sse.cpp

#include "stdafx.h"

#include <iostream>
#include <chrono>
#include <xmmintrin.h>
#include <cstring>

void test_sse( unsigned char *output, int *input )
{
	//Create alligned arrays on the stack and the pointers needed to access them
	const float factor = 255.f / 1024.f;
	__declspec(align(16)) float results[8] = { 0.0f };

	//Const cast the values so they can be used by sse
	const __m256i* wrd_ptr = ( __m256i* )input;

	//Load the integers into an xmm register
	__m256i loaded_ints = _mm256_load_si256( wrd_ptr );

	//Convert the integers to float values
	__m256 values_reg = _mm256_cvtepi32_ps( loaded_ints );

	//Set the factors
	__m256 factors_reg = _mm256_set1_ps( factor );

	//Apply the multiplication
	__m256 result_reg = _mm256_mul_ps( values_reg, factors_reg );

	_mm256_store_ps( results, result_reg );

	for( int i = 0; i < 8; ++i )
	{
		output[i] = static_cast<unsigned char>( results[i] );
	}
}

void test( unsigned char *output, int *input )
{
	__declspec( align( 16 ) ) float results[8] = { 0.0f };

	const float factor = 255 / 1024.f;

	results[0] = input[0] * factor;
	results[1] = input[1] * factor;
	results[2] = input[2] * factor;
	results[3] = input[3] * factor;
	results[4] = input[4] * factor;
	results[5] = input[5] * factor;
	results[6] = input[6] * factor;
	results[7] = input[7] * factor;

	for( int i = 0; i < 8; ++i )
	{
		output[i] = static_cast<unsigned char>( results[i] );
	}
}
int main( int argc, const char * argv[] )
{
	unsigned char *output = new unsigned char[8];
	int *values = new int[8];


	auto start = std::chrono::high_resolution_clock::now();

	for( int i = 0; i < 10000000; ++i )
	{
		values[0] = 1;
		values[1] = 254;
		values[2] = 777;
		values[3] = 1024;
		values[4] = 1;
		values[5] = 254;
		values[6] = 777;
		values[7] = 1024;

		//test( output, values );
		test_sse( output, values );
	}

	auto end = std::chrono::high_resolution_clock::now();
	auto duration = end - start;

	std::cout << std::chrono::duration_cast<std::chrono::milliseconds>(duration ).count() <<  std::endl;

	//for( int i = 0; i<8; ++i )
	//{
	//	std::cout << (int) output[i] << std::endl;
	//}

	delete[] values;
	delete[] output;
	return 0;
}