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

/* 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_R	16 + STACK + ARGS(%esi)

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

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

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

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

#define OLD_LDC	44 + STACK + ARGS(%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 BX	40(%esp)

#define OLD_STACK 44(%esp)

#define OFFSET  48(%esp)

#define KK	52(%esp)

#define KKK	56(%esp)

#define BUFFER 256(%esp)

#define PREFETCH_R    (8 * 16 + 0)

#define PREFETCH_W    (PREFETCH_R * 2)

#define PREFETCHSIZE  (8 * 7 + 4)

#define PREFETCH     prefetcht0

#define AA	%edx

#define BB	%ecx

#define LDC	%ebp

#define B	%edi

#define C1	%esi

#define I	%ebx

	PROLOGUE

	PROFCODE

	pushl	%ebp

	pushl	%edi

	pushl	%esi

	pushl	%ebx

	movl	%esp, %esi	# save old stack

	subl	$512 + LOCAL_BUFFER_SIZE, %esp

	andl	$-4096, %esp	# align stack

	STACK_TOUCHING

	movl	OLD_M, %ebx

	movl	OLD_N, %eax

	movl	OLD_K, %ecx

	movl	OLD_A, %edx

	movsd	OLD_ALPHA_R,  %xmm0

	movhps	OLD_ALPHA_I,  %xmm0

	movl	%ebx, M

	movl	%eax, N

	movl	%ecx, K

	movl	%edx, A

	movl	%esi, OLD_STACK

	movl	OLD_B, B

	movl	OLD_C, %ebx

	movaps	 %xmm0, ALPHA

	movl	%ebx, C

	movl	OLD_LDC, LDC

	subl	$-16 * SIZE, A

	subl	$-16 * SIZE, B

	sall	$ZBASE_SHIFT, LDC

	sarl	$1, %eax

	movl	%eax, J

	jle	.L40

	ALIGN_4

.L01:

	leal	16 * SIZE + BUFFER, BB

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

	movl	OFFSET, %eax

	movl	%eax, KK

#endif	

	movl	K, %eax

	sarl	$2, %eax

	jle	.L05

	ALIGN_4

.L02:

	movddup	 -16 * SIZE(B), %xmm0

	movddup	 -15 * SIZE(B), %xmm1

	movddup	 -14 * SIZE(B), %xmm2

	movddup	 -13 * SIZE(B), %xmm3

	movddup	 -12 * SIZE(B), %xmm4

	movddup	 -11 * SIZE(B), %xmm5

	movddup	 -10 * SIZE(B), %xmm6

	movddup	  -9 * SIZE(B), %xmm7

	prefetcht0	(PREFETCH_R + 0) * SIZE(B)

	movapd	%xmm0,  -16 * SIZE(BB)

	movapd	%xmm1,  -14 * SIZE(BB)

	movapd	%xmm2,  -12 * SIZE(BB)

	movapd	%xmm3,  -10 * SIZE(BB)

	movapd	%xmm4,   -8 * SIZE(BB)

	movapd	%xmm5,   -6 * SIZE(BB)

	movapd	%xmm6,   -4 * SIZE(BB)

	movapd	%xmm7,   -2 * SIZE(BB)

	addl	$ 8 * SIZE, B

	addl	$16 * SIZE, BB

	decl	%eax

	jne	.L02

	ALIGN_4

.L05:

	movl	K, %eax

	andl	$3, %eax

	BRANCH

	jle	.L10

	ALIGN_4

.L06:

	movddup	 -16 * SIZE(B), %xmm0

	movddup	 -15 * SIZE(B), %xmm1

	movapd	%xmm0,  -16 * SIZE(BB)

	movapd	%xmm1,  -14 * SIZE(BB)

	addl	$2 * SIZE, B

	addl	$4 * SIZE, BB

	decl	%eax

	jne	.L06

	ALIGN_4

.L10:

	movl	B, BX

	movl	C, C1

	movl	A, AA

	movl	M,  I

	sarl	$2, I

	jle	.L20

	ALIGN_4

.L11:

#if !defined(TRMMKERNEL) || \

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

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

	leal	16 * SIZE + BUFFER, BB

#else

	leal	16 * SIZE + BUFFER, BB

	movl	KK, %eax

	leal	(, %eax, SIZE), %eax

	leal	(AA, %eax, 4), AA

	leal	(BB, %eax, 4), BB /* because it's doubled */

#endif

	movapd	-16 * SIZE(AA), %xmm0

	pxor	%xmm4, %xmm4

	movapd	-16 * SIZE(BB), %xmm1

	pxor	%xmm5, %xmm5

	movapd	 -8 * SIZE(AA), %xmm3

	pxor	%xmm6, %xmm6

	prefetcht0	3 * SIZE(C1)

	pxor	%xmm7, %xmm7

	prefetcht0	3 * SIZE(C1, LDC)

	movapd	%xmm1,  %xmm2

	movl	BX, %eax

	prefetcht0   (%eax)

	subl	$-8 * SIZE, %eax

	movl	%eax, BX

#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	$2, %eax

#endif

	movl	%eax, KKK

#endif

	sarl	$3, %eax

	je	.L15

	ALIGN_4

.L12:

	mulpd	%xmm0,  %xmm1

	addpd	%xmm1,  %xmm4

	movapd	-14 * SIZE(BB), %xmm1

	mulpd	%xmm1,  %xmm0

	addpd	%xmm0,  %xmm5

	movapd	-14 * SIZE(AA), %xmm0

	mulpd	%xmm0,  %xmm2

	addpd	%xmm2,  %xmm6

	movapd	-12 * SIZE(BB), %xmm2

	mulpd	%xmm0,  %xmm1

	movapd	-12 * SIZE(AA), %xmm0

	addpd	%xmm1,  %xmm7

	PADDING;

	movapd	%xmm2,  %xmm1

	mulpd	%xmm0,  %xmm2

	addpd	%xmm2,  %xmm4

	movapd	-10 * SIZE(BB), %xmm2

	mulpd	%xmm2,  %xmm0

	addpd	%xmm0,  %xmm5

	movapd	-10 * SIZE(AA), %xmm0

	mulpd	%xmm0,  %xmm1

	addpd	%xmm1,  %xmm6

	movapd	 -8 * SIZE(BB), %xmm1

	mulpd	%xmm0,  %xmm2

	PADDING;

	movapd	  0 * SIZE(AA), %xmm0

	addpd	%xmm2,  %xmm7

	PADDING;

	movapd	%xmm1,  %xmm2

	mulpd	%xmm3,  %xmm1

	addpd	%xmm1,  %xmm4

	movapd	 -6 * SIZE(BB), %xmm1

	mulpd	%xmm1,  %xmm3

	addpd	%xmm3,  %xmm5

	movapd	 -6 * SIZE(AA), %xmm3

	mulpd	%xmm3,  %xmm2

	addpd	%xmm2,  %xmm6

	movapd	 -4 * SIZE(BB), %xmm2

	mulpd	%xmm3,  %xmm1

	movapd	 -4 * SIZE(AA), %xmm3

	addpd	%xmm1,  %xmm7

	PADDING;

	movapd	%xmm2,  %xmm1

	mulpd	%xmm3,  %xmm2

	addpd	%xmm2,  %xmm4

	movapd	 -2 * SIZE(BB), %xmm2

	mulpd	%xmm2,  %xmm3

	addpd	%xmm3,  %xmm5

	movapd	 -2 * SIZE(AA), %xmm3

	mulpd	%xmm3,  %xmm1

	addpd	%xmm1,  %xmm6

	PADDING;

	movapd	  0 * SIZE(BB), %xmm1

	mulpd	%xmm3,  %xmm2

	movapd	  8 * SIZE(AA), %xmm3

	addpd	%xmm2,  %xmm7

	PADDING;

