/*********************************************************************/ /* Copyright 2009, 2010 The University of Texas at Austin. */ /* All rights reserved. */ /* */ /* Redistribution and use in source and binary forms, with or */ /* without modification, are permitted provided that the following */ /* conditions are met: */ /* */ /* 1. Redistributions of source code must retain the above */ /* copyright notice, this list of conditions and the following */ /* disclaimer. */ /* */ /* 2. Redistributions in binary form must reproduce the above */ /* copyright notice, this list of conditions and the following */ /* disclaimer in the documentation and/or other materials */ /* provided with the distribution. */ /* */ /* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */ /* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */ /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */ /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */ /* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */ /* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */ /* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */ /* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */ /* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */ /* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */ /* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */ /* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */ /* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */ /* POSSIBILITY OF SUCH DAMAGE. */ /* */ /* The views and conclusions contained in the software and */ /* documentation are those of the authors and should not be */ /* interpreted as representing official policies, either expressed */ /* or implied, of The University of Texas at Austin. */ /*********************************************************************/ #define ASSEMBLER #include "common.h" #if defined(PENTIUM4) || defined(GENERIC) #define PREFETCHSIZE 16 #define PREFETCH prefetcht0 #define PREFETCHW prefetcht0 #endif #if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) #define PREFETCHSIZE 16 #define PREFETCH prefetcht0 #define PREFETCHW prefetcht0 #endif #ifdef ATOM #define PREFETCHSIZE 16 #define PREFETCH prefetcht0 #define PREFETCHW prefetcht0 #endif #ifdef NANO #define PREFETCHSIZE 16 #define PREFETCH prefetcht0 #define PREFETCHW prefetcht0 #endif #ifdef OPTERON #define PREFETCHSIZE 16 #define PREFETCH prefetch #define PREFETCHW prefetchw #endif #ifdef GENERIC #define PREFETCHSIZE 16 #define PREFETCH prefetcht0 #define PREFETCHW prefetcht0 #endif #ifndef WINDOWS_ABI #define M ARG1 /* rdi */ #define N ARG2 /* rsi */ #define A ARG3 /* rdx */ #define