/*********************************************************************/
/* 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 ARGS 16
#define M 4 + STACK + ARGS(%esp)
#define N 8 + STACK + ARGS(%esp)
#define K 12 + STACK + ARGS(%esp)
#define ALPHA_R 16 + STACK + ARGS(%esp)
#define ALPHA_I 24 + STACK + ARGS(%esp)
#define A 32 + STACK + ARGS(%esp)
#define ARG_B 36 + STACK + ARGS(%esp)
#define C 40 + STACK + ARGS(%esp)
#define ARG_LDC 44 + STACK + ARGS(%esp)
#define OFFSET 48 + STACK + ARGS(%esp)
#define J 0 + STACK(%esp)
#define BX 4 + STACK(%esp)
#define KK 8 + STACK(%esp)
#define KKK 12 + STACK(%esp)
#ifdef PENTIUM4
#define PREFETCH_R (8 * 4)
#define PREFETCH prefetcht1
#define PREFETCHSIZE 84
#endif
#ifdef PENTIUMM
#define PREFETCH_R (8 * 4)
#define PREFETCH prefetcht1
#define PREFETCHSIZE 84
#endif
#define AA %edx
#define BB %ecx
#define LDC %ebp
#define B %edi
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC) || \
defined(RN) || defined(RT) || defined(CN) || defined(CT)
#define ADDSUB addpd
#else
#define ADDSUB subpd
#endif
#define KERNEL1(address) \
mulpd %xmm0, %xmm2; \
PREFETCH (PREFETCHSIZE + 0) * SIZE + (address) * 1 * SIZE(AA); \
addpd %xmm2, %xmm4; \
movddup 1 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm0, %xmm2; \
ADDSUB %xmm2, %xmm5; \
movddup 2 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm0, %xmm2; \
addpd %xmm2, %xmm6; \
movddup 3 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm0, %xmm2; \
movapd 2 * SIZE + (address) * 1 * SIZE(AA), %xmm0; \
ADDSUB %xmm2, %xmm7; \
movddup 4 * SIZE + (address) * 2 * SIZE(BB), %xmm2
#define KERNEL2(address) \
mulpd %xmm0, %xmm2; \
addpd %xmm2, %xmm4; \
movddup 5 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm0, %xmm2; \
ADDSUB %xmm2, %xmm5; \
movddup 6 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm0, %xmm2; \
addpd %xmm2, %xmm6; \
movddup 7 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm0, %xmm2; \
movapd 4 * SIZE + (address) * 1 * SIZE(AA), %xmm0; \
ADDSUB %xmm2, %xmm7; \
movddup 16 * SIZE + (address) * 2 * SIZE(BB), %xmm2
#define KERNEL3(address) \
mulpd %xmm0, %xmm3; \
addpd %xmm3, %xmm4; \
movddup 9 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm0, %xmm3; \
ADDSUB %xmm3, %xmm5; \
movddup 10 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm0, %xmm3; \
addpd %xmm3, %xmm6; \
movddup 11 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm0, %xmm3; \
movapd 6 * SIZE + (address) * 1 * SIZE(AA), %xmm0; \
ADDSUB %xmm3, %xmm7; \
