/*********************************************************************/
/* 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"
#define STACK 16
#define OLD_M 4 + STACK(%esi)
#define OLD_N 8 + STACK(%esi)
#define OLD_K 12 + STACK(%esi)
#define OLD_ALPHA_R 16 + STACK(%esi)
#define OLD_ALPHA_I 20 + STACK(%esi)
#define OLD_A 24 + STACK(%esi)
#define OLD_B 28 + STACK(%esi)
#define OLD_C 32 + STACK(%esi)
#define OLD_LDC 36 + STACK(%esi)
#define ALPHA 0(%esp)
#define K 16(%esp)
#define N 20(%esp)
#define M 24(%esp)
#define A 28(%esp)
#define C 32(%esp)
#define J 36(%esp)
#define BX 40(%esp)
#define OLD_STACK 44(%esp)
#define OFFSET 48(%esp)
#define KK 52(%esp)
#define KKK 56(%esp)
#define BUFFER 128(%esp)
#ifdef ATHLON
#define PREFETCH prefetch
#define PREFETCHSIZE 64
#endif
#if defined(OPTERON) || defined(BARCELONA)
#define PREFETCH prefetch
#define PREFETCHSIZE (16 * 10 + 8)
#endif
#ifdef PENTIUM4
#define PREFETCH prefetcht0
#define PREFETCHSIZE 96
#endif
#define AA %edx
#define BB %ecx
#define LDC %ebp
#if defined(OPTERON) || defined(BARCELONA)
#define movsd movlps
#endif
#if defined(OPTERON) || defined(BARCELONA)
#define KERNEL1(address) \
mulps %xmm0, %xmm2; \
addps %xmm2, %xmm4; \
PREFETCH (PREFETCHSIZE + 0) * SIZE + (address) * 1 * SIZE(AA); \
movaps 4 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm0, %xmm2; \
addps %xmm2, %xmm5; \
movaps 8 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm0, %xmm2; \
mulps 12 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \
addps %xmm2, %xmm6; \
movaps 32 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
addps %xmm0, %xmm7; \
movaps 4 * SIZE + (address) * 1 * SIZE(AA), %xmm0
#define KERNEL2(address) \
mulps %xmm0, %xmm3; \
addps %xmm3, %xmm4; \
movaps 20 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm0, %xmm3; \
addps %xmm3, %xmm5; \
movaps 24 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm0, %xmm3; \
mulps 28 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \
addps %xmm3, %xmm6; \
movaps 48 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
addps %xmm0, %xmm7; \
movaps 8 * SIZE + (address) * 1 * SIZE(AA), %xmm0
#define KERNEL3(address) \
mulps %xmm0, %xmm2; \
addps %xmm2, %xmm4; \
movaps 36 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm0, %xmm2; \
addps %xmm2, %xmm5; \
movaps 40 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm0, %xmm2; \
mulps 44 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \
addps %xmm2, %xmm6; \
movaps 64 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
addps %xmm0, %xmm7; \
movaps 12 * SIZE + (address) * 1 * SIZE(AA), %xmm0
#define KERNEL4(address) \
mulps %xmm0, %xmm3; \
addps %xmm3, %xmm4; \
movaps 52 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm0, %xmm3; \
addps %xmm3, %xmm5; \
movaps 56 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm0, %xmm3; \
mulps 60 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \
addps %xmm3, %xmm6; \
movaps 80 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
addps %xmm0, %xmm7; \
movaps 32 * SIZE + (address) * 1 * SIZE(AA), %xmm0
#define KERNEL5(address) \
PREFETCH (PREFETCHSIZE + 16) * SIZE + (address) * 1 * SIZE(AA); \
mulps %xmm1, %xmm2; \
addps %xmm2, %xmm4; \
movaps 68 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm1, %xmm2; \
addps %xmm2, %xmm5; \
movaps 72 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm1, %xmm2; \
mulps 76 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \
addps %xmm2, %xmm6; \
movaps 96 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
addps %xmm1, %xmm7; \
movaps 20 * SIZE + (address) * 1 * SIZE(AA), %xmm1
#define KERNEL6(address) \
mulps %xmm1, %xmm3; \
addps %xmm3, %xmm4; \
movaps 84 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm1, %xmm3; \
addps %xmm3, %xmm5; \
movaps 88 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm1, %xmm3; \
mulps 92 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \
addps %xmm3, %xmm6; \
movaps 112 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
addps %xmm1, %xmm7; \
movaps 24 * SIZE + (address) * 1 * SIZE(AA), %xmm1
#define KERNEL7(address) \
mulps %xmm1, %xmm2; \
addps %xmm2, %xmm4; \
movaps 100 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm1, %xmm2; \
addps %xmm2, %xmm5; \
movaps 104 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm1, %xmm2; \
mulps 108 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \
addps %xmm2, %xmm6; \
movaps 128 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
addps %xmm1, %xmm7; \
movaps 28 * SIZE + (address) * 1 * SIZE(AA), %xmm1
#define KERNEL8(address) \
mulps %xmm1, %xmm3; \
addps %xmm3, %xmm4; \
movaps 116 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm1, %xmm3; \
addps %xmm3, %xmm5; \
movaps 120 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm1, %xmm3; \
mulps 124 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \
addps %xmm3, %xmm6; \
movaps 144 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
addps %xmm1, %xmm7; \
movaps 48 * SIZE + (address) * 1 * SIZE(AA), %xmm1;
#endif
#ifdef PENTIUM4
#define KERNEL1(address) \
mulps %xmm0, %xmm2; \
addps %xmm2, %xmm4; \
movaps 4 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm0, %xmm2; \
PREFETCH (PREFETCHSIZE + 0) * SIZE + (address) * 1 * SIZE(AA); \
addps %xmm2, %xmm5; \
movaps 8 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm0, %xmm2; \
mulps 12 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \
addps %xmm2, %xmm6; \
movaps 32 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
addps %xmm0, %xmm7; \
movaps 4 * SIZE + (address) * 1 * SIZE(AA), %xmm0
#define KERNEL2(address) \
mulps %xmm0, %xmm3; \
addps %xmm3, %xmm4; \
movaps 20 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm0, %xmm3; \
addps %xmm3, %xmm5; \
