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

/* 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 M	%rdi

#define N	%rsi

#define K	%rdx

#define A	%rcx

#define B	%r8

#define C	%r9

#define LDC	%r10

#define I	%r11

#define J	%r12

#define AO	%r13

#define BO	%r14

#define	CO1	%r15

#define CO2	%rbp

#ifndef WINDOWS_ABI

#define STACKSIZE 64

#define OLD_LDC		 8 + STACKSIZE(%rsp)

#define OLD_OFFSET	16 + STACKSIZE(%rsp)

#else

#define STACKSIZE 256

#define OLD_A		40 + STACKSIZE(%rsp)

#define OLD_B		48 + STACKSIZE(%rsp)

#define OLD_C		56 + STACKSIZE(%rsp)

#define OLD_LDC		64 + STACKSIZE(%rsp)

#define OLD_OFFSET	72 + STACKSIZE(%rsp)

#endif

#define ALPHA	  0(%rsp)

#define OFFSET	 16(%rsp)

#define KK	 24(%rsp)

#define KKK	 32(%rsp)

#define AORIG	 40(%rsp)

#define BORIG	 48(%rsp)

#define BUFFER	128(%rsp)

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

#define PREFETCH     prefetch

#define PREFETCHW    prefetchw

#define PREFETCHNTA  prefetchnta

#ifndef ALLOC_HUGETLB

#define PREFETCHSIZE (8 * 4 + 4)

#else

#define PREFETCHSIZE (8 * 2 + 4)

#endif

#endif

#ifdef GENERIC

#define PREFETCH     prefetcht0

#define PREFETCHW    prefetcht0

#define PREFETCHNTA  prefetchnta

#define PREFETCHSIZE (8 * 4 + 4)

