diff --git a/Makefile b/Makefile
index 5b0ca0dc5..397836da1 100644
--- a/Makefile
+++ b/Makefile
@@ -23,7 +23,7 @@ endif
 SUBDIRS_ALL = $(SUBDIRS) test ctest utest exports benchmark ../laswp ../bench
 
 .PHONY : all libs netlib test ctest shared install
-.NOTPARALLEL : all libs prof lapack-test install
+.NOTPARALLEL : all libs prof lapack-test install blas-test
 
 all :: libs netlib tests shared
 	@echo
@@ -282,6 +282,11 @@ lapack-test :
 	make -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING xeigtstc  xeigtstd  xeigtsts  xeigtstz  xlintstc  xlintstd  xlintstds  xlintstrfd  xlintstrfz  xlintsts  xlintstz  xlintstzc xlintstrfs xlintstrfc
 	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r )
 
+blas-test:
+	(cd $(NETLIB_LAPACK_DIR)/BLAS && rm -f x* *.out)
+	make -j 1 -C $(NETLIB_LAPACK_DIR) blas_testing
+	(cd $(NETLIB_LAPACK_DIR)/BLAS && cat *.out)
+
 
 dummy :
 
diff --git a/kernel/Makefile.L1 b/kernel/Makefile.L1
index bd31503a4..7c7cb2770 100644
--- a/kernel/Makefile.L1
+++ b/kernel/Makefile.L1
@@ -687,15 +687,27 @@ $(KDIR)ddot_k$(TSUFFIX).$(SUFFIX) $(KDIR)ddot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNEL
 $(KDIR)qdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)qdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QDOTKERNEL)
 	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
 
-$(KDIR)dsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)dsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
-
 $(KDIR)sdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)sdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL)
 	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
 
+ifdef DSDOTKERNEL
+
+$(KDIR)dsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)dsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DSDOTKERNEL)
+	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
+
+$(KDIR)sdsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)sdsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DSDOTKERNEL)
+	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
+
+else
+
+$(KDIR)dsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)dsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL)
+	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
+
 $(KDIR)sdsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)sdsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL)
 	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
 
+endif
+
 $(KDIR)zdotu_k$(TSUFFIX).$(SUFFIX) $(KDIR)zdotu_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZDOTKERNEL)
 	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -UCONJ $< -o $@
 
diff --git a/kernel/x86_64/KERNEL b/kernel/x86_64/KERNEL
index d3486283e..fa6282c53 100644
--- a/kernel/x86_64/KERNEL
+++ b/kernel/x86_64/KERNEL
@@ -119,9 +119,15 @@ XCOPYKERNEL = zcopy.S
 endif
 
 ifndef SDOTKERNEL
-SDOTKERNEL = ../arm/dot.c
+SDOTKERNEL = dot_sse.S
 endif
 
+
+ifndef DSDOTKERNEL
+DSDOTKERNEL = ../arm/dot.c 
+endif
+
+
 ifndef DDOTKERNEL
 DDOTKERNEL =  dot_sse2.S
 endif
diff --git a/kernel/x86_64/KERNEL.BARCELONA b/kernel/x86_64/KERNEL.BARCELONA
index b1e099e4c..7c7359551 100644
--- a/kernel/x86_64/KERNEL.BARCELONA
+++ b/kernel/x86_64/KERNEL.BARCELONA
@@ -2,7 +2,7 @@ SGEMVNKERNEL = sgemv_n.S
 SGEMVTKERNEL = sgemv_t.S
 
 ZGEMVNKERNEL = zgemv_n_dup.S
-ZGEMVTKERNEL = zgemv_t_dup.S
+ZGEMVTKERNEL = zgemv_t.S
 
 SGEMMKERNEL    =  gemm_kernel_8x4_barcelona.S
 SGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
diff --git a/kernel/x86_64/KERNEL.BOBCAT b/kernel/x86_64/KERNEL.BOBCAT
index 2b6b2fe59..313c62d7c 100644
--- a/kernel/x86_64/KERNEL.BOBCAT
+++ b/kernel/x86_64/KERNEL.BOBCAT
@@ -1,5 +1,5 @@
 ZGEMVNKERNEL = zgemv_n_dup.S
-ZGEMVTKERNEL = zgemv_t_dup.S
+ZGEMVTKERNEL = zgemv_t.S
 
 SGEMMKERNEL    =  gemm_kernel_8x4_barcelona.S
 SGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
diff --git a/kernel/x86_64/KERNEL.BULLDOZER b/kernel/x86_64/KERNEL.BULLDOZER
index f8ac3db0e..d9b9f84f3 100644
--- a/kernel/x86_64/KERNEL.BULLDOZER
+++ b/kernel/x86_64/KERNEL.BULLDOZER
@@ -2,7 +2,7 @@ SGEMVNKERNEL = sgemv_n.S
 SGEMVTKERNEL = sgemv_t.S
 
 ZGEMVNKERNEL = zgemv_n_dup.S
-ZGEMVTKERNEL = zgemv_t_dup.S
+ZGEMVTKERNEL = zgemv_t.S
 
 DGEMVNKERNEL = dgemv_n_bulldozer.S
 DGEMVTKERNEL = dgemv_t_bulldozer.S
diff --git a/kernel/x86_64/KERNEL.OPTERON_SSE3 b/kernel/x86_64/KERNEL.OPTERON_SSE3
index 9367bd62a..7218ad6d1 100644
--- a/kernel/x86_64/KERNEL.OPTERON_SSE3
+++ b/kernel/x86_64/KERNEL.OPTERON_SSE3
@@ -1,5 +1,5 @@
 ZGEMVNKERNEL = zgemv_n_dup.S
-ZGEMVTKERNEL = zgemv_t_dup.S
+ZGEMVTKERNEL = zgemv_t.S
 
 SGEMMKERNEL    =  gemm_kernel_8x4_sse.S
 SGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
diff --git a/kernel/x86_64/KERNEL.PILEDRIVER b/kernel/x86_64/KERNEL.PILEDRIVER
index a06a04cfa..b083b5d53 100644
--- a/kernel/x86_64/KERNEL.PILEDRIVER
+++ b/kernel/x86_64/KERNEL.PILEDRIVER
@@ -2,7 +2,7 @@ SGEMVNKERNEL = sgemv_n.S
 SGEMVTKERNEL = sgemv_t.S
 
 ZGEMVNKERNEL = zgemv_n_dup.S
-ZGEMVTKERNEL = zgemv_t_dup.S
+ZGEMVTKERNEL = zgemv_t.S
 
 DGEMVNKERNEL = dgemv_n_bulldozer.S
 DGEMVTKERNEL = dgemv_t_bulldozer.S
diff --git a/kernel/x86_64/KERNEL.PRESCOTT b/kernel/x86_64/KERNEL.PRESCOTT
index 03d6664e8..9b3e514d1 100644
--- a/kernel/x86_64/KERNEL.PRESCOTT
+++ b/kernel/x86_64/KERNEL.PRESCOTT
@@ -1,5 +1,5 @@
 ZGEMVNKERNEL = zgemv_n_dup.S
-ZGEMVTKERNEL = zgemv_t_dup.S
+ZGEMVTKERNEL = zgemv_t.S
 
