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

/* 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 STACK_M	 4 + STACK + ARGS(%esi)

#define STACK_N	 8 + STACK + ARGS(%esi)

#define STACK_K	12 + STACK + ARGS(%esi)

#define STACK_ALPHA_R	16 + STACK + ARGS(%esi)

#define STACK_ALPHA_I	20 + STACK + ARGS(%esi)

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

#define STACK_B	28 + STACK + ARGS(%esi)

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

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

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

#define POSINV	 0(%esp)

#define ALPHA_R	16(%esp)

#define ALPHA_I	32(%esp)

#define K	48(%esp)

#define N	52(%esp)

#define M	56(%esp)

#define A	60(%esp)

#define C	64(%esp)

#define J	68(%esp)

#define OLD_STACK 72(%esp)

#define OFFSET  76(%esp)

#define KK	80(%esp)

#define KKK	84(%esp)

#define BUFFER 128(%esp)

#define B	%edi

#define	LDC	%ebp

#define AA	%edx

#define BB	%ecx

#ifdef PENTIUM4

#define PREFETCH prefetcht0

#define PREFETCHSIZE 168

#endif

#ifdef PENTIUMM

#define PREFETCH prefetcht0

#define PREFETCHSIZE 168

#endif

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \

    defined(RN) || defined(RT) || defined(CN) || defined(CT)

#define ADDSUB	addps

#else

#define ADDSUB	subps

#endif

#define KERNEL1(address) \

	mulps	%xmm0, %xmm2; \

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

	addps	%xmm2, %xmm4; \

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

	mulps	%xmm0, %xmm2; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%xmm2, %xmm7; \

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

#define KERNEL6(address) \

	mulps	%xmm1, %xmm2; \

	addps	%xmm2, %xmm4; \

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

	mulps	%xmm1, %xmm2; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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; \

	ADDSUB	%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

	andl	$-1024, %esp	# align stack

	STACK_TOUCHING

	movl	STACK_M, %ebx

	movl	STACK_N, %eax

	movl	STACK_K, %ecx

	movl	STACK_A, %edx

	movl	%ebx, M

	movl	%eax, N

	movl	%ecx, K

	movl	%edx, A

	movl	%esi, OLD_STACK

	movl	STACK_B, %edi

	movl	STACK_C, %ebx

#ifdef TRMMKERNEL

	movss	STACK_OFFT, %xmm4

#endif

	movss	STACK_ALPHA_R, %xmm0

	movss	STACK_ALPHA_I, %xmm1

	pxor	%xmm7, %xmm7

	cmpeqps	%xmm7, %xmm7

	pslld	$31, %xmm7	# Generate mask

	shufps	$0, %xmm0, %xmm0

	movaps	 %xmm0,  0 + ALPHA_R

	movss	 %xmm1,  4 + ALPHA_I

	movss	 %xmm1, 12 + ALPHA_I

	xorps	 %xmm7, %xmm1

	movss	 %xmm1,  0 + ALPHA_I

	movss	 %xmm1,  8 + ALPHA_I

	movl	%ebx, C

	movl	STACK_LDC, LDC

#ifdef TRMMKERNEL

	movss	%xmm4, OFFSET

	movss	%xmm4, KK

#ifndef LEFT

	negl	KK

#endif	

#endif

	sall	$ZBASE_SHIFT, LDC

	movl	%eax, J			# j = n

	sarl	$1, J

	jle	.L100

	ALIGN_4

.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	.L03

	ALIGN_4

.L02:

	movddup	 0 * SIZE(B), %xmm0

	movddup	 2 * SIZE(B), %xmm1

	movddup	 4 * SIZE(B), %xmm2

	movddup	 6 * SIZE(B), %xmm3

	movddup	 8 * SIZE(B), %xmm4

	movddup	10 * SIZE(B), %xmm5

	movddup	12 * SIZE(B), %xmm6

	movddup	14 * SIZE(B), %xmm7

	movaps	%xmm0,  0 * SIZE(BB)

	movaps	%xmm1,  4 * SIZE(BB)

	movaps	%xmm2,  8 * SIZE(BB)

	movaps	%xmm3, 12 * SIZE(BB)

	movaps	%xmm4, 16 * SIZE(BB)

	movaps	%xmm5, 20 * SIZE(BB)

	movaps	%xmm6, 24 * SIZE(BB)

	movaps	%xmm7, 28 * SIZE(BB)

#	prefetcht1	128 * SIZE(%ecx)

	prefetcht0	112 * SIZE(%edi)

	addl	$16 * SIZE, B

	addl	$32 * SIZE, BB

	decl	%eax

	jne	.L02

	ALIGN_4

.L03:

	movl	K, %eax

	andl	$3, %eax

	BRANCH

	jle	.L05

	ALIGN_4

.L04:

	movddup	0 * SIZE(B), %xmm0

	movddup	2 * SIZE(B), %xmm1

	movaps	%xmm0,  0 * SIZE(BB)

	movaps	%xmm1,  4 * SIZE(BB)

	addl	$4 * SIZE, B

	addl	$8 * SIZE, BB

	decl	%eax

	jne	.L04

	ALIGN_4

.L05:

	movl	C,  %esi

	movl	A,  %edx

	movl	M,  %ebx

	sarl	$1, %ebx

	jle	.L30

	ALIGN_4

.L10:

