/*********************************************************************/

/* 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	 0

#define OLD_M	 4 + STACK + ARGS(%esi)

#define OLD_N	 8 + STACK + ARGS(%esi)

#define OLD_K	12 + STACK + ARGS(%esi)

#define OLD_ALPHA	16 + STACK + ARGS(%esi)

#define OLD_A	24 + STACK + ARGS(%esi)

#define OLD_B	28 + STACK + ARGS(%esi)

#define OLD_C	32 + STACK + ARGS(%esi)

#define OLD_LDC	36 + STACK + ARGS(%esi)

#define OLD_OFFT	40 + STACK + ARGS(%esi)

#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 OLD_STACK 40(%esp)

#define OFFSET  44(%esp)

#define KK	48(%esp)

#define KKK	52(%esp)

#define AORIG	56(%esp)

#define BORIG	60(%esp)

#define BUFFER 128(%esp)

#define STACK_ALIGN	4096

#define STACK_OFFSET	1024

#if defined(OPTERON) || defined(BARCELONA)

#define PREFETCH     prefetch

#define PREFETCHSIZE (8 * 10 + 4)

#endif

#define B	%edi

#define AA	%edx

#define BB	%ecx

#define LDC	%ebp

#define CO1	%esi

#define KERNEL1(address) \

	mulpd	%xmm0, %xmm2; \

	addpd	%xmm2, %xmm4; \

	PREFETCH (PREFETCHSIZE +  0) * SIZE + (address) * 1 * SIZE(AA); \

	movapd	 2 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	mulpd	%xmm0, %xmm2; \

	addpd	%xmm2, %xmm5; \

	movapd	 4 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	mulpd	%xmm0, %xmm2; \

	mulpd	 6 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \

	addpd	%xmm2, %xmm6; \

	movapd	16 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	addpd	%xmm0, %xmm7; \

	movapd	 2 * SIZE + (address) * 1 * SIZE(AA), %xmm0

#define KERNEL2(address) \

	mulpd	%xmm0, %xmm3; \

	addpd	%xmm3, %xmm4; \

	movapd	10 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	mulpd	%xmm0, %xmm3; \

	addpd	%xmm3, %xmm5; \

	movapd	12 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	mulpd	%xmm0, %xmm3; \

	mulpd	14 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \

	addpd	%xmm3, %xmm6; \

	movapd	24 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	addpd	%xmm0, %xmm7; \

	movapd	 4 * SIZE + (address) * 1 * SIZE(AA), %xmm0

#define KERNEL3(address) \

	mulpd	%xmm0, %xmm2; \

	addpd	%xmm2, %xmm4; \

	movapd	18 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	mulpd	%xmm0, %xmm2; \

	addpd	%xmm2, %xmm5; \

	movapd	20 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	mulpd	%xmm0, %xmm2; \

	mulpd	22 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \

	addpd	%xmm2, %xmm6; \

	movapd	32 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	addpd	%xmm0, %xmm7; \

	movapd	 6 * SIZE + (address) * 1 * SIZE(AA), %xmm0

#define KERNEL4(address) \

	mulpd	%xmm0, %xmm3; \

	addpd	%xmm3, %xmm4; \

	movapd	26 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	mulpd	%xmm0, %xmm3; \

	addpd	%xmm3, %xmm5; \

	movapd	28 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	mulpd	%xmm0, %xmm3; \

	mulpd	30 * SIZE + (address) * 4 * SIZE(BB), %xmm0; \

	addpd	%xmm3, %xmm6; \

	movapd	40 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	addpd	%xmm0, %xmm7; \

	movapd	16 * SIZE + (address) * 1 * SIZE(AA), %xmm0

#define KERNEL5(address) \

	PREFETCH (PREFETCHSIZE + 8) * SIZE + (address) * 1 * SIZE(AA); \

	mulpd	%xmm1, %xmm2; \

	addpd	%xmm2, %xmm4; \

	movapd	34 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	mulpd	%xmm1, %xmm2; \

	addpd	%xmm2, %xmm5; \

	movapd	36 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	mulpd	%xmm1, %xmm2; \

	mulpd	38 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \

	addpd	%xmm2, %xmm6; \

	movapd	48 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	addpd	%xmm1, %xmm7; \