 SGEMMKERNEL    =  gemm_kernel_8x4_sse3.S
 SGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
diff --git a/kernel/x86_64/KERNEL.SANDYBRIDGE b/kernel/x86_64/KERNEL.SANDYBRIDGE
index 1b4228c84..b84a5f374 100644
--- a/kernel/x86_64/KERNEL.SANDYBRIDGE
+++ b/kernel/x86_64/KERNEL.SANDYBRIDGE
@@ -21,11 +21,11 @@ DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
 DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
 DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-CGEMMKERNEL    =  zgemm_kernel_2x4_nehalem.S
-CGEMMINCOPY    =  zgemm_ncopy_2.S
-CGEMMITCOPY    =  zgemm_tcopy_2.S
-CGEMMONCOPY    =  ../generic/zgemm_ncopy_4.c
-CGEMMOTCOPY    =  ../generic/zgemm_tcopy_4.c
+CGEMMKERNEL    =  cgemm_kernel_8x2_sandy.S
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_8.c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_8.c
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
 CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
 CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
 CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
diff --git a/kernel/x86_64/cgemm_kernel_4x2_bulldozer.S b/kernel/x86_64/cgemm_kernel_4x2_bulldozer.S
index 33d3d29dd..97958a88f 100644
--- a/kernel/x86_64/cgemm_kernel_4x2_bulldozer.S
+++ b/kernel/x86_64/cgemm_kernel_4x2_bulldozer.S
@@ -522,16 +522,16 @@
 #ifdef WINDOWS_ABI
 	movq	%rdi,    48(%rsp)
 	movq	%rsi,    56(%rsp)
-	movups	%xmm6,   64(%rsp)
-	movups	%xmm7,   80(%rsp)
-	movups	%xmm8,   96(%rsp)
-	movups	%xmm9,  112(%rsp)
-	movups	%xmm10, 128(%rsp)
-	movups	%xmm11, 144(%rsp)
-	movups	%xmm12, 160(%rsp)
-	movups	%xmm13, 176(%rsp)
-	movups	%xmm14, 192(%rsp)
-	movups	%xmm15, 208(%rsp)
+	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
@@ -541,14 +541,15 @@
 	movq	OLD_C,     C
 	movq	OLD_LDC,   LDC
 #ifdef TRMMKERNEL
-	movsd	OLD_OFFSET, %xmm12
+	vmovsd	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
+	vmovsd	STACKSIZE + 16(%rsp), %xmm12
 #endif
 
 #endif
@@ -1865,6 +1866,8 @@
 
 
 .L999:
+	vzeroupper
+
 	movq   		SP, %rsp
 	movq	   (%rsp), %rbx
 	movq	  8(%rsp), %rbp
@@ -1876,16 +1879,16 @@
 #ifdef WINDOWS_ABI
 	movq	 48(%rsp), %rdi
 	movq	 56(%rsp), %rsi
-	movups	 64(%rsp), %xmm6
-	movups	 80(%rsp), %xmm7
-	movups	 96(%rsp), %xmm8
-	movups	112(%rsp), %xmm9
-	movups	128(%rsp), %xmm10
-	movups	144(%rsp), %xmm11
-	movups	160(%rsp), %xmm12
-	movups	176(%rsp), %xmm13
-	movups	192(%rsp), %xmm14
-	movups	208(%rsp), %xmm15
+	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
diff --git a/kernel/x86_64/cgemm_kernel_4x2_piledriver.S b/kernel/x86_64/cgemm_kernel_4x2_piledriver.S
index 76d6810fa..72deee12f 100644
--- a/kernel/x86_64/cgemm_kernel_4x2_piledriver.S
+++ b/kernel/x86_64/cgemm_kernel_4x2_piledriver.S
@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/
 /*********************************************************************
 *
-* 2013/10/31 Saar
+* 2014/06/28 Saar
 *        BLASTEST               : OK
 *        CTEST                  : OK
 *        TEST                   : OK
@@ -546,16 +546,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	%rdi,    48(%rsp)
 	movq	%rsi,    56(%rsp)
-	movups	%xmm6,   64(%rsp)
-	movups	%xmm7,   80(%rsp)
-	movups	%xmm8,   96(%rsp)
-	movups	%xmm9,  112(%rsp)
-	movups	%xmm10, 128(%rsp)
-	movups	%xmm11, 144(%rsp)
-	movups	%xmm12, 160(%rsp)
-	movups	%xmm13, 176(%rsp)
-	movups	%xmm14, 192(%rsp)
-	movups	%xmm15, 208(%rsp)
+	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
@@ -568,6 +568,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	movsd	OLD_OFFSET, %xmm12
 #endif
 	vmovaps	%xmm3, %xmm0
+	vmovsd   OLD_ALPHA_I, %xmm1
 
 #else
 	movq	STACKSIZE +  8(%rsp), LDC
@@ -1889,6 +1890,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .L999:
+	vzeroupper
+
 	movq   		SP, %rsp
 	movq	   (%rsp), %rbx
 	movq	  8(%rsp), %rbp
@@ -1900,16 +1903,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	 48(%rsp), %rdi
 	movq	 56(%rsp), %rsi
-	movups	 64(%rsp), %xmm6
-	movups	 80(%rsp), %xmm7
-	movups	 96(%rsp), %xmm8
-	movups	112(%rsp), %xmm9
-	movups	128(%rsp), %xmm10
-	movups	144(%rsp), %xmm11
-	movups	160(%rsp), %xmm12
-	movups	176(%rsp), %xmm13
-	movups	192(%rsp), %xmm14
-	movups	208(%rsp), %xmm15
+	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
diff --git a/kernel/x86_64/cgemm_kernel_8x2_haswell.S b/kernel/x86_64/cgemm_kernel_8x2_haswell.S
index 285a9f09d..baee3cd2f 100644
--- a/kernel/x86_64/cgemm_kernel_8x2_haswell.S
+++ b/kernel/x86_64/cgemm_kernel_8x2_haswell.S
@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 **********************************************************************************/
 
 /*********************************************************************
-* 2013/11/13 Saar
+* 2014/06/28 Saar
 *        BLASTEST               : OK
 *        CTEST                  : OK
 *        TEST                   : OK
@@ -816,16 +816,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	%rdi,    48(%rsp)
 	movq	%rsi,    56(%rsp)
-	movups	%xmm6,   64(%rsp)
-	movups	%xmm7,   80(%rsp)
-	movups	%xmm8,   96(%rsp)
-	movups	%xmm9,  112(%rsp)
-	movups	%xmm10, 128(%rsp)
-	movups	%xmm11, 144(%rsp)
-	movups	%xmm12, 160(%rsp)
-	movups	%xmm13, 176(%rsp)
-	movups	%xmm14, 192(%rsp)
-	movups	%xmm15, 208(%rsp)
+	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
@@ -838,6 +838,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	movsd	OLD_OFFSET, %xmm12
 #endif
 	vmovaps	%xmm3, %xmm0
+	vmovsd   OLD_ALPHA_I, %xmm1
 