LDA ARG4 /* rcx */ #define B ARG5 /* r8 */ #define I %r9 #else #define STACKSIZE 256 #define M ARG1 /* rcx */ #define N ARG2 /* rdx */ #define A ARG3 /* r8 */ #define LDA ARG4 /* r9 */ #define OLD_B 40 + 32 + STACKSIZE(%rsp) #define B %r14 #define I %r15 #endif #define J %r10 #define AO1 %r11 #define AO2 %r12 #define MM %r13 PROLOGUE PROFCODE #ifdef WINDOWS_ABI pushq %r15 pushq %r14 #endif pushq %r13 pushq %r12 #ifdef WINDOWS_ABI subq $STACKSIZE, %rsp movups %xmm6, 0(%rsp) movups %xmm7, 16(%rsp) movq OLD_B, B #endif leaq (,LDA, SIZE), LDA subq $-16 * SIZE, B movq M, MM leaq -1(M), %rax testq $SIZE, A cmovne %rax, MM testq $SIZE, LDA jne .L50 movq N, J sarq $2, J jle .L20 ALIGN_4 .L11: movq A, AO1 leaq (A, LDA, 2), AO2 leaq (A, LDA, 4), A testq $SIZE, A je .L12 movsd 0 * SIZE(AO1), %xmm0 movsd 0 * SIZE(AO1, LDA), %xmm1 movsd 0 * SIZE(AO2), %xmm2 movsd 0 * SIZE(AO2, LDA), %xmm3 unpcklpd %xmm1, %xmm0 unpcklpd %xmm3, %xmm2 movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) addq $1 * SIZE, AO1 addq $1 * SIZE, AO2 subq $-4 * SIZE, B ALIGN_3 .L12: movq MM, I sarq $3, I jle .L14 ALIGN_4 .L13: #ifdef PREFETCH PREFETCH PREFETCHSIZE * SIZE(AO1) #endif movapd 0 * SIZE(AO1), %xmm0 movapd 0 * SIZE(AO1, LDA), %xmm1 movapd 0 * SIZE(AO2), %xmm2 movapd 0 * SIZE(AO2, LDA), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) movapd %xmm4, -12 * SIZE(B) movapd %xmm6, -10 * SIZE(B) #ifdef PREFETCH PREFETCH PREFETCHSIZE * SIZE(AO1, LDA) #endif movapd 2 * SIZE(AO1), %xmm0 movapd 2 * SIZE(AO1, LDA), %xmm1 movapd 2 * SIZE(AO2), %xmm2 movapd 2 * SIZE(AO2, LDA), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 8) * SIZE(B) #endif movapd %xmm0, -8 * SIZE(B) movapd %xmm2, -6 * SIZE(B) movapd %xmm4, -4 * SIZE(B) movapd %xmm6, -2 * SIZE(B) #ifdef PREFETCH PREFETCH PREFETCHSIZE * SIZE(AO2) #endif movapd 4 * SIZE(AO1), %xmm0 movapd 4 * SIZE(AO1, LDA), %xmm1 movapd 4 * SIZE(AO2), %xmm2 movapd 4 * SIZE(AO2, LDA), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 16) * SIZE(B) #endif movapd %xmm0, 0 * SIZE(B) movapd %xmm2, 2 * SIZE(B) movapd %xmm4, 4 * SIZE(B) movapd %xmm6, 6 * SIZE(B) #ifdef PREFETCH PREFETCH PREFETCHSIZE * SIZE(AO2, LDA) #endif movapd 6 * SIZE(AO1), %xmm0 movapd 6 * SIZE(AO1, LDA), %xmm1 movapd 6 * SIZE(AO2), %xmm2 movapd 6 * SIZE(AO2, LDA), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 24) * SIZE(B) #endif movapd %xmm0, 8 * SIZE(B) movapd %xmm2, 10 * SIZE(B) movapd %xmm4, 12 * SIZE(B) movapd %xmm6, 14 * SIZE(B) addq $8 * SIZE, AO1 addq $8 * SIZE, AO2 subq $-32 * SIZE, B decq I jg .L13 ALIGN_4 .L14: testq $4, MM jle .L16 movapd 0 * SIZE(AO1), %xmm0 movapd 0 * SIZE(AO1, LDA), %xmm1 movapd 0 * SIZE(AO2), %xmm2 