	movapd	10 * SIZE + (address) * 1 * SIZE(AA), %xmm1

#define KERNEL6(address) \

	mulpd	%xmm1, %xmm3; \

	addpd	%xmm3, %xmm4; \

	movapd	42 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	mulpd	%xmm1, %xmm3; \

	addpd	%xmm3, %xmm5; \

	movapd	44 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	mulpd	%xmm1, %xmm3; \

	mulpd	46 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \

	addpd	%xmm3, %xmm6; \

	movapd	56 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	addpd	%xmm1, %xmm7; \

	movapd	12 * SIZE + (address) * 1 * SIZE(AA), %xmm1

#define KERNEL7(address) \

	mulpd	%xmm1, %xmm2; \

	addpd	%xmm2, %xmm4; \

	movapd	50 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	mulpd	%xmm1, %xmm2; \

	addpd	%xmm2, %xmm5; \

	movapd	52 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	mulpd	%xmm1, %xmm2; \

	mulpd	54 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \

	addpd	%xmm2, %xmm6; \

	movapd	64 * SIZE + (address) * 4 * SIZE(BB), %xmm2; \

	addpd	%xmm1, %xmm7; \

	movapd	14 * SIZE + (address) * 1 * SIZE(AA), %xmm1

#define KERNEL8(address) \

	mulpd	%xmm1, %xmm3; \

	addpd	%xmm3, %xmm4; \

	movapd	58 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	mulpd	%xmm1, %xmm3; \

	addpd	%xmm3, %xmm5; \

	movapd	60 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	mulpd	%xmm1, %xmm3; \

	mulpd	62 * SIZE + (address) * 4 * SIZE(BB), %xmm1; \

	addpd	%xmm3, %xmm6; \

	movapd	72 * SIZE + (address) * 4 * SIZE(BB), %xmm3; \

	addpd	%xmm1, %xmm7; \

	movapd	24 * SIZE + (address) * 1 * SIZE(AA), %xmm1

	PROLOGUE

	pushl	%ebp

	pushl	%edi

	pushl	%esi

	pushl	%ebx

	PROFCODE

	EMMS

	movl	%esp, %esi	# save old stack

	subl	$128 + LOCAL_BUFFER_SIZE + STACK_OFFSET, %esp

	andl	$-STACK_ALIGN, %esp

	addl	$STACK_OFFSET, %esp

	STACK_TOUCHING

	movl	OLD_M, %ebx

	movl	OLD_N, %eax

	movl	OLD_K, %ecx

	movl	OLD_A, %edx

	movl	%ebx, M

	movl	%eax, N

	movl	%ecx, K

	movl	%edx, A

	movl	%esi, OLD_STACK

	movd	OLD_OFFT, %mm4

	movl	OLD_B, B

	movl	OLD_C, %ebx

	movl	%ebx, C

	movl	OLD_LDC, LDC

	movd	%mm4, OFFSET

	movd	%mm4, KK

	leal	(, LDC, SIZE), LDC

#ifdef LN

       movl	M, %eax

       leal	(, %eax, SIZE), %eax

       addl	%eax, C

       imull	K, %eax

       addl	%eax, A

#endif

#ifdef RT

       movl	N, %eax

       leal	(, %eax, SIZE), %eax

       imull	K, %eax

       addl	%eax, B

       movl	N, %eax

       imull	LDC, %eax

       addl	%eax, C

#endif

#ifdef RN

	negl	KK

#endif	

#ifdef RT

       movl	N, %eax

       subl	OFFSET, %eax

       movl	%eax, KK

#endif

	testl	$1, N

	je	.L30

#ifdef LN

	movl	OFFSET, %eax

	addl	M, %eax

	movl	%eax, KK

#endif	

	leal	BUFFER, BB

#ifdef RT

       movl	K, %eax

       sall	$BASE_SHIFT, %eax

       subl	%eax, B

#endif

#if defined(LN) || defined(RT)

	movl	KK, %eax

	movl	B, BORIG

	leal	(, %eax, SIZE), %eax

	leal	(B,  %eax, 1), B

	leal	(BB, %eax, 2), BB

#endif	

#ifdef LT

	movl	OFFSET, %eax

	movl	%eax, KK

#endif

#if defined(LT) || defined(RN)

	movl	KK, %eax

#else

	movl	K, %eax

	subl	KK, %eax

#endif

	sarl	$3, %eax

	jle	.L65

	ALIGN_4

.L62:

#define COPYPREFETCH 40

	prefetchnta	(COPYPREFETCH) * SIZE(B)

	movq	 0 * SIZE(B), %mm0

	movq	 1 * SIZE(B), %mm1

	movq	 2 * SIZE(B), %mm2

	movq	 3 * SIZE(B), %mm3

	movq	 4 * SIZE(B), %mm4

	movq	 5 * SIZE(B), %mm5

	movq	 6 * SIZE(B), %mm6

	movq	 7 * SIZE(B), %mm7

	movq	%mm0,  0 * SIZE(BB)

	movq	%mm0,  1 * SIZE(BB)

	movq	%mm1,  2 * SIZE(BB)

	movq	%mm1,  3 * SIZE(BB)

	movq	%mm2,  4 * SIZE(BB)

	movq	%mm2,  5 * SIZE(BB)

	movq	%mm3,  6 * SIZE(BB)

	movq	%mm3,  7 * SIZE(BB)

	movq	%mm4,  8 * SIZE(BB)

	movq	%mm4,  9 * SIZE(BB)

	movq	%mm5, 10 * SIZE(BB)

	movq	%mm5, 11 * SIZE(BB)

	movq	%mm6, 12 * SIZE(BB)

	movq	%mm6, 13 * SIZE(BB)

	movq	%mm7, 14 * SIZE(BB)

	movq	%mm7, 15 * SIZE(BB)

	addl	$ 8 * SIZE, B

	addl	$16 * SIZE, BB

	decl	%eax

	jne	.L62

	ALIGN_2

.L65:

#if defined(LT) || defined(RN)

	movl	KK, %eax

#else

	movl	K, %eax

	subl	KK, %eax

#endif

	andl	$7, %eax

	BRANCH

	jle	.L70

	ALIGN_2

.L66:

	movq	 0 * SIZE(B), %mm0

	movq	%mm0,  0 * SIZE(BB)

	movq	%mm0,  1 * SIZE(BB)

	addl	$1 * SIZE, B

	addl	$2 * SIZE, BB

	decl	%eax

	jne	.L66

	ALIGN_4

.L70:

#if defined(LT) || defined(RN)

	movl	A, AA

#else

	movl	A, %eax

	movl	%eax, AORIG

#endif

#ifdef RT

	subl	LDC, C

#endif

	movl	C, CO1

#ifndef RT

	addl	LDC, C

#endif

	movl	M,  %ebx

	sarl	$1, %ebx	# i = (m >> 2)

	jle	.L80

	ALIGN_4

.L71:

#ifdef LN

       movl	K, %eax

       sall	$1 + BASE_SHIFT, %eax

       subl	%eax, AORIG

#endif

#if defined(LN) || defined(RT)

	movl	KK, %eax

	movl	AORIG, AA

	leal	(, %eax, SIZE), %eax

	leal	(AA, %eax, 2), AA

#endif

	leal	BUFFER, BB

#if defined(LN) || defined(RT)

	movl	KK, %eax

	sall	$1 + BASE_SHIFT, %eax

	addl	%eax, BB

#endif	

	pxor	%xmm4, %xmm4

	pxor	%xmm5, %xmm5

	pxor	%xmm6, %xmm6

	pxor	%xmm7, %xmm7

	movapd	 0 * SIZE(AA), %xmm0

	movapd	 8 * SIZE(AA), %xmm1

	movapd	 0 * SIZE(BB), %xmm2

	movapd	 8 * SIZE(BB), %xmm3

#ifdef LN

	prefetchw	-2 * SIZE(CO1)

#else

	prefetchw	 1 * SIZE(CO1)

