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

/* 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	16 + STACK(%esi)

#define OLD_A	20 + STACK(%esi)

#define OLD_B	24 + STACK(%esi)

#define OLD_C	28 + STACK(%esi)

#define OLD_LDC	32 + STACK(%esi)

#define STACK_OFFT	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 OLD_STACK 40(%esp)

#define OFFSET  44(%esp)

#define KK	48(%esp)

#define KKK	52(%esp)

#define BUFFER 128(%esp)

#define PREFETCH     prefetch

#define PREFETCHSIZE  (16 * 17 + 0)

#define RPREFETCHSIZE (16 *  9 + 0)

#define WPREFETCHSIZE (16 *  9 + 0)

#define AA	%edx

#define	BB	%ecx

#define LDC	%ebp

#define KERNEL1(address) \

	mulps	%xmm0, %xmm2; \

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

	addps	%xmm2, %xmm4; \

	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;

	PROLOGUE

	pushl	%ebp

	pushl	%edi

	pushl	%esi

	pushl	%ebx

	PROFCODE

	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,  %xmm3

#ifdef TRMMKERNEL

	movss	STACK_OFFT, %xmm4

#endif

	movl	%ebx, M

	movl	%eax, N

	movl	%ecx, K

	movl	%edx, A

	movl	%esi, OLD_STACK

	shufps	$0, %xmm3, %xmm3

	movl	OLD_B, %edi

	movl	OLD_C, %ebx

	movaps	 %xmm3, ALPHA

	movl	%ebx, C

	movl	OLD_LDC, LDC

#ifdef TRMMKERNEL

	movss	%xmm4, OFFSET

	movss	%xmm4, KK

#ifndef LEFT

	negl	KK

#endif	

#endif

	leal	(, LDC, SIZE), 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:

	prefetch	 (RPREFETCHSIZE +  0)  * SIZE(%edi)

	movaps	 0 * SIZE(%edi), %xmm3

	movaps	 4 * SIZE(%edi), %xmm7

	prefetchw	 (WPREFETCHSIZE +  0)  * SIZE(%ecx)

	pshufd	 $0x00, %xmm3, %xmm0

	pshufd	 $0x55, %xmm3, %xmm1

	pshufd	 $0xaa, %xmm3, %xmm2

	pshufd	 $0xff, %xmm3, %xmm3

	prefetchw	 (WPREFETCHSIZE + 16)  * SIZE(%ecx)

	pshufd	 $0x00, %xmm7, %xmm4

	pshufd	 $0x55, %xmm7, %xmm5

	pshufd	 $0xaa, %xmm7, %xmm6

	pshufd	 $0xff, %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)

	addl	$  8 * SIZE, %edi

	subl	$-32 * SIZE, %ecx

	decl	%eax

	jne	.L02

	ALIGN_2

.L05:

	movl	K, %eax

	andl	$1, %eax

	BRANCH

	jle	.L10

	movaps	 0 * SIZE(%edi), %xmm3

	pshufd	 $0x00, %xmm3, %xmm0

	pshufd	 $0x55, %xmm3, %xmm1

	pshufd	 $0xaa, %xmm3, %xmm2

	pshufd	 $0xff, %xmm3, %xmm3

	movaps	%xmm0,  0 * SIZE(%ecx)

	movaps	%xmm1,  4 * SIZE(%ecx)

	movaps	%xmm2,  8 * SIZE(%ecx)

	movaps	%xmm3, 12 * SIZE(%ecx)

	addl	$4 * SIZE, %edi

	ALIGN_4

.L10:

	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	

	movaps	 0 * SIZE(AA), %xmm0

	xorps	%xmm4, %xmm4

	movaps	16 * SIZE(AA), %xmm1

	xorps	%xmm5, %xmm5

	movaps	 0 * SIZE(BB), %xmm2

	xorps	%xmm6, %xmm6

	movaps	16 * SIZE(BB), %xmm3

	xorps	%xmm7, %xmm7

	leal	(%esi, LDC, 2), %eax

	prefetchw	3 * SIZE(%esi)

	prefetchw	3 * SIZE(%esi, LDC)

	prefetchw	3 * SIZE(%eax)

	prefetchw	3 * SIZE(%eax, LDC)

