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

/* 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"

#include "l2param.h"

#ifndef WINDOWS_ABI

#define STACKSIZE	64

#define OLD_INCX	 8 + STACKSIZE(%rsp)

#define OLD_Y		16 + STACKSIZE(%rsp)

#define OLD_INCY	24 + STACKSIZE(%rsp)

#define OLD_BUFFER	32 + STACKSIZE(%rsp)

#define ALPHA_R		48	      (%rsp)

#define ALPHA_I		56	      (%rsp)

#define M	  %rdi

#define N	  %rsi

#define A	  %rcx

#define LDA	  %r8

#define X	  %r9

#define INCX	  %rdx

#define Y	  %rbp

#define INCY	  %r10

#else

#define STACKSIZE	256

#define OLD_ALPHA_I	 40 + STACKSIZE(%rsp)

#define OLD_A		 48 + STACKSIZE(%rsp)

#define OLD_LDA		 56 + STACKSIZE(%rsp)

#define OLD_X		 64 + STACKSIZE(%rsp)

#define OLD_INCX	 72 + STACKSIZE(%rsp)

#define OLD_Y		 80 + STACKSIZE(%rsp)

#define OLD_INCY	 88 + STACKSIZE(%rsp)

#define OLD_BUFFER	 96 + STACKSIZE(%rsp)

#define ALPHA_R		224	       (%rsp)

#define ALPHA_I		232	       (%rsp)

#define M	  %rcx

#define N	  %rdx

#define A	  %r8

#define LDA	  %r9

#define X	  %rdi

#define INCX	  %rsi

#define Y	  %rbp

#define INCY	  %r10

#endif

#define I	%rax

#define A1	%r12

#define A2	%r13

#define Y1	%r14

#define BUFFER	%r15

#define J	%r11

#undef SUBPD

#if (!defined(CONJ) && !defined(XCONJ)) || (defined(CONJ) && defined(XCONJ))

#define SUBPD	   subpd

#else

#define SUBPD	   addpd

#endif

	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	OLD_A,     A

	movq	OLD_LDA,   LDA

	movq	OLD_X,     X

	movapd	%xmm3,       %xmm0

	movsd	OLD_ALPHA_I, %xmm1

#endif

	movq	OLD_INCX,  INCX

	movq	OLD_Y,     Y

	movq	OLD_INCY,  INCY

	movq	OLD_BUFFER, BUFFER

	salq	$ZBASE_SHIFT,   LDA

	salq	$ZBASE_SHIFT,   INCX

	salq	$ZBASE_SHIFT,   INCY

	movlpd	%xmm0, ALPHA_R

	movlpd	%xmm1, ALPHA_I

	subq	$-16 * SIZE, A

	testq	M, M

	jle	.L999

	testq	N, N

	jle	.L999

	ALIGN_3

	movq	BUFFER, Y1

	pxor	%xmm4, %xmm4

	movq	M,   %rax

	addq	$8,  %rax

	sarq	$3,  %rax

	ALIGN_3

.L01:

	movapd	%xmm4,  0 * SIZE(Y1)

	movapd	%xmm4,  2 * SIZE(Y1)

	movapd	%xmm4,  4 * SIZE(Y1)

	movapd	%xmm4,  6 * SIZE(Y1)

	movapd	%xmm4,  8 * SIZE(Y1)

	movapd	%xmm4, 10 * SIZE(Y1)

	movapd	%xmm4, 12 * SIZE(Y1)

	movapd	%xmm4, 14 * SIZE(Y1)

	subq	$-16 * SIZE, Y1

	decq	%rax

	jg	.L01

	ALIGN_3

.L10:

#ifdef ALIGNED_ACCESS

	testq	$SIZE, A

	jne	.L100

#endif

#if GEMV_UNROLL >= 4

	cmpq	$4, N

	jl	.L20

	ALIGN_3

.L11:

	subq	$4, N

	leaq	16 * SIZE(BUFFER), Y1

	movq	A,  A1

	leaq	(A,  LDA, 2), A2

	leaq	(A,  LDA, 4), A

	movsd	0 * SIZE(X), %xmm8

	movhpd	1 * SIZE(X), %xmm8

	addq	INCX, X

	movsd	0 * SIZE(X), %xmm10

	movhpd	1 * SIZE(X), %xmm10

	addq	INCX, X

	movsd	0 * SIZE(X), %xmm12

	movhpd	1 * SIZE(X), %xmm12

	addq	INCX, X

	movsd	0 * SIZE(X), %xmm14

	movhpd	1 * SIZE(X), %xmm14

	addq	INCX, X

	pcmpeqb	%xmm5, %xmm5

	psllq	$63,   %xmm5

	shufps	$0xc0, %xmm5, %xmm5	

	pshufd	$0x4e, %xmm8,  %xmm9

	pshufd	$0x4e, %xmm10, %xmm11

	pshufd	$0x4e, %xmm12, %xmm13

	pshufd	$0x4e, %xmm14, %xmm15

#ifdef HAVE_SSE3

	movddup	ALPHA_R, %xmm6

	movddup	ALPHA_I, %xmm7

#else

	movsd	ALPHA_R, %xmm6

	unpcklpd %xmm6, %xmm6

	movsd	ALPHA_I, %xmm7

	unpcklpd %xmm7, %xmm7

#endif

	xorpd	 %xmm5, %xmm9

	xorpd	 %xmm5, %xmm11

	xorpd	 %xmm5, %xmm13

	xorpd	 %xmm5, %xmm15

	mulpd	 %xmm6, %xmm8

	mulpd	 %xmm7, %xmm9

	mulpd	 %xmm6, %xmm10

	mulpd	 %xmm7, %xmm11

	mulpd	 %xmm6, %xmm12

	mulpd	 %xmm7, %xmm13

	mulpd	 %xmm6, %xmm14

	mulpd	 %xmm7, %xmm15

#ifndef XCONJ

	subpd	 %xmm9,  %xmm8

	subpd	 %xmm11, %xmm10

	subpd	 %xmm13, %xmm12

	subpd	 %xmm15, %xmm14

#else

	addpd	 %xmm9,  %xmm8

	addpd	 %xmm11, %xmm10

	addpd	 %xmm13, %xmm12

	addpd	 %xmm15, %xmm14

#endif

	pshufd	 $0xee, %xmm8,  %xmm9

	pshufd	 $0x44, %xmm8,  %xmm8

	pshufd	 $0xee, %xmm10, %xmm11

	pshufd	 $0x44, %xmm10, %xmm10

	pshufd	 $0xee, %xmm12, %xmm13

	pshufd	 $0x44, %xmm12, %xmm12

	pshufd	 $0xee, %xmm14, %xmm15

	pshufd	 $0x44, %xmm14, %xmm14

#ifndef CONJ

	xorpd	 %xmm5, %xmm9

	xorpd	 %xmm5, %xmm11

	xorpd	 %xmm5, %xmm13

	xorpd	 %xmm5, %xmm15

#else

	xorpd	 %xmm5, %xmm8

	xorpd	 %xmm5, %xmm10

	xorpd	 %xmm5, %xmm12

	xorpd	 %xmm5, %xmm14

#endif

	MOVUPS_YL1(-16 * SIZE, Y1, %xmm0)

	MOVUPS_YL1(-14 * SIZE, Y1, %xmm1)

	MOVUPS_YL1(-12 * SIZE, Y1, %xmm2)

	MOVUPS_YL1(-10 * SIZE, Y1, %xmm3)

	ALIGN_3

	movq	M,   I

	sarq	$2,  I

	jle	.L15

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)

	MOVUPS_A1(-14 * SIZE, A1, %xmm6)

	decq	 I

	jle	 .L14

	ALIGN_3

.L13:

#ifdef PREFETCH

	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A1)

#endif

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm8,  %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1(-12 * SIZE, A1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm8,  %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A1(-10 * SIZE, A1, %xmm6)

	mulpd	 %xmm9,  %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm9,  %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm8,  %xmm4

	addpd	 %xmm4,  %xmm2

	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm8,  %xmm6

	addpd	 %xmm6,  %xmm3

	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm6)

	mulpd	 %xmm9,  %xmm5

	SUBPD	 %xmm5,  %xmm2

	mulpd	 %xmm9,  %xmm7

	SUBPD	 %xmm7,  %xmm3

#ifdef PREFETCH

	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A1, LDA)

#endif

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm10, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A2(-12 * SIZE, A1, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm10, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A2(-10 * SIZE, A1, LDA, 1, %xmm6)

	mulpd	 %xmm11, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm11, %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm10, %xmm4

	addpd	 %xmm4,  %xmm2

	MOVUPS_A1(-16 * SIZE, A2, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm10, %xmm6

	addpd	 %xmm6,  %xmm3

	MOVUPS_A1(-14 * SIZE, A2, %xmm6)

	mulpd	 %xmm11, %xmm5

	SUBPD	 %xmm5,  %xmm2

	mulpd	 %xmm11, %xmm7

	SUBPD	 %xmm7,  %xmm3

#ifdef PREFETCH

	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A2)

#endif

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm12, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1(-12 * SIZE, A2, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm12, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A1(-10 * SIZE, A2, %xmm6)

	mulpd	 %xmm13, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm13, %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm12, %xmm4

	addpd	 %xmm4,  %xmm2

	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm12, %xmm6

	addpd	 %xmm6,  %xmm3

	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm6)

	mulpd	 %xmm13, %xmm5

	SUBPD	 %xmm5,  %xmm2

	mulpd	 %xmm13, %xmm7

	SUBPD	 %xmm7,  %xmm3

#ifdef PREFETCH

	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A2, LDA)

#endif

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm14, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A2(-12 * SIZE, A2, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm14, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A2(-10 * SIZE, A2, LDA, 1, %xmm6)

	mulpd	 %xmm15, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm15, %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm14, %xmm4

	addpd	 %xmm4,  %xmm2

	MOVUPS_A1( -8 * SIZE, A1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm14, %xmm6

	addpd	 %xmm6,  %xmm3

	MOVUPS_A1( -6 * SIZE, A1, %xmm6)

	mulpd	 %xmm15, %xmm5

	SUBPD	 %xmm5,  %xmm2

	mulpd	 %xmm15, %xmm7

	SUBPD	 %xmm7,  %xmm3

#ifdef PREFETCHW

	PREFETCHW	(PREFETCHSIZE) - 128 + PREOFFSET(Y1)

#endif

	MOVUPS_YS1(-16 * SIZE, Y1, %xmm0)

	MOVUPS_YS1(-14 * SIZE, Y1, %xmm1)

	MOVUPS_YS1(-12 * SIZE, Y1, %xmm2)

	MOVUPS_YS1(-10 * SIZE, Y1, %xmm3)

	MOVUPS_YL1( -8 * SIZE, Y1, %xmm0)

	MOVUPS_YL1( -6 * SIZE, Y1, %xmm1)

	MOVUPS_YL1( -4 * SIZE, Y1, %xmm2)

	MOVUPS_YL1( -2 * SIZE, Y1, %xmm3)

	subq	 $-8 * SIZE, A1

	subq	 $-8 * SIZE, A2

	subq	 $-8 * SIZE, Y1

	subq	 $1, I

	BRANCH

	jg	.L13

	ALIGN_3

.L14:

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm8,  %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1(-12 * SIZE, A1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm8,  %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A1(-10 * SIZE, A1, %xmm6)

	mulpd	 %xmm9,  %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm9,  %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm8,  %xmm4

	addpd	 %xmm4,  %xmm2

	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm8,  %xmm6

	addpd	 %xmm6,  %xmm3

	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm6)

	mulpd	 %xmm9,  %xmm5

	SUBPD	 %xmm5,  %xmm2

	mulpd	 %xmm9,  %xmm7

	SUBPD	 %xmm7,  %xmm3

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm10, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A2(-12 * SIZE, A1, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm10, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A2(-10 * SIZE, A1, LDA, 1, %xmm6)

	mulpd	 %xmm11, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm11, %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm10, %xmm4

	addpd	 %xmm4,  %xmm2

	MOVUPS_A1(-16 * SIZE, A2, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm10, %xmm6

	addpd	 %xmm6,  %xmm3

	MOVUPS_A1(-14 * SIZE, A2, %xmm6)

	mulpd	 %xmm11, %xmm5

	SUBPD	 %xmm5,  %xmm2

	mulpd	 %xmm11, %xmm7

	SUBPD	 %xmm7,  %xmm3

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm12, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1(-12 * SIZE, A2, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm12, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A1(-10 * SIZE, A2, %xmm6)

	mulpd	 %xmm13, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm13, %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm12, %xmm4

	addpd	 %xmm4,  %xmm2

	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm12, %xmm6

	addpd	 %xmm6,  %xmm3

	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm6)

	mulpd	 %xmm13, %xmm5

	SUBPD	 %xmm5,  %xmm2

	mulpd	 %xmm13, %xmm7

	SUBPD	 %xmm7,  %xmm3

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm14, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A2(-12 * SIZE, A2, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm14, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A2(-10 * SIZE, A2, LDA, 1, %xmm6)

	mulpd	 %xmm15, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm15, %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm14, %xmm4

	addpd	 %xmm4,  %xmm2

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm14, %xmm6

	addpd	 %xmm6,  %xmm3

	mulpd	 %xmm15, %xmm5

	SUBPD	 %xmm5,  %xmm2

	mulpd	 %xmm15, %xmm7

	SUBPD	 %xmm7,  %xmm3

	MOVUPS_YS1(-16 * SIZE, Y1, %xmm0)

	MOVUPS_YS1(-14 * SIZE, Y1, %xmm1)

	MOVUPS_YS1(-12 * SIZE, Y1, %xmm2)

	MOVUPS_YS1(-10 * SIZE, Y1, %xmm3)

	MOVUPS_YL1( -8 * SIZE, Y1, %xmm0)

	MOVUPS_YL1( -6 * SIZE, Y1, %xmm1)

	MOVUPS_YL1( -4 * SIZE, Y1, %xmm2)

	MOVUPS_YL1( -2 * SIZE, Y1, %xmm3)

	subq	 $-8 * SIZE, A1

	subq	 $-8 * SIZE, A2

	subq	 $-8 * SIZE, Y1

	ALIGN_3

.L15:

	testq	$2, M

	je	.L17

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)

	MOVUPS_A1(-14 * SIZE, A1, %xmm6)

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm8,  %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm8,  %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm6)

	mulpd	 %xmm9,  %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm9,  %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm10, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1(-16 * SIZE, A2, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm10, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A1(-14 * SIZE, A2, %xmm6)

	mulpd	 %xmm11, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm11, %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm12, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm4)

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm12, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm6)

	mulpd	 %xmm13, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm13, %xmm7

	SUBPD	 %xmm7,  %xmm1

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm14, %xmm4

	addpd	 %xmm4,  %xmm0

	mulpd	 %xmm15, %xmm5

	SUBPD	 %xmm5,  %xmm0

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm14, %xmm6

	addpd	 %xmm6,  %xmm1

	mulpd	 %xmm15, %xmm7

	SUBPD	 %xmm7,  %xmm1

	MOVUPS_YS1(-16 * SIZE, Y1, %xmm0)

	MOVUPS_YS1(-14 * SIZE, Y1, %xmm1)

	movapd	 %xmm2, %xmm0

	addq	 $4 * SIZE, A1

	addq	 $4 * SIZE, A2

	addq	 $4 * SIZE, Y1

	ALIGN_3

.L17:

	testq	$1, M

	je	.L19

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)