#endif

#if defined(LT) || defined(RN)

	movl	KK, %eax

#else

	movl	K, %eax

	subl	KK, %eax

#endif

	sarl	$3, %eax

	je	.L75

	ALIGN_4

.L72:

	mulpd	%xmm0, %xmm2

	addpd	%xmm2, %xmm4

	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)

	movapd	16 * SIZE(BB), %xmm2

	movapd	 2 * SIZE(AA), %xmm0

	mulpd	 2 * SIZE(BB), %xmm0

	addpd	%xmm0, %xmm4

	movapd	 4 * SIZE(AA), %xmm0

	mulpd	 4 * SIZE(BB), %xmm0

	addpd	%xmm0, %xmm4

	movapd	 6 * SIZE(AA), %xmm0

	mulpd	 6 * SIZE(BB), %xmm0

	addpd	%xmm0, %xmm4

	movapd	16 * SIZE(AA), %xmm0

	prefetcht0 (PREFETCHSIZE  + 8) * SIZE(AA)

	mulpd	%xmm1, %xmm3

	addpd	%xmm3, %xmm4

	movapd	24 * SIZE(BB), %xmm3

	movapd	10 * SIZE(AA), %xmm1

	mulpd	10 * SIZE(BB), %xmm1

	addpd	%xmm1, %xmm4

	movapd	12 * SIZE(AA), %xmm1

	mulpd	12 * SIZE(BB), %xmm1

	addpd	%xmm1, %xmm4

	movapd	14 * SIZE(AA), %xmm1

	mulpd	14 * SIZE(BB), %xmm1

	addpd	%xmm1, %xmm4

	movapd	24 * SIZE(AA), %xmm1

	addl   $16 * SIZE, AA

	addl   $16 * SIZE, BB

	decl   %eax

	jne    .L72

	ALIGN_4

.L75:

#if defined(LT) || defined(RN)

	movl	KK, %eax

#else

	movl	K, %eax

	subl	KK, %eax

#endif

	andl	$7, %eax		# if (k & 1)

	BRANCH

	je .L78

	ALIGN_3

.L76:

	mulpd	%xmm0, %xmm2

	addpd	%xmm2, %xmm4

	movapd	 2 * SIZE(AA), %xmm0

	movapd	 2 * SIZE(BB), %xmm2

	addl	$2 * SIZE, AA

	addl	$2 * SIZE, BB

	decl	%eax

	jg	.L76

	ALIGN_4

.L78:

#if defined(LN) || defined(RT)

	movl	KK, %eax

#ifdef LN

	subl	$2, %eax

#else

	subl	$1, %eax

#endif

	movl	AORIG, AA

	movl	BORIG, B

	leal	BUFFER, BB

	leal	(, %eax, SIZE), %eax

	leal	(AA, %eax, 2), AA

	leal	(B,  %eax, 1), B

	leal	(BB, %eax, 2), BB

#endif

#if defined(LN) || defined(LT)

	movapd	 0 * SIZE(B), %xmm2

	subpd	%xmm4,  %xmm2

#else

	movapd	 0 * SIZE(AA), %xmm0

	subpd	%xmm4, %xmm0

#endif

#ifdef LN

	movapd	 %xmm2, %xmm3

	unpckhpd %xmm3, %xmm3

	movlpd	 3 * SIZE(AA), %xmm4

	mulsd	 %xmm4, %xmm3

	movlpd	 2 * SIZE(AA), %xmm4

	mulsd	 %xmm3, %xmm4

	subsd	 %xmm4, %xmm2

	movlpd	 0 * SIZE(AA), %xmm4

	mulsd	 %xmm4, %xmm2

	unpcklpd %xmm3, %xmm2

#endif

#ifdef LT

	movapd	 %xmm2, %xmm3

	unpckhpd %xmm3, %xmm3

	movlpd	 0 * SIZE(AA), %xmm4

	mulsd	 %xmm4, %xmm2

	movlpd	 1 * SIZE(AA), %xmm4

	mulsd	 %xmm2, %xmm4

	subsd	 %xmm4, %xmm3

	movlpd	 3 * SIZE(AA), %xmm4

	mulsd	 %xmm4, %xmm3

	unpcklpd %xmm3, %xmm2

#endif

#ifdef RN

	movlpd	 0 * SIZE(B), %xmm4

	movhpd	 0 * SIZE(B), %xmm4

	mulpd	 %xmm4, %xmm0

#endif

#ifdef RT

	movlpd	 0 * SIZE(B), %xmm4

	movhpd	 0 * SIZE(B), %xmm4

	mulpd	 %xmm4, %xmm0

#endif

#if defined(LN) || defined(LT)

	movapd	%xmm2,   0 * SIZE(B)

	movlpd	%xmm2,   0 * SIZE(BB)

	movlpd	%xmm2,   1 * SIZE(BB)

	movhpd	%xmm2,   2 * SIZE(BB)

	movhpd	%xmm2,   3 * SIZE(BB)

