floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Eliminate loop code when called as/from dsdot

This commit is contained in:
Martin Kroeker 2017-10-25 16:45:41 +02:00 committed by GitHub
parent 5e3e91d0fc
commit a07807caac
1 changed files with 2 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
kernel/x86_64/sdot_microk_haswell-2.c
View File

@ -53,9 +53,11 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
"vfmadd231ps 64(%3,%0,4), %%ymm14, %%ymm6 \n\t" // 2 * y "vfmadd231ps 64(%3,%0,4), %%ymm14, %%ymm6 \n\t" // 2 * y
"vfmadd231ps 96(%3,%0,4), %%ymm15, %%ymm7 \n\t" // 2 * y "vfmadd231ps 96(%3,%0,4), %%ymm15, %%ymm7 \n\t" // 2 * y
#ifndef DSDOT
"addq $32 , %0 \n\t" "addq $32 , %0 \n\t"
"subq $32 , %1 \n\t" "subq $32 , %1 \n\t"
"jnz 1b \n\t" "jnz 1b \n\t"
#endif
"vextractf128 $1 , %%ymm4 , %%xmm12 \n\t" "vextractf128 $1 , %%ymm4 , %%xmm12 \n\t"
"vextractf128 $1 , %%ymm5 , %%xmm13 \n\t" "vextractf128 $1 , %%ymm5 , %%xmm13 \n\t"