#endif

#ifdef OPTERON	

#define movsd	movlpd

#endif

#define KERNEL1(xx) \

	mulpd	%xmm8, %xmm9 ;\

	addpd	%xmm9, %xmm0 ;\

	movapd	 0 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\

	mulpd	%xmm8, %xmm11 ;\

	PREFETCH	(PREFETCHSIZE     +  0) * SIZE + 1 * (xx) * SIZE(AO) ;\

	addpd	%xmm11, %xmm1 ;\

	movapd	 2 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\

	mulpd	%xmm8, %xmm13 ;\

	mulpd	 6 * SIZE + 2 * (xx) * SIZE(BO), %xmm8 ;\

	addpd	%xmm13, %xmm2 ;\

	movapd	 4 * SIZE + 2 * (xx) * SIZE(BO), %xmm13 ;\

	addpd	%xmm8, %xmm3 ;\

	movapd	 8 * SIZE + 1 * (xx) * SIZE(AO), %xmm8

#define KERNEL2(xx) \

	mulpd	%xmm10, %xmm9 ;\

	addpd	%xmm9, %xmm4 ;\

	movapd	16 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\

	mulpd	%xmm10, %xmm11 ;\

	addpd	%xmm11, %xmm5 ;\

	movapd	10 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\

	mulpd	%xmm10, %xmm13 ;\

	mulpd	 6 * SIZE + 2 * (xx) * SIZE(BO), %xmm10 ;\

	addpd	%xmm13, %xmm6 ;\

	movapd	12 * SIZE + 2 * (xx) * SIZE(BO), %xmm13 ;\

	addpd	%xmm10, %xmm7 ;\

	movapd	10 * SIZE + 1 * (xx) * SIZE(AO), %xmm10

#define KERNEL3(xx) \

	mulpd	%xmm12, %xmm15 ;\

	addpd	%xmm15, %xmm0 ;\

	movapd	 8 * SIZE + 2 * (xx) * SIZE(BO), %xmm15 ;\

	mulpd	%xmm12, %xmm11 ;\

	addpd	%xmm11, %xmm1 ;\

	movapd	10 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\

	mulpd	%xmm12, %xmm13 ;\

	mulpd	14 * SIZE + 2 * (xx) * SIZE(BO), %xmm12 ;\

	addpd	%xmm13, %xmm2 ;\

	movapd	12 * SIZE + 2 * (xx) * SIZE(BO), %xmm13 ;\

	addpd	%xmm12, %xmm3 ;\

	movapd	12 * SIZE + 1 * (xx) * SIZE(AO), %xmm12

#define KERNEL4(xx) \

	mulpd	%xmm14, %xmm15 ;\

	addpd	%xmm15, %xmm4 ;\

	movapd	24 * SIZE + 2 * (xx) * SIZE(BO), %xmm15 ;\

	mulpd	%xmm14, %xmm11 ;\

	addpd	%xmm11, %xmm5 ;\

	movapd	18 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\

	mulpd	%xmm14, %xmm13 ;\

	mulpd	14 * SIZE + 2 * (xx) * SIZE(BO), %xmm14 ;\

	addpd	%xmm13, %xmm6 ;\

	movapd	20 * SIZE + 2 * (xx) * SIZE(BO), %xmm13 ;\

	addpd	%xmm14, %xmm7 ;\

	movapd	14 * SIZE + 1 * (xx) * SIZE(AO), %xmm14

#define KERNEL5(xx) \

	mulpd	%xmm8, %xmm9 ;\

	addpd	%xmm9, %xmm0 ;\

	movapd	16 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\

	mulpd	%xmm8, %xmm11 ;\

	PREFETCH	(PREFETCHSIZE     +  8) * SIZE + 1 * (xx) * SIZE(AO) ;\

	addpd	%xmm11, %xmm1 ;\

	movapd	18 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\

	mulpd	%xmm8, %xmm13 ;\

	mulpd	22 * SIZE + 2 * (xx) * SIZE(BO), %xmm8 ;\

	addpd	%xmm13, %xmm2 ;\

	movapd	20 * SIZE + 2 * (xx) * SIZE(BO), %xmm13 ;\

	addpd	%xmm8, %xmm3 ;\

	movapd	16 * SIZE + 1 * (xx) * SIZE(AO), %xmm8

#define KERNEL6(xx) \

	mulpd	%xmm10, %xmm9 ;\

	addpd	%xmm9, %xmm4 ;\

	movapd	32 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\

	mulpd	%xmm10, %xmm11 ;\

	addpd	%xmm11, %xmm5 ;\

	movapd	26 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\

	mulpd	%xmm10, %xmm13 ;\

	mulpd	22 * SIZE + 2 * (xx) * SIZE(BO), %xmm10 ;\

	addpd	%xmm13, %xmm6 ;\

	movapd	28 * SIZE + 2 * (xx) * SIZE(BO), %xmm13 ;\

	addpd	%xmm10, %xmm7 ;\

	movapd	18 * SIZE + 1 * (xx) * SIZE(AO), %xmm10

#define KERNEL7(xx) \

	mulpd	%xmm12, %xmm15 ;\

	addpd	%xmm15, %xmm0 ;\

	movapd	24 * SIZE + 2 * (xx) * SIZE(BO), %xmm15 ;\

	mulpd	%xmm12, %xmm11 ;\

	addpd	%xmm11, %xmm1 ;\

	movapd	26 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\

	mulpd	%xmm12, %xmm13 ;\

	mulpd	30 * SIZE + 2 * (xx) * SIZE(BO), %xmm12 ;\

	addpd	%xmm13, %xmm2 ;\

	movapd	28 * SIZE + 2 * (xx) * SIZE(BO), %xmm13 ;\

	addpd	%xmm12, %xmm3 ;\

	movapd	20 * SIZE + 1 * (xx) * SIZE(AO), %xmm12

#define KERNEL8(xx) \

	mulpd	%xmm14, %xmm15 ;\

	addpd	%xmm15, %xmm4 ;\

	movapd	40 * SIZE + 2 * (xx) * SIZE(BO), %xmm15 ;\

	mulpd	%xmm14, %xmm11 ;\

	addpd	%xmm11, %xmm5 ;\

	movapd	34 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\

	mulpd	%xmm14, %xmm13 ;\

	mulpd	30 * SIZE + 2 * (xx) * SIZE(BO), %xmm14 ;\

	addpd	%xmm13, %xmm6 ;\

	movapd	36 * SIZE + 2 * (xx) * SIZE(BO), %xmm13 ;\

	addpd	%xmm14, %xmm7 ;\

	movapd	22 * SIZE + 1 * (xx) * SIZE(AO), %xmm14

	PROLOGUE

	PROFCODE

	subq	$STACKSIZE, %rsp

	movq	%rbx,  0(%rsp)

	movq	%rbp,  8(%rsp)

	movq	%r12, 16(%rsp)

	movq	%r13, 24(%rsp)

	movq	%r14, 32(%rsp)

	movq	%r15, 40(%rsp)

