Здравствуйте, Евгений!
Прошу... Полная аналогия...
[code h=200]// SSEIntrinsics.cpp : Defines the entry point for the console application.
#include "stdafx.h"
#include <cstdio>
#include <iostream>
#include <cmath>
#include <windows.h>
#include <xmmintrin.h>
//calc function cpp: f(x,y,z) = x*x + y - sqrt(z*z + 1)
void ComputeArrayCpp(
float* pArray1, // [in] first source array
float* pArray2, // [in] second source array
float* pArray3, // [in] third source array
float* pResult, // [out] result array
int nSize) // [in] size of all arrays
{
float* pSource1 = pArray1;
float* pSource2 = pArray2;
float* pSource3 = pArray3;
float* pDest = pResult;
for ( int i = 0; i < nSize; i++ )
{
*pDest = (float)((*pSource1) * (*pSource1) + (*pSource2) -
sqrt((*pSource3) * (*pSource3)) + 1.f);
pSource1++;
pSource2++;
pSource3++;
pDest++;
}
}
//calc function sse: f(x,y,z) = x*x + y - sqrt(z*z + 1)
void ComputeArrayCppSSE(
float* pArray1, // [in] first source array
float* pArray2, // [in] second source array
float* pArray3, // [in] third source array
float* pResult, // [out] result array
int nSize) // [in] size of all arrays
{
int nLoop = nSize/ 4;
__m128 m1, m2, m3, m4;
__m128* pSrc1 = (__m128*) pArray1;
__m128* pSrc2 = (__m128*) pArray2;
__m128* pSrc3 = (__m128*) pArray3;
__m128* pDest = (__m128*) pResult;
__m128 m1_0 = _mm_set_ps1(1.0f); // m1_0[0, 1, 2, 3] = 1.0
for ( int i = 0; i < nLoop; i++ )
{
m1 = _mm_mul_ps(*pSrc1, *pSrc1);// m1 = x * x
m2 = _mm_mul_ps(*pSrc3, *pSrc3);// m2 = z * z
m3 = _mm_add_ps(m2, m1_0); // m3 = z*z + 1.0
m4 = _mm_sqrt_ps(m3); // m4 = sqrt(z*z+1)
m2 = _mm_add_ps(m1, *pSrc2); // m2 = x*x + y
*pDest = _mm_sub_ps(m2, m4); // *pDest = x*x + y - sqrt(z*z+1)
pSrc1++;
pSrc2++;
pSrc3++;
pDest++;
}
}
//init array with sin(rnd())
void init(float* a, int size)
{
for(int i = 0; i < size; i++)
{
float x = (float)rand()/RAND_MAX;
a[i] = sin(x);
}
}
int _tmain(int argc, _TCHAR* argv[])
{
setlocale(LC_ALL, "Russian");
const int MAX_SIZE = 10000000;
float *a = (float*)_aligned_malloc(sizeof(float)*MAX_SIZE, 16);
float *b = (float*)_aligned_malloc(sizeof(float)*MAX_SIZE, 16);
float *c = (float*)_aligned_malloc(sizeof(float)*MAX_SIZE, 16);
float *f = (float*)_aligned_malloc(sizeof(float)*MAX_SIZE, 16);
DWORD startTime, endTime;
startTime = GetTickCount();
init(a, MAX_SIZE);
init(b, MAX_SIZE);
init(c, MAX_SIZE);
endTime = GetTickCount();
printf("Инициализация массивов: %d мс\n", endTime - startTime);
startTime = GetTickCount();
ComputeArrayCpp(a, b, c, f, MAX_SIZE);
endTime = GetTickCount();
printf("Вычисление средствами C++: %d мс\n", endTime - startTime);
startTime = GetTickCount();
ComputeArrayCppSSE(a, b, c, f, MAX_SIZE);
endTime = GetTickCount();
printf("Вычисление средствами SSE: %d мс\n", endTime - startTime);
_aligned_free(a);
_aligned_free(b);
_aligned_free(c);
_aligned_free(f);
return 0;
}[/code]