movaps 24 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm0, %xmm3; \
mulps 28 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \
addps %xmm3, %xmm6; \
movaps 48 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
addps %xmm0, %xmm7; \
movaps 8 * SIZE + (address) * 1 * SIZE(AA), %xmm0
#define KERNEL3(address) \
mulps %xmm0, %xmm2; \
addps %xmm2, %xmm4; \
movaps 36 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm0, %xmm2; \
addps %xmm2, %xmm5; \
movaps 40 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm0, %xmm2; \
mulps 44 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \
addps %xmm2, %xmm6; \
movaps 64 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
addps %xmm0, %xmm7; \
movaps 12 * SIZE + (address) * 1 * SIZE(AA), %xmm0
#define KERNEL4(address) \
mulps %xmm0, %xmm3; \
addps %xmm3, %xmm4; \
movaps 52 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm0, %xmm3; \
addps %xmm3, %xmm5; \
movaps 56 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm0, %xmm3; \
mulps 60 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \
addps %xmm3, %xmm6; \
movaps 80 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
addps %xmm0, %xmm7; \
movaps 32 * SIZE + (address) * 1 * SIZE(AA), %xmm0
#define KERNEL5(address) \
mulps %xmm1, %xmm2; \
addps %xmm2, %xmm4; \
movaps 68 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm1, %xmm2; \
addps %xmm2, %xmm5; \
movaps 72 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm1, %xmm2; \
mulps 76 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \
addps %xmm2, %xmm6; \
movaps 96 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
addps %xmm1, %xmm7; \
movaps 20 * SIZE + (address) * 1 * SIZE(AA), %xmm1
#define KERNEL6(address) \
mulps %xmm1, %xmm3; \
addps %xmm3, %xmm4; \
movaps 84 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm1, %xmm3; \
addps %xmm3, %xmm5; \
movaps 88 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm1, %xmm3; \
mulps 92 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \
addps %xmm3, %xmm6; \
movaps 112 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
addps %xmm1, %xmm7; \
movaps 24 * SIZE + (address) * 1 * SIZE(AA), %xmm1
#define KERNEL7(address) \
mulps %xmm1, %xmm2; \
addps %xmm2, %xmm4; \
movaps 100 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm1, %xmm2; \
addps %xmm2, %xmm5; \
movaps 104 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
mulps %xmm1, %xmm2; \
mulps 108 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \
addps %xmm2, %xmm6; \
movaps 128 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \
addps %xmm1, %xmm7; \
movaps 28 * SIZE + (address) * 1 * SIZE(AA), %xmm1
#define KERNEL8(address) \
mulps %xmm1, %xmm3; \
addps %xmm3, %xmm4; \
movaps 116 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm1, %xmm3; \
addps %xmm3, %xmm5; \
movaps 120 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
mulps %xmm1, %xmm3; \
mulps 124 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \
addps %xmm3, %xmm6; \
movaps 144 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \
addps %xmm1, %xmm7; \
movaps 48 * SIZE + (address) * 1 * SIZE(AA), %xmm1
#endif
PROLOGUE
pushl %ebp
pushl %edi
pushl %esi
pushl %ebx
PROFCODE
EMMS
movl %esp, %esi # save old stack
subl $128 + LOCAL_BUFFER_SIZE, %esp
movl OLD_M, %ebx
andl $-1024, %esp # align stack
STACK_TOUCHING
movl OLD_N, %eax
movl OLD_K, %ecx
movl OLD_A, %edx
movss OLD_ALPHA_R, %xmm0
movss OLD_ALPHA_I, %xmm1
movl %ebx, M
movl %eax, N
movl %ecx, K
movl %edx, A
movl %esi, OLD_STACK
movl OLD_B, %edi
movl OLD_C, %ebx
unpcklps %xmm1, %xmm0
movlhps %xmm0, %xmm0
movaps %xmm0, ALPHA
movl %ebx, C
movl OLD_LDC, LDC
sall $ZBASE_SHIFT, LDC
sarl $2, %eax
movl %eax, J
jle .L40
.L01:
#if defined(TRMMKERNEL) && defined(LEFT)
movl OFFSET, %eax
movl %eax, KK
#endif
/* Copying to Sub Buffer */
leal BUFFER, %ecx
movl K, %eax
sarl $1, %eax
jle .L05
ALIGN_4
.L02:
#ifdef HAVE_SSE2
movss 0 * SIZE(%edi), %xmm0
movss 1 * SIZE(%edi), %xmm1
movss 2 * SIZE(%edi), %xmm2
movss 3 * SIZE(%edi), %xmm3
movss 4 * SIZE(%edi), %xmm4
movss 5 * SIZE(%edi), %xmm5
movss 6 * SIZE(%edi), %xmm6
movss 7 * SIZE(%edi), %xmm7
shufps $0, %xmm0, %xmm0
shufps $0, %xmm1, %xmm1
shufps $0, %xmm2, %xmm2
shufps $0, %xmm3, %xmm3
shufps $0, %xmm4, %xmm4
shufps $0, %xmm5, %xmm5
shufps $0, %xmm6, %xmm6
shufps $0, %xmm7, %xmm7
movaps %xmm0, 0 * SIZE(%ecx)
movaps %xmm1, 4 * SIZE(%ecx)
movaps %xmm2, 8 * SIZE(%ecx)
movaps %xmm3, 12 * SIZE(%ecx)
movaps %xmm4, 16 * SIZE(%ecx)
movaps %xmm5, 20 * SIZE(%ecx)
movaps %xmm6, 24 * SIZE(%ecx)
movaps %xmm7, 28 * SIZE(%ecx)
#else
movd 0 * SIZE(%edi), %mm0
movd 1 * SIZE(%edi), %mm1
movd 2 * SIZE(%edi), %mm2
movd 3 * SIZE(%edi), %mm3
movd 4 * SIZE(%edi), %mm4
movd 5 * SIZE(%edi), %mm5
movd 6 * SIZE(%edi), %mm6
movd 7 * SIZE(%edi), %mm7
movd %mm0, 0 * SIZE(%ecx)
movd %mm0, 1 * SIZE(%ecx)
movd %mm0, 2 * SIZE(%ecx)
movd %mm0, 3 * SIZE(%ecx)
movd %mm1, 4 * SIZE(%ecx)
movd %mm1, 5 * SIZE(%ecx)
movd %mm1, 6 * SIZE(%ecx)
movd %mm1, 7 * SIZE(%ecx)
movd %mm2, 8 * SIZE(%ecx)
movd %mm2, 9 * SIZE(%ecx)
movd %mm2, 10 * SIZE(%ecx)
movd %mm2, 11 * SIZE(%ecx)
movd %mm3, 12 * SIZE(%ecx)
movd %mm3, 13 * SIZE(%ecx)
movd %mm3, 14 * SIZE(%ecx)
movd %mm3, 15 * SIZE(%ecx)
movd %mm4, 16 * SIZE(%ecx)
movd %mm4, 17 * SIZE(%ecx)
movd %mm4, 18 * SIZE(%ecx)
movd %mm4, 19 * SIZE(%ecx)
movd %mm5, 20 * SIZE(%ecx)
movd %mm5, 21 * SIZE(%ecx)
movd %mm5, 22 * SIZE(%ecx)
movd %mm5, 23 * SIZE(%ecx)
movd %mm6, 24 * SIZE(%ecx)
movd %mm6, 25 * SIZE(%ecx)
movd %mm6, 26 * SIZE(%ecx)
movd %mm6, 27 * SIZE(%ecx)
movd %mm7, 28 * SIZE(%ecx)
movd %mm7, 29 * SIZE(%ecx)
movd %mm7, 30 * SIZE(%ecx)
movd %mm7, 31 * SIZE(%ecx)
#endif
#ifdef PENTIUM4
prefetcht2 112 * SIZE(%ecx)
#endif
#if defined(OPTERON) || defined(BARCELONA)