	movapd	%xmm1,  %xmm2

	mulpd	%xmm0,  %xmm1

	addpd	%xmm1,  %xmm4

	movapd	  2 * SIZE(BB), %xmm1

	mulpd	%xmm1,  %xmm0

	addpd	%xmm0,  %xmm5

	movapd	  2 * SIZE(AA), %xmm0

	mulpd	%xmm0,  %xmm2

	addpd	%xmm2,  %xmm6

	movapd	  4 * SIZE(BB), %xmm2

	mulpd	%xmm0,  %xmm1

	movapd	  4 * SIZE(AA), %xmm0

	addpd	%xmm1,  %xmm7

	PADDING;

	movapd	%xmm2,  %xmm1

	mulpd	%xmm0,  %xmm2

	addpd	%xmm2,  %xmm4

	movapd	  6 * SIZE(BB), %xmm2

	mulpd	%xmm2,  %xmm0

	addpd	%xmm0,  %xmm5

	movapd	  6 * SIZE(AA), %xmm0

	mulpd	%xmm0,  %xmm1

	addpd	%xmm1,  %xmm6

	movapd	  8 * SIZE(BB), %xmm1

	mulpd	%xmm0,  %xmm2

	movapd	 16 * SIZE(AA), %xmm0

	addpd	%xmm2,  %xmm7

	PADDING;

	movapd	%xmm1,  %xmm2

	mulpd	%xmm3,  %xmm1

	addpd	%xmm1,  %xmm4

	movapd	 10 * SIZE(BB), %xmm1

	mulpd	%xmm1,  %xmm3

	addpd	%xmm3,  %xmm5

	movapd	 10 * SIZE(AA), %xmm3

	mulpd	%xmm3,  %xmm2

	addpd	%xmm2,  %xmm6

	movapd	 12 * SIZE(BB), %xmm2

	mulpd	%xmm3,  %xmm1

	movapd	 12 * SIZE(AA), %xmm3

	addpd	%xmm1,  %xmm7

	PADDING;

	movapd	%xmm2,  %xmm1

	mulpd	%xmm3,  %xmm2

	addpd	%xmm2,  %xmm4

	movapd	 14 * SIZE(BB), %xmm2

	mulpd	%xmm2,  %xmm3

	subl   $-32 * SIZE, BB

	addpd	%xmm3,  %xmm5

	movapd	 14 * SIZE(AA), %xmm3

	mulpd	%xmm3,  %xmm1

	addpd	%xmm1,  %xmm6

	movapd	-16 * SIZE(BB), %xmm1

	mulpd	%xmm3,  %xmm2

	movapd	 24 * SIZE(AA), %xmm3

	addpd	%xmm2,  %xmm7

	PADDING;

	movapd	%xmm1,  %xmm2

	subl   $-32 * SIZE, AA

	decl   %eax

	BRANCH

	jne  .L12

	ALIGN_4

.L15:

#ifndef TRMMKERNEL

	movl	K, %eax

#else

	movl	KKK, %eax

#endif

	movaps	ALPHA,  %xmm3

	andl	$7, %eax

	BRANCH

	je .L18

	ALIGN_4

.L16:

	mulpd	%xmm0,  %xmm1

	addpd	%xmm1,  %xmm4

	movapd	-14 * SIZE(BB), %xmm1

	mulpd	%xmm1,  %xmm0

	addpd	%xmm0,  %xmm5

	movapd	-14 * SIZE(AA), %xmm0

	mulpd	%xmm0,  %xmm2

	mulpd	%xmm0,  %xmm1

	movapd	-12 * SIZE(AA), %xmm0

	addpd	%xmm2,  %xmm6

	addpd	%xmm1,  %xmm7

	movapd	-12 * SIZE(BB), %xmm1

	movapd	%xmm1,  %xmm2

	addl	$4 * SIZE, AA

	addl	$4 * SIZE, BB

	decl	%eax

	jg	.L16

	ALIGN_4

.L18:

	movsd	0 * SIZE(%esi), %xmm0

	movhps	1 * SIZE(%esi), %xmm0

	movsd	2 * SIZE(%esi), %xmm1

	movhps	3 * SIZE(%esi), %xmm1

	pshufd  $0x44,  %xmm4,  %xmm2

	unpckhpd %xmm4, %xmm4

	mulpd	 %xmm3,  %xmm2

	addpd	 %xmm2,  %xmm0

	mulpd	 %xmm3,  %xmm4

	addpd	 %xmm4,  %xmm1

	movlps	%xmm0, 0 * SIZE(%esi)