movddup 12 * SIZE + (address) * 2 * SIZE(BB), %xmm3
#define KERNEL4(address) \
mulpd %xmm0, %xmm3; \
addpd %xmm3, %xmm4; \
movddup 13 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm0, %xmm3; \
ADDSUB %xmm3, %xmm5; \
movddup 14 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm0, %xmm3; \
addpd %xmm3, %xmm6; \
movddup 15 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm0, %xmm3; \
movapd 16 * SIZE + (address) * 1 * SIZE(AA), %xmm0; \
ADDSUB %xmm3, %xmm7; \
movddup 24 * SIZE + (address) * 2 * SIZE(BB), %xmm3
#define KERNEL5(address) \
mulpd %xmm1, %xmm2; \
addpd %xmm2, %xmm4; \
movddup 17 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm1, %xmm2; \
ADDSUB %xmm2, %xmm5; \
movddup 18 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm1, %xmm2; \
addpd %xmm2, %xmm6; \
movddup 19 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm1, %xmm2; \
movapd 10 * SIZE + (address) * 1 * SIZE(AA), %xmm1; \
ADDSUB %xmm2, %xmm7; \
movddup 20 * SIZE + (address) * 2 * SIZE(BB), %xmm2
#define KERNEL6(address) \
mulpd %xmm1, %xmm2; \
addpd %xmm2, %xmm4; \
movddup 21 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm1, %xmm2; \
ADDSUB %xmm2, %xmm5; \
movddup 22 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm1, %xmm2; \
addpd %xmm2, %xmm6; \
movddup 23 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm1, %xmm2; \
movapd 12 * SIZE + (address) * 1 * SIZE(AA), %xmm1; \
ADDSUB %xmm2, %xmm7
#define KERNEL7(address) \
movddup 32 * SIZE + (address) * 2 * SIZE(BB), %xmm2; \
mulpd %xmm1, %xmm3; \
addpd %xmm3, %xmm4; \
movddup 25 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm1, %xmm3; \
ADDSUB %xmm3, %xmm5; \
movddup 26 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm1, %xmm3; \
addpd %xmm3, %xmm6; \
movddup 27 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm1, %xmm3; \
movapd 14 * SIZE + (address) * 1 * SIZE(AA), %xmm1; \
ADDSUB %xmm3, %xmm7; \
movddup 28 * SIZE + (address) * 2 * SIZE(BB), %xmm3
#define KERNEL8(address) \
mulpd %xmm1, %xmm3; \
addpd %xmm3, %xmm4; \
movddup 29 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm1, %xmm3; \
ADDSUB %xmm3, %xmm5; \
movddup 30 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm1, %xmm3; \
addpd %xmm3, %xmm6; \
movddup 31 * SIZE + (address) * 2 * SIZE(BB), %xmm3; \
mulpd %xmm1, %xmm3; \
movapd 24 * SIZE + (address) * 1 * SIZE(AA), %xmm1; \
ADDSUB %xmm3, %xmm7; \
movddup 40 * SIZE + (address) * 2 * SIZE(BB), %xmm3
PROLOGUE
subl $ARGS, %esp
pushl %ebp
pushl %edi
pushl %esi