prefetchnta 80 * SIZE(%edi)
prefetchw 112 * SIZE(%ecx)
prefetchw 120 * SIZE(%ecx)
#endif
addl $ 8 * SIZE, %edi
addl $32 * SIZE, %ecx
decl %eax
jne .L02
ALIGN_2
.L05:
movl K, %eax
andl $1, %eax
BRANCH
jle .L10
#ifdef HAVE_SSE2
movss 0 * SIZE(%edi), %xmm0
movss 1 * SIZE(%edi), %xmm1
movss 2 * SIZE(%edi), %xmm2
movss 3 * SIZE(%edi), %xmm3
shufps $0, %xmm0, %xmm0
shufps $0, %xmm1, %xmm1
shufps $0, %xmm2, %xmm2
shufps $0, %xmm3, %xmm3
movaps %xmm0, 0 * SIZE(%ecx)
movaps %xmm1, 4 * SIZE(%ecx)
movaps %xmm2, 8 * SIZE(%ecx)
movaps %xmm3, 12 * SIZE(%ecx)
#else
movd 0 * SIZE(%edi), %mm0
movd 1 * SIZE(%edi), %mm1
movd 2 * SIZE(%edi), %mm2
movd 3 * SIZE(%edi), %mm3
movd %mm0, 0 * SIZE(%ecx)
movd %mm0, 1 * SIZE(%ecx)
movd %mm0, 2 * SIZE(%ecx)
movd %mm0, 3 * SIZE(%ecx)
movd %mm1, 4 * SIZE(%ecx)
movd %mm1, 5 * SIZE(%ecx)
movd %mm1, 6 * SIZE(%ecx)
movd %mm1, 7 * SIZE(%ecx)
movd %mm2, 8 * SIZE(%ecx)
movd %mm2, 9 * SIZE(%ecx)
movd %mm2, 10 * SIZE(%ecx)
movd %mm2, 11 * SIZE(%ecx)
movd %mm3, 12 * SIZE(%ecx)
movd %mm3, 13 * SIZE(%ecx)
movd %mm3, 14 * SIZE(%ecx)
movd %mm3, 15 * SIZE(%ecx)
#endif
addl $4 * SIZE, %edi
ALIGN_4
.L10:
movl %edi, BX
movl C, %esi # coffset = c
movl A, %edx # aoffset = a
movl M, %ebx
sarl $2, %ebx # i = (m >> 2)
jle .L20
ALIGN_4
.L11:
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 8), %eax
leal (AA, %eax, 2), AA
leal (BB, %eax, 8), BB
#endif
movl BX, %eax
#ifdef HAVE_SSE
prefetcht2 0 * SIZE(%eax)
prefetcht2 4 * SIZE(%eax)
#if L2_SIZE > 262144
subl $-8 * SIZE, BX
#elif L2_SIZE > 131072
prefetcht2 8 * SIZE(%eax)
prefetcht2 12 * SIZE(%eax)
subl $-16 * SIZE, BX
#else
prefetcht2 16 * SIZE(%eax)
prefetcht2 20 * SIZE(%eax)
prefetcht2 24 * SIZE(%eax)
prefetcht2 28 * SIZE(%eax)
subl $-32 * SIZE, BX
#endif
#endif
movaps 0 * SIZE(AA), %xmm0
pxor %xmm4, %xmm4
movaps 16 * SIZE(AA), %xmm1
pxor %xmm5, %xmm5
movaps 0 * SIZE(BB), %xmm2
pxor %xmm6, %xmm6
movaps 16 * SIZE(BB), %xmm3
pxor %xmm7, %xmm7
leal (LDC, LDC, 2), %eax
#if defined(OPTERON) || defined(BARCELONA)
prefetchw 4 * SIZE(%esi)
prefetchw 4 * SIZE(%esi, LDC)
prefetchw 4 * SIZE(%esi, LDC, 2)
prefetchw 4 * SIZE(%esi, %eax)
#endif
#ifdef PENTIUM4
prefetchnta 4 * SIZE(%esi)
prefetchnta 4 * SIZE(%esi, LDC)
prefetchnta 4 * SIZE(%esi, LDC, 2)
prefetchnta 4 * SIZE(%esi, %eax)
#endif
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $4, %eax
#else
addl $4, %eax
#endif
movl %eax, KKK
#endif
#if 1
andl $-8, %eax
sall $4, %eax
je .L15
.L1X:
KERNEL1(32 * 0)
KERNEL2(32 * 0)
KERNEL3(32 * 0)
KERNEL4(32 * 0)
KERNEL5(32 * 0)
KERNEL6(32 * 0)
KERNEL7(32 * 0)
KERNEL8(32 * 0)
cmpl $128 * 1, %eax
jle .L12
KERNEL1(32 * 1)
KERNEL2(32 * 1)
KERNEL3(32 * 1)
KERNEL4(32 * 1)
KERNEL5(32 * 1)
KERNEL6(32 * 1)
KERNEL7(32 * 1)
KERNEL8(32 * 1)
cmpl $128 * 2, %eax
jle .L12
KERNEL1(32 * 2)
KERNEL2(32 * 2)
KERNEL3(32 * 2)
KERNEL4(32 * 2)
KERNEL5(32 * 2)
KERNEL6(32 * 2)
KERNEL7(32 * 2)
KERNEL8(32 * 2)
cmpl $128 * 3, %eax
jle .L12
KERNEL1(32 * 3)
KERNEL2(32 * 3)
KERNEL3(32 * 3)
KERNEL4(32 * 3)
KERNEL5(32 * 3)
KERNEL6(32 * 3)
KERNEL7(32 * 3)
KERNEL8(32 * 3)
cmpl $128 * 4, %eax
jle .L12
KERNEL1(32 * 4)
KERNEL2(32 * 4)
KERNEL3(32 * 4)
KERNEL4(32 * 4)
KERNEL5(32 * 4)
KERNEL6(32 * 4)
KERNEL7(32 * 4)
KERNEL8(32 * 4)
cmpl $128 * 5, %eax
jle .L12
KERNEL1(32 * 5)
KERNEL2(32 * 5)
KERNEL3(32 * 5)
KERNEL4(32 * 5)
KERNEL5(32 * 5)
KERNEL6(32 * 5)
KERNEL7(32 * 5)
KERNEL8(32 * 5)
cmpl $128 * 6, %eax
jle .L12
KERNEL1(32 * 6)
KERNEL2(32 * 6)
KERNEL3(32 * 6)
KERNEL4(32 * 6)
KERNEL5(32 * 6)
KERNEL6(32 * 6)
KERNEL7(32 * 6)
KERNEL8(32 * 6)
cmpl $128 * 7, %eax
jle .L12
KERNEL1(32 * 7)
KERNEL2(32 * 7)
KERNEL3(32 * 7)
KERNEL4(32 * 7)
KERNEL5(32 * 7)
KERNEL6(32 * 7)
KERNEL7(32 * 7)
KERNEL8(32 * 7)
addl $128 * 8 * SIZE, BB
addl $128 * 2 * SIZE, AA
subl $128 * 8, %eax
jg .L1X
jmp .L15
.L12:
leal (AA, %eax, 1), AA
leal (BB, %eax, 4), BB
ALIGN_4
#else
sarl $3, %eax
je .L15
ALIGN_4
.L12:
KERNEL1(32 * 0)
KERNEL2(32 * 0)
KERNEL3(32 * 0)
KERNEL4(32 * 0)
KERNEL5(32 * 0)
KERNEL6(32 * 0)
KERNEL7(32 * 0)
KERNEL8(32 * 0)
addl $128 * SIZE, BB
addl $32 * SIZE, AA
decl %eax
jne .L12
ALIGN_4
#endif
.L15:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L18
ALIGN_4
.L16:
mulps %xmm0, %xmm2
addps %xmm2, %xmm4
movaps 4 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
addps %xmm2, %xmm5
movaps 8 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
mulps 12 * SIZE(BB), %xmm0
addps %xmm2, %xmm6
movaps 16 * SIZE(BB), %xmm2
addps %xmm0, %xmm7
movaps 4 * SIZE(AA), %xmm0
addl $ 4 * SIZE, AA
addl $16 * SIZE, BB
decl %eax
jg .L16
ALIGN_4
.L18:
leal (LDC, LDC, 2), %eax
movsd 0 * SIZE(%esi), %xmm0
movhps 2 * SIZE(%esi), %xmm0
movsd 4 * SIZE(%esi), %xmm1
movhps 6 * SIZE(%esi), %xmm1
pshufd $0x50, %xmm4, %xmm2
pshufd $0xfa, %xmm4, %xmm4
mulps %xmm3, %xmm2
mulps %xmm3, %xmm4
addps %xmm2, %xmm0
addps %xmm4, %xmm1
movlps %xmm0, 0 * SIZE(%esi)
movhps %xmm0, 2 * SIZE(%esi)
movlps %xmm1, 4 * SIZE(%esi)
movhps %xmm1, 6 * SIZE(%esi)
movsd 0 * SIZE(%esi, LDC), %xmm0
movhps 2 * SIZE(%esi, LDC), %xmm0
movsd 4 * SIZE(%esi, LDC), %xmm1
movhps 6 * SIZE(%esi, LDC), %xmm1
pshufd $0x50, %xmm5, %xmm2
pshufd $0xfa, %xmm5, %xmm5
mulps %xmm3, %xmm2
mulps %xmm3, %xmm5
addps %xmm2, %xmm0
addps %xmm5, %xmm1
movlps %xmm0, 0 * SIZE(%esi, LDC)
movhps %xmm0, 2 * SIZE(%esi, LDC)
movlps %xmm1, 4 * SIZE(%esi, LDC)
movhps %xmm1, 6 * SIZE(%esi, LDC)
movsd 0 * SIZE(%esi, LDC, 2), %xmm0
movhps 2 * SIZE(%esi, LDC, 2), %xmm0
movsd 4 * SIZE(%esi, LDC, 2), %xmm1
movhps 6 * SIZE(%esi, LDC, 2), %xmm1
pshufd $0x50, %xmm6, %xmm2
pshufd $0xfa, %xmm6, %xmm6
mulps %xmm3, %xmm2
mulps %xmm3, %xmm6
addps %xmm2, %xmm0
addps %xmm6, %xmm1
movlps %xmm0, 0 * SIZE(%esi, LDC, 2)
movhps %xmm0, 2 * SIZE(%esi, LDC, 2)
movlps %xmm1, 4 * SIZE(%esi, LDC, 2)
movhps %xmm1, 6 * SIZE(%esi, LDC, 2)
movsd 0 * SIZE(%esi, %eax), %xmm0
