【贝能高性价比ATSAMD51评估板】基准性能测试之三:线性浮点运算能力测试linpack

qinyunti

【贝能高性价比ATSAMD51评估板】基准性能测试之三:线性浮点运算能力测试linpack [复制链接]

前言

  Linpack是国际上使用最广泛的测试高性能计算机系统浮点性能的基准测试。通过对高性能计算机采用高斯消元法求解一元 N次稠密线性代数方程组的测试，评价高性能计算机的浮点计算性能。Linpack的结果按每秒浮点运算次数（flops）表示。

 

过程

添加代码

获取代码

http://www.roylongbottom.org.uk/classic_benchmarks.tar.gz

解压classic_benchmarks.tar.gz将\classic_benchmarks\classic_benchmarks\source_code\linpack文件夹复制到自己的工程。

添加代码

 

将\classic_benchmarks\source_code\linpack文件夹复制到工程目录，并添加工程中

 

 

移植接口

linpack.c中

 

62行注释掉//#include "cpuidh.h"

 

81行main (int argc, char *argv[])

改为

int linpack_main (int argc, char *argv[])

 

95行        int         nopause = 1;改为

        int         nopause = 0;

 

注释掉113行的内容

#if 0         

        getDetails();

        for (i=1; i<10; i++)

        {

            printf("%s\n", configdata);

        }

        printf("\n");

        printf("##########################################\n");

 

        outfile = fopen("Linpack.txt","a+");

        if (outfile == NULL)

        {

            printf (" Cannot open results file \n\n");

            printf(" Press Enter\n\n");

            int g = getchar();

            exit (0);

        }

        fprintf (outfile, " ########################################################\n\n");                     

        for (i=1; i<10; i++)

        {

            fprintf(outfile, "%s \n", configdata);

        }

        fprintf (outfile, "\n");

#endif

 

start_time();改为

uint32_t s_stime_u32 = SYSTICK_GetTickCounter();

end_time();改为

uint32_t s_etime_u32 = SYSTICK_GetTickCounter();

uint32_t secs = (s_etime_u32-s_stime_u32)/1000.0;

 

添加

#include <stdint.h>

 #include "definitions.h"

 

注释掉432行

        //local_time();

 

注释掉438行

    //fprintf (outfile, " ########################################################\n\n");                     

    //fprintf (outfile, " Linpack %s Precision %s Benchmark n @ 100\n", PREC, ROLLING);

//fprintf (outfile, " Optimisation %s, %s\n", options, timeday);

 

注释掉507行

#if 0

    for (i=0; i<5; i++)

    {

        if (strcmp (expect, was)  != 0)

        {

            fprintf(outfile, " Variable %s Non-standard result was %s instead of %s\n",

                               title, was, expect);  

            errors = errors + 1;

        }

    }

#endif

 

532行添加 return 0;

 

 

编译ram可能不够可以修改

#define LDA  21

#define LDAA 20

#define N 10

 

        lda = 201;

        ldaa = 200;

        cray = .056;

        n = 100;改为

 

        static REAL aa[200*200],a[200*201],b[200],x[200];  

改为

        static REAL aa[LDAA*LDAA],a[LDAA*LDA],b[LDAA],x[LDAA];

 

static int ipvt[200]

改为

static int ipvt[LDAA]

测试

main.c中

申明 void linpack_main (int argc, char *argv[]);

int main ( void )

{

    /* Initialize all modules */

    SYS_Initialize ( NULL );

  SYSTICK_TimerStart();

    PORT_REGS->GROUP[1].PORT_PINCFG[24] = 0x1U;

    PORT_REGS->GROUP[1].PORT_PINCFG[25] = 0x1U;

    PORT_REGS->GROUP[1].PORT_PMUX[12] = 0x33U;

  linpack_main(0,0);

    while ( true )

    {

        /* Maintain state machines of all polled MPLAB Harmony modules. */

        SYS_Tasks ( );

    }

 

    /* Execution should not come here during normal operation */

 

    return ( EXIT_FAILURE );

}

 

 

结果如下

 

 

http://www.roylongbottom.org.uk/linpack%20results.htm

对比得分比AMD 80386高一些

 

 

 

总结

本文进行了线性浮点数运算性能测试，也和其他芯片对比，可以横向参考下芯片的性能。