Merge pull request #571 from wernsaar/develop

added optimized cscal and zscal functions

kernel/x86_64/KERNEL.BULLDOZER

@@ -1,4 +1,5 @@
 DSCALKERNEL = dscal.c
+ZSCALKERNEL = zscal.c
 
 DAXPYKERNEL = daxpy.c
 CAXPYKERNEL = caxpy.c

kernel/x86_64/KERNEL.HASWELL

@@ -1,4 +1,6 @@
 DSCALKERNEL = dscal.c
+CSCALKERNEL = cscal.c
+ZSCALKERNEL = zscal.c
 
 SGEMVNKERNEL = sgemv_n_4.c
 SGEMVTKERNEL = sgemv_t_4.c

kernel/x86_64/cscal.c

@@ -0,0 +1,446 @@
+/***************************************************************************
+Copyright (c) 2013 - 2015, 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(HASWELL)
+#include "cscal_microk_haswell-2.c"
+#endif
+
+
+#if !defined(HAVE_KERNEL_16)
+
+static void cscal_kernel_16( BLASLONG n, FLOAT *alpha , FLOAT *x )  __attribute__ ((noinline));
+static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha , FLOAT *x )  __attribute__ ((noinline));
+static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha , FLOAT *x )  __attribute__ ((noinline));
+static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha , FLOAT *x )  __attribute__ ((noinline));
+
+static void cscal_kernel_16( BLASLONG n, FLOAT *alpha , FLOAT *x )
+{
+
+	BLASLONG i;
+	FLOAT da_r = alpha[0];
+	FLOAT da_i = alpha[1];
+	FLOAT t0,t1,t2,t3;
+
+	for( i=0; i<n; i+=4 )
+	{
+		t0 = da_r *x[0] - da_i *x[1];	
+		t1 = da_r *x[2] - da_i *x[3];	
+		t2 = da_r *x[4] - da_i *x[5];	
+		t3 = da_r *x[6] - da_i *x[7];	
+
+		x[1] = da_r * x[1] + da_i * x[0];
+		x[3] = da_r * x[3] + da_i * x[2];
+		x[5] = da_r * x[5] + da_i * x[4];
+		x[7] = da_r * x[7] + da_i * x[6];
+		
+		x[0] = t0;
+		x[2] = t1;
+		x[4] = t2;
+		x[6] = t3;
+
+		x+=8;
+	}
+
+}
+
+
+static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha , FLOAT *x )
+{
+
+	BLASLONG i;
+	FLOAT da_i = alpha[1];
+	FLOAT t0,t1,t2,t3;
+
+	for( i=0; i<n; i+=4 )
+	{
+		t0 =  - da_i *x[1];	
+		t1 =  - da_i *x[3];	
+		t2 =  - da_i *x[5];	
+		t3 =  - da_i *x[7];	
+
+		x[1] =  da_i * x[0];
+		x[3] =  da_i * x[2];
+		x[5] =  da_i * x[4];
+		x[7] =  da_i * x[6];
+		
+		x[0] = t0;
+		x[2] = t1;
+		x[4] = t2;
+		x[6] = t3;
+
+		x+=8;
+	}
+
+}
+
+
+static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha , FLOAT *x )
+{
+
+	BLASLONG i;
+	FLOAT da_r = alpha[0];
+	FLOAT t0,t1,t2,t3;
+
+	for( i=0; i<n; i+=4 )
+	{
+		t0 = da_r *x[0];	
+		t1 = da_r *x[2];	
+		t2 = da_r *x[4];	
+		t3 = da_r *x[6];	
+
+		x[1] = da_r * x[1];
+		x[3] = da_r * x[3];
+		x[5] = da_r * x[5];
+		x[7] = da_r * x[7];
+		
+		x[0] = t0;
+		x[2] = t1;
+		x[4] = t2;
+		x[6] = t3;
+
+		x+=8;
+	}
+
+}
+
+
+
+static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha , FLOAT *x )
+{
+
+	BLASLONG i;
+	for( i=0; i<n; i+=4 )
+	{
+		x[0] = 0.0;	
+		x[1] = 0.0;	
+		x[2] = 0.0;	
+		x[3] = 0.0;	
+		x[4] = 0.0;	
+		x[5] = 0.0;	
+		x[6] = 0.0;	
+		x[7] = 0.0;	
+		x+=8;
+	}
+
+}
+
+#endif
+
+static void cscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_x)  __attribute__ ((noinline));
+
+static void cscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_x)
+{
+
+	BLASLONG i;
+	BLASLONG inc_x2 = 2 * inc_x;
+	BLASLONG inc_x3 = inc_x2 + inc_x;
+	FLOAT t0,t1,t2,t3;
+	FLOAT da_r = alpha[0];
+	FLOAT da_i = alpha[1];
+
+	for ( i=0; i<n; i+=4 )
+	{
+		t0 = da_r * x[0]         - da_i *x[1];	
+		t1 = da_r * x[inc_x]     - da_i *x[inc_x  + 1];	
+		t2 = da_r * x[inc_x2]    - da_i *x[inc_x2 + 1];	
+		t3 = da_r * x[inc_x3]    - da_i *x[inc_x3 + 1];	
+
+		x[1]               = da_i * x[0]       + da_r * x[1];
+		x[inc_x  +1]       = da_i * x[inc_x]   + da_r * x[inc_x  +1];
+		x[inc_x2 +1]       = da_i * x[inc_x2]  + da_r * x[inc_x2 +1];
+		x[inc_x3 +1]       = da_i * x[inc_x3]  + da_r * x[inc_x3 +1];
+
+		x[0]        = t0;
