From 44f2bf9bae7b25356fc0179d6b935de4edadc637 Mon Sep 17 00:00:00 2001
From: wernsaar <wernsaar@googlemail.com>
Date: Tue, 9 Sep 2014 13:34:22 +0200
Subject: [PATCH] added optimized dgemv_t kernel for haswell

---
 kernel/x86_64/dgemv_t_4.c                | 623 +++++++++++++++++++++++
 kernel/x86_64/dgemv_t_microk_haswell-4.c | 127 +++++
 2 files changed, 750 insertions(+)
 create mode 100644 kernel/x86_64/dgemv_t_4.c
 create mode 100644 kernel/x86_64/dgemv_t_microk_haswell-4.c

diff --git a/kernel/x86_64/dgemv_t_4.c b/kernel/x86_64/dgemv_t_4.c
new file mode 100644
index 000000000..0d0409bec
--- /dev/null
+++ b/kernel/x86_64/dgemv_t_4.c
@@ -0,0 +1,623 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+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.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "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 OPENBLAS PROJECT 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.
+*****************************************************************************/
+
+
+#include "common.h"
+
+/*
+#if defined(NEHALEM)
+#include "dgemv_t_microk_nehalem-4.c"
+#elif defined(BULLDOZER) || defined(PILEDRIVER)
+#include "dgemv_t_microk_bulldozer-4.c"
+#elif defined(SANDYBRIDGE)
+#include "dgemv_t_microk_sandy-4.c"
+#elif defined(HASWELL)
+#include "dgemv_t_microk_haswell-4.c"
+#endif
+*/
+
+#define NBMAX 2048
+
+#ifndef HAVE_KERNEL_4x4
+
+static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
+{
+	BLASLONG i;
+	FLOAT *a0,*a1,*a2,*a3;
+	a0 = ap[0];
+	a1 = ap[1];
+	a2 = ap[2];
+	a3 = ap[3];
+	FLOAT temp0 = 0.0;
+	FLOAT temp1 = 0.0;
+	FLOAT temp2 = 0.0;
+	FLOAT temp3 = 0.0;
+
+	for ( i=0; i< n; i+=4 )
+	{
+		temp0 += a0[i]*x[i] + a0[i+1]*x[i+1] + a0[i+2]*x[i+2] + a0[i+3]*x[i+3];		
+		temp1 += a1[i]*x[i] + a1[i+1]*x[i+1] + a1[i+2]*x[i+2] + a1[i+3]*x[i+3];		
+		temp2 += a2[i]*x[i] + a2[i+1]*x[i+1] + a2[i+2]*x[i+2] + a2[i+3]*x[i+3];		
+		temp3 += a3[i]*x[i] + a3[i+1]*x[i+1] + a3[i+2]*x[i+2] + a3[i+3]*x[i+3];		
+	}
+	y[0] = temp0;
+	y[1] = temp1;
+	y[2] = temp2;
+	y[3] = temp3;
+}
+	
+#endif
+
+static void dgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT *y)  __attribute__ ((noinline));
+
+static void dgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT *y)
+{
+	BLASLONG i;
+
+	i=0;
+
+        __asm__  __volatile__
+ 	(
+	"xorpd %%xmm10 , %%xmm10		\n\t"
+	"xorpd %%xmm11 , %%xmm11		\n\t"
+		
+	"testq	$2 , %1				\n\t"
+	"jz	.L01LABEL%=			\n\t"
