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

Update dgemm_kernel_4x8_haswell.S

This commit is contained in:
wjc404 2019-07-17 22:39:15 +08:00 committed by GitHub
parent 1733f927e6
commit 211ab03b14
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
kernel/x86_64/dgemm_kernel_4x8_haswell.S
View File

@ -267,23 +267,24 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.macro SAVE4x12
prefetcht0 128(%rsp) /*BUFFER 1*/
vbroadcastsd ALPHA, %ymm0
vmulpd %ymm0 , %ymm4 , %ymm4
vmulpd %ymm0 , %ymm5 , %ymm5
vmulpd %ymm0 , %ymm6 , %ymm6
vmulpd %ymm0 , %ymm7 , %ymm7
prefetcht0 192(%rsp)
vmulpd %ymm0 , %ymm8 , %ymm8
vmulpd %ymm0 , %ymm9 , %ymm9
vmulpd %ymm0 , %ymm10, %ymm10
vmulpd %ymm0 , %ymm11, %ymm11
prefetcht0 256(%rsp)
vmulpd %ymm0 , %ymm12, %ymm12
vmulpd %ymm0 , %ymm13, %ymm13
vmulpd %ymm0 , %ymm14, %ymm14
vmulpd %ymm0 , %ymm15, %ymm15
prefetcht0 320(%rsp)
vpermilpd $ 0x05 , %ymm5, %ymm5
vpermilpd $ 0x05 , %ymm7, %ymm7