Assembly 流固有特性会降低性能
我在玩弄_mm_stream_ps的内在特性,在理解它的性能方面遇到了一些问题 下面是我正在处理的代码片段。。。 流版本:Assembly 流固有特性会降低性能,assembly,vectorization,sse,intrinsics,avx,Assembly,Vectorization,Sse,Intrinsics,Avx,我在玩弄_mm_stream_ps的内在特性,在理解它的性能方面遇到了一些问题 下面是我正在处理的代码片段。。。 流版本: #include <stdio.h> #include <stdint.h> #include <stdlib.h> #include <omp.h> #include <immintrin.h> #define NUM_ELEMENTS 10000000L static void copy_temporal
#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>
#include <omp.h>
#include <immintrin.h>
#define NUM_ELEMENTS 10000000L
static void copy_temporal(float* restrict x, float* restrict y)
{
for(uint64_t i = 0; i < NUM_ELEMENTS/2; ++i){
_mm_store_ps(y,_mm_load_ps(x));
_mm_store_ps(y+4,_mm_load_ps(x+4));
x+=8;
y+=8;
}
}
static void copy_nontemporal(float* restrict x, float* restrict y)
{
for(uint64_t i = 0; i < NUM_ELEMENTS/2; ++i){
_mm_stream_ps(y,_mm_load_ps(x));
_mm_stream_ps(y+4,_mm_load_ps(x+4));
x+=8;
y+=8;
}
}
int main(int argc, char** argv)
{
uint64_t sizeX = sizeof(float) * 4 * NUM_ELEMENTS;
float *x = (float*) _mm_malloc(sizeX,32);
float *y = (float*) _mm_malloc(sizeX,32);
//initialization
for(uint64_t i = 0 ; i < 4 * NUM_ELEMENTS; ++i){
x[i] = (float)rand()/RAND_MAX;
y[i] = 0;
}
printf("%g MB allocated\n",(2 * sizeX)/1024.0/1024.0);
double start = omp_get_wtime();
copy_nontemporal(x, y);
double time = omp_get_wtime() - start;
printf("Bandwidth (non-temporal): %g GB/s\n",((3 * sizeX)/1024.0/1024.0/1024.0)/time);
start = omp_get_wtime();
copy_temporal(x, y);
time = omp_get_wtime() - start;
printf("Bandwidth: %g GB/s\n",((3 * sizeX)/1024.0/1024.0/1024.0)/time);
_mm_free(x);
_mm_free(y);
return 0;
}
真正让我困惑的是,如果我使用不对齐的加载和存储(即storeu_ps/loadu_ps),我会在Xeon CPU(而不是笔记本电脑)上看到更好的性能:
我希望流版本比非流版本更快——因为y的冗余负载。然而,测量表明流版本实际上比非流版本慢两倍
你对此有什么解释吗
使用的编译器:英特尔14.0.1;
编译器标志:-O3-限制-xAVX;
使用的CPU:英特尔至强E5-2650
谢谢。流变体创建直接到DRAM的流水线突发写入。速度应大致与DRAM的速度相匹配。标准存储写入缓存(但如果数据不在缓存中,则先将其读入缓存)。如果数据已经在缓存中,则标准存储以缓存写入的速度运行。通常,使用流方法时,大小远大于最后一级缓存大小的写入速度要快得多。使用标准存储时,小型写入通常更快。尝试使用几GB的缓冲区大小运行测试。stream方法应该更快 这里有一个基准来证明:
#define __USE_MINGW_ANSI_STDIO 1
#include <stdlib.h>
#include <intrin.h>
#include <windows.h>
#include <stdio.h>
#include <stdint.h>
//-----------------------------------------------------------------------------
//
// queryPerformanceCounter - similar to QueryPerformanceCounter, but returns
// count directly.
uint64_t queryPerformanceCounter (void)
{
LARGE_INTEGER int64;
QueryPerformanceCounter (&int64);
return int64.QuadPart;
}
//-----------------------------------------------------------------------------
//
// queryPerformanceFrequency - same as QueryPerformanceFrequency, but returns count direcly.
