diff --git a/benchmark/Makefile b/benchmark/Makefile
index db183c8ad..fefd99026 100644
--- a/benchmark/Makefile
+++ b/benchmark/Makefile
@@ -18,12 +18,12 @@ ATLAS=/usr/lib64/atlas
 LIBATLAS = -fopenmp $(ATLAS)/liblapack.a  $(ATLAS)/libptcblas.a  $(ATLAS)/libptf77blas.a  $(ATLAS)/libatlas.a -lgfortran -lm
 
 # Intel standard
-MKL=/opt/intel/mkl/lib/intel64
-LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm
+# MKL=/opt/intel/mkl/lib/intel64
+# LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm
 
 # Intel custom
-#MKL=/home/saar/intel_mkl
-#LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm
+MKL=/home/saar/intel_mkl
+LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm
 
 
 
@@ -34,6 +34,8 @@ goto :: slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
        strsm.goto dtrsm.goto ctrsm.goto ztrsm.goto \
        ssyrk.goto dsyrk.goto csyrk.goto zsyrk.goto \
        ssyr2k.goto dsyr2k.goto csyr2k.goto zsyr2k.goto \
+       sger.goto dger.goto \
+       ssymv.goto dsymv.goto \
        chemm.goto zhemm.goto \
        cherk.goto zherk.goto \
        cher2k.goto zher2k.goto \
@@ -47,6 +49,8 @@ acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
        strsm.acml dtrsm.acml ctrsm.acml ztrsm.acml \
        ssyrk.acml dsyrk.acml csyrk.acml zsyrk.acml \
        ssyr2k.acml dsyr2k.acml csyr2k.acml zsyr2k.acml \
+       sger.acml dger.acml \
+       ssymv.acml dsymv.acml \
        chemm.acml zhemm.acml \
        cherk.acml zherk.acml \
        cher2k.acml zher2k.acml \
@@ -60,6 +64,9 @@ atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
        strsm.atlas dtrsm.atlas ctrsm.atlas ztrsm.atlas \
        ssyrk.atlas dsyrk.atlas csyrk.atlas zsyrk.atlas \
        ssyr2k.atlas dsyr2k.atlas csyr2k.atlas zsyr2k.atlas \
+       sger.atlas dger.atlas \
+       ssymv.atlas dsymv.atlas \
+       chemm.acml zhemm.acml \
        chemm.atlas zhemm.atlas \
        cherk.atlas zherk.atlas \
        cher2k.atlas zher2k.atlas \
@@ -73,6 +80,8 @@ mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
        strsm.mkl dtrsm.mkl ctrsm.mkl ztrsm.mkl \
        ssyrk.mkl dsyrk.mkl csyrk.mkl zsyrk.mkl \
        ssyr2k.mkl dsyr2k.mkl csyr2k.mkl zsyr2k.mkl \
+       sger.mkl dger.mkl \
+       ssymv.mkl dsymv.mkl \
        chemm.mkl zhemm.mkl \
        cherk.mkl zherk.mkl \
        cher2k.mkl zher2k.mkl \
@@ -659,6 +668,58 @@ zgemv.atlas : zgemv.$(SUFFIX)
 zgemv.mkl : zgemv.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+##################################### Sger ####################################################
+sger.goto : sger.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) -lm
+
+sger.acml : sger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sger.atlas : sger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sger.mkl : sger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dger ####################################################
+dger.goto : dger.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) -lm
+
+dger.acml : dger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dger.atlas : dger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dger.mkl : dger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ssymv ####################################################
+ssymv.goto : ssymv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) -lm
+
+ssymv.acml : ssymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssymv.atlas : ssymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssymv.mkl : ssymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dsymv ####################################################
+dsymv.goto : dsymv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) -lm
+
+dsymv.acml : dsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsymv.atlas : dsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsymv.mkl : dsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 
 ###################################################################################################
 
@@ -788,6 +849,18 @@ cgemv.$(SUFFIX) : gemv.c
 zgemv.$(SUFFIX) : gemv.c
 	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
 
+sger.$(SUFFIX) : ger.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dger.$(SUFFIX) : ger.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ssymv.$(SUFFIX) : symv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dsymv.$(SUFFIX) : symv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
 
 clean ::
 	@rm -f *.goto *.mkl *.acml *.atlas
diff --git a/benchmark/ger.c b/benchmark/ger.c
new file mode 100644
index 000000000..5085389da
--- /dev/null
+++ b/benchmark/ger.c
@@ -0,0 +1,218 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef GER
+
+
+#ifdef DOUBLE
+#define GER   BLASFUNC(dger)
+#else
+#define GER   BLASFUNC(sger)
+#endif
+
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int MAIN__(int argc, char *argv[]){
+
+  FLOAT *a, *x, *y;
+  FLOAT alpha[] = {1.0, 1.0};
+  blasint m, i, j;
+  blasint inc_x=1,inc_y=1;
+  blasint n=0;
+  int has_param_n = 0;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+  if ((p = getenv("OPENBLAS_INCY")))   inc_y = atoi(p);
+  if ((p = getenv("OPENBLAS_PARAM_N"))) {
+	  n = atoi(p);
+	  if ((n>0) && (n<=to)) has_param_n = 1;
+  }
+
+  if ( has_param_n == 1 )
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d N = %d Inc_x = %d Inc_y = %d Loops = %d\n", from, to, step,n,inc_x,inc_y,loops);
+  else
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Inc_y = %d Loops = %d\n", from, to, step,inc_x,inc_y,loops);
+
+  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   if ( has_param_n == 0 ) n = m;
+
+   fprintf(stderr, " %6dx%d : ", (int)m,(int)n);
+
+   for(j = 0; j < m; j++){
+      		for(i = 0; i < n * COMPSIZE; i++){
+			a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+      		}
+   }
+
+
+   for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   }
+
+   for(i = 0; i < n * COMPSIZE * abs(inc_y); i++){
+			y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   }
+
+    for (l=0; l<loops; l++)
+    {
+
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	GER (&m, &n, alpha, x, &inc_x, y, &inc_y, a , &m);
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops\n",
+	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)n / timeg * 1.e-6);
+
+  }
+
+  return 0;
+}
+
+void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+
diff --git a/benchmark/plot-filter.sh b/benchmark/plot-filter.sh
new file mode 100755
index 000000000..b47535b6f
--- /dev/null
+++ b/benchmark/plot-filter.sh
@@ -0,0 +1,38 @@
+#!/bin/sh
+# **********************************************************************************
+# Copyright (c) 2014, The OpenBLAS Project
+# All rights reserved.
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+# 1. Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+# 2. Redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in
+# the documentation and/or other materials provided with the
+# distribution.
+# 3. Neither the name of the OpenBLAS project nor the names of
+# its contributors may be used to endorse or promote products
+# derived from this software without specific prior written permission.
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+# USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+# **********************************************************************************
+
+# ************************************************************************
+# sample filter for data output from benchmark programs
+#
+# usage example: 
+#   ./dgemm.goto 2>&1|./plotfilter.sh >OpenBLAS
+# ************************************************************************
+
+
+awk '/MFlops/ { print $1,int($3) }'|tail --lines=+2
+
diff --git a/benchmark/plot-header b/benchmark/plot-header
new file mode 100644
index 000000000..070c10d0a
--- /dev/null
+++ b/benchmark/plot-header
@@ -0,0 +1,42 @@
+# **********************************************************************************
+# Copyright (c) 2014, The OpenBLAS Project
+# All rights reserved.
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+# 1. Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+# 2. Redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in
+# the documentation and/or other materials provided with the
+# distribution.
+# 3. Neither the name of the OpenBLAS project nor the names of
+# its contributors may be used to endorse or promote products
+# derived from this software without specific prior written permission.
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+# USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+# **********************************************************************************
+
+set term x11 font sans;
+set ylabel "MFlops";
+set xlabel "Size";
+set grid xtics;
+set grid ytics;
+set key left;
+set timestamp "generated on %Y-%m-%d by `whoami`"
+set title "Dtrsm\nUPLO=U TRANS=N SIDE=L\nBulldozer 1 Thread"
+plot 'OpenBLAS' smooth bezier, 'ACML' smooth bezier, 'MKL' smooth bezier;
+set output "print.png";
+show title;
+show plot;
+show output;
+
+
diff --git a/benchmark/symv.c b/benchmark/symv.c
new file mode 100644
index 000000000..4bcfb411b
--- /dev/null
+++ b/benchmark/symv.c
@@ -0,0 +1,218 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef SYMV
+
+#ifndef COMPLEX
+
+#ifdef DOUBLE
+#define SYMV   BLASFUNC(dsymv)
+#else
+#define SYMV   BLASFUNC(ssymv)
+#endif
+
+#else
+
+#ifdef DOUBLE
+#define SYMV   BLASFUNC(zsymv)
+#else
+#define SYMV   BLASFUNC(csymv)
+#endif
+
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int MAIN__(int argc, char *argv[]){
+
+  FLOAT *a, *x, *y;
+  FLOAT alpha[] = {1.0, 1.0};
+  FLOAT beta [] = {1.0, 1.0};
+  char uplo='L';
+  blasint m, i, j;
+  blasint inc_x=1,inc_y=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+  if ((p = getenv("OPENBLAS_INCY")))   inc_y = atoi(p);
+  if ((p = getenv("OPENBLAS_UPLO")))  uplo=*p;
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = '%c' Inc_x = %d Inc_y = %d Loops = %d\n", from, to, step,uplo,inc_x,inc_y,loops);
+
+  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6dx%d : ", (int)m,(int)m);
+
+   for(j = 0; j < m; j++){
+      		for(i = 0; i < m * COMPSIZE; i++){
+			a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+      		}
+   }
+
+
+    for (l=0; l<loops; l++)
+    {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
+			y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	SYMV (&uplo, &m, alpha, a, &m, x, &inc_x, beta, y, &inc_y );
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops\n",
+	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m / timeg * 1.e-6);
+
+  }
+
+  return 0;
+}
+
+void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/common_param.h b/common_param.h
index 863216406..1c362e8cb 100644
--- a/common_param.h
+++ b/common_param.h
@@ -919,14 +919,22 @@ extern gotoblas_t *gotoblas;
 #define	SGEMM_R		SGEMM_DEFAULT_R
 #define SGEMM_UNROLL_M	SGEMM_DEFAULT_UNROLL_M
 #define SGEMM_UNROLL_N	SGEMM_DEFAULT_UNROLL_N
+#ifdef  SGEMM_DEFAULT_UNROLL_MN
+#define SGEMM_UNROLL_MN	SGEMM_DEFAULT_UNROLL_MN
+#else
 #define SGEMM_UNROLL_MN	MAX((SGEMM_UNROLL_M), (SGEMM_UNROLL_N))
+#endif
 