#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

	prefetchnta 4 * SIZE(%esi)

	prefetchnta 4 * SIZE(%esi, 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	$2, %eax

#else

	addl	$2, %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

.L11:

	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    .L11

	ALIGN_4

#endif

.L15:

#ifndef TRMMKERNEL

	movl	K, %eax

#else

	movl	KKK, %eax

#endif

	movaps	ALPHA_R, %xmm1

	movaps	ALPHA_I, %xmm3

	andl	$7, %eax		# if (k & 1)

	BRANCH

	je .L14

	ALIGN_4

.L13:

	mulps	%xmm0, %xmm2

	addps	%xmm2, %xmm4

	movshdup  0 * SIZE(BB), %xmm2

	mulps	%xmm0, %xmm2

	ADDSUB	%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

	ADDSUB	%xmm2, %xmm7

	movsldup  8 * SIZE(BB), %xmm2

	addl	$4 * SIZE, AA

	addl	$8 * SIZE, BB

	decl	%eax

	jg	.L13

	ALIGN_4

.L14:

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \

    defined(NR) || defined(NC) || defined(TR) || defined(TC)

	shufps	$0xb1, %xmm5, %xmm5

	shufps	$0xb1, %xmm7, %xmm7

	addsubps	%xmm5, %xmm4

	addsubps	%xmm7, %xmm6

	movaps	%xmm4, %xmm5

	movaps	%xmm6, %xmm7

	shufps	$0xb1, %xmm4, %xmm4

	shufps	$0xb1, %xmm6, %xmm6

#else

	shufps	$0xb1, %xmm4, %xmm4

	shufps	$0xb1, %xmm6, %xmm6

	addsubps	%xmm4, %xmm5

	addsubps	%xmm6, %xmm7

	movaps	%xmm5, %xmm4

	movaps	%xmm7, %xmm6

	shufps	$0xb1, %xmm5, %xmm5

	shufps	$0xb1, %xmm7, %xmm7

#endif

	mulps	%xmm1, %xmm5

	mulps	%xmm3, %xmm4

	mulps	%xmm1, %xmm7

	mulps	%xmm3, %xmm6

	addps	%xmm5, %xmm4

	addps	%xmm7, %xmm6

#ifndef TRMMKERNEL

	shufps	$0xe4, %xmm0, %xmm0

	movsd	0 * SIZE(%esi), %xmm0

	movhps	2 * SIZE(%esi), %xmm0

	shufps	$0xe4, %xmm2, %xmm2

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

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

	addps	%xmm0, %xmm4

	addps	%xmm2, %xmm6

#endif

	movsd	%xmm4, 0 * SIZE(%esi)

	movhps	%xmm4, 2 * SIZE(%esi)

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

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

#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, 4), BB

#endif

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

	addl	$2, KK

#endif

	addl	$4 * SIZE, %esi		# coffset += 4

	decl	%ebx			# i --

	jg	.L10

	ALIGN_4

.L30:

	movl	M,  %ebx

	andl	$1, %ebx

	jle	.L99

	ALIGN_4

.L40:

#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	$1, %eax

#else

	addl	$2, %eax

#endif

	movl	%eax, KKK

#endif

	sarl	$3, %eax

	je	.L42

	ALIGN_4

.L41:

	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	.L41

	ALIGN_4

.L42:

#ifndef TRMMKERNEL

	movl	K, %eax

#else

	movl	KKK, %eax

#endif

	movaps	ALPHA_R, %xmm1

	movaps	ALPHA_I, %xmm3

	andl	$7, %eax		# if (k & 1)

	BRANCH

	je .L44

	ALIGN_4

.L43:

	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	.L43

	ALIGN_4

.L44:

	movaps	%xmm4, %xmm6

	movlhps	%xmm5, %xmm4

	movhlps %xmm6, %xmm5

#if  defined(NR) || defined(NC) || defined(TR) || defined(TC) || \

     defined(RR) || defined(RC) || defined(CR) || defined(CC) 

	cmpeqps	%xmm7, %xmm7

	pslld	$31,   %xmm7

	xorps	%xmm7, %xmm5

#endif     

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \

    defined(NR) || defined(NC) || defined(TR) || defined(TC)

	shufps	$0xb1, %xmm5, %xmm5

	addsubps	%xmm5, %xmm4

	movaps	%xmm4, %xmm5

	shufps	$0xb1, %xmm4, %xmm4

#else

	shufps	$0xb1, %xmm4, %xmm4