movhps 2 * SIZE(%esi, %eax), %xmm0
movsd 4 * SIZE(%esi, %eax), %xmm1
movhps 6 * SIZE(%esi, %eax), %xmm1
pshufd $0x50, %xmm7, %xmm2
pshufd $0xfa, %xmm7, %xmm7
mulps %xmm3, %xmm2
mulps %xmm3, %xmm7
addps %xmm2, %xmm0
addps %xmm7, %xmm1
movlps %xmm0, 0 * SIZE(%esi, %eax)
movhps %xmm0, 2 * SIZE(%esi, %eax)
movlps %xmm1, 4 * SIZE(%esi, %eax)
movhps %xmm1, 6 * SIZE(%esi, %eax)
addl $8 * SIZE, %esi # coffset += 2
decl %ebx # i --
jg .L11
ALIGN_4
.L20:
testl $2, M
je .L30
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 8), %eax
leal (AA, %eax, 1), AA
leal (BB, %eax, 8), BB
#endif
movsd 0 * SIZE(AA), %xmm0
pxor %xmm4, %xmm4
movsd 8 * SIZE(AA), %xmm1
pxor %xmm5, %xmm5
movsd 0 * SIZE(BB), %xmm2
pxor %xmm6, %xmm6
movsd 16 * SIZE(BB), %xmm3
pxor %xmm7, %xmm7
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $2, %eax
#else
addl $4, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je .L25
ALIGN_4
.L22:
mulps %xmm0, %xmm2
addps %xmm2, %xmm4
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 0) * SIZE(AA)
#endif
movsd 4 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
addps %xmm2, %xmm5
movsd 8 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
addps %xmm2, %xmm6
movsd 12 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 2 * SIZE(AA), %xmm0
addps %xmm2, %xmm7
movsd 32 * SIZE(BB), %xmm2
mulps %xmm0, %xmm3
addps %xmm3, %xmm4
movsd 20 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
addps %xmm3, %xmm5
movsd 24 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
addps %xmm3, %xmm6
movsd 28 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
movsd 4 * SIZE(AA), %xmm0
addps %xmm3, %xmm7
movsd 48 * SIZE(BB), %xmm3
mulps %xmm0, %xmm2
addps %xmm2, %xmm4
movsd 36 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
addps %xmm2, %xmm5
movsd 40 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
addps %xmm2, %xmm6
movsd 44 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 6 * SIZE(AA), %xmm0
addps %xmm2, %xmm7
movsd 64 * SIZE(BB), %xmm2
mulps %xmm0, %xmm3
addps %xmm3, %xmm4
movsd 52 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
addps %xmm3, %xmm5
movsd 56 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
addps %xmm3, %xmm6
movsd 60 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
movsd 16 * SIZE(AA), %xmm0
addps %xmm3, %xmm7
movsd 80 * SIZE(BB), %xmm3
mulps %xmm1, %xmm2
addps %xmm2, %xmm4
movsd 68 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
addps %xmm2, %xmm5
movsd 72 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
addps %xmm2, %xmm6
movsd 76 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
movsd 10 * SIZE(AA), %xmm1
addps %xmm2, %xmm7
movsd 96 * SIZE(BB), %xmm2
mulps %xmm1, %xmm3
addps %xmm3, %xmm4
movsd 84 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
addps %xmm3, %xmm5
movsd 88 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
addps %xmm3, %xmm6
movsd 92 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
movsd 12 * SIZE(AA), %xmm1
addps %xmm3, %xmm7
movsd 112 * SIZE(BB), %xmm3
mulps %xmm1, %xmm2
addps %xmm2, %xmm4
movsd 100 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
addps %xmm2, %xmm5
movsd 104 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
addps %xmm2, %xmm6
movsd 108 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
movsd 14 * SIZE(AA), %xmm1
addps %xmm2, %xmm7
movsd 128 * SIZE(BB), %xmm2
mulps %xmm1, %xmm3
addps %xmm3, %xmm4
movsd 116 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
addps %xmm3, %xmm5
movsd 120 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
addps %xmm3, %xmm6
movsd 124 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
movsd 24 * SIZE(AA), %xmm1
addps %xmm3, %xmm7
movsd 144 * SIZE(BB), %xmm3
addl $ 16 * SIZE, AA
addl $128 * SIZE, BB
decl %eax
jne .L22
ALIGN_4
.L25:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L28
ALIGN_4
.L26:
mulps %xmm0, %xmm2
addps %xmm2, %xmm4
movsd 4 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
addps %xmm2, %xmm5
movsd 8 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
addps %xmm2, %xmm6
movsd 12 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 2 * SIZE(AA), %xmm0
addps %xmm2, %xmm7
movsd 16 * SIZE(BB), %xmm2
addl $ 2 * SIZE, AA
addl $16 * SIZE, BB
decl %eax
jg .L26
ALIGN_4
.L28:
leal (LDC, LDC, 2), %eax
movsd 0 * SIZE(%esi), %xmm0
movhps 2 * SIZE(%esi), %xmm0
movsd 0 * SIZE(%esi, LDC), %xmm1
movhps 2 * SIZE(%esi, LDC), %xmm1
shufps $0x50, %xmm4, %xmm4
shufps $0x50, %xmm5, %xmm5
mulps %xmm3, %xmm4
mulps %xmm3, %xmm5
addps %xmm4, %xmm0
addps %xmm5, %xmm1
movlps %xmm0, 0 * SIZE(%esi)
movhps %xmm0, 2 * SIZE(%esi)
movlps %xmm1, 0 * SIZE(%esi, LDC)
movhps %xmm1, 2 * SIZE(%esi, LDC)
movsd 0 * SIZE(%esi, LDC, 2), %xmm0
movhps 2 * SIZE(%esi, LDC, 2), %xmm0
movsd 0 * SIZE(%esi, %eax), %xmm1
movhps 2 * SIZE(%esi, %eax), %xmm1
shufps $0x50, %xmm6, %xmm6
shufps $0x50, %xmm7, %xmm7
mulps %xmm3, %xmm6
mulps %xmm3, %xmm7
addps %xmm6, %xmm0
addps %xmm7, %xmm1
movlps %xmm0, 0 * SIZE(%esi, LDC, 2)
movhps %xmm0, 2 * SIZE(%esi, LDC, 2)
movlps %xmm1, 0 * SIZE(%esi, %eax)
movhps %xmm1, 2 * SIZE(%esi, %eax)
addl $4 * SIZE, %esi # coffset += 2
ALIGN_4
.L30:
testl $1, M
je .L39
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 4), %eax
leal (AA, %eax, 1), AA
leal (BB, %eax, 8), BB
leal (BB, %eax, 8), BB
#endif
movss 0 * SIZE(AA), %xmm0
pxor %xmm4, %xmm4
movss 4 * SIZE(AA), %xmm1
pxor %xmm5, %xmm5
movss 0 * SIZE(BB), %xmm2
pxor %xmm6, %xmm6
movss 16 * SIZE(BB), %xmm3
pxor %xmm7, %xmm7
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $1, %eax
#else
addl $4, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je .L35
ALIGN_4
.L32:
mulss %xmm0, %xmm2
addss %xmm2, %xmm4
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 0) * SIZE(AA)
#endif
movss 4 * SIZE(BB), %xmm2
mulss %xmm0, %xmm2