 #define	DGEMM_P		DGEMM_DEFAULT_P
 #define	DGEMM_Q		DGEMM_DEFAULT_Q
 #define	DGEMM_R		DGEMM_DEFAULT_R
 #define DGEMM_UNROLL_M	DGEMM_DEFAULT_UNROLL_M
 #define DGEMM_UNROLL_N	DGEMM_DEFAULT_UNROLL_N
+#ifdef  DGEMM_DEFAULT_UNROLL_MN
+#define DGEMM_UNROLL_MN	DGEMM_DEFAULT_UNROLL_MN
+#else
 #define DGEMM_UNROLL_MN	MAX((DGEMM_UNROLL_M), (DGEMM_UNROLL_N))
+#endif
 
 #define	QGEMM_P		QGEMM_DEFAULT_P
 #define	QGEMM_Q		QGEMM_DEFAULT_Q
@@ -940,14 +948,22 @@ extern gotoblas_t *gotoblas;
 #define	CGEMM_R		CGEMM_DEFAULT_R
 #define CGEMM_UNROLL_M	CGEMM_DEFAULT_UNROLL_M
 #define CGEMM_UNROLL_N	CGEMM_DEFAULT_UNROLL_N
+#ifdef  CGEMM_DEFAULT_UNROLL_MN
+#define CGEMM_UNROLL_MN	CGEMM_DEFAULT_UNROLL_MN
+#else
 #define CGEMM_UNROLL_MN	MAX((CGEMM_UNROLL_M), (CGEMM_UNROLL_N))
+#endif
 
 #define	ZGEMM_P		ZGEMM_DEFAULT_P
 #define	ZGEMM_Q		ZGEMM_DEFAULT_Q
 #define	ZGEMM_R		ZGEMM_DEFAULT_R
 #define ZGEMM_UNROLL_M	ZGEMM_DEFAULT_UNROLL_M
 #define ZGEMM_UNROLL_N	ZGEMM_DEFAULT_UNROLL_N
+#ifdef  ZGEMM_DEFAULT_UNROLL_MN
+#define ZGEMM_UNROLL_MN	ZGEMM_DEFAULT_UNROLL_MN
+#else
 #define ZGEMM_UNROLL_MN	MAX((ZGEMM_UNROLL_M), (ZGEMM_UNROLL_N))
+#endif
 
 #define	XGEMM_P		XGEMM_DEFAULT_P
 #define	XGEMM_Q		XGEMM_DEFAULT_Q
diff --git a/driver/level3/level3.c b/driver/level3/level3.c
index 261204099..70a6500b6 100644
--- a/driver/level3/level3.c
+++ b/driver/level3/level3.c
@@ -333,16 +333,10 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
 
-#if ( defined(BULLDOZER) || defined(PILEDRIVER) || defined(HASWELL) ) && defined(ARCH_X86_64) && !defined(XDOUBLE) && !defined(COMPLEX)
-                if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
-                else
-                        if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
-                        else
-                                if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
-#else
+        if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
+        else
+          if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
-        if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
-#endif
 
 
 	START_RPCC();
diff --git a/driver/level3/level3_syrk_threaded.c b/driver/level3/level3_syrk_threaded.c
index 01c7b23ed..5119baad3 100644
--- a/driver/level3/level3_syrk_threaded.c
+++ b/driver/level3/level3_syrk_threaded.c
@@ -538,10 +538,10 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   mask  = MAX(QGEMM_UNROLL_M, QGEMM_UNROLL_N) - 1;
 #elif defined(DOUBLE)
   mode  =  BLAS_DOUBLE  | BLAS_REAL;
-  mask  = MAX(DGEMM_UNROLL_M, DGEMM_UNROLL_N) - 1;
+  mask  = DGEMM_UNROLL_MN - 1;
 #else
   mode  =  BLAS_SINGLE  | BLAS_REAL;
-  mask  = MAX(SGEMM_UNROLL_M, SGEMM_UNROLL_N) - 1;
+  mask  = SGEMM_UNROLL_MN - 1;
 #endif
 #else
 #ifdef XDOUBLE
@@ -549,10 +549,10 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   mask  = MAX(XGEMM_UNROLL_M, XGEMM_UNROLL_N) - 1;
 #elif defined(DOUBLE)
   mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
-  mask  = MAX(ZGEMM_UNROLL_M, ZGEMM_UNROLL_N) - 1;
+  mask  = ZGEMM_UNROLL_MN - 1;
 #else
   mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
-  mask  = MAX(CGEMM_UNROLL_M, CGEMM_UNROLL_N) - 1;
+  mask  = CGEMM_UNROLL_MN - 1;
 #endif
 #endif
 
diff --git a/driver/level3/level3_thread.c b/driver/level3/level3_thread.c
index 95860d0c0..6162a9f0d 100644
--- a/driver/level3/level3_thread.c
+++ b/driver/level3/level3_thread.c
@@ -367,16 +367,10 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
       for(jjs = xxx; jjs < MIN(n_to, xxx + div_n); jjs += min_jj){
 	min_jj = MIN(n_to, xxx + div_n) - jjs;
 
-#if ( defined(BULLDOZER) || defined(PILEDRIVER) || defined(HASWELL) ) && defined(ARCH_X86_64) && !defined(XDOUBLE) && !defined(COMPLEX)
-		if (min_jj >= 6*GEMM_UNROLL_N) min_jj = 6*GEMM_UNROLL_N;
-		else
-			if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
-			else
-				if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
-#else
+	if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
+	else
+		if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
-#endif
 
 	START_RPCC();
 
diff --git a/driver/level3/syrk_thread.c b/driver/level3/syrk_thread.c
index 0d9bdf209..94274be72 100644
--- a/driver/level3/syrk_thread.c
+++ b/driver/level3/syrk_thread.c
@@ -57,10 +57,10 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
 
     switch (mode & BLAS_PREC) {
     case BLAS_SINGLE:
-      mask = MAX(SGEMM_UNROLL_M, SGEMM_UNROLL_N) - 1;
+      mask = SGEMM_UNROLL_MN - 1;
       break;
     case BLAS_DOUBLE:
-      mask = MAX(DGEMM_UNROLL_M, DGEMM_UNROLL_N) - 1;
+      mask = DGEMM_UNROLL_MN - 1;
       break;
 #ifdef EXPRECISION
     case BLAS_XDOUBLE:
@@ -71,10 +71,10 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
   } else {
     switch (mode & BLAS_PREC) {
     case BLAS_SINGLE:
-      mask = MAX(CGEMM_UNROLL_M, CGEMM_UNROLL_N) - 1;
+      mask = CGEMM_UNROLL_MN - 1;
       break;
     case BLAS_DOUBLE:
-      mask = MAX(ZGEMM_UNROLL_M, ZGEMM_UNROLL_N) - 1;
+      mask = ZGEMM_UNROLL_MN - 1;
       break;
 #ifdef EXPRECISION
     case BLAS_XDOUBLE:
diff --git a/driver/level3/trmm_L.c b/driver/level3/trmm_L.c
index c0a822b51..8a81d31a0 100644
--- a/driver/level3/trmm_L.c
+++ b/driver/level3/trmm_L.c
@@ -135,7 +135,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
     for(jjs = js; jjs < js + min_j; jjs += min_jj){
       min_jj = min_j + js - jjs;
-      if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+      if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+      else
+      	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
       START_RPCC();
 
@@ -199,7 +201,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+        if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+        else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 	START_RPCC();
 
