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

/* 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	%rbx

#define KK	%rbp

#ifndef WINDOWS_ABI

#define STACKSIZE 128

#define OLD_LDC		 8 + STACKSIZE(%rsp)

#define OLD_OFFSET	16 + STACKSIZE(%rsp)

#define OFFSET	 48(%rsp)

#define KKK	 56(%rsp)

#define AORIG    64(%rsp)

#else

#define STACKSIZE 256

#define OLD_A		48 + STACKSIZE(%rsp)

#define OLD_B		56 + STACKSIZE(%rsp)

#define OLD_C		64 + STACKSIZE(%rsp)

#define OLD_LDC		72 + STACKSIZE(%rsp)

#define OLD_OFFSET	80 + STACKSIZE(%rsp)

#define OFFSET	 224(%rsp)

#define KKK	 232(%rsp)

#define AORIG    240(%rsp)

#endif

#define PREFETCH     prefetcht1

#define PREFETCHSIZE (16 * 12 + 3)

#define PREFETCH_R    (4 *  4 + 0)

#ifndef CONJ

#define ADD1	  addpd

#define ADD2	  addpd

#else

#define ADD1	  subpd

#define ADD2	  addpd

#endif

#define KERNEL1(address) \

	mulpd	%xmm8, %xmm9;\

	PREFETCH  (PREFETCHSIZE +  0) * SIZE + (address) * 2 * SIZE(AO);\

	ADD1	%xmm9, %xmm0;\

	movddup	 1 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	ADD2	%xmm9, %xmm1;\

	movddup	 2 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	ADD1	%xmm9, %xmm2;\

	movddup	 3 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	movapd	 2 * SIZE + (address) * 2 * SIZE(AO), %xmm8;\

	ADD2	%xmm9, %xmm3;\

	movddup	 0 * SIZE + (address) * 2 * SIZE(BO), %xmm9

#define KERNEL2(address) \

	mulpd	%xmm8, %xmm9;\

	ADD1	%xmm9, %xmm4;\

	movddup	 1 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	ADD2	%xmm9, %xmm5;\

	movddup	 2 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	ADD1	%xmm9, %xmm6;\

	movddup	 3 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	movapd	 4 * SIZE + (address) * 2 * SIZE(AO), %xmm8;\

	ADD2	%xmm9, %xmm7;\

	movddup	 4 * SIZE + (address) * 2 * SIZE(BO), %xmm9

#define KERNEL3(address) \

	mulpd	%xmm8, %xmm9;\

	ADD1	%xmm9, %xmm0;\

	movddup	 5 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	ADD2	%xmm9, %xmm1;\

	movddup	 6 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	ADD1	%xmm9, %xmm2;\

	movddup	 7 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	movapd	 6 * SIZE + (address) * 2 * SIZE(AO), %xmm8;\

	ADD2	%xmm9, %xmm3;\

	movddup	 4 * SIZE + (address) * 2 * SIZE(BO), %xmm9

#define KERNEL4(address) \

	mulpd	%xmm8, %xmm9;\

	ADD1	%xmm9, %xmm4;\

	movddup	 5 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	ADD2	%xmm9, %xmm5;\

	movddup	 6 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	ADD1	%xmm9, %xmm6;\

	movddup	 7 * SIZE + (address) * 2 * SIZE(BO), %xmm9;\

	mulpd	%xmm8, %xmm9;\

	movapd	32 * SIZE + (address) * 2 * SIZE(AO), %xmm8;\

	ADD2	%xmm9, %xmm7;\

	movddup	32 * SIZE + (address) * 2 * SIZE(BO), %xmm9

#define KERNEL5(address) \

	mulpd	%xmm10, %xmm11;\

	ADD1	%xmm11, %xmm0;\

	movddup	 9 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	ADD2	%xmm11, %xmm1;\

	movddup	10 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	ADD1	%xmm11, %xmm2;\

	movddup	11 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	movapd	10 * SIZE + (address) * 2 * SIZE(AO), %xmm10;\

