AVX 标量操作要快得多 [英] AVX scalar operations are much faster

问题描述

我测试了下面的简单函数

I test the following simple function

void mul(double *a, double *b) {
  for (int i = 0; i<N; i++) a[i] *= b[i];
}

具有非常大的数组，因此它受内存带宽限制.我使用的测试代码如下.当我用 -O2 编译时，它需要 1.7 秒.当我用 -O2 -mavx 编译时，它只需要 1.0 秒.非 vex 编码的标量操作慢 70%！这是为什么?

with very large arrays so that it is memory bandwidth bound. The test code I use is below. When I compile with -O2 it takes 1.7 seconds. When I compile with -O2 -mavx it takes only 1.0 seconds. The non vex-encoded scalar operations are 70% slower! Why is this?

这是-O2 和-O2 -mavx 的程序集.

https://godbolt.org/g/w4p60f

系统:i7-6700HQ@2.60GHz (Skylake) 32 GB 内存，Ubuntu 16.10，GCC 6.3

System: i7-6700HQ@2.60GHz (Skylake) 32 GB mem, Ubuntu 16.10, GCC 6.3

测试代码

//gcc -O2 -fopenmp test.c
//or
//gcc -O2 -mavx -fopenmp test.c
#include <string.h>
#include <stdio.h>
#include <x86intrin.h>
#include <omp.h>

#define N 1000000
#define R 1000

void mul(double *a, double *b) {
  for (int i = 0; i<N; i++) a[i] *= b[i];
}

int main() {
  double *a = (double*)_mm_malloc(sizeof *a * N, 32);
  double *b = (double*)_mm_malloc(sizeof *b * N, 32);

  //b must be initialized to get the correct bandwidth!!!
  memset(a, 1, sizeof *a * N);
  memset(b, 1, sizeof *b * N);

  double dtime;
  const double mem = 3*sizeof(double)*N*R/1024/1024/1024;
  const double maxbw = 34.1;
  dtime = -omp_get_wtime();
  for(int i=0; i<R; i++) mul(a,b);
  dtime += omp_get_wtime();
  printf("time %.2f s, %.1f GB/s, efficency %.1f%%
", dtime, mem/dtime, 100*mem/dtime/maxbw);

  _mm_free(a), _mm_free(b);
}

推荐答案

该问题与调用 omp_get_wtime() 后 AVX 寄存器的上半部分变脏有关.这对于 Skylake 处理器来说尤其是一个问题.

The problem is related to a dirty upper half of an AVX register after calling omp_get_wtime(). This is a problem particularly for Skylake processors.

我第一次看到这个问题是在这里.从那时起其他人已经观察到这个问题:这里 和 此处.

The first time I read about this problem was here. Since then other people have observed this problem: here and here.

使用gdb 我发现omp_get_wtime() 调用clock_gettime.我重写了我的代码以使用 clock_gettime() 并且我看到了同样的问题.

Using gdb I found that omp_get_wtime() calls clock_gettime. I rewrote my code to use clock_gettime() and I see the same problem.

void fix_avx() { __asm__ __volatile__ ( "vzeroupper" : : : ); }
void fix_sse() { }
void (*fix)();

double get_wtime() {
  struct timespec time;
  clock_gettime(CLOCK_MONOTONIC, &time);
  #ifndef  __AVX__ 
  fix();
  #endif
  return time.tv_sec + 1E-9*time.tv_nsec;
}

void dispatch() {
  fix = fix_sse;
  #if defined(__INTEL_COMPILER)
  if (_may_i_use_cpu_feature (_FEATURE_AVX)) fix = fix_avx;
  #else
  #if defined(__GNUC__) && !defined(__clang__)
  __builtin_cpu_init();
  #endif
  if(__builtin_cpu_supports("avx")) fix = fix_avx;
  #endif
}

使用gdb 单步执行代码我看到第一次调用clock_gettime 时它调用了_dl_runtime_resolve_avx().我相信问题出在基于 此评论.此函数似乎仅在第一次调用 clock_gettime 时调用.