	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm6)

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm8,  %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1(-16 * SIZE, A2, %xmm4)

	mulpd	 %xmm9,  %xmm5

	SUBPD	 %xmm5,  %xmm0

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm10, %xmm6

	addpd	 %xmm6,  %xmm0

	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm6)

	mulpd	 %xmm11, %xmm7

	SUBPD	 %xmm7,  %xmm0

	pshufd	 $0x4e, %xmm4, %xmm5

	mulpd	 %xmm12, %xmm4

	addpd	 %xmm4,  %xmm0

	mulpd	 %xmm13, %xmm5

	SUBPD	 %xmm5,  %xmm0

	pshufd	 $0x4e, %xmm6, %xmm7

	mulpd	 %xmm14, %xmm6

	addpd	 %xmm6,  %xmm0

	mulpd	 %xmm15, %xmm7

	SUBPD	 %xmm7,  %xmm0

	MOVUPS_YS1(-16 * SIZE, Y1, %xmm0)

	ALIGN_3

.L19:

	cmpq	$4, N

	jge	.L11

	ALIGN_3

.L20:

#endif

#if GEMV_UNROLL >= 2

	cmpq	$2, N

	jl	.L30

#if GEMV_UNROLL == 2

	ALIGN_3

.L21:

#endif

	subq	$2, N

	leaq	16 * SIZE(BUFFER), Y1

	movq	A,  A1

	leaq	(A,  LDA, 1), A2

	leaq	(A,  LDA, 2), A

	movsd	0 * SIZE(X), %xmm12

	movhpd	1 * SIZE(X), %xmm12

	addq	INCX, X

	movsd	0 * SIZE(X), %xmm14

	movhpd	1 * SIZE(X), %xmm14

	addq	INCX, X

	pcmpeqb	%xmm11, %xmm11

	psllq	$63,    %xmm11

	shufps	$0xc0, %xmm11, %xmm11	

	pshufd	$0x4e, %xmm12, %xmm13

	pshufd	$0x4e, %xmm14, %xmm15

#ifdef HAVE_SSE3

	movddup	ALPHA_R, %xmm8

	movddup	ALPHA_I, %xmm9

#else

	movsd	ALPHA_R, %xmm8

	unpcklpd %xmm8, %xmm8

	movsd	ALPHA_I, %xmm9

	unpcklpd %xmm9, %xmm9

#endif

	xorpd	 %xmm11, %xmm13

	xorpd	 %xmm11, %xmm15

	mulpd	 %xmm8, %xmm12

	mulpd	 %xmm9, %xmm13

	mulpd	 %xmm8, %xmm14

	mulpd	 %xmm9, %xmm15

#ifndef XCONJ

	subpd	 %xmm13, %xmm12

	subpd	 %xmm15, %xmm14

#else

	addpd	 %xmm13, %xmm12

	addpd	 %xmm15, %xmm14

#endif

	pshufd	 $0xee, %xmm12, %xmm13

	pshufd	 $0x44, %xmm12, %xmm12

	pshufd	 $0xee, %xmm14, %xmm15

	pshufd	 $0x44, %xmm14, %xmm14

#ifndef CONJ

	xorpd	 %xmm11, %xmm13

	xorpd	 %xmm11, %xmm15

#else

	xorpd	 %xmm11, %xmm12

	xorpd	 %xmm11, %xmm14

#endif

	MOVUPS_YL1(-16 * SIZE, Y1, %xmm0)

	MOVUPS_YL1(-14 * SIZE, Y1, %xmm1)

	MOVUPS_YL1(-12 * SIZE, Y1, %xmm2)

	MOVUPS_YL1(-10 * SIZE, Y1, %xmm3)

	ALIGN_3

	movq	M,   I

	sarq	$2,  I

	jle	.L25

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)

