How to calculate approximate reciprocal square root of an array faster on a cpu with popcnt and SSE4.2?
The input is positive integers (ranges from 0 to about 200,000) stored in an array of floats.
The output is an array of floats.
Both arrays have correct memory alignment for sse.
The code below only use 1 xmm register, runs on linux, and can be compiled by gcc -O3 code.cpp -lrt -msse4.2
Thank you.
#include <iostream>
#include <emmintrin.h>
#include <time.h>
using namespace std;
void print_xmm(__m128 xmm){
float out[4];
_mm_storeu_ps(out,xmm);
int i;
for (i = 0; i < 4; ++i) std::cout << out[i] << " ";
std::cout << std::endl;
}
void print_arr(float* ptr, size_t size){
size_t i;
for(i = 0; i < size; ++i){
cout << ptr[i] << " ";
}
cout << endl;
}
int main(void){
size_t size = 25000 * 4;
// this has to be multiple of 4
size_t repeat = 10000;
// test 10000 cycles of the code
float* ar_in = (float*)aligned_alloc(16, size*sizeof(float));
float* ar_out = (float*)aligned_alloc(16, size*sizeof(float));
//fill test data into the input array
//the data is an array of positive numbers.
size_t i;
for (i = 0; i < size; ++i){
ar_in[i] = (i+1) * (i+1);
}
//prepare for recipical square root.
__m128 xmm0;
size_t size_fix = size*sizeof(float)/sizeof(__m128);
float* ar_in_end = ar_in + size_fix;
float* ar_out_now;
float* ar_in_now;
//timing
struct timespec tp_start, tp_end;
i = repeat;
clock_gettime(CLOCK_MONOTONIC, &tp_start);
//start timing
while(--i){
ar_out_now = ar_out;
for(ar_in_now = ar_in;
ar_in_now != ar_in_end;
ar_in_now += 4, ar_out_now+=4){
//4 = sizeof(__m128)/sizeof(float);
xmm0 = _mm_load_ps(ar_in_now);
//cout << "load xmm: ";
//print_xmm(xmm0);
xmm0 = _mm_rsqrt_ps(xmm0);
//cout << "rsqrt xmm: ";
//print_xmm(xmm0);
_mm_store_ps(ar_out_now,xmm0);
}
}
//end timing
clock_gettime(CLOCK_MONOTONIC, &tp_end);
double timing;
const double nano = 0.000000001;
timing = ((double)(tp_end.tv_sec - tp_start.tv_sec )
+ (tp_end.tv_nsec - tp_start.tv_nsec) * nano)/repeat;
cout << " timing per cycle: " << timing << endl;
/*
cout << "input array: ";
print_arr(ar_in, size);
cout << "output array: ";
print_arr(ar_out,size);
*/
//free mem
free(ar_in);
free(ar_out);
return 0;
}