pushl %ebx
PROFCODE
movl ARG_B, B
movl ARG_LDC, LDC
#ifdef TRMMKERNEL
movl OFFSET, %eax
#ifndef LEFT
negl %eax
#endif
movl %eax, KK
#endif
sall $ZBASE_SHIFT, LDC
movl N, %eax
sarl $1, %eax
movl %eax, J # j = n
jle .L100
ALIGN_4
.L01:
#if defined(TRMMKERNEL) && defined(LEFT)
movl OFFSET, %eax
movl %eax, KK
#endif
movl B, BX
movl C, %esi # coffset = c
movl A, AA # aoffset = a
movl M, %ebx
testl %ebx, %ebx
jle .L100
ALIGN_4
.L10:
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
movl B, BB
#else
movl KK, %eax
leal (, %eax, SIZE), %eax
leal (AA, %eax, 2), AA
leal (B, %eax, 4), BB
#endif
movl BX, %eax
prefetcht2 (PREFETCH_R + 0) * SIZE(%eax)
prefetcht2 (PREFETCH_R + 16) * SIZE(%eax)
subl $-8 * SIZE, BX
movapd 0 * SIZE(AA), %xmm0
pxor %xmm4, %xmm4
movapd 8 * SIZE(AA), %xmm1
pxor %xmm5, %xmm5
movddup 0 * SIZE(BB), %xmm2
pxor %xmm6, %xmm6
movddup 8 * SIZE(BB), %xmm3
pxor %xmm7, %xmm7
#ifdef PENTIUM4
prefetchnta 3 * SIZE(%esi)
prefetchnta 3 * 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 $1, %eax
#else
addl $2, %eax
#endif
movl %eax, KKK
#endif
#ifdef CORE_PRESCOTT
andl $-8, %eax
sall $4, %eax
je .L12
.L1X:
KERNEL1(16 * 0)
KERNEL2(16 * 0)
KERNEL3(16 * 0)
KERNEL4(16 * 0)
KERNEL5(16 * 0)
KERNEL6(16 * 0)
KERNEL7(16 * 0)
KERNEL8(16 * 0)
cmpl $128 * 1, %eax
jle .L11
KERNEL1(16 * 1)
KERNEL2(16 * 1)
KERNEL3(16 * 1)
KERNEL4(16 * 1)
KERNEL5(16 * 1)
KERNEL6(16 * 1)
KERNEL7(16 * 1)
KERNEL8(16 * 1)
cmpl $128 * 2, %eax
jle .L11
KERNEL1(16 * 2)
KERNEL2(16 * 2)
KERNEL3(16 * 2)
KERNEL4(16 * 2)
KERNEL5(16 * 2)
KERNEL6(16 * 2)
KERNEL7(16 * 2)
KERNEL8(16 * 2)
cmpl $128 * 3, %eax
jle .L11
KERNEL1(16 * 3)
KERNEL2(16 * 3)
KERNEL3(16 * 3)
KERNEL4(16 * 3)
KERNEL5(16 * 3)
KERNEL6(16 * 3)
KERNEL7(16 * 3)
KERNEL8(16 * 3)
cmpl $128 * 4, %eax
jle .L11
KERNEL1(16 * 4)
KERNEL2(16 * 4)
KERNEL3(16 * 4)
KERNEL4(16 * 4)
KERNEL5(16 * 4)
KERNEL6(16 * 4)
KERNEL7(16 * 4)
KERNEL8(16 * 4)
cmpl $128 * 5, %eax
jle .L11
KERNEL1(16 * 5)
KERNEL2(16 * 5)
KERNEL3(16 * 5)
KERNEL4(16 * 5)
KERNEL5(16 * 5)
KERNEL6(16 * 5)
KERNEL7(16 * 5)
KERNEL8(16 * 5)
cmpl $128 * 6, %eax
jle .L11
KERNEL1(16 * 6)
KERNEL2(16 * 6)
KERNEL3(16 * 6)
KERNEL4(16 * 6)
KERNEL5(16 * 6)
KERNEL6(16 * 6)
KERNEL7(16 * 6)
KERNEL8(16 * 6)
cmpl $128 * 7, %eax
jle .L11
KERNEL1(16 * 7)
KERNEL2(16 * 7)
KERNEL3(16 * 7)
KERNEL4(16 * 7)
KERNEL5(16 * 7)
KERNEL6(16 * 7)
KERNEL7(16 * 7)
KERNEL8(16 * 7)
#if 1
cmpl $128 * 8, %eax
jle .L11
KERNEL1(16 * 8)
KERNEL2(16 * 8)
KERNEL3(16 * 8)
KERNEL4(16 * 8)
KERNEL5(16 * 8)
KERNEL6(16 * 8)