	MOVUPS_A1(-14 * SIZE, A1, %xmm6)

	MOVUPS_A1(-12 * SIZE, A1, %xmm8)

	MOVUPS_A1(-10 * SIZE, A1, %xmm10)

	decq	 I

	jle	 .L24

	ALIGN_3

.L23:

#ifdef PREFETCH

	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(A2)

#endif

	pshufd	 $0x4e, %xmm4,  %xmm5

	mulpd	 %xmm12, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1(-16 * SIZE, A2, %xmm4)

	pshufd	 $0x4e, %xmm6,  %xmm7

	mulpd	 %xmm12, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A1(-14 * SIZE, A2, %xmm6)

	pshufd	 $0x4e, %xmm8,  %xmm9

	mulpd	 %xmm12, %xmm8

	addpd	 %xmm8,  %xmm2

	MOVUPS_A1(-12 * SIZE, A2, %xmm8)

	pshufd	 $0x4e, %xmm10, %xmm11

	mulpd	 %xmm12, %xmm10

	addpd	 %xmm10, %xmm3

	MOVUPS_A1(-10 * SIZE, A2, %xmm10)

	mulpd	 %xmm13, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm13, %xmm7

	SUBPD	 %xmm7,  %xmm1

	mulpd	 %xmm13, %xmm9

	SUBPD	 %xmm9,  %xmm2

	mulpd	 %xmm13, %xmm11

	SUBPD	 %xmm11, %xmm3

#ifdef PREFETCH

	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(A1)

#endif

	pshufd	 $0x4e, %xmm4,  %xmm5

	mulpd	 %xmm14, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1( -8 * SIZE, A1, %xmm4)

	pshufd	 $0x4e, %xmm6,  %xmm7

	mulpd	 %xmm14, %xmm6

	addpd	 %xmm6,  %xmm1

	MOVUPS_A1( -6 * SIZE, A1, %xmm6)

	pshufd	 $0x4e, %xmm8,  %xmm9

	mulpd	 %xmm14, %xmm8

	addpd	 %xmm8,  %xmm2

	MOVUPS_A1( -4 * SIZE, A1, %xmm8)

	pshufd	 $0x4e, %xmm10, %xmm11

	mulpd	 %xmm14, %xmm10

	addpd	 %xmm10, %xmm3

	MOVUPS_A1( -2 * SIZE, A1, %xmm10)

	mulpd	 %xmm15, %xmm5

	SUBPD	 %xmm5,  %xmm0

	mulpd	 %xmm15, %xmm7

	SUBPD	 %xmm7,  %xmm1

	mulpd	 %xmm15, %xmm9

	SUBPD	 %xmm9,  %xmm2

	mulpd	 %xmm15, %xmm11

	SUBPD	 %xmm11, %xmm3

#ifdef PREFETCHW

	PREFETCHW	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(Y1)

#endif

	MOVUPS_YS1(-16 * SIZE, Y1, %xmm0)

	MOVUPS_YS1(-14 * SIZE, Y1, %xmm1)

	MOVUPS_YS1(-12 * SIZE, Y1, %xmm2)

	MOVUPS_YS1(-10 * SIZE, Y1, %xmm3)

	MOVUPS_YL1( -8 * SIZE, Y1, %xmm0)

	MOVUPS_YL1( -6 * SIZE, Y1, %xmm1)

	MOVUPS_YL1( -4 * SIZE, Y1, %xmm2)

	MOVUPS_YL1( -2 * SIZE, Y1, %xmm3)

	subq	 $-8 * SIZE, A1

	subq	 $-8 * SIZE, A2

	subq	 $-8 * SIZE, Y1

	subq	 $1, I

	BRANCH

	jg	.L23

	ALIGN_3

.L24:

	pshufd	 $0x4e, %xmm4,  %xmm5

	mulpd	 %xmm12, %xmm4

	addpd	 %xmm4,  %xmm0

	MOVUPS_A1(-16 * SIZE, A2, %xmm4)