diff --git a/CMakeLists.txt b/CMakeLists.txt
index 50da721cd..74db77135 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -6,7 +6,7 @@ cmake_minimum_required(VERSION 2.8.5)
 project(OpenBLAS C ASM)
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 7.dev)
+set(OpenBLAS_PATCH_VERSION 8.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
@@ -211,7 +211,8 @@ if (USE_THREAD)
   target_link_libraries(${OpenBLAS_LIBNAME} ${CMAKE_THREAD_LIBS_INIT})
 endif()
 
-if (MSVC OR NOT NOFORTRAN)
+#if (MSVC OR NOT NOFORTRAN)
+if (NOT NO_CBLAS)
   # Broken without fortran on unix
   add_subdirectory(utest)
 endif()
diff --git a/Changelog.txt b/Changelog.txt
index 8df35d5c3..f160a4e13 100644
--- a/Changelog.txt
+++ b/Changelog.txt
@@ -1,4 +1,46 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.7
+11-Aug 2019
+
+common:
+        * having the gmake special variables TARGET_ARCH or TARGET_MACH
+          defined no longer causes build failures in ctest or utest
+        * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
+          has the same effect as setting them to 1
+        * a new test program was added to allow checking the library for
+          thread safety
+        * a new option USE_LOCKING was added to ensure thread safety when
+          OpenBLAS itself is built without multithreading but will be 
+          called from multiple threads.
+        * a build failure on Linux with glibc versions earlier than 2.5
+          was fixed
+        * a runtime error with CPU enumeration (and NO_AFFINITY not set)
+          on glibc 2.6 was fixed
+        * NO_AFFINITY was added to the CMAKE options (and defaults to being
+          active on Linux, as in the gmake builds)
+
+x86_64:
+        * the build-time logic for detection of AVX512 availability in
+          the processor and compiler was fixed
+        * gmake builds on OSX now set the internal name of the library to 
+          libopenblas.0.dylib (consistent with CMAKE)
+        * the Haswell DGEMM kernel received a significant speedup through
+          improved prefetch and load instructions
+        * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
+          increased by avoiding vpermpd instructions
+        * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
+          to fix remaining errors in DGEMM, DSYMM and DTRMM
+
+## POWER:
+        * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
+        * added optimized kernels for POWER9 SGEMM and STRMM
+
+## ARMV7:
+        * fixed the softfp implementations of xAMAX and IxAMAX
+        * removed the predefined -march= flags on both ARMV5 and ARMV6 as
+          they were appropriate for only a subset of platforms
+
 ====================================================================
 Version 0.3.6
 29-Apr-2019
diff --git a/Makefile b/Makefile
index 07b08439e..60f189ef2 100644
--- a/Makefile
+++ b/Makefile
@@ -109,6 +109,7 @@ endif
 ifeq ($(OSNAME), Darwin)
 	@$(MAKE) -C exports dyn
 	@ln -fs $(LIBDYNNAME) $(LIBPREFIX).dylib
+	@ln -fs $(LIBDYNNAME) $(LIBPREFIX).$(MAJOR_VERSION).dylib
 endif
 ifeq ($(OSNAME), WINNT)
 	@$(MAKE) -C exports dll
diff --git a/Makefile.arm b/Makefile.arm
index eedd39b73..b5d80f8e6 100644
--- a/Makefile.arm
+++ b/Makefile.arm
@@ -1,7 +1,7 @@
 ifeq ($(CORE), $(filter $(CORE),ARMV7 CORTEXA9 CORTEXA15))
 ifeq ($(OSNAME), Android)
-CCOMMON_OPT += -mfpu=neon  -march=armv7-a
-FCOMMON_OPT += -mfpu=neon  -march=armv7-a
+CCOMMON_OPT += -mfpu=neon
+FCOMMON_OPT += -mfpu=neon
 else
 CCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
 FCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
@@ -9,11 +9,6 @@ endif
 endif
 
 ifeq ($(CORE), ARMV6)
-CCOMMON_OPT += -mfpu=vfp -march=armv6
-FCOMMON_OPT += -mfpu=vfp -march=armv6
-endif
-
-ifeq ($(CORE), ARMV5)
-CCOMMON_OPT += -march=armv5
-FCOMMON_OPT += -march=armv5
+CCOMMON_OPT += -mfpu=vfp
+FCOMMON_OPT += -mfpu=vfp
 endif
diff --git a/Makefile.install b/Makefile.install
index fefecd98d..8070b4729 100644
--- a/Makefile.install
+++ b/Makefile.install
@@ -83,7 +83,8 @@ ifeq ($(OSNAME), Darwin)
 	@-cp $(LIBDYNNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@-install_name_tool -id "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)/$(LIBDYNNAME)" "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)/$(LIBDYNNAME)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
-	ln -fs $(LIBDYNNAME) $(LIBPREFIX).dylib
+	ln -fs $(LIBDYNNAME) $(LIBPREFIX).dylib ; \
+	ln -fs $(LIBDYNNAME) $(LIBPREFIX).$(MAJOR_VERSION).dylib
 endif
 ifeq ($(OSNAME), WINNT)
 	@-cp $(LIBDLLNAME) "$(DESTDIR)$(OPENBLAS_BINARY_DIR)"
diff --git a/Makefile.rule b/Makefile.rule
index a299588e0..c0941e488 100644
--- a/Makefile.rule
+++ b/Makefile.rule
@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.7.dev
+VERSION = 0.3.8.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
diff --git a/Makefile.system b/Makefile.system
index 16791bcc2..a54282f6c 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -142,9 +142,9 @@ endif
 endif
 
 
-# On x86_64 build getarch with march=native. This is required to detect AVX512 support in getarch.
+# On x86_64 build getarch with march=native unless the compiler is PGI. This is required to detect AVX512 support in getarch.
 ifeq ($(ARCH), x86_64)
-ifneq ($(C_COMPILER), PGI)
+ifeq ($(findstring pgcc,$(HOSTCC)),)
 GETARCH_FLAGS += -march=native
 endif
 endif
@@ -267,9 +267,10 @@ OBJCOPY = $(CROSS_SUFFIX)objcopy
 OBJCONV = $(CROSS_SUFFIX)objconv
 
 
-# For detect fortran failed, only build BLAS.
+# When fortran support was either not detected or actively deselected, only build BLAS.
 ifeq ($(NOFORTRAN), 1)
 NO_LAPACK = 1
+override FEXTRALIB = 
 endif
 
 #
@@ -1124,8 +1125,12 @@ endif
 endif
 
 ifdef NO_AFFINITY
+ifeq ($(NO_AFFINITY), 0)
+override undefine NO_AFFINITY
+else
 CCOMMON_OPT	+= -DNO_AFFINITY
 endif
+endif
 
 ifdef FUNCTION_PROFILE
 CCOMMON_OPT	+= -DFUNCTION_PROFILE
diff --git a/appveyor.yml b/appveyor.yml
index 44a616aaa..2f9cc7b0b 100644
--- a/appveyor.yml
+++ b/appveyor.yml
@@ -35,7 +35,14 @@ environment:
       DYNAMIC_ARCH: ON
       WITH_FORTRAN: no
     - COMPILER: cl
-
+    - COMPILER: MinGW64-gcc-7.2.0-mingw
+      DYNAMIC_ARCH: OFF
+      WITH_FORTRAN: ignore
+    - COMPILER: MinGW64-gcc-7.2.0
+    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
+      COMPILER: MinGW-gcc-5.3.0
+      WITH_FORTRAN: ignore
+    
 install:
   - if [%COMPILER%]==[clang-cl] call %CONDA_INSTALL_LOCN%\Scripts\activate.bat
   - if [%COMPILER%]==[clang-cl] conda config --add channels conda-forge --force
@@ -52,7 +59,14 @@ install:
 before_build:
   - ps: if (-Not (Test-Path .\build)) { mkdir build }
   - cd build
+  - set PATH=%PATH:C:\Program Files\Git\usr\bin;=%
+  - if [%COMPILER%]==[MinGW-gcc-5.3.0] set PATH=C:\MinGW\bin;C:\msys64\usr\bin;C:\mingw-w64\x86_64-7.2.0-posix-seh-rt_v5-rev1\mingw64\bin;%PATH%
+  - if [%COMPILER%]==[MinGW64-gcc-7.2.0-mingw] set PATH=C:\MinGW\bin;C:\mingw-w64\x86_64-7.2.0-posix-seh-rt_v5-rev1\mingw64\bin;%PATH%
+  - if [%COMPILER%]==[MinGW64-gcc-7.2.0] set PATH=C:\msys64\usr\bin;C:\mingw-w64\x86_64-7.2.0-posix-seh-rt_v5-rev1\mingw64\bin;%PATH%
   - if [%COMPILER%]==[cl] cmake -G "Visual Studio 15 2017 Win64" ..
+  - if [%COMPILER%]==[MinGW64-gcc-7.2.0-mingw] cmake -G "MinGW Makefiles" -DNOFORTRAN=1 ..
+  - if [%COMPILER%]==[MinGW64-gcc-7.2.0] cmake -G "MSYS Makefiles"  -DBINARY=32 -DNOFORTRAN=1 ..
+  - if [%COMPILER%]==[MinGW-gcc-5.3.0] cmake -G "MSYS Makefiles" -DNOFORTRAN=1 ..
   - if [%WITH_FORTRAN%]==[no] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DMSVC_STATIC_CRT=ON ..
   - if [%WITH_FORTRAN%]==[yes] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DBUILD_WITHOUT_LAPACK=no -DNOFORTRAN=0 ..
   - if [%DYNAMIC_ARCH%]==[ON] cmake -DDYNAMIC_ARCH=ON -DDYNAMIC_LIST='CORE2;NEHALEM;SANDYBRIDGE;BULLDOZER;HASWELL' ..
@@ -64,3 +78,4 @@ test_script:
   - echo Running Test
   - cd utest
   - openblas_utest
+  
diff --git a/cmake/arch.cmake b/cmake/arch.cmake
index b4547b7c9..5a7434551 100644
--- a/cmake/arch.cmake
+++ b/cmake/arch.cmake
@@ -81,7 +81,8 @@ if (DYNAMIC_ARCH)
   endif ()
 
   if (NOT DYNAMIC_CORE)
-    unset(DYNAMIC_ARCH)
+    message (STATUS "DYNAMIC_ARCH is not supported on this architecture, removing from options")
+    unset(DYNAMIC_ARCH CACHE)
   endif ()
 endif ()
 
diff --git a/cmake/prebuild.cmake b/cmake/prebuild.cmake
index a67c44bf5..e508a46c2 100644
--- a/cmake/prebuild.cmake
+++ b/cmake/prebuild.cmake
@@ -59,6 +59,9 @@ set(FU "")
 if (APPLE OR (MSVC AND NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang"))
   set(FU "_")
 endif()
+if(MINGW AND NOT MINGW64)
+  set(FU "_")
+endif()
 
 set(COMPILER_ID ${CMAKE_C_COMPILER_ID})
 if (${COMPILER_ID} STREQUAL "GNU")
@@ -82,6 +85,11 @@ endif ()
 # f_check
 if (NOT NOFORTRAN)
   include("${PROJECT_SOURCE_DIR}/cmake/f_check.cmake")
+else ()
+ file(APPEND ${TARGET_CONF_TEMP}
+   "#define BUNDERSCORE _\n"
+   "#define NEEDBUNDERSCORE 1\n")
+ set(BU "_")
 endif ()
 
 # Cannot run getarch on target if we are cross-compiling
diff --git a/cmake/system.cmake b/cmake/system.cmake
index 7f3696286..4f8011603 100644
--- a/cmake/system.cmake
+++ b/cmake/system.cmake
@@ -66,10 +66,17 @@ if (DEFINED TARGET)
 endif ()
 
 # On x86_64 build getarch with march=native. This is required to detect AVX512 support in getarch.
-if (X86_64)
+if (X86_64 AND NOT ${CMAKE_C_COMPILER_ID} STREQUAL "PGI")
   set(GETARCH_FLAGS "${GETARCH_FLAGS} -march=native")
 endif ()
 
+# On x86 no AVX support is available
+if (X86 OR X86_64)
+if ((DEFINED BINARY AND BINARY EQUAL 32) OR ("$CMAKE_SIZEOF_VOID_P}" EQUAL "4"))
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX -DNO_AVX2 -DNO_AVX512")
+endif ()
+endif ()
+
 if (INTERFACE64)
   message(STATUS "Using 64-bit integers.")
   set(GETARCH_FLAGS	"${GETARCH_FLAGS} -DUSE64BITINT")
@@ -148,7 +155,9 @@ else()
 endif ()
 
 include("${PROJECT_SOURCE_DIR}/cmake/prebuild.cmake")
-
+if (DEFINED BINARY)
+  message(STATUS "Compiling a ${BINARY}-bit binary.")
+endif ()
 if (NOT DEFINED NEED_PIC)
   set(NEED_PIC 1)
 endif ()
@@ -165,6 +174,9 @@ include("${PROJECT_SOURCE_DIR}/cmake/cc.cmake")
 if (NOT NOFORTRAN)
   # Fortran Compiler dependent settings
   include("${PROJECT_SOURCE_DIR}/cmake/fc.cmake")
+else ()
+set(NO_LAPACK 1)
+set(NO_LAPACKE 1)
 endif ()
 
 if (BINARY64)
@@ -190,9 +202,14 @@ if (NEED_PIC)
 endif ()
 
 if (DYNAMIC_ARCH)
-  set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_ARCH")
-  if (DYNAMIC_OLDER)
-    set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_OLDER")
+  if (X86 OR X86_64 OR ARM64 OR PPC)
+    set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_ARCH")
+    if (DYNAMIC_OLDER)
+      set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_OLDER")
+    endif ()
+  else ()
+    unset (DYNAMIC_ARCH)
+    message (STATUS "DYNAMIC_ARCH is not supported on the target architecture, removing")
   endif ()
 endif ()
 
diff --git a/cmake/system_check.cmake b/cmake/system_check.cmake
index 94d3ba643..610f689e0 100644
--- a/cmake/system_check.cmake
+++ b/cmake/system_check.cmake
@@ -15,7 +15,7 @@ if (${HOST_OS} STREQUAL "LINUX")
     EXECUTE_PROCESS( COMMAND uname -o COMMAND tr -d '\n' OUTPUT_VARIABLE OPERATING_SYSTEM)
       if(${OPERATING_SYSTEM} MATCHES "Android")
         set(HOST_OS ANDROID)
-      endif(${OPERATING_SYSTEM} MATCHES "Android")
+      endif()
 endif()
 
 
diff --git a/cpuid_arm64.c b/cpuid_arm64.c
index a5e731d74..e8aa29813 100644
--- a/cpuid_arm64.c
+++ b/cpuid_arm64.c
@@ -94,7 +94,7 @@ int get_feature(char *search)
 	if( p == NULL ) return 0;
 
 	t = strtok(p," ");
-	while( t = strtok(NULL," "))
+	while( (t = strtok(NULL," ")))
 	{
 		if (!strcmp(t, search))   { return(1); }
 	}
@@ -344,7 +344,7 @@ void get_features(void)
 	if( p == NULL ) return;
 
 	t = strtok(p," ");
-	while( t = strtok(NULL," "))
+	while( (t = strtok(NULL," ")))
 	{
 	}
 
diff --git a/ctest/Makefile b/ctest/Makefile
index 569a5dda3..f562c9bb3 100644
--- a/ctest/Makefile
+++ b/ctest/Makefile
@@ -6,6 +6,8 @@ TOPDIR = ..
 include $(TOPDIR)/Makefile.system
 
 override CFLAGS += -DADD$(BU) -DCBLAS
+override TARGET_ARCH=
+override TARGET_MACH=
 
 LIB = $(TOPDIR)/$(LIBNAME)
 
diff --git a/driver/others/dynamic.c b/driver/others/dynamic.c
index 045fc65b8..f1cd3c6e6 100644
--- a/driver/others/dynamic.c
+++ b/driver/others/dynamic.c
@@ -585,9 +585,27 @@ static gotoblas_t *get_coretype(void){
 	  }
         }
         return NULL;  
+      case 7:
+        if (model == 14) {
+	// Ice Lake
+          if (support_avx512()) 
+	    return &gotoblas_SKYLAKEX;
+	  if(support_avx2()){
+	    openblas_warning(FALLBACK_VERBOSE, HASWELL_FALLBACK);
+	    return &gotoblas_HASWELL;
+          }
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+          openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+          return &gotoblas_NEHALEM;
+          }
+        }
+        return NULL;  
       case 9:
       case 8:
-	if (model == 14 ) { // Kaby Lake
+	if (model == 14 ) { // Kaby Lake, Coffee Lake
 	  if(support_avx2())
 	    return &gotoblas_HASWELL;
 	  if(support_avx()) {
diff --git a/driver/others/memory.c b/driver/others/memory.c
index f67cb01f4..534d6d9fc 100644
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@@ -2041,8 +2041,12 @@ static BLASULONG  alloc_lock = 0UL;
 
 static void alloc_mmap_free(struct release_t *release){
 
+if (!release->address) return;
+
   if (munmap(release -> address, BUFFER_SIZE)) {
-    printf("OpenBLAS : munmap failed\n");
+      int errsv=errno;
+       perror("OpenBLAS : munmap failed:");
+       printf("error code=%d,\trelease->address=%lx\n",errsv,release->address);
   }
 }
 
@@ -2073,6 +2077,12 @@ static void *alloc_mmap(void *address){
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
     UNLOCK_COMMAND(&alloc_lock);
 #endif    
+  } else {
+#ifdef DEBUG  
+        int errsv=errno;
+       perror("OpenBLAS : mmap failed:");
+       printf("error code=%d,\tmap_address=%lx\n",errsv,map_address);
+#endif
   }
 
 #ifdef OS_LINUX
diff --git a/exports/Makefile b/exports/Makefile
index b1348bd4a..d32e449df 100644
--- a/exports/Makefile
+++ b/exports/Makefile
@@ -105,6 +105,10 @@ $(LIBPREFIX).def : gensymbol
 libgoto_hpl.def : gensymbol
 	perl ./gensymbol win2khpl $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
 