uint64_t queryPerformanceFrequency (void)
{
LARGE_INTEGER int64;
QueryPerformanceFrequency (&int64);
return int64.QuadPart;
}
//---------------------------------------------------------------------------
static void testNontemporal (float *x, float *y, uint64_t numberOfVectors)
{
uint64_t i;
for(i = 0; i < numberOfVectors / 2; ++i)
{
_mm_stream_ps(y,_mm_load_ps(x));
_mm_stream_ps(y+4,_mm_load_ps(x+4));
y+=8; x+=8;
}
}
//---------------------------------------------------------------------------
static void testTemporal (float *x, float *y, uint64_t numberOfVectors)
{
uint64_t i;
for(i = 0; i < numberOfVectors / 2; ++i)
{
_mm_store_ps(y,_mm_load_ps(x));
_mm_store_ps(y+4,_mm_load_ps(x+4));
y+=8; x+=8;
}
}
//---------------------------------------------------------------------------
static void runtests (int nonTemporal)
{
uint64_t startCount, elapsed, index;
float *x, *y;
uint64_t numberOfBytes = 400 * 0x100000ull;
uint64_t numberOfFloats = numberOfBytes / sizeof *x;
uint64_t numberOfVectors = numberOfFloats / 4;
double gbPerSecond;
x = _mm_malloc (numberOfBytes, 32);
y = _mm_malloc (numberOfBytes, 32);
if (x == NULL || y == NULL) exit (1);
// put valid floating point data into the source buffer
// to avoid performance penalty
for (index = 0; index < numberOfFloats; index++)
x [index] = (float) index, y [index] = 0;
startCount = queryPerformanceCounter ();
if (nonTemporal)
testNontemporal (x, y, numberOfVectors);
else
testTemporal (x, y, numberOfVectors);
elapsed = queryPerformanceCounter () - startCount;
gbPerSecond = (double) numberOfBytes / 0x40000000 * queryPerformanceFrequency () / elapsed;
printf ("%.2f GB/s\n", gbPerSecond);
_mm_free (x);
_mm_free (y);
}
//---------------------------------------------------------------------------
int main (void)
{
// raise our priority to increase measurement accuracy
SetPriorityClass (GetCurrentProcess (), REALTIME_PRIORITY_CLASS);
printf ("using temporal stores\n");
runtests (0);
printf ("using non-temporal stores\n");
runtests (1);
return 0;
}
//---------------------------------------------------------------------------
另外,非临时存储从所有缓存中删除目标缓存线。如果在该行被自然删除之前再次触碰它,那么您将损失惨重。正如ScottD所指出的,问题的答案在于生成的汇编代码。 显然,英特尔编译器足够智能,能够检测访问模式,并自动生成非临时加载,即使是临时版本 以下是编译器为临时版本生成的汇编代码的示例:
..___tag_value___Z13copy_temporalPfS_.35: #
xor edx, edx #22.4
xor eax, eax #
..B2.2: # Preds ..B2.2 ..B2.1
vmovups xmm0, XMMWORD PTR [rax+rdi] #23.34
inc rdx #22.4
vmovntps XMMWORD PTR [rax+rsi], xmm0 #23.20
vmovups xmm1, XMMWORD PTR [16+rax+rdi] #24.36
vmovntps XMMWORD PTR [16+rax+rsi], xmm1 #24.20
vmovups xmm2, XMMWORD PTR [32+rax+rdi] #23.34
vmovntps XMMWORD PTR [32+rax+rsi], xmm2 #23.20
vmovups xmm3, XMMWORD PTR [48+rax+rdi] #24.36
vmovntps XMMWORD PTR [48+rax+rsi], xmm3 #24.20
add rax, 64 #22.4
cmp rdx, 5000000 #22.4
jb ..B2.2 # Prob 99% #22.4
仍然存在的问题如下:
为什么CPU E5-2650(见上文)的非对齐时间版本的性能优于非时间版本。我已经看过生成的汇编代码,编译器确实生成了vmovups指令(由于不存在对齐) 谢谢你的回复。我已经在使用大小为400MB的缓冲区(也就是说,比我系统中的任何缓存都大得多)。此外,为了读取一些硬件计数器,我对代码进行了检测,结果是决定性的(即使用stream_ps可以减少二级写入未命中)。但是,我仍然无法解释这两个版本之间的巨大性能差异。我将添加一个示例基准测试,试图展示大型缓冲区的非时间(流)优势。这是一个有点快和肮脏,但我认为是正确的。使用了不可移植的(Windows)计时功能。我已经更新了我的原始帖子,但我无法复制您的结果(即使我将您的代码移植到Linux,结果仍然相同)。你知道为什么会这样吗?此外,你对不结盟版本更快的原因有何解释?这实际上可能指向真正的问题,因为流确实需要对齐。@ScottD,我在MSVC中运行了您的测试。运行它们几次,结果就会前后交换。这两种方法在不确定性范围内具有相同的速度。@Z,您使用的是优化的发布版本还是调试版本?我得到了您在运行调试构建时描述的内容。但是发布版本(或命令行cl-Ox stream.c)的工作与预期的一样。我想说的是,这取决于所讨论的缓冲区的大小,但是您对ScottD的评论表明它们非常大。在这一点上,我不确定发生了什么。您可以尝试各种方法,如注释掉
#pragma
、在不使用-xAVX
的情况下编译等,寻找常规存储和非临时存储之间性能比率的变化。无需展开循环。循环展开仅在依赖链中有用,并且没有依赖链。CPU可以帮你处理这个问题。不过我有个问题。带宽计算中的因子3是多少?两次读取+一次写入。尽管非时态版本只进行一次阅读,但为了简化比较,我保留了三个因子。如果ICC做了与你所说的不同的事情,那就令人失望了。我更希望它能实现你想要的内在目标。
#define __USE_MINGW_ANSI_STDIO 1
#include <stdlib.h>
#include <intrin.h>
#include <windows.h>
#include <stdio.h>
#include <stdint.h>
//-----------------------------------------------------------------------------
//
// queryPerformanceCounter - similar to QueryPerformanceCounter, but returns
// count directly.