+		x[inc_x]    = t1;
+		x[inc_x2]   = t2;
+		x[inc_x3]   = t3;
+
+		x+=4*inc_x;
+
+	}
+
+
+}
+
+
+int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
+{
+	BLASLONG i=0,j=0;
+	FLOAT temp0;
+	FLOAT temp1;
+	FLOAT alpha[2];
+
+	if ( inc_x != 1 )
+	{
+		inc_x <<= 1;
+
+		if ( da_r == 0.0 )
+		{
+
+			BLASLONG n1 = n & -2;
+
+			if ( da_i == 0.0 )
+			{
+
+				while(j < n1)
+				{
+			
+					x[i]=0.0;
+					x[i+1]=0.0;
+					x[i+inc_x]=0.0;
+					x[i+1+inc_x]=0.0;
+					i += 2*inc_x ;
+					j+=2;
+
+				}
+
+				while(j < n)
+				{
+			
+					x[i]=0.0;
+					x[i+1]=0.0;
+					i += inc_x ;
+					j++;
+
+				}
+
+			}
+			else
+			{
+
+				while(j < n1)
+				{
+			
+					temp0        = -da_i * x[i+1];
+					x[i+1]       =  da_i * x[i];
+					x[i]         =  temp0;
+					temp1        = -da_i * x[i+1+inc_x];
+					x[i+1+inc_x] =  da_i * x[i+inc_x];
+					x[i+inc_x]   =  temp1;
+					i += 2*inc_x ;
+					j+=2;
+
+				}
+
+				while(j < n)
+				{
+			
+					temp0        = -da_i * x[i+1];
+					x[i+1]       =  da_i * x[i];
+					x[i]         =  temp0;
+					i += inc_x ;
+					j++;
+
+				}
+
+
+
+			}
+
+		}
+		else
+		{
+
+
+			if ( da_i == 0.0 )
+			{
+				BLASLONG n1 = n & -2;
+
+				while(j < n1)
+				{
+			
+					temp0        =  da_r * x[i];
+					x[i+1]       =  da_r * x[i+1];
+					x[i]         =  temp0;
+					temp1        =  da_r * x[i+inc_x];
+					x[i+1+inc_x] =  da_r * x[i+1+inc_x];
+					x[i+inc_x]   =  temp1;
+					i += 2*inc_x ;
+					j+=2;
+
+				}
+
+				while(j < n)
+				{
+			
+					temp0        =  da_r * x[i];
+					x[i+1]       =  da_r * x[i+1];
+					x[i]         =  temp0;
+					i += inc_x ;
+					j++;
+
+				}
+
+			}
+			else
+			{
+
+				BLASLONG n1 = n & -8;
+				if ( n1 > 0 )
+				{
+					alpha[0] = da_r;
+			                alpha[1] = da_i;
+					cscal_kernel_inc_8(n1, alpha, x, inc_x);
+					j = n1 ;
+					i = n1 * inc_x;
+				}
+
+				while(j < n)
+				{
+
+					temp0        =  da_r * x[i]   - da_i * x[i+1];
+					x[i+1]       =  da_r * x[i+1] + da_i * x[i];
+					x[i]         =  temp0;
+					i += inc_x ;
+					j++;
+
+				}
+
+			}
+
+		}
+
+		return(0);
+	}
+
+
+	BLASLONG n1 = n & -16;
+	if ( n1 > 0 )
+	{
+
+		alpha[0] = da_r;
+		alpha[1] = da_i;
+	
+		if ( da_r == 0.0 )
+			if ( da_i == 0 )
+				cscal_kernel_16_zero(n1 , alpha , x);
+			else
+				cscal_kernel_16_zero_r(n1 , alpha , x);
+		else
+			if ( da_i == 0 )
+				cscal_kernel_16_zero_i(n1 , alpha , x);
+			else
+				cscal_kernel_16(n1 , alpha , x);
+
+		i = n1 << 1;
+		j = n1;
+	}
+
+
+	if ( da_r == 0.0 )
+	{
+
+		if ( da_i == 0.0 )
+		{
+
+			while(j < n)
+			{
+		
+					x[i]=0.0;
+					x[i+1]=0.0;
+					i += 2 ;
+					j++;
+
+			}
+
+		}
+		else
+		{
+
+			while(j < n)
+			{
+			
+				temp0        = -da_i * x[i+1];
+				x[i+1]       =  da_i * x[i];
+				x[i]         =  temp0;
+				i += 2 ;
+				j++;
+
+			}
+
+		}
+
+	}
+	else
+	{
+
+		if ( da_i == 0.0 )
+		{
+
+			while(j < n)
+			{
+			
+					temp0        =  da_r * x[i];
+					x[i+1]       =  da_r * x[i+1];
+					x[i]         =  temp0;
+					i += 2 ;
+					j++;
+
+			}
+
+		}
+		else
+		{
+
+			BLASLONG n2 = n & -2;
+
+			while(j < n2)
+			{
+
+				temp0        =  da_r * x[i]   - da_i * x[i+1];
+				temp1        =  da_r * x[i+2] - da_i * x[i+3];
+				x[i+1]       =  da_r * x[i+1] + da_i * x[i];
+				x[i+3]       =  da_r * x[i+3] + da_i * x[i+2];
+				x[i]         =  temp0;
+				x[i+2]       =  temp1;
+				i += 4 ;
+				j+=2;
+
+			}
+
+			while(j < n)
+			{
+
+				temp0        =  da_r * x[i]   - da_i * x[i+1];
+				x[i+1]       =  da_r * x[i+1] + da_i * x[i];
+				x[i]         =  temp0;
+				i += 2 ;
+				j++;
+
+			}
+
+		}		
+
+	}
+
+	return(0);
+}
+
+

kernel/x86_64/cscal_microk_haswell-2.c

@@ -0,0 +1,348 @@
+/***************************************************************************
+Copyright (c) 2014-2015, 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_16 1
+
+static void cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vbroadcastss		(%2), %%ymm0		    \n\t"  // da_r	
+	"vbroadcastss          4(%2), %%ymm1		    \n\t"  // da_i 	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%ymm4		    \n\t"
+	"vmovups	 -96(%1), %%ymm5		    \n\t"
+	"vmovups	 -64(%1), %%ymm6		    \n\t"
+	"vmovups	 -32(%1), %%ymm7		    \n\t"
+
+	"vpermilps	$0xb1 , %%ymm4, %%ymm12		    \n\t"
+	"vpermilps	$0xb1 , %%ymm5, %%ymm13		    \n\t"
+	"vpermilps	$0xb1 , %%ymm6, %%ymm14 	    \n\t"
+	"vpermilps	$0xb1 , %%ymm7, %%ymm15		    \n\t"
+
+	"subq	        $16, %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmulps		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmovups	   0(%1), %%ymm4		    \n\t"
+	"vmulps		%%ymm0, %%ymm5 , %%ymm9		    \n\t"
+	"vmovups	  32(%1), %%ymm5		    \n\t"
+	"vmulps		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 
+	"vmovups	  64(%1), %%ymm6		    \n\t"
+	"vmulps		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 
+	"vmovups	  96(%1), %%ymm7		    \n\t"
+
+	"vmulps		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubps	%%ymm12 , %%ymm8 , %%ymm8	    \n\t"
+	"vmulps		%%ymm1, %%ymm13, %%ymm13	    \n\t" 
+	"vaddsubps	%%ymm13 , %%ymm9 , %%ymm9	    \n\t"
+	"vmulps		%%ymm1, %%ymm14, %%ymm14	    \n\t" 
+	"vaddsubps	%%ymm14 , %%ymm10, %%ymm10	    \n\t"
+	"vmulps		%%ymm1, %%ymm15, %%ymm15	    \n\t" 
+	"vaddsubps	%%ymm15 , %%ymm11, %%ymm11	    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vpermilps	$0xb1 , %%ymm4, %%ymm12		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vpermilps	$0xb1 , %%ymm5, %%ymm13		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vpermilps	$0xb1 , %%ymm6, %%ymm14 	    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+	"vpermilps	$0xb1 , %%ymm7, %%ymm15		    \n\t"
+
+	"addq		$128 ,%1  	 	            \n\t"
+	"subq	        $16, %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+
+	"vmulps		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmulps		%%ymm0, %%ymm5 , %%ymm9		    \n\t"
+	"vmulps		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 
+	"vmulps		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 
+
+	"vmulps		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubps	%%ymm12 , %%ymm8 , %%ymm8	    \n\t"
+	"vmulps		%%ymm1, %%ymm13, %%ymm13	    \n\t" 
+	"vaddsubps	%%ymm13 , %%ymm9 , %%ymm9	    \n\t"
+	"vmulps		%%ymm1, %%ymm14, %%ymm14	    \n\t" 
+	"vaddsubps	%%ymm14 , %%ymm10, %%ymm10	    \n\t"
+	"vmulps		%%ymm1, %%ymm15, %%ymm15	    \n\t" 
+	"vaddsubps	%%ymm15 , %%ymm11, %%ymm11	    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "0", "1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vxorps	           %%ymm0, %%ymm0, %%ymm0	    \n\t"
+	"vbroadcastss          4(%2), %%ymm1		    \n\t"  // da_i 	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%ymm4		    \n\t"
+	"vmovups	 -96(%1), %%ymm5		    \n\t"
+	"vmovups	 -64(%1), %%ymm6		    \n\t"
+	"vmovups	 -32(%1), %%ymm7		    \n\t"
+
+	"vpermilps	$0xb1 , %%ymm4, %%ymm12		    \n\t"
+	"vpermilps	$0xb1 , %%ymm5, %%ymm13		    \n\t"
+	"vpermilps	$0xb1 , %%ymm6, %%ymm14 	    \n\t"
+	"vpermilps	$0xb1 , %%ymm7, %%ymm15		    \n\t"
+
+	"subq	        $16, %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmovups	   0(%1), %%ymm4		    \n\t"
+	"vmovups	  32(%1), %%ymm5		    \n\t"
+	"vmovups	  64(%1), %%ymm6		    \n\t"
+	"vmovups	  96(%1), %%ymm7		    \n\t"
+
+	"vmulps		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubps	%%ymm12 , %%ymm0 , %%ymm8	    \n\t"
+	"vmulps		%%ymm1, %%ymm13, %%ymm13	    \n\t" 
+	"vaddsubps	%%ymm13 , %%ymm0 , %%ymm9	    \n\t"
+	"vmulps		%%ymm1, %%ymm14, %%ymm14	    \n\t" 
+	"vaddsubps	%%ymm14 , %%ymm0 , %%ymm10	    \n\t"
+	"vmulps		%%ymm1, %%ymm15, %%ymm15	    \n\t" 
+	"vaddsubps	%%ymm15 , %%ymm0 , %%ymm11	    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vpermilps	$0xb1 , %%ymm4, %%ymm12		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vpermilps	$0xb1 , %%ymm5, %%ymm13		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vpermilps	$0xb1 , %%ymm6, %%ymm14 	    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+	"vpermilps	$0xb1 , %%ymm7, %%ymm15		    \n\t"
+
+	"addq		$128 ,%1  	 	            \n\t"
+	"subq	        $16, %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+	"vmulps		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubps	%%ymm12 , %%ymm0 , %%ymm8	    \n\t"
+	"vmulps		%%ymm1, %%ymm13, %%ymm13	    \n\t" 
+	"vaddsubps	%%ymm13 , %%ymm0 , %%ymm9	    \n\t"
+	"vmulps		%%ymm1, %%ymm14, %%ymm14	    \n\t" 
+	"vaddsubps	%%ymm14 , %%ymm0 , %%ymm10	    \n\t"
+	"vmulps		%%ymm1, %%ymm15, %%ymm15	    \n\t" 
+	"vaddsubps	%%ymm15 , %%ymm0 , %%ymm11	    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "0", "1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+
+static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vbroadcastss		(%2), %%ymm0		    \n\t"  // da_r	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%ymm4		    \n\t"
+	"vmovups	 -96(%1), %%ymm5		    \n\t"
+	"vmovups	 -64(%1), %%ymm6		    \n\t"
+	"vmovups	 -32(%1), %%ymm7		    \n\t"
+
+
+	"subq	        $16, %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmulps		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmovups	   0(%1), %%ymm4		    \n\t"
+	"vmulps		%%ymm0, %%ymm5 , %%ymm9		    \n\t"
+	"vmovups	  32(%1), %%ymm5		    \n\t"
+	"vmulps		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 
+	"vmovups	  64(%1), %%ymm6		    \n\t"
+	"vmulps		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 
+	"vmovups	  96(%1), %%ymm7		    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+
+	"addq		$128 ,%1  	 	            \n\t"
+	"subq	        $16, %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+
+	"vmulps		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmulps		%%ymm0, %%ymm5 , %%ymm9		    \n\t"
+	"vmulps		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 
+	"vmulps		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vxorps	           %%ymm0, %%ymm0, %%ymm0	    \n\t"
+
+	"addq	$128, %1				    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmovups	%%ymm0 , -128(%1)		    \n\t"
+	"vmovups	%%ymm0 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm0 ,  -64(%1)		    \n\t"
+	"vmovups	%%ymm0 ,  -32(%1)		    \n\t"
+
+	"addq		$128 ,%1  	 	            \n\t"
+	"subq	        $16, %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "0", "1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+