+
+	"movups  (%5,%0,8) , %%xmm14		\n\t" // x
+	"movups  (%3,%0,8) , %%xmm12		\n\t" // ap0
+	"movups  (%4,%0,8) , %%xmm13		\n\t" // ap1
+	"mulpd   %%xmm14   , %%xmm12 		\n\t"
+	"mulpd   %%xmm14   , %%xmm13 		\n\t"
+        "addq           $2 , %0                 \n\t"
+	"addpd   %%xmm12   , %%xmm10		\n\t"
+        "subq           $2 , %1                 \n\t"
+	"addpd   %%xmm13   , %%xmm11		\n\t"
+
+        ".L01LABEL%=:                           \n\t"
+
+	"cmpq	$0, %1				\n\t"
+	"je	.L01END%=			\n\t"
+
+        ".align 16                              \n\t"
+        ".L01LOOP%=:                            \n\t"
+
+	"movups  (%5,%0,8) , %%xmm14		\n\t" // x
+	"movups  (%3,%0,8) , %%xmm12		\n\t" // ap0
+	"movups  (%4,%0,8) , %%xmm13		\n\t" // ap1
+	"mulpd   %%xmm14   , %%xmm12 		\n\t"
+	"mulpd   %%xmm14   , %%xmm13 		\n\t"
+	"addpd   %%xmm12   , %%xmm10		\n\t"
+	"addpd   %%xmm13   , %%xmm11		\n\t"
+
+	"movups  16(%5,%0,8) , %%xmm14		\n\t" // x
+	"movups  16(%3,%0,8) , %%xmm12		\n\t" // ap0
+	"movups  16(%4,%0,8) , %%xmm13		\n\t" // ap1
+	"mulpd   %%xmm14   , %%xmm12 		\n\t"
+	"mulpd   %%xmm14   , %%xmm13 		\n\t"
+	"addpd   %%xmm12   , %%xmm10		\n\t"
+	"addpd   %%xmm13   , %%xmm11		\n\t"
+
+        "addq           $4 , %0                 \n\t"
+        "subq           $4 , %1                 \n\t"
+        "jnz            .L01LOOP%=              \n\t"
+
+        ".L01END%=:                             \n\t"
+
+	"haddpd        %%xmm10, %%xmm10         \n\t"
+	"haddpd        %%xmm11, %%xmm11         \n\t"
+
+	"movsd	       %%xmm10, (%2)	        \n\t"
+	"movsd	       %%xmm11,8(%2)	        \n\t"
+
+        :
+   	:
+	"r" (i),	 // 0
+	"r" (n),	 // 1
+        "r" (y),         // 2    
+        "r" (ap0),       // 3
+        "r" (ap1),       // 4
+        "r" (x)          // 5
+        : "cc",
+       	"%xmm4", "%xmm5", "%xmm10", "%xmm11",
+       	"%xmm12", "%xmm13", "%xmm14", "%xmm15",
+       	"memory"
+       	);
+
+
+}
+	
+static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)  __attribute__ ((noinline));
+
+static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
+{
+	BLASLONG i;
+
+	i=0;
+
+        __asm__  __volatile__
+ 	(
+	"xorpd %%xmm9  , %%xmm9 		\n\t"
+	"xorpd %%xmm10 , %%xmm10		\n\t"
+	
+	"testq	$2 , %1				\n\t"
+	"jz	.L01LABEL%=			\n\t"
+
+	"movups  (%3,%0,8) , %%xmm12		\n\t"
+	"movups  (%4,%0,8) , %%xmm11		\n\t"
+	"mulpd   %%xmm11   , %%xmm12 		\n\t"