 #else
 	movq	STACKSIZE +  8(%rsp), LDC
@@ -2253,6 +2254,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .L999:
+	vzeroupper
+
 	movq   		SP, %rsp
 	movq	   (%rsp), %rbx
 	movq	  8(%rsp), %rbp
@@ -2264,16 +2267,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	 48(%rsp), %rdi
 	movq	 56(%rsp), %rsi
-	movups	 64(%rsp), %xmm6
-	movups	 80(%rsp), %xmm7
-	movups	 96(%rsp), %xmm8
-	movups	112(%rsp), %xmm9
-	movups	128(%rsp), %xmm10
-	movups	144(%rsp), %xmm11
-	movups	160(%rsp), %xmm12
-	movups	176(%rsp), %xmm13
-	movups	192(%rsp), %xmm14
-	movups	208(%rsp), %xmm15
+	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
diff --git a/kernel/x86_64/cgemm_kernel_8x2_sandy.S b/kernel/x86_64/cgemm_kernel_8x2_sandy.S
new file mode 100644
index 000000000..564b73380
--- /dev/null
+++ b/kernel/x86_64/cgemm_kernel_8x2_sandy.S
@@ -0,0 +1,2281 @@
+/*********************************************************************************
+Copyright (c) 2013, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+
+
+#define ASSEMBLER
+#include "common.h"
+ 
+#define OLD_M	%rdi
+#define OLD_N	%rsi
+#define M	%r13
+#define J	%r14
+#define OLD_K	%rdx
+
+#define A	%rcx
+#define B	%r8
+#define C	%r9
+#define LDC	%r10
+	
+#define I	%r11
+#define AO	%rdi
+#define BO	%rsi
+#define	CO1	%r15
+#define K	%r12
+#define BI	%rbp
+#define	SP	%rbx
+
+#define BO1	%rdi
+#define BO2	%r15
+
+#ifndef WINDOWS_ABI
+
+#define STACKSIZE 96
+
+#else
+
+#define STACKSIZE 320
+
+#define OLD_ALPHA_I     40 + STACKSIZE(%rsp)
+#define OLD_A           48 + STACKSIZE(%rsp)
+#define OLD_B           56 + STACKSIZE(%rsp)
+#define OLD_C           64 + STACKSIZE(%rsp)
+#define OLD_LDC         72 + STACKSIZE(%rsp)
+#define OLD_OFFSET      80 + STACKSIZE(%rsp)
+
+#endif
+
+#define L_BUFFER_SIZE 8192
+
+#define Ndiv6	 24(%rsp)
+#define Nmod6	 32(%rsp)
+#define N	 40(%rsp)
+#define ALPHA_R  48(%rsp)
+#define ALPHA_I  56(%rsp)
+#define OFFSET   64(%rsp)
+#define KK       72(%rsp)
+#define KKK      80(%rsp)
+#define BUFFER1	           128(%rsp)
+
+#if defined(OS_WINDOWS)
+#if   L_BUFFER_SIZE > 16384
+#define STACK_TOUCH \
+        movl    $ 0,  4096 * 4(%rsp);\
+        movl    $ 0,  4096 * 3(%rsp);\
+        movl    $ 0,  4096 * 2(%rsp);\
+        movl    $ 0,  4096 * 1(%rsp);
+#elif L_BUFFER_SIZE > 12288
+#define STACK_TOUCH \
+        movl    $ 0,  4096 * 3(%rsp);\
+        movl    $ 0,  4096 * 2(%rsp);\
+        movl    $ 0,  4096 * 1(%rsp);
+#elif L_BUFFER_SIZE > 8192
+#define STACK_TOUCH \
+        movl    $ 0,  4096 * 2(%rsp);\
+        movl    $ 0,  4096 * 1(%rsp);
+#elif L_BUFFER_SIZE > 4096
+#define STACK_TOUCH \
+        movl    $ 0,  4096 * 1(%rsp);
+#else
+#define STACK_TOUCH
+#endif
+#else
+#define STACK_TOUCH
+#endif
+
+
+
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+
+#define	VFMADDPS_YR( y0,y1,y2 ) \
+                               vmulps y1,y2,%ymm2;\
+                               vaddps y0,%ymm2,y0
+
+#define	VFMADDPS_YI( y0,y1,y2 ) \
+                               vmulps y1,y2,%ymm3;\
+                               vaddps y0,%ymm3,y0
+
+#define	VFMADDPS_R( y0,y1,y2 ) \
+                               vmulps y1,y2,%xmm2;\
+                               vaddps y0,%xmm2,y0
+
+#define	VFMADDPS_I( y0,y1,y2 ) \
+                               vmulps y1,y2,%xmm3;\
+                               vaddps y0,%xmm3,y0
+
+
+#elif defined(RN) || defined(RT) || defined(CN) || defined(CT)
+
+#define	VFMADDPS_YR( y0,y1,y2 ) \
+                               vmulps y1,y2,%ymm2;\
+                               vsubps %ymm2,y0,y0
+
+#define	VFMADDPS_YI( y0,y1,y2 ) \
+                               vmulps y1,y2,%ymm3;\
+                               vaddps y0,%ymm3,y0
+
+#define	VFMADDPS_R( y0,y1,y2 ) \
+                               vmulps y1,y2,%xmm2;\
+                               vsubps %xmm2,y0,y0
+
+#define	VFMADDPS_I( y0,y1,y2 ) \
+                               vmulps y1,y2,%xmm3;\
+                               vaddps y0,%xmm3,y0
+
+
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+#define	VFMADDPS_YR( y0,y1,y2 ) \
+                               vmulps y1,y2,%ymm2;\
+                               vaddps y0,%ymm2,y0
+
+#define	VFMADDPS_YI( y0,y1,y2 ) \
+                               vmulps y1,y2,%ymm3;\
+                               vsubps %ymm3,y0,y0
+
+#define	VFMADDPS_R( y0,y1,y2 ) \
+                               vmulps y1,y2,%xmm2;\
+                               vaddps y0,%xmm2,y0
+
+#define	VFMADDPS_I( y0,y1,y2 ) \
+                               vmulps y1,y2,%xmm3;\
+                               vsubps %xmm3,y0,y0
+
+
+#else
+
+#define	VFMADDPS_YR( y0,y1,y2 ) \
+                               vmulps y1,y2,%ymm2;\
+                               vsubps %ymm2,y0,y0
+
+#define	VFMADDPS_YI( y0,y1,y2 ) \
+                               vmulps y1,y2,%ymm3;\
+                               vsubps %ymm3,y0,y0
+
+#define	VFMADDPS_R( y0,y1,y2 ) \
+                               vmulps y1,y2,%xmm2;\
+                               vsubps %xmm2,y0,y0
+
+#define	VFMADDPS_I( y0,y1,y2 ) \
+                               vmulps y1,y2,%xmm3;\
+                               vsubps %xmm3,y0,y0
+
+
+#endif
+
+
+#define	A_PR1	512
+#define	B_PR1	512
+
+/***************************************************************************************************************************/
+
+.macro KERNEL8x2_SUB
+
+        vmovups         -16 * SIZE(AO, %rax, SIZE), %ymm0
+        vbroadcastss         -8 * SIZE(BO, BI, SIZE), %ymm4
+        VFMADDPS_YR(        %ymm8,%ymm4,%ymm0  )
+        vmovups          -8 * SIZE(AO, %rax, SIZE), %ymm1
+        VFMADDPS_YR(        %ymm12,%ymm4,%ymm1 )
+        vbroadcastss         -7 * SIZE(BO, BI, SIZE), %ymm5
+        VFMADDPS_YI(        %ymm9,%ymm5,%ymm0  )
+        VFMADDPS_YI(        %ymm13,%ymm5,%ymm1 )
+        vbroadcastss         -6 * SIZE(BO, BI, SIZE), %ymm6
+        VFMADDPS_YR(        %ymm10,%ymm6,%ymm0 )
+        VFMADDPS_YR(        %ymm14,%ymm6,%ymm1 )
+        vbroadcastss         -5 * SIZE(BO, BI, SIZE), %ymm7
+        VFMADDPS_YI(        %ymm11,%ymm7,%ymm0 )
+        VFMADDPS_YI(        %ymm15,%ymm7,%ymm1 )
+        addq    $ 4 , BI                           
+        addq    $ 16, %rax                         
+.endm
+
+.macro SAVE8x2
+
