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

Refs #314. Fixed clang compiling bug on OSX.

This commit is contained in:
Zhang Xianyi 2013-11-07 08:12:03 +08:00
parent 73770e60b8
commit 2f5fdd2000
2 changed files with 16 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
kernel/x86_64/dtrsm_kernel_LT_8x2_bulldozer.S
View File

@ -103,7 +103,7 @@
vmovups -10*SIZE(AO,%rax,8), %xmm6
vfmaddpd %xmm14, %xmm6 , %xmm1 , %xmm14
vfmaddpd %xmm15, %xmm6 , %xmm2 , %xmm15
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
.macro SOLVE_8x2
@ -265,7 +265,7 @@
vmovups -14*SIZE(AO,%rax,4), %xmm0
vfmaddpd %xmm10, %xmm0 , %xmm1 , %xmm10
vfmaddpd %xmm11, %xmm0 , %xmm2 , %xmm11
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
@ -338,7 +338,7 @@
vmovups -16*SIZE(AO,%rax,2), %xmm0
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
vfmaddpd %xmm9 , %xmm0 , %xmm2 , %xmm9
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
@ -378,7 +378,7 @@
vmovups -16*SIZE(BO,%rax,2), %xmm1
vmovddup -16*SIZE(AO,%rax,1), %xmm0
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
.macro SOLVE_1x2
@ -411,7 +411,7 @@
vfmaddpd %xmm10, %xmm0 , %xmm1 , %xmm10
vmovups -10*SIZE(AO,%rax,8), %xmm0
vfmaddpd %xmm11, %xmm0 , %xmm1 , %xmm11
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
.macro SOLVE_8x1
@ -510,7 +510,7 @@
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
vmovups -14*SIZE(AO,%rax,4), %xmm0
vfmaddpd %xmm9 , %xmm0 , %xmm1 , %xmm9
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
@ -560,7 +560,7 @@
vmovddup -16*SIZE(BO,%rax,1), %xmm1
vmovups -16*SIZE(AO,%rax,2), %xmm0
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
@ -592,7 +592,7 @@
vmovsd -16*SIZE(BO,%rax,1), %xmm1
vmovsd -16*SIZE(AO,%rax,1), %xmm0
vfmaddsd %xmm8 , %xmm0 , %xmm1 , %xmm8
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
.macro SOLVE_1x1

16
kernel/x86_64/dtrsm_kernel_RN_8x2_bulldozer.S
View File

@ -103,7 +103,7 @@
vmovups -10*SIZE(AO,%rax,8), %xmm6
vfmaddpd %xmm14, %xmm6 , %xmm1 , %xmm14
vfmaddpd %xmm15, %xmm6 , %xmm2 , %xmm15
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
.macro SOLVE_8x2
@ -177,7 +177,7 @@
vmovups -14*SIZE(AO,%rax,4), %xmm0
vfmaddpd %xmm10, %xmm0 , %xmm1 , %xmm10
vfmaddpd %xmm11, %xmm0 , %xmm2 , %xmm11
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
@ -226,7 +226,7 @@
vmovups -16*SIZE(AO,%rax,2), %xmm0
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
vfmaddpd %xmm9 , %xmm0 , %xmm2 , %xmm9
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
@ -262,7 +262,7 @@
vmovups -16*SIZE(BO,%rax,2), %xmm1
vmovddup -16*SIZE(AO,%rax,1), %xmm0
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
.macro SOLVE_1x2
@ -306,7 +306,7 @@
vfmaddpd %xmm10, %xmm0 , %xmm1 , %xmm10
vmovups -10*SIZE(AO,%rax,8), %xmm0
vfmaddpd %xmm11, %xmm0 , %xmm1 , %xmm11
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
.macro SOLVE_8x1
@ -347,7 +347,7 @@
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
vmovups -14*SIZE(AO,%rax,4), %xmm0
vfmaddpd %xmm9 , %xmm0 , %xmm1 , %xmm9
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
@ -377,7 +377,7 @@
vmovddup -16*SIZE(BO,%rax,1), %xmm1
vmovups -16*SIZE(AO,%rax,2), %xmm0
vfmaddpd %xmm8 , %xmm0 , %xmm1 , %xmm8
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
@ -402,7 +402,7 @@
vmovsd -16*SIZE(BO,%rax,1), %xmm1
vmovsd -16*SIZE(AO,%rax,1), %xmm0
vfmaddsd %xmm8 , %xmm0 , %xmm1 , %xmm8
addq $SIZE, %rax
addq $ SIZE, %rax
.endm
.macro SOLVE_1x1