KERNEL7(16 * 8)
KERNEL8(16 * 8)
cmpl $128 * 9, %eax
jle .L11
KERNEL1(16 * 9)
KERNEL2(16 * 9)
KERNEL3(16 * 9)
KERNEL4(16 * 9)
KERNEL5(16 * 9)
KERNEL6(16 * 9)
KERNEL7(16 * 9)
KERNEL8(16 * 9)
cmpl $128 * 10, %eax
jle .L11
KERNEL1(16 * 10)
KERNEL2(16 * 10)
KERNEL3(16 * 10)
KERNEL4(16 * 10)
KERNEL5(16 * 10)
KERNEL6(16 * 10)
KERNEL7(16 * 10)
KERNEL8(16 * 10)
cmpl $128 * 11, %eax
jle .L11
KERNEL1(16 * 11)
KERNEL2(16 * 11)
KERNEL3(16 * 11)
KERNEL4(16 * 11)
KERNEL5(16 * 11)
KERNEL6(16 * 11)
KERNEL7(16 * 11)
KERNEL8(16 * 11)
cmpl $128 * 12, %eax
jle .L11
KERNEL1(16 * 12)
KERNEL2(16 * 12)
KERNEL3(16 * 12)
KERNEL4(16 * 12)
KERNEL5(16 * 12)
KERNEL6(16 * 12)
KERNEL7(16 * 12)
KERNEL8(16 * 12)
cmpl $128 * 13, %eax
jle .L11
KERNEL1(16 * 13)
KERNEL2(16 * 13)
KERNEL3(16 * 13)
KERNEL4(16 * 13)
KERNEL5(16 * 13)
KERNEL6(16 * 13)
KERNEL7(16 * 13)
KERNEL8(16 * 13)
cmpl $128 * 14, %eax
jle .L11
KERNEL1(16 * 14)
KERNEL2(16 * 14)
KERNEL3(16 * 14)
KERNEL4(16 * 14)
KERNEL5(16 * 14)
KERNEL6(16 * 14)
KERNEL7(16 * 14)
KERNEL8(16 * 14)
cmpl $128 * 15, %eax
jle .L11
KERNEL1(16 * 15)
KERNEL2(16 * 15)
KERNEL3(16 * 15)
KERNEL4(16 * 15)
KERNEL5(16 * 15)
KERNEL6(16 * 15)
KERNEL7(16 * 15)
KERNEL8(16 * 15)
#else
addl $32 * 4 * SIZE, AA
addl $32 * 8 * SIZE, BB
subl $128 * 8, %eax
jg .L1X
#endif
.L11:
leal (AA, %eax, 1), AA # * 16
leal (BB, %eax, 2), BB # * 64
#else
sarl $3, %eax
je .L12
ALIGN_4
.L11:
KERNEL1(16 * 0)
KERNEL2(16 * 0)
KERNEL3(16 * 0)
KERNEL4(16 * 0)
KERNEL5(16 * 0)
KERNEL6(16 * 0)
KERNEL7(16 * 0)
KERNEL8(16 * 0)
addl $32 * SIZE, BB
addl $16 * SIZE, AA
decl %eax
jne .L11
ALIGN_4
#endif
.L12:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movddup ALPHA_R, %xmm1
movddup ALPHA_I, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je .L14
ALIGN_4
.L13:
mulpd %xmm0, %xmm2
addpd %xmm2, %xmm4
movddup 1 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
ADDSUB %xmm2, %xmm5
movddup 2 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm2, %xmm6
movddup 3 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
movapd 2 * SIZE(AA), %xmm0
ADDSUB %xmm2, %xmm7
movddup 4 * SIZE(BB), %xmm2
addl $2 * SIZE, AA
addl $4 * SIZE, BB
decl %eax
jg .L13
ALIGN_4
.L14:
pcmpeqb %xmm0, %xmm0
SHUFPD_1 %xmm5, %xmm5
psllq $63, %xmm0
SHUFPD_1 %xmm7, %xmm7
#if defined(NN) || defined(NT) || defined(TN) || defined(TT)
shufps $0x04, %xmm0, %xmm0
pxor %xmm0, %xmm5
pxor %xmm0, %xmm7
#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
shufps $0x40, %xmm0, %xmm0
pxor %xmm0, %xmm5
pxor %xmm0, %xmm7