@@ -288,7 +292,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
     for(jjs = js; jjs < js + min_j; jjs += min_jj){
       min_jj = min_j + js - jjs;
-      if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+      if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+      else
+        if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
       START_RPCC();
 
@@ -352,7 +358,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+        if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+        else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 	START_RPCC();
 
diff --git a/driver/level3/trmm_R.c b/driver/level3/trmm_R.c
index 6012386c8..bdd9370cd 100644
--- a/driver/level3/trmm_R.c
+++ b/driver/level3/trmm_R.c
@@ -119,7 +119,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < ls - js; jjs += min_jj){
 	min_jj = ls - js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	GEMM_ONCOPY(min_l, min_jj, a + (ls + (js + jjs) * lda) * COMPSIZE, lda, sb + min_l * jjs * COMPSIZE);
@@ -137,7 +139,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < min_l; jjs += min_jj){
 	min_jj = min_l - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	TRMM_OLNCOPY(min_l, min_jj, a, lda, ls, ls + jjs, sb + min_l * (ls - js + jjs) * COMPSIZE);
@@ -188,7 +192,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	GEMM_ONCOPY(min_l, min_jj, a + (ls + jjs * lda) * COMPSIZE, lda, sb + min_l * (jjs - js) * COMPSIZE);
@@ -239,7 +245,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < min_l; jjs += min_jj){
 	min_jj = min_l - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	TRMM_OUNCOPY(min_l, min_jj, a, lda, ls, ls + jjs, sb + min_l * jjs * COMPSIZE);
@@ -258,7 +266,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < js - ls - min_l; jjs += min_jj){
 	min_jj = js - ls - min_l - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	GEMM_ONCOPY(min_l, min_jj, a + (ls + (ls + min_l + jjs) * lda) * COMPSIZE, lda,
@@ -313,7 +323,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	GEMM_ONCOPY(min_l, min_jj, a + (ls + (jjs - min_j) * lda) * COMPSIZE, lda, sb + min_l * (jjs - js) * COMPSIZE);
diff --git a/driver/level3/trsm_L.c b/driver/level3/trsm_L.c
index fa3b0d580..78da0eb6c 100644
--- a/driver/level3/trsm_L.c
+++ b/driver/level3/trsm_L.c
@@ -128,7 +128,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 	GEMM_ONCOPY(min_l, min_jj, b + (ls + jjs * ldb) * COMPSIZE, ldb, sb + min_l * (jjs - js) * COMPSIZE);
 
@@ -194,7 +196,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 	GEMM_ONCOPY(min_l, min_jj, b + (ls - min_l + jjs * ldb) * COMPSIZE, ldb, sb + min_l * (jjs - js) * COMPSIZE);
 
diff --git a/driver/level3/trsm_R.c b/driver/level3/trsm_R.c
index b6ee95654..169441d1e 100644
--- a/driver/level3/trsm_R.c
+++ b/driver/level3/trsm_R.c
@@ -123,7 +123,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	GEMM_ONCOPY(min_l, min_jj, a + (ls + jjs * lda) * COMPSIZE, lda, sb + min_l * (jjs - js) * COMPSIZE);
@@ -177,7 +179,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < min_j - min_l - ls + js; jjs += min_jj){
 	min_jj = min_j - min_l - ls + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
       GEMM_ONCOPY (min_l, min_jj, a + (ls + (ls + min_l + jjs) * lda) * COMPSIZE, lda,
@@ -238,7 +242,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = js; jjs < js + min_j; jjs += min_jj){
 	min_jj = min_j + js - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	GEMM_ONCOPY(min_l, min_jj, a + (ls + (jjs - min_j) * lda) * COMPSIZE, lda, sb + min_l * (jjs - js) * COMPSIZE);
@@ -297,7 +303,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
       for(jjs = 0; jjs < min_j - js + ls; jjs += min_jj){
 	min_jj = min_j - js + ls - jjs;
-	if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+	if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
+	else
+	  if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
 
 #ifndef TRANSA
 	GEMM_ONCOPY (min_l, min_jj, a + (ls + (js - min_j + jjs) * lda) * COMPSIZE, lda,
diff --git a/kernel/setparam-ref.c b/kernel/setparam-ref.c
index 5086420c1..b1beeae5c 100644
--- a/kernel/setparam-ref.c
+++ b/kernel/setparam-ref.c
@@ -54,7 +54,14 @@ gotoblas_t TABLE_NAME = {
   GEMM_DEFAULT_OFFSET_A, GEMM_DEFAULT_OFFSET_B, GEMM_DEFAULT_ALIGN,
 
   0, 0, 0,
-  SGEMM_DEFAULT_UNROLL_M, SGEMM_DEFAULT_UNROLL_N, MAX(SGEMM_DEFAULT_UNROLL_M, SGEMM_DEFAULT_UNROLL_N),
+  SGEMM_DEFAULT_UNROLL_M, SGEMM_DEFAULT_UNROLL_N,
+#ifdef SGEMM_DEFAULT_UNROLL_MN
+ SGEMM_DEFAULT_UNROLL_MN,
+#else
+ MAX(SGEMM_DEFAULT_UNROLL_M, SGEMM_DEFAULT_UNROLL_N),
+#endif
+
+
 #ifdef HAVE_EXCLUSIVE_CACHE
   1,
 #else
@@ -110,7 +117,12 @@ gotoblas_t TABLE_NAME = {
 #endif
 
   0, 0, 0,
-  DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N, MAX(DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N),
+  DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N,
+#ifdef DGEMM_DEFAULT_UNROLL_MN
+ DGEMM_DEFAULT_UNROLL_MN,
+#else
+ MAX(DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N),
+#endif
 
   damax_kTS,  damin_kTS,  dmax_kTS,  dmin_kTS,
   idamax_kTS, idamin_kTS, idmax_kTS, idmin_kTS,
@@ -214,7 +226,12 @@ gotoblas_t TABLE_NAME = {
 #endif
 
   0, 0, 0,
-  CGEMM_DEFAULT_UNROLL_M, CGEMM_DEFAULT_UNROLL_N, MAX(CGEMM_DEFAULT_UNROLL_M, CGEMM_DEFAULT_UNROLL_N),
+  CGEMM_DEFAULT_UNROLL_M, CGEMM_DEFAULT_UNROLL_N,
+#ifdef CGEMM_DEFAULT_UNROLL_MN
+ CGEMM_DEFAULT_UNROLL_MN,
+#else
+ MAX(CGEMM_DEFAULT_UNROLL_M, CGEMM_DEFAULT_UNROLL_N),
+#endif
 
   camax_kTS, camin_kTS, icamax_kTS, icamin_kTS,
   cnrm2_kTS, casum_kTS, ccopy_kTS,
@@ -307,7 +324,12 @@ gotoblas_t TABLE_NAME = {
 #endif
 
   0, 0, 0,
-  ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N, MAX(ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N),
+  ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N,
+#ifdef ZGEMM_DEFAULT_UNROLL_MN
+ ZGEMM_DEFAULT_UNROLL_MN,
+#else
+ MAX(ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N),
+#endif
 
   zamax_kTS, zamin_kTS, izamax_kTS, izamin_kTS,
   znrm2_kTS, zasum_kTS, zcopy_kTS,
diff --git a/kernel/x86_64/zgemm_kernel_4x2_haswell.S b/kernel/x86_64/zgemm_kernel_4x2_haswell.S
index a71fff7af..e23e09ecc 100644
--- a/kernel/x86_64/zgemm_kernel_4x2_haswell.S
+++ b/kernel/x86_64/zgemm_kernel_4x2_haswell.S
@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 **********************************************************************************/
 
 /********************************************************************************
-* 2014/06/28 Saar
+* 2014/07/28 Saar
 *        BLASTEST               : OK
 *        CTEST                  : OK
 *        TEST                   : OK
@@ -40,12 +40,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *	A_PR1			512
 *	B_PR1			512
 *
-*
+* 2014/07/28 Saar
 * Performance at 4608x4608x4608:
-*       1 thread:       43 GFLOPS	(SANDYBRIDGE:  29)	(MKL:   53)
-*       2 threads:      85 GFLOPS	(SANDYBRIDGE:  59)	(MKL:  100)
-*       3 threads:     122 GFLOPS	(SANDYBRIDGE:  86)	(MKL:  138)
-*       4 threads:     156 GFLOPS	(SANDYBRIDGE: 108)	(MKL:  172)
+*       1 thread:       53 GFLOPS	(SANDYBRIDGE:  29)	(MKL:   53)
+*       2 threads:     101 GFLOPS	(SANDYBRIDGE:  59)	(MKL:  100)
+*       3 threads:     146 GFLOPS	(SANDYBRIDGE:  86)	(MKL:  138)
+*       4 threads:     184 GFLOPS	(SANDYBRIDGE: 108)	(MKL:  172)
 *
 ********************************************************************************/
 
