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

/* 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_R	16 + STACK(%esi)

#define OLD_ALPHA_I	20 + STACK(%esi)

#define OLD_A	24 + STACK(%esi)

#define OLD_B	28 + STACK(%esi)

#define OLD_C	32 + STACK(%esi)

#define OLD_LDC	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)

#if defined(PENRYN) || defined(DUNNINGTON)

#define PREFETCH     prefetcht0

#define PREFETCHSIZE   96

#endif

#ifdef PENTIUM4

#define PREFETCH     prefetcht0

#define PREFETCHSIZE   96

#endif

#ifdef PENTIUMM

#define PREFETCH     prefetcht0

#define PREFETCHSIZE   96

#endif

#define AA	%edx

#define	BB	%ecx

#define LDC	%ebp

#define KERNEL1(address) \

	mulps	%xmm0, %xmm2; \

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

	addps	%xmm2, %xmm4; \

	movshdup  0 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm0, %xmm2; \

	addps	%xmm2, %xmm5; \

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

	mulps	%xmm0, %xmm2; \

	addps	%xmm2, %xmm6; \

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

	mulps	%xmm0, %xmm2; \

	movaps	  4 * SIZE + 1 * (address) * SIZE(AA), %xmm0; \

	addps	%xmm2, %xmm7; \

	movsldup  8 * SIZE + 2 * (address) * SIZE(BB), %xmm2

#define KERNEL2(address) \

	mulps	%xmm0, %xmm2; \

	addps	%xmm2, %xmm4; \

	movshdup  8 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm0, %xmm2; \

	addps	%xmm2, %xmm5; \

	movsldup 12 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm0, %xmm2; \

	addps	%xmm2, %xmm6; \

	movshdup 12 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm0, %xmm2; \

	movaps	  8 * SIZE + 1 * (address) * SIZE(AA), %xmm0; \

	addps	%xmm2, %xmm7; \

	movsldup 32 * SIZE + 2 * (address) * SIZE(BB), %xmm2

#define KERNEL3(address) \

	mulps	%xmm0, %xmm3; \

	addps	%xmm3, %xmm4; \

	movshdup 16 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm0, %xmm3; \

	addps	%xmm3, %xmm5; \

	movsldup 20 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm0, %xmm3; \

	addps	%xmm3, %xmm6; \

	movshdup 20 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm0, %xmm3; \

	movaps	 12 * SIZE + 1 * (address) * SIZE(AA), %xmm0; \

	addps	%xmm3, %xmm7; \

	movsldup 24 * SIZE + 2 * (address) * SIZE(BB), %xmm3

#define KERNEL4(address) \

	mulps	%xmm0, %xmm3; \

	addps	%xmm3, %xmm4; \

	movshdup 24 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm0, %xmm3; \

	addps	%xmm3, %xmm5; \

	movsldup 28 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm0, %xmm3; \

	addps	%xmm3, %xmm6; \

	movshdup 28 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm0, %xmm3; \

	movaps	 32 * SIZE + 1 * (address) * SIZE(AA), %xmm0; \

	addps	%xmm3, %xmm7; \

	movsldup 48 * SIZE + 2 * (address) * SIZE(BB), %xmm3

#define KERNEL5(address) \

	mulps	%xmm1, %xmm2; \

	addps	%xmm2, %xmm4; \

	movshdup 32 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm1, %xmm2; \

	addps	%xmm2, %xmm5; \

	movsldup 36 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm1, %xmm2; \

	addps	%xmm2, %xmm6; \

	movshdup 36 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm1, %xmm2; \

	movaps	 20 * SIZE + 1 * (address) * SIZE(AA), %xmm1; \

	addps	%xmm2, %xmm7

#define KERNEL6(address) \

