Optimized standard Blas Level-1,2 (excluding nrm2 functions) for z13 (double precision)

CONTRIBUTORS.md

@@ -166,5 +166,5 @@ In chronological order:
   * [2017-01-01] dgemm and dtrmm kernels for IBM z13
   * [2017-02-26] ztrmm kernel for IBM z13
   * [2017-03-13] strmm and ctrmm kernel for IBM z13
-
+  * [2017-09-01] initial Blas Level-1,2 (double precision) for IBM z13
 

README.md

@@ -107,7 +107,7 @@ Please read GotoBLAS_01Readme.txt
 - **ARM Cortex-A57**: Experimental
 
 #### IBM zEnterprise System:
-- **Z13**: Optimized Level-3 BLAS 
+- **Z13**: Optimized Level-3 BLAS and Level-1,2 (double precision)
      
 
 ### Support OS:

interface/axpy.c

@@ -40,8 +40,12 @@
 #include "common.h"
 #ifdef FUNCTION_PROFILE
 #include "functable.h"
+#endif 
+#if  defined(Z13)
+#define MULTI_THREAD_MINIMAL  200000
+#else
+#define MULTI_THREAD_MINIMAL  10000        
 #endif
-
 #ifndef CBLAS
 
 void NAME(blasint *N, FLOAT *ALPHA, FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY){
@@ -88,7 +92,7 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx, FLOAT *y, blasint inc
 
   //Temporarily work-around the low performance issue with small imput size &
   //multithreads.
-  if (n <= 10000)
+  if (n <= MULTI_THREAD_MINIMAL)
 	  nthreads = 1;
 
   if (nthreads == 1) {

kernel/zarch/KERNEL.Z13

@@ -15,14 +15,14 @@ SMINKERNEL   = ../arm/min.c
 DMINKERNEL   = ../arm/min.c
 
 ISAMAXKERNEL = ../arm/iamax.c
-IDAMAXKERNEL = ../arm/iamax.c
+IDAMAXKERNEL = idamax.c
 ICAMAXKERNEL = ../arm/izamax.c
-IZAMAXKERNEL = ../arm/izamax.c
+IZAMAXKERNEL = izamax.c
 
 ISAMINKERNEL = ../arm/iamin.c
-IDAMINKERNEL = ../arm/iamin.c
+IDAMINKERNEL = idamin.c
 ICAMINKERNEL = ../arm/izamin.c
-IZAMINKERNEL = ../arm/izamin.c
+IZAMINKERNEL = izamin.c
 
 ISMAXKERNEL  = ../arm/imax.c
 IDMAXKERNEL  = ../arm/imax.c
@@ -31,24 +31,24 @@ ISMINKERNEL  = ../arm/imin.c
 IDMINKERNEL  = ../arm/imin.c
 
 SASUMKERNEL  = ../arm/asum.c
-DASUMKERNEL  = ../arm/asum.c
+DASUMKERNEL  = dasum.c
 CASUMKERNEL  = ../arm/zasum.c
-ZASUMKERNEL  = ../arm/zasum.c
+ZASUMKERNEL  = zasum.c
 
 SAXPYKERNEL  = ../arm/axpy.c
-DAXPYKERNEL  = ../arm/axpy.c
+DAXPYKERNEL  = daxpy.c
 CAXPYKERNEL  = ../arm/zaxpy.c
-ZAXPYKERNEL  = ../arm/zaxpy.c
+ZAXPYKERNEL  = zaxpy.c
 
 SCOPYKERNEL  = ../arm/copy.c
-DCOPYKERNEL  = ../arm/copy.c
+DCOPYKERNEL  = dcopy.c
 CCOPYKERNEL  = ../arm/zcopy.c
-ZCOPYKERNEL  = ../arm/zcopy.c
+ZCOPYKERNEL  = zcopy.c
 
 SDOTKERNEL   = ../arm/dot.c
-DDOTKERNEL   = ../arm/dot.c
+DDOTKERNEL   = ddot.c
 CDOTKERNEL   = ../arm/zdot.c
-ZDOTKERNEL   = ../arm/zdot.c
+ZDOTKERNEL   = zdot.c
 
 SNRM2KERNEL  = ../arm/nrm2.c
 DNRM2KERNEL  = ../arm/nrm2.c
@@ -56,29 +56,29 @@ CNRM2KERNEL  = ../arm/znrm2.c
 ZNRM2KERNEL  = ../arm/znrm2.c
 
 SROTKERNEL   = ../arm/rot.c
-DROTKERNEL   = ../arm/rot.c
+DROTKERNEL   = drot.c
 CROTKERNEL   = ../arm/zrot.c
-ZROTKERNEL   = ../arm/zrot.c
+ZROTKERNEL   = zrot.c
 
 SSCALKERNEL  = ../arm/scal.c
-DSCALKERNEL  = ../arm/scal.c
+DSCALKERNEL  = dscal.c
 CSCALKERNEL  = ../arm/zscal.c
-ZSCALKERNEL  = ../arm/zscal.c
+ZSCALKERNEL  = zscal.c
 
 SSWAPKERNEL  = ../arm/swap.c
-DSWAPKERNEL  = ../arm/swap.c
+DSWAPKERNEL  = dswap.c
 CSWAPKERNEL  = ../arm/zswap.c
-ZSWAPKERNEL  = ../arm/zswap.c
+ZSWAPKERNEL  = zswap.c
 
 SGEMVNKERNEL = ../arm/gemv_n.c
-DGEMVNKERNEL = ../arm/gemv_n.c
+DGEMVNKERNEL = dgemv_n_4.c
 CGEMVNKERNEL = ../arm/zgemv_n.c
-ZGEMVNKERNEL = ../arm/zgemv_n.c
+ZGEMVNKERNEL = zgemv_n_4.c
 
 SGEMVTKERNEL = ../arm/gemv_t.c
-DGEMVTKERNEL = ../arm/gemv_t.c
+DGEMVTKERNEL = dgemv_t_4.c
 CGEMVTKERNEL = ../arm/zgemv_t.c
-ZGEMVTKERNEL = ../arm/zgemv_t.c
+ZGEMVTKERNEL = zgemv_t_4.c
 
 STRMMKERNEL	= strmm8x4V.S
 DTRMMKERNEL	= trmm8x4V.S

kernel/zarch/dasum.c

@@ -0,0 +1,159 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+#include <math.h>
+
+#if defined(DOUBLE) 
+#define ABS fabs 
+#else 
+#define ABS fabsf 
+#endif
+
+
+static FLOAT __attribute__ ((noinline)) dasum_kernel_32(BLASLONG n, FLOAT *x) {
+
+    __asm__ (
+            "pfd     1, 0(%1)   \n\t"
+            "sllg    %%r0,%0,3  \n\t"
+            "agr     %%r0,%1    \n\t"   
+            "vzero   %%v0       \n\t"
+            "vzero   %%v1       \n\t"
+            "vzero   %%v2       \n\t"
+            "vzero   %%v3       \n\t"   
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd     1, 256(%1 ) \n\t"
+            "vlm     %%v24,%%v31, 0(%1 ) \n\t"  
+    
+            "vflpdb  %%v24, %%v24 \n\t"
+            "vflpdb  %%v25, %%v25 \n\t"
+            "vflpdb  %%v26, %%v26 \n\t"
+            "vflpdb  %%v27, %%v27 \n\t"
+            "vflpdb  %%v28, %%v28 \n\t"
+            "vflpdb  %%v29, %%v29 \n\t"
+            "vflpdb  %%v30, %%v30 \n\t"
+            "vflpdb  %%v31, %%v31 \n\t"
+    
+            "vfadb   %%v0,%%v0,%%v24    \n\t"
+            "vfadb   %%v1,%%v1,%%v25    \n\t"
+            "vfadb   %%v2,%%v2,%%v26    \n\t"
+            "vfadb   %%v3,%%v3,%%v27    \n\t" 
+            "vfadb   %%v0,%%v0,%%v28    \n\t"
+            "vfadb   %%v1,%%v1,%%v29    \n\t"
+            "vfadb   %%v2,%%v2,%%v30    \n\t"
+            "vfadb   %%v3,%%v3,%%v31    \n\t" 
+    
+            "vlm     %%v24,%%v31, 128(%1) \n\t"  
+    
+            "vflpdb  %%v24, %%v24       \n\t"
+            "vflpdb  %%v25, %%v25       \n\t"
+            "vflpdb  %%v26, %%v26       \n\t"
+            "vflpdb  %%v27, %%v27       \n\t"
+            "vflpdb  %%v28, %%v28       \n\t"
+            "vflpdb  %%v29, %%v29       \n\t"
+            "vflpdb  %%v30, %%v30       \n\t"
+            "vflpdb  %%v31, %%v31       \n\t"
+            "la      %1,256(%1)         \n\t"  
+            "vfadb   %%v0,%%v0,%%v24    \n\t"
+            "vfadb   %%v1,%%v1,%%v25    \n\t"
+            "vfadb   %%v2,%%v2,%%v26    \n\t"
+            "vfadb   %%v3,%%v3,%%v27    \n\t" 
+            "vfadb   %%v0,%%v0,%%v28    \n\t"
+            "vfadb   %%v1,%%v1,%%v29    \n\t"
+            "vfadb   %%v2,%%v2,%%v30    \n\t"
+            "vfadb   %%v3,%%v3,%%v31    \n\t"  
+            
+            "clgrjl  %1,%%r0,1b         \n\t"
+            "vfadb   %%v24,%%v0,%%v1    \n\t"
+            "vfadb   %%v25,%%v2,%%v3    \n\t"
+            "vfadb   %%v0,%%v25,%%v24   \n\t"
+            "vrepg   %%v1,%%v0,1        \n\t"
+            "adbr    %%f0,%%f1          \n\t"
+            :
+            : "r"(n), "a"(x)
+            : "cc", "memory","r0","f0","f1","v0","v1","v2","v3","v24","v25","v26","v27","v28","v29","v30","v31"
+            );
+
+}
+
+
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
+    BLASLONG i = 0;
+    BLASLONG j = 0;
+    FLOAT sumf = 0.0;
+    BLASLONG n1;
+
+    if (n <= 0 || inc_x <= 0) return sumf;
+
+    if (inc_x == 1) {
+
+        n1 = n & -32;
+               
+        if (n1 > 0) {
+
+            sumf = dasum_kernel_32(n1, x);
+            i = n1;
+        }
+
+        while (i < n) {
+            sumf += ABS(x[i]);
+            i++;
+        }
+
+    } else {
+        BLASLONG n1 = n & -4;
+        register FLOAT sum1, sum2;
+        sum1 = 0.0;
+        sum2 = 0.0;
+        while (j < n1) {
+
+            sum1 += ABS(x[i]);
+            sum2 += ABS(x[i + inc_x]);
+            sum1 += ABS(x[i + 2 * inc_x]);
+            sum2 += ABS(x[i + 3 * inc_x]);
+
+            i += inc_x * 4;
+            j += 4;
+
+        }
+        sumf = sum1 + sum2;
+        while (j < n) {
+
+            sumf += ABS(x[i]);
+            i += inc_x;
+            j++;
+        }
+
+
+    }
+    return sumf;
+}
+
+

kernel/zarch/daxpy.c

@@ -0,0 +1,386 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+#define Z13_D 1
+#define PREFETCH_INS 1
+#if defined(Z13_A)
+#include <vecintrin.h>
+
+static void daxpy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG  i = 0;
+    __vector double v_a = {*alpha,*alpha}; 
+    __vector double * v_y=(__vector double *)y;
+    __vector double * v_x=(__vector double *)x;
+        
+    for(; i<n/2; i+=16){
+
+        v_y[i]    += v_a * v_x[i];
+        v_y[i+1]  += v_a * v_x[i+1];
+        v_y[i+2]  += v_a * v_x[i+2];
+        v_y[i+3]  += v_a * v_x[i+3];
+        v_y[i+4]  += v_a * v_x[i+4];
+        v_y[i+5]  += v_a * v_x[i+5];
+        v_y[i+6]  += v_a * v_x[i+6];
+        v_y[i+7]  += v_a * v_x[i+7]; 
+        v_y[i+8]  += v_a * v_x[i+8];
+        v_y[i+9]  += v_a * v_x[i+9];
+        v_y[i+10] += v_a * v_x[i+10];
+        v_y[i+11] += v_a * v_x[i+11];
+        v_y[i+12] += v_a * v_x[i+12];
+        v_y[i+13] += v_a * v_x[i+13];
+        v_y[i+14] += v_a * v_x[i+14];
+        v_y[i+15] += v_a * v_x[i+15];
+    }
+
+}
+#elif  defined(Z13_B)
+static void __attribute__ ((noinline))  daxpy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{ 
+ 
+
+             __asm__ volatile( 
+#if defined(PREFETCH_INS)         
+            "pfd 1, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+#endif     
+            "vlrepg %%v0 , 0(%3) \n\t"
+            "srlg %3,%0,5        \n\t" 
+            "xgr %%r1,%%r1       \n\t"
+             "vlr %%v1,%%v0      \n\t"
+            ".align 16 \n\t"    
+            "1: \n\t"
+#if defined(PREFETCH_INS)         
+            "pfd 1, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t" 
+#endif 
+
+            "vl  %%v24, 0(%%r1,%2) \n\t"
+            "vl  %%v16, 0(%%r1,%1) \n\t"
+            "vfmadb    %%v16,%%v0,%%v16,%%v24  \n\t" 
+            "vst  %%v16, 0(%%r1,%2)  \n\t" 
+            "vl  %%v25, 16(%%r1,%2)  \n\t"
+            "vl  %%v17, 16(%%r1,%1)  \n\t" 
+            "vfmadb    %%v17,%%v0,%%v17,%%v25  \n\t" 
+            "vst  %%v17, 16(%%r1,%2) \n\t"
+            "vl  %%v26, 32(%%r1,%2)  \n\t"
+            "vl  %%v18, 32(%%r1,%1)  \n\t" 
+            "vfmadb    %%v18,%%v0,%%v18,%%v26  \n\t" 
+            "vst  %%v18, 32(%%r1,%2) \n\t"
+            "vl  %%v27, 48(%%r1,%2)  \n\t"
+            "vl  %%v19, 48(%%r1,%1)  \n\t" 
+            "vfmadb    %%v19,%%v0,%%v19,%%v27  \n\t" 
+            "vst  %%v19, 48(%%r1,%2) \n\t"   
+
+            "vl  %%v24,( 0+64)(%%r1,%2) \n\t"
+            "vl  %%v16,( 0+64)(%%r1,%1) \n\t"
+            "vfmadb    %%v16,%%v0,%%v16,%%v24  \n\t" 
+            "vst  %%v16,( 0+64)(%%r1,%2) \n\t" 
+            "vl  %%v25, (16+64)(%%r1,%2) \n\t"
+            "vl  %%v17, (16+64)(%%r1,%1) \n\t" 
+            "vfmadb    %%v17,%%v0,%%v17,%%v25  \n\t" 
+            "vst  %%v17, (16+64)(%%r1,%2) \n\t"
+            "vl  %%v26, (32+64)(%%r1,%2) \n\t"
+            "vl  %%v18, (32+64)(%%r1,%1) \n\t" 
+            "vfmadb    %%v18,%%v0,%%v18,%%v26  \n\t" 
+            "vst  %%v18, (32+64)(%%r1,%2) \n\t"
+            "vl  %%v27, (48+64)(%%r1,%2) \n\t"
+            "vl  %%v19, (48+64)(%%r1,%1) \n\t" 
+            "vfmadb    %%v19,%%v0,%%v19,%%v27  \n\t" 
+            "vst  %%v19, (48+64)(%%r1,%2) \n\t"       
+            
+            "vl  %%v24,( 0+128)(%%r1,%2) \n\t"
+            "vl  %%v16,( 0+128)(%%r1,%1) \n\t"
+            "vfmadb    %%v16,%%v0,%%v16,%%v24  \n\t" 
+            "vst  %%v16,( 0+128)(%%r1,%2) \n\t" 
+            "vl  %%v25, (16+128)(%%r1,%2) \n\t"
+            "vl  %%v17, (16+128)(%%r1,%1) \n\t" 
+            "vfmadb    %%v17,%%v0,%%v17,%%v25  \n\t" 
+            "vst  %%v17, (16+128)(%%r1,%2) \n\t"
+            "vl  %%v26, (32+128)(%%r1,%2)  \n\t"
+            "vl  %%v18, (32+128)(%%r1,%1)  \n\t" 
+            "vfmadb    %%v18,%%v0,%%v18,%%v26  \n\t" 
+            "vst  %%v18, (32+128)(%%r1,%2) \n\t"
+            "vl  %%v27, (48+128)(%%r1,%2)  \n\t"
+            "vl  %%v19, (48+128)(%%r1,%1)  \n\t" 
+            "vfmadb    %%v19,%%v0,%%v19,%%v27  \n\t" 
+            "vst  %%v19, (48+128)(%%r1,%2) \n\t"         
+            
+            "vl  %%v24,( 0+192)(%%r1,%2) \n\t"
+            "vl  %%v16,( 0+192)(%%r1,%1) \n\t"
+            "vfmadb    %%v16,%%v0,%%v16,%%v24  \n\t" 
+            "vst  %%v16,( 0+192)(%%r1,%2) \n\t" 
+            "vl  %%v25, (16+192)(%%r1,%2) \n\t"
+            "vl  %%v17, (16+192)(%%r1,%1) \n\t" 
+            "vfmadb    %%v17,%%v0,%%v17,%%v25  \n\t" 
+            "vst  %%v17, (16+192)(%%r1,%2) \n\t"
+            "vl  %%v26, (32+192)(%%r1,%2)  \n\t"
+            "vl  %%v18, (32+192)(%%r1,%1)  \n\t" 
+            "vfmadb    %%v18,%%v0,%%v18,%%v26  \n\t" 
+            "vst  %%v18, (32+192)(%%r1,%2) \n\t"
+            "vl  %%v27, (48+192)(%%r1,%2) \n\t"
+            "vl  %%v19, (48+192)(%%r1,%1) \n\t" 
+            "vfmadb    %%v19,%%v0,%%v19,%%v27  \n\t" 
+            "vst  %%v19, (48+192)(%%r1,%2) \n\t"                                                              
+ 
+              
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %3,1b"
+            : 
+            :"r"(n),"a"(x),"a"(y),"a"(alpha)
+            :"cc", "memory", "r1" ,"v0" ,"v16","v17","v18","v19", "v24","v25","v26","v27" 
+            );
+}
+ 
+#elif  defined(Z13_C)
+static void __attribute__ ((noinline))  daxpy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{ 
+ 
+             __asm__ volatile( 
+#if defined(PREFETCH_INS)         
+            "pfd 1, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+#endif     
+            "vlrepg %%v0 , 0(%3)   \n\t"
+            "srlg %3,%0,5          \n\t" 
+            "xgr %%r1,%%r1         \n\t"
+             "vlr %%v1,%%v0        \n\t"
+            ".align 16 \n\t"    
+            "1: \n\t"
+#if defined(PREFETCH_INS)         
+            "pfd 1, 256(%%r1,%1)    \n\t"
+            "pfd 2, 256(%%r1,%2)    \n\t" 
+#endif 
+            "vl  %%v16, 0(%%r1,%1)  \n\t"
+            "vl  %%v17, 16(%%r1,%1) \n\t"
+            "vl  %%v18, 32(%%r1,%1) \n\t"
+            "vl  %%v19, 48(%%r1,%1) \n\t"
+
+            "vl  %%v24, 0(%%r1,%2)  \n\t"
+            "vl  %%v25, 16(%%r1,%2) \n\t"
+            "vl  %%v26, 32(%%r1,%2) \n\t"
+            "vl  %%v27, 48(%%r1,%2) \n\t" 
+            "vfmadb    %%v16,%%v0,%%v16,%%v24  \n\t" 
+            "vfmadb    %%v17,%%v1,%%v17,%%v25  \n\t" 
+            "vfmadb    %%v18,%%v0,%%v18,%%v26  \n\t"
+            "vfmadb    %%v19,%%v1,%%v19,%%v27  \n\t" 
+            "vst  %%v16, 0(%%r1,%2)  \n\t"  
+            "vst  %%v17, 16(%%r1,%2) \n\t" 
+            "vst  %%v18, 32(%%r1,%2) \n\t" 
+            "vst  %%v19, 48(%%r1,%2) \n\t" 
+             
+            "vl  %%v24, 64(%%r1,%1)  \n\t"
+            "vl  %%v25, 80(%%r1,%1)  \n\t"
+            "vl  %%v26, 96(%%r1,%1)  \n\t"
+            "vl  %%v27, 112(%%r1,%1) \n\t"
+             
+            "vl  %%v16, 64(%%r1,%2)  \n\t"
+            "vl  %%v17, 80(%%r1,%2)  \n\t"
+            "vl  %%v18, 96(%%r1,%2)  \n\t"
+            "vl  %%v19, 112(%%r1,%2) \n\t"             
+
+
+            "vfmadb   %%v24,%%v0,%%v24,%%v16   \n\t" 
+            "vfmadb    %%v25,%%v1,%%v25,%%v17  \n\t" 
+            "vfmadb    %%v26,%%v0,%%v26,%%v18  \n\t"
+            "vfmadb    %%v27,%%v1,%%v27,%%v19  \n\t" 
+
+            "vst  %%v24, 64(%%r1,%2)  \n\t"
+            "vst  %%v25, 80(%%r1,%2)  \n\t"
+            "vst  %%v26, 96(%%r1,%2)  \n\t"
+            "vst  %%v27, 112(%%r1,%2) \n\t" 
+
+            "vl  %%v16, (0+128)(%%r1,%1)  \n\t"
+            "vl  %%v17, (16+128)(%%r1,%1) \n\t"
+            "vl  %%v18, (32+128)(%%r1,%1) \n\t"
+            "vl  %%v19, (48+128)(%%r1,%1) \n\t"
+             
+            "vl  %%v24, (0+128)(%%r1,%2)  \n\t"
+            "vl  %%v25, (16+128)(%%r1,%2) \n\t"
+            "vl  %%v26, (32+128)(%%r1,%2) \n\t"
+            "vl  %%v27, (48+128)(%%r1,%2) \n\t" 
+ 
+            "vfmadb    %%v16,%%v0,%%v16,%%v24  \n\t" 
+            "vfmadb    %%v17,%%v1,%%v17,%%v25  \n\t" 
+            "vfmadb    %%v18,%%v0,%%v18,%%v26  \n\t"
+            "vfmadb    %%v19,%%v1,%%v19,%%v27  \n\t" 
+            "vst  %%v16, (0+128)(%%r1,%2)  \n\t"  
+            "vst  %%v17, (16+128)(%%r1,%2) \n\t" 
+            "vst  %%v18, (32+128)(%%r1,%2) \n\t" 
+            "vst  %%v19, (48+128)(%%r1,%2) \n\t" 
+             
+            "vl  %%v24, (64+128)(%%r1,%1)  \n\t"
+            "vl  %%v25, (80+128)(%%r1,%1)  \n\t"
+            "vl  %%v26, (96+128)(%%r1,%1)  \n\t"
+            "vl  %%v27, (112+128)(%%r1,%1) \n\t"
+             
+            "vl  %%v16, (64+128)(%%r1,%2)  \n\t"
+            "vl  %%v17, (80+128)(%%r1,%2)  \n\t"
+            "vl  %%v18, (96+128)(%%r1,%2)  \n\t"
+            "vl  %%v19, (112+128)(%%r1,%2) \n\t"             
+ 
+            "vfmadb   %%v24,%%v0,%%v24,%%v16   \n\t" 
+            "vfmadb    %%v25,%%v1,%%v25,%%v17  \n\t" 
+            "vfmadb    %%v26,%%v0,%%v26,%%v18  \n\t"
+            "vfmadb    %%v27,%%v1,%%v27,%%v19  \n\t" 
+
+            "vst  %%v24, (64+128)(%%r1,%2)  \n\t"
+            "vst  %%v25, (80+128)(%%r1,%2)  \n\t"
+            "vst  %%v26, (96+128)(%%r1,%2)  \n\t"
+            "vst  %%v27, (112+128)(%%r1,%2) \n\t"            
+                    
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %3,1b"
+            : 
+            :"r"(n),"a"(x),"a"(y),"a"(alpha)
+            :"cc", "memory", "r1" ,"v0","v1","v16","v17","v18","v19", "v24","v25","v26","v27" 
+            );
+} 
+ 
+ 
+#elif defined(Z13_D)
+static void __attribute__ ((noinline))  daxpy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{ 
+        
+         __asm__ volatile( 
+#if defined(PREFETCH_INS)         
+                   "pfd 1, 0(%1) \n\t"
+                   "pfd 2, 0(%2) \n\t"
+#endif         
+                    "vlrepg %%v0 , 0(%3) \n\t"
+                    "srlg %3,%0,5       \n\t" 
+                    "vlr %%v1,%%v0  \n\t"
+                    ".align 16 \n\t"    
+                    "1: \n\t"
+#if defined(PREFETCH_INS)         
+                    "pfd 1, 256(%1) \n\t"
+                    "pfd 2, 256(%2) \n\t"
+#endif                  
+                    "vlm      %%v16,%%v23,  0(%1)     \n\t"
+                    "vlm      %%v24, %%v31, 0(%2)     \n\t"
+                    "vfmadb   %%v16,%%v0,%%v16,%%v24  \n\t"
+                    "vfmadb   %%v17,%%v1,%%v17,%%v25  \n\t"  
+                    "vfmadb   %%v18,%%v0,%%v18,%%v26  \n\t"
+                    "vfmadb   %%v19,%%v1,%%v19,%%v27  \n\t"
+                    "vfmadb   %%v20,%%v0,%%v20,%%v28  \n\t"
+                    "vfmadb   %%v21,%%v1,%%v21,%%v29  \n\t"  
+                    "vfmadb   %%v22,%%v0,%%v22,%%v30  \n\t"
+                    "vfmadb   %%v23,%%v1,%%v23,%%v31  \n\t"
+                    "vstm     %%v16,%%v23, 0(%2)      \n\t" 
+                    "vlm      %%v24,%%v31, 128(%1)    \n\t"
+                    "vlm      %%v16,%%v23,  128(%2)   \n\t"
+                    "vfmadb   %%v24,%%v0,%%v24,%%v16  \n\t"
+                    "vfmadb   %%v25,%%v1,%%v25,%%v17  \n\t"  
+                    "vfmadb   %%v26,%%v0,%%v26,%%v18  \n\t"
+                    "vfmadb   %%v27,%%v1,%%v27,%%v19  \n\t"
+                    "vfmadb   %%v28,%%v0,%%v28,%%v20  \n\t"
+                    "vfmadb   %%v29,%%v1,%%v29,%%v21  \n\t"  
+                    "vfmadb   %%v30,%%v0,%%v30,%%v22  \n\t"
+                    "vfmadb   %%v31,%%v1,%%v31,%%v23  \n\t"  
+                    "la %1,256(%1)                    \n\t"
+                    "vstm %%v24, %%v31, 128(%2)       \n\t"
+                    "la %2,256(%2)                    \n\t" 
+                    "brctg %3,1b"
+                    : 
+                    :"r"(n),"a"(x),"a"(y),"a"(alpha)
+                    :"cc", "memory",  "v0","v1","v16","v17","v18","v19","v20","v21",
+                    "v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+                 );
+ 
+
+}
+ 
+ 
+
+#endif
+ 
+
+int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
+{
+    BLASLONG i=0;
+    BLASLONG ix=0,iy=0;
+
+    if ( n <= 0 )  return 0 ;
+
+    if ( (inc_x == 1) && (inc_y == 1) )
+    {
+
+        BLASLONG n1 = n & -32;
+
+        if ( n1 )
+            daxpy_kernel_32(n1, x, y , &da );
+
+        i = n1;
+        while(i < n)
+        {
+
+            y[i] += da * x[i] ;
+            i++ ;
+
+        }
+        return 0 ;
+
+
+    }
+
+    BLASLONG n1 = n & -4;
+
+    while(i < n1)
+    {
+
+        FLOAT m1      = da * x[ix] ;
+        FLOAT m2      = da * x[ix+inc_x] ;
+        FLOAT m3      = da * x[ix+2*inc_x] ;
+        FLOAT m4      = da * x[ix+3*inc_x] ;
+
+        y[iy]         += m1 ;
+        y[iy+inc_y]   += m2 ;
+        y[iy+2*inc_y] += m3 ;
+        y[iy+3*inc_y] += m4 ;
+
+        ix  += inc_x*4 ;
+        iy  += inc_y*4 ;
+        i+=4 ;
+
+    }
+
+    while(i < n)
+    {
+
+        y[iy] += da * x[ix] ;
+        ix  += inc_x ;
+        iy  += inc_y ;
+        i++ ;
+
+    }
+    return 0 ;
+
+}
+
+