+ifeq ($(OSNAME), Darwin)
+INTERNALNAME = $(LIBPREFIX).$(MAJOR_VERSION).dylib
+endif
+
 ifeq (, $(SYMBOLPREFIX)$(SYMBOLSUFFIX))
 $(LIBDYNNAME) : ../$(LIBNAME) osx.def
 else
@@ -114,9 +118,9 @@ $(LIBDYNNAME) : ../$(LIBNAME).osx.renamed osx.def
 endif
 ifneq (,$(filter 1 2,$(NOFORTRAN)))
 #only build without Fortran
-	$(CC) $(CFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(LIBDYNNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+	$(CC) $(CFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(INTERNALNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 else
-	$(FC) $(FFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(LIBDYNNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+	$(FC) $(FFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(INTERNALNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 endif
 
 dllinit.$(SUFFIX) : dllinit.c
diff --git a/kernel/x86/lsame.S b/kernel/x86/lsame.S
index 3ac7a7314..2a2ab2bb5 100644
--- a/kernel/x86/lsame.S
+++ b/kernel/x86/lsame.S
@@ -56,13 +56,13 @@
 #ifndef HAVE_CMOV
 	movl	%eax, %ecx
 	subl	$32,  %ecx
-	jle	.L1
+	jl	.L1
 	movl	%ecx, %eax
 .L1:
 
 	movl	%edx, %ecx
 	subl	$32,  %ecx
-	jle	.L2
+	jl	.L2
 	movl	%ecx, %edx
 .L2:
 	subl	%eax, %edx
diff --git a/kernel/x86_64/dgemm_kernel_4x8_haswell.S b/kernel/x86_64/dgemm_kernel_4x8_haswell.S
index c84b599ce..19e32ef2c 100644
--- a/kernel/x86_64/dgemm_kernel_4x8_haswell.S
+++ b/kernel/x86_64/dgemm_kernel_4x8_haswell.S
@@ -106,7 +106,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 #define	A_PR1	512
-#define	B_PR1	512
+#define	B_PR1	160
+#define BROADCASTKERNEL
 
 /*******************************************************************************************
 * Macro definitions
@@ -133,7 +134,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	prefetcht0	A_PR1(AO)
 	vmovups		-12 * SIZE(BO), %ymm1
 	prefetcht0	B_PR1(BO)
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+# else
 	vmovups 	-16 * SIZE(AO), %ymm0
+# endif
 	prefetcht0	B_PR1+64(BO)
 	vmovups		 -8 * SIZE(BO), %ymm2
 	prefetcht0	B_PR1+128(BO)
@@ -143,17 +148,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmulpd  	%ymm0 ,%ymm2  , %ymm8
 	vmulpd  	%ymm0 ,%ymm3  , %ymm12
 	prefetcht0	B_PR1+256(BO)
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm5
 	vmulpd  	%ymm0 ,%ymm2  , %ymm9
 	vmulpd  	%ymm0 ,%ymm3  , %ymm13
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+# else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+# endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm6
 	vmulpd  	%ymm0 ,%ymm2  , %ymm10
 
 	addq		$ 12*SIZE, BO
 	vmulpd  	%ymm0 ,%ymm3  , %ymm14
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -13 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		-12 * SIZE(BO), %ymm1
 	vmulpd  	%ymm0 ,%ymm2  , %ymm11
@@ -165,23 +182,38 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL4x12_M1
 	prefetcht0	A_PR1(AO)
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+# else
 	vmovups 	-16 * SIZE(AO), %ymm0
+# endif
 	prefetcht0	B_PR1(BO)
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
 	prefetcht0	B_PR1+64(BO)
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm8
 	prefetcht0	B_PR1+128(BO)
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm12
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm9
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm13
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+# else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm10
-
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm14
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -13 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		-12 * SIZE(BO), %ymm1
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm11
@@ -192,21 +224,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .endm
 
 .macro KERNEL4x12_M2
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -12 * SIZE(AO), %ymm0
+# else
 	vmovups 	-12 * SIZE(AO), %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm8
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm12
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -11 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm9
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm13
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -10 * SIZE(AO), %ymm0
+# else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm10
 
 	addq		$ 8*SIZE, AO
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm14
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		  0 * SIZE(BO), %ymm1
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm11
@@ -218,21 +266,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .macro KERNEL4x12_E
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -12 * SIZE(AO), %ymm0
+# else
 	vmovups 	-12 * SIZE(AO), %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm8
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm12
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -11 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm9
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm13
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -10 * SIZE(AO), %ymm0
+# else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm10
 
 	addq		$ 8*SIZE, AO
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm14
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm11
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm15
@@ -241,23 +305,39 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL4x12_SUB
 	vmovups		-12 * SIZE(BO), %ymm1
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+# else
 	vmovups 	-16 * SIZE(AO), %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
 	vmovups		 -8 * SIZE(BO), %ymm2
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm8
 	vmovups		 -4 * SIZE(BO), %ymm3
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm12
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm9
 	addq		$ 12*SIZE, BO
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm13
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+# else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm10
 	addq		$ 4*SIZE, AO
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm14
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+# if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+# else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+# endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm11
 	vfmadd231pd  	%ymm0 ,%ymm3  , %ymm15
@@ -267,43 +347,83 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro SAVE4x12
 
+        prefetcht0      BUFFER1
 	vbroadcastsd	ALPHA, %ymm0
 
 	vmulpd	%ymm0 , %ymm4 , %ymm4
 	vmulpd	%ymm0 , %ymm5 , %ymm5
 	vmulpd	%ymm0 , %ymm6 , %ymm6
 	vmulpd	%ymm0 , %ymm7 , %ymm7
-
+        prefetcht0      64 + BUFFER1
 	vmulpd	%ymm0 , %ymm8 , %ymm8
 	vmulpd	%ymm0 , %ymm9 , %ymm9
 	vmulpd	%ymm0 , %ymm10, %ymm10
 	vmulpd	%ymm0 , %ymm11, %ymm11
-
+#if B_PR1 > 32
+        prefetcht0      128 + BUFFER1
+#endif
 	vmulpd	%ymm0 , %ymm12, %ymm12
 	vmulpd	%ymm0 , %ymm13, %ymm13
 	vmulpd	%ymm0 , %ymm14, %ymm14
 	vmulpd	%ymm0 , %ymm15, %ymm15
+#if B_PR1 > 96
+        prefetcht0      192 + BUFFER1
+#endif
 
-	vpermpd $ 0xb1 , %ymm5, %ymm5
-	vpermpd $ 0xb1 , %ymm7, %ymm7
+#if defined BROADCASTKERNEL
+        vperm2f128 $ 0x20 , %ymm6, %ymm4 , %ymm0
+        vperm2f128 $ 0x20 , %ymm7, %ymm5 , %ymm1
+        vperm2f128 $ 0x31 , %ymm6, %ymm4 , %ymm2
+        vperm2f128 $ 0x31 , %ymm7, %ymm5 , %ymm3
+#else
+	vpermilpd $ 0x05 , %ymm5, %ymm5
+	vpermilpd $ 0x05 , %ymm7, %ymm7
+#endif
 
+#if B_PR1 > 160
+        prefetcht0      256 + BUFFER1
+#endif
+
+#if defined BROADCASTKERNEL
+        vunpcklpd %ymm1, %ymm0, %ymm4
+        vunpckhpd %ymm1, %ymm0, %ymm5
+        vunpcklpd %ymm3, %ymm2, %ymm6
+        vunpckhpd %ymm3, %ymm2, %ymm7
+#else
 	vblendpd $ 0x0a, %ymm5, %ymm4, %ymm0
 	vblendpd $ 0x05, %ymm5, %ymm4, %ymm1
 	vblendpd $ 0x0a, %ymm7, %ymm6, %ymm2
 	vblendpd $ 0x05, %ymm7, %ymm6, %ymm3
+#endif
 
-	vpermpd $ 0x1b , %ymm2, %ymm2
-	vpermpd $ 0x1b , %ymm3, %ymm3
-	vpermpd $ 0xb1 , %ymm2, %ymm2
-	vpermpd $ 0xb1 , %ymm3, %ymm3
+#if B_PR1 > 224
+        prefetcht0      320 + BUFFER1
+#endif
 
+#ifndef BROADCASTKERNEL
+	vperm2f128 $ 0x01 , %ymm2, %ymm2 , %ymm2
+	vperm2f128 $ 0x01 , %ymm3, %ymm3 , %ymm3
+#endif
+
+#if B_PR1 > 288
+        prefetcht0      384 + BUFFER1
+#endif
+
+#ifndef BROADCASTKERNEL
 	vblendpd $ 0x03, %ymm0, %ymm2 , %ymm4
 	vblendpd $ 0x03, %ymm1, %ymm3 , %ymm5
 	vblendpd $ 0x03, %ymm2, %ymm0 , %ymm6
 	vblendpd $ 0x03, %ymm3, %ymm1 , %ymm7
+#endif
 
+#if B_PR1 > 352
+        prefetcht0      448 + BUFFER1
+#endif
         leaq    (CO1, LDC, 2), %rax     
 	
+#if B_PR1 > 416
+        prefetcht0      512 + BUFFER1
+#endif
 
 #if !defined(TRMMKERNEL)
 
@@ -319,29 +439,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmovups	%ymm6 ,  	(%rax)
 	vmovups	%ymm7 ,  	(%rax, LDC)
 
-	prefetcht0	32(CO1)
-	prefetcht0	32(CO1,LDC)
-	prefetcht0	32(%rax)
-	prefetcht0	32(%rax,LDC)
+	prefetcht1	56(CO1)
+	prefetcht1	56(CO1,LDC)
+	prefetcht1	56(%rax)
+	prefetcht1	56(%rax,LDC)
 
-	vpermpd $ 0xb1 , %ymm9 , %ymm9
-	vpermpd $ 0xb1 , %ymm11, %ymm11
+#if defined BROADCASTKERNEL
+        vperm2f128 $ 0x20 , %ymm10, %ymm8 , %ymm0
+        vperm2f128 $ 0x20 , %ymm11, %ymm9 , %ymm1
+        vperm2f128 $ 0x31 , %ymm10, %ymm8 , %ymm2
+        vperm2f128 $ 0x31 , %ymm11, %ymm9 , %ymm3
+        vunpcklpd %ymm1, %ymm0, %ymm4
+        vunpckhpd %ymm1, %ymm0, %ymm5
+        vunpcklpd %ymm3, %ymm2, %ymm6
+        vunpckhpd %ymm3, %ymm2, %ymm7
+#else
+	vpermilpd $ 0x05 , %ymm9, %ymm9
+	vpermilpd $ 0x05 , %ymm11, %ymm11
 
-	vblendpd $ 0x0a, %ymm9 , %ymm8 , %ymm0
-	vblendpd $ 0x05, %ymm9 , %ymm8 , %ymm1
+	vblendpd $ 0x0a, %ymm9, %ymm8, %ymm0
+	vblendpd $ 0x05, %ymm9, %ymm8, %ymm1
 	vblendpd $ 0x0a, %ymm11, %ymm10, %ymm2
 	vblendpd $ 0x05, %ymm11, %ymm10, %ymm3
 
-	vpermpd $ 0x1b , %ymm2, %ymm2
-	vpermpd $ 0x1b , %ymm3, %ymm3
-	vpermpd $ 0xb1 , %ymm2, %ymm2
-	vpermpd $ 0xb1 , %ymm3, %ymm3
+	vperm2f128 $ 0x01 , %ymm2, %ymm2 , %ymm2
+	vperm2f128 $ 0x01 , %ymm3, %ymm3 , %ymm3
 
 	vblendpd $ 0x03, %ymm0, %ymm2 , %ymm4
 	vblendpd $ 0x03, %ymm1, %ymm3 , %ymm5
 	vblendpd $ 0x03, %ymm2, %ymm0 , %ymm6
 	vblendpd $ 0x03, %ymm3, %ymm1 , %ymm7
-
+#endif
 
 	leaq	(%rax, LDC, 2), %rax
 	leaq	(%rax, LDC, 2), %rbp
@@ -360,29 +488,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmovups	%ymm6 ,  	(%rbp)
 	vmovups	%ymm7 ,  	(%rbp, LDC)
 
-	prefetcht0	32(%rax)
-	prefetcht0	32(%rax,LDC)
-	prefetcht0	32(%rbp)
-	prefetcht0	32(%rbp,LDC)
+	prefetcht1	56(%rax)
+	prefetcht1	56(%rax,LDC)
+	prefetcht1	56(%rbp)
+	prefetcht1	56(%rbp,LDC)
 
-	vpermpd $ 0xb1 , %ymm13, %ymm13
-	vpermpd $ 0xb1 , %ymm15, %ymm15
+#if defined BROADCASTKERNEL
+        vperm2f128 $ 0x20 , %ymm14, %ymm12 , %ymm0
+        vperm2f128 $ 0x20 , %ymm15, %ymm13 , %ymm1
+        vperm2f128 $ 0x31 , %ymm14, %ymm12 , %ymm2
+        vperm2f128 $ 0x31 , %ymm15, %ymm13 , %ymm3
+        vunpcklpd %ymm1, %ymm0, %ymm4
+        vunpckhpd %ymm1, %ymm0, %ymm5
+        vunpcklpd %ymm3, %ymm2, %ymm6
+        vunpckhpd %ymm3, %ymm2, %ymm7
+#else
+	vpermilpd $ 0x05 , %ymm13, %ymm13
+	vpermilpd $ 0x05 , %ymm15, %ymm15
 
 	vblendpd $ 0x0a, %ymm13, %ymm12, %ymm0
 	vblendpd $ 0x05, %ymm13, %ymm12, %ymm1
 	vblendpd $ 0x0a, %ymm15, %ymm14, %ymm2
 	vblendpd $ 0x05, %ymm15, %ymm14, %ymm3
 
-	vpermpd $ 0x1b , %ymm2, %ymm2
-	vpermpd $ 0x1b , %ymm3, %ymm3
-	vpermpd $ 0xb1 , %ymm2, %ymm2
-	vpermpd $ 0xb1 , %ymm3, %ymm3
+	vperm2f128 $ 0x01 , %ymm2, %ymm2 , %ymm2
+	vperm2f128 $ 0x01 , %ymm3, %ymm3 , %ymm3
 
 	vblendpd $ 0x03, %ymm0, %ymm2 , %ymm4
 	vblendpd $ 0x03, %ymm1, %ymm3 , %ymm5
 	vblendpd $ 0x03, %ymm2, %ymm0 , %ymm6
 	vblendpd $ 0x03, %ymm3, %ymm1 , %ymm7
-
+#endif
 
 	leaq	(%rax, LDC, 4), %rax
 	leaq	(%rbp, LDC, 4), %rbp
@@ -401,10 +537,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmovups	%ymm6 ,  	(%rbp)
 	vmovups	%ymm7 ,  	(%rbp, LDC)
 
-	prefetcht0	32(%rax)
-	prefetcht0	32(%rax,LDC)
-	prefetcht0	32(%rbp)
-	prefetcht0	32(%rbp,LDC)
+	prefetcht1	56(%rax)
+	prefetcht1	56(%rax,LDC)
+	prefetcht1	56(%rbp)
+	prefetcht1	56(%rbp,LDC)
 
 	addq	$ 4*SIZE, CO1
 .endm
@@ -683,19 +819,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL4x8_I
 	vmovups		-12 * SIZE(BO), %ymm1
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+#else
 	vmovups 	-16 * SIZE(AO), %ymm0
+#endif
 	vmovups		 -8 * SIZE(BO), %ymm2
 	vmulpd  	%ymm0 ,%ymm1  , %ymm4
 	vmulpd  	%ymm0 ,%ymm2  , %ymm8
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm5
 	vmulpd  	%ymm0 ,%ymm2  , %ymm9
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm6
 	vmulpd  	%ymm0 ,%ymm2  , %ymm10
 
 	addq		$  8*SIZE, BO
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -13 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		-12 * SIZE(BO), %ymm1
 	vmulpd  	%ymm0 ,%ymm2  , %ymm11
@@ -705,19 +857,34 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL4x8_M1
 	prefetcht0	A_PR1(AO)
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+#else
 	vmovups 	-16 * SIZE(AO), %ymm0
+#endif
 	prefetcht0	B_PR1(BO)
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
 	prefetcht0	B_PR1+64(BO)
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm8
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm9
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm10
-
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -13 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		-12 * SIZE(BO), %ymm1
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm11
@@ -726,18 +893,34 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .endm
 
 .macro KERNEL4x8_M2
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -12 * SIZE(AO), %ymm0
+#else
 	vmovups 	-12 * SIZE(AO), %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm8
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -11 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm9
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -10 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm10
 
 	addq		$ 8*SIZE, AO
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		 -4 * SIZE(BO), %ymm1
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm11
@@ -747,18 +930,34 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .macro KERNEL4x8_E
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -12 * SIZE(AO), %ymm0
+#else
 	vmovups 	-12 * SIZE(AO), %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm8
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -11 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm9
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -10 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm10
 
 	addq		$ 8*SIZE, AO
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm11
 	addq		$  8*SIZE, BO
@@ -766,19 +965,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL4x8_SUB
 	vmovups		-12 * SIZE(BO), %ymm1
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+#else
 	vmovups 	-16 * SIZE(AO), %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
 	vmovups		 -8 * SIZE(BO), %ymm2
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm8
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm9
 	addq		$  8*SIZE, BO
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm10
 	addq		$ 4*SIZE, AO
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm11
 
@@ -799,23 +1014,32 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmulpd	%ymm0 , %ymm10, %ymm10
 	vmulpd	%ymm0 , %ymm11, %ymm11
 
-	vpermpd $ 0xb1 , %ymm5, %ymm5
-	vpermpd $ 0xb1 , %ymm7, %ymm7
+#if defined BROADCASTKERNEL
+        vperm2f128 $ 0x20 , %ymm6, %ymm4 , %ymm0
+        vperm2f128 $ 0x20 , %ymm7, %ymm5 , %ymm1
+        vperm2f128 $ 0x31 , %ymm6, %ymm4 , %ymm2
+        vperm2f128 $ 0x31 , %ymm7, %ymm5 , %ymm3
+        vunpcklpd %ymm1, %ymm0, %ymm4
+        vunpckhpd %ymm1, %ymm0, %ymm5
+        vunpcklpd %ymm3, %ymm2, %ymm6
+        vunpckhpd %ymm3, %ymm2, %ymm7
+#else
+	vpermilpd $ 0x05 , %ymm5, %ymm5
+	vpermilpd $ 0x05 , %ymm7, %ymm7
 
 	vblendpd $ 0x0a, %ymm5, %ymm4, %ymm0
 	vblendpd $ 0x05, %ymm5, %ymm4, %ymm1
 	vblendpd $ 0x0a, %ymm7, %ymm6, %ymm2
 	vblendpd $ 0x05, %ymm7, %ymm6, %ymm3
 
-	vpermpd $ 0x1b , %ymm2, %ymm2
-	vpermpd $ 0x1b , %ymm3, %ymm3
-	vpermpd $ 0xb1 , %ymm2, %ymm2
-	vpermpd $ 0xb1 , %ymm3, %ymm3
+	vperm2f128 $ 0x01 , %ymm2, %ymm2 , %ymm2
+	vperm2f128 $ 0x01 , %ymm3, %ymm3 , %ymm3
 
 	vblendpd $ 0x03, %ymm0, %ymm2 , %ymm4
 	vblendpd $ 0x03, %ymm1, %ymm3 , %ymm5
 	vblendpd $ 0x03, %ymm2, %ymm0 , %ymm6
 	vblendpd $ 0x03, %ymm3, %ymm1 , %ymm7
+#endif
 
         leaq    (CO1, LDC, 2), %rax     
 	
@@ -834,29 +1058,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmovups	%ymm6 ,  	(%rax)
 	vmovups	%ymm7 ,  	(%rax, LDC)
 
-	prefetcht0	32(CO1)
-	prefetcht0	32(CO1,LDC)
-	prefetcht0	32(%rax)
-	prefetcht0	32(%rax,LDC)
+	prefetcht0	56(CO1)
+	prefetcht0	56(CO1,LDC)
+	prefetcht0	56(%rax)
+	prefetcht0	56(%rax,LDC)
 
-	vpermpd $ 0xb1 , %ymm9 , %ymm9
-	vpermpd $ 0xb1 , %ymm11, %ymm11
+#if defined BROADCASTKERNEL
+        vperm2f128 $ 0x20 , %ymm10, %ymm8 , %ymm0
+        vperm2f128 $ 0x20 , %ymm11, %ymm9 , %ymm1
+        vperm2f128 $ 0x31 , %ymm10, %ymm8 , %ymm2
+        vperm2f128 $ 0x31 , %ymm11, %ymm9 , %ymm3
+        vunpcklpd %ymm1, %ymm0, %ymm4
+        vunpckhpd %ymm1, %ymm0, %ymm5
+        vunpcklpd %ymm3, %ymm2, %ymm6
+        vunpckhpd %ymm3, %ymm2, %ymm7
+#else
+	vpermilpd $ 0x05 , %ymm9 , %ymm9
+	vpermilpd $ 0x05 , %ymm11, %ymm11
 
 	vblendpd $ 0x0a, %ymm9 , %ymm8 , %ymm0
 	vblendpd $ 0x05, %ymm9 , %ymm8 , %ymm1
 	vblendpd $ 0x0a, %ymm11, %ymm10, %ymm2
 	vblendpd $ 0x05, %ymm11, %ymm10, %ymm3
 
-	vpermpd $ 0x1b , %ymm2, %ymm2
-	vpermpd $ 0x1b , %ymm3, %ymm3
-	vpermpd $ 0xb1 , %ymm2, %ymm2
-	vpermpd $ 0xb1 , %ymm3, %ymm3
+	vperm2f128 $ 0x01 , %ymm2, %ymm2 , %ymm2
+	vperm2f128 $ 0x01 , %ymm3, %ymm3 , %ymm3
 
 	vblendpd $ 0x03, %ymm0, %ymm2 , %ymm4
 	vblendpd $ 0x03, %ymm1, %ymm3 , %ymm5
 	vblendpd $ 0x03, %ymm2, %ymm0 , %ymm6
 	vblendpd $ 0x03, %ymm3, %ymm1 , %ymm7
-
+#endif
 
 	leaq	(%rax, LDC, 2), %rax
 	leaq	(%rax, LDC, 2), %rbp
@@ -875,10 +1107,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmovups	%ymm6 ,  	(%rbp)
 	vmovups	%ymm7 ,  	(%rbp, LDC)
 
-	prefetcht0	32(%rax)
-	prefetcht0	32(%rax,LDC)
-	prefetcht0	32(%rbp)
-	prefetcht0	32(%rbp,LDC)
+	prefetcht0	56(%rax)
+	prefetcht0	56(%rax,LDC)
+	prefetcht0	56(%rbp)
+	prefetcht0	56(%rbp,LDC)
 
 	addq	$ 4*SIZE, CO1
 .endm
@@ -1082,15 +1314,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL4x4_I
 	prefetcht0	A_PR1(AO)
 	vmovups		-12 * SIZE(BO), %ymm1
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+#else
 	vmovups 	-16 * SIZE(AO), %ymm0
+#endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm4
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm5
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm6
 
 	addq		$ 4*SIZE, BO
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -13 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vmulpd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		-12 * SIZE(BO), %ymm1
 
@@ -1098,29 +1346,60 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL4x4_M1
 	prefetcht0	A_PR1(AO)
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+#else
 	vmovups 	-16 * SIZE(AO), %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
-
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -13 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		-12 * SIZE(BO), %ymm1
 
 .endm
 
 .macro KERNEL4x4_M2
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -12 * SIZE(AO), %ymm0
+#else
 	vmovups 	-12 * SIZE(AO), %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -11 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -10 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 
 	addq		$ 8*SIZE, AO
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	vmovups		 -8 * SIZE(BO), %ymm1
 	addq		$ 8*SIZE, BO
@@ -1128,30 +1407,62 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .macro KERNEL4x4_E
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -12 * SIZE(AO), %ymm0
+#else
 	vmovups 	-12 * SIZE(AO), %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -11 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -10 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 
 	addq		$ 8*SIZE, AO
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 	addq		$ 4*SIZE, BO
 .endm
 
 .macro KERNEL4x4_SUB
 	vmovups		-12 * SIZE(BO), %ymm1
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -16 * SIZE(AO), %ymm0
+#else
 	vmovups 	-16 * SIZE(AO), %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm4
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -15 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm5
 	addq		$ 4*SIZE, BO
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -14 * SIZE(AO), %ymm0
+#else
 	vpermpd		$ 0x1b, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm6
 	addq		$ 4*SIZE, AO
-	vpermpd		$ 0xb1, %ymm0  , %ymm0
+#if defined BROADCASTKERNEL
+        vbroadcastsd    -17 * SIZE(AO), %ymm0
+#else
+	vpermilpd	$ 0x05, %ymm0  , %ymm0
+#endif
 	vfmadd231pd  	%ymm0 ,%ymm1  , %ymm7
 
 .endm
@@ -1165,23 +1476,32 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vmulpd	%ymm0 , %ymm5 , %ymm5
 	vmulpd	%ymm0 , %ymm6 , %ymm6
 
-	vpermpd $ 0xb1 , %ymm5, %ymm5
-	vpermpd $ 0xb1 , %ymm7, %ymm7
+#if defined BROADCASTKERNEL
+        vperm2f128 $ 0x20 , %ymm6, %ymm4 , %ymm0
+        vperm2f128 $ 0x20 , %ymm7, %ymm5 , %ymm1
+        vperm2f128 $ 0x31 , %ymm6, %ymm4 , %ymm2
+        vperm2f128 $ 0x31 , %ymm7, %ymm5 , %ymm3
+        vunpcklpd %ymm1, %ymm0, %ymm4
+        vunpckhpd %ymm1, %ymm0, %ymm5
+        vunpcklpd %ymm3, %ymm2, %ymm6
+        vunpckhpd %ymm3, %ymm2, %ymm7
+#else
+	vpermilpd $ 0x05 , %ymm5, %ymm5
+	vpermilpd $ 0x05 , %ymm7, %ymm7
 
 	vblendpd $ 0x0a, %ymm5, %ymm4, %ymm0
 	vblendpd $ 0x05, %ymm5, %ymm4, %ymm1
 	vblendpd $ 0x0a, %ymm7, %ymm6, %ymm2
 	vblendpd $ 0x05, %ymm7, %ymm6, %ymm3
 
-	vpermpd $ 0x1b , %ymm2, %ymm2
-	vpermpd $ 0x1b , %ymm3, %ymm3
-	vpermpd $ 0xb1 , %ymm2, %ymm2
-	vpermpd $ 0xb1 , %ymm3, %ymm3
+	vperm2f128 $ 0x01 , %ymm2, %ymm2 , %ymm2
+	vperm2f128 $ 0x01 , %ymm3, %ymm3 , %ymm3
 
 	vblendpd $ 0x03, %ymm0, %ymm2 , %ymm4
 	vblendpd $ 0x03, %ymm1, %ymm3 , %ymm5
 	vblendpd $ 0x03, %ymm2, %ymm0 , %ymm6
 	vblendpd $ 0x03, %ymm3, %ymm1 , %ymm7
+#endif
 
         leaq    (CO1, LDC, 2), %rax     
 	
@@ -1617,6 +1937,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .endm
 
 
+.macro PREFETCHT0_C
+        prefetcht0 (CO1)
+        prefetcht0 24(CO1)
+        prefetcht0 (CO1,LDC,4)
+        prefetcht0 24(CO1,LDC,4)
+        prefetcht0 (CO1,LDC,8)
+        prefetcht0 24(CO1,LDC,8)
+.endm
 /*******************************************************************************************/
 
 #if !defined(TRMMKERNEL)
@@ -1784,12 +2112,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	dec	%rax
 	jne	.L12_12
-
+	
 .L12_12a:
-
+        prefetcht0 ALPHA
+        PREFETCHT0_C
+        addq  LDC,CO1
 	KERNEL4x12_M1
+        PREFETCHT0_C
+        leaq  (CO1,LDC,2),CO1
 	KERNEL4x12_M2
+        PREFETCHT0_C
+        subq  LDC,CO1
 	KERNEL4x12_M1
+        PREFETCHT0_C
+        subq  LDC,CO1
+        subq  LDC,CO1
 	KERNEL4x12_M2
 
 	KERNEL4x12_M1
@@ -1844,6 +2181,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	SAVE4x12
 
+        /* here for the prefetch of next b source block */
+	/* the increment should be proportional to GEMM_Q/GEMM_P */
+
+        salq    $3, K
+#ifdef WINDOWS_ABI /* GEMM_P == GEMM_Q * 4 */
+        prefetcht2 32(B)
+        prefetcht2 32(B, K, 8)
+        addq    $64, B /* increment */
+#else /* GEMM_P == GEMM_Q * 2 under linux x86_64 */
+        prefetcht2 32(B)
+        prefetcht2 32(B, K, 8)
+        prefetcht2 96(B)
+        prefetcht2 96(B, K, 8)
+        addq    $128, B /* increment */
+#endif
+        sarq    $3, K
+
 	decq	I			# i --
 	jne	.L12_11
 	ALIGN_4	
@@ -1851,6 +2205,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /**************************************************************************
 * Rest of M 
 ***************************************************************************/
+
+        /* recover the original value of pointer B after prefetch */
+        movq    M, I
+        sarq    $2, I
+#ifdef WINDOWS_ABI /* GEMM_P == GEMM_Q * 4 */
+        salq    $6, I
+#else /* GEMM_P == GEMM_Q * 2 under linux x86_64 */
+        salq    $7, I
+#endif
+        subq    I, B
+
 .L12_20:
 	// Test rest of M
 
@@ -2068,10 +2433,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	jne	.L13_12
 
 .L13_12a:
-
+        prefetcht0 ALPHA
+        PREFETCHT0_C
+        addq  LDC,CO1
 	KERNEL4x12_M1
+        PREFETCHT0_C
+        leaq  (CO1,LDC,2),CO1
 	KERNEL4x12_M2
+        PREFETCHT0_C
+        subq  LDC,CO1
 	KERNEL4x12_M1
+        PREFETCHT0_C
+        subq  LDC,CO1
+        subq  LDC,CO1
 	KERNEL4x12_M2
 
 	KERNEL4x12_M1
@@ -2081,7 +2455,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	jmp .L13_16
 
-
 .L13_13:
 
 	test $1, %rax
@@ -2126,6 +2499,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	SAVE4x12
 
+        /* here for the prefetch of next b source block */
+	/* the increment should be proportional to GEMM_Q/GEMM_P */
+
+        salq    $3, K
+#ifdef WINDOWS_ABI /* GEMM_P == GEMM_Q * 4 */
+        prefetcht2 (B)
+        prefetcht2 (B, K, 8)
+        addq    $64, B /* increment */
+#else /* GEMM_P == GEMM_Q * 2 under linux x86_64 */
+        prefetcht2 (B)
+        prefetcht2 (B, K, 8)
+        prefetcht2 64(B)
+        prefetcht2 64(B, K, 8)
+        addq    $128, B /* increment */
+#endif
+        sarq    $3, K
+
 	decq	I			# i --
 	jne	.L13_11
 	ALIGN_4	
@@ -2133,6 +2523,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /**************************************************************************
 * Rest of M 
 ***************************************************************************/
+        /* recover the original value of pointer B */
+        movq    M, I
+        sarq    $2, I
+#ifdef WINDOWS_ABI /* GEMM_P == GEMM_Q * 4 */
+        salq    $6, I
+#else /* GEMM_P == GEMM_Q * 2 under linux x86_64 */
+        salq    $7, I
+#endif
+        subq    I, B
+
 .L13_20:
 	// Test rest of M
 
diff --git a/kernel/x86_64/dtrmm_kernel_4x8_haswell.c b/kernel/x86_64/dtrmm_kernel_4x8_haswell.c
index 651736b89..2acdc4615 100644
--- a/kernel/x86_64/dtrmm_kernel_4x8_haswell.c
+++ b/kernel/x86_64/dtrmm_kernel_4x8_haswell.c
@@ -33,7 +33,7 @@ static void dtrmm_kernel_4x8( BLASLONG n, FLOAT *alpha ,FLOAT *a, FLOAT *b, FLOA
 		"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm4		\n\t"
 		"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm8		\n\t"
 
-		"	vpermpd         $0xb1  , %%ymm0 , %%ymm0		\n\t"
+		"	vpermilpd         $0x05  , %%ymm0 , %%ymm0		\n\t"
 		"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm5		\n\t"
 		"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm9		\n\t"
 
@@ -41,7 +41,7 @@ static void dtrmm_kernel_4x8( BLASLONG n, FLOAT *alpha ,FLOAT *a, FLOAT *b, FLOA
 		"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm6		\n\t"
 		"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm10		\n\t"
 
-		"	vpermpd         $0xb1  , %%ymm0 , %%ymm0		\n\t"
+		"	vpermilpd         $0x05  , %%ymm0 , %%ymm0		\n\t"
 		"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm7		\n\t"
 		"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm11		\n\t"
 
@@ -62,18 +62,16 @@ static void dtrmm_kernel_4x8( BLASLONG n, FLOAT *alpha ,FLOAT *a, FLOAT *b, FLOA
 		"	vmulpd		%%ymm0 , %%ymm10, %%ymm10		\n\t"
 		"	vmulpd		%%ymm0 , %%ymm11, %%ymm11		\n\t"
 
-		"	vpermpd 	$0xb1  , %%ymm5 , %%ymm5		\n\t"
-		"	vpermpd 	$0xb1  , %%ymm7 , %%ymm7		\n\t"
+		"	vpermilpd 	$0x05  , %%ymm5 , %%ymm5		\n\t"
+		"	vpermilpd 	$0x05  , %%ymm7 , %%ymm7		\n\t"
 
 		"	vblendpd 	$0x0a  , %%ymm5 , %%ymm4 , %%ymm0	\n\t"
 		"	vblendpd 	$0x05  , %%ymm5 , %%ymm4 , %%ymm1	\n\t"
 		"	vblendpd 	$0x0a  , %%ymm7 , %%ymm6 , %%ymm2	\n\t"
 		"	vblendpd 	$0x05  , %%ymm7 , %%ymm6 , %%ymm3	\n\t"
 
-		"	vpermpd 	$0x1b  , %%ymm2 , %%ymm2		\n\t"
-		"	vpermpd 	$0x1b  , %%ymm3 , %%ymm3		\n\t"
-		"	vpermpd 	$0xb1  , %%ymm2 , %%ymm2		\n\t"
-		"	vpermpd 	$0xb1  , %%ymm3 , %%ymm3		\n\t"
+		"	vperm2f128 	$0x01  , %%ymm2 , %%ymm2 , %%ymm2	\n\t"
+		"	vperm2f128 	$0x01  , %%ymm3 , %%ymm3 , %%ymm3	\n\t"
 
 		"	vblendpd 	$0x03  , %%ymm0 , %%ymm2 , %%ymm4	\n\t"
 		"	vblendpd 	$0x03  , %%ymm1 , %%ymm3 , %%ymm5	\n\t"
@@ -85,18 +83,16 @@ static void dtrmm_kernel_4x8( BLASLONG n, FLOAT *alpha ,FLOAT *a, FLOAT *b, FLOA
 		"	vmovups		%%ymm6 , (%7)				\n\t"
 		"	vmovups		%%ymm7 , (%8)				\n\t"
 