	movhps	%xmm0, 1 * SIZE(%esi)

	movlps	%xmm1, 2 * SIZE(%esi)

	movhps	%xmm1, 3 * SIZE(%esi)

	movsd	4 * SIZE(%esi), %xmm0

	movhps	5 * SIZE(%esi), %xmm0

	movsd	6 * SIZE(%esi), %xmm1

	movhps	7 * SIZE(%esi), %xmm1

	pshufd  $0x44,  %xmm6,  %xmm2

	unpckhpd %xmm6, %xmm6

	mulpd	 %xmm3,  %xmm2

	addpd	 %xmm2,  %xmm0

	mulpd	 %xmm3,  %xmm6

	addpd	 %xmm6,  %xmm1

	movlps	%xmm0, 4 * SIZE(%esi)

	movhps	%xmm0, 5 * SIZE(%esi)

	movlps	%xmm1, 6 * SIZE(%esi)

	movhps	%xmm1, 7 * SIZE(%esi)

	movsd	0 * SIZE(%esi, LDC), %xmm0

	movhps	1 * SIZE(%esi, LDC), %xmm0

	movsd	2 * SIZE(%esi, LDC), %xmm1

	movhps	3 * SIZE(%esi, LDC), %xmm1

	pshufd  $0x44,  %xmm5,  %xmm2

	unpckhpd %xmm5, %xmm5

	mulpd	 %xmm3,  %xmm2

	addpd	 %xmm2,  %xmm0

	mulpd	 %xmm3,  %xmm5

	addpd	 %xmm5,  %xmm1

	movlps	%xmm0, 0 * SIZE(%esi, LDC)

	movhps	%xmm0, 1 * SIZE(%esi, LDC)

	movlps	%xmm1, 2 * SIZE(%esi, LDC)

	movhps	%xmm1, 3 * SIZE(%esi, LDC)

	movsd	4 * SIZE(%esi, LDC), %xmm0

	movhps	5 * SIZE(%esi, LDC), %xmm0

	movsd	6 * SIZE(%esi, LDC), %xmm1

	movhps	7 * SIZE(%esi, LDC), %xmm1

	pshufd  $0x44,  %xmm7,  %xmm2

	unpckhpd %xmm7, %xmm7

	mulpd	 %xmm3,  %xmm2

	addpd	 %xmm2,  %xmm0

	mulpd	 %xmm3,  %xmm7

	addpd	 %xmm7,  %xmm1

	movlps	%xmm0, 4 * SIZE(%esi, LDC)

	movhps	%xmm0, 5 * SIZE(%esi, LDC)

	movlps	%xmm1, 6 * SIZE(%esi, LDC)

	movhps	%xmm1, 7 * SIZE(%esi, LDC)

	addl	$8 * SIZE, C1

	decl	I

	jg	.L11

	ALIGN_4

.L20:

	movl	M,  I

	testl	$2, I

	jle	.L30

.L21:

#if !defined(TRMMKERNEL) || \

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

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

	leal	16 * SIZE + BUFFER, BB

#else

	leal	16 * SIZE + BUFFER, BB

	movl	KK, %eax

	leal	(, %eax, SIZE), %eax

	leal	(AA, %eax, 2), AA

	leal	(BB, %eax, 4), BB /* because it's doubled */

#endif

	movapd	-16 * SIZE(AA), %xmm0

	pxor	%xmm4, %xmm4

	movapd	-16 * SIZE(BB), %xmm1

	pxor	%xmm5, %xmm5

	movapd	 -8 * SIZE(AA), %xmm2

	pxor	%xmm6, %xmm6

	movapd	 -8 * SIZE(BB), %xmm3

	pxor	%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

	addl	$2, %eax

	movl	%eax, KKK

#endif

	sarl	$3, %eax

	je	.L25

	ALIGN_4

.L22:

	mulpd	%xmm0,  %xmm1

	mulpd	-14 * SIZE(BB), %xmm0

	addpd	%xmm1, %xmm4

	movapd	-12 * SIZE(BB), %xmm1

	addpd	%xmm0, %xmm5

	movapd	-14 * SIZE(AA), %xmm0

	mulpd	%xmm0,  %xmm1

	mulpd	-10 * SIZE(BB), %xmm0

	addpd	%xmm1, %xmm6

	movapd	  0 * SIZE(BB), %xmm1

	addpd	%xmm0, %xmm7

	movapd	-12 * SIZE(AA), %xmm0

	mulpd	%xmm0,  %xmm3

	mulpd	 -6 * SIZE(BB), %xmm0

	addpd	%xmm3, %xmm4

	movapd	 -4 * SIZE(BB), %xmm3

	addpd	%xmm0, %xmm5

	movapd	-10 * SIZE(AA), %xmm0

	mulpd	%xmm0,  %xmm3

	mulpd	 -2 * SIZE(BB), %xmm0

	addpd	%xmm3, %xmm6

	movapd	  8 * SIZE(BB), %xmm3

	addpd	%xmm0, %xmm7

	movapd	  0 * SIZE(AA), %xmm0

	mulpd	%xmm2,  %xmm1

	mulpd	  2 * SIZE(BB), %xmm2

	addpd	%xmm1, %xmm4

	movapd	  4 * SIZE(BB), %xmm1

	addpd	%xmm2, %xmm5

	movapd	 -6 * SIZE(AA), %xmm2

	mulpd	%xmm2,  %xmm1

	mulpd	  6 * SIZE(BB), %xmm2

	addpd	%xmm1, %xmm6

	movapd	 16 * SIZE(BB), %xmm1

	addpd	%xmm2, %xmm7

	movapd	 -4 * SIZE(AA), %xmm2

	mulpd	%xmm2,  %xmm3

	mulpd	 10 * SIZE(BB), %xmm2

	addpd	%xmm3, %xmm4

	movapd	 12 * SIZE(BB), %xmm3

	addpd	%xmm2, %xmm5

	movapd	 -2 * SIZE(AA), %xmm2

	mulpd	%xmm2,  %xmm3

	mulpd	 14 * SIZE(BB), %xmm2

	addpd	%xmm3, %xmm6

	movapd	 24 * SIZE(BB), %xmm3

	addpd	%xmm2, %xmm7

	movapd	  8 * SIZE(AA), %xmm2

	subl   $-16 * SIZE, AA

	addl   $ 32 * SIZE, BB

	decl   %eax

	jne    .L22

	ALIGN_4

.L25:

	movaps	ALPHA,  %xmm3

#ifndef TRMMKERNEL

	movl	K, %eax

#else

	movl	KKK, %eax

#endif

	andl	$7, %eax

	BRANCH

	je .L28

	ALIGN_4

.L26:

	mulpd	%xmm0,  %xmm1

	mulpd	-14 * SIZE(BB), %xmm0

	addpd	%xmm1, %xmm4

	movapd	-12 * SIZE(BB), %xmm1

	addpd	%xmm0, %xmm5

	movapd	-14 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA

	addl	$4 * SIZE, BB

	decl	%eax

	jg	.L26

	ALIGN_4

.L28:

	addpd	%xmm6, %xmm4

	addpd	%xmm7, %xmm5

	movsd	0 * SIZE(%esi), %xmm0

	movhps	1 * SIZE(%esi), %xmm0

	movsd	2 * SIZE(%esi), %xmm1

	movhps	3 * SIZE(%esi), %xmm1

	pshufd  $0x44,  %xmm4,  %xmm2

	unpckhpd %xmm4, %xmm4

	mulpd	 %xmm3,  %xmm2

	addpd	 %xmm2,  %xmm0

	mulpd	 %xmm3,  %xmm4

	addpd	 %xmm4,  %xmm1

	movlps	%xmm0, 0 * SIZE(%esi)

	movhps	%xmm0, 1 * SIZE(%esi)

	movlps	%xmm1, 2 * SIZE(%esi)