movapd 0 * SIZE(AO2, LDA), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) movapd %xmm4, -12 * SIZE(B) movapd %xmm6, -10 * SIZE(B) movapd 2 * SIZE(AO1), %xmm0 movapd 2 * SIZE(AO1, LDA), %xmm1 movapd 2 * SIZE(AO2), %xmm2 movapd 2 * SIZE(AO2, LDA), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 movapd %xmm0, -8 * SIZE(B) movapd %xmm2, -6 * SIZE(B) movapd %xmm4, -4 * SIZE(B) movapd %xmm6, -2 * SIZE(B) addq $4 * SIZE, AO1 addq $4 * SIZE, AO2 subq $-16 * SIZE, B ALIGN_4 .L16: testq $2, MM jle .L18 movapd 0 * SIZE(AO1), %xmm0 movapd 0 * SIZE(AO1, LDA), %xmm1 movapd 0 * SIZE(AO2), %xmm2 movapd 0 * SIZE(AO2, LDA), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) movapd %xmm4, -12 * SIZE(B) movapd %xmm6, -10 * SIZE(B) addq $2 * SIZE, AO1 addq $2 * SIZE, AO2 subq $-8 * SIZE, B ALIGN_4 .L18: testq $1, MM jle .L19 movsd 0 * SIZE(AO1), %xmm0 movsd 0 * SIZE(AO1, LDA), %xmm1 movsd 0 * SIZE(AO2), %xmm2 movsd 0 * SIZE(AO2, LDA), %xmm3 unpcklpd %xmm1, %xmm0 unpcklpd %xmm3, %xmm2 movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) subq $-4 * SIZE, B ALIGN_4 .L19: decq J jg .L11 ALIGN_4 .L20: testq $2, N jle .L30 movq A, AO1 leaq (A, LDA), AO2 leaq (A, LDA, 2), A testq $SIZE, A je .L22 movsd 0 * SIZE(AO1), %xmm0 movsd 0 * SIZE(AO2), %xmm1 unpcklpd %xmm1, %xmm0 movapd %xmm0, -16 * SIZE(B) addq $1 * SIZE, AO1 addq $1 * SIZE, AO2 subq $-2 * SIZE, B ALIGN_3 .L22: movq MM, I sarq $3, I jle .L24 ALIGN_4 .L23: #ifdef PREFETCH PREFETCH PREFETCHSIZE * 2 * SIZE(AO1) #endif movapd 0 * SIZE(AO1), %xmm0 movapd 0 * SIZE(AO2), %xmm1 movapd 2 * SIZE(AO1), %xmm2 movapd 2 * SIZE(AO2), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm0, -16 * SIZE(B) movapd %xmm4, -14 * SIZE(B) movapd %xmm2, -12 * SIZE(B) movapd %xmm6, -10 * SIZE(B) #ifdef PREFETCH PREFETCH PREFETCHSIZE * 2 * SIZE(AO2) #endif movapd 4 * SIZE(AO1), %xmm0 movapd 4 * SIZE(AO2), %xmm1 movapd 6 * SIZE(AO1), %xmm2 movapd 6 * SIZE(AO2), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm1, %xmm4 unpckhpd %xmm3, %xmm6 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 8) * SIZE(B) #endif movapd %xmm0, -8 * SIZE(B) movapd %xmm4, -6 * SIZE(B) movapd %xmm2, -4 * SIZE(B) movapd %xmm6, -2 * SIZE(B) addq $8 * SIZE, AO1 addq $8 * SIZE, AO2 subq $-16 * SIZE, B decq I jg .L23 ALIGN_4 .L24: testq $4, MM jle .L26 movapd 0 * SIZE(AO1), %xmm0 movapd 0 * SIZE(AO2), %xmm1 movapd 2 * SIZE(AO1), %xmm2 movapd 2 * SIZE(AO2), %xmm3 movapd %xmm0, %xmm4 unpcklpd %xmm1, %xmm0 unpckhpd %xmm1, %xmm4 movapd %xmm2, %xmm6 unpcklpd %xmm3, %xmm2 unpckhpd %xmm3, %xmm6 movapd %xmm0, -16 * SIZE(B) movapd %xmm4, -14 * SIZE(B) movapd %xmm2, -12 * SIZE(B) movapd %xmm6, -10 * SIZE(B) addq $4 * SIZE, AO1 addq $4 * SIZE, AO2 subq $-8 * SIZE, B ALIGN_4 .L26: testq $2, MM jle .L28 movapd 0 * SIZE(AO1), %xmm0 movapd 0 * SIZE(AO2), %xmm1 movapd %xmm0, %xmm2 unpcklpd %xmm1, %xmm0 unpckhpd %xmm1, %xmm2 movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) addq $2 * SIZE, AO1 addq $2 * SIZE, AO2 subq $-4 * SIZE, B ALIGN_4 .L28: testq $1, MM jle .L30 movsd 0 * SIZE(AO1), %xmm0 movsd 0 * SIZE(AO2), %xmm1 unpcklpd %xmm1, %xmm0 movapd %xmm0, -16 * SIZE(B) subq $-2 * SIZE, B ALIGN_4 .L30: testq $1, N jle .L999 movq A, AO1 testq $SIZE, A jne .L35 movq MM, I sarq $3, I jle .L32 ALIGN_4 .L31: #ifdef PREFETCH PREFETCH PREFETCHSIZE * 4 * SIZE(AO1) #endif movapd 0 * SIZE(AO1), %xmm0 movapd 2 * SIZE(AO1), %xmm1 movapd 4 * SIZE(AO1), %xmm2 movapd 6 * SIZE(AO1), %xmm3 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm0, -16 * SIZE(B) movapd %xmm1, -14 * SIZE(B) movapd %xmm2, -12 * SIZE(B) movapd %xmm3, -10 * SIZE(B) addq $8 * SIZE, AO1 subq $-8 * SIZE, B decq I jg .L31 ALIGN_4 .L32: testq $4, MM jle .L33 movapd 0 * SIZE(AO1), %xmm0 movapd 2 * SIZE(AO1), %xmm1 movapd %xmm0, -16 * SIZE(B) movapd %xmm1, -14 * SIZE(B) addq $4 * SIZE, AO1 subq $-4 * SIZE, B ALIGN_4 .L33: testq $2, MM jle .L34 movapd 0 * SIZE(AO1), %xmm0 movapd %xmm0, -16 * SIZE(B) addq $2 * SIZE, AO1 subq $-2 * SIZE, B ALIGN_4 .L34: testq $1, MM jle .L999 movsd 0 * SIZE(AO1), %xmm0 movlpd %xmm0, -16 * SIZE(B) jmp .L999 ALIGN_4 .L35: movapd -1 * SIZE(AO1), %xmm0 movq MM, I sarq $3, I jle .L36 ALIGN_4 .L36: #ifdef PREFETCH PREFETCH PREFETCHSIZE * 4 * SIZE(AO1) #endif movapd 1 * SIZE(AO1), %xmm1 movapd 3 * SIZE(AO1), %xmm2 movapd 5 * SIZE(AO1), %xmm3 movapd 7 * SIZE(AO1), %xmm4 shufpd $1, %xmm1, %xmm0 shufpd $1, %xmm2, %xmm1 shufpd $1, %xmm3, %xmm2 shufpd $1, %xmm4, %xmm3 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm0, -16 * SIZE(B) movapd %xmm1, -14 * SIZE(B) movapd %xmm2, -12 * SIZE(B) movapd %xmm3, -10 * SIZE(B) movapd %xmm4, %xmm0 addq $8 * SIZE, AO1 subq $-8 * SIZE, B decq I jg .L36 ALIGN_4 .L37: testq $4, MM jle .L38 movapd 1 * SIZE(AO1), %xmm1 movapd 3 * SIZE(AO1), %xmm2 shufpd $1, %xmm1, %xmm0 shufpd $1, %xmm2, %xmm1 movapd %xmm0, -16 * SIZE(B) movapd %xmm1, -14 * SIZE(B) movapd %xmm2, %xmm0 addq $4 * SIZE, AO1 addq $4 * SIZE, B ALIGN_4 .L38: testq $2, MM jle .L39 movapd 1 * SIZE(AO1), %xmm1 shufpd $1, %xmm1, %xmm0 movapd %xmm0, -16 * SIZE(B) movapd %xmm1, %xmm0 addq $2 * SIZE, AO1 subq $-2 * SIZE, B ALIGN_4 .L39: testq $1, MM jle .L999 shufpd $1, %xmm0, %xmm0 movlpd %xmm0, -16 * SIZE(B) jmp .L999 ALIGN_4 .L50: movq N, J sarq $2, J jle .L60 ALIGN_4 .L51: movq A, AO1 leaq (A, LDA, 2), AO2 leaq (A, LDA, 4), A testq $SIZE, A je .L52 movsd 0 * SIZE(AO1), %xmm0 movsd 0 * SIZE(AO1, LDA), %xmm1 movsd 0 * SIZE(AO2), %xmm2 movsd 0 * SIZE(AO2, LDA), %xmm3 unpcklpd %xmm1, %xmm0 unpcklpd %xmm3, %xmm2 movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) addq $1 * SIZE, AO1 addq $1 * SIZE, AO2 subq $-4 * SIZE, B ALIGN_3 .L52: movapd -1 * SIZE(AO1, LDA), %xmm5 movapd -1 * SIZE(AO2, LDA), %xmm7 movq MM, I sarq $3, I jle .L54 ALIGN_4 .L53: #ifdef PREFETCH PREFETCH PREFETCHSIZE * SIZE(AO1) #endif movapd 0 * SIZE(AO1), %xmm0 movapd 1 * SIZE(AO1, LDA), %xmm1 movapd 0 * SIZE(AO2), %xmm2 movapd 1 * SIZE(AO2, LDA), %xmm3 movsd %xmm0, %xmm5 movsd %xmm2, %xmm7 shufpd $1, %xmm1, %xmm0 shufpd $1, %xmm3, %xmm2 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm5, -16 * SIZE(B) movapd %xmm7, -14 * SIZE(B) movapd %xmm0, -12 * SIZE(B) movapd %xmm2, -10 * SIZE(B) #ifdef PREFETCH PREFETCH PREFETCHSIZE * SIZE(AO1, LDA) #endif movapd 2 * SIZE(AO1), %xmm0 movapd 3 * SIZE(AO1, LDA), %xmm5 movapd 2 * SIZE(AO2), %xmm2 movapd 3 * SIZE(AO2, LDA), %xmm7 movsd %xmm0, %xmm1 movsd %xmm2, %xmm3 shufpd $1, %xmm5, %xmm0 shufpd $1, %xmm7, %xmm2 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 8) * SIZE(B) #endif movapd %xmm1, -8 * SIZE(B) movapd %xmm3, -6 * SIZE(B) movapd %xmm0, -4 * SIZE(B) movapd %xmm2, -2 * SIZE(B) #ifdef PREFETCH PREFETCH PREFETCHSIZE * SIZE(AO2) #endif movapd 4 * SIZE(AO1), %xmm0 movapd 5 * SIZE(AO1, LDA), %xmm1 movapd 4 * SIZE(AO2), %xmm2 movapd 5 * SIZE(AO2, LDA), %xmm3 movsd %xmm0, %xmm5 movsd %xmm2, %xmm7 shufpd $1, %xmm1, %xmm0 shufpd $1, %xmm3, %xmm2 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 16) * SIZE(B) #endif movapd %xmm5, 0 * SIZE(B) movapd %xmm7, 2 * SIZE(B) movapd %xmm0, 4 * SIZE(B) movapd %xmm2, 6 * SIZE(B) #ifdef PREFETCH PREFETCH PREFETCHSIZE * SIZE(AO2, LDA) #endif movapd 6 * SIZE(AO1), %xmm0 movapd 7 * SIZE(AO1, LDA), %xmm5 movapd 6 * SIZE(AO2), %xmm2 movapd 7 * SIZE(AO2, LDA), %xmm7 movsd %xmm0, %xmm1 movsd %xmm2, %xmm3 shufpd $1, %xmm5, %xmm0 shufpd $1, %xmm7, %xmm2 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 24) * SIZE(B) #endif movapd %xmm1, 8 * SIZE(B) movapd %xmm3, 10 * SIZE(B) movapd %xmm0, 12 * SIZE(B) movapd %xmm2, 14 * SIZE(B) addq $8 * SIZE, AO1 addq $8 * SIZE, AO2 subq $-32 * SIZE, B decq I jg .L53 ALIGN_4 .L54: testq $4, MM jle .L56 movapd 0 * SIZE(AO1), %xmm0 movapd 1 * SIZE(AO1, LDA), %xmm1 movapd 0 * SIZE(AO2), %xmm2 movapd 1 * SIZE(AO2, LDA), %xmm3 movsd %xmm0, %xmm5 shufpd $1, %xmm1, %xmm0 movsd %xmm2, %xmm7 shufpd $1, %xmm3, %xmm2 movapd %xmm5, -16 * SIZE(B) movapd %xmm7, -14 * SIZE(B) movapd %xmm0, -12 * SIZE(B) movapd %xmm2, -10 * SIZE(B) movapd 2 * SIZE(AO1), %xmm0 movapd 3 * SIZE(AO1, LDA), %xmm5 