+	vbroadcastss	ALPHA_R, %ymm0
+	vbroadcastss	ALPHA_I, %ymm1
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %ymm9 , %ymm9, %ymm9
+        vshufps $ 0xb1, %ymm11, %ymm11, %ymm11
+        vshufps $ 0xb1, %ymm13, %ymm13, %ymm13
+        vshufps $ 0xb1, %ymm15, %ymm15, %ymm15
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %ymm9, %ymm8 , %ymm8
+        vaddsubps %ymm11,%ymm10, %ymm10
+        vaddsubps %ymm13,%ymm12, %ymm12
+        vaddsubps %ymm15,%ymm14, %ymm14
+
+        vshufps $ 0xb1, %ymm8 , %ymm8, %ymm9
+        vshufps $ 0xb1, %ymm10, %ymm10, %ymm11
+        vshufps $ 0xb1, %ymm12, %ymm12, %ymm13
+        vshufps $ 0xb1, %ymm14, %ymm14, %ymm15
+
+#else
+        vaddsubps %ymm8,  %ymm9 ,%ymm9
+        vaddsubps %ymm10, %ymm11,%ymm11
+        vaddsubps %ymm12, %ymm13,%ymm13
+        vaddsubps %ymm14, %ymm15,%ymm15
+
+        vmovaps   %ymm9,  %ymm8
+        vmovaps   %ymm11, %ymm10
+        vmovaps   %ymm13, %ymm12
+        vmovaps   %ymm15, %ymm14
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %ymm9 , %ymm9, %ymm9
+        vshufps $ 0xb1, %ymm11, %ymm11, %ymm11
+        vshufps $ 0xb1, %ymm13, %ymm13, %ymm13
+        vshufps $ 0xb1, %ymm15, %ymm15, %ymm15
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %ymm8 , %ymm0, %ymm8
+        vmulps  %ymm10, %ymm0, %ymm10
+        vmulps  %ymm12, %ymm0, %ymm12
+        vmulps  %ymm14, %ymm0, %ymm14
+
+	// multiply with ALPHA_I
+        vmulps  %ymm9 , %ymm1, %ymm9
+        vmulps  %ymm11, %ymm1, %ymm11
+        vmulps  %ymm13, %ymm1, %ymm13
+        vmulps  %ymm15, %ymm1, %ymm15
+
+	vaddsubps %ymm9, %ymm8 , %ymm8
+        vaddsubps %ymm11,%ymm10, %ymm10
+        vaddsubps %ymm13,%ymm12, %ymm12
+        vaddsubps %ymm15,%ymm14, %ymm14
+
+
+
+#ifndef TRMMKERNEL
+
+	vaddps 	 	(CO1), %ymm8 , %ymm8
+	vaddps  8 * SIZE(CO1), %ymm12, %ymm12
+
+	vaddps 	 	(CO1, LDC), %ymm10, %ymm10
+	vaddps  8 * SIZE(CO1, LDC), %ymm14, %ymm14
+
+#endif
+
+	vmovups	%ymm8 ,  	(CO1)
+	vmovups	%ymm12 , 8 * SIZE(CO1)
+
+	vmovups	%ymm10 ,  	(CO1, LDC)
+	vmovups	%ymm14 , 8 * SIZE(CO1, LDC)
+
+	prefetcht0	64(CO1)
+	prefetcht0	64(CO1, LDC)
+
+.endm
+
+/***************************************************************************************************************************/
+
+.macro KERNEL4x2_SUB
+        vmovups         -16 * SIZE(AO, %rax, SIZE), %xmm0
+        vbroadcastss         -8 * SIZE(BO, BI, SIZE), %xmm4
+        VFMADDPS_R(        %xmm8,%xmm4,%xmm0  )
+        vmovups         -12 * SIZE(AO, %rax, SIZE), %xmm1
+        VFMADDPS_R(        %xmm12,%xmm4,%xmm1 )
+        vbroadcastss         -7 * SIZE(BO, BI, SIZE), %xmm5
+        VFMADDPS_I(        %xmm9,%xmm5,%xmm0  )
+        VFMADDPS_I(        %xmm13,%xmm5,%xmm1 )
+        vbroadcastss         -6 * SIZE(BO, BI, SIZE), %xmm6
+        VFMADDPS_R(        %xmm10,%xmm6,%xmm0 )
+        VFMADDPS_R(        %xmm14,%xmm6,%xmm1 )
+        vbroadcastss         -5 * SIZE(BO, BI, SIZE), %xmm7
+        VFMADDPS_I(        %xmm11,%xmm7,%xmm0 )
+        VFMADDPS_I(        %xmm15,%xmm7,%xmm1 )
+        addq    $ 4, BI                           
+        addq    $ 8, %rax                         
+.endm
+
+.macro SAVE4x2
+
+	vbroadcastss	ALPHA_R, %xmm0
+	vbroadcastss	ALPHA_I, %xmm1
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
+        vshufps $ 0xb1, %xmm13, %xmm13, %xmm13
+        vshufps $ 0xb1, %xmm15, %xmm15, %xmm15
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm11,%xmm10, %xmm10
+        vaddsubps %xmm13,%xmm12, %xmm12
+        vaddsubps %xmm15,%xmm14, %xmm14
+
+        vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
+        vshufps $ 0xb1, %xmm10, %xmm10, %xmm11
+        vshufps $ 0xb1, %xmm12, %xmm12, %xmm13
+        vshufps $ 0xb1, %xmm14, %xmm14, %xmm15
+
+#else
+        vaddsubps %xmm8,  %xmm9 ,%xmm9
+        vaddsubps %xmm10, %xmm11,%xmm11
+        vaddsubps %xmm12, %xmm13,%xmm13
+        vaddsubps %xmm14, %xmm15,%xmm15
+
+        vmovaps   %xmm9,  %xmm8
+        vmovaps   %xmm11, %xmm10
+        vmovaps   %xmm13, %xmm12
+        vmovaps   %xmm15, %xmm14
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
+        vshufps $ 0xb1, %xmm13, %xmm13, %xmm13
+        vshufps $ 0xb1, %xmm15, %xmm15, %xmm15
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %xmm8 , %xmm0, %xmm8
+        vmulps  %xmm10, %xmm0, %xmm10
+        vmulps  %xmm12, %xmm0, %xmm12
+        vmulps  %xmm14, %xmm0, %xmm14
+
+	// multiply with ALPHA_I
+        vmulps  %xmm9 , %xmm1, %xmm9
+        vmulps  %xmm11, %xmm1, %xmm11
+        vmulps  %xmm13, %xmm1, %xmm13
+        vmulps  %xmm15, %xmm1, %xmm15
+
+	vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm11,%xmm10, %xmm10
+        vaddsubps %xmm13,%xmm12, %xmm12
+        vaddsubps %xmm15,%xmm14, %xmm14
+
+#ifndef TRMMKERNEL
+
+	vaddps 	 	(CO1), %xmm8 , %xmm8
+	vaddps  4 * SIZE(CO1), %xmm12, %xmm12
+
+	vaddps 	 	(CO1, LDC), %xmm10, %xmm10
+	vaddps  4 * SIZE(CO1, LDC), %xmm14, %xmm14
+
+#endif
+
+	vmovups	%xmm8 ,  	(CO1)
+	vmovups	%xmm12 , 4 * SIZE(CO1)
+
+	vmovups	%xmm10 ,  	(CO1, LDC)
+	vmovups	%xmm14 , 4 * SIZE(CO1, LDC)
+
+.endm
+
+/************************************************************************************************/
+
+.macro KERNEL2x2_SUB
+        vmovups         -16 * SIZE(AO, %rax, SIZE), %xmm0
+        vbroadcastss         -8 * SIZE(BO, BI, SIZE), %xmm4
+        VFMADDPS_R(        %xmm8,%xmm4,%xmm0  )
+        vbroadcastss         -7 * SIZE(BO, BI, SIZE), %xmm5
+        VFMADDPS_I(        %xmm9,%xmm5,%xmm0  )
+        vbroadcastss         -6 * SIZE(BO, BI, SIZE), %xmm6
+        VFMADDPS_R(        %xmm10,%xmm6,%xmm0 )
+        vbroadcastss         -5 * SIZE(BO, BI, SIZE), %xmm7
+        VFMADDPS_I(        %xmm11,%xmm7,%xmm0 )
+        addq    $ 4, BI                           
+        addq    $ 4, %rax                         
+.endm
+
+.macro SAVE2x2
+
+	vbroadcastss	ALPHA_R, %xmm0
+	vbroadcastss	ALPHA_I, %xmm1
+
+	// swap high and low 4 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm11,%xmm10, %xmm10
+
+        vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
+        vshufps $ 0xb1, %xmm10, %xmm10, %xmm11
+
+#else
+        vaddsubps %xmm8,  %xmm9 ,%xmm9
+        vaddsubps %xmm10, %xmm11,%xmm11
+
+        vmovaps   %xmm9,  %xmm8
+        vmovaps   %xmm11, %xmm10
+