	ADD2	%xmm11, %xmm3;\

	movddup	 8 * SIZE + (address) * 2 * SIZE(BO), %xmm11

#define KERNEL6(address) \

	mulpd	%xmm10, %xmm11;\

	ADD1	%xmm11, %xmm4;\

	movddup	 9 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	ADD2	%xmm11, %xmm5;\

	movddup	10 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	ADD1	%xmm11, %xmm6;\

	movddup	11 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	movapd	12 * SIZE + (address) * 2 * SIZE(AO), %xmm10;\

	ADD2	%xmm11, %xmm7;\

	movddup	12 * SIZE + (address) * 2 * SIZE(BO), %xmm11

#define KERNEL7(address) \

	mulpd	%xmm10, %xmm11;\

	ADD1	%xmm11, %xmm0;\

	movddup	13 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	ADD2	%xmm11, %xmm1;\

	movddup	14 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	ADD1	%xmm11, %xmm2;\

	movddup	15 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	movapd	14 * SIZE + (address) * 2 * SIZE(AO), %xmm10;\

	ADD2	%xmm11, %xmm3;\

	movddup	12 * SIZE + (address) * 2 * SIZE(BO), %xmm11

#define KERNEL8(address) \

	mulpd	%xmm10, %xmm11;\

	ADD1	%xmm11, %xmm4;\

	movddup	13 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	ADD2	%xmm11, %xmm5;\

	movddup	14 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	ADD1	%xmm11, %xmm6;\

	movddup	15 * SIZE + (address) * 2 * SIZE(BO), %xmm11;\

	mulpd	%xmm10, %xmm11;\

	movapd	40 * SIZE + (address) * 2 * SIZE(AO), %xmm10;\

	ADD2	%xmm11, %xmm7;\

	movddup	40 * SIZE + (address) * 2 * SIZE(BO), %xmm11

#define KERNEL9(address) \

	mulpd	%xmm12, %xmm13;\

	PREFETCH  (PREFETCHSIZE + 16) * SIZE + (address) * 2 * SIZE(AO);\

	ADD1	%xmm13, %xmm0;\

	movddup	17 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	ADD2	%xmm13, %xmm1;\

	movddup	18 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	ADD1	%xmm13, %xmm2;\

	movddup	19 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	movapd	18 * SIZE + (address) * 2 * SIZE(AO), %xmm12;\

	ADD2	%xmm13, %xmm3;\

	movddup	16 * SIZE + (address) * 2 * SIZE(BO), %xmm13

#define KERNEL10(address) \

	mulpd	%xmm12, %xmm13;\

	ADD1	%xmm13, %xmm4;\

	movddup	17 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	ADD2	%xmm13, %xmm5;\

	movddup	18 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	ADD1	%xmm13, %xmm6;\

	movddup	19 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	movapd	20 * SIZE + (address) * 2 * SIZE(AO), %xmm12;\

	ADD2	%xmm13, %xmm7;\

	movddup	20 * SIZE + (address) * 2 * SIZE(BO), %xmm13

#define KERNEL11(address) \

	mulpd	%xmm12, %xmm13;\

	ADD1	%xmm13, %xmm0;\

	movddup	21 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	ADD2	%xmm13, %xmm1;\

	movddup	22 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	ADD1	%xmm13, %xmm2;\

	movddup	23 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	movapd	22 * SIZE + (address) * 2 * SIZE(AO), %xmm12;\

	ADD2	%xmm13, %xmm3;\

	movddup	20 * SIZE + (address) * 2 * SIZE(BO), %xmm13

#define KERNEL12(address) \

	mulpd	%xmm12, %xmm13;\

	ADD1	%xmm13, %xmm4;\

	movddup 21 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	ADD2	%xmm13, %xmm5;\

	movddup	22 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	ADD1	%xmm13, %xmm6;\

	movddup	23 * SIZE + (address) * 2 * SIZE(BO), %xmm13;\

	mulpd	%xmm12, %xmm13;\

	movapd	48 * SIZE + (address) * 2 * SIZE(AO), %xmm12;\

	ADD2	%xmm13, %xmm7;\

	movddup	48 * SIZE + (address) * 2 * SIZE(BO), %xmm13

#define KERNEL13(address) \

	mulpd	%xmm14, %xmm15;\

	ADD1	%xmm15, %xmm0;\

	movddup	25 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	ADD2	%xmm15, %xmm1;\

	movddup	26 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	ADD1	%xmm15, %xmm2;\

	movddup	27 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	movapd	26 * SIZE + (address) * 2 * SIZE(AO), %xmm14;\

