Created
March 31, 2017 05:13
-
-
Save 9il/e678e40f4432f9a8f52b6efbf1b79c74 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
.section __TEXT,__text,regular,pure_instructions | |
.globl _dot_reg_basic_generic | |
.p2align 4, 0x90 | |
_dot_reg_basic_generic: | |
.cfi_startproc | |
vxorps %ymm14, %ymm14, %ymm14 | |
vxorps %ymm1, %ymm1, %ymm1 | |
vxorps %ymm9, %ymm9, %ymm9 | |
vxorps %ymm2, %ymm2, %ymm2 | |
vxorps %ymm7, %ymm7, %ymm7 | |
vxorps %ymm3, %ymm3, %ymm3 | |
vxorps %ymm10, %ymm10, %ymm10 | |
vxorps %ymm4, %ymm4, %ymm4 | |
vxorps %ymm8, %ymm8, %ymm8 | |
vxorps %ymm5, %ymm5, %ymm5 | |
vxorps %ymm11, %ymm11, %ymm11 | |
vxorps %ymm6, %ymm6, %ymm6 | |
movq %rdx, %r8 | |
movq %rdi, %rax | |
.p2align 4, 0x90 | |
LBB0_1: | |
vmovaps %ymm11, %ymm12 | |
vmovaps %ymm8, %ymm13 | |
vmovups %ymm10, -120(%rsp) | |
vmovups %ymm7, -88(%rsp) | |
vmovups %ymm9, -56(%rsp) | |
prefetcht0 512(%rax) | |
vmovaps (%rax), %ymm0 | |
vmovaps 32(%rax), %ymm15 | |
vbroadcastss (%rsi), %ymm11 | |
vbroadcastss 4(%rsi), %ymm8 | |
vfmadd231ps %ymm0, %ymm11, %ymm6 | |
vfmadd213ps %ymm12, %ymm15, %ymm11 | |
vfmadd231ps %ymm0, %ymm8, %ymm5 | |
vfmadd213ps %ymm13, %ymm15, %ymm8 | |
vbroadcastss 8(%rsi), %ymm10 | |
vbroadcastss 12(%rsi), %ymm7 | |
vfmadd231ps %ymm0, %ymm10, %ymm4 | |
vfmadd213ps -120(%rsp), %ymm15, %ymm10 | |
vfmadd231ps %ymm0, %ymm7, %ymm3 | |
vfmadd213ps -88(%rsp), %ymm15, %ymm7 | |
vbroadcastss 16(%rsi), %ymm9 | |
vbroadcastss 20(%rsi), %ymm12 | |
vfmadd231ps %ymm0, %ymm9, %ymm2 | |
vfmadd213ps -56(%rsp), %ymm15, %ymm9 | |
vfmadd231ps %ymm0, %ymm12, %ymm1 | |
vfmadd231ps %ymm15, %ymm12, %ymm14 | |
addq $64, %rax | |
addq $24, %rsi | |
addq $-1, %r8 | |
jne LBB0_1 | |
shlq $6, %rdx | |
addq %rdx, %rdi | |
vmovaps %ymm6, (%rcx) | |
vmovaps %ymm11, 32(%rcx) | |
vmovaps %ymm5, 64(%rcx) | |
vmovaps %ymm8, 96(%rcx) | |
vmovaps %ymm4, 128(%rcx) | |
vmovaps %ymm10, 160(%rcx) | |
vmovaps %ymm3, 192(%rcx) | |
vmovaps %ymm7, 224(%rcx) | |
vmovaps %ymm2, 256(%rcx) | |
vmovaps %ymm9, 288(%rcx) | |
vmovaps %ymm1, 320(%rcx) | |
vmovaps %ymm14, 352(%rcx) | |
movq %rdi, %rax | |
vzeroupper | |
retq | |
.cfi_endproc | |
.subsections_via_symbols |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment