/*********************************************************************/
/* 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"

#if GEMV_UNROLL < 2
#undef  GEMV_UNROLL
#define GEMV_UNROLL 2
#endif

#ifndef WINDOWS_ABI

#define STACKSIZE	128

#define OLD_M	  %rdi
#define OLD_N	  %rsi
#define OLD_A	  %rcx
#define OLD_LDA	  %r8
#define STACK_INCX	 8 + STACKSIZE(%rsp)
#define STACK_Y		16 + STACKSIZE(%rsp)
#define STACK_INCY	24 + STACKSIZE(%rsp)
#define STACK_BUFFER	32 + STACKSIZE(%rsp)
#define MMM	56(%rsp)
#define NN	64(%rsp)
#define AA	72(%rsp)
#define LDAX	80(%rsp)
#else

#define STACKSIZE	256

#define OLD_M	  %rcx
#define OLD_N	  %rdx
#define OLD_A		 40 + STACKSIZE(%rsp)
#define OLD_LDA		 48 + STACKSIZE(%rsp)
#define OLD_X		 56 + STACKSIZE(%rsp)
#define STACK_INCX	 64 + STACKSIZE(%rsp)
#define STACK_Y		 72 + STACKSIZE(%rsp)
#define STACK_INCY	 80 + STACKSIZE(%rsp)
#define STACK_BUFFER	 88 + STACKSIZE(%rsp)
//Temp variables for M,N,A,LDA
#define MMM	224(%rsp)
#define NN	232(%rsp)
#define AA	240(%rsp)
#define LDAX	248(%rsp)

#endif

#define LDA	%r8
#define X	%r9

#define INCX	%rsi
#define INCY	%rdi

#define M	%r10
#define N	%r11
#define A	%r12
#define Y	%r14
#define BUFFER	%r13

#define I	%rax
#define A1	%rbx
#define A2	%rcx
#define LDA3	%rdx
#define Y1	%rbp
#define X1	%r15

#ifdef ALIGNED_ACCESS
#define MM	INCX
#else
#define MM	M
#endif

#define ALPHA	%xmm15

	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_M,	      M
	movq	OLD_N,        N
	movq	OLD_A,        A
	movq	OLD_LDA,      LDA
	movq	OLD_X,        X

	movq	M,	      MMM
	movq	N,            NN
	movq	A,            AA
	movq	LDA,	      LDAX

#else
	movq	OLD_M,	      MMM
	movq	OLD_N,        NN
	movq	OLD_A,        AA
	movq	OLD_LDA,      LDAX
#endif
#ifdef HAVE_SSE3
#ifndef WINDOWS_ABI
	movddup	%xmm0, ALPHA
#else
	movddup	%xmm3, ALPHA
#endif
#else
#ifndef WINDOWS_ABI
	movapd	%xmm0, ALPHA
#else
	movapd	%xmm3, ALPHA
#endif
	unpcklpd ALPHA, ALPHA
#endif



.L0x:
	xorq	M,M
	addq	$1,M
	salq	$21,M
	subq	M,MMM
	jge .L00

	movq	MMM,%rax
	addq	M,%rax
	jle	.L999x
	movq	%rax,M

.L00:
	movq	LDAX,LDA
	movq	NN,N
	movq	AA,A
	movq	STACK_INCX,   INCX
	movq	STACK_Y,      Y
	movq	STACK_INCY,   INCY
	movq	STACK_BUFFER, BUFFER

	leaq	-1(INCX), %rax

	leaq	(,LDA,  SIZE), LDA
	leaq	(,INCX, SIZE), INCX
	leaq	(,INCY, SIZE), INCY

	leaq	(LDA, LDA, 2), LDA3

	subq	$-16 * SIZE, A

	testq	M, M
	jle	.L999
	testq	N, N
	jle	.L999

	movq	BUFFER, X1

#ifdef ALIGNED_ACCESS
	testq	$SIZE, A
	je	.L01

	movsd	(X), %xmm0
	addq	INCX, X

	movsd	%xmm0, 1 * SIZE(BUFFER)
	addq	$1 * SIZE, BUFFER
	addq	$2 * SIZE, X1
	decq	M
	jle	.L10
	ALIGN_4

.L01:
#endif

	movq	M,  I
	sarq	$3, I
	jle	.L05
	ALIGN_4

.L02:
	movsd	(X), %xmm0
	addq	INCX, X
	movhpd	(X), %xmm0
	addq	INCX, X

	movsd	(X), %xmm1
	addq	INCX, X
	movhpd	(X), %xmm1
	addq	INCX, X

	movsd	(X), %xmm2
	addq	INCX, X
	movhpd	(X), %xmm2
	addq	INCX, X

	movsd	(X), %xmm3
	addq	INCX, X
	movhpd	(X), %xmm3
	addq	INCX, X

	movapd	%xmm0, 0 * SIZE(X1)
	movapd	%xmm1, 2 * SIZE(X1)
	movapd	%xmm2, 4 * SIZE(X1)
	movapd	%xmm3, 6 * SIZE(X1)

	addq	$8 * SIZE, X1
	decq	I
	jg	.L02
	ALIGN_4

.L05:
	movq	M,  I
	andq	$7, I
	jle	.L10
	ALIGN_2

.L06:
	movsd	(X), %xmm0
	addq	INCX, X
	movsd	%xmm0, 0 * SIZE(X1)
	addq	$SIZE, X1
	decq	I
	jg	.L06
	ALIGN_4

.L10:
	movq	Y, Y1

#ifdef ALIGNED_ACCESS
	testq	$SIZE, LDA
	jne	.L50
#endif

#if GEMV_UNROLL >= 8
	cmpq	$8, N
	jl	.L20
	ALIGN_3

.L11:
	subq	$8, N

	leaq	16 * SIZE(BUFFER), X1

	movq	A, A1
	leaq	(A1, LDA, 4), A2
	leaq	(A1, LDA, 8), A

	xorps	%xmm0, %xmm0
	xorps	%xmm1, %xmm1
	xorps	%xmm2, %xmm2
	xorps	%xmm3, %xmm3
	xorps	%xmm4, %xmm4
	xorps	%xmm5, %xmm5
	xorps	%xmm6, %xmm6
	xorps	%xmm7, %xmm7

#ifdef PREFETCHW
	PREFETCHW	7 * SIZE(Y1)
#endif

#ifdef ALIGNED_ACCESS
	testq	$SIZE, A
	je	.L1X

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm0
	movsd	-16 * SIZE(A1, LDA), %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm1
	movsd	-16 * SIZE(A1, LDA, 2), %xmm10
	mulsd	%xmm12, %xmm10
	addsd	%xmm10, %xmm2
	movsd	-16 * SIZE(A1, LDA3), %xmm11
	mulsd	%xmm12, %xmm11
	addsd	%xmm11, %xmm3
	movsd	-16 * SIZE(A2), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm4
	movsd	-16 * SIZE(A2, LDA), %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm5
	movsd	-16 * SIZE(A2, LDA, 2), %xmm10
	mulsd	%xmm12, %xmm10
	addsd	%xmm10, %xmm6
	movsd	-16 * SIZE(A2, LDA3), %xmm11
	mulsd	%xmm12, %xmm11
	addsd	%xmm11, %xmm7

	addq	 $SIZE, A1
	addq	 $SIZE, A2
	addq	 $SIZE, X1
	ALIGN_3

.L1X:
#endif

	movq	M,  I
	sarq	$3,  I
	jle	.L15

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm9)
	MOVUPS_A2(-16 * SIZE, A1, LDA, 2, %xmm10)
	MOVUPS_A2(-16 * SIZE, A1, LDA3, 1, %xmm11)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	decq	I
	jle	.L13
	ALIGN_4

.L12:
#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A1)
#endif

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-16 * SIZE, A2, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-16 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-16 * SIZE, A2, LDA3, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A1, LDA)
#endif

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm4
	MOVUPS_A1(-14 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm5
	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm6
	MOVUPS_A2(-14 * SIZE, A1, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm11, %xmm7
	MOVUPS_A2(-14 * SIZE, A1, LDA3, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A1, LDA, 2)
#endif

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-14 * SIZE, A2, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-14 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-14 * SIZE, A2, LDA3, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A1, LDA3)
#endif

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm4
	MOVUPS_A1(-12 * SIZE, A1, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm5
	MOVUPS_A2(-12 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm6
	MOVUPS_A2(-12 * SIZE, A1, LDA, 2, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm7
	MOVUPS_A2(-12 * SIZE, A1, LDA3, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A2)
#endif

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-12 * SIZE, A2, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-12 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-12 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-12 * SIZE, A2, LDA3, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A2, LDA)
#endif

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm4
	MOVUPS_A1(-10 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm5
	MOVUPS_A2(-10 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm6
	MOVUPS_A2(-10 * SIZE, A1, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	MOVUPS_XL1(-8 * SIZE, X1, %xmm12)
	addpd	%xmm11, %xmm7
	MOVUPS_A2(-10 * SIZE, A1, LDA3, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A2, LDA, 2)
#endif

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-10 * SIZE, A2, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-10 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-10 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-10 * SIZE, A2, LDA3, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(A2, LDA3)
#endif

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm4
	MOVUPS_A1(-8 * SIZE, A1, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm5
	MOVUPS_A2(-8 * SIZE, A1, LDA, 1, %xmm9)

#ifdef PREFETCHW
	PREFETCH	(PREFETCHSIZE) - 128 + PREOFFSET(X1)
#endif

	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm6
	MOVUPS_A2(-8 * SIZE, A1, LDA, 2, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1(-6 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm7
	MOVUPS_A2(-8 * SIZE, A1, LDA3, 1, %xmm11)

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1

	decq	I
	jg	.L12
	ALIGN_4

.L13:
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-16 * SIZE, A2, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-16 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-16 * SIZE, A2, LDA3, 1, %xmm11)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm4
	MOVUPS_A1(-14 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm5
	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm6
	MOVUPS_A2(-14 * SIZE, A1, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm11, %xmm7
	MOVUPS_A2(-14 * SIZE, A1, LDA3, 1, %xmm11)

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-14 * SIZE, A2, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-14 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-14 * SIZE, A2, LDA3, 1, %xmm11)

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm4
	MOVUPS_A1(-12 * SIZE, A1, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm5
	MOVUPS_A2(-12 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm6
	MOVUPS_A2(-12 * SIZE, A1, LDA, 2, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm7
	MOVUPS_A2(-12 * SIZE, A1, LDA3, 1, %xmm11)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-12 * SIZE, A2, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-12 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-12 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-12 * SIZE, A2, LDA3, 1, %xmm11)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm4
	MOVUPS_A1(-10 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm5
	MOVUPS_A2(-10 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm6
	MOVUPS_A2(-10 * SIZE, A1, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm7
	MOVUPS_A2(-10 * SIZE, A1, LDA3, 1, %xmm11)

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-10 * SIZE, A2, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-10 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-10 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-10 * SIZE, A2, LDA3, 1, %xmm11)

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm4
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm5
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm6
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm7

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1
	ALIGN_4

.L15:
	testq	$4, M
	jle	.L16

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm9)
	MOVUPS_A2(-16 * SIZE, A1, LDA, 2, %xmm10)
	MOVUPS_A2(-16 * SIZE, A1, LDA3, 1, %xmm11)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-16 * SIZE, A2, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-16 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-16 * SIZE, A2, LDA3, 1, %xmm11)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm4
	MOVUPS_A1(-14 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm5
	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm6
	MOVUPS_A2(-14 * SIZE, A1, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm7
	MOVUPS_A2(-14 * SIZE, A1, LDA3, 1, %xmm11)

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-14 * SIZE, A2, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-14 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-14 * SIZE, A2, LDA3, 1, %xmm11)

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm4
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm5
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm6
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm7

	addq	$4 * SIZE, A1
	addq	$4 * SIZE, A2
	addq	$4 * SIZE, X1
	ALIGN_4

.L16:
	testq	$2, M
	jle	.L17

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm9)
	MOVUPS_A2(-16 * SIZE, A1, LDA, 2, %xmm10)
	MOVUPS_A2(-16 * SIZE, A1, LDA3, 1, %xmm11)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-16 * SIZE, A2, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-16 * SIZE, A2, LDA, 2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-16 * SIZE, A2, LDA3, 1, %xmm11)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm4
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm5
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm6
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm7

	addq	$2 * SIZE, A1
	addq	$2 * SIZE, A2
	addq	$2 * SIZE, X1
	ALIGN_4

.L17:
	testq	$1, M
	je	.L18

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm0
	movsd	-16 * SIZE(A1, LDA), %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm1
	movsd	-16 * SIZE(A1, LDA, 2), %xmm10
	mulsd	%xmm12, %xmm10
	addsd	%xmm10, %xmm2
	movsd	-16 * SIZE(A1, LDA3), %xmm11
	mulsd	%xmm12, %xmm11
	addsd	%xmm11, %xmm3
	movsd	-16 * SIZE(A2), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm4
	movsd	-16 * SIZE(A2, LDA), %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm5
	movsd	-16 * SIZE(A2, LDA, 2), %xmm10
	mulsd	%xmm12, %xmm10
	addsd	%xmm10, %xmm6
	movsd	-16 * SIZE(A2, LDA3), %xmm11
	mulsd	%xmm12, %xmm11
	addsd	%xmm11, %xmm7
	ALIGN_4

.L18:
#ifdef HAVE_SSE3
	haddpd	%xmm1, %xmm0
	haddpd	%xmm3, %xmm2
	haddpd	%xmm5, %xmm4
	haddpd	%xmm7, %xmm6
#else
	movapd	%xmm0, %xmm8
	unpcklpd %xmm1, %xmm0
	unpckhpd %xmm1, %xmm8

	movapd	%xmm2, %xmm9
	unpcklpd %xmm3, %xmm2
	unpckhpd %xmm3, %xmm9

	movapd	%xmm4, %xmm10
	unpcklpd %xmm5, %xmm4
	unpckhpd %xmm5, %xmm10

	movapd	%xmm6, %xmm11
	unpcklpd %xmm7, %xmm6
	unpckhpd %xmm7, %xmm11

	addpd	%xmm8,  %xmm0
	addpd	%xmm9,  %xmm2
	addpd	%xmm10, %xmm4
	addpd	%xmm11, %xmm6
#endif

	mulpd	ALPHA, %xmm0
	mulpd	ALPHA, %xmm2
	mulpd	ALPHA, %xmm4
	mulpd	ALPHA, %xmm6

	cmpq	$SIZE, INCY
	jne	.L19

	movsd	 0 * SIZE(Y), %xmm8
	movhpd	 1 * SIZE(Y), %xmm8
	movsd	 2 * SIZE(Y), %xmm9
	movhpd	 3 * SIZE(Y), %xmm9
	movsd	 4 * SIZE(Y), %xmm10
	movhpd	 5 * SIZE(Y), %xmm10
	movsd	 6 * SIZE(Y), %xmm11
	movhpd	 7 * SIZE(Y), %xmm11
	addq	$8 * SIZE, Y

	addpd	%xmm8, %xmm0
	addpd	%xmm9,  %xmm2
	addpd	%xmm10, %xmm4
	addpd	%xmm11, %xmm6

	movlpd	%xmm0,  0 * SIZE(Y1)
	movhpd	%xmm0,  1 * SIZE(Y1)
	movlpd	%xmm2,  2 * SIZE(Y1)
	movhpd	%xmm2,  3 * SIZE(Y1)
	movlpd	%xmm4,  4 * SIZE(Y1)
	movhpd	%xmm4,  5 * SIZE(Y1)
	movlpd	%xmm6,  6 * SIZE(Y1)
	movhpd	%xmm6,  7 * SIZE(Y1)
	addq	$8 * SIZE, Y1

	cmpq	$8, N
	jge	.L11
	jmp	.L20
	ALIGN_4

.L19:
	movsd	(Y), %xmm8
	addq	INCY, Y
	movhpd	(Y), %xmm8
	addq	INCY, Y
	movsd	(Y), %xmm9
	addq	INCY, Y
	movhpd	(Y), %xmm9
	addq	INCY, Y
	movsd	(Y), %xmm10
	addq	INCY, Y
	movhpd	(Y), %xmm10
	addq	INCY, Y
	movsd	(Y), %xmm11
	addq	INCY, Y
	movhpd	(Y), %xmm11
	addq	INCY, Y

	addpd	%xmm8,  %xmm0
	addpd	%xmm9,  %xmm2
	addpd	%xmm10, %xmm4
	addpd	%xmm11, %xmm6

	movlpd	%xmm0, (Y1)
	addq	INCY, Y1
	movhpd	%xmm0, (Y1)
	addq	INCY, Y1
	movlpd	%xmm2, (Y1)
	addq	INCY, Y1
	movhpd	%xmm2, (Y1)
	addq	INCY, Y1
	movlpd	%xmm4, (Y1)
	addq	INCY, Y1
	movhpd	%xmm4, (Y1)
	addq	INCY, Y1
	movlpd	%xmm6, (Y1)
	addq	INCY, Y1
	movhpd	%xmm6, (Y1)
	addq	INCY, Y1

	cmpq	$8, N
	jge	.L11
	ALIGN_4

.L20:
#endif

#if GEMV_UNROLL >= 4

	cmpq	$4, N
	jl	.L30

#if GEMV_UNROLL == 4
	ALIGN_3

.L21:
#endif
	subq	$4, N

	leaq	16 * SIZE(BUFFER), X1

	movq	A, A1
	leaq	(A1, LDA, 2), A2
	leaq	(A1, LDA, 4), A

	xorps	%xmm0, %xmm0
	xorps	%xmm1, %xmm1
	xorps	%xmm2, %xmm2
	xorps	%xmm3, %xmm3

#if (GEMV_UNROLL == 4 ) && defined(PREFETCHW)
	PREFETCHW	3 * SIZE(Y1)
#endif

#ifdef ALIGNED_ACCESS
	testq	$SIZE, A
	je	.L2X

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm0
	movsd	-16 * SIZE(A1, LDA), %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm1
	movsd	-16 * SIZE(A2), %xmm10
	mulsd	%xmm12, %xmm10
	addsd	%xmm10, %xmm2
	movsd	-16 * SIZE(A2, LDA), %xmm11
	mulsd	%xmm12, %xmm11
	addsd	%xmm11, %xmm3

	addq	 $SIZE, A1
	addq	 $SIZE, A2
	addq	 $SIZE, X1
	ALIGN_3

.L2X:
#endif

	movq	M,  I
	sarq	$3,  I
	jle	.L25

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm9)
	MOVUPS_A1(-16 * SIZE, A2, %xmm10)
	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm11)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	decq	I
	jle	.L23
	ALIGN_4

.L22:
#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(A1)
#endif

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-14 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-14 * SIZE, A2, %xmm10)
	mulpd	%xmm12, %xmm11
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(A1, LDA)
#endif

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-12 * SIZE, A1, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-12 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-12 * SIZE, A2, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-12 * SIZE, A2, LDA, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(A2)
#endif

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-10 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-10 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-10 * SIZE, A2, %xmm10)
	mulpd	%xmm12, %xmm11
	MOVUPS_XL1( -8 * SIZE, X1, %xmm12)
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-10 * SIZE, A2, LDA, 1, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(A2, LDA)
#endif

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1( -8 * SIZE, A1, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2( -8 * SIZE, A1, LDA, 1, %xmm9)

#ifdef PREFETCHW
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(X1)
#endif

	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1( -8 * SIZE, A2, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1( -6 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm3
	MOVUPS_A2( -8 * SIZE, A2, LDA, 1, %xmm11)

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1

	decq	I
	jg	.L22
	ALIGN_4

.L23:
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-14 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-14 * SIZE, A2, %xmm10)
	mulpd	%xmm12, %xmm11
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm11)

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-12 * SIZE, A1, %xmm8)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-12 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-12 * SIZE, A2, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-12 * SIZE, A2, LDA, 1, %xmm11)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-10 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A2(-10 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-10 * SIZE, A2, %xmm10)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3
	MOVUPS_A2(-10 * SIZE, A2, LDA, 1, %xmm11)

	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1
	ALIGN_4

.L25:
	testq	$4, M
	jle	.L26

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A1(-16 * SIZE, A2, %xmm10)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm11)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3

	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	MOVUPS_A1(-14 * SIZE, A1, %xmm8)
	mulpd	%xmm13, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A2(-14 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A1(-14 * SIZE, A2, %xmm10)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-14 * SIZE, A2, LDA, 1, %xmm11)
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3

	addq	$4 * SIZE, A1
	addq	$4 * SIZE, A2
	addq	$4 * SIZE, X1
	ALIGN_4

.L26:
	testq	$2, M
	jle	.L27

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A2(-16 * SIZE, A1, LDA, 1, %xmm9)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1
	MOVUPS_A1(-16 * SIZE, A2, %xmm10)
	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A2(-16 * SIZE, A2, LDA, 1, %xmm11)
	mulpd	%xmm12, %xmm11
	addpd	%xmm11, %xmm3

	addq	$2 * SIZE, A1
	addq	$2 * SIZE, A2
	addq	$2 * SIZE, X1
	ALIGN_4

.L27:
	testq	$1, M
	je	.L28

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm0
	movsd	-16 * SIZE(A1, LDA), %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm1
	movsd	-16 * SIZE(A2), %xmm10
	mulsd	%xmm12, %xmm10
	addsd	%xmm10, %xmm2
	movsd	-16 * SIZE(A2, LDA), %xmm11
	mulsd	%xmm12, %xmm11
	addsd	%xmm11, %xmm3
	ALIGN_4

.L28:
#ifdef HAVE_SSE3
	haddpd	%xmm1, %xmm0
	haddpd	%xmm3, %xmm2
#else
	movapd	%xmm0, %xmm8
	unpcklpd %xmm1, %xmm0
	unpckhpd %xmm1, %xmm8

	movapd	%xmm2, %xmm9
	unpcklpd %xmm3, %xmm2
	unpckhpd %xmm3, %xmm9

	addpd	%xmm8, %xmm0
	addpd	%xmm9, %xmm2
#endif

	mulpd	ALPHA, %xmm0
	mulpd	ALPHA, %xmm2

	cmpq	$SIZE, INCY
	jne	.L29

	movsd	 0 * SIZE(Y), %xmm4
	movhpd	 1 * SIZE(Y), %xmm4
	movsd	 2 * SIZE(Y), %xmm5
	movhpd	 3 * SIZE(Y), %xmm5
	addq	$4 * SIZE, Y

	addpd	%xmm4, %xmm0
	addpd	%xmm5, %xmm2

	movlpd	%xmm0,  0 * SIZE(Y1)
	movhpd	%xmm0,  1 * SIZE(Y1)
	movlpd	%xmm2,  2 * SIZE(Y1)
	movhpd	%xmm2,  3 * SIZE(Y1)
	addq	$4 * SIZE, Y1

#if GEMV_UNROLL == 4
	cmpq	$4, N
	jge	.L21
#endif
	jmp	.L30
	ALIGN_4

.L29:
	movsd	(Y), %xmm4
	addq	INCY, Y
	movhpd	(Y), %xmm4
	addq	INCY, Y
	movsd	(Y), %xmm5
	addq	INCY, Y
	movhpd	(Y), %xmm5
	addq	INCY, Y

	addpd	%xmm4, %xmm0
	addpd	%xmm5, %xmm2

	movlpd	%xmm0, (Y1)
	addq	INCY, Y1
	movhpd	%xmm0, (Y1)
	addq	INCY, Y1
	movlpd	%xmm2, (Y1)
	addq	INCY, Y1
	movhpd	%xmm2, (Y1)
	addq	INCY, Y1

#if GEMV_UNROLL == 4
	cmpq	$4, N
	jge	.L21
#endif
	ALIGN_4

.L30:
#endif

#if GEMV_UNROLL >= 2

	cmpq	$2, N
	jl	.L40

#if GEMV_UNROLL == 2
	ALIGN_3

.L31:
#endif

	subq	$2, N

	leaq	16 * SIZE(BUFFER), X1

	movq	A, A1
	leaq	(A1, LDA), A2
	leaq	(A1, LDA, 2), A

	xorps	%xmm0, %xmm0
	xorps	%xmm1, %xmm1
	xorps	%xmm2, %xmm2
	xorps	%xmm3, %xmm3

#if (GEMV_UNROLL == 2 ) && defined(PREFETCHW)
	PREFETCHW	2 * SIZE(Y1)
#endif

#ifdef ALIGNED_ACCESS
	testq	$SIZE, A
	je	.L3X

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm0
	movsd	-16 * SIZE(A2), %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm1

	addq	 $SIZE, A1
	addq	 $SIZE, A2
	addq	 $SIZE, X1
	ALIGN_3

.L3X:
#endif

	movq	M,  I
	sarq	$3,  I
	jle	.L35

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	MOVUPS_A1(-16 * SIZE, A2, %xmm9)
	MOVUPS_A1(-14 * SIZE, A1, %xmm10)
	MOVUPS_A1(-14 * SIZE, A2, %xmm11)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	decq	I
	jle	.L33
	ALIGN_4

.L32:
#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 4 - 128 + PREOFFSET(A1)
#endif

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-12 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm9, %xmm1
	MOVUPS_A1(-12 * SIZE, A2, %xmm9)

	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-10 * SIZE, A1, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm3
	MOVUPS_A1(-10 * SIZE, A2, %xmm11)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 4 - 128 + PREOFFSET(A2)
#endif

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1( -8 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	MOVUPS_XL1( -8 * SIZE, X1, %xmm12)
	addpd	%xmm9, %xmm1
	MOVUPS_A1( -8 * SIZE, A2, %xmm9)

#ifdef PREFETCHW
	PREFETCH	(PREFETCHSIZE) * 4 - 128 + PREOFFSET(X1)
#endif

	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1( -6 * SIZE, A1, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1( -6 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm3
	MOVUPS_A1( -6 * SIZE, A2, %xmm11)

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1

	decq	I
	jg	.L32
	ALIGN_4

.L33:
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-12 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm9
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm9, %xmm1
	MOVUPS_A1(-12 * SIZE, A2, %xmm9)

	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-10 * SIZE, A1, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm3
	MOVUPS_A1(-10 * SIZE, A2, %xmm11)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1

	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1
	ALIGN_4

.L35:
	testq	$4, M
	jle	.L36

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-16 * SIZE, A2, %xmm9)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1

	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	MOVUPS_A1(-14 * SIZE, A1, %xmm10)
	mulpd	%xmm13, %xmm10
	addpd	%xmm10, %xmm2
	MOVUPS_A1(-14 * SIZE, A2, %xmm11)
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm3

	addq	$4 * SIZE, A1
	addq	$4 * SIZE, A2
	addq	$4 * SIZE, X1
	ALIGN_4

.L36:
	testq	$2, M
	jle	.L37

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0
	MOVUPS_A1(-16 * SIZE, A2, %xmm9)
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm1

	addq	$2 * SIZE, A1
	addq	$2 * SIZE, A2
	addq	$2 * SIZE, X1
	ALIGN_4

.L37:
	testq	$1, M
	je	.L38

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm0
	movsd	-16 * SIZE(A2), %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm1
	ALIGN_4

.L38:
	addpd	%xmm2, %xmm0
	addpd	%xmm3, %xmm1

#ifdef HAVE_SSE3
	haddpd	%xmm1, %xmm0
#else
	movapd	%xmm0, %xmm8
	unpcklpd %xmm1, %xmm0
	unpckhpd %xmm1, %xmm8

	addpd	%xmm8, %xmm0
#endif

	mulpd	ALPHA, %xmm0

	movsd	(Y), %xmm4
	addq	INCY, Y
	movhpd	(Y), %xmm4
	addq	INCY, Y

	addpd	%xmm4, %xmm0

	movlpd	%xmm0, (Y1)
	addq	INCY, Y1
	movhpd	%xmm0, (Y1)
	addq	INCY, Y1

#if GEMV_UNROLL == 2
	cmpq	$2, N
	jge	.L31
#endif
	ALIGN_4

.L40:
	cmpq	$1, N
	jl	.L999

#endif

	leaq	16 * SIZE(BUFFER), X1

	movq	A, A1

	xorps	%xmm0, %xmm0
	xorps	%xmm1, %xmm1
	xorps	%xmm2, %xmm2
	xorps	%xmm3, %xmm3

#ifdef ALIGNED_ACCESS
	testq	$SIZE, A
	je	.L4X

	movsd	-16 * SIZE(X1), %xmm12
	movsd	-16 * SIZE(A1), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm0

	addq	 $SIZE, A1
	addq	 $SIZE, X1
	ALIGN_3

.L4X:
#endif

	movq	M,  I
	sarq	$3,  I
	jle	.L45

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	MOVUPS_A1(-14 * SIZE, A1, %xmm9)
	MOVUPS_A1(-12 * SIZE, A1, %xmm10)
	MOVUPS_A1(-10 * SIZE, A1, %xmm11)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	decq	I
	jle	.L43
	ALIGN_4

.L42:
#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 8 - 128 + PREOFFSET(A1)
#endif

	mulpd	%xmm12, %xmm8
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm8,  %xmm0
	MOVUPS_A1( -8 * SIZE, A1, %xmm8)
	mulpd	%xmm13, %xmm9
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm9,  %xmm2
	MOVUPS_A1( -6 * SIZE, A1, %xmm9)

#ifdef PREFETCHW
	PREFETCH	(PREFETCHSIZE) * 8 - 128 + PREOFFSET(X1)
#endif

	mulpd	%xmm12, %xmm10
	MOVUPS_XL1( -8 * SIZE, X1, %xmm12)
	addpd	%xmm10, %xmm0
	MOVUPS_A1( -4 * SIZE, A1, %xmm10)
	mulpd	%xmm13, %xmm11
	MOVUPS_XL1( -6 * SIZE, X1, %xmm13)
	addpd	%xmm11, %xmm2
	MOVUPS_A1( -2 * SIZE, A1, %xmm11)

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, X1

	decq	I
	jg	.L42
	ALIGN_4

.L43:
	mulpd	%xmm12, %xmm8
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm8,  %xmm0
	mulpd	%xmm13, %xmm9
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm9,  %xmm2

	mulpd	%xmm12, %xmm10
	addpd	%xmm10, %xmm0
	mulpd	%xmm13, %xmm11
	addpd	%xmm11, %xmm2

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, X1
	ALIGN_4

.L45:
	testq	$4, M
	jle	.L46

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	MOVUPS_A1(-14 * SIZE, A1, %xmm9)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	mulpd	%xmm12, %xmm8
	addpd	%xmm8,  %xmm0
	mulpd	%xmm13, %xmm9
	addpd	%xmm9,  %xmm2

	addq	$4 * SIZE, A1
	addq	$4 * SIZE, X1
	ALIGN_4

.L46:
	testq	$2, M
	jle	.L47

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	MOVUPS_A1(-16 * SIZE, A1, %xmm8)
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm0

	addq	$2 * SIZE, A1
	addq	$2 * SIZE, X1
	ALIGN_4

.L47:
	testq	$1, M
	je	.L48

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm0
	ALIGN_4

.L48:
	addpd	%xmm2, %xmm0
	addpd	%xmm3, %xmm1

	addpd	%xmm1, %xmm0

#ifdef HAVE_SSE3
	haddpd	%xmm1, %xmm0
#else
	movapd	%xmm0, %xmm8
	unpcklpd %xmm1, %xmm0
	unpckhpd %xmm1, %xmm8

	addsd	%xmm8, %xmm0
#endif

	mulsd	ALPHA, %xmm0

	movsd	(Y), %xmm4
	addq	INCY, Y

	addsd	%xmm4, %xmm0

	movlpd	%xmm0, (Y1)
	addq	INCY, Y1

#ifdef ALIGNED_ACCESS
	jmp	.L999
	ALIGN_4

.L50:
#if GEMV_UNROLL >= 4

	cmpq	$4, N
	jl	.L60
	ALIGN_3

.L51:
	subq	$4, N

	leaq	16 * SIZE(BUFFER), X1

	movq	A, A1
	leaq	(A1, LDA, 2), A2
	leaq	(A1, LDA, 4), A

	xorps	%xmm0, %xmm0
	xorps	%xmm1, %xmm1
	xorps	%xmm2, %xmm2
	xorps	%xmm3, %xmm3

#ifdef PREFETCHW
	PREFETCHW	3 * SIZE(Y1)
#endif

#ifdef ALIGNED_ACCESS
	testq	$SIZE, A
	je	.L5X

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm4
	mulsd	%xmm12, %xmm4
	addsd	%xmm4, %xmm0
	movsd	-16 * SIZE(A1, LDA), %xmm5
	mulsd	%xmm12, %xmm5
	addsd	%xmm5, %xmm1
	movsd	-16 * SIZE(A2), %xmm6
	mulsd	%xmm12, %xmm6
	addsd	%xmm6, %xmm2
	movsd	-16 * SIZE(A2, LDA), %xmm7
	mulsd	%xmm12, %xmm7
	addsd	%xmm7, %xmm3

	addq	 $SIZE, A1
	addq	 $SIZE, A2
	addq	 $SIZE, X1
	ALIGN_3

.L5X:
#endif

	movhpd	-16 * SIZE(A1, LDA), %xmm8
	movhpd	-16 * SIZE(A2, LDA), %xmm9

	movq	M,  I
	sarq	$3,  I
	jle	.L55

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	MOVUPS_A2(-15 * SIZE, A1, LDA, 1, %xmm5)
	MOVUPS_A1(-16 * SIZE, A2, %xmm6)
	MOVUPS_A2(-15 * SIZE, A2, LDA, 1, %xmm7)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	decq	I
	jle	.L53
	ALIGN_4

.L52:
#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(A1)
#endif

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-14 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm1
	MOVUPS_A2(-13 * SIZE, A1, LDA, 1, %xmm8)

	mulpd	%xmm12, %xmm6
	addpd	%xmm6, %xmm2
	MOVUPS_A1(-14 * SIZE, A2, %xmm6)
	shufpd	$1, %xmm7, %xmm9
	mulpd	%xmm12, %xmm9
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm9, %xmm3
	MOVUPS_A2(-13 * SIZE, A2, LDA, 1, %xmm9)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET + 8(A1, LDA)
#endif
	mulpd	%xmm13, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-12 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm8, %xmm5
	mulpd	%xmm13, %xmm5
	addpd	%xmm5, %xmm1
	MOVUPS_A2(-11 * SIZE, A1, LDA, 1, %xmm5)

	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm2
	MOVUPS_A1(-12 * SIZE, A2, %xmm6)
	shufpd	$1, %xmm9, %xmm7
	mulpd	%xmm13, %xmm7
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm7, %xmm3
	MOVUPS_A2(-11 * SIZE, A2, LDA, 1, %xmm7)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET(A2)
#endif

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-10 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm1
	MOVUPS_A2( -9 * SIZE, A1, LDA, 1, %xmm8)

	mulpd	%xmm12, %xmm6
	addpd	%xmm6, %xmm2
	MOVUPS_A1(-10 * SIZE, A2, %xmm6)
	shufpd	$1, %xmm7, %xmm9
	mulpd	%xmm12, %xmm9
	MOVUPS_XL1(-8 * SIZE, X1, %xmm12)
	addpd	%xmm9, %xmm3
	MOVUPS_A2( -9 * SIZE, A2, LDA, 1, %xmm9)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET + 8(A2, LDA)
#endif

	mulpd	%xmm13, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-8 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm8, %xmm5
	mulpd	%xmm13, %xmm5
	addpd	%xmm5, %xmm1
	MOVUPS_A2(-7 * SIZE, A1, LDA, 1, %xmm5)

#ifdef PREFETCHW
	PREFETCH	(PREFETCHSIZE) * 2 - 128 + PREOFFSET + 8(X1)
#endif

	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm2
	MOVUPS_A1(-8 * SIZE, A2, %xmm6)
	shufpd	$1, %xmm9, %xmm7
	mulpd	%xmm13, %xmm7
	MOVUPS_XL1(-6 * SIZE, X1, %xmm13)
	addpd	%xmm7, %xmm3
	MOVUPS_A2(-7 * SIZE, A2, LDA, 1, %xmm7)

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1

	decq	I
	jg	.L52
	ALIGN_4

.L53:
	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-14 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm1
	MOVUPS_A2(-13 * SIZE, A1, LDA, 1, %xmm8)

	mulpd	%xmm12, %xmm6
	addpd	%xmm6, %xmm2
	MOVUPS_A1(-14 * SIZE, A2, %xmm6)
	shufpd	$1, %xmm7, %xmm9
	mulpd	%xmm12, %xmm9
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm9, %xmm3
	MOVUPS_A2(-13 * SIZE, A2, LDA, 1, %xmm9)

	mulpd	%xmm13, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-12 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm8, %xmm5
	mulpd	%xmm13, %xmm5
	addpd	%xmm5, %xmm1
	MOVUPS_A2(-11 * SIZE, A1, LDA, 1, %xmm5)
	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm2
	MOVUPS_A1(-12 * SIZE, A2, %xmm6)
	shufpd	$1, %xmm9, %xmm7
	mulpd	%xmm13, %xmm7
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm7, %xmm3
	MOVUPS_A2(-11 * SIZE, A2, LDA, 1, %xmm7)

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-10 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm1
	MOVUPS_A2( -9 * SIZE, A1, LDA, 1, %xmm8)

	mulpd	%xmm12, %xmm6
	addpd	%xmm6, %xmm2
	MOVUPS_A1(-10 * SIZE, A2, %xmm6)
	shufpd	$1, %xmm7, %xmm9
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm3
	MOVUPS_A2( -9 * SIZE, A2, LDA, 1, %xmm9)

	mulpd	%xmm13, %xmm4
	addpd	%xmm4, %xmm0
	shufpd	$1, %xmm8, %xmm5
	mulpd	%xmm13, %xmm5
	addpd	%xmm5, %xmm1

	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm2
	shufpd	$1, %xmm9, %xmm7
	mulpd	%xmm13, %xmm7
	addpd	%xmm7, %xmm3

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1
	ALIGN_4

.L55:
	testq	$4, M
	jle	.L56

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	MOVUPS_A2(-15 * SIZE, A1, LDA, 1, %xmm5)
	MOVUPS_A1(-16 * SIZE, A2, %xmm6)
	MOVUPS_A2(-15 * SIZE, A2, LDA, 1, %xmm7)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-14 * SIZE, A1, %xmm4)

	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm1
	MOVUPS_A2(-13 * SIZE, A1, LDA, 1, %xmm8)

	mulpd	%xmm12, %xmm6
	addpd	%xmm6, %xmm2
	MOVUPS_A1(-14 * SIZE, A2, %xmm6)

	shufpd	$1, %xmm7, %xmm9
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm3
	MOVUPS_A2(-13 * SIZE, A2, LDA, 1, %xmm9)

	mulpd	%xmm13, %xmm4
	addpd	%xmm4, %xmm0
	shufpd	$1, %xmm8, %xmm5
	mulpd	%xmm13, %xmm5
	addpd	%xmm5, %xmm1
	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm2
	shufpd	$1, %xmm9, %xmm7
	mulpd	%xmm13, %xmm7
	addpd	%xmm7, %xmm3

	addq	$4 * SIZE, A1
	addq	$4 * SIZE, A2
	addq	$4 * SIZE, X1
	ALIGN_4

.L56:
	testq	$2, M
	jle	.L57

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	MOVUPS_A2(-15 * SIZE, A1, LDA, 1, %xmm5)
	MOVUPS_A1(-16 * SIZE, A2, %xmm6)
	MOVUPS_A2(-15 * SIZE, A2, LDA, 1, %xmm7)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm1
	movaps	%xmm5, %xmm8
	mulpd	%xmm12, %xmm6
	addpd	%xmm6, %xmm2
	shufpd	$1, %xmm7, %xmm9
	mulpd	%xmm12, %xmm9
	addpd	%xmm9, %xmm3
	movaps	%xmm7, %xmm9

	addq	$2 * SIZE, A1
	addq	$2 * SIZE, A2
	addq	$2 * SIZE, X1
	ALIGN_4

.L57:
	testq	$1, M
	je	.L58

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm4
	mulsd	%xmm12, %xmm4
	addsd	%xmm4, %xmm0
	shufpd	$1, %xmm8, %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm1
	movsd	-16 * SIZE(A2), %xmm6
	mulsd	%xmm12, %xmm6
	addsd	%xmm6, %xmm2
	shufpd	$1, %xmm9, %xmm9
	mulsd	%xmm12, %xmm9
	addsd	%xmm9, %xmm3
	ALIGN_4