-		"	vpermpd 	$0xb1  , %%ymm9 , %%ymm9		\n\t"
-		"	vpermpd 	$0xb1  , %%ymm11, %%ymm11		\n\t"
+		"	vpermilpd 	$0x05  , %%ymm9 , %%ymm9		\n\t"
+		"	vpermilpd 	$0x05  , %%ymm11, %%ymm11		\n\t"
 
 		"	vblendpd 	$0x0a  , %%ymm9 , %%ymm8 , %%ymm0	\n\t"
 		"	vblendpd 	$0x05  , %%ymm9 , %%ymm8 , %%ymm1	\n\t"
 		"	vblendpd 	$0x0a  , %%ymm11, %%ymm10, %%ymm2	\n\t"
 		"	vblendpd 	$0x05  , %%ymm11, %%ymm10, %%ymm3	\n\t"
 
-		"	vpermpd 	$0x1b  , %%ymm2 , %%ymm2		\n\t"
-		"	vpermpd 	$0x1b  , %%ymm3 , %%ymm3		\n\t"
-		"	vpermpd 	$0xb1  , %%ymm2 , %%ymm2		\n\t"
-		"	vpermpd 	$0xb1  , %%ymm3 , %%ymm3		\n\t"
+		"	vperm2f128 	$0x01  , %%ymm2 , %%ymm2 , %%ymm2	\n\t"
+		"	vperm2f128 	$0x01  , %%ymm3 , %%ymm3 , %%ymm3	\n\t"
 
 		"	vblendpd 	$0x03  , %%ymm0 , %%ymm2 , %%ymm4	\n\t"
 		"	vblendpd 	$0x03  , %%ymm1 , %%ymm3 , %%ymm5	\n\t"
diff --git a/kernel/x86_64/dtrsm_kernel_RN_haswell.c b/kernel/x86_64/dtrsm_kernel_RN_haswell.c
index 9ab78fc8e..cb939e762 100644
--- a/kernel/x86_64/dtrsm_kernel_RN_haswell.c
+++ b/kernel/x86_64/dtrsm_kernel_RN_haswell.c
@@ -132,7 +132,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"1:									\n\t"
 
 	"	vmovups         (%8,%1,4), %%ymm4				\n\t"	// read a
-        "       vpermpd         $0xb1  , %%ymm0 , %%ymm3                	\n\t"
+        "       vpermilpd         $0x05  , %%ymm0 , %%ymm3                	\n\t"   // was vpermpd 0xb1
 
 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm8			\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm12			\n\t"
@@ -143,7 +143,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 
         "       vpermpd         $0x1b  , %%ymm3 , %%ymm0                	\n\t"
 	"	vmovups       32(%9,%1,8), %%ymm6				\n\t"	// read b1
-        "       vpermpd         $0xb1  , %%ymm0 , %%ymm3                	\n\t"
+        "       vpermilpd         $0x05  , %%ymm0 , %%ymm3                	\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm10			\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm14			\n\t"
 
@@ -160,7 +160,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm8			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm12			\n\t"
 
-        "       vpermpd         $0xb1  , %%ymm4 , %%ymm4                	\n\t"
+        "       vpermilpd         $0x05  , %%ymm4 , %%ymm4                	\n\t"
 	"	vmovups         (%9,%1,8), %%ymm1				\n\t"	// read b0
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm9			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm13			\n\t"
@@ -170,7 +170,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm10			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm14			\n\t"
 
-        "       vpermpd         $0xb1  , %%ymm4 , %%ymm4                	\n\t"
+        "       vpermilpd         $0x05  , %%ymm4 , %%ymm4                	\n\t"
 	"	addq		$8, %1						\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm11			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm15			\n\t"
@@ -185,7 +185,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm8			\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm12			\n\t"
 
-        "       vpermpd         $0xb1  , %%ymm0 , %%ymm0                	\n\t"
+        "       vpermilpd         $0x05  , %%ymm0 , %%ymm0                	\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm9			\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm13			\n\t"
 
@@ -193,7 +193,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm10			\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm14			\n\t"
 
-        "       vpermpd         $0xb1  , %%ymm0 , %%ymm0                	\n\t"
+        "       vpermilpd         $0x05  , %%ymm0 , %%ymm0                	\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm11			\n\t"
 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm15			\n\t"
 
@@ -204,7 +204,7 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm8			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm12			\n\t"
 
-        "       vpermpd         $0xb1  , %%ymm4 , %%ymm4                	\n\t"
+        "       vpermilpd         $0x05  , %%ymm4 , %%ymm4                	\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm9			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm13			\n\t"
 
@@ -212,42 +212,38 @@ static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLON
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm10			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm14			\n\t"
 
-        "       vpermpd         $0xb1  , %%ymm4 , %%ymm4                	\n\t"
+        "       vpermilpd         $0x05  , %%ymm4 , %%ymm4                	\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm11			\n\t"
 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm15			\n\t"
 
 	"3:								\n\t"	
 
-        "       vpermpd         $0xb1  , %%ymm9 , %%ymm9                \n\t"
-        "       vpermpd         $0xb1  , %%ymm11, %%ymm11               \n\t"
+        "       vpermilpd         $0x05  , %%ymm9 , %%ymm9                \n\t"
+        "       vpermilpd         $0x05  , %%ymm11, %%ymm11               \n\t"
 
         "       vblendpd        $0x0a  , %%ymm9 , %%ymm8 , %%ymm0       \n\t"
         "       vblendpd        $0x05  , %%ymm9 , %%ymm8 , %%ymm1       \n\t"
         "       vblendpd        $0x0a  , %%ymm11, %%ymm10, %%ymm2       \n\t"
         "       vblendpd        $0x05  , %%ymm11, %%ymm10, %%ymm3       \n\t"
 
-        "       vpermpd         $0x1b  , %%ymm2 , %%ymm2                \n\t"
-        "       vpermpd         $0x1b  , %%ymm3 , %%ymm3                \n\t"
-        "       vpermpd         $0xb1  , %%ymm2 , %%ymm2                \n\t"
-        "       vpermpd         $0xb1  , %%ymm3 , %%ymm3                \n\t"
+        "       vperm2f128         $0x01  , %%ymm2 , %%ymm2 , %%ymm2    \n\t"
+        "       vperm2f128         $0x01  , %%ymm3 , %%ymm3 , %%ymm3    \n\t"
 
         "       vblendpd        $0x03  , %%ymm0 , %%ymm2 , %%ymm8       \n\t"
         "       vblendpd        $0x03  , %%ymm1 , %%ymm3 , %%ymm9       \n\t"
         "       vblendpd        $0x03  , %%ymm2 , %%ymm0 , %%ymm10      \n\t"
         "       vblendpd        $0x03  , %%ymm3 , %%ymm1 , %%ymm11      \n\t"
 
-        "       vpermpd         $0xb1  , %%ymm13, %%ymm13               \n\t"
-        "       vpermpd         $0xb1  , %%ymm15, %%ymm15               \n\t"
+        "       vpermilpd         $0x05  , %%ymm13, %%ymm13               \n\t"
+        "       vpermilpd         $0x05  , %%ymm15, %%ymm15               \n\t"
 
         "       vblendpd        $0x0a  , %%ymm13, %%ymm12, %%ymm0       \n\t"
         "       vblendpd        $0x05  , %%ymm13, %%ymm12, %%ymm1       \n\t"
         "       vblendpd        $0x0a  , %%ymm15, %%ymm14, %%ymm2       \n\t"
         "       vblendpd        $0x05  , %%ymm15, %%ymm14, %%ymm3       \n\t"
 
-        "       vpermpd         $0x1b  , %%ymm2 , %%ymm2                \n\t"
-        "       vpermpd         $0x1b  , %%ymm3 , %%ymm3                \n\t"
-        "       vpermpd         $0xb1  , %%ymm2 , %%ymm2                \n\t"
-        "       vpermpd         $0xb1  , %%ymm3 , %%ymm3                \n\t"
+        "       vperm2f128         $0x01  , %%ymm2 , %%ymm2 , %%ymm2       \n\t"
+        "       vperm2f128         $0x01  , %%ymm3 , %%ymm3 , %%ymm3       \n\t"
 
         "       vblendpd        $0x03  , %%ymm0 , %%ymm2 , %%ymm12      \n\t"
         "       vblendpd        $0x03  , %%ymm1 , %%ymm3 , %%ymm13      \n\t"
diff --git a/kernel/x86_64/zdot.c b/kernel/x86_64/zdot.c
index ef12569c8..48f855b0e 100644
--- a/kernel/x86_64/zdot.c
+++ b/kernel/x86_64/zdot.c
@@ -86,18 +86,26 @@ static void zdot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
 
 #endif
 
-OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
+
+#if defined(SMP)
+extern int blas_level1_thread_with_return_value(int mode, BLASLONG m, BLASLONG n,
+        BLASLONG k, void *alpha, void *a, BLASLONG lda, void *b, BLASLONG ldb,
+        void *c, BLASLONG ldc, int (*function)(), int nthreads);
+#endif
+                
+                
+
+static void zdot_compute (BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y,OPENBLAS_COMPLEX_FLOAT *result)
 {
 	BLASLONG i;
 	BLASLONG ix,iy;
 	FLOAT  dot[4] = { 0.0, 0.0, 0.0 , 0.0 } ; 
-
+	
 	if ( n <= 0 ) 
 	{
-//	        CREAL(result) = 0.0 ;
-//        	CIMAG(result) = 0.0 ;
-		OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
-		return(result);
+		OPENBLAS_COMPLEX_FLOAT res=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
+		*result=res;
+		return;
 
