diff --git a/.github/workflows/nightly-Homebrew-build.yml b/.github/workflows/nightly-Homebrew-build.yml
index b025f8634..29ec96f73 100644
--- a/.github/workflows/nightly-Homebrew-build.yml
+++ b/.github/workflows/nightly-Homebrew-build.yml
@@ -43,11 +43,6 @@ jobs:
       - name: Update Homebrew
         if: github.event_name != 'pull_request'
         run: brew update || true
-
-      - name: unlink installed gcc to allow updating
-        run: |
-          brew unlink gcc@8
-          brew unlink gcc@9
           
       - name: Install prerequisites
         run: brew install --fetch-HEAD --HEAD --only-dependencies --keep-tmp openblas
diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
index be9a32a7c..6be41960c 100644
--- a/CONTRIBUTORS.md
+++ b/CONTRIBUTORS.md
@@ -194,3 +194,6 @@ In chronological order:
 
 * PingTouGe Semiconductor Co., Ltd.
   * [2020-10] Add RISC-V Vector (0.7.1) support. Optimize BLAS kernels for Xuantie C910
+
+* River Dillon <oss@outerpassage.net>
+  * [2021-07-10] fix compilation with musl libc
diff --git a/Changelog.txt b/Changelog.txt
index 6c5cf573e..8cd101699 100644
--- a/Changelog.txt
+++ b/Changelog.txt
@@ -1,4 +1,52 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.16
+ 11-Jul-2021
+
+common:
+ - drastically reduced the stack size requirements for running the LAPACK
+   testsuite (Reference-LAPACK PR 553)
+ - fixed spurious test failures in the LAPACK testsuite (Reference-LAPACK
+   PR 564)
+ - expressly setting DYNAMIC_ARCH=0 no longer enables dynamic_arch mode 
+ - improved performance of xGER, xSPR, xSPR2, xSYR, xSYR2, xTRSV, SGEMV_N 
+   and DGEMV_N, for small input sizes and consecutive arguments
+ - improved performance of xGETRF, xPORTF and xPOTRI for small input sizes
+   by disabling multithreading
+ - fixed installing with BSD versions of the "install" utility
+
+RISCV:
+ - fixed the implementation of xIMIN
+ - improved the performance of DSDOT
+ - fixed linking of the tests on C910V with current vendor gcc
+
+POWER:
+- fixed SBGEMM computation for some odd value inputs
+- fixed compilation for PPCG4, PPC970, POWER3, POWER4 and POWER5
+
+x86_64:
+ - improved performance of SGEMV_N and SGEMV_T for small N on AVX512-capable cpus
+ - worked around a miscompilation of ZGEMM/ZTRMM on Sandybridge with old gcc
+   versions
+ - fixed compilation with MS Visual Studio versions older than 2017
+ - fixed macro name collision with winnt.h from the latest Win10 SDK
+ - added cpu type autodetection for Intel Ice Lake SP
+ - fixed cpu type autodetection for Intel Tiger Lake
+ - added cpu type autodetection for recent Centaur/Zhaoxin models
+ - fixed compilation with musl libc
+
+ARM64:
+- fixed compilation with gcc/gfortran on the Apple M1
+- fixed linking of the tests on FreeBSD
+- fixed missing restore of a register in the recently rewritten DNRM2 kernel
+  for ThunderX2 and Neoverse N1 that could cause spurious failures in e.g.
+  DGEEV
+- added compiler optimization flags for the EMAG8180
+- added initial support for Cortex A55
+
+ARM:
+- fixed linking of the tests on FreeBSD
+
 ====================================================================
 Version 0.3.15
   2-May-2021
diff --git a/Makefile.arm64 b/Makefile.arm64
index 23362b4e5..c23a0876e 100644
--- a/Makefile.arm64
+++ b/Makefile.arm64
@@ -57,6 +57,28 @@ endif
 endif
 endif
 
+# Use a53 tunings because a55 is only available in GCC>=8.1
+ifeq ($(CORE), CORTEXA55)
+ifeq ($(GCCVERSIONGTEQ7), 1)
+ifeq ($(GCCVERSIONGTEQ8), 1)
+CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a55
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a55
+endif
+else
+CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a53
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a53
+endif
+endif
+else
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
+endif
+endif
+endif
+
 ifeq ($(CORE), THUNDERX)
 CCOMMON_OPT += -march=armv8-a -mtune=thunderx
 ifneq ($(F_COMPILER), NAG)
@@ -107,4 +129,13 @@ FCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
 endif
 endif
 endif
+
+ifeq ($(GCCVERSIONGTEQ9), 1)
+ifeq ($(CORE), EMAG8180)
+CCOMMON_OPT += -march=armv8-a -mtune=emag
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a -mtune=emag
+endif
+endif
+endif
 endif
diff --git a/Makefile.install b/Makefile.install
index e8b64465f..28727de37 100644
--- a/Makefile.install
+++ b/Makefile.install
@@ -74,17 +74,17 @@ endif
 ifneq ($(OSNAME), AIX)
 ifndef NO_LAPACKE
 	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapack.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapack.h"
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h"
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_config.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_config.h"
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_mangling_with_flags.h.in "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h"
-	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_utils.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_utils.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapack.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapack.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_config.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_config.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_mangling_with_flags.h.in "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h"
+	@-install -m644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_utils.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_utils.h"
 endif
 
 #for install static library
 ifneq ($(NO_STATIC),1)
 	@echo Copying the static library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
-	@install -pm644 $(LIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+	@install -m644 $(LIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 endif
@@ -92,7 +92,7 @@ endif
 ifneq ($(NO_SHARED),1)
 	@echo Copying the shared library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
 ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku FreeBSD DragonFly))
-	@install -pm755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+	@install -m755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
diff --git a/Makefile.system b/Makefile.system
index ae703e4d9..bb8c60e91 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -333,6 +333,7 @@ GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
 GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
 GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
 GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
+GCCVERSIONGTEQ8 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 8)
 GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
 GCCVERSIONGTEQ11 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 11)
 GCCVERSIONGTEQ10 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 10)
@@ -380,6 +381,12 @@ ifeq ($(OSNAME), AIX)
 EXTRALIB	+= -lm
 endif
 
+ifeq ($(OSNAME), FreeBSD)
+ifeq ($(ARCH), $(filter $(ARCH),arm arm64))
+EXTRALIB	+= -lm
+endif
+endif
+
 ifeq ($(OSNAME), WINNT)
 NEED_PIC = 0
 NO_EXPRECISION = 1
@@ -619,6 +626,7 @@ DYNAMIC_CORE += CORTEXA57
 DYNAMIC_CORE += CORTEXA72
 DYNAMIC_CORE += CORTEXA73
 DYNAMIC_CORE += NEOVERSEN1
+DYNAMIC_CORE += CORTEXA55
 DYNAMIC_CORE += FALKOR
 DYNAMIC_CORE += THUNDERX
 DYNAMIC_CORE += THUNDERX2T99
diff --git a/Makefile.x86 b/Makefile.x86
index 893379c33..25ca660bd 100644
--- a/Makefile.x86
+++ b/Makefile.x86
@@ -1,6 +1,6 @@
 # COMPILER_PREFIX = mingw32-
 
-ifndef DYNAMIC_ARCH
+ifneq ($(DYNAMIC_ARCH),1)
 ADD_CPUFLAGS = 1
 else
 ifdef TARGET_CORE
diff --git a/Makefile.x86_64 b/Makefile.x86_64
index f62ab9e5e..307cbe1d9 100644
--- a/Makefile.x86_64
+++ b/Makefile.x86_64
@@ -9,7 +9,7 @@ endif
 endif
 
 
-ifndef DYNAMIC_ARCH
+ifneq ($(DYNAMIC_ARCH),1)
 ADD_CPUFLAGS = 1
 else
 ifdef TARGET_CORE
diff --git a/README.md b/README.md
index 174f951f4..d7e0d60a7 100644
--- a/README.md
+++ b/README.md
@@ -27,7 +27,7 @@ We provide official binary packages for the following platform:
 
   * Windows x86/x86_64
 
-You can download them from [file hosting on sourceforge.net](https://sourceforge.net/projects/openblas/files/).
+You can download them from [file hosting on sourceforge.net](https://sourceforge.net/projects/openblas/files/) or from the Releases section of the github project page, [https://github.com/xianyi/OpenBLAS/releases](https://github.com/xianyi/OpenBLAS/releases).
 
 ## Installation from Source
 
diff --git a/TargetList.txt b/TargetList.txt
index d19964916..f93a629d8 100644
--- a/TargetList.txt
+++ b/TargetList.txt
@@ -92,6 +92,7 @@ CORTEXA57
 CORTEXA72
 CORTEXA73
 NEOVERSEN1
+CORTEXA55
 EMAG8180
 FALKOR
 THUNDERX
diff --git a/appveyor.yml b/appveyor.yml
index c9b2fa3a1..d575c5b7f 100644
--- a/appveyor.yml
+++ b/appveyor.yml
@@ -47,6 +47,7 @@ environment:
 install:
   - if [%COMPILER%]==[clang-cl] call %CONDA_INSTALL_LOCN%\Scripts\activate.bat
   - if [%COMPILER%]==[clang-cl] conda config --add channels conda-forge --force
+  - if [%COMPILER%]==[clang-cl] conda config --set auto_update_conda false
   - if [%COMPILER%]==[clang-cl] conda install --yes --quiet clangdev cmake ninja flang=11.0.1
   - if [%COMPILER%]==[clang-cl] call "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\vcvarsall.bat" x64
   - if [%COMPILER%]==[clang-cl] set "LIB=%CONDA_INSTALL_LOCN%\Library\lib;%LIB%"
diff --git a/azure-pipelines.yml b/azure-pipelines.yml
index 56a3fd4ae..889b920e3 100644
--- a/azure-pipelines.yml
+++ b/azure-pipelines.yml
@@ -137,3 +137,31 @@ jobs:
       source /opt/intel/oneapi/setvars.sh
       make CC=/usr/local/opt/llvm/bin/clang FC=ifort
  
+- job: OSX_NDK_ARMV7
+  pool:
+     vmImage: 'macOS-10.15'
+  steps:   
+  - script: | 
+      brew update
+      brew install --cask android-ndk
+      export ANDROID_NDK_HOME=/usr/local/share/android-ndk
+      make TARGET=ARMV7 ONLY_CBLAS=1 CC=$ANDROID_NDK_HOME/toolchains/llvm/prebuilt/darwin-x86_64/bin/armv7a-linux-androideabi21-clang AR=$ANDROID_NDK_HOME/toolchains/llvm/prebuilt/darwin-x86_64/bin/arm-linux-androideabi-ar HOSTCC=gcc ARM_SOFTFP_ABI=1 -j4
+ 
+- job: ALPINE_MUSL
+  pool:
+     vmImage: 'ubuntu-latest'
+  steps:
+  - script: |
+        wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.9.0/alpine-chroot-install' \
+          && echo 'e5dfbbdc0c4b3363b99334510976c86bfa6cb251  alpine-chroot-install' | sha1sum -c || exit 1
+        alpine() { /alpine/enter-chroot -u "$USER" "$@"; }
+        sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers sudo'
+        alpine make DYNAMIC_ARCH=1 BINARY=64
+        alpine make DYNAMIC_ARCH=1 BINARY=64 PREFIX=mytestdir install
+        alpine ls -l mytestdir/include
+        alpine echo "// tests that inclusion of openblas_config.h works with musl" >test_install.c
+        alpine echo "#include <openblas_config.h>" >>test_install.c
+        alpine echo "int main(){" >> test_install.c
+        alpine echo "cpu_set_t* cpu_set = NULL;}" >>test_install.c
+        alpine gcc -Imytestdir/include test_install.c -Lmytestdir/lib -lopenblas -lpthread -lgfortran -o test_install
+        
diff --git a/benchmark/getri.c b/benchmark/getri.c
index 98a860906..4c8891226 100644
--- a/benchmark/getri.c
+++ b/benchmark/getri.c
@@ -72,13 +72,17 @@ int main(int argc, char *argv[]){
   FLOAT *a,*work;
   FLOAT wkopt[4];
   blasint *ipiv;
-  blasint m, i, j, info,lwork;
+  blasint m, i, j, l, info,lwork;
 
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
-  double time1;
+  double time1,timeg;
+  
+  char *p;
+  char btest = 'I';
 
   argc--;argv++;
 
@@ -86,6 +90,9 @@ int main(int argc, char *argv[]){
   if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
   if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
 
+  if ((p = getenv("OPENBLAS_TEST"))) btest=*p;
+  
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
 
   fprintf(stderr, "From : %3d  To : %3d Step = %3d\n", from, to, step);
 
@@ -124,32 +131,41 @@ int main(int argc, char *argv[]){
   fprintf(stderr, "   SIZE           FLops           Time          Lwork\n");
 
   for(m = from; m <= to; m += step){
-
+    timeg = 0.;
     fprintf(stderr, " %6d : ", (int)m);
 
-    GETRF (&m, &m, a, &m, ipiv, &info);
+    for (l = 0; l < loops; l++) {
 
+    if (btest == 'F') begin();
+    GETRF (&m, &m, a, &m, ipiv, &info);
+    if (btest == 'F') {
+      end();
+      timeg += getsec();
+    }
     if (info) {
       fprintf(stderr, "Matrix is not singular .. %d\n", info);
       exit(1);
     }
 
-    begin();
+    if (btest == 'I') begin();
 
     lwork = -1;
     GETRI(&m, a, &m, ipiv, wkopt, &lwork, &info);
 
     lwork = (blasint)wkopt[0];
     GETRI(&m, a, &m, ipiv, work, &lwork, &info);
-    end();
+    if (btest == 'I') end();
 
     if (info) {
       fprintf(stderr, "failed compute inverse matrix .. %d\n", info);
       exit(1);
     }
 
-    time1 = getsec();
-
+    if (btest == 'I') 
+      timeg += getsec();
+    
+    } // loops
+    time1 = timeg/(double)loops;
     fprintf(stderr,
 	    " %10.2f MFlops : %10.2f Sec : %d\n",
 	    COMPSIZE * COMPSIZE * (4.0/3.0 * (double)m * (double)m *(double)m - (double)m *(double)m + 5.0/3.0* (double)m) / time1 * 1.e-6,time1,lwork);
diff --git a/benchmark/linpack.c b/benchmark/linpack.c
index 202035245..32ccb0386 100644
--- a/benchmark/linpack.c
+++ b/benchmark/linpack.c
@@ -72,17 +72,21 @@ int main(int argc, char *argv[]){
   FLOAT *a, *b;
   blasint *ipiv;
 
-  blasint m, i, j, info;
+  blasint m, i, j, l, info;
   blasint unit =   1;
 
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
   FLOAT maxerr;
 
-  double time1, time2;
+  double time1, time2, timeg1,timeg2;
 
+  char *p;
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
+  
   argc--;argv++;
 
   if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
@@ -110,9 +114,9 @@ int main(int argc, char *argv[]){
   fprintf(stderr, "   SIZE       Residual     Decompose            Solve           Total\n");
 
   for(m = from; m <= to; m += step){
-
+    timeg1 = timeg2 = 0.;
     fprintf(stderr, " %6d : ", (int)m);
-
+    for (l = 0; l < loops; l++) {
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
 	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
@@ -138,7 +142,7 @@ int main(int argc, char *argv[]){
       exit(1);
     }
 
-    time1 = getsec();
+    timeg1 += getsec();
 
     begin();
 
@@ -151,8 +155,10 @@ int main(int argc, char *argv[]){
       exit(1);
     }
 
-    time2 = getsec();
-
+    timeg2 += getsec();
+    } //loops
+    time1=timeg1/(double)loops;
+    time2=timeg2/(double)loops;
     maxerr = 0.;
 
     for(i = 0; i < m; i++){
diff --git a/benchmark/potrf.c b/benchmark/potrf.c
index 116d0cca5..8808203a5 100644
--- a/benchmark/potrf.c
+++ b/benchmark/potrf.c
@@ -99,14 +99,15 @@ int main(int argc, char *argv[]){
   char *p;
   char btest = 'F';
 
-  blasint m, i, j, info, uplos=0;
-  double flops;
+  blasint m, i, j, l, info, uplos=0;
+  double flops = 0.;
 
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
-  double time1;
+  double time1, timeg;
 
   argc--;argv++;
 
@@ -119,6 +120,8 @@ int main(int argc, char *argv[]){
 
   if ((p = getenv("OPENBLAS_TEST"))) btest=*p;
 
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
+
   fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c\n", from, to, step,*uplo[uplos]);
 
   if (( a    = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
@@ -129,19 +132,21 @@ int main(int argc, char *argv[]){
     fprintf(stderr,"Out of Memory!!\n");exit(1);
   }
 
-  for(m = from; m <= to; m += step){
 
+  for(m = from; m <= to; m += step){
+    timeg=0.;
+	  for (l = 0; l < loops; l++) {
 #ifndef COMPLEX
       if (uplos & 1) {
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i < j; i++)     a[(long)i + (long)j * (long)m] = 0.;
-	                             a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
 	  for(i = j + 1; i < m; i++) a[(long)i + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) - 0.5;
 	}
       } else {
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i < j; i++)     a[(long)i + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) - 0.5;
-	                             a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
 	  for(i = j + 1; i < m; i++) a[(long)i + (long)j * (long)m] = 0.;
 	}
       }
@@ -192,8 +197,8 @@ int main(int argc, char *argv[]){
 	exit(1);
       }
 
-      time1 = getsec();
-      flops = COMPSIZE * COMPSIZE * (1.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 1.0/6.0* (double)m) / time1 * 1.e-6;
+      if ( btest == 'F')
+	      timeg += getsec();
 
       if ( btest == 'S' )
       {
@@ -214,9 +219,7 @@ int main(int argc, char *argv[]){
 		fprintf(stderr, "Potrs info = %d\n", info);
 		exit(1);
         }
-        time1 = getsec();
-        flops = COMPSIZE * COMPSIZE * (2.0 * (double)m * (double)m *(double)m ) / time1 * 1.e-6;
-
+        timeg += getsec();
       }
 	
       if ( btest == 'I' )
@@ -232,11 +235,17 @@ int main(int argc, char *argv[]){
 		fprintf(stderr, "Potri info = %d\n", info);
 		exit(1);
         }
-
-        time1 = getsec();
-        flops = COMPSIZE * COMPSIZE * (2.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 5.0/6.0* (double)m) / time1 * 1.e-6;
+        timeg += getsec();
       }
-	
+    } // loops
+
+      time1 = timeg/(double)loops;
+	if ( btest == 'F')
+		flops = COMPSIZE * COMPSIZE * (1.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 1.0/6.0* (double)m) / time1 * 1.e-6;
+	if ( btest == 'S')
+		flops = COMPSIZE * COMPSIZE * (2.0 * (double)m * (double)m *(double)m ) / time1 * 1.e-6;
+	if ( btest == 'I')
+		flops = COMPSIZE * COMPSIZE * (2.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 5.0/6.0* (double)m) / time1 * 1.e-6;
       fprintf(stderr, "%8d : %10.2f MFlops : %10.3f Sec : Test=%c\n",m,flops ,time1,btest);
 
 
diff --git a/benchmark/syr2.c b/benchmark/syr2.c
index acbc86987..61d1036ea 100644
--- a/benchmark/syr2.c
+++ b/benchmark/syr2.c
@@ -46,14 +46,17 @@ int main(int argc, char *argv[]){
 
   if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p; 
 
-  blasint m, i, j;
+  blasint m, i, j, l;
   blasint inc_x= 1;
   blasint inc_y= 1;
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
-  double time1;
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
+
+  double time1,timeg;
 
   argc--;argv++;
 
@@ -85,8 +88,9 @@ int main(int argc, char *argv[]){
 
   for(m = from; m <= to; m += step)
   {
-	
+    timeg = 0.;	
     fprintf(stderr, " %6d : ", (int)m);
+    for (l = 0; l < loops; l++) {
 	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
 	x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
    	}
@@ -107,8 +111,10 @@ int main(int argc, char *argv[]){
 
     end();
 
-    time1 = getsec();
+    timeg += getsec();
+    } // loops
 
+    time1 = timeg/(double)loops;
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m / time1 * 1.e-6);
diff --git a/benchmark/syrk.c b/benchmark/syrk.c
index 82606a21a..fa0f24666 100644
--- a/benchmark/syrk.c
+++ b/benchmark/syrk.c
@@ -56,17 +56,20 @@ int main(int argc, char *argv[]){
 
   char uplo='U';
   char trans='N';
-
+  
   if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
   if ((p = getenv("OPENBLAS_TRANS"))) trans=*p;
 
-  blasint m, i, j;
+  blasint m, i, j, l;
 
   int from =   1;
   int to   = 200;
   int step =   1;
+  int loops =  1;
 
-  double time1;
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
+
+  double time1,timeg;
 
   argc--;argv++;
 
@@ -95,9 +98,12 @@ int main(int argc, char *argv[]){
 
   for(m = from; m <= to; m += step)
   {
+    timeg = 0.;
 
     fprintf(stderr, " %6d : ", (int)m);
 
+    for(l = 0; l < loops; l++) {
+
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
 	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
@@ -111,8 +117,10 @@ int main(int argc, char *argv[]){
 
     end();
 
-    time1 = getsec();
-
+    timeg += getsec();
+    
+    } //loops
+    time1 = timeg / (double)loops;
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 1. * (double)m * (double)m * (double)m / time1 * 1.e-6);
diff --git a/cmake/arch.cmake b/cmake/arch.cmake
index 4451f9eaa..154e59db6 100644
--- a/cmake/arch.cmake
+++ b/cmake/arch.cmake
@@ -44,7 +44,7 @@ endif ()
 
 if (DYNAMIC_ARCH)
   if (ARM64)
-    set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
+	  set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA55 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
     if (DYNAMIC_LIST)
 	    set(DYNAMIC_CORE ARMV8 ${DYNAMIC_LIST})
     endif ()
diff --git a/cmake/prebuild.cmake b/cmake/prebuild.cmake
index da7686c33..d86e10035 100644
--- a/cmake/prebuild.cmake
+++ b/cmake/prebuild.cmake
@@ -177,7 +177,7 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
     set(ZGEMM_UNROLL_M 4)
     set(ZGEMM_UNROLL_N 4)
     set(SYMV_P 16)
-  elseif ("${TCORE}" STREQUAL "CORTEXA57" OR "${TCORE}" STREQUAL "CORTEXA53")
+  elseif ("${TCORE}" STREQUAL "CORTEXA57" OR "${TCORE}" STREQUAL "CORTEXA53" OR "${TCORE}" STREQUAL "CORTEXA55")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_CODE_SIZE\t32768\n"
       "#define L1_CODE_LINESIZE\t64\n"
diff --git a/cmake/system.cmake b/cmake/system.cmake
index d6c71b774..34874827c 100644
--- a/cmake/system.cmake
+++ b/cmake/system.cmake
@@ -39,7 +39,7 @@ if (DEFINED BINARY AND DEFINED TARGET AND BINARY EQUAL 32)
   if (${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "PILEDRIVER" OR ${TARGET} STREQUAL "ZEN")
     set(TARGET "BARCELONA")
   endif ()
-  if (${TARGET} STREQUAL "ARMV8" OR ${TARGET} STREQUAL "CORTEXA57" OR ${TARGET} STREQUAL "CORTEXA53")
+  if (${TARGET} STREQUAL "ARMV8" OR ${TARGET} STREQUAL "CORTEXA57" OR ${TARGET} STREQUAL "CORTEXA53" OR ${TARGET} STREQUAL "CORTEXA55")
     set(TARGET "ARMV7")
   endif ()
 endif ()
diff --git a/cmake/utils.cmake b/cmake/utils.cmake
index 29b5a067b..794d73d06 100644
--- a/cmake/utils.cmake
+++ b/cmake/utils.cmake
@@ -254,6 +254,19 @@ function(GenerateNamedObjects sources_in)
       # now add the object and set the defines
       set(obj_defines ${defines_in})
 
+      list(FIND obj_defines "RC" def_idx)
+      if (${def_idx} GREATER -1) 
+	      #	      list(REMOVE_AT ${obj_defines} ${def_idx})
+	      list (REMOVE_ITEM obj_defines "RC")
+	      list(APPEND obj_defines  "RC=RC")
+      endif ()
+      list(FIND obj_defines "CR" def_idx)
+      if (${def_idx} GREATER -1) 
+	      #	      list(REMOVE_AT ${obj_defines} ${def_idx})
+	      list (REMOVE_ITEM obj_defines "CR")
+	      list(APPEND obj_defines  "CR=CR")
+      endif ()
+
       if (use_cblas)
         set(obj_name "cblas_${obj_name}")
         list(APPEND obj_defines "CBLAS")
diff --git a/common_interface.h b/common_interface.h
index b9ebb2772..318827920 100644
--- a/common_interface.h
+++ b/common_interface.h
@@ -709,6 +709,13 @@ int BLASFUNC(cpotrf)(char *, blasint *, float  *, blasint *, blasint *);
 int BLASFUNC(zpotrf)(char *, blasint *, double *, blasint *, blasint *);
 int BLASFUNC(xpotrf)(char *, blasint *, xdouble *, blasint *, blasint *);
 
+int BLASFUNC(spotri)(char *, blasint *, float  *, blasint *, blasint *);
+int BLASFUNC(dpotri)(char *, blasint *, double *, blasint *, blasint *);
+int BLASFUNC(qpotri)(char *, blasint *, xdouble *, blasint *, blasint *);
+int BLASFUNC(cpotri)(char *, blasint *, float  *, blasint *, blasint *);
+int BLASFUNC(zpotri)(char *, blasint *, double *, blasint *, blasint *);
+int BLASFUNC(xpotri)(char *, blasint *, xdouble *, blasint *, blasint *);
+
 int BLASFUNC(spotrs)(char *, blasint *, blasint *, float   *, blasint *, float   *, blasint *, blasint *);
 int BLASFUNC(dpotrs)(char *, blasint *, blasint *, double  *, blasint *, double  *, blasint *, blasint *);
 int BLASFUNC(qpotrs)(char *, blasint *, blasint *, xdouble *, blasint *, xdouble *, blasint *, blasint *);
diff --git a/cpuid_arm64.c b/cpuid_arm64.c
index 5f5d7771b..2a9399f7d 100644
--- a/cpuid_arm64.c
+++ b/cpuid_arm64.c
@@ -36,6 +36,7 @@ size_t length=sizeof(value);
 #define CPU_ARMV8       	1
 // Arm
 #define CPU_CORTEXA53     2
+#define CPU_CORTEXA55     14
 #define CPU_CORTEXA57     3
 #define CPU_CORTEXA72     4
 #define CPU_CORTEXA73     5
@@ -67,7 +68,8 @@ static char *cpuname[] = {
   "EMAG8180",
   "NEOVERSEN1",
   "THUNDERX3T110",
-  "VORTEX"	
+  "VORTEX",
+  "CORTEXA55"
 };
 
 static char *cpuname_lower[] = {
@@ -84,7 +86,8 @@ static char *cpuname_lower[] = {
   "emag8180",
   "neoversen1",
   "thunderx3t110",
-  "vortex"	
+  "vortex",
+  "cortexa55"
 };
 
 int get_feature(char *search)
@@ -161,6 +164,8 @@ int detect(void)
         return CPU_CORTEXA73;
       else if (strstr(cpu_part, "0xd0c"))
         return CPU_NEOVERSEN1;
+      else if (strstr(cpu_part, "0xd05"))
+	return CPU_CORTEXA55;
     }
     // Qualcomm
     else if (strstr(cpu_implementer, "0x51") && strstr(cpu_part, "0xc00"))
@@ -281,6 +286,7 @@ void get_cpuconfig(void)
 	{
 
     case CPU_CORTEXA53:
+    case CPU_CORTEXA55:			
       printf("#define %s\n", cpuname[d]);
       // Fall-through
 		case CPU_ARMV8:
diff --git a/cpuid_x86.c b/cpuid_x86.c
index 44704fcd9..00fc8baa0 100644
--- a/cpuid_x86.c
+++ b/cpuid_x86.c
@@ -283,6 +283,7 @@ int get_vendor(void){
   if (!strcmp(vendor, "CyrixInstead")) return VENDOR_CYRIX;
   if (!strcmp(vendor, "NexGenDriven")) return VENDOR_NEXGEN;
   if (!strcmp(vendor, "CentaurHauls")) return VENDOR_CENTAUR;
+  if (!strcmp(vendor, "  Shanghai  ")) return VENDOR_CENTAUR;
   if (!strcmp(vendor, "RiseRiseRise")) return VENDOR_RISE;
   if (!strcmp(vendor, " SiS SiS SiS")) return VENDOR_SIS;
   if (!strcmp(vendor, "GenuineTMx86")) return VENDOR_TRANSMETA;
@@ -1398,6 +1399,17 @@ int get_cpuname(void){
 	    return CPUTYPE_SANDYBRIDGE;
 	  else
 	  return CPUTYPE_NEHALEM;
+	case 10: // Ice Lake SP
+	  if(support_avx512_bf16())
+            return CPUTYPE_COOPERLAKE;	
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	  return CPUTYPE_NEHALEM;	
         }
       break;
       case 7: // family 6 exmodel 7
@@ -1620,7 +1632,9 @@ int get_cpuname(void){
     case 0x6:
       return CPUTYPE_NANO;
       break;
-
+    case 0x7:
+      return CPUTYPE_NEHALEM;
+      break;
     }
     return CPUTYPE_VIAC3;
   }
@@ -2112,7 +2126,22 @@ int get_coretype(void){
 #endif
 	  else
 	    return CORE_NEHALEM;
-#endif			
+#endif
+	if (model == 10)
+#ifndef NO_AVX512
+	  if(support_avx512_bf16())
+            return CORE_COOPERLAKE;
+	  return CORE_SKYLAKEX;
+#else
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CORE_HASWELL;
+#else
+	    return CORE_SANDYBRIDGE;
+#endif
+	  else
+	    return CORE_NEHALEM;
+#endif	
         break;    	
       case 7:
         if (model == 10) 
@@ -2135,13 +2164,13 @@ int get_coretype(void){
       case 8:
        if (model == 12) { // Tiger Lake
           if(support_avx512())
-            return CPUTYPE_SKYLAKEX;
+            return CORE_SKYLAKEX;
           if(support_avx2())
-            return CPUTYPE_HASWELL;
+            return CORE_HASWELL;
           if(support_avx())
-            return CPUTYPE_SANDYBRIDGE;
+            return CORE_SANDYBRIDGE;
           else
-          return CPUTYPE_NEHALEM;
+          return CORE_NEHALEM;
         }
         if (model == 14) { // Kaby Lake 
 	  if(support_avx())
@@ -2259,6 +2288,9 @@ int get_coretype(void){
     case 0x6:
       return CORE_NANO;
       break;
+    case 0x7:
+      return CORE_NEHALEM;
+      break;
     }
     return CORE_VIAC3;
   }
diff --git a/driver/level3/Makefile b/driver/level3/Makefile
index 78f32b961..b8465d4ed 100644
--- a/driver/level3/Makefile
+++ b/driver/level3/Makefile
@@ -425,7 +425,7 @@ cgemm_rr.$(SUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm_rc.$(SUFFIX) : gemm.c level3.c  ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm_cn.$(SUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -473,7 +473,7 @@ zgemm_rr.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm_rc.$(SUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm_cn.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -521,7 +521,7 @@ xgemm_rr.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm_rc.$(SUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm_cn.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -632,7 +632,7 @@ cgemm_thread_rr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm_thread_rc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm_thread_cn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -680,7 +680,7 @@ zgemm_thread_rr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm_thread_rc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm_thread_cn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -728,7 +728,7 @@ xgemm_thread_rr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm_thread_rc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm_thread_cn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -1895,7 +1895,7 @@ cgemm3m_rr.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm3m_rc.$(SUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm3m_cn.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -1943,7 +1943,7 @@ zgemm3m_rr.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm3m_rc.$(SUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm3m_cn.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -1991,7 +1991,7 @@ xgemm3m_rr.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm3m_rc.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm3m_cn.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2048,7 +2048,7 @@ cgemm3m_thread_rr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm3m_thread_rc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm3m_thread_cn.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2096,7 +2096,7 @@ zgemm3m_thread_rr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm3m_thread_rc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm3m_thread_cn.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2144,7 +2144,7 @@ xgemm3m_thread_rr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm3m_thread_rc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm3m_thread_cn.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2817,7 +2817,7 @@ cgemm_rr.$(PSUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm_rc.$(PSUFFIX) : gemm.c level3.c  ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm_cn.$(PSUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2865,7 +2865,7 @@ zgemm_rr.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm_rc.$(PSUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm_cn.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -2913,7 +2913,7 @@ xgemm_rr.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm_rc.$(PSUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm_cn.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -3025,7 +3025,7 @@ cgemm_thread_rr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm_thread_rc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm_thread_cn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -3073,7 +3073,7 @@ zgemm_thread_rr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm_thread_rc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm_thread_cn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -3121,7 +3121,7 @@ xgemm_thread_rr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm_thread_rc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm_thread_cn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4288,7 +4288,7 @@ cgemm3m_rr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm3m_rc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm3m_cn.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4336,7 +4336,7 @@ zgemm3m_rr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm3m_rc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm3m_cn.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4384,7 +4384,7 @@ xgemm3m_rr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm3m_rc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm3m_cn.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4441,7 +4441,7 @@ cgemm3m_thread_rr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 cgemm3m_thread_rc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 cgemm3m_thread_cn.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4489,7 +4489,7 @@ zgemm3m_thread_rr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 zgemm3m_thread_rc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 zgemm3m_thread_cn.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCN $< -o $(@F)
@@ -4537,7 +4537,7 @@ xgemm3m_thread_rr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRR $< -o $(@F)
 
 xgemm3m_thread_rc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DRC=RC $< -o $(@F)
 
 xgemm3m_thread_cn.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCN $< -o $(@F)
diff --git a/driver/others/dynamic.c b/driver/others/dynamic.c
index 158e1b3da..1a33870db 100644
--- a/driver/others/dynamic.c
+++ b/driver/others/dynamic.c
@@ -404,6 +404,7 @@ static int get_vendor(void){
   if (!strcmp(vendor.vchar, "GenuineIntel")) return VENDOR_INTEL;
   if (!strcmp(vendor.vchar, "AuthenticAMD")) return VENDOR_AMD;
   if (!strcmp(vendor.vchar, "CentaurHauls")) return VENDOR_CENTAUR;
+  if (!strcmp(vendor.vchar, "  Shanghai  ")) return VENDOR_CENTAUR;
   if (!strcmp(vendor.vchar, "HygonGenuine")) return VENDOR_HYGON;
 
   if ((eax == 0) || ((eax & 0x500) != 0)) return VENDOR_INTEL;
@@ -621,6 +622,22 @@ static gotoblas_t *get_coretype(void){
 	    return &gotoblas_NEHALEM;
 	  }
         }
+	if (model == 10) {
+          // Ice Lake SP
+	   if(support_avx512_bf16())
+             return &gotoblas_COOPERLAKE;
+          if (support_avx512()) 
+	    return &gotoblas_SKYLAKEX;
+	  if(support_avx2())
+	    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 7:
 	if (model == 10) // Goldmont Plus
@@ -808,6 +825,9 @@ static gotoblas_t *get_coretype(void){
     switch (family) {
     case 0x6:
       return &gotoblas_NANO;
+      break;
+    case 0x7:
+      return &gotoblas_NEHALEM;
     }
   }
 
diff --git a/driver/others/dynamic_arm64.c b/driver/others/dynamic_arm64.c
index 0b623c3ac..04ceaaf6d 100644
--- a/driver/others/dynamic_arm64.c
+++ b/driver/others/dynamic_arm64.c
@@ -99,6 +99,11 @@ extern gotoblas_t  gotoblas_NEOVERSEN1;
 #else
 #define gotoblas_NEOVERSEN1 gotoblas_ARMV8
 #endif
+#ifdef DYN_CORTEX_A55
+extern gotoblas_t  gotoblas_CORTEXA55;
+#else
+#define gotoblas_CORTEXA55 gotoblas_ARMV8
+#endif
 #else
 extern gotoblas_t  gotoblas_CORTEXA53;
 extern gotoblas_t  gotoblas_CORTEXA57;
@@ -111,11 +116,12 @@ extern gotoblas_t  gotoblas_TSV110;
 extern gotoblas_t  gotoblas_EMAG8180;
 extern gotoblas_t  gotoblas_NEOVERSEN1;
 extern gotoblas_t  gotoblas_THUNDERX3T110;
+extern gotoblas_t  gotoblas_CORTEXA55;
 #endif
 
 extern void openblas_warning(int verbose, const char * msg);
 
-#define NUM_CORETYPES   12
+#define NUM_CORETYPES   13
 
 /*
  * In case asm/hwcap.h is outdated on the build system, make sure
@@ -142,6 +148,7 @@ static char *corename[] = {
   "emag8180",
   "neoversen1",
   "thunderx3t110",
+  "cortexa55",
   "unknown"
 };
 
@@ -158,6 +165,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_EMAG8180)     return corename[ 9];
   if (gotoblas == &gotoblas_NEOVERSEN1)   return corename[10];
   if (gotoblas == &gotoblas_THUNDERX3T110) return corename[11];
+  if (gotoblas == &gotoblas_CORTEXA55)    return corename[12];
   return corename[NUM_CORETYPES];
 }
 
@@ -189,6 +197,7 @@ static gotoblas_t *force_coretype(char *coretype) {
     case  9: return (&gotoblas_EMAG8180);
     case 10: return (&gotoblas_NEOVERSEN1);
     case 11: return (&gotoblas_THUNDERX3T110);
+    case 12: return (&gotoblas_CORTEXA55);
   }
   snprintf(message, 128, "Core not found: %s\n", coretype);
   openblas_warning(1, message);
@@ -247,6 +256,8 @@ static gotoblas_t *get_coretype(void) {
           return &gotoblas_CORTEXA73;
         case 0xd0c: // Neoverse N1
           return &gotoblas_NEOVERSEN1;
+	case 0xd05: // Cortex A55
+	  return &gotoblas_CORTEXA55;
       }
       break;
     case 0x42: // Broadcom
diff --git a/driver/others/memory.c b/driver/others/memory.c
index 63fa6a566..6e654ccf2 100644
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@@ -1702,7 +1702,6 @@ inline int atoi(const char *str) { return 0; }
 #include <sys/sysinfo.h>
 #include <sched.h>
 #include <errno.h>
-#include <linux/unistd.h>
 #include <sys/syscall.h>
 #include <sys/time.h>
 #include <sys/resource.h>
diff --git a/f_check b/f_check
index 2c0d7fcb9..4825fb09a 100644
--- a/f_check
+++ b/f_check
@@ -314,11 +314,11 @@ if ($link ne "") {
 
     $link =~ s/\-Y\sP\,/\-Y/g;
     
-    $link =~ s/\-R\s*/\-rpath\@/g;
+    $link =~ s/\-R\s*/\-rpath\%/g;
 
-    $link =~ s/\-rpath\s+/\-rpath\@/g;
+    $link =~ s/\-rpath\s+/\-rpath\%/g;
 
-    $link =~ s/\-rpath-link\s+/\-rpath-link\@/g;
+    $link =~ s/\-rpath-link\s+/\-rpath-link\%/g;
 
     @flags = split(/[\s\,\n]/, $link);
     # remove leading and trailing quotes from each flag.
@@ -344,13 +344,13 @@ if ($link ne "") {
 	}
 
 
-	if ($flags =~ /^\-rpath\@/) {
-	    $flags =~ s/\@/\,/g;
+	if ($flags =~ /^\-rpath\%/) {
+	    $flags =~ s/\%/\,/g;
 	    $linker_L .= "-Wl,". $flags . " " ;
 	}
 
-	if ($flags =~ /^\-rpath-link\@/) {
-	    $flags =~ s/\@/\,/g;
+	if ($flags =~ /^\-rpath-link\%/) {
+	    $flags =~ s/\%/\,/g;
 	    $linker_L .= "-Wl,". $flags . " " ;
 	}
 	if ($flags =~ /-lgomp/ && $ENV{"CC"} =~ /clang/) {
diff --git a/getarch.c b/getarch.c
index f48944f36..3bc8a0c3d 100644
--- a/getarch.c
+++ b/getarch.c
@@ -1159,6 +1159,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_CORTEXA55
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "CORTEXA55"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DCORTEXA55 " \
+       "-DL1_CODE_SIZE=16384 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=3 " \
+       "-DL1_DATA_SIZE=16384 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=2 " \
+       "-DL2_SIZE=65536 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
+#define LIBNAME   "cortexa55"
+#define CORENAME  "CORTEXA55"
+#else
+#endif
 
 #ifdef FORCE_FALKOR
 #define FORCE
diff --git a/interface/gemm.c b/interface/gemm.c
index 6fde69049..10426fd8f 100644
--- a/interface/gemm.c
+++ b/interface/gemm.c
@@ -49,6 +49,8 @@
 #define ERROR_NAME "QGEMM "
 #elif defined(DOUBLE)
 #define ERROR_NAME "DGEMM "
+#elif defined(BFLOAT16)
+#define ERROR_NAME "SBGEMM "
 #else
 #define ERROR_NAME "SGEMM "
 #endif
@@ -124,6 +126,7 @@ void NAME(char *TRANSA, char *TRANSB,
 
 #ifdef SMP
   double MNK;
+#if defined(USE_SIMPLE_THREADED_LEVEL3) || !defined(NO_AFFINITY)
 #ifndef COMPLEX
 #ifdef XDOUBLE
   int mode  =  BLAS_XDOUBLE | BLAS_REAL;
@@ -142,6 +145,7 @@ void NAME(char *TRANSA, char *TRANSB,
 #endif
 #endif
 #endif
+#endif
 
 #if defined(SMP) && !defined(NO_AFFINITY) && !defined(USE_SIMPLE_THREADED_LEVEL3)
   int nodes;
diff --git a/interface/gemv.c b/interface/gemv.c
index d5d739fb1..1f14cdb2c 100644
--- a/interface/gemv.c
+++ b/interface/gemv.c
@@ -202,6 +202,11 @@ void CNAME(enum CBLAS_ORDER order,
 
   if (alpha == ZERO) return;
 
+  if (trans == 0 && incx == 1 && incy == 1 && m*n < 2304 *GEMM_MULTITHREAD_THRESHOLD) {
+    GEMV_N(m, n, 0, alpha, a, lda, x, incx, y, incy, NULL);
+    return;
+  }    
+
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
diff --git a/interface/ger.c b/interface/ger.c
index 8cf1614e3..af6ae8606 100644
--- a/interface/ger.c
+++ b/interface/ger.c
@@ -164,6 +164,11 @@ void CNAME(enum CBLAS_ORDER order,
   if (m == 0 || n == 0) return;
   if (alpha == 0.) return;
 
+  if (incx == 1 && incy == 1 && 1L*m*n <= 2048 *GEMM_MULTITHREAD_THRESHOLD) {
+    GER(m, n, 0, alpha, x, incx, y, incy, a, lda, NULL);
+    return;
+  }  
+
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
diff --git a/interface/lapack/getrf.c b/interface/lapack/getrf.c
index 02bb124b3..323370ebc 100644
--- a/interface/lapack/getrf.c
+++ b/interface/lapack/getrf.c
@@ -95,7 +95,14 @@ int NAME(blasint *M, blasint *N, FLOAT *a, blasint *ldA, blasint *ipiv, blasint
 
 #ifdef SMP
   args.common = NULL;
-  args.nthreads = num_cpu_avail(4);
+#ifndef DOUBLE
+  if (args.m*args.n < 40000)
+#else
+  if (args.m*args.n < 10000)
+#endif
+	args.nthreads=1;
+  else
+	args.nthreads = num_cpu_avail(4);
 
   if (args.nthreads == 1) {
 #endif
diff --git a/interface/lapack/potrf.c b/interface/lapack/potrf.c
index dbd55f62f..3abc80133 100644
--- a/interface/lapack/potrf.c
+++ b/interface/lapack/potrf.c
@@ -112,6 +112,13 @@ int NAME(char *UPLO, blasint *N, FLOAT *a, blasint *ldA, blasint *Info){
 
 #ifdef SMP
   args.common = NULL;
+#ifndef DOUBLE
+  if (args.n <128)
+#else
+  if (args.n <64)
+#endif
+    args.nthreads = 1;
+  else
   args.nthreads = num_cpu_avail(4);
 
   if (args.nthreads == 1) {
diff --git a/interface/lapack/potri.c b/interface/lapack/potri.c
index 2c0c64b6f..eb0fcbe70 100644
--- a/interface/lapack/potri.c
+++ b/interface/lapack/potri.c
@@ -121,6 +121,9 @@ int NAME(char *UPLO, blasint *N, FLOAT *a, blasint *ldA, blasint *Info){
 
 #ifdef SMP
   args.common = NULL;
+  if (args.n < 180)
+    args.nthreads = 1;
+  else
   args.nthreads = num_cpu_avail(4);
 
   if (args.nthreads == 1) {
diff --git a/interface/lapack/zgetrf.c b/interface/lapack/zgetrf.c
index 7f8db94f6..d03541fad 100644
--- a/interface/lapack/zgetrf.c
+++ b/interface/lapack/zgetrf.c
@@ -95,7 +95,10 @@ int NAME(blasint *M, blasint *N, FLOAT *a, blasint *ldA, blasint *ipiv, blasint
 
 #ifdef SMP
   args.common = NULL;
-  args.nthreads = num_cpu_avail(4);
+  if (args.m*args.n <10000)
+	args.nthreads = 1;
+  else
+	args.nthreads = num_cpu_avail(4);
 
   if (args.nthreads == 1) {
 #endif
diff --git a/interface/lapack/zpotrf.c b/interface/lapack/zpotrf.c
index c4cd99bf6..298efbbc1 100644
--- a/interface/lapack/zpotrf.c
+++ b/interface/lapack/zpotrf.c
@@ -112,6 +112,13 @@ int NAME(char *UPLO, blasint *N, FLOAT *a, blasint *ldA, blasint *Info){
 
 #ifdef SMP
   args.common = NULL;
+#ifndef DOUBLE
+  if (args.n < 64)
+#else
+  if (args.n < 64)
+#endif
+    args.nthreads = 1;
+  else
   args.nthreads = num_cpu_avail(4);
 
   if (args.nthreads == 1) {
diff --git a/interface/lapack/zpotri.c b/interface/lapack/zpotri.c
index 8da211683..8748c6352 100644
--- a/interface/lapack/zpotri.c
+++ b/interface/lapack/zpotri.c
@@ -121,6 +121,15 @@ int NAME(char *UPLO, blasint *N, FLOAT *a, blasint *ldA, blasint *Info){
 
 #ifdef SMP
   args.nthreads = num_cpu_avail(4);
+#ifndef DOUBLE
+  if (args.n < 200)
+#else
+  if (args.n < 150)
+#endif
+	  args.nthreads=1;
+  else
+#endif
+  args.nthreads = num_cpu_avail(4);
 
   if (args.nthreads == 1) {
 #endif
diff --git a/interface/spr.c b/interface/spr.c
index 1956986e9..8aafc9f85 100644
--- a/interface/spr.c
+++ b/interface/spr.c
@@ -167,6 +167,26 @@ void CNAME(enum CBLAS_ORDER order,
 
   FUNCTION_PROFILE_START();
 
+  if (incx == 1 && n <100) {
+    blasint i;
+    if (uplo==0) {
+      for (i = 0; i < n; i++){
+        if (x[i] != ZERO) {
+          AXPYU_K(i + 1, 0, 0, alpha * x[i], x,     1, a, 1, NULL, 0);
+        }
+        a += i + 1;
+      }
+    } else { 
+      for (i = 0; i < n; i++){
+        if (x[i] != ZERO) {
+          AXPYU_K(n - i, 0, 0, alpha * x[i], x + i, 1, a, 1, NULL, 0);
+        }
+        a += n - i;
+      }
+    }
+    return;
+  }
+
   if (incx < 0 ) x -= (n - 1) * incx;
 
   buffer = (FLOAT *)blas_memory_alloc(1);
diff --git a/interface/spr2.c b/interface/spr2.c
index 73a811c3e..b5aab1767 100644
--- a/interface/spr2.c
+++ b/interface/spr2.c
@@ -168,6 +168,24 @@ void CNAME(enum CBLAS_ORDER order,
 
   if (alpha == ZERO) return;
 
+  if (incx == 1 && incy == 1 && n < 50) {
+    blasint i;
+    if (!uplo) {
+      for (i = 0; i < n; i++){
+        AXPYU_K(i + 1, 0, 0, alpha * x[i], y,     1, a, 1, NULL, 0);
+        AXPYU_K(i + 1, 0, 0, alpha * y[i], x,     1, a, 1, NULL, 0);
+        a += i + 1;
+      }
+    } else {
+      for (i = 0; i < n; i++){
+	AXPYU_K(n - i, 0, 0, alpha * x[i], y + i, 1, a, 1, NULL, 0);
+        AXPYU_K(n - i, 0, 0, alpha * y[i], x + i, 1, a, 1, NULL, 0);
+        a += n - i;
+      }
+    }
+    return;
+  }
+
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
diff --git a/interface/syr.c b/interface/syr.c
index 1374bcc69..ad75264b1 100644
--- a/interface/syr.c
+++ b/interface/syr.c
@@ -168,7 +168,28 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, blasint n, FLOAT alpha,
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
+#if 1
+  if (incx == 1 && n < 100) {
+    BLASLONG i;
 
+    if (uplo == 0) {
+      for (i = 0; i < n; i++){
+        if (x[i] != ZERO) {
+          AXPYU_K(i + 1, 0, 0, alpha * x[i], x,     1, a, 1, NULL, 0);
+        }
+        a += lda;
+      }  
+    } else {
+      for (i = 0; i < n; i++){
+        if (x[i] != ZERO) {
+          AXPYU_K(n - i, 0, 0, alpha * x[i], x + i, 1, a, 1, NULL, 0);
+        }
+        a += 1 + lda;
+      }
+    }
+    return;
+  } 
+#endif
   if (incx < 0 ) x -= (n - 1) * incx;
 
   buffer = (FLOAT *)blas_memory_alloc(1);
diff --git a/interface/syr2.c b/interface/syr2.c
index 08fd47e57..632906d28 100644
--- a/interface/syr2.c
+++ b/interface/syr2.c
@@ -170,6 +170,25 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, blasint n, FLOAT alpha,
 
   IDEBUG_START;
 
+  if (incx == 1 && incy == 1 && n < 100) {
+    blasint i;
+    if (!uplo) {
+      for (i = 0; i < n; i++){
+        AXPYU_K(i + 1, 0, 0, alpha * x[i], y,     1, a, 1, NULL, 0);
+        AXPYU_K(i + 1, 0, 0, alpha * y[i], x,     1, a, 1, NULL, 0);
+        a += lda;
+      }
+    } else {
+      for (i = 0; i < n; i++){
+        AXPYU_K(n - i, 0, 0, alpha * x[i], y + i, 1, a, 1, NULL, 0);
+        AXPYU_K(n - i, 0, 0, alpha * y[i], x + i, 1, a, 1, NULL, 0);
+        a += 1 + lda;
+      }
+    }
+    return;
+  }
+
+	  
   FUNCTION_PROFILE_START();
 
   if (incx < 0 ) x -= (n - 1) * incx;
diff --git a/interface/syrk.c b/interface/syrk.c
index 7699db683..edb113d6c 100644
--- a/interface/syrk.c
+++ b/interface/syrk.c
@@ -354,6 +354,17 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, enum CBLAS_TRANSPOSE Tr
 #endif
 
   args.common = NULL;
+#ifndef COMPLEX
+#ifdef DOUBLE
+  if (args.n < 100)
+#else
+  if (args.n < 200)
+#endif
+#else
+  if (args.n < 65)
+#endif
+  args.nthreads = 1;
+  else
   args.nthreads = num_cpu_avail(3);
 
   if (args.nthreads == 1) {
diff --git a/interface/trsv.c b/interface/trsv.c
index a054d8eeb..6a6e8f8ba 100644
--- a/interface/trsv.c
+++ b/interface/trsv.c
@@ -188,6 +188,12 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 
   if (n == 0) return;
 
+  if (incx == 1 && trans == 0 && n < 50) {
+    buffer = NULL;
+    (trsv[(trans<<2) | (uplo<<1) | unit])(n, a, lda, x, incx, buffer);
+    return;  
+  }
+
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
diff --git a/interface/zsyr.c b/interface/zsyr.c
index 09b1de578..71d4dbf29 100644
--- a/interface/zsyr.c
+++ b/interface/zsyr.c
@@ -172,6 +172,32 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, int n, FLOAT alpha, FLO
 
   if ((alpha_r == ZERO) && (alpha_i == ZERO)) return;
 
+  if (incx == 1 && n < 50) {
+    blasint i;
+    if (!uplo) {
+      for (i = 0; i < n; i++){
+        if ((x[i * 2 + 0] != ZERO) || (x[i * 2 + 1] != ZERO)) {
+          AXPYU_K(i + 1, 0, 0,
+              alpha_r * x[i * 2 + 0] - alpha_i * x[i * 2 + 1],
+              alpha_i * x[i * 2 + 0] + alpha_r * x[i * 2 + 1],
+              x,         1, a, 1, NULL, 0);
+        }
+        a += lda;
+      }
+    } else {
+      for (i = 0; i < n; i++){
+        if ((x[i * 2 + 0] != ZERO) || (x[i * 2 + 1] != ZERO)) {
+          AXPYU_K(n - i, 0, 0,
+              alpha_r * x[i * 2 + 0] - alpha_i * x[i * 2 + 1],
+              alpha_i * x[i * 2 + 0] + alpha_r * x[i * 2 + 1],
+              x + i * 2, 1, a, 1, NULL, 0);
+        }
+        a += 2 + lda;
+      }
+    }
+    return;
+  }
+
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
diff --git a/interface/ztrsv.c b/interface/ztrsv.c
index cbb7bba13..cf750b0b0 100644
--- a/interface/ztrsv.c
+++ b/interface/ztrsv.c
@@ -199,6 +199,12 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 
   if (n == 0) return;
 
+  if (incx == 1 && trans == 0 && n < 50) {
+    buffer = NULL;
+    (trsv[(trans<<2) | (uplo<<1) | unit])(n, a, lda, x, incx, buffer);
+    return;
+  }
+
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
diff --git a/kernel/Makefile.L3 b/kernel/Makefile.L3
index d8d739965..2d9e3ec36 100644
--- a/kernel/Makefile.L3
+++ b/kernel/Makefile.L3
@@ -818,6 +818,8 @@ ifeq ($(OS), AIX)
 	m4 zgemm_kernel_n.s > zgemm_kernel_n_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN zgemm_kernel_n_nomacros.s -o $@
 	rm zgemm_kernel_n.s zgemm_kernel_n_nomacros.s
+else ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
 endif
@@ -828,6 +830,8 @@ ifeq ($(OS), AIX)
 	m4 zgemm_kernel_l.s > zgemm_kernel_l_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCN zgemm_kernel_l_nomacros.s -o $@
 	rm zgemm_kernel_l.s zgemm_kernel_l_nomacros.s
+else ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DDOUBLE -DCOMPLEX -DCN $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $@
 endif
@@ -838,6 +842,8 @@ ifeq ($(OS), AIX)
 	m4 zgemm_kernel_r.s > zgemm_kernel_r_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNC zgemm_kernel_r_nomacros.s -o $@
 	rm zgemm_kernel_r.s zgemm_kernel_r_nomacros.s
+else ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DDOUBLE -DCOMPLEX -DNC $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNC $< -o $@
 endif
@@ -848,6 +854,8 @@ ifeq ($(OS), AIX)
 	m4 zgemm_kernel_b.s > zgemm_kernel_b_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCC zgemm_kernel_b_nomacros.s -o $@
 	rm zgemm_kernel_b.s zgemm_kernel_b_nomacros.s
+else ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DDOUBLE -DCOMPLEX -DCC $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCC $< -o $@
 endif
@@ -1044,6 +1052,8 @@ ifeq ($(OS), AIX)
 	m4 ztrmm_kernel_ln.s > ztrmm_kernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_ln_nomacros.s -o $@
 	rm ztrmm_kernel_ln.s ztrmm_kernel_ln_nomacros.s
+else ifeq ($(CORE), SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 endif
@@ -1054,6 +1064,8 @@ ifeq ($(OS), AIX)
 	m4 ztrmm_kernel_lt.s > ztrmm_kernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_lt_nomacros.s -o $@
 	rm ztrmm_kernel_lt.s ztrmm_kernel_lt_nomacros.s
+else ifeq ($(CORE), SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 endif
@@ -1064,6 +1076,8 @@ ifeq ($(OS), AIX)
 	m4 ztrmm_kernel_lr.s > ztrmm_kernel_lr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN ztrmm_kernel_lr_nomacros.s -o $@
 	rm ztrmm_kernel_lr.s ztrmm_kernel_lr_nomacros.s
+else ifeq ($(CORE), SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 endif
@@ -1074,6 +1088,8 @@ ifeq ($(OS), AIX)
 	m4 ztrmm_kernel_lc.s >ztrmm_kernel_lc_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ztrmm_kernel_lc_nomacros.s -o $@
 	rm ztrmm_kernel_lc.s ztrmm_kernel_lc_nomacros.s 
+else ifeq ($(CORE), SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 endif
@@ -1084,6 +1100,8 @@ ifeq ($(OS), AIX)
 	m4 ztrmm_kernel_rn.s > ztrmm_kernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_rn_nomacros.s -o $@
 	rm ztrmm_kernel_rn.s ztrmm_kernel_rn_nomacros.s
+else ifeq ($(CORE), SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 endif
@@ -1094,6 +1112,8 @@ ifeq ($(OS), AIX)
 	m4 ztrmm_kernel_rt.s > ztrmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_rt_nomacros.s -o $@
 	rm ztrmm_kernel_rt.s ztrmm_kernel_rt_nomacros.s
+else ifeq ($(CORE), SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 endif
@@ -1104,6 +1124,8 @@ ifeq ($(OS), AIX)
 	m4 ztrmm_kernel_rr.s > ztrmm_kernel_rr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ztrmm_kernel_rr_nomacros.s -o $@
 	rm ztrmm_kernel_rr.s ztrmm_kernel_rr_nomacros.s
+else ifeq ($(CORE), SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 endif
@@ -1114,6 +1136,8 @@ ifeq ($(OS), AIX)
 	m4 ztrmm_kernel_rc.s > ztrmm_kernel_rc_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ztrmm_kernel_rc_nomacros.s -o $@
 	rm ztrmm_kernel_rc.s ztrmm_kernel_rc_nomacros.s
+else ifeq ($(CORE), SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 endif
@@ -1187,29 +1211,55 @@ $(KDIR)ctrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)ztrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
+ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+endif
 
 $(KDIR)ztrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
+ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
-
+endif
 $(KDIR)ztrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
+ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
-
+endif
 $(KDIR)ztrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
+ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
-
+endif
 $(KDIR)ztrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
+ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
-
+endif
 $(KDIR)ztrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
+ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
-
+endif
 $(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
+ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
-
+endif
 $(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
+ifeq ($(CORE),SANDYBRIDGE)
+	$(CC) $(filter-out -mavx,$(CFLAGS)) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+else
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 endif
+endif
 
 
 
diff --git a/kernel/arm64/KERNEL.CORTEXA55 b/kernel/arm64/KERNEL.CORTEXA55
new file mode 100644
index 000000000..db322dd0d
--- /dev/null
+++ b/kernel/arm64/KERNEL.CORTEXA55
@@ -0,0 +1,196 @@
+SAMINKERNEL  = ../arm/amin.c
+DAMINKERNEL  = ../arm/amin.c
+CAMINKERNEL  = ../arm/zamin.c
+ZAMINKERNEL  = ../arm/zamin.c
+
+SMAXKERNEL   = ../arm/max.c
+DMAXKERNEL   = ../arm/max.c
+
+SMINKERNEL   = ../arm/min.c
+DMINKERNEL   = ../arm/min.c
+
+ISAMINKERNEL = ../arm/iamin.c
+IDAMINKERNEL = ../arm/iamin.c
+ICAMINKERNEL = ../arm/izamin.c
+IZAMINKERNEL = ../arm/izamin.c
+
+ISMAXKERNEL  = ../arm/imax.c
+IDMAXKERNEL  = ../arm/imax.c
+
+ISMINKERNEL  = ../arm/imin.c
+IDMINKERNEL  = ../arm/imin.c
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+SAMAXKERNEL  = amax.S
+DAMAXKERNEL  = amax.S
+CAMAXKERNEL  = zamax.S
+ZAMAXKERNEL  = zamax.S
+
+SAXPYKERNEL  = axpy.S
+DAXPYKERNEL  = axpy.S
+CAXPYKERNEL  = zaxpy.S
+ZAXPYKERNEL  = zaxpy.S
+
+SROTKERNEL   = rot.S
+DROTKERNEL   = rot.S
+CROTKERNEL   = zrot.S
+ZROTKERNEL   = zrot.S
+
+SSCALKERNEL  = scal.S
+DSCALKERNEL  = scal.S
+CSCALKERNEL  = zscal.S
+ZSCALKERNEL  = zscal.S
+
+SGEMVNKERNEL = gemv_n.S
+DGEMVNKERNEL = gemv_n.S
+CGEMVNKERNEL = zgemv_n.S
+ZGEMVNKERNEL = zgemv_n.S
+
+SGEMVTKERNEL = gemv_t.S
+DGEMVTKERNEL = gemv_t.S
+CGEMVTKERNEL = zgemv_t.S
+ZGEMVTKERNEL = zgemv_t.S
+
+
+SASUMKERNEL    = asum.S
+DASUMKERNEL    = asum.S
+CASUMKERNEL    = casum.S
+ZASUMKERNEL    = zasum.S
+
+SCOPYKERNEL    = copy.S
+DCOPYKERNEL    = copy.S
+CCOPYKERNEL    = copy.S
+ZCOPYKERNEL    = copy.S
+
+SSWAPKERNEL    = swap.S
+DSWAPKERNEL    = swap.S
+CSWAPKERNEL    = swap.S
+ZSWAPKERNEL    = swap.S
+
+ISAMAXKERNEL   = iamax.S
+IDAMAXKERNEL   = iamax.S
+ICAMAXKERNEL   = izamax.S
+IZAMAXKERNEL   = izamax.S
+
+SNRM2KERNEL    = nrm2.S
+DNRM2KERNEL    = nrm2.S
+CNRM2KERNEL    = znrm2.S
+ZNRM2KERNEL    = znrm2.S
+
+ifneq ($(C_COMPILER), PGI)
+SDOTKERNEL   = ../generic/dot.c
+else
+SDOTKERNEL   = dot.S
+endif
+DDOTKERNEL   = dot.S
+ifneq ($(C_COMPILER), PGI)
+CDOTKERNEL   = zdot.S
+ZDOTKERNEL   = zdot.S
+else
+CDOTKERNEL = ../arm/zdot.c
+ZDOTKERNEL = ../arm/zdot.c
+endif
+DSDOTKERNEL  = dot.S
+
+DGEMM_BETA     = dgemm_beta.S
+SGEMM_BETA     = sgemm_beta.S
+
+ifeq ($(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N), 8x8)
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_cortexa53.S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_cortexa53.S
+else
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+endif
+ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
+ifeq ($(SGEMM_UNROLL_M), 16)
+SGEMMITCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_M).S
+else
+SGEMMITCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
+endif
+ifeq ($(SGEMM_UNROLL_M), 4)
+SGEMMINCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_M).S
+else
+SGEMMINCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
+endif
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+SGEMMOTCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_N).S
+SGEMMONCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_N).S
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+DGEMMKERNEL    =  dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+DTRMMKERNEL    =  dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+
+ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
+
+ifeq ($(DGEMM_UNROLL_M), 8)
+DGEMMINCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_M).S
+DGEMMITCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_M).S
+else
+DGEMMINCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
+endif
+
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifeq ($(DGEMM_UNROLL_N), 4)
+DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
+DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
+else
+DGEMMONCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
+endif
+
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CGEMMKERNEL    =  cgemm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+CTRMMKERNEL    =  ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
diff --git a/kernel/arm64/dznrm2_thunderx2t99.c b/kernel/arm64/dznrm2_thunderx2t99.c
index b021a2832..fba2fe8ce 100644
--- a/kernel/arm64/dznrm2_thunderx2t99.c
+++ b/kernel/arm64/dznrm2_thunderx2t99.c
@@ -321,7 +321,7 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
 	: "cc",
 	  "memory",
 	  "x0", "x1", "x2", "x3", "x4", "x5", "x6",
-	  "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d8"
+	  "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d8", REGINF
 	);
 
 }
diff --git a/kernel/power/KERNEL.POWER5 b/kernel/power/KERNEL.POWER5
index fbef79e59..bea7b17c8 100644
--- a/kernel/power/KERNEL.POWER5
+++ b/kernel/power/KERNEL.POWER5
@@ -54,3 +54,8 @@ ZTRSMKERNEL_LN	=  ztrsm_kernel_LN.S
 ZTRSMKERNEL_LT	=  ztrsm_kernel_LT.S
 ZTRSMKERNEL_RN	=  ztrsm_kernel_LT.S
 ZTRSMKERNEL_RT	=  ztrsm_kernel_RT.S
+
+CROTKERNEL = ../arm/zrot.c
+ZROTKERNEL = ../arm/zrot.c
+SGEMVNKERNEL = ../arm/gemv_n.c
+SGEMVTKERNEL = ../arm/gemv_t.c
diff --git a/kernel/power/KERNEL.PPC440 b/kernel/power/KERNEL.PPC440
index 677af5f21..fd9a8c780 100644
--- a/kernel/power/KERNEL.PPC440
+++ b/kernel/power/KERNEL.PPC440
@@ -16,11 +16,11 @@ ZASUMKERNEL	= zasum_ppc440.S
 SAXPYKERNEL	=  axpy_ppc440.S
 DAXPYKERNEL	=  axpy_ppc440.S
 ifneq ($(__BYTE_ORDER__),__ORDER_BIG_ENDIAN__)
-CAXPYKERNEL	= ../arm/zaxpy.c
-ZAXPYKERNEL	= ../arm/zaxpy.c
-else
 CAXPYKERNEL	= zaxpy_ppc440.S
 ZAXPYKERNEL	= zaxpy_ppc440.S
+else
+CAXPYKERNEL	= ../arm/zaxpy.c
+ZAXPYKERNEL	= ../arm/zaxpy.c
 endif
 
 SDOTKERNEL	=  dot_ppc440.S
diff --git a/kernel/power/KERNEL.PPCG4 b/kernel/power/KERNEL.PPCG4
index 54660b54d..1bdd3119e 100644
--- a/kernel/power/KERNEL.PPCG4
+++ b/kernel/power/KERNEL.PPCG4
@@ -15,8 +15,13 @@ ZASUMKERNEL	= zasum_ppc440.S
 
 SAXPYKERNEL	=  axpy_ppc440.S
 DAXPYKERNEL	=  axpy_ppc440.S
+ifneq ($(__BYTE_ORDER__),__ORDER_BIG_ENDIAN__)
 CAXPYKERNEL	= zaxpy_ppc440.S
 ZAXPYKERNEL	= zaxpy_ppc440.S
+else
+CAXPYKERNEL    = ../arm/zaxpy.c
+ZAXPYKERNEL    = ../arm/zaxpy.c
+endif
 
 SDOTKERNEL	=  dot_ppc440.S
 DDOTKERNEL	=  dot_ppc440.S
diff --git a/kernel/power/gemv_n.S b/kernel/power/gemv_n.S
index abc61b62e..9c6f87639 100644
--- a/kernel/power/gemv_n.S
+++ b/kernel/power/gemv_n.S
@@ -159,6 +159,11 @@
 #define PREFETCHSIZE_C  16
 #endif
 
+#ifdef POWER3
+#define PREFETCHSIZE_A  16
+#define PREFETCHSIZE_C  16
+#endif
+
 #ifdef POWER4
 #define PREFETCHSIZE_A  16
 #define PREFETCHSIZE_C  16
diff --git a/kernel/power/gemv_t.S b/kernel/power/gemv_t.S
index 25a4dd01b..accdad702 100644
--- a/kernel/power/gemv_t.S
+++ b/kernel/power/gemv_t.S
@@ -124,6 +124,11 @@
 #define PREFETCHSIZE_C  16
 #endif
 
+#ifdef POWER3
+#define PREFETCHSIZE_A  16
+#define PREFETCHSIZE_C  16
+#endif
+
 #ifdef POWER4
 #define PREFETCHSIZE_A  48
 #define PREFETCHSIZE_C  16
diff --git a/kernel/power/sbgemm_kernel_power10.c b/kernel/power/sbgemm_kernel_power10.c
index d15586703..134929ec1 100644
--- a/kernel/power/sbgemm_kernel_power10.c
+++ b/kernel/power/sbgemm_kernel_power10.c
@@ -49,17 +49,11 @@ typedef __vector unsigned char  vec_t;
 typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
 typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
 
-vector char mask =
-  { 0x0, 0x1, 0x8, 0x9, 0x2, 0x3, 0xa, 0xb, 0x4, 0x5, 0xc, 0xd, 0x6, 0x7, 0xe,
-  0xf
-};
-
 /* 
  * BFLOAT16 xvbf16ger2pp instruction needs 4×2 matrix of
  * bfloat16 floating-point values as input. Hence this
  * merging is needed on A and B matrices. 
  */
-#define MERGE_ROW(x) vec_perm(x, x, mask)
 #define MERGE_HIGH(x, y) (vec_t) vec_mergeh ((vector short)x, (vector short)y)
 #define MERGE_LOW(x, y) (vec_t) vec_mergel ((vector short)x, (vector short)y)
 
@@ -104,6 +98,30 @@ vector char mask =
 	  rowC = (v2sf_t *) &CO[7* ldc+J]; \
           rowC[0] += result[6] * alpha;
 
+ #define  SAVE4x2_ACC_SCALAR(ACC) {                             \
+           __builtin_mma_disassemble_acc ((void *)result, ACC); \
+           res[0] = result[0] * alpha;                          \
+           res[1] = result[1] * alpha;                          \
+           res[2] = result[2] * alpha;                          \
+           res[3] = result[3] * alpha;                          \
+           CO[0 * ldc] += res[0][0];                            \
+           CO[1 * ldc] += res[1][0];                            \
+           CO[2 * ldc] += res[2][0];                            \
+           CO[3 * ldc] += res[3][0];                            \
+ }
+
+ #define  SAVE4x2_ACC1_SCALAR(ACC) {                            \
+           __builtin_mma_disassemble_acc ((void *)result, ACC); \
+           res[0] = result[0] * alpha;                          \
+           res[1] = result[1] * alpha;                          \
+           res[2] = result[2] * alpha;                          \
+           res[3] = result[3] * alpha;                          \
+           CO[4 * ldc] += res[0][0];                            \
+           CO[5 * ldc] += res[1][0];                            \
+           CO[6 * ldc] += res[2][0];                            \
+           CO[7 * ldc] += res[3][0];                            \
+}
+
 #define MMA __builtin_mma_xvbf16ger2pp
 
 #define  SAVE2x4_ACC(ACC, J)  \
@@ -179,8 +197,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 4;
 	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
 	      vec_t *rowB = (vec_t *) & (BO[l]);
-	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
-	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], vzero);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], vzero);
 	      vec_t rowA_h = MERGE_HIGH (rowA[0], vzero);
 	      vec_t rowA_l = MERGE_LOW (rowA[0], vzero);
 	      vec_t rowA2_h = MERGE_HIGH (rowA[1], vzero);
@@ -231,8 +249,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 4;
 	      vec_t *rowA = (vec_t *) & (AO[l]);
 	      vec_t *rowB = (vec_t *) & (BO[l]);
-	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
-	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], vzero);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], vzero);
 	      vec_t rowA_h = MERGE_HIGH (rowA[0], vzero);
 	      vec_t rowA_l = MERGE_LOW (rowA[0], vzero);
 	      MMA (&acc0, rowB_h, rowA_h);
@@ -271,8 +289,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	      vector short rowA =
 		{ AO[l + 0], 0, AO[l + 1], 0, AO[l + 2], 0, AO[l + 3], 0 };
 	      vec_t *rowB = (vec_t *) & (BO[l << 1]);
-	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
-	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], vzero), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], vzero), (vec_t) rowA);
 	    }
 	  SAVE_ACC (&acc0, 0);
 	  SAVE_ACC1 (&acc1, 0);
@@ -306,8 +324,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 2;
 	      vector short rowA = { AO[l + 0], 0, AO[l + 1], 0, 0, 0, 0, 0 };
 	      vec_t *rowB = (vec_t *) & (BO[(l << 2)]);
-	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
-	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], vzero), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], vzero), (vec_t) rowA);
 	    }
 	  SAVE4x2_ACC (&acc0, 0);
 	  SAVE4x2_ACC1 (&acc1, 0);
@@ -319,7 +337,7 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	{
 	  IFLOAT *BO = B;
 	  v2sf_t *rowC;
-	  v2sf_t result[8];
+	  v4sf_t result[4], res[4];
 	  __vector_quad acc0, acc1;
 	  __builtin_mma_xxsetaccz (&acc0);
 	  __builtin_mma_xxsetaccz (&acc1);
@@ -338,11 +356,11 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 1;
 	      vector short rowA = { AO[l], 0, 0, 0, 0, 0, 0, 0 };
 	      vec_t *rowB = (vec_t *) & (BO[(l << 3)]);
-	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
-	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], vzero), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], vzero), (vec_t) rowA);
 	    }
-	  SAVE4x2_ACC (&acc0, 0);
-	  SAVE4x2_ACC1 (&acc1, 0);
+	  SAVE4x2_ACC_SCALAR (&acc0);
+	  SAVE4x2_ACC1_SCALAR (&acc1);
 	  CO += 1;
 	  AO += k;
 	  BO += (k << 3);
@@ -387,16 +405,16 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 3;
 	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
 	      vec_t *rowA1 = (vec_t *) & (A1[(l << 2)]);
-	      vec_t *rowB = (vec_t *) & (BO[l]);
-	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
-	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
-	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
-	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], vzero));
-	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], vzero));
-	      MMA (&acc4, rowB_mrg, MERGE_HIGH (rowA1[0], vzero));
-	      MMA (&acc5, rowB_mrg, MERGE_LOW (rowA1[0], vzero));
-	      MMA (&acc6, rowB_mrg, MERGE_HIGH (rowA1[1], vzero));
-	      MMA (&acc7, rowB_mrg, MERGE_LOW (rowA1[1], vzero));
+	      vector short rowB_mrg =
+		{ BO[l], 0, BO[l + 1], 0, BO[l + 2], 0, BO[l + 3], 0 };
+	      MMA (&acc0, (vec_t)rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, (vec_t)rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, (vec_t)rowB_mrg, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, (vec_t)rowB_mrg, MERGE_LOW (rowA[1], vzero));
+	      MMA (&acc4, (vec_t)rowB_mrg, MERGE_HIGH (rowA1[0], vzero));
+	      MMA (&acc5, (vec_t)rowB_mrg, MERGE_LOW (rowA1[0], vzero));
+	      MMA (&acc6, (vec_t)rowB_mrg, MERGE_HIGH (rowA1[1], vzero));
+	      MMA (&acc7, (vec_t)rowB_mrg, MERGE_LOW (rowA1[1], vzero));
 	    }
 
 	  SAVE_ACC (&acc0, 0);
@@ -436,12 +454,12 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	      if (k > 1)
 		l = (k / 2) << 3;
 	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
-	      vec_t *rowB = (vec_t *) & (BO[l]);
-	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
-	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
-	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
-	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], vzero));
-	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], vzero));
+	      vector short rowB_mrg =
+		{ BO[l], 0, BO[l + 1], 0, BO[l + 2], 0, BO[l + 3], 0 };
+	      MMA (&acc0, (vec_t)rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, (vec_t)rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, (vec_t)rowB_mrg, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, (vec_t)rowB_mrg, MERGE_LOW (rowA[1], vzero));
 	    }
 
 	  SAVE_ACC (&acc0, 0);
@@ -475,9 +493,10 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 3;
 	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
 	      vec_t *rowB = (vec_t *) & (BO[l]);
-	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
-	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
-	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	      vector short rowB_mrg =
+		{ BO[l], 0, BO[l + 1], 0, BO[l + 2], 0, BO[l + 3], 0 };
+	      MMA (&acc0, (vec_t)rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, (vec_t)rowB_mrg, MERGE_LOW (rowA[0], vzero));
 	    }
 	  SAVE_ACC (&acc0, 0);
 	  SAVE_ACC (&acc1, 4);
@@ -505,8 +524,9 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 3;
 	      vector short rowA =
 		{ AO[l], 0, AO[l + 1], 0, AO[l + 2], 0, AO[l + 3], 0 };
-	      vec_t *rowB = (vec_t *) & (BO[l]);
-	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	      vector short rowB_mrg =
+		{ BO[l], 0, BO[l + 1], 0, BO[l + 2], 0, BO[l + 3], 0 };
+	      MMA (&acc0, (vec_t)(rowB_mrg), (vec_t) rowA);
 	    }
 	  SAVE_ACC (&acc0, 0);
 	  CO += 4;
@@ -536,8 +556,11 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	      if (k > 1)
 		l = (k / 2) << 2;
 	      vector short rowA = { AO[l], 0, AO[l + 1], 0, 0, 0, 0, 0 };
-	      vec_t *rowB = (vec_t *) & (BO[l << 1]);
-	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	      vector short rowB_mrg =
+		{ BO[(l<<1)], 0, BO[(l<<1) + 1], 0, BO[(l<<1) + 2], 0,
+		BO[(l<<1) + 3], 0
+	      };
+	      MMA (&acc0, (vec_t)(rowB_mrg), (vec_t) rowA);
 	    }
 	  SAVE4x2_ACC (&acc0, 0);
 	  CO += 2;
@@ -548,7 +571,7 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	{
 	  IFLOAT *BO = B;
 	  v2sf_t *rowC;
-	  v2sf_t result[8];
+	  v4sf_t result[4], res[4];
 	  __vector_quad acc0;
 	  BLASLONG l = 0;
 	  __builtin_mma_xxsetaccz (&acc0);
@@ -566,10 +589,13 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	      if (k > 1)
 		l = (k / 2) << 1;
 	      vector short rowA = { AO[l], 0, 0, 0, 0, 0, 0, 0 };
-	      vec_t *rowB = (vec_t *) & (BO[l << 2]);
-	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	      vector short rowB_mrg =
+		{ BO[(l<<2) + 0], 0, BO[(l<<2) + 1], 0, BO[(l <<2) + 2], 0,
+		BO[(l<<2) + 3], 0
+	      };
+	      MMA (&acc0, (vec_t)(rowB_mrg), (vec_t) rowA);
 	    }
-	  SAVE4x2_ACC (&acc0, 0);
+	  SAVE4x2_ACC_SCALAR (&acc0);
 	  AO += k;
 	  BO += (k << 2);
 	  CO += 1;
@@ -620,14 +646,14 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
 	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
 	      vec_t *rowA1 = (vec_t *) & (A1[l << 3]);
-	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
-	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
-	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
-	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
-	      MMA (&acc4, (vec_t) rowB, MERGE_HIGH (rowA1[0], rowA1[2]));
-	      MMA (&acc5, (vec_t) rowB, MERGE_LOW (rowA1[0], rowA1[2]));
-	      MMA (&acc6, (vec_t) rowB, MERGE_HIGH (rowA1[1], rowA1[3]));
-	      MMA (&acc7, (vec_t) rowB, MERGE_LOW (rowA1[1], rowA1[3]));
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], vzero));
+	      MMA (&acc4, (vec_t) rowB, MERGE_HIGH (rowA1[0], vzero));
+	      MMA (&acc5, (vec_t) rowB, MERGE_LOW (rowA1[0], vzero));
+	      MMA (&acc6, (vec_t) rowB, MERGE_HIGH (rowA1[1], vzero));
+	      MMA (&acc7, (vec_t) rowB, MERGE_LOW (rowA1[1], vzero));
 	    }
 	  SAVE2x4_ACC (&acc0, 0);
 	  SAVE2x4_ACC (&acc1, 4);
@@ -669,10 +695,10 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 2;
 	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
 	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
-	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
-	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
-	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
-	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], vzero ));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], vzero));
 	    }
 	  SAVE2x4_ACC (&acc0, 0);
 	  SAVE2x4_ACC (&acc1, 4);
@@ -708,8 +734,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 2;
 	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
 	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
-	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[1]));
-	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[1]));
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], vzero));
 	    }
 	  SAVE2x4_ACC (&acc0, 0);
 	  SAVE2x4_ACC (&acc1, 4);
@@ -740,8 +766,10 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	      if (k > 1)
 		l = (k / 2) << 2;
 	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
-	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
-	      MMA (&acc0, (vec_t) rowB, MERGE_ROW (rowA[0]));
+	      vector short rowA =
+	        { AO[(l << 1)], 0, AO[(l << 1) + 1] , 0 , AO[(l<<1) + 2],
+	        0, AO[(l << 1) + 3], 0 };
+	      MMA (&acc0, (vec_t) rowB, (vec_t)(rowA));
 	    }
 	  SAVE2x4_ACC (&acc0, 0);
 	  CO += 4;
@@ -829,10 +857,10 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 1;
 	      vector short rowB = { BO[l], 0, 0, 0, 0, 0, 0, 0 };
 	      vec_t *rowA = (vec_t *) & (AO[(l << 4)]);
-	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
-	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
-	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
-	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], vzero));
 	    }
 	  rowC = (v4sf_t *) &CO[0];
 	  __builtin_mma_disassemble_acc ((void *)result, &acc0);
@@ -871,8 +899,8 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 		l = (k / 2) << 1;
 	      vector short rowB = { BO[l], 0, 0, 0, 0, 0, 0, 0 };
 	      vec_t *rowA = (vec_t *) & (AO[(l << 3)]);
-	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[1]));
-	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[1]));
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], vzero));
 	    }
 	  rowC = (v4sf_t *) &CO[0];
 	  __builtin_mma_disassemble_acc ((void *)result, &acc0);
@@ -904,8 +932,10 @@ CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
 	      if (k > 1)
 		l = (k / 2) << 1;
 	      vector short rowB = { BO[l], 0, 0, 0, 0, 0, 0, 0 };
-	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
-	      MMA (&acc0, (vec_t) rowB, MERGE_ROW (rowA[0]));
+	      vector short rowA =
+	        { AO[(l << 2)], 0, AO[(l << 2) + 1] , 0 ,
+		AO[(l << 2) + 2], 0, AO[(l << 2) + 3], 0 };
+	      MMA (&acc0, (vec_t) rowB, (vec_t)(rowA));
 	    }
 	  rowC = (v4sf_t *) &CO[0];
 	  __builtin_mma_disassemble_acc ((void *)result, &acc0);
diff --git a/kernel/power/zgemv_n.S b/kernel/power/zgemv_n.S
index 708f1318d..48f49f97b 100644
--- a/kernel/power/zgemv_n.S
+++ b/kernel/power/zgemv_n.S
@@ -155,6 +155,11 @@
 #define PREFETCHSIZE_C  16
 #endif
 
+#ifdef POWER3
+#define PREFETCHSIZE_A  34
+#define PREFETCHSIZE_C  16
+#endif
+
 #ifdef POWER4
 #define PREFETCHSIZE_A  34
 #define PREFETCHSIZE_C  16
diff --git a/kernel/power/zgemv_t.S b/kernel/power/zgemv_t.S
index d82fab16a..314cf5e6e 100644
--- a/kernel/power/zgemv_t.S
+++ b/kernel/power/zgemv_t.S
@@ -129,6 +129,11 @@
 #define PREFETCHSIZE_C  16
 #endif
 
+#ifdef POWER3
+#define PREFETCHSIZE_A  34
+#define PREFETCHSIZE_C  16
+#endif
+
 #ifdef POWER4
 #define PREFETCHSIZE_A  34
 #define PREFETCHSIZE_C  16
diff --git a/kernel/riscv64/KERNEL.RISCV64_GENERIC b/kernel/riscv64/KERNEL.RISCV64_GENERIC
index ea6a8cf21..61a8a2b91 100644
--- a/kernel/riscv64/KERNEL.RISCV64_GENERIC
+++ b/kernel/riscv64/KERNEL.RISCV64_GENERIC
@@ -54,6 +54,7 @@ SDOTKERNEL   = ../riscv64/dot.c
 DDOTKERNEL   = ../riscv64/dot.c
 CDOTKERNEL   = ../riscv64/zdot.c
 ZDOTKERNEL   = ../riscv64/zdot.c
+DSDOTKERNEL  = ../generic/dot.c
 
