I have implemented an inline function (_mm256_concat_epi16). It concatenates two AVX2 vector containing 16-bit values. It works fine for first 8 numbers. If I want to use it for the rest of the vector I should change the implementation. But It would be better to use a single inline function in my main program.
The question is : Is there any better solution than mine or any suggestion to make this inline function more general which works on 16 values instead of my solution that works on 8 values? My solution concatenate 2 vectors but only 8 states of 16 possible state is solved.
**EDIT*My current solution for this question is using unaligned load function which exactly can read from any part from memory. But, when data is ready in register it might be better to reuse it. However, it might cause bottlenecks on port 5 which issues shuffle, permute, etc. But throughput might be enough (haven't test yet).
#include <stdio.h>
#include <x86intrin.h>
inline _mm256_print_epi16(__m256i a, char* name){
short temp[16], i;
_mm256_storeu_si256((__m256i *) &temp[0], a);
for(i=0; i<16; i++)
printf("%s[%d]=%4d , ",name,i+1,temp[i]);
printf("\n");
}
inline __m256i _mm256_concat_epi16(__m256i a, __m256i b, const int indx){
return _mm256_alignr_epi8(_mm256_permute2x128_si256(a,b,0x21),a,indx*2);
}
int main()
{
__m256i a = _mm256_setr_epi16(101,102,103,104,105,106,107,108,109,1010,1011,1012,1013,1014,1015,1016);_mm256_print_epi16(a, "a");
__m256i b = _mm256_setr_epi16(201,202,203,204,205,206,207,208,209,2010,2011,2012,2013,2014,2015,2016);_mm256_print_epi16(b, "b");
_mm256_print_epi16(_mm256_concat_epi16(a,b,8), "c");//numbers: 0-8
return 0;
}
The out put is :
// icc -march=native -O3 -D _GNU_SOURCE -o "concat" "concat.c"
[fedora@localhost concatination]$ "./concat"
a[1]= 101 , a[2]= 102 , a[3]= 103 , a[4]= 104 , a[5]= 105 , a[6]= 106 , a[7]= 107 , a[8]= 108 , a[9]= 109 , a[10]=1010 , a[11]=1011 , a[12]=1012 , a[13]=1013 , a[14]=1014 , a[15]=1015 , a[16]=1016 ,
b[1]= 201 , b[2]= 202 , b[3]= 203 , b[4]= 204 , b[5]= 205 , b[6]= 206 , b[7]= 207 , b[8]= 208 , b[9]= 209 , b[10]=2010 , b[11]=2011 , b[12]=2012 , b[13]=2013 , b[14]=2014 , b[15]=2015 , b[16]=2016 ,
c[1]= 109 , c[2]=1010 , c[3]=1011 , c[4]=1012 , c[5]=1013 , c[6]=1014 , c[7]=1015 , c[8]=1016 , c[9]= 201 , c[10]= 202 , c[11]= 203 , c[12]= 204 , c[13]= 205 , c[14]= 206 , c[15]= 207 , c[16]= 208 ,