#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

	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

.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

#ifndef TRMMKERNEL

	shufps	$0xe4, %xmm0, %xmm0

	shufps	$0xe4, %xmm1, %xmm1

	shufps	$0xe4, %xmm2, %xmm2

	shufps	$0xe4, %xmm3, %xmm3

	mulps	%xmm3, %xmm4

	movsd	0 * SIZE(%esi), %xmm0

	movhps	2 * SIZE(%esi), %xmm0

	mulps	%xmm3, %xmm5

	movsd	0 * SIZE(%esi, LDC, 1), %xmm1

	movhps	2 * SIZE(%esi, LDC, 1), %xmm1

	mulps	%xmm3, %xmm6

	movsd	0 * SIZE(%esi, LDC, 2), %xmm2

	movhps	2 * SIZE(%esi, LDC, 2), %xmm2

	mulps	%xmm3, %xmm7

	movsd	0 * SIZE(%esi, %eax, 1), %xmm3

	movhps	2 * SIZE(%esi, %eax, 1), %xmm3

	addps	%xmm0, %xmm4

	addps	%xmm1, %xmm5

	addps	%xmm2, %xmm6

	addps	%xmm3, %xmm7

#else

	mulps	%xmm3, %xmm4

	mulps	%xmm3, %xmm5

	mulps	%xmm3, %xmm6

	mulps	%xmm3, %xmm7

#endif

	movsd	%xmm4, 0 * SIZE(%esi)

	movhps	%xmm4, 2 * SIZE(%esi)

	movsd	%xmm5, 0 * SIZE(%esi, LDC, 1)

	movhps	%xmm5, 2 * SIZE(%esi, LDC, 1)

	movsd	%xmm6, 0 * SIZE(%esi, LDC, 2)

	movhps	%xmm6, 2 * SIZE(%esi, LDC, 2)

	movsd	%xmm7, 0 * SIZE(%esi, %eax, 1)

	movhps	%xmm7, 2 * SIZE(%esi, %eax, 1)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \

    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	movl	K, %eax

	subl	KKK, %eax

	leal	(,%eax, 8), %eax

	leal	(AA, %eax, 2), AA

	leal	(BB, %eax, 8), BB

#endif

#if defined(TRMMKERNEL) && defined(LEFT)

	addl	$4, KK

#endif

	addl	$4 * 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

	xorps	%xmm4, %xmm4

	movsd	 8 * SIZE(AA), %xmm1

	xorps	%xmm5, %xmm5

	movsd	 0 * SIZE(BB), %xmm2

	xorps	%xmm6, %xmm6

	movsd	16 * SIZE(BB), %xmm3

	xorps	%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

#ifndef TRMMKERNEL

	mulps	%xmm3, %xmm4

	movsd	0 * SIZE(%esi), %xmm0

	mulps	%xmm3, %xmm5

	movsd	0 * SIZE(%esi, LDC, 1), %xmm1

	mulps	%xmm3, %xmm6

	movsd	0 * SIZE(%esi, LDC, 2), %xmm2

	mulps	%xmm3, %xmm7

	movsd	0 * SIZE(%esi, %eax, 1), %xmm3

	addps	%xmm0, %xmm4

	addps	%xmm1, %xmm5

	addps	%xmm2, %xmm6

	addps	%xmm3, %xmm7

#else

	mulps	%xmm3, %xmm4

	mulps	%xmm3, %xmm5

	mulps	%xmm3, %xmm6

	mulps	%xmm3, %xmm7

#endif

	movsd	%xmm4, 0 * SIZE(%esi)

	movsd	%xmm5, 0 * SIZE(%esi, LDC, 1)

	movsd	%xmm6, 0 * SIZE(%esi, LDC, 2)

	movsd	%xmm7, 0 * SIZE(%esi, %eax, 1)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \

    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	movl	K, %eax

	subl	KKK, %eax

	leal	(,%eax, 8), %eax

	leal	(AA, %eax, 1), AA

	leal	(BB, %eax, 8), BB

#endif

#if defined(TRMMKERNEL) && defined(LEFT)

	addl	$2, KK

#endif

	addl	$2 * 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

	xorps	%xmm4, %xmm4

	movss	 4 * SIZE(AA), %xmm1

	xorps	%xmm5, %xmm5

	movss	 0 * SIZE(BB), %xmm2

	xorps	%xmm6, %xmm6

	movss	16 * SIZE(BB), %xmm3

	xorps	%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