 SNRM2KERNEL  = ../riscv64/nrm2.c
 DNRM2KERNEL  = ../riscv64/nrm2.c
diff --git a/kernel/riscv64/Makefile b/kernel/riscv64/Makefile
new file mode 100644
index 000000000..520349bd6
--- /dev/null
+++ b/kernel/riscv64/Makefile
@@ -0,0 +1 @@
+clean ::
diff --git a/kernel/riscv64/amax_vector.c b/kernel/riscv64/amax_vector.c
index b6aec131e..5312f9ef0 100644
--- a/kernel/riscv64/amax_vector.c
+++ b/kernel/riscv64/amax_vector.c
@@ -29,29 +29,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <math.h>
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFMAXVV_FLOAT vfmaxvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFMAXVV_FLOAT vfmax_vv_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFMAXVV_FLOAT vfmaxvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFMAXVV_FLOAT vfmax_vv_f64m8
 #endif
 
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@@ -62,19 +66,25 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	if (n <= 0 || inc_x <= 0) return(maxf);
         unsigned int gvl = 0;
         FLOAT_V_T v0, v1, v_max;
+        FLOAT_V_T_M1 v_res, v_zero;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_zero = VFMVVF_FLOAT_M1(0, gvl);
 
         MASK_T mask0, mask1;
         FLOAT zero = 0.0;
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl <= n/2){
                         v_max = VFMVVF_FLOAT(0, gvl);
                         for(i=0,j=0; i<n/(gvl*2); i++){
                                 v0 = VLEV_FLOAT(&x[j], gvl);
+                                v1 = VLEV_FLOAT(&x[j+gvl], gvl);
                                 mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
                                 //v0 = VFRSUBVF_MASK_FLOAT(v0, 0, mask0, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -83,6 +93,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -98,6 +109,7 @@ asm volatile(
                                 //v1 = VFRSUBVF_MASK_FLOAT(v1, 0, mask1, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -106,6 +118,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -117,17 +130,17 @@ asm volatile(
                                 v_max = VFMAXVV_FLOAT(v_max, v1, gvl);
                                 j += gvl*2;
                         }
-                        v0 = VFMVVF_FLOAT(0, gvl);
-                        v0 = VFREDMAXVS_FLOAT(v_max, v0, gvl);
-                        maxf = v0[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_zero, gvl);
+                        maxf = v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
                         //v0 = VFRSUBVF_MASK_FLOAT(v0, 0, mask0, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -136,6 +149,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -144,14 +158,13 @@ asm volatile(
         :"v0");
 #endif
 
-                        v1 = VFMVVF_FLOAT(0, gvl);
-                        v0 = VFREDMAXVS_FLOAT(v0, v1, gvl);
-                        if(v0[0] > maxf)
-                                maxf = v0[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v0, v_zero, gvl);
+                        if(v_res[0] > maxf)
+                                maxf = v_res[0];
                         j += gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * sizeof(FLOAT);
                 if(gvl <= n/2){
                         BLASLONG inc_xv = inc_x * gvl;
@@ -162,6 +175,7 @@ asm volatile(
                                 //v0 = VFRSUBVF_MASK_FLOAT(v0, 0, mask0, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -170,6 +184,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -185,6 +200,7 @@ asm volatile(
                                 //v1 = VFRSUBVF_MASK_FLOAT(v1, 0, mask1, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -193,6 +209,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -205,17 +222,17 @@ asm volatile(
                                 j += gvl*2;
                                 ix += inc_xv*2;
                         }
-                        v0 = VFMVVF_FLOAT(0, gvl);
-                        v0 = VFREDMAXVS_FLOAT(v_max, v0, gvl);
-                        maxf = v0[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_zero, gvl);
+                        maxf = v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
                         //v0 = VFRSUBVF_MASK_FLOAT(v0, 0, mask0, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -224,6 +241,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -232,10 +250,9 @@ asm volatile(
         :"v0");
 #endif
 
-                        v1 = VFMVVF_FLOAT(0, gvl);
-                        v0 = VFREDMAXVS_FLOAT(v0, v1, gvl);
-                        if(v0[0] > maxf)
-                                maxf = v0[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v0, v_zero, gvl);
+                        if(v_res[0] > maxf)
+                                maxf = v_res[0];
                         j += gvl;
                 }
         }
diff --git a/kernel/riscv64/amin_vector.c b/kernel/riscv64/amin_vector.c
index 53243ad56..ae2867ef8 100644
--- a/kernel/riscv64/amin_vector.c
+++ b/kernel/riscv64/amin_vector.c
@@ -30,29 +30,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <float.h>
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMINVS_FLOAT vfredminvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFMINVV_FLOAT vfminvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFMINVV_FLOAT vfmin_vv_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMINVS_FLOAT vfredminvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFMINVV_FLOAT vfminvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFMINVV_FLOAT vfmin_vv_f64m8
 #endif
 
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@@ -62,11 +66,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	FLOAT minf=FLT_MAX;
         unsigned int gvl = 0;
         FLOAT_V_T v0, v1, v_min;
+        FLOAT_V_T_M1 v_res, v_max;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_max = VFMVVF_FLOAT_M1(FLT_MAX, gvl);
 
         MASK_T mask0, mask1;
-	FLOAT zero = 0.0;
+	    FLOAT zero = 0.0;
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl <= n/2){
                         v_min = VFMVVF_FLOAT(FLT_MAX, gvl);
                         for(i=0,j=0; i<n/(gvl*2); i++){
@@ -75,6 +83,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 //v0 = VFRSUBVF_MASK_FLOAT(v0, 0, mask0, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -83,6 +92,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -97,6 +107,7 @@ asm volatile(
                                 //v1 = VFRSUBVF_MASK_FLOAT(v1, 0, mask1, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -105,6 +116,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -116,17 +128,17 @@ asm volatile(
                                 v_min = VFMINVV_FLOAT(v_min, v1, gvl);
                                 j += gvl*2;
                         }
-                        v1 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        v0 = VFREDMINVS_FLOAT(v_min, v1, gvl);
-                        minf = v0[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                        minf = v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
                         //v0 = VFRSUBVF_MASK_FLOAT(v0, 0, mask0, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -135,6 +147,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -142,14 +155,13 @@ asm volatile(
         :"v"(mask0), "f"(zero), "r"(gvl)
         :"v0");
 #endif
-                        v1 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        v0 = VFREDMINVS_FLOAT(v0, v1, gvl);
-                        if(v0[0] < minf)
-                                minf = v0[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v0, v_max, gvl);
+                        if(v_res[0] < minf)
+                                minf = v_res[0];
                         j += gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * sizeof(FLOAT);
                 if(gvl <= n/2){
                         BLASLONG idx = 0, inc_xv = inc_x * gvl;
@@ -160,6 +172,7 @@ asm volatile(
                                 //v0 = VFRSUBVF_MASK_FLOAT(v0, 0, mask0, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -168,6 +181,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -182,6 +196,7 @@ asm volatile(
                                 //v1 = VFRSUBVF_MASK_FLOAT(v1, 0, mask1, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -190,6 +205,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -202,17 +218,17 @@ asm volatile(
                                 j += gvl*2;
                                 idx += inc_xv*2;
                         }
-                        v1 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        v0 = VFREDMINVS_FLOAT(v_min, v1, gvl);
-                        minf = v0[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                        minf = v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
                         //v0 = VFRSUBVF_MASK_FLOAT(v0, 0, mask0, gvl);
 #if defined(DOUBLE)
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e64,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -221,6 +237,7 @@ asm volatile(
         :"v0");
 #else
 asm volatile(
+        "vsetvli    zero, zero, e8, m1\n\t"
         "vor.vv     v0, %1, %1\n\t"
         "vsetvli    x0, %3, e32,m8 \n\t"
         "vfrsub.vf  %0, %0, %2, v0.t \n\t"
@@ -228,10 +245,9 @@ asm volatile(
         :"v"(mask0), "f"(zero), "r"(gvl)
         :"v0");
 #endif
-                        v1 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        v0 = VFREDMINVS_FLOAT(v0, v1, gvl);
-                        if(v0[0] < minf)
-                                minf = v0[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v0, v_max, gvl);
+                        if(v_res[0] < minf)
+                                minf = v_res[0];
                         j += gvl;
                 }
         }
diff --git a/kernel/riscv64/asum_vector.c b/kernel/riscv64/asum_vector.c
index 7ab7484e8..a623e9313 100644
--- a/kernel/riscv64/asum_vector.c
+++ b/kernel/riscv64/asum_vector.c
@@ -29,29 +29,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <math.h>
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDSUMVS_FLOAT vfredsumvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFADDVV_FLOAT vfaddvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDSUMVS_FLOAT vfredsum_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFADDVV_FLOAT vfadd_vv_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDSUMVS_FLOAT vfredsumvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFADDVV_FLOAT vfaddvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDSUMVS_FLOAT vfredsum_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFADDVV_FLOAT vfadd_vv_f64m8
 #endif
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 {
@@ -61,39 +65,43 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	if (n <= 0 || inc_x <= 0) return(asumf);
         unsigned int gvl = 0;
         FLOAT_V_T v0, v1, v_zero,v_sum;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
 
         MASK_T mask0, mask1;
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 v_zero = VFMVVF_FLOAT(0, gvl);
                 if(gvl <= n/2){
                         v_sum = VFMVVF_FLOAT(0, gvl);
                         for(i=0,j=0; i<n/(gvl*2); i++){
                                 v0 = VLEV_FLOAT(&x[j], gvl);
                                 mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                                v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                                v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                                 v_sum = VFADDVV_FLOAT(v_sum, v0, gvl);
 
                                 v1 = VLEV_FLOAT(&x[j+gvl], gvl);
                                 mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                                v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                                v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
                                 v_sum = VFADDVV_FLOAT(v_sum, v1, gvl);
                                 j += gvl * 2;
                         }
-                        v0 = VFREDSUMVS_FLOAT(v_sum, v_zero, gvl);
-                        asumf += v0[0];
+                        v_res = VFREDSUMVS_FLOAT(v_res, v_sum, v_z0, gvl);
+                        asumf += v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
-                        v0 = VFREDSUMVS_FLOAT(v0, v_zero, gvl);
-                        asumf += v0[0];
+                        v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
+                        v_res = VFREDSUMVS_FLOAT(v_res, v0, v_z0, gvl);
+                        asumf += v_res[0];
                         j += gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
                 v_zero = VFMVVF_FLOAT(0, gvl);
                 if(gvl <= n/2){
@@ -102,26 +110,26 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         for(i=0,j=0; i<n/(gvl*2); i++){
                                 v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                 mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                                v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                                v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                                 v_sum = VFADDVV_FLOAT(v_sum, v0, gvl);
 
                                 v1 = VLSEV_FLOAT(&x[ix+inc_xv], stride_x, gvl);
                                 mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                                v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                                v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
                                 v_sum = VFADDVV_FLOAT(v_sum, v1, gvl);
                                 j += gvl * 2;
                                 inc_xv += inc_xv * 2;
                         }
-                        v0 = VFREDSUMVS_FLOAT(v_sum, v_zero, gvl);
-                        asumf += v0[0];
+                        v_res = VFREDSUMVS_FLOAT(v_res, v_sum, v_z0, gvl);
+                        asumf += v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
-                        v0 = VFREDSUMVS_FLOAT(v0, v_zero, gvl);
-                        asumf += v0[0];
+                        v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
+                        v_res = VFREDSUMVS_FLOAT(v_res, v0, v_z0, gvl);
+                        asumf += v_res[0];
                         j += gvl;
                 }
         }
diff --git a/kernel/riscv64/axpby_vector.c b/kernel/riscv64/axpby_vector.c
index 432708db7..988c57ec2 100644
--- a/kernel/riscv64/axpby_vector.c
+++ b/kernel/riscv64/axpby_vector.c
@@ -28,27 +28,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSEV_FLOAT vsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFMULVF_FLOAT vfmulvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define FLOAT_V_T vfloat32m4_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSEV_FLOAT vse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMULVF_FLOAT vfmul_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSEV_FLOAT vsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFMULVF_FLOAT vfmulvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define FLOAT_V_T vfloat64m4_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSEV_FLOAT vse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMULVF_FLOAT vfmul_vf_f64m4
 #endif
 
 int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *y, BLASLONG inc_y)
@@ -65,7 +63,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
         if(beta == 0.0){
                 if(alpha == 0.0){//alpha == 0 && beta == 0
                         if(inc_y == 1){
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 if(gvl <= n/2){
                                         vy0 = VFMVVF_FLOAT(0.0, gvl);
                                         for(i=0,j=0;i<n/(gvl*2);i++){
@@ -75,13 +73,13 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vy0 = VFMVVF_FLOAT(0.0, gvl);
                                         VSEV_FLOAT(&y[j], vy0, gvl);
                                         j += gvl;
                                 }
                         }else{
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 stride_y = inc_y * sizeof(FLOAT);
                                 if(gvl <= n/2){
                                         vy0 = VFMVVF_FLOAT(0.0, gvl);
@@ -94,7 +92,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vy0 = VFMVVF_FLOAT(0.0, gvl);
                                         VSSEV_FLOAT(&y[j*inc_y], stride_y, vy0, gvl);
                                         j += gvl;
@@ -103,7 +101,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
 
                 }else{//alpha != 0 && beta == 0, y = ax
 			if(inc_x == 1 && inc_y == 1){
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 if(gvl <= n/2){
                                         for(i=0,j=0;i<n/(2*gvl);i++){
                                                 vx0 = VLEV_FLOAT(&x[j], gvl);
@@ -117,14 +115,14 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vx0 = VLEV_FLOAT(&x[j], gvl);
                                         vy0 = VFMULVF_FLOAT(vx0, alpha, gvl);
                                         VSEV_FLOAT(&y[j], vy0, gvl);
                                         j += gvl;
                                 }
 			}else if(inc_y == 1){
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 stride_x = inc_x * sizeof(FLOAT);
                                 if(gvl <= n/2){
                                         BLASLONG inc_xv = inc_x * gvl;
@@ -141,14 +139,14 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vx0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                                         vy0 = VFMULVF_FLOAT(vx0, alpha, gvl);
                                         VSEV_FLOAT(&y[j], vy0, gvl);
                                         j += gvl;
                                 }
                         }else if(inc_x == 1){
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 stride_y = inc_y * sizeof(FLOAT);
                                 if(gvl <= n/2){
                                         BLASLONG inc_yv = inc_y * gvl;
@@ -165,14 +163,14 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vx0 = VLEV_FLOAT(&x[j], gvl);
                                         vy0 = VFMULVF_FLOAT(vx0, alpha, gvl);
                                         VSSEV_FLOAT(&y[j*inc_y], stride_y, vy0, gvl);
                                         j += gvl;
                                 }
                         }else{//inc_x !=1 && inc_y != 1
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 stride_x = inc_x * sizeof(FLOAT);
                                 stride_y = inc_y * sizeof(FLOAT);
                                 if(gvl <= n/2){
@@ -192,7 +190,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vx0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                                         vy0 = VFMULVF_FLOAT(vx0, alpha, gvl);
                                         VSSEV_FLOAT(&y[j*inc_y], stride_y, vy0, gvl);
@@ -203,7 +201,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
         }else{//beta != 0
 		if(alpha == 0.0){//alpha == 0 && beta != 0; y = by
 			if(inc_y == 1){
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 if(gvl <= n/2){
                                         for(i=0,j=0;i<n/(2*gvl);i++){
                                                 vy0 = VLEV_FLOAT(&y[j], gvl);
@@ -217,14 +215,14 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vy0 = VLEV_FLOAT(&y[j], gvl);
                                         vy0 = VFMULVF_FLOAT(vy0, beta, gvl);
                                         VSEV_FLOAT(&y[j], vy0, gvl);
                                         j += gvl;
                                 }
 			}else{
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 stride_y = inc_y * sizeof(FLOAT);
                                 if(gvl <= n/2){
                                         BLASLONG inc_yv = inc_y * gvl;
@@ -241,7 +239,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vy0 = VLSEV_FLOAT(&y[j*inc_y], stride_y, gvl);
                                         vy0 = VFMULVF_FLOAT(vy0, beta, gvl);
                                         VSSEV_FLOAT(&y[j*inc_y], stride_y, vy0, gvl);
@@ -251,7 +249,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
 
 		}else{//alpha != 0 && beta != 0; y = ax + by
 			if(inc_x == 1 && inc_y == 1){
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 if(gvl <= n/2){
                                         for(i=0,j=0;i<n/(2*gvl);i++){
                                                 vx0 = VLEV_FLOAT(&x[j], gvl);
@@ -269,7 +267,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vx0 = VLEV_FLOAT(&x[j], gvl);
                                         vx0 = VFMULVF_FLOAT(vx0, alpha, gvl);
                                         vy0 = VLEV_FLOAT(&y[j], gvl);
@@ -278,7 +276,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         j += gvl;
                                 }
 			}else if(inc_y == 1){
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 stride_x = inc_x * sizeof(FLOAT);
                                 if(gvl <= n/2){
                                         BLASLONG inc_xv = inc_x * gvl;
@@ -299,7 +297,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vx0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                                         vx0 = VFMULVF_FLOAT(vx0, alpha, gvl);
                                         vy0 = VLEV_FLOAT(&y[j], gvl);
@@ -308,7 +306,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         j += gvl;
                                 }
                         }else if(inc_x == 1){
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 stride_y = inc_y * sizeof(FLOAT);
                                 if(gvl <= n/2){
                                         BLASLONG inc_yv = inc_y * gvl;
@@ -329,7 +327,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vx0 = VLEV_FLOAT(&x[j], gvl);
                                         vx0 = VFMULVF_FLOAT(vx0, alpha, gvl);
                                         vy0 = VLSEV_FLOAT(&y[j*inc_y], stride_y, gvl);
@@ -338,7 +336,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         j += gvl;
                                 }
                         }else{//inc_x != 1 && inc_y != 1
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 stride_x = inc_x * sizeof(FLOAT);
                                 stride_y = inc_y * sizeof(FLOAT);
                                 if(gvl <= n/2){
@@ -362,7 +360,7 @@ int CNAME(BLASLONG n, FLOAT alpha, FLOAT *x, BLASLONG inc_x, FLOAT beta, FLOAT *
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vx0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                                         vx0 = VFMULVF_FLOAT(vx0, alpha, gvl);
                                         vy0 = VLSEV_FLOAT(&y[j*inc_y], stride_y, gvl);
diff --git a/kernel/riscv64/axpy_vector.c b/kernel/riscv64/axpy_vector.c
index 5a7ba4b08..98b9f6814 100644
--- a/kernel/riscv64/axpy_vector.c
+++ b/kernel/riscv64/axpy_vector.c
@@ -28,23 +28,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSEV_FLOAT vsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define FLOAT_V_T vfloat32m4_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSEV_FLOAT vse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSEV_FLOAT vsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define FLOAT_V_T vfloat64m4_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSEV_FLOAT vse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
 #endif
 
 int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
@@ -60,7 +58,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 
 	if (inc_x == 1 && inc_y == 1) {
 
-		gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+		gvl = VSETVL(n);
 
 		if (gvl <= n/2) {
 			for (i = 0, j=0; i < n/(2*gvl); i++, j+=2*gvl) {
@@ -77,7 +75,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 		}
 		//tail
 		for (; j < n; ) {
-			gvl = vsetvli(n - j, RVV_EFLOAT, RVV_M);
+			gvl = VSETVL(n - j);
 			vx0 = VLEV_FLOAT(&x[j], gvl);
 			vy0 = VLEV_FLOAT(&y[j], gvl);
 			vy0 = VFMACCVF_FLOAT(vy0, da, vx0, gvl);
@@ -87,7 +85,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 		}
 	}else if (inc_y == 1) {
 		stride_x = inc_x * sizeof(FLOAT);
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl <= n/2){
                         BLASLONG inc_xv = inc_x * gvl;
                         for(i=0,j=0; i<n/(2*gvl); i++){
@@ -106,7 +104,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
                         }
                 }
 		for (; j<n; ) {
-			gvl = vsetvli(n - j, RVV_EFLOAT, RVV_M);
+			gvl = VSETVL(n - j);
 			vx0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
 			vy0 = VLEV_FLOAT(&y[j], gvl);
 			vy0 = VFMACCVF_FLOAT(vy0, da, vx0, gvl);
@@ -115,7 +113,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 		}
         }else if(inc_x == 1){
 		stride_y = inc_y * sizeof(FLOAT);
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl <= n/2){
                         BLASLONG inc_yv = inc_y * gvl;
                         for(i=0,j=0; i<n/(2*gvl); i++){
@@ -134,7 +132,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
                         }
                 }
 		for (; j<n; ) {
-			gvl = vsetvli(n - j, RVV_EFLOAT, RVV_M);
+			gvl = VSETVL(n - j);
 			vx0 = VLEV_FLOAT(&x[j], gvl);
 			vy0 = VLSEV_FLOAT(&y[j*inc_y], stride_y, gvl);
 			vy0 = VFMACCVF_FLOAT(vy0, da, vx0, gvl);
@@ -144,7 +142,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 	}else{
 		stride_x = inc_x * sizeof(FLOAT);
 		stride_y = inc_y * sizeof(FLOAT);
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl <= n/2){
                         BLASLONG inc_xv = inc_x * gvl;
                         BLASLONG inc_yv = inc_y * gvl;
@@ -165,7 +163,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
                         }
                 }
 		for (; j<n; ) {
-			gvl = vsetvli(n - j, RVV_EFLOAT, RVV_M);
+			gvl = VSETVL(n - j);
 			vx0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
 			vy0 = VLSEV_FLOAT(&y[j*inc_y], stride_y, gvl);
 			vy0 = VFMACCVF_FLOAT(vy0, da, vx0, gvl);
diff --git a/kernel/riscv64/copy_vector.c b/kernel/riscv64/copy_vector.c
index ce44a20f3..b4ac110f1 100644
--- a/kernel/riscv64/copy_vector.c
+++ b/kernel/riscv64/copy_vector.c
@@ -26,21 +26,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VSEV_FLOAT vsev_float32xm8
-#define VSSEV_FLOAT vssev_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define FLOAT_V_T vfloat32m8_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VSEV_FLOAT vse_v_f32m8
+#define VSSEV_FLOAT vsse_v_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VSEV_FLOAT vsev_float64xm8
-#define VSSEV_FLOAT vssev_float64xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define FLOAT_V_T vfloat64m8_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VSEV_FLOAT vse_v_f64m8
+#define VSSEV_FLOAT vsse_v_f64m8
 #endif
 
 int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
@@ -56,7 +54,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
         if(inc_x == 1 && inc_y == 1){
                 memcpy(&y[0], &x[0], n*sizeof(FLOAT));
         }else if (inc_y == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 stride_x = inc_x * sizeof(FLOAT);
                 if(gvl <= n/4){
                         BLASLONG inc_xv = inc_x * gvl;
@@ -77,13 +75,13 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                         }
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         VSEV_FLOAT(&y[j], v0, gvl);
                         j += gvl;
                 }
         }else if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 stride_y = inc_y * sizeof(FLOAT);
                 if(gvl <= n/4){
                         BLASLONG inc_yv = inc_y * gvl;
@@ -104,14 +102,14 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                         }
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         VSSEV_FLOAT(&y[j*inc_y], stride_y, v0, gvl);
                         j += gvl;
                 }
 
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 stride_x = inc_x * sizeof(FLOAT);
                 stride_y = inc_y * sizeof(FLOAT);
                 if(gvl <= n/4){
@@ -136,7 +134,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                         }
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         VSSEV_FLOAT(&y[j*inc_y], stride_y, v0, gvl);
                         j += gvl;
diff --git a/kernel/riscv64/dot_vector.c b/kernel/riscv64/dot_vector.c
index 8b1ae278c..1d92699e0 100644
--- a/kernel/riscv64/dot_vector.c
+++ b/kernel/riscv64/dot_vector.c
@@ -27,25 +27,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFDOTVV_FLOAT vfdotvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFDOTVV_FLOAT vfdot_vv_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFDOTVV_FLOAT vfdotvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFDOTVV_FLOAT vfdot_vv_f64m4
 #endif
 
 #if defined(DSDOT)
@@ -61,8 +65,13 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
 
         FLOAT_V_T vr, vx, vy;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
+
         if(inc_x == 1 && inc_y == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 vr = VFMVVF_FLOAT(0, gvl);
                 for(i=0,j=0; i<n/gvl; i++){
                         vx = VLEV_FLOAT(&x[j], gvl);
@@ -71,23 +80,22 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                         j += gvl;
                 }
                 if(j > 0){
-                        vx = VFMVVF_FLOAT(0, gvl);
-                        vx = VFREDSUM_FLOAT(vr, vx, gvl);
-                        dot += vx[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        dot += v_res[0];
                 }
                 //tail
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLEV_FLOAT(&x[j], gvl);
                         vy = VLEV_FLOAT(&y[j], gvl);
                         FLOAT_V_T vz = VFMVVF_FLOAT(0, gvl);
                         //vr = VFDOTVV_FLOAT(vx, vy, gvl);
                         vr = VFMACCVV_FLOAT(vz, vx, vy, gvl);
-                        vx = VFREDSUM_FLOAT(vr, vz, gvl);
-                        dot += vx[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        dot += v_res[0];
                 }
         }else if(inc_y == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 vr = VFMVVF_FLOAT(0, gvl);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
                 for(i=0,j=0; i<n/gvl; i++){
@@ -97,23 +105,22 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                         j += gvl;
                 }
                 if(j > 0){
-                        vx = VFMVVF_FLOAT(0, gvl);
-                        vx = VFREDSUM_FLOAT(vr, vx, gvl);
-                        dot += vx[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        dot += v_res[0];
                 }
                 //tail
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         vy = VLEV_FLOAT(&y[j], gvl);
                         FLOAT_V_T vz = VFMVVF_FLOAT(0, gvl);
                         //vr = VFDOTVV_FLOAT(vx, vy, gvl);
                         vr = VFMACCVV_FLOAT(vz, vx, vy, gvl);
-                        vx = VFREDSUM_FLOAT(vr, vz, gvl);
-                        dot += vx[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        dot += v_res[0];
                 }
         }else if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 vr = VFMVVF_FLOAT(0, gvl);
                 unsigned int stride_y = inc_y * sizeof(FLOAT);
                 for(i=0,j=0; i<n/gvl; i++){
@@ -123,23 +130,22 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                         j += gvl;
                 }
                 if(j > 0){
-                        vx = VFMVVF_FLOAT(0, gvl);
-                        vx = VFREDSUM_FLOAT(vr, vx, gvl);
-                        dot += vx[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        dot += v_res[0];
                 }
                 //tail
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLEV_FLOAT(&x[j], gvl);
                         vy = VLSEV_FLOAT(&y[j*inc_y], stride_y, gvl);
                         FLOAT_V_T vz = VFMVVF_FLOAT(0, gvl);
                         //vr = VFDOTVV_FLOAT(vx, vy, gvl);
                         vr = VFMACCVV_FLOAT(vz, vx, vy, gvl);
-                        vx = VFREDSUM_FLOAT(vr, vz, gvl);
-                        dot += vx[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        dot += v_res[0];
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 vr = VFMVVF_FLOAT(0, gvl);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
                 unsigned int stride_y = inc_y * sizeof(FLOAT);
@@ -150,20 +156,19 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                         j += gvl;
                 }
                 if(j > 0){
-                        vx = VFMVVF_FLOAT(0, gvl);
-                        vx = VFREDSUM_FLOAT(vr, vx, gvl);
-                        dot += vx[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        dot += v_res[0];
                 }
                 //tail
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         vy = VLSEV_FLOAT(&y[j*inc_y], stride_y, gvl);
                         FLOAT_V_T vz = VFMVVF_FLOAT(0, gvl);
                         //vr = VFDOTVV_FLOAT(vx, vy, gvl);
                         vr = VFMACCVV_FLOAT(vz, vx, vy, gvl);
-                        vx = VFREDSUM_FLOAT(vr, vz, gvl);
-                        dot += vx[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        dot += v_res[0];
                 }
         }
 	return(dot);
diff --git a/kernel/riscv64/gemv_n_vector.c b/kernel/riscv64/gemv_n_vector.c
index bd4d23eae..32ca8618b 100644
--- a/kernel/riscv64/gemv_n_vector.c
+++ b/kernel/riscv64/gemv_n_vector.c
@@ -27,23 +27,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSEV_FLOAT vsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define FLOAT_V_T vfloat32m4_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSEV_FLOAT vse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSEV_FLOAT vsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define FLOAT_V_T vfloat64m4_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSEV_FLOAT vse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
 #endif
 
 int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@@ -57,7 +55,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
         FLOAT_V_T va0, va1, vy0, vy1;
         unsigned int gvl = 0;
         if(inc_y == 1){
-                gvl = vsetvli(m, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(m);
                 if(gvl <= m/2){
                         for(k=0,j=0; k<m/(2*gvl); k++){
                                 a_ptr = a;
@@ -81,7 +79,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
                 }
                 //tail
                 for(;j < m;){
-                        gvl = vsetvli(m-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(m-j);
                         a_ptr = a;
                         ix = 0;
                         vy0 = VLEV_FLOAT(&y[j], gvl);
@@ -98,7 +96,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
                 }
         }else{
                 BLASLONG stride_y = inc_y * sizeof(FLOAT);
-                gvl = vsetvli(m, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(m);
                 if(gvl <= m/2){
                         BLASLONG inc_yv = inc_y * gvl;
                         for(k=0,j=0; k<m/(2*gvl); k++){
@@ -124,7 +122,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
                 }
                 //tail
                 for(;j < m;){
-                        gvl = vsetvli(m-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(m-j);
                         a_ptr = a;
                         ix = 0;
                         vy0 = VLSEV_FLOAT(&y[j*inc_y], stride_y, gvl);
diff --git a/kernel/riscv64/gemv_t_vector.c b/kernel/riscv64/gemv_t_vector.c
index beca8dc0f..ceba107cc 100644
--- a/kernel/riscv64/gemv_t_vector.c
+++ b/kernel/riscv64/gemv_t_vector.c
@@ -27,41 +27,50 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFDOTVV_FLOAT vfdotvv_float32xm4
-#define VFMULVV_FLOAT vfmulvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFDOTVV_FLOAT vfdot_vv_f32m4
+#define VFMULVV_FLOAT vfmul_vv_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFDOTVV_FLOAT vfdotvv_float64xm4
-#define VFMULVV_FLOAT vfmulvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFDOTVV_FLOAT vfdot_vv_f64m4
+#define VFMULVV_FLOAT vfmul_vv_f64m4
 #endif
 
 int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
 {
-	BLASLONG i = 0, j = 0, k = 0;
-	BLASLONG ix = 0, iy = 0;
-	FLOAT *a_ptr = a;
+	    BLASLONG i = 0, j = 0, k = 0;
+	    BLASLONG ix = 0, iy = 0;
+	    FLOAT *a_ptr = a;
         FLOAT temp;
 
         FLOAT_V_T va, vr, vx;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
+
         if(inc_x == 1){
                 for(i = 0; i < n; i++){
-                        gvl = vsetvli(m, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(m);
                         j = 0;
                         vr = VFMVVF_FLOAT(0, gvl);
                         for(k = 0; k < m/gvl; k++){
@@ -70,29 +79,26 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
                                 vr = VFMACCVV_FLOAT(vr, va, vx, gvl);
                                 j += gvl;
                         }
-                        va = VFMVVF_FLOAT(0, gvl);
-                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                        temp = va[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        temp = v_res[0];
                         if(j < m){
-                                gvl = vsetvli(m-j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(m-j);
                                 va = VLEV_FLOAT(&a_ptr[j], gvl);
                                 vx = VLEV_FLOAT(&x[j], gvl);
                                 vr = VFMULVV_FLOAT(va, vx, gvl);
 
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp += va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp += v_res[0];
                         }
                         y[iy] += alpha * temp;
                         iy += inc_y;
                         a_ptr += lda;
                 }
         }else{
-                gvl = vsetvli(m, RVV_EFLOAT, RVV_M);
                 BLASLONG stride_x = inc_x * sizeof(FLOAT);
                 BLASLONG inc_xv = inc_x * gvl;
                 for(i = 0; i < n; i++){
-                        gvl = vsetvli(m, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(m);
                         j = 0;
                         ix = 0;
                         vr = VFMVVF_FLOAT(0, gvl);
@@ -103,18 +109,16 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
                                 j += gvl;
                                 ix += inc_xv;
                         }
-                        va = VFMVVF_FLOAT(0, gvl);
-                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                        temp = va[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        temp = v_res[0];
                         if(j < m){
-                                gvl = vsetvli(m-j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(m-j);
                                 va = VLEV_FLOAT(&a_ptr[j], gvl);
                                 vx = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                 vr = VFMULVV_FLOAT(va, vx, gvl);
 
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp += va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp += v_res[0];
                         }
                         y[iy] += alpha * temp;
                         iy += inc_y;
diff --git a/kernel/riscv64/iamax_vector.c b/kernel/riscv64/iamax_vector.c
index 3aa64afc9..056c0aaa8 100644
--- a/kernel/riscv64/iamax_vector.c
+++ b/kernel/riscv64/iamax_vector.c
@@ -31,49 +31,53 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(DOUBLE)
 
 #define ABS fabs
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VMFLTVV_FLOAT vmfltvv_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFMAXVV_FLOAT vfmaxvv_float64xm8
-#define VMFGEVF_FLOAT vmfgevf_e64xm8_float64xm8
-#define VMFIRSTM vmfirstm_e64xm8
-#define UINT_V_T uint64xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint64xm8
-#define VIDV_UINT vidv_uint64xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint64xm8
-#define VADDVX_UINT vaddvx_uint64xm8
-#define VMVVX_UINT vmvvx_uint64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFMAXVV_FLOAT vfmax_vv_f64m8
+#define VMFGEVF_FLOAT vmfge_vf_f64m8_b8
+#define VMFIRSTM vmfirst_m_b8
+#define UINT_V_T vuint64m8_t
+#define VIDV_MASK_UINT vid_v_u64m8_m
+#define VIDV_UINT vid_v_u64m8
+#define VADDVX_MASK_UINT vadd_vx_u64m8_m
+#define VADDVX_UINT vadd_vx_u64m8
+#define VMVVX_UINT vmv_v_x_u64m8
 #else
 
 #define ABS fabsf
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VMFLTVV_FLOAT vmfltvv_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFMAXVV_FLOAT vfmaxvv_float32xm8
-#define VMFGEVF_FLOAT vmfgevf_e32xm8_float32xm8
-#define VMFIRSTM vmfirstm_e32xm8
-#define UINT_V_T uint32xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint32xm8
-#define VIDV_UINT vidv_uint32xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint32xm8
-#define VADDVX_UINT vaddvx_uint32xm8
-#define VMVVX_UINT vmvvx_uint32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFMAXVV_FLOAT vfmax_vv_f32m8
+#define VMFGEVF_FLOAT vmfge_vf_f32m8_b4
+#define VMFIRSTM vmfirst_m_b4
+#define UINT_V_T vuint32m8_t
+#define VIDV_MASK_UINT vid_v_u32m8_m
+#define VIDV_UINT vid_v_u32m8
+#define VADDVX_MASK_UINT vadd_vx_u32m8_m
+#define VADDVX_UINT vadd_vx_u32m8
+#define VMVVX_UINT vmv_v_x_u32m8
 #endif
 
 
@@ -88,42 +92,45 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         UINT_V_T v_max_index;
         MASK_T mask;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
+
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 v_max_index = VMVVX_UINT(0, gvl);
                 v_max = VFMVVF_FLOAT(-1, gvl);
                 for(i=0,j=0; i < n/gvl; i++){
                         vx = VLEV_FLOAT(&x[j], gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(vx, 0, gvl);
-                        vx = VFRSUBVF_MASK_FLOAT(vx, vx, 0, mask, gvl);
+                        vx = VFRSUBVF_MASK_FLOAT(mask, vx, vx, 0, gvl);
 
                         //index where element greater than v_max
                         mask = VMFLTVV_FLOAT(v_max, vx, gvl);
-                        v_max_index = VIDV_MASK_UINT(v_max_index, mask, gvl);
-                        v_max_index = VADDVX_MASK_UINT(v_max_index, v_max_index, j, mask, gvl);
+                        v_max_index = VIDV_MASK_UINT(mask, v_max_index, gvl);
+                        v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j,gvl);
 
                         //update v_max and start_index j
                         v_max = VFMAXVV_FLOAT(v_max, vx, gvl);
                         j += gvl;
                 }
-                vx = VFMVVF_FLOAT(0, gvl);
-                vx = VFREDMAXVS_FLOAT(v_max, vx, gvl);
-                maxf = vx[0];
+                v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, gvl);
+                maxf = v_res[0];
                 mask = VMFGEVF_FLOAT(v_max, maxf, gvl);
                 max_index = VMFIRSTM(mask,gvl);
                 max_index = v_max_index[max_index];
 
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLEV_FLOAT(&x[j], gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(vx, 0, gvl);
-                        v_max = VFRSUBVF_MASK_FLOAT(vx, vx, 0, mask, gvl);
+                        v_max = VFRSUBVF_MASK_FLOAT(mask, vx, vx, 0, gvl);
 
-                        vx = VFMVVF_FLOAT(0, gvl);
-                        vx = VFREDMAXVS_FLOAT(v_max, vx, gvl);
-                        FLOAT cur_maxf = vx[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, gvl);
+                        FLOAT cur_maxf = v_res[0];
                         if(cur_maxf > maxf){
                                 //tail index
                                 v_max_index = VIDV_UINT(gvl);
@@ -135,7 +142,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         }
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
                 unsigned int idx = 0, inc_v = gvl * inc_x;
 
@@ -145,35 +152,33 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         vx = VLSEV_FLOAT(&x[idx], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(vx, 0, gvl);
-                        vx = VFRSUBVF_MASK_FLOAT(vx, vx, 0, mask, gvl);
+                        vx = VFRSUBVF_MASK_FLOAT(mask, vx, vx, 0, gvl);
 
                         //index where element greater than v_max
                         mask = VMFLTVV_FLOAT(v_max, vx, gvl);
-                        v_max_index = VIDV_MASK_UINT(v_max_index, mask, gvl);
-                        v_max_index = VADDVX_MASK_UINT(v_max_index, v_max_index, j, mask, gvl);
+                        v_max_index = VIDV_MASK_UINT(mask, v_max_index, gvl);
+                        v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, gvl);
 
                         //update v_max and start_index j
                         v_max = VFMAXVV_FLOAT(v_max, vx, gvl);
                         j += gvl;
                         idx += inc_v;
                 }
-                vx = VFMVVF_FLOAT(0, gvl);
-                vx = VFREDMAXVS_FLOAT(v_max, vx, gvl);
-                maxf = vx[0];
+                v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, gvl);
+                maxf = v_res[0];
                 mask = VMFGEVF_FLOAT(v_max, maxf, gvl);
                 max_index = VMFIRSTM(mask,gvl);
                 max_index = v_max_index[max_index];
 
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLSEV_FLOAT(&x[idx], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(vx, 0, gvl);
-                        v_max = VFRSUBVF_MASK_FLOAT(vx, vx, 0, mask, gvl);
+                        v_max = VFRSUBVF_MASK_FLOAT(mask, vx, vx, 0, gvl);
 
-                        vx = VFMVVF_FLOAT(0, gvl);
-                        vx = VFREDMAXVS_FLOAT(v_max, vx, gvl);
-                        FLOAT cur_maxf = vx[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, gvl);
+                        FLOAT cur_maxf = v_res[0];
                         if(cur_maxf > maxf){
                                 //tail index
                                 v_max_index = VIDV_UINT(gvl);
diff --git a/kernel/riscv64/iamin_vector.c b/kernel/riscv64/iamin_vector.c
index 608f19a00..5bcffece5 100644
--- a/kernel/riscv64/iamin_vector.c
+++ b/kernel/riscv64/iamin_vector.c
@@ -32,49 +32,53 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(DOUBLE)
 
 #define ABS fabs
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMINVS_FLOAT vfredminvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VMFLTVV_FLOAT vmfltvv_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFMINVV_FLOAT vfminvv_float64xm8
-#define VMFLEVF_FLOAT vmflevf_e64xm8_float64xm8
-#define VMFIRSTM vmfirstm_e64xm8
-#define UINT_V_T uint64xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint64xm8
-#define VIDV_UINT vidv_uint64xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint64xm8
-#define VADDVX_UINT vaddvx_uint64xm8
-#define VMVVX_UINT vmvvx_uint64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFMINVV_FLOAT vfmin_vv_f64m8
+#define VMFLEVF_FLOAT vmfle_vf_f64m8_b8
+#define VMFIRSTM vmfirst_m_b8
+#define UINT_V_T vuint64m8_t
+#define VIDV_MASK_UINT vid_v_u64m8_m
+#define VIDV_UINT vid_v_u64m8
+#define VADDVX_MASK_UINT vadd_vx_u64m8_m
+#define VADDVX_UINT vadd_vx_u64m8
+#define VMVVX_UINT vmv_v_x_u64m8
 #else
 
 #define ABS fabsf
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMINVS_FLOAT vfredminvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VMFLTVV_FLOAT vmfltvv_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFMINVV_FLOAT vfminvv_float32xm8
-#define VMFLEVF_FLOAT vmflevf_e32xm8_float32xm8
-#define VMFIRSTM vmfirstm_e32xm8
-#define UINT_V_T uint32xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint32xm8
-#define VIDV_UINT vidv_uint32xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint32xm8
-#define VADDVX_UINT vaddvx_uint32xm8
-#define VMVVX_UINT vmvvx_uint32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFMINVV_FLOAT vfmin_vv_f32m8
+#define VMFLEVF_FLOAT vmfle_vf_f32m8_b4
+#define VMFIRSTM vmfirst_m_b4
+#define UINT_V_T vuint32m8_t
+#define VIDV_MASK_UINT vid_v_u32m8_m
+#define VIDV_UINT vid_v_u32m8
+#define VADDVX_MASK_UINT vadd_vx_u32m8_m
+#define VADDVX_UINT vadd_vx_u32m8
+#define VMVVX_UINT vmv_v_x_u32m8
 #endif
 
 
@@ -89,42 +93,45 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         UINT_V_T v_min_index;
         MASK_T mask;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_max;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_max = VFMVVF_FLOAT_M1(FLT_MAX, gvl);
+
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 v_min = VFMVVF_FLOAT(FLT_MAX, gvl);
                 v_min_index = VMVVX_UINT(0, gvl);
                 for(i=0,j=0; i < n/gvl; i++){
                         vx = VLEV_FLOAT(&x[j], gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(vx, 0, gvl);
-                        vx = VFRSUBVF_MASK_FLOAT(vx, vx, 0, mask, gvl);
+                        vx = VFRSUBVF_MASK_FLOAT(mask, vx, vx, 0, gvl);
 
                         //index where element less than v_min
                         mask = VMFLTVV_FLOAT(vx, v_min, gvl);
-                        v_min_index = VIDV_MASK_UINT(v_min_index, mask, gvl);
-                        v_min_index = VADDVX_MASK_UINT(v_min_index, v_min_index, j, mask, gvl);
+                        v_min_index = VIDV_MASK_UINT(mask, v_min_index, gvl);
+                        v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, gvl);
 
                         //update v_min and start_index j
                         v_min = VFMINVV_FLOAT(v_min, vx, gvl);
                         j += gvl;
                 }
-                vx = VFMVVF_FLOAT(FLT_MAX, gvl);
-                vx = VFREDMINVS_FLOAT(v_min, vx, gvl);
-                minf = vx[0];
+                v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                minf = v_res[0];
                 mask = VMFLEVF_FLOAT(v_min, minf, gvl);
                 min_index = VMFIRSTM(mask,gvl);
                 min_index = v_min_index[min_index];
 
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLEV_FLOAT(&x[j], gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(vx, 0, gvl);
-                        v_min = VFRSUBVF_MASK_FLOAT(vx, vx, 0, mask, gvl);
+                        v_min = VFRSUBVF_MASK_FLOAT(mask, vx, vx, 0, gvl);
 
-                        vx = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        vx = VFREDMINVS_FLOAT(v_min, vx, gvl);
-                        FLOAT cur_minf = vx[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                        FLOAT cur_minf = v_res[0];
                         if(cur_minf < minf){
                                 //tail index
                                 v_min_index = VIDV_UINT(gvl);
@@ -136,7 +143,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         }
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
                 unsigned int idx = 0, inc_v = gvl * inc_x;
 
@@ -146,35 +153,33 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         vx = VLSEV_FLOAT(&x[idx], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(vx, 0, gvl);
-                        vx = VFRSUBVF_MASK_FLOAT(vx, vx, 0, mask, gvl);
+                        vx = VFRSUBVF_MASK_FLOAT(mask, vx, vx, 0, gvl);
 
                         //index where element less than v_min
                         mask = VMFLTVV_FLOAT(vx, v_min, gvl);
-                        v_min_index = VIDV_MASK_UINT(v_min_index, mask, gvl);
-                        v_min_index = VADDVX_MASK_UINT(v_min_index, v_min_index, j, mask, gvl);
+                        v_min_index = VIDV_MASK_UINT(mask, v_min_index, gvl);
+                        v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, gvl);
 
                         //update v_min and start_index j
                         v_min = VFMINVV_FLOAT(v_min, vx, gvl);
                         j += gvl;
                         idx += inc_v;
                 }
-                vx = VFMVVF_FLOAT(FLT_MAX, gvl);
-                vx = VFREDMINVS_FLOAT(v_min, vx, gvl);
-                minf = vx[0];
+                v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                minf = v_res[0];
                 mask = VMFLEVF_FLOAT(v_min, minf, gvl);
                 min_index = VMFIRSTM(mask,gvl);
                 min_index = v_min_index[min_index];
 
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLSEV_FLOAT(&x[idx], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(vx, 0, gvl);
-                        v_min = VFRSUBVF_MASK_FLOAT(vx, vx, 0, mask, gvl);
+                        v_min = VFRSUBVF_MASK_FLOAT(mask, vx, vx, 0, gvl);
 
-                        vx = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        vx = VFREDMINVS_FLOAT(v_min, vx, gvl);
-                        FLOAT cur_minf = vx[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                        FLOAT cur_minf = v_res[0];
                         if(cur_minf < minf){
                                 //tail index
                                 v_min_index = VIDV_UINT(gvl);
diff --git a/kernel/riscv64/imax_vector.c b/kernel/riscv64/imax_vector.c
index 44af7101b..42705f5de 100644
--- a/kernel/riscv64/imax_vector.c
+++ b/kernel/riscv64/imax_vector.c
@@ -32,45 +32,49 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(DOUBLE)
 
 #define ABS fabs
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVV_FLOAT vmfltvv_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFMAXVV_FLOAT vfmaxvv_float64xm8
-#define VMFGEVF_FLOAT vmfgevf_e64xm8_float64xm8
-#define VMFIRSTM vmfirstm_e64xm8
-#define UINT_V_T uint64xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint64xm8
-#define VIDV_UINT vidv_uint64xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint64xm8
-#define VADDVX_UINT vaddvx_uint64xm8
-#define VMVVX_UINT vmvvx_uint64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFMAXVV_FLOAT vfmax_vv_f64m8
+#define VMFGEVF_FLOAT vmfge_vf_f64m8_b8
+#define VMFIRSTM vmfirst_m_b8
+#define UINT_V_T vuint64m8_t
+#define VIDV_MASK_UINT vid_v_u64m8_m
+#define VIDV_UINT vid_v_u64m8
+#define VADDVX_MASK_UINT vadd_vx_u64m8_m
+#define VADDVX_UINT vadd_vx_u64m8
+#define VMVVX_UINT vmv_v_x_u64m8
 #else
 
 #define ABS fabsf
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVV_FLOAT vmfltvv_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFMAXVV_FLOAT vfmaxvv_float32xm8
-#define VMFGEVF_FLOAT vmfgevf_e32xm8_float32xm8
-#define VMFIRSTM vmfirstm_e32xm8
-#define UINT_V_T uint32xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint32xm8
-#define VIDV_UINT vidv_uint32xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint32xm8
-#define VADDVX_UINT vaddvx_uint32xm8
-#define VMVVX_UINT vmvvx_uint32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFMAXVV_FLOAT vfmax_vv_f32m8
+#define VMFGEVF_FLOAT vmfge_vf_f32m8_b4
+#define VMFIRSTM vmfirst_m_b4
+#define UINT_V_T vuint32m8_t
+#define VIDV_MASK_UINT vid_v_u32m8_m
+#define VIDV_UINT vid_v_u32m8
+#define VADDVX_MASK_UINT vadd_vx_u32m8_m
+#define VADDVX_UINT vadd_vx_u32m8
+#define VMVVX_UINT vmv_v_x_u32m8
 #endif
 
 
@@ -85,8 +89,13 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         UINT_V_T v_max_index;
         MASK_T mask;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_min;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_min = VFMVVF_FLOAT_M1(-FLT_MAX, gvl);
+
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 v_max_index = VMVVX_UINT(0, gvl);
                 v_max = VFMVVF_FLOAT(-FLT_MAX, gvl);
                 for(i=0,j=0; i < n/gvl; i++){
@@ -94,27 +103,25 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 
                         //index where element greater than v_max
                         mask = VMFLTVV_FLOAT(v_max, vx, gvl);
-                        v_max_index = VIDV_MASK_UINT(v_max_index, mask, gvl);
-                        v_max_index = VADDVX_MASK_UINT(v_max_index, v_max_index, j, mask, gvl);
+                        v_max_index = VIDV_MASK_UINT(mask, v_max_index, gvl);
+                        v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j,gvl);
 
                         //update v_max and start_index j
                         v_max = VFMAXVV_FLOAT(v_max, vx, gvl);
                         j += gvl;
                 }
-                vx = VFMVVF_FLOAT(-FLT_MAX, gvl);
-                vx = VFREDMAXVS_FLOAT(v_max, vx, gvl);
-                maxf = vx[0];
+                v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_min, gvl);
+                maxf = v_res[0];
                 mask = VMFGEVF_FLOAT(v_max, maxf, gvl);
                 max_index = VMFIRSTM(mask,gvl);
                 max_index = v_max_index[max_index];
 
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v_max = VLEV_FLOAT(&x[j], gvl);
 
-                        vx = VFMVVF_FLOAT(-FLT_MAX, gvl);
-                        vx = VFREDMAXVS_FLOAT(v_max, vx, gvl);
-                        FLOAT cur_maxf = vx[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_min, gvl);
+                        FLOAT cur_maxf = v_res[0];
                         if(cur_maxf > maxf){
                                 //tail index
                                 v_max_index = VIDV_UINT(gvl);
@@ -126,7 +133,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         }
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
                 unsigned int idx = 0, inc_v = gvl * inc_x;
 
@@ -137,28 +144,26 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 
                         //index where element greater than v_max
                         mask = VMFLTVV_FLOAT(v_max, vx, gvl);
-                        v_max_index = VIDV_MASK_UINT(v_max_index, mask, gvl);
-                        v_max_index = VADDVX_MASK_UINT(v_max_index, v_max_index, j, mask, gvl);
+                        v_max_index = VIDV_MASK_UINT(mask, v_max_index, gvl);
+                        v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j,gvl);
 
                         //update v_max and start_index j
                         v_max = VFMAXVV_FLOAT(v_max, vx, gvl);
                         j += gvl;
                         idx += inc_v;
                 }
-                vx = VFMVVF_FLOAT(-FLT_MAX, gvl);
-                vx = VFREDMAXVS_FLOAT(v_max, vx, gvl);
-                maxf = vx[0];
+                v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_min, gvl);
+                maxf = v_res[0];
                 mask = VMFGEVF_FLOAT(v_max, maxf, gvl);
                 max_index = VMFIRSTM(mask,gvl);
                 max_index = v_max_index[max_index];
 
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v_max = VLSEV_FLOAT(&x[idx], stride_x, gvl);
 
-                        vx = VFMVVF_FLOAT(-FLT_MAX, gvl);
-                        vx = VFREDMAXVS_FLOAT(v_max, vx, gvl);
-                        FLOAT cur_maxf = vx[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_min, gvl);
+                        FLOAT cur_maxf = v_res[0];
                         if(cur_maxf > maxf){
                                 //tail index
                                 v_max_index = VIDV_UINT(gvl);
diff --git a/kernel/riscv64/imin.c b/kernel/riscv64/imin.c
index 598cba387..ffc65226e 100644
--- a/kernel/riscv64/imin.c
+++ b/kernel/riscv64/imin.c
@@ -53,7 +53,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 
 	while(i < n)
 	{
-		if( x[ix] > minf )
+		if( x[ix] < minf )
 		{
 			min = i;
 			minf = x[ix];
diff --git a/kernel/riscv64/imin_vector.c b/kernel/riscv64/imin_vector.c
index e6e0e9f9f..3afa74dd6 100644
--- a/kernel/riscv64/imin_vector.c
+++ b/kernel/riscv64/imin_vector.c
@@ -32,45 +32,49 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(DOUBLE)
 
 #define ABS fabs
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMINVS_FLOAT vfredminvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVV_FLOAT vmfltvv_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFMINVV_FLOAT vfminvv_float64xm8
-#define VMFLEVF_FLOAT vmflevf_e64xm8_float64xm8
-#define VMFIRSTM vmfirstm_e64xm8
-#define UINT_V_T uint64xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint64xm8
-#define VIDV_UINT vidv_uint64xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint64xm8
-#define VADDVX_UINT vaddvx_uint64xm8
-#define VMVVX_UINT vmvvx_uint64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFMINVV_FLOAT vfmin_vv_f64m8
+#define VMFLEVF_FLOAT vmfle_vf_f64m8_b8
+#define VMFIRSTM vmfirst_m_b8
+#define UINT_V_T vuint64m8_t
+#define VIDV_MASK_UINT vid_v_u64m8_m
+#define VIDV_UINT vid_v_u64m8
+#define VADDVX_MASK_UINT vadd_vx_u64m8_m
+#define VADDVX_UINT vadd_vx_u64m8
+#define VMVVX_UINT vmv_v_x_u64m8
 #else
 
 #define ABS fabsf
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMINVS_FLOAT vfredminvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVV_FLOAT vmfltvv_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFMINVV_FLOAT vfminvv_float32xm8
-#define VMFLEVF_FLOAT vmflevf_e32xm8_float32xm8
-#define VMFIRSTM vmfirstm_e32xm8
-#define UINT_V_T uint32xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint32xm8
-#define VIDV_UINT vidv_uint32xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint32xm8
-#define VADDVX_UINT vaddvx_uint32xm8
-#define VMVVX_UINT vmvvx_uint32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFMINVV_FLOAT vfmin_vv_f32m8
+#define VMFLEVF_FLOAT vmfle_vf_f32m8_b4
+#define VMFIRSTM vmfirst_m_b4
+#define UINT_V_T vuint32m8_t
+#define VIDV_MASK_UINT vid_v_u32m8_m
+#define VIDV_UINT vid_v_u32m8
+#define VADDVX_MASK_UINT vadd_vx_u32m8_m
+#define VADDVX_UINT vadd_vx_u32m8
+#define VMVVX_UINT vmv_v_x_u32m8
 #endif
 
 
@@ -85,15 +89,20 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         UINT_V_T v_min_index;
         MASK_T mask;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_max;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_max = VFMVVF_FLOAT_M1(FLT_MAX, gvl);
+
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 v_min = VFMVVF_FLOAT(FLT_MAX, gvl);
                 v_min_index = VMVVX_UINT(0, gvl);
                 for(i=0,j=0; i < n/gvl; i++){
                         vx = VLEV_FLOAT(&x[j], gvl);
                         //index where element less than v_min
                         mask = VMFLTVV_FLOAT(vx, v_min, gvl);
-                        v_min_index = VIDV_MASK_UINT(v_min_index, mask, gvl);
+                        v_min_index = VIDV_MASK_UINT(mask, v_min_index, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -113,26 +122,24 @@ asm volatile(
         :"v0");
 #endif
 */
-                        v_min_index = VADDVX_MASK_UINT(v_min_index, v_min_index, j, mask, gvl);
+                        v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j,gvl);
 
                         //update v_min and start_index j
                         v_min = VFMINVV_FLOAT(v_min, vx, gvl);
                         j += gvl;
                 }
-                vx = VFMVVF_FLOAT(FLT_MAX, gvl);
-                vx = VFREDMINVS_FLOAT(v_min, vx, gvl);
-                minf = vx[0];
+                v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                minf = v_res[0];
                 mask = VMFLEVF_FLOAT(v_min, minf, gvl);
                 min_index = VMFIRSTM(mask,gvl);
                 min_index = v_min_index[min_index];
 
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v_min = VLEV_FLOAT(&x[j], gvl);
 
-                        vx = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        vx = VFREDMINVS_FLOAT(v_min, vx, gvl);
-                        FLOAT cur_minf = vx[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                        FLOAT cur_minf = v_res[0];
                         if(cur_minf < minf){
                                 //tail index
                                 v_min_index = VIDV_UINT(gvl);
@@ -143,7 +150,7 @@ asm volatile(
                         }
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
                 unsigned int idx = 0, inc_v = gvl * inc_x;
 
@@ -154,7 +161,7 @@ asm volatile(
 
                         //index where element less than v_min
                         mask = VMFLTVV_FLOAT(vx, v_min, gvl);
-                        v_min_index = VIDV_MASK_UINT(v_min_index, mask, gvl);
+                        v_min_index = VIDV_MASK_UINT(mask, v_min_index, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -175,27 +182,25 @@ asm volatile(
 #endif
 */
 
-                        v_min_index = VADDVX_MASK_UINT(v_min_index, v_min_index, j, mask, gvl);
+                        v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j,gvl);
 
                         //update v_min and start_index j
                         v_min = VFMINVV_FLOAT(v_min, vx, gvl);
                         j += gvl;
                         idx += inc_v;
                 }
-                vx = VFMVVF_FLOAT(FLT_MAX, gvl);
-                vx = VFREDMINVS_FLOAT(v_min, vx, gvl);
-                minf = vx[0];
+                v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                minf = v_res[0];
                 mask = VMFLEVF_FLOAT(v_min, minf, gvl);
                 min_index = VMFIRSTM(mask,gvl);
                 min_index = v_min_index[min_index];
 
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v_min = VLSEV_FLOAT(&x[idx], stride_x, gvl);
 
-                        vx = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        vx = VFREDMINVS_FLOAT(v_min, vx, gvl);
-                        FLOAT cur_minf = vx[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                        FLOAT cur_minf = v_res[0];
                         if(cur_minf < minf){
                                 //tail index
                                 v_min_index = VIDV_UINT(gvl);
diff --git a/kernel/riscv64/izamax_vector.c b/kernel/riscv64/izamax_vector.c
index 62c95d973..ddb5eabde 100644
--- a/kernel/riscv64/izamax_vector.c
+++ b/kernel/riscv64/izamax_vector.c
@@ -30,47 +30,53 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(DOUBLE)
 
-#define RVV_EFLOAT RVV_E64
-#define FLOAT_V_T float64xm8_t
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VMFLTVV_FLOAT vmfltvv_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFMAXVV_FLOAT vfmaxvv_float64xm8
-#define VMFGEVF_FLOAT vmfgevf_e64xm8_float64xm8
-#define VMFIRSTM vmfirstm_e64xm8
-#define UINT_V_T uint64xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint64xm8
-#define VIDV_UINT vidv_uint64xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint64xm8
-#define VADDVX_UINT vaddvx_uint64xm8
-#define VFADDVV_FLOAT vfaddvv_float64xm8
-#define VMVVX_UINT vmvvx_uint64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFMAXVV_FLOAT vfmax_vv_f64m8
+#define VMFGEVF_FLOAT vmfge_vf_f64m8_b8
+#define VMFIRSTM vmfirst_m_b8
+#define UINT_V_T vuint64m8_t
+#define VIDV_MASK_UINT vid_v_u64m8_m
+#define VIDV_UINT vid_v_u64m8
+#define VADDVX_MASK_UINT vadd_vx_u64m8_m
+#define VADDVX_UINT vadd_vx_u64m8
+#define VFADDVV_FLOAT vfadd_vv_f64m8
+#define VMVVX_UINT vmv_v_x_u64m8
 #else
 
 #define ABS fabsf
-#define RVV_EFLOAT RVV_E32
-#define FLOAT_V_T float32xm8_t
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VMFLTVV_FLOAT vmfltvv_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFMAXVV_FLOAT vfmaxvv_float32xm8
-#define VMFGEVF_FLOAT vmfgevf_e32xm8_float32xm8
-#define VMFIRSTM vmfirstm_e32xm8
-#define UINT_V_T uint32xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint32xm8
-#define VIDV_UINT vidv_uint32xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint32xm8
-#define VADDVX_UINT vaddvx_uint32xm8
-#define VFADDVV_FLOAT vfaddvv_float32xm8
-#define VMVVX_UINT vmvvx_uint32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFMAXVV_FLOAT vfmax_vv_f32m8
+#define VMFGEVF_FLOAT vmfge_vf_f32m8_b4
+#define VMFIRSTM vmfirst_m_b4
+#define UINT_V_T vuint32m8_t
+#define VIDV_MASK_UINT vid_v_u32m8_m
+#define VIDV_UINT vid_v_u32m8
+#define VADDVX_MASK_UINT vadd_vx_u32m8_m
+#define VADDVX_UINT vadd_vx_u32m8
+#define VFADDVV_FLOAT vfadd_vv_f32m8
+#define VMVVX_UINT vmv_v_x_u32m8
 #endif
 
 #define RVV_M RVV_M8
@@ -86,7 +92,12 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         UINT_V_T v_max_index;
         MASK_T mask0, mask1;
         unsigned int gvl = 0;
-        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
+
+        gvl = VSETVL(n);
         v_max_index = VMVVX_UINT(0, gvl);
         v_max = VFMVVF_FLOAT(-1, gvl);
         BLASLONG stride_x = inc_x * 2 * sizeof(FLOAT);
@@ -96,7 +107,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                 vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 //fabs(vector)
                 mask0 = VMFLTVF_FLOAT(vx0, 0, gvl);
-                vx0 = VFRSUBVF_MASK_FLOAT(vx0, vx0, 0, mask0, gvl);
+                vx0 = VFRSUBVF_MASK_FLOAT(mask0, vx0, vx0, 0, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -119,7 +130,7 @@ asm volatile(
                 vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 //fabs(vector)
                 mask1 = VMFLTVF_FLOAT(vx1, 0, gvl);
-                vx1 = VFRSUBVF_MASK_FLOAT(vx1, vx1, 0, mask1, gvl);
+                vx1 = VFRSUBVF_MASK_FLOAT(mask1, vx1, vx1, 0, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -143,7 +154,7 @@ asm volatile(
 
                 //index where element greater than v_max
                 mask0 = VMFLTVV_FLOAT(v_max, vx0, gvl);
-                v_max_index = VIDV_MASK_UINT(v_max_index, mask0, gvl);
+                v_max_index = VIDV_MASK_UINT(mask0, v_max_index, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -163,7 +174,7 @@ asm volatile(
         :"v0");
 #endif
 */
-                v_max_index = VADDVX_MASK_UINT(v_max_index, v_max_index, j, mask0, gvl);
+                v_max_index = VADDVX_MASK_UINT(mask0, v_max_index, v_max_index, j, gvl);
 
                 //update v_max and start_index j
                 v_max = VFMAXVV_FLOAT(v_max, vx0, gvl);
@@ -171,19 +182,19 @@ asm volatile(
                 ix += inc_xv;
         }
         vx0 = VFMVVF_FLOAT(0, gvl);
-        vx0 = VFREDMAXVS_FLOAT(v_max, vx0, gvl);
-        maxf = vx0[0];
+        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, gvl);
+        maxf = v_res[0];
         mask0 = VMFGEVF_FLOAT(v_max, maxf, gvl);
         max_index = VMFIRSTM(mask0,gvl);
         max_index = v_max_index[max_index];
 
         if(j < n){
-                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n-j);
                 v_max_index = VMVVX_UINT(0, gvl);
                 vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 //fabs(vector)
                 mask0 = VMFLTVF_FLOAT(vx0, 0, gvl);
-                vx0 = VFRSUBVF_MASK_FLOAT(vx0, vx0, 0, mask0, gvl);
+                vx0 = VFRSUBVF_MASK_FLOAT(mask0, vx0, vx0, 0, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -206,7 +217,7 @@ asm volatile(
                 vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 //fabs(vector)
                 mask1 = VMFLTVF_FLOAT(vx1, 0, gvl);
-                vx1 = VFRSUBVF_MASK_FLOAT(vx1, vx1, 0, mask1, gvl);
+                vx1 = VFRSUBVF_MASK_FLOAT(mask1, vx1, vx1, 0, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -227,9 +238,8 @@ asm volatile(
 #endif
 */
                 v_max = VFADDVV_FLOAT(vx0, vx1, gvl);
-                vx0 = VFMVVF_FLOAT(0, gvl);
-                vx0 = VFREDMAXVS_FLOAT(v_max, vx0, gvl);
-                FLOAT cur_maxf = vx0[0];
+                v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, gvl);
+                FLOAT cur_maxf = v_res[0];
                 if(cur_maxf > maxf){
                         //tail index
                         v_max_index = VIDV_UINT(gvl);
diff --git a/kernel/riscv64/izamin_vector.c b/kernel/riscv64/izamin_vector.c
index 38eccf1b5..6e328dc31 100644
--- a/kernel/riscv64/izamin_vector.c
+++ b/kernel/riscv64/izamin_vector.c
@@ -31,50 +31,55 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(DOUBLE)
 
-#define RVV_EFLOAT RVV_E64
-#define FLOAT_V_T float64xm8_t
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMINVS_FLOAT vfredminvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VMFLTVV_FLOAT vmfltvv_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFMINVV_FLOAT vfminvv_float64xm8
-#define VMFLEVF_FLOAT vmflevf_e64xm8_float64xm8
-#define VMFIRSTM vmfirstm_e64xm8
-#define UINT_V_T uint64xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint64xm8
-#define VIDV_UINT vidv_uint64xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint64xm8
-#define VADDVX_UINT vaddvx_uint64xm8
-#define VFADDVV_FLOAT vfaddvv_float64xm8
-#define VMVVX_UINT vmvvx_uint64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VMFLTVV_FLOAT vmflt_vv_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFMINVV_FLOAT vfmin_vv_f64m8
+#define VMFLEVF_FLOAT vmfle_vf_f64m8_b8
+#define VMFIRSTM vmfirst_m_b8
+#define UINT_V_T vuint64m8_t
+#define VIDV_MASK_UINT vid_v_u64m8_m
+#define VIDV_UINT vid_v_u64m8
+#define VADDVX_MASK_UINT vadd_vx_u64m8_m
+#define VADDVX_UINT vadd_vx_u64m8
+#define VFADDVV_FLOAT vfadd_vv_f64m8
+#define VMVVX_UINT vmv_v_x_u64m8
 #else
 
 #define ABS fabsf
-#define RVV_EFLOAT RVV_E32
-#define FLOAT_V_T float32xm8_t
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMINVS_FLOAT vfredminvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VMFLTVV_FLOAT vmfltvv_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFMINVV_FLOAT vfminvv_float32xm8
-#define VMFLEVF_FLOAT vmflevf_e32xm8_float32xm8
-#define VMFIRSTM vmfirstm_e32xm8
-#define UINT_V_T uint32xm8_t
-#define VIDV_MASK_UINT vidv_mask_uint32xm8
-#define VIDV_UINT vidv_uint32xm8
-#define VADDVX_MASK_UINT vaddvx_mask_uint32xm8
-#define VADDVX_UINT vaddvx_uint32xm8
-#define VFADDVV_FLOAT vfaddvv_float32xm8
-#define VMVVX_UINT vmvvx_uint32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VMFLTVV_FLOAT vmflt_vv_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFMINVV_FLOAT vfmin_vv_f32m8
+#define VMFLEVF_FLOAT vmfle_vf_f32m8_b4
+#define VMFIRSTM vmfirst_m_b4
+#define UINT_V_T vuint32m8_t
+#define VIDV_MASK_UINT vid_v_u32m8_m
+#define VIDV_UINT vid_v_u32m8
+#define VADDVX_MASK_UINT vadd_vx_u32m8_m
+#define VADDVX_UINT vadd_vx_u32m8
+#define VFADDVV_FLOAT vfadd_vv_f32m8
+#define VMVVX_UINT vmv_v_x_u32m8
 #endif
 
-#define RVV_M RVV_M8
 
 BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 {
@@ -87,7 +92,12 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         UINT_V_T v_min_index;
         MASK_T mask0, mask1;
         unsigned int gvl = 0;
-        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+        FLOAT_V_T_M1 v_res, v_max;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_max = VFMVVF_FLOAT_M1(FLT_MAX, gvl);
+
+        gvl = VSETVL(n);
         v_min_index = VMVVX_UINT(0, gvl);
         v_min = VFMVVF_FLOAT(FLT_MAX, gvl);
         BLASLONG stride_x = inc_x * 2 * sizeof(FLOAT);
@@ -97,7 +107,7 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                 vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 //fabs(vector)
                 mask0 = VMFLTVF_FLOAT(vx0, 0, gvl);
-                vx0 = VFRSUBVF_MASK_FLOAT(vx0, vx0, 0, mask0, gvl);
+                vx0 = VFRSUBVF_MASK_FLOAT(mask0, vx0, vx0, 0, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -120,7 +130,7 @@ asm volatile(
                 vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 //fabs(vector)
                 mask1 = VMFLTVF_FLOAT(vx1, 0, gvl);
-                vx1 = VFRSUBVF_MASK_FLOAT(vx1, vx1, 0, mask1, gvl);
+                vx1 = VFRSUBVF_MASK_FLOAT(mask1, vx1, vx1, 0, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -144,7 +154,7 @@ asm volatile(
 
                 //index where element less than v_min
                 mask0 = VMFLTVV_FLOAT(vx0, v_min, gvl);
-                v_min_index = VIDV_MASK_UINT(v_min_index, mask0, gvl);
+                v_min_index = VIDV_MASK_UINT(mask0, v_min_index, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -164,27 +174,26 @@ asm volatile(
         :"v0");
 #endif
 */
-                v_min_index = VADDVX_MASK_UINT(v_min_index, v_min_index, j, mask0, gvl);
+                v_min_index = VADDVX_MASK_UINT(mask0, v_min_index, v_min_index, j, gvl);
 
                 //update v_min and start_index j
                 v_min = VFMINVV_FLOAT(v_min, vx0, gvl);
                 j += gvl;
                 ix += inc_xv;
         }
-        vx0 = VFMVVF_FLOAT(FLT_MAX, gvl);
-        vx0 = VFREDMINVS_FLOAT(v_min, vx0, gvl);
-        minf = vx0[0];
+        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+        minf = v_res[0];
         mask0 = VMFLEVF_FLOAT(v_min, minf, gvl);
         min_index = VMFIRSTM(mask0,gvl);
         min_index = v_min_index[min_index];
 
         if(j < n){
-                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n-j);
                 v_min_index = VMVVX_UINT(0, gvl);
                 vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 //fabs(vector)
                 mask0 = VMFLTVF_FLOAT(vx0, 0, gvl);
-                vx0 = VFRSUBVF_MASK_FLOAT(vx0, vx0, 0, mask0, gvl);
+                vx0 = VFRSUBVF_MASK_FLOAT(mask0, vx0, vx0, 0, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -207,7 +216,7 @@ asm volatile(
                 vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 //fabs(vector)
                 mask1 = VMFLTVF_FLOAT(vx1, 0, gvl);
-                vx1 = VFRSUBVF_MASK_FLOAT(vx1, vx1, 0, mask1, gvl);
+                vx1 = VFRSUBVF_MASK_FLOAT(mask1, vx1, vx1, 0, gvl);
 /*
 #if defined(DOUBLE)
 asm volatile(
@@ -228,9 +237,8 @@ asm volatile(
 #endif
 */
                 v_min = VFADDVV_FLOAT(vx0, vx1, gvl);
-                vx0 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                vx0 = VFREDMINVS_FLOAT(v_min, vx0, gvl);
-                FLOAT cur_minf = vx0[0];
+                v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                FLOAT cur_minf = v_res[0];
                 if(cur_minf < minf){
                         //tail index
                         v_min_index = VIDV_UINT(gvl);
diff --git a/kernel/riscv64/max_vector.c b/kernel/riscv64/max_vector.c
index 4ef75452d..0fc59b74c 100644
--- a/kernel/riscv64/max_vector.c
+++ b/kernel/riscv64/max_vector.c
@@ -29,23 +29,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <math.h>
 #include <float.h>
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFMAXVV_FLOAT vfmaxvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFMAXVV_FLOAT vfmax_vv_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFMAXVV_FLOAT vfmaxvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFMAXVV_FLOAT vfmax_vv_f64m8
 #endif
 
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@@ -55,9 +59,13 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	FLOAT maxf=-FLT_MAX;
         unsigned int gvl = 0;
         FLOAT_V_T v0, v1, v_max;
+        FLOAT_V_T_M1 v_res, v_min;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_min = VFMVVF_FLOAT_M1(-FLT_MAX, gvl);
 
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl <= n/2){
                         v_max = VFMVVF_FLOAT(-FLT_MAX, gvl);
                         for(i=0,j=0; i<n/(gvl*2); i++){
@@ -68,21 +76,19 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 v_max = VFMAXVV_FLOAT(v_max, v1, gvl);
                                 j += gvl * 2;
                         }
-                        v1 = VFMVVF_FLOAT(-FLT_MAX, gvl);
-                        v0 = VFREDMAXVS_FLOAT(v_max, v1, gvl);
-                        maxf = v0[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_min, gvl);
+                        maxf = v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
-                        v1 = VFMVVF_FLOAT(-FLT_MAX, gvl);
-                        v0 = VFREDMAXVS_FLOAT(v0, v1, gvl);
-                        if(v0[0] > maxf)
-                                maxf = v0[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v0, v_min, gvl);
+                        if(v_res[0] > maxf)
+                                maxf = v_res[0];
                         j += gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * sizeof(FLOAT);
                 if(gvl <= n/2){
                         v_max = VFMVVF_FLOAT(-FLT_MAX, gvl);
@@ -96,17 +102,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 j += gvl * 2;
                                 idx += inc_xv * 2;
                         }
-                        v1 = VFMVVF_FLOAT(-FLT_MAX, gvl);
-                        v0 = VFREDMAXVS_FLOAT(v_max, v1, gvl);
-                        maxf = v0[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_min, gvl);
+                        maxf = v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
-                        v1 = VFMVVF_FLOAT(-FLT_MAX, gvl);
-                        v0 = VFREDMAXVS_FLOAT(v0, v1, gvl);
-                        if(v0[0] > maxf)
-                                maxf = v0[0];
+                        v_res = VFREDMAXVS_FLOAT(v_res, v0, v_min, gvl);
+                        if(v_res[0] > maxf)
+                                maxf = v_res[0];
                         j += gvl;
                 }
         }
diff --git a/kernel/riscv64/min_vector.c b/kernel/riscv64/min_vector.c
index 83c965bfa..8223fa87a 100644
--- a/kernel/riscv64/min_vector.c
+++ b/kernel/riscv64/min_vector.c
@@ -29,23 +29,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <math.h>
 #include <float.h>
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMINVS_FLOAT vfredminvs_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFMINVV_FLOAT vfminvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFMINVV_FLOAT vfmin_vv_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMINVS_FLOAT vfredminvs_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFMINVV_FLOAT vfminvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFMINVV_FLOAT vfmin_vv_f64m8
 #endif
 
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@@ -55,9 +59,13 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	FLOAT minf=FLT_MAX;
         unsigned int gvl = 0;
         FLOAT_V_T v0, v1, v_min;
+        FLOAT_V_T_M1 v_res, v_max;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_max = VFMVVF_FLOAT_M1(FLT_MAX, gvl);
 
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl <= n/2){
                         v_min = VFMVVF_FLOAT(FLT_MAX, gvl);
                         for(i=0,j=0; i<n/(gvl*2); i++){
@@ -68,21 +76,19 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 v_min = VFMINVV_FLOAT(v_min, v1, gvl);
                                 j += gvl * 2;
                         }
-                        v1 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        v0 = VFREDMINVS_FLOAT(v_min, v1, gvl);
-                        minf = v0[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                        minf = v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
-                        v1 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        v0 = VFREDMINVS_FLOAT(v0, v1, gvl);
-                        if(v0[0] < minf)
-                                minf = v0[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v0, v_max, gvl);
+                        if(v_res[0] < minf)
+                                minf = v_res[0];
                         j += gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * sizeof(FLOAT);
                 if(gvl <= n/2){
                         v_min = VFMVVF_FLOAT(FLT_MAX, gvl);
@@ -96,17 +102,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 j += gvl * 2;
                                 idx += inc_xv * 2;
                         }
-                        v1 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        v0 = VFREDMINVS_FLOAT(v_min, v1, gvl);
-                        minf = v0[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+                        minf = v_res[0];
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
-                        v1 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                        v0 = VFREDMINVS_FLOAT(v0, v1, gvl);
-                        if(v0[0] < minf)
-                                minf = v0[0];
+                        v_res = VFREDMINVS_FLOAT(v_res, v0, v_max, gvl);
+                        if(v_res[0] < minf)
+                                minf = v_res[0];
                         j += gvl;
                 }
         }
diff --git a/kernel/riscv64/nrm2_vector.c b/kernel/riscv64/nrm2_vector.c
index 785c0d2f8..c461f410c 100644
--- a/kernel/riscv64/nrm2_vector.c
+++ b/kernel/riscv64/nrm2_vector.c
@@ -27,41 +27,45 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFDOTVV_FLOAT vfdotvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFDOTVV_FLOAT vfdot_vv_f32m4
 #define ABS fabsf
-#define MASK_T e32xm4_t
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm4
-#define VMFGTVF_FLOAT vmfgtvf_e32xm4_float32xm4
-#define VMFIRSTM vmfirstm_e32xm4
-#define VFDIVVF_FLOAT vfdivvf_float32xm4
-#define VMFLTVF_FLOAT vmfltvf_e32xm4_float32xm4
-#define VFREDMAXVS_FLOAT vfredmaxvs_float32xm4
+#define MASK_T vbool8_t
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m4_m
+#define VMFGTVF_FLOAT vmfgt_vf_f32m4_b8
+#define VMFIRSTM vmfirst_m_b8
+#define VFDIVVF_FLOAT vfdiv_vf_f32m4
+#define VMFLTVF_FLOAT vmflt_vf_f32m4_b8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f32m4_f32m1
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFDOTVV_FLOAT vfdotvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFDOTVV_FLOAT vfdot_vv_f64m4
 #define ABS fabs
-#define MASK_T e64xm4_t
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm4
-#define VMFGTVF_FLOAT vmfgtvf_e64xm4_float64xm4
-#define VMFIRSTM vmfirstm_e64xm4
-#define VFDIVVF_FLOAT vfdivvf_float64xm4
-#define VMFLTVF_FLOAT vmfltvf_e64xm4_float64xm4
-#define VFREDMAXVS_FLOAT vfredmaxvs_float64xm4
+#define MASK_T vbool16_t
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m4_m
+#define VMFGTVF_FLOAT vmfgt_vf_f64m4_b16
+#define VMFIRSTM vmfirst_m_b16
+#define VFDIVVF_FLOAT vfdiv_vf_f64m4
+#define VMFLTVF_FLOAT vmflt_vf_f64m4_b16
+#define VFREDMAXVS_FLOAT vfredmax_vs_f64m4_f64m1
 #endif
 
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@@ -73,18 +77,23 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 
         FLOAT_V_T vr, v0, v_zero;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
+
         FLOAT scale = 0.0, ssq = 0.0;
         MASK_T mask;
         BLASLONG index = 0;
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 vr = VFMVVF_FLOAT(0, gvl);
                 v_zero = VFMVVF_FLOAT(0, gvl);
                 for(i=0,j=0; i<n/gvl; i++){
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -95,15 +104,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 }
                         }else{//found greater element
                                 //ssq in vector vr: vr[0]
-                                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                                 //total ssq before current vector
-                                ssq += vr[0];
+                                ssq += v_res[0];
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 //ssq in vector vr
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                                 vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
@@ -111,17 +120,17 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         j += gvl;
                 }
                 //ssq in vector vr: vr[0]
-                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                 //total ssq now
-                ssq += vr[0];
+                ssq += v_res[0];
 
                 //tail
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -130,21 +139,21 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                         v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                         }else{//found greater element
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                         }
                         vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
                         //ssq in vector vr: vr[0]
-                        vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                         //total ssq now
-                        ssq += vr[0];
+                        ssq += v_res[0];
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 vr = VFMVVF_FLOAT(0, gvl);
                 v_zero = VFMVVF_FLOAT(0, gvl);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
@@ -153,7 +162,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         v0 = VLSEV_FLOAT(&x[idx], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -164,15 +173,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 }
                         }else{//found greater element
                                 //ssq in vector vr: vr[0]
-                                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                                 //total ssq before current vector
-                                ssq += vr[0];
+                                ssq += v_res[0];
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 //ssq in vector vr
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                                 vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
@@ -181,17 +190,17 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         idx += inc_v;
                 }
                 //ssq in vector vr: vr[0]
-                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                 //total ssq now
-                ssq += vr[0];
+                ssq += v_res[0];
 
                 //tail
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[idx], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -200,18 +209,18 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                         v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                         }else{//found greater element
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
                                 scale = vr[0];
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                         }
                         vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
                         //ssq in vector vr: vr[0]
-                        vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                         //total ssq now
-                        ssq += vr[0];
+                        ssq += v_res[0];
                 }
         }
 	return(scale * sqrt(ssq));
diff --git a/kernel/riscv64/nrm2_vector_dot.c b/kernel/riscv64/nrm2_vector_dot.c
index a3d15406c..62dfe97d6 100644
--- a/kernel/riscv64/nrm2_vector_dot.c
+++ b/kernel/riscv64/nrm2_vector_dot.c
@@ -27,26 +27,30 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDSUM_FLOAT vfredsumvs_float32xm8
-#define VFMACCVV_FLOAT vfmaccvv_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFDOTVV_FLOAT vfdotvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDSUM_FLOAT vfredsum_vs_f32m8_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m8
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFDOTVV_FLOAT vfdot_vv_f32m8
 #define ABS fabsf
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDSUM_FLOAT vfredsumvs_float64xm8
-#define VFMACCVV_FLOAT vfmaccvv_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFDOTVV_FLOAT vfdotvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDSUM_FLOAT vfredsum_vs_f64m8_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFDOTVV_FLOAT vfdot_vv_f64m8
 #define ABS fabs
 #endif
 
@@ -60,8 +64,13 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 
         FLOAT_V_T vr, v0, v1;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
+
         if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl < n/2){
                         vr = VFMVVF_FLOAT(0, gvl);
                         for(i=0,j=0; i<n/(2*gvl); i++){
@@ -73,25 +82,24 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 vr = VFMACCVV_FLOAT(vr, v1, v1, gvl);
                                 j += gvl;
                         }
-                        v0 = VFMVVF_FLOAT(0, gvl);
-                        v0 = VFREDSUM_FLOAT(vr, v0, gvl);
-                        len += v0[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        len += v_res[0];
                 }
                 //tail
                 for(;j < n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         //v1 = 0
-                        v1 = VFMVVF_FLOAT(0, gvl);
+                        //v1 = VFMVVF_FLOAT(0, gvl);
                         //vr = VFDOTVV_FLOAT(v0, v0, gvl);
                         vr = VFMACCVV_FLOAT(v1, v0, v0, gvl);
-                        v0 = VFREDSUM_FLOAT(vr, v1, gvl);
-                        len += v0[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        len += v_res[0];
 
                         j += gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT);
                 if(gvl < n/2){
                         vr = VFMVVF_FLOAT(0, gvl);
@@ -104,20 +112,19 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 vr = VFMACCVV_FLOAT(vr, v1, v1, gvl);
                                 j += gvl;
                         }
-                        v0 = VFMVVF_FLOAT(0, gvl);
-                        v0 = VFREDSUM_FLOAT(vr, v0, gvl);
-                        len += v0[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        len += v_res[0];
                 }
                 //tail
                 for(;j < n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         //v1 = 0
-                        v1 = VFMVVF_FLOAT(0, gvl);
+                        //v1 = VFMVVF_FLOAT(0, gvl);
                         //vr = VFDOTVV_FLOAT(v0, v0, gvl);
                         vr = VFMACCVV_FLOAT(v1, v0, v0, gvl);
-                        v0 = VFREDSUM_FLOAT(vr, v1, gvl);
-                        len += v0[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        len += v_res[0];
 