.L58:
#ifdef HAVE_SSE3
	haddpd	%xmm1, %xmm0
	haddpd	%xmm3, %xmm2
#else
	movapd	%xmm0, %xmm4
	unpcklpd %xmm1, %xmm0
	unpckhpd %xmm1, %xmm4

	movapd	%xmm2, %xmm5
	unpcklpd %xmm3, %xmm2
	unpckhpd %xmm3, %xmm5

	addpd	%xmm4, %xmm0
	addpd	%xmm5, %xmm2
#endif

	mulpd	ALPHA, %xmm0
	mulpd	ALPHA, %xmm2

	cmpq	$SIZE, INCY
	jne	.L59

	movsd	 0 * SIZE(Y), %xmm4
	movhpd	 1 * SIZE(Y), %xmm4
	movsd	 2 * SIZE(Y), %xmm5
	movhpd	 3 * SIZE(Y), %xmm5
	addq	$4 * SIZE, Y

	addpd	%xmm4, %xmm0
	addpd	%xmm5, %xmm2

	movlpd	%xmm0,  0 * SIZE(Y1)
	movhpd	%xmm0,  1 * SIZE(Y1)
	movlpd	%xmm2,  2 * SIZE(Y1)
	movhpd	%xmm2,  3 * SIZE(Y1)
	addq	$4 * SIZE, Y1

	cmpq	$4, N
	jge	.L51
	jmp	.L60
	ALIGN_4

.L59:
	movsd	(Y), %xmm4
	addq	INCY, Y
	movhpd	(Y), %xmm4
	addq	INCY, Y
	movsd	(Y), %xmm5
	addq	INCY, Y
	movhpd	(Y), %xmm5
	addq	INCY, Y

	addpd	%xmm4, %xmm0
	addpd	%xmm5, %xmm2

	movlpd	%xmm0, (Y1)
	addq	INCY, Y1
	movhpd	%xmm0, (Y1)
	addq	INCY, Y1
	movlpd	%xmm2, (Y1)
	addq	INCY, Y1
	movhpd	%xmm2, (Y1)
	addq	INCY, Y1
	cmpq	$4, N
	jge	.L51
	ALIGN_4

.L60:
#endif

#if GEMV_UNROLL >= 2

	cmpq	$2, N
	jl	.L70

#if GEMV_UNROLL == 2
	ALIGN_3

.L61:
#endif

	subq	$2, N

	leaq	16 * SIZE(BUFFER), X1

	movq	A, A1
	leaq	(A1, LDA), A2
	leaq	(A1, LDA, 2), A

	xorps	%xmm0, %xmm0
	xorps	%xmm1, %xmm1
	xorps	%xmm2, %xmm2
	xorps	%xmm3, %xmm3

#if (GEMV_UNROLL == 2 ) && defined(PREFETCHW)
	PREFETCHW	2 * SIZE(Y1)
#endif

#ifdef ALIGNED_ACCESS
	testq	$SIZE, A
	je	.L6X

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm4
	mulsd	%xmm12, %xmm4
	addsd	%xmm4, %xmm0
	movsd	-16 * SIZE(A2), %xmm5
	mulsd	%xmm12, %xmm5
	addsd	%xmm5, %xmm1

	addq	 $SIZE, A1
	addq	 $SIZE, A2
	addq	 $SIZE, X1
	ALIGN_3

.L6X:
#endif

	movhpd	-16 * SIZE(A2), %xmm8

	movq	M,  I
	sarq	$3,  I
	jle	.L65

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	MOVUPS_A1(-15 * SIZE, A2, %xmm5)
	MOVUPS_A1(-14 * SIZE, A1, %xmm6)
	MOVUPS_A1(-13 * SIZE, A2, %xmm7)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	decq	I
	jle	.L63
	ALIGN_4

.L62:
#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 4 - 128 + PREOFFSET(A1)
#endif

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-12 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm8, %xmm1
	MOVUPS_A1(-11 * SIZE, A2, %xmm9)

	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm0
	MOVUPS_A1(-10 * SIZE, A1, %xmm6)
	shufpd	$1, %xmm7, %xmm5
	mulpd	%xmm13, %xmm5
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm5, %xmm1
	MOVUPS_A1( -9 * SIZE, A2, %xmm8)

#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 4 - 128 + PREOFFSET + 8(A2)
#endif

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-8 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm9, %xmm7
	mulpd	%xmm12, %xmm7
	MOVUPS_XL1(-8 * SIZE, X1, %xmm12)
	addpd	%xmm7, %xmm1
	MOVUPS_A1(-7 * SIZE, A2, %xmm5)

#ifdef PREFETCHW
	PREFETCH	(PREFETCHSIZE) * 4 - 128 + PREOFFSET + 8(X1)
#endif

	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm0
	MOVUPS_A1(-6 * SIZE, A1, %xmm6)
	shufpd	$1, %xmm8, %xmm9
	mulpd	%xmm13, %xmm9
	MOVUPS_XL1(-6 * SIZE, X1, %xmm13)
	addpd	%xmm9, %xmm1
	MOVUPS_A1(-5 * SIZE, A2, %xmm7)

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1

	decq	I
	jg	.L62
	ALIGN_4

.L63:
	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	MOVUPS_A1(-12 * SIZE, A1, %xmm4)
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm8, %xmm1
	MOVUPS_A1(-11 * SIZE, A2, %xmm9)

	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm0
	MOVUPS_A1(-10 * SIZE, A1, %xmm6)
	shufpd	$1, %xmm7, %xmm5
	mulpd	%xmm13, %xmm5
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm5, %xmm1
	MOVUPS_A1( -9 * SIZE, A2, %xmm8)

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	shufpd	$1, %xmm9, %xmm7
	mulpd	%xmm12, %xmm7
	addpd	%xmm7, %xmm1

	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm0
	shufpd	$1, %xmm8, %xmm9
	mulpd	%xmm13, %xmm9
	addpd	%xmm9, %xmm1

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, A2
	addq	$8 * SIZE, X1
	ALIGN_4

.L65:
	testq	$4, M
	jle	.L66

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	MOVUPS_A1(-15 * SIZE, A2, %xmm5)
	MOVUPS_A1(-14 * SIZE, A1, %xmm6)
	MOVUPS_A1(-13 * SIZE, A2, %xmm7)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm1

	mulpd	%xmm13, %xmm6
	addpd	%xmm6, %xmm0
	shufpd	$1, %xmm7, %xmm5
	movaps	%xmm7, %xmm8
	mulpd	%xmm13, %xmm5
	addpd	%xmm5, %xmm1

	addq	$4 * SIZE, A1
	addq	$4 * SIZE, A2
	addq	$4 * SIZE, X1
	ALIGN_4

.L66:
	testq	$2, M
	jle	.L67

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	MOVUPS_A1(-15 * SIZE, A2, %xmm5)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0
	shufpd	$1, %xmm5, %xmm8
	mulpd	%xmm12, %xmm8
	addpd	%xmm8, %xmm1
	movaps	%xmm5, %xmm8

	addq	$2 * SIZE, A1
	addq	$2 * SIZE, A2
	addq	$2 * SIZE, X1
	ALIGN_4

.L67:
	testq	$1, M
	je	.L68

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm4
	mulsd	%xmm12, %xmm4
	addsd	%xmm4, %xmm0
	shufpd	$1, %xmm8, %xmm8
	mulsd	%xmm12, %xmm8
	addsd	%xmm8, %xmm1
	ALIGN_4