uint64_t queryPerformanceCounter (void)
{
LARGE_INTEGER int64;
QueryPerformanceCounter (&int64);
return int64.QuadPart;
}
//-----------------------------------------------------------------------------
//
// queryPerformanceFrequency - same as QueryPerformanceFrequency, but returns count direcly.
uint64_t queryPerformanceFrequency (void)
{
LARGE_INTEGER int64;
QueryPerformanceFrequency (&int64);
return int64.QuadPart;
}
//---------------------------------------------------------------------------
static void testNontemporal (float *x, float *y, uint64_t numberOfVectors)
{
uint64_t i;
for(i = 0; i < numberOfVectors / 2; ++i)
{
_mm_stream_ps(y,_mm_load_ps(x));
_mm_stream_ps(y+4,_mm_load_ps(x+4));
y+=8; x+=8;
}
}
//---------------------------------------------------------------------------
static void testTemporal (float *x, float *y, uint64_t numberOfVectors)
{
uint64_t i;
for(i = 0; i < numberOfVectors / 2; ++i)
{
_mm_store_ps(y,_mm_load_ps(x));
_mm_store_ps(y+4,_mm_load_ps(x+4));
y+=8; x+=8;
}
}
//---------------------------------------------------------------------------
static void runtests (int nonTemporal)
{
uint64_t startCount, elapsed, index;
float *x, *y;
uint64_t numberOfBytes = 400 * 0x100000ull;
uint64_t numberOfFloats = numberOfBytes / sizeof *x;
uint64_t numberOfVectors = numberOfFloats / 4;
double gbPerSecond;
x = _mm_malloc (numberOfBytes, 32);
y = _mm_malloc (numberOfBytes, 32);
if (x == NULL || y == NULL) exit (1);
// put valid floating point data into the source buffer
// to avoid performance penalty
for (index = 0; index < numberOfFloats; index++)
x [index] = (float) index, y [index] = 0;
startCount = queryPerformanceCounter ();
if (nonTemporal)
testNontemporal (x, y, numberOfVectors);
else
testTemporal (x, y, numberOfVectors);
elapsed = queryPerformanceCounter () - startCount;
gbPerSecond = (double) numberOfBytes / 0x40000000 * queryPerformanceFrequency () / elapsed;
printf ("%.2f GB/s\n", gbPerSecond);
_mm_free (x);
_mm_free (y);
}
//---------------------------------------------------------------------------
int main (void)
{
// raise our priority to increase measurement accuracy
SetPriorityClass (GetCurrentProcess (), REALTIME_PRIORITY_CLASS);
printf ("using temporal stores\n");
runtests (0);
printf ("using non-temporal stores\n");
runtests (1);
return 0;
}
//---------------------------------------------------------------------------
using temporal stores
5.57 GB/s
using non-temporal stores
8.35 GB/s
..___tag_value___Z13copy_temporalPfS_.35: #
xor edx, edx #22.4
xor eax, eax #
..B2.2: # Preds ..B2.2 ..B2.1
vmovups xmm0, XMMWORD PTR [rax+rdi] #23.34
inc rdx #22.4
vmovntps XMMWORD PTR [rax+rsi], xmm0 #23.20
vmovups xmm1, XMMWORD PTR [16+rax+rdi] #24.36
vmovntps XMMWORD PTR [16+rax+rsi], xmm1 #24.20
vmovups xmm2, XMMWORD PTR [32+rax+rdi] #23.34
vmovntps XMMWORD PTR [32+rax+rsi], xmm2 #23.20
vmovups xmm3, XMMWORD PTR [48+rax+rdi] #24.36
vmovntps XMMWORD PTR [48+rax+rsi], xmm3 #24.20
add rax, 64 #22.4
cmp rdx, 5000000 #22.4
jb ..B2.2 # Prob 99% #22.4