	ADD2	%xmm15, %xmm3;\

	movddup	24 * SIZE + (address) * 2 * SIZE(BO), %xmm15

#define KERNEL14(address) \

	mulpd	%xmm14, %xmm15;\

	ADD1	%xmm15, %xmm4;\

	movddup	25 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	ADD2	%xmm15, %xmm5;\

	movddup	26 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	ADD1	%xmm15, %xmm6;\

	movddup	27 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	movapd	28 * SIZE + (address) * 2 * SIZE(AO), %xmm14;\

	ADD2	%xmm15, %xmm7;\

	movddup	28 * SIZE + (address) * 2 * SIZE(BO), %xmm15

#define KERNEL15(address) \

	mulpd	%xmm14, %xmm15;\

	ADD1	%xmm15, %xmm0;\

	movddup	29 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	ADD2	%xmm15, %xmm1;\

	movddup	30 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	ADD1	%xmm15, %xmm2;\

	movddup	31 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	movapd	30 * SIZE + (address) * 2 * SIZE(AO), %xmm14;\

	ADD2	%xmm15, %xmm3;\

	movddup	28 * SIZE + (address) * 2 * SIZE(BO), %xmm15

#define KERNEL16(address) \

	mulpd	%xmm14, %xmm15;\

	ADD1	%xmm15, %xmm4;\

	movddup	29 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	ADD2	%xmm15, %xmm5;\

	movddup	30 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	ADD1	%xmm15, %xmm6;\

	movddup	31 * SIZE + (address) * 2 * SIZE(BO), %xmm15;\

	mulpd	%xmm14, %xmm15;\

	movapd	56 * SIZE + (address) * 2 * SIZE(AO), %xmm14;\

	ADD2	%xmm15, %xmm7;\

	movddup	56 * SIZE + (address) * 2 * SIZE(BO), %xmm15

	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

#endif

	movq	OLD_LDC,    LDC

	movq	OLD_OFFSET, KK

	movq	KK, OFFSET

	salq	$ZBASE_SHIFT, LDC

#ifdef LN

       movq	M, %rax

       salq	$ZBASE_SHIFT, %rax

       addq	%rax, C

       imulq	K, %rax

       addq	%rax, A

#endif

#ifdef RT

       movq	N, %rax

       salq	$ZBASE_SHIFT, %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, KK

       subq	OFFSET, KK

#endif

	movq	N,  J

	sarq	$1, J		# j = (n >> 2)

	jle	.L100

	ALIGN_4

.L01:

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

	movq	A, AO

#else

	movq	A, AORIG

#endif

#ifdef RT

       movq	K, %rax

       salq	$1 + ZBASE_SHIFT, %rax

       subq	%rax, B

       leaq	(, LDC, 2), %rax

       subq	%rax, C

#endif

	movq	C, CO1			# coffset1 = c

	leaq	(C, LDC, 1), CO2	# coffset2 = c + ldc

#ifndef RT

	leaq	(C, LDC, 2), C

#endif

#ifdef LN

	movq	OFFSET, KK

	addq	M, KK

#endif	

#ifdef LT

	movq	OFFSET, KK

#endif

	movq	M,  I

	sarq	$1, I		# i = (m >> 2)

	jle	.L30

	ALIGN_4

.L10:

#ifdef LN

       movq	K, %rax

       salq	$1 + ZBASE_SHIFT, %rax

       subq	%rax, AORIG

#endif

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

	movq	KK, %rax

	leaq	(, %rax, SIZE), %rax

	movq	AORIG, AO

	leaq	(AO, %rax, 4), AO

	leaq	(B,  %rax, 4), BO

#else

	movq	B, BO

#endif	

	movapd	 0 * SIZE(AO), %xmm8

	pxor	%xmm0, %xmm0

	movddup	 0 * SIZE(BO), %xmm9

	pxor	%xmm1, %xmm1

	movapd	 8 * SIZE(AO), %xmm10

	pxor	%xmm2, %xmm2

	movddup	 8 * SIZE(BO), %xmm11

	pxor	%xmm3, %xmm3

	movapd	16 * SIZE(AO), %xmm12

	movddup 16 * SIZE(BO), %xmm13

	movapd	24 * SIZE(AO), %xmm14

	movddup	24 * SIZE(BO), %xmm15

	prefetchnta     4 * SIZE(CO1)

	pxor	%xmm4, %xmm4

	prefetchnta     4 * SIZE(CO2)

	pxor	%xmm5, %xmm5

	pxor	%xmm6, %xmm6

	pxor	%xmm7, %xmm7

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

	movq	KK, %rax

#else

	movq	K, %rax

	subq	KK, %rax

#endif

	andq	$-8, %rax

	salq	$4, %rax

	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)

	KERNEL9 (16  *  0)

	KERNEL10(16  *  0)

	KERNEL11(16  *  0)

	KERNEL12(16  *  0)

	KERNEL13(16  *  0)

	KERNEL14(16  *  0)

	KERNEL15(16  *  0)

	KERNEL16(16  *  0)

	cmpq	$128 *  1, %rax

	NOBRANCH

	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)

	KERNEL9 (16  *  1)

	KERNEL10(16  *  1)

	KERNEL11(16  *  1)

	KERNEL12(16  *  1)

	KERNEL13(16  *  1)

	KERNEL14(16  *  1)

	KERNEL15(16  *  1)

	KERNEL16(16  *  1)

	cmpq	$128 *  2, %rax

	NOBRANCH

	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)

	KERNEL9 (16  *  2)

	KERNEL10(16  *  2)

	KERNEL11(16  *  2)

	KERNEL12(16  *  2)

	KERNEL13(16  *  2)

	KERNEL14(16  *  2)

	KERNEL15(16  *  2)

	KERNEL16(16  *  2)

	cmpq	$128 *  3, %rax

	NOBRANCH

	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)

	KERNEL9 (16  *  3)

	KERNEL10(16  *  3)

	KERNEL11(16  *  3)

	KERNEL12(16  *  3)

	KERNEL13(16  *  3)

	KERNEL14(16  *  3)

	KERNEL15(16  *  3)

	KERNEL16(16  *  3)

	cmpq	$128 *  4, %rax

	NOBRANCH

	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)

	KERNEL9 (16  *  4)

	KERNEL10(16  *  4)

	KERNEL11(16  *  4)

	KERNEL12(16  *  4)

	KERNEL13(16  *  4)

	KERNEL14(16  *  4)

	KERNEL15(16  *  4)

	KERNEL16(16  *  4)

	cmpq	$128 *  5, %rax

	NOBRANCH

	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)

	KERNEL9 (16  *  5)

	KERNEL10(16  *  5)

	KERNEL11(16  *  5)

	KERNEL12(16  *  5)

	KERNEL13(16  *  5)

	KERNEL14(16  *  5)

	KERNEL15(16  *  5)

	KERNEL16(16  *  5)

	cmpq	$128 *  6, %rax

	NOBRANCH

	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)

	KERNEL9 (16  *  6)

	KERNEL10(16  *  6)

	KERNEL11(16  *  6)

	KERNEL12(16  *  6)

	KERNEL13(16  *  6)

	KERNEL14(16  *  6)

	KERNEL15(16  *  6)

	KERNEL16(16  *  6)

	cmpq	$128 *  7, %rax

	NOBRANCH

	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)

	KERNEL9 (16  *  7)

	KERNEL10(16  *  7)

	KERNEL11(16  *  7)

	KERNEL12(16  *  7)

	KERNEL13(16  *  7)

	KERNEL14(16  *  7)

	KERNEL15(16  *  7)

	KERNEL16(16  *  7)

	addq	$32 * 8  * SIZE, AO

	addq	$32 * 8  * SIZE, BO

	subq	$128 * 8, %rax

	jg	.L1X

.L11:

	leaq	(AO, %rax, 2), AO	# * 16

	leaq	(BO, %rax, 2), BO	# * 64

	ALIGN_4

.L12:

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

	movq	KK, %rax

#else

	movq	K, %rax

	subq	KK, %rax

#endif

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

	BRANCH

	je .L14

	ALIGN_4

.L13:

	mulpd	%xmm8, %xmm9

	movapd	 2 * SIZE(AO), %xmm10

	ADD1	%xmm9, %xmm0

	movddup	 1 * SIZE(BO), %xmm9

	mulpd	%xmm8, %xmm9

	movddup	 0 * SIZE(BO), %xmm11

	ADD2	%xmm9, %xmm1

	movddup	 2 * SIZE(BO), %xmm9

	mulpd	%xmm8, %xmm9

	ADD1	%xmm9, %xmm2

	movddup	 3 * SIZE(BO), %xmm9

	mulpd	%xmm8, %xmm9

	movapd	 4 * SIZE(AO), %xmm8

	ADD2	%xmm9, %xmm3

	movddup	 4 * SIZE(BO), %xmm9

	mulpd	%xmm10, %xmm11

	ADD1	%xmm11, %xmm4

	movddup	 1 * SIZE(BO), %xmm11

	mulpd	%xmm10, %xmm11

	ADD2	%xmm11, %xmm5

	movddup	 2 * SIZE(BO), %xmm11

	mulpd	%xmm10, %xmm11

	ADD1	%xmm11, %xmm6

	movddup	 3 * SIZE(BO), %xmm11

	mulpd	%xmm10, %xmm11

	ADD2	%xmm11, %xmm7

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

	addq	$4 * SIZE, BO		# boffset1 += 8

	decq	%rax

	jg	.L13

	ALIGN_4

.L14:

	SHUFPD_1 %xmm1, %xmm1

	SHUFPD_1 %xmm3, %xmm3

	SHUFPD_1 %xmm5, %xmm5

	SHUFPD_1 %xmm7, %xmm7

#ifndef CONJ

	addsubpd	%xmm1, %xmm0

	addsubpd	%xmm3, %xmm2

	addsubpd	%xmm5, %xmm4

	addsubpd	%xmm7, %xmm6

#else

	addsubpd	%xmm0, %xmm1

	addsubpd	%xmm2, %xmm3

	addsubpd	%xmm4, %xmm5

	addsubpd	%xmm6, %xmm7

#endif

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

	movq	KK, %rax

	subq	$2, %rax

	leaq	(, %rax, SIZE), %rax

	movq	AORIG, AO

	leaq	(AO, %rax, 4), AO

	leaq	(B,  %rax, 4), BO

#endif

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

	movapd	 0 * SIZE(BO), %xmm8

	movapd	 2 * SIZE(BO), %xmm9

	movapd	 4 * SIZE(BO), %xmm10

	movapd	 6 * SIZE(BO), %xmm11

#else

	movapd	 0 * SIZE(AO), %xmm8

	movapd	 2 * SIZE(AO), %xmm9

	movapd	 4 * SIZE(AO), %xmm10

	movapd	 6 * SIZE(AO), %xmm11

#endif

#if   (defined(LN) || defined(LT)) && !defined(CONJ)

	subpd	%xmm0,  %xmm8

	subpd	%xmm2,  %xmm9

	subpd	%xmm4,  %xmm10

	subpd	%xmm6,  %xmm11

#elif (defined(LN) || defined(LT)) &&  defined(CONJ)

	subpd	%xmm1,  %xmm8

	subpd	%xmm3,  %xmm9

	subpd	%xmm5,  %xmm10

	subpd	%xmm7,  %xmm11

#elif (defined(RN) || defined(RT)) && !defined(CONJ)