+	// swap high and low 4 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %xmm8 , %xmm0, %xmm8
+        vmulps  %xmm10, %xmm0, %xmm10
+
+	// multiply with ALPHA_I
+        vmulps  %xmm9 , %xmm1, %xmm9
+        vmulps  %xmm11, %xmm1, %xmm11
+
+	vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm11,%xmm10, %xmm10
+
+#ifndef TRMMKERNEL
+
+	vaddps 	 	(CO1), %xmm8 , %xmm8
+
+	vaddps 	 	(CO1, LDC), %xmm10, %xmm10
+
+#endif
+
+	vmovups	%xmm8 ,  	(CO1)
+
+	vmovups	%xmm10 ,  	(CO1, LDC)
+
+.endm
+
+/************************************************************************************************/
+
+.macro KERNEL1x2_SUB
+        vmovsd         -16 * SIZE(AO, %rax, SIZE), %xmm0
+        vbroadcastss         -8 * SIZE(BO, BI, SIZE), %xmm4
+        VFMADDPS_R(        %xmm8,%xmm4,%xmm0  )
+        vbroadcastss         -7 * SIZE(BO, BI, SIZE), %xmm5
+        VFMADDPS_I(        %xmm9,%xmm5,%xmm0  )
+        vbroadcastss         -6 * SIZE(BO, BI, SIZE), %xmm6
+        VFMADDPS_R(        %xmm10,%xmm6,%xmm0 )
+        vbroadcastss         -5 * SIZE(BO, BI, SIZE), %xmm7
+        VFMADDPS_I(        %xmm11,%xmm7,%xmm0 )
+        addq    $ 4, BI                           
+        addq    $ 2, %rax                         
+.endm
+
+.macro SAVE1x2
+
+	vbroadcastss	ALPHA_R, %xmm0
+	vbroadcastss	ALPHA_I, %xmm1
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm11,%xmm10, %xmm10
+
+        vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
+        vshufps $ 0xb1, %xmm10, %xmm10, %xmm11
+
+#else
+        vaddsubps %xmm8,  %xmm9 ,%xmm9
+        vaddsubps %xmm10, %xmm11,%xmm11
+
+        vmovaps   %xmm9,  %xmm8
+        vmovaps   %xmm11, %xmm10
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %xmm8 , %xmm0, %xmm8
+        vmulps  %xmm10, %xmm0, %xmm10
+
+	// multiply with ALPHA_I
+        vmulps  %xmm9 , %xmm1, %xmm9
+        vmulps  %xmm11, %xmm1, %xmm11
+
+	vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm11,%xmm10, %xmm10
+
+#ifndef TRMMKERNEL
+
+	vmovsd		(CO1), %xmm14
+	vaddps 	 	%xmm14, %xmm8 , %xmm8
+
+	vmovsd		(CO1, LDC), %xmm15
+	vaddps 	 	%xmm15, %xmm10, %xmm10
+
+#endif
+
+	vmovsd	%xmm8 ,  	(CO1)
+	vmovsd	%xmm10 ,  	(CO1, LDC)
+
+.endm
+
+/************************************************************************************************/
+
+.macro KERNEL8x1_SUB
+        vmovups         -16 * SIZE(AO, %rax, SIZE), %ymm0
+        vmovups          -8 * SIZE(AO, %rax, SIZE), %ymm1
+        vbroadcastss         -4 * SIZE(BO, BI, SIZE), %ymm4
+        VFMADDPS_YR(        %ymm8,%ymm4,%ymm0  )
+        VFMADDPS_YR(        %ymm12,%ymm4,%ymm1 )
+        vbroadcastss         -3 * SIZE(BO, BI, SIZE), %ymm5
+        VFMADDPS_YI(        %ymm9,%ymm5,%ymm0  )
+        VFMADDPS_YI(        %ymm13,%ymm5,%ymm1 )
+        addq    $ 2 , BI                           
+        addq    $ 16, %rax                         
+.endm
+
+.macro SAVE8x1
+
+	vbroadcastss	ALPHA_R, %ymm0
+	vbroadcastss	ALPHA_I, %ymm1
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %ymm9 , %ymm9, %ymm9
+        vshufps $ 0xb1, %ymm13, %ymm13, %ymm13
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %ymm9, %ymm8 , %ymm8
+        vaddsubps %ymm13,%ymm12, %ymm12
+
+        vshufps $ 0xb1, %ymm8 , %ymm8, %ymm9
+        vshufps $ 0xb1, %ymm12, %ymm12, %ymm13
+
+#else
+        vaddsubps %ymm8,  %ymm9 ,%ymm9
+        vaddsubps %ymm12, %ymm13,%ymm13
+
+        vmovaps   %ymm9,  %ymm8
+        vmovaps   %ymm13, %ymm12
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %ymm9 , %ymm9, %ymm9
+        vshufps $ 0xb1, %ymm13, %ymm13, %ymm13
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %ymm8 , %ymm0, %ymm8
+        vmulps  %ymm12, %ymm0, %ymm12
+
+	// multiply with ALPHA_I
+        vmulps  %ymm9 , %ymm1, %ymm9
+        vmulps  %ymm13, %ymm1, %ymm13
+
+	vaddsubps %ymm9, %ymm8 , %ymm8
+        vaddsubps %ymm13,%ymm12, %ymm12
+
+
+
+#ifndef TRMMKERNEL
+
+	vaddps 	 	(CO1), %ymm8 , %ymm8
+	vaddps  8 * SIZE(CO1), %ymm12, %ymm12
+
+#endif
+
+	vmovups	%ymm8 ,  	(CO1)
+	vmovups	%ymm12 , 8 * SIZE(CO1)
+
+.endm
+
+
+/************************************************************************************************/
+
+.macro KERNEL4x1_SUB
+        vmovups         -16 * SIZE(AO, %rax, SIZE), %xmm0
+        vbroadcastss         -4 * SIZE(BO, BI, SIZE), %xmm4
+        VFMADDPS_R(        %xmm8,%xmm4,%xmm0  )
+        vmovups         -12 * SIZE(AO, %rax, SIZE), %xmm1
+        VFMADDPS_R(        %xmm12,%xmm4,%xmm1 )
+        vbroadcastss         -3 * SIZE(BO, BI, SIZE), %xmm5
+        VFMADDPS_I(        %xmm9,%xmm5,%xmm0  )
+        VFMADDPS_I(        %xmm13,%xmm5,%xmm1 )
+        addq    $ 2, BI                           
+        addq    $ 8, %rax                         
+.endm
+
+.macro SAVE4x1
+
+	vbroadcastss	ALPHA_R, %xmm0
+	vbroadcastss	ALPHA_I, %xmm1
+
+	// swap high and low 4 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm13, %xmm13, %xmm13
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm13,%xmm12, %xmm12
+
+        vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
+        vshufps $ 0xb1, %xmm12, %xmm12, %xmm13
+
+#else
+        vaddsubps %xmm8,  %xmm9 ,%xmm9
+        vaddsubps %xmm12, %xmm13,%xmm13
+
+        vmovaps   %xmm9,  %xmm8
+        vmovaps   %xmm13, %xmm12
+
+	// swap high and low 4 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm13, %xmm13, %xmm13
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %xmm8 , %xmm0, %xmm8
+        vmulps  %xmm12, %xmm0, %xmm12
+
+	// multiply with ALPHA_I
+        vmulps  %xmm9 , %xmm1, %xmm9
+        vmulps  %xmm13, %xmm1, %xmm13
+
+	vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm13,%xmm12, %xmm12
+
+#ifndef TRMMKERNEL
+
+	vaddps 	 	(CO1), %xmm8 , %xmm8
+	vaddps  4 * SIZE(CO1), %xmm12, %xmm12
+
+#endif
+
+	vmovups	%xmm8 ,  	(CO1)
+	vmovups	%xmm12 , 4 * SIZE(CO1)
+
+.endm
+
+/************************************************************************************************/
+
+.macro KERNEL2x1_SUB
+        vmovups         -16 * SIZE(AO, %rax, SIZE), %xmm0
+        vbroadcastss         -4 * SIZE(BO, BI, SIZE), %xmm4
+        VFMADDPS_R(        %xmm8,%xmm4,%xmm0  )
+        vbroadcastss         -3 * SIZE(BO, BI, SIZE), %xmm5
+        VFMADDPS_I(        %xmm9,%xmm5,%xmm0  )
+        addq    $ 2, BI                           
+        addq    $ 4, %rax                         
