Test memory read rate

1. Ideas

At ordinary times CPU Read data from cache Or read data from memory . If you want to simply read data from memory , So you need to take cache close . Then read the array sequentially from memory . Read rate bandwidth、 Array size arraysize、 Read time time The calculation formula between is as follows ：
$$bandwidth=\frac{arraysize}{time}$$

2. close cache

First, in the Linux System shutdown cache, Use the following code to cache close .（ Code reference 1, Code reference 2）

#include <linux/init.h>
#include <linux/module.h>
MODULE_LICENSE("Dual BSD/GPL");
static int disableCache_init(void)
{
    
        printk(KERN_ALERT "Disabling L1 and L2 caches.\n");
        __asm__(".intel_syntax noprefix\n\t"
                "mov rax,cr0\n\t"
                "or rax,(1 << 30)\n\t"
                "mov cr0,rax\n\t"
                "wbinvd\n\t"
                ".att_syntax noprefix\n\t"
        : : : "rax" );
        return 0;
}
static void disableCache_exit(void)
{
    
        printk(KERN_ALERT "Enabling L1 and L2 caches.\n");
        __asm__(".intel_syntax noprefix\n\t"
                "mov rax,cr0\n\t"
                "and rax,~(1 << 30)\n\t"
                "mov cr0,rax\n\t"
                "wbinvd\n\t"
                ".att_syntax noprefix\n\t"
        : : : "rax" );
}
module_init(disableCache_init);
module_exit(disableCache_exit);

makefile as follows

  EXTRA_CFLAGS = -m64
  obj-m += disableCache.o
  
  all:
  	make -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules

Use the following Linux Command to compile source code

  $ make

After compilation, use the command insmod Load the program into kernel module in , And use dmesg Check whether the load is successful

  $ insmod disableCache.ko
  $ dmesg

Post load system log The following output will appear in

3. Start testing bandwidth

Closing cache Use the following code to test bandwidth

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

//  Test read BIG_ARR How long does it take , And then calculate bandwidth
#define BIG_ARR 999999

int main()
{
    
    int *block = (int *)calloc(BIG_ARR, sizeof(int));

    int i, temp;
    clock_t start = clock(), total_time;
    for(i = 0; i < BIG_ARR; ++i)
    temp += block[i];

    
    total_time = clock() - start;
    double sec = (double)total_time / (double) CLOCKS_PER_SEC;
    printf("read %d times from main memory need %lf sec\n\ The bandwidth is %lfbyte/sec\n",\
    BIG_ARR, sec, BIG_ARR / sec);
    return 0;
}

The running results are as follows （ Here the code output forgets to multiply the number of bytes sizeof(int), So the final result needs to be multiplied by 4）

obtain $bandwidth=359453271int/sec\approx 1.44M/sec$

4. Enable cache

Will actually close cache load kernel module Just uninstall the module , Use the following instructions

  $ rmmod disableCache.ko
  $ dmesg

Output in the system log You can see the following output in , Explain that you have cache Enable .

notes ： This test is in Intel i3 The complete ,AMD Need to be right cr0 Register for other operations , Code reference 2.