@@ -191,6 +191,379 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define	A_PR1	512
 #define	B_PR1	512
+
+
+
+/***************************************************************************************************/
+
+.macro KERNEL4x3_SUB
+        vmovups                   (AO), %ymm0
+        vmovups           4 * SIZE(AO), %ymm1
+	prefetcht0	  A_PR1(AO)
+
+        vbroadcastsd              (BO),   %ymm2
+        vbroadcastsd      1 * SIZE(BO),   %ymm3
+        VFMADDPD_R(        %ymm8 ,%ymm2,%ymm0 )
+        VFMADDPD_R(        %ymm12,%ymm2,%ymm1 )
+        VFMADDPD_I(        %ymm9 ,%ymm3,%ymm0 )
+        VFMADDPD_I(        %ymm13,%ymm3,%ymm1 )
+
+        vbroadcastsd      2 * SIZE(BO),   %ymm2
+        vbroadcastsd      3 * SIZE(BO),   %ymm3
+        VFMADDPD_R(        %ymm10,%ymm2,%ymm0 )
+        VFMADDPD_R(        %ymm14,%ymm2,%ymm1 )
+        VFMADDPD_I(        %ymm11,%ymm3,%ymm0 )
+        VFMADDPD_I(        %ymm15,%ymm3,%ymm1 )
+
+        vbroadcastsd      4 * SIZE(BO),   %ymm2
+        vbroadcastsd      5 * SIZE(BO),   %ymm3
+        VFMADDPD_R(        %ymm4 ,%ymm2,%ymm0 )
+        VFMADDPD_R(        %ymm6 ,%ymm2,%ymm1 )
+        VFMADDPD_I(        %ymm5 ,%ymm3,%ymm0 )
+        VFMADDPD_I(        %ymm7 ,%ymm3,%ymm1 )
+
+        addq    $6*SIZE, BO                           
+        addq    $8*SIZE, AO                           
+        decq	%rax                         
+.endm
+
+.macro SAVE4x3
+
+	vbroadcastsd	ALPHA_R, %ymm0
+	vbroadcastsd	ALPHA_I, %ymm1
+
+	// swap high and low 8 bytes
+        vshufpd $ 0x05, %ymm9 , %ymm9, %ymm9
+        vshufpd $ 0x05, %ymm11, %ymm11, %ymm11
+        vshufpd $ 0x05, %ymm13, %ymm13, %ymm13
+        vshufpd $ 0x05, %ymm15, %ymm15, %ymm15
+        vshufpd $ 0x05, %ymm5 , %ymm5 , %ymm5
+        vshufpd $ 0x05, %ymm7 , %ymm7 , %ymm7
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubpd %ymm9, %ymm8 , %ymm8
+        vaddsubpd %ymm11,%ymm10, %ymm10
+        vaddsubpd %ymm13,%ymm12, %ymm12
+        vaddsubpd %ymm15,%ymm14, %ymm14
+        vaddsubpd %ymm5 ,%ymm4 , %ymm4
+        vaddsubpd %ymm7 ,%ymm6 , %ymm6
+
+        vshufpd $ 0x05, %ymm8 , %ymm8 , %ymm9
+        vshufpd $ 0x05, %ymm10, %ymm10, %ymm11
+        vshufpd $ 0x05, %ymm12, %ymm12, %ymm13
+        vshufpd $ 0x05, %ymm14, %ymm14, %ymm15
+        vshufpd $ 0x05, %ymm4 , %ymm4 , %ymm5
+        vshufpd $ 0x05, %ymm6 , %ymm6 , %ymm7
+
+#else
+        vaddsubpd %ymm8,  %ymm9 ,%ymm9
+        vaddsubpd %ymm10, %ymm11,%ymm11
+        vaddsubpd %ymm12, %ymm13,%ymm13
+        vaddsubpd %ymm14, %ymm15,%ymm15
+        vaddsubpd %ymm4 , %ymm5 ,%ymm5
+        vaddsubpd %ymm6 , %ymm7 ,%ymm7
+
+        vmovapd   %ymm9,  %ymm8
+        vmovapd   %ymm11, %ymm10
+        vmovapd   %ymm13, %ymm12
+        vmovapd   %ymm15, %ymm14
+        vmovapd   %ymm5 , %ymm4
+        vmovapd   %ymm7 , %ymm6
+
+	// swap high and low 8 bytes
+        vshufpd $ 0x05, %ymm9 , %ymm9, %ymm9
+        vshufpd $ 0x05, %ymm11, %ymm11, %ymm11
+        vshufpd $ 0x05, %ymm13, %ymm13, %ymm13
+        vshufpd $ 0x05, %ymm15, %ymm15, %ymm15
+        vshufpd $ 0x05, %ymm5 , %ymm5 , %ymm5
+        vshufpd $ 0x05, %ymm7 , %ymm7 , %ymm7
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulpd  %ymm8 , %ymm0, %ymm8
+        vmulpd  %ymm10, %ymm0, %ymm10
+        vmulpd  %ymm12, %ymm0, %ymm12
+        vmulpd  %ymm14, %ymm0, %ymm14
+        vmulpd  %ymm4 , %ymm0, %ymm4
+        vmulpd  %ymm6 , %ymm0, %ymm6
+
+	// multiply with ALPHA_I
+        vmulpd  %ymm9 , %ymm1, %ymm9
+        vmulpd  %ymm11, %ymm1, %ymm11
+        vmulpd  %ymm13, %ymm1, %ymm13
+        vmulpd  %ymm15, %ymm1, %ymm15
+        vmulpd  %ymm5 , %ymm1, %ymm5
+        vmulpd  %ymm7 , %ymm1, %ymm7
+
+	vaddsubpd %ymm9, %ymm8 , %ymm8
+        vaddsubpd %ymm11,%ymm10, %ymm10
+        vaddsubpd %ymm13,%ymm12, %ymm12
+        vaddsubpd %ymm15,%ymm14, %ymm14
+        vaddsubpd %ymm5 ,%ymm4 , %ymm4
+        vaddsubpd %ymm7 ,%ymm6 , %ymm6
+
+
+
+#ifndef TRMMKERNEL
+
+	vaddpd 	 	(CO1), %ymm8 , %ymm8
+	vaddpd  4 * SIZE(CO1), %ymm12, %ymm12
+
+	vaddpd 	 	(CO1, LDC), %ymm10, %ymm10
+	vaddpd  4 * SIZE(CO1, LDC), %ymm14, %ymm14
+
+	vaddpd 	 	(CO1, LDC,2), %ymm4 , %ymm4
+	vaddpd  4 * SIZE(CO1, LDC,2), %ymm6 , %ymm6
+#endif
+
+	vmovups	%ymm8 ,  	 (CO1)
+	vmovups	%ymm12 , 4 * SIZE(CO1)
+
+	vmovups	%ymm10 ,  	 (CO1, LDC)
+	vmovups	%ymm14 , 4 * SIZE(CO1, LDC)
+
+	vmovups	%ymm4  ,  	 (CO1, LDC, 2)
+	vmovups	%ymm6  , 4 * SIZE(CO1, LDC, 2)
+
+	prefetcht0	64(CO1)
+	prefetcht0	64(CO1, LDC)
+
+.endm
+
+
+
+/***************************************************************************************************/
+
+.macro KERNEL2x3_SUB
+        vmovups                  (AO), %xmm0
+        vmovups          2 * SIZE(AO), %xmm1
+        vmovddup                 (BO), %xmm2
+        vmovddup         1 * SIZE(BO), %xmm3
+
+        VFMADDPD_R(        %xmm8 ,%xmm2,%xmm0 )
+        VFMADDPD_R(        %xmm12,%xmm2,%xmm1 )
+        VFMADDPD_I(        %xmm9 ,%xmm3,%xmm0 )
+        VFMADDPD_I(        %xmm13,%xmm3,%xmm1 )
+
+        vmovddup         2 * SIZE(BO), %xmm2
+        vmovddup         3 * SIZE(BO), %xmm3
+        VFMADDPD_R(        %xmm10,%xmm2,%xmm0 )
+        VFMADDPD_R(        %xmm14,%xmm2,%xmm1 )
+        VFMADDPD_I(        %xmm11,%xmm3,%xmm0 )
+        VFMADDPD_I(        %xmm15,%xmm3,%xmm1 )
+
+        vmovddup         4 * SIZE(BO), %xmm2
+        vmovddup         5 * SIZE(BO), %xmm3
+        VFMADDPD_R(        %xmm4 ,%xmm2,%xmm0 )
+        VFMADDPD_R(        %xmm6 ,%xmm2,%xmm1 )
+        VFMADDPD_I(        %xmm5 ,%xmm3,%xmm0 )
+        VFMADDPD_I(        %xmm7 ,%xmm3,%xmm1 )
+
+        addq    $6*SIZE, BO                           
+        addq    $4*SIZE, AO                           