+        "addq           $2 , %0                 \n\t"
+	"addpd   %%xmm12   , %%xmm10		\n\t"
+        "subq           $2 , %1                 \n\t"
+
+        ".L01LABEL%=:                           \n\t"
+
+	"cmpq	$0, %1				\n\t"
+	"je	.L01END%=			\n\t"
+
+        ".align 16                              \n\t"
+        ".L01LOOP%=:                            \n\t"
+
+	"movups    (%3,%0,8) , %%xmm12		\n\t"
+	"movups  16(%3,%0,8) , %%xmm14		\n\t"
+	"movups    (%4,%0,8) , %%xmm11		\n\t"
+	"movups  16(%4,%0,8) , %%xmm13		\n\t"
+	"mulpd   %%xmm11   , %%xmm12 		\n\t"
+	"mulpd   %%xmm13   , %%xmm14 		\n\t"
+        "addq           $4 , %0                 \n\t"
+	"addpd   %%xmm12   , %%xmm10		\n\t"
+        "subq           $4 , %1                 \n\t"
+	"addpd   %%xmm14   , %%xmm9 		\n\t"
+
+        "jnz            .L01LOOP%=              \n\t"
+
+        ".L01END%=:                             \n\t"
+
+	"addpd	       %%xmm9 , %%xmm10         \n\t"
+	"haddpd        %%xmm10, %%xmm10         \n\t"
+
+	"movsd	       %%xmm10, (%2)	        \n\t"
+
+        :
+   	:
+	"r" (i),	 // 0
+	"r" (n),	 // 1
+        "r" (y),         // 2    
+        "r" (ap),        // 3
+        "r" (x)          // 4
+        : "cc",
+       	"%xmm9", "%xmm10" ,
+       	"%xmm11", "%xmm12", "%xmm13", "%xmm14",
+       	"memory"
+       	);
+
+
+}
+	
+static void copy_x(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_src)
+{
+        BLASLONG i;
+        for ( i=0; i<n; i++ )
+        {
+                *dest = *src;
+                dest++;
+                src += inc_src;
+        }
+}
+
+static void add_y(BLASLONG n, FLOAT da , FLOAT *src, FLOAT *dest, BLASLONG inc_dest) __attribute__ ((noinline));
+
+static void add_y(BLASLONG n, FLOAT da , FLOAT *src, FLOAT *dest, BLASLONG inc_dest)
+{
+
+        BLASLONG i;
+
+	if ( inc_dest != 1 )
+	{
+        	for ( i=0; i<n; i++ )
+        	{
+                	*dest += src[i]  * da;
+                	dest  += inc_dest;
+		}
+		return;
+        }
+
+	i=0;
+
+        __asm__  __volatile__
+ 	(
+	"movsd	 (%2) , %%xmm10                 \n\t"
+	"shufpd  $0 , %%xmm10 , %%xmm10		\n\t"
+
+        ".align 16                              \n\t"
+        ".L01LOOP%=:                            \n\t"
+
+	"movups  (%3,%0,8) , %%xmm12		\n\t"
+	"movups  (%4,%0,8) , %%xmm11		\n\t"
+	"mulpd   %%xmm10   , %%xmm12 		\n\t"
+        "addq           $2 , %0                 \n\t"
+	"addpd   %%xmm12   , %%xmm11		\n\t"