	movsldup 40 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm1, %xmm2; \

	addps	%xmm2, %xmm4; \

	movshdup 40 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm1, %xmm2; \

	addps	%xmm2, %xmm5; \

	movsldup 44 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm1, %xmm2; \

	addps	%xmm2, %xmm6; \

	movshdup 44 * SIZE + 2 * (address) * SIZE(BB), %xmm2; \

	mulps	%xmm1, %xmm2; \

	movaps	 24 * SIZE + 1 * (address) * SIZE(AA), %xmm1; \

	addps	%xmm2, %xmm7; \

	movsldup 64 * SIZE + 2 * (address) * SIZE(BB), %xmm2

#define KERNEL7(address) \

	mulps	%xmm1, %xmm3; \

	addps	%xmm3, %xmm4; \

	movshdup 48 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm1, %xmm3; \

	addps	%xmm3, %xmm5; \

	movsldup 52 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm1, %xmm3; \

	addps	%xmm3, %xmm6; \

	movshdup 52 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm1, %xmm3; \

	movaps	 28 * SIZE + 1 * (address) * SIZE(AA), %xmm1; \

	addps	%xmm3, %xmm7; \

	movsldup 56 * SIZE + 2 * (address) * SIZE(BB), %xmm3

#define KERNEL8(address) \

	mulps	%xmm1, %xmm3; \

	addps	%xmm3, %xmm4; \

	movshdup 56 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm1, %xmm3; \

	addps	%xmm3, %xmm5; \

	movsldup 60 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm1, %xmm3; \

	addps	%xmm3, %xmm6; \

	movshdup 60 * SIZE + 2 * (address) * SIZE(BB), %xmm3; \

	mulps	%xmm1, %xmm3; \

	movaps	 48 * SIZE + 1 * (address) * SIZE(AA), %xmm1; \

	addps	%xmm3, %xmm7; \

	movsldup 80 * SIZE + 2 * (address) * SIZE(BB), %xmm3

	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_R,  %xmm0

	movss	OLD_ALPHA_I,  %xmm1

	movl	%ebx, M

	movl	%eax, N

	movl	%ecx, K

	movl	%edx, A

	movl	%esi, OLD_STACK

	movl	OLD_B, %edi

	movl	OLD_C, %ebx

	unpcklps %xmm1, %xmm0

	movlhps	 %xmm0, %xmm0

	movaps	 %xmm0, ALPHA

	movl	%ebx, C

	movl	OLD_LDC, LDC

#ifdef TRMMKERNEL

	movss	%xmm4, OFFSET

	movss	%xmm4, KK

#ifndef LEFT

	negl	KK

#endif	

#endif

	sall	$ZBASE_SHIFT, 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	$2, %eax

	jle	.L05

	ALIGN_4

.L02:

	movddup	 0 * SIZE(%edi), %xmm0

	movddup	 2 * SIZE(%edi), %xmm1

	movddup	 4 * SIZE(%edi), %xmm2

	movddup	 6 * SIZE(%edi), %xmm3

	movddup	 8 * SIZE(%edi), %xmm4

	movddup	10 * SIZE(%edi), %xmm5

	movddup	12 * SIZE(%edi), %xmm6

	movddup	14 * SIZE(%edi), %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)

#	prefetcht1	128 * SIZE(%ecx)

	prefetcht0	112 * SIZE(%edi)

	addl	$16 * SIZE, %edi

	addl	$32 * SIZE, %ecx

	decl	%eax

	jne	.L02

	ALIGN_2

.L05:

	movl	K, %eax

	andl	$3, %eax

	BRANCH

	jle	.L10

	ALIGN_2

.L06:

	movddup	 0 * SIZE(%edi), %xmm0

	movddup	 2 * SIZE(%edi), %xmm1

	movaps	%xmm0,  0 * SIZE(%ecx)

	movaps	%xmm1,  4 * SIZE(%ecx)

	addl	$4 * SIZE, %edi

	addl	$8 * SIZE, %ecx

	decl	%eax

	jne	.L06

	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, 4), BB

#endif	

	movaps	 0 * SIZE(AA), %xmm0

	pxor	%xmm4, %xmm4

	movaps	16 * SIZE(AA), %xmm1

	pxor	%xmm5, %xmm5

	movsldup  0 * SIZE(BB), %xmm2

	pxor	%xmm6, %xmm6

	movsldup 16 * SIZE(BB), %xmm3

	pxor	%xmm7, %xmm7

	leal	(LDC, LDC, 2), %eax

	prefetchnta	4 * SIZE(%esi)

	prefetchnta	4 * SIZE(%esi, LDC)

	prefetchnta	4 * SIZE(%esi, LDC, 2)

	prefetchnta	4 * SIZE(%esi, %eax)