.L68:
	addpd	%xmm2, %xmm0
	addpd	%xmm3, %xmm1

#ifdef HAVE_SSE3
	haddpd	%xmm1, %xmm0
#else
	movapd	%xmm0, %xmm4
	unpcklpd %xmm1, %xmm0
	unpckhpd %xmm1, %xmm4

	addpd	%xmm4, %xmm0
#endif

	mulpd	ALPHA, %xmm0

	movsd	(Y), %xmm4
	addq	INCY, Y
	movhpd	(Y), %xmm4
	addq	INCY, Y

	addpd	%xmm4, %xmm0

	movlpd	%xmm0, (Y1)
	addq	INCY, Y1
	movhpd	%xmm0, (Y1)
	addq	INCY, Y1

#if GEMV_UNROLL == 2
	cmpq	$2, N
	jge	.L61
#endif
	ALIGN_4

.L70:
	cmpq	$1, N
	jl	.L999

#endif

	leaq	16 * SIZE(BUFFER), X1

	movq	A, A1

	xorps	%xmm0, %xmm0
	xorps	%xmm1, %xmm1
	xorps	%xmm2, %xmm2
	xorps	%xmm3, %xmm3

#ifdef ALIGNED_ACCESS
	testq	$SIZE, A
	je	.L7X

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm4
	mulsd	%xmm12, %xmm4
	addsd	%xmm4, %xmm0

	addq	 $SIZE, A1
	addq	 $SIZE, X1
	ALIGN_3

.L7X:
#endif
	movq	M,  I
	sarq	$3,  I
	jle	.L75

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	MOVUPS_A1(-14 * SIZE, A1, %xmm5)
	MOVUPS_A1(-12 * SIZE, A1, %xmm6)
	MOVUPS_A1(-10 * SIZE, A1, %xmm7)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	decq	I
	jle	.L73
	ALIGN_4

.L72:
#ifdef PREFETCH
	PREFETCH	(PREFETCHSIZE) * 8 - 128 + PREOFFSET(A1)
#endif

	mulpd	%xmm12, %xmm4
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm4,  %xmm0
	MOVUPS_A1( -8 * SIZE, A1, %xmm4)
	mulpd	%xmm13, %xmm5
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm5,  %xmm2
	MOVUPS_A1( -6 * SIZE, A1, %xmm5)

#ifdef PREFETCHW
	PREFETCH	(PREFETCHSIZE) * 8 - 128 + PREOFFSET(X1)
#endif

	mulpd	%xmm12, %xmm6
	MOVUPS_XL1( -8 * SIZE, X1, %xmm12)
	addpd	%xmm6, %xmm0
	MOVUPS_A1( -4 * SIZE, A1, %xmm6)
	mulpd	%xmm13, %xmm7
	MOVUPS_XL1( -6 * SIZE, X1, %xmm13)
	addpd	%xmm7, %xmm2
	MOVUPS_A1( -2 * SIZE, A1, %xmm7)

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, X1

	decq	I
	jg	.L72
	ALIGN_4

.L73:
	mulpd	%xmm12, %xmm4
	MOVUPS_XL1(-12 * SIZE, X1, %xmm12)
	addpd	%xmm4,  %xmm0
	mulpd	%xmm13, %xmm5
	MOVUPS_XL1(-10 * SIZE, X1, %xmm13)
	addpd	%xmm5,  %xmm2

	mulpd	%xmm12, %xmm6
	addpd	%xmm6, %xmm0
	mulpd	%xmm13, %xmm7
	addpd	%xmm7, %xmm2

	addq	$8 * SIZE, A1
	addq	$8 * SIZE, X1
	ALIGN_4

.L75:
	testq	$4, M
	jle	.L76

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	MOVUPS_A1(-14 * SIZE, A1, %xmm5)

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)
	MOVUPS_XL1(-14 * SIZE, X1, %xmm13)

	mulpd	%xmm12, %xmm4
	addpd	%xmm4,  %xmm0
	mulpd	%xmm13, %xmm5
	addpd	%xmm5,  %xmm2

	addq	$4 * SIZE, A1
	addq	$4 * SIZE, X1
	ALIGN_4

.L76:
	testq	$2, M
	jle	.L77

	MOVUPS_XL1(-16 * SIZE, X1, %xmm12)

	MOVUPS_A1(-16 * SIZE, A1, %xmm4)
	mulpd	%xmm12, %xmm4
	addpd	%xmm4, %xmm0

	addq	$2 * SIZE, A1
	addq	$2 * SIZE, X1
	ALIGN_4

.L77:
	testq	$1, M
	je	.L78

	movsd	-16 * SIZE(X1), %xmm12

	movsd	-16 * SIZE(A1), %xmm4
	mulsd	%xmm12, %xmm4
	addsd	%xmm4, %xmm0
	ALIGN_4

.L78:
	addpd	%xmm2, %xmm0
	addpd	%xmm3, %xmm1

	addpd	%xmm1, %xmm0

#ifdef HAVE_SSE3
	haddpd	%xmm1, %xmm0
#else
	movapd	%xmm0, %xmm4
	unpcklpd %xmm1, %xmm0
	unpckhpd %xmm1, %xmm4

	addsd	%xmm4, %xmm0
#endif

	mulsd	ALPHA, %xmm0

	movsd	(Y), %xmm4
	addq	INCY, Y

	addsd	%xmm4, %xmm0

	movlpd	%xmm0, (Y1)
	addq	INCY, Y1
#endif
	ALIGN_4

.L999:
	leaq	(, M, SIZE), %rax
	addq %rax,AA
	jmp .L0x;
	ALIGN_4

.L999x:
	movq	  0(%rsp), %rbx
	movq	  8(%rsp), %rbp
	movq	 16(%rsp), %r12
	movq	 24(%rsp), %r13
	movq	 32(%rsp), %r14
	movq	 40(%rsp), %r15

#ifdef WINDOWS_ABI
	movq	 48(%rsp), %rdi
	movq	 56(%rsp), %rsi
	movups	 64(%rsp), %xmm6
	movups	 80(%rsp), %xmm7
	movups	 96(%rsp), %xmm8
	movups	112(%rsp), %xmm9
	movups	128(%rsp), %xmm10
	movups	144(%rsp), %xmm11
	movups	160(%rsp), %xmm12
	movups	176(%rsp), %xmm13
	movups	192(%rsp), %xmm14
	movups	208(%rsp), %xmm15
#endif

	addq	$STACKSIZE, %rsp
	ret
	ALIGN_4

	EPILOGUE