addss %xmm2, %xmm5
movss 8 * SIZE(BB), %xmm2
mulss %xmm0, %xmm2
mulss 12 * SIZE(BB), %xmm0
addss %xmm2, %xmm6
movss 32 * SIZE(BB), %xmm2
addss %xmm0, %xmm7
movss 1 * SIZE(AA), %xmm0
mulss %xmm0, %xmm3
addss %xmm3, %xmm4
movss 20 * SIZE(BB), %xmm3
mulss %xmm0, %xmm3
addss %xmm3, %xmm5
movss 24 * SIZE(BB), %xmm3
mulss %xmm0, %xmm3
mulss 28 * SIZE(BB), %xmm0
addss %xmm3, %xmm6
movss 48 * SIZE(BB), %xmm3
addss %xmm0, %xmm7
movss 2 * SIZE(AA), %xmm0
mulss %xmm0, %xmm2
addss %xmm2, %xmm4
movss 36 * SIZE(BB), %xmm2
mulss %xmm0, %xmm2
addss %xmm2, %xmm5
movss 40 * SIZE(BB), %xmm2
mulss %xmm0, %xmm2
mulss 44 * SIZE(BB), %xmm0
addss %xmm2, %xmm6
movss 64 * SIZE(BB), %xmm2
addss %xmm0, %xmm7
movss 3 * SIZE(AA), %xmm0
mulss %xmm0, %xmm3
addss %xmm3, %xmm4
movss 52 * SIZE(BB), %xmm3
mulss %xmm0, %xmm3
addss %xmm3, %xmm5
movss 56 * SIZE(BB), %xmm3
mulss %xmm0, %xmm3
mulss 60 * SIZE(BB), %xmm0
addss %xmm3, %xmm6
movss 80 * SIZE(BB), %xmm3
addss %xmm0, %xmm7
movss 8 * SIZE(AA), %xmm0
mulss %xmm1, %xmm2
addss %xmm2, %xmm4
movss 68 * SIZE(BB), %xmm2
mulss %xmm1, %xmm2
addss %xmm2, %xmm5
movss 72 * SIZE(BB), %xmm2
mulss %xmm1, %xmm2
mulss 76 * SIZE(BB), %xmm1
addss %xmm2, %xmm6
movss 96 * SIZE(BB), %xmm2
addss %xmm1, %xmm7
movss 5 * SIZE(AA), %xmm1
mulss %xmm1, %xmm3
addss %xmm3, %xmm4
movss 84 * SIZE(BB), %xmm3
mulss %xmm1, %xmm3
addss %xmm3, %xmm5
movss 88 * SIZE(BB), %xmm3
mulss %xmm1, %xmm3
mulss 92 * SIZE(BB), %xmm1
addss %xmm3, %xmm6
movss 112 * SIZE(BB), %xmm3
addss %xmm1, %xmm7
movss 6 * SIZE(AA), %xmm1
mulss %xmm1, %xmm2
addss %xmm2, %xmm4
movss 100 * SIZE(BB), %xmm2
mulss %xmm1, %xmm2
addss %xmm2, %xmm5
movss 104 * SIZE(BB), %xmm2
mulss %xmm1, %xmm2
mulss 108 * SIZE(BB), %xmm1
addss %xmm2, %xmm6
movss 128 * SIZE(BB), %xmm2
addss %xmm1, %xmm7
movss 7 * SIZE(AA), %xmm1
mulss %xmm1, %xmm3
addss %xmm3, %xmm4
movss 116 * SIZE(BB), %xmm3
mulss %xmm1, %xmm3
addss %xmm3, %xmm5
movss 120 * SIZE(BB), %xmm3
mulss %xmm1, %xmm3
mulss 124 * SIZE(BB), %xmm1
addss %xmm3, %xmm6
movss 144 * SIZE(BB), %xmm3
addss %xmm1, %xmm7
movss 12 * SIZE(AA), %xmm1
addl $ 8 * SIZE, AA
addl $128 * SIZE, BB
decl %eax
jne .L32
ALIGN_4
.L35:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L38
ALIGN_4
.L36:
mulss %xmm0, %xmm2
addss %xmm2, %xmm4
movss 4 * SIZE(BB), %xmm2
mulss %xmm0, %xmm2
addss %xmm2, %xmm5
movss 8 * SIZE(BB), %xmm2
mulss %xmm0, %xmm2
mulss 12 * SIZE(BB), %xmm0
addss %xmm2, %xmm6
movss 16 * SIZE(BB), %xmm2
addss %xmm0, %xmm7
movss 1 * SIZE(AA), %xmm0
addl $ 1 * SIZE, AA
addl $16 * SIZE, BB
decl %eax
jg .L36
ALIGN_4
.L38:
leal (LDC, LDC, 2), %eax
movsd (%esi), %xmm0
movhps (%esi, LDC), %xmm0
shufps $0, %xmm5, %xmm4
mulps %xmm3, %xmm4
addps %xmm4, %xmm0
movlps %xmm0, (%esi)
movhps %xmm0, (%esi, LDC)
movsd (%esi, LDC, 2), %xmm0
movhps (%esi, %eax), %xmm0
shufps $0, %xmm7, %xmm6
mulps %xmm3, %xmm6
addps %xmm6, %xmm0
movlps %xmm0, (%esi, LDC, 2)
movhps %xmm0, (%esi, %eax)
ALIGN_4
.L39:
#if defined(TRMMKERNEL) && !defined(LEFT)
addl $4, KK
#endif
leal (, LDC, 4), %eax
addl %eax, C # c += 4 * ldc
decl J # j --
jg .L01
ALIGN_4
.L40:
testl $2, N
je .L80
#if defined(TRMMKERNEL) && defined(LEFT)
movl OFFSET, %eax
movl %eax, KK
#endif
movl K, %eax
leal BUFFER, %ecx
sarl $2, %eax
jle .L45
ALIGN_4
.L42:
prefetchnta 80 * SIZE(%edi)
#if defined(OPTERON) || defined(BARCELONA)
prefetchw 112 * SIZE(%ecx)
prefetchw 120 * SIZE(%ecx)
#endif
#ifdef PENTIUM4
prefetcht1 112 * SIZE(%ecx)
#endif
#ifdef HAVE_SSE2
movss 0 * SIZE(%edi), %xmm0
movss 1 * SIZE(%edi), %xmm1
movss 2 * SIZE(%edi), %xmm2
movss 3 * SIZE(%edi), %xmm3
movss 4 * SIZE(%edi), %xmm4
movss 5 * SIZE(%edi), %xmm5
movss 6 * SIZE(%edi), %xmm6
movss 7 * SIZE(%edi), %xmm7
shufps $0, %xmm0, %xmm0
shufps $0, %xmm1, %xmm1
shufps $0, %xmm2, %xmm2
shufps $0, %xmm3, %xmm3
shufps $0, %xmm4, %xmm4
shufps $0, %xmm5, %xmm5
shufps $0, %xmm6, %xmm6
shufps $0, %xmm7, %xmm7
movaps %xmm0, 0 * SIZE(%ecx)
movaps %xmm1, 4 * SIZE(%ecx)
movaps %xmm2, 8 * SIZE(%ecx)
movaps %xmm3, 12 * SIZE(%ecx)
movaps %xmm4, 16 * SIZE(%ecx)
movaps %xmm5, 20 * SIZE(%ecx)
movaps %xmm6, 24 * SIZE(%ecx)
movaps %xmm7, 28 * SIZE(%ecx)
#else
movd 0 * SIZE(%edi), %mm0
movd 1 * SIZE(%edi), %mm1
movd 2 * SIZE(%edi), %mm2
movd 3 * SIZE(%edi), %mm3
movd 4 * SIZE(%edi), %mm4
movd 5 * SIZE(%edi), %mm5
movd 6 * SIZE(%edi), %mm6
movd 7 * SIZE(%edi), %mm7
movd %mm0, 0 * SIZE(%ecx)
movd %mm0, 1 * SIZE(%ecx)
movd %mm0, 2 * SIZE(%ecx)
movd %mm0, 3 * SIZE(%ecx)
movd %mm1, 4 * SIZE(%ecx)
movd %mm1, 5 * SIZE(%ecx)
movd %mm1, 6 * SIZE(%ecx)
movd %mm1, 7 * SIZE(%ecx)
movd %mm2, 8 * SIZE(%ecx)
movd %mm2, 9 * SIZE(%ecx)
movd %mm2, 10 * SIZE(%ecx)
movd %mm2, 11 * SIZE(%ecx)
movd %mm3, 12 * SIZE(%ecx)
movd %mm3, 13 * SIZE(%ecx)
movd %mm3, 14 * SIZE(%ecx)
movd %mm3, 15 * SIZE(%ecx)
movd %mm4, 16 * SIZE(%ecx)
movd %mm4, 17 * SIZE(%ecx)
movd %mm4, 18 * SIZE(%ecx)
movd %mm4, 19 * SIZE(%ecx)
movd %mm5, 20 * SIZE(%ecx)
movd %mm5, 21 * SIZE(%ecx)
movd %mm5, 22 * SIZE(%ecx)
movd %mm5, 23 * SIZE(%ecx)
movd %mm6, 24 * SIZE(%ecx)
movd %mm6, 25 * SIZE(%ecx)
movd %mm6, 26 * SIZE(%ecx)
movd %mm6, 27 * SIZE(%ecx)
movd %mm7, 28 * SIZE(%ecx)
movd %mm7, 29 * SIZE(%ecx)
movd %mm7, 30 * SIZE(%ecx)
movd %mm7, 31 * SIZE(%ecx)
#endif
addl $ 8 * SIZE, %edi
addl $32 * SIZE, %ecx
decl %eax
jne .L42
ALIGN_4
.L45:
movl K, %eax
andl $3, %eax
BRANCH
jle .L50
ALIGN_4
.L46:
#ifdef HAVE_SSE2
movss 0 * SIZE(%edi), %xmm0
movss 1 * SIZE(%edi), %xmm1
shufps $0, %xmm0, %xmm0
shufps $0, %xmm1, %xmm1
movaps %xmm0, 0 * SIZE(%ecx)
movaps %xmm1, 4 * SIZE(%ecx)
#else
movd 0 * SIZE(%edi), %mm0
movd 1 * SIZE(%edi), %mm1
movd %mm0, 0 * SIZE(%ecx)
movd %mm0, 1 * SIZE(%ecx)
movd %mm0, 2 * SIZE(%ecx)
movd %mm0, 3 * SIZE(%ecx)
movd %mm1, 4 * SIZE(%ecx)
movd %mm1, 5 * SIZE(%ecx)
movd %mm1, 6 * SIZE(%ecx)
movd %mm1, 7 * SIZE(%ecx)
