接上文：linux编程的108种奇淫巧计-11(乱序)

用了支持SSE4的CPU，intel core i3，因为支持了palignr指令，所以把上文的代码改用了palignr指令重写了一下如下：

可能由于在虚拟机上运行的原因，性能提升并不显著。

#include<stdio.h> #include<stdlib.h> #include <stddef.h> #include <stdint.h> asm(" .text "); asm(" .type shl_7, @function "); asm("shl_7: push %rbp "); asm(" mov %rsp,%rbp "); asm("loop: sub $0x80, %rdx "); asm(" movaps -0x07(%rsi), %xmm1 "); asm(" movaps 0x09(%rsi), %xmm2 "); asm(" movaps 0x19(%rsi), %xmm3 "); asm(" movaps 0x29(%rsi), %xmm4 "); asm(" movaps 0x39(%rsi), %xmm5 "); asm(" movaps 0x49(%rsi), %xmm6 "); asm(" movaps 0x59(%rsi), %xmm7 "); asm(" movaps 0x69(%rsi), %xmm8 "); asm(" movaps 0x79(%rsi), %xmm9 "); asm(" lea 0x80(%rsi), %rsi "); asm(" palignr $7, %xmm8, %xmm9 "); asm(" palignr $7, %xmm7, %xmm8 "); asm(" palignr $7, %xmm6, %xmm7 "); asm(" palignr $7, %xmm5, %xmm6 "); asm(" palignr $7, %xmm4, %xmm5 "); asm(" palignr $7, %xmm3, %xmm4 "); asm(" palignr $7, %xmm2, %xmm3 "); asm(" palignr $7, %xmm1, %xmm2 "); asm(" movaps %xmm9, 0x70(%rdi) "); asm(" movaps %xmm8, 0x60(%rdi) "); asm(" movaps %xmm7, 0x50(%rdi) "); asm(" movaps %xmm6, 0x40(%rdi) "); asm(" movaps %xmm5, 0x30(%rdi) "); asm(" movaps %xmm4, 0x20(%rdi) "); asm(" movaps %xmm3, 0x10(%rdi) "); asm(" movaps %xmm2, (%rdi) "); asm(" leaveq "); asm(" retq "); asm(" .text "); asm(" .type shl_7_f, @function "); asm("shl_7_f: push %rbp "); asm(" mov %rsp,%rbp "); asm("loop_f: sub $0x80, %rdx "); asm(" movaps -0x07(%rsi), %xmm1 "); asm(" movaps 0x09(%rsi), %xmm2 "); asm(" movaps 0x19(%rsi), %xmm3 "); asm(" movaps 0x29(%rsi), %xmm4 "); asm(" movaps 0x39(%rsi), %xmm5 "); asm(" movaps 0x49(%rsi), %xmm6 "); asm(" movaps 0x59(%rsi), %xmm7 "); asm(" movaps 0x69(%rsi), %xmm8 "); asm(" movaps 0x79(%rsi), %xmm9 "); asm(" lea 0x80(%rsi), %rsi "); asm(" palignr $7, %xmm8, %xmm9 "); asm(" movaps %xmm9, 0x70(%rdi) "); asm(" palignr $7, %xmm7, %xmm8 "); asm(" movaps %xmm8, 0x60(%rdi) "); asm(" palignr $7, %xmm6, %xmm7 "); asm(" movaps %xmm7, 0x50(%rdi) "); asm(" palignr $7, %xmm5, %xmm6 "); asm(" movaps %xmm6, 0x40(%rdi) "); asm(" palignr $7, %xmm4, %xmm5 "); asm(" movaps %xmm5, 0x30(%rdi) "); asm(" palignr $7, %xmm3, %xmm4 "); asm(" movaps %xmm4, 0x20(%rdi) "); asm(" palignr $7, %xmm2, %xmm3 "); asm(" movaps %xmm3, 0x10(%rdi) "); asm(" palignr $7, %xmm1, %xmm2 "); asm(" movaps %xmm2, (%rdi) "); asm(" leaveq "); asm(" retq "); int main(void) { uint8_t * src = (uint8_t*)malloc(sizeof(uint8_t)*(128+7)); uint8_t * des = (uint8_t*)malloc(sizeof(uint8_t)*128); int i = 0; for(;i<128;++i) { src[i] = 0xFF; } src[0] = 0x0; src[1]=0x1; src[2]=0x2; src[7]=0x7; src[8]=0x8; src[9]=0x9; src[10]=0xA; src[126]=0xB; src[127]=0xA; src+=7; shl_7(des,src); i = 0; for(;i<128;++i) { //printf("%d,%d/n",i,des[i]); } i = 0; for(;i<100000000;++i) { #ifdef _SLOW shl_7(des,src); #endif #ifdef _FAST shl_7_f(des,src); #endif } return 0; }