kernel/x86_64/zscal.c

@@ -0,0 +1,432 @@
+/***************************************************************************
+Copyright (c) 2013 - 2015, 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(HASWELL)
+#include "zscal_microk_haswell-2.c"
+#elif defined(BULLDOZER)
+#include "zscal_microk_bulldozer-2.c"
+#endif
+
+
+#if !defined(HAVE_KERNEL_8)
+
+static void zscal_kernel_8( BLASLONG n, FLOAT *alpha , FLOAT *x )  __attribute__ ((noinline));
+static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha , FLOAT *x )  __attribute__ ((noinline));
+static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha , FLOAT *x )  __attribute__ ((noinline));
+static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha , FLOAT *x )  __attribute__ ((noinline));
+
+static void zscal_kernel_8( BLASLONG n, FLOAT *alpha , FLOAT *x )
+{
+
+	BLASLONG i;
+	FLOAT da_r = alpha[0];
+	FLOAT da_i = alpha[1];
+	FLOAT t0,t1,t2,t3;
+
+	for( i=0; i<n; i+=4 )
+	{
+		t0 = da_r *x[0] - da_i *x[1];	
+		t1 = da_r *x[2] - da_i *x[3];	
+		t2 = da_r *x[4] - da_i *x[5];	
+		t3 = da_r *x[6] - da_i *x[7];	
+
+		x[1] = da_r * x[1] + da_i * x[0];
+		x[3] = da_r * x[3] + da_i * x[2];
+		x[5] = da_r * x[5] + da_i * x[4];
+		x[7] = da_r * x[7] + da_i * x[6];
+		
+		x[0] = t0;
+		x[2] = t1;
+		x[4] = t2;
+		x[6] = t3;
+
+		x+=8;
+	}
+
+}
+
+
+static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha , FLOAT *x )
+{
+
+	BLASLONG i;
+	FLOAT da_i = alpha[1];
+	FLOAT t0,t1,t2,t3;
+
+	for( i=0; i<n; i+=4 )
+	{
+		t0 =  - da_i *x[1];	
+		t1 =  - da_i *x[3];	
+		t2 =  - da_i *x[5];	
+		t3 =  - da_i *x[7];	
+
+		x[1] =  da_i * x[0];
+		x[3] =  da_i * x[2];
+		x[5] =  da_i * x[4];
+		x[7] =  da_i * x[6];
+		
+		x[0] = t0;
+		x[2] = t1;
+		x[4] = t2;
+		x[6] = t3;
+
+		x+=8;
+	}
+
+}
+
+
+static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha , FLOAT *x )
+{
+
+	BLASLONG i;
+	FLOAT da_r = alpha[0];
+	FLOAT t0,t1,t2,t3;
+
+	for( i=0; i<n; i+=4 )
+	{
+		t0 = da_r *x[0];	
+		t1 = da_r *x[2];	
+		t2 = da_r *x[4];	
+		t3 = da_r *x[6];	
+
+		x[1] = da_r * x[1];
+		x[3] = da_r * x[3];
+		x[5] = da_r * x[5];
+		x[7] = da_r * x[7];
+		
+		x[0] = t0;
+		x[2] = t1;
+		x[4] = t2;
+		x[6] = t3;
+
+		x+=8;
+	}
+
+}
+
+
+
+static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha , FLOAT *x )
+{
+
+	BLASLONG i;
+	for( i=0; i<n; i+=4 )
+	{
+		x[0] = 0.0;	
+		x[1] = 0.0;	
+		x[2] = 0.0;	
+		x[3] = 0.0;	
+		x[4] = 0.0;	
+		x[5] = 0.0;	
+		x[6] = 0.0;	
+		x[7] = 0.0;	
+		x+=8;
+	}
+
+}
+
+#endif
+
+static void zscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_x)  __attribute__ ((noinline));
+
+static void zscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_x)
+{
+
+	BLASLONG i;
+	BLASLONG inc_x2 = 2 * inc_x;
+	BLASLONG inc_x3 = inc_x2 + inc_x;
+	FLOAT t0,t1,t2,t3;
+	FLOAT da_r = alpha[0];
+	FLOAT da_i = alpha[1];
+
+	for ( i=0; i<n; i+=4 )
+	{
+		t0 = da_r * x[0]         - da_i *x[1];	
+		t1 = da_r * x[inc_x]     - da_i *x[inc_x  + 1];	
+		t2 = da_r * x[inc_x2]    - da_i *x[inc_x2 + 1];	
+		t3 = da_r * x[inc_x3]    - da_i *x[inc_x3 + 1];	
+
+		x[1]               = da_i * x[0]       + da_r * x[1];
+		x[inc_x  +1]       = da_i * x[inc_x]   + da_r * x[inc_x  +1];
+		x[inc_x2 +1]       = da_i * x[inc_x2]  + da_r * x[inc_x2 +1];
+		x[inc_x3 +1]       = da_i * x[inc_x3]  + da_r * x[inc_x3 +1];
+
