (%6.3f,%6.3f)", a[j + i * lda].real, a[j + i * lda].imag);
			}
		}
		printf("\n");
	}
}

5.3 生成单位矩阵

int n=500; // 矩阵维度
float *IdentityMatrix; 
IdentityMatrix = (float *)mkl_calloc(n*n, sizeof(float), 64);

for (int i=0; i<n*n; i++) {
    IdentityMatrix[i*(n+1)] = 1.0f; 
}

mkl_free(IdentityMatrix);

  博主不定期发布『保研/推免、C/C++、5G移动通信、Linux、生活随笔』系列文章，如果觉得本文对你有帮助，可以『点赞＋关注』支持一下哦！

随遇而安...
原创文章 75 获赞 176 访问量 4万+
关注 私信

猜你喜欢

转载自blog.csdn.net/weixin_43871127/article/details/105345514
今日推荐
基于大语言模型的开源知识库问答系统 MaxKB GitHub Star 数量突破 5,000 个！
美国拟限制 AI 大模型出口中国和俄罗斯
苹果将与 OpenAI 达成协议，将 ChatGPT 应用于 iPhone
openKylin 社区生态委员会第六次会议圆满召开
阿里云正式发布通义千问 2.5
Python 3.13 发布首个 Beta：实验性自由线程模式和 JIT、改进交互式解释器
Stack Overflow 拿我的代码去训练 AI 大模型，还封了我的账号
Pop!_OS 的 COSMIC 桌面完成 App Store 上架工作
《2024 年一季度互联网投融资运行情况》研究报告
报告：Django 仍然是 74% 开发者的首选
15 年前上了“FFmpeg 耻辱柱”，今天他还得谢谢咱——腾讯QQPlayer一雪前耻？
TIOBE 5 月榜单：Fortran “复活”进入 Top 10
周排行
记一下去大梅沙的准备（2018-05-26）
Spring 注解 事务
基于HTTP协议的客户端缓存
阿里云rds 备份和还原
[PHP] 几个拖慢 PHP 程序/API 运行速度的点
python 代码风格------------PEP8规则
js控制json生成菜单——自制菜单（一）
将字符串: 'k:1|k1:2|k2:3|k3:4 ' ,处理成 python 字典: {'k':1, 'k1':2, ...}
微信小程序转支付宝小程序
Qt551.窗口滚动条
每日归档
更多
2024-05-13(18)
2024-05-12(0)
2024-05-11(38)
2024-05-10(38)
2024-05-09(35)
2024-05-08(42)
2024-05-07(14)
2024-05-06(40)
2024-05-05(0)
2024-05-04(7)

代码天地 © 2018 codetd.com

境外服务器   最新电视剧2022