kernel/zarch/dcopy.c

@@ -0,0 +1,169 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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(Z13mvc)
+
+static void __attribute__ ((noinline)) dcopy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y) {
+
+    __asm__ volatile(
+            "pfd 1, 0(%1)    \n\t"
+            "pfd 2, 0(%2)    \n\t"
+            "srlg %%r0,%0,5  \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "mvc 0(256,%2),0(%1) \n\t"
+            "la %1,256(%1)       \n\t"
+            "la %2,256(%2)       \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y)
+            : "cc", "memory","r0"
+            );
+    return;
+
+}
+#else
+
+static void __attribute__ ((noinline)) dcopy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y) {
+
+    __asm__ volatile(
+            "pfd 1, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+            "srlg %%r0,%0,5       \n\t"
+            "xgr %%r1,%%r1       \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+
+            "vl   %%v24, 0(%%r1,%1)   \n\t"
+            "vst  %%v24, 0(%%r1,%2)   \n\t"
+            "vl   %%v25, 16(%%r1,%1)  \n\t"
+            "vst  %%v25, 16(%%r1,%2)  \n\t"
+            "vl   %%v26, 32(%%r1,%1)  \n\t"
+            "vst  %%v26, 32(%%r1,%2)  \n\t"
+            "vl   %%v27, 48(%%r1,%1)  \n\t"
+            "vst  %%v27, 48(%%r1,%2)  \n\t"
+
+            "vl   %%v24, 64(%%r1,%1)  \n\t"
+            "vst  %%v24, 64(%%r1,%2)  \n\t"
+            "vl   %%v25, 80(%%r1,%1)  \n\t"
+            "vst  %%v25, 80(%%r1,%2)  \n\t"
+            "vl   %%v26, 96(%%r1,%1)  \n\t"
+            "vst  %%v26, 96(%%r1,%2)  \n\t"
+            "vl   %%v27, 112(%%r1,%1) \n\t"
+            "vst  %%v27, 112(%%r1,%2) \n\t"
+
+
+            "vl   %%v24, 128(%%r1,%1) \n\t"
+            "vst  %%v24, 128(%%r1,%2) \n\t"
+
+            "vl   %%v25, 144(%%r1,%1) \n\t"
+            "vst  %%v25, 144(%%r1,%2) \n\t"
+
+            "vl   %%v26, 160(%%r1,%1) \n\t"
+            "vst  %%v26, 160(%%r1,%2) \n\t"
+
+            "vl   %%v27, 176(%%r1,%1) \n\t"
+            "vst  %%v27, 176(%%r1,%2) \n\t"
+
+            "vl   %%v24, 192(%%r1,%1) \n\t"
+            "vst  %%v24, 192(%%r1,%2) \n\t"
+            "vl   %%v25, 208(%%r1,%1) \n\t"
+            "vst  %%v25, 208(%%r1,%2) \n\t"
+            "vl   %%v26, 224(%%r1,%1) \n\t"
+            "vst  %%v26, 224(%%r1,%2) \n\t"
+            "vl   %%v27, 240(%%r1,%1) \n\t"
+            "vst  %%v27, 240(%%r1,%2) \n\t"
+            "la   %%r1,256(%%r1)      \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y)
+            : "cc", "memory","r0","r1", "v24","v25","v26","v27"
+            );
+    return;
+
+}
+#endif
+
+int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y) {
+    BLASLONG i = 0;
+    BLASLONG ix = 0, iy = 0;
+
+    if (n <= 0) return 0;
+
+    if ((inc_x == 1) && (inc_y == 1)) {
+
+        BLASLONG n1 = n & -32;
+        if (n1 > 0) {
+            dcopy_kernel_32(n1, x, y);
+            i = n1;
+        }
+
+        while (i < n) {
+            y[i] = x[i];
+            i++;
+
+        }
+
+
+    } else {
+
+        BLASLONG n1 = n & -4;
+
+        while (i < n1) {
+
+            y[iy] = x[ix];
+            y[iy + inc_y] = x[ix + inc_x];
+            y[iy + 2 * inc_y] = x[ix + 2 * inc_x];
+            y[iy + 3 * inc_y] = x[ix + 3 * inc_x];
+
+            ix += inc_x * 4;
+            iy += inc_y * 4;
+            i += 4;
+
+        }
+
+        while (i < n) {
+
+            y[iy] = x[ix];
+            ix += inc_x;
+            iy += inc_y;
+            i++;
+
+        }
+
+    }
+    return 0;
+
+
+}
+
+

kernel/zarch/ddot.c

@@ -0,0 +1,194 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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(Z13)
+static void __attribute__ ((noinline))   ddot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
+{
+   
+                 __asm__ volatile( 
+                    "pfd 1, 0(%1) \n\t"
+                    "pfd 1, 0(%2) \n\t"      
+                    "vzero %%v24  \n\t"
+                    "vzero %%v25  \n\t" 
+                    "vzero %%v26  \n\t"
+                    "vzero %%v27  \n\t"                  
+                    "srlg  %%r0,%0,4    \n\t" 
+                    "xgr   %%r1,%%r1    \n\t"
+                    ".align 16 \n\t"    
+                    "1: \n\t"
+                    "pfd 1, 256(%%r1,%1) \n\t"
+                    "pfd 1, 256(%%r1,%2) \n\t"                
+                    "vl  %%v16, 0(%%r1,%1)  \n\t"
+                    "vl  %%v17, 16(%%r1,%1) \n\t"
+                    "vl  %%v18, 32(%%r1,%1) \n\t"
+                    "vl  %%v19, 48(%%r1,%1) \n\t"
+
+                    "vl  %%v28, 0(%%r1,%2) \n\t"
+                    "vfmadb    %%v24,%%v16,%%v28,%%v24  \n\t"  
+                    "vl  %%v29, 16(%%r1,%2) \n\t"
+                    "vfmadb    %%v25,%%v17,%%v29,%%v25  \n\t"   
+             
+                    "vl  %%v30, 32(%%r1,%2) \n\t"
+                    "vfmadb    %%v26,%%v18,%%v30,%%v26  \n\t"      
+                    "vl  %%v31, 48(%%r1,%2) \n\t" 
+                    "vfmadb    %%v27,%%v19,%%v31,%%v27  \n\t"   
+         
+                    "vl  %%v16, 64(%%r1,%1)  \n\t"
+                    "vl  %%v17, 80(%%r1,%1)  \n\t"
+                    "vl  %%v18, 96(%%r1,%1)  \n\t"
+                    "vl  %%v19, 112(%%r1,%1) \n\t"
+
+                    "vl  %%v28, 64(%%r1,%2) \n\t"
+                    "vfmadb    %%v24,%%v16,%%v28,%%v24  \n\t"  
+                    "vl  %%v29, 80(%%r1,%2) \n\t"
+                    "vfmadb    %%v25,%%v17,%%v29,%%v25  \n\t"  
+                  
+             
+                    "vl  %%v30, 96(%%r1,%2)  \n\t"
+                    "vfmadb    %%v26,%%v18,%%v30,%%v26  \n\t" 
+                    "vl  %%v31, 112(%%r1,%2) \n\t" 
+                    "vfmadb    %%v27,%%v19,%%v31,%%v27  \n\t"  
+                     
+                    
+                    "la %%r1,128(%%r1) \n\t"
+                    "brctg %%r0,1b \n\t"
+                    "vfadb   %%v24,%%v25,%%v24    \n\t"
+                    "vfadb   %%v24,%%v26,%%v24    \n\t"
+                    "vfadb   %%v24,%%v27,%%v24    \n\t"                 
+                    "vrepg   %%v1,%%v24,1         \n\t"
+                    "vfadb   %%v1,%%v24,%%v1      \n\t"  
+                    " std    %%f1,0(%3)           \n\t"  
+                    :
+                    :"r"(n),"a"(x),"a"(y),"a"(d)
+                    :"cc" , "memory" ,"r0","r1","v16", "v17","v18","v19","v20","v21","v22","v23",
+                    "v24","v25","v26","v27","v28","v29","v30","v31"
+
+                 );
+
+}
+
+
+#else
+
+static void ddot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
+{
+    BLASLONG register i = 0;
+    FLOAT dot = 0.0;
+
+    while(i < n)
+        {
+              dot += y[i]  * x[i]
+                  + y[i+1] * x[i+1]
+                  + y[i+2] * x[i+2]
+                  + y[i+3] * x[i+3]
+                  + y[i+4] * x[i+4]
+                  + y[i+5] * x[i+5]
+                  + y[i+6] * x[i+6]
+                  + y[i+7] * x[i+7] ;
+
+              i+=8 ;
+
+       }
+       *d += dot;
+
+}
+
+#endif
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
+{
+    BLASLONG i=0;
+    BLASLONG ix=0,iy=0;
+
+    FLOAT  dot = 0.0 ;
+
+    if ( n <= 0 )  return(dot);
+
+    if ( (inc_x == 1) && (inc_y == 1) )
+    {
+
+        BLASLONG n1 = n & -16;
+
+        if ( n1 )
+            ddot_kernel_8(n1, x, y , &dot );
+
+        i = n1;
+        while(i < n)
+        {
+
+            dot += y[i] * x[i] ;
+            i++ ;
+
+        } 
+        return(dot);
+
+
+    }
+
+    FLOAT temp1 = 0.0;
+    FLOAT temp2 = 0.0;
+
+    BLASLONG n1 = n & -4;    
+
+    while(i < n1)
+    {
+
+        FLOAT m1 = y[iy]       * x[ix] ;
+        FLOAT m2 = y[iy+inc_y] * x[ix+inc_x] ;
+
+        FLOAT m3 = y[iy+2*inc_y] * x[ix+2*inc_x] ;
+        FLOAT m4 = y[iy+3*inc_y] * x[ix+3*inc_x] ;
+
+        ix  += inc_x*4 ;
+        iy  += inc_y*4 ;
+
+        temp1 += m1+m3;
+        temp2 += m2+m4;
+
+        i+=4 ;
+
+    }
+
+    while(i < n)
+    {
+
+        temp1 += y[iy] * x[ix] ;
+        ix  += inc_x ;
+        iy  += inc_y ;
+        i++ ;
+
+    }
+    dot = temp1 + temp2;
+    return(dot);
+
+}
+
+

kernel/zarch/dgemv_n_4.c

@@ -0,0 +1,487 @@
+/***************************************************************************
+Copyright (c) 2017, 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"
+
+#define NBMAX 2048
+
+#define HAVE_KERNEL_4x4_VEC 1
+#define HAVE_KERNEL_4x2_VEC 1
+#define HAVE_KERNEL_4x1_VEC 1
+
+#if defined(HAVE_KERNEL_4x4_VEC) || defined(HAVE_KERNEL_4x2_VEC) || defined(HAVE_KERNEL_4x1_VEC)
+ #include <vecintrin.h>
+#endif
+
+#ifdef HAVE_KERNEL_4x4
+
+#elif HAVE_KERNEL_4x4_VEC
+
+static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *xo, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT x0,x1,x2,x3;
+    x0 = xo[0] * *alpha;
+    x1 = xo[1] * *alpha;
+    x2 = xo[2] * *alpha;
+    x3 = xo[3] * *alpha;
+    __vector double   v_x0 = {x0,x0};
+    __vector double   v_x1 = {x1,x1};
+    __vector double   v_x2 = {x2,x2};
+    __vector double   v_x3 = {x3,x3};
+    __vector double* v_y =(__vector double*)y;      
+    __vector double* va0 = (__vector double*)ap[0];
+    __vector double* va1 = (__vector double*)ap[1];
+    __vector double* va2 = (__vector double*)ap[2];
+    __vector double* va3 = (__vector double*)ap[3]; 
+
+    for ( i=0; i< n/2; i+=2 )
+    {
+        v_y[i]   += v_x0 * va0[i]   +  v_x1 * va1[i]   + v_x2 * va2[i]   + v_x3 * va3[i] ;
+        v_y[i+1] += v_x0 * va0[i+1] +  v_x1 * va1[i+1] + v_x2 * va2[i+1] + v_x3 * va3[i+1] ;        
+    }
+}
+
+#else
+
+static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *xo, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0,*a1,*a2,*a3;
+    FLOAT x[4]  __attribute__ ((aligned (16)));
+    a0 = ap[0];
+    a1 = ap[1];
+    a2 = ap[2];
+    a3 = ap[3];
+
+    for ( i=0; i<4; i++)
+        x[i] = xo[i] * *alpha;
+
+    for ( i=0; i< n; i+=4 )
+    {
+        y[i] += a0[i]*x[0] + a1[i]*x[1] + a2[i]*x[2] + a3[i]*x[3];        
+        y[i+1] += a0[i+1]*x[0] + a1[i+1]*x[1] + a2[i+1]*x[2] + a3[i+1]*x[3];        
+        y[i+2] += a0[i+2]*x[0] + a1[i+2]*x[1] + a2[i+2]*x[2] + a3[i+2]*x[3];        
+        y[i+3] += a0[i+3]*x[0] + a1[i+3]*x[1] + a2[i+3]*x[2] + a3[i+3]*x[3];        
+    }
+}
+
+
+#endif
+
+#ifdef HAVE_KERNEL_4x2
+
+#elif HAVE_KERNEL_4x2_VEC
+
+static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *xo, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT x0,x1;
+    x0 = xo[0] * *alpha;
+    x1 = xo[1] * *alpha; 
+    __vector double   v_x0 = {x0,x0};
+    __vector double   v_x1 = {x1,x1}; 
+    __vector double* v_y =(__vector double*)y;      
+    __vector double* va0 = (__vector double*)ap[0];
+    __vector double* va1 = (__vector double*)ap[1]; 
+
+    for ( i=0; i< n/2; i+=2 )
+    {
+        v_y[i]   += v_x0 * va0[i] +  v_x1 * va1[i]   ;
+        v_y[i+1] += v_x0 * va0[i+1] +  v_x1 * va1[i+1]  ;        
+    } 
+}
+#else
+
+static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *xo, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0,*a1;
+    FLOAT x[4]  __attribute__ ((aligned (16)));
+    a0 = ap[0];
+    a1 = ap[1];
+
+    for ( i=0; i<2; i++)
+        x[i] = xo[i] * *alpha;
+
+    for ( i=0; i< n; i+=4 )
+    {
+        y[i] += a0[i]*x[0] + a1[i]*x[1];        
+        y[i+1] += a0[i+1]*x[0] + a1[i+1]*x[1];        
+        y[i+2] += a0[i+2]*x[0] + a1[i+2]*x[1];        
+        y[i+3] += a0[i+3]*x[0] + a1[i+3]*x[1];        
+    }
+}
+
+
+#endif
+
+#ifdef HAVE_KERNEL_4x1
+
+#elif HAVE_KERNEL_4x1_VEC
+static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *xo, FLOAT *y, FLOAT *alpha)
+{
+    
+    BLASLONG i;
+    FLOAT x0;
+    x0 = xo[0] * *alpha;
+    __vector double   v_x0 = {x0,x0};
+    __vector double* v_y =(__vector double*)y;      
+    __vector double* va0 = (__vector double*)ap;
+
+    for ( i=0; i< n/2; i+=2 )
+    {
+        v_y[i] += v_x0 * va0[i]    ;
+        v_y[i+1] += v_x0 * va0[i+1]  ;        
+    }
+        
+ 
+}
+
+#else
+static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *xo, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0;
+    FLOAT x[4]  __attribute__ ((aligned (16)));
+    a0 = ap;
+
+    for ( i=0; i<1; i++)
+        x[i] = xo[i] * *alpha;
+
+    for ( i=0; i< n; i+=4 )
+    {
+        y[i] += a0[i]*x[0];        
+        y[i+1] += a0[i+1]*x[0];        
+        y[i+2] += a0[i+2]*x[0];        
+        y[i+3] += a0[i+3]*x[0];        
+    }
+}
+
+
+#endif
+
+
+ 
+static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest) __attribute__ ((noinline));
+
+static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest)
+{
+    BLASLONG i;
+        
+    for ( i=0; i<n; i++ ){
+            *dest += *src;
+            src++;
+            dest += inc_dest;
+    }
+    return;
+     
+}
+
+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 i;
+    BLASLONG j;
+    FLOAT *a_ptr;
+    FLOAT *x_ptr;
+    FLOAT *y_ptr;
+    FLOAT *ap[4];
+    BLASLONG n1;
+    BLASLONG m1;
+    BLASLONG m2;
+    BLASLONG m3;
+    BLASLONG n2;
+    BLASLONG lda4 =  lda << 2;
+    FLOAT xbuffer[8],*ybuffer;
+
+    if ( m < 1 ) return(0);
+    if ( n < 1 ) return(0);
+
+    ybuffer = buffer;
+    
+    n1 = n >> 2 ;
+    n2 = n &  3 ;
+
+    m3 = m & 3  ;
+    m1 = m & -4 ;
+    m2 = (m & (NBMAX-1)) - m3 ;
+
+    y_ptr = y;
+
+    BLASLONG NB = NBMAX;
+
+    while ( NB == NBMAX )
+    {
+        
+        m1 -= NB;
+        if ( m1 < 0)
+        {
+            if ( m2 == 0 ) break;    
+            NB = m2;
+        }
+        
+        a_ptr = a;
+        x_ptr = x;
+        
+        ap[0] = a_ptr;
+        ap[1] = a_ptr + lda;
+        ap[2] = ap[1] + lda;
+        ap[3] = ap[2] + lda;
+
+        if ( inc_y != 1 )
+            memset(ybuffer,0,NB*8);
+        else
+            ybuffer = y_ptr;
+
+        if ( inc_x == 1 )
+        {
+
+
+            for( i = 0; i < n1 ; i++)
+            {
+                dgemv_kernel_4x4(NB,ap,x_ptr,ybuffer,&alpha);
+                ap[0] += lda4; 
+                ap[1] += lda4; 
+                ap[2] += lda4; 
+                ap[3] += lda4; 
+                a_ptr += lda4;
+                x_ptr += 4;    
+            }
+
+            if ( n2 & 2 )
+            {
+                dgemv_kernel_4x2(NB,ap,x_ptr,ybuffer,&alpha);
+                a_ptr += lda*2;
+                x_ptr += 2;    
+            }
+
+
+            if ( n2 & 1 )
+            {
+                dgemv_kernel_4x1(NB,a_ptr,x_ptr,ybuffer,&alpha);
+                a_ptr += lda;
+                x_ptr += 1;    
+
+            }
+
+
+        }
+        else
+        {
+
+            for( i = 0; i < n1 ; i++)
+            {
+                xbuffer[0] = x_ptr[0];
+                x_ptr += inc_x;    
+                xbuffer[1] =  x_ptr[0];
+                x_ptr += inc_x;    
+                xbuffer[2] =  x_ptr[0];
+                x_ptr += inc_x;    
+                xbuffer[3] = x_ptr[0];
+                x_ptr += inc_x;    
+                dgemv_kernel_4x4(NB,ap,xbuffer,ybuffer,&alpha);
+                ap[0] += lda4; 
+                ap[1] += lda4; 
+                ap[2] += lda4; 
+                ap[3] += lda4; 
+                a_ptr += lda4;
+            }
+
+            for( i = 0; i < n2 ; i++)
+            {
+                xbuffer[0] = x_ptr[0];
+                x_ptr += inc_x;    
+                dgemv_kernel_4x1(NB,a_ptr,xbuffer,ybuffer,&alpha);
+                a_ptr += lda;
+
+            }
+
+        }
+
+        a     += NB;
+        if ( inc_y != 1 )
+        {
+            add_y(NB,ybuffer,y_ptr,inc_y);
+            y_ptr += NB * inc_y;
+        }
+        else
+            y_ptr += NB ;
+
+    }
+
+    if ( m3 == 0 ) return(0);
+
+    if ( m3 == 3 )
+    {
+        a_ptr = a;
+        x_ptr = x;
+        FLOAT temp0 = 0.0;
+        FLOAT temp1 = 0.0;
+        FLOAT temp2 = 0.0;
+        if ( lda == 3 && inc_x ==1 )
+        {
+
+            for( i = 0; i < ( n & -4 ); i+=4 )
+            {
+
+                temp0 += a_ptr[0] * x_ptr[0] + a_ptr[3] * x_ptr[1];
+                temp1 += a_ptr[1] * x_ptr[0] + a_ptr[4] * x_ptr[1];
+                temp2 += a_ptr[2] * x_ptr[0] + a_ptr[5] * x_ptr[1];
+
+                temp0 += a_ptr[6] * x_ptr[2] + a_ptr[9]  * x_ptr[3];
+                temp1 += a_ptr[7] * x_ptr[2] + a_ptr[10] * x_ptr[3];
+                temp2 += a_ptr[8] * x_ptr[2] + a_ptr[11] * x_ptr[3];
+
+                a_ptr += 12;
+                x_ptr += 4;
+            }
+
+            for( ; i < n; i++ )
+            {
+                temp0 += a_ptr[0] * x_ptr[0];
+                temp1 += a_ptr[1] * x_ptr[0];
+                temp2 += a_ptr[2] * x_ptr[0];
+                a_ptr += 3;
+                x_ptr ++;
+            }
+
+        }
+        else
+        {
+
+            for( i = 0; i < n; i++ )
+            {
+                temp0 += a_ptr[0] * x_ptr[0];
+                temp1 += a_ptr[1] * x_ptr[0];
+                temp2 += a_ptr[2] * x_ptr[0];
+                a_ptr += lda;
+                x_ptr += inc_x;
+
+
+            }
+
+        }
+        y_ptr[0] += alpha * temp0;
+        y_ptr += inc_y;
+        y_ptr[0] += alpha * temp1;
+        y_ptr += inc_y;
+        y_ptr[0] += alpha * temp2;
+        return(0);
+    }
+
+
+    if ( m3 == 2 )
+    {
+        a_ptr = a;
+        x_ptr = x;
+        FLOAT temp0 = 0.0;
+        FLOAT temp1 = 0.0;
+        if ( lda == 2 && inc_x ==1 )
+        {
+
+            for( i = 0; i < (n & -4) ; i+=4 )
+            {
+                temp0 += a_ptr[0] * x_ptr[0] + a_ptr[2] * x_ptr[1];
+                temp1 += a_ptr[1] * x_ptr[0] + a_ptr[3] * x_ptr[1];
+                temp0 += a_ptr[4] * x_ptr[2] + a_ptr[6] * x_ptr[3];
+                temp1 += a_ptr[5] * x_ptr[2] + a_ptr[7] * x_ptr[3];
+                a_ptr += 8;
+                x_ptr += 4;
+
+            }
+
+
+            for( ; i < n; i++ )
+            {
+                temp0 += a_ptr[0]   * x_ptr[0];
+                temp1 += a_ptr[1]   * x_ptr[0];
+                a_ptr += 2;
+                x_ptr ++;
+            }
+
+        }
+        else
+        {
+
+            for( i = 0; i < n; i++ )
+            {
+                temp0 += a_ptr[0] * x_ptr[0];
+                temp1 += a_ptr[1] * x_ptr[0];
+                a_ptr += lda;
+                x_ptr += inc_x;
+
+
+            }
+
+        }
+        y_ptr[0] += alpha * temp0;
+        y_ptr += inc_y;
+        y_ptr[0] += alpha * temp1;
+        return(0);
+    }
+
+    if ( m3 == 1 )
+    {
+        a_ptr = a;
+        x_ptr = x;
+        FLOAT temp = 0.0;
+        if ( lda == 1 && inc_x ==1 )
+        {
+
+            for( i = 0; i < (n & -4); i+=4 )
+            {
+                temp += a_ptr[i] * x_ptr[i] + a_ptr[i+1] * x_ptr[i+1] + a_ptr[i+2] * x_ptr[i+2] + a_ptr[i+3] * x_ptr[i+3];
+    
+            }
+
+            for( ; i < n; i++ )
+            {
+                temp += a_ptr[i] * x_ptr[i];
+            }
+
+        }
+        else
+        {
+
+            for( i = 0; i < n; i++ )
+            {
+                temp += a_ptr[0] * x_ptr[0];
+                a_ptr += lda;
+                x_ptr += inc_x;
+            }
+
+        }
+        y_ptr[0] += alpha * temp;
+        return(0);
+    }
+
+
+    return(0);
+}
+
+

kernel/zarch/dgemv_t_4.c

@@ -0,0 +1,541 @@
+/***************************************************************************
+Copyright (c) 2017, 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"
+
+#define HAVE_KERNEL_4x4_VEC 1
+#define HAVE_KERNEL_4x2_VEC 1
+#define HAVE_KERNEL_4x1_VEC 1
+
+#if defined(HAVE_KERNEL_4x4_VEC) || defined(HAVE_KERNEL_4x2_VEC) || defined(HAVE_KERNEL_4x1_VEC)
+ #include <vecintrin.h>
+#endif
+#define NBMAX 2048
+
+#ifdef HAVE_KERNEL_4x4
+
+#elif HAVE_KERNEL_4x4_VEC
+
+static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
+{
+    BLASLONG i; 
+    __vector double* va0 = (__vector double*)ap[0];
+    __vector double* va1 = (__vector double*)ap[1];
+    __vector double* va2 = (__vector double*)ap[2];
+    __vector double* va3 = (__vector double*)ap[3];     
+    __vector double* v_x =(__vector double*)x;           
+    __vector double temp0 = {0,0};      
+    __vector double temp1 = {0,0};       
+    __vector double temp2 = {0,0};       
+    __vector double temp3 = {0,0};       
+
+    for ( i=0; i< n/2; i+=2 )
+    {
+        temp0 += v_x[i] * va0[i]  + v_x[i+1] * va0[i+1] ;        
+        temp1 += v_x[i] * va1[i]  + v_x[i+1] * va1[i+1] ;    
+        temp2 += v_x[i] * va2[i]  + v_x[i+1] * va2[i+1] ;    
+        temp3 += v_x[i] * va3[i]  + v_x[i+1] * va3[i+1] ;        
+    }
+        
+    y[0] = temp0[0] + temp0[1];
+    y[1] = temp1[0] + temp1[1];
+    y[2] = temp2[0] + temp2[1];
+    y[3] = temp3[0] + temp3[1];; 
+}
+#else
+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
+ 
+#ifdef HAVE_KERNEL_4x2
+
+#elif HAVE_KERNEL_4x2_VEC
+
+static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
+{
+    BLASLONG i; 
+    __vector double* va0 = (__vector double*)ap[0];
+    __vector double* va1 = (__vector double*)ap[1];     
+    __vector double* v_x =(__vector double*)x;           
+    __vector double temp0 = {0,0};      
+    __vector double temp1 = {0,0};         
+
+    for ( i=0; i< n/2; i+=2 )
+    {
+        temp0 += v_x[i] * va0[i]  + v_x[i+1] * va0[i+1] ;        
+        temp1 += v_x[i] * va1[i]  + v_x[i+1] * va1[i+1] ;        
+    }
+        
+    y[0] = temp0[0] + temp0[1];
+    y[1] = temp1[0] + temp1[1]; 
+}
+#else
+static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
+{
+
+    BLASLONG i;
+    FLOAT *a0,*a1;
+    a0 = ap[0];
+    a1 = ap[1]; 
+    FLOAT temp0 = 0.0;
+    FLOAT temp1 = 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];         
+    }
+    y[0] = temp0;
+    y[1] = temp1; 
+
+}
+#endif    
+
+#ifdef HAVE_KERNEL_4x1
+
+#elif HAVE_KERNEL_4x1_VEC
+
+static void dgemv_kernel_4x1(BLASLONG n, FLOAT *a0, FLOAT *x, FLOAT *y)
+{
+    BLASLONG i; 
+    __vector double* va0 = (__vector double*)a0;      
+    __vector double* v_x =(__vector double*)x;           
+    __vector double temp0 = {0,0};          
+
+    for ( i=0; i< n/2; i+=2 )
+    {
+        temp0 += v_x[i] * va0[i]  + v_x[i+1] * va0[i+1] ;         
+    }
+        
+    y[0] = temp0[0] + temp0[1]; 
+}
+#else
+static void dgemv_kernel_4x1(BLASLONG n, FLOAT *a0, FLOAT *x, FLOAT *y)
+{
+    BLASLONG i;
+    
+     
+    FLOAT temp0 = 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];         
+    }
+    y[0] = temp0; 
+}
+#endif
+    
+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)
+{
+
+        BLASLONG i;
+ 
+        for ( i=0; i<n; i++ )
+        {
+                *dest += src[i]  * da;
+                dest  += inc_dest;
+        }
+        return; 
+         
+}
+
+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 + (m < NBMAX ? m : 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,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);
+}
+
+