+        decq    %rax                         
+.endm
+
+.macro SAVE2x3
+
+	vmovddup	ALPHA_R, %xmm0
+	vmovddup	ALPHA_I, %xmm1
+
+	// swap high and low 64 bytes
+        vshufpd $ 0x01, %xmm9 , %xmm9, %xmm9
+        vshufpd $ 0x01, %xmm11, %xmm11, %xmm11
+        vshufpd $ 0x01, %xmm13, %xmm13, %xmm13
+        vshufpd $ 0x01, %xmm15, %xmm15, %xmm15
+        vshufpd $ 0x01, %xmm5 , %xmm5 , %xmm5
+        vshufpd $ 0x01, %xmm7 , %xmm7 , %xmm7
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubpd %xmm9, %xmm8 , %xmm8
+        vaddsubpd %xmm11,%xmm10, %xmm10
+        vaddsubpd %xmm13,%xmm12, %xmm12
+        vaddsubpd %xmm15,%xmm14, %xmm14
+        vaddsubpd %xmm5, %xmm4 , %xmm4
+        vaddsubpd %xmm7, %xmm6 , %xmm6
+
+        vshufpd $ 0x01, %xmm8 , %xmm8, %xmm9
+        vshufpd $ 0x01, %xmm10, %xmm10, %xmm11
+        vshufpd $ 0x01, %xmm12, %xmm12, %xmm13
+        vshufpd $ 0x01, %xmm14, %xmm14, %xmm15
+        vshufpd $ 0x01, %xmm4 , %xmm4, %xmm5
+        vshufpd $ 0x01, %xmm6 , %xmm6, %xmm7
+
+#else
+        vaddsubpd %xmm8,  %xmm9 ,%xmm9
+        vaddsubpd %xmm10, %xmm11,%xmm11
+        vaddsubpd %xmm12, %xmm13,%xmm13
+        vaddsubpd %xmm14, %xmm15,%xmm15
+        vaddsubpd %xmm4,  %xmm5 ,%xmm5
+        vaddsubpd %xmm6,  %xmm7 ,%xmm7
+
+        vmovapd   %xmm9,  %xmm8
+        vmovapd   %xmm11, %xmm10
+        vmovapd   %xmm13, %xmm12
+        vmovapd   %xmm15, %xmm14
+        vmovapd   %xmm5,  %xmm4
+        vmovapd   %xmm7,  %xmm6
+
+	// swap high and low 64 bytes
+        vshufpd $ 0x01, %xmm9 , %xmm9, %xmm9
+        vshufpd $ 0x01, %xmm11, %xmm11, %xmm11
+        vshufpd $ 0x01, %xmm13, %xmm13, %xmm13
+        vshufpd $ 0x01, %xmm15, %xmm15, %xmm15
+        vshufpd $ 0x01, %xmm5 , %xmm5, %xmm5
+        vshufpd $ 0x01, %xmm7 , %xmm7, %xmm7
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulpd  %xmm8 , %xmm0, %xmm8
+        vmulpd  %xmm10, %xmm0, %xmm10
+        vmulpd  %xmm12, %xmm0, %xmm12
+        vmulpd  %xmm14, %xmm0, %xmm14
+        vmulpd  %xmm4 , %xmm0, %xmm4
+        vmulpd  %xmm6 , %xmm0, %xmm6
+
+	// multiply with ALPHA_I
+        vmulpd  %xmm9 , %xmm1, %xmm9
+        vmulpd  %xmm11, %xmm1, %xmm11
+        vmulpd  %xmm13, %xmm1, %xmm13
+        vmulpd  %xmm15, %xmm1, %xmm15
+        vmulpd  %xmm5 , %xmm1, %xmm5
+        vmulpd  %xmm7 , %xmm1, %xmm7
+
+	vaddsubpd %xmm9, %xmm8 , %xmm8
+        vaddsubpd %xmm11,%xmm10, %xmm10
+        vaddsubpd %xmm13,%xmm12, %xmm12
+        vaddsubpd %xmm15,%xmm14, %xmm14
+	vaddsubpd %xmm5, %xmm4 , %xmm4
+	vaddsubpd %xmm7, %xmm6 , %xmm6
+
+#ifndef TRMMKERNEL
+
+	vaddpd 	 	(CO1), %xmm8 , %xmm8
+	vaddpd  2 * SIZE(CO1), %xmm12, %xmm12
+
+	vaddpd 	 	(CO1, LDC), %xmm10, %xmm10
+	vaddpd  2 * SIZE(CO1, LDC), %xmm14, %xmm14
+
+	vaddpd 	 	(CO1, LDC,2), %xmm4 , %xmm4
+	vaddpd  2 * SIZE(CO1, LDC,2), %xmm6 , %xmm6
+
+#endif
+
+	vmovups	%xmm8 ,  	(CO1)
+	vmovups	%xmm12 , 2 * SIZE(CO1)
+
+	vmovups	%xmm10 ,  	(CO1, LDC)
+	vmovups	%xmm14 , 2 * SIZE(CO1, LDC)
+
+	vmovups	%xmm4  ,  	(CO1, LDC,2)
+	vmovups	%xmm6  , 2 * SIZE(CO1, LDC,2)
+
+.endm
+
+
+/************************************************************************************************/
+
+
+.macro KERNEL1x3_SUB
+        vmovups                  (AO), %xmm0
+        vmovddup                 (BO), %xmm2
+        vmovddup         1 * SIZE(BO), %xmm3
+
+        VFMADDPD_R(        %xmm8,%xmm2,%xmm0 )
+        VFMADDPD_I(        %xmm9,%xmm3,%xmm0 )
+
+        vmovddup         2 * SIZE(BO), %xmm2
+        vmovddup         3 * SIZE(BO), %xmm3
+        VFMADDPD_R(        %xmm10,%xmm2,%xmm0 )
+        VFMADDPD_I(        %xmm11,%xmm3,%xmm0 )
+
+        vmovddup         4 * SIZE(BO), %xmm2
+        vmovddup         5 * SIZE(BO), %xmm3
+        VFMADDPD_R(        %xmm4 ,%xmm2,%xmm0 )
+        VFMADDPD_I(        %xmm5 ,%xmm3,%xmm0 )
+
+        addq    $6*SIZE, BO                           
+        addq    $2*SIZE, AO                           
+        decq    %rax                         
+.endm
+
+.macro SAVE1x3
+
+	vmovddup	ALPHA_R, %xmm0
+	vmovddup	ALPHA_I, %xmm1
+
+	// swap high and low 64 bytes
+        vshufpd $ 0x01, %xmm9 , %xmm9, %xmm9
+        vshufpd $ 0x01, %xmm11, %xmm11, %xmm11
+        vshufpd $ 0x01, %xmm5 , %xmm5, %xmm5
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubpd %xmm9, %xmm8 , %xmm8
+        vaddsubpd %xmm11,%xmm10, %xmm10
+        vaddsubpd %xmm5, %xmm4 , %xmm4
+
+        vshufpd $ 0x01, %xmm8 , %xmm8, %xmm9
+        vshufpd $ 0x01, %xmm10, %xmm10, %xmm11
+        vshufpd $ 0x01, %xmm4 , %xmm4, %xmm5
+
+#else
+        vaddsubpd %xmm8, %xmm9, %xmm9
+        vaddsubpd %xmm10,%xmm11, %xmm11
+        vaddsubpd %xmm4, %xmm5, %xmm5
+
+        vmovapd   %xmm9,  %xmm8
+        vmovapd   %xmm11, %xmm10
+        vmovapd   %xmm5,  %xmm4
+
+	// swap high and low 64 bytes
+        vshufpd $ 0x01, %xmm9 , %xmm9, %xmm9
+        vshufpd $ 0x01, %xmm11, %xmm11, %xmm11
+        vshufpd $ 0x01, %xmm5 , %xmm5, %xmm5
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulpd  %xmm8 , %xmm0, %xmm8
+        vmulpd  %xmm10, %xmm0, %xmm10
+        vmulpd  %xmm4 , %xmm0, %xmm4
+
+	// multiply with ALPHA_I
+        vmulpd  %xmm9 , %xmm1, %xmm9
+        vmulpd  %xmm11, %xmm1, %xmm11
+        vmulpd  %xmm5 , %xmm1, %xmm5
+
+	vaddsubpd %xmm9, %xmm8 , %xmm8
+        vaddsubpd %xmm11,%xmm10, %xmm10
+	vaddsubpd %xmm5, %xmm4 , %xmm4
+
+#ifndef TRMMKERNEL
+
+	vaddpd 	 	(CO1)        , %xmm8 , %xmm8
+	vaddpd 	 	(CO1, LDC)   , %xmm10, %xmm10
+	vaddpd 	 	(CO1, LDC,2) , %xmm4 , %xmm4
+
+#endif
+
+	vmovups	%xmm8 ,  	(CO1)
+	vmovups	%xmm10 ,  	(CO1, LDC)
+	vmovups	%xmm4  ,  	(CO1, LDC,2)
+
+.endm
+
+
+
+
 /***************************************************************************************************/
 