+.endm
+
+.macro SAVE2x1
+
+	vbroadcastss	ALPHA_R, %xmm0
+	vbroadcastss	ALPHA_I, %xmm1
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %xmm9, %xmm8 , %xmm8
+
+        vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
+
+#else
+        vaddsubps %xmm8,  %xmm9 ,%xmm9
+
+        vmovaps   %xmm9,  %xmm8
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %xmm8 , %xmm0, %xmm8
+
+	// multiply with ALPHA_I
+        vmulps  %xmm9 , %xmm1, %xmm9
+
+	vaddsubps %xmm9, %xmm8 , %xmm8
+
+#ifndef TRMMKERNEL
+
+	vaddps 	 	(CO1), %xmm8 , %xmm8
+
+#endif
+
+	vmovups	%xmm8 ,  	(CO1)
+
+.endm
+
+/************************************************************************************************/
+
+.macro KERNEL1x1_SUB
+        vmovsd         -16 * SIZE(AO, %rax, SIZE), %xmm0
+        vbroadcastss         -4 * SIZE(BO, BI, SIZE), %xmm4
+        VFMADDPS_R(        %xmm8,%xmm4,%xmm0 )
+        vbroadcastss         -3 * SIZE(BO, BI, SIZE), %xmm5
+        VFMADDPS_I(        %xmm9,%xmm5,%xmm0 )
+        addq    $ 2, BI                           
+        addq    $ 2, %rax                         
+.endm
+
+.macro SAVE1x1
+
+	vbroadcastss	ALPHA_R, %xmm0
+	vbroadcastss	ALPHA_I, %xmm1
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %xmm9, %xmm8 , %xmm8
+
+        vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
+
+#else
+        vaddsubps %xmm8,  %xmm9 ,%xmm9
+
+        vmovaps   %xmm9,  %xmm8
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %xmm8 , %xmm0, %xmm8
+
+	// multiply with ALPHA_I
+        vmulps  %xmm9 , %xmm1, %xmm9
+
+	vaddsubps %xmm9, %xmm8 , %xmm8
+
+#ifndef TRMMKERNEL
+
+	vmovsd		(CO1), %xmm14
+	vaddps 	 	%xmm14, %xmm8 , %xmm8
+
+#endif
+
+	vmovsd	%xmm8 ,  	(CO1)
+
+.endm
+
+/************************************************************************************************/
+
+
+
+
+	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
+
+	vmovss	 %xmm0, ALPHA_R
+	vmovss	 %xmm1, ALPHA_I
+
+	salq	$ ZBASE_SHIFT, LDC
+
+	movq    N, %rax
+        xorq    %rdx, %rdx
+        movq    $ 2,  %rdi
+        divq    %rdi                    //    N / 2
+        movq    %rax, Ndiv6             //    N / 2
+        movq    %rdx, Nmod6             //    N % 2
+
+	
+
+#ifdef TRMMKERNEL
+	vmovsd	%xmm12, OFFSET
+	vmovsd	%xmm12, KK
+#ifndef LEFT
+	negq	KK
+#endif	
+#endif
+
+.L2_0:
+
+	movq	Ndiv6,  J
+	cmpq	$ 0, J
+	je	.L1_0
+	ALIGN_4
+
+
+
+.L2_01:
+	// copy to sub buffer
+	movq	B, BO1
+	leaq    BUFFER1, BO		// first buffer to BO
+	movq	K, %rax
+	ALIGN_4
+
+.L2_02b:
+
+	vmovups	(BO1), %xmm0
+	vmovups	%xmm0,       (BO)
+	addq	$ 4*SIZE,BO1
+	addq	$ 4*SIZE,BO
+	decq	%rax
+	jnz	.L2_02b
+
+.L2_02c:
+
+	movq	BO1, B			// next offset of B
+
+.L2_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	$ 16 * SIZE, AO
+
+	movq	M,  I
+	sarq	$ 3, I			// i = (m >> 3)
+	je	.L2_4_10
+
+	ALIGN_4
+/**********************************************************************************************************/
+
+.L2_8_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	$ 4, %rax			// rax = rax *16 ; 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    $ 8, %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_8_16
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 4, %rax			// rax = rax *16 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L2_8_12:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+
+	je	.L2_8_16
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x2_SUB
+
+	je	.L2_8_16
+
+	jmp	.L2_8_12
+	ALIGN_4
+
+.L2_8_16:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L2_8_19
+
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 4, %rax			// rax = rax *16 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L2_8_17:
+
+	KERNEL8x2_SUB
+
+	jl	.L2_8_17
+	ALIGN_4
+
+
+.L2_8_19:
+
+	SAVE8x2
+
+
+#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	$ 4, %rax			// rax = rax *16 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        addq    $ 8, KK
+#endif
+
+	addq	$ 16 * SIZE, CO1		# coffset += 16
+	decq	I			# i --
+	jg	.L2_8_11
+	ALIGN_4	
+
+
+/**********************************************************************************************************/
+
+
+
+
+.L2_4_10:
+	testq	$ 7, M		
+	jz	.L2_4_60		// to next 2 lines of N
+
+	testq	$ 4, M		
+	jz	.L2_4_20
+	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
+	KERNEL4x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	je	.L2_4_16
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x2_SUB
+	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
+	ALIGN_4	
+
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+
+.L2_4_20:
+
+	testq	$ 2, M		
+	jz	.L2_4_40
+	ALIGN_4
+
+.L2_4_21:
+
+#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_4_26
+	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_4_22:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	je	.L2_4_26
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	je	.L2_4_26
+
+	jmp	.L2_4_22
+	ALIGN_4
+
+.L2_4_26:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L2_4_29
+
+	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_4_27:
+
+	KERNEL2x2_SUB
+
+	jl	.L2_4_27
+	ALIGN_4
+
+
+.L2_4_29:
+
+	vbroadcastss	ALPHA_R, %xmm0
+	vbroadcastss	ALPHA_I, %xmm1
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
+
+#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
+    defined(NR) || defined(NC) || defined(TR) || defined(TC)
+
+        vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm11,%xmm10, %xmm10
+
+        vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
+        vshufps $ 0xb1, %xmm10, %xmm10, %xmm11
+
+#else
+        vaddsubps %xmm8,  %xmm9 ,%xmm9
+        vaddsubps %xmm10, %xmm11,%xmm11
+
+        vmovaps   %xmm9,  %xmm8
+        vmovaps   %xmm11, %xmm10
+
+	// swap high and low 64 bytes
+        vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
+        vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
+
+#endif
+
+	// multiply with ALPHA_R
+        vmulps  %xmm8 , %xmm0, %xmm8
+        vmulps  %xmm10, %xmm0, %xmm10
+
+	// multiply with ALPHA_I
+        vmulps  %xmm9 , %xmm1, %xmm9
+        vmulps  %xmm11, %xmm1, %xmm11
+
+	vaddsubps %xmm9, %xmm8 , %xmm8
+        vaddsubps %xmm11,%xmm10, %xmm10
+
+
+
+#ifndef TRMMKERNEL
+
+	vaddps 	 	(CO1), %xmm8 , %xmm8
+
+	vaddps 	 	(CO1, LDC), %xmm10, %xmm10
+
+#endif
+