kernel/zarch/drot.c

@@ -0,0 +1,270 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+
+  
+
+
+static void __attribute__ ((noinline))  drot_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *c, FLOAT *s)
+{
+          __asm__  (
+            "pfd 2, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+             
+            "vlrepg %%v0,0(%3) \n\t"
+            "vlrepg %%v1,0(%4) \n\t"
+            "srlg   %%r0,%0,5  \n\t"
+            "xgr    %%r1,%%r1  \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 2, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+            "vl  %%v24, 0(%%r1,%1)  \n\t" 
+            "vl  %%v25, 16(%%r1,%1) \n\t" 
+            "vl  %%v26, 32(%%r1,%1) \n\t" 
+            "vl  %%v27, 48(%%r1,%1) \n\t" 
+            "vl  %%v16, 0(%%r1,%2)  \n\t" 
+            "vl  %%v17, 16(%%r1,%2) \n\t" 
+            "vl  %%v18, 32(%%r1,%2) \n\t" 
+            "vl  %%v19, 48(%%r1,%2) \n\t"  
+           
+            "vfmdb %%v28,%%v24,%%v0 \n\t"
+            "vfmdb %%v29,%%v25,%%v0 \n\t"
+            "vfmdb %%v20,%%v24,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v21,%%v25,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v30,%%v26,%%v0 \n\t"
+            "vfmdb %%v22,%%v26,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v31,%%v27,%%v0 \n\t"
+            "vfmdb %%v23,%%v27,%%v1 \n\t" /* yn=x*s  */
+            /* 2nd parts*/
+            "vfmadb %%v28,%%v16,%%v1,%%v28 \n\t"  
+            "vfmsdb %%v20,%%v16,%%v0,%%v20 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v29,%%v17,%%v1,%%v29 \n\t"  
+            "vfmsdb %%v21,%%v17,%%v0,%%v21 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v30,%%v18,%%v1,%%v30 \n\t" 
+            "vfmsdb %%v22,%%v18,%%v0,%%v22 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v31,%%v19,%%v1,%%v31 \n\t" 
+            "vfmsdb %%v23,%%v19,%%v0,%%v23 \n\t"  /* yn=y*c-yn */
+           
+           
+           
+            "vst  %%v28, 0(%%r1,%1)  \n\t" 
+            "vst  %%v29, 16(%%r1,%1) \n\t" 
+            "vst  %%v30, 32(%%r1,%1) \n\t" 
+            "vst  %%v31, 48(%%r1,%1) \n\t" 
+            "vst  %%v20, 0(%%r1,%2)  \n\t" 
+            "vst  %%v21, 16(%%r1,%2) \n\t" 
+            "vst  %%v22, 32(%%r1,%2) \n\t" 
+            "vst  %%v23, 48(%%r1,%2) \n\t"  
+           
+            "vl  %%v24, 64(%%r1,%1)  \n\t" 
+            "vl  %%v25, 80(%%r1,%1)  \n\t" 
+            "vl  %%v26, 96(%%r1,%1)  \n\t" 
+            "vl  %%v27, 112(%%r1,%1) \n\t" 
+            "vl  %%v16, 64(%%r1,%2)  \n\t" 
+            "vl  %%v17, 80(%%r1,%2)  \n\t" 
+            "vl  %%v18, 96(%%r1,%2)  \n\t" 
+            "vl  %%v19, 112(%%r1,%2) \n\t"  
+           
+            "vfmdb %%v28,%%v24,%%v0 \n\t"
+            "vfmdb %%v29,%%v25,%%v0 \n\t"
+            "vfmdb %%v20,%%v24,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v21,%%v25,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v30,%%v26,%%v0 \n\t"
+            "vfmdb %%v22,%%v26,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v31,%%v27,%%v0 \n\t"
+            "vfmdb %%v23,%%v27,%%v1 \n\t" /* yn=x*s  */
+            /* 2nd parts*/
+            "vfmadb %%v28,%%v16,%%v1,%%v28 \n\t"  
+            "vfmsdb %%v20,%%v16,%%v0,%%v20 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v29,%%v17,%%v1,%%v29 \n\t"  
+            "vfmsdb %%v21,%%v17,%%v0,%%v21 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v30,%%v18,%%v1,%%v30 \n\t" 
+            "vfmsdb %%v22,%%v18,%%v0,%%v22 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v31,%%v19,%%v1,%%v31 \n\t" 
+            "vfmsdb %%v23,%%v19,%%v0,%%v23 \n\t"  /* yn=y*c-yn */
+           
+           
+           
+            "vst  %%v28, 64(%%r1,%1)  \n\t" 
+            "vst  %%v29, 80(%%r1,%1)  \n\t" 
+            "vst  %%v30, 96(%%r1,%1)  \n\t" 
+            "vst  %%v31, 112(%%r1,%1) \n\t" 
+            "vst  %%v20, 64(%%r1,%2)  \n\t" 
+            "vst  %%v21, 80(%%r1,%2)  \n\t" 
+            "vst  %%v22, 96(%%r1,%2)  \n\t" 
+            "vst  %%v23, 112(%%r1,%2) \n\t"
+           
+            "vl  %%v24, 128(%%r1,%1) \n\t" 
+            "vl  %%v25, 144(%%r1,%1) \n\t" 
+            "vl  %%v26, 160(%%r1,%1) \n\t" 
+            "vl  %%v27, 176(%%r1,%1) \n\t" 
+            "vl  %%v16, 128(%%r1,%2) \n\t" 
+            "vl  %%v17, 144(%%r1,%2) \n\t" 
+            "vl  %%v18, 160(%%r1,%2) \n\t" 
+            "vl  %%v19, 176(%%r1,%2) \n\t"  
+           
+            "vfmdb %%v28,%%v24,%%v0 \n\t"
+            "vfmdb %%v29,%%v25,%%v0 \n\t"
+            "vfmdb %%v20,%%v24,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v21,%%v25,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v30,%%v26,%%v0 \n\t"
+            "vfmdb %%v22,%%v26,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v31,%%v27,%%v0 \n\t"
+            "vfmdb %%v23,%%v27,%%v1 \n\t" /* yn=x*s  */
+            /* 2nd parts*/
+            "vfmadb %%v28,%%v16,%%v1,%%v28 \n\t"  
+            "vfmsdb %%v20,%%v16,%%v0,%%v20 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v29,%%v17,%%v1,%%v29 \n\t"  
+            "vfmsdb %%v21,%%v17,%%v0,%%v21 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v30,%%v18,%%v1,%%v30 \n\t" 
+            "vfmsdb %%v22,%%v18,%%v0,%%v22 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v31,%%v19,%%v1,%%v31 \n\t" 
+            "vfmsdb %%v23,%%v19,%%v0,%%v23 \n\t"  /* yn=y*c-yn */
+           
+           
+           
+            "vst  %%v28, 128(%%r1,%1) \n\t" 
+            "vst  %%v29, 144(%%r1,%1) \n\t" 
+            "vst  %%v30, 160(%%r1,%1) \n\t" 
+            "vst  %%v31, 176(%%r1,%1) \n\t" 
+            "vst  %%v20, 128(%%r1,%2) \n\t" 
+            "vst  %%v21, 144(%%r1,%2) \n\t" 
+            "vst  %%v22, 160(%%r1,%2) \n\t" 
+            "vst  %%v23, 176(%%r1,%2) \n\t"  
+           
+            "vl  %%v24, 192(%%r1,%1) \n\t" 
+            "vl  %%v25, 208(%%r1,%1) \n\t" 
+            "vl  %%v26, 224(%%r1,%1) \n\t" 
+            "vl  %%v27, 240(%%r1,%1) \n\t" 
+            "vl  %%v16, 192(%%r1,%2) \n\t" 
+            "vl  %%v17, 208(%%r1,%2) \n\t" 
+            "vl  %%v18, 224(%%r1,%2) \n\t" 
+            "vl  %%v19, 240(%%r1,%2) \n\t"  
+           
+            "vfmdb %%v28,%%v24,%%v0 \n\t"
+            "vfmdb %%v29,%%v25,%%v0 \n\t"
+            "vfmdb %%v20,%%v24,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v21,%%v25,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v30,%%v26,%%v0 \n\t"
+            "vfmdb %%v22,%%v26,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v31,%%v27,%%v0 \n\t"
+            "vfmdb %%v23,%%v27,%%v1 \n\t" /* yn=x*s  */
+            /* 2nd parts*/
+            "vfmadb %%v28,%%v16,%%v1,%%v28 \n\t"  
+            "vfmsdb %%v20,%%v16,%%v0,%%v20 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v29,%%v17,%%v1,%%v29 \n\t"  
+            "vfmsdb %%v21,%%v17,%%v0,%%v21 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v30,%%v18,%%v1,%%v30 \n\t" 
+            "vfmsdb %%v22,%%v18,%%v0,%%v22 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v31,%%v19,%%v1,%%v31 \n\t" 
+            "vfmsdb %%v23,%%v19,%%v0,%%v23 \n\t"  /* yn=y*c-yn */
+           
+           
+           
+            "vst  %%v28, 192(%%r1,%1) \n\t" 
+            "vst  %%v29, 208(%%r1,%1) \n\t" 
+            "vst  %%v30, 224(%%r1,%1) \n\t" 
+            "vst  %%v31, 240(%%r1,%1) \n\t" 
+            "vst  %%v20, 192(%%r1,%2) \n\t" 
+            "vst  %%v21, 208(%%r1,%2) \n\t" 
+            "vst  %%v22, 224(%%r1,%2) \n\t" 
+            "vst  %%v23, 240(%%r1,%2) \n\t"
+           
+           
+          
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y),"a"(c),"a"(s)
+            : "cc", "memory","r0","r1" ,"v0","v1","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+            );
+      return;
+
+}
+
+
+ 
+
+int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s)
+{
+    BLASLONG i=0;
+    BLASLONG ix=0,iy=0;
+     
+    FLOAT temp;
+
+    if ( n <= 0 )  return(0);
+
+    if ( (inc_x == 1) && (inc_y == 1) )
+    {
+
+        BLASLONG n1 = n & -32;
+        if ( n1 > 0 )
+        {
+            FLOAT cosa,sina;
+            cosa=c;
+            sina=s;
+            drot_kernel_32(n1, x, y, &cosa, &sina);
+            i=n1;
+        }
+
+        while(i < n)
+        {
+            temp  = c*x[i] + s*y[i] ;
+            y[i]  = c*y[i] - s*x[i] ;
+            x[i]  = temp ;
+
+            i++ ;
+
+        }
+
+
+    }
+    else
+    {
+
+        while(i < n)
+        {
+            temp   = c*x[ix] + s*y[iy] ;
+            y[iy]  = c*y[iy] - s*x[ix] ;
+            x[ix]  = temp ;
+
+            ix += inc_x ;
+            iy += inc_y ;
+            i++ ;
+
+        }
+
+    }
+    return(0);
+
+}
+
+

kernel/zarch/dscal.c

@@ -0,0 +1,210 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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(Z13)
+static void __attribute__ ((noinline))  dscal_kernel_8( BLASLONG n, FLOAT  da , FLOAT *x )
+{
+   
+             __asm__ ("pfd 2, 0(%1) \n\t"      
+                    "vrepg %%v0 , %%v0,0   \n\t"
+                    "sllg  %%r0,%0,3       \n\t" 
+                    "xgr   %%r1,%%r1       \n\t"
+                    ".align 16 \n\t"    
+                    "1: \n\t" 
+                    "pfd 2, 256(%%r1,%1)   \n\t"     
+                    "vl  %%v24, 0(%%r1,%1) \n\t"
+                    "vfmdb    %%v24,%%v24,%%v0 \n\t"
+                    "vst  %%v24, 0(%%r1,%1)  \n\t" 
+                    "vl   %%v25, 16(%%r1,%1) \n\t"
+                    "vfmdb    %%v25,%%v25,%%v0  \n\t"  
+                    "vst  %%v25, 16(%%r1,%1) \n\t" 
+                    "vl   %%v26, 32(%%r1,%1) \n\t"
+                    "vfmdb    %%v26,%%v26,%%v0 \n\t"
+                    "vst  %%v26, 32(%%r1,%1) \n\t"  
+                    "vl   %%v27, 48(%%r1,%1) \n\t" 
+                    "vfmdb    %%v27,%%v27,%%v0  \n\t"  
+                    "vst  %%v27, 48(%%r1,%1) \n\t"  
+                    "vl   %%v24, 64(%%r1,%1) \n\t"
+                    "vfmdb    %%v24,%%v24,%%v0  \n\t"
+                    "vst  %%v24, 64(%%r1,%1) \n\t" 
+                    "vl   %%v25, 80(%%r1,%1) \n\t"
+                    "vfmdb    %%v25,%%v25,%%v0  \n\t"  
+                    "vst  %%v25, 80(%%r1,%1) \n\t" 
+                    "vl   %%v26, 96(%%r1,%1) \n\t"
+                    "vfmdb    %%v26,%%v26,%%v0  \n\t"
+                    "vst  %%v26, 96(%%r1,%1)  \n\t"  
+                    "vl   %%v27, 112(%%r1,%1) \n\t" 
+                    "vfmdb    %%v27,%%v27,%%v0  \n\t"  
+                    "vst  %%v27, 112(%%r1,%1) \n\t"   
+                    "la %%r1,128(%%r1) \n\t"
+                    "clgrjl %%r1,%%r0,1b \n\t"
+                    : 
+                    :"r"(n),"a"(x),"f"(da)
+                    :"cc" , "memory" ,"r0","r1","v0","v24","v25","v26","v27"
+                 );
+
+}
+
+static void __attribute__ ((noinline))  dscal_kernel_8_zero( BLASLONG n, FLOAT da , FLOAT *x )
+{
+   
+             __asm__ ("pfd 2, 0(%1) \n\t"      
+                    "vzero %%v0     \n\t"
+                    "sllg %%r0,%0,3   \n\t" 
+                    "xgr %%r1,%%r1    \n\t"
+                    ".align 16 \n\t"    
+                    "1: \n\t" 
+                    "pfd 2, 256(%%r1,%1) \n\t"     
+                    "vst  %%v0, 0(%%r1,%1)   \n\t" 
+                    "vst  %%v0, 16(%%r1,%1)  \n\t" 
+                    "vst  %%v0, 32(%%r1,%1)  \n\t"   
+                    "vst  %%v0, 48(%%r1,%1)  \n\t"  
+                    "vst  %%v0, 64(%%r1,%1)  \n\t" 
+                    "vst  %%v0, 80(%%r1,%1)  \n\t" 
+                    "vst  %%v0, 96(%%r1,%1)  \n\t"  
+                    "vst  %%v0, 112(%%r1,%1) \n\t"   
+                    "la %%r1,128(%%r1) \n\t"
+                    "clgrjl %%r1,%%r0,1b \n\t"
+                    : 
+                    :"r"(n),"a"(x),"f"(da)
+                    :"cc" , "memory" ,"r0","r1","v0"
+                 );
+}
+ 
+
+#endif
+
+int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
+{
+    BLASLONG i=0,j=0;
+    if ( n <= 0 || inc_x <=0 )
+        return(0);
+
+ 
+    if ( inc_x == 1 )
+    {
+
+        if ( da == 0.0 )
+        {        
+
+            BLASLONG n1 = n & -16;
+            if ( n1 > 0 )
+            {
+                
+                dscal_kernel_8_zero(n1 , da , x);
+                j=n1;
+            }
+
+            while(j < n)
+            {
+
+                x[j]=0.0;
+                j++;
+            }
+
+        }
+        else
+        {
+
+            BLASLONG n1 = n & -16;
+            if ( n1 > 0 )
+            { 
+                dscal_kernel_8(n1 , da , x);
+                j=n1;
+            }
+            while(j < n)
+            {
+
+                x[j] = da * x[j] ;
+                j++;
+            }
+        }
+
+
+    }
+    else
+    {
+
+        if ( da == 0.0 )
+        {        
+
+                        BLASLONG n1 = n & -4;
+
+                        while (j < n1) {
+
+                            x[i]=0.0;
+                            x[i + inc_x]=0.0;
+                            x[i + 2 * inc_x]=0.0;
+                            x[i + 3 * inc_x]=0.0;
+
+                            i += inc_x * 4; 
+                            j += 4;
+
+                        } 
+            while(j < n)
+            {
+
+                x[i]=0.0;
+                i += inc_x ;
+                j++;
+            }
+
+        }
+        else
+        {
+                        BLASLONG n1 = n & -4;
+
+                        while (j < n1) {
+
+                            x[i] = da * x[i] ;
+                            x[i + inc_x] = da * x[i + inc_x];
+                            x[i + 2 * inc_x] = da *  x[i + 2 * inc_x];
+                            x[i + 3 * inc_x] = da * x[i + 3 * inc_x];
+
+                            i += inc_x * 4; 
+                            j += 4;
+
+                        }  
+
+            while(j < n)
+            {
+
+                x[i] = da * x[i] ;
+                i += inc_x ;
+                j++;
+            }
+        }
+
+    }
+    return 0;
+
+}
+
+

kernel/zarch/dswap.c

@@ -0,0 +1,382 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+
+#define Z13_SWAP_C 1
+
+#if defined(Z13_SWAP_A)
+static void  __attribute__ ((noinline))  dswap_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y)
+{
+         __asm__ volatile(
+            "pfd 1, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+            "srlg %%r0,%0,5      \n\t"
+            "xgr  %%r1,%%r1      \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 2, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+            
+            "vl  %%v24, 0(%%r1,%1) \n\t" 
+            "vl  %%v16, 0(%%r1,%2) \n\t" 
+            "vst %%v24, 0(%%r1,%2) \n\t" 
+            "vst %%v16, 0(%%r1,%1) \n\t" 
+
+            "vl  %%v25, 16(%%r1,%1) \n\t" 
+            "vl  %%v17, 16(%%r1,%2) \n\t" 
+            "vst %%v25, 16(%%r1,%2) \n\t" 
+            "vst %%v17, 16(%%r1,%1) \n\t" 
+
+            "vl  %%v26, 32(%%r1,%1) \n\t" 
+            "vl  %%v18, 32(%%r1,%2) \n\t" 
+            "vst %%v26, 32(%%r1,%2) \n\t" 
+            "vst %%v18, 32(%%r1,%1) \n\t"           
+
+            "vl  %%v27, 48(%%r1,%1) \n\t" 
+            "vl  %%v19, 48(%%r1,%2) \n\t" 
+            "vst %%v27, 48(%%r1,%2) \n\t" 
+            "vst %%v19, 48(%%r1,%1) \n\t"    
+
+            "vl  %%v28, 64(%%r1,%1) \n\t" 
+            "vl  %%v20, 64(%%r1,%2) \n\t" 
+            "vst %%v28, 64(%%r1,%2) \n\t" 
+            "vst %%v20, 64(%%r1,%1) \n\t"   
+
+            "vl  %%v29, 80(%%r1,%1) \n\t" 
+            "vl  %%v21, 80(%%r1,%2) \n\t" 
+            "vst %%v29, 80(%%r1,%2) \n\t" 
+            "vst %%v21, 80(%%r1,%1) \n\t" 
+
+            "vl  %%v30, 96(%%r1,%1) \n\t" 
+            "vl  %%v22, 96(%%r1,%2) \n\t" 
+            "vst %%v30, 96(%%r1,%2) \n\t" 
+            "vst %%v22, 96(%%r1,%1) \n\t" 
+
+            "vl  %%v31, 112(%%r1,%1) \n\t" 
+            "vl  %%v23, 112(%%r1,%2) \n\t" 
+            "vst %%v31, 112(%%r1,%2) \n\t" 
+            "vst %%v23, 112(%%r1,%1) \n\t" 
+
+            "vl  %%v24, 128(%%r1,%1) \n\t" 
+            "vl  %%v16, 128(%%r1,%2) \n\t" 
+            "vst %%v24, 128(%%r1,%2) \n\t" 
+            "vst %%v16, 128(%%r1,%1) \n\t" 
+
+            "vl  %%v25, 144(%%r1,%1) \n\t" 
+            "vl  %%v17, 144(%%r1,%2) \n\t" 
+            "vst %%v25, 144(%%r1,%2) \n\t" 
+            "vst %%v17, 144(%%r1,%1) \n\t" 
+
+            "vl  %%v26, 160(%%r1,%1) \n\t" 
+            "vl  %%v18, 160(%%r1,%2) \n\t" 
+            "vst %%v26, 160(%%r1,%2) \n\t" 
+            "vst %%v18, 160(%%r1,%1) \n\t"           
+
+            "vl  %%v27, 176(%%r1,%1) \n\t" 
+            "vl  %%v19, 176(%%r1,%2) \n\t" 
+            "vst %%v27, 176(%%r1,%2) \n\t" 
+            "vst %%v19, 176(%%r1,%1) \n\t"    
+
+            "vl  %%v28, 192(%%r1,%1) \n\t" 
+            "vl  %%v20, 192(%%r1,%2) \n\t" 
+            "vst %%v28, 192(%%r1,%2) \n\t" 
+            "vst %%v20, 192(%%r1,%1) \n\t"   
+
+            "vl  %%v29, 208(%%r1,%1) \n\t" 
+            "vl  %%v21, 208(%%r1,%2) \n\t" 
+            "vst %%v29, 208(%%r1,%2) \n\t" 
+            "vst %%v21, 208(%%r1,%1) \n\t" 
+
+            "vl  %%v30, 224(%%r1,%1) \n\t" 
+            "vl  %%v22, 224(%%r1,%2) \n\t" 
+            "vst %%v30, 224(%%r1,%2) \n\t" 
+            "vst %%v22, 224(%%r1,%1) \n\t" 
+
+            "vl  %%v31, 240(%%r1,%1) \n\t" 
+            "vl  %%v23, 240(%%r1,%2) \n\t" 
+            "vst %%v31, 240(%%r1,%2) \n\t" 
+            "vst %%v23, 240(%%r1,%1) \n\t"    
+          
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y)
+            : "cc", "memory" ,"r0","r1", "v16","v17","v18","v19","v20","v21","v22","v23"
+              ,"v24","v25","v26","v27","v28","v29","v30","v31"
+            );
+    return;
+
+}
+
+#elif defined(Z13_SWAP_B) 
+static void  __attribute__ ((noinline))  dswap_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y)
+{
+         __asm__ volatile(
+            "pfd 2, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+            "srlg %%r0,%0,5       \n\t"
+            "xgr %%r1,%%r1       \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 2, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+            "vl  %%v24, 0(%%r1,%1)   \n\t" 
+            "vl  %%v25, 16(%%r1,%1)  \n\t" 
+            "vl  %%v26, 32(%%r1,%1)  \n\t" 
+            "vl  %%v27, 48(%%r1,%1)  \n\t"  
+            "vl  %%v28, 64(%%r1,%1)  \n\t" 
+            "vl  %%v29, 80(%%r1,%1)  \n\t" 
+            "vl  %%v30, 96(%%r1,%1)  \n\t" 
+            "vl  %%v31, 112(%%r1,%1) \n\t"
+            "vl  %%v16, 0(%%r1,%2)   \n\t" 
+            "vl  %%v17, 16(%%r1,%2)  \n\t" 
+            "vl  %%v18, 32(%%r1,%2)  \n\t" 
+            "vl  %%v19, 48(%%r1,%2)  \n\t"  
+            "vl  %%v20, 64(%%r1,%2)  \n\t" 
+            "vl  %%v21, 80(%%r1,%2)  \n\t" 
+            "vl  %%v22, 96(%%r1,%2)  \n\t" 
+            "vl  %%v23, 112(%%r1,%2) \n\t"
+
+            "vst  %%v24, 0(%%r1,%2)  \n\t" 
+            "vst  %%v25, 16(%%r1,%2) \n\t" 
+            "vst  %%v26, 32(%%r1,%2) \n\t" 
+            "vst  %%v27, 48(%%r1,%2) \n\t"  
+            "vst  %%v28, 64(%%r1,%2) \n\t" 
+            "vst  %%v29, 80(%%r1,%2) \n\t" 
+            "vst  %%v30, 96(%%r1,%2) \n\t" 
+            "vst  %%v31, 112(%%r1,%2)\n\t"
+            "vst  %%v16, 0(%%r1,%1)  \n\t" 
+            "vst  %%v17, 16(%%r1,%1) \n\t" 
+            "vst  %%v18, 32(%%r1,%1) \n\t" 
+            "vst  %%v19, 48(%%r1,%1) \n\t"  
+            "vst  %%v20, 64(%%r1,%1) \n\t" 
+            "vst  %%v21, 80(%%r1,%1) \n\t" 
+            "vst  %%v22, 96(%%r1,%1) \n\t" 
+            "vst  %%v23, 112(%%r1,%1)\n\t"
+            
+
+
+            "vl  %%v24, 128(%%r1,%1) \n\t" 
+            "vl  %%v25, 144(%%r1,%1) \n\t" 
+            "vl  %%v26, 160(%%r1,%1) \n\t" 
+            "vl  %%v27, 176(%%r1,%1) \n\t" 
+            "vl  %%v28, 192(%%r1,%1) \n\t" 
+            "vl  %%v29, 208(%%r1,%1) \n\t" 
+            "vl  %%v30, 224(%%r1,%1) \n\t" 
+            "vl  %%v31, 240(%%r1,%1) \n\t"
+
+            "vl  %%v16, 128(%%r1,%2) \n\t" 
+            "vl  %%v17, 144(%%r1,%2) \n\t" 
+            "vl  %%v18, 160(%%r1,%2) \n\t" 
+            "vl  %%v19, 176(%%r1,%2) \n\t"  
+            "vl  %%v20, 192(%%r1,%2) \n\t" 
+            "vl  %%v21, 208(%%r1,%2) \n\t" 
+            "vl  %%v22, 224(%%r1,%2) \n\t" 
+            "vl  %%v23, 240(%%r1,%2) \n\t"
+
+
+            "vst  %%v24, 128(%%r1,%2) \n\t" 
+            "vst  %%v25, 144(%%r1,%2) \n\t" 
+            "vst  %%v26, 160(%%r1,%2) \n\t" 
+            "vst  %%v27, 176(%%r1,%2) \n\t" 
+            "vst  %%v28, 192(%%r1,%2) \n\t" 
+            "vst  %%v29, 208(%%r1,%2) \n\t" 
+            "vst  %%v30, 224(%%r1,%2) \n\t" 
+            "vst  %%v31, 240(%%r1,%2) \n\t"
+
+
+            "vst  %%v16, 128(%%r1,%1) \n\t" 
+            "vst  %%v17, 144(%%r1,%1) \n\t" 
+            "vst  %%v18, 160(%%r1,%1) \n\t" 
+            "vst  %%v19, 176(%%r1,%1) \n\t"  
+            "vst  %%v20, 192(%%r1,%1) \n\t" 
+            "vst  %%v21, 208(%%r1,%1) \n\t" 
+            "vst  %%v22, 224(%%r1,%1) \n\t" 
+            "vst  %%v23, 240(%%r1,%1) \n\t"
+ 
+          
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y)
+            : "cc", "memory","r0","r1", "v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+
+            );
+    return;
+
+}
+
+#elif defined(Z13_SWAP_C)  
+static void  __attribute__ ((noinline))  dswap_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y)
+{
+         __asm__ volatile(
+            "pfd 2, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+            "srlg %%r0,%0,5       \n\t"
+            "xgr %%r1,%%r1       \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 2, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+            
+            "vl  %%v16, 0(%%r1,%1)   \n\t" 
+            "vl  %%v17, 16(%%r1,%1)  \n\t" 
+            "vl  %%v18, 32(%%r1,%1)  \n\t" 
+            "vl  %%v19, 48(%%r1,%1)  \n\t"  
+            "vl  %%v20, 64(%%r1,%1)  \n\t" 
+            "vl  %%v21, 80(%%r1,%1)  \n\t" 
+            "vl  %%v22, 96(%%r1,%1)  \n\t" 
+            "vl  %%v23, 112(%%r1,%1) \n\t"
+            "vl  %%v24, 128(%%r1,%1) \n\t" 
+            "vl  %%v25, 144(%%r1,%1) \n\t" 
+            "vl  %%v26, 160(%%r1,%1) \n\t" 
+            "vl  %%v27, 176(%%r1,%1) \n\t" 
+            "vl  %%v28, 192(%%r1,%1) \n\t" 
+            "vl  %%v29, 208(%%r1,%1) \n\t" 
+            "vl  %%v30, 224(%%r1,%1) \n\t" 
+            "vl  %%v31, 240(%%r1,%1) \n\t"
+
+
+            "vl  %%v0, 0(%%r1,%2)    \n\t" 
+            "vl  %%v1, 16(%%r1,%2)   \n\t" 
+            "vl  %%v2, 32(%%r1,%2)   \n\t" 
+            "vl  %%v3, 48(%%r1,%2)   \n\t"  
+            "vl  %%v4, 64(%%r1,%2)   \n\t" 
+            "vl  %%v5, 80(%%r1,%2)   \n\t" 
+            "vl  %%v6, 96(%%r1,%2)   \n\t" 
+            "vl  %%v7, 112(%%r1,%2)  \n\t"
+            "vst  %%v0, 0(%%r1,%1)   \n\t" 
+            "vst  %%v1, 16(%%r1,%1)  \n\t" 
+            "vst  %%v2, 32(%%r1,%1)  \n\t" 
+            "vst  %%v3, 48(%%r1,%1)  \n\t"  
+            "vst  %%v4, 64(%%r1,%1)  \n\t" 
+            "vst  %%v5, 80(%%r1,%1)  \n\t" 
+            "vst  %%v6, 96(%%r1,%1)  \n\t" 
+            "vst  %%v7, 112(%%r1,%1) \n\t" 
+
+            "vl  %%v0, 128(%%r1,%2)  \n\t" 
+            "vl  %%v1, 144(%%r1,%2)  \n\t" 
+            "vl  %%v2, 160(%%r1,%2)  \n\t" 
+            "vl  %%v3, 176(%%r1,%2)  \n\t"  
+            "vl  %%v4, 192(%%r1,%2)  \n\t" 
+            "vl  %%v5, 208(%%r1,%2)  \n\t" 
+            "vl  %%v6, 224(%%r1,%2)  \n\t" 
+            "vl  %%v7, 240(%%r1,%2)  \n\t" 
+            "vst  %%v0, 128(%%r1,%1) \n\t" 
+            "vst  %%v1, 144(%%r1,%1) \n\t" 
+            "vst  %%v2, 160(%%r1,%1) \n\t" 
+            "vst  %%v3, 176(%%r1,%1) \n\t"  
+            "vst  %%v4, 192(%%r1,%1) \n\t" 
+            "vst  %%v5, 208(%%r1,%1) \n\t" 
+            "vst  %%v6, 224(%%r1,%1) \n\t" 
+            "vst  %%v7, 240(%%r1,%1) \n\t"
+
+            "vst  %%v16, 0(%%r1,%2)   \n\t" 
+            "vst  %%v17, 16(%%r1,%2)  \n\t" 
+            "vst  %%v18, 32(%%r1,%2)  \n\t" 
+            "vst  %%v19, 48(%%r1,%2)  \n\t"  
+            "vst  %%v20, 64(%%r1,%2)  \n\t" 
+            "vst  %%v21, 80(%%r1,%2)  \n\t" 
+            "vst  %%v22, 96(%%r1,%2)  \n\t" 
+            "vst  %%v23, 112(%%r1,%2) \n\t"
+            "vst  %%v24, 128(%%r1,%2) \n\t" 
+            "vst  %%v25, 144(%%r1,%2) \n\t" 
+            "vst  %%v26, 160(%%r1,%2) \n\t" 
+            "vst  %%v27, 176(%%r1,%2) \n\t" 
+            "vst  %%v28, 192(%%r1,%2) \n\t" 
+            "vst  %%v29, 208(%%r1,%2) \n\t" 
+            "vst  %%v30, 224(%%r1,%2) \n\t" 
+            "vst  %%v31, 240(%%r1,%2) \n\t"
+ 
+          
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y)
+            : "cc", "memory","r0","r1", "v0","v1","v2","v3","v4","v5","v6","v7","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31" 
+            );
+    return;
+
+}
+
+#endif
+
+int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
+{
+    BLASLONG i=0;
+    BLASLONG ix=0,iy=0;
+    FLOAT temp;
+
+    if ( n <= 0     )  return(0);
+
+    if ( (inc_x == 1) && (inc_y == 1 ))
+    {
+
+        BLASLONG n1 = n & -32;
+        if ( n1 > 0 )
+        {
+            dswap_kernel_32(n1, x, y);
+            i=n1;
+        }
+
+        while(i < n)
+        {
+            temp = y[i];    
+            y[i] = x[i] ;
+            x[i] = temp;
+            i++ ;
+
+        }
+
+
+    }
+    else
+    {
+
+        while(i < n)
+        {
+            temp  = y[iy];
+            y[iy] = x[ix] ;
+            x[ix] = temp;
+            ix += inc_x ;
+            iy += inc_y ;
+            i++ ;
+
+        }
+
+    }
+    return(0);
+    
+
+}
+
+