 .macro KERNEL4x2_SUB
@@ -676,6 +1049,1697 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 
+#if !defined(TRMMKERNEL)
+
+
+	PROLOGUE
+	PROFCODE
+	
+	subq	$ STACKSIZE, %rsp
+	movq	%rbx,   (%rsp)
+	movq	%rbp,  8(%rsp)
+	movq	%r12, 16(%rsp)
+	movq	%r13, 24(%rsp)
+	movq	%r14, 32(%rsp)
+	movq	%r15, 40(%rsp)
+
+	vzeroupper
+
+#ifdef WINDOWS_ABI
+	movq	%rdi,    48(%rsp)
+	movq	%rsi,    56(%rsp)
+	vmovups	%xmm6,   64(%rsp)
+	vmovups	%xmm7,   80(%rsp)
+	vmovups	%xmm8,   96(%rsp)
+	vmovups	%xmm9,  112(%rsp)
+	vmovups	%xmm10, 128(%rsp)
+	vmovups	%xmm11, 144(%rsp)
+	vmovups	%xmm12, 160(%rsp)
+	vmovups	%xmm13, 176(%rsp)
+	vmovups	%xmm14, 192(%rsp)
+	vmovups	%xmm15, 208(%rsp)
+
+	movq	ARG1,      OLD_M
+	movq	ARG2,      OLD_N
+	movq	ARG3,      OLD_K
+	movq	OLD_A,     A
+	movq	OLD_B,     B
+	movq	OLD_C,     C
+	movq	OLD_LDC,   LDC
+#ifdef TRMMKERNEL
+	movsd	OLD_OFFSET, %xmm12
+#endif
+	vmovaps	%xmm3, %xmm0
+	vmovsd   OLD_ALPHA_I, %xmm1
+
+#else
+	movq	STACKSIZE +  8(%rsp), LDC
+#ifdef TRMMKERNEL
+	movsd	STACKSIZE + 16(%rsp), %xmm12
+#endif
+
+#endif
+
+	movq    %rsp, SP      # save old stack
+        subq    $ 128 + L_BUFFER_SIZE, %rsp
+        andq    $ -4096, %rsp    # align stack
+
+        STACK_TOUCH
+
+	cmpq	$ 0, OLD_M
+	je	.L999
+
+	cmpq	$ 0, OLD_N
+	je	.L999
+
+	cmpq	$ 0, OLD_K
+	je	.L999
+
+	movq	OLD_M, M
+	movq	OLD_N, N
+	movq	OLD_K, K
+
+	vmovsd	 %xmm0, ALPHA_R
+	vmovsd	 %xmm1, ALPHA_I
+
+	salq	$ ZBASE_SHIFT, LDC
+
+	movq    N, %rax
+        xorq    %rdx, %rdx
+        movq    $ 6,  %rdi
+        divq    %rdi                    //    N / 6
+        movq    %rax, Ndiv6             //    N / 6
+        movq    %rdx, Nmod6             //    N % 6
+
+	
+
+/************************************************************************************************/
+.L6_00_0:
+
+	movq	Ndiv6,  J
+	cmpq	$ 0, J
+	je	.L2_00_0
+	ALIGN_4
+
+
+
+.L6_00_01:
+	// copy to sub buffer
+	movq	B, BO1
+	leaq    BUFFER1, BO		// first buffer to BO
+	movq	K, %rax
+	salq	$2, %rax		// 2 * COMPSIZE
+	leaq	(B, %rax,8), BO2 
+	movq	BO2, B			// next offset of B
+	movq	K, %rax
+	ALIGN_4
+
+.L6_00_02b:
+
+	vmovups		(BO1), %xmm0
+	vmovups	2 * SIZE(BO1), %xmm1
+	vmovups	        (BO2), %xmm2
+	vmovups	%xmm0,         (BO)
+	vmovups	%xmm1, 2 * SIZE(BO)
+	vmovups	%xmm2, 4 * SIZE(BO)
+	addq	$ 4*SIZE,BO1
+	addq	$ 4*SIZE,BO2
+	addq	$ 6*SIZE,BO
+	decq	%rax
+	jnz	.L6_00_02b
+
+.L6_00_02c:
+
+
+
+.L6_00_10:
+	movq	C, CO1
+	leaq	(C, LDC, 2), C		// c += 2 * ldc
+	leaq	(C, LDC, 1), C		// c += 1 * ldc
+
+	movq	A, AO		 	// aoffset = a
+
+	movq	M,  I
+	sarq	$ 2, I			// i = (m >> 2)
+	je	.L6_2_10
+
+	ALIGN_4
+
+/******************************************************************************************************************/
+
+.L6_4_11:
+
+	leaq	BUFFER1, BO		// first buffer to BO
+
+	vzeroall
+
+        movq    K, %rax
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L6_4_16
+	ALIGN_4
+
+.L6_4_12:
+
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+
+	je	.L6_4_16
+
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+
+	je	.L6_4_16
+
+	jmp	.L6_4_12
+	ALIGN_4
+
+.L6_4_16:
+        movq    K, %rax
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L6_4_19
+	ALIGN_4
+
+.L6_4_17:
+
+	KERNEL4x3_SUB
+
+	jnz	.L6_4_17
+	ALIGN_4
+
+
+.L6_4_19:
+
+	SAVE4x3
+
+	addq	$ 8 * SIZE, CO1		# coffset += 8
+	decq	I			# i --
+	jg	.L6_4_11
+	ALIGN_4	
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+
+
+/******************************************************************************************************************/
+.L6_2_10:
+	testq	$ 2, M		
+	jz	.L6_2_40		// to next 2 lines of N
+
+.L6_2_11:
+
+	leaq	BUFFER1, BO		// first buffer to BO
+
+	vzeroall
+
+        movq    K, %rax
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L6_2_16
+	ALIGN_4
+
+.L6_2_12:
+
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+
+	je	.L6_2_16
+
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+
+	je	.L6_2_16
+
+	jmp	.L6_2_12
+	ALIGN_4
+
+.L6_2_16:
+        movq    K, %rax
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L6_2_19
+	ALIGN_4
+
+.L6_2_17:
+
+	KERNEL2x3_SUB
+
+	jnz	.L6_2_17
+	ALIGN_4
+
+
+.L6_2_19:
+
+	SAVE2x3
+
+	addq	$ 4 * SIZE, CO1		# coffset += 4
+	ALIGN_4	
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+.L6_2_40:
+	testq	$ 1, M		
+	jz	.L6_2_60		// to next 2 lines of N
+
+	ALIGN_4
+
+.L6_2_41:
+
+	leaq	BUFFER1, BO		// first buffer to BO
+
+	vzeroall
+
+        movq    K, %rax
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L6_2_46
+
+	ALIGN_4
+
+.L6_2_42:
+
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+
+	je	.L6_2_46
+
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+
+	je	.L6_2_46
+
+	jmp	.L6_2_42
+	ALIGN_4
+
+.L6_2_46:
+        movq    K, %rax
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L6_2_49
+
+	ALIGN_4
+
+.L6_2_47:
+
+	KERNEL1x3_SUB
+
+	jnz	.L6_2_47
+	ALIGN_4
+
+
+.L6_2_49:
+
+	SAVE1x3
+
+	addq	$ 2 * SIZE, CO1		# coffset += 2
+	decq	I			# i --
+	jg	.L6_2_41
+	ALIGN_4	
+
+
+
+	
+.L6_2_60:
+
+
+/************************************************************************************************/
+
+/************************************************************************************************/
+
+
+.L7_00_01:
+	// copy to sub buffer
+	movq	B, BO1
+	leaq    BUFFER1, BO		// first buffer to BO
+	movq	K, %rax
+	salq	$2, %rax		// 2 * COMPSIZE
+	leaq	(B, %rax,8), BO2 
+	movq	K, %rax
+	ALIGN_4
+
+.L7_00_02b:
+
+	vmovups	2 * SIZE(BO1), %xmm0
+	vmovups	        (BO2), %xmm1
+	vmovups	2 * SIZE(BO2), %xmm2
+	vmovups	%xmm0,         (BO)
+	vmovups	%xmm1, 2 * SIZE(BO)
+	vmovups	%xmm2, 4 * SIZE(BO)
+	addq	$ 4*SIZE,BO1
+	addq	$ 4*SIZE,BO2
+	addq	$ 6*SIZE,BO
+	decq	%rax
+	jnz	.L7_00_02b
+
+.L7_00_02c:
+
+	movq	BO2, B			// next offset of B
+
+
+.L7_00_10:
+	movq	C, CO1
+	leaq	(C, LDC, 2), C		// c += 2 * ldc
+	leaq	(C, LDC, 1), C		// c += 1 * ldc
+
+	movq	A, AO		 	// aoffset = a
+
+	movq	M,  I
+	sarq	$ 2, I			// i = (m >> 2)
+	je	.L7_2_10
+
+	ALIGN_4
+
+/******************************************************************************************************************/
+
+.L7_4_11:
+
+	leaq	BUFFER1, BO		// first buffer to BO
+
+	vzeroall
+
+        movq    K, %rax
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L7_4_16
+	ALIGN_4
+
+.L7_4_12:
+
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+
+	je	.L7_4_16
+
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+	KERNEL4x3_SUB
+
+	je	.L7_4_16
+
+	jmp	.L7_4_12
+	ALIGN_4
+
+.L7_4_16:
+        movq    K, %rax
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L7_4_19
+
+	ALIGN_4
+
+.L7_4_17:
+
+	KERNEL4x3_SUB
+
+	jnz	.L7_4_17
+	ALIGN_4
+
+
+.L7_4_19:
+
+	SAVE4x3
+
+	addq	$ 8 * SIZE, CO1		# coffset += 8
+	decq	I			# i --
+	jg	.L7_4_11
+	ALIGN_4	
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+
+
+/******************************************************************************************************************/
+.L7_2_10:
+	testq	$ 2, M		
+	jz	.L7_2_40		// to next 2 lines of N
+
+.L7_2_11:
+
+	leaq	BUFFER1, BO		// first buffer to BO
+
+	vzeroall
+
+        movq    K, %rax
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L7_2_16
+	ALIGN_4
+
+.L7_2_12:
+
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+
+	je	.L7_2_16
+
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+	KERNEL2x3_SUB
+
+	je	.L7_2_16
+
+	jmp	.L7_2_12
+	ALIGN_4
+
+.L7_2_16:
+        movq    K, %rax
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L7_2_19
+
+	ALIGN_4
+
+.L7_2_17:
+
+	KERNEL2x3_SUB
+
+	jnz	.L7_2_17
+	ALIGN_4
+
+
+.L7_2_19:
+
+	SAVE2x3
+
+	addq	$ 4 * SIZE, CO1		# coffset += 4
+	ALIGN_4	
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+.L7_2_40:
+	testq	$ 1, M		
+	jz	.L7_2_60		// to next 2 lines of N
+
+	ALIGN_4
+
+.L7_2_41:
+
+	leaq	BUFFER1, BO		// first buffer to BO
+
+	vzeroall
+
+        movq    K, %rax
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L7_2_46
+
+	ALIGN_4
+
+.L7_2_42:
+
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+
+	je	.L7_2_46
+
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+	KERNEL1x3_SUB
+
+	je	.L7_2_46
+
+	jmp	.L7_2_42
+	ALIGN_4
+
+.L7_2_46:
+        movq    K, %rax
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L7_2_49
+	ALIGN_4
+
+.L7_2_47:
+
+	KERNEL1x3_SUB
+
+	jnz	.L7_2_47
+	ALIGN_4
+
+
+.L7_2_49:
+
+	SAVE1x3
+
+	addq	$ 2 * SIZE, CO1		# coffset += 2
+	decq	I			# i --