                         j += gvl;
                 }
diff --git a/kernel/riscv64/rot_vector.c b/kernel/riscv64/rot_vector.c
index aeabca1ba..9b48d1c69 100644
--- a/kernel/riscv64/rot_vector.c
+++ b/kernel/riscv64/rot_vector.c
@@ -28,27 +28,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSEV_FLOAT vsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMULVF_FLOAT vfmulvf_float32xm4
-#define VFMSACVF_FLOAT vfmsacvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSEV_FLOAT vse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMULVF_FLOAT vfmul_vf_f32m4
+#define VFMSACVF_FLOAT vfmsac_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSEV_FLOAT vsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMULVF_FLOAT vfmulvf_float64xm4
-#define VFMSACVF_FLOAT vfmsacvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSEV_FLOAT vse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMULVF_FLOAT vfmul_vf_f64m4
+#define VFMSACVF_FLOAT vfmsac_vf_f64m4
 #endif
 
 int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s)
@@ -61,7 +61,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
         FLOAT_V_T v0, v1, vx, vy;
 
         if(inc_x == 1 && inc_y == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 for(i=0,j=0; i<n/gvl; i++){
                         vx = VLEV_FLOAT(&x[j], gvl);
                         vy = VLEV_FLOAT(&y[j], gvl);
@@ -77,7 +77,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
                         j += gvl;
                 }
                 if(j<n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLEV_FLOAT(&x[j], gvl);
                         vy = VLEV_FLOAT(&y[j], gvl);
 
@@ -90,7 +90,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
                         VSEV_FLOAT(&y[j], v1, gvl);
                 }
         }else if(inc_y == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * sizeof(FLOAT);
                 BLASLONG inc_xv = inc_x * gvl;
                 for(i=0,j=0; i<n/gvl; i++){
@@ -109,7 +109,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
                         ix += inc_xv;
                 }
                 if(j<n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLSEV_FLOAT(&x[j*inc_x], stride_x, gvl);
                         vy = VLEV_FLOAT(&y[j], gvl);
 
@@ -122,7 +122,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
                         VSEV_FLOAT(&y[j], v1, gvl);
                 }
         }else if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_y = inc_y * sizeof(FLOAT);
                 BLASLONG inc_yv = inc_y * gvl;
                 for(i=0,j=0; i<n/gvl; i++){
@@ -141,7 +141,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
                         iy += inc_yv;
                 }
                 if(j<n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLEV_FLOAT(&x[j], gvl);
                         vy = VLSEV_FLOAT(&y[j*inc_y],stride_y, gvl);
 
@@ -154,7 +154,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
                         VSSEV_FLOAT(&y[j*inc_y], stride_y, v1, gvl);
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * sizeof(FLOAT);
                 BLASLONG stride_y = inc_y * sizeof(FLOAT);
                 BLASLONG inc_xv = inc_x * gvl;
@@ -176,7 +176,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
                         iy += inc_yv;
                 }
                 if(j<n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx = VLSEV_FLOAT(&x[j*inc_x],stride_x, gvl);
                         vy = VLSEV_FLOAT(&y[j*inc_y],stride_y, gvl);
 
diff --git a/kernel/riscv64/scal_vector.c b/kernel/riscv64/scal_vector.c
index 5152eea06..e554fd040 100644
--- a/kernel/riscv64/scal_vector.c
+++ b/kernel/riscv64/scal_vector.c
@@ -27,25 +27,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VSEV_FLOAT vsev_float32xm8
-#define VSSEV_FLOAT vssev_float32xm8
-#define VFMULVF_FLOAT vfmulvf_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VSEV_FLOAT vse_v_f32m8
+#define VSSEV_FLOAT vsse_v_f32m8
+#define VFMULVF_FLOAT vfmul_vf_f32m8
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VSEV_FLOAT vsev_float64xm8
-#define VSSEV_FLOAT vssev_float64xm8
-#define VFMULVF_FLOAT vfmulvf_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VSEV_FLOAT vse_v_f64m8
+#define VSSEV_FLOAT vsse_v_f64m8
+#define VFMULVF_FLOAT vfmul_vf_f64m8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
 #endif
 
 int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
@@ -61,7 +61,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
                 if(da == 0.0){
                         memset(&x[0], 0, n * sizeof(FLOAT));
                 }else{
-                        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n);
                         if(gvl <= n / 2){
                                 for(i = 0, j = 0; i < n/(2*gvl); i++, j+=2*gvl){
                                         v0 = VLEV_FLOAT(&x[j], gvl);
@@ -75,7 +75,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
                         }
                         //tail
                         for(; j <n; ){
-                                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n-j);
                                 v0 = VLEV_FLOAT(&x[j], gvl);
                                 v0 = VFMULVF_FLOAT(v0, da, gvl);
                                 VSEV_FLOAT(&x[j], v0, gvl);
@@ -84,7 +84,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
                 }
         }else{
                 if(da == 0.0){
-                        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n);
                         if(gvl <= n / 2){
                                 v0 = VFMVVF_FLOAT(0, gvl);
                                 for(i = 0, j = 0; i < n/(2*gvl); i++, j+=2*gvl){
@@ -94,13 +94,13 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
                         }
                         //tail
                         for(; j <n; ){
-                                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n-j);
                                 v0 = VFMVVF_FLOAT(0, gvl);
                                 VSEV_FLOAT(&x[j], v0, gvl);
                                 j += gvl;
                         }
                 }else{
-                        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n);
                         BLASLONG stride_x = inc_x * sizeof(FLOAT);
                         BLASLONG ix = 0;
                         if(gvl < n / 2){
@@ -118,7 +118,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
                         }
                         //tail
                         for(; j <n; ){
-                                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n-j);
                                 v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                 v0 = VFMULVF_FLOAT(v0, da, gvl);
                                 VSSEV_FLOAT(&x[ix], stride_x, v0, gvl);
diff --git a/kernel/riscv64/swap_vector.c b/kernel/riscv64/swap_vector.c
index 9377bf4b9..d9421e2f1 100644
--- a/kernel/riscv64/swap_vector.c
+++ b/kernel/riscv64/swap_vector.c
@@ -28,21 +28,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 #include <stdio.h>
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VSEV_FLOAT vsev_float32xm8
-#define VSSEV_FLOAT vssev_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VSEV_FLOAT vse_v_f32m8
+#define VSSEV_FLOAT vsse_v_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VSEV_FLOAT vsev_float64xm8
-#define VSSEV_FLOAT vssev_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VSEV_FLOAT vse_v_f64m8
+#define VSSEV_FLOAT vsse_v_f64m8
 #endif
 
 int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
@@ -55,7 +55,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
 
 	if (n < 0)  return(0);
         if(inc_x == 1 && inc_y == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 if(gvl <= n/2){
                         for(i=0,j=0; i<n/(2*gvl); i++){
                                 vx0 = VLEV_FLOAT(&x[j], gvl);
@@ -71,7 +71,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
                         }
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx0 = VLEV_FLOAT(&x[j], gvl);
                         vy0 = VLEV_FLOAT(&y[j], gvl);
                         VSEV_FLOAT(&x[j], vy0, gvl);
@@ -79,7 +79,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
                         j+=gvl;
                 }
         }else if (inc_y == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 stride_x = inc_x * sizeof(FLOAT);
                 if(gvl <= n/2){
                         BLASLONG inc_xv = inc_x * gvl;
@@ -98,7 +98,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
                         }
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         vy0 = VLEV_FLOAT(&y[j], gvl);
                         VSSEV_FLOAT(&x[ix], stride_x, vy0, gvl);
@@ -107,7 +107,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
                         ix += inc_x * gvl;
                 }
         }else if(inc_x == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 stride_y = inc_y * sizeof(FLOAT);
                 if(gvl <= n/2){
                         BLASLONG inc_yv = inc_y * gvl;
@@ -126,7 +126,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
                         }
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx0 = VLEV_FLOAT(&x[j], gvl);
                         vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
                         VSEV_FLOAT(&x[j], vy0, gvl);
@@ -135,7 +135,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
                         iy += inc_y * gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 stride_x = inc_x * sizeof(FLOAT);
                 stride_y = inc_y * sizeof(FLOAT);
                 if(gvl <= n/2){
@@ -157,7 +157,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT *x,
                         }
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
                         VSSEV_FLOAT(&x[ix], stride_x, vy0, gvl);
diff --git a/kernel/riscv64/symv_L_vector.c b/kernel/riscv64/symv_L_vector.c
index 3c2647026..de8917532 100644
--- a/kernel/riscv64/symv_L_vector.c
+++ b/kernel/riscv64/symv_L_vector.c
@@ -27,31 +27,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSEV_FLOAT vsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFMULVV_FLOAT vfmulvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSEV_FLOAT vse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFMULVV_FLOAT vfmul_vv_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSEV_FLOAT vsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFMULVV_FLOAT vfmulvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSEV_FLOAT vse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFMULVV_FLOAT vfmul_vv_f64m4
 #endif
 
 int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@@ -63,6 +67,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
         FLOAT temp2;
         FLOAT *a_ptr = a;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
 
         FLOAT_V_T va, vx, vy, vr;
         BLASLONG stride_x, stride_y, inc_xv, inc_yv, len;
@@ -76,7 +84,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                         i = j + 1;
                         len = m - i;
                         if(len > 0){
-                                gvl = vsetvli(len, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(len);
                                 vr = VFMVVF_FLOAT(0, gvl);
                                 for(k = 0; k < len / gvl; k++){
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
@@ -89,11 +97,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         i += gvl;
                                 }
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp2 = va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp2 = v_res[0];
                                 if(i < m){
-                                        gvl = vsetvli(m-i, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(m-i);
                                         vy = VLEV_FLOAT(&y[i], gvl);
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
                                         vy = VFMACCVF_FLOAT(vy, temp1, va, gvl);
@@ -101,9 +108,8 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         vx = VLEV_FLOAT(&x[i], gvl);
                                         vr = VFMULVV_FLOAT(vx, va, gvl);
-                                        va = VFMVVF_FLOAT(0, gvl);
-                                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                                        temp2 += va[0];
+                                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                        temp2 += v_res[0];
                                 }
 			}
                         y[j] += alpha * temp2;
@@ -121,7 +127,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                         i = j + 1;
                         len = m - i;
                         if(len > 0){
-                                gvl = vsetvli(len, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(len);
                                 inc_yv = inc_y * gvl;
                                 vr = VFMVVF_FLOAT(0, gvl);
                                 for(k = 0; k < len / gvl; k++){
@@ -136,11 +142,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                                         i += gvl;
                                         iy += inc_yv;
                                 }
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp2 = va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp2 = v_res[0];
                                 if(i < m){
-                                        gvl = vsetvli(m-i, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(m-i);
                                         vy = VLSEV_FLOAT(&y[iy], stride_y, gvl);
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
                                         vy = VFMACCVF_FLOAT(vy, temp1, va, gvl);
@@ -148,9 +153,8 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         vx = VLEV_FLOAT(&x[i], gvl);
                                         vr = VFMULVV_FLOAT(vx, va, gvl);
-                                        va = VFMVVF_FLOAT(0, gvl);
-                                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                                        temp2 += va[0];
+                                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                        temp2 += v_res[0];
                                 }
 			}
                         y[jy] += alpha * temp2;
@@ -169,7 +173,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                         i = j + 1;
                         len = m - i;
                         if(len > 0){
-                                gvl = vsetvli(len, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(len);
                                 vr = VFMVVF_FLOAT(0, gvl);
                                 inc_xv = inc_x * gvl;
                                 for(k = 0; k < len / gvl; k++){
@@ -184,11 +188,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                                         i += gvl;
                                         ix += inc_xv;
                                 }
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp2 = va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp2 = v_res[0];
                                 if(i < m){
-                                        gvl = vsetvli(m-i, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(m-i);
                                         vy = VLEV_FLOAT(&y[i], gvl);
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
                                         vy = VFMACCVF_FLOAT(vy, temp1, va, gvl);
@@ -196,9 +199,8 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         vx = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                         vr = VFMULVV_FLOAT(vx, va, gvl);
-                                        va = VFMVVF_FLOAT(0, gvl);
-                                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                                        temp2 += va[0];
+                                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                        temp2 += v_res[0];
                                 }
 			}
                         y[j] += alpha * temp2;
@@ -220,7 +222,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                         i = j + 1;
                         len = m - i;
                         if(len > 0){
-                                gvl = vsetvli(len, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(len);
                                 inc_xv = inc_x * gvl;
                                 inc_yv = inc_y * gvl;
                                 vr = VFMVVF_FLOAT(0, gvl);
@@ -237,11 +239,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                                         ix += inc_xv;
                                         iy += inc_yv;
                                 }
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp2 = va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp2 = v_res[0];
                                 if(i < m){
-                                        gvl = vsetvli(m-i, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(m-i);
                                         vy = VLSEV_FLOAT(&y[iy], stride_y, gvl);
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
                                         vy = VFMACCVF_FLOAT(vy, temp1, va, gvl);
@@ -249,9 +250,8 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         vx = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                         vr = VFMULVV_FLOAT(vx, va, gvl);
-                                        va = VFMVVF_FLOAT(0, gvl);
-                                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                                        temp2 += va[0];
+                                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                        temp2 += v_res[0];
                                 }
 			}
                         y[jy] += alpha * temp2;
diff --git a/kernel/riscv64/symv_U_vector.c b/kernel/riscv64/symv_U_vector.c
index 29e0e4b65..7229a48b1 100644
--- a/kernel/riscv64/symv_U_vector.c
+++ b/kernel/riscv64/symv_U_vector.c
@@ -27,33 +27,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSEV_FLOAT vsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFDOTVV_FLOAT vfdotvv_float32xm4
-#define VFMULVV_FLOAT vfmulvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSEV_FLOAT vse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFDOTVV_FLOAT vfdot_vv_f32m4
+#define VFMULVV_FLOAT vfmul_vv_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSEV_FLOAT vsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFDOTVV_FLOAT vfdotvv_float64xm4
-#define VFMULVV_FLOAT vfmulvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSEV_FLOAT vse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFDOTVV_FLOAT vfdot_vv_f64m4
+#define VFMULVV_FLOAT vfmul_vv_f64m4
 #endif
 
 int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@@ -65,6 +69,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
         FLOAT temp2;
         FLOAT *a_ptr = a;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
 
         FLOAT_V_T va, vx, vy, vr;
         BLASLONG stride_x, stride_y, inc_xv, inc_yv;
@@ -78,7 +86,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                         temp2 = 0.0;
                         if(j > 0){
                                 i = 0;
-                                gvl = vsetvli(j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(j);
                                 vr = VFMVVF_FLOAT(0, gvl);
                                 for(k = 0; k < j / gvl; k++){
                                         vy = VLEV_FLOAT(&y[i], gvl);
@@ -91,11 +99,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         i += gvl;
                                 }
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp2 = va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp2 = v_res[0];
                                 if(i < j){
-                                        gvl = vsetvli(j-i, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(j-i);
                                         vy = VLEV_FLOAT(&y[i], gvl);
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
                                         vy = VFMACCVF_FLOAT(vy, temp1, va, gvl);
@@ -103,9 +110,8 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         vx = VLEV_FLOAT(&x[i], gvl);
                                         vr = VFMULVV_FLOAT(vx, va, gvl);
-                                        va = VFMVVF_FLOAT(0, gvl);
-                                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                                        temp2 += va[0];
+                                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                        temp2 += v_res[0];
                                 }
                         }
                         y[j] += temp1 * a_ptr[j] + alpha * temp2;
@@ -122,7 +128,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                         if(j > 0){
                                 iy = 0;
                                 i = 0;
-                                gvl = vsetvli(j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(j);
                                 inc_yv = inc_y * gvl;
                                 vr = VFMVVF_FLOAT(0, gvl);
                                 for(k = 0; k < j / gvl; k++){
@@ -137,11 +143,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                                         i += gvl;
                                         iy += inc_yv;
                                 }
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp2 = va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp2 = v_res[0];
                                 if(i < j){
-                                        gvl = vsetvli(j-i, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(j-i);
                                         vy = VLSEV_FLOAT(&y[iy], stride_y, gvl);
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
                                         vy = VFMACCVF_FLOAT(vy, temp1, va, gvl);
@@ -149,9 +154,8 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         vx = VLEV_FLOAT(&x[i], gvl);
                                         vr = VFMULVV_FLOAT(vx, va, gvl);
-                                        va = VFMVVF_FLOAT(0, gvl);
-                                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                                        temp2 += va[0];
+                                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                        temp2 += v_res[0];
                                 }
                         }
                         y[jy] += temp1 * a_ptr[j] + alpha * temp2;
@@ -169,7 +173,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                         if(j > 0){
                                 ix = 0;
                                 i = 0;
-                                gvl = vsetvli(j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(j);
                                 inc_xv = inc_x * gvl;
                                 vr = VFMVVF_FLOAT(0, gvl);
                                 for(k = 0; k < j / gvl; k++){
@@ -184,11 +188,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                                         i += gvl;
                                         ix += inc_xv;
                                 }
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp2 = va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp2 = v_res[0];
                                 if(i < j){
-                                        gvl = vsetvli(j-i, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(j-i);
                                         vy = VLEV_FLOAT(&y[i], gvl);
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
                                         vy = VFMACCVF_FLOAT(vy, temp1, va, gvl);
@@ -196,9 +199,8 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         vx = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                         vr = VFMULVV_FLOAT(vx, va, gvl);
-                                        va = VFMVVF_FLOAT(0, gvl);
-                                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                                        temp2 += va[0];
+                                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                        temp2 += v_res[0];
                                 }
                         }
                         y[j] += temp1 * a_ptr[j] + alpha * temp2;
@@ -219,7 +221,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                                 ix = 0;
                                 iy = 0;
                                 i = 0;
-                                gvl = vsetvli(j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(j);
                                 inc_xv = inc_x * gvl;
                                 inc_yv = inc_y * gvl;
                                 vr = VFMVVF_FLOAT(0, gvl);
@@ -236,11 +238,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
                                         ix += inc_xv;
                                         iy += inc_yv;
                                 }
-                                va = VFMVVF_FLOAT(0, gvl);
-                                va = VFREDSUM_FLOAT(vr, va, gvl);
-                                temp2 = va[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                temp2 = v_res[0];
                                 if(i < j){
-                                        gvl = vsetvli(j-i, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(j-i);
                                         vy = VLSEV_FLOAT(&y[iy], stride_y, gvl);
                                         va = VLEV_FLOAT(&a_ptr[i], gvl);
                                         vy = VFMACCVF_FLOAT(vy, temp1, va, gvl);
@@ -248,9 +249,8 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOA
 
                                         vx = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                         vr = VFMULVV_FLOAT(vx, va, gvl);
-                                        va = VFMVVF_FLOAT(0, gvl);
-                                        va = VFREDSUM_FLOAT(vr, va, gvl);
-                                        temp2 += va[0];
+                                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                                        temp2 += v_res[0];
                                 }
                         }
                         y[jy] += temp1 * a_ptr[j] + alpha * temp2;
diff --git a/kernel/riscv64/zamax_vector.c b/kernel/riscv64/zamax_vector.c
index a6c742b14..5cd65b225 100644
--- a/kernel/riscv64/zamax_vector.c
+++ b/kernel/riscv64/zamax_vector.c
@@ -29,29 +29,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <math.h>
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFMAXVV_FLOAT vfmaxvv_float32xm8
-#define VFADDVV_FLOAT vfaddvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFMAXVV_FLOAT vfmax_vv_f32m8
+#define VFADDVV_FLOAT vfadd_vv_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMAXVS_FLOAT vfredmaxvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFMAXVV_FLOAT vfmaxvv_float64xm8
-#define VFADDVV_FLOAT vfaddvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFMAXVV_FLOAT vfmax_vv_f64m8
+#define VFADDVV_FLOAT vfadd_vv_f64m8
 #endif
 
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@@ -62,19 +66,23 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	if (n <= 0 || inc_x <= 0) return(maxf);
         unsigned int gvl = 0;
         FLOAT_V_T v0, v1, v_max;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
 
         MASK_T mask0, mask1;
         BLASLONG stride_x = inc_x * sizeof(FLOAT) * 2;
-        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+        gvl = VSETVL(n);
         v_max = VFMVVF_FLOAT(0, gvl);
         BLASLONG inc_xv = inc_x * gvl * 2;
         for(; i<n/gvl; i++){
                 v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                 mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
 
                 v0 = VFADDVV_FLOAT(v0, v1, gvl);
                 v_max = VFMAXVV_FLOAT(v_max, v0, gvl);
@@ -82,23 +90,21 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                 j += gvl;
                 ix += inc_xv;
         }
-        v0 = VFMVVF_FLOAT(0, gvl);
-        v_max = VFREDMAXVS_FLOAT(v_max, v0, gvl);
-        maxf = v_max[0];
+        v_res = VFREDMAXVS_FLOAT(v_res, v_max, v_z0, gvl);
+        maxf = v_res[0];
 
         if(j<n){
-                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n-j);
                 v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                 mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
                 v1 = VFADDVV_FLOAT(v0, v1, gvl);
-                v0 = VFMVVF_FLOAT(0, gvl);
-                v_max = VFREDMAXVS_FLOAT(v1, v0, gvl);
-                if(v_max[0] > maxf)
-                        maxf = v_max[0];
+                v_res = VFREDMAXVS_FLOAT(v_res, v1, v_z0, gvl);
+                if(v_res[0] > maxf)
+                        maxf = v_res[0];
         }
         return(maxf);
 }
diff --git a/kernel/riscv64/zamin_vector.c b/kernel/riscv64/zamin_vector.c
index 44a7cf1dc..9d567b3da 100644
--- a/kernel/riscv64/zamin_vector.c
+++ b/kernel/riscv64/zamin_vector.c
@@ -30,29 +30,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <float.h>
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDMINVS_FLOAT vfredminvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFMINVV_FLOAT vfminvv_float32xm8
-#define VFADDVV_FLOAT vfaddvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFMINVV_FLOAT vfmin_vv_f32m8
+#define VFADDVV_FLOAT vfadd_vv_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDMINVS_FLOAT vfredminvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFMINVV_FLOAT vfminvv_float64xm8
-#define VFADDVV_FLOAT vfaddvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDMINVS_FLOAT vfredmin_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFMINVV_FLOAT vfmin_vv_f64m8
+#define VFADDVV_FLOAT vfadd_vv_f64m8
 #endif
 
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@@ -63,18 +67,23 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	FLOAT minf=FLT_MAX;
         unsigned int gvl = 0;
         FLOAT_V_T v0, v1, v_min;
+        FLOAT_V_T_M1 v_res, v_max;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_max = VFMVVF_FLOAT_M1(FLT_MAX, gvl);
+
         MASK_T mask0, mask1;
         BLASLONG stride_x = inc_x * sizeof(FLOAT) * 2;
-        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+        gvl = VSETVL(n);
         v_min = VFMVVF_FLOAT(FLT_MAX, gvl);
         BLASLONG inc_xv = inc_x * gvl * 2;
         for(; i<n/gvl; i++){
                 v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                 mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
 
                 v0 = VFADDVV_FLOAT(v0, v1, gvl);
                 v_min = VFMINVV_FLOAT(v_min, v0, gvl);
@@ -82,23 +91,21 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                 j += gvl;
                 ix += inc_xv;
         }
-        v0 = VFMVVF_FLOAT(FLT_MAX, gvl);
-        v_min = VFREDMINVS_FLOAT(v_min, v0, gvl);
-        minf = v_min[0];
+        v_res = VFREDMINVS_FLOAT(v_res, v_min, v_max, gvl);
+        minf = v_res[0];
 
         if(j<n){
-                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n-j);
                 v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                 mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
                 v1 = VFADDVV_FLOAT(v0, v1, gvl);
-                v0 = VFMVVF_FLOAT(FLT_MAX, gvl);
-                v_min = VFREDMINVS_FLOAT(v1, v0, gvl);
-                if(v_min[0] < minf)
-                        minf = v_min[0];
+                v_res = VFREDMINVS_FLOAT(v_res, v1, v_max, gvl);
+                if(v_res[0] < minf)
+                        minf = v_res[0];
         }
         return(minf);
 }
diff --git a/kernel/riscv64/zasum_vector.c b/kernel/riscv64/zasum_vector.c
index d9fa88971..12527f9dc 100644
--- a/kernel/riscv64/zasum_vector.c
+++ b/kernel/riscv64/zasum_vector.c
@@ -29,29 +29,33 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <math.h>
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VFREDSUMVS_FLOAT vfredsumvs_float32xm8
-#define MASK_T e32xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e32xm8_float32xm8
-#define VFMVVF_FLOAT vfmvvf_float32xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm8
-#define VFADDVV_FLOAT vfaddvv_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VFREDSUMVS_FLOAT vfredsum_vs_f32m8_f32m1
+#define MASK_T vbool4_t
+#define VMFLTVF_FLOAT vmflt_vf_f32m8_b4
+#define VFMVVF_FLOAT vfmv_v_f_f32m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m8_m
+#define VFADDVV_FLOAT vfadd_vv_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VFREDSUMVS_FLOAT vfredsumvs_float64xm8
-#define MASK_T e64xm8_t
-#define VMFLTVF_FLOAT vmfltvf_e64xm8_float64xm8
-#define VFMVVF_FLOAT vfmvvf_float64xm8
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm8
-#define VFADDVV_FLOAT vfaddvv_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VFREDSUMVS_FLOAT vfredsum_vs_f64m8_f64m1
+#define MASK_T vbool8_t
+#define VMFLTVF_FLOAT vmflt_vf_f64m8_b8
+#define VFMVVF_FLOAT vfmv_v_f_f64m8
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m8_m
+#define VFADDVV_FLOAT vfadd_vv_f64m8
 #endif
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 {
@@ -61,40 +65,44 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	if (n <= 0 || inc_x <= 0) return(asumf);
         unsigned int gvl = 0;
         FLOAT_V_T v0, v1, v_zero,v_sum;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
 
         MASK_T mask0, mask1;
         if(inc_x == 1){
                 BLASLONG n2 = n * 2;
-                gvl = vsetvli(n2, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n2);
                 v_zero = VFMVVF_FLOAT(0, gvl);
                 if(gvl <= n2/2){
                         v_sum = VFMVVF_FLOAT(0, gvl);
                         for(i=0,j=0; i<n2/(gvl*2); i++){
                                 v0 = VLEV_FLOAT(&x[j], gvl);
                                 mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                                v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                                v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                                 v_sum = VFADDVV_FLOAT(v_sum, v0, gvl);
 
                                 v1 = VLEV_FLOAT(&x[j+gvl], gvl);
                                 mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                                v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                                v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
                                 v_sum = VFADDVV_FLOAT(v_sum, v1, gvl);
                                 j += gvl * 2;
                         }
-                        v0 = VFREDSUMVS_FLOAT(v_sum, v_zero, gvl);
-                        asumf += v0[0];
+                        v_res = VFREDSUMVS_FLOAT(v_res, v_sum, v_z0, gvl);
+                        asumf += v_res[0];
                 }
                 for(;j<n2;){
-                        gvl = vsetvli(n2-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n2-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
-                        v0 = VFREDSUMVS_FLOAT(v0, v_zero, gvl);
-                        asumf += v0[0];
+                        v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
+                        v_res = VFREDSUMVS_FLOAT(v_res, v0, v_z0, gvl);
+                        asumf += v_res[0];
                         j += gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT) * 2;
                 v_zero = VFMVVF_FLOAT(0, gvl);
 
@@ -103,31 +111,31 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                 for(i=0,j=0; i<n/gvl; i++){
                         v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                         v_sum = VFADDVV_FLOAT(v_sum, v0, gvl);
 
                         v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                         mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                        v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                        v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
                         v_sum = VFADDVV_FLOAT(v_sum, v1, gvl);
 
                         j += gvl;
                         ix += inc_xv;
                 }
-                v0 = VFREDSUMVS_FLOAT(v_sum, v_zero, gvl);
-                asumf += v0[0];
+                v_res = VFREDSUMVS_FLOAT(v_res, v_sum, v_z0, gvl);
+                asumf += v_res[0];
                 if(j<n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         mask0 = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask0, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask0, v0, v0, 0, gvl);
                         v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
 
                         mask1 = VMFLTVF_FLOAT(v1, 0, gvl);
-                        v1 = VFRSUBVF_MASK_FLOAT(v1, v1, 0, mask1, gvl);
+                        v1 = VFRSUBVF_MASK_FLOAT(mask1, v1, v1, 0, gvl);
                         v_sum = VFADDVV_FLOAT(v0, v1, gvl);
-                        v_sum = VFREDSUMVS_FLOAT(v_sum, v_zero, gvl);
-                        asumf += v_sum[0];
+                        v_res = VFREDSUMVS_FLOAT(v_res, v_sum, v_z0, gvl);
+                        asumf += v_res[0];
                 }
         }
 	return(asumf);
diff --git a/kernel/riscv64/zaxpby_vector.c b/kernel/riscv64/zaxpby_vector.c
index 1897ce417..3eca20415 100644
--- a/kernel/riscv64/zaxpby_vector.c
+++ b/kernel/riscv64/zaxpby_vector.c
@@ -28,27 +28,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFMULVF_FLOAT vfmulvf_float32xm4
-#define VFMSACVF_FLOAT vfmsacvf_float32xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define FLOAT_V_T vfloat32m4_t
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMULVF_FLOAT vfmul_vf_f32m4
+#define VFMSACVF_FLOAT vfmsac_vf_f32m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFMULVF_FLOAT vfmulvf_float64xm4
-#define VFMSACVF_FLOAT vfmsacvf_float64xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define FLOAT_V_T vfloat64m4_t
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMULVF_FLOAT vfmul_vf_f64m4
+#define VFMSACVF_FLOAT vfmsac_vf_f64m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4
 #endif
 
 int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FLOAT beta_r, FLOAT beta_i, FLOAT *y, BLASLONG inc_y)
@@ -69,7 +67,7 @@ int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FL
                         if(inc_y == 1){
                                 memset(&y[0], 0, 2 * n * sizeof(FLOAT));
                         }else{
-                                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n);
                                 if(gvl <= n/2){
                                         vy0 = VFMVVF_FLOAT(0.0, gvl);
                                         BLASLONG inc_yv = inc_y * gvl * 2;
@@ -83,7 +81,7 @@ int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FL
                                         }
                                 }
                                 for(;j<n;){
-                                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                        gvl = VSETVL(n-j);
                                         vy0 = VFMVVF_FLOAT(0.0, gvl);
                                         VSSEV_FLOAT(&y[iy], stride_y, vy0, gvl);
                                         VSSEV_FLOAT(&y[iy+1], stride_y, vy0, gvl);
@@ -92,7 +90,7 @@ int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FL
                                 }
                         }
 		}else{
-                        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n);
                         BLASLONG inc_xv = inc_x * gvl * 2;
                         BLASLONG inc_yv = inc_y * gvl * 2;
                         for(i=0,j=0; i<n/gvl; i++){
@@ -110,7 +108,7 @@ int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FL
                                 iy += inc_yv;
                         }
                         if(j<n){
-                                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n-j);
                                 vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                 vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                                 vy0 = VFMULVF_FLOAT(vx1, alpha_i, gvl);
@@ -124,7 +122,7 @@ int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FL
         }else{
 	        FLOAT_V_T v0, v1;
                 if(alpha_r == 0.0 && alpha_i == 0.0){
-                        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n);
                         BLASLONG inc_yv = inc_y * gvl * 2;
                         for(i=0,j=0;i<n/gvl;i++){
                                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
@@ -139,7 +137,7 @@ int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FL
                                 iy += inc_yv;
                         }
                         if(j<n){
-                                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n-j);
                                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
                                 vy1 = VLSEV_FLOAT(&y[iy+1], stride_y, gvl);
                                 v0 = VFMULVF_FLOAT(vy1, beta_i, gvl);
@@ -150,7 +148,7 @@ int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FL
                                 VSSEV_FLOAT(&y[iy+1], stride_y, v1, gvl);
                         }
 		}else{
-                        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n);
                         BLASLONG inc_xv = inc_x * gvl * 2;
                         BLASLONG inc_yv = inc_y * gvl * 2;
                         for(i=0,j=0; i<n/gvl; i++){
@@ -174,7 +172,7 @@ int CNAME(BLASLONG n, FLOAT alpha_r, FLOAT alpha_i, FLOAT *x, BLASLONG inc_x, FL
                                 iy += inc_yv;
                         }
                         if(j<n){
-                                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                                gvl = VSETVL(n-j);
                                 vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                                 vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
diff --git a/kernel/riscv64/zaxpy_vector.c b/kernel/riscv64/zaxpy_vector.c
index fb2656a1d..303d3541e 100644
--- a/kernel/riscv64/zaxpy_vector.c
+++ b/kernel/riscv64/zaxpy_vector.c
@@ -28,21 +28,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define FLOAT_V_T vfloat32m4_t
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define FLOAT_V_T vfloat64m4_t
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4
 #endif
 
 int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
@@ -56,7 +54,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
         BLASLONG stride_y = inc_y * 2 * sizeof(FLOAT);
 
         FLOAT_V_T vx0, vx1, vy0, vy1;
-        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+        gvl = VSETVL(n);
         BLASLONG inc_xv = inc_x * 2 * gvl;
         BLASLONG inc_yv = inc_y * 2 * gvl;
         for(i=0,j=0; i < n/gvl; i++){
@@ -82,7 +80,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
                 iy += inc_yv;
         }
         if(j < n){
-                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n-j);
                 vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
diff --git a/kernel/riscv64/zcopy_vector.c b/kernel/riscv64/zcopy_vector.c
index 6ed430931..600f02bba 100644
--- a/kernel/riscv64/zcopy_vector.c
+++ b/kernel/riscv64/zcopy_vector.c
@@ -27,17 +27,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define FLOAT_V_T vfloat32m4_t
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define FLOAT_V_T vfloat64m4_t
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
 #endif
 
 
@@ -52,7 +50,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                 memcpy(&y[0], &x[0], n * 2 * sizeof(FLOAT));
         }else{
                 FLOAT_V_T vx0, vx1, vx2, vx3;
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * 2 * sizeof(FLOAT);
                 BLASLONG stride_y = inc_y * 2 * sizeof(FLOAT);
                 if(gvl <= n/2){
@@ -75,7 +73,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
                         }
                 }
                 for(;j<n;){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                         VSSEV_FLOAT(&y[iy], stride_y, vx0, gvl);
diff --git a/kernel/riscv64/zdot_vector.c b/kernel/riscv64/zdot_vector.c
index 33efd07e7..1928a897c 100644
--- a/kernel/riscv64/zdot_vector.c
+++ b/kernel/riscv64/zdot_vector.c
@@ -27,31 +27,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFDOTVV_FLOAT vfdotvv_float32xm4
-#define VFMULVV_FLOAT vfmulvv_float32xm4
-#define VFMSACVV_FLOAT vfmsacvv_float32xm4
-#define VFNMSACVV_FLOAT vfnmsacvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFDOTVV_FLOAT vfdot_vv_f32m4
+#define VFMULVV_FLOAT vfmul_vv_f32m4
+#define VFMSACVV_FLOAT vfmsac_vv_f32m4
+#define VFNMSACVV_FLOAT vfnmsac_vv_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFDOTVV_FLOAT vfdotvv_float64xm4
-#define VFMULVV_FLOAT vfmulvv_float64xm4
-#define VFMSACVV_FLOAT vfmsacvv_float64xm4
-#define VFNMSACVV_FLOAT vfnmsacvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFDOTVV_FLOAT vfdot_vv_f64m4
+#define VFMULVV_FLOAT vfmul_vv_f64m4
+#define VFMSACVV_FLOAT vfmsac_vv_f64m4
+#define VFNMSACVV_FLOAT vfnmsac_vv_f64m4
 #endif
 
 OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
@@ -70,9 +74,13 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
         if ( n < 1 )  return(result);
 
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
 
         FLOAT_V_T vr0, vr1, vx0, vx1, vy0, vy1;
-        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+        gvl = VSETVL(n);
         vr0 = VFMVVF_FLOAT(0, gvl);
         vr1 = VFMVVF_FLOAT(0, gvl);
         BLASLONG stride_x = inc_x * 2 * sizeof(FLOAT);
@@ -99,14 +107,13 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
                 ix += inc_xv;
                 iy += inc_yv;
         }
-        vx0 = VFMVVF_FLOAT(0, gvl);
-        vr0 = VFREDSUM_FLOAT(vr0, vx0, gvl);
-        dot[0] += vr0[0];
-        vr1 = VFREDSUM_FLOAT(vr1, vx0, gvl);
-        dot[1] += vr1[0];
+        v_res = VFREDSUM_FLOAT(v_res, vr0, v_z0, gvl);
+        dot[0] += v_res[0];
+        v_res = VFREDSUM_FLOAT(v_res, vr1, v_z0, gvl);
+        dot[1] += v_res[0];
         //tail
         if(j < n){
-                gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n-j);
                 vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                 vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
@@ -123,11 +130,10 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
                 vr1 = VFMULVV_FLOAT(vx1, vy0, gvl);
                 vr1 = VFMSACVV_FLOAT(vr1, vx0, vy1, gvl);
 #endif
-                vx0 = VFMVVF_FLOAT(0, gvl);
-                vr0 = VFREDSUM_FLOAT(vr0, vx0, gvl);
-                dot[0] += vr0[0];
-                vr1 = VFREDSUM_FLOAT(vr1, vx0, gvl);
-                dot[1] += vr1[0];
+                v_res = VFREDSUM_FLOAT(v_res, vr0, v_z0, gvl);
+                dot[0] += v_res[0];
+                v_res = VFREDSUM_FLOAT(v_res, vr1, v_z0, gvl);
+                dot[1] += v_res[0];
         }
         CREAL(result) = dot[0];
         CIMAG(result) = dot[1];
diff --git a/kernel/riscv64/zgemv_n_vector.c b/kernel/riscv64/zgemv_n_vector.c
index 31cbbe6bb..b5ee1f054 100644
--- a/kernel/riscv64/zgemv_n_vector.c
+++ b/kernel/riscv64/zgemv_n_vector.c
@@ -27,25 +27,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSEV_FLOAT vsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define FLOAT_V_T vfloat32m4_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSEV_FLOAT vse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSEV_FLOAT vsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define FLOAT_V_T vfloat64m4_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSEV_FLOAT vse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4
 #endif
 
 int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@@ -58,7 +56,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
         unsigned int gvl = 0;
         BLASLONG stride_a = sizeof(FLOAT) * 2;
         BLASLONG stride_y = inc_y * sizeof(FLOAT) * 2;
-        gvl = vsetvli(m, RVV_EFLOAT, RVV_M);
+        gvl = VSETVL(m);
         BLASLONG inc_yv = inc_y * gvl * 2;
         BLASLONG inc_x2 = inc_x * 2;
         BLASLONG lda2 = lda * 2;
@@ -117,7 +115,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
         }
         //tail
         if(j/2 < m){
-                gvl = vsetvli(m-j/2, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(m-j/2);
                 a_ptr = a;
                 ix = 0;
                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
diff --git a/kernel/riscv64/zgemv_t_vector.c b/kernel/riscv64/zgemv_t_vector.c
index b23a4d8a3..214861eca 100644
--- a/kernel/riscv64/zgemv_t_vector.c
+++ b/kernel/riscv64/zgemv_t_vector.c
@@ -27,25 +27,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFNMSACVV_FLOAT vfnmsacvv_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFMULVV_FLOAT vfmulvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFNMSACVV_FLOAT vfnmsac_vv_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFMULVV_FLOAT vfmul_vv_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFNMSACVV_FLOAT vfnmsacvv_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFMULVV_FLOAT vfmulvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFNMSACVV_FLOAT vfnmsac_vv_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFMULVV_FLOAT vfmul_vv_f64m4
 #endif
 
 int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@@ -57,15 +61,20 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
 
         FLOAT_V_T va0, va1, vx0, vx1, vr, vi;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
+
         BLASLONG stride_x = inc_x * sizeof(FLOAT) * 2;
         BLASLONG stride_a = sizeof(FLOAT) * 2;
-        gvl = vsetvli(m, RVV_EFLOAT, RVV_M);
+        gvl = VSETVL(m);
         BLASLONG inc_xv = inc_x * gvl * 2;
         BLASLONG inc_av = gvl * 2;
         BLASLONG inc_y2 = inc_y * 2;
         BLASLONG lda2 = lda * 2;
         for(i = 0; i < n; i++){
-                gvl = vsetvli(m, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(m);
                 j = 0;
                 ix = 0;
                 vr = VFMVVF_FLOAT(0, gvl);
@@ -90,13 +99,12 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
                         j += inc_av;
                         ix += inc_xv;
                 }
-                va0 = VFMVVF_FLOAT(0, gvl);
-                vx0 = VFREDSUM_FLOAT(vr, va0, gvl);
-                temp_r = vx0[0];
-                vx1 = VFREDSUM_FLOAT(vi, va0, gvl);
-                temp_i = vx1[0];
+                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                temp_r = v_res[0];
+                v_res = VFREDSUM_FLOAT(v_res, vi, v_z0, gvl);
+                temp_i = v_res[0];
                 if(j/2 < m){
-                        gvl = vsetvli(m-j/2, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(m-j/2);
                         va0 = VLSEV_FLOAT(&a_ptr[j], stride_a, gvl);
                         va1 = VLSEV_FLOAT(&a_ptr[j+1], stride_a, gvl);
                         vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
@@ -113,11 +121,10 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
                         vi = VFNMSACVV_FLOAT(vi, va1, vx0, gvl);
 
 #endif
-                        va0 = VFMVVF_FLOAT(0, gvl);
-                        vx0 = VFREDSUM_FLOAT(vr, va0, gvl);
-                        temp_r += vx0[0];
-                        vx1 = VFREDSUM_FLOAT(vi, va0, gvl);
-                        temp_i += vx1[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
+                        temp_r += v_res[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vi, v_z0, gvl);
+                        temp_i += v_res[0];
                 }
 #if !defined(XCONJ)
                 y[iy]   += alpha_r * temp_r - alpha_i * temp_i;
diff --git a/kernel/riscv64/zhemv_LM_vector.c b/kernel/riscv64/zhemv_LM_vector.c
index aa9ac85d5..c8c2178d0 100644
--- a/kernel/riscv64/zhemv_LM_vector.c
+++ b/kernel/riscv64/zhemv_LM_vector.c
@@ -27,31 +27,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFMULVV_FLOAT vfmulvv_float32xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float32xm4
-#define VFNMSACVV_FLOAT vfnmsacvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFMULVV_FLOAT vfmul_vv_f32m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4
+#define VFNMSACVV_FLOAT vfnmsac_vv_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFMULVV_FLOAT vfmulvv_float64xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float64xm4
-#define VFNMSACVV_FLOAT vfnmsacvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFMULVV_FLOAT vfmul_vv_f64m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4
+#define VFNMSACVV_FLOAT vfnmsac_vv_f64m4
 #endif
 
 int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, FLOAT *buffer){
@@ -62,7 +66,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, B
         FLOAT temp_r2, temp_i2;
         FLOAT *a_ptr = a;
         unsigned int gvl = 0;
-
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
 