#endif
addl $2 * SIZE, %edi
addl $8 * SIZE, %ecx
decl %eax
jne .L46
ALIGN_4
.L50:
movl C, %esi # coffset = c
movl A, %edx # aoffset = a
movl M, %ebx
sarl $2, %ebx # i = (m >> 2)
jle .L60
ALIGN_4
.L51:
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 8), %eax
leal (AA, %eax, 2), AA
leal (BB, %eax, 4), BB
#endif
pxor %xmm4, %xmm4
pxor %xmm5, %xmm5
pxor %xmm6, %xmm6
pxor %xmm7, %xmm7
movaps 0 * SIZE(AA), %xmm0
movaps 16 * SIZE(AA), %xmm1
movaps 0 * SIZE(BB), %xmm2
movaps 16 * SIZE(BB), %xmm3
#ifdef HAVE_3DNOW
prefetchw 4 * SIZE(%esi)
prefetchw 4 * SIZE(%esi, LDC)
#elif defined(HAVE_SSE) || defined(HAVE_SSE2)
prefetcht2 4 * SIZE(%esi)
prefetcht2 4 * SIZE(%esi, LDC)
#endif
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $4, %eax
#else
addl $2, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je .L55
ALIGN_4
.L52:
mulps %xmm0, %xmm2
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 0) * SIZE(AA)
#endif
mulps 4 * SIZE(BB), %xmm0
addps %xmm2, %xmm4
movaps 8 * SIZE(BB), %xmm2
addps %xmm0, %xmm5
movaps 4 * SIZE(AA), %xmm0
mulps %xmm0, %xmm2
mulps 12 * SIZE(BB), %xmm0
addps %xmm2, %xmm4
movaps 32 * SIZE(BB), %xmm2
addps %xmm0, %xmm5
movaps 8 * SIZE(AA), %xmm0
mulps %xmm0, %xmm3
mulps 20 * SIZE(BB), %xmm0
addps %xmm3, %xmm4
movaps 24 * SIZE(BB), %xmm3
addps %xmm0, %xmm5
movaps 12 * SIZE(AA), %xmm0
mulps %xmm0, %xmm3
mulps 28 * SIZE(BB), %xmm0
addps %xmm3, %xmm4
movaps 48 * SIZE(BB), %xmm3
addps %xmm0, %xmm5
movaps 32 * SIZE(AA), %xmm0
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 16) * SIZE(AA)
#endif
mulps %xmm1, %xmm2
mulps 36 * SIZE(BB), %xmm1
addps %xmm2, %xmm4
movaps 40 * SIZE(BB), %xmm2
addps %xmm1, %xmm5
movaps 20 * SIZE(AA), %xmm1
mulps %xmm1, %xmm2
mulps 44 * SIZE(BB), %xmm1
addps %xmm2, %xmm4
movaps 64 * SIZE(BB), %xmm2
addps %xmm1, %xmm5
movaps 24 * SIZE(AA), %xmm1
mulps %xmm1, %xmm3
mulps 52 * SIZE(BB), %xmm1
addps %xmm3, %xmm4
movaps 56 * SIZE(BB), %xmm3
addps %xmm1, %xmm5
movaps 28 * SIZE(AA), %xmm1
mulps %xmm1, %xmm3
mulps 60 * SIZE(BB), %xmm1
addps %xmm3, %xmm4
movaps 80 * SIZE(BB), %xmm3
addps %xmm1, %xmm5
movaps 48 * SIZE(AA), %xmm1
addl $32 * SIZE, AA
addl $64 * SIZE, BB
decl %eax
jne .L52
ALIGN_4
.L55:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L58
ALIGN_4
.L56:
mulps %xmm0, %xmm2
mulps 4 * SIZE(BB), %xmm0
addps %xmm2, %xmm4
movaps 8 * SIZE(BB), %xmm2
addps %xmm0, %xmm5
movaps 4 * SIZE(AA), %xmm0
addl $4 * SIZE, AA
addl $8 * SIZE, BB
decl %eax
jg .L56
ALIGN_4
.L58:
movsd 0 * SIZE(%esi), %xmm0
movhps 2 * SIZE(%esi), %xmm0
movsd 4 * SIZE(%esi), %xmm1
movhps 6 * SIZE(%esi), %xmm1
pshufd $0x50, %xmm4, %xmm2
pshufd $0xfa, %xmm4, %xmm4
mulps %xmm3, %xmm2
mulps %xmm3, %xmm4
addps %xmm2, %xmm0
addps %xmm4, %xmm1
movlps %xmm0, 0 * SIZE(%esi)
movhps %xmm0, 2 * SIZE(%esi)
movlps %xmm1, 4 * SIZE(%esi)
movhps %xmm1, 6 * SIZE(%esi)
movsd 0 * SIZE(%esi, LDC), %xmm0
movhps 2 * SIZE(%esi, LDC), %xmm0
movsd 4 * SIZE(%esi, LDC), %xmm1
movhps 6 * SIZE(%esi, LDC), %xmm1
pshufd $0x50, %xmm5, %xmm2
pshufd $0xfa, %xmm5, %xmm5
mulps %xmm3, %xmm2
mulps %xmm3, %xmm5
addps %xmm2, %xmm0
addps %xmm5, %xmm1
movlps %xmm0, 0 * SIZE(%esi, LDC)
movhps %xmm0, 2 * SIZE(%esi, LDC)
movlps %xmm1, 4 * SIZE(%esi, LDC)
movhps %xmm1, 6 * SIZE(%esi, LDC)
addl $8 * SIZE, %esi # coffset += 2
decl %ebx # i --
jg .L51
ALIGN_4
.L60:
testl $2, M
je .L70
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 8), %eax
leal (AA, %eax, 1), AA
leal (BB, %eax, 4), BB
#endif
pxor %xmm4, %xmm4
pxor %xmm5, %xmm5
pxor %xmm6, %xmm6
pxor %xmm7, %xmm7
movsd 0 * SIZE(AA), %xmm0
movsd 8 * SIZE(AA), %xmm1
movsd 0 * SIZE(BB), %xmm2
movsd 16 * SIZE(BB), %xmm3
leal (LDC, LDC, 2), %eax
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $2, %eax
#else
addl $2, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je .L65
ALIGN_4
.L62:
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 0) * SIZE(AA)
#endif
mulps %xmm0, %xmm2
addps %xmm2, %xmm4
movsd 4 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 2 * SIZE(AA), %xmm0
addps %xmm2, %xmm5
movsd 8 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
addps %xmm2, %xmm6
movsd 12 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 4 * SIZE(AA), %xmm0
addps %xmm2, %xmm7
movsd 32 * SIZE(BB), %xmm2
mulps %xmm0, %xmm3
addps %xmm3, %xmm4
movsd 20 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
movsd 6 * SIZE(AA), %xmm0
addps %xmm3, %xmm5
movsd 24 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
addps %xmm3, %xmm6
movsd 28 * SIZE(BB), %xmm3
mulps %xmm0, %xmm3
movsd 16 * SIZE(AA), %xmm0
addps %xmm3, %xmm7
movsd 48 * SIZE(BB), %xmm3
mulps %xmm1, %xmm2
addps %xmm2, %xmm4
movsd 36 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
movsd 10 * SIZE(AA), %xmm1
addps %xmm2, %xmm5
movsd 40 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
addps %xmm2, %xmm6
movsd 44 * SIZE(BB), %xmm2
mulps %xmm1, %xmm2
movsd 12 * SIZE(AA), %xmm1
addps %xmm2, %xmm7
movsd 64 * SIZE(BB), %xmm2
mulps %xmm1, %xmm3
addps %xmm3, %xmm4
movsd 52 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
movsd 14 * SIZE(AA), %xmm1
addps %xmm3, %xmm5
movsd 56 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
addps %xmm3, %xmm6
movsd 60 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
movsd 24 * SIZE(AA), %xmm1
addps %xmm3, %xmm7
movsd 80 * SIZE(BB), %xmm3
addl $16 * SIZE, AA
addl $64 * SIZE, BB
decl %eax
jne .L62
ALIGN_4
.L65:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L68
ALIGN_4
.L66:
mulps %xmm0, %xmm2
addps %xmm2, %xmm4
movsd 4 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 2 * SIZE(AA), %xmm0
addps %xmm2, %xmm5
movsd 8 * SIZE(BB), %xmm2
addl $2 * SIZE, AA
addl $8 * SIZE, BB
decl %eax
jg .L66
ALIGN_4
.L68:
addps %xmm6, %xmm4
addps %xmm7, %xmm5
movsd 0 * SIZE(%esi), %xmm0
movhps 2 * SIZE(%esi), %xmm0