+		x[0]        = t0;
+		x[inc_x]    = t1;
+		x[inc_x2]   = t2;
+		x[inc_x3]   = t3;
+
+		x+=4*inc_x;
+
+	}
+
+
+}
+
+
+int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
+{
+	BLASLONG i=0,j=0;
+	FLOAT temp0;
+	FLOAT temp1;
+	FLOAT alpha[2];
+
+	if ( inc_x != 1 )
+	{
+		inc_x <<= 1;
+
+		if ( da_r == 0.0 )
+		{
+
+			BLASLONG n1 = n & -2;
+
+			if ( da_i == 0.0 )
+			{
+
+				while(j < n1)
+				{
+			
+					x[i]=0.0;
+					x[i+1]=0.0;
+					x[i+inc_x]=0.0;
+					x[i+1+inc_x]=0.0;
+					i += 2*inc_x ;
+					j+=2;
+
+				}
+
+				while(j < n)
+				{
+			
+					x[i]=0.0;
+					x[i+1]=0.0;
+					i += inc_x ;
+					j++;
+
+				}
+
+			}
+			else
+			{
+
+				while(j < n1)
+				{
+			
+					temp0        = -da_i * x[i+1];
+					x[i+1]       =  da_i * x[i];
+					x[i]         =  temp0;
+					temp1        = -da_i * x[i+1+inc_x];
+					x[i+1+inc_x] =  da_i * x[i+inc_x];
+					x[i+inc_x]   =  temp1;
+					i += 2*inc_x ;
+					j+=2;
+
+				}
+
+				while(j < n)
+				{
+			
+					temp0        = -da_i * x[i+1];
+					x[i+1]       =  da_i * x[i];
+					x[i]         =  temp0;
+					i += inc_x ;
+					j++;
+
+				}
+
+
+
+			}
+
+		}
+		else
+		{
+
+
+			if ( da_i == 0.0 )
+			{
+				BLASLONG n1 = n & -2;
+
+				while(j < n1)
+				{
+			
+					temp0        =  da_r * x[i];
+					x[i+1]       =  da_r * x[i+1];
+					x[i]         =  temp0;
+					temp1        =  da_r * x[i+inc_x];
+					x[i+1+inc_x] =  da_r * x[i+1+inc_x];
+					x[i+inc_x]   =  temp1;
+					i += 2*inc_x ;
+					j+=2;
+
+				}
+
+				while(j < n)
+				{
+			
+					temp0        =  da_r * x[i];
+					x[i+1]       =  da_r * x[i+1];
+					x[i]         =  temp0;
+					i += inc_x ;
+					j++;
+
+				}
+
+			}
+			else
+			{
+
+				BLASLONG n1 = n & -8;
+				if ( n1 > 0 )
+				{
+					alpha[0] = da_r;
+			                alpha[1] = da_i;
+					zscal_kernel_inc_8(n1, alpha, x, inc_x);
+					j = n1 ;
+					i = n1 * inc_x;
+				}
+
+				while(j < n)
+				{
+
+					temp0        =  da_r * x[i]   - da_i * x[i+1];
+					x[i+1]       =  da_r * x[i+1] + da_i * x[i];
+					x[i]         =  temp0;
+					i += inc_x ;
+					j++;
+
+				}
+
+			}
+
+		}
+
+		return(0);
+	}
+
+
+	BLASLONG n1 = n & -8;
+	if ( n1 > 0 )
+	{
+
+		alpha[0] = da_r;
+		alpha[1] = da_i;
+	
+		if ( da_r == 0.0 )
+			if ( da_i == 0 )
+				zscal_kernel_8_zero(n1 , alpha , x);
+			else
+				zscal_kernel_8_zero_r(n1 , alpha , x);
+		else
+			if ( da_i == 0 )
+				zscal_kernel_8_zero_i(n1 , alpha , x);
+			else
+				zscal_kernel_8(n1 , alpha , x);
+
+		i = n1 << 1;
+		j = n1;
+	}
+
+
+	if ( da_r == 0.0 )
+	{
+
+		if ( da_i == 0.0 )
+		{
+
+			while(j < n)
+			{
+		
+					x[i]=0.0;
+					x[i+1]=0.0;
+					i += 2 ;
+					j++;
+
+			}
+
+		}
+		else
+		{
+
+			while(j < n)
+			{
+			
+				temp0        = -da_i * x[i+1];
+				x[i+1]       =  da_i * x[i];
+				x[i]         =  temp0;
+				i += 2 ;
+				j++;
+
+			}
+
+		}
+
+	}
+	else
+	{
+
+		if ( da_i == 0.0 )
+		{
+
+			while(j < n)
+			{
+			
+					temp0        =  da_r * x[i];
+					x[i+1]       =  da_r * x[i+1];
+					x[i]         =  temp0;
+					i += 2 ;
+					j++;
+
+			}
+
+		}
+		else
+		{
+
+			while(j < n)
+			{
+
+				temp0        =  da_r * x[i]   - da_i * x[i+1];
+				x[i+1]       =  da_r * x[i+1] + da_i * x[i];
+				x[i]         =  temp0;
+				i += 2 ;
+				j++;
+
+			}
+
+		}		
+
+	}
+
+	return(0);
+}
+
+