kernel/zarch/idamax.c

@@ -0,0 +1,249 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+#include <math.h>
+
+#if defined(DOUBLE)
+
+#define ABS fabs
+
+#else
+
+#define ABS fabsf
+
+#endif
+
+static BLASLONG __attribute__((noinline)) diamax_kernel_32_TUNED(BLASLONG n, FLOAT *x, FLOAT *maxf) {
+
+    __asm__(
+            "pfd 1, 0(%1) \n\t"
+            "sllg   %%r0,%0,3  \n\t"
+            "agr    %%r0,%1    \n\t"
+            "VLEIG  %%v20,0,0  \n\t"
+            "VLEIG  %%v20,1,1  \n\t"
+            "VLEIG  %%v21,2,0  \n\t"
+            "VLEIG  %%v21,3,1  \n\t"
+            "VLEIG  %%v22,4,0  \n\t"
+            "VLEIG  %%v22,5,1  \n\t"
+            "VLEIG  %%v23,6,0  \n\t"
+            "VLEIG  %%v23,7,1  \n\t"
+            "VREPIG %%v4,8     \n\t"
+            "vzero %%v5        \n\t"
+            "vzero %%v18       \n\t"
+            "vzero %%v19       \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%1 ) \n\t"
+            "vlm  %%v24,%%v31, 0(%1 ) \n\t"
+
+            "vflpdb  %%v24, %%v24 \n\t"
+            "vflpdb  %%v25, %%v25 \n\t"
+            "vflpdb  %%v26, %%v26 \n\t"
+            "vflpdb  %%v27, %%v27 \n\t"
+            "vflpdb  %%v28, %%v28 \n\t"
+            "vflpdb  %%v29, %%v29 \n\t"
+            "vflpdb  %%v30, %%v30 \n\t"
+            "vflpdb  %%v31, %%v31 \n\t"
+
+            "vfchdb  %%v16,%%v25,%%v24  \n\t "
+            "vfchdb  %%v17,%%v27,%%v26  \n\t "
+            "vsel    %%v1,%%v21,%%v20,%%v16 \n\t"
+            "vsel    %%v0,%%v25,%%v24,%%v16 \n\t"
+            "vsel    %%v2,%%v23,%%v22,%%v17 \n\t"
+            "vsel    %%v3,%%v27,%%v26,%%v17 \n\t"
+            "vfchdb  %%v16,%%v29,%%v28  \n\t "
+            "vfchdb  %%v17,%%v31,%%v30  \n\t"
+            "vsel    %%v24,%%v21,%%v20,%%v16 \n\t"
+            "vsel    %%v25,%%v29,%%v28,%%v16 \n\t"
+            "vsel    %%v26,%%v23,%%v22,%%v17 \n\t"
+            "vsel    %%v27,%%v31,%%v30,%%v17 \n\t"
+
+
+            "vfchdb  %%v28, %%v3,%%v0         \n\t"
+            "vfchdb  %%v29,%%v27, %%v25       \n\t"
+            "vsel    %%v1,%%v2,%%v1,%%v28     \n\t"
+            "vsel    %%v0,%%v3,%%v0,%%v28     \n\t"
+            "vsel    %%v24,%%v26,%%v24,%%v29  \n\t"
+            "vsel    %%v25,%%v27,%%v25,%%v29  \n\t"
+
+            "VAG %%v1,%%v1,%%v5    \n\t"
+            "VAG %%v24,%%v24,%%v5  \n\t"
+            "VAG %%v24,%%v24,%%v4  \n\t"
+
+            "vfchdb %%v16,%%v25 , %%v0     \n\t"
+            "VAG    %%v5,%%v5,%%v4         \n\t"
+            "vsel   %%v29,%%v25,%%v0,%%v16 \n\t"
+            "vsel   %%v28,%%v24,%%v1,%%v16 \n\t"
+
+            "vfchdb %%v17, %%v29,%%v18   \n\t"
+            "vsel  %%v19,%%v28,%%v19,%%v17 \n\t"
+            "vsel    %%v18,%%v29,%%v18,%%v17 \n\t"
+
+            "VAG %%v5,%%v5,%%v4        \n\t"
+            "vlm  %%v24,%%v31,128(%1 ) \n\t"
+
+            "vflpdb  %%v24, %%v24 \n\t"
+            "vflpdb  %%v25, %%v25 \n\t"
+            "vflpdb  %%v26, %%v26 \n\t"
+            "vflpdb  %%v27, %%v27 \n\t"
+            "vflpdb  %%v28, %%v28 \n\t"
+            "vflpdb  %%v29, %%v29 \n\t"
+            "vflpdb  %%v30, %%v30 \n\t"
+            "vflpdb  %%v31, %%v31 \n\t"
+
+            "vfchdb  %%v16,%%v25,%%v24  \n\t "
+            "vfchdb  %%v17,%%v27,%%v26  \n\t "
+            "vsel    %%v1,%%v21,%%v20,%%v16 \n\t"
+            "vsel    %%v0,%%v25,%%v24,%%v16 \n\t"
+            "vsel    %%v2,%%v23,%%v22,%%v17 \n\t"
+            "vsel    %%v3,%%v27,%%v26,%%v17 \n\t"
+            "vfchdb  %%v16,%%v29,%%v28  \n\t "
+            "vfchdb  %%v17,%%v31,%%v30  \n\t"
+            "vsel    %%v24,%%v21,%%v20,%%v16 \n\t"
+            "vsel    %%v25,%%v29,%%v28,%%v16 \n\t"
+            "vsel    %%v26,%%v23,%%v22,%%v17 \n\t"
+            "vsel    %%v27,%%v31,%%v30,%%v17 \n\t"
+
+
+            "vfchdb  %%v28, %%v3,%%v0     \n\t"
+            "vfchdb  %%v29,%%v27, %%v25   \n\t"
+            "vsel    %%v1,%%v2,%%v1,%%v28 \n\t"
+            "vsel    %%v0,%%v3,%%v0,%%v28 \n\t"
+            "vsel    %%v24,%%v26,%%v24,%%v29 \n\t"
+            "vsel    %%v25,%%v27,%%v25,%%v29 \n\t"
+
+            "VAG %%v1,%%v1,%%v5     \n\t"
+            "VAG %%v24,%%v24,%%v5   \n\t"
+            "la %1,256(%1)          \n\t"
+            "VAG %%v24,%%v24,%%v4   \n\t"
+
+            "vfchdb  %%v16,%%v25 , %%v0  \n\t"
+            "VAG     %%v5,%%v5,%%v4      \n\t"
+            "vsel    %%v29,%%v25,%%v0,%%v16 \n\t"
+            "vsel    %%v28,%%v24,%%v1,%%v16 \n\t"
+
+            "vfchdb  %%v17, %%v29,%%v18      \n\t"
+            "vsel    %%v19,%%v28,%%v19,%%v17 \n\t"
+            "vsel    %%v18,%%v29,%%v18,%%v17 \n\t"
+
+            "VAG %%v5,%%v5,%%v4 \n\t"
+
+            "clgrjl %1,%%r0,1b  \n\t"
+
+
+            "vrepg %%v26,%%v18,1     \n\t"
+            "vrepg %%v5,%%v19,1      \n\t"
+            "wfcdb  %%v26,%%v18      \n\t"
+            "jne 2f  \n\t"
+            "VSTEG   %%v18,0(%2),0   \n\t"
+            "VMNLG   %%v1,%%v5,%%v19 \n\t"
+            "VLGVG   %%r2,%%v1,0     \n\t"
+            "br %%r14  \n\t"
+            "2: \n\t"
+            "wfchdb  %%v16,%%v26,%%v18      \n\t"
+            "vsel    %%v1,%%v5,%%v19,%%v16  \n\t"
+            "vsel    %%v0,%%v26,%%v18,%%v16 \n\t"
+            "VLGVG  %%r2,%%v1,0  \n\t"
+            "std %%f0,0(%2)    \n\t"
+
+            :
+            : "r"(n), "a"(x), "a"(maxf)
+            : "cc", "memory","r0","r1","r2","f0","v0","v1","v2","v3","v4","v5","v6","v7","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+
+            );
+
+}
+ 
+BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
+    BLASLONG i = 0;
+    BLASLONG j = 0;
+    BLASLONG ix = 0;
+    FLOAT maxf = 0.0;
+    BLASLONG max = 0;
+
+    if (n <= 0 || inc_x <= 0) return (max);
+
+    if (inc_x == 1) {
+
+        BLASLONG n1 = n & -32;
+        if (n1 > 0) {
+
+            max = diamax_kernel_32_TUNED(n1, x, &maxf);
+
+            i = n1;
+        }
+
+        while (i < n) {
+            if (ABS(x[i]) > maxf) {
+                max = i;
+                maxf = ABS(x[i]);
+            }
+            i++;
+        }
+        return (max + 1);
+
+    } else {
+
+        BLASLONG n1 = n & -4;
+        while (j < n1) {
+
+            if (ABS(x[i]) > maxf) {
+                max = j;
+                maxf = ABS(x[i]);
+            }
+            if (ABS(x[i + inc_x]) > maxf) {
+                max = j + 1;
+                maxf = ABS(x[i + inc_x]);
+            }
+            if (ABS(x[i + 2 * inc_x]) > maxf) {
+                max = j + 2;
+                maxf = ABS(x[i + 2 * inc_x]);
+            }
+            if (ABS(x[i + 3 * inc_x]) > maxf) {
+                max = j + 3;
+                maxf = ABS(x[i + 3 * inc_x]);
+            }
+
+            i += inc_x * 4;
+
+            j += 4;
+
+        }
+
+
+        while (j < n) {
+            if (ABS(x[i]) > maxf) {
+                max = j;
+                maxf = ABS(x[i]);
+            }
+            i += inc_x;
+            j++;
+        }
+        return (max + 1);
+    }
+}

kernel/zarch/idamin.c

@@ -0,0 +1,249 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+#include <math.h>
+
+#if defined(DOUBLE)
+
+#define ABS fabs
+
+#else
+
+#define ABS fabsf
+
+#endif
+
+static BLASLONG __attribute__((noinline)) diamin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *maxf) {
+
+    __asm__(
+            "pfd 1, 0(%1) \n\t"
+            "sllg %%r0,%0,3    \n\t"
+            "agr %%r0,%1  \n\t"
+            "VLEIG  %%v20,0,0  \n\t"
+            "VLEIG  %%v20,1,1  \n\t"
+            "VLEIG  %%v21,2,0  \n\t"
+            "VLEIG  %%v21,3,1  \n\t"
+            "VLEIG  %%v22,4,0  \n\t"
+            "VLEIG  %%v22,5,1  \n\t"
+            "VLEIG  %%v23,6,0  \n\t"
+            "VLEIG  %%v23,7,1  \n\t"
+            "VREPIG %%v4,8     \n\t"
+            "vzero %%v5        \n\t"
+            "vlrepg %%v18,0(%1)   \n\t"
+            "vzero %%v19          \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%1 ) \n\t"
+            "vlm  %%v24,%%v31, 0(%1 ) \n\t"
+
+            "vflpdb  %%v24, %%v24 \n\t"
+            "vflpdb  %%v25, %%v25 \n\t"
+            "vflpdb  %%v26, %%v26 \n\t"
+            "vflpdb  %%v27, %%v27 \n\t"
+            "vflpdb  %%v28, %%v28 \n\t"
+            "vflpdb  %%v29, %%v29 \n\t"
+            "vflpdb  %%v30, %%v30 \n\t"
+            "vflpdb  %%v31, %%v31 \n\t"
+
+            "vfchdb  %%v16,%%v24,%%v25  \n\t "
+            "vfchdb  %%v17,%%v26 ,%%v27 \n\t "
+            "vsel    %%v1,%%v21,%%v20,%%v16 \n\t"
+            "vsel    %%v0,%%v25,%%v24,%%v16 \n\t"
+            "vsel    %%v2,%%v23,%%v22,%%v17 \n\t"
+            "vsel    %%v3,%%v27,%%v26,%%v17 \n\t"
+            "vfchdb  %%v16,%%v28, %%v29 \n\t "
+            "vfchdb  %%v17,%%v30,%%v31  \n\t"
+            "vsel    %%v24,%%v21,%%v20,%%v16 \n\t"
+            "vsel    %%v25,%%v29,%%v28,%%v16 \n\t"
+            "vsel    %%v26,%%v23,%%v22,%%v17 \n\t"
+            "vsel    %%v27,%%v31,%%v30,%%v17 \n\t"
+
+
+            "vfchdb  %%v28,%%v0 , %%v3       \n\t"
+            "vfchdb  %%v29, %%v25,%%v27      \n\t"
+            "vsel    %%v1,%%v2,%%v1,%%v28    \n\t"
+            "vsel    %%v0,%%v3,%%v0,%%v28    \n\t"
+            "vsel    %%v24,%%v26,%%v24,%%v29 \n\t"
+            "vsel    %%v25,%%v27,%%v25,%%v29 \n\t"
+
+            "VAG %%v1,%%v1,%%v5   \n\t"
+            "VAG %%v24,%%v24,%%v5   \n\t"
+            "VAG %%v24,%%v24,%%v4   \n\t"
+
+            "vfchdb %%v16, %%v0,%%v25   \n\t"
+            "VAG %%v5,%%v5,%%v4 \n\t"
+            "vsel    %%v29,%%v25,%%v0,%%v16 \n\t"
+            "vsel  %%v28,%%v24,%%v1,%%v16 \n\t"
+
+            "vfchdb %%v17,%%v18, %%v29  \n\t"
+            "vsel  %%v19,%%v28,%%v19,%%v17 \n\t"
+            "vsel    %%v18,%%v29,%%v18,%%v17 \n\t"
+
+            "VAG %%v5,%%v5,%%v4 \n\t"
+
+            "vlm  %%v24,%%v31,128(%1 ) \n\t"
+            "vflpdb  %%v24, %%v24 \n\t"
+            "vflpdb  %%v25, %%v25 \n\t"
+            "vflpdb  %%v26, %%v26 \n\t"
+            "vflpdb  %%v27, %%v27 \n\t"
+            "vflpdb  %%v28, %%v28 \n\t"
+            "vflpdb  %%v29, %%v29 \n\t"
+            "vflpdb  %%v30, %%v30 \n\t"
+            "vflpdb  %%v31, %%v31 \n\t"
+
+            "vfchdb  %%v16,%%v24,%%v25  \n\t"
+            "vfchdb  %%v17,%%v26 ,%%v27 \n\t"
+            "vsel    %%v1,%%v21,%%v20,%%v16 \n\t"
+            "vsel    %%v0,%%v25,%%v24,%%v16 \n\t"
+            "vsel    %%v2,%%v23,%%v22,%%v17 \n\t"
+            "vsel    %%v3,%%v27,%%v26,%%v17 \n\t"
+            "vfchdb  %%v16,%%v28 ,%%v29 \n\t"
+            "vfchdb  %%v17,%%v30,%%v31  \n\t"
+            "vsel    %%v24,%%v21,%%v20,%%v16 \n\t"
+            "vsel    %%v25,%%v29,%%v28,%%v16 \n\t"
+            "vsel    %%v26,%%v23,%%v22,%%v17 \n\t"
+            "vsel    %%v27,%%v31,%%v30,%%v17 \n\t"
+
+
+            "vfchdb  %%v28,%%v0 , %%v3       \n\t"
+            "vfchdb  %%v29, %%v25,%%v27      \n\t"
+            "vsel    %%v1,%%v2,%%v1,%%v28    \n\t"
+            "vsel    %%v0,%%v3,%%v0,%%v28    \n\t"
+            "vsel    %%v24,%%v26,%%v24,%%v29 \n\t"
+            "vsel    %%v25,%%v27,%%v25,%%v29 \n\t"
+
+            "VAG     %%v1,%%v1,%%v5     \n\t"
+            "VAG     %%v24,%%v24,%%v5   \n\t"
+            "la %1,256(%1)   \n\t"
+            "VAG     %%v24,%%v24,%%v4   \n\t"
+
+            "vfchdb  %%v16, %%v0,%%v25      \n\t"
+            "VAG     %%v5,%%v5,%%v4         \n\t"
+            "vsel    %%v29,%%v25,%%v0,%%v16 \n\t"
+            "vsel    %%v28,%%v24,%%v1,%%v16 \n\t"
+
+            "vfchdb  %%v17,%%v18, %%v29      \n\t"
+            "vsel    %%v19,%%v28,%%v19,%%v17 \n\t"
+            "vsel    %%v18,%%v29,%%v18,%%v17 \n\t"
+
+            "VAG     %%v5,%%v5,%%v4 \n\t"
+
+            "clgrjl %1,%%r0,1b \n\t"
+
+
+            "vrepg   %%v26,%%v18,1   \n\t"
+            "vrepg   %%v5,%%v19,1    \n\t"
+            "wfcdb   %%v26,%%v18     \n\t"
+            "jne 2f  \n\t"
+            "VSTEG   %%v18,0(%2),0  \n\t"
+            "VMNLG   %%v1,%%v5,%%v19 \n\t"
+            "VLGVG   %%r2,%%v1,0  \n\t"
+            "br %%r14  \n\t"
+            "2: \n\t"
+            "wfchdb  %%v16,%%v18 ,%%v26     \n\t "
+            "vsel    %%v1,%%v5,%%v19,%%v16  \n\t"
+            "vsel    %%v0,%%v26,%%v18,%%v16 \n\t"
+            "VLGVG   %%r2,%%v1,0  \n\t"
+            "std %%f0,0(%2)    \n\t"
+
+            :
+            : "r"(n), "a"(x), "a"(maxf)
+            : "cc", "memory","r0","r1","r2","f0","v0","v1","v2","v3","v4","v5","v6","v7","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+
+            );
+
+}
+
+BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
+    BLASLONG i = 0;
+    BLASLONG j = 0;
+    BLASLONG ix = 0;
+    FLOAT minf = 0.0;
+    BLASLONG min = 0;
+
+    if (n <= 0 || inc_x <= 0) return (min);
+
+    if (inc_x == 1) {
+
+        BLASLONG n1 = n & -32;
+        if (n1 > 0) {
+
+            min = diamin_kernel_32(n1, x, &minf);
+
+            i = n1;
+        }
+
+        while (i < n) {
+            if (ABS(x[i]) < minf) {
+                min = i;
+                minf = ABS(x[i]);
+            }
+            i++;
+        }
+        return (min + 1);
+
+    } else {
+
+        BLASLONG n1 = n & -4;
+        while (j < n1) {
+
+            if (ABS(x[i]) < minf) {
+                min = j;
+                minf = ABS(x[i]);
+            }
+            if (ABS(x[i + inc_x]) < minf) {
+                min = j + 1;
+                minf = ABS(x[i + inc_x]);
+            }
+            if (ABS(x[i + 2 * inc_x]) < minf) {
+                min = j + 2;
+                minf = ABS(x[i + 2 * inc_x]);
+            }
+            if (ABS(x[i + 3 * inc_x]) < minf) {
+                min = j + 3;
+                minf = ABS(x[i + 3 * inc_x]);
+            }
+
+            i += inc_x * 4;
+
+            j += 4;
+
+        }
+
+
+        while (j < n) {
+            if (ABS(x[i]) < minf) {
+                min = j;
+                minf = ABS(x[i]);
+            }
+            i += inc_x;
+            j++;
+        }
+        return (min + 1);
+    }
+}

kernel/zarch/izamax.c

@@ -0,0 +1,257 @@
+/***************************************************************************
+Copyright (c) 2017, 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"
+#include <math.h>
+ 
+#define ABS fabs 
+#define CABS1(x,i)    ABS(x[i])+ABS(x[i+1])
+
+ 
+
+static BLASLONG __attribute__((noinline)) ziamax_kernel_8_TUNED(BLASLONG n, FLOAT *x, FLOAT *maxf) { 
+
+    __asm__(
+            "pfd 1, 0(%1) \n\t" 
+            "VLEIG  %%v16,0,0  \n\t"
+            "VLEIG  %%v16,1,1  \n\t"
+            "VLEIG  %%v17,2,0  \n\t"
+            "VLEIG  %%v17,3,1  \n\t"
+            "VLEIG  %%v18,4,0  \n\t"
+            "VLEIG  %%v18,5,1  \n\t"
+            "VLEIG  %%v19,6,0  \n\t"
+            "VLEIG  %%v19,7,1  \n\t" 
+            "VLEIG  %%v20,8,0  \n\t"
+            "VLEIG  %%v20,9,1  \n\t"
+            "VLEIG  %%v21,10,0 \n\t"
+            "VLEIG  %%v21,11,1 \n\t"
+            "VLEIG  %%v22,12,0 \n\t"
+            "VLEIG  %%v22,13,1 \n\t"
+            "VLEIG  %%v23,14,0 \n\t"
+            "VLEIG  %%v23,15,1 \n\t" 
+    
+    
+            "sllg %%r0,%0,4 \n\t"
+            "agr %%r0,%1    \n\t" 
+            "vzero %%v6     \n\t"
+            "vzero %%v7     \n\t"
+            "VREPIG %%v4,16 \n\t"
+            "vzero %%v5     \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%1 ) \n\t"
+        
+            "vleg %%v24 ,  0( %1),0 \n\t" 
+            "vleg %%v25 ,  8( %1),0 \n\t"
+            "vleg %%v24 , 16( %1),1 \n\t" 
+            "vleg %%v25 , 24( %1),1 \n\t"
+            "vleg %%v26 , 32( %1),0 \n\t"  
+            "vleg %%v27 , 40( %1),0 \n\t"
+            "vleg %%v26 , 48( %1),1 \n\t" 
+            "vleg %%v27 , 56( %1),1 \n\t" 
+            "vleg %%v28 , 64( %1),0 \n\t" 
+            "vleg %%v29 , 72( %1),0 \n\t" 
+            "vleg %%v28 , 80( %1),1 \n\t"
+            "vleg %%v29 , 88( %1),1 \n\t" 
+            "vleg %%v30 , 96( %1),0 \n\t"  
+            "vleg %%v31 ,104( %1),0 \n\t"
+            "vleg %%v30 ,112( %1),1 \n\t"
+            "vleg %%v31 ,120( %1),1 \n\t"  
+            "vflpdb  %%v24, %%v24   \n\t" 
+            "vflpdb  %%v25, %%v25   \n\t" 
+            "vflpdb  %%v26, %%v26   \n\t" 
+            "vflpdb  %%v27, %%v27   \n\t" 
+            "vflpdb  %%v28, %%v28   \n\t" 
+            "vflpdb  %%v29, %%v29   \n\t"
+            "vflpdb  %%v30, %%v30   \n\t" 
+            "vflpdb  %%v31, %%v31   \n\t"    
+     
+            "vfadb %%v0,%%v24,%%v25 \n\t"
+            "vfadb %%v1,%%v26,%%v27 \n\t"
+            "vfadb %%v2,%%v28,%%v29 \n\t"
+            "vfadb %%v3,%%v30,%%v31 \n\t"
+     
+    
+            "vleg %%v24 , 128( %1),0  \n\t" 
+            "vleg %%v25 , 136( %1),0  \n\t"
+            "vleg %%v24 , 144( %1),1  \n\t" 
+            "vleg %%v25 , 152( %1),1  \n\t"
+            "vleg %%v26 , 160( %1),0  \n\t"  
+            "vleg %%v27 , 168( %1),0  \n\t"
+            "vleg %%v26 , 176( %1),1  \n\t" 
+            "vleg %%v27 , 184( %1),1  \n\t" 
+            "vleg %%v28 , 192( %1),0  \n\t" 
+            "vleg %%v29 , 200( %1),0  \n\t" 
+            "vleg %%v28 , 208( %1),1  \n\t"
+            "vleg %%v29 , 216( %1),1  \n\t" 
+            "vleg %%v30 , 224( %1),0  \n\t"  
+            "vleg %%v31 , 232( %1),0  \n\t"
+            "vleg %%v30 , 240( %1),1  \n\t"
+            "vleg %%v31 , 248( %1),1  \n\t"  
+            "vflpdb  %%v24, %%v24 \n\t" 
+            "vflpdb  %%v25, %%v25 \n\t" 
+            "vflpdb  %%v26, %%v26 \n\t" 
+            "vflpdb  %%v27, %%v27 \n\t" 
+            "vflpdb  %%v28, %%v28 \n\t" 
+            "vflpdb  %%v29, %%v29 \n\t"
+            "vflpdb  %%v30, %%v30 \n\t" 
+            "vflpdb  %%v31, %%v31 \n\t"    
+     
+            "vfadb %%v24,%%v24,%%v25  \n\t"
+            "vfadb %%v26,%%v26,%%v27  \n\t"
+            "vfadb %%v28,%%v28,%%v29  \n\t"
+            "vfadb %%v30,%%v30,%%v31  \n\t"
+    
+            "vfchdb  %%v25,%%v1,%%v0  \n\t" 
+            "vsel  %%v29,%%v17,%%v16,%%v25 \n\t"
+            "vsel  %%v31,%%v1,%%v0,%%v25   \n\t"  
+    
+            "vfchdb  %%v27,%%v3,%%v2 \n\t "   
+            "vsel  %%v0,%%v19,%%v18,%%v27 \n\t"
+            "vsel  %%v1,%%v3,%%v2,%%v27   \n\t"  
+    
+            "vfchdb  %%v25,%%v26,%%v24  \n\t " 
+            "vsel  %%v2,%%v21,%%v20,%%v25 \n\t"
+            "vsel  %%v3,%%v26,%%v24,%%v25 \n\t" 
+    
+            "vfchdb  %%v27,%%v30,%%v28 \n\t "   
+            "vsel  %%v25,%%v23,%%v22,%%v27 \n\t"
+            "vsel  %%v27,%%v30,%%v28,%%v27 \n\t"  
+    
+            "vfchdb %%v24, %%v1,%%v31      \n\t" 
+            "vsel  %%v26,%%v0,%%v29,%%v24  \n\t" 
+            "vsel  %%v28,%%v1,%%v31,%%v24  \n\t"
+    
+            "vfchdb %%v30, %%v27,%%v3      \n\t" 
+            "vsel  %%v29,%%v25,%%v2,%%v30  \n\t" 
+            "vsel  %%v31,%%v27,%%v3 ,%%v30 \n\t" 
+    
+            "la %1,256(%1)      \n\t"  
+           
+            "vfchdb %%v0, %%v31,%%v28     \n\t" 
+            "vsel  %%v25,%%v29,%%v26,%%v0 \n\t" 
+            "vsel  %%v27,%%v31,%%v28,%%v0 \n\t"
+    
+            "VAG %%v25,%%v25,%%v5 \n\t"
+    
+             //cmp with previous
+            "vfchdb %%v30, %%v27,%%v6    \n\t"
+            "vsel  %%v7,%%v25,%%v7,%%v30 \n\t" 
+            "vsel  %%v6,%%v27,%%v6,%%v30 \n\t"
+    
+            "VAG %%v5,%%v5,%%v4 \n\t" 
+    
+            "clgrjl %1,%%r0,1b \n\t"
+
+            //xtract index
+            "vrepg %%v26,%%v6,1      \n\t"
+            "vrepg %%v5,%%v7,1       \n\t"
+            "wfcdb  %%v26,%%v6       \n\t"
+            "jne 2f  \n\t"
+            "VSTEG   %%v6,0(%2),0  \n\t"
+            "VMNLG  %%v1,%%v5,%%v7 \n\t"
+            "VLGVG  %%r2,%%v1,0  \n\t"
+            "br %%r14  \n\t"
+            "2: \n\t"
+            "wfchdb  %%v16,%%v26,%%v6      \n\t"
+            "vsel    %%v1,%%v5,%%v7,%%v16  \n\t"
+            "vsel    %%v0,%%v26,%%v6,%%v16 \n\t"
+            "VLGVG  %%r2,%%v1,0  \n\t"
+            "std %%f0,0(%2)    \n\t"
+
+            :
+            : "r"(n), "a"(x), "a"(maxf)
+            : "cc", "memory","r0","r1","r2","f0","v0","v1","v2","v3","v4","v5","v6","v7","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+
+            );
+
+}
+
+
+  
+
+ 
+ 
+
+BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
+{
+    BLASLONG i = 0;
+    BLASLONG ix = 0;
+    FLOAT maxf = 0;
+    BLASLONG max = 0;
+    BLASLONG inc_x2;
+
+    if (n <= 0 || inc_x <= 0) return(max);
+     
+    if (inc_x == 1) {
+
+      BLASLONG n1 = n & -8;
+      if (n1 > 0) {
+
+            max = ziamax_kernel_8_TUNED(n1, x, &maxf);
+
+            i = n1;
+      }
+
+      while(i < n)
+    {
+        if( CABS1(x,ix) > maxf )
+        {
+            max = i;
+            maxf = CABS1(x,ix);
+        }
+        ix += 2;
+        i++;
+    }
+        return (max + 1);
+
+    } else {
+ 
+      inc_x2 = 2 * inc_x;
+
+    maxf = CABS1(x,0);
+    ix += inc_x2;
+    i++;
+
+    while(i < n)
+    {
+        if( CABS1(x,ix) > maxf )
+        {
+            max = i;
+            maxf = CABS1(x,ix);
+        }
+        ix += inc_x2;
+        i++;
+    }
+        return (max + 1);
+    }
+ 
+}
+
+