Stepping through code with gdb I see that the first time clock_gettime is called it calls _dl_runtime_resolve_avx(). I believe the problem is in this function based on this comment. This function appears to only be called the first time clock_gettime is called.

在 GCC 中，使用 //__asm__ __volatile__ ("vzeroupper" : : ); 在第一次调用 clock_gettime 之后问题消失了，但是使用 Clang(使用 clang -O2 -fno-vectorize 因为 Clang 即使在 -O2 处也进行矢量化)它只会在每次调用 clock_gettime 后使用它才会消失.

With GCC the problem goes away using //__asm__ __volatile__ ( "vzeroupper" : : : ); after the first call with clock_gettime however with Clang (using clang -O2 -fno-vectorize since Clang vectorizes even at -O2) it only goes away using it after every call to clock_gettime.

这是我用来测试的代码(使用 GCC 6.3 和 Clang 3.8)

Here is the code I used to test this (with GCC 6.3 and Clang 3.8)

#include <string.h>
#include <stdio.h>
#include <x86intrin.h>
#include <time.h>

void fix_avx() { __asm__ __volatile__ ( "vzeroupper" : : : ); }
void fix_sse() { }
void (*fix)();

double get_wtime() {
  struct timespec time;
  clock_gettime(CLOCK_MONOTONIC, &time);
  #ifndef  __AVX__ 
  fix();
  #endif
  return time.tv_sec + 1E-9*time.tv_nsec;
}

void dispatch() {
  fix = fix_sse;
  #if defined(__INTEL_COMPILER)
  if (_may_i_use_cpu_feature (_FEATURE_AVX)) fix = fix_avx;
  #else
  #if defined(__GNUC__) && !defined(__clang__)
  __builtin_cpu_init();
  #endif
  if(__builtin_cpu_supports("avx")) fix = fix_avx;
  #endif
}

#define N 1000000
#define R 1000

void mul(double *a, double *b) {
  for (int i = 0; i<N; i++) a[i] *= b[i];
}

int main() {
  dispatch();
  const double mem = 3*sizeof(double)*N*R/1024/1024/1024;
  const double maxbw = 34.1;

  double *a = (double*)_mm_malloc(sizeof *a * N, 32);
  double *b = (double*)_mm_malloc(sizeof *b * N, 32);

  //b must be initialized to get the correct bandwidth!!!
  memset(a, 1, sizeof *a * N);
  memset(b, 1, sizeof *b * N);

  double dtime;
  //dtime = get_wtime(); // call once to fix GCC
  //printf("%f
", dtime);
  //fix = fix_sse;

  dtime = -get_wtime();
  for(int i=0; i<R; i++) mul(a,b);
  dtime += get_wtime();
  printf("time %.2f s, %.1f GB/s, efficency %.1f%%
", dtime, mem/dtime, 100*mem/dtime/maxbw);

  _mm_free(a), _mm_free(b);
}

<小时>

如果我使用 -z now 禁用延迟函数调用解析(例如 clang -O2 -fno-vectorize -z now foo.c)然后 Clang 只需要 __asm__ __volatile__ ( "vzeroupper" :: : ); 在第一次调用 clock_gettime 之后，就像 GCC 一样.

---

If I disable lazy function call resolution with -z now (e.g. clang -O2 -fno-vectorize -z now foo.c) then Clang only needs __asm__ __volatile__ ( "vzeroupper" : : : ); after the first call to clock_gettime just like GCC.

我希望现在使用 -z 我只需要 __asm__ __volatile__ ( "vzeroupper" : : ); 紧跟 main()> 但在第一次调用 clock_gettime 后我仍然需要它.

I expected that with -z now I would only need __asm__ __volatile__ ( "vzeroupper" : : : ); right after main() but I still need it after the first call to clock_gettime.

这篇关于AVX 标量操作要快得多的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！