kernel/x86_64/zscal_microk_bulldozer-2.c

@@ -0,0 +1,348 @@
+/***************************************************************************
+Copyright (c) 2014-2015, 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_8 1
+
+static void zscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void zscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vmovddup    		(%2), %%xmm0		    \n\t"  // da_r	
+	"vmovddup              8(%2), %%xmm1		    \n\t"  // da_i 	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%xmm4		    \n\t"
+	"vmovups	-112(%1), %%xmm5		    \n\t"
+	"vmovups	 -96(%1), %%xmm6		    \n\t"
+	"vmovups	 -80(%1), %%xmm7		    \n\t"
+
+	"vpermilpd	$0x01 , %%xmm4, %%xmm12		    \n\t"
+	"vpermilpd	$0x01 , %%xmm5, %%xmm13		    \n\t"
+	"vpermilpd	$0x01 , %%xmm6, %%xmm14 	    \n\t"
+	"vpermilpd	$0x01 , %%xmm7, %%xmm15		    \n\t"
+
+	"subq	        $4 , %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	"prefetcht0     192(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmulpd		%%xmm0, %%xmm4 , %%xmm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmovups	 -64(%1), %%xmm4		    \n\t"
+	"vmulpd		%%xmm0, %%xmm5 , %%xmm9		    \n\t"
+	"vmovups	 -48(%1), %%xmm5		    \n\t"
+	"vmulpd		%%xmm0, %%xmm6 , %%xmm10	    \n\t" 
+	"vmovups	 -32(%1), %%xmm6		    \n\t"
+	"vmulpd		%%xmm0, %%xmm7 , %%xmm11	    \n\t" 
+	"vmovups	 -16(%1), %%xmm7		    \n\t"
+
+	"vmulpd		%%xmm1, %%xmm12, %%xmm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubpd	%%xmm12 , %%xmm8 , %%xmm8	    \n\t"
+	"vmulpd		%%xmm1, %%xmm13, %%xmm13	    \n\t" 
+	"vaddsubpd	%%xmm13 , %%xmm9 , %%xmm9	    \n\t"
+	"vmulpd		%%xmm1, %%xmm14, %%xmm14	    \n\t" 
+	"vaddsubpd	%%xmm14 , %%xmm10, %%xmm10	    \n\t"
+	"vmulpd		%%xmm1, %%xmm15, %%xmm15	    \n\t" 
+	"vaddsubpd	%%xmm15 , %%xmm11, %%xmm11	    \n\t"
+
+	"vmovups	%%xmm8 , -128(%1)		    \n\t"
+	"vmovups	%%xmm9 , -112(%1)		    \n\t"
+	"vpermilpd	$0x01 , %%xmm4, %%xmm12		    \n\t"
+	"vpermilpd	$0x01 , %%xmm5, %%xmm13		    \n\t"
+	"vmovups	%%xmm10,  -96(%1)		    \n\t"
+	"vmovups	%%xmm11,  -80(%1)		    \n\t"
+	"vpermilpd	$0x01 , %%xmm6, %%xmm14 	    \n\t"
+	"vpermilpd	$0x01 , %%xmm7, %%xmm15		    \n\t"
+
+	"addq		$64  ,%1  	 	            \n\t"
+	"subq	        $4 , %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+
+	"vmulpd		%%xmm0, %%xmm4 , %%xmm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmulpd		%%xmm0, %%xmm5 , %%xmm9		    \n\t"
+	"vmulpd		%%xmm0, %%xmm6 , %%xmm10	    \n\t" 
+	"vmulpd		%%xmm0, %%xmm7 , %%xmm11	    \n\t" 
+
+	"vmulpd		%%xmm1, %%xmm12, %%xmm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubpd	%%xmm12 , %%xmm8 , %%xmm8	    \n\t"
+	"vmulpd		%%xmm1, %%xmm13, %%xmm13	    \n\t" 
+	"vaddsubpd	%%xmm13 , %%xmm9 , %%xmm9	    \n\t"
+	"vmulpd		%%xmm1, %%xmm14, %%xmm14	    \n\t" 
+	"vaddsubpd	%%xmm14 , %%xmm10, %%xmm10	    \n\t"
+	"vmulpd		%%xmm1, %%xmm15, %%xmm15	    \n\t" 
+	"vaddsubpd	%%xmm15 , %%xmm11, %%xmm11	    \n\t"
+
+	"vmovups	%%xmm8 , -128(%1)		    \n\t"
+	"vmovups	%%xmm9 , -112(%1)		    \n\t"
+	"vmovups	%%xmm10,  -96(%1)		    \n\t"
+	"vmovups	%%xmm11,  -80(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vxorpd	           %%xmm0, %%xmm0, %%xmm0	    \n\t"
+	"vmovddup              8(%2), %%xmm1		    \n\t"  // da_i 	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%xmm4		    \n\t"
+	"vmovups	-112(%1), %%xmm5		    \n\t"
+	"vmovups	 -96(%1), %%xmm6		    \n\t"
+	"vmovups	 -80(%1), %%xmm7		    \n\t"
+
+	"vpermilpd	$0x01 , %%xmm4, %%xmm12		    \n\t"
+	"vpermilpd	$0x01 , %%xmm5, %%xmm13		    \n\t"
+	"vpermilpd	$0x01 , %%xmm6, %%xmm14 	    \n\t"
+	"vpermilpd	$0x01 , %%xmm7, %%xmm15		    \n\t"