kernel/zarch/izamin.c

@@ -0,0 +1,259 @@
+/***************************************************************************
+Copyright (c) 2017, 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"
+#include <math.h>
+ 
+#define ABS fabs 
+#define CABS1(x,i)    ABS(x[i])+ABS(x[i+1])
+
+ 
+
+static BLASLONG __attribute__((noinline)) ziamin_kernel_8_TUNED(BLASLONG n, FLOAT *x, FLOAT *minf) { 
+
+    __asm__(
+            "pfd 1, 0(%1) \n\t" 
+            "VLEIG  %%v16,0,0  \n\t"
+            "VLEIG  %%v16,1,1  \n\t"
+            "VLEIG  %%v17,2,0  \n\t"
+            "VLEIG  %%v17,3,1  \n\t"
+            "VLEIG  %%v18,4,0  \n\t"
+            "VLEIG  %%v18,5,1  \n\t"
+            "VLEIG  %%v19,6,0  \n\t"
+            "VLEIG  %%v19,7,1  \n\t" 
+            "VLEIG  %%v20,8,0  \n\t"
+            "VLEIG  %%v20,9,1  \n\t"
+            "VLEIG  %%v21,10,0 \n\t"
+            "VLEIG  %%v21,11,1 \n\t"
+            "VLEIG  %%v22,12,0 \n\t"
+            "VLEIG  %%v22,13,1 \n\t"
+            "VLEIG  %%v23,14,0 \n\t"
+            "VLEIG  %%v23,15,1 \n\t" 
+            "ld %%f6,0(%1)     \n\t"
+            "lpdbr %%f6,%%f6   \n\t" 
+            "ld %%f7,8(%1)     \n\t"
+            "lpdbr %%f7,%%f7   \n\t"   
+            "adbr %%f6,%%f7    \n\t"      
+            "sllg %%r0,%0,4    \n\t"
+            "agr %%r0,%1       \n\t" 
+            "vrepg %%v6,%%v6,0 \n\t"
+            "vzero %%v7        \n\t"
+            "VREPIG %%v4,16    \n\t"
+            "vzero %%v5        \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%1 ) \n\t"
+        
+            "vleg %%v24 ,  0( %1),0 \n\t" 
+            "vleg %%v25 ,  8( %1),0 \n\t"
+            "vleg %%v24 , 16( %1),1 \n\t" 
+            "vleg %%v25 , 24( %1),1 \n\t"
+            "vleg %%v26 , 32( %1),0 \n\t"  
+            "vleg %%v27 , 40( %1),0 \n\t"
+            "vleg %%v26 , 48( %1),1 \n\t" 
+            "vleg %%v27 , 56( %1),1 \n\t" 
+            "vleg %%v28 , 64( %1),0 \n\t" 
+            "vleg %%v29 , 72( %1),0 \n\t" 
+            "vleg %%v28 , 80( %1),1 \n\t"
+            "vleg %%v29 , 88( %1),1 \n\t" 
+            "vleg %%v30 , 96( %1),0 \n\t"  
+            "vleg %%v31 ,104( %1),0 \n\t"
+            "vleg %%v30 ,112( %1),1 \n\t"
+            "vleg %%v31 ,120( %1),1 \n\t"  
+            "vflpdb  %%v24, %%v24   \n\t" 
+            "vflpdb  %%v25, %%v25   \n\t" 
+            "vflpdb  %%v26, %%v26   \n\t" 
+            "vflpdb  %%v27, %%v27   \n\t" 
+            "vflpdb  %%v28, %%v28   \n\t" 
+            "vflpdb  %%v29, %%v29   \n\t"
+            "vflpdb  %%v30, %%v30   \n\t" 
+            "vflpdb  %%v31, %%v31   \n\t"    
+     
+            "vfadb %%v0,%%v24,%%v25 \n\t"
+            "vfadb %%v1,%%v26,%%v27 \n\t"
+            "vfadb %%v2,%%v28,%%v29 \n\t"
+            "vfadb %%v3,%%v30,%%v31 \n\t"
+     
+    
+            "vleg %%v24 ,128( %1),0 \n\t" 
+            "vleg %%v25 ,136( %1),0 \n\t"
+            "vleg %%v24 ,144( %1),1 \n\t" 
+            "vleg %%v25 ,152( %1),1 \n\t"
+            "vleg %%v26 ,160( %1),0 \n\t"  
+            "vleg %%v27 ,168( %1),0 \n\t"
+            "vleg %%v26 ,176( %1),1 \n\t" 
+            "vleg %%v27 ,184( %1),1 \n\t" 
+            "vleg %%v28 ,192( %1),0 \n\t" 
+            "vleg %%v29 ,200( %1),0 \n\t" 
+            "vleg %%v28 ,208( %1),1 \n\t"
+            "vleg %%v29 ,216( %1),1 \n\t" 
+            "vleg %%v30 ,224( %1),0 \n\t"  
+            "vleg %%v31 ,232( %1),0 \n\t"
+            "vleg %%v30 ,240( %1),1 \n\t"
+            "vleg %%v31 ,248( %1),1 \n\t"  
+            "vflpdb  %%v24, %%v24   \n\t" 
+            "vflpdb  %%v25, %%v25   \n\t" 
+            "vflpdb  %%v26, %%v26   \n\t" 
+            "vflpdb  %%v27, %%v27   \n\t" 
+            "vflpdb  %%v28, %%v28   \n\t" 
+            "vflpdb  %%v29, %%v29   \n\t"
+            "vflpdb  %%v30, %%v30   \n\t" 
+            "vflpdb  %%v31, %%v31   \n\t"    
+     
+            "vfadb %%v24,%%v24,%%v25  \n\t"
+            "vfadb %%v26,%%v26,%%v27  \n\t"
+            "vfadb %%v28,%%v28,%%v29  \n\t"
+            "vfadb %%v30,%%v30,%%v31  \n\t"
+    
+    
+            "vfchdb  %%v25,%%v0 ,%%v1      \n\t" 
+            "vsel  %%v29,%%v17,%%v16,%%v25 \n\t"
+            "vsel  %%v31,%%v1,%%v0,%%v25   \n\t"  
+    
+            "vfchdb  %%v27,%%v2,%%v3       \n\t"   
+            "vsel  %%v0,%%v19,%%v18,%%v27  \n\t"
+            "vsel  %%v1,%%v3,%%v2,%%v27    \n\t"  
+    
+            "vfchdb  %%v25,%%v24,%%v26     \n\t" 
+            "vsel  %%v2,%%v21,%%v20,%%v25  \n\t"
+            "vsel  %%v3,%%v26,%%v24,%%v25  \n\t" 
+    
+            "vfchdb  %%v27,%%v28,%%v30     \n\t"   
+            "vsel  %%v25,%%v23,%%v22,%%v27 \n\t"
+            "vsel  %%v27,%%v30,%%v28,%%v27 \n\t"  
+    
+            "vfchdb %%v24,%%v31, %%v1      \n\t" 
+            "vsel  %%v26,%%v0,%%v29,%%v24  \n\t" 
+            "vsel  %%v28,%%v1,%%v31,%%v24  \n\t"
+    
+            "vfchdb %%v30,%%v3, %%v27      \n\t" 
+            "vsel  %%v29,%%v25,%%v2,%%v30  \n\t" 
+            "vsel  %%v31,%%v27,%%v3 ,%%v30 \n\t" 
+    
+            "la %1,256(%1)      \n\t"  
+           
+            "vfchdb %%v0,%%v28, %%v31      \n\t" 
+            "vsel  %%v25,%%v29,%%v26,%%v0  \n\t" 
+            "vsel  %%v27,%%v31,%%v28,%%v0  \n\t"
+    
+            "VAG %%v25,%%v25,%%v5 \n\t"
+    
+             //cmp with previous
+            "vfchdb %%v30,%%v6 , %%v27     \n\t"
+            "vsel  %%v7,%%v25,%%v7,%%v30   \n\t" 
+            "vsel  %%v6,%%v27,%%v6,%%v30   \n\t"
+    
+            "VAG %%v5,%%v5,%%v4 \n\t" 
+    
+            "clgrjl %1,%%r0,1b  \n\t"
+
+            //xtract index
+            "vrepg %%v26,%%v6,1      \n\t"
+            "vrepg %%v5,%%v7,1       \n\t"
+            "wfcdb  %%v26,%%v6       \n\t"
+            "jne 2f  \n\t"
+            "VSTEG   %%v6,0(%2),0    \n\t"
+            "VMNLG  %%v1,%%v5,%%v7   \n\t"
+            "VLGVG  %%r2,%%v1,0      \n\t"
+            "br %%r14  \n\t"
+            "2: \n\t"
+            "wfchdb  %%v16,%%v6 ,%%v26     \n\t"
+            "vsel    %%v1,%%v5,%%v7,%%v16  \n\t"
+            "vsel    %%v0,%%v26,%%v6,%%v16 \n\t"
+            "VLGVG   %%r2,%%v1,0  \n\t"
+            "std %%f0,0(%2)       \n\t"
+
+            :
+            : "r"(n), "a"(x), "a"(minf)
+            : "cc", "memory","r0","r1","r2","f0","v0","v1","v2","v3","v4","v5","v6","v7","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+
+            );
+
+}
+
+ 
+ 
+
+BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
+{
+    BLASLONG i=0;
+    BLASLONG ix=0;
+    FLOAT minf;
+    BLASLONG min=0;
+    BLASLONG inc_x2;
+
+    if (n <= 0 || inc_x <= 0) return(min);
+    
+
+    if (inc_x == 1) {
+
+        BLASLONG n1 = n & -8;
+        if (n1 > 0) {
+
+            min = ziamin_kernel_8_TUNED(n1, x, &minf);
+
+            i = n1;
+        }
+
+        while(i < n)
+        {
+            if( CABS1(x,ix) < minf )
+            {
+                min = i;
+                minf = CABS1(x,ix);
+            }
+            ix += 2;
+            i++;
+        }
+        return (min + 1);
+
+    } else {
+ 
+        inc_x2 = 2 * inc_x;
+
+        minf = CABS1(x,0);
+        ix += inc_x2;
+        i++;
+
+        while(i < n)
+        {
+            if( CABS1(x,ix) < minf )
+            {
+                min = i;
+                minf = CABS1(x,ix);
+            }
+            ix += inc_x2;
+            i++;
+        }
+        return (min + 1);
+    }
+ 
+}
+
+

kernel/zarch/zasum.c

@@ -0,0 +1,156 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+#include <math.h>
+
+#if defined(DOUBLE)
+
+#define ABS fabs
+
+#else
+
+#define ABS fabsf
+
+#endif
+
+
+static FLOAT __attribute__ ((noinline)) zasum_kernel_16(BLASLONG n, FLOAT *x) {
+
+    __asm__ (
+            "pfd 1, 0(%1) \n\t"
+            "sllg  %%r0,%0,4 \n\t"
+            "agr   %%r0,%1   \n\t"   
+            "vzero %%v0      \n\t"
+            "vzero %%v1      \n\t"
+            "vzero %%v22     \n\t"
+            "vzero %%v23     \n\t"   
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%1 ) \n\t"
+            "vlm %%v24,%%v31,0(%1) \n\t"  
+    
+            "vflpdb  %%v24, %%v24 \n\t"
+            "vflpdb  %%v25, %%v25 \n\t"
+            "vflpdb  %%v26, %%v26 \n\t"
+            "vflpdb  %%v27, %%v27 \n\t"
+            "vflpdb  %%v28, %%v28 \n\t"
+            "vflpdb  %%v29, %%v29 \n\t"
+            "vflpdb  %%v30, %%v30 \n\t"
+            "vflpdb  %%v31, %%v31 \n\t"
+    
+            "vfadb   %%v0,%%v0,%%v24    \n\t"
+            "vfadb   %%v1,%%v1,%%v25    \n\t"
+            "vfadb   %%v23,%%v23,%%v26  \n\t"
+            "vfadb   %%v22,%%v22,%%v27  \n\t" 
+            "vfadb   %%v0,%%v0,%%v28    \n\t"
+            "vfadb   %%v1,%%v1,%%v29    \n\t"
+            "vfadb   %%v23,%%v23,%%v30  \n\t"
+            "vfadb   %%v22,%%v22,%%v31  \n\t" 
+    
+            "vlm  %%v24,%%v31, 128(%1 ) \n\t"  
+    
+            "vflpdb  %%v24, %%v24 \n\t"
+            "vflpdb  %%v25, %%v25 \n\t"
+            "vflpdb  %%v26, %%v26 \n\t"
+            "vflpdb  %%v27, %%v27 \n\t"
+            "vflpdb  %%v28, %%v28 \n\t"
+            "vflpdb  %%v29, %%v29 \n\t"
+            "vflpdb  %%v30, %%v30 \n\t"
+            "vflpdb  %%v31, %%v31 \n\t"
+            "la %1,256(%1) \n\t"  
+            "vfadb   %%v0,%%v0,%%v24   \n\t"
+            "vfadb   %%v1,%%v1,%%v25   \n\t"
+            "vfadb   %%v23,%%v23,%%v26 \n\t"
+            "vfadb   %%v22,%%v22,%%v27 \n\t" 
+            "vfadb   %%v0,%%v0,%%v28   \n\t"
+            "vfadb   %%v1,%%v1,%%v29   \n\t"
+            "vfadb   %%v23,%%v23,%%v30 \n\t"
+            "vfadb   %%v22,%%v22,%%v31 \n\t"  
+            
+            "clgrjl %1,%%r0,1b \n\t"
+            "vfadb   %%v24,%%v0,%%v1   \n\t"
+            "vfadb   %%v25,%%v23,%%v22 \n\t"
+            "vfadb   %%v0,%%v25,%%v24  \n\t"
+            "vrepg %%v1,%%v0,1         \n\t"
+            "adbr %%f0,%%f1    \n\t"
+            :
+            : "r"(n), "a"(x)
+            : "cc", "memory","r0","f0","f1","v0","v1","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+            );
+
+}
+
+ 
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
+{
+    BLASLONG i=0;
+    BLASLONG ip=0;
+    FLOAT sumf = 0.0; 
+    BLASLONG n1;
+    BLASLONG inc_x2;
+
+    if (n <= 0 || inc_x <= 0) return(sumf);
+
+    if ( inc_x == 1 )
+    {
+
+        n1 = n & -16;
+        if ( n1 > 0 )
+        {
+
+            sumf=zasum_kernel_16(n1, x ); 
+            i=n1;
+            ip=2*n1;
+        }
+
+        while(i < n)
+        {
+            sumf += ABS(x[ip]) + ABS(x[ip+1]);
+            i++;
+            ip+=2;
+        }
+
+    }
+    else
+    {
+        inc_x2 = 2* inc_x;
+
+        while(i < n)
+        {
+            sumf += ABS(x[ip]) + ABS(x[ip+1]);
+            ip+=inc_x2;
+            i++;
+        }
+
+    }
+    return(sumf);
+}
+
+

kernel/zarch/zaxpy.c

@@ -0,0 +1,207 @@
+/***************************************************************************
+Copyright (c) 2017, 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"
+
+static void __attribute__ ((noinline)) zaxpy_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) {
+    __asm__ ("pfd 1, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+            "vlrepg %%v28 , 0(%3) \n\t"
+            "vlrepg %%v29, 8(%3)  \n\t"
+            "srlg %3,%0,3         \n\t"
+            "xgr  %%r1,%%r1       \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+            "vleg %%v16 ,  0(%%r1,%2),0  \n\t"
+            "vleg %%v17 ,  8(%%r1,%2),0  \n\t"
+            "vleg %%v16 , 16(%%r1,%2),1  \n\t"
+            "vleg %%v17 , 24(%%r1,%2),1  \n\t"
+
+            "vleg %%v18 , 32(%%r1,%2),0  \n\t"
+            "vleg %%v19 , 40(%%r1,%2),0  \n\t"
+            "vleg %%v18 , 48(%%r1,%2),1  \n\t"
+            "vleg %%v19 , 56(%%r1,%2),1  \n\t"
+
+            "vleg %%v24 ,  0(%%r1,%1),0  \n\t"
+            "vleg %%v25 ,  8(%%r1,%1),0  \n\t"
+            "vleg %%v24 , 16(%%r1,%1),1  \n\t"
+            "vleg %%v25 , 24(%%r1,%1),1  \n\t"
+
+            "vleg %%v26 , 32(%%r1,%1),0  \n\t"
+            "vleg %%v27 , 40(%%r1,%1),0  \n\t"
+            "vleg %%v26 , 48(%%r1,%1),1  \n\t"
+            "vleg %%v27 , 56(%%r1,%1),1  \n\t"
+#if !defined(CONJ)
+            "vfmsdb %%v16,  %%v25, %%v29,%%v16  \n\t"
+            "vfmadb %%v17,  %%v24, %%v29, %%v17 \n\t"
+            "vfmsdb %%v18,  %%v27, %%v29, %%v18 \n\t"
+            "vfmadb %%v19,  %%v26, %%v29, %%v19 \n\t"
+
+            "vfmsdb %%v16, %%v24, %%v28 ,%%v16  \n\t"
+            "vfmadb %%v17, %%v25, %%v28, %%v17  \n\t"
+            "vfmsdb %%v18, %%v26, %%v28, %%v18  \n\t"
+            "vfmadb %%v19, %%v27, %%v28, %%v19  \n\t"
+#else
+            "vfmadb %%v16, %%v25, %%v29, %%v16  \n\t"
+            "vfmsdb %%v17, %%v25, %%v28, %%v17  \n\t"
+            "vfmadb %%v18, %%v27, %%v29, %%v18  \n\t"
+            "vfmsdb %%v19, %%v27, %%v28, %%v19  \n\t"
+            "vfmadb %%v16, %%v24, %%v28, %%v16  \n\t"
+            "vfmsdb %%v17, %%v24, %%v29, %%v17  \n\t"
+            "vfmadb %%v18, %%v26, %%v28, %%v18  \n\t"
+            "vfmsdb %%v19, %%v26, %%v29, %%v19  \n\t"
+
+#endif 
+            "vsteg %%v16 ,  0(%%r1,%2),0  \n\t"
+            "vsteg %%v17 ,  8(%%r1,%2),0  \n\t"
+            "vsteg %%v16 , 16(%%r1,%2),1  \n\t"
+            "vsteg %%v17 , 24(%%r1,%2),1  \n\t"
+
+            "vsteg %%v18 , 32(%%r1,%2),0  \n\t"
+            "vsteg %%v19 , 40(%%r1,%2),0  \n\t"
+            "vsteg %%v18 , 48(%%r1,%2),1  \n\t"
+            "vsteg %%v19 , 56(%%r1,%2),1  \n\t"
+
+            "vleg %%v20 , 64(%%r1,%2),0   \n\t"
+            "vleg %%v21 , 72(%%r1,%2),0   \n\t"
+            "vleg %%v20 , 80(%%r1,%2),1   \n\t"
+            "vleg %%v21 , 88(%%r1,%2),1   \n\t"
+
+            "vleg %%v22 ,  96(%%r1,%2),0  \n\t"
+            "vleg %%v23 , 104(%%r1,%2),0  \n\t"
+            "vleg %%v22 , 112(%%r1,%2),1  \n\t"
+            "vleg %%v23 , 120(%%r1,%2),1  \n\t"
+
+            "vleg %%v24 , 64(%%r1,%1),0   \n\t"
+            "vleg %%v25 , 72(%%r1,%1),0   \n\t"
+            "vleg %%v24 , 80(%%r1,%1),1   \n\t"
+            "vleg %%v25 , 88(%%r1,%1),1   \n\t"
+
+            "vleg %%v26 ,  96(%%r1,%1),0  \n\t"
+            "vleg %%v27 , 104(%%r1,%1),0  \n\t"
+            "vleg %%v26 , 112(%%r1,%1),1  \n\t"
+            "vleg %%v27 , 120(%%r1,%1),1  \n\t"
+#if !defined(CONJ)
+            "vfmsdb %%v20,  %%v25, %%v29,%%v20   \n\t"
+            "vfmadb %%v21,  %%v24, %%v29, %%v21  \n\t"
+            "vfmsdb %%v22,  %%v27, %%v29, %%v22  \n\t"
+            "vfmadb %%v23,  %%v26, %%v29, %%v23  \n\t"
+
+            "vfmsdb %%v20, %%v24, %%v28 ,%%v20   \n\t"
+            "vfmadb %%v21, %%v25, %%v28, %%v21   \n\t"
+            "vfmsdb %%v22, %%v26, %%v28, %%v22   \n\t"
+            "vfmadb %%v23, %%v27, %%v28, %%v23   \n\t"
+#else
+            "vfmadb %%v20, %%v25, %%v29, %%v20   \n\t"
+            "vfmsdb %%v21, %%v25, %%v28, %%v21   \n\t"
+            "vfmadb %%v22, %%v27, %%v29, %%v22   \n\t"
+            "vfmsdb %%v23, %%v27, %%v28, %%v23   \n\t"
+            "vfmadb %%v20, %%v24, %%v28, %%v20   \n\t"
+            "vfmsdb %%v21, %%v24, %%v29, %%v21   \n\t"
+            "vfmadb %%v22, %%v26, %%v28, %%v22   \n\t"
+            "vfmsdb %%v23, %%v26, %%v29, %%v23   \n\t"
+#endif 
+            "vsteg %%v20 , 64(%%r1,%2),0   \n\t"
+            "vsteg %%v21 , 72(%%r1,%2),0   \n\t"
+            "vsteg %%v20 , 80(%%r1,%2),1   \n\t"
+            "vsteg %%v21 , 88(%%r1,%2),1   \n\t"
+
+            "vsteg %%v22 ,  96(%%r1,%2),0  \n\t"
+            "vsteg %%v23 , 104(%%r1,%2),0  \n\t"
+            "vsteg %%v22 , 112(%%r1,%2),1  \n\t"
+            "vsteg %%v23 , 120(%%r1,%2),1  \n\t"
+
+            "la %%r1,128(%%r1) \n\t"
+            "brctg %3,1b"
+            :
+            : "r"(n), "a"(x), "a"(y), "a"(alpha)
+            : "cc", "memory", "r1","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29"
+            );
+
+}
+
+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;
+    BLASLONG ix = 0, iy = 0;
+    FLOAT da[2];
+
+    if (n <= 0) return (0);
+
+    if ((inc_x == 1) && (inc_y == 1)) {
+
+        BLASLONG n1 = n & -8;
+
+        if (n1) {
+            da[0] = da_r;
+            da[1] = da_i;
+            zaxpy_kernel_8(n1, x, y, da);
+            ix = 2 * n1;
+        }
+        i = n1;
+        while (i < n) {
+#if !defined(CONJ)
+            y[ix] += (da_r * x[ix] - da_i * x[ix + 1]);
+            y[ix + 1] += (da_r * x[ix + 1] + da_i * x[ix]);
+#else
+            y[ix] += (da_r * x[ix] + da_i * x[ix + 1]);
+            y[ix + 1] -= (da_r * x[ix + 1] - da_i * x[ix]);
+#endif
+            i++;
+            ix += 2;
+
+        }
+        return (0);
+
+
+    }
+
+    inc_x *= 2;
+    inc_y *= 2;
+
+    while (i < n) {
+
+#if !defined(CONJ)
+        y[iy] += (da_r * x[ix] - da_i * x[ix + 1]);
+        y[iy + 1] += (da_r * x[ix + 1] + da_i * x[ix]);
+#else
+        y[iy] += (da_r * x[ix] + da_i * x[ix + 1]);
+        y[iy + 1] -= (da_r * x[ix + 1] - da_i * x[ix]);
+#endif
+        ix += inc_x;
+        iy += inc_y;
+        i++;
+
+    }
+    return (0);
+
+}
+
+

kernel/zarch/zcopy.c

@@ -0,0 +1,145 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+ 
+static void __attribute__ ((noinline)) zcopy_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y) {
+
+    __asm__ volatile(
+            "pfd 1, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+            "srlg %%r0,%0,4      \n\t"
+            "xgr  %%r1,%%r1       \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+
+            "vl  %%v24,  0(%%r1,%1) \n\t"
+            "vst %%v24,  0(%%r1,%2) \n\t"
+            "vl  %%v25, 16(%%r1,%1) \n\t"
+            "vst %%v25, 16(%%r1,%2) \n\t"
+            "vl  %%v26, 32(%%r1,%1) \n\t"
+            "vst %%v26, 32(%%r1,%2) \n\t"
+            "vl  %%v27, 48(%%r1,%1) \n\t"
+            "vst %%v27, 48(%%r1,%2) \n\t"
+
+            "vl  %%v28, 64(%%r1,%1) \n\t"
+            "vst %%v28, 64(%%r1,%2) \n\t"
+            "vl  %%v29, 80(%%r1,%1) \n\t"
+            "vst %%v29, 80(%%r1,%2) \n\t"
+            "vl  %%v30, 96(%%r1,%1) \n\t"
+            "vst %%v30, 96(%%r1,%2) \n\t"
+            "vl  %%v31,112(%%r1,%1) \n\t"
+            "vst %%v31,112(%%r1,%2) \n\t"
+
+
+            "vl  %%v24,128(%%r1,%1) \n\t"
+            "vst %%v24,128(%%r1,%2) \n\t"
+
+            "vl  %%v25,144(%%r1,%1) \n\t"
+            "vst %%v25,144(%%r1,%2) \n\t"
+
+            "vl  %%v26,160(%%r1,%1) \n\t"
+            "vst %%v26,160(%%r1,%2) \n\t"
+
+            "vl  %%v27,176(%%r1,%1) \n\t"
+            "vst %%v27,176(%%r1,%2) \n\t"
+
+            "vl  %%v28, 192(%%r1,%1) \n\t"
+            "vst %%v28, 192(%%r1,%2) \n\t"
+            "vl  %%v29, 208(%%r1,%1) \n\t"
+            "vst %%v29, 208(%%r1,%2) \n\t"
+            "vl  %%v30, 224(%%r1,%1) \n\t"
+            "vst %%v30, 224(%%r1,%2) \n\t"
+            "vl  %%v31, 240(%%r1,%1) \n\t"
+            "vst %%v31, 240(%%r1,%2) \n\t"
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y)
+            : "cc", "memory","r0","r1","v24","v25","v26","v27","v28","v29","v30","v31" 
+            );
+    return;
+
+}
+
+
+int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
+{
+    BLASLONG i=0;
+    BLASLONG ix=0,iy=0;
+
+    if ( n <= 0     )  return(0);
+
+    if ( (inc_x == 1) && (inc_y == 1 ))
+    {
+
+        BLASLONG n1 = n & -16;
+        if ( n1 > 0 )
+        {
+            zcopy_kernel_16(n1, x, y);
+            i=n1;
+            ix=n1*2;
+            iy=n1*2;
+        }
+
+        while(i < n)
+        {
+            y[iy] = x[iy] ;
+            y[iy+1] = x[ix+1] ;
+            ix+=2;
+            iy+=2;
+            i++ ;
+
+        }
+
+
+    }
+    else
+    {
+
+        BLASLONG inc_x2 = 2 * inc_x;
+        BLASLONG inc_y2 = 2 * inc_y;
+
+        while(i < n)
+        {
+            y[iy] = x[ix] ;
+            y[iy+1] = x[ix+1] ;
+            ix += inc_x2 ;
+            iy += inc_y2 ;
+            i++ ;
+
+        }
+
+    }
+    return(0);
+    
+
+}
+
+

kernel/zarch/zdot.c

@@ -0,0 +1,216 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+
+static void __attribute__ ((noinline)) zdot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d) {
+
+    __asm__ volatile(
+            "pfd 1, 0(%1) \n\t"
+            "pfd 1, 0(%2) \n\t"
+            "vzero %%v24  \n\t"
+            "vzero %%v25  \n\t"
+            "vzero %%v26  \n\t"
+            "vzero %%v27  \n\t"
+            "srlg %%r0,%0,3      \n\t"
+            "xgr %%r1,%%r1       \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 1, 256(%%r1,%1)     \n\t"
+            "pfd 1, 256(%%r1,%2)     \n\t"
+            "vl  %%v16,  0(%%r1,%1)  \n\t"
+            "vl  %%v17, 16(%%r1,%1)  \n\t"
+            "vl  %%v18, 32(%%r1,%1)  \n\t"
+            "vl  %%v19, 48(%%r1,%1)  \n\t"
+            "vl  %%v28,  0(%%r1,%2)  \n\t"
+            "vl  %%v29, 16(%%r1,%2)  \n\t"
+            "vl  %%v30, 32(%%r1,%2)  \n\t"
+            "vl  %%v31, 48(%%r1,%2)  \n\t"
+            "vpdi %%v20,%%v16,%%v16,4 \n\t"
+            "vpdi %%v21,%%v17,%%v17,4 \n\t"
+            "vpdi %%v22,%%v18,%%v18,4 \n\t"
+            "vpdi %%v23,%%v19,%%v19,4 \n\t"
+
+
+            "vfmadb    %%v24,%%v16,%%v28,%%v24  \n\t"
+            "vfmadb    %%v25,%%v20,%%v28,%%v25  \n\t"
+            "vfmadb    %%v26,%%v17,%%v29,%%v26  \n\t"
+            "vfmadb    %%v27,%%v21,%%v29,%%v27  \n\t"
+            "vfmadb    %%v24,%%v18,%%v30,%%v24  \n\t"
+            "vfmadb    %%v25,%%v22,%%v30,%%v25  \n\t"
+            "vfmadb    %%v26,%%v19,%%v31,%%v26  \n\t"
+            "vfmadb    %%v27,%%v23,%%v31,%%v27  \n\t"
+
+
+
+            "vl  %%v16, 64(%%r1,%1) \n\t"
+            "vl  %%v17, 80(%%r1,%1) \n\t"
+            "vl  %%v18, 96(%%r1,%1) \n\t"
+            "vl  %%v19,112(%%r1,%1) \n\t"
+            "vl  %%v28, 64(%%r1,%2) \n\t"
+            "vl  %%v29, 80(%%r1,%2) \n\t"
+            "vl  %%v30, 96(%%r1,%2) \n\t"
+            "vl  %%v31,112(%%r1,%2) \n\t"
+            "vpdi %%v20,%%v16,%%v16,4 \n\t"
+            "vpdi %%v21,%%v17,%%v17,4 \n\t"
+            "vpdi %%v22,%%v18,%%v18,4 \n\t"
+            "vpdi %%v23,%%v19,%%v19,4 \n\t"
+            "vfmadb    %%v24,%%v16,%%v28,%%v24  \n\t"
+            "vfmadb    %%v25,%%v20,%%v28,%%v25  \n\t"
+            "vfmadb    %%v26,%%v17,%%v29,%%v26  \n\t"
+            "vfmadb    %%v27,%%v21,%%v29,%%v27  \n\t"
+            "vfmadb    %%v24,%%v18,%%v30,%%v24  \n\t"
+            "vfmadb    %%v25,%%v22,%%v30,%%v25  \n\t"
+            "vfmadb    %%v26,%%v19,%%v31,%%v26  \n\t"
+            "vfmadb    %%v27,%%v23,%%v31,%%v27  \n\t"
+
+
+            "la %%r1,128(%%r1) \n\t"
+            "brctg %%r0,1b     \n\t"
+            "vfadb %%v24,%%v26,%%v24 \n\t"
+            "vfadb %%v25,%%v25,%%v27 \n\t"
+            "vsteg %%v24,0(%3),0     \n\t"
+            "vsteg %%v24,8(%3),1     \n\t"
+            "vsteg %%v25,16(%3),1    \n\t"
+            "vsteg %%v25,24(%3),0    \n\t"
+            :
+            : "r"(n), "a"(x), "a"(y), "a"(d)
+            : "cc", "memory","r0","r1","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31" 
+            );
+
+}
+
+static __attribute__ ((noinline)) void zdot_kernel_8n(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d) {
+    BLASLONG register i = 0;
+    FLOAT dot[4] = {0.0, 0.0, 0.0, 0.0};
+    BLASLONG j = 0;
+
+    while (i < n) {
+
+        dot[0] += x[j] * y[j];
+        dot[1] += x[j + 1] * y[j + 1];
+        dot[2] += x[j] * y[j + 1];
+        dot[3] += x[j + 1] * y[j];
+
+        dot[0] += x[j + 2] * y[j + 2];
+        dot[1] += x[j + 3] * y[j + 3];
+        dot[2] += x[j + 2] * y[j + 3];
+        dot[3] += x[j + 3] * y[j + 2];
+
+        dot[0] += x[j + 4] * y[j + 4];
+        dot[1] += x[j + 5] * y[j + 5];
+        dot[2] += x[j + 4] * y[j + 5];
+        dot[3] += x[j + 5] * y[j + 4];
+
+        dot[0] += x[j + 6] * y[j + 6];
+        dot[1] += x[j + 7] * y[j + 7];
+        dot[2] += x[j + 6] * y[j + 7];
+        dot[3] += x[j + 7] * y[j + 6];
+
+        j += 8;
+        i += 4;
+
+    }
+    d[0] = dot[0];
+    d[1] = dot[1];
+    d[2] = dot[2];
+    d[3] = dot[3];
+
+}
+
+OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y) {
+    BLASLONG i;
+    BLASLONG ix, iy;
+    OPENBLAS_COMPLEX_FLOAT result;
+    FLOAT dot[4] __attribute__ ((aligned(16))) = {0.0, 0.0, 0.0, 0.0};
+
+    if (n <= 0) {
+        CREAL(result) = 0.0;
+        CIMAG(result) = 0.0;
+        return (result);
+
+    }
+
+    if ((inc_x == 1) && (inc_y == 1)) {
+
+        BLASLONG n1 = n & -16;
+
+        if (n1)
+            zdot_kernel_8(n1, x, y, dot);
+
+        i = n1;
+        BLASLONG j = i * 2;
+
+        while (i < n) {
+
+            dot[0] += x[j] * y[j];
+            dot[1] += x[j + 1] * y[j + 1];
+            dot[2] += x[j] * y[j + 1];
+            dot[3] += x[j + 1] * y[j];
+
+            j += 2;
+            i++;
+
+        }
+
+
+    } else {
+        i = 0;
+        ix = 0;
+        iy = 0;
+        inc_x <<= 1;
+        inc_y <<= 1;
+        while (i < n) {
+
+            dot[0] += x[ix] * y[iy];
+            dot[1] += x[ix + 1] * y[iy + 1];
+            dot[2] += x[ix] * y[iy + 1];
+            dot[3] += x[ix + 1] * y[iy];
+
+            ix += inc_x;
+            iy += inc_y;
+            i++;
+
+        }
+    }
+
+#if !defined(CONJ)
+    CREAL(result) = dot[0] - dot[1];
+    CIMAG(result) = dot[2] + dot[3];
+#else
+    CREAL(result) = dot[0] + dot[1];
+    CIMAG(result) = dot[2] - dot[3];
+
+#endif
+
+    return (result);
+
+}
+
+