#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

#if 1

	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)

#if 1

	cmpl	$128 *  8, %eax

	jle	.L12

	KERNEL1(32 *  8)

	KERNEL2(32 *  8)

	KERNEL3(32 *  8)

	KERNEL4(32 *  8)

	KERNEL5(32 *  8)

	KERNEL6(32 *  8)

	KERNEL7(32 *  8)

	KERNEL8(32 *  8)

	cmpl	$128 *  9, %eax

	jle	.L12

	KERNEL1(32 *  9)

	KERNEL2(32 *  9)

	KERNEL3(32 *  9)

	KERNEL4(32 *  9)

	KERNEL5(32 *  9)

	KERNEL6(32 *  9)

	KERNEL7(32 *  9)

	KERNEL8(32 *  9)

	cmpl	$128 * 10, %eax

	jle	.L12

	KERNEL1(32 * 10)

	KERNEL2(32 * 10)

	KERNEL3(32 * 10)

	KERNEL4(32 * 10)

	KERNEL5(32 * 10)

	KERNEL6(32 * 10)

	KERNEL7(32 * 10)

	KERNEL8(32 * 10)

	cmpl	$128 * 11, %eax

	jle	.L12

	KERNEL1(32 * 11)

	KERNEL2(32 * 11)

	KERNEL3(32 * 11)

	KERNEL4(32 * 11)

	KERNEL5(32 * 11)

	KERNEL6(32 * 11)

	KERNEL7(32 * 11)

	KERNEL8(32 * 11)

	cmpl	$128 * 12, %eax

	jle	.L12

	KERNEL1(32 * 12)

	KERNEL2(32 * 12)

	KERNEL3(32 * 12)

	KERNEL4(32 * 12)

	KERNEL5(32 * 12)

	KERNEL6(32 * 12)

	KERNEL7(32 * 12)

	KERNEL8(32 * 12)

	cmpl	$128 * 13, %eax

	jle	.L12

	KERNEL1(32 * 13)

	KERNEL2(32 * 13)

	KERNEL3(32 * 13)

	KERNEL4(32 * 13)

	KERNEL5(32 * 13)

	KERNEL6(32 * 13)

	KERNEL7(32 * 13)

	KERNEL8(32 * 13)

	cmpl	$128 * 14, %eax

	jle	.L12

	KERNEL1(32 * 14)

	KERNEL2(32 * 14)

	KERNEL3(32 * 14)

	KERNEL4(32 * 14)

	KERNEL5(32 * 14)

	KERNEL6(32 * 14)

	KERNEL7(32 * 14)

	KERNEL8(32 * 14)

	cmpl	$128 * 15, %eax

	jle	.L12

	KERNEL1(32 * 15)

	KERNEL2(32 * 15)

	KERNEL3(32 * 15)

	KERNEL4(32 * 15)

	KERNEL5(32 * 15)

	KERNEL6(32 * 15)

	KERNEL7(32 * 15)

	KERNEL8(32 * 15)

#else

	addl	$128 * 4  * SIZE, BB

	addl	$128 * 2  * SIZE, AA

	subl	$128 * 8, %eax

	jg	.L1X

	jmp	.L15

#endif

.L12:

	leal	(AA, %eax, 1), AA

	leal	(BB, %eax, 2), BB

	ALIGN_4

#else

	sarl	$3, %eax

	je	.L15

	ALIGN_4

.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   $32 * SIZE, AA

	addl   $64 * SIZE, BB

	decl   %eax

	jne    .L12

	ALIGN_4

#endif

.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

	movshdup  0 * SIZE(BB), %xmm2

	mulps	%xmm0, %xmm2

	addps	%xmm2, %xmm5

	movsldup  4 * SIZE(BB), %xmm2

	mulps	%xmm0, %xmm2

	addps	%xmm2, %xmm6

	movshdup  4 * SIZE(BB), %xmm2

	mulps	%xmm0, %xmm2

	movaps	  4 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm7

	movsldup  8 * SIZE(BB), %xmm2

	addl	$4 * SIZE, AA

	addl	$8 * SIZE, BB

	decl	%eax

	jg	.L16

	ALIGN_4

.L18:

	leal	(LDC, LDC, 2), %eax

	movsd	 0 * SIZE(%esi), %xmm0

	movhps	 2 * SIZE(%esi), %xmm0

	movsd	 4 * SIZE(%esi), %xmm1

	movhps	 6 * SIZE(%esi), %xmm1

	pshufd	$0x50, %xmm4,  %xmm2

	pshufd	$0xfa, %xmm4,  %xmm4

	mulps	%xmm3, %xmm2

	mulps	%xmm3, %xmm4

	addps	%xmm2, %xmm0

	addps	%xmm4, %xmm1

	movlps	%xmm0,   0 * SIZE(%esi)

	movhps	%xmm0,   2 * SIZE(%esi)

	movlps	%xmm1,   4 * SIZE(%esi)

	movhps	%xmm1,   6 * SIZE(%esi)

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

	movhps	 2 * SIZE(%esi, LDC), %xmm0

	movsd	 4 * SIZE(%esi, LDC), %xmm1

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

	pshufd	$0x50, %xmm5,  %xmm2

	pshufd	$0xfa, %xmm5,  %xmm5

	mulps	%xmm3, %xmm2

	mulps	%xmm3, %xmm5

	addps	%xmm2, %xmm0

	addps	%xmm5, %xmm1

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

	movhps	%xmm0,   2 * SIZE(%esi, LDC)

	movlps	%xmm1,   4 * SIZE(%esi, LDC)

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

	movsd	 0 * SIZE(%esi, LDC, 2), %xmm0

	movhps	 2 * SIZE(%esi, LDC, 2), %xmm0

	movsd	 4 * SIZE(%esi, LDC, 2), %xmm1

	movhps	 6 * SIZE(%esi, LDC, 2), %xmm1

	pshufd	$0x50, %xmm6,  %xmm2

	pshufd	$0xfa, %xmm6,  %xmm6

	mulps	%xmm3, %xmm2

	mulps	%xmm3, %xmm6

	addps	%xmm2, %xmm0

	addps	%xmm6, %xmm1

	movlps	%xmm0,   0 * SIZE(%esi, LDC, 2)

	movhps	%xmm0,   2 * SIZE(%esi, LDC, 2)

	movlps	%xmm1,   4 * SIZE(%esi, LDC, 2)

	movhps	%xmm1,   6 * SIZE(%esi, LDC, 2)

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

	movhps	 2 * SIZE(%esi, %eax), %xmm0

	movsd	 4 * SIZE(%esi, %eax), %xmm1

	movhps	 6 * SIZE(%esi, %eax), %xmm1

	pshufd	$0x50, %xmm7,  %xmm2

	pshufd	$0xfa, %xmm7,  %xmm7

	mulps	%xmm3, %xmm2

	mulps	%xmm3, %xmm7

	addps	%xmm2, %xmm0

	addps	%xmm7, %xmm1

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

	movhps	%xmm0,   2 * SIZE(%esi, %eax)

	movlps	%xmm1,   4 * SIZE(%esi, %eax)

	movhps	%xmm1,   6 * SIZE(%esi, %eax)

	addl	$8 * 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, 4), BB

#endif	

	movddup	  0 * SIZE(AA), %xmm0

	pxor	%xmm4, %xmm4

	movddup	  8 * SIZE(AA), %xmm1

	pxor	%xmm5, %xmm5

	movsd     0 * SIZE(BB), %xmm2

	movsd    16 * SIZE(BB), %xmm3

#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:

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm0, %xmm2

	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AA)

	addps	%xmm2, %xmm4

	movsd     4 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm0, %xmm2

	movddup	  2 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm5

	movsd     8 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm0, %xmm2

	addps	%xmm2, %xmm4

	movsd    12 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm0, %xmm2

	movddup	  4 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm5

	movsd    32 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm3, %xmm3

	mulps	%xmm0, %xmm3

	addps	%xmm3, %xmm4

	movsd    20 * SIZE(BB), %xmm3

	shufps	$0x50, %xmm3, %xmm3

	mulps	%xmm0, %xmm3

	movddup	  6 * SIZE(AA), %xmm0

	addps	%xmm3, %xmm5

	movsd    24 * SIZE(BB), %xmm3

	shufps	$0x50, %xmm3, %xmm3

	mulps	%xmm0, %xmm3

	addps	%xmm3, %xmm4

	movsd    28 * SIZE(BB), %xmm3

	shufps	$0x50, %xmm3, %xmm3

	mulps	%xmm0, %xmm3

	movddup	 16 * SIZE(AA), %xmm0

	addps	%xmm3, %xmm5

	movsd    48 * SIZE(BB), %xmm3

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm1, %xmm2

	addps	%xmm2, %xmm4

	movsd    36 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm1, %xmm2

	movddup	 10 * SIZE(AA), %xmm1

	addps	%xmm2, %xmm5

	movsd    40 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm1, %xmm2

	addps	%xmm2, %xmm4

	movsd    44 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm1, %xmm2

	movddup	 12 * SIZE(AA), %xmm1

	addps	%xmm2, %xmm5

	movsd    64 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm3, %xmm3

	mulps	%xmm1, %xmm3

	addps	%xmm3, %xmm4

	movsd    52 * SIZE(BB), %xmm3

	shufps	$0x50, %xmm3, %xmm3

	mulps	%xmm1, %xmm3

	movddup	 14 * SIZE(AA), %xmm1

	addps	%xmm3, %xmm5

	movsd    56 * SIZE(BB), %xmm3

	shufps	$0x50, %xmm3, %xmm3

	mulps	%xmm1, %xmm3

	addps	%xmm3, %xmm4

	movsd    60 * SIZE(BB), %xmm3

	shufps	$0x50, %xmm3, %xmm3

	mulps	%xmm1, %xmm3

	movddup	 24 * SIZE(AA), %xmm1

	addps	%xmm3, %xmm5

	movsd    80 * SIZE(BB), %xmm3

	addl	$16 * SIZE, AA

	addl	$64 * 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:

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm0, %xmm2

	addps	%xmm2, %xmm4

	movsd     4 * SIZE(BB), %xmm2

	shufps	$0x50, %xmm2, %xmm2

	mulps	%xmm0, %xmm2

 	movddup	  2 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm5

	movsd     8 * SIZE(BB), %xmm2

	addl	$2 * SIZE, AA

	addl	$8 * SIZE, BB

	decl	%eax

	jg	.L26

	ALIGN_4

.L28:

	leal	(LDC, LDC, 2), %eax

	movsd	 0 * SIZE(%esi), %xmm0

	movhps	 2 * SIZE(%esi), %xmm0

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

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

	pshufd	$0x50, %xmm4,  %xmm2

	pshufd	$0xfa, %xmm4,  %xmm4

	mulps	%xmm3, %xmm2

	mulps	%xmm3, %xmm4

	addps	%xmm2, %xmm0

	addps	%xmm4, %xmm1

	movlps	%xmm0,   0 * SIZE(%esi)

	movhps	%xmm0,   2 * SIZE(%esi)

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

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

	movsd	 0 * SIZE(%esi, LDC, 2), %xmm0

	movhps	 2 * SIZE(%esi, LDC, 2), %xmm0

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

	movhps	 2 * SIZE(%esi, %eax), %xmm1

	pshufd	$0x50, %xmm5,  %xmm2

	pshufd	$0xfa, %xmm5,  %xmm5

	mulps	%xmm3, %xmm2

	mulps	%xmm3, %xmm5

	addps	%xmm2, %xmm0

	addps	%xmm5, %xmm1