+        "subq           $2 , %1                 \n\t"
+	"movups  %%xmm11, -16(%4,%0,8)		\n\t"
+
+        "jnz            .L01LOOP%=              \n\t"
+
+        :
+   	:
+	"r" (i),	  // 0
+	"r" (n),	  // 1
+        "r" (&da),        // 2    
+        "r" (src),        // 3
+        "r" (dest)        // 4
+        : "cc",
+       	"%xmm10", "%xmm11", "%xmm12",
+       	"memory"
+       	);
+
+
+}
+
+int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
+{
+	BLASLONG register i;
+	BLASLONG register j;
+	FLOAT *a_ptr;
+	FLOAT *x_ptr;
+	FLOAT *y_ptr;
+	BLASLONG n0;
+	BLASLONG n1;
+	BLASLONG m1;
+	BLASLONG m2;
+	BLASLONG m3;
+	BLASLONG n2;
+	FLOAT ybuffer[4],*xbuffer;
+	FLOAT *ytemp;
+
+        if ( m < 1 ) return(0);
+        if ( n < 1 ) return(0);
+
+	xbuffer = buffer;
+	ytemp   = buffer + NBMAX;
+	
+	n0 = n / NBMAX;
+        n1 = (n % NBMAX)  >> 2 ;
+        n2 = n & 3  ;
+
+	m3 = m & 3  ;
+        m1 = m & -4 ;
+        m2 = (m & (NBMAX-1)) - m3 ;
+
+
+	BLASLONG NB = NBMAX;
+
+	while ( NB == NBMAX )
+	{
+		
+		m1 -= NB;
+		if ( m1 < 0)
+		{
+			if ( m2 == 0 ) break;	
+			NB = m2;
+		}
+		
+		y_ptr = y;
+		a_ptr = a;
+		x_ptr = x;
+
+		if ( inc_x == 1 )
+			xbuffer = x_ptr;
+		else
+			copy_x(NB,x_ptr,xbuffer,inc_x);
+
+
+		FLOAT *ap[4];
+		FLOAT *yp;
+		BLASLONG register lda4 = 4 * lda;
+		ap[0] = a_ptr;
+		ap[1] = a_ptr + lda;
+		ap[2] = ap[1] + lda;
+		ap[3] = ap[2] + lda;
+
+		if ( n0 > 0 )
+		{
+			BLASLONG nb1 = NBMAX / 4;
+			for( j=0; j<n0; j++)
+			{
+
+				yp = ytemp;
+				for( i = 0; i < nb1  ; i++)
+				{
+					dgemv_kernel_4x4(NB,ap,xbuffer,yp);
+					ap[0] += lda4 ;
+					ap[1] += lda4 ;
+					ap[2] += lda4 ;
+					ap[3] += lda4 ;
+					yp += 4;
+				}
+				add_y(nb1*4, alpha, ytemp, y_ptr, inc_y );
+				y_ptr += nb1 * inc_y * 4;
+				a_ptr += nb1 * lda4 ;
+
+			}
+
+		}
+
+
+		yp = ytemp;
+
+		for( i = 0; i < n1 ; i++)
+		{
+			dgemv_kernel_4x4(NB,ap,xbuffer,yp);
+			ap[0] += lda4 ;
+			ap[1] += lda4 ;
+			ap[2] += lda4 ;
+			ap[3] += lda4 ;
+			yp += 4;
+		}
+		if ( n1 > 0 )
+		{
+			add_y(n1*4, alpha, ytemp, y_ptr, inc_y );
+			y_ptr += n1 * inc_y * 4;
+			a_ptr += n1 * lda4 ;
+		}
+
+		if ( n2 & 2 )
+		{
+
+			dgemv_kernel_4x2(NB,ap[0],ap[1],xbuffer,ybuffer);
+			a_ptr  += lda * 2;
+			*y_ptr += ybuffer[0] * alpha;
+			y_ptr  += inc_y;
+			*y_ptr += ybuffer[1] * alpha;
+			y_ptr  += inc_y;
+
+		}
+
+		if ( n2 & 1 )
+		{
+
+			dgemv_kernel_4x1(NB,a_ptr,xbuffer,ybuffer);