kernel/zarch/zgemv_n_4.c

@@ -0,0 +1,919 @@
+/***************************************************************************
+Copyright (c) 2017, 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 <stdlib.h>
+#include <stdio.h>
+#include "common.h"
+
+ 
+#define HAVE_KERNEL_4x4_VEC 1
+#define HAVE_KERNEL_4x2_VEC 1
+#define HAVE_KERNEL_4x1_VEC 1
+#define HAVE_KERNEL_ADDY 1
+
+#if defined(HAVE_KERNEL_4x4_VEC) || defined(HAVE_KERNEL_4x2_VEC) || defined(HAVE_KERNEL_4x1_VEC)
+ #include <vecintrin.h> 
+#endif
+
+/**
+ * if define IGNORE_TEMP_PERM we store and use ybuffer as {real,real} {img;img}
+ * of not we will retrieve and store normal way
+ */
+#if (defined(HAVE_KERNEL_4x4_VEC_ASM) || defined(HAVE_KERNEL_4x4_VEC) ) && defined(HAVE_KERNEL_4x2_VEC) && defined(HAVE_KERNEL_4x1_VEC) && defined(HAVE_KERNEL_ADDY) 
+ // #define  IGNORE_TEMP_PERM 1
+#endif   
+ 
+#define NBMAX 1024
+
+#ifdef HAVE_KERNEL_4x4_VEC_ASM
+
+#elif HAVE_KERNEL_4x4_VEC
+static void zgemv_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];
+        
+    register __vector double  vx0_r = {x[0],x[0]};
+    register __vector double  vx0_i = {x[1],x[1]}; 
+    register __vector double  vx1_r = {x[2],x[2]};
+    register __vector double  vx1_i = {x[3],x[3]}; 
+    register __vector double  vx2_r = {x[4],x[4]};
+    register __vector double  vx2_i = {x[5],x[5]}; 
+    register __vector double  vx3_r = {x[6],x[6]};
+    register __vector double  vx3_i = {x[7],x[7]}; 
+
+#ifdef IGNORE_TEMP_PERM 
+    register __vector double *vy = (__vector double *)y;
+    register BLASLONG j=0;
+#endif        
+
+    for ( i=0; i< 2*n; i+=4 )
+    {
+            
+#ifdef IGNORE_TEMP_PERM 
+        register __vector double  vresult_r = vy[j];
+        register __vector double  vresult_i = vy[j+1];
+
+#else            
+        register __vector double  vresult_r = {y[i],y[i+2]};
+        register __vector double  vresult_i = {y[i+1],y[i+3]};
+#endif                
+        register __vector double  va0_r= {a0[i],a0[i+2]};
+        register __vector double  va0_i= {a0[i+1],a0[i+3]};
+        register __vector double  va1_r= {a1[i],a1[i+2]};
+        register __vector double  va1_i= {a1[i+1],a1[i+3]};    
+        register __vector double  va2_r= {a2[i],a2[i+2]};
+        register __vector double  va2_i= {a2[i+1],a2[i+3]};
+        register __vector double  va3_r= {a3[i],a3[i+2]};
+        register __vector double  va3_i= {a3[i+1],a3[i+3]};                 
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) ) 
+                
+        vresult_r = va0_r * vx0_r - (va0_i*vx0_i -vresult_r) ;
+        vresult_i = vresult_i + va0_r * vx0_i  +  va0_i * vx0_r ;
+        vresult_r = va1_r * vx1_r - (va1_i*vx1_i -vresult_r) ;
+        vresult_i = vresult_i + va1_r * vx1_i  +  va1_i * vx1_r ;
+        vresult_r = va2_r * vx2_r - (va2_i*vx2_i -vresult_r) ;
+        vresult_i = vresult_i + va2_r * vx2_i  +  va2_i * vx2_r ;
+        vresult_r = va3_r * vx3_r - (va3_i*vx3_i -vresult_r) ;
+        vresult_i = vresult_i + va3_r * vx3_i  +  va3_i * vx3_r ;                  
+        
+#else  
+        vresult_r = vresult_r + va0_r * vx0_r +  va0_i*vx0_i   ; 
+        vresult_i =  va0_r * vx0_i  -  ( va0_i * vx0_r  - vresult_i) ;      
+        vresult_r = vresult_r + va1_r * vx1_r +  va1_i*vx1_i   ; 
+        vresult_i =  va1_r * vx1_i  -  ( va1_i * vx1_r  - vresult_i) ; 
+        vresult_r = vresult_r + va2_r * vx2_r +  va2_i*vx2_i   ; 
+        vresult_i =  va2_r * vx2_i  -  ( va2_i * vx2_r  - vresult_i) ;      
+        vresult_r = vresult_r + va3_r * vx3_r +  va3_i*vx3_i   ; 
+        vresult_i =  va3_r * vx3_i  -  ( va3_i * vx3_r  - vresult_i) ;                
+#endif
+
+#ifdef IGNORE_TEMP_PERM 
+        vy[j] =   vresult_r ; 
+        vy[j+1] = vresult_i ;
+        j+=2;
+
+#else
+        y[i]       =  vresult_r[0];
+        y[i+1]     =  vresult_i[0];
+        y[i +2 ]   =  vresult_r[1];
+        y[i + 3 ]  =  vresult_i[1];
+#endif                
+ 
+    }        
+ 
+}
+    
+#else
+static void zgemv_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];
+
+    for ( i=0; i< 2*n; i+=2 )
+    {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+        y[i]   += a0[i]*x[0] - a0[i+1] * x[1];
+        y[i+1] += a0[i]*x[1] + a0[i+1] * x[0];
+        y[i]   += a1[i]*x[2] - a1[i+1] * x[3];
+        y[i+1] += a1[i]*x[3] + a1[i+1] * x[2];
+        y[i]   += a2[i]*x[4] - a2[i+1] * x[5];
+        y[i+1] += a2[i]*x[5] + a2[i+1] * x[4];
+        y[i]   += a3[i]*x[6] - a3[i+1] * x[7];
+        y[i+1] += a3[i]*x[7] + a3[i+1] * x[6];
+#else 
+        y[i]   += a0[i]*x[0] + a0[i+1] * x[1];
+        y[i+1] += a0[i]*x[1] - a0[i+1] * x[0];
+        y[i]   += a1[i]*x[2] + a1[i+1] * x[3];
+        y[i+1] += a1[i]*x[3] - a1[i+1] * x[2];
+        y[i]   += a2[i]*x[4] + a2[i+1] * x[5];
+        y[i+1] += a2[i]*x[5] - a2[i+1] * x[4];
+        y[i]   += a3[i]*x[6] + a3[i+1] * x[7];
+        y[i+1] += a3[i]*x[7] - a3[i+1] * x[6];
+#endif
+    }
+}
+    
+#endif
+
+
+
+#ifdef  HAVE_KERNEL_4x2_VEC
+
+static void zgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
+{
+    BLASLONG i;
+    FLOAT *a0,*a1;
+    a0 = ap[0];
+    a1 = ap[1];
+         
+    register __vector double  vx0_r = {x[0],x[0]};
+    register __vector double  vx0_i = {x[1],x[1]}; 
+    register __vector double  vx1_r = {x[2],x[2]};
+    register __vector double  vx1_i = {x[3],x[3]}; 
+#ifdef IGNORE_TEMP_PERM 
+    register __vector double *vy = (__vector double *)y;
+    register BLASLONG j=0;
+#endif        
+
+    for ( i=0; i< 2*n; i+=4 )
+    {
+#ifdef IGNORE_TEMP_PERM 
+        register __vector double  vresult_r = vy[j];
+        register __vector double  vresult_i = vy[j+1];
+
+#else            
+        register __vector double  vresult_r = {y[i],y[i+2]};
+        register __vector double  vresult_i = {y[i+1],y[i+3]};
+#endif      
+        register __vector double  va0_r= {a0[i],a0[i+2]};
+        register __vector double  va0_i= {a0[i+1],a0[i+3]};
+        register __vector double  va1_r= {a1[i],a1[i+2]};
+        register __vector double  va1_i= {a1[i+1],a1[i+3]};             
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) ) 
+                
+        vresult_r = va0_r * vx0_r - (va0_i*vx0_i -vresult_r) ;
+        vresult_i = vresult_i + va0_r * vx0_i  +  va0_i * vx0_r ;
+        vresult_r = va1_r * vx1_r - (va1_i*vx1_i -vresult_r) ;
+        vresult_i = vresult_i + va1_r * vx1_i  +  va1_i * vx1_r ;      
+        
+#else  
+        vresult_r = vresult_r + va0_r * vx0_r +  va0_i*vx0_i   ; 
+        vresult_i =  va0_r * vx0_i  -  ( va0_i * vx0_r  - vresult_i) ;      
+        vresult_r = vresult_r + va1_r * vx1_r +  va1_i*vx1_i   ; 
+        vresult_i =  va1_r * vx1_i  -  ( va1_i * vx1_r  - vresult_i) ;                 
+#endif
+                
+#ifdef IGNORE_TEMP_PERM 
+        vy[j] =   vresult_r ; 
+        vy[j+1] = vresult_i ;
+        j+=2;
+
+#else
+        y[i]       =  vresult_r[0];
+        y[i+1]     =  vresult_i[0];
+        y[i +2 ]   =  vresult_r[1];
+        y[i + 3 ]  =  vresult_i[1];
+#endif     
+ 
+    }
+}
+
+#else
+static void zgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
+{
+    BLASLONG i;
+    FLOAT *a0,*a1;
+    a0 = ap[0];
+    a1 = ap[1];
+
+    for ( i=0; i< 2*n; i+=2 )
+    {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+        y[i]   += a0[i]*x[0] - a0[i+1] * x[1];
+        y[i+1] += a0[i]*x[1] + a0[i+1] * x[0];
+        y[i]   += a1[i]*x[2] - a1[i+1] * x[3];
+        y[i+1] += a1[i]*x[3] + a1[i+1] * x[2];
+#else 
+        y[i]   += a0[i]*x[0] + a0[i+1] * x[1];
+        y[i+1] += a0[i]*x[1] - a0[i+1] * x[0];
+        y[i]   += a1[i]*x[2] + a1[i+1] * x[3];
+        y[i+1] += a1[i]*x[3] - a1[i+1] * x[2];
+#endif
+    }
+}
+    
+#endif
+
+
+
+
+#ifdef  HAVE_KERNEL_4x1_VEC
+
+static void zgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
+{
+    BLASLONG i;
+    FLOAT *a0;
+    a0 = ap;
+        
+    register __vector double  vx_r = {x[0],x[0]};
+    register __vector double  vx_i = {x[1],x[1]}; 
+
+#ifdef IGNORE_TEMP_PERM 
+    register __vector double *vy = (__vector double *)y;
+    register BLASLONG j=0;
+#endif                
+
+    for ( i=0; i< 2*n; i+=4 )
+    {
+#ifdef IGNORE_TEMP_PERM 
+        register __vector double  vresult_r = vy[j];
+        register __vector double  vresult_i = vy[j+1];
+
+#else            
+        register __vector double  vresult_r = {y[i],y[i+2]};
+        register __vector double  vresult_i = {y[i+1],y[i+3]};
+#endif      
+        register __vector double  va0_r= {a0[i],a0[i+2]};
+        register __vector double  va0_i= {a0[i+1],a0[i+3]};            
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+
+        vresult_r = va0_r * vx_r - (va0_i*vx_i -vresult_r) ;
+        vresult_i = vresult_i + va0_r * vx_i  +  va0_i * vx_r ;     
+    
+#else  
+        vresult_r = vresult_r + va0_r * vx_r +  va0_i*vx_i   ; 
+        vresult_i =  va0_r * vx_i  -  ( va0_i * vx_r  - vresult_i) ;  
+                
+//        y[i]   += a0[i]*x[0] + a0[i+1] * x[1];
+//        y[i+1] += a0[i]*x[1] - a0[i+1] * x[0];
+#endif
+#ifndef IGNORE_TEMP_PERM 
+        y[i]       =  vresult_r[0];
+        y[i+1]     =  vresult_i[0];
+        y[i +2 ]   =  vresult_r[1];
+        y[i + 3 ]  =  vresult_i[1];
+#else
+        vy[j]   =   vresult_r ; 
+        vy[j+1] =   vresult_i ;
+        j+=2;
+#endif        
+
+    }
+}
+
+#else
+
+static void zgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
+{
+    BLASLONG i;
+    FLOAT *a0;
+    a0 = ap;
+
+    for ( i=0; i< 2*n; i+=2 )
+    {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+        y[i]   += a0[i]*x[0] - a0[i+1] * x[1];
+        y[i+1] += a0[i]*x[1] + a0[i+1] * x[0];
+#else 
+        y[i]   += a0[i]*x[0] + a0[i+1] * x[1];
+        y[i+1] += a0[i]*x[1] - a0[i+1] * x[0];
+#endif
+
+    }
+}
+
+
+#endif
+
+
+#ifdef HAVE_KERNEL_ADDY
+static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest,FLOAT alpha_r, FLOAT alpha_i)
+{
+    BLASLONG i;
+#ifdef IGNORE_TEMP_PERM 
+    register __vector double *src_vec = (__vector double *)src;
+#endif       
+    register __vector double  valpha_r = {alpha_r,alpha_r};
+    register __vector double  valpha_i = {alpha_i,alpha_i}; 
+    register __vector double vresult_r;
+    register __vector double  vresult_i;                
+    if ( inc_dest != 2 )
+    {
+ 
+                
+        for ( i=0; i<n; i+=2 )
+        {
+                    
+#ifdef IGNORE_TEMP_PERM 
+            register __vector double  vsrc_r= src_vec[i];
+            register __vector double  vsrc_i= src_vec[i+1]; 
+#else
+            register __vector double  vsrc_r= {src[0],src[2]};
+            register __vector double  vsrc_i= {src[1],src[3]};                
+                
+#endif      
+            
+#if !defined(XCONJ) 
+
+            vresult_r = vsrc_r * valpha_r;
+            vresult_r-=  vsrc_i*valpha_i   ;
+            vresult_i = vsrc_r * valpha_i  +  vsrc_i * valpha_r ;     
+    
+#else  
+            vresult_r =  vsrc_r * valpha_r +  vsrc_i*valpha_i   ; 
+            vresult_i =  vsrc_r * valpha_i ;
+            vresult_i -=   vsrc_i * valpha_r    ; 
+#endif 
+
+            *dest += vresult_r[0];
+            *(dest+1) += vresult_i[0];
+            *(dest + inc_dest) += vresult_r[1];
+            *(dest+inc_dest+1) += vresult_i[1];                        
+                         
+#ifndef IGNORE_TEMP_PERM                         
+            src+=4;
+#endif                        
+            dest += 2*inc_dest;
+        }
+        return;
+    }
+
+        for ( i=0; i<n; i+=2 )
+        {
+                    
+
+#ifdef IGNORE_TEMP_PERM 
+            register __vector double  vsrc_r= src_vec[i];
+            register __vector double  vsrc_i= src_vec[i+1]; 
+#else
+            register __vector double  vsrc_r= {src[0],src[2]};
+            register __vector double  vsrc_i= {src[1],src[3]};                
+                
+#endif           
+#if !defined(XCONJ) 
+
+            vresult_r = vsrc_r * valpha_r;
+            vresult_r-=  vsrc_i*valpha_i   ;
+            vresult_i = vsrc_r * valpha_i  +  vsrc_i * valpha_r ;     
+    
+#else  
+            vresult_r =  vsrc_r * valpha_r +  vsrc_i*valpha_i   ; 
+            vresult_i =  vsrc_r * valpha_i ;
+            vresult_i -=   vsrc_i * valpha_r    ; 
+#endif 
+
+            *dest += vresult_r[0];
+            *(dest+1) += vresult_i[0];
+            *(dest + 2) += vresult_r[1];
+            *(dest+3) += vresult_i[1];                        
+                         
+                        
+#ifndef IGNORE_TEMP_PERM                         
+            src+=4;
+#endif    
+            dest += 4;
+        }
+        return;
+ 
+    return;
+
+}
+
+#else
+static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest,FLOAT alpha_r, FLOAT alpha_i)
+{
+    BLASLONG i;
+
+    if ( inc_dest != 2 )
+    {
+
+        FLOAT temp_r;
+        FLOAT temp_i;
+        for ( i=0; i<n; i++ )
+        {
+#if !defined(XCONJ) 
+            temp_r = alpha_r * src[0] - alpha_i * src[1];
+            temp_i = alpha_r * src[1] + alpha_i * src[0];
+#else
+            temp_r =  alpha_r * src[0] + alpha_i * src[1];
+            temp_i = -alpha_r * src[1] + alpha_i * src[0];
+#endif
+
+            *dest += temp_r;
+            *(dest+1) += temp_i;
+
+            src+=2;
+            dest += inc_dest;
+        }
+        return;
+    }
+
+    FLOAT temp_r0;
+    FLOAT temp_i0;
+    FLOAT temp_r1;
+    FLOAT temp_i1;
+    FLOAT temp_r2;
+    FLOAT temp_i2;
+    FLOAT temp_r3;
+    FLOAT temp_i3;
+    for ( i=0; i<n; i+=4 )
+    {
+#if !defined(XCONJ) 
+        temp_r0 = alpha_r * src[0] - alpha_i * src[1];
+        temp_i0 = alpha_r * src[1] + alpha_i * src[0];
+        temp_r1 = alpha_r * src[2] - alpha_i * src[3];
+        temp_i1 = alpha_r * src[3] + alpha_i * src[2];
+        temp_r2 = alpha_r * src[4] - alpha_i * src[5];
+        temp_i2 = alpha_r * src[5] + alpha_i * src[4];
+        temp_r3 = alpha_r * src[6] - alpha_i * src[7];
+        temp_i3 = alpha_r * src[7] + alpha_i * src[6];
+#else
+        temp_r0 =  alpha_r * src[0] + alpha_i * src[1];
+        temp_i0 = -alpha_r * src[1] + alpha_i * src[0];
+        temp_r1 =  alpha_r * src[2] + alpha_i * src[3];
+        temp_i1 = -alpha_r * src[3] + alpha_i * src[2];
+        temp_r2 =  alpha_r * src[4] + alpha_i * src[5];
+        temp_i2 = -alpha_r * src[5] + alpha_i * src[4];
+        temp_r3 =  alpha_r * src[6] + alpha_i * src[7];
+        temp_i3 = -alpha_r * src[7] + alpha_i * src[6];
+#endif
+
+        dest[0]   += temp_r0;
+        dest[1]   += temp_i0;
+        dest[2]   += temp_r1;
+        dest[3]   += temp_i1;
+        dest[4]   += temp_r2;
+        dest[5]   += temp_i2;
+        dest[6]   += temp_r3;
+        dest[7]   += temp_i3;
+
+        src  += 8;
+        dest += 8;
+    }
+    return;
+
+}
+
+#endif
+
+int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r,FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
+{
+    BLASLONG i;
+    BLASLONG j;
+    FLOAT *a_ptr;
+    FLOAT *x_ptr;
+    FLOAT *y_ptr;
+    FLOAT *ap[4];
+    BLASLONG n1;
+    BLASLONG m1;
+    BLASLONG m2;
+    BLASLONG m3;
+    BLASLONG n2;
+    BLASLONG lda4;
+    FLOAT xbuffer[8],*ybuffer;
+
+
+ 
+
+    if ( m < 1 ) return(0);
+    if ( n < 1 ) return(0);
+
+    ybuffer = buffer;
+    
+    inc_x *= 2;
+    inc_y *= 2;
+    lda   *= 2;
+    lda4  = 4 * lda;
+
+    n1 = n / 4 ;
+    n2 = n % 4 ;
+    
+    m3 = m % 4;
+    m1 = m - ( m % 4 );
+    m2 = (m % NBMAX) - (m % 4) ;
+    
+    y_ptr = y;
+
+    BLASLONG NB = NBMAX;
+
+    while ( NB == NBMAX )
+    {
+        
+        m1 -= NB;
+        if ( m1 < 0)
+        {
+            if ( m2 == 0 ) break;    
+            NB = m2;
+        }
+        
+        a_ptr = a;
+        ap[0] = a_ptr;
+        ap[1] = a_ptr + lda;
+        ap[2] = ap[1] + lda;
+        ap[3] = ap[2] + lda;
+        x_ptr = x;
+        //zero_y(NB,ybuffer);
+        memset(ybuffer,0,NB*16);
+
+        if ( inc_x == 2 )
+        {
+
+            for( i = 0; i < n1 ; i++)
+            {
+                zgemv_kernel_4x4(NB,ap,x_ptr,ybuffer);
+                ap[0] += lda4;
+                ap[1] += lda4;
+                ap[2] += lda4;
+                ap[3] += lda4;
+                a_ptr += lda4;
+                x_ptr += 8;    
+            }
+
+            if ( n2 & 2 )
+            {
+                zgemv_kernel_4x2(NB,ap,x_ptr,ybuffer);
+                x_ptr += 4;    
+                a_ptr += 2 * lda;
+
+            }
+
+            if ( n2 & 1 )
+            {
+                zgemv_kernel_4x1(NB,a_ptr,x_ptr,ybuffer);
+                x_ptr += 2;    
+                a_ptr += lda;
+
+            }
+        }
+        else
+        {
+
+            for( i = 0; i < n1 ; i++)
+            {
+
+                xbuffer[0] = x_ptr[0];
+                xbuffer[1] = x_ptr[1];
+                x_ptr += inc_x;    
+                xbuffer[2] = x_ptr[0];
+                xbuffer[3] = x_ptr[1];
+                x_ptr += inc_x;    
+                xbuffer[4] = x_ptr[0];
+                xbuffer[5] = x_ptr[1];
+                x_ptr += inc_x;    
+                xbuffer[6] = x_ptr[0];
+                xbuffer[7] = x_ptr[1];
+                x_ptr += inc_x;    
+
+                zgemv_kernel_4x4(NB,ap,xbuffer,ybuffer);
+                ap[0] += lda4;
+                ap[1] += lda4;
+                ap[2] += lda4;
+                ap[3] += lda4;
+                a_ptr += lda4;
+            }
+
+            for( i = 0; i < n2 ; i++)
+            {
+                xbuffer[0] = x_ptr[0];
+                xbuffer[1] = x_ptr[1];
+                x_ptr += inc_x;    
+                zgemv_kernel_4x1(NB,a_ptr,xbuffer,ybuffer);
+                a_ptr += 1 * lda;
+
+            }
+
+        }
+
+        add_y(NB,ybuffer,y_ptr,inc_y,alpha_r,alpha_i);
+        a     += 2 * NB;
+        y_ptr += NB * inc_y;
+    }
+
+    if ( m3 == 0 ) return(0);
+
+    if ( m3 == 1 )
+    {
+        a_ptr = a;
+        x_ptr = x;
+        FLOAT temp_r = 0.0;
+        FLOAT temp_i = 0.0;
+
+        if ( lda == 2 && inc_x == 2 )
+        {
+
+
+            for( i=0 ; i < (n & -2); i+=2 )
+            {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+                temp_r += a_ptr[0] * x_ptr[0] - a_ptr[1] * x_ptr[1];
+                temp_i += a_ptr[0] * x_ptr[1] + a_ptr[1] * x_ptr[0];
+                temp_r += a_ptr[2] * x_ptr[2] - a_ptr[3] * x_ptr[3];
+                temp_i += a_ptr[2] * x_ptr[3] + a_ptr[3] * x_ptr[2];
+#else
+                temp_r += a_ptr[0] * x_ptr[0] + a_ptr[1] * x_ptr[1];
+                temp_i += a_ptr[0] * x_ptr[1] - a_ptr[1] * x_ptr[0];
+                temp_r += a_ptr[2] * x_ptr[2] + a_ptr[3] * x_ptr[3];
+                temp_i += a_ptr[2] * x_ptr[3] - a_ptr[3] * x_ptr[2];
+#endif
+
+                a_ptr += 4;
+                x_ptr += 4;
+            }
+
+
+
+            for( ; i < n; i++ )
+            {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+                temp_r += a_ptr[0] * x_ptr[0] - a_ptr[1] * x_ptr[1];
+                temp_i += a_ptr[0] * x_ptr[1] + a_ptr[1] * x_ptr[0];
+#else
+                temp_r += a_ptr[0] * x_ptr[0] + a_ptr[1] * x_ptr[1];
+                temp_i += a_ptr[0] * x_ptr[1] - a_ptr[1] * x_ptr[0];
+#endif
+
+                a_ptr += 2;
+                x_ptr += 2;
+            }
+
+
+        }
+        else
+        {
+
+            for( i = 0; i < n; i++ )
+            {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+                temp_r += a_ptr[0] * x_ptr[0] - a_ptr[1] * x_ptr[1];
+                temp_i += a_ptr[0] * x_ptr[1] + a_ptr[1] * x_ptr[0];
+#else
+                temp_r += a_ptr[0] * x_ptr[0] + a_ptr[1] * x_ptr[1];
+                temp_i += a_ptr[0] * x_ptr[1] - a_ptr[1] * x_ptr[0];
+#endif
+
+                a_ptr += lda;
+                x_ptr += inc_x;
+            }
+
+        }
+#if !defined(XCONJ) 
+        y_ptr[0] += alpha_r * temp_r - alpha_i * temp_i;
+        y_ptr[1] += alpha_r * temp_i + alpha_i * temp_r;
+#else
+        y_ptr[0] += alpha_r * temp_r + alpha_i * temp_i;
+        y_ptr[1] -= alpha_r * temp_i - alpha_i * temp_r;
+#endif
+        return(0);
+    }
+
+    if ( m3 == 2 )
+    {
+        a_ptr = a;
+        x_ptr = x;
+        FLOAT temp_r0 = 0.0;
+        FLOAT temp_i0 = 0.0;
+        FLOAT temp_r1 = 0.0;
+        FLOAT temp_i1 = 0.0;
+
+        if ( lda == 4 && inc_x == 2 )
+        {
+
+            for( i = 0; i < (n & -2); i+=2 )
+            {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+
+                temp_r0 += a_ptr[0] * x_ptr[0] - a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] + a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] - a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] + a_ptr[3] * x_ptr[0];
+
+                temp_r0 += a_ptr[4] * x_ptr[2] - a_ptr[5] * x_ptr[3];
+                temp_i0 += a_ptr[4] * x_ptr[3] + a_ptr[5] * x_ptr[2];
+                temp_r1 += a_ptr[6] * x_ptr[2] - a_ptr[7] * x_ptr[3];
+                temp_i1 += a_ptr[6] * x_ptr[3] + a_ptr[7] * x_ptr[2];
+
+#else
+                temp_r0 += a_ptr[0] * x_ptr[0] + a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] - a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] + a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] - a_ptr[3] * x_ptr[0];
+
+                temp_r0 += a_ptr[4] * x_ptr[2] + a_ptr[5] * x_ptr[3];
+                temp_i0 += a_ptr[4] * x_ptr[3] - a_ptr[5] * x_ptr[2];
+                temp_r1 += a_ptr[6] * x_ptr[2] + a_ptr[7] * x_ptr[3];
+                temp_i1 += a_ptr[6] * x_ptr[3] - a_ptr[7] * x_ptr[2];
+
+#endif
+
+                a_ptr += 8;
+                x_ptr += 4;
+            }
+
+
+            for( ; i < n; i++ )
+            {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+                temp_r0 += a_ptr[0] * x_ptr[0] - a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] + a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] - a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] + a_ptr[3] * x_ptr[0];
+#else
+                temp_r0 += a_ptr[0] * x_ptr[0] + a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] - a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] + a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] - a_ptr[3] * x_ptr[0];
+#endif
+
+                a_ptr += 4;
+                x_ptr += 2;
+            }
+
+
+        }
+        else
+        {
+
+            for( i=0 ; i < n; i++ )
+            {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+                temp_r0 += a_ptr[0] * x_ptr[0] - a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] + a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] - a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] + a_ptr[3] * x_ptr[0];
+#else
+                temp_r0 += a_ptr[0] * x_ptr[0] + a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] - a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] + a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] - a_ptr[3] * x_ptr[0];
+#endif
+
+                a_ptr += lda;
+                x_ptr += inc_x;
+            }
+
+
+        }
+#if !defined(XCONJ) 
+        y_ptr[0] += alpha_r * temp_r0 - alpha_i * temp_i0;
+        y_ptr[1] += alpha_r * temp_i0 + alpha_i * temp_r0;
+        y_ptr    += inc_y;
+        y_ptr[0] += alpha_r * temp_r1 - alpha_i * temp_i1;
+        y_ptr[1] += alpha_r * temp_i1 + alpha_i * temp_r1;
+#else
+        y_ptr[0] += alpha_r * temp_r0 + alpha_i * temp_i0;
+        y_ptr[1] -= alpha_r * temp_i0 - alpha_i * temp_r0;
+        y_ptr    += inc_y;
+        y_ptr[0] += alpha_r * temp_r1 + alpha_i * temp_i1;
+        y_ptr[1] -= alpha_r * temp_i1 - alpha_i * temp_r1;
+#endif
+        return(0);
+    }
+
+
+    if ( m3 == 3 )
+    {
+        a_ptr = a;
+        x_ptr = x;
+        FLOAT temp_r0 = 0.0;
+        FLOAT temp_i0 = 0.0;
+        FLOAT temp_r1 = 0.0;
+        FLOAT temp_i1 = 0.0;
+        FLOAT temp_r2 = 0.0;
+        FLOAT temp_i2 = 0.0;
+
+        if ( lda == 6 && inc_x == 2 )
+        {
+
+            for( i=0 ; i < n; i++ )
+            {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+                temp_r0 += a_ptr[0] * x_ptr[0] - a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] + a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] - a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] + a_ptr[3] * x_ptr[0];
+                temp_r2 += a_ptr[4] * x_ptr[0] - a_ptr[5] * x_ptr[1];
+                temp_i2 += a_ptr[4] * x_ptr[1] + a_ptr[5] * x_ptr[0];
+#else
+                temp_r0 += a_ptr[0] * x_ptr[0] + a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] - a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] + a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] - a_ptr[3] * x_ptr[0];
+                temp_r2 += a_ptr[4] * x_ptr[0] + a_ptr[5] * x_ptr[1];
+                temp_i2 += a_ptr[4] * x_ptr[1] - a_ptr[5] * x_ptr[0];
+#endif
+
+                a_ptr += 6;
+                x_ptr += 2;
+            }
+
+
+        }
+        else
+        {
+
+            for( i = 0; i < n; i++ )
+            {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+                temp_r0 += a_ptr[0] * x_ptr[0] - a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] + a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] - a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] + a_ptr[3] * x_ptr[0];
+                temp_r2 += a_ptr[4] * x_ptr[0] - a_ptr[5] * x_ptr[1];
+                temp_i2 += a_ptr[4] * x_ptr[1] + a_ptr[5] * x_ptr[0];
+#else
+                temp_r0 += a_ptr[0] * x_ptr[0] + a_ptr[1] * x_ptr[1];
+                temp_i0 += a_ptr[0] * x_ptr[1] - a_ptr[1] * x_ptr[0];
+                temp_r1 += a_ptr[2] * x_ptr[0] + a_ptr[3] * x_ptr[1];
+                temp_i1 += a_ptr[2] * x_ptr[1] - a_ptr[3] * x_ptr[0];
+                temp_r2 += a_ptr[4] * x_ptr[0] + a_ptr[5] * x_ptr[1];
+                temp_i2 += a_ptr[4] * x_ptr[1] - a_ptr[5] * x_ptr[0];
+#endif
+
+                a_ptr += lda;
+                x_ptr += inc_x;
+            }
+
+        }
+#if !defined(XCONJ) 
+        y_ptr[0] += alpha_r * temp_r0 - alpha_i * temp_i0;
+        y_ptr[1] += alpha_r * temp_i0 + alpha_i * temp_r0;
+        y_ptr    += inc_y;
+        y_ptr[0] += alpha_r * temp_r1 - alpha_i * temp_i1;
+        y_ptr[1] += alpha_r * temp_i1 + alpha_i * temp_r1;
+        y_ptr    += inc_y;
+        y_ptr[0] += alpha_r * temp_r2 - alpha_i * temp_i2;
+        y_ptr[1] += alpha_r * temp_i2 + alpha_i * temp_r2;
+#else
+        y_ptr[0] += alpha_r * temp_r0 + alpha_i * temp_i0;
+        y_ptr[1] -= alpha_r * temp_i0 - alpha_i * temp_r0;
+        y_ptr    += inc_y;
+        y_ptr[0] += alpha_r * temp_r1 + alpha_i * temp_i1;
+        y_ptr[1] -= alpha_r * temp_i1 - alpha_i * temp_r1;
+        y_ptr    += inc_y;
+        y_ptr[0] += alpha_r * temp_r2 + alpha_i * temp_i2;
+        y_ptr[1] -= alpha_r * temp_i2 - alpha_i * temp_r2;
+#endif
+        return(0);
+    }
+
+
+
+
+
+    return(0);
+}
+
+