movsd 0 * SIZE(%esi, LDC), %xmm1
movhps 2 * SIZE(%esi, LDC), %xmm1
shufps $0x50, %xmm4, %xmm4
shufps $0x50, %xmm5, %xmm5
mulps %xmm3, %xmm4
mulps %xmm3, %xmm5
addps %xmm4, %xmm0
addps %xmm5, %xmm1
movlps %xmm0, 0 * SIZE(%esi)
movhps %xmm0, 2 * SIZE(%esi)
movlps %xmm1, 0 * SIZE(%esi, LDC)
movhps %xmm1, 2 * SIZE(%esi, LDC)
addl $4 * SIZE, %esi
ALIGN_4
.L70:
testl $1, M
je .L79
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 4), %eax
leal (AA, %eax, 1), AA
leal (BB, %eax, 8), BB
#endif
pxor %xmm4, %xmm4
pxor %xmm5, %xmm5
pxor %xmm6, %xmm6
pxor %xmm7, %xmm7
movss 0 * SIZE(AA), %xmm0
movss 4 * SIZE(AA), %xmm1
movss 0 * SIZE(BB), %xmm2
movss 16 * SIZE(BB), %xmm3
leal (LDC, LDC, 2), %eax
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $1, %eax
#else
addl $2, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je .L75
ALIGN_4
.L72:
mulss %xmm0, %xmm2
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 0) * SIZE(AA)
#endif
mulss 4 * SIZE(BB), %xmm0
addss %xmm2, %xmm4
movss 8 * SIZE(BB), %xmm2
addss %xmm0, %xmm5
movss 1 * SIZE(AA), %xmm0
mulss %xmm0, %xmm2
mulss 12 * SIZE(BB), %xmm0
addss %xmm2, %xmm6
movss 32 * SIZE(BB), %xmm2
addss %xmm0, %xmm7
movss 2 * SIZE(AA), %xmm0
mulss %xmm0, %xmm3
mulss 20 * SIZE(BB), %xmm0
addss %xmm3, %xmm4
movss 24 * SIZE(BB), %xmm3
addss %xmm0, %xmm5
movss 3 * SIZE(AA), %xmm0
mulss %xmm0, %xmm3
mulss 28 * SIZE(BB), %xmm0
addss %xmm3, %xmm6
movss 48 * SIZE(BB), %xmm3
addss %xmm0, %xmm7
movss 8 * SIZE(AA), %xmm0
mulss %xmm1, %xmm2
mulss 36 * SIZE(BB), %xmm1
addss %xmm2, %xmm4
movss 40 * SIZE(BB), %xmm2
addss %xmm1, %xmm5
movss 5 * SIZE(AA), %xmm1
mulss %xmm1, %xmm2
mulss 44 * SIZE(BB), %xmm1
addss %xmm2, %xmm6
movss 64 * SIZE(BB), %xmm2
addss %xmm1, %xmm7
movss 6 * SIZE(AA), %xmm1
mulss %xmm1, %xmm3
mulss 52 * SIZE(BB), %xmm1
addss %xmm3, %xmm4
movss 56 * SIZE(BB), %xmm3
addss %xmm1, %xmm5
movss 7 * SIZE(AA), %xmm1
mulss %xmm1, %xmm3
mulss 60 * SIZE(BB), %xmm1
addss %xmm3, %xmm6
movss 80 * SIZE(BB), %xmm3
addss %xmm1, %xmm7
movss 12 * SIZE(AA), %xmm1
addl $ 8 * SIZE, AA
addl $64 * SIZE, BB
decl %eax
jne .L72
ALIGN_4
.L75:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L78
ALIGN_4
.L76:
mulss %xmm0, %xmm2
mulss 4 * SIZE(BB), %xmm0
addss %xmm2, %xmm4
movss 8 * SIZE(BB), %xmm2
addss %xmm0, %xmm5
movss 1 * SIZE(AA), %xmm0
addl $ 1 * SIZE, AA
addl $ 8 * SIZE, BB
decl %eax
jg .L76
ALIGN_4
.L78:
addss %xmm6, %xmm4
addss %xmm7, %xmm5
movsd (%esi), %xmm0
movhps (%esi, LDC), %xmm0
shufps $0, %xmm5, %xmm4
mulps %xmm3, %xmm4
addps %xmm4, %xmm0
movlps %xmm0, (%esi)
movhps %xmm0, (%esi, LDC)
ALIGN_4
.L79:
#if defined(TRMMKERNEL) && !defined(LEFT)
addl $2, KK
#endif
leal (, LDC, 2), %eax
addl %eax, C
ALIGN_4
.L80:
testl $1, N
je .L999
#if defined(TRMMKERNEL) && defined(LEFT)
movl OFFSET, %eax
movl %eax, KK
#endif
movl K, %eax
leal BUFFER, %ecx
sarl $3, %eax
jle .L85
ALIGN_4
.L82:
prefetchnta 80 * SIZE(%edi)
#if defined(OPTERON) || defined(BARCELONA)
prefetchw 112 * SIZE(%ecx)
prefetchw 120 * SIZE(%ecx)
#endif
#ifdef PENTIUM4
prefetcht1 112 * SIZE(%ecx)
#endif
#ifdef HAVE_SSE2
movss 0 * SIZE(%edi), %xmm0
movss 1 * SIZE(%edi), %xmm1
movss 2 * SIZE(%edi), %xmm2
movss 3 * SIZE(%edi), %xmm3
movss 4 * SIZE(%edi), %xmm4
movss 5 * SIZE(%edi), %xmm5
movss 6 * SIZE(%edi), %xmm6
movss 7 * SIZE(%edi), %xmm7
shufps $0, %xmm0, %xmm0
shufps $0, %xmm1, %xmm1
shufps $0, %xmm2, %xmm2
shufps $0, %xmm3, %xmm3
shufps $0, %xmm4, %xmm4
shufps $0, %xmm5, %xmm5
shufps $0, %xmm6, %xmm6
shufps $0, %xmm7, %xmm7
movaps %xmm0, 0 * SIZE(%ecx)
movaps %xmm1, 4 * SIZE(%ecx)
movaps %xmm2, 8 * SIZE(%ecx)
movaps %xmm3, 12 * SIZE(%ecx)
movaps %xmm4, 16 * SIZE(%ecx)
movaps %xmm5, 20 * SIZE(%ecx)
movaps %xmm6, 24 * SIZE(%ecx)
movaps %xmm7, 28 * SIZE(%ecx)
#else
movd 0 * SIZE(%edi), %mm0
movd 1 * SIZE(%edi), %mm1
movd 2 * SIZE(%edi), %mm2
movd 3 * SIZE(%edi), %mm3
movd 4 * SIZE(%edi), %mm4
movd 5 * SIZE(%edi), %mm5
movd 6 * SIZE(%edi), %mm6
movd 7 * SIZE(%edi), %mm7
movd %mm0, 0 * SIZE(%ecx)
movd %mm0, 1 * SIZE(%ecx)
movd %mm0, 2 * SIZE(%ecx)
movd %mm0, 3 * SIZE(%ecx)
movd %mm1, 4 * SIZE(%ecx)
movd %mm1, 5 * SIZE(%ecx)
movd %mm1, 6 * SIZE(%ecx)
movd %mm1, 7 * SIZE(%ecx)
movd %mm2, 8 * SIZE(%ecx)
movd %mm2, 9 * SIZE(%ecx)
movd %mm2, 10 * SIZE(%ecx)
movd %mm2, 11 * SIZE(%ecx)
movd %mm3, 12 * SIZE(%ecx)
movd %mm3, 13 * SIZE(%ecx)
movd %mm3, 14 * SIZE(%ecx)
movd %mm3, 15 * SIZE(%ecx)
movd %mm4, 16 * SIZE(%ecx)
movd %mm4, 17 * SIZE(%ecx)
movd %mm4, 18 * SIZE(%ecx)
movd %mm4, 19 * SIZE(%ecx)
movd %mm5, 20 * SIZE(%ecx)
movd %mm5, 21 * SIZE(%ecx)
movd %mm5, 22 * SIZE(%ecx)
movd %mm5, 23 * SIZE(%ecx)
movd %mm6, 24 * SIZE(%ecx)
movd %mm6, 25 * SIZE(%ecx)
movd %mm6, 26 * SIZE(%ecx)
movd %mm6, 27 * SIZE(%ecx)
movd %mm7, 28 * SIZE(%ecx)
movd %mm7, 29 * SIZE(%ecx)
movd %mm7, 30 * SIZE(%ecx)
movd %mm7, 31 * SIZE(%ecx)
#endif
addl $ 8 * SIZE, %edi
addl $32 * SIZE, %ecx
decl %eax
jne .L82
ALIGN_4
.L85:
movl K, %eax
andl $7, %eax
BRANCH
jle .L90
ALIGN_4
.L86:
#ifdef HAVE_SSE2
movss 0 * SIZE(%edi), %xmm0
shufps $0, %xmm0, %xmm0
movaps %xmm0, 0 * SIZE(%ecx)
#else
movd 0 * SIZE(%edi), %mm0
movd %mm0, 0 * SIZE(%ecx)
movd %mm0, 1 * SIZE(%ecx)
movd %mm0, 2 * SIZE(%ecx)
movd %mm0, 3 * SIZE(%ecx)
#endif
addl $1 * SIZE, %edi
addl $4 * SIZE, %ecx
decl %eax
jne .L86
ALIGN_4
.L90:
movl C, %esi # coffset = c
movl A, %edx # aoffset = a
movl M, %ebx
sarl $2, %ebx # i = (m >> 2)
jle .L100
ALIGN_4
.L91:
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 8), %eax
leal (AA, %eax, 2), AA
leal (BB, %eax, 2), BB
#endif
pxor %xmm4, %xmm4
pxor %xmm5, %xmm5
pxor %xmm6, %xmm6
pxor %xmm7, %xmm7
movaps 0 * SIZE(AA), %xmm0
movaps 16 * SIZE(AA), %xmm1
movaps 0 * SIZE(BB), %xmm2
movaps 16 * SIZE(BB), %xmm3
#ifdef HAVE_3DNOW
prefetchw 4 * SIZE(%esi)
#elif defined(HAVE_SSE) || defined(HAVE_SSE2)