#elif defined(RN) || defined(RT) || defined(CN) || defined(CT)
shufps $0x40, %xmm0, %xmm0
pxor %xmm0, %xmm4
pxor %xmm0, %xmm6
#else
shufps $0x40, %xmm0, %xmm0
pxor %xmm0, %xmm4
pxor %xmm0, %xmm6
#endif
addpd %xmm5, %xmm4
addpd %xmm7, %xmm6
movaps %xmm4, %xmm5
movaps %xmm6, %xmm7
SHUFPD_1 %xmm5, %xmm5
SHUFPD_1 %xmm7, %xmm7
mulpd %xmm1, %xmm4
mulpd %xmm1, %xmm6
mulpd %xmm3, %xmm5
mulpd %xmm3, %xmm7
addsubpd %xmm5, %xmm4
addsubpd %xmm7, %xmm6
#ifndef TRMMKERNEL
movsd 0 * SIZE(%esi), %xmm0
movhpd 1 * SIZE(%esi), %xmm0
movsd 0 * SIZE(%esi, LDC), %xmm2
movhpd 1 * SIZE(%esi, LDC), %xmm2
addpd %xmm0, %xmm4
addpd %xmm2, %xmm6
#endif
movsd %xmm4, 0 * SIZE(%esi)
movhpd %xmm4, 1 * SIZE(%esi)
movsd %xmm6, 0 * SIZE(%esi, LDC)
movhpd %xmm6, 1 * SIZE(%esi, LDC)
#if (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
movl K, %eax
subl KKK, %eax
leal (,%eax, SIZE), %eax
leal (AA, %eax, 2), AA
leal (BB, %eax, 4), BB
#endif
#if defined(TRMMKERNEL) && defined(LEFT)
addl $1, KK
#endif
addl $2 * SIZE, %esi # coffset += 4
decl %ebx # i --
jg .L10
ALIGN_4
.L99:
#if defined(TRMMKERNEL) && !defined(LEFT)
addl $2, KK
#endif
leal (, LDC, 2), %eax
movl BB, B
addl %eax, C # c += ldc
decl J # j --
jg .L01
ALIGN_4
.L100:
movl N, %eax
testl $1, %eax
jle .L500
movl C, %esi # coffset = c
movl A, AA # aoffset = a
#if defined(TRMMKERNEL) && defined(LEFT)
movl OFFSET, %eax
movl %eax, KK
#endif
movl M, %ebx
testl %ebx, %ebx
jle .L500
ALIGN_4
L110:
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
movl B, BB
#else
movl KK, %eax
leal (, %eax, SIZE), %eax
leal (AA, %eax, 2), AA
leal (B, %eax, 2), BB
#endif
movapd 0 * SIZE(AA), %xmm0
pxor %xmm4, %xmm4
movapd 8 * SIZE(AA), %xmm1
pxor %xmm5, %xmm5
movddup 0 * SIZE(BB), %xmm2
pxor %xmm6, %xmm6
movddup 8 * SIZE(BB), %xmm3
pxor %xmm7, %xmm7
#ifdef PENTIUM4
prefetchnta 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 $1, %eax
#else
addl $1, %eax
#endif
movl %eax, KKK
#endif
sarl $3, %eax
je L112
ALIGN_4
L111:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AA)
mulpd %xmm0, %xmm2
addpd %xmm2, %xmm4
movddup 1 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
movapd 2 * SIZE(AA), %xmm0
ADDSUB %xmm2, %xmm5
movddup 2 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm2, %xmm6
movddup 3 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
movapd 4 * SIZE(AA), %xmm0
ADDSUB %xmm2, %xmm7
movddup 4 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm2, %xmm4
movddup 5 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