+
+	"subq	        $4 , %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmovups	 -64(%1), %%xmm4		    \n\t"
+	"vmovups	 -48(%1), %%xmm5		    \n\t"
+	"vmovups	 -32(%1), %%xmm6		    \n\t"
+	"vmovups	 -16(%1), %%xmm7		    \n\t"
+
+	"vmulpd		%%xmm1, %%xmm12, %%xmm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubpd	%%xmm12 , %%xmm0 , %%xmm8	    \n\t"
+	"vmulpd		%%xmm1, %%xmm13, %%xmm13	    \n\t" 
+	"vaddsubpd	%%xmm13 , %%xmm0 , %%xmm9	    \n\t"
+	"vmulpd		%%xmm1, %%xmm14, %%xmm14	    \n\t" 
+	"vaddsubpd	%%xmm14 , %%xmm0 , %%xmm10	    \n\t"
+	"vmulpd		%%xmm1, %%xmm15, %%xmm15	    \n\t" 
+	"vaddsubpd	%%xmm15 , %%xmm0 , %%xmm11	    \n\t"
+
+	"vmovups	%%xmm8 , -128(%1)		    \n\t"
+	"vpermilpd	$0x01 , %%xmm4, %%xmm12		    \n\t"
+	"vmovups	%%xmm9 , -112(%1)		    \n\t"
+	"vpermilpd	$0x01 , %%xmm5, %%xmm13		    \n\t"
+	"vmovups	%%xmm10,  -96(%1)		    \n\t"
+	"vpermilpd	$0x01 , %%xmm6, %%xmm14 	    \n\t"
+	"vmovups	%%xmm11,  -80(%1)		    \n\t"
+	"vpermilpd	$0x01 , %%xmm7, %%xmm15		    \n\t"
+
+	"addq		$64  ,%1  	 	            \n\t"
+	"subq	        $4 , %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+	"vmulpd		%%xmm1, %%xmm12, %%xmm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubpd	%%xmm12 , %%xmm0 , %%xmm8	    \n\t"
+	"vmulpd		%%xmm1, %%xmm13, %%xmm13	    \n\t" 
+	"vaddsubpd	%%xmm13 , %%xmm0 , %%xmm9	    \n\t"
+	"vmulpd		%%xmm1, %%xmm14, %%xmm14	    \n\t" 
+	"vaddsubpd	%%xmm14 , %%xmm0 , %%xmm10	    \n\t"
+	"vmulpd		%%xmm1, %%xmm15, %%xmm15	    \n\t" 
+	"vaddsubpd	%%xmm15 , %%xmm0 , %%xmm11	    \n\t"
+
+	"vmovups	%%xmm8 , -128(%1)		    \n\t"
+	"vmovups	%%xmm9 , -112(%1)		    \n\t"
+	"vmovups	%%xmm10,  -96(%1)		    \n\t"
+	"vmovups	%%xmm11,  -80(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+
+static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vmovddup    		(%2), %%xmm0		    \n\t"  // da_r	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%xmm4		    \n\t"
+	"vmovups	-112(%1), %%xmm5		    \n\t"
+	"vmovups	 -96(%1), %%xmm6		    \n\t"
+	"vmovups	 -80(%1), %%xmm7		    \n\t"
+
+
+	"subq	        $4 , %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmulpd		%%xmm0, %%xmm4 , %%xmm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmovups	 -64(%1), %%xmm4		    \n\t"
+	"vmulpd		%%xmm0, %%xmm5 , %%xmm9		    \n\t"
+	"vmovups	 -48(%1), %%xmm5		    \n\t"
+	"vmulpd		%%xmm0, %%xmm6 , %%xmm10	    \n\t" 
+	"vmovups	 -32(%1), %%xmm6		    \n\t"
+	"vmulpd		%%xmm0, %%xmm7 , %%xmm11	    \n\t" 
+	"vmovups	 -16(%1), %%xmm7		    \n\t"
+
+	"vmovups	%%xmm8 , -128(%1)		    \n\t"
+	"vmovups	%%xmm9 , -112(%1)		    \n\t"
+	"vmovups	%%xmm10,  -96(%1)		    \n\t"
+	"vmovups	%%xmm11,  -80(%1)		    \n\t"
+
+	"addq		$64  ,%1  	 	            \n\t"
+	"subq	        $4 , %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+
+	"vmulpd		%%xmm0, %%xmm4 , %%xmm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmulpd		%%xmm0, %%xmm5 , %%xmm9		    \n\t"
+	"vmulpd		%%xmm0, %%xmm6 , %%xmm10	    \n\t" 
+	"vmulpd		%%xmm0, %%xmm7 , %%xmm11	    \n\t" 
+
+	"vmovups	%%xmm8 , -128(%1)		    \n\t"
+	"vmovups	%%xmm9 , -112(%1)		    \n\t"
+	"vmovups	%%xmm10,  -96(%1)		    \n\t"
+	"vmovups	%%xmm11,  -80(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vxorpd	           %%xmm0, %%xmm0, %%xmm0	    \n\t"
+
+	"addq	$128, %1				    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmovups	%%xmm0 , -128(%1)		    \n\t"
+	"vmovups	%%xmm0 , -112(%1)		    \n\t"
+	"vmovups	%%xmm0 ,  -96(%1)		    \n\t"
+	"vmovups	%%xmm0 ,  -80(%1)		    \n\t"
+
+	"addq		$64  ,%1  	 	            \n\t"
+	"subq	        $4 , %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+