kernel/zarch/zgemv_t_4.c

@@ -0,0 +1,788 @@
+/***************************************************************************
+Copyright (c) 2017, 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"
+
+ 
+#define NBMAX 1024
+#define HAVE_KERNEL_4x4_VEC 1
+#define HAVE_KERNEL_4x2_VEC 1
+#define HAVE_KERNEL_4x1_VEC 1
+
+
+#if defined(HAVE_KERNEL_4x4_VEC) || defined(HAVE_KERNEL_4x2_VEC) || defined(HAVE_KERNEL_4x1_VEC)
+ #include <vecintrin.h> 
+#endif
+
+#ifdef HAVE_KERNEL_4x4_VEC_ASM
+
+#elif HAVE_KERNEL_4x4_VEC
+static void zgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0,*a1,*a2,*a3;
+    a0 = ap[0];
+    a1 = ap[1];
+    a2 = ap[2];
+    a3 = ap[3];
+    register __vector double  vtemp0_r = {0.0,0.0};  
+    register __vector double  vtemp0_i = {0.0,0.0};    
+    register __vector double  vtemp1_r = {0.0,0.0};  
+    register __vector double  vtemp1_i = {0.0,0.0};   
+    register __vector double  vtemp2_r = {0.0,0.0};  
+    register __vector double  vtemp2_i = {0.0,0.0};    
+    register __vector double  vtemp3_r = {0.0,0.0};  
+    register __vector double  vtemp3_i = {0.0,0.0};        
+    for ( i=0; i< 2*n; i+=4 )
+    {
+        register __vector double  vx_r = {x[i],x[i+2]};
+        register __vector double  vx_i = {x[i+1],x[i+3]};    
+        register __vector double  va0_r= {a0[i],a0[i+2]};
+        register __vector double  va0_i= {a0[i+1],a0[i+3]}; 
+        register __vector double  va1_r= {a1[i],a1[i+2]};
+        register __vector double  va1_i= {a1[i+1],a1[i+3]};    
+        register __vector double  va2_r= {a2[i],a2[i+2]};
+        register __vector double  va2_i= {a2[i+1],a2[i+3]}; 
+        register __vector double  va3_r= {a3[i],a3[i+2]};
+        register __vector double  va3_i= {a3[i+1],a3[i+3]};     
+           
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+
+        vtemp0_r = va0_r * vx_r - (va0_i*vx_i -vtemp0_r) ;
+        vtemp0_i = vtemp0_i + va0_r * vx_i  +  va0_i * vx_r ;
+        vtemp1_r = va1_r * vx_r - (va1_i*vx_i -vtemp1_r) ;
+        vtemp1_i = vtemp1_i + va1_r * vx_i  +  va1_i * vx_r ;
+
+        vtemp2_r = va2_r * vx_r - (va2_i*vx_i -vtemp2_r) ;
+        vtemp2_i = vtemp2_i + va2_r * vx_i  +  va2_i * vx_r ;
+        vtemp3_r = va3_r * vx_r - (va3_i*vx_i -vtemp3_r) ;
+        vtemp3_i = vtemp3_i + va3_r * vx_i  +  va3_i * vx_r ;                
+#else
+
+
+        vtemp0_r = vtemp0_r + va0_r * vx_r +  va0_i*vx_i   ; 
+        vtemp0_i =  va0_r * vx_i  -  ( va0_i * vx_r  - vtemp0_i) ;
+        vtemp1_r = vtemp1_r + va1_r * vx_r +  va1_i*vx_i   ; 
+        vtemp1_i =  va1_r * vx_i  -  ( va1_i * vx_r  - vtemp1_i);  
+        vtemp2_r = vtemp2_r + va2_r * vx_r +  va2_i*vx_i   ; 
+        vtemp2_i =  va2_r * vx_i  -  ( va2_i * vx_r  - vtemp2_i) ;
+        vtemp3_r = vtemp3_r + va3_r * vx_r +  va3_i*vx_i   ; 
+        vtemp3_i =  va3_r * vx_i  -  ( va3_i * vx_r  - vtemp3_i);                           
+#endif
+    }
+
+    register FLOAT  alpha_r = alpha[0] ;
+    register FLOAT  alpha_i = alpha[1] ;      
+    register FLOAT  temp_r0 = vtemp0_r[0]+vtemp0_r[1] ;
+    register FLOAT  temp_i0 = vtemp0_i[0]+vtemp0_i[1] ; 
+    register FLOAT  temp_r1 = vtemp1_r[0]+vtemp1_r[1] ;
+    register FLOAT  temp_i1 = vtemp1_i[0]+vtemp1_i[1] ;  
+
+    register FLOAT  temp_r2 = vtemp2_r[0]+vtemp2_r[1] ;
+    register FLOAT  temp_i2 = vtemp2_i[0]+vtemp2_i[1] ; 
+    register FLOAT  temp_r3 = vtemp3_r[0]+vtemp3_r[1] ;
+    register FLOAT  temp_i3 = vtemp3_i[0]+vtemp3_i[1] ;  
+
+#if !defined(XCONJ)
+
+    y[0] +=  alpha_r * temp_r0 - alpha_i * temp_i0;
+    y[1] +=  alpha_r * temp_i0 + alpha_i * temp_r0;
+    y[2] +=  alpha_r * temp_r1 - alpha_i * temp_i1;
+    y[3] +=  alpha_r * temp_i1 + alpha_i * temp_r1;
+    y[4] +=  alpha_r * temp_r2 - alpha_i * temp_i2;
+    y[5] +=  alpha_r * temp_i2 + alpha_i * temp_r2;
+    y[6] +=  alpha_r * temp_r3 - alpha_i * temp_i3;
+    y[7] +=  alpha_r * temp_i3 + alpha_i * temp_r3;
+
+#else
+
+    y[0] +=  alpha_r * temp_r0 + alpha_i * temp_i0;
+    y[1] -=  alpha_r * temp_i0 - alpha_i * temp_r0;
+    y[2] +=  alpha_r * temp_r1 + alpha_i * temp_i1;
+    y[3] -=  alpha_r * temp_i1 - alpha_i * temp_r1;
+    y[4] +=  alpha_r * temp_r2 + alpha_i * temp_i2;
+    y[5] -=  alpha_r * temp_i2 - alpha_i * temp_r2;
+    y[6] +=  alpha_r * temp_r3 + alpha_i * temp_i3;
+    y[7] -=  alpha_r * temp_i3 - alpha_i * temp_r3;
+
+#endif
+}
+
+#else
+
+static void zgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0,*a1,*a2,*a3;
+    a0 = ap[0];
+    a1 = ap[1];
+    a2 = ap[2];
+    a3 = ap[3];
+    FLOAT alpha_r = alpha[0];
+    FLOAT alpha_i = alpha[1];
+    FLOAT temp_r0 = 0.0;
+    FLOAT temp_r1 = 0.0;
+    FLOAT temp_r2 = 0.0;
+    FLOAT temp_r3 = 0.0;
+    FLOAT temp_i0 = 0.0;
+    FLOAT temp_i1 = 0.0;
+    FLOAT temp_i2 = 0.0;
+    FLOAT temp_i3 = 0.0;
+
+
+    for ( i=0; i< 2*n; i+=2 )
+    {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+        temp_r0 += a0[i]*x[i]   - a0[i+1]*x[i+1];        
+        temp_i0 += a0[i]*x[i+1] + a0[i+1]*x[i];        
+        temp_r1 += a1[i]*x[i]   - a1[i+1]*x[i+1];        
+        temp_i1 += a1[i]*x[i+1] + a1[i+1]*x[i];        
+        temp_r2 += a2[i]*x[i]   - a2[i+1]*x[i+1];        
+        temp_i2 += a2[i]*x[i+1] + a2[i+1]*x[i];        
+        temp_r3 += a3[i]*x[i]   - a3[i+1]*x[i+1];        
+        temp_i3 += a3[i]*x[i+1] + a3[i+1]*x[i];        
+#else
+        temp_r0 += a0[i]*x[i]   + a0[i+1]*x[i+1];        
+        temp_i0 += a0[i]*x[i+1] - a0[i+1]*x[i];        
+        temp_r1 += a1[i]*x[i]   + a1[i+1]*x[i+1];        
+        temp_i1 += a1[i]*x[i+1] - a1[i+1]*x[i];        
+        temp_r2 += a2[i]*x[i]   + a2[i+1]*x[i+1];        
+        temp_i2 += a2[i]*x[i+1] - a2[i+1]*x[i];        
+        temp_r3 += a3[i]*x[i]   + a3[i+1]*x[i+1];        
+        temp_i3 += a3[i]*x[i+1] - a3[i+1]*x[i];        
+#endif
+    }
+
+#if !defined(XCONJ)
+
+    y[0] +=  alpha_r * temp_r0 - alpha_i * temp_i0;
+    y[1] +=  alpha_r * temp_i0 + alpha_i * temp_r0;
+    y[2] +=  alpha_r * temp_r1 - alpha_i * temp_i1;
+    y[3] +=  alpha_r * temp_i1 + alpha_i * temp_r1;
+    y[4] +=  alpha_r * temp_r2 - alpha_i * temp_i2;
+    y[5] +=  alpha_r * temp_i2 + alpha_i * temp_r2;
+    y[6] +=  alpha_r * temp_r3 - alpha_i * temp_i3;
+    y[7] +=  alpha_r * temp_i3 + alpha_i * temp_r3;
+
+#else
+
+    y[0] +=  alpha_r * temp_r0 + alpha_i * temp_i0;
+    y[1] -=  alpha_r * temp_i0 - alpha_i * temp_r0;
+    y[2] +=  alpha_r * temp_r1 + alpha_i * temp_i1;
+    y[3] -=  alpha_r * temp_i1 - alpha_i * temp_r1;
+    y[4] +=  alpha_r * temp_r2 + alpha_i * temp_i2;
+    y[5] -=  alpha_r * temp_i2 - alpha_i * temp_r2;
+    y[6] +=  alpha_r * temp_r3 + alpha_i * temp_i3;
+    y[7] -=  alpha_r * temp_i3 - alpha_i * temp_r3;
+
+#endif
+}
+    
+#endif
+
+#ifdef HAVE_KERNEL_4x2_VEC
+
+
+static void zgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0,*a1;
+    a0 = ap[0];
+    a1 = ap[1]; 
+    register __vector double  vtemp0_r = {0.0,0.0};  
+    register __vector double  vtemp0_i = {0.0,0.0};    
+    register __vector double  vtemp1_r = {0.0,0.0};  
+    register __vector double  vtemp1_i = {0.0,0.0};        
+    for ( i=0; i< 2*n; i+=4 )
+    {
+        register __vector double  vx_r = {x[i],x[i+2]};
+        register __vector double  vx_i = {x[i+1],x[i+3]};    
+        register __vector double  va0_r= {a0[i],a0[i+2]};
+        register __vector double  va0_i= {a0[i+1],a0[i+3]}; 
+        register __vector double  va1_r= {a1[i],a1[i+2]};
+        register __vector double  va1_i= {a1[i+1],a1[i+3]};    
+
+
+           
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+
+        vtemp0_r = va0_r * vx_r - (va0_i*vx_i -vtemp0_r) ;
+        vtemp0_i = vtemp0_i + va0_r * vx_i  +  va0_i * vx_r ;
+        vtemp1_r = va1_r * vx_r - (va1_i*vx_i -vtemp1_r) ;
+        vtemp1_i = vtemp1_i + va1_r * vx_i  +  va1_i * vx_r ;
+#else
+        vtemp0_r = vtemp0_r + va0_r * vx_r +  va0_i*vx_i   ; 
+        vtemp0_i =  va0_r * vx_i  -  ( va0_i * vx_r  - vtemp0_i) ;
+        vtemp1_r = vtemp1_r + va1_r * vx_r +  va1_i*vx_i   ; 
+        vtemp1_i =  va1_r * vx_i  -  ( va1_i * vx_r  - vtemp1_i);            
+#endif
+    }
+
+    register FLOAT  temp_r0 = vtemp0_r[0]+vtemp0_r[1] ;
+    register FLOAT  temp_i0 = vtemp0_i[0]+vtemp0_i[1] ; 
+    register FLOAT  temp_r1 = vtemp1_r[0]+vtemp1_r[1] ;
+    register FLOAT  temp_i1 = vtemp1_i[0]+vtemp1_i[1] ;        
+    register FLOAT  alpha_r = alpha[0] ;
+    register FLOAT  alpha_i = alpha[1] ;        
+
+#if !defined(XCONJ)
+
+    y[0] +=  alpha_r * temp_r0 - alpha_i * temp_i0;
+    y[1] +=  alpha_r * temp_i0 + alpha_i * temp_r0;
+    y[2] +=  alpha_r * temp_r1 - alpha_i * temp_i1;
+    y[3] +=  alpha_r * temp_i1 + alpha_i * temp_r1;
+
+#else
+
+    y[0] +=  alpha_r * temp_r0 + alpha_i * temp_i0;
+    y[1] -=  alpha_r * temp_i0 - alpha_i * temp_r0;
+    y[2] +=  alpha_r * temp_r1 + alpha_i * temp_i1;
+    y[3] -=  alpha_r * temp_i1 - alpha_i * temp_r1;
+
+#endif
+}
+    
+#else
+
+static void zgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0,*a1;
+    a0 = ap[0];
+    a1 = ap[1];
+    FLOAT alpha_r = alpha[0];
+    FLOAT alpha_i = alpha[1];
+    FLOAT temp_r0 = 0.0;
+    FLOAT temp_r1 = 0.0;
+    FLOAT temp_i0 = 0.0;
+    FLOAT temp_i1 = 0.0;
+
+
+    for ( i=0; i< 2*n; i+=2 )
+    {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+        temp_r0 += a0[i]*x[i]   - a0[i+1]*x[i+1];        
+        temp_i0 += a0[i]*x[i+1] + a0[i+1]*x[i];        
+        temp_r1 += a1[i]*x[i]   - a1[i+1]*x[i+1];        
+        temp_i1 += a1[i]*x[i+1] + a1[i+1]*x[i];        
+#else
+        temp_r0 += a0[i]*x[i]   + a0[i+1]*x[i+1];        
+        temp_i0 += a0[i]*x[i+1] - a0[i+1]*x[i];        
+        temp_r1 += a1[i]*x[i]   + a1[i+1]*x[i+1];        
+        temp_i1 += a1[i]*x[i+1] - a1[i+1]*x[i];        
+#endif
+    }
+
+#if !defined(XCONJ)
+
+    y[0] +=  alpha_r * temp_r0 - alpha_i * temp_i0;
+    y[1] +=  alpha_r * temp_i0 + alpha_i * temp_r0;
+    y[2] +=  alpha_r * temp_r1 - alpha_i * temp_i1;
+    y[3] +=  alpha_r * temp_i1 + alpha_i * temp_r1;
+
+#else
+
+    y[0] +=  alpha_r * temp_r0 + alpha_i * temp_i0;
+    y[1] -=  alpha_r * temp_i0 - alpha_i * temp_r0;
+    y[2] +=  alpha_r * temp_r1 + alpha_i * temp_i1;
+    y[3] -=  alpha_r * temp_i1 - alpha_i * temp_r1;
+
+#endif
+}
+
+#endif
+
+
+#ifdef HAVE_KERNEL_4x1_VEC
+static void zgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0;
+    a0 = ap;
+ 
+
+    register __vector double  vtemp_r = {0.0,0.0};  
+    register __vector double  vtemp_i = {0.0,0.0};    
+        
+    for ( i=0; i< 2*n; i+=4 )
+    {
+        register __vector double  va0_r= {a0[i],a0[i+2]};
+        register __vector double  va0_i= {a0[i+1],a0[i+3]};   
+        register __vector double  vx0_r = {x[i],x[i+2]};
+        register __vector double  vx0_i = {x[i+1],x[i+3]}; 
+
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+
+        vtemp_r = va0_r * vx0_r - (va0_i*vx0_i -vtemp_r) ;
+        vtemp_i = vtemp_i + va0_r * vx0_i  +  va0_i * vx0_r ;
+#else
+        vtemp_r = vtemp_r + va0_r * vx0_r +  va0_i*vx0_i   ; 
+        vtemp_i =  va0_r * vx0_i  -  ( va0_i * vx0_r  - vtemp_i) ;      
+#endif
+    }
+        
+    register FLOAT  temp_r0 = vtemp_r[0]+vtemp_r[1] ;
+    register FLOAT  temp_i0 = vtemp_i[0]+vtemp_i[1] ; 
+    register FLOAT  alpha_r = alpha[0] ;
+    register FLOAT  alpha_i = alpha[1] ;        
+
+#if !defined(XCONJ)
+
+    y[0] +=  alpha_r * temp_r0 - alpha_i * temp_i0;
+    y[1] +=  alpha_r * temp_i0 + alpha_i * temp_r0;
+
+#else
+
+    y[0] +=  alpha_r * temp_r0 + alpha_i * temp_i0;
+    y[1] -=  alpha_r * temp_i0 - alpha_i * temp_r0;
+
+#endif
+
+
+}
+
+#else
+
+static void zgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{
+    BLASLONG i;
+    FLOAT *a0;
+    a0 = ap;
+    FLOAT alpha_r = alpha[0];
+    FLOAT alpha_i = alpha[1];
+    FLOAT temp_r0 = 0.0;
+    FLOAT temp_i0 = 0.0;
+
+    for ( i=0; i< 2*n; i+=2 )
+    {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+        temp_r0 += a0[i]*x[i]   - a0[i+1]*x[i+1];        
+        temp_i0 += a0[i]*x[i+1] + a0[i+1]*x[i];        
+#else
+        temp_r0 += a0[i]*x[i]   + a0[i+1]*x[i+1];        
+        temp_i0 += a0[i]*x[i+1] - a0[i+1]*x[i];        
+#endif
+    }
+
+#if !defined(XCONJ)
+
+    y[0] +=  alpha_r * temp_r0 - alpha_i * temp_i0;
+    y[1] +=  alpha_r * temp_i0 + alpha_i * temp_r0;
+
+#else
+
+    y[0] +=  alpha_r * temp_r0 + alpha_i * temp_i0;
+    y[1] -=  alpha_r * temp_i0 - alpha_i * temp_r0;
+
+#endif
+
+
+}
+
+#endif
+
+
+static void copy_x(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_src)
+{
+        BLASLONG i;
+        for ( i=0; i<n; i++ )
+        {
+            *dest     = *src;
+            *(dest+1) = *(src+1);
+            dest+=2;
+            src += inc_src;
+        }
+}
+
+
+int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
+{
+    BLASLONG i;
+    BLASLONG j;
+    FLOAT *a_ptr;
+    FLOAT *x_ptr;
+    FLOAT *y_ptr;
+    FLOAT *ap[8];
+    BLASLONG n1;
+    BLASLONG m1;
+    BLASLONG m2;
+    BLASLONG m3;
+    BLASLONG n2;
+    BLASLONG lda4;
+    FLOAT ybuffer[8],*xbuffer;
+    FLOAT alpha[2];
+
+    if ( m < 1 ) return(0);
+    if ( n < 1 ) return(0);
+
+    inc_x <<= 1;
+    inc_y <<= 1;
+    lda   <<= 1;
+    lda4    = lda << 2;
+
+    xbuffer = buffer;
+    
+    n1 = n  >> 2 ;
+    n2 = n  &  3 ;
+    
+    m3 = m & 3 ;
+    m1 = m - m3;
+    m2 = (m & (NBMAX-1)) - m3 ;
+    
+    alpha[0] = alpha_r;
+    alpha[1] = alpha_i;
+
+    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;
+        ap[0] = a_ptr;
+        ap[1] = a_ptr + lda;
+        ap[2] = ap[1] + lda;
+        ap[3] = ap[2] + lda;
+        if ( inc_x != 2 )
+            copy_x(NB,x_ptr,xbuffer,inc_x);
+        else
+            xbuffer = x_ptr;
+        
+        if ( inc_y == 2 )
+        {
+
+            for( i = 0; i < n1 ; i++)
+            {
+                zgemv_kernel_4x4(NB,ap,xbuffer,y_ptr,alpha);
+                ap[0] += lda4;
+                ap[1] += lda4;
+                ap[2] += lda4;
+                ap[3] += lda4;
+                a_ptr += lda4;
+                y_ptr += 8;
+                
+            }
+
+            if ( n2 & 2 )
+            {
+                zgemv_kernel_4x2(NB,ap,xbuffer,y_ptr,alpha);
+                a_ptr += lda * 2;
+                y_ptr += 4;
+
+            }
+
+            if ( n2 & 1 )
+            {
+                zgemv_kernel_4x1(NB,a_ptr,xbuffer,y_ptr,alpha);
+                a_ptr += lda;
+                y_ptr += 2;
+
+            }
+
+        }
+        else
+        {
+
+            for( i = 0; i < n1 ; i++)
+            {
+                memset(ybuffer,0,64);
+                zgemv_kernel_4x4(NB,ap,xbuffer,ybuffer,alpha);
+                ap[0] += lda4;
+                ap[1] += lda4;
+                ap[2] += lda4;
+                ap[3] += lda4;
+                a_ptr += lda4;
+
+                y_ptr[0] += ybuffer[0];
+                y_ptr[1] += ybuffer[1];
+                y_ptr  += inc_y;
+                y_ptr[0] += ybuffer[2];
+                y_ptr[1] += ybuffer[3];
+                y_ptr  += inc_y;
+                y_ptr[0] += ybuffer[4];
+                y_ptr[1] += ybuffer[5];
+                y_ptr  += inc_y;
+                y_ptr[0] += ybuffer[6];
+                y_ptr[1] += ybuffer[7];
+                y_ptr  += inc_y;
+
+            }
+
+            for( i = 0; i < n2 ; i++)
+            {
+                memset(ybuffer,0,64);
+                zgemv_kernel_4x1(NB,a_ptr,xbuffer,ybuffer,alpha);
+                a_ptr += lda;
+                y_ptr[0] += ybuffer[0];
+                y_ptr[1] += ybuffer[1];
+                y_ptr  += inc_y;
+
+            }
+
+        }
+        a += 2 * NB;
+        x += NB * inc_x;    
+    }
+
+
+
+    if ( m3 == 0 ) return(0);
+
+        x_ptr = x;
+        j=0;
+        a_ptr = a;
+        y_ptr = y;
+
+    if ( m3 == 3 )
+    {
+
+        FLOAT temp_r ;
+        FLOAT temp_i ;
+        FLOAT x0 = x_ptr[0];
+        FLOAT x1 = x_ptr[1];
+        x_ptr += inc_x;
+        FLOAT x2 = x_ptr[0];
+        FLOAT x3 = x_ptr[1];
+        x_ptr += inc_x;
+        FLOAT x4 = x_ptr[0];
+        FLOAT x5 = x_ptr[1];
+        while ( j < n)
+        {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+            temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 
+            temp_r += a_ptr[2] * x2 - a_ptr[3] * x3; 
+            temp_i += a_ptr[2] * x3 + a_ptr[3] * x2; 
+            temp_r += a_ptr[4] * x4 - a_ptr[5] * x5;
+            temp_i += a_ptr[4] * x5 + a_ptr[5] * x4;
+#else
+
+            temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 
+            temp_r += a_ptr[2] * x2 + a_ptr[3] * x3; 
+            temp_i += a_ptr[2] * x3 - a_ptr[3] * x2; 
+            temp_r += a_ptr[4] * x4 + a_ptr[5] * x5;
+            temp_i += a_ptr[4] * x5 - a_ptr[5] * x4;
+#endif
+
+#if !defined(XCONJ) 
+            y_ptr[0] += alpha_r * temp_r - alpha_i * temp_i;
+            y_ptr[1] += alpha_r * temp_i + alpha_i * temp_r;
+#else
+            y_ptr[0] += alpha_r * temp_r + alpha_i * temp_i;
+            y_ptr[1] -= alpha_r * temp_i - alpha_i * temp_r;
+#endif
+
+            a_ptr += lda;
+            y_ptr += inc_y;
+            j++;
+        }
+        return(0);
+    }
+
+
+    if ( m3 == 2 )
+    {
+
+        FLOAT temp_r ;
+        FLOAT temp_i ;
+        FLOAT temp_r1 ;
+        FLOAT temp_i1 ;
+        FLOAT x0 = x_ptr[0];
+        FLOAT x1 = x_ptr[1];
+        x_ptr += inc_x;
+        FLOAT x2 = x_ptr[0];
+        FLOAT x3 = x_ptr[1];
+        FLOAT ar = alpha[0];
+        FLOAT ai = alpha[1];
+
+        while ( j < ( n & -2 ))
+        {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+            temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 
+            temp_r += a_ptr[2] * x2 - a_ptr[3] * x3; 
+            temp_i += a_ptr[2] * x3 + a_ptr[3] * x2; 
+            a_ptr += lda;
+            temp_r1  = a_ptr[0] * x0 - a_ptr[1] * x1; 
+            temp_i1  = a_ptr[0] * x1 + a_ptr[1] * x0; 
+            temp_r1 += a_ptr[2] * x2 - a_ptr[3] * x3; 
+            temp_i1 += a_ptr[2] * x3 + a_ptr[3] * x2; 
+#else
+
+            temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 
+            temp_r += a_ptr[2] * x2 + a_ptr[3] * x3; 
+            temp_i += a_ptr[2] * x3 - a_ptr[3] * x2; 
+            a_ptr += lda;
+            temp_r1  = a_ptr[0] * x0 + a_ptr[1] * x1; 
+            temp_i1  = a_ptr[0] * x1 - a_ptr[1] * x0; 
+            temp_r1 += a_ptr[2] * x2 + a_ptr[3] * x3; 
+            temp_i1 += a_ptr[2] * x3 - a_ptr[3] * x2; 
+#endif
+
+#if !defined(XCONJ) 
+            y_ptr[0] += ar * temp_r - ai * temp_i;
+            y_ptr[1] += ar * temp_i + ai * temp_r;
+            y_ptr += inc_y;
+            y_ptr[0] += ar * temp_r1 - ai * temp_i1;
+            y_ptr[1] += ar * temp_i1 + ai * temp_r1;
+#else
+            y_ptr[0] += ar * temp_r + ai * temp_i;
+            y_ptr[1] -= ar * temp_i - ai * temp_r;
+            y_ptr += inc_y;
+            y_ptr[0] += ar * temp_r1 + ai * temp_i1;
+            y_ptr[1] -= ar * temp_i1 - ai * temp_r1;
+#endif
+
+            a_ptr += lda;
+            y_ptr += inc_y;
+            j+=2;
+        }
+
+
+        while ( j < n)
+        {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+            temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 
+            temp_r += a_ptr[2] * x2 - a_ptr[3] * x3; 
+            temp_i += a_ptr[2] * x3 + a_ptr[3] * x2; 
+#else
+
+            temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 
+            temp_r += a_ptr[2] * x2 + a_ptr[3] * x3; 
+            temp_i += a_ptr[2] * x3 - a_ptr[3] * x2; 
+#endif
+
+#if !defined(XCONJ) 
+            y_ptr[0] += ar * temp_r - ai * temp_i;
+            y_ptr[1] += ar * temp_i + ai * temp_r;
+#else
+            y_ptr[0] += ar * temp_r + ai * temp_i;
+            y_ptr[1] -= ar * temp_i - ai * temp_r;
+#endif
+
+            a_ptr += lda;
+            y_ptr += inc_y;
+            j++;
+        }
+
+        return(0);
+    }
+
+
+    if ( m3 == 1 )
+    {
+
+        FLOAT temp_r ;
+        FLOAT temp_i ;
+        FLOAT temp_r1 ;
+        FLOAT temp_i1 ;
+        FLOAT x0 = x_ptr[0];
+        FLOAT x1 = x_ptr[1];
+        FLOAT ar = alpha[0];
+        FLOAT ai = alpha[1];
+
+        while ( j < ( n & -2 ))
+        {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+            temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 
+            a_ptr += lda;
+            temp_r1  = a_ptr[0] * x0 - a_ptr[1] * x1; 
+            temp_i1  = a_ptr[0] * x1 + a_ptr[1] * x0; 
+#else
+
+            temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 
+            a_ptr += lda;
+            temp_r1  = a_ptr[0] * x0 + a_ptr[1] * x1; 
+            temp_i1  = a_ptr[0] * x1 - a_ptr[1] * x0; 
+#endif
+
+#if !defined(XCONJ) 
+            y_ptr[0] += ar * temp_r - ai * temp_i;
+            y_ptr[1] += ar * temp_i + ai * temp_r;
+            y_ptr += inc_y;
+            y_ptr[0] += ar * temp_r1 - ai * temp_i1;
+            y_ptr[1] += ar * temp_i1 + ai * temp_r1;
+#else
+            y_ptr[0] += ar * temp_r + ai * temp_i;
+            y_ptr[1] -= ar * temp_i - ai * temp_r;
+            y_ptr += inc_y;
+            y_ptr[0] += ar * temp_r1 + ai * temp_i1;
+            y_ptr[1] -= ar * temp_i1 - ai * temp_r1;
+#endif
+
+            a_ptr += lda;
+            y_ptr += inc_y;
+            j+=2;
+        }
+
+        while ( j < n)
+        {
+#if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
+            temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 
+#else
+
+            temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 
+            temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 
+#endif
+
+#if !defined(XCONJ) 
+            y_ptr[0] += ar * temp_r - ai * temp_i;
+            y_ptr[1] += ar * temp_i + ai * temp_r;
+#else
+            y_ptr[0] += ar * temp_r + ai * temp_i;
+            y_ptr[1] -= ar * temp_i - ai * temp_r;
+#endif
+
+            a_ptr += lda;
+            y_ptr += inc_y;
+            j++;
+        }
+        return(0);
+    }
+
+    return(0);
+
+
+}
+
+