movapd 6 * SIZE(AA), %xmm0
ADDSUB %xmm2, %xmm5
movddup 6 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm2, %xmm6
movddup 7 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
movapd 16 * SIZE(AA), %xmm0
ADDSUB %xmm2, %xmm7
movddup 16 * SIZE(BB), %xmm2
mulpd %xmm1, %xmm3
addpd %xmm3, %xmm4
movddup 9 * SIZE(BB), %xmm3
mulpd %xmm1, %xmm3
movapd 10 * SIZE(AA), %xmm1
ADDSUB %xmm3, %xmm5
movddup 10 * SIZE(BB), %xmm3
mulpd %xmm1, %xmm3
addpd %xmm3, %xmm6
movddup 11 * SIZE(BB), %xmm3
mulpd %xmm1, %xmm3
movapd 12 * SIZE(AA), %xmm1
ADDSUB %xmm3, %xmm7
movddup 12 * SIZE(BB), %xmm3
mulpd %xmm1, %xmm3
addpd %xmm3, %xmm4
movddup 13 * SIZE(BB), %xmm3
mulpd %xmm1, %xmm3
movapd 14 * SIZE(AA), %xmm1
ADDSUB %xmm3, %xmm5
movddup 14 * SIZE(BB), %xmm3
mulpd %xmm1, %xmm3
addpd %xmm3, %xmm6
movddup 15 * SIZE(BB), %xmm3
mulpd %xmm1, %xmm3
movapd 24 * SIZE(AA), %xmm1
ADDSUB %xmm3, %xmm7
movddup 24 * SIZE(BB), %xmm3
addl $16 * SIZE, AA
addl $16 * SIZE, BB
decl %eax
jne L111
ALIGN_4
L112:
#ifndef TRMMKERNEL
movl K, %eax
#else
movl KKK, %eax
#endif
movddup ALPHA_R, %xmm1
movddup ALPHA_I, %xmm3
andl $7, %eax # if (k & 1)
BRANCH
je L114
ALIGN_4
L113:
mulpd %xmm0, %xmm2
addpd %xmm2, %xmm4
movddup 1 * SIZE(BB), %xmm2
mulpd %xmm0, %xmm2
movapd 2 * SIZE(AA), %xmm0
ADDSUB %xmm2, %xmm5
movddup 2 * SIZE(BB), %xmm2
addl $2 * SIZE, AA
addl $2 * SIZE, BB
decl %eax
jg L113
ALIGN_4
L114:
addpd %xmm6, %xmm4
addpd %xmm7, %xmm5
pcmpeqb %xmm0, %xmm0
SHUFPD_1 %xmm5, %xmm5
psllq $63, %xmm0
#if defined(NN) || defined(NT) || defined(TN) || defined(TT)
shufps $0x04, %xmm0, %xmm0
pxor %xmm0, %xmm5
#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
shufps $0x40, %xmm0, %xmm0
pxor %xmm0, %xmm5
#elif defined(RN) || defined(RT) || defined(CN) || defined(CT)
shufps $0x40, %xmm0, %xmm0
pxor %xmm0, %xmm4
#else
shufps $0x40, %xmm0, %xmm0
pxor %xmm0, %xmm4
#endif
addpd %xmm5, %xmm4
movaps %xmm4, %xmm5
SHUFPD_1 %xmm5, %xmm5
mulpd %xmm1, %xmm4
mulpd %xmm3, %xmm5
addsubpd %xmm5, %xmm4
#ifndef TRMMKERNEL
movsd 0 * SIZE(%esi), %xmm0
movhpd 1 * SIZE(%esi), %xmm0
addpd %xmm0, %xmm4
#endif
movsd %xmm4, 0 * SIZE(%esi)
movhpd %xmm4, 1 * SIZE(%esi)
#if (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
movl K, %eax
subl KKK, %eax
leal (,%eax, SIZE), %eax
leal (AA, %eax, 2), AA
leal (BB, %eax, 2), BB
#endif
#if defined(TRMMKERNEL) && defined(LEFT)
addl $1, KK
#endif
addl $2 * SIZE, %esi # coffset += 4
decl %ebx # i --
jg L110
ALIGN_4
.L500:
popl %ebx
popl %esi
popl %edi
popl %ebp
addl $ARGS, %esp
ret
EPILOGUE