	movhps	%xmm1, 3 * SIZE(%esi)

	movsd	0 * SIZE(%esi, LDC), %xmm0

	movhps	1 * SIZE(%esi, LDC), %xmm0

	movsd	2 * SIZE(%esi, LDC), %xmm1

	movhps	3 * SIZE(%esi, LDC), %xmm1

	pshufd  $0x44,  %xmm5,  %xmm2

	unpckhpd %xmm5, %xmm5

	mulpd	 %xmm3,  %xmm2

	addpd	 %xmm2,  %xmm0

	mulpd	 %xmm3,  %xmm5

	addpd	 %xmm5,  %xmm1

	movlps	%xmm0, 0 * SIZE(%esi, LDC)

	movhps	%xmm0, 1 * SIZE(%esi, LDC)

	movlps	%xmm1, 2 * SIZE(%esi, LDC)

	movhps	%xmm1, 3 * SIZE(%esi, LDC)

	addl	$4 * SIZE, C1

	ALIGN_4

.L30:

	movl	M,  I

	testl	$1, I

	jle	.L39

.L31:

#if !defined(TRMMKERNEL) || \

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

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

	leal	16 * SIZE + BUFFER, BB

#else

	leal	16 * SIZE + BUFFER, BB

	movl	KK, %eax

	leal	(, %eax, SIZE), %eax

	leal	(AA, %eax, 1), AA

	leal	(BB, %eax, 4), BB /* because it's doubled */

#endif

	movsd	-16 * SIZE(AA), %xmm0

	pxor	%xmm4, %xmm4

	movsd	-16 * SIZE(BB), %xmm1

	pxor	%xmm5, %xmm5

	movsd	-12 * SIZE(AA), %xmm2

	pxor	%xmm6, %xmm6

	movsd	 -8 * SIZE(BB), %xmm3

	pxor	%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	$2, %eax

#endif

	movl	%eax, KKK

#endif

	sarl	$3, %eax

	je	.L35

	ALIGN_4

.L32:

	mulsd	%xmm0,  %xmm1

	mulsd	-14 * SIZE(BB), %xmm0

	addsd	%xmm1, %xmm4

 	movsd	-12 * SIZE(BB), %xmm1

	addsd	%xmm0, %xmm5

	movsd	-15 * SIZE(AA), %xmm0

	mulsd	%xmm0,  %xmm1

	mulsd	-10 * SIZE(BB), %xmm0

	addsd	%xmm1, %xmm6

	movsd	  0 * SIZE(BB), %xmm1

	addsd	%xmm0, %xmm7

	movsd	-14 * SIZE(AA), %xmm0

	mulsd	%xmm0,  %xmm3

	mulsd	 -6 * SIZE(BB), %xmm0

	addsd	%xmm3, %xmm4

	movsd	 -4 * SIZE(BB), %xmm3

	addsd	%xmm0, %xmm5

	movsd	-13 * SIZE(AA), %xmm0

	mulsd	%xmm0,  %xmm3

	mulsd	 -2 * SIZE(BB), %xmm0

	addsd	%xmm3, %xmm6

	movsd	  8 * SIZE(BB), %xmm3

	addsd	%xmm0, %xmm7

	movsd	 -8 * SIZE(AA), %xmm0

	mulsd	%xmm2,  %xmm1

	mulsd	  2 * SIZE(BB), %xmm2

	addsd	%xmm1, %xmm4

	movsd	  4 * SIZE(BB), %xmm1

	addsd	%xmm2, %xmm5

	movsd	-11 * SIZE(AA), %xmm2

	mulsd	%xmm2,  %xmm1

	mulsd	  6 * SIZE(BB), %xmm2

	addsd	%xmm1, %xmm6

	movsd	 16 * SIZE(BB), %xmm1

	addsd	%xmm2, %xmm7

	movsd	-10 * SIZE(AA), %xmm2

	mulsd	%xmm2,  %xmm3

	mulsd	 10 * SIZE(BB), %xmm2

	addsd	%xmm3, %xmm4

	movsd	 12 * SIZE(BB), %xmm3

	addsd	%xmm2, %xmm5

	movsd	 -9 * SIZE(AA), %xmm2

	mulsd	%xmm2,  %xmm3

	mulsd	 14 * SIZE(BB), %xmm2

	addsd	%xmm3, %xmm6

	movsd	 24 * SIZE(BB), %xmm3

	addsd	%xmm2, %xmm7

	movsd	 -4 * SIZE(AA), %xmm2

	subl   $-8 * SIZE, AA

	addl   $32 * SIZE, BB

	decl   %eax

	jne    .L32

	ALIGN_4

.L35:

	movaps	ALPHA,  %xmm3

#ifndef TRMMKERNEL

	movl	K, %eax

#else

	movl	KKK, %eax

#endif

	andl	$7, %eax

	BRANCH

	je .L38

	ALIGN_4

.L36:

	mulsd	%xmm0,  %xmm1

	mulsd	-14 * SIZE(BB), %xmm0

	addsd	%xmm1, %xmm4

	movsd	-12 * SIZE(BB), %xmm1

	addsd	%xmm0, %xmm5

	movsd	-15 * SIZE(AA), %xmm0

	addl	$1 * SIZE, AA

	addl	$4 * SIZE, BB

	decl	%eax

	jg	.L36

	ALIGN_4

.L38:

	addsd	%xmm6, %xmm4

	addsd	%xmm7, %xmm5

	movsd	0 * SIZE(%esi), %xmm0

	movhps	1 * SIZE(%esi), %xmm0

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

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

	unpcklpd  %xmm4, %xmm4

	unpcklpd  %xmm5, %xmm5

	mulpd	 %xmm3,  %xmm4

	addpd	 %xmm4,  %xmm0

	mulpd	 %xmm3,  %xmm5

	addpd	 %xmm5,  %xmm1

	movlps	%xmm0, 0 * SIZE(%esi)

	movhps	%xmm0, 1 * SIZE(%esi)

	movlps	%xmm1, 0 * SIZE(%esi, LDC)

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

	ALIGN_4

.L39:

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

	addl	$2, KK

#endif

	leal	(, LDC, 2), %eax

	addl	%eax, C

	decl	J

	jg	.L01

	ALIGN_4

.L40:

	movl	N, %eax

	testl	$1, %eax

	jle	.L999

	ALIGN_4

.L41:

	leal	16 * SIZE + BUFFER, BB

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

	movl	OFFSET, %eax

	movl	%eax, KK

#endif	

	movl	K, %eax

	sarl	$3, %eax

	jle	.L45

	ALIGN_4

.L42:

	movddup	 -16 * SIZE(B), %xmm0

	movddup	 -15 * SIZE(B), %xmm1

	movddup	 -14 * SIZE(B), %xmm2

	movddup	 -13 * SIZE(B), %xmm3

	movddup	 -12 * SIZE(B), %xmm4

	movddup	 -11 * SIZE(B), %xmm5

	movddup	 -10 * SIZE(B), %xmm6

	movddup	  -9 * SIZE(B), %xmm7

	movapd	%xmm0,  -16 * SIZE(BB)

	movapd	%xmm1,  -14 * SIZE(BB)

	movapd	%xmm2,  -12 * SIZE(BB)

	movapd	%xmm3,  -10 * SIZE(BB)

	movapd	%xmm4,   -8 * SIZE(BB)

	movapd	%xmm5,   -6 * SIZE(BB)

	movapd	%xmm6,   -4 * SIZE(BB)

	movapd	%xmm7,   -2 * SIZE(BB)

	addl	$ 8 * SIZE, B

	addl	$16 * SIZE, BB

	decl	%eax

	jne	.L42

	ALIGN_4

.L45:

	movl	K, %eax

	andl	$7, %eax

	BRANCH

	jle	.L50

	ALIGN_4

.L46:

	movddup	 -16 * SIZE(B), %xmm0

	movapd	%xmm0,  -16 * SIZE(BB)

	addl	$1 * SIZE, B

	addl	$2 * SIZE, BB

	decl	%eax

	jne	.L46

	ALIGN_4