 	}
 
@@ -150,18 +158,68 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
 	}
 
 #if !defined(CONJ)
-	OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]);
-//	CREAL(result) = dot[0] - dot[1];
-//	CIMAG(result) = dot[2] + dot[3];
+	OPENBLAS_COMPLEX_FLOAT res=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]);
 #else
-	OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]);
-//	CREAL(result) = dot[0] + dot[1];
-//	CIMAG(result) = dot[2] - dot[3];
-
+	OPENBLAS_COMPLEX_FLOAT res=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]);
 #endif
-
-	return(result);
-
+        *result=res;
+	return;
 }
 
+#if defined(SMP)
+static int zdot_thread_function(BLASLONG n, BLASLONG dummy0,
+BLASLONG dummy1, FLOAT dummy2, FLOAT *x, BLASLONG inc_x, FLOAT *y,
+BLASLONG inc_y, FLOAT *result, BLASLONG dummy3)
+{
+        zdot_compute(n, x, inc_x, y, inc_y, (void *)result);
+        return 0;
+}
+#endif
+
+OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
+{
+#if defined(SMP)
+	int nthreads;
+	FLOAT dummy_alpha;
+#endif
+	OPENBLAS_COMPLEX_FLOAT zdot;
+       CREAL(zdot) = 0.0;
+       CIMAG(zdot) = 0.0;
+
+#if defined(SMP)
+	if (inc_x == 0 || inc_y == 0 || n <= 10000)
+		nthreads = 1;
+	else
+		nthreads = num_cpu_avail(1);
+
+	if (nthreads == 1) {
+		zdot_compute(n, x, inc_x, y, inc_y, &zdot);
+	} else {
+		int mode, i;
+		char result[MAX_CPU_NUMBER * sizeof(double) * 2];
+		OPENBLAS_COMPLEX_FLOAT *ptr;
+
+#if !defined(DOUBLE)
+		mode = BLAS_SINGLE  | BLAS_COMPLEX;
+#else
+		mode = BLAS_DOUBLE  | BLAS_COMPLEX;
+#endif
+
+		blas_level1_thread_with_return_value(mode, n, 0, 0, &dummy_alpha,
+				   x, inc_x, y, inc_y, result, 0,
+				   ( void *)zdot_thread_function, nthreads);
+
+		ptr = (OPENBLAS_COMPLEX_FLOAT *)result;
+		for (i = 0; i < nthreads; i++) {
+			CREAL(zdot) = CREAL(zdot) + CREAL(*ptr);
+			CIMAG(zdot) = CIMAG(zdot) + CIMAG(*ptr);
+			ptr = (void *)(((char *)ptr) + sizeof(double) * 2);
+		}
+	}
+#else
+	zdot_compute(n, x, inc_x, y, inc_y, &zdot);
+#endif
+
+	return zdot;
+}
 
diff --git a/kernel/x86_64/zdot_microk_haswell-2.c b/kernel/x86_64/zdot_microk_haswell-2.c
index 9f2fc2c1d..4eade7bfd 100644
--- a/kernel/x86_64/zdot_microk_haswell-2.c
+++ b/kernel/x86_64/zdot_microk_haswell-2.c
@@ -66,13 +66,17 @@ static void zdot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 
 	"vfmadd231pd       %%ymm8 , %%ymm12, %%ymm0     \n\t"  // x_r * y_r, x_i * y_i
 	"vfmadd231pd       %%ymm9 , %%ymm13, %%ymm1     \n\t"  // x_r * y_r, x_i * y_i
-	"vpermpd      $0xb1 , %%ymm12, %%ymm12               \n\t"
-	"vpermpd      $0xb1 , %%ymm13, %%ymm13               \n\t"
+	"vpermilpd      $0x05 , %%ymm12, %%ymm12               \n\t"
+	"vpermilpd      $0x05 , %%ymm13, %%ymm13               \n\t"
+//	"vpermpd      $0xb1 , %%ymm12, %%ymm12               \n\t"
+//	"vpermpd      $0xb1 , %%ymm13, %%ymm13               \n\t"
 
 	"vfmadd231pd       %%ymm10, %%ymm14, %%ymm2     \n\t"  // x_r * y_r, x_i * y_i
 	"vfmadd231pd       %%ymm11, %%ymm15, %%ymm3     \n\t"  // x_r * y_r, x_i * y_i
-	"vpermpd      $0xb1 , %%ymm14, %%ymm14               \n\t"
-	"vpermpd      $0xb1 , %%ymm15, %%ymm15               \n\t"
+	"vpermilpd      $0x05 , %%ymm14, %%ymm14               \n\t"
+	"vpermilpd      $0x05 , %%ymm15, %%ymm15               \n\t"
+//	"vpermpd      $0xb1 , %%ymm14, %%ymm14               \n\t"
+//	"vpermpd      $0xb1 , %%ymm15, %%ymm15               \n\t"
 
 	"vfmadd231pd       %%ymm8 , %%ymm12, %%ymm4     \n\t"  // x_r * y_i, x_i * y_r
 	"addq		$16 , %0	  	 	             \n\t"
@@ -151,13 +155,17 @@ static void zdot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 
 	"vfmadd231pd       %%ymm8 , %%ymm12, %%ymm0     \n\t"  // x_r * y_r, x_i * y_i
 	"vfmadd231pd       %%ymm9 , %%ymm13, %%ymm1     \n\t"  // x_r * y_r, x_i * y_i
-	"vpermpd      $0xb1 , %%ymm12, %%ymm12               \n\t"
-	"vpermpd      $0xb1 , %%ymm13, %%ymm13               \n\t"
+	"vpermilpd      $0x05 , %%ymm12, %%ymm12               \n\t"
+	"vpermilpd      $0x05 , %%ymm13, %%ymm13               \n\t"
+//	"vpermpd      $0xb1 , %%ymm12, %%ymm12               \n\t"
+//	"vpermpd      $0xb1 , %%ymm13, %%ymm13               \n\t"
 
 	"vfmadd231pd       %%ymm10, %%ymm14, %%ymm2     \n\t"  // x_r * y_r, x_i * y_i
 	"vfmadd231pd       %%ymm11, %%ymm15, %%ymm3     \n\t"  // x_r * y_r, x_i * y_i
-	"vpermpd      $0xb1 , %%ymm14, %%ymm14               \n\t"
-	"vpermpd      $0xb1 , %%ymm15, %%ymm15               \n\t"
+	"vpermilpd      $0x05 , %%ymm14, %%ymm14               \n\t"
+	"vpermilpd      $0x05 , %%ymm15, %%ymm15               \n\t"
+//	"vpermpd      $0xb1 , %%ymm14, %%ymm14               \n\t"
+//	"vpermpd      $0xb1 , %%ymm15, %%ymm15               \n\t"
 
 	"vfmadd231pd       %%ymm8 , %%ymm12, %%ymm4     \n\t"  // x_r * y_i, x_i * y_r
 	"addq		$16 , %0	  	 	             \n\t"
diff --git a/utest/CMakeLists.txt b/utest/CMakeLists.txt
index 4e647cadc..1e3051a8f 100644
--- a/utest/CMakeLists.txt
+++ b/utest/CMakeLists.txt
@@ -38,9 +38,14 @@ if (NOT NO_LAPACK)
 set(OpenBLAS_utest_src
   ${OpenBLAS_utest_src}
   test_potrs.c
+  )
+if (NOT NO_CBLAS AND NOT NO_LAPACKE)
+set(OpenBLAS_utest_src
+  ${OpenBLAS_utest_src}
   test_kernel_regress.c
   )
 endif()
+endif()
 
 set(OpenBLAS_utest_bin openblas_utest)
 add_executable(${OpenBLAS_utest_bin} ${OpenBLAS_utest_src})
diff --git a/utest/Makefile b/utest/Makefile
index cbe639cdb..8c7e6b9f8 100644
--- a/utest/Makefile
+++ b/utest/Makefile
@@ -1,6 +1,9 @@
 UTEST_CHECK = 1
 TOPDIR	= ..
 
+override TARGET_ARCH=
+override TARGET_MACH=
+
 UTESTBIN=openblas_utest
 
 .PHONY : all
@@ -13,8 +16,12 @@ OBJS=utest_main.o test_amax.o test_rotmg.o test_axpy.o test_dotu.o test_dsdot.o
 
 ifneq ($(NO_LAPACK), 1)
 OBJS += test_potrs.o
+ifneq ($(NO_CBLAS), 1)
+ifneq ($(NO_LAPACKE), 1)
 OBJS += test_kernel_regress.o
 endif
+endif
+endif
 
 #this does not work with OpenMP nor with native Windows or Android threads
 # FIXME TBD if this works on OSX, SunOS, POWER and zarch