+			a_ptr  += lda;
+			*y_ptr += ybuffer[0] * alpha;
+			y_ptr  += inc_y;
+
+		}
+		a += NB;
+		x += NB * inc_x;	
+	}
+
+	if ( m3 == 0 ) return(0);
+
+	x_ptr = x;
+	a_ptr = a;
+	if ( m3 == 3 )
+	{
+		FLOAT xtemp0 = *x_ptr * alpha;
+		x_ptr += inc_x;
+		FLOAT xtemp1 = *x_ptr * alpha;
+		x_ptr += inc_x;
+		FLOAT xtemp2 = *x_ptr * alpha;
+
+		FLOAT *aj = a_ptr;
+		y_ptr = y;
+
+		if ( lda == 3 && inc_y == 1 )
+		{
+
+			for ( j=0; j< ( n & -4) ; j+=4 )
+			{
+
+				y_ptr[j]   += aj[0] * xtemp0 + aj[1]  * xtemp1 + aj[2]  * xtemp2;
+				y_ptr[j+1] += aj[3] * xtemp0 + aj[4]  * xtemp1 + aj[5]  * xtemp2;
+				y_ptr[j+2] += aj[6] * xtemp0 + aj[7]  * xtemp1 + aj[8]  * xtemp2;
+				y_ptr[j+3] += aj[9] * xtemp0 + aj[10] * xtemp1 + aj[11] * xtemp2;
+			 	aj        += 12;
+			}
+
+			for ( ; j<n; j++ )
+			{
+				y_ptr[j]  += aj[0] * xtemp0 + aj[1] * xtemp1 + aj[2] * xtemp2;
+			 	aj        += 3;
+			}
+
+		}
+		else
+		{
+
+			if ( inc_y == 1 )
+			{
+
+				BLASLONG register lda2 = lda << 1;
+				BLASLONG register lda4 = lda << 2;
+				BLASLONG register lda3 = lda2 + lda;
+
+				for ( j=0; j< ( n & -4 ); j+=4 )
+				{
+
+					y_ptr[j]    += *aj        * xtemp0 + *(aj+1)      * xtemp1 + *(aj+2)      * xtemp2;
+					y_ptr[j+1]  += *(aj+lda)  * xtemp0 + *(aj+lda+1)  * xtemp1 + *(aj+lda+2)  * xtemp2;
+					y_ptr[j+2]  += *(aj+lda2) * xtemp0 + *(aj+lda2+1) * xtemp1 + *(aj+lda2+2) * xtemp2;
+					y_ptr[j+3]  += *(aj+lda3) * xtemp0 + *(aj+lda3+1) * xtemp1 + *(aj+lda3+2) * xtemp2;
+			 		aj          += lda4;
+				}
+
+				for ( ; j< n ; j++ )
+				{
+
+					y_ptr[j]    += *aj * xtemp0 + *(aj+1) * xtemp1 + *(aj+2) * xtemp2 ;
+			 		aj          += lda;
+				}
+
+			}
+			else
+			{
+
+				for ( j=0; j<n; j++ )
+				{
+					*y_ptr += *aj * xtemp0 + *(aj+1) * xtemp1 + *(aj+2) * xtemp2;
+				 	y_ptr += inc_y;
+			 		aj    += lda;
+				}
+
+
+			}
+
+		}
+		return(0);
+	}
+
+	if ( m3 == 2 )
+	{
+		FLOAT xtemp0 = *x_ptr * alpha;
+		x_ptr += inc_x;
+		FLOAT xtemp1 = *x_ptr * alpha;
+
+		FLOAT *aj = a_ptr;
+		y_ptr = y;
+
+		if ( lda == 2 && inc_y == 1 )
+		{
+
+			for ( j=0; j< ( n & -4) ; j+=4 )
+			{
+				y_ptr[j]   += aj[0] * xtemp0 + aj[1] * xtemp1 ;
+				y_ptr[j+1] += aj[2] * xtemp0 + aj[3] * xtemp1 ;
+				y_ptr[j+2] += aj[4] * xtemp0 + aj[5] * xtemp1 ;
+				y_ptr[j+3] += aj[6] * xtemp0 + aj[7] * xtemp1 ;
+			 	aj         += 8;
+
+			}
+
+			for ( ; j<n; j++ )
+			{