#ifdef WINDOWS_ABI

	movq	%rdi,    48(%rsp)

	movq	%rsi,    56(%rsp)

	movups	%xmm6,   64(%rsp)

	movups	%xmm7,   80(%rsp)

	movups	%xmm8,   96(%rsp)

	movups	%xmm9,  112(%rsp)

	movups	%xmm10, 128(%rsp)

	movups	%xmm11, 144(%rsp)

	movups	%xmm12, 160(%rsp)

	movups	%xmm13, 176(%rsp)

	movups	%xmm14, 192(%rsp)

	movups	%xmm15, 208(%rsp)

	movq	ARG1,      M

	movq	ARG2,      N

	movq	ARG3,      K

	movq	OLD_A,     A

	movq	OLD_B,     B

	movq	OLD_C,     C

	movq	OLD_LDC,   LDC

	movsd	OLD_OFFSET, %xmm4

	movaps	%xmm3, %xmm0

#else

	movq	OLD_LDC,   LDC

	movsd	OLD_OFFSET, %xmm4

#endif

	movq	%rsp, %rbx	# save old stack

	subq	$128 + LOCAL_BUFFER_SIZE, %rsp

	andq	$-4096, %rsp	# align stack

	STACK_TOUCHING

	movsd	%xmm4, OFFSET

	movsd	%xmm4, KK

	leaq	(, LDC, SIZE), LDC

#ifdef LN

       leaq	(, M, SIZE), %rax

       addq	%rax, C

       imulq	K, %rax

       addq	%rax, A

#endif

#ifdef RT

       leaq	(, N, SIZE), %rax

       imulq	K, %rax

       addq	%rax, B

       movq	N, %rax

       imulq	LDC, %rax

       addq	%rax, C

#endif

#ifdef RN

	negq	KK

#endif	

#ifdef RT

       movq	N, %rax

       subq	OFFSET, %rax

       movq	%rax, KK

#endif

	testq	$1, N

	je	.L40

	ALIGN_4

.L81:

/* Copying to Sub Buffer */

#ifdef LN

	movq	OFFSET, %rax

	addq	M, %rax

	movq	%rax, KK

#endif	

	leaq	BUFFER, BO

#ifdef RT

       movq	K, %rax

       salq	$0 + BASE_SHIFT, %rax

       subq	%rax, B

#endif

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

	movq	KK, %rax

	movq	B, BORIG

	leaq	(, %rax, SIZE), %rax

	leaq	(B,  %rax, 1), B

	leaq	(BO, %rax, 2), BO

#endif	

#ifdef LT

	movq	OFFSET, %rax

	movq	%rax, KK

#endif

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

	movq	KK, %rax

#else

	movq	K, %rax

	subq	KK, %rax

#endif

	sarq	$3, %rax

	jle	.L83

	ALIGN_4

.L82:

	PREFETCH	 56 * SIZE(B)

	movsd	 0 * SIZE(B), %xmm0

	movsd	 1 * SIZE(B), %xmm1

	movsd	 2 * SIZE(B), %xmm2

	movsd	 3 * SIZE(B), %xmm3

	movsd	 4 * SIZE(B), %xmm4

	movsd	 5 * SIZE(B), %xmm5

	movsd	 6 * SIZE(B), %xmm6

	movsd	 7 * SIZE(B), %xmm7

	addq	$ 8 * SIZE, B

	addq	$16 * SIZE, BO

	movsd	%xmm0, -16 * SIZE(BO)

	movsd	%xmm0, -15 * SIZE(BO)

	movsd	%xmm1, -14 * SIZE(BO)

	movsd	%xmm1, -13 * SIZE(BO)

	movsd	%xmm2, -12 * SIZE(BO)

	movsd	%xmm2, -11 * SIZE(BO)

	movsd	%xmm3, -10 * SIZE(BO)

	movsd	%xmm3,  -9 * SIZE(BO)

	movsd	%xmm4,  -8 * SIZE(BO)

	movsd	%xmm4,  -7 * SIZE(BO)

	movsd	%xmm5,  -6 * SIZE(BO)

	movsd	%xmm5,  -5 * SIZE(BO)

	movsd	%xmm6,  -4 * SIZE(BO)

	movsd	%xmm6,  -3 * SIZE(BO)

	movsd	%xmm7,  -2 * SIZE(BO)

	movsd	%xmm7,  -1 * SIZE(BO)

	decq	%rax

	jne	.L82

	ALIGN_4

.L83:

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

	movq	KK, %rax

#else

	movq	K, %rax

	subq	KK, %rax

#endif

	andq	$7, %rax

	BRANCH

	jle	.L90

	ALIGN_4

.L84:

	movsd	 0 * SIZE(B), %xmm0

	movsd	%xmm0,  0 * SIZE(BO)

	movsd	%xmm0,  1 * SIZE(BO)

	addq	$1 * SIZE, B

	addq	$2 * SIZE, BO

	decq	%rax

	jne	.L84

	ALIGN_4

.L90:

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

	movq	A, AO

#else

	movq	A, AORIG

#endif

#ifdef RT

	subq	LDC, C

#endif

	movq	C, CO1			# coffset1 = c

#ifndef RT

	addq	LDC, C

#endif

	movq	M,  I

	sarq	$2, I	# i = (m >> 2)

	jle	.L100

	ALIGN_4

.L91:

#ifdef LN

       movq	K, %rax

       salq	$2 + BASE_SHIFT, %rax

       subq	%rax, AORIG

#endif

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

	movq	KK, %rax

	movq	AORIG, AO

	leaq	(, %rax, SIZE), %rax

	leaq	(AO, %rax, 4), AO

#endif

	leaq	BUFFER, BO

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

	movq	KK, %rax

	salq	$0 + BASE_SHIFT, %rax

	leaq	(BO, %rax, 2), BO

#endif	

	movapd	 0 * SIZE(AO), %xmm8

	pxor	%xmm0, %xmm0

	movapd	 0 * SIZE(BO), %xmm9

	pxor	%xmm1, %xmm1

	movapd	 8 * SIZE(AO), %xmm10

	pxor	%xmm2, %xmm2

	movapd	 8 * SIZE(BO), %xmm11

	pxor	%xmm3, %xmm3

	movapd	16 * SIZE(AO), %xmm12

	movapd	24 * SIZE(AO), %xmm14

	PREFETCHW      4 * SIZE(CO1)

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

	movq	KK, %rax

#else

	movq	K, %rax

	subq	KK, %rax

#endif

	sarq	$3, %rax

	je	.L95

	ALIGN_4

.L92:

	mulpd	%xmm9, %xmm8

	PREFETCH	(PREFETCHSIZE     +  0) * SIZE(AO)

	mulpd	 2 * SIZE(AO), %xmm9

	addpd	%xmm8, %xmm0

	movapd	 4 * SIZE(AO), %xmm8

	addpd	%xmm9, %xmm1

	movapd	 2 * SIZE(BO), %xmm9

	mulpd	%xmm9, %xmm8

	mulpd	 6 * SIZE(AO), %xmm9

	addpd	%xmm8, %xmm2

	movapd	32 * SIZE(AO), %xmm8

	PREFETCH	(PREFETCHSIZE     +  8) * SIZE(AO)

	addpd	%xmm9, %xmm3

	movapd	 4 * SIZE(BO), %xmm9

	mulpd	%xmm9, %xmm10

	mulpd	10 * SIZE(AO), %xmm9

	addpd	%xmm10, %xmm0

	movapd	12 * SIZE(AO), %xmm10

	addpd	%xmm9, %xmm1

	movapd	 6 * SIZE(BO), %xmm9

	mulpd	%xmm9, %xmm10

	mulpd	14 * SIZE(AO), %xmm9

	addpd	%xmm10, %xmm2

	movapd	40 * SIZE(AO), %xmm10

	PREFETCH	(PREFETCHSIZE     + 16) * SIZE(AO)