	subpd	%xmm0,  %xmm8

	subpd	%xmm4,  %xmm9

	subpd	%xmm2,  %xmm10

	subpd	%xmm6,  %xmm11

#else

	addsubpd %xmm1,  %xmm8

	addsubpd %xmm5,  %xmm9

	addsubpd %xmm3,  %xmm10

	addsubpd %xmm7,  %xmm11

#endif

#ifdef CONJ

	pcmpeqb	%xmm7,  %xmm7

	psllq	$63,    %xmm7

#endif

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

#ifdef LN

	movddup	 6 * SIZE(AO), %xmm0

	movddup	 7 * SIZE(AO), %xmm1

	movddup	 4 * SIZE(AO), %xmm2

	movddup	 5 * SIZE(AO), %xmm3

	movddup	 0 * SIZE(AO), %xmm4

	movddup	 1 * SIZE(AO), %xmm5

#else

	movddup	 6 * SIZE(BO), %xmm0

	movddup	 7 * SIZE(BO), %xmm1

	movddup	 4 * SIZE(BO), %xmm2

	movddup	 5 * SIZE(BO), %xmm3

	movddup	 0 * SIZE(BO), %xmm4

	movddup	 1 * SIZE(BO), %xmm5

#endif

#ifdef CONJ

	xorpd	%xmm7, %xmm1

	xorpd	%xmm7, %xmm3

	xorpd	%xmm7, %xmm5

#endif

	movapd	%xmm10,  %xmm12

	movapd	%xmm11,  %xmm13

	SHUFPD_1 %xmm12, %xmm12

	SHUFPD_1 %xmm13, %xmm13

	mulpd	 %xmm0, %xmm10

	mulpd	 %xmm0, %xmm11

	mulpd	 %xmm1, %xmm12

	mulpd	 %xmm1, %xmm13

	addsubpd %xmm12, %xmm10

	addsubpd %xmm13, %xmm11

	movapd	%xmm10,  %xmm12

	movapd	%xmm10,  %xmm13

	movapd	%xmm11,  %xmm14

	movapd	%xmm11,  %xmm15

	SHUFPD_1 %xmm13, %xmm13

	SHUFPD_1 %xmm15, %xmm15

	mulpd	%xmm2, %xmm12

	mulpd	%xmm2, %xmm14

	mulpd	%xmm3, %xmm13

	mulpd	%xmm3, %xmm15

	addsubpd %xmm13, %xmm12

	addsubpd %xmm15, %xmm14

	subpd	 %xmm12, %xmm8

	subpd	 %xmm14, %xmm9

	movapd	%xmm8,  %xmm12

	movapd	%xmm9,  %xmm13

	SHUFPD_1 %xmm12, %xmm12

	SHUFPD_1 %xmm13, %xmm13

	mulpd	 %xmm4, %xmm8

	mulpd	 %xmm4, %xmm9

	mulpd	 %xmm5, %xmm12

	mulpd	 %xmm5, %xmm13

	addsubpd %xmm12, %xmm8

	addsubpd %xmm13, %xmm9

#endif

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

#ifdef LT

	movddup	 0 * SIZE(AO), %xmm0

	movddup	 1 * SIZE(AO), %xmm1

	movddup	 2 * SIZE(AO), %xmm2

	movddup	 3 * SIZE(AO), %xmm3

	movddup	 6 * SIZE(AO), %xmm4

	movddup	 7 * SIZE(AO), %xmm5

#else

	movddup	 0 * SIZE(BO), %xmm0

	movddup	 1 * SIZE(BO), %xmm1

	movddup	 2 * SIZE(BO), %xmm2

	movddup	 3 * SIZE(BO), %xmm3

	movddup	 6 * SIZE(BO), %xmm4

	movddup	 7 * SIZE(BO), %xmm5

#endif

#ifdef CONJ

	xorpd	%xmm7, %xmm1

	xorpd	%xmm7, %xmm3

	xorpd	%xmm7, %xmm5

#endif

	movapd	%xmm8,  %xmm12

	movapd	%xmm9,  %xmm13

	SHUFPD_1 %xmm12, %xmm12

	SHUFPD_1 %xmm13, %xmm13

	mulpd	 %xmm0, %xmm8

	mulpd	 %xmm0, %xmm9

	mulpd	 %xmm1, %xmm12

	mulpd	 %xmm1, %xmm13

	addsubpd %xmm12, %xmm8

	addsubpd %xmm13, %xmm9