+	vmovups	%xmm8 ,  	(CO1)
+
+	vmovups	%xmm10 ,  	(CO1, LDC)
+
+
+
+#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
+	decq	I			# i --
+	jg	.L2_4_21
+	ALIGN_4	
+
+
+
+/**************************************************************************/
+.L2_4_40:
+	testq	$ 1, M		
+	jz	.L2_4_60		// to next 2 lines of N
+
+	ALIGN_4
+
+.L2_4_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_4_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_4_42:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	je	.L2_4_46
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	je	.L2_4_46
+
+	jmp	.L2_4_42
+	ALIGN_4
+
+.L2_4_46:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L2_4_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_4_47:
+
+	KERNEL1x2_SUB
+
+	jl	.L2_4_47
+	ALIGN_4
+
+
+.L2_4_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_4_41
+	ALIGN_4	
+
+
+
+	
+.L2_4_60:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+        addq    $ 2, KK
+#endif
+
+	decq	J			// j --
+	jg	.L2_01			// next 2 lines of N
+
+
+
+.L1_0:
+
+/************************************************************************************************
+* Loop for Nmod6 % 2 > 0
+*************************************************************************************************/
+
+	movq	Nmod6, J		
+	andq	$ 1, J			// j % 2
+	je	.L999
+	ALIGN_4
+
+.L1_01:
+	// copy to sub buffer
+	movq	B, BO1
+	leaq    BUFFER1, BO		// first buffer to BO
+	movq	K, %rax
+	ALIGN_4
+
+.L1_02b:
+
+	vmovsd		(BO1), %xmm0
+	vmovsd	%xmm0,       (BO)
+	addq	$ 2*SIZE,BO1
+	addq	$ 2*SIZE,BO
+	decq	%rax
+	jnz	.L1_02b
+
+.L1_02c:
+
+	movq	BO1, B			// next offset of B
+
+.L1_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	$ 16 * SIZE, AO
+
+	movq	M,  I
+	sarq	$ 3, I			// i = (m >> 3)
+	je	.L1_4_10
+
+	ALIGN_4
+
+/**************************************************************************************************/
+
+.L1_8_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	$ 4, %rax			// rax = rax *16 ; 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    $ 8, %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_8_16
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
+
+	salq	$ 4, %rax			// rax = rax *16 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L1_8_12:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+
+	je	.L1_8_16
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL8x1_SUB
+
+	je	.L1_8_16
+
+	jmp	.L1_8_12
+	ALIGN_4
+
+.L1_8_16:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L1_8_19
+
+	movq    %rax, BI                        //  Index for BO
+        leaq    ( ,BI,2), BI                    //  BI = BI * 4 ; number of values
+
+	salq	$ 4, %rax			// rax = rax *16 ; number of values
+	leaq	(AO, %rax, SIZE), AO
+	leaq	(BO, BI, SIZE), BO
+	negq	BI
+	negq	%rax
+	ALIGN_4
+
+.L1_8_17:
+
+	KERNEL8x1_SUB
+
+	jl	.L1_8_17
+	ALIGN_4
+
+
+.L1_8_19:
+
+	SAVE8x1
+
+
+#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	$ 4, %rax			// rax = rax *16 ; number of values
+        leaq    (AO, %rax, SIZE), AO
+#endif
+
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+        addq    $ 8, KK
+#endif
+
+	addq	$ 16 * SIZE, CO1		# coffset += 16
+	decq	I			# i --
+	jg	.L1_8_11
+	ALIGN_4	
+
+
+
+/**************************************************************************************************/
+.L1_4_10:
+
+	testq	$ 7, M		
+	jz	.L999
+
+	testq	$ 4, M		
+	jz	.L1_4_20
+
+
+.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:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	je	.L1_4_16
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	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 * 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
+
+.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
+	ALIGN_4	
+
+
+
+/**************************************************************************
+* Rest of M 
+***************************************************************************/
+
+.L1_4_20:
+
+	testq	$ 2, M		
+	jz	.L1_4_40
+	ALIGN_4
+
+.L1_4_21:
+
+#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_4_26
+	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_4_22:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	je	.L1_4_26
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	je	.L1_4_26
+
+	jmp	.L1_4_22
+	ALIGN_4
+
+.L1_4_26:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L1_4_29
+
+	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_4_27:
+
+	KERNEL2x1_SUB
+
+	jl	.L1_4_27
+	ALIGN_4
+
+
+.L1_4_29:
+
+	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	
+
+
+
+/**************************************************************************/
+.L1_4_40:
+	testq	$ 1, M		
+	jz	.L999		// to next 2 lines of N
+
+	ALIGN_4
+
+.L1_4_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_4_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_4_42:
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	je	.L1_4_46
+
+	prefetcht0	A_PR1(AO,%rax,SIZE)
+	prefetcht0	B_PR1(BO,BI,SIZE)
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	je	.L1_4_46
+
+	jmp	.L1_4_42
+	ALIGN_4
+
+.L1_4_46:
+#ifndef TRMMKERNEL
+        movq    K, %rax
+#else
+        movq    KKK, %rax
+#endif
+
+	andq	$ 7, %rax		# if (k & 1)
+	je .L1_4_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_4_47:
+
+	KERNEL1x1_SUB
+
+	jl	.L1_4_47
+	ALIGN_4
+
+
+.L1_4_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
+	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
diff --git a/kernel/x86_64/zgemm_kernel_2x2_bulldozer.S b/kernel/x86_64/zgemm_kernel_2x2_bulldozer.S
index 7cf1fda8e..94e2f6117 100644
--- a/kernel/x86_64/zgemm_kernel_2x2_bulldozer.S
+++ b/kernel/x86_64/zgemm_kernel_2x2_bulldozer.S
@@ -412,16 +412,16 @@
 #ifdef WINDOWS_ABI
 	movq	%rdi,    48(%rsp)
 	movq	%rsi,    56(%rsp)
-	movups	%xmm6,   64(%rsp)
-	movups	%xmm7,   80(%rsp)
-	movups	%xmm8,   96(%rsp)
-	movups	%xmm9,  112(%rsp)
-	movups	%xmm10, 128(%rsp)
-	movups	%xmm11, 144(%rsp)
-	movups	%xmm12, 160(%rsp)
-	movups	%xmm13, 176(%rsp)
-	movups	%xmm14, 192(%rsp)
-	movups	%xmm15, 208(%rsp)
+	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
@@ -431,14 +431,15 @@
 	movq	OLD_C,     C
 	movq	OLD_LDC,   LDC
 #ifdef TRMMKERNEL
-	movsd	OLD_OFFSET, %xmm12
+	vmovsd	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
+	vmovsd	STACKSIZE + 16(%rsp), %xmm12
 #endif
 