kernel/zarch/zrot.c

@@ -0,0 +1,276 @@
+/***************************************************************************
+Copyright (c) 2017, 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"
+
+static void __attribute__ ((noinline))  zrot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *c, FLOAT *s)
+{
+          __asm__  (
+            "pfd 2, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+             
+            "vlrepg %%v0,0(%3) \n\t"
+            "vlrepg %%v1,0(%4) \n\t"
+            "srlg %%r0,%0,4    \n\t"
+            "xgr %%r1,%%r1     \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 2, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+            "vl  %%v24,  0(%%r1,%1) \n\t" 
+            "vl  %%v25, 16(%%r1,%1) \n\t" 
+            "vl  %%v26, 32(%%r1,%1) \n\t" 
+            "vl  %%v27, 48(%%r1,%1) \n\t" 
+            "vl  %%v16,  0(%%r1,%2) \n\t" 
+            "vl  %%v17, 16(%%r1,%2) \n\t" 
+            "vl  %%v18, 32(%%r1,%2) \n\t" 
+            "vl  %%v19, 48(%%r1,%2) \n\t"  
+           
+            "vfmdb %%v28,%%v24,%%v0 \n\t"
+            "vfmdb %%v29,%%v25,%%v0 \n\t"
+            "vfmdb %%v20,%%v24,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v21,%%v25,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v30,%%v26,%%v0 \n\t"
+            "vfmdb %%v22,%%v26,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v31,%%v27,%%v0 \n\t"
+            "vfmdb %%v23,%%v27,%%v1 \n\t" /* yn=x*s  */
+            /* 2nd parts*/
+            "vfmadb %%v28,%%v16,%%v1,%%v28 \n\t"  
+            "vfmsdb %%v20,%%v16,%%v0,%%v20 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v29,%%v17,%%v1,%%v29 \n\t"  
+            "vfmsdb %%v21,%%v17,%%v0,%%v21 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v30,%%v18,%%v1,%%v30 \n\t" 
+            "vfmsdb %%v22,%%v18,%%v0,%%v22 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v31,%%v19,%%v1,%%v31 \n\t" 
+            "vfmsdb %%v23,%%v19,%%v0,%%v23 \n\t"  /* yn=y*c-yn */
+           
+           
+           
+            "vst  %%v28, 0(%%r1,%1) \n\t" 
+            "vst  %%v29, 16(%%r1,%1) \n\t" 
+            "vst  %%v30, 32(%%r1,%1) \n\t" 
+            "vst  %%v31, 48(%%r1,%1) \n\t" 
+            "vst  %%v20, 0(%%r1,%2) \n\t" 
+            "vst  %%v21, 16(%%r1,%2) \n\t" 
+            "vst  %%v22, 32(%%r1,%2) \n\t" 
+            "vst  %%v23, 48(%%r1,%2) \n\t"  
+           
+            "vl  %%v24, 64(%%r1,%1) \n\t" 
+            "vl  %%v25, 80(%%r1,%1) \n\t" 
+            "vl  %%v26, 96(%%r1,%1) \n\t" 
+            "vl  %%v27,112(%%r1,%1) \n\t" 
+            "vl  %%v16, 64(%%r1,%2) \n\t" 
+            "vl  %%v17, 80(%%r1,%2) \n\t" 
+            "vl  %%v18, 96(%%r1,%2) \n\t" 
+            "vl  %%v19,112(%%r1,%2) \n\t"  
+           
+            "vfmdb %%v28,%%v24,%%v0 \n\t"
+            "vfmdb %%v29,%%v25,%%v0 \n\t"
+            "vfmdb %%v20,%%v24,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v21,%%v25,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v30,%%v26,%%v0 \n\t"
+            "vfmdb %%v22,%%v26,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v31,%%v27,%%v0 \n\t"
+            "vfmdb %%v23,%%v27,%%v1 \n\t" /* yn=x*s  */
+            /* 2nd parts*/
+            "vfmadb %%v28,%%v16,%%v1,%%v28 \n\t"  
+            "vfmsdb %%v20,%%v16,%%v0,%%v20 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v29,%%v17,%%v1,%%v29 \n\t"  
+            "vfmsdb %%v21,%%v17,%%v0,%%v21 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v30,%%v18,%%v1,%%v30 \n\t" 
+            "vfmsdb %%v22,%%v18,%%v0,%%v22 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v31,%%v19,%%v1,%%v31 \n\t" 
+            "vfmsdb %%v23,%%v19,%%v0,%%v23 \n\t"  /* yn=y*c-yn */
+           
+           
+           
+            "vst  %%v28, 64(%%r1,%1) \n\t" 
+            "vst  %%v29, 80(%%r1,%1) \n\t" 
+            "vst  %%v30, 96(%%r1,%1) \n\t" 
+            "vst  %%v31, 112(%%r1,%1) \n\t" 
+            "vst  %%v20, 64(%%r1,%2) \n\t" 
+            "vst  %%v21, 80(%%r1,%2) \n\t" 
+            "vst  %%v22, 96(%%r1,%2) \n\t" 
+            "vst  %%v23, 112(%%r1,%2) \n\t"
+           
+            "vl  %%v24, 128(%%r1,%1) \n\t" 
+            "vl  %%v25, 144(%%r1,%1) \n\t" 
+            "vl  %%v26, 160(%%r1,%1) \n\t" 
+            "vl  %%v27, 176(%%r1,%1) \n\t" 
+            "vl  %%v16, 128(%%r1,%2) \n\t" 
+            "vl  %%v17, 144(%%r1,%2) \n\t" 
+            "vl  %%v18, 160(%%r1,%2) \n\t" 
+            "vl  %%v19, 176(%%r1,%2) \n\t"  
+           
+            "vfmdb %%v28,%%v24,%%v0 \n\t"
+            "vfmdb %%v29,%%v25,%%v0 \n\t"
+            "vfmdb %%v20,%%v24,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v21,%%v25,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v30,%%v26,%%v0 \n\t"
+            "vfmdb %%v22,%%v26,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v31,%%v27,%%v0 \n\t"
+            "vfmdb %%v23,%%v27,%%v1 \n\t" /* yn=x*s  */
+            /* 2nd parts*/
+            "vfmadb %%v28,%%v16,%%v1,%%v28 \n\t"  
+            "vfmsdb %%v20,%%v16,%%v0,%%v20 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v29,%%v17,%%v1,%%v29 \n\t"  
+            "vfmsdb %%v21,%%v17,%%v0,%%v21 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v30,%%v18,%%v1,%%v30 \n\t" 
+            "vfmsdb %%v22,%%v18,%%v0,%%v22 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v31,%%v19,%%v1,%%v31 \n\t" 
+            "vfmsdb %%v23,%%v19,%%v0,%%v23 \n\t"  /* yn=y*c-yn */
+           
+           
+           
+            "vst  %%v28, 128(%%r1,%1) \n\t" 
+            "vst  %%v29, 144(%%r1,%1) \n\t" 
+            "vst  %%v30, 160(%%r1,%1) \n\t" 
+            "vst  %%v31, 176(%%r1,%1) \n\t" 
+            "vst  %%v20, 128(%%r1,%2) \n\t" 
+            "vst  %%v21, 144(%%r1,%2) \n\t" 
+            "vst  %%v22, 160(%%r1,%2) \n\t" 
+            "vst  %%v23, 176(%%r1,%2) \n\t"  
+           
+            "vl  %%v24, 192(%%r1,%1) \n\t" 
+            "vl  %%v25, 208(%%r1,%1) \n\t" 
+            "vl  %%v26, 224(%%r1,%1) \n\t" 
+            "vl  %%v27, 240(%%r1,%1) \n\t" 
+            "vl  %%v16, 192(%%r1,%2) \n\t" 
+            "vl  %%v17, 208(%%r1,%2) \n\t" 
+            "vl  %%v18, 224(%%r1,%2) \n\t" 
+            "vl  %%v19, 240(%%r1,%2) \n\t"  
+           
+            "vfmdb %%v28,%%v24,%%v0 \n\t"
+            "vfmdb %%v29,%%v25,%%v0 \n\t"
+            "vfmdb %%v20,%%v24,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v21,%%v25,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v30,%%v26,%%v0 \n\t"
+            "vfmdb %%v22,%%v26,%%v1 \n\t" /* yn=x*s  */
+            "vfmdb %%v31,%%v27,%%v0 \n\t"
+            "vfmdb %%v23,%%v27,%%v1 \n\t" /* yn=x*s  */
+            /* 2nd parts*/
+            "vfmadb %%v28,%%v16,%%v1,%%v28 \n\t"  
+            "vfmsdb %%v20,%%v16,%%v0,%%v20 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v29,%%v17,%%v1,%%v29 \n\t"  
+            "vfmsdb %%v21,%%v17,%%v0,%%v21 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v30,%%v18,%%v1,%%v30 \n\t" 
+            "vfmsdb %%v22,%%v18,%%v0,%%v22 \n\t"  /* yn=y*c-yn */ 
+            "vfmadb %%v31,%%v19,%%v1,%%v31 \n\t" 
+            "vfmsdb %%v23,%%v19,%%v0,%%v23 \n\t"  /* yn=y*c-yn */
+           
+           
+           
+            "vst  %%v28, 192(%%r1,%1) \n\t" 
+            "vst  %%v29, 208(%%r1,%1) \n\t" 
+            "vst  %%v30, 224(%%r1,%1) \n\t" 
+            "vst  %%v31, 240(%%r1,%1) \n\t" 
+            "vst  %%v20, 192(%%r1,%2) \n\t" 
+            "vst  %%v21, 208(%%r1,%2) \n\t" 
+            "vst  %%v22, 224(%%r1,%2) \n\t" 
+            "vst  %%v23, 240(%%r1,%2) \n\t"
+           
+           
+          
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y),"a"(c),"a"(s)
+            : "cc", "memory","r0","r1" ,"v0","v1","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+            );
+      return;
+
+}
+
+
+
+int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s)
+{
+    BLASLONG i=0;
+    BLASLONG ix=0,iy=0;
+    FLOAT temp[2];
+    BLASLONG inc_x2;
+    BLASLONG inc_y2;
+
+    if ( n <= 0     )  return(0); 
+
+    if ( (inc_x == 1) && (inc_y == 1) )
+    {
+
+        BLASLONG n1 = n & -16;
+        if ( n1 > 0 )
+        {
+            FLOAT cosa,sina;
+            cosa=c;
+            sina=s;
+            zrot_kernel_16(n1, x, y, &cosa, &sina);
+            i=n1; 
+            ix=2*n1; 
+        }
+
+         while(i < n)
+           {
+                temp[0]   = c*x[ix]   + s*y[ix] ;
+                temp[1]   = c*x[ix+1] + s*y[ix+1] ;
+                y[ix]     = c*y[ix]   - s*x[ix] ;
+                y[ix+1]   = c*y[ix+1] - s*x[ix+1] ;
+                x[ix]     = temp[0] ;
+                x[ix+1]   = temp[1] ;
+
+                ix += 2 ; 
+                i++ ;
+
+            }
+
+
+    }
+    else
+    {
+        inc_x2 = 2 * inc_x ;
+        inc_y2 = 2 * inc_y ;
+        while(i < n)
+        {
+            temp[0]   = c*x[ix]   + s*y[iy] ;
+            temp[1]   = c*x[ix+1] + s*y[iy+1] ;
+            y[iy]     = c*y[iy]   - s*x[ix] ;
+            y[iy+1]   = c*y[iy+1] - s*x[ix+1] ;
+            x[ix]     = temp[0] ;
+            x[ix+1]   = temp[1] ;
+
+            ix += inc_x2 ;
+            iy += inc_y2 ;
+            i++ ;
+
+        }
+
+    }
+    return(0);
+ 
+}
+
+

kernel/zarch/zscal.c

@@ -0,0 +1,483 @@
+/***************************************************************************
+Copyright (c) 2013 - 2017, 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"
+
+ 
+
+static void  __attribute__ ((noinline))  zscal_kernel_8(BLASLONG n, FLOAT *alpha, FLOAT *x) {
+    __asm__(
+
+            "pfd 1, 0(%1) \n\t"
+            "sllg %%r0,%0,4       \n\t"
+            "agr  %%r0,%2  \n\t"
+            "vlrepg %%v24,0(%1)  \n\t"
+            "vlrepg %%v25,8(%1)  \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 2, 256(%2 ) \n\t"
+
+            "vleg %%v20 ,  0(%2),0  \n\t" 
+            "vleg %%v21 ,  8(%2),0  \n\t"
+            "vleg %%v20 , 16(%2),1  \n\t"
+            "vleg %%v21 , 24(%2),1  \n\t"
+
+            "vleg %%v22 , 32(%2),0  \n\t" 
+            "vleg %%v23 , 40(%2),0  \n\t"
+            "vleg %%v22 , 48(%2),1  \n\t"
+            "vleg %%v23 , 56(%2),1  \n\t"
+
+            "vfmdb %%v16,  %%v21, %%v25  \n\t"
+            "vfmdb %%v17,  %%v20, %%v25  \n\t"
+            "vfmdb %%v18,  %%v23, %%v25  \n\t"
+            "vfmdb %%v19,  %%v22, %%v25  \n\t"
+
+            "vfmsdb %%v16, %%v20, %%v24 ,%%v16  \n\t"
+            "vfmadb %%v17, %%v21, %%v24, %%v17  \n\t"
+            "vfmsdb %%v18, %%v22, %%v24, %%v18  \n\t"
+            "vfmadb %%v19, %%v23, %%v24, %%v19  \n\t"
+
+            "vsteg %%v16 ,  0(%2),0  \n\t" 
+            "vsteg %%v17 ,  8(%2),0  \n\t"
+            "vsteg %%v16 , 16(%2),1  \n\t"
+            "vsteg %%v17 , 24(%2),1  \n\t"
+
+            "vsteg %%v18 , 32(%2),0  \n\t" 
+            "vsteg %%v19 , 40(%2),0  \n\t"
+            "vsteg %%v18 , 48(%2),1  \n\t"
+            "vsteg %%v19 , 56(%2),1  \n\t"
+            
+            "vleg %%v20 ,  64(%2),0  \n\t" 
+            "vleg %%v21 , 72(%2),0   \n\t"
+            "vleg %%v20 , 80(%2),1   \n\t"
+            "vleg %%v21 , 88(%2),1   \n\t"
+
+            "vleg %%v22 , 96(%2),0   \n\t" 
+            "vleg %%v23 , 104(%2),0  \n\t"
+            "vleg %%v22 , 112(%2),1  \n\t"
+            "vleg %%v23 , 120(%2),1  \n\t"
+
+            "vfmdb %%v16,  %%v21, %%v25  \n\t"
+            "vfmdb %%v17,  %%v20, %%v25  \n\t"
+            "vfmdb %%v18,  %%v23, %%v25  \n\t"
+            "vfmdb %%v19,  %%v22, %%v25  \n\t"
+
+            "vfmsdb %%v16, %%v20, %%v24 ,%%v16  \n\t"
+            "vfmadb %%v17, %%v21, %%v24, %%v17  \n\t"
+            "vfmsdb %%v18, %%v22, %%v24, %%v18  \n\t"
+            "vfmadb %%v19, %%v23, %%v24, %%v19  \n\t"
+
+            "vsteg %%v16 , 64(%2),0  \n\t" 
+            "vsteg %%v17 , 72(%2),0  \n\t"
+            "vsteg %%v16 , 80(%2),1  \n\t"
+            "vsteg %%v17 , 88(%2),1  \n\t"
+
+            "vsteg %%v18 ,  96(%2),0 \n\t" 
+            "vsteg %%v19 , 104(%2),0 \n\t"
+            "vsteg %%v18 , 112(%2),1 \n\t"
+            "vsteg %%v19 , 120(%2),1 \n\t"
+   
+            "la %2,128(%2)     \n\t"
+            "clgrjl %2,%%r0,1b \n\t"
+            :
+            : "r"(n), "a"(alpha), "a"(x)
+            : "cc", "memory","r0","v16","v17","v18","v19","v20","v21","v22","v23","v24","v25"
+            );
+
+
+}
+ 
+static void   __attribute__ ((noinline))  zscal_kernel_8_zero_r(BLASLONG n, FLOAT *alpha, FLOAT *x) {
+ 
+        __asm__ (   "pfd 2, 0(%1)   \n\t" 
+                    "ld %%f0,8(%2)  \n\t"
+                    "lcdbr %%f1,%%f0     \n\t"
+                    "lgdr %%r0,%%f1      \n\t"
+                    "vlvgg %%v0,%%r0,1   \n\t"
+                    "vlr %%v16,%%v0      \n\t"
+                    "vlr %%v17 ,%%v0     \n\t"
+                    "vlr %%v1,%%v0       \n\t" 
+                    "sllg %%r0,%0,4      \n\t"  
+                    "agr %%r0,%1  \n\t"
+                    ".align 16    \n\t"    
+                    "1: \n\t"  
+                    "vl     %%v24, 0(%1)      \n\t"
+                    "vfmdb  %%v24,%%v24,%%v0  \n\t"
+                    "vsteg  %%v24, 0(%1),1    \n\t" 
+                    "vsteg  %%v24, 8(%1),0    \n\t" 
+                    "vl     %%v25, 16(%1)     \n\t"
+                    "vfmdb  %%v25,%%v25,%%v1  \n\t"  
+                    "vsteg  %%v25, 16(%1),1   \n\t" 
+                    "vsteg  %%v25, 24(%1),0   \n\t" 
+                    "vl     %%v26, 32(%1)     \n\t"
+                    "vfmdb  %%v26,%%v26,%%v16 \n\t"
+                    "vsteg  %%v26, 32(%1),1   \n\t" 
+                    "vsteg  %%v26, 40(%1),0   \n\t"   
+                    "vl     %%v27, 48(%1)     \n\t" 
+                    "vfmdb  %%v27,%%v27,%%v17 \n\t"  
+                    "vsteg  %%v27, 40(%1),1   \n\t" 
+                    "vsteg  %%v27, 48(%1),0   \n\t" 
+                    "vl     %%v28, 64(%1)     \n\t"
+                    "vfmdb  %%v28,%%v28,%%v0  \n\t"
+                    "vsteg  %%v28, 64(%1),1   \n\t" 
+                    "vsteg  %%v28, 72(%1),0   \n\t" 
+                    "vl     %%v29, 80(%1)     \n\t"
+                    "vfmdb  %%v29,%%v29,%%v1  \n\t"  
+                    "vsteg  %%v29, 80(%1),1   \n\t" 
+                    "vsteg  %%v29, 88(%1),0   \n\t" 
+                    "vl     %%v30, 96(%1)     \n\t"
+                    "vfmdb  %%v30,%%v30,%%v16 \n\t"
+                    "vsteg  %%v27,  96(%1),1  \n\t" 
+                    "vsteg  %%v27, 104(%1),0  \n\t"  
+                    "vl     %%v31, 112(%1)    \n\t" 
+                    "vfmdb  %%v31,%%v31,%%v17 \n\t"  
+                    "vsteg  %%v31, 112(%1),1  \n\t" 
+                    "vsteg  %%v31, 120(%1),0  \n\t" 
+                    "la %1,128(%1) \n\t"
+                    "clgrjl %1,%%r0,1b \n\t"
+                    : 
+                    :"r"(n),"a"(x) ,"a"(alpha)
+                    :"cc", "memory","r0","f0", "f1","v0","v1","v18","v19","v24","v25","v26","v27","v28","v29","v30","v31" 
+                 );
+
+
+}
+
+static void  __attribute__ ((noinline)) zscal_kernel_8_zero_i(BLASLONG n, FLOAT *alpha, FLOAT *x) {
+           __asm__ ("pfd 2, 0(%1) \n\t"      
+                    "vlrepg %%v18,0(%2) \n\t"
+                    "vlr %%v19,%%v18    \n\t"
+                    "vlr %%v16 ,%%v18   \n\t"
+                    "vlr %%v17,%%v18    \n\t" 
+                    "sllg %%r0,%0,4     \n\t"  
+                    "agr %%r0,%1  \n\t"
+                    ".align 16 \n\t"    
+                    "1: \n\t"  
+                    "vl     %%v24, 0(%1)      \n\t"
+                    "vfmdb  %%v24,%%v24,%%v18 \n\t"
+                    "vst    %%v24, 0(%1)      \n\t" 
+                    "vl     %%v25, 16(%1)     \n\t"
+                    "vfmdb  %%v25,%%v25,%%v19 \n\t"  
+                    "vst    %%v25, 16(%1)     \n\t" 
+                    "vl     %%v26, 32(%1)     \n\t"
+                    "vfmdb  %%v26,%%v26,%%v16 \n\t"
+                    "vst    %%v26, 32(%1)     \n\t"  
+                    "vl     %%v27, 48(%1)     \n\t" 
+                    "vfmdb  %%v27,%%v27,%%v17 \n\t"  
+                    "vst    %%v27, 48(%1)     \n\t"  
+                    "vl     %%v28, 64(%1)     \n\t"
+                    "vfmdb  %%v28,%%v28,%%v18 \n\t"
+                    "vst    %%v28, 64(%1)     \n\t" 
+                    "vl     %%v29, 80(%1)     \n\t"
+                    "vfmdb  %%v29,%%v29,%%v19 \n\t"  
+                    "vst    %%v29, 80(%1)     \n\t" 
+                    "vl     %%v30, 96(%1)     \n\t"
+                    "vfmdb  %%v30,%%v30,%%v16 \n\t"
+                    "vst    %%v30, 96(%1)     \n\t"  
+                    "vl     %%v31, 112(%1)    \n\t" 
+                    "vfmdb  %%v31,%%v31,%%v17 \n\t"  
+                    "vst    %%v31, 112(%1)    \n\t"
+                    "la    %1,128(%1)         \n\t"
+                    "clgrjl %1,%%r0,1b        \n\t"
+                    : 
+                    :"r"(n),"a"(x) ,"a"(alpha)
+                    :"cc", "memory","r0","v16", "v17","v18","v19","v24","v25","v26","v27","v28","v29","v30","v31" 
+                 );
+
+}
+
+static void  __attribute__ ((noinline)) zscal_kernel_8_zero(BLASLONG n,  FLOAT *x) {
+
+     __asm__ (      "pfd 2, 0(%1)    \n\t"      
+                    "vzero %%v24     \n\t"
+                    "vzero %%v25     \n\t"
+                    "vzero %%v26     \n\t"
+                    "vzero %%v27     \n\t" 
+                    "sllg  %%r0,%0,4 \n\t"  
+                    "agr   %%r0,%1   \n\t"
+                    ".align 16 \n\t"    
+                    "1: \n\t" 
+                    "pfd 2, 256( %1)      \n\t"     
+                    "vst  %%v24,  0( %1)  \n\t" 
+                    "vst  %%v25, 16( %1)  \n\t" 
+                    "vst  %%v26, 32( %1)  \n\t"   
+                    "vst  %%v27, 48( %1)  \n\t"  
+                    "vst  %%v24, 64( %1)  \n\t" 
+                    "vst  %%v25, 80( %1)  \n\t" 
+                    "vst  %%v26, 96( %1)  \n\t"  
+                    "vst  %%v27,112( %1)  \n\t"  
+              
+                    "la     %1,128(%1) \n\t"
+                    "clgrjl %1,%%r0,1b \n\t"
+                    : 
+                    :"r"(n),"a"(x) 
+                    :"cc" , "memory" ,"r0","v24","v25","v26","v27"
+                 );
+
+}
+
+
+
+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,  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/zarch/zswap.c

@@ -0,0 +1,198 @@
+/***************************************************************************
+Copyright (c) 2013-2017, 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"
+
+
+static void  __attribute__ ((noinline))  zswap_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y)
+{
+         __asm__ volatile(
+            "pfd 2, 0(%1) \n\t"
+            "pfd 2, 0(%2) \n\t"
+            "srlg %%r0,%0,4      \n\t"
+            "xgr %%r1,%%r1       \n\t"
+            ".align 16 \n\t"
+            "1: \n\t"
+            "pfd 2, 256(%%r1,%1) \n\t"
+            "pfd 2, 256(%%r1,%2) \n\t"
+            
+            "vl  %%v16,   0(%%r1,%1) \n\t" 
+            "vl  %%v17,  16(%%r1,%1) \n\t" 
+            "vl  %%v18,  32(%%r1,%1) \n\t" 
+            "vl  %%v19,  48(%%r1,%1) \n\t"  
+            "vl  %%v20,  64(%%r1,%1) \n\t" 
+            "vl  %%v21,  80(%%r1,%1) \n\t" 
+            "vl  %%v22,  96(%%r1,%1) \n\t" 
+            "vl  %%v23, 112(%%r1,%1) \n\t"
+            "vl  %%v24, 128(%%r1,%1) \n\t" 
+            "vl  %%v25, 144(%%r1,%1) \n\t" 
+            "vl  %%v26, 160(%%r1,%1) \n\t" 
+            "vl  %%v27, 176(%%r1,%1) \n\t" 
+            "vl  %%v28, 192(%%r1,%1) \n\t" 
+            "vl  %%v29, 208(%%r1,%1) \n\t" 
+            "vl  %%v30, 224(%%r1,%1) \n\t" 
+            "vl  %%v31, 240(%%r1,%1) \n\t"
+
+
+            "vl  %%v0,    0(%%r1,%2) \n\t" 
+            "vl  %%v1,   16(%%r1,%2) \n\t" 
+            "vl  %%v2,   32(%%r1,%2) \n\t" 
+            "vl  %%v3,   48(%%r1,%2) \n\t"  
+            "vl  %%v4,   64(%%r1,%2) \n\t" 
+            "vl  %%v5,   80(%%r1,%2) \n\t" 
+            "vl  %%v6,   96(%%r1,%2) \n\t" 
+            "vl  %%v7,  112(%%r1,%2) \n\t"
+            "vst %%v0,    0(%%r1,%1) \n\t" 
+            "vst %%v1,   16(%%r1,%1) \n\t" 
+            "vst %%v2,   32(%%r1,%1) \n\t" 
+            "vst %%v3,   48(%%r1,%1) \n\t"  
+            "vst %%v4,   64(%%r1,%1) \n\t" 
+            "vst %%v5,   80(%%r1,%1) \n\t" 
+            "vst %%v6,   96(%%r1,%1) \n\t" 
+            "vst %%v7,  112(%%r1,%1) \n\t" 
+
+            "vl  %%v0,  128(%%r1,%2) \n\t" 
+            "vl  %%v1,  144(%%r1,%2) \n\t" 
+            "vl  %%v2,  160(%%r1,%2) \n\t" 
+            "vl  %%v3,  176(%%r1,%2) \n\t"  
+            "vl  %%v4,  192(%%r1,%2) \n\t" 
+            "vl  %%v5,  208(%%r1,%2) \n\t" 
+            "vl  %%v6,  224(%%r1,%2) \n\t" 
+            "vl  %%v7,  240(%%r1,%2) \n\t" 
+            "vst %%v0,  128(%%r1,%1) \n\t" 
+            "vst %%v1,  144(%%r1,%1) \n\t" 
+            "vst %%v2,  160(%%r1,%1) \n\t" 
+            "vst %%v3,  176(%%r1,%1) \n\t"  
+            "vst %%v4,  192(%%r1,%1) \n\t" 
+            "vst %%v5,  208(%%r1,%1) \n\t" 
+            "vst %%v6,  224(%%r1,%1) \n\t" 
+            "vst %%v7,  240(%%r1,%1) \n\t"
+
+            "vst %%v16,   0(%%r1,%2) \n\t" 
+            "vst %%v17,  16(%%r1,%2) \n\t" 
+            "vst %%v18,  32(%%r1,%2) \n\t" 
+            "vst %%v19,  48(%%r1,%2) \n\t"  
+            "vst %%v20,  64(%%r1,%2) \n\t" 
+            "vst %%v21,  80(%%r1,%2) \n\t" 
+            "vst %%v22,  96(%%r1,%2) \n\t" 
+            "vst %%v23, 112(%%r1,%2) \n\t"
+            "vst %%v24, 128(%%r1,%2) \n\t" 
+            "vst %%v25, 144(%%r1,%2) \n\t" 
+            "vst %%v26, 160(%%r1,%2) \n\t" 
+            "vst %%v27, 176(%%r1,%2) \n\t" 
+            "vst %%v28, 192(%%r1,%2) \n\t" 
+            "vst %%v29, 208(%%r1,%2) \n\t" 
+            "vst %%v30, 224(%%r1,%2) \n\t" 
+            "vst %%v31, 240(%%r1,%2) \n\t"
+ 
+          
+            "la %%r1,256(%%r1) \n\t"
+            "brctg %%r0,1b"
+            :
+            : "r"(n), "a"(x), "a"(y)
+            :"cc", "memory","r0","r1", "v0","v1","v2","v3","v4","v5","v6","v7","v16",
+            "v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
+            );
+    return;
+
+}
+
+
+
+ 
+
+
+int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dummy4, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
+{
+    BLASLONG i=0;
+    BLASLONG ix=0,iy=0;
+    FLOAT temp[2];
+    BLASLONG inc_x2, inc_y2;
+
+    if ( n <= 0     )  return(0);
+
+    if ( (inc_x == 1) && (inc_y == 1 ))
+    {
+
+        BLASLONG n1 = n & -16;
+        if ( n1 > 0 )
+        {
+            zswap_kernel_16(n1, x, y);
+            i=n1;
+            ix = 2* n1;
+            iy = 2* n1;
+        }
+
+        while(i < n)
+        {
+
+            temp[0]  = x[ix]   ;
+            temp[1]  = x[ix+1] ;
+            x[ix]    = y[iy]   ;
+            x[ix+1]  = y[iy+1] ;
+            y[iy]    = temp[0] ;
+            y[iy+1]  = temp[1] ;
+
+            ix += 2 ;
+            iy += 2 ;
+            i++ ;
+
+
+        }
+
+
+    }
+    else
+    {
+
+        inc_x2 = 2 * inc_x;
+        inc_y2 = 2 * inc_y;
+
+        while(i < n)
+        {
+
+            temp[0]  = x[ix]   ;
+            temp[1]  = x[ix+1] ;
+            x[ix]    = y[iy]   ;
+            x[ix+1]  = y[iy+1] ;
+            y[iy]    = temp[0] ;
+            y[iy+1]  = temp[1] ;
+
+            ix += inc_x2 ;
+            iy += inc_y2 ;
+            i++ ;
+
+        }
+
+    }
+    return(0);
+    
+
+}
+
+