+				y_ptr[j] += aj[0] * xtemp0 + aj[1] * xtemp1 ;
+			 	aj       += 2;
+			}
+
+		}
+		else
+		{
+			if ( inc_y == 1 )
+			{
+
+				BLASLONG register lda2 = lda << 1;
+				BLASLONG register lda4 = lda << 2;
+				BLASLONG register lda3 = lda2 + lda;
+
+				for ( j=0; j< ( n & -4 ); j+=4 )
+				{
+
+					y_ptr[j]    += *aj        * xtemp0 + *(aj+1)      * xtemp1 ;
+					y_ptr[j+1]  += *(aj+lda)  * xtemp0 + *(aj+lda+1)  * xtemp1 ;
+					y_ptr[j+2]  += *(aj+lda2) * xtemp0 + *(aj+lda2+1) * xtemp1 ;
+					y_ptr[j+3]  += *(aj+lda3) * xtemp0 + *(aj+lda3+1) * xtemp1 ;
+			 		aj          += lda4;
+				}
+
+				for ( ; j< n ; j++ )
+				{
+
+					y_ptr[j]    += *aj * xtemp0 + *(aj+1) * xtemp1 ;
+			 		aj          += lda;
+				}
+
+			}
+			else
+			{
+				for ( j=0; j<n; j++ )
+				{
+					*y_ptr += *aj * xtemp0 + *(aj+1) * xtemp1 ;
+			 		y_ptr += inc_y;
+			 		aj    += lda;
+				}
+			}
+
+		}
+		return(0);
+
+	}
+
+	FLOAT xtemp = *x_ptr * alpha;
+	FLOAT *aj = a_ptr;
+	y_ptr = y;
+	if ( lda == 1 && inc_y == 1 )
+	{
+		for ( j=0; j< ( n & -4) ; j+=4 )
+		{
+			y_ptr[j]   += aj[j]   * xtemp;
+			y_ptr[j+1] += aj[j+1] * xtemp;
+			y_ptr[j+2] += aj[j+2] * xtemp;
+			y_ptr[j+3] += aj[j+3] * xtemp;
+		}
+		for ( ; j<n   ; j++ )
+		{
+			y_ptr[j] += aj[j] * xtemp;
+		}
+
+
+
+	}
+	else
+	{
+		if ( inc_y == 1 )
+		{
+
+			BLASLONG register lda2 = lda << 1;
+			BLASLONG register lda4 = lda << 2;
+			BLASLONG register lda3 = lda2 + lda;
+			for ( j=0; j< ( n & -4 ); j+=4 )
+			{
+				y_ptr[j]    += *aj        * xtemp;
+				y_ptr[j+1]  += *(aj+lda)  * xtemp;
+				y_ptr[j+2]  += *(aj+lda2) * xtemp;
+				y_ptr[j+3]  += *(aj+lda3) * xtemp;
+		 		aj          += lda4  ;
+			}
+
+			for ( ; j<n; j++ )
+			{
+				y_ptr[j]  += *aj * xtemp;
+		 		aj        += lda;
+			}
+
+		}
+		else
+		{
+			for ( j=0; j<n; j++ )
+			{
+				*y_ptr += *aj * xtemp;
+		 		y_ptr += inc_y;
+		 		aj    += lda;
+			}
+
+		}
+	}
+
+	return(0);
+}
+
+
diff --git a/kernel/x86_64/dgemv_t_microk_haswell-4.c b/kernel/x86_64/dgemv_t_microk_haswell-4.c
new file mode 100644
index 000000000..410225500
--- /dev/null
+++ b/kernel/x86_64/dgemv_t_microk_haswell-4.c
@@ -0,0 +1,127 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+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.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "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 OPENBLAS PROJECT 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.