+	jg	.L7_2_41
+	ALIGN_4	
+
+
+
+	
+.L7_2_60:
+
+	decq	J			// j --
+	jg	.L6_00_01		// next 6 lines of N
+
+/************************************************************************************************/
+
+
+
+/************************************************************************************************/
+.L2_00_0:
+
+	movq	Nmod6,  J
+	sarq	$1, J		// j = j / 2
+	cmpq	$ 0, J
+	je	.L1_2_0
+	ALIGN_4
+
+
+
+.L2_00_01:
+	// copy to sub buffer
+	movq	B, BO1
+	leaq    BUFFER1, BO		// first buffer to BO
+	movq	K, %rax
+	ALIGN_4
+
+.L2_00_02b:
+
+	vmovups		(BO1), %xmm0
+	vmovups	2 * SIZE(BO1), %xmm1
+	vmovups	%xmm0,       (BO)
+	vmovups	%xmm1, 2 * SIZE(BO)
+	addq	$ 4*SIZE,BO1
+	addq	$ 4*SIZE,BO
+	decq	%rax
+	jnz	.L2_00_02b
+
+.L2_00_02c:
+
+	movq	BO1, B			// next offset of B
+
+
+.L2_00_10:
+	movq	C, CO1
+	leaq	(C, LDC, 2), C		// c += 2 * ldc
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        movq    OFFSET, %rax
+        movq    %rax, KK
+#endif
+	
+	movq	A, AO		 	// aoffset = a
+	addq	$ 8 * SIZE, AO
+
+	movq	M,  I
+	sarq	$ 2, I			// i = (m >> 2)
+	je	.L2_2_10
+
+	ALIGN_4
+
+/******************************************************************************************************************/
+
+.L2_4_11:
+
+#if !defined(TRMMKERNEL) || \
+        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+	leaq	BUFFER1, BO		// first buffer to BO
+	addq	$ 8 * SIZE, BO
+#else
+        movq    KK, %rax
+	leaq	BUFFER1, BO			// first buffer to BO
+	addq	$ 8 * SIZE, BO
+	movq    %rax, BI                        //  Index for BO
+        leaq    (,BI,4), BI                     //  BI = BI * 4 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 3, %rax			// rax = rax * 8 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+	vzeroall
+
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+        movq    K, %rax
+        subq    KK, %rax
+        movq    %rax, KKK
+#else
+        movq    KK, %rax
+#ifdef LEFT
+        addq    $ 4, %rax        // number of values in AO
+#else
+        addq    $ 2, %rax        // number of values in BO
+#endif
+        movq    %rax, KKK
+#endif
+
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L2_4_16
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 3, %rax			// rax = rax * 8 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L2_4_12:
+
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+        prefetcht0      B_PR1(BO,BI  ,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+        prefetcht0      B_PR1(BO,BI  ,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+        prefetcht0      B_PR1(BO,BI  ,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+        prefetcht0      B_PR1(BO,BI  ,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+
+	je	.L2_4_16
+
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+        prefetcht0      B_PR1(BO,BI  ,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+        prefetcht0      B_PR1(BO,BI  ,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+        prefetcht0      B_PR1(BO,BI  ,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+        prefetcht0      B_PR1(BO,BI  ,SIZE)
+	KERNEL4x2_SUB
+        prefetcht0      A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+
+	je	.L2_4_16
+
+	jmp	.L2_4_12
+	ALIGN_4
+
+.L2_4_16:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L2_4_19
+
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 3, %rax			// rax = rax * 8 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L2_4_17:
+
+	KERNEL4x2_SUB
+
+	jl	.L2_4_17
+	ALIGN_4
+
+
+.L2_4_19:
+
+	SAVE4x2
+
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+        movq    K, %rax
+        subq    KKK, %rax
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 3, %rax			// rax = rax * 8 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        addq    $ 4, KK
+#endif
+
+	addq	$ 8 * SIZE, CO1		# coffset += 8
+	decq	I			# i --
+	jg	.L2_4_11
+	ALIGN_4	
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+
+
+/******************************************************************************************************************/
+.L2_2_10:
+	testq	$ 2, M		
+	jz	.L2_2_40		// to next 2 lines of N
+
+.L2_2_11:
+
+#if !defined(TRMMKERNEL) || \
+        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+	leaq	BUFFER1, BO		// first buffer to BO
+	addq	$ 8 * SIZE, BO
+#else
+        movq    KK, %rax
+	leaq	BUFFER1, BO			// first buffer to BO
+	addq	$ 8 * SIZE, BO
+	movq    %rax, BI                        //  Index for BO
+        leaq    (,BI,4), BI                     //  BI = BI * 4 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 2, %rax			// rax = rax * 4 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+	vzeroall
+
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+        movq    K, %rax
+        subq    KK, %rax
+        movq    %rax, KKK
+#else
+        movq    KK, %rax
+#ifdef LEFT
+        addq    $ 2, %rax        // number of values in AO
+#else
+        addq    $ 2, %rax        // number of values in BO
+#endif
+        movq    %rax, KKK
+#endif
+
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L2_2_16
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 2, %rax			// rax = rax * 4 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L2_2_12:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	je	.L2_2_16
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	je	.L2_2_16
+
+	jmp	.L2_2_12
+	ALIGN_4
+
+.L2_2_16:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L2_2_19
+
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 2, %rax			// rax = rax * 4 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L2_2_17:
+
+	KERNEL2x2_SUB
+
+	jl	.L2_2_17
+	ALIGN_4
+
+
+.L2_2_19:
+
+	SAVE2x2
+
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+        movq    K, %rax
+        subq    KKK, %rax
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 2, %rax			// rax = rax * 4 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        addq    $ 2, KK
+#endif
+
+	addq	$ 4 * SIZE, CO1		# coffset += 4
+	ALIGN_4	
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+.L2_2_40:
+	testq	$ 1, M		
+	jz	.L2_2_60		// to next 2 lines of N
+
+	ALIGN_4
+
+.L2_2_41:
+
+#if !defined(TRMMKERNEL) || \
+        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+	leaq	BUFFER1, BO		// first buffer to BO
+	addq	$ 8 * SIZE, BO
+#else
+        movq    KK, %rax
+	leaq	BUFFER1, BO			// first buffer to BO
+	addq	$ 8 * SIZE, BO
+	movq    %rax, BI                        //  Index for BO
+        leaq    (,BI,4), BI                     //  BI = BI * 4 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 1, %rax			// rax = rax * 2 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+	vzeroall
+
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+        movq    K, %rax
+        subq    KK, %rax
+        movq    %rax, KKK
+#else
+        movq    KK, %rax
+#ifdef LEFT
+        addq    $ 1, %rax        // number of values in AO
+#else
+        addq    $ 2, %rax        // number of values in BO
+#endif
+        movq    %rax, KKK
+#endif
+
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L2_2_46
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 1, %rax			// rax = rax * 2 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L2_2_42:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	je	.L2_2_46
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	je	.L2_2_46
+
+	jmp	.L2_2_42
+	ALIGN_4
+
+.L2_2_46:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L2_2_49
+
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 1, %rax			// rax = rax * 2 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L2_2_47:
+
+	KERNEL1x2_SUB
+
+	jl	.L2_2_47
+	ALIGN_4
+
+
+.L2_2_49:
+
+	SAVE1x2
+
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+        movq    K, %rax
+        subq    KKK, %rax
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 1, %rax			// rax = rax * 2 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        addq    $ 1, KK
+#endif
+
+	addq	$ 2 * SIZE, CO1		# coffset += 2
+	decq	I			# i --
+	jg	.L2_2_41
+	ALIGN_4	
+
+
+
+	
+.L2_2_60:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+        addq    $ 2, KK
+#endif
+
+	decq	J			// j --
+	jg	.L2_00_01			// next 2 lines of N
+
+
+
+.L1_2_0:
+
+/************************************************************************************************