 #endif
@@ -1372,6 +1373,8 @@
 
 
 .L999:
+	vzeroupper
+
 	movq   		SP, %rsp
 	movq	   (%rsp), %rbx
 	movq	  8(%rsp), %rbp
@@ -1383,16 +1386,16 @@
 #ifdef WINDOWS_ABI
 	movq	 48(%rsp), %rdi
 	movq	 56(%rsp), %rsi
-	movups	 64(%rsp), %xmm6
-	movups	 80(%rsp), %xmm7
-	movups	 96(%rsp), %xmm8
-	movups	112(%rsp), %xmm9
-	movups	128(%rsp), %xmm10
-	movups	144(%rsp), %xmm11
-	movups	160(%rsp), %xmm12
-	movups	176(%rsp), %xmm13
-	movups	192(%rsp), %xmm14
-	movups	208(%rsp), %xmm15
+	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
diff --git a/kernel/x86_64/zgemm_kernel_2x2_piledriver.S b/kernel/x86_64/zgemm_kernel_2x2_piledriver.S
index b4c416b64..848b6f237 100644
--- a/kernel/x86_64/zgemm_kernel_2x2_piledriver.S
+++ b/kernel/x86_64/zgemm_kernel_2x2_piledriver.S
@@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 /*********************************************************************
 *
-* 2013/10/30 Saar
+* 2014/06/28 Saar
 *        BLASTEST               : OK
 *        CTEST                  : OK
 *        TEST                   : OK
@@ -437,16 +437,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	%rdi,    48(%rsp)
 	movq	%rsi,    56(%rsp)
-	movups	%xmm6,   64(%rsp)
-	movups	%xmm7,   80(%rsp)
-	movups	%xmm8,   96(%rsp)
-	movups	%xmm9,  112(%rsp)
-	movups	%xmm10, 128(%rsp)
-	movups	%xmm11, 144(%rsp)
-	movups	%xmm12, 160(%rsp)
-	movups	%xmm13, 176(%rsp)
-	movups	%xmm14, 192(%rsp)
-	movups	%xmm15, 208(%rsp)
+	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
@@ -456,14 +456,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	movq	OLD_C,     C
 	movq	OLD_LDC,   LDC
 #ifdef TRMMKERNEL
-	movsd	OLD_OFFSET, %xmm12
+	vmovsd	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
+	vmovsd	STACKSIZE + 16(%rsp), %xmm12
 #endif
 
 #endif
@@ -1397,6 +1398,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .L999:
+	vzeroupper
+
 	movq   		SP, %rsp
 	movq	   (%rsp), %rbx
 	movq	  8(%rsp), %rbp
@@ -1408,16 +1411,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	 48(%rsp), %rdi
 	movq	 56(%rsp), %rsi
-	movups	 64(%rsp), %xmm6
-	movups	 80(%rsp), %xmm7
-	movups	 96(%rsp), %xmm8
-	movups	112(%rsp), %xmm9
-	movups	128(%rsp), %xmm10
-	movups	144(%rsp), %xmm11
-	movups	160(%rsp), %xmm12
-	movups	176(%rsp), %xmm13
-	movups	192(%rsp), %xmm14
-	movups	208(%rsp), %xmm15
+	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
diff --git a/kernel/x86_64/zgemm_kernel_4x2_haswell.S b/kernel/x86_64/zgemm_kernel_4x2_haswell.S
index 289fbbfe9..a71fff7af 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.
 **********************************************************************************/
 
 /********************************************************************************
-* 2013/11/13 Saar
+* 2014/06/28 Saar
 *        BLASTEST               : OK
 *        CTEST                  : OK
 *        TEST                   : OK
@@ -693,16 +693,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	%rdi,    48(%rsp)
 	movq	%rsi,    56(%rsp)
-	movups	%xmm6,   64(%rsp)
-	movups	%xmm7,   80(%rsp)
-	movups	%xmm8,   96(%rsp)
-	movups	%xmm9,  112(%rsp)
-	movups	%xmm10, 128(%rsp)
-	movups	%xmm11, 144(%rsp)
-	movups	%xmm12, 160(%rsp)
-	movups	%xmm13, 176(%rsp)
-	movups	%xmm14, 192(%rsp)
-	movups	%xmm15, 208(%rsp)
+	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
@@ -715,6 +715,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	movsd	OLD_OFFSET, %xmm12
 #endif
 	vmovaps	%xmm3, %xmm0
+	vmovsd   OLD_ALPHA_I, %xmm1
 
 #else
 	movq	STACKSIZE +  8(%rsp), LDC
@@ -1781,6 +1782,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .L999:
+	vzeroupper
+
 	movq   		SP, %rsp
 	movq	   (%rsp), %rbx
 	movq	  8(%rsp), %rbp
@@ -1792,16 +1795,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	 48(%rsp), %rdi
 	movq	 56(%rsp), %rsi
-	movups	 64(%rsp), %xmm6
-	movups	 80(%rsp), %xmm7
-	movups	 96(%rsp), %xmm8
-	movups	112(%rsp), %xmm9
-	movups	128(%rsp), %xmm10
-	movups	144(%rsp), %xmm11
-	movups	160(%rsp), %xmm12
-	movups	176(%rsp), %xmm13
-	movups	192(%rsp), %xmm14
-	movups	208(%rsp), %xmm15
+	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
diff --git a/make.inc b/make.inc
index affae3a2d..da430b7d6 100644
--- a/make.inc
+++ b/make.inc
@@ -1,7 +1,7 @@
 SHELL = /bin/sh
 PLAT = _LINUX
 DRVOPTS  = $(OPTS)
-LOADER   = $(FORTRAN) -pthread
+LOADER   = $(FORTRAN)
 ARCHFLAGS= -ru
 #RANLIB   = ranlib
 
diff --git a/param.h b/param.h
index 71af5c664..880219b7c 100644
--- a/param.h
+++ b/param.h
@@ -1111,14 +1111,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SGEMM_DEFAULT_UNROLL_M 16
 #define DGEMM_DEFAULT_UNROLL_M 8
 #define QGEMM_DEFAULT_UNROLL_M 2
-#define CGEMM_DEFAULT_UNROLL_M 2
+#define CGEMM_DEFAULT_UNROLL_M 8
 #define ZGEMM_DEFAULT_UNROLL_M 4
 #define XGEMM_DEFAULT_UNROLL_M 1
 
 #define SGEMM_DEFAULT_UNROLL_N 4
 #define DGEMM_DEFAULT_UNROLL_N 4
 #define QGEMM_DEFAULT_UNROLL_N 2
-#define CGEMM_DEFAULT_UNROLL_N 4
+#define CGEMM_DEFAULT_UNROLL_N 2
 #define ZGEMM_DEFAULT_UNROLL_N 4
 #define XGEMM_DEFAULT_UNROLL_N 1
 #endif
@@ -1134,7 +1134,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define QGEMM_DEFAULT_P 504
 #define QGEMM_DEFAULT_R qgemm_r
 
-#define CGEMM_DEFAULT_P 128
+#define CGEMM_DEFAULT_P 384
 //#define CGEMM_DEFAULT_R cgemm_r
 #define CGEMM_DEFAULT_R 1024
 
@@ -1148,7 +1148,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SGEMM_DEFAULT_Q 384
 #define DGEMM_DEFAULT_Q 256
 #define QGEMM_DEFAULT_Q 128
-#define CGEMM_DEFAULT_Q 256
+#define CGEMM_DEFAULT_Q 192
 #define ZGEMM_DEFAULT_Q 192
 #define XGEMM_DEFAULT_Q 128