	addpd	%xmm9, %xmm3

	movapd	16 * SIZE(BO), %xmm9

	mulpd	%xmm11, %xmm12

	mulpd	18 * SIZE(AO), %xmm11

	addpd	%xmm12, %xmm0

	movapd	20 * SIZE(AO), %xmm12

	addpd	%xmm11, %xmm1

	movapd	10 * SIZE(BO), %xmm11

	mulpd	%xmm11, %xmm12

	mulpd	22 * SIZE(AO), %xmm11

	addpd	%xmm12, %xmm2

	movapd	48 * SIZE(AO), %xmm12

	PREFETCH	(PREFETCHSIZE     + 24) * SIZE(AO)

	addpd	%xmm11, %xmm3

	movapd	12 * SIZE(BO), %xmm11

	mulpd	%xmm11, %xmm14

	mulpd	26 * SIZE(AO), %xmm11

	addpd	%xmm14, %xmm0

	movapd	28 * SIZE(AO), %xmm14

	addpd	%xmm11, %xmm1

	movapd	14 * SIZE(BO), %xmm11

	mulpd	%xmm11, %xmm14

	mulpd	30 * SIZE(AO), %xmm11

	addpd	%xmm14, %xmm2

	movapd	56 * SIZE(AO), %xmm14

	addpd	%xmm11, %xmm3

	movapd	24 * SIZE(BO), %xmm11

	addq   $32 * SIZE, AO

	addq   $16 * SIZE, BO

	decq   %rax

	jne    .L92

	ALIGN_4

.L95:

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

	movq	KK, %rax

#else

	movq	K, %rax

	subq	KK, %rax

#endif

	andq	$7, %rax		# if (k & 1)

	BRANCH

	je .L99

	ALIGN_4

.L96:

	mulpd	%xmm9, %xmm8

	mulpd	 2 * SIZE(AO), %xmm9

	addpd	%xmm8, %xmm0

	movapd	 4 * SIZE(AO), %xmm8

	addpd	%xmm9, %xmm1

	movapd	 2 * SIZE(BO), %xmm9

	addq	$4 * SIZE, AO		# aoffset  += 4

	addq	$2 * SIZE, BO		# boffset1 += 8

	decq	%rax

	jg	.L96

	ALIGN_4

.L99:

	addpd	%xmm2, %xmm0

	addpd	%xmm3, %xmm1

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

	movq	KK, %rax

#ifdef LN

	subq	$4, %rax

#else

	subq	$1, %rax

#endif

	movq	AORIG, AO

	movq	BORIG, B

	leaq	BUFFER, BO

	leaq	(, %rax, SIZE), %rax

	leaq	(AO, %rax, 4), AO

	leaq	(B,  %rax, 1), B

	leaq	(BO, %rax, 2), BO

#endif

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

	movapd	 0 * SIZE(B), %xmm2

	movapd	 2 * SIZE(B), %xmm3

	subpd	%xmm0,  %xmm2

	subpd	%xmm1,  %xmm3

#else

	movapd	 0 * SIZE(AO), %xmm2

	movapd	 2 * SIZE(AO), %xmm3

	subpd	%xmm0, %xmm2

	subpd	%xmm1, %xmm3

#endif

#ifdef LN

	movapd	%xmm2, %xmm0

        unpckhpd %xmm0, %xmm0

	movapd	%xmm3, %xmm1

        unpckhpd %xmm1, %xmm1

	movsd	15 * SIZE(AO), %xmm4

	mulsd	 %xmm4, %xmm1

	movsd	14 * SIZE(AO), %xmm5

	mulsd	 %xmm1, %xmm5

	subsd	 %xmm5, %xmm3

	movsd	13 * SIZE(AO), %xmm6

	mulsd	 %xmm1, %xmm6

	subsd	 %xmm6, %xmm0

	movsd	12 * SIZE(AO), %xmm7

	mulsd	 %xmm1, %xmm7

	subsd	 %xmm7, %xmm2

	movsd	10 * SIZE(AO), %xmm4

	mulsd	 %xmm4, %xmm3

	movsd	 9 * SIZE(AO), %xmm5

	mulsd	 %xmm3, %xmm5

	subsd	 %xmm5, %xmm0

	movsd	 8 * SIZE(AO), %xmm6

	mulsd	 %xmm3, %xmm6

	subsd	 %xmm6, %xmm2

	movsd	 5 * SIZE(AO), %xmm4

	mulsd	 %xmm4, %xmm0

	movsd	 4 * SIZE(AO), %xmm5

	mulsd	 %xmm0, %xmm5

	subsd	 %xmm5, %xmm2

	movsd	 0 * SIZE(AO), %xmm4

	mulsd	 %xmm4, %xmm2

	unpcklpd %xmm0, %xmm2

	unpcklpd %xmm1, %xmm3

#endif

#ifdef LT

	movapd	%xmm2, %xmm0

        unpckhpd %xmm0, %xmm0

	movapd	%xmm3, %xmm1

        unpckhpd %xmm1, %xmm1

	movsd	 0 * SIZE(AO), %xmm4

	mulsd	 %xmm4, %xmm2

	movsd	 1 * SIZE(AO), %xmm5

	mulsd	 %xmm2, %xmm5

	subsd	 %xmm5, %xmm0

	movsd	 2 * SIZE(AO), %xmm6

	mulsd	 %xmm2, %xmm6

	subsd	 %xmm6, %xmm3

	movsd	 3 * SIZE(AO), %xmm7

	mulsd	 %xmm2, %xmm7

	subsd	 %xmm7, %xmm1

	movsd	 5 * SIZE(AO), %xmm4

	mulsd	 %xmm4, %xmm0

	movsd	 6 * SIZE(AO), %xmm5

	mulsd	 %xmm0, %xmm5

	subsd	 %xmm5, %xmm3

	movsd	 7 * SIZE(AO), %xmm6

	mulsd	 %xmm0, %xmm6

	subsd	 %xmm6, %xmm1

	movsd	10 * SIZE(AO), %xmm4

	mulsd	 %xmm4, %xmm3

	movsd	11 * SIZE(AO), %xmm5

	mulsd	 %xmm3, %xmm5

	subsd	 %xmm5, %xmm1

	movsd	15 * SIZE(AO), %xmm4

	mulsd	 %xmm4, %xmm1

	unpcklpd %xmm0, %xmm2

	unpcklpd %xmm1, %xmm3

#endif

#ifdef RN

	movlpd	 0 * SIZE(B), %xmm0

	movhpd	 0 * SIZE(B), %xmm0

	mulpd	 %xmm0, %xmm2

	mulpd	 %xmm0, %xmm3

#endif

#ifdef RT

	movlpd	 0 * SIZE(B), %xmm0

	movhpd	 0 * SIZE(B), %xmm0

	mulpd	 %xmm0, %xmm2

	mulpd	 %xmm0, %xmm3

#endif

#ifdef LN

	subq	$4 * SIZE, CO1

#endif

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

	movsd	%xmm2,  0 * SIZE(CO1)

	movhpd	%xmm2,  1 * SIZE(CO1)

	movsd	%xmm3,  2 * SIZE(CO1)

	movhpd	%xmm3,  3 * SIZE(CO1)

#else

	movsd	%xmm2,  0 * SIZE(CO1)

	movhpd	%xmm2,  1 * SIZE(CO1)

	movsd	%xmm3,  2 * SIZE(CO1)

	movhpd	%xmm3,  3 * SIZE(CO1)

#endif

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

	movapd	%xmm2,   0 * SIZE(B)

	movapd	%xmm3,   2 * SIZE(B)

	movlpd	%xmm2,   0 * SIZE(BO)

	movlpd	%xmm2,   1 * SIZE(BO)

	movhpd	%xmm2,   2 * SIZE(BO)

	movhpd	%xmm2,   3 * SIZE(BO)

	movlpd	%xmm3,   4 * SIZE(BO)

	movlpd	%xmm3,   5 * SIZE(BO)

	movhpd	%xmm3,   6 * SIZE(BO)

	movhpd	%xmm3,   7 * SIZE(BO)

#else

	movapd	%xmm2,   0 * SIZE(AO)

	movapd	%xmm3,   2 * SIZE(AO)