+* Loop for Nmod6 % 2 > 0
+*************************************************************************************************/
+
+	movq	Nmod6, J		
+	andq	$ 1, J			// j % 2
+	je	.L999
+	ALIGN_4
+
+.L1_00_01:
+	// copy to sub buffer
+	movq	B, BO1
+	leaq    BUFFER1, BO		// first buffer to BO
+	movq	K, %rax
+	ALIGN_4
+
+.L1_00_02b:
+
+	vmovups		(BO1), %xmm0
+	vmovups	%xmm0,       (BO)
+	addq	$ 2*SIZE,BO1
+	addq	$ 2*SIZE,BO
+	decq	%rax
+	jnz	.L1_00_02b
+
+.L1_00_02c:
+
+	movq	BO1, B			// next offset of B
+
+.L1_00_10:
+	movq	C, CO1
+	leaq	(C, LDC, 1), C		// c += 1 * ldc
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        movq    OFFSET, %rax
+        movq    %rax, KK
+#endif
+	
+	movq	A, AO		 	// aoffset = a
+	addq	$ 8 * SIZE, AO
+
+	movq	M,  I
+	sarq	$ 2, I			// i = (m >> 2)
+	je	.L1_2_10
+
+	ALIGN_4
+
+/*******************************************************************************************************/
+
+
+.L1_4_11:
+
+#if !defined(TRMMKERNEL) || \
+        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+	leaq	BUFFER1, BO		// first buffer to BO
+	addq	$ 4 * SIZE, BO
+#else
+        movq    KK, %rax
+	leaq	BUFFER1, BO			// first buffer to BO
+	addq	$ 4 * SIZE, BO
+	movq    %rax, BI                        //  Index for BO
+        leaq    (,BI,2), BI                     //  BI = BI * 2 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 3, %rax			// rax = rax * 8 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+	vzeroall
+
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+        movq    K, %rax
+        subq    KK, %rax
+        movq    %rax, KKK
+#else
+        movq    KK, %rax
+#ifdef LEFT
+        addq    $ 4, %rax        // number of values in AO
+#else
+        addq    $ 1, %rax        // number of values in BO
+#endif
+        movq    %rax, KKK
+#endif
+
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L1_4_16
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+
+	salq	$ 3, %rax			// rax = rax * 8 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L1_4_12:
+
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	je	.L1_4_16
+
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	je	.L1_4_16
+
+	jmp	.L1_4_12
+	ALIGN_4
+
+.L1_4_16:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L1_4_19
+
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+
+	salq	$ 3, %rax			// rax = rax * 8 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L1_4_17:
+
+	KERNEL4x1_SUB
+
+	jl	.L1_4_17
+	ALIGN_4
+
+
+.L1_4_19:
+
+	SAVE4x1
+
+
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+        movq    K, %rax
+        subq    KKK, %rax
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 3, %rax			// rax = rax * 8 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        addq    $ 4, KK
+#endif
+
+	addq	$ 8 * SIZE, CO1		# coffset += 8
+	decq	I			# i --
+	jg	.L1_4_11
+	ALIGN_4	
+
+
+
+
+/*******************************************************************************************************/
+.L1_2_10:
+	testq	$ 2, M		
+	jz	.L1_2_40
+
+
+.L1_2_11:
+
+#if !defined(TRMMKERNEL) || \
+        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+	leaq	BUFFER1, BO		// first buffer to BO
+	addq	$ 4 * SIZE, BO
+#else
+        movq    KK, %rax
+	leaq	BUFFER1, BO			// first buffer to BO
+	addq	$ 4 * SIZE, BO
+	movq    %rax, BI                        //  Index for BO
+        leaq    (,BI,2), BI                     //  BI = BI * 2 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 2, %rax			// rax = rax * 4 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+	vzeroall
+
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+        movq    K, %rax
+        subq    KK, %rax
+        movq    %rax, KKK
+#else
+        movq    KK, %rax
+#ifdef LEFT
+        addq    $ 2, %rax        // number of values in AO
+#else
+        addq    $ 1, %rax        // number of values in BO
+#endif
+        movq    %rax, KKK
+#endif
+
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L1_2_16
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+
+	salq	$ 2, %rax			// rax = rax * 4 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L1_2_12:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	je	.L1_2_16
+
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	je	.L1_2_16
+
+	jmp	.L1_2_12
+	ALIGN_4
+
+.L1_2_16:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L1_2_19
+
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+
+	salq	$ 2, %rax			// rax = rax * 4 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L1_2_17:
+
+	KERNEL2x1_SUB
+
+	jl	.L1_2_17
+	ALIGN_4
+
+
+.L1_2_19:
+
+	SAVE2x1
+
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+        movq    K, %rax
+        subq    KKK, %rax
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 2, %rax			// rax = rax * 4 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        addq    $ 2, KK
+#endif
+
+	addq	$ 4 * SIZE, CO1		# coffset += 4
+
+	ALIGN_4	
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+.L1_2_40:
+	testq	$ 1, M		
+	jz	.L999
+
+	ALIGN_4
+
+.L1_2_41:
+
+#if !defined(TRMMKERNEL) || \
+        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+	leaq	BUFFER1, BO		// first buffer to BO
+	addq	$ 4 * SIZE, BO
+#else
+        movq    KK, %rax
+	leaq	BUFFER1, BO			// first buffer to BO
+	addq	$ 4 * SIZE, BO
+	movq    %rax, BI                        //  Index for BO
+        leaq    (,BI,2), BI                     //  BI = BI * 2 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 1, %rax			// rax = rax * 2 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+	vzeroall
+
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+        movq    K, %rax
+        subq    KK, %rax
+        movq    %rax, KKK
+#else
+        movq    KK, %rax
+#ifdef LEFT
+        addq    $ 1, %rax        // number of values in AO
+#else
+        addq    $ 1, %rax        // number of values in BO
+#endif
+        movq    %rax, KKK
+#endif
+
+
+	andq	$ -8, %rax			//  K = K - ( K % 8 )
+	je	.L1_2_46
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+
+	salq	$ 1, %rax			// rax = rax * 2 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L1_2_42:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	je	.L1_2_46
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	je	.L1_2_46
+
+	jmp	.L1_2_42
+	ALIGN_4
+
+.L1_2_46:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L1_2_49
+
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+
+	salq	$ 1, %rax			// rax = rax * 2 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L1_2_47:
+
+	KERNEL1x1_SUB
+
+	jl	.L1_2_47
+	ALIGN_4
+
+
+.L1_2_49:
+
+	SAVE1x1
+
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
+    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+        movq    K, %rax
+        subq    KKK, %rax
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+        leaq    (BO, BI, SIZE), BO
+	salq	$ 1, %rax			// rax = rax * 2 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        addq    $ 1, KK
+#endif
+
+	addq	$ 2 * SIZE, CO1		# coffset += 2
+	decq	I			# i --
+	jg	.L1_2_41
+	ALIGN_4	
+
+
+
+
+
+
+.L999:
+	vzeroupper
+
+	movq   		SP, %rsp
+	movq	   (%rsp), %rbx
+	movq	  8(%rsp), %rbp
+	movq	 16(%rsp), %r12
+	movq	 24(%rsp), %r13
+	movq	 32(%rsp), %r14
+	movq	 40(%rsp), %r15
+
+#ifdef WINDOWS_ABI
+	movq	 48(%rsp), %rdi
+	movq	 56(%rsp), %rsi
+	vmovups	 64(%rsp), %xmm6
+	vmovups	 80(%rsp), %xmm7
+	vmovups	 96(%rsp), %xmm8
+	vmovups	112(%rsp), %xmm9
+	vmovups	128(%rsp), %xmm10
+	vmovups	144(%rsp), %xmm11
+	vmovups	160(%rsp), %xmm12
+	vmovups	176(%rsp), %xmm13
+	vmovups	192(%rsp), %xmm14
+	vmovups	208(%rsp), %xmm15
+#endif
+
+	addq	$ STACKSIZE, %rsp
+	ret
+
+	EPILOGUE
+
+
+#else
+/************************************************************************************************
+ TRMM Kernel
+************************************************************************************************/
 
 	PROLOGUE
 	PROFCODE
@@ -1811,3 +3875,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	ret
 
 	EPILOGUE
+
+#endif
+
+
diff --git a/param.h b/param.h
index 880219b7c..863e83c32 100644
--- a/param.h
+++ b/param.h
@@ -1206,6 +1206,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ZGEMM_DEFAULT_UNROLL_N 2
 #define XGEMM_DEFAULT_UNROLL_N 1
 
+#define DGEMM_DEFAULT_UNROLL_MN 16
 #endif
 
 #ifdef ARCH_X86