         FLOAT_V_T va0, va1, vx0, vx1, vy0, vy1, vr0, vr1;
         BLASLONG stride_x, stride_y, stride_a, inc_xv, inc_yv, inc_av, len, lda2;
@@ -90,7 +97,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, B
                 i = j + 1;
                 len = m - i;
                 if(len > 0){
-                        gvl = vsetvli(len, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(len);
                         inc_xv = incx * gvl * 2;
                         inc_yv = incy * gvl * 2;
                         inc_av = gvl * 2;
@@ -134,13 +141,12 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, B
                                 iy += inc_yv;
                                 ia += inc_av;
                         }
-                        va0 = VFMVVF_FLOAT(0, gvl);
-                        vx0 = VFREDSUM_FLOAT(vr0, va0, gvl);
-                        temp_r2 = vx0[0];
-                        vx1 = VFREDSUM_FLOAT(vr1, va0, gvl);
-                        temp_i2 = vx1[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr0, v_z0, gvl);
+                        temp_r2 = v_res[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr1, v_z0, gvl);
+                        temp_i2 = v_res[0];
                         if(i < m){
-				gvl = vsetvli(m-i, RVV_EFLOAT, RVV_M);
+				                gvl = VSETVL(m-i);
                                 va0 = VLSEV_FLOAT(&a_ptr[ia], stride_a, gvl);
                                 va1 = VLSEV_FLOAT(&a_ptr[ia+1], stride_a, gvl);
                                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
@@ -173,11 +179,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, B
                                 vr1 = VFMACCVV_FLOAT(vr1, vx0, va1, gvl);
 #endif
 
-                                va0 = VFMVVF_FLOAT(0, gvl);
-                                vx0 = VFREDSUM_FLOAT(vr0, va0, gvl);
-                                temp_r2 += vx0[0];
-                                vx1 = VFREDSUM_FLOAT(vr1, va0, gvl);
-                                temp_i2 += vx1[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr0, v_z0, gvl);
+                                temp_r2 += v_res[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr1, v_z0, gvl);
+                                temp_i2 += v_res[0];
                         }
                 }
 		y[jy] += alpha_r * temp_r2 - alpha_i * temp_i2;
diff --git a/kernel/riscv64/zhemv_UV_vector.c b/kernel/riscv64/zhemv_UV_vector.c
index 6fe12c76c..40cd9cd64 100644
--- a/kernel/riscv64/zhemv_UV_vector.c
+++ b/kernel/riscv64/zhemv_UV_vector.c
@@ -27,31 +27,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFMULVV_FLOAT vfmulvv_float32xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float32xm4
-#define VFNMSACVV_FLOAT vfnmsacvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFMULVV_FLOAT vfmul_vv_f32m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4
+#define VFNMSACVV_FLOAT vfnmsac_vv_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFMULVV_FLOAT vfmulvv_float64xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float64xm4
-#define VFNMSACVV_FLOAT vfnmsacvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFMULVV_FLOAT vfmul_vv_f64m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4
+#define VFNMSACVV_FLOAT vfnmsac_vv_f64m4
 #endif
 
 int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, FLOAT *buffer){
@@ -62,7 +66,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, B
         FLOAT temp_r2, temp_i2;
         FLOAT *a_ptr = a;
         unsigned int gvl = 0;
-
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
 
         FLOAT_V_T va0, va1, vx0, vx1, vy0, vy1, vr0, vr1;
         BLASLONG stride_x, stride_y, stride_a, inc_xv, inc_yv, inc_av, lda2;
@@ -89,7 +96,7 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, B
                 ia = 0;
                 i = 0;
                 if(j > 0){
-                        gvl = vsetvli(j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(j);
                         inc_xv = incx * gvl * 2;
                         inc_yv = incy * gvl * 2;
                         inc_av = gvl * 2;
@@ -133,13 +140,12 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, B
                                 iy += inc_yv;
                                 ia += inc_av;
                         }
-                        va0 = VFMVVF_FLOAT(0, gvl);
-                        vx0 = VFREDSUM_FLOAT(vr0, va0, gvl);
-                        temp_r2 = vx0[0];
-                        vx1 = VFREDSUM_FLOAT(vr1, va0, gvl);
-                        temp_i2 = vx1[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr0, v_z0, gvl);
+                        temp_r2 = v_res[0];
+                        v_res = VFREDSUM_FLOAT(v_res, vr1, v_z0, gvl);
+                        temp_i2 = v_res[0];
                         if(i < j){
-				gvl = vsetvli(j-i, RVV_EFLOAT, RVV_M);
+				                gvl = VSETVL(j-i);
                                 va0 = VLSEV_FLOAT(&a_ptr[ia], stride_a, gvl);
                                 va1 = VLSEV_FLOAT(&a_ptr[ia+1], stride_a, gvl);
                                 vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
@@ -172,11 +178,10 @@ int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, B
                                 vr1 = VFMACCVV_FLOAT(vr1, vx0, va1, gvl);
 #endif
 
-                                va0 = VFMVVF_FLOAT(0, gvl);
-                                vx0 = VFREDSUM_FLOAT(vr0, va0, gvl);
-                                temp_r2 += vx0[0];
-                                vx1 = VFREDSUM_FLOAT(vr1, va0, gvl);
-                                temp_i2 += vx1[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr0, v_z0, gvl);
+                                temp_r2 += v_res[0];
+                                v_res = VFREDSUM_FLOAT(v_res, vr1, v_z0, gvl);
+                                temp_i2 += v_res[0];
                         }
                 }
                 y[jy] += temp_r1 * a_ptr[ja];
diff --git a/kernel/riscv64/znrm2_vector.c b/kernel/riscv64/znrm2_vector.c
index b0ebfa5f4..5ac62eb80 100644
--- a/kernel/riscv64/znrm2_vector.c
+++ b/kernel/riscv64/znrm2_vector.c
@@ -27,41 +27,45 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VFREDSUM_FLOAT vfredsumvs_float32xm4
-#define VFMACCVV_FLOAT vfmaccvv_float32xm4
-#define VFMVVF_FLOAT vfmvvf_float32xm4
-#define VFDOTVV_FLOAT vfdotvv_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define FLOAT_V_T_M1 vfloat32m1_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VFREDSUM_FLOAT vfredsum_vs_f32m4_f32m1
+#define VFMACCVV_FLOAT vfmacc_vv_f32m4
+#define VFMVVF_FLOAT vfmv_v_f_f32m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f32m1
+#define VFDOTVV_FLOAT vfdot_vv_f32m4
 #define ABS fabsf
-#define MASK_T e32xm4_t
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float32xm4
-#define VMFGTVF_FLOAT vmfgtvf_e32xm4_float32xm4
-#define VMFIRSTM vmfirstm_e32xm4
-#define VFDIVVF_FLOAT vfdivvf_float32xm4
-#define VMFLTVF_FLOAT vmfltvf_e32xm4_float32xm4
-#define VFREDMAXVS_FLOAT vfredmaxvs_float32xm4
+#define MASK_T vbool8_t
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f32m4_m
+#define VMFGTVF_FLOAT vmfgt_vf_f32m4_b8
+#define VMFIRSTM vmfirst_m_b8
+#define VFDIVVF_FLOAT vfdiv_vf_f32m4
+#define VMFLTVF_FLOAT vmflt_vf_f32m4_b8
+#define VFREDMAXVS_FLOAT vfredmax_vs_f32m4_f32m1
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VFREDSUM_FLOAT vfredsumvs_float64xm4
-#define VFMACCVV_FLOAT vfmaccvv_float64xm4
-#define VFMVVF_FLOAT vfmvvf_float64xm4
-#define VFDOTVV_FLOAT vfdotvv_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define FLOAT_V_T_M1 vfloat64m1_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VFREDSUM_FLOAT vfredsum_vs_f64m4_f64m1
+#define VFMACCVV_FLOAT vfmacc_vv_f64m4
+#define VFMVVF_FLOAT vfmv_v_f_f64m4
+#define VFMVVF_FLOAT_M1 vfmv_v_f_f64m1
+#define VFDOTVV_FLOAT vfdot_vv_f64m4
 #define ABS fabs
-#define MASK_T e64xm4_t
-#define VFRSUBVF_MASK_FLOAT vfrsubvf_mask_float64xm4
-#define VMFGTVF_FLOAT vmfgtvf_e64xm4_float64xm4
-#define VMFIRSTM vmfirstm_e64xm4
-#define VFDIVVF_FLOAT vfdivvf_float64xm4
-#define VMFLTVF_FLOAT vmfltvf_e64xm4_float64xm4
-#define VFREDMAXVS_FLOAT vfredmaxvs_float64xm4
+#define MASK_T vbool16_t
+#define VFRSUBVF_MASK_FLOAT vfrsub_vf_f64m4_m
+#define VMFGTVF_FLOAT vmfgt_vf_f64m4_b16
+#define VMFIRSTM vmfirst_m_b16
+#define VFDIVVF_FLOAT vfdiv_vf_f64m4
+#define VMFLTVF_FLOAT vmflt_vf_f64m4_b16
+#define VFREDMAXVS_FLOAT vfredmax_vs_f64m4_f64m1
 #endif
 
 FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
@@ -73,19 +77,24 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 
         FLOAT_V_T vr, v0, v_zero;
         unsigned int gvl = 0;
+        FLOAT_V_T_M1 v_res, v_z0;
+        gvl = VSETVL_MAX;
+        v_res = VFMVVF_FLOAT_M1(0, gvl);
+        v_z0 = VFMVVF_FLOAT_M1(0, gvl);
+
         FLOAT scale = 0.0, ssq = 0.0;
         MASK_T mask;
         BLASLONG index = 0;
         if(inc_x == 1){
                 BLASLONG n2 = n * 2;
-                gvl = vsetvli(n2, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n2);
                 vr = VFMVVF_FLOAT(0, gvl);
                 v_zero = VFMVVF_FLOAT(0, gvl);
                 for(i=0,j=0; i<n2/gvl; i++){
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -96,15 +105,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 }
                         }else{//found greater element
                                 //ssq in vector vr: vr[0]
-                                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                                 //total ssq before current vector
-                                ssq += vr[0];
+                                ssq += v_res[0];
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 //ssq in vector vr
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                                 vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
@@ -112,17 +121,17 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         j += gvl;
                 }
                 //ssq in vector vr: vr[0]
-                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                 //total ssq now
-                ssq += vr[0];
+                ssq += v_res[0];
 
                 //tail
                 if(j < n2){
-                        gvl = vsetvli(n2-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n2-j);
                         v0 = VLEV_FLOAT(&x[j], gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -131,21 +140,21 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                         v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                         }else{//found greater element
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                         }
                         vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
                         //ssq in vector vr: vr[0]
-                        vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                         //total ssq now
-                        ssq += vr[0];
+                        ssq += v_res[0];
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 vr = VFMVVF_FLOAT(0, gvl);
                 v_zero = VFMVVF_FLOAT(0, gvl);
                 unsigned int stride_x = inc_x * sizeof(FLOAT) * 2;
@@ -154,7 +163,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         v0 = VLSEV_FLOAT(&x[idx], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -165,15 +174,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 }
                         }else{//found greater element
                                 //ssq in vector vr: vr[0]
-                                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                                 //total ssq before current vector
-                                ssq += vr[0];
+                                ssq += v_res[0];
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 //ssq in vector vr
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                                 vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
@@ -182,7 +191,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         v0 = VLSEV_FLOAT(&x[idx+1], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -193,15 +202,15 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 }
                         }else{//found greater element
                                 //ssq in vector vr: vr[0]
-                                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                                 //total ssq before current vector
-                                ssq += vr[0];
+                                ssq += v_res[0];
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 //ssq in vector vr
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                                 vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
@@ -210,17 +219,17 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         idx += inc_v;
                 }
                 //ssq in vector vr: vr[0]
-                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                 //total ssq now
-                ssq += vr[0];
+                ssq += v_res[0];
 
                 //tail
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[idx], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -231,11 +240,11 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 }
                         }else{//found greater element
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                                 vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
                         }
@@ -243,7 +252,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                         v0 = VLSEV_FLOAT(&x[idx+1], stride_x, gvl);
                         //fabs(vector)
                         mask = VMFLTVF_FLOAT(v0, 0, gvl);
-                        v0 = VFRSUBVF_MASK_FLOAT(v0, v0, 0, mask, gvl);
+                        v0 = VFRSUBVF_MASK_FLOAT(mask, v0, v0, 0, gvl);
                         //if scale change
                         mask = VMFGTVF_FLOAT(v0, scale, gvl);
                         index = VMFIRSTM(mask, gvl);
@@ -254,22 +263,22 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
                                 }
                         }else{//found greater element
                                 //ssq in vector vr: vr[0]
-                                vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                                v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                                 //total ssq before current vector
-                                ssq += vr[0];
+                                ssq += v_res[0];
                                 //find max
-                                vr = VFREDMAXVS_FLOAT(v0, v_zero, gvl);
+                                v_res = VFREDMAXVS_FLOAT(v_res, v0, v_z0, gvl);
                                 //update ssq before max_index
-                                ssq = ssq * (scale/vr[0])*(scale/vr[0]);
+                                ssq = ssq * (scale/v_res[0])*(scale/v_res[0]);
                                 //update scale
-                                scale = vr[0];
+                                scale = v_res[0];
                                 v0 = VFDIVVF_FLOAT(v0, scale, gvl);
                                 vr = VFMACCVV_FLOAT(v_zero, v0, v0, gvl);
                         }
                         //ssq in vector vr: vr[0]
-                        vr = VFREDSUM_FLOAT(vr, v_zero, gvl);
+                        v_res = VFREDSUM_FLOAT(v_res, vr, v_z0, gvl);
                         //total ssq now
-                        ssq += vr[0];
+                        ssq += v_res[0];
                 }
         }
 	return(scale * sqrt(ssq));
diff --git a/kernel/riscv64/zrot_vector.c b/kernel/riscv64/zrot_vector.c
index a3fdda45a..2fdd8135a 100644
--- a/kernel/riscv64/zrot_vector.c
+++ b/kernel/riscv64/zrot_vector.c
@@ -27,27 +27,27 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLEV_FLOAT vlev_float32xm4
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSEV_FLOAT vsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMULVF_FLOAT vfmulvf_float32xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define VLEV_FLOAT vle_v_f32m4
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSEV_FLOAT vse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMULVF_FLOAT vfmul_vf_f32m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLEV_FLOAT vlev_float64xm4
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSEV_FLOAT vsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMULVF_FLOAT vfmulvf_float64xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define VLEV_FLOAT vle_v_f64m4
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSEV_FLOAT vse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMULVF_FLOAT vfmul_vf_f64m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4
 #endif
 
 int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT c, FLOAT s)
@@ -59,7 +59,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
         unsigned int gvl = 0;
 
         FLOAT_V_T vt0, vt1, vx0, vx1, vy0, vy1;
-        gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+        gvl = VSETVL(n);
         BLASLONG stride_x = inc_x * 2 * sizeof(FLOAT);
         BLASLONG stride_y = inc_y * 2 * sizeof(FLOAT);
         BLASLONG inc_xv = inc_x * 2 * gvl;
@@ -90,7 +90,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
 			ix += 2*gvl;
 		}
 		if(j < n){
-			gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+			gvl = VSETVL(n-j);
 						vx0 = VLEV_FLOAT(&x[ix], gvl);
 			vx1 = VLEV_FLOAT(&x[ix+gvl], gvl);
 			vy0 = VLEV_FLOAT(&y[ix], gvl);
@@ -137,7 +137,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT
 			iy += inc_yv;
 		}
 		if(j < n){
-			gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+			gvl = VSETVL(n-j);
 			vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
 			vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
 			vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
diff --git a/kernel/riscv64/zscal_vector.c b/kernel/riscv64/zscal_vector.c
index 796c52a02..02a76f168 100644
--- a/kernel/riscv64/zscal_vector.c
+++ b/kernel/riscv64/zscal_vector.c
@@ -27,23 +27,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M4
-#define FLOAT_V_T float32xm4_t
-#define VLSEV_FLOAT vlsev_float32xm4
-#define VSSEV_FLOAT vssev_float32xm4
-#define VFMACCVF_FLOAT vfmaccvf_float32xm4
-#define VFMULVF_FLOAT vfmulvf_float32xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float32xm4
+#define VSETVL(n) vsetvl_e32m4(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m4_t
+#define VLSEV_FLOAT vlse_v_f32m4
+#define VSSEV_FLOAT vsse_v_f32m4
+#define VFMACCVF_FLOAT vfmacc_vf_f32m4
+#define VFMULVF_FLOAT vfmul_vf_f32m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f32m4
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M4
-#define FLOAT_V_T float64xm4_t
-#define VLSEV_FLOAT vlsev_float64xm4
-#define VSSEV_FLOAT vssev_float64xm4
-#define VFMACCVF_FLOAT vfmaccvf_float64xm4
-#define VFMULVF_FLOAT vfmulvf_float64xm4
-#define VFNMSACVF_FLOAT vfnmsacvf_float64xm4
+#define VSETVL(n) vsetvl_e64m4(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m4_t
+#define VLSEV_FLOAT vlse_v_f64m4
+#define VSSEV_FLOAT vsse_v_f64m4
+#define VFMACCVF_FLOAT vfmacc_vf_f64m4
+#define VFMULVF_FLOAT vfmul_vf_f64m4
+#define VFNMSACVF_FLOAT vfnmsac_vf_f64m4
 #endif
 
 int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
@@ -60,7 +60,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, F
         if(da_r == 0.0 && da_i == 0.0){
                 memset(&x[0], 0, n * 2 * sizeof(FLOAT));
         }else if(da_r == 0.0){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * 2 * sizeof(FLOAT);
                 BLASLONG inc_xv = inc_x * 2 * gvl;
                 for(i=0,j=0; i < n/gvl; i++){
@@ -77,7 +77,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, F
                         ix += inc_xv;
                 }
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
 
@@ -88,7 +88,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, F
                         VSSEV_FLOAT(&x[ix+1], stride_x, v1, gvl);
                 }
         }else if(da_i == 0.0){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * 2 * sizeof(FLOAT);
                 BLASLONG inc_xv = inc_x * 2 * gvl;
                 for(i=0,j=0; i < n/gvl; i++){
@@ -105,7 +105,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, F
                         ix += inc_xv;
                 }
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
 
@@ -116,7 +116,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, F
                         VSSEV_FLOAT(&x[ix+1], stride_x, v1, gvl);
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG stride_x = inc_x * 2 * sizeof(FLOAT);
                 BLASLONG inc_xv = inc_x * 2 * gvl;
                 for(i=0,j=0; i < n/gvl; i++){
@@ -135,7 +135,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, F
                         ix += inc_xv;
                 }
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         v0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         v1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
 
diff --git a/kernel/riscv64/zswap_vector.c b/kernel/riscv64/zswap_vector.c
index b655a968c..7550294b5 100644
--- a/kernel/riscv64/zswap_vector.c
+++ b/kernel/riscv64/zswap_vector.c
@@ -28,21 +28,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 #include <stdio.h>
 #if !defined(DOUBLE)
-#define RVV_EFLOAT RVV_E32
-#define RVV_M RVV_M8
-#define FLOAT_V_T float32xm8_t
-#define VLEV_FLOAT vlev_float32xm8
-#define VLSEV_FLOAT vlsev_float32xm8
-#define VSEV_FLOAT vsev_float32xm8
-#define VSSEV_FLOAT vssev_float32xm8
+#define VSETVL(n) vsetvl_e32m8(n)
+#define VSETVL_MAX vsetvlmax_e32m1()
+#define FLOAT_V_T vfloat32m8_t
+#define VLEV_FLOAT vle_v_f32m8
+#define VLSEV_FLOAT vlse_v_f32m8
+#define VSEV_FLOAT vse_v_f32m8
+#define VSSEV_FLOAT vsse_v_f32m8
 #else
-#define RVV_EFLOAT RVV_E64
-#define RVV_M RVV_M8
-#define FLOAT_V_T float64xm8_t
-#define VLEV_FLOAT vlev_float64xm8
-#define VLSEV_FLOAT vlsev_float64xm8
-#define VSEV_FLOAT vsev_float64xm8
-#define VSSEV_FLOAT vssev_float64xm8
+#define VSETVL(n) vsetvl_e64m8(n)
+#define VSETVL_MAX vsetvlmax_e64m1()
+#define FLOAT_V_T vfloat64m8_t
+#define VLEV_FLOAT vle_v_f64m8
+#define VLSEV_FLOAT vlse_v_f64m8
+#define VSEV_FLOAT vse_v_f64m8
+#define VSSEV_FLOAT vsse_v_f64m8
 #endif
 
 int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dummy4, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
@@ -55,7 +55,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dumm
 
 	if (n < 0)  return(0);
         if(inc_x == 1 && inc_y == 1){
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 BLASLONG n2 = n * 2;
                 if(gvl <= n2/2){
                         for(i=0,j=0; i<n2/(2*gvl); i++){
@@ -72,7 +72,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dumm
                         }
                 }
                 for(;j<n2;){
-                        gvl = vsetvli(n2-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n2-j);
                         vx0 = VLEV_FLOAT(&x[j], gvl);
                         vy0 = VLEV_FLOAT(&y[j], gvl);
                         VSEV_FLOAT(&x[j], vy0, gvl);
@@ -80,7 +80,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dumm
                         j += gvl;
                 }
         }else{
-                gvl = vsetvli(n, RVV_EFLOAT, RVV_M);
+                gvl = VSETVL(n);
                 stride_x = inc_x * 2 * sizeof(FLOAT);
                 stride_y = inc_y * 2 * sizeof(FLOAT);
                 BLASLONG inc_xv = inc_x * gvl * 2;
@@ -100,7 +100,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT dummy3, FLOAT dumm
                         iy += inc_yv;
                 }
                 if(j < n){
-                        gvl = vsetvli(n-j, RVV_EFLOAT, RVV_M);
+                        gvl = VSETVL(n-j);
                         vx0 = VLSEV_FLOAT(&x[ix], stride_x, gvl);
                         vx1 = VLSEV_FLOAT(&x[ix+1], stride_x, gvl);
                         vy0 = VLSEV_FLOAT(&y[iy], stride_y, gvl);
diff --git a/kernel/x86_64/cdot.c b/kernel/x86_64/cdot.c
index f2bf19dcd..654cd351a 100644
--- a/kernel/x86_64/cdot.c
+++ b/kernel/x86_64/cdot.c
@@ -27,7 +27,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #include "common.h"
-#include <complex.h>
 
 
 #if defined(BULLDOZER)
diff --git a/kernel/x86_64/dgemm_kernel_16x2_skylakex.c b/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
index 9f2bf24e2..15185d7fc 100644
--- a/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
+++ b/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
@@ -149,6 +149,7 @@
  #define KERNEL_h_k1m16n2 \
   "vmovddup (%0),%%zmm1; vmovddup 8(%0),%%zmm2; vmovddup 64(%0),%%zmm3; vmovddup 72(%0),%%zmm4; addq $128,%0;"\
   unit_acc_m16n2(8,9,10,11,%1)
+
 #endif
 #define KERNEL_k1m16n2 KERNEL_h_k1m16n2 "addq $16,%1;"
 #define KERNEL_h_k1m16n4 KERNEL_h_k1m16n2 "prefetcht0 384(%0);" unit_acc_m16n2(12,13,14,15,%1,%%r12,1)
@@ -283,7 +284,32 @@
 #define KERNEL_h_k1m4n10 KERNEL_h_k1m4n8 unit_acc_m4n2(12,13,%%r15,%%r12,1)
 #define KERNEL_k1m4n10 KERNEL_h_k1m4n10 "addq $16,%%r15;"
 #define KERNEL_h_k1m4n12 KERNEL_h_k1m4n10 unit_acc_m4n2(14,15,%%r15,%%r12,2)
-#define KERNEL_k1m4n12 KERNEL_h_k1m4n12 "addq $16,%%r15;"
+//#define KERNEL_k1m4n12 KERNEL_h_k1m4n12 "addq $16,%%r15;"
+#define unit_acc_k2m4n2(c1_no,c2_no,...)\
+  "vbroadcastf64x4 ("#__VA_ARGS__"),%%zmm3; vpermpd %%zmm3,%%zmm30,%%zmm3;"\
+  "vfmadd231pd %%zmm1,%%zmm3,%%zmm"#c1_no"; vfmadd231pd %%zmm2,%%zmm3,%%zmm"#c2_no";"
+
+#define unit_merge_to_ymm(c1_no) \
+  "vextractf64x4 $1,%%zmm"#c1_no",%%ymm30; vaddpd %%ymm"#c1_no",%%ymm30,%%ymm"#c1_no";"
+
+#define KERNEL_k1m4n12 \
+  "cmpq $2, %5; jb 104912f;"\
+  "vmovupd 64+%11,%%zmm30;"\
+  "\n204912:"\
+  "vmovddup (%0),%%zmm1; vmovddup 8(%0),%%zmm2; addq $64,%0;" \
+  unit_acc_k2m4n2(4,5,%1) unit_acc_k2m4n2(6,7,%1,%%r12,1) unit_acc_k2m4n2(8, 9, %1, %%r12, 2) "addq $32,%1;" \
+  unit_acc_k2m4n2(10,11,%%r15) unit_acc_k2m4n2(12,13,%%r15,%%r12,1) unit_acc_k2m4n2(14,15,%%r15,%%r12,2) "addq $32,%%r15;" \
+  "subq $2, %5; cmpq $2, %5; jnb 204912b;"\
+  unit_merge_to_ymm(4) unit_merge_to_ymm(5) unit_merge_to_ymm(6) unit_merge_to_ymm(7) \
+  unit_merge_to_ymm(8) unit_merge_to_ymm(9) unit_merge_to_ymm(10) unit_merge_to_ymm(11) \
+  unit_merge_to_ymm(12) unit_merge_to_ymm(13) unit_merge_to_ymm(14) unit_merge_to_ymm(15) \
+  "testq %5, %5; jz 1004912f;"\
+  "\n104912:"\
+  KERNEL_h_k1m4n12 "addq $16,%%r15;"\
+  "decq %5; jnz 104912b;"\
+  "\n1004912:"\
+  "incq %5;"
+
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m4 "vmovddup (%0,%3,1),%%ymm1; vmovddup 8(%0,%3,1),%%ymm2; addq $32,%3;"
   #define acc_kend_nc2_k1m4(boff1) unit_acc_gen_m4n2(6,7,boff1,%1,%%r12,1)
@@ -336,7 +362,31 @@
 #define KERNEL_h_k1m2n10 KERNEL_h_k1m2n8 unit_acc_m2n2(12,13,%%r15,%%r12,1)
 #define KERNEL_k1m2n10 KERNEL_h_k1m2n10 "addq $16,%%r15;"
 #define KERNEL_h_k1m2n12 KERNEL_h_k1m2n10 unit_acc_m2n2(14,15,%%r15,%%r12,2)
-#define KERNEL_k1m2n12 KERNEL_h_k1m2n12 "addq $16,%%r15;"
+//#define KERNEL_k1m2n12 KERNEL_h_k1m2n12 "addq $16,%%r15;"
+
+#define unit_acc_k4m2n2(c1_no,c2_no,...) \
+  "vmovupd ("#__VA_ARGS__"),%%zmm3; vfmadd231pd %%zmm1,%%zmm3,%%zmm"#c1_no"; vfmadd231pd %%zmm2,%%zmm3,%%zmm"#c2_no";"
+
+#define unit_merge_to_xmm(c1_no) \
+  "vextractf64x2 $0,%%zmm"#c1_no",%%xmm20; vextractf64x2 $1,%%zmm"#c1_no",%%xmm21; vextractf64x2 $2,%%zmm"#c1_no",%%xmm22; vextractf64x2 $3,%%zmm"#c1_no",%%xmm23;"\
+  "vaddpd %%xmm20,%%xmm21,%%xmm20; vaddpd %%xmm22,%%xmm23,%%xmm22; vaddpd %%xmm20,%%xmm22,%%xmm"#c1_no";"
+
+#define KERNEL_k1m2n12 \
+  "cmpq $4,%5; jb 102912f;"\
+  "\n402912:"\
+  "vmovddup (%0),%%zmm1; vmovddup 8(%0),%%zmm2; addq $64,%0;" \
+  unit_acc_k4m2n2(4,5,%1) unit_acc_k4m2n2(6,7,%1,%%r12,1) unit_acc_k4m2n2(8,9,%1,%%r12,2) "addq $64,%1;" \
+  unit_acc_k4m2n2(10,11,%%r15) unit_acc_k4m2n2(12,13,%%r15,%%r12,1) unit_acc_k4m2n2(14,15,%%r15,%%r12,2) "addq $64,%%r15;" \
+  "subq $4,%5; cmpq $4,%5; jnb 402912b;"\
+  unit_merge_to_xmm(4) unit_merge_to_xmm(5) unit_merge_to_xmm(6) unit_merge_to_xmm(7) unit_merge_to_xmm(8) unit_merge_to_xmm(9) \
+  unit_merge_to_xmm(10) unit_merge_to_xmm(11) unit_merge_to_xmm(12) unit_merge_to_xmm(13) unit_merge_to_xmm(14) unit_merge_to_xmm(15) \
+  "testq %5,%5; jz 1002912f;"\
+  "\n102912:"\
+  KERNEL_h_k1m2n12 "addq $16,%%r15;" \
+  "decq %5; jnz 102912b;" \
+  "\n1002912:"\
+  "incq %5;"
+
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m2 "vmovddup (%0,%3,1),%%xmm1; vmovddup 8(%0,%3,1),%%xmm2; addq $16,%3;"
   #define acc_kend_nc2_k1m2(boff1) unit_acc_gen_m2n2(6,7,boff1,%1,%%r12,1)
@@ -387,7 +437,24 @@
 #define KERNEL_h_k1m1n10 KERNEL_h_k1m1n8 "vfmadd231pd (%%r15,%%r12,1),%%xmm1,%%xmm8;"
 #define KERNEL_k1m1n10 KERNEL_h_k1m1n10 "addq $16,%%r15;"
 #define KERNEL_h_k1m1n12 KERNEL_h_k1m1n10 "vfmadd231pd (%%r15,%%r12,2),%%xmm1,%%xmm9;"
-#define KERNEL_k1m1n12 KERNEL_h_k1m1n12 "addq $16,%%r15;"
+//#define KERNEL_k1m1n12 KERNEL_h_k1m1n12 "addq $16,%%r15;"
+#define KERNEL_k1m1n12 \
+  "cmpq $4,%5; jb 101912f;" \
+  "vmovupd %11,%%zmm2;"\
+  "\n401912:"\
+  "vmovupd (%0),%%ymm1; vpermpd %%zmm1,%%zmm2,%%zmm1; addq $32,%0;" \
+  "vfmadd231pd (%1),%%zmm1,%%zmm4; vfmadd231pd (%1,%%r12,1),%%zmm1,%%zmm5; vfmadd231pd (%1,%%r12,2),%%zmm1,%%zmm6; addq $64,%1;"\
+  "vfmadd231pd (%%r15),%%zmm1,%%zmm7; vfmadd231pd (%%r15,%%r12,1),%%zmm1,%%zmm8; vfmadd231pd (%%r15,%%r12,2),%%zmm1,%%zmm9; addq $64,%%r15;"\
+  "subq $4,%5; cmpq $4,%5; jnb 401912b;"\
+  unit_merge_to_xmm(4) unit_merge_to_xmm(5) unit_merge_to_xmm(6) \
+  unit_merge_to_xmm(7) unit_merge_to_xmm(8) unit_merge_to_xmm(9) \
+  "testq %5,%5; jz 1001912f;"\
+  "\n101912:"\
+  KERNEL_h_k1m1n12 "addq $16,%%r15;" \
+  "decq %5; jnz 101912b;" \
+  "\n1001912:"\
+  "incq %5;"
+
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m1 "vmovddup (%0,%3,1),%%xmm1; addq $8,%3;"
   #define acc_kend_nc2_k1m1(boff1) "vfmadd231pd "#boff1"(%1,%%r12,1),%%xmm1,%%xmm5;"
@@ -480,7 +547,7 @@
     COMPUTE_SIMPLE(1,ndim) "subq $1,%%r11;"\
     #ndim"33106:\n\t"\
     "movq %%r14,%1;"\
-  :"+r"(a_ptr),"+r"(b_ptr),"+r"(c_ptr),"+r"(c_tmp),"+r"(ldc_in_bytes),"+r"(k_count),"+r"(b_pref):"m"(M),"m"(ALPHA),"m"(off),"m"(K):"r10","r11","r12","r13","r14","r15","cc","memory",\
+  :"+r"(a_ptr),"+r"(b_ptr),"+r"(c_ptr),"+r"(c_tmp),"+r"(ldc_in_bytes),"+r"(k_count),"+r"(b_pref):"m"(M),"m"(ALPHA),"m"(off),"m"(K), "o"(permute_table):"r10","r11","r12","r13","r14","r15","cc","memory",\
     "zmm0","zmm1","zmm2","zmm3","zmm4","zmm5","zmm6","zmm7","zmm8","zmm9","zmm10","zmm11","zmm12","zmm13","zmm14","zmm15",\
     "zmm16","zmm17","zmm18","zmm19","zmm20","zmm21","zmm22","zmm23","zmm24","zmm25","zmm26","zmm27","zmm28","zmm29","zmm30","zmm31");\
   a_ptr -= M * K; b_ptr += ndim * K; c_ptr += ndim * ldc - M; TAIL_SET_OFF(ndim)\
@@ -501,6 +568,10 @@ CNAME(BLASLONG m, BLASLONG n, BLASLONG k, double alpha, double * __restrict__ A,
     int64_t M = (int64_t)m, K = (int64_t)k, k_count = 0;
     BLASLONG n_count = n, off = 0;
     double *a_ptr = A,*b_ptr = B,*c_ptr = C,*c_tmp = C,*b_pref = B;
+    int64_t permute_table[] = {
+	    0, 0, 1, 1, 2, 2, 3, 3,  // abcdxxxx -> aabbccdd
+	    0, 1, 0, 1, 2, 3, 2, 3,  // abcdxxxx -> ababcdcd
+    };
 #ifdef TRMMKERNEL
   #ifdef LEFT
     off = offset;
diff --git a/kernel/x86_64/sgemv_n_4.c b/kernel/x86_64/sgemv_n_4.c
index 3eec21774..06de28d97 100644
--- a/kernel/x86_64/sgemv_n_4.c
+++ b/kernel/x86_64/sgemv_n_4.c
@@ -35,8 +35,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "sgemv_n_microk_nehalem-4.c"
 #elif defined(SANDYBRIDGE)
 #include "sgemv_n_microk_sandy-4.c"
-#elif defined(HASWELL) || defined(ZEN) || defined (SKYLAKEX) || defined (COOPERLAKE)
+#elif defined(HASWELL) || defined(ZEN)
 #include "sgemv_n_microk_haswell-4.c"
+#elif defined (SKYLAKEX) || defined (COOPERLAKE)
+#include "sgemv_n_microk_haswell-4.c"
+#include "sgemv_n_microk_skylakex-8.c"
 #endif
 
 #if defined(STEAMROLLER)  || defined(EXCAVATOR)
@@ -291,6 +294,41 @@ static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest)
 
 int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
 {
+    if ( m < 1 || n < 1) return(0);
+
+    #ifdef HAVE_SGEMV_N_SKYLAKE_KERNEL
+    if (m <= 16384 && n <= 48 && !(n == 4))
+    {
+        FLOAT * xbuffer_align = x;
+        FLOAT * ybuffer_align = y;
+
+        FLOAT * xbuffer = NULL;
+        FLOAT * ybuffer = NULL;
+
+        if (inc_x != 1) {
+            xbuffer_align = buffer;
+            for(BLASLONG i=0; i<n; i++) {
+                 xbuffer_align[i] = x[i*inc_x];
+            }
+        }
+
+        if (inc_y != 1) {
+            ybuffer_align = buffer + n;
+            for(BLASLONG i=0; i<m; i++) {
+                ybuffer_align[i] = y[i*inc_y];
+            }
+        }
+        sgemv_kernel_n_128(m, n , alpha, a, lda, xbuffer_align, ybuffer_align);
+
+        if(inc_y != 1) {
+            for(BLASLONG i=0; i<m; i++) {
+                   y[i*inc_y] = ybuffer_align[i];
+            }
+        }
+        return(0);
+    }
+
+    #endif
 	BLASLONG i;
 	FLOAT *a_ptr;
 	FLOAT *x_ptr;
@@ -305,9 +343,6 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
 	BLASLONG lda8 =  lda << 3;
 	FLOAT xbuffer[8],*ybuffer;
 
-        if ( m < 1 ) return(0);
-        if ( n < 1 ) return(0);
-
 	ybuffer = buffer;
 	
         if ( inc_x == 1 )
@@ -322,10 +357,9 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
 
 	}
 	
-        m3 = m & 3  ;
-        m1 = m & -4 ;
-        m2 = (m & (NBMAX-1)) - m3 ;
-
+    m3 = m & 3  ;
+    m1 = m & -4 ;
+    m2 = (m & (NBMAX-1)) - m3 ;
 
 	y_ptr = y;
 