movapd 2 * SIZE(AO2), %xmm2 movapd 3 * SIZE(AO2, LDA), %xmm7 movsd %xmm0, %xmm1 shufpd $1, %xmm5, %xmm0 movsd %xmm2, %xmm3 shufpd $1, %xmm7, %xmm2 movapd %xmm1, -8 * SIZE(B) movapd %xmm3, -6 * SIZE(B) movapd %xmm0, -4 * SIZE(B) movapd %xmm2, -2 * SIZE(B) addq $4 * SIZE, AO1 addq $4 * SIZE, AO2 subq $-16 * SIZE, B ALIGN_4 .L56: testq $2, MM jle .L58 movapd 0 * SIZE(AO1), %xmm0 movapd 1 * SIZE(AO1, LDA), %xmm1 movapd 0 * SIZE(AO2), %xmm2 movapd 1 * SIZE(AO2, LDA), %xmm3 movsd %xmm0, %xmm5 movsd %xmm2, %xmm7 shufpd $1, %xmm1, %xmm0 shufpd $1, %xmm3, %xmm2 movapd %xmm5, -16 * SIZE(B) movapd %xmm7, -14 * SIZE(B) movapd %xmm0, -12 * SIZE(B) movapd %xmm2, -10 * SIZE(B) addq $2 * SIZE, AO1 addq $2 * SIZE, AO2 subq $-8 * SIZE, B ALIGN_4 .L58: testq $1, MM jle .L59 movsd 0 * SIZE(AO1), %xmm0 movsd 0 * SIZE(AO1, LDA), %xmm1 movsd 0 * SIZE(AO2), %xmm2 movsd 0 * SIZE(AO2, LDA), %xmm3 unpcklpd %xmm1, %xmm0 unpcklpd %xmm3, %xmm2 movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) subq $-4 * SIZE, B ALIGN_4 .L59: decq J jg .L51 ALIGN_4 .L60: testq $2, N jle .L70 movq A, AO1 leaq (A, LDA), AO2 leaq (A, LDA, 2), A testq $SIZE, A je .L62 movsd 0 * SIZE(AO1), %xmm0 movsd 0 * SIZE(AO2), %xmm1 unpcklpd %xmm1, %xmm0 movapd %xmm0, -16 * SIZE(B) addq $1 * SIZE, AO1 addq $1 * SIZE, AO2 subq $-2 * SIZE, B ALIGN_3 .L62: movapd -1 * SIZE(AO2), %xmm5 movq MM, I sarq $3, I jle .L64 ALIGN_4 .L63: #ifdef PREFETCH PREFETCH PREFETCHSIZE * 2 * SIZE(AO1) #endif movapd 0 * SIZE(AO1), %xmm0 movapd 1 * SIZE(AO2), %xmm1 movapd 2 * SIZE(AO1), %xmm2 movapd 3 * SIZE(AO2), %xmm3 movsd %xmm0, %xmm5 shufpd $1, %xmm1, %xmm0 movsd %xmm2, %xmm1 shufpd $1, %xmm3, %xmm2 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm5, -16 * SIZE(B) movapd %xmm0, -14 * SIZE(B) movapd %xmm1, -12 * SIZE(B) movapd %xmm2, -10 * SIZE(B) #ifdef PREFETCH PREFETCH PREFETCHSIZE * 2 * SIZE(AO2) #endif movapd 4 * SIZE(AO1), %xmm0 movapd 5 * SIZE(AO2), %xmm1 movapd 6 * SIZE(AO1), %xmm2 movapd 7 * SIZE(AO2), %xmm5 movsd %xmm0, %xmm3 shufpd $1, %xmm1, %xmm0 movsd %xmm2, %xmm1 shufpd $1, %xmm5, %xmm2 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm3, -8 * SIZE(B) movapd %xmm0, -6 * SIZE(B) movapd %xmm1, -4 * SIZE(B) movapd %xmm2, -2 * SIZE(B) addq $8 * SIZE, AO1 addq $8 * SIZE, AO2 subq $-16 * SIZE, B decq I jg .L63 ALIGN_4 .L64: testq $4, MM jle .L66 movapd 0 * SIZE(AO1), %xmm0 movapd 1 * SIZE(AO2), %xmm1 movapd 2 * SIZE(AO1), %xmm2 movapd 3 * SIZE(AO2), %xmm3 movsd %xmm0, %xmm5 shufpd $1, %xmm1, %xmm0 movsd %xmm2, %xmm1 shufpd $1, %xmm3, %xmm2 movapd %xmm5, -16 * SIZE(B) movapd %xmm0, -14 * SIZE(B) movapd %xmm1, -12 * SIZE(B) movapd %xmm2, -10 * SIZE(B) movaps %xmm3, %xmm5 