prefetcht2 4 * SIZE(%esi)
#endif
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $4, %eax
#else
addl $1, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je .L95
ALIGN_4
.L92:
mulps %xmm0, %xmm2
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 0) * SIZE(AA)
#endif
movaps 4 * SIZE(AA), %xmm0
addps %xmm2, %xmm4
movaps 32 * SIZE(BB), %xmm2
mulps 4 * SIZE(BB), %xmm0
addps %xmm0, %xmm5
movaps 8 * SIZE(AA), %xmm0
mulps 8 * SIZE(BB), %xmm0
addps %xmm0, %xmm6
movaps 12 * SIZE(AA), %xmm0
mulps 12 * SIZE(BB), %xmm0
addps %xmm0, %xmm7
movaps 32 * SIZE(AA), %xmm0
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 16) * SIZE(AA)
#endif
mulps %xmm1, %xmm3
movaps 20 * SIZE(AA), %xmm1
addps %xmm3, %xmm4
movaps 48 * SIZE(BB), %xmm3
mulps 20 * SIZE(BB), %xmm1
addps %xmm1, %xmm5
movaps 24 * SIZE(AA), %xmm1
mulps 24 * SIZE(BB), %xmm1
addps %xmm1, %xmm6
movaps 28 * SIZE(AA), %xmm1
mulps 28 * SIZE(BB), %xmm1
addps %xmm1, %xmm7
movaps 48 * SIZE(AA), %xmm1
addl $32 * SIZE, AA
addl $32 * SIZE, BB
decl %eax
jne .L92
ALIGN_4
.L95:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L98
ALIGN_4
.L96:
mulps %xmm0, %xmm2
addps %xmm2, %xmm4
movaps 4 * SIZE(AA), %xmm0
movaps 4 * SIZE(BB), %xmm2
addl $4 * SIZE, AA
addl $4 * SIZE, BB
decl %eax
jg .L96
ALIGN_4
.L98:
addps %xmm5, %xmm4
addps %xmm7, %xmm6
addps %xmm6, %xmm4
movsd 0 * SIZE(%esi), %xmm0
movhps 2 * SIZE(%esi), %xmm0
movsd 4 * SIZE(%esi), %xmm1
movhps 6 * SIZE(%esi), %xmm1
pshufd $0x50, %xmm4, %xmm2
pshufd $0xfa, %xmm4, %xmm4
mulps %xmm3, %xmm2
mulps %xmm3, %xmm4
addps %xmm2, %xmm0
addps %xmm4, %xmm1
movlps %xmm0, 0 * SIZE(%esi)
movhps %xmm0, 2 * SIZE(%esi)
movlps %xmm1, 4 * SIZE(%esi)
movhps %xmm1, 6 * SIZE(%esi)
addl $8 * SIZE, %esi # coffset += 2
decl %ebx # i --
jg .L91
ALIGN_4
.L100:
testl $2, M
je .L110
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 8), %eax
leal (AA, %eax, 1), AA
leal (BB, %eax, 2), BB
#endif
pxor %xmm4, %xmm4
pxor %xmm5, %xmm5
pxor %xmm6, %xmm6
pxor %xmm7, %xmm7
movsd 0 * SIZE(AA), %xmm0
movsd 8 * SIZE(AA), %xmm1
movsd 0 * SIZE(BB), %xmm2
movsd 16 * SIZE(BB), %xmm3
leal (LDC, LDC, 2), %eax
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $2, %eax
#else
addl $1, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je .L105
ALIGN_4
.L102:
mulps %xmm0, %xmm2
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 0) * SIZE(AA)
#endif
movsd 2 * SIZE(AA), %xmm0
addps %xmm2, %xmm4
movsd 4 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 4 * SIZE(AA), %xmm0
addps %xmm2, %xmm5
movsd 8 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 6 * SIZE(AA), %xmm0
addps %xmm2, %xmm6
movsd 12 * SIZE(BB), %xmm2
mulps %xmm0, %xmm2
movsd 16 * SIZE(AA), %xmm0
addps %xmm2, %xmm7
movsd 32 * SIZE(BB), %xmm2
mulps %xmm1, %xmm3
movsd 10 * SIZE(AA), %xmm1
addps %xmm3, %xmm4
movsd 20 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
movsd 12 * SIZE(AA), %xmm1
addps %xmm3, %xmm5
movsd 24 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
movsd 14 * SIZE(AA), %xmm1
addps %xmm3, %xmm6
movsd 28 * SIZE(BB), %xmm3
mulps %xmm1, %xmm3
movsd 24 * SIZE(AA), %xmm1
addps %xmm3, %xmm7
movsd 48 * SIZE(BB), %xmm3
addl $16 * SIZE, AA
addl $32 * SIZE, BB
decl %eax
jne .L102
ALIGN_4
.L105:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L108
ALIGN_4
.L106:
mulps %xmm0, %xmm2
addps %xmm2, %xmm4
movsd 2 * SIZE(AA), %xmm0
movsd 4 * SIZE(BB), %xmm2
addl $2 * SIZE, AA
addl $4 * SIZE, BB
decl %eax
jg .L106
ALIGN_4
.L108:
addps %xmm5, %xmm4
addps %xmm7, %xmm6
addps %xmm6, %xmm4
movsd 0 * SIZE(%esi), %xmm0
movhps 2 * SIZE(%esi), %xmm0
shufps $0x50, %xmm4, %xmm4
mulps %xmm3, %xmm4
addps %xmm4, %xmm0
movlps %xmm0, 0 * SIZE(%esi)
movhps %xmm0, 2 * SIZE(%esi)
addl $4 * SIZE, %esi
ALIGN_4
.L110:
testl $1, M
je .L999
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leal BUFFER, BB # boffset1 = boffset
#else
leal BUFFER, BB # boffset1 = boffset
movl KK, %eax
leal (, %eax, 4), %eax
leal (AA, %eax, 1), AA
leal (BB, %eax, 4), BB
#endif
pxor %xmm4, %xmm4
pxor %xmm5, %xmm5
pxor %xmm6, %xmm6
pxor %xmm7, %xmm7
movss 0 * SIZE(AA), %xmm0
movss 4 * SIZE(AA), %xmm1
movss 0 * SIZE(BB), %xmm2
movss 16 * SIZE(BB), %xmm3
leal (LDC, LDC, 2), %eax
#ifndef TRMMKERNEL
movl K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
movl K, %eax
subl KK, %eax
movl %eax, KKK
#else
movl KK, %eax
#ifdef LEFT
addl $1, %eax
#else
addl $1, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je .L115
ALIGN_4
.L112:
mulss %xmm0, %xmm2
#if defined(OPTERON) || defined(BARCELONA)
prefetcht0 (PREFETCHSIZE + 0) * SIZE(AA)
#endif
movss 1 * SIZE(AA), %xmm0
addss %xmm2, %xmm4
movss 32 * SIZE(BB), %xmm2
mulss 4 * SIZE(BB), %xmm0
addss %xmm0, %xmm5
movss 2 * SIZE(AA), %xmm0
mulss 8 * SIZE(BB), %xmm0
addss %xmm0, %xmm6
movss 3 * SIZE(AA), %xmm0
mulss 12 * SIZE(BB), %xmm0
addss %xmm0, %xmm7
movss 8 * SIZE(AA), %xmm0
mulss %xmm1, %xmm3
movss 5 * SIZE(AA), %xmm1
addss %xmm3, %xmm4
movss 48 * SIZE(BB), %xmm3
mulss 20 * SIZE(BB), %xmm1
addss %xmm1, %xmm5
movss 6 * SIZE(AA), %xmm1
mulss 24 * SIZE(BB), %xmm1
addss %xmm1, %xmm6
movss 7 * SIZE(AA), %xmm1
mulss 28 * SIZE(BB), %xmm1
addss %xmm1, %xmm7
movss 12 * SIZE(AA), %xmm1
addl $ 8 * SIZE, AA
addl $32 * SIZE, BB
decl %eax
jne .L112
ALIGN_4
.L115:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movaps ALPHA, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L118
ALIGN_4
.L116:
mulss %xmm0, %xmm2
movss 1 * SIZE(AA), %xmm0
addss %xmm2, %xmm4
movss 4 * SIZE(BB), %xmm2
addl $ 1 * SIZE, AA
addl $ 4 * SIZE, BB
decl %eax
jg .L116
ALIGN_4
.L118:
addss %xmm5, %xmm4
addss %xmm7, %xmm6
addss %xmm6, %xmm4
movsd (%esi), %xmm0
shufps $0, %xmm5, %xmm4
mulps %xmm3, %xmm4
addps %xmm4, %xmm0
movlps %xmm0, (%esi)
ALIGN_4
.L999:
EMMS
movl OLD_STACK, %esp
popl %ebx
popl %esi
popl %edi
popl %ebp
ret
EPILOGUE