#else

	movapd	%xmm0,   0 * SIZE(AA)

#endif

#ifdef LN

	subl	$2 * SIZE, CO1

#endif

#if defined(LN) || defined(LT)

	movlpd	%xmm2, 0 * SIZE(CO1)

	movhpd	%xmm2, 1 * SIZE(CO1)

#else

	movlpd	%xmm0, 0 * SIZE(CO1)

	movhpd	%xmm0, 1 * SIZE(CO1)

#endif

#ifndef LN

	addl	$2 * SIZE, CO1

#endif

#if defined(LT) || defined(RN)

	movl	K,  %eax

	subl	KK, %eax

	leal	(,%eax, SIZE), %eax

	leal	(AA, %eax, 2), AA

#ifdef LT

	addl	$2 * SIZE, B

#endif

#endif

#ifdef LN

	subl	$2, KK

	movl	BORIG, B

#endif

#ifdef LT

	addl	$2, KK

#endif

#ifdef RT

	movl	K, %eax

	movl	BORIG, B

	sall	$1 + BASE_SHIFT, %eax

	addl	%eax, AORIG

#endif

	decl	%ebx			# i --

	jg	.L71

	ALIGN_4

.L80:

	movl	M,  %ebx

	testl	$1, %ebx	# i = (m >> 2)

	jle	.L99

#ifdef LN

       movl	K, %eax

       sall	$BASE_SHIFT, %eax

       subl	%eax, AORIG

#endif

#if defined(LN) || defined(RT)

	movl	KK, %eax

	movl	AORIG, AA

	leal	(AA, %eax, SIZE), AA

#endif

	leal	BUFFER, BB

#if defined(LN) || defined(RT)

	movl	KK, %eax

	sall	$1 + BASE_SHIFT, %eax

	addl	%eax, BB

#endif	

	pxor	%xmm4, %xmm4

	pxor	%xmm5, %xmm5

	pxor	%xmm6, %xmm6

	pxor	%xmm7, %xmm7

	movlpd	 0 * SIZE(AA), %xmm0

	movlpd	 4 * SIZE(AA), %xmm1

	movlpd	 0 * SIZE(BB), %xmm2

	movlpd	 8 * SIZE(BB), %xmm3

#if defined(LT) || defined(RN)

	movl	KK, %eax

#else

	movl	K, %eax

	subl	KK, %eax

#endif

	sarl	$3, %eax

	je	.L85

	ALIGN_4

.L82:

	mulsd	%xmm0, %xmm2

	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)

	movlpd	 1 * SIZE(AA), %xmm0

	mulsd	 2 * SIZE(BB), %xmm0

	addsd	%xmm2, %xmm4

	movlpd	16 * SIZE(BB), %xmm2

	addsd	%xmm0, %xmm5

	movlpd	 2 * SIZE(AA), %xmm0

	mulsd	 4 * SIZE(BB), %xmm0

	addsd	%xmm0, %xmm6

	movlpd	 3 * SIZE(AA), %xmm0

	mulsd	 6 * SIZE(BB), %xmm0

	addsd	%xmm0, %xmm7

	movlpd	 8 * SIZE(AA), %xmm0

	mulsd	%xmm1, %xmm3

	movlpd	 5 * SIZE(AA), %xmm1

	mulsd	10 * SIZE(BB), %xmm1

	addsd	%xmm3, %xmm4

	movlpd	24 * SIZE(BB), %xmm3

	addsd	%xmm1, %xmm5

	movlpd	 6 * SIZE(AA), %xmm1

	mulsd	12 * SIZE(BB), %xmm1

	addsd	%xmm1, %xmm6

	movlpd	 7 * SIZE(AA), %xmm1

	mulsd	14 * SIZE(BB), %xmm1

	addsd	%xmm1, %xmm7

	movlpd	12 * SIZE(AA), %xmm1

	addl   $ 8 * SIZE, AA

	addl   $16 * SIZE, BB

	decl   %eax

	jne    .L82

	ALIGN_4

.L85:

#if defined(LT) || defined(RN)

	movl	KK, %eax

#else

	movl	K, %eax

	subl	KK, %eax

#endif

	andl	$7, %eax		# if (k & 1)

	BRANCH

	je .L88

.L86:

	mulsd	%xmm0, %xmm2

	addsd	%xmm2, %xmm4

	movlpd	 2 * SIZE(BB), %xmm2

	movlpd	 1 * SIZE(AA), %xmm0

	addl	$1 * SIZE, AA

	addl	$2 * SIZE, BB

	decl	%eax

	jg	.L86

	ALIGN_4

.L88:

	addsd	%xmm5, %xmm4

	addsd	%xmm7, %xmm6

	addsd	%xmm6, %xmm4

#if defined(LN) || defined(RT)

	movl	KK, %eax

#ifdef LN

	subl	$1, %eax

#else

	subl	$1, %eax

#endif

	movl	AORIG, AA

	movl	BORIG, B

	leal	BUFFER, BB

	leal	(, %eax, SIZE), %eax

	addl	%eax, AA

	addl	%eax, B

	leal	(BB, %eax, 2), BB

#endif

#if defined(LN) || defined(LT)

	movlpd	 0 * SIZE(B), %xmm2

	subsd	%xmm4, %xmm2

#else

	movlpd	 0 * SIZE(AA), %xmm0

	subsd	%xmm4, %xmm0

#endif

#ifdef LN

	movlpd	 0 * SIZE(AA), %xmm4

	mulsd	 %xmm4, %xmm2

#endif

#ifdef LT

	movlpd	 0 * SIZE(AA), %xmm4

	mulsd	 %xmm4, %xmm2

#endif

#ifdef RN

	movlpd	 0 * SIZE(B), %xmm4

	mulsd	 %xmm4, %xmm0

#endif

#ifdef RT

	movlpd	 0 * SIZE(B), %xmm4

	mulsd	 %xmm4, %xmm0

#endif

#if defined(LN) || defined(LT)

	movlpd	%xmm2,   0 * SIZE(B)

	movlpd	%xmm2,   0 * SIZE(BB)

	movlpd	%xmm2,   1 * SIZE(BB)

#else

	movlpd	%xmm0,   0 * SIZE(AA)

#endif

#ifdef LN

	subl	$1 * SIZE, CO1

#endif

#if defined(LN) || defined(LT)

	movlpd	%xmm2, 0 * SIZE(CO1)

#else

	movlpd	%xmm0, 0 * SIZE(CO1)

#endif

#ifndef LN

	addl	$1 * SIZE, CO1

#endif

#if defined(LT) || defined(RN)

	movl	K,  %eax

	subl	KK, %eax

	leal	(AA,%eax, SIZE), AA

#ifdef LT

	addl	$1 * SIZE, B

#endif

#endif

#ifdef LN

	subl	$1, KK

	movl	BORIG, B

#endif

#ifdef LT

	addl	$1, KK

#endif

#ifdef RT

	movl	K, %eax

	movl	BORIG, B

	sall	$BASE_SHIFT, %eax

	addl	%eax, AORIG

#endif

	ALIGN_4

.L99:

#ifdef LN

       movl	K, %eax

       leal	(B, %eax, SIZE), B

#endif

#if defined(LT) || defined(RN)

	movl	K,  %eax

	subl	KK, %eax

	leal	(B,%eax, SIZE), B

#endif

#ifdef RN

	addl	$1, KK

#endif

#ifdef RT

	subl	$1, KK

#endif

	ALIGN_4

.L30:

	testl	$2, N

	je	.L60

#ifdef LN

	movl	OFFSET, %eax

	addl	M, %eax

	movl	%eax, KK

#endif	

	leal	BUFFER, BB

#ifdef RT

       movl	K, %eax

       sall	$1 + BASE_SHIFT, %eax

       subl	%eax, B

#endif

#if defined(LN) || defined(RT)

	movl	KK, %eax

	movl	B, BORIG