strchr.c

#include <immintrin.h>
#include <stdio.h>
#include <time.h>
#include <stdint.h>
#include <stdbool.h>
#include <string.h>
#include <x86intrin.h>


long long lzcnt(long long l)
{
    return __lzcnt64(l);
}

int main() {
    
    size_t N = 1e9;
    bool found = false; //Not really used ...
    size_t char_index1 = 0;
    size_t char_index2 = 0;
    char * str = malloc(N);
    char * s;
    char * str2;
    memset(str,'a',N);
    clock_t clock_begin;
    clock_t clock_end;
    
    __m256i char_match;
    __m256i str_simd;
    __m256i result;
    __m256i* pSrc1;
    
    int simd_mask;
    
    double time_spent_std1;
    double time_spent_std2;
    double time_spent_simd;
    double speed_up = 0;
    
    str[(size_t)5e8] = 'b';
    
    for (int k = 0; k < 10; k++){
        
      	//=========================================
        //strchr
        //--------------------
        clock_begin = clock();
        str2 = strchr(str,'b');
        if (str2!=NULL){
            found = true;
            char_index2 = str2 - str;
        }
        clock_end = clock();
        time_spent_std2 = (double)(clock_end - clock_begin) / CLOCKS_PER_SEC;
        //=========================================
        
        
        
        //=========================================
        //Manual search 
        //----------------------------------------------
        //- useful if we know the charcter is present ...
        clock_begin = clock();
        s = str;
        while (*s != 'b'){
            ++s;
        }
        char_index1 = s - str;
        
        clock_end = clock();

        time_spent_std1 = (double)(clock_end - clock_begin) / CLOCKS_PER_SEC;
        //=========================================

        printf("STD1: %0.3f, I: %zu\n",time_spent_std1,char_index1);
        printf("STD2: %0.3f, I: %zu\n",time_spent_std2,char_index2);
        printf("strchr vs naive: %0.1fv\n",time_spent_std1/time_spent_std2);
        
        //SIMD
        //------------------------
        //__m256i _mm256_cmpeq_epi8 (__m256i a, __m256i b)
        
                clock_begin = clock();

                
        //Notes
        //-----------
        //=> placing 1 in while loop led to speedup
        //=> increment pointer as mm256i       
        
        char_match = _mm256_set1_epi8('b');
        result = _mm256_set1_epi32(0);
        
        simd_mask = 0;

        pSrc1 = (__m256i *)str;
        
        while (1){
            str_simd  = _mm256_lddqu_si256(pSrc1);
            result = _mm256_cmpeq_epi8(str_simd, char_match);
            simd_mask = _mm256_movemask_epi8(result);   
            if (simd_mask != 0){
                break;
            }
            pSrc1++;
        }
        
        //TODO: go from pSrc to local pointer ...
     	clock_end = clock();

      	time_spent_simd = (double)(clock_end - clock_begin) / CLOCKS_PER_SEC;
        printf("SIMD: %0.3f\n",time_spent_simd);

        
        printf("mask:%d, I:%ld , off: %lld\n",simd_mask,s - str,lzcnt((long long)simd_mask));
        
    }


    
    
return 0;
}