+*****************************************************************************/
+
+#define HAVE_KERNEL_4x4 1
+static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) __attribute__ ((noinline));
+
+static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
+{
+
+	BLASLONG register i = 0;
+
+	__asm__  __volatile__
+	(
+	"vzeroupper			         \n\t"
+	"vxorpd		%%ymm4 , %%ymm4, %%ymm4  \n\t"
+	"vxorpd		%%ymm5 , %%ymm5, %%ymm5  \n\t"
+	"vxorpd		%%ymm6 , %%ymm6, %%ymm6  \n\t"
+	"vxorpd		%%ymm7 , %%ymm7, %%ymm7  \n\t"
+
+        "testq          $0x04, %1                      \n\t"
+        "jz             .L08LABEL%=                    \n\t"
+
+	"vmovups	(%2,%0,8), %%ymm12       \n\t"	// 4 * x
+
+	"vfmadd231pd   (%4,%0,8), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231pd   (%5,%0,8), %%ymm12, %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%0,8), %%ymm12, %%ymm6      \n\t" 
+	"vfmadd231pd   (%7,%0,8), %%ymm12, %%ymm7      \n\t" 
+
+        "addq		$4 , %0	  	 	      \n\t"
+	"subq	        $4 , %1			      \n\t"		
+
+        ".L08LABEL%=:                                  \n\t"
+
+        "cmpq           $0, %1                         \n\t"
+        "je             .L16END%=                      \n\t"
+
+	".align 16				 \n\t"
+	".L01LOOP%=:				 \n\t"
+	"prefetcht0	 384(%2,%0,8)		 \n\t"
+	"vmovups	(%2,%0,8), %%ymm12       \n\t"	// 4 * x
+	"vmovups      32(%2,%0,8), %%ymm13       \n\t"	// 4 * x
+
+	"prefetcht0	 384(%4,%0,8)		       \n\t"
+	"vfmadd231pd   (%4,%0,8), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231pd   (%5,%0,8), %%ymm12, %%ymm5      \n\t" 
+	"prefetcht0	 384(%5,%0,8)		       \n\t"
+	"vfmadd231pd 32(%4,%0,8), %%ymm13, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%5,%0,8), %%ymm13, %%ymm5      \n\t" 
+	"prefetcht0	 384(%6,%0,8)		       \n\t"
+	"vfmadd231pd   (%6,%0,8), %%ymm12, %%ymm6      \n\t" 
+	"vfmadd231pd   (%7,%0,8), %%ymm12, %%ymm7      \n\t" 
+	"prefetcht0	 384(%7,%0,8)		       \n\t"
+	"vfmadd231pd 32(%6,%0,8), %%ymm13, %%ymm6      \n\t" 
+	"vfmadd231pd 32(%7,%0,8), %%ymm13, %%ymm7      \n\t" 
+
+        "addq		$8 , %0	  	 	      \n\t"
+	"subq	        $8 , %1			      \n\t"		
+	"jnz		.L01LOOP%=		      \n\t"
+
+        ".L16END%=:                                   \n\t"
+
+	"vextractf128   $1 , %%ymm4, %%xmm12	      \n\t"
+	"vextractf128   $1 , %%ymm5, %%xmm13	      \n\t"
+	"vextractf128   $1 , %%ymm6, %%xmm14	      \n\t"
+	"vextractf128   $1 , %%ymm7, %%xmm15	      \n\t"
+
+	"vaddpd		%%xmm4, %%xmm12, %%xmm4       \n\t"
+	"vaddpd		%%xmm5, %%xmm13, %%xmm5       \n\t"
+	"vaddpd		%%xmm6, %%xmm14, %%xmm6       \n\t"
+	"vaddpd		%%xmm7, %%xmm15, %%xmm7       \n\t"
+
+        "vhaddpd        %%xmm4, %%xmm4, %%xmm4  \n\t"
+        "vhaddpd        %%xmm5, %%xmm5, %%xmm5  \n\t"
+        "vhaddpd        %%xmm6, %%xmm6, %%xmm6  \n\t"
+        "vhaddpd        %%xmm7, %%xmm7, %%xmm7  \n\t"
+
+        "vmovsd         %%xmm4,    (%3)         \n\t"
+        "vmovsd         %%xmm5,   8(%3)         \n\t"
+        "vmovsd         %%xmm6,  16(%3)         \n\t"
+        "vmovsd         %%xmm7,  24(%3)         \n\t"
+
+	"vzeroupper			 \n\t"
+
+	:
+        : 
+          "r" (i),	// 0	
+	  "r" (n),  	// 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (ap[0]),  // 4
+          "r" (ap[1]),  // 5
+          "r" (ap[2]),  // 6
+          "r" (ap[3])   // 7
+	: "cc", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7",
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+