GCC vector 叠加示例

一个简单的vector叠加示例

GCC version 4.7.0

 

#include <stdlib.h>
#include <stdint.h>
#include <unistd.h>
#include <x86intrin.h>

typedef union i4factor{
    __v4si v;
    int u4[4];
}
__attribute__((aligned(16)))
i4factor_t;
//typedef int v4sf __attribute__ ((mode(V4SF))); // vector of four single floats
#if 0
int a[4] = {1, 2, 3, 4};
int b[4] = {4, 3, 2, 1};

void vectorAdd(void)
{
    int i = 0;
    for (i = 0; i < 4; i++) {
        a[i] = a[i] + b[i];
    }
}
#else
i4factor_t a, b, c;

void vectorAdd(void)
{
    //c = __builtin_addv4si(a, b);
    c.v = a.v + b.v;
}
#endif
int main(int argc, char **argv)
{
    a.u4[0] = 1;
    a.u4[1] = 2;
    a.u4[2] = 3;
    a.u4[3] = 4;

    b.u4[0] = 4;
    b.u4[1] = 3;
    b.u4[2] = 2;
    b.u4[3] = 1;
    vectorAdd();
    //printf("%d, %d, %d, %d\n", a[0], a[1], a[2], a[3]);
    printf("%d, %d, %d, %d\n", c.u4[0], c.u4[1], c.u4[2], c.u4[3]);
    return 0;
}

 

编译

# gcc -march=core2 -O2 -pipe -ggdb -c v4si.c -o v4si.o
# gcc v4si.o  -o v4si

执行

# ./v4si
5, 5, 5, 5

objdump 查看是使用了SIMD指令

# objdump -dS v4si.o |grep -22 c.v |tail -25

void vectorAdd(void)
{
    //c = __builtin_addv4si(a, b);
        c.v = a.v + b.v;
  63:   66 0f fe 05 00 00 00    paddd  0x0(%rip),%xmm0        # 6b <main+0x6b>
  6a:   00
  6b:   66 0f 7f 05 00 00 00    movdqa %xmm0,0x0(%rip)        # 73 <main+0x73>
  72:   00