kernel/x86_64/zscal_microk_haswell-2.c

@@ -0,0 +1,348 @@
+/***************************************************************************
+Copyright (c) 2014-2015, 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_8 1
+
+static void zscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void zscal_kernel_8( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vbroadcastsd		(%2), %%ymm0		    \n\t"  // da_r	
+	"vbroadcastsd          8(%2), %%ymm1		    \n\t"  // da_i 	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%ymm4		    \n\t"
+	"vmovups	 -96(%1), %%ymm5		    \n\t"
+	"vmovups	 -64(%1), %%ymm6		    \n\t"
+	"vmovups	 -32(%1), %%ymm7		    \n\t"
+
+	"vpermilpd	$0x05 , %%ymm4, %%ymm12		    \n\t"
+	"vpermilpd	$0x05 , %%ymm5, %%ymm13		    \n\t"
+	"vpermilpd	$0x05 , %%ymm6, %%ymm14 	    \n\t"
+	"vpermilpd	$0x05 , %%ymm7, %%ymm15		    \n\t"
+
+	"subq	        $8 , %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmulpd		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmovups	   0(%1), %%ymm4		    \n\t"
+	"vmulpd		%%ymm0, %%ymm5 , %%ymm9		    \n\t"
+	"vmovups	  32(%1), %%ymm5		    \n\t"
+	"vmulpd		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 
+	"vmovups	  64(%1), %%ymm6		    \n\t"
+	"vmulpd		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 
+	"vmovups	  96(%1), %%ymm7		    \n\t"
+
+	"vmulpd		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubpd	%%ymm12 , %%ymm8 , %%ymm8	    \n\t"
+	"vmulpd		%%ymm1, %%ymm13, %%ymm13	    \n\t" 
+	"vaddsubpd	%%ymm13 , %%ymm9 , %%ymm9	    \n\t"
+	"vmulpd		%%ymm1, %%ymm14, %%ymm14	    \n\t" 
+	"vaddsubpd	%%ymm14 , %%ymm10, %%ymm10	    \n\t"
+	"vmulpd		%%ymm1, %%ymm15, %%ymm15	    \n\t" 
+	"vaddsubpd	%%ymm15 , %%ymm11, %%ymm11	    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vpermilpd	$0x05 , %%ymm4, %%ymm12		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vpermilpd	$0x05 , %%ymm5, %%ymm13		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vpermilpd	$0x05 , %%ymm6, %%ymm14 	    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+	"vpermilpd	$0x05 , %%ymm7, %%ymm15		    \n\t"
+
+	"addq		$128 ,%1  	 	            \n\t"
+	"subq	        $8 , %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+
+	"vmulpd		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmulpd		%%ymm0, %%ymm5 , %%ymm9		    \n\t"
+	"vmulpd		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 
+	"vmulpd		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 
+
+	"vmulpd		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubpd	%%ymm12 , %%ymm8 , %%ymm8	    \n\t"
+	"vmulpd		%%ymm1, %%ymm13, %%ymm13	    \n\t" 
+	"vaddsubpd	%%ymm13 , %%ymm9 , %%ymm9	    \n\t"
+	"vmulpd		%%ymm1, %%ymm14, %%ymm14	    \n\t" 
+	"vaddsubpd	%%ymm14 , %%ymm10, %%ymm10	    \n\t"
+	"vmulpd		%%ymm1, %%ymm15, %%ymm15	    \n\t" 
+	"vaddsubpd	%%ymm15 , %%ymm11, %%ymm11	    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vxorpd	           %%ymm0, %%ymm0, %%ymm0	    \n\t"
+	"vbroadcastsd          8(%2), %%ymm1		    \n\t"  // da_i 	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%ymm4		    \n\t"
+	"vmovups	 -96(%1), %%ymm5		    \n\t"
+	"vmovups	 -64(%1), %%ymm6		    \n\t"
+	"vmovups	 -32(%1), %%ymm7		    \n\t"
+
+	"vpermilpd	$0x05 , %%ymm4, %%ymm12		    \n\t"
+	"vpermilpd	$0x05 , %%ymm5, %%ymm13		    \n\t"
+	"vpermilpd	$0x05 , %%ymm6, %%ymm14 	    \n\t"
+	"vpermilpd	$0x05 , %%ymm7, %%ymm15		    \n\t"
+
+	"subq	        $8 , %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmovups	   0(%1), %%ymm4		    \n\t"
+	"vmovups	  32(%1), %%ymm5		    \n\t"
+	"vmovups	  64(%1), %%ymm6		    \n\t"
+	"vmovups	  96(%1), %%ymm7		    \n\t"
+
+	"vmulpd		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubpd	%%ymm12 , %%ymm0 , %%ymm8	    \n\t"
+	"vmulpd		%%ymm1, %%ymm13, %%ymm13	    \n\t" 
+	"vaddsubpd	%%ymm13 , %%ymm0 , %%ymm9	    \n\t"
+	"vmulpd		%%ymm1, %%ymm14, %%ymm14	    \n\t" 
+	"vaddsubpd	%%ymm14 , %%ymm0 , %%ymm10	    \n\t"
+	"vmulpd		%%ymm1, %%ymm15, %%ymm15	    \n\t" 
+	"vaddsubpd	%%ymm15 , %%ymm0 , %%ymm11	    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vpermilpd	$0x05 , %%ymm4, %%ymm12		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vpermilpd	$0x05 , %%ymm5, %%ymm13		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vpermilpd	$0x05 , %%ymm6, %%ymm14 	    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+	"vpermilpd	$0x05 , %%ymm7, %%ymm15		    \n\t"
+
+	"addq		$128 ,%1  	 	            \n\t"
+	"subq	        $8 , %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+	"vmulpd		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0
+	"vaddsubpd	%%ymm12 , %%ymm0 , %%ymm8	    \n\t"
+	"vmulpd		%%ymm1, %%ymm13, %%ymm13	    \n\t" 
+	"vaddsubpd	%%ymm13 , %%ymm0 , %%ymm9	    \n\t"
+	"vmulpd		%%ymm1, %%ymm14, %%ymm14	    \n\t" 
+	"vaddsubpd	%%ymm14 , %%ymm0 , %%ymm10	    \n\t"
+	"vmulpd		%%ymm1, %%ymm15, %%ymm15	    \n\t" 
+	"vaddsubpd	%%ymm15 , %%ymm0 , %%ymm11	    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+
+static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vbroadcastsd		(%2), %%ymm0		    \n\t"  // da_r	
+
+	"addq	$128, %1				    \n\t"
+
+	"vmovups	-128(%1), %%ymm4		    \n\t"
+	"vmovups	 -96(%1), %%ymm5		    \n\t"
+	"vmovups	 -64(%1), %%ymm6		    \n\t"
+	"vmovups	 -32(%1), %%ymm7		    \n\t"
+
+
+	"subq	        $8 , %0			            \n\t"		
+	"jz	2f					    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmulpd		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmovups	   0(%1), %%ymm4		    \n\t"
+	"vmulpd		%%ymm0, %%ymm5 , %%ymm9		    \n\t"
+	"vmovups	  32(%1), %%ymm5		    \n\t"
+	"vmulpd		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 
+	"vmovups	  64(%1), %%ymm6		    \n\t"
+	"vmulpd		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 
+	"vmovups	  96(%1), %%ymm7		    \n\t"
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+
+	"addq		$128 ,%1  	 	            \n\t"
+	"subq	        $8 , %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"2:				            	    \n\t"
+
+
+	"vmulpd		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1
+	"vmulpd		%%ymm0, %%ymm5 , %%ymm9		    \n\t"
+	"vmulpd		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 
+	"vmulpd		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 
+
+	"vmovups	%%ymm8 , -128(%1)		    \n\t"
+	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm10,  -64(%1)		    \n\t"
+	"vmovups	%%ymm11,  -32(%1)		    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));
+
+static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)
+{
+
+
+	__asm__  __volatile__
+	(
+	"vxorpd	           %%ymm0, %%ymm0, %%ymm0	    \n\t"
+
+	"addq	$128, %1				    \n\t"
+
+	".align 16				            \n\t"
+	"1:				            	    \n\t"
+
+	//"prefetcht0     128(%1)				    \n\t"
+	// ".align 2				            \n\t"
+
+	"vmovups	%%ymm0 , -128(%1)		    \n\t"
+	"vmovups	%%ymm0 ,  -96(%1)		    \n\t"
+	"vmovups	%%ymm0 ,  -64(%1)		    \n\t"
+	"vmovups	%%ymm0 ,  -32(%1)		    \n\t"
+
+	"addq		$128 ,%1  	 	            \n\t"
+	"subq	        $8 , %0			            \n\t"		
+	"jnz		1b		             	    \n\t"
+
+	"vzeroupper					    \n\t"
+
+	:
+        : 
+	  "r" (n),  	// 0
+          "r" (x),      // 1
+          "r" (alpha)   // 2
+	: "cc", "%0", "%1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+