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

modified dtrsm_kernel_LT_8x2_bulldozer.S

This commit is contained in:
wernsaar 2013-08-04 10:15:33 +02:00
parent 87fc9de572
commit 8aeec32ea0
1 changed files with 9 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
kernel/x86_64/dtrsm_kernel_LT_8x2_bulldozer.S
View File

@ -84,12 +84,6 @@
#endif #endif
#define PREFETCH prefetch
#define PREFETCHSIZE (8 * 7 + 0)
#define movlpd vmovsd
#define movapd vmovups
#define movupd vmovups
.macro KERNEL8x2_SUB .macro KERNEL8x2_SUB
vmovddup -16*SIZE(BO,%rax,2), %xmm1 vmovddup -16*SIZE(BO,%rax,2), %xmm1
@ -97,15 +91,15 @@
vmovups -16*SIZE(AO,%rax,8), %xmm0 vmovups -16*SIZE(AO,%rax,8), %xmm0
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8 vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
vfmaddpd %xmm9 , %xmm0 , %xmm2 , %xmm9 vfmaddpd %xmm9 , %xmm0 , %xmm2 , %xmm9
vmovups -14*SIZE(AO,%rax,8), %xmm0 vmovups -14*SIZE(AO,%rax,8), %xmm4
vfmaddpd %xmm10, %xmm0 , %xmm1 , %xmm10 vfmaddpd %xmm10, %xmm4 , %xmm1 , %xmm10
vfmaddpd %xmm11, %xmm0 , %xmm2 , %xmm11 vfmaddpd %xmm11, %xmm4 , %xmm2 , %xmm11
vmovups -12*SIZE(AO,%rax,8), %xmm0 vmovups -12*SIZE(AO,%rax,8), %xmm5
vfmaddpd %xmm12, %xmm0 , %xmm1 , %xmm12 vfmaddpd %xmm12, %xmm5 , %xmm1 , %xmm12
vfmaddpd %xmm13, %xmm0 , %xmm2 , %xmm13 vfmaddpd %xmm13, %xmm5 , %xmm2 , %xmm13
vmovups -10*SIZE(AO,%rax,8), %xmm0 vmovups -10*SIZE(AO,%rax,8), %xmm6
vfmaddpd %xmm14, %xmm0 , %xmm1 , %xmm14 vfmaddpd %xmm14, %xmm6 , %xmm1 , %xmm14
vfmaddpd %xmm15, %xmm0 , %xmm2 , %xmm15 vfmaddpd %xmm15, %xmm6 , %xmm2 , %xmm15
addq $SIZE, %rax addq $SIZE, %rax
.endm .endm