Update dgemm_kernel_4x8_haswell.S

This commit is contained in:
wjc404 2019-07-17 17:02:35 +08:00 committed by GitHub
parent 7a9050d681
commit 182b06d6ad
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 20 additions and 20 deletions

View File

@ -317,10 +317,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
vmovups %ymm6 , (%rax)
vmovups %ymm7 , (%rax, LDC)
prefetcht0 32(CO1)
prefetcht0 32(CO1,LDC)
prefetcht0 32(%rax)
prefetcht0 32(%rax,LDC)
prefetcht0 56(CO1)
prefetcht0 56(CO1,LDC)
prefetcht0 56(%rax)
prefetcht0 56(%rax,LDC)
vpermilpd $ 0x05 , %ymm9 , %ymm9
vpermilpd $ 0x05 , %ymm11, %ymm11
@ -356,10 +356,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
vmovups %ymm6 , (%rbp)
vmovups %ymm7 , (%rbp, LDC)
prefetcht0 32(%rax)
prefetcht0 32(%rax,LDC)
prefetcht0 32(%rbp)
prefetcht0 32(%rbp,LDC)
prefetcht0 56(%rax)
prefetcht0 56(%rax,LDC)
prefetcht0 56(%rbp)
prefetcht0 56(%rbp,LDC)
vpermilpd $ 0x05 , %ymm13, %ymm13
vpermilpd $ 0x05 , %ymm15, %ymm15
@ -395,10 +395,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
vmovups %ymm6 , (%rbp)
vmovups %ymm7 , (%rbp, LDC)
prefetcht0 32(%rax)
prefetcht0 32(%rax,LDC)
prefetcht0 32(%rbp)
prefetcht0 32(%rbp,LDC)
prefetcht0 56(%rax)
prefetcht0 56(%rax,LDC)
prefetcht0 56(%rbp)
prefetcht0 56(%rbp,LDC)
addq $ 4*SIZE, CO1
.endm
@ -826,10 +826,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
vmovups %ymm6 , (%rax)
vmovups %ymm7 , (%rax, LDC)
prefetcht0 32(CO1)
prefetcht0 32(CO1,LDC)
prefetcht0 32(%rax)
prefetcht0 32(%rax,LDC)
prefetcht0 56(CO1)
prefetcht0 56(CO1,LDC)
prefetcht0 56(%rax)
prefetcht0 56(%rax,LDC)
vpermilpd $ 0x05 , %ymm9 , %ymm9
vpermilpd $ 0x05 , %ymm11, %ymm11
@ -865,10 +865,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
vmovups %ymm6 , (%rbp)
vmovups %ymm7 , (%rbp, LDC)
prefetcht0 32(%rax)
prefetcht0 32(%rax,LDC)
prefetcht0 32(%rbp)
prefetcht0 32(%rbp,LDC)
prefetcht0 56(%rax)
prefetcht0 56(%rax,LDC)
prefetcht0 56(%rbp)
prefetcht0 56(%rbp,LDC)
addq $ 4*SIZE, CO1
.endm