addq $4 * SIZE, AO1 addq $4 * SIZE, AO2 subq $-8 * SIZE, B ALIGN_4 .L66: testq $2, MM jle .L68 movapd 0 * SIZE(AO1), %xmm0 movapd 1 * SIZE(AO2), %xmm1 movsd %xmm0, %xmm5 shufpd $1, %xmm1, %xmm0 movapd %xmm5, -16 * SIZE(B) movapd %xmm0, -14 * SIZE(B) addq $2 * SIZE, AO1 addq $2 * SIZE, AO2 subq $-4 * SIZE, B ALIGN_4 .L68: testq $1, MM jle .L70 movsd 0 * SIZE(AO1), %xmm0 movsd 0 * SIZE(AO2), %xmm1 unpcklpd %xmm1, %xmm0 movapd %xmm0, -16 * SIZE(B) subq $-2 * SIZE, B ALIGN_4 .L70: testq $1, N jle .L999 movq A, AO1 testq $SIZE, A jne .L75 movq MM, I sarq $3, I jle .L72 ALIGN_4 .L71: #ifdef PREFETCH PREFETCH PREFETCHSIZE * 4 * SIZE(AO1) #endif movapd 0 * SIZE(AO1), %xmm0 movapd 2 * SIZE(AO1), %xmm2 movapd 4 * SIZE(AO1), %xmm4 movapd 6 * SIZE(AO1), %xmm6 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) movapd %xmm4, -12 * SIZE(B) movapd %xmm6, -10 * SIZE(B) addq $8 * SIZE, AO1 subq $-8 * SIZE, B decq I jg .L71 ALIGN_4 .L72: testq $4, MM jle .L73 movapd 0 * SIZE(AO1), %xmm0 movapd 2 * SIZE(AO1), %xmm2 movapd %xmm0, -16 * SIZE(B) movapd %xmm2, -14 * SIZE(B) addq $4 * SIZE, AO1 subq $-4 * SIZE, B ALIGN_4 .L73: testq $2, MM jle .L74 movapd 0 * SIZE(AO1), %xmm0 movapd %xmm0, -16 * SIZE(B) addq $2 * SIZE, AO1 subq $-2 * SIZE, B ALIGN_4 .L74: testq $1, MM jle .L999 movsd 0 * SIZE(AO1), %xmm0 movlpd %xmm0, -16 * SIZE(B) jmp .L999 ALIGN_4 .L75: movapd -1 * SIZE(AO1), %xmm0 movq MM, I sarq $3, I jle .L76 ALIGN_4 .L76: #ifdef PREFETCH PREFETCH PREFETCHSIZE * 4 * SIZE(AO1) #endif movapd 1 * SIZE(AO1), %xmm1 movapd 3 * SIZE(AO1), %xmm2 movapd 5 * SIZE(AO1), %xmm3 movapd 7 * SIZE(AO1), %xmm4 shufpd $1, %xmm1, %xmm0 shufpd $1, %xmm2, %xmm1 shufpd $1, %xmm3, %xmm2 shufpd $1, %xmm4, %xmm3 #ifdef PREFETCHW PREFETCHW (PREFETCHSIZE * 4 + 0) * SIZE(B) #endif movapd %xmm0, -16 * SIZE(B) movapd %xmm1, -14 * SIZE(B) movapd %xmm2, -12 * SIZE(B) movapd %xmm3, -10 * SIZE(B) movapd %xmm4, %xmm0 addq $8 * SIZE, AO1 subq $-8 * SIZE, B decq I jg .L76 ALIGN_4 .L77: testq $4, MM jle .L78 movapd 1 * SIZE(AO1), %xmm1 movapd 3 * SIZE(AO1), %xmm2 shufpd $1, %xmm1, %xmm0 shufpd $1, %xmm2, %xmm1 movapd %xmm0, -16 * SIZE(B) movapd %xmm1, -14 * SIZE(B) movapd %xmm2, %xmm0 addq $4 * SIZE, AO1 addq $4 * SIZE, B ALIGN_4 .L78: testq $2, MM jle .L79 movapd 1 * SIZE(AO1), %xmm1 shufpd $1, %xmm1, %xmm0 movapd %xmm0, -16 * SIZE(B) movapd %xmm1, %xmm0 addq $2 * SIZE, AO1 subq $-2 * SIZE, B ALIGN_4 .L79: testq $1, MM jle .L999 shufpd $1, %xmm0, %xmm0 movlpd %xmm0, -16 * SIZE(B) ALIGN_4 .L999: #ifdef WINDOWS_ABI movups 0(%rsp), %xmm6 movups 16(%rsp), %xmm7 addq $STACKSIZE, %rsp #endif popq %r12 popq %r13 #ifdef WINDOWS_ABI popq %r14 popq %r15 #endif ret EPILOGUE