@@ -383,7 +417,11 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
 
 			if ( n2 & 2 )
 			{
+#ifdef HAVE_SGEMV_N_SKYLAKE_KERNEL				
+				sgemv_kernel_n_64(NB, 2, alpha, a_ptr, lda, x_ptr, ybuffer);
+#else
 				sgemv_kernel_4x2(NB,ap,x_ptr,ybuffer,&alpha);
+#endif
 				a_ptr += lda*2;
 				x_ptr += 2;	
 			}
@@ -391,7 +429,11 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
 
 			if ( n2 & 1 )
 			{
+#ifdef HAVE_SGEMV_N_SKYLAKE_KERNEL
+				sgemv_kernel_n_64(NB, 1, alpha, a_ptr, lda, x_ptr, ybuffer);
+#else
 				sgemv_kernel_4x1(NB,a_ptr,x_ptr,ybuffer,&alpha);
+#endif
 				/* a_ptr += lda;
 				x_ptr += 1a; */
 
diff --git a/kernel/x86_64/sgemv_n_microk_skylakex-8.c b/kernel/x86_64/sgemv_n_microk_skylakex-8.c
new file mode 100644
index 000000000..fba9cedcd
--- /dev/null
+++ b/kernel/x86_64/sgemv_n_microk_skylakex-8.c
@@ -0,0 +1,258 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+/* need a new enough GCC for avx512 support */
+#if (( defined(__GNUC__)  && __GNUC__   >= 6 && defined(__AVX512CD__)) || (defined(__clang__) && __clang_major__ >= 6))
+
+#define HAVE_SGEMV_N_SKYLAKE_KERNEL 1
+#include "common.h"
+#include <immintrin.h>
+static int sgemv_kernel_n_128(BLASLONG m, BLASLONG n, float alpha, float *a, BLASLONG lda, float *x, float *y)
+{
+    __m512 matrixArray_0, matrixArray_1, matrixArray_2, matrixArray_3, matrixArray_4, matrixArray_5, matrixArray_6, matrixArray_7;
+    __m512 accum512_0, accum512_1, accum512_2, accum512_3, accum512_4, accum512_5, accum512_6, accum512_7;
+    __m512 xArray_0;
+    __m512  ALPHAVECTOR = _mm512_set1_ps(alpha);
+    BLASLONG tag_m_128x = m & (~127);
+    BLASLONG tag_m_64x = m & (~63);
+    BLASLONG tag_m_32x = m & (~31);
+    BLASLONG tag_m_16x = m & (~15);
+
+    for (BLASLONG idx_m = 0; idx_m < tag_m_128x; idx_m+=128) {
+        accum512_0 = _mm512_setzero_ps();
+        accum512_1 = _mm512_setzero_ps();
+        accum512_2 = _mm512_setzero_ps();
+        accum512_3 = _mm512_setzero_ps();
+        accum512_4 = _mm512_setzero_ps();
+        accum512_5 = _mm512_setzero_ps();
+        accum512_6 = _mm512_setzero_ps();
+        accum512_7 = _mm512_setzero_ps();
+
+        for (BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+            xArray_0 = _mm512_set1_ps(x[idx_n]);
+
+            matrixArray_0 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 0]);
+            matrixArray_1 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 16]);
+            matrixArray_2 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 32]);
+            matrixArray_3 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 48]);
+            matrixArray_4 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 64]);
+            matrixArray_5 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 80]);
+            matrixArray_6 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 96]);
+            matrixArray_7 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 112]);
+
+            accum512_0 = _mm512_fmadd_ps(matrixArray_0, xArray_0, accum512_0);
+            accum512_1 = _mm512_fmadd_ps(matrixArray_1, xArray_0, accum512_1);
+            accum512_2 = _mm512_fmadd_ps(matrixArray_2, xArray_0, accum512_2);
+            accum512_3 = _mm512_fmadd_ps(matrixArray_3, xArray_0, accum512_3);
+            accum512_4 = _mm512_fmadd_ps(matrixArray_4, xArray_0, accum512_4);
+            accum512_5 = _mm512_fmadd_ps(matrixArray_5, xArray_0, accum512_5);
+            accum512_6 = _mm512_fmadd_ps(matrixArray_6, xArray_0, accum512_6);
+            accum512_7 = _mm512_fmadd_ps(matrixArray_7, xArray_0, accum512_7);
+        }
+
+        _mm512_storeu_ps(&y[idx_m + 0], _mm512_fmadd_ps(accum512_0, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 0])));
+        _mm512_storeu_ps(&y[idx_m + 16], _mm512_fmadd_ps(accum512_1, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 16])));
+        _mm512_storeu_ps(&y[idx_m + 32], _mm512_fmadd_ps(accum512_2, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 32])));
+        _mm512_storeu_ps(&y[idx_m + 48], _mm512_fmadd_ps(accum512_3, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 48])));
+        _mm512_storeu_ps(&y[idx_m + 64], _mm512_fmadd_ps(accum512_4, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 64])));
+        _mm512_storeu_ps(&y[idx_m + 80], _mm512_fmadd_ps(accum512_5, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 80])));
+        _mm512_storeu_ps(&y[idx_m + 96], _mm512_fmadd_ps(accum512_6, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 96])));
+        _mm512_storeu_ps(&y[idx_m + 112], _mm512_fmadd_ps(accum512_7, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 112])));
+    }
+    if (tag_m_128x != m) {
+        for (BLASLONG idx_m = tag_m_128x; idx_m < tag_m_64x; idx_m+=64) {
+            accum512_0 = _mm512_setzero_ps();
+            accum512_1 = _mm512_setzero_ps();
+            accum512_2 = _mm512_setzero_ps();
+            accum512_3 = _mm512_setzero_ps();
+    
+            for (BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+                xArray_0 = _mm512_set1_ps(x[idx_n]);
+    
+                matrixArray_0 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 0]);
+                matrixArray_1 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 16]);
+                matrixArray_2 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 32]);
+                matrixArray_3 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 48]);
+    
+                accum512_0 = _mm512_fmadd_ps(matrixArray_0, xArray_0, accum512_0);
+                accum512_1 = _mm512_fmadd_ps(matrixArray_1, xArray_0, accum512_1);
+                accum512_2 = _mm512_fmadd_ps(matrixArray_2, xArray_0, accum512_2);
+                accum512_3 = _mm512_fmadd_ps(matrixArray_3, xArray_0, accum512_3);
+            }
+    
+            _mm512_storeu_ps(&y[idx_m + 0], _mm512_fmadd_ps(accum512_0, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 0])));
+            _mm512_storeu_ps(&y[idx_m + 16], _mm512_fmadd_ps(accum512_1, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 16])));
+            _mm512_storeu_ps(&y[idx_m + 32], _mm512_fmadd_ps(accum512_2, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 32])));
+            _mm512_storeu_ps(&y[idx_m + 48], _mm512_fmadd_ps(accum512_3, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 48])));
+        }
+
+        if(tag_m_64x != m) {
+            for (BLASLONG idx_m = tag_m_64x; idx_m < tag_m_32x; idx_m+=32) {
+                accum512_0 = _mm512_setzero_ps();
+                accum512_1 = _mm512_setzero_ps();
+        
+                for (BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+                    xArray_0 = _mm512_set1_ps(x[idx_n]);
+        
+                    matrixArray_0 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 0]);
+                    matrixArray_1 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 16]);
+        
+                    accum512_0 = _mm512_fmadd_ps(matrixArray_0, xArray_0, accum512_0);
+                    accum512_1 = _mm512_fmadd_ps(matrixArray_1, xArray_0, accum512_1);
+                }
+        
+                _mm512_storeu_ps(&y[idx_m + 0], _mm512_fmadd_ps(accum512_0, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 0])));
+                _mm512_storeu_ps(&y[idx_m + 16], _mm512_fmadd_ps(accum512_1, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 16])));
+            }
+
+            if(tag_m_32x != m) {    
+        
+                for (BLASLONG idx_m = tag_m_32x; idx_m < tag_m_16x; idx_m+=16) {
+                    accum512_0 = _mm512_setzero_ps();
+            
+                    for (BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+                        xArray_0 = _mm512_set1_ps(x[idx_n]);
+            
+                        matrixArray_0 = _mm512_loadu_ps(&a[idx_n * lda + idx_m + 0]);
+            
+                        accum512_0 = _mm512_fmadd_ps(matrixArray_0, xArray_0, accum512_0);
+                    }
+            
+                    _mm512_storeu_ps(&y[idx_m + 0], _mm512_fmadd_ps(accum512_0, ALPHAVECTOR, _mm512_loadu_ps(&y[idx_m + 0])));
+                }       
+            
+                if (tag_m_16x != m) {
+                    accum512_0 = _mm512_setzero_ps();
+            
+                    unsigned short tail_mask_value = (((unsigned int)0xffff) >> (16-(m&15)));
+                    __mmask16 tail_mask = *((__mmask16*) &tail_mask_value);
+            
+                    for(BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+                        xArray_0 = _mm512_set1_ps(x[idx_n]);
+                        matrixArray_0 = _mm512_maskz_loadu_ps(tail_mask, &a[idx_n * lda + tag_m_16x]);
+            
+                        accum512_0 = _mm512_fmadd_ps(matrixArray_0, xArray_0, accum512_0);
+                    }
+            
+                    _mm512_mask_storeu_ps(&y[tag_m_16x], tail_mask, _mm512_fmadd_ps(accum512_0, ALPHAVECTOR, _mm512_maskz_loadu_ps(tail_mask, &y[tag_m_16x])));
+            
+                }
+            }
+        }
+    }
+    return 0;
+}
+
+static int sgemv_kernel_n_64(BLASLONG m, BLASLONG n, float alpha, float *a, BLASLONG lda, float *x, float *y)
+{
+    __m256 ma0, ma1, ma2, ma3, ma4, ma5, ma6, ma7;
+    __m256 as0, as1, as2, as3, as4, as5, as6, as7;
+    __m256 alphav = _mm256_set1_ps(alpha);
+    __m256 xv;
+    BLASLONG tag_m_32x = m & (~31);
+    BLASLONG tag_m_16x = m & (~15);
+    BLASLONG tag_m_8x = m & (~7);
+    __mmask8 one_mask = 0xff;
+
+    for (BLASLONG idx_m = 0; idx_m < tag_m_32x; idx_m+=32) {
+        as0 = _mm256_setzero_ps();
+        as1 = _mm256_setzero_ps();
+        as2 = _mm256_setzero_ps();
+        as3 = _mm256_setzero_ps();
+
+        for (BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+            xv = _mm256_set1_ps(x[idx_n]);
+            ma0 = _mm256_maskz_loadu_ps(one_mask, &a[idx_n * lda + idx_m +0]);
+            ma1 = _mm256_maskz_loadu_ps(one_mask, &a[idx_n * lda + idx_m +8]);
+            ma2 = _mm256_maskz_loadu_ps(one_mask, &a[idx_n * lda + idx_m +16]);
+            ma3 = _mm256_maskz_loadu_ps(one_mask, &a[idx_n * lda + idx_m +24]);
+
+            as0 = _mm256_maskz_fmadd_ps(one_mask, ma0, xv, as0);
+            as1 = _mm256_maskz_fmadd_ps(one_mask, ma1, xv, as1);
+            as2 = _mm256_maskz_fmadd_ps(one_mask, ma2, xv, as2);
+            as3 = _mm256_maskz_fmadd_ps(one_mask, ma3, xv, as3);
+        }
+        _mm256_mask_storeu_ps(&y[idx_m], one_mask, _mm256_maskz_fmadd_ps(one_mask, as0, alphav, _mm256_maskz_loadu_ps(one_mask, &y[idx_m])));
+        _mm256_mask_storeu_ps(&y[idx_m + 8], one_mask, _mm256_maskz_fmadd_ps(one_mask, as1, alphav, _mm256_maskz_loadu_ps(one_mask, &y[idx_m + 8])));
+        _mm256_mask_storeu_ps(&y[idx_m + 16], one_mask, _mm256_maskz_fmadd_ps(one_mask, as2, alphav, _mm256_maskz_loadu_ps(one_mask, &y[idx_m + 16])));
+        _mm256_mask_storeu_ps(&y[idx_m + 24], one_mask, _mm256_maskz_fmadd_ps(one_mask, as3, alphav, _mm256_maskz_loadu_ps(one_mask, &y[idx_m + 24])));
+ 
+    }    
+ 
+    if (tag_m_32x != m ) {
+            for (BLASLONG idx_m = tag_m_32x; idx_m < tag_m_16x; idx_m+=16) {
+            as4 = _mm256_setzero_ps();
+            as5 = _mm256_setzero_ps();
+    
+            for (BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+                xv = _mm256_set1_ps(x[idx_n]);
+                ma4 = _mm256_maskz_loadu_ps(one_mask, &a[idx_n * lda + idx_m +0]);
+                ma5 = _mm256_maskz_loadu_ps(one_mask, &a[idx_n * lda + idx_m +8]);
+    
+                as4 = _mm256_maskz_fmadd_ps(one_mask, ma4, xv, as4);
+                as5 = _mm256_maskz_fmadd_ps(one_mask, ma5, xv, as5);
+            }
+            _mm256_mask_storeu_ps(&y[idx_m], one_mask, _mm256_maskz_fmadd_ps(one_mask, as4, alphav, _mm256_maskz_loadu_ps(one_mask, &y[idx_m])));
+            _mm256_mask_storeu_ps(&y[idx_m + 8], one_mask, _mm256_maskz_fmadd_ps(one_mask, as5, alphav, _mm256_maskz_loadu_ps(one_mask, &y[idx_m + 8])));
+        }
+    
+        if (tag_m_16x != m ) {
+            for (BLASLONG idx_m = tag_m_16x; idx_m < tag_m_8x; idx_m+=8) {
+                as6 = _mm256_setzero_ps();
+    
+                for (BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+                    xv = _mm256_set1_ps(x[idx_n]);
+                    ma6 = _mm256_maskz_loadu_ps(one_mask, &a[idx_n * lda + idx_m]);
+                    as6 = _mm256_maskz_fmadd_ps(one_mask, ma6, xv, as6);
+                }
+                _mm256_mask_storeu_ps(&y[idx_m], one_mask, _mm256_maskz_fmadd_ps(one_mask, as6, alphav, _mm256_maskz_loadu_ps(one_mask, &y[idx_m])));
+            }
+        
+            if (tag_m_8x != m) {
+                as7 = _mm256_setzero_ps();
+    
+                unsigned char tail_mask_uint = (((unsigned char)0xff) >> (8-(m&7)));
+                __mmask8 tail_mask = *((__mmask8*) &tail_mask_uint);
+    
+                for(BLASLONG idx_n = 0; idx_n < n; idx_n++) {
+                    xv = _mm256_set1_ps(x[idx_n]);
+                    ma7 = _mm256_maskz_loadu_ps(tail_mask, &a[idx_n * lda + tag_m_8x]);
+    
+                    as7 = _mm256_maskz_fmadd_ps(tail_mask, ma7, xv, as7);
+                }
+    
+                _mm256_mask_storeu_ps(&y[tag_m_8x], tail_mask, _mm256_maskz_fmadd_ps(tail_mask, as7, alphav, _mm256_maskz_loadu_ps(tail_mask, &y[tag_m_8x])));
+    
+            }
+        }
+    }
+    
+    return 0;
+}
+
+
+#endif
\ No newline at end of file
diff --git a/kernel/x86_64/sgemv_t_4.c b/kernel/x86_64/sgemv_t_4.c
index fe886f57f..a36c8ace9 100644
--- a/kernel/x86_64/sgemv_t_4.c
+++ b/kernel/x86_64/sgemv_t_4.c
@@ -34,8 +34,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "sgemv_t_microk_bulldozer-4.c"
 #elif defined(SANDYBRIDGE)
 #include "sgemv_t_microk_sandy-4.c"
-#elif defined(HASWELL) || defined(ZEN) || defined (SKYLAKEX) || defined (COOPERLAKE)
+#elif defined(HASWELL) || defined(ZEN)
 #include "sgemv_t_microk_haswell-4.c"
+#elif defined (SKYLAKEX) || defined (COOPERLAKE)
+#include "sgemv_t_microk_haswell-4.c"
+#include "sgemv_t_microk_skylakex.c"
 #endif
 
 #if defined(STEAMROLLER) || defined(EXCAVATOR)
@@ -305,6 +308,37 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
         if ( m < 1 ) return(0);
         if ( n < 1 ) return(0);
 
+	#ifdef HAVE_SGEMV_T_SKYLAKE_KERNEL
+    if (lda == m && n <= 16384 && m <= 8)
+    {
+        FLOAT * xbuffer_align = x;
+        FLOAT * ybuffer_align = y;
+
+        if (inc_x != 1) {
+            xbuffer_align = buffer;
+            for(BLASLONG i=0; i<m; i++) {
+                 xbuffer_align[i] = x[i*inc_x];
+            }
+        }
+
+        if (inc_y != 1) {
+            ybuffer_align = buffer + m;
+            for(BLASLONG i=0; i<n; i++) {
+                ybuffer_align[i] = y[i*inc_y];
+            }
+        }
+        sgemv_kernel_t(m, n , alpha, a, xbuffer_align, ybuffer_align);
+
+        if(inc_y != 1) {
+            for(BLASLONG i=0; i<n; i++) {
+                   y[i*inc_y] = ybuffer_align[i];
+            }
+        }
+        return(0);
+    }
+
+    #endif
+
 	xbuffer = buffer;
 	ytemp   = buffer + (m < NBMAX ? m : NBMAX);
 	