#endif

#ifndef LN

	addq	$4 * SIZE, CO1

#endif

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

	movq	K,  %rax

	subq	KK, %rax

	leaq	(,%rax, SIZE), %rax

	leaq	(AO, %rax, 4), AO

#ifdef LT

	addq	$4 * SIZE, B

#endif

#endif

#ifdef LN

	subq	$4, KK

	movq	BORIG, B

#endif

#ifdef LT

	addq	$4, KK

#endif

#ifdef RT

       movq	K, %rax

       movq	BORIG, B

       salq	$2 + BASE_SHIFT, %rax

       addq	%rax, AORIG

#endif

	decq	I			# i --

	jg	.L91

	ALIGN_4	

.L100:

	testq	$2, M

	je	.L110

	ALIGN_4

.L101:

#ifdef LN

       movq	K, %rax

       salq	$1 + BASE_SHIFT, %rax

       subq	%rax, AORIG

#endif

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

	movq	KK, %rax

	movq	AORIG, AO

	leaq	(, %rax, SIZE), %rax

	leaq	(AO, %rax, 2), AO

#endif

	leaq	BUFFER, BO

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

	movq	KK, %rax

	salq	$0 + BASE_SHIFT, %rax

	leaq	(BO, %rax, 2), BO

#endif	

	movapd	 0 * SIZE(AO), %xmm8

	pxor	%xmm0, %xmm0

	movapd	 0 * SIZE(BO), %xmm9

	pxor	%xmm1, %xmm1

	movapd	 8 * SIZE(AO), %xmm10

	pxor	%xmm2, %xmm2

	movapd	 8 * SIZE(BO), %xmm11

	pxor	%xmm3, %xmm3

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

	movq	KK, %rax

#else

	movq	K, %rax

	subq	KK, %rax

#endif

	sarq	$3, %rax

	je	.L105

	ALIGN_4

.L102:

	mulpd	%xmm8, %xmm9

	PREFETCH	(PREFETCHSIZE     +  0) * SIZE(AO)

	movapd	 2 * SIZE(AO), %xmm8

	mulpd	 2 * SIZE(BO), %xmm8

	addpd	%xmm9, %xmm0

	movapd	16 * SIZE(BO), %xmm9

	addpd	%xmm8, %xmm1

	movapd	 4 * SIZE(AO), %xmm8

	mulpd	 4 * SIZE(BO), %xmm8

	addpd	%xmm8, %xmm2

	movapd	 6 * SIZE(AO), %xmm8

	mulpd	 6 * SIZE(BO), %xmm8

	addpd	%xmm8, %xmm3

	movapd	16 * SIZE(AO), %xmm8

	PREFETCH	(PREFETCHSIZE     +  0) * SIZE(AO)

	mulpd	%xmm10, %xmm11

	movapd	10 * SIZE(AO), %xmm10

	mulpd	10 * SIZE(BO), %xmm10

	addpd	%xmm11, %xmm0

	movapd	24 * SIZE(BO), %xmm11

	addpd	%xmm10, %xmm1

	movapd	12 * SIZE(AO), %xmm10

	mulpd	12 * SIZE(BO), %xmm10

	addpd	%xmm10, %xmm2

	movapd	14 * SIZE(AO), %xmm10

	mulpd	14 * SIZE(BO), %xmm10

	addpd	%xmm10, %xmm3

	movapd	24 * SIZE(AO), %xmm10

	addq   $16 * SIZE, AO

	addq   $16 * SIZE, BO

	decq   %rax

	jne    .L102

	ALIGN_4

.L105:

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

	movq	KK, %rax

#else

	movq	K, %rax

	subq	KK, %rax

#endif

	andq	$7, %rax		# if (k & 1)

	BRANCH

	je .L109

	ALIGN_4

.L106:

	mulpd	%xmm8, %xmm9

	addpd	%xmm9, %xmm0

	movapd	 2 * SIZE(AO), %xmm8

	movapd	 2 * SIZE(BO), %xmm9

	addq	$2 * SIZE, AO		# aoffset  += 4

	addq	$2 * SIZE, BO		# boffset1 += 8

	decq	%rax

	jg	.L106

	ALIGN_4

.L109:

	addpd	%xmm1, %xmm0

	addpd	%xmm3, %xmm2

	addpd	%xmm2, %xmm0

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

	movq	KK, %rax

#ifdef LN

	subq	$2, %rax

#else

	subq	$1, %rax

#endif