diff --git a/kernel/x86_64/sgemv_t_microk_skylakex.c b/kernel/x86_64/sgemv_t_microk_skylakex.c
new file mode 100644
index 000000000..dca12acfc
--- /dev/null
+++ b/kernel/x86_64/sgemv_t_microk_skylakex.c
@@ -0,0 +1,60 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+/* need a new enough GCC for avx512 support */
+#if (( defined(__GNUC__)  && __GNUC__   >= 6 && defined(__AVX512CD__)) || (defined(__clang__) && __clang_major__ >= 6))
+
+#define HAVE_SGEMV_T_SKYLAKE_KERNEL 1
+#include "common.h"
+#include <immintrin.h>
+#include "sgemv_t_microk_skylakex_template.c"
+
+//sgemv_t:
+// ----- m -----
+// |<-----------
+// |<-----------
+// n
+// |<-----------
+// |<-----------
+
+static int sgemv_kernel_t(BLASLONG m, BLASLONG n, float alpha, float *a, float *x, float *y)
+{    
+    switch(m) {
+        case 1:  sgemv_kernel_t_1(n, alpha, a, x, y); break;
+        case 2:  sgemv_kernel_t_2(n, alpha, a, x, y); break;
+        case 3:  sgemv_kernel_t_3(n, alpha, a, x, y); break;
+        case 4:  sgemv_kernel_t_4(n, alpha, a, x, y); break;
+        case 5:  sgemv_kernel_t_5(n, alpha, a, x, y); break;
+        case 6:  sgemv_kernel_t_6(n, alpha, a, x, y); break;
+        case 7:  sgemv_kernel_t_7(n, alpha, a, x, y); break;
+        case 8:  sgemv_kernel_t_8(n, alpha, a, x, y); break;
+        default: break;
+    }
+    return 0;
+}
+
+#endif
diff --git a/kernel/x86_64/sgemv_t_microk_skylakex_template.c b/kernel/x86_64/sgemv_t_microk_skylakex_template.c
new file mode 100644
index 000000000..34415054c
--- /dev/null
+++ b/kernel/x86_64/sgemv_t_microk_skylakex_template.c
@@ -0,0 +1,1120 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#include <immintrin.h>
+#include "common.h"
+
+//Here the m means n in sgemv_t:
+// ----- n -----
+// |
+// |
+// m
+// |
+// |
+static int sgemv_kernel_t_1(BLASLONG m, float alpha, float *a, float *x, float *y)
+{
+    //printf("enter into t_1 kernel\n");
+    //printf("m = %ld\n", m);
+    __m512 matrixArray_0, matrixArray_1, matrixArray_2, matrixArray_3, matrixArray_4, matrixArray_5, matrixArray_6, matrixArray_7;
+    float alphaX = alpha * (*x);
+    __m512  ALPHAXVECTOR = _mm512_set1_ps(alphaX);
+    
+    BLASLONG tag_m_128x = m & (~127);
+    BLASLONG tag_m_64x = m & (~63);
+    BLASLONG tag_m_32x = m & (~31);
+    BLASLONG tag_m_16x = m & (~15);
+
+    for (BLASLONG idx_m = 0; idx_m < tag_m_128x; idx_m+=128) {
+            matrixArray_0 = _mm512_loadu_ps(&a[idx_m + 0]);
+            matrixArray_1 = _mm512_loadu_ps(&a[idx_m + 16]);
+            matrixArray_2 = _mm512_loadu_ps(&a[idx_m + 32]);
+            matrixArray_3 = _mm512_loadu_ps(&a[idx_m + 48]);
+            matrixArray_4 = _mm512_loadu_ps(&a[idx_m + 64]);
+            matrixArray_5 = _mm512_loadu_ps(&a[idx_m + 80]);
+            matrixArray_6 = _mm512_loadu_ps(&a[idx_m + 96]);
+            matrixArray_7 = _mm512_loadu_ps(&a[idx_m + 112]);
+            
+        _mm512_storeu_ps(&y[idx_m + 0], _mm512_fmadd_ps(matrixArray_0, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 0])));
+        _mm512_storeu_ps(&y[idx_m + 16], _mm512_fmadd_ps(matrixArray_1, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 16])));
+        _mm512_storeu_ps(&y[idx_m + 32], _mm512_fmadd_ps(matrixArray_2, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 32])));
+        _mm512_storeu_ps(&y[idx_m + 48], _mm512_fmadd_ps(matrixArray_3, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 48])));
+        _mm512_storeu_ps(&y[idx_m + 64], _mm512_fmadd_ps(matrixArray_4, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 64])));
+        _mm512_storeu_ps(&y[idx_m + 80], _mm512_fmadd_ps(matrixArray_5, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 80])));
+        _mm512_storeu_ps(&y[idx_m + 96], _mm512_fmadd_ps(matrixArray_6, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 96])));
+        _mm512_storeu_ps(&y[idx_m + 112], _mm512_fmadd_ps(matrixArray_7, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 112])));
+
+    }
+
+    if (tag_m_128x != m) {
+        for (BLASLONG idx_m = tag_m_128x; idx_m < tag_m_64x; idx_m+=64) {
+            matrixArray_0 = _mm512_loadu_ps(&a[idx_m + 0]);
+            matrixArray_1 = _mm512_loadu_ps(&a[idx_m + 16]);
+            matrixArray_2 = _mm512_loadu_ps(&a[idx_m + 32]);
+            matrixArray_3 = _mm512_loadu_ps(&a[idx_m + 48]);
+            
+            _mm512_storeu_ps(&y[idx_m + 0], _mm512_fmadd_ps(matrixArray_0, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 0])));
+            _mm512_storeu_ps(&y[idx_m + 16], _mm512_fmadd_ps(matrixArray_1, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 16])));
+            _mm512_storeu_ps(&y[idx_m + 32], _mm512_fmadd_ps(matrixArray_2, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 32])));
+            _mm512_storeu_ps(&y[idx_m + 48], _mm512_fmadd_ps(matrixArray_3, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 48])));
+  
+        }
+
+        if (tag_m_64x != m) {
+            for (BLASLONG idx_m = tag_m_64x; idx_m < tag_m_32x; idx_m+=32) {
+                matrixArray_0 = _mm512_loadu_ps(&a[idx_m + 0]);
+                matrixArray_1 = _mm512_loadu_ps(&a[idx_m + 16]);
+
+                _mm512_storeu_ps(&y[idx_m + 0], _mm512_fmadd_ps(matrixArray_0, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 0])));
+                _mm512_storeu_ps(&y[idx_m + 16], _mm512_fmadd_ps(matrixArray_1, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 16])));
+ 
+            }
+
+            if (tag_m_32x != m) {
+                for (BLASLONG idx_m = tag_m_64x; idx_m < tag_m_16x; idx_m+=32) {
+                    matrixArray_0 = _mm512_loadu_ps(&a[idx_m + 0]);
+            
+                    _mm512_storeu_ps(&y[idx_m + 0], _mm512_fmadd_ps(matrixArray_0, ALPHAXVECTOR, _mm512_loadu_ps(&y[idx_m + 0])));
+                }
+            
+                if (tag_m_16x != m) {
+                    unsigned short tail_mask_value = (((unsigned int)0xffff) >> (16-(m&15)));
+                    __mmask16 tail_mask = *((__mmask16*) &tail_mask_value);
+                    matrixArray_0 = _mm512_maskz_loadu_ps(tail_mask, &a[tag_m_16x]);
+
+                    _mm512_mask_storeu_ps(&y[tag_m_16x], tail_mask, _mm512_fmadd_ps(matrixArray_0, ALPHAXVECTOR, _mm512_maskz_loadu_ps(tail_mask, &y[tag_m_16x])));
+
+                }
+
+ 
+            }
+        }
+    }
+
+    return 0;
+}
+
+static int sgemv_kernel_t_2(BLASLONG m, float alpha, float *a, float *x, float *y)
+{
+    __m512 m0, m1, m2, m3, col0_1, col0_2, col1_1, col1_2, x1Array, x2Array;
+    float x1a = x[0] * alpha;
+    float x2a = x[1] * alpha;
+    x1Array = _mm512_set1_ps(x1a);
+    x2Array = _mm512_set1_ps(x2a);
+    BLASLONG tag_m_32x = m & (~31);
+    BLASLONG tag_m_16x = m & (~15);
+    BLASLONG tag_m_8x = m & (~7);
+    __m512i M512_EPI32_1 = _mm512_set1_epi32(1);
+    __m512i idx_base_0 = _mm512_set_epi32(30, 28, 26, 24, 22, 20, 18, 16, 14, 12, 10, 8, 6, 4, 2, 0);
+    __m512i idx_base_1 = _mm512_add_epi32(idx_base_0, M512_EPI32_1);
+
+    for (BLASLONG idx_m = 0; idx_m < tag_m_32x; idx_m+=32) {
+        m0 = _mm512_loadu_ps(&a[idx_m*2]);
+        m1 = _mm512_loadu_ps(&a[idx_m*2 + 16]);
+        m2 = _mm512_loadu_ps(&a[idx_m*2 + 32]);
+        m3 = _mm512_loadu_ps(&a[idx_m*2 + 48]);
+        col0_1 = _mm512_permutex2var_ps(m0, idx_base_0, m1);
+        col0_2 = _mm512_permutex2var_ps(m0, idx_base_1, m1);
+        col1_1 = _mm512_permutex2var_ps(m2, idx_base_0, m3);
+        col1_2 = _mm512_permutex2var_ps(m2, idx_base_1, m3);
+
+        _mm512_storeu_ps(&y[idx_m], _mm512_add_ps(_mm512_fmadd_ps(x2Array, col0_2, _mm512_mul_ps(col0_1, x1Array)), _mm512_loadu_ps(&y[idx_m])));
+        _mm512_storeu_ps(&y[idx_m + 16], _mm512_add_ps(_mm512_fmadd_ps(x2Array, col1_2, _mm512_mul_ps(col1_1, x1Array)), _mm512_loadu_ps(&y[idx_m + 16])));
+    }
+    if (tag_m_32x != m) {
+        for (BLASLONG idx_m = tag_m_32x; idx_m < tag_m_16x; idx_m+=16) {
+            m0 = _mm512_loadu_ps(&a[idx_m]);
+            m1 = _mm512_loadu_ps(&a[idx_m + 16]);
+            col1_1 = _mm512_permutex2var_ps(m0, idx_base_0, m1);
+            col1_2 = _mm512_permutex2var_ps(m0, idx_base_1, m1);
+            _mm512_storeu_ps(&y[idx_m], _mm512_add_ps(_mm512_fmadd_ps(x2Array, col1_2, _mm512_mul_ps(col1_1, x1Array)), _mm512_loadu_ps(&y[idx_m])));
+        }
+        if (tag_m_16x != m) {
+            __m512  ALPHAVECTOR = _mm512_set1_ps(alpha);
+            unsigned char load_mask_value = (((unsigned char)0xff) >> 6);
+            __mmask8 load_mask = *((__mmask8*) &load_mask_value);
+            x1Array = _mm512_broadcast_f32x2(_mm_maskz_loadu_ps(load_mask, x));
+            for (BLASLONG idx_m = tag_m_16x; idx_m < tag_m_8x; idx_m+=8) {
+                m0 = _mm512_loadu_ps(&a[idx_m]);
+                m1 = _mm512_mul_ps(_mm512_mul_ps(m0, x1Array), ALPHAVECTOR);
+                m2 = _mm512_permutexvar_ps(_mm512_set_epi32(15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0), m1);
+                __m256 ret = _mm256_add_ps(_mm512_extractf32x8_ps(m2, 1), _mm512_extractf32x8_ps(m2, 0));
+                _mm256_storeu_ps(&y[idx_m], _mm256_add_ps(ret, _mm256_loadu_ps(&y[idx_m])));
+                 
+            }
+
+            if (tag_m_8x != m) {
+                unsigned short tail_mask_value = (((unsigned int)0xffff) >> (16-((m-tag_m_8x)*2)&15));
+                __mmask16 a_mask = *((__mmask16*) &tail_mask_value);
+                unsigned char y_mask_value = (((unsigned char)0xff) >> (8-(m-tag_m_8x)));
+                __mmask8 y_mask = *((__mmask8*) &y_mask_value);
+
+                m0 = _mm512_maskz_loadu_ps(a_mask, &a[tag_m_8x]);
+                m1 = _mm512_mul_ps(_mm512_mul_ps(m0, x1Array), ALPHAVECTOR);
+                m2 = _mm512_permutexvar_ps(_mm512_set_epi32(15, 13, 11, 9, 7, 5, 3, 1, 14, 12, 10, 8, 6, 4, 2, 0), m1);
+                __m256 ret = _mm256_add_ps(_mm512_extractf32x8_ps(m2, 1), _mm512_extractf32x8_ps(m2, 0));
+                _mm256_mask_storeu_ps(&y[tag_m_8x], y_mask, _mm256_add_ps(ret, _mm256_maskz_loadu_ps(y_mask, &y[tag_m_8x])));
+            }                  
+        }        
+    }
+    return 0;
+}
+
+static int sgemv_kernel_t_3(BLASLONG m, float alpha, float *a, float *x, float *y)
+{
+    __m512 m0, m1, m2, c1, c2, c3, tmp, x1Array, x2Array, x3Array;
+    float x1a = x[0] * alpha;
+    float x2a = x[1] * alpha;
+    float x3a = x[2] * alpha;
+    x1Array = _mm512_set1_ps(x1a);
+    x2Array = _mm512_set1_ps(x2a);
+    x3Array = _mm512_set1_ps(x3a);
+    BLASLONG tag_m_16x = m & (~15);
+    BLASLONG tag_m_8x = m & (~7);
+    BLASLONG tag_m_4x = m & (~3);
+    BLASLONG tag_m_2x = m & (~1);
+
+    __m512i M512_EPI32_1 = _mm512_set1_epi32(1);
+    __m512i M512_EPI32_s1 = _mm512_set1_epi32(-1);
+    __m512i idx_c1_1 = _mm512_set_epi32(0, 0, 0, 0, 0, 30, 27, 24, 21, 18, 15, 12, 9, 6, 3, 0);
+    __m512i idx_c2_1 = _mm512_add_epi32(idx_c1_1, M512_EPI32_1);
+    __m512i idx_c3_1 = _mm512_add_epi32(idx_c2_1, M512_EPI32_1);
+
+    __m512i idx_c3_2 = _mm512_set_epi32(31, 28, 25, 22, 19, 16, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+    __m512i idx_c2_2 = _mm512_add_epi32(idx_c3_2, M512_EPI32_s1);
+    __m512i idx_c1_2 = _mm512_add_epi32(idx_c2_2, M512_EPI32_s1);
+
+    __mmask16 step_1 = 0x07ff;
+    __mmask16 step_2 = 0xf800;
+    __mmask16 c31 = 0x03ff;
+
+    for (BLASLONG idx_m = 0; idx_m < tag_m_16x; idx_m+=16) {
+        m0 = _mm512_loadu_ps(&a[idx_m*3]);
+        m1 = _mm512_loadu_ps(&a[idx_m*3 + 16]);
+        m2 = _mm512_loadu_ps(&a[idx_m*3 + 32]);
+
+        tmp = _mm512_mask_permutex2var_ps(m0, step_1, idx_c1_1, m1);
+        c1 = _mm512_mask_permutex2var_ps(tmp, step_2, idx_c1_2, m2);
+        tmp = _mm512_mask_permutex2var_ps(m0, step_1, idx_c2_1, m1);
+        c2 = _mm512_mask_permutex2var_ps(tmp, step_2, idx_c2_2, m2);
+        tmp = _mm512_mask_permutex2var_ps(m0, c31, idx_c3_1, m1);
+        c3 = _mm512_permutex2var_ps(tmp, idx_c3_2, m2);
+
+        tmp = _mm512_fmadd_ps(x2Array, c2, _mm512_mul_ps(c1, x1Array));
+        _mm512_storeu_ps(&y[idx_m], _mm512_add_ps(_mm512_fmadd_ps(x3Array, c3, tmp), _mm512_loadu_ps(&y[idx_m])));
+    }
+
+    if(tag_m_16x != m) {
+        __mmask8 a_mask = 0xff;
+        __m256i M256_EPI32_1 = _mm256_maskz_set1_epi32(a_mask, 1);
+        __m256i M256_EPI32_s1 = _mm256_maskz_set1_epi32(a_mask, -1);
+        __m256i idx_c1_1 = _mm256_set_epi32(0, 0, 15, 12, 9, 6, 3, 0);
+        __m256i idx_c2_1 = _mm256_add_epi32(idx_c1_1, M256_EPI32_1);
+        __m256i idx_c3_1 = _mm256_add_epi32(idx_c2_1, M256_EPI32_1);
+
+        __m256i idx_c3_2 = _mm256_set_epi32(15, 12, 9, 0, 0, 0, 0, 0);
+        __m256i idx_c2_2 = _mm256_add_epi32(idx_c3_2, M256_EPI32_s1);
+        __m256i idx_c1_2 = _mm256_add_epi32(idx_c2_2, M256_EPI32_s1);
+
+        __mmask8 step_1 = 0x1f;
+        __mmask8 step_2 = 0xe0;
+        __mmask8 c12 = 0xc0;
+        
+        __m256 m256_0, m256_1, m256_2, tmp256, c256_1, c256_2, c256_3, x256_1, x256_2, x256_3;
+        x256_1 = _mm256_set1_ps(x1a);
+        x256_2 = _mm256_set1_ps(x2a);
+        x256_3 = _mm256_set1_ps(x3a);
+
+        for (BLASLONG idx_m = tag_m_16x; idx_m < tag_m_8x; idx_m+=8) {
+            m256_0 = _mm256_loadu_ps(&a[idx_m*3]);
+            m256_1 = _mm256_loadu_ps(&a[idx_m*3 + 8]);
+            m256_2 = _mm256_loadu_ps(&a[idx_m*3 + 16]);
+
+            tmp256 = _mm256_permutex2var_ps(m256_0, idx_c1_1, m256_1);
+            c256_1 = _mm256_mask_permutex2var_ps(tmp256, c12, idx_c1_2, m256_2);
+            tmp256 = _mm256_mask_permutex2var_ps(m256_0, step_1, idx_c2_1, m256_1);
+            c256_2 = _mm256_mask_permutex2var_ps(tmp256, step_2, idx_c2_2, m256_2);
+            tmp256 = _mm256_mask_permutex2var_ps(m256_0, step_1, idx_c3_1, m256_1);
+            c256_3 = _mm256_mask_permutex2var_ps(tmp256, step_2, idx_c3_2, m256_2);
+
+            tmp256 = _mm256_fmadd_ps(x256_2, c256_2, _mm256_mul_ps(c256_1, x256_1));
+            _mm256_storeu_ps(&y[idx_m], _mm256_maskz_add_ps(a_mask, _mm256_fmadd_ps(x256_3, c256_3, tmp256), _mm256_loadu_ps(&y[idx_m])));
+        }
+
+        if(tag_m_8x != m){
+            for (BLASLONG idx_m = tag_m_8x; idx_m < tag_m_4x; idx_m+=4){
+                m0 = _mm512_maskz_loadu_ps(0x0fff, &a[tag_m_8x*3]);
+                m256_0 = _mm512_extractf32x8_ps(m0, 0);
+                m256_1 = _mm512_extractf32x8_ps(m0, 1);
+                __m256i idx1 = _mm256_set_epi32(10, 7, 4, 1, 9, 6, 3, 0);
+                __m256i M256_EPI32_2 = _mm256_maskz_set1_epi32(0x0f, 2);
+                __m256i idx2 = _mm256_add_epi32(idx1, M256_EPI32_2);
+
+                c256_1 = _mm256_mask_permutex2var_ps(m256_0, 0xff, idx1, m256_1);
+                c256_2 = _mm256_mask_permutex2var_ps(m256_0, 0x0f, idx2, m256_1);
+
+                __m128 c128_1 = _mm256_extractf32x4_ps(c256_1, 0);
+                __m128 c128_2 = _mm256_extractf32x4_ps(c256_1, 1);
+                __m128 c128_3 = _mm256_extractf32x4_ps(c256_2, 0);
+
+                __m128 x128_1 = _mm_set1_ps(x1a);
+                __m128 x128_2 = _mm_set1_ps(x2a);
+                __m128 x128_3 = _mm_set1_ps(x3a);
+
+                __m128 tmp128 = _mm_maskz_fmadd_ps(0x0f, c128_1, x128_1, _mm_maskz_mul_ps(0x0f, c128_2, x128_2));
+                _mm_mask_storeu_ps(&y[idx_m], 0x0f, _mm_maskz_add_ps(0x0f, _mm_maskz_fmadd_ps(0x0f, c128_3, x128_3, tmp128), _mm_maskz_loadu_ps(0x0f, &y[idx_m])));
+            }
+
+            if(tag_m_4x != m) {
+                for (BLASLONG idx_m = tag_m_4x; idx_m < tag_m_2x; idx_m+=2) {
+                    m256_0 = _mm256_maskz_loadu_ps(0x3f, &a[idx_m*3]);
+                    __m128 a128_1 = _mm256_extractf32x4_ps(m256_0, 0);
+                    __m128 a128_2 = _mm256_extractf32x4_ps(m256_0, 1);
+                    __m128 x128 = _mm_maskz_loadu_ps(0x07, x);
+
+                    __m128i idx128_1= _mm_set_epi32(0, 2, 1, 0);
+                    __m128i M128_EPI32_3 = _mm_maskz_set1_epi32(0x07, 3);
+                    __m128i idx128_2 = _mm_add_epi32(idx128_1, M128_EPI32_3);
+
+                    __m128 c128_1 = _mm_maskz_permutex2var_ps(0x07, a128_1, idx128_1, a128_2);
+                    __m128 c128_2 = _mm_maskz_permutex2var_ps(0x07, a128_1, idx128_2, a128_2);
+
+                    __m128 tmp128 = _mm_hadd_ps(_mm_maskz_mul_ps(0x07, c128_1, x128), _mm_maskz_mul_ps(0x07, c128_2, x128));
+                    float ret[4];
+                    _mm_mask_storeu_ps(ret, 0x0f, tmp128);
+                    y[idx_m] += alpha *(ret[0] + ret[1]);
+                    y[idx_m+1] += alpha * (ret[2] + ret[3]);
+                }
+
+                if(tag_m_2x != m) {
+                    y[tag_m_2x] += alpha*(a[tag_m_2x*3]*x[0] + a[tag_m_2x*3+1]*x[1] + a[tag_m_2x*3+2]*x[2]);
+                }
+            }
+        }
+    }
+
+    return 0;
+}
+
+static int sgemv_kernel_t_4(BLASLONG m, float alpha, float *a, float *x, float *y)
+{
+    BLASLONG tag_m_4x = m & (~3);
+    BLASLONG tag_m_2x = m & (~1);
+    __m512 m0, m1, m2;
+    __m256 m256_0, m256_1, c256_1, c256_2;
+    __m128 c1, c2, c3, c4, ret;
+    __m128 xarray = _mm_maskz_loadu_ps(0x0f, x);
+    __m512 x512 = _mm512_broadcast_f32x4(xarray);
+    __m512 alphavector = _mm512_set1_ps(alpha);
+    __m512 xa512 = _mm512_mul_ps(x512, alphavector);
+    __m256i idx1 = _mm256_set_epi32(13, 9, 5, 1, 12, 8, 4, 0);
+    __m256i idx2 = _mm256_set_epi32(15, 11, 7, 3, 14, 10, 6, 2);
+
+
+    for (BLASLONG idx_m = 0; idx_m < tag_m_4x; idx_m+=4) {
+        m0 = _mm512_loadu_ps(&a[idx_m*4]);
+        m1 = _mm512_mul_ps(m0, xa512);
+        m256_0 = _mm512_extractf32x8_ps(m1, 0);
+        m256_1 = _mm512_extractf32x8_ps(m1, 1);
+        c256_1 = _mm256_mask_permutex2var_ps(m256_0, 0xff, idx1, m256_1);
+        c256_2 = _mm256_mask_permutex2var_ps(m256_0, 0xff, idx2, m256_1);
+
+        c1 = _mm256_extractf32x4_ps(c256_1, 0);
+        c2 = _mm256_extractf32x4_ps(c256_1, 1);
+        c3 = _mm256_extractf32x4_ps(c256_2, 0);
+        c4 = _mm256_extractf32x4_ps(c256_2, 1);
+
+        ret = _mm_maskz_add_ps(0xff, _mm_maskz_add_ps(0xff, _mm_maskz_add_ps(0xff, c1, c2), _mm_maskz_add_ps(0xff, c3, c4)), _mm_maskz_loadu_ps(0xff, y));
+        _mm_mask_storeu_ps(&y[idx_m], 0xff, ret);
+    }
+    
+    if(tag_m_4x != m) {
+        float result[4];
+        for(BLASLONG idx_m=tag_m_4x; idx_m < tag_m_2x; idx_m+=2) {
+            m256_0 = _mm256_maskz_loadu_ps(0xff, &a[idx_m*4]);
+            c1 = _mm256_maskz_extractf32x4_ps(0xff, m256_0, 0);
+            c2 = _mm256_maskz_extractf32x4_ps(0xff, m256_0, 1);
+
+            c3 = _mm_maskz_mul_ps(0x0f, c1, xarray);
+            c4 = _mm_maskz_mul_ps(0x0f, c2, xarray);
+
+            ret = _mm_hadd_ps(c3, c4);            
+            _mm_mask_storeu_ps(result, 0x0f, ret);
+            y[idx_m] += alpha *(result[0] + result[1]);
+            y[idx_m+1] += alpha * (result[2] + result[3]);
+        }
+
+        if(tag_m_2x != m ) {
+            c1 = _mm_maskz_loadu_ps(0x0f, &a[tag_m_2x * 4]);
+            c2 = _mm_maskz_mul_ps(0x0f, c1, xarray);
+            _mm_mask_storeu_ps(result, 0x0f, c2);
+            y[tag_m_2x] += alpha *(result[0] + result[1] + result[2] + result[3]);
+        }
+    }
+
+    return 0;
+}
+
+static int sgemv_kernel_t_5(BLASLONG m, float alpha, float *a, float *x, float *y)
+{
+    BLASLONG tag_m_16x = m & (~15);
+    BLASLONG tag_m_8x = m & (~7);
+    BLASLONG tag_m_4x = m & (~3);
+    BLASLONG tag_m_2x = m & (~1);
+    __m512 m0, m1, m2, m3, m4, tmp0, tmp1, tmp2, accum, c0, c1, c2, c3, c4;
+    __m512 x0_512 = _mm512_set1_ps(x[0]);
+    __m512 x1_512 = _mm512_set1_ps(x[1]);
+    __m512 x2_512 = _mm512_set1_ps(x[2]);
+    __m512 x3_512 = _mm512_set1_ps(x[3]);
+    __m512 x4_512 = _mm512_set1_ps(x[4]);
+    __m512 alpha_512 = _mm512_set1_ps(alpha);
+
+
+    __m512i M512_EPI32_1 = _mm512_set1_epi32(1);
+    __m512i M512_EPI32_16 = _mm512_set1_epi32(16);
+    __m512i M512_EPI32_0 = _mm512_setzero_epi32();
+
+    __m512i idx_c0 = _mm512_set_epi32(27, 22, 17, 28, 23, 18, 13, 8, 3, 30, 25, 20, 15, 10, 5, 0);
+    __m512i idx_c1 = _mm512_add_epi32(idx_c0, M512_EPI32_1);
+    __m512i idx_c2 = _mm512_add_epi32(idx_c1, M512_EPI32_1);
+    idx_c2 = _mm512_mask_blend_epi32(0x0040, idx_c2, M512_EPI32_0);
+    __m512i idx_c3 = _mm512_add_epi32(idx_c2, M512_EPI32_1);
+    __m512i idx_c4 = _mm512_add_epi32(idx_c3, M512_EPI32_1);
+    idx_c4 = _mm512_mask_blend_epi32(0x1000, idx_c4, M512_EPI32_16);
+
+    for (BLASLONG idx_m=0; idx_m < tag_m_16x; idx_m+=16) {      
+        m0 = _mm512_loadu_ps(&a[idx_m*5]);
+        m1 = _mm512_loadu_ps(&a[idx_m*5 + 16]);
+        m2 = _mm512_loadu_ps(&a[idx_m*5 + 32]);
+        m3 = _mm512_loadu_ps(&a[idx_m*5 + 48]);
+        m4 = _mm512_loadu_ps(&a[idx_m*5 + 64]);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x007f, m0, idx_c0, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x1f80, m2, idx_c0, m3);
+        c0 = _mm512_mask_blend_ps(0x1f80, tmp0, tmp1);
+        c0 = _mm512_mask_permutex2var_ps(c0, 0xe000, idx_c0, m4);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x007f, m0, idx_c1, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x1f80, m2, idx_c1, m3);
+        c1 = _mm512_mask_blend_ps(0x1f80, tmp0, tmp1);
+        c1 = _mm512_mask_permutex2var_ps(c1, 0xe000, idx_c1, m4);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x003f, m0, idx_c2, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x1fc0, m2, idx_c2, m3);
+        c2 = _mm512_mask_blend_ps(0x1fc0, tmp0, tmp1);
+        c2 = _mm512_mask_permutex2var_ps(c2, 0xe000, idx_c2, m4);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x003f, m0, idx_c3, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x1fc0, m2, idx_c3, m3);
+        c3 = _mm512_mask_blend_ps(0x1fc0, tmp0, tmp1);
+        c3 = _mm512_mask_permutex2var_ps(c3, 0xe000, idx_c3, m4);
+       
+        tmp0 = _mm512_maskz_permutex2var_ps(0x003f, m0, idx_c4, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x0fc0, m2, idx_c4, m3);
+        c4 = _mm512_mask_blend_ps(0x0fc0, tmp0, tmp1);
+        c4 = _mm512_mask_permutex2var_ps(c4, 0xf000, idx_c4, m4);
+       
+        accum = _mm512_fmadd_ps(c1, x1_512, _mm512_mul_ps(c0, x0_512));
+        accum = _mm512_fmadd_ps(c2, x2_512, accum);
+        accum = _mm512_fmadd_ps(c3, x3_512, accum);
+        accum = _mm512_fmadd_ps(c4, x4_512, accum);
+        accum = _mm512_fmadd_ps(accum, alpha_512, _mm512_loadu_ps(&y[idx_m]));
+        _mm512_storeu_ps(&y[idx_m], accum);
+
+    }
+    if(tag_m_16x !=m) {
+        __m512i idx_c0c2 = _mm512_set_epi32(0, 0, 27, 22, 17, 12, 7, 2 , 0, 30, 25, 20, 15, 10, 5, 0);
+        __m512i idx_c1c3 = _mm512_add_epi32(idx_c0c2, M512_EPI32_1);
+        idx_c4 = _mm512_add_epi32(idx_c1c3, M512_EPI32_1);
+        __m256i idx_c0m4 = _mm256_set_epi32(11, 6, 0, 0, 0, 0, 0, 0);
+        __m256i M256_EPI32_1 = _mm256_set1_epi32(1);
+        __m256i idx_c1m4 = _mm256_add_epi32(idx_c0m4, M256_EPI32_1);
+        __m256i idx_c2m4 = _mm256_add_epi32(idx_c1m4, M256_EPI32_1);
+        __m256i idx_c3m4 = _mm256_add_epi32(idx_c2m4, M256_EPI32_1);
+        __m256i idx_c4m4 = _mm256_add_epi32(idx_c3m4, M256_EPI32_1);
+        //TODO: below can change to use extract to decrease the latency
+        __m256 x0_256 = _mm256_set1_ps(x[0]);
+        __m256 x1_256 = _mm256_set1_ps(x[1]);
+        __m256 x2_256 = _mm256_set1_ps(x[2]);
+        __m256 x3_256 = _mm256_set1_ps(x[3]);
+        __m256 x4_256 = _mm256_set1_ps(x[4]);
+        __m256 alpha256 = _mm256_set1_ps(alpha);
+        __m256 accum_256, m256_4;
+
+        for(BLASLONG idx_m=tag_m_16x; idx_m < tag_m_8x; idx_m+=8) {
+            m0 = _mm512_loadu_ps(&a[idx_m*5]);
+            m1 = _mm512_loadu_ps(&a[idx_m*5 + 16]);
+            m256_4 = _mm256_loadu_ps(&a[idx_m*5 + 32]);
+            tmp0 = _mm512_permutex2var_ps(m0, idx_c0c2, m1);
+            tmp1 = _mm512_permutex2var_ps(m0, idx_c1c3, m1);
+            tmp2 = _mm512_permutex2var_ps(m0, idx_c4, m1);
+
+            __m256 c256_0 = _mm512_extractf32x8_ps(tmp0, 0);
+            __m256 c256_2 = _mm512_extractf32x8_ps(tmp0, 1);
+            __m256 c256_1 = _mm512_extractf32x8_ps(tmp1, 0);
+            __m256 c256_3 = _mm512_extractf32x8_ps(tmp1, 1);
+            __m256 c256_4 = _mm512_extractf32x8_ps(tmp2, 1);
+
+            c256_0 = _mm256_mask_permutex2var_ps(c256_0, 0x80, idx_c0m4, m256_4);
+            c256_1 = _mm256_mask_permutex2var_ps(c256_1, 0x80, idx_c1m4, m256_4);
+            c256_2 = _mm256_mask_permutex2var_ps(c256_2, 0xc0, idx_c2m4, m256_4);
+            c256_3 = _mm256_mask_permutex2var_ps(c256_3, 0xc0, idx_c3m4, m256_4);
+            c256_4 = _mm256_mask_permutex2var_ps(c256_4, 0xc0, idx_c4m4, m256_4);
+            
+            accum_256 = _mm256_fmadd_ps(c256_1, x1_256, _mm256_mul_ps(c256_0, x0_256));
+            accum_256 = _mm256_fmadd_ps(c256_2, x2_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_3, x3_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_4, x4_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(accum_256, alpha256, _mm256_loadu_ps(&y[idx_m]));
+            _mm256_storeu_ps(&y[idx_m], accum_256);
+        }
+        if(tag_m_8x != m) {
+            __m256i idx_c02 = _mm256_set_epi32(17, 12, 7, 2, 15, 10, 5, 0);
+            __m256i idx_c13 = _mm256_add_epi32(idx_c02, M256_EPI32_1);
+            __m256i idx_4 = _mm256_add_epi32(idx_c13, M256_EPI32_1); 
+            __m128 accum_128;
+            __m256 m256_0, m256_1, tmp256_0, tmp256_1;
+            for (BLASLONG idx_m = tag_m_8x; idx_m < tag_m_4x; idx_m+=4){
+                m256_0 = _mm256_loadu_ps(&a[idx_m*5]);
+                m256_1 = _mm256_loadu_ps(&a[idx_m*5 + 8]);
+                __m128 m128_4 = _mm_maskz_loadu_ps(0x0f, &a[idx_m*5 + 16]);
+
+                tmp256_0 = _mm256_permutex2var_ps(m256_0, idx_c02, m256_1);
+                tmp256_1 = _mm256_permutex2var_ps(m256_0, idx_c13, m256_1);
+                __m256 tmp256_2 = _mm256_maskz_permutex2var_ps(0xf0, m256_0, idx_4, m256_1);
+
+                __m128 c128_0 = _mm256_extractf32x4_ps(tmp256_0, 0);
+                __m128 c128_1 = _mm256_extractf32x4_ps(tmp256_1, 0);
+                __m128 c128_2 = _mm256_extractf32x4_ps(tmp256_0, 1);
+                __m128 c128_3 = _mm256_extractf32x4_ps(tmp256_1, 1);
+                __m128 c128_4 = _mm256_extractf32x4_ps(tmp256_2, 1);
+
+                __m128i idx_c14 = _mm_set_epi32(4, 0, 0, 0);
+                __m128i M128_EPI32_1 = _mm_set1_epi32(1);
+                __m128i idx_c24 = _mm_add_epi32(idx_c14, M128_EPI32_1);
+                __m128i idx_c34 = _mm_add_epi32(idx_c24, M128_EPI32_1);
+                __m128i idx_c44 = _mm_add_epi32(idx_c34, M128_EPI32_1);
+
+                c128_1 = _mm_mask_permutex2var_ps(c128_1, 0x08, idx_c14, m128_4);
+                c128_2 = _mm_mask_permutex2var_ps(c128_2, 0x08, idx_c24, m128_4);
+                c128_3 = _mm_mask_permutex2var_ps(c128_3, 0x08, idx_c34, m128_4);
+                c128_4 = _mm_mask_permutex2var_ps(c128_4, 0x08, idx_c44, m128_4);
+
+                __m128 x128_0 = _mm256_extractf32x4_ps(x0_256, 0);
+                __m128 x128_1 = _mm256_extractf32x4_ps(x1_256, 0);
+                __m128 x128_2 = _mm256_extractf32x4_ps(x2_256, 0);
+                __m128 x128_3 = _mm256_extractf32x4_ps(x3_256, 0);
+                __m128 x128_4 = _mm256_extractf32x4_ps(x4_256, 0);
+
+                __m128 alpha_128 = _mm256_extractf32x4_ps(alpha256, 0);
+                accum_128 = _mm_maskz_fmadd_ps(0x0f, c128_1, x128_1, _mm_maskz_mul_ps(0x0f, c128_0, x128_0));
+                accum_128 = _mm_maskz_fmadd_ps(0x0f, c128_2, x128_2, accum_128);
+                accum_128 = _mm_maskz_fmadd_ps(0x0f, c128_3, x128_3, accum_128);
+                accum_128 = _mm_maskz_fmadd_ps(0x0f, c128_4, x128_4, accum_128);
+                accum_128 = _mm_maskz_fmadd_ps(0x0f, accum_128, alpha_128, _mm_maskz_loadu_ps(0x0f, &y[idx_m]));
+                _mm_mask_storeu_ps(&y[idx_m], 0x0f, accum_128);
+
+            }
+
+            if(tag_m_4x !=m ){
+                x0_256 = _mm256_maskz_loadu_ps(0x1f, x);
+                x0_256 = _mm256_mul_ps(x0_256, alpha256);
+                float ret8[8];
+
+                for(BLASLONG idx_m = tag_m_4x; idx_m < tag_m_2x; idx_m+=2){
+                    m256_0 = _mm256_maskz_loadu_ps(0x1f, &a[idx_m*5]);
+                    m256_1 = _mm256_maskz_loadu_ps(0x1f, &a[idx_m*5 + 5]);
+
+                    m256_0 = _mm256_mul_ps(m256_0, x0_256);
+                    m256_1 = _mm256_mul_ps(m256_1, x0_256);
+
+                    _mm256_mask_storeu_ps(ret8, 0x1f, m256_0);
+                    y[idx_m] += ret8[0] + ret8[1] + ret8[2] + ret8[3] + ret8[4];
+                    _mm256_mask_storeu_ps(ret8, 0x1f, m256_1);
+                    y[idx_m+1] += ret8[0] + ret8[1] + ret8[2] + ret8[3] + ret8[4];                    
+
+                }
+
+                if(tag_m_2x != m){
+                    m256_0 = _mm256_maskz_loadu_ps(0x1f, &a[tag_m_2x*5]);
+                    m256_0 = _mm256_mul_ps(m256_0, x0_256);
+
+
+                    _mm256_mask_storeu_ps(ret8, 0x1f, m256_0);
+                    y[tag_m_2x] += ret8[0] + ret8[1] + ret8[2] + ret8[3] + ret8[4];
+
+                }
+            }
+        }
+
+    }
+    return 0;
+}
+
+static int sgemv_kernel_t_6(BLASLONG m, float alpha, float *a, float *x, float *y)
+{
+    BLASLONG tag_m_16x = m & (~15);
+    BLASLONG tag_m_8x = m & (~7);
+    BLASLONG tag_m_4x = m & (~3);
+    BLASLONG tag_m_2x = m & (~1);
+
+    __m512 m0, m1, m2, m3, m4, m5, c0, c1, c2, c3, c4, c5, tmp0, tmp1, tmp2, accum;
+    __m512i idx_c0 = _mm512_set_epi32(26, 20, 14, 8, 2, 28, 22, 16, 10, 4, 30, 24, 18, 12, 6, 0);
+    __m512i M512_EPI32_1 = _mm512_set1_epi32(1);
+    __m512i M512_EPI32_0 = _mm512_setzero_epi32();
+    __m512i M512_EPI32_16 = _mm512_set1_epi32(16);
+    __m512i idx_c1 = _mm512_add_epi32(idx_c0, M512_EPI32_1);
+    __m512i idx_c2 = _mm512_add_epi32(idx_c1, M512_EPI32_1);
+    idx_c2 = _mm512_mask_blend_epi32(0x0020, idx_c2, M512_EPI32_0);
+    __m512i idx_c3 = _mm512_add_epi32(idx_c2, M512_EPI32_1);
+    __m512i idx_c4 = _mm512_add_epi32(idx_c3, M512_EPI32_1);
+    idx_c4 = _mm512_mask_blend_epi32(0x0400, idx_c4, M512_EPI32_0);
+    __m512i idx_c5 = _mm512_add_epi32(idx_c4, M512_EPI32_1);
+
+    __m512 x0_512 = _mm512_set1_ps(x[0]);
+    __m512 x1_512 = _mm512_set1_ps(x[1]);
+    __m512 x2_512 = _mm512_set1_ps(x[2]);
+    __m512 x3_512 = _mm512_set1_ps(x[3]);
+    __m512 x4_512 = _mm512_set1_ps(x[4]);
+    __m512 x5_512 = _mm512_set1_ps(x[5]);
+    __m512 alpha_512 = _mm512_set1_ps(alpha);
+
+    for (BLASLONG idx_m=0; idx_m < tag_m_16x; idx_m+=16) {      
+        m0 = _mm512_loadu_ps(&a[idx_m*6]);
+        m1 = _mm512_loadu_ps(&a[idx_m*6 + 16]);
+        m2 = _mm512_loadu_ps(&a[idx_m*6 + 32]);
+        m3 = _mm512_loadu_ps(&a[idx_m*6 + 48]);
+        m4 = _mm512_loadu_ps(&a[idx_m*6 + 64]);
+        m5 = _mm512_loadu_ps(&a[idx_m*6 + 80]);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x003f, m0, idx_c0, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x07c0, m2, idx_c0, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0xf800, m4, idx_c0, m5);
+        c0 = _mm512_mask_blend_ps(0x07c0, tmp0, tmp1);
+        c0 = _mm512_mask_blend_ps(0xf800, c0, tmp2);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x003f, m0, idx_c1, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x07c0, m2, idx_c1, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0xf800, m4, idx_c1, m5);
+        c1 = _mm512_mask_blend_ps(0x07c0, tmp0, tmp1);
+        c1 = _mm512_mask_blend_ps(0xf800, c1, tmp2);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x001f, m0, idx_c2, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x07e0, m2, idx_c2, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0xf800, m4, idx_c2, m5);
+        c2 = _mm512_mask_blend_ps(0x07e0, tmp0, tmp1);
+        c2 = _mm512_mask_blend_ps(0xf800, c2, tmp2);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x001f, m0, idx_c3, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x07e0, m2, idx_c3, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0xf800, m4, idx_c3, m5);
+        c3 = _mm512_mask_blend_ps(0x07e0, tmp0, tmp1);
+        c3 = _mm512_mask_blend_ps(0xf800, c3, tmp2);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x001f, m0, idx_c4, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x03e0, m2, idx_c4, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0xfc00, m4, idx_c4, m5);
+        c4 = _mm512_mask_blend_ps(0x03e0, tmp0, tmp1);
+        c4 = _mm512_mask_blend_ps(0xfc00, c4, tmp2);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x001f, m0, idx_c5 , m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x03e0, m2, idx_c5 , m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0xfc00, m4, idx_c5 , m5);
+        c5 = _mm512_mask_blend_ps(0x03e0, tmp0, tmp1);
+        c5 = _mm512_mask_blend_ps(0xfc00, c5, tmp2);
+
+        accum = _mm512_fmadd_ps(c1, x1_512, _mm512_mul_ps(c0, x0_512));
+        accum = _mm512_fmadd_ps(c2, x2_512, accum);
+        accum = _mm512_fmadd_ps(c3, x3_512, accum);
+        accum = _mm512_fmadd_ps(c4, x4_512, accum);
+        accum = _mm512_fmadd_ps(c5, x5_512, accum);
+        accum = _mm512_fmadd_ps(accum, alpha_512, _mm512_loadu_ps(&y[idx_m]));
+        _mm512_storeu_ps(&y[idx_m], accum);
+    }
+
+    if(tag_m_16x != m) {
+        __m512i idx_c0c3 = _mm512_set_epi32(29, 23, 17, 27, 21, 15, 9, 3, 26, 20, 30, 24, 18, 12, 6, 0);
+        __m512i idx_c1c4 = _mm512_add_epi32(idx_c0c3, M512_EPI32_1);
+        __m512i idx_c2c5 = _mm512_add_epi32(idx_c1c4, M512_EPI32_1);
+        idx_c2c5 = _mm512_mask_blend_epi32(0x0020, idx_c2c5, M512_EPI32_16);
+        __m256 c256_0, c256_1, c256_2, c256_3, c256_4, c256_5;
+
+        __m256 x0_256 = _mm256_set1_ps(x[0]);
+        __m256 x1_256 = _mm256_set1_ps(x[1]);
+        __m256 x2_256 = _mm256_set1_ps(x[2]);
+        __m256 x3_256 = _mm256_set1_ps(x[3]);
+        __m256 x4_256 = _mm256_set1_ps(x[4]);
+        __m256 x5_256 = _mm256_set1_ps(x[5]);
+        __m256 alpha256 = _mm256_set1_ps(alpha);
+        __m256 accum_256;
+
+        for(BLASLONG idx_m = tag_m_16x; idx_m <tag_m_8x; idx_m+=8){
+            m0 = _mm512_loadu_ps(&a[idx_m*6]);
+            m1 = _mm512_loadu_ps(&a[idx_m*6 + 16]);
+            m2 = _mm512_loadu_ps(&a[idx_m*6 + 32]);
+
+            tmp0 = _mm512_maskz_permutex2var_ps(0x1f3f, m0, idx_c0c3 , m1);
+            tmp1 = _mm512_mask_permutex2var_ps(tmp0, 0xe0c0, idx_c0c3 , m2);
+            c256_0 = _mm512_extractf32x8_ps(tmp1, 0);
+            c256_3 = _mm512_extractf32x8_ps(tmp1, 1);
+
+            tmp0 = _mm512_maskz_permutex2var_ps(0x1f3f, m0, idx_c1c4 , m1);
+            tmp1 = _mm512_mask_permutex2var_ps(tmp0, 0xe0c0, idx_c1c4 , m2);
+            c256_1 = _mm512_extractf32x8_ps(tmp1, 0);
+            c256_4 = _mm512_extractf32x8_ps(tmp1, 1);
+
+            tmp0 = _mm512_maskz_permutex2var_ps(0x1f1f, m0, idx_c2c5 , m1);
+            tmp1 = _mm512_mask_permutex2var_ps(tmp0, 0xe0e0, idx_c2c5 , m2);
+            c256_2 = _mm512_extractf32x8_ps(tmp1, 0);
+            c256_5 = _mm512_extractf32x8_ps(tmp1, 1);            
+
+            accum_256 = _mm256_fmadd_ps(c256_1, x1_256, _mm256_mul_ps(c256_0, x0_256));
+            accum_256 = _mm256_fmadd_ps(c256_2, x2_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_3, x3_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_4, x4_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_5, x5_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(accum_256, alpha256, _mm256_loadu_ps(&y[idx_m]));
+            _mm256_storeu_ps(&y[idx_m], accum_256);
+        }
+
+        if(tag_m_8x != m) {
+            __m256 m256_0, m256_1, m256_2;
+            __m128 c128_0, c128_1;
+            idx_c0 = _mm512_set_epi32(22, 16, 10, 4, 19, 13, 7, 1, 21, 15, 9, 3, 18, 12, 6, 0);
+            idx_c1 = _mm512_add_epi32(idx_c0, M512_EPI32_1);
+            m2 = _mm512_set_ps(x[4], x[4], x[4], x[4], x[1], x[1], x[1], x[1], x[3], x[3], x[3], x[3], x[0], x[0], x[0], x[0]);
+            m3 = _mm512_set_ps(x[5], x[5], x[5], x[5], x[2], x[2], x[2], x[2], 0, 0, 0, 0, 0, 0, 0, 0);
+            for(BLASLONG idx_m=tag_m_8x; idx_m < tag_m_4x; idx_m+=4) {
+                m0 = _mm512_loadu_ps(&a[idx_m*6]);
+                m1 = _mm512_maskz_loadu_ps(0x00ff, &a[idx_m*6+16]);
+
+                tmp0 = _mm512_permutex2var_ps(m0, idx_c0, m1);
+                tmp1 = _mm512_maskz_permutex2var_ps(0xff00, m0, idx_c1, m1);
+
+                tmp0 = _mm512_mul_ps(tmp0, m2);
+                tmp1 = _mm512_mul_ps(tmp1, m3);
+                
+                tmp0 = _mm512_add_ps(tmp0, tmp1);
+
+                m256_0 = _mm512_extractf32x8_ps(tmp0, 0);
+                m256_1 = _mm512_extractf32x8_ps(tmp0, 1);
+
+                m256_0 = _mm256_add_ps(m256_0, m256_1);
+                m256_0 = _mm256_mul_ps(m256_0, alpha256);
+                c128_0 = _mm256_extractf32x4_ps(m256_0, 0);
+                c128_1 = _mm256_extractf32x4_ps(m256_0, 1);
+
+                c128_0 = _mm_maskz_add_ps(0x0f, c128_0, c128_1);
+                c128_0 = _mm_maskz_add_ps(0x0f, c128_0, _mm_maskz_loadu_ps(0x0f, &y[idx_m]));
+                _mm_mask_storeu_ps(&y[idx_m], 0x0f, c128_0);
+            }
+
+            if(tag_m_4x != m) {
+                //m256_2 is x*alpha
+                m256_2 = _mm256_maskz_loadu_ps(0x3f, x);
+                m256_2 = _mm256_mul_ps(m256_2, alpha256);
+                float ret8[8];
+                for(BLASLONG idx_m=tag_m_4x; idx_m < tag_m_2x;idx_m+=2) {
+                    m256_0 = _mm256_maskz_loadu_ps(0x3f, &a[idx_m*6]);
+                    m256_1 = _mm256_maskz_loadu_ps(0x3f, &a[idx_m*6 + 6]);
+                    
+                    m256_0 = _mm256_mul_ps(m256_0, m256_2);
+                    m256_1 = _mm256_mul_ps(m256_1, m256_2);
+
+                    _mm256_storeu_ps(ret8, m256_0);
+                    for(int i=0; i<6;i++){
+                        y[idx_m]+=ret8[i];
+                    }
+
+                    _mm256_storeu_ps(ret8, m256_1);
+                    for(int i=0; i<6;i++){
+                        y[idx_m+1]+=ret8[i];
+                    }
+                }
+                
+                if(tag_m_2x !=m) {
+                    m256_0 = _mm256_maskz_loadu_ps(0x3f, &a[tag_m_2x*6]);
+                    m256_0 = _mm256_mul_ps(m256_0, m256_2);
+                    
+                    _mm256_storeu_ps(ret8, m256_0);
+                    for(int i=0; i<6;i++){
+                        y[tag_m_2x]+=ret8[i];
+                    }
+                }
+            }
+        }
+    }
+    
+    return 0;
+}
+
+static int sgemv_kernel_t_7(BLASLONG m, float alpha, float *a, float *x, float *y)
+{
+    BLASLONG tag_m_16x = m & (~15);
+    BLASLONG tag_m_8x = m & (~7);
+    BLASLONG tag_m_4x = m & (~3);
+    BLASLONG tag_m_2x = m & (~1);
+
+    __m512 m0, m1, m2, m3, m4, m5, m6, tmp0, tmp1, tmp2, c0, c1, c2, c3, c4, c5, c6, accum;
+    __m512i idx_c0 = _mm512_set_epi32(25, 18, 27, 20, 13, 6, 31, 24, 17, 10, 3, 28, 21, 14, 7, 0);
+
+    __m512i M512_EPI32_1 = _mm512_set1_epi32(1);
+    __m512i M512_EPI32_0 = _mm512_setzero_epi32();
+    __m512i M512_EPI32_16 = _mm512_set1_epi32(16);
+
+    __m512i idx_c1 = _mm512_add_epi32(idx_c0, M512_EPI32_1);
+    idx_c1 = _mm512_mask_blend_epi32(0x0200, idx_c1, M512_EPI32_0);
+    __m512i idx_c2 = _mm512_add_epi32(idx_c1, M512_EPI32_1);
+    __m512i idx_c3 = _mm512_add_epi32(idx_c2, M512_EPI32_1);
+    __m512i idx_c4 = _mm512_add_epi32(idx_c3, M512_EPI32_1);
+    idx_c4 = _mm512_mask_blend_epi32(0x0010, idx_c4, M512_EPI32_0);
+    __m512i idx_c5 = _mm512_add_epi32(idx_c4, M512_EPI32_1);
+    idx_c5 = _mm512_mask_blend_epi32(0x2000, idx_c5, M512_EPI32_16);
+    __m512i idx_c6 = _mm512_add_epi32(idx_c5, M512_EPI32_1);
+
+    __m512 x0_512 = _mm512_set1_ps(x[0]);
+    __m512 x1_512 = _mm512_set1_ps(x[1]);
+    __m512 x2_512 = _mm512_set1_ps(x[2]);
+    __m512 x3_512 = _mm512_set1_ps(x[3]);
+    __m512 x4_512 = _mm512_set1_ps(x[4]);
+    __m512 x5_512 = _mm512_set1_ps(x[5]);
+    __m512 x6_512 = _mm512_set1_ps(x[6]);
+    __m512 alpha_512 = _mm512_set1_ps(alpha);
+
+    for (BLASLONG idx_m=0; idx_m < tag_m_16x; idx_m+=16) {
+        m0 = _mm512_loadu_ps(&a[idx_m*7]);
+        m1 = _mm512_loadu_ps(&a[idx_m*7 + 16]);
+        m2 = _mm512_loadu_ps(&a[idx_m*7 + 32]);
+        m3 = _mm512_loadu_ps(&a[idx_m*7 + 48]);
+        m4 = _mm512_loadu_ps(&a[idx_m*7 + 64]);
+        m5 = _mm512_loadu_ps(&a[idx_m*7 + 80]);
+        m6 = _mm512_loadu_ps(&a[idx_m*7 + 96]);
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x001f, m0, idx_c0, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x03e0, m2, idx_c0, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0x3c00, m4, idx_c0, m5);
+        c0 = _mm512_mask_blend_ps(0x03e0, tmp0, tmp1);
+        c0 = _mm512_mask_blend_ps(0x3c00, c0, tmp2);        
+        c0 = _mm512_mask_permutex2var_ps(c0, 0xc000, idx_c0, m6);        
+        
+        tmp0 = _mm512_maskz_permutex2var_ps(0x001f, m0, idx_c1, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x01e0, m2, idx_c1, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0x3e00, m4, idx_c1, m5);
+        c1 = _mm512_mask_blend_ps(0x01e0, tmp0, tmp1);
+        c1 = _mm512_mask_blend_ps(0x3e00, c1, tmp2);        
+        c1 = _mm512_mask_permutex2var_ps(c1, 0xc000, idx_c1, m6); 
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x001f, m0, idx_c2, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x01e0, m2, idx_c2, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0x3e00, m4, idx_c2, m5);
+        c2 = _mm512_mask_blend_ps(0x01e0, tmp0, tmp1);
+        c2 = _mm512_mask_blend_ps(0x3e00, c2, tmp2);        
+        c2 = _mm512_mask_permutex2var_ps(c2, 0xc000, idx_c2, m6); 
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x001f, m0, idx_c3, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x01e0, m2, idx_c3, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0x3e00, m4, idx_c3, m5);
+        c3 = _mm512_mask_blend_ps(0x01e0, tmp0, tmp1);
+        c3 = _mm512_mask_blend_ps(0x3e00, c3, tmp2);        
+        c3 = _mm512_mask_permutex2var_ps(c3, 0xc000, idx_c3, m6); 
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x000f, m0, idx_c4, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x01f0, m2, idx_c4, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0x3e00, m4, idx_c4, m5);
+        c4 = _mm512_mask_blend_ps(0x01f0, tmp0, tmp1);
+        c4 = _mm512_mask_blend_ps(0x3e00, c4, tmp2);        
+        c4 = _mm512_mask_permutex2var_ps(c4, 0xc000, idx_c4, m6); 
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x000f, m0, idx_c5, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x01f0, m2, idx_c5, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0x1e00, m4, idx_c5, m5);
+        c5 = _mm512_mask_blend_ps(0x01f0, tmp0, tmp1);
+        c5 = _mm512_mask_blend_ps(0x1e00, c5, tmp2);        
+        c5 = _mm512_mask_permutex2var_ps(c5, 0xe000, idx_c5, m6); 
+
+        tmp0 = _mm512_maskz_permutex2var_ps(0x000f, m0, idx_c6, m1);
+        tmp1 = _mm512_maskz_permutex2var_ps(0x01f0, m2, idx_c6, m3);
+        tmp2 = _mm512_maskz_permutex2var_ps(0x1e00, m4, idx_c6, m5);
+        c6 = _mm512_mask_blend_ps(0x01f0, tmp0, tmp1);
+        c6 = _mm512_mask_blend_ps(0x1e00, c6, tmp2);        
+        c6 = _mm512_mask_permutex2var_ps(c6, 0xe000, idx_c6, m6); 
+
+        accum = _mm512_fmadd_ps(c1, x1_512, _mm512_mul_ps(c0, x0_512));
+        accum = _mm512_fmadd_ps(c2, x2_512, accum);
+        accum = _mm512_fmadd_ps(c3, x3_512, accum);
+        accum = _mm512_fmadd_ps(c4, x4_512, accum);
+        accum = _mm512_fmadd_ps(c5, x5_512, accum);
+        accum = _mm512_fmadd_ps(c6, x6_512, accum);
+        accum = _mm512_fmadd_ps(accum, alpha_512, _mm512_loadu_ps(&y[idx_m]));
+        _mm512_storeu_ps(&y[idx_m], accum);
+
+    }
+
+    if(tag_m_16x != m){
+        //this is idx of c0c3
+        idx_c0 = _mm512_set_epi32(20, 13, 6, 31, 24, 17, 10, 3, 17, 10, 3, 28, 21, 14, 7, 0);
+        //this is idx of c1c4
+        idx_c1 = _mm512_add_epi32(idx_c0, M512_EPI32_1);
+        idx_c1 = _mm512_mask_blend_epi32(0x1000, idx_c1, M512_EPI32_0);
+        //this is idx of c2c5
+        idx_c2 = _mm512_add_epi32(idx_c1, M512_EPI32_1);
+        idx_c6 = _mm512_add_epi32(idx_c2, M512_EPI32_1);
+        __m256 c256_0, c256_1, c256_2, c256_3, c256_4, c256_5, c256_6;
+        __m256 x0_256 = _mm256_set1_ps(x[0]);
+        __m256 x1_256 = _mm256_set1_ps(x[1]);
+        __m256 x2_256 = _mm256_set1_ps(x[2]);
+        __m256 x3_256 = _mm256_set1_ps(x[3]);
+        __m256 x4_256 = _mm256_set1_ps(x[4]);
+        __m256 x5_256 = _mm256_set1_ps(x[5]);
+        __m256 x6_256 = _mm256_set1_ps(x[6]);
+        __m256 alpha256 = _mm256_set1_ps(alpha);
+        __m256 accum_256;
+        for (BLASLONG idx_m=tag_m_16x; idx_m < tag_m_8x; idx_m+=8) {
+            m0 = _mm512_loadu_ps(&a[idx_m*7]);
+            m1 = _mm512_loadu_ps(&a[idx_m*7 + 16]);
+            m2 = _mm512_loadu_ps(&a[idx_m*7 + 32]);
+            m3 = _mm512_maskz_loadu_ps(0x00ff, &a[idx_m*7 + 48]);
+
+            tmp0 = _mm512_maskz_permutex2var_ps(0x1f1f, m0, idx_c0, m1);
+            tmp1 = _mm512_maskz_permutex2var_ps(0xe0e0, m2, idx_c0, m3);
+            //this is c0c3
+            c0 = _mm512_mask_blend_ps(0xe0e0, tmp0, tmp1);
+            c256_0 = _mm512_extractf32x8_ps(c0, 0);
+            c256_3 = _mm512_extractf32x8_ps(c0, 1);
+
+            tmp0 = _mm512_maskz_permutex2var_ps(0x0f1f, m0, idx_c1, m1);
+            tmp1 = _mm512_maskz_permutex2var_ps(0xf0e0, m2, idx_c1, m3);
+            //this is c1c4
+            c1 = _mm512_mask_blend_ps(0xf0e0, tmp0, tmp1);
+            c256_1 = _mm512_extractf32x8_ps(c1, 0);
+            c256_4 = _mm512_extractf32x8_ps(c1, 1);
+
+            tmp0 = _mm512_maskz_permutex2var_ps(0x0f1f, m0, idx_c2, m1);
+            tmp1 = _mm512_maskz_permutex2var_ps(0xf0e0, m2, idx_c2, m3);
+            //this is c2c5
+            c2 = _mm512_mask_blend_ps(0xf0e0, tmp0, tmp1);
+            c256_2 = _mm512_extractf32x8_ps(c2, 0);
+            c256_5 = _mm512_extractf32x8_ps(c2, 1);
+
+            tmp0 = _mm512_maskz_permutex2var_ps(0x0f00, m0, idx_c6, m1);
+            tmp1 = _mm512_maskz_permutex2var_ps(0xf000, m2, idx_c6, m3);
+            //this is c2c5
+            c6 = _mm512_mask_blend_ps(0xf000, tmp0, tmp1);  
+            c256_6 = _mm512_extractf32x8_ps(c6, 1);
+            accum_256 = _mm256_fmadd_ps(c256_1, x1_256, _mm256_mul_ps(c256_0, x0_256));
+            accum_256 = _mm256_fmadd_ps(c256_2, x2_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_3, x3_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_4, x4_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_5, x5_256, accum_256);
+            accum_256 = _mm256_fmadd_ps(c256_6, x6_256, accum_256);
+
+            accum_256 = _mm256_fmadd_ps(accum_256, alpha256, _mm256_loadu_ps(&y[idx_m]));
+            _mm256_storeu_ps(&y[idx_m], accum_256);
+        }
+
+        if(tag_m_8x!=m) {
+            idx_c0 = _mm512_set_epi32(27, 20, 13, 6, 25, 18, 11, 4, 23, 16, 9, 2 ,21, 14, 7, 0);
+            idx_c1 = _mm512_add_epi32(idx_c0, M512_EPI32_1);
+
+            for (BLASLONG idx_m=tag_m_8x; idx_m < tag_m_4x; idx_m+=4) {
+                m0 = _mm512_loadu_ps(&a[idx_m*7]);
+                m1 = _mm512_maskz_loadu_ps(0x0fff, &a[idx_m*7 + 16]);
+                //this is x
+                m2 = _mm512_set_ps(x[6], x[6], x[6], x[6], x[4], x[4], x[4], x[4], x[2], x[2], x[2], x[2], x[0], x[0], x[0], x[0]);
+                //this is x
+                m4 = _mm512_set_ps(0, 0, 0, 0, x[5], x[5], x[5], x[5], x[3], x[3], x[3], x[3], x[1], x[1], x[1], x[1]);
+
+                tmp0 = _mm512_permutex2var_ps(m0, idx_c0, m1);
+                tmp1 = _mm512_maskz_permutex2var_ps(0x0fff, m0, idx_c1, m1);
+
+                tmp0 = _mm512_mul_ps(tmp0, m2);
+                tmp1 = _mm512_mul_ps(tmp1, m4);
+
+                tmp0 = _mm512_add_ps(tmp0, tmp1);
+
+                c256_0 = _mm512_extractf32x8_ps(tmp0, 0);
+                c256_1 = _mm512_extractf32x8_ps(tmp0, 1);
+
+                c256_0 = _mm256_add_ps(c256_0, c256_1);
+
+                __m128 c128_0 = _mm256_extractf32x4_ps(c256_0, 0);
+                __m128 c128_1 = _mm256_extractf32x4_ps(c256_0, 1);
+
+                c128_0 = _mm_maskz_add_ps(0x0f, c128_0, c128_1);
+                c128_0 = _mm_maskz_add_ps(0x0f, c128_0, _mm_maskz_loadu_ps(0x0f, &y[idx_m]));
+                _mm_mask_storeu_ps(&y[idx_m], 0x0f, c128_0);
+
+            }
+
+            if(tag_m_4x != m) {
+                //c256_2 is x*alpha
+                c256_2 = _mm256_maskz_loadu_ps(0x7f, x);
+                c256_2 = _mm256_mul_ps(c256_2, alpha256);
+                float ret8[8];
+                for(BLASLONG idx_m=tag_m_4x; idx_m < tag_m_2x;idx_m+=2) {
+                    c256_0 = _mm256_maskz_loadu_ps(0x7f, &a[idx_m*7]);
+                    c256_1 = _mm256_maskz_loadu_ps(0x7f, &a[idx_m*7 + 7]);
+                    
+                    c256_0 = _mm256_mul_ps(c256_0, c256_2);
+                    c256_1 = _mm256_mul_ps(c256_1, c256_2);
+
+                    _mm256_storeu_ps(ret8, c256_0);
+                    for(int i=0; i<7;i++){
+                        y[idx_m]+=ret8[i];
+                    }
+
+                    _mm256_storeu_ps(ret8, c256_1);
+                    for(int i=0; i<7;i++){
+                        y[idx_m+1]+=ret8[i];
+                    }
+                }
+                
+                if(tag_m_2x !=m) {
+                    c256_0 = _mm256_maskz_loadu_ps(0x7f, &a[tag_m_2x*7]);
+                    c256_0 = _mm256_mul_ps(c256_0, c256_2);
+                    
+                    _mm256_storeu_ps(ret8, c256_0);
+                    for(int i=0; i<7;i++){
+                        y[tag_m_2x]+=ret8[i];
+                    }
+                }
+            }
+        }
+    }
+
+    return 0;
+}
+
+static int sgemv_kernel_t_8(BLASLONG m, float alpha, float *a, float *x, float *y)
+{
+    BLASLONG tag_m_8x = m & (~7);
+    BLASLONG tag_m_4x = m & (~3);
+    BLASLONG tag_m_2x = m & (~1);
+
+    __m512 m0, m1, m2, m3;
+    __m256 r0, r1, r2, r3, r4, r5, r6, r7, tmp0, tmp1, tmp2, tmp3;
+    __m128 c128_0, c128_1, c128_2, c128_3;
+    __m128 alpha128 = _mm_set1_ps(alpha);
+
+    __m256 x256 = _mm256_loadu_ps(x);
+    __m512 x512 = _mm512_broadcast_f32x8(x256);
+
+    for(BLASLONG idx_m=0; idx_m<tag_m_8x; idx_m+=8) {
+        m0 = _mm512_loadu_ps(&a[idx_m*8]);
+        m1 = _mm512_loadu_ps(&a[idx_m*8 + 16]);
+        m2 = _mm512_loadu_ps(&a[idx_m*8 + 32]);
+        m3 = _mm512_loadu_ps(&a[idx_m*8 + 48]);
+        m0 = _mm512_mul_ps(m0, x512);
+        m1 = _mm512_mul_ps(m1, x512);
+        m2 = _mm512_mul_ps(m2, x512);
+        m3 = _mm512_mul_ps(m3, x512);
+
+        r0 = _mm512_extractf32x8_ps(m0, 0);
+        r1 = _mm512_extractf32x8_ps(m0, 1);
+        r2 = _mm512_extractf32x8_ps(m1, 0);
+        r3 = _mm512_extractf32x8_ps(m1, 1);
+        r4 = _mm512_extractf32x8_ps(m2, 0);
+        r5 = _mm512_extractf32x8_ps(m2, 1);
+        r6 = _mm512_extractf32x8_ps(m3, 0);
+        r7 = _mm512_extractf32x8_ps(m3, 1);
+
+        tmp0 = _mm256_hadd_ps(r0, r1);
+        tmp1 = _mm256_hadd_ps(r2, r3);
+        tmp2 = _mm256_hadd_ps(r4, r5);
+        tmp3 = _mm256_hadd_ps(r6, r7);
+        tmp1 = _mm256_hadd_ps(tmp0, tmp1);
+        tmp3 = _mm256_hadd_ps(tmp2, tmp3);
+        c128_0 = _mm256_extractf32x4_ps(tmp1, 0);
+        c128_1 = _mm256_extractf32x4_ps(tmp1, 1);
+        c128_2 = _mm256_extractf32x4_ps(tmp3, 0);
+        c128_3 = _mm256_extractf32x4_ps(tmp3, 1);
+
+        c128_0 = _mm_add_ps(c128_0, c128_1);
+        c128_2 = _mm_add_ps(c128_2, c128_3);
+        _mm_storeu_ps(&y[idx_m], _mm_fmadd_ps(c128_0, alpha128, _mm_loadu_ps(&y[idx_m])));        
+        _mm_storeu_ps(&y[idx_m+4], _mm_fmadd_ps(c128_2, alpha128, _mm_loadu_ps(&y[idx_m+4])));        
+    }
+
+    if (tag_m_8x !=m ){
+        for(BLASLONG idx_m=tag_m_8x; idx_m<tag_m_4x; idx_m+=4) {
+            m0 = _mm512_loadu_ps(&a[idx_m*8]);
+            m1 = _mm512_loadu_ps(&a[idx_m*8 + 16]);
+
+            m0 = _mm512_mul_ps(m0, x512);
+            m1 = _mm512_mul_ps(m1, x512);
+
+            r0 = _mm512_extractf32x8_ps(m0, 0);
+            r1 = _mm512_extractf32x8_ps(m0, 1);
+            r2 = _mm512_extractf32x8_ps(m1, 0);
+            r3 = _mm512_extractf32x8_ps(m1, 1);
+
+            tmp0 = _mm256_hadd_ps(r0, r1);
+            tmp1 = _mm256_hadd_ps(r2, r3);
+
+            tmp1 = _mm256_hadd_ps(tmp0, tmp1);
+            c128_0 = _mm256_extractf32x4_ps(tmp1, 0);
+            c128_1 = _mm256_extractf32x4_ps(tmp1, 1);
+
+            c128_0 = _mm_add_ps(c128_0, c128_1);
+            _mm_storeu_ps(&y[idx_m], _mm_fmadd_ps(c128_0, alpha128, _mm_loadu_ps(&y[idx_m])));
+
+        }
+
+        if(tag_m_4x != m) {
+            float ret[4];
+            for(BLASLONG idx_m=tag_m_4x; idx_m<tag_m_2x; idx_m+=2) {
+                m0 = _mm512_loadu_ps(&a[idx_m*8]);
+                m0 = _mm512_mul_ps(m0, x512);
+                r0 = _mm512_extractf32x8_ps(m0, 0);
+                r1 = _mm512_extractf32x8_ps(m0, 1);
+                tmp0 = _mm256_hadd_ps(r0, r1);
+                c128_0 = _mm256_extractf32x4_ps(tmp0, 0);
+                c128_1 = _mm256_extractf32x4_ps(tmp0, 1);                
+
+                c128_0 = _mm_add_ps(c128_0, c128_1);
+                c128_0 = _mm_mul_ps(c128_0, alpha128);
+
+                _mm_storeu_ps(ret, c128_0);
+                y[idx_m] += (ret[0]+ret[1]);
+                y[idx_m+1] += (ret[2]+ret[3]);
+
+            }
+
+            if(tag_m_2x!=m) {
+                r0 = _mm256_loadu_ps(&a[tag_m_2x*8]);
+                r0 = _mm256_mul_ps(r0, x256);
+
+                c128_0 = _mm256_extractf32x4_ps(r0, 0);
+                c128_1 = _mm256_extractf32x4_ps(r0, 1);   
+
+                c128_0 = _mm_add_ps(c128_0, c128_1);
+                _mm_storeu_ps(ret, c128_0);
+                y[tag_m_2x] += (ret[0] + ret[1] + ret[2] + ret[3]);
+
+            }
+        }
+    }
+    return 0;
+}
diff --git a/lapack-netlib/SRC/slasq2.f b/lapack-netlib/SRC/slasq2.f
index 219797c4a..c0c71b82e 100644
--- a/lapack-netlib/SRC/slasq2.f
+++ b/lapack-netlib/SRC/slasq2.f
@@ -185,7 +185,7 @@
 *
          IF( Z( 1 ).LT.ZERO ) THEN
             INFO = -201
-            CALL XERBLA( 'DLASQ2', 2 )
+            CALL XERBLA( 'SLASQ2', 2 )
             RETURN
          ELSE IF( Z( 2 ).LT.ZERO ) THEN
             INFO = -202
diff --git a/lapack-netlib/TESTING/EIG/cbdt05.f b/lapack-netlib/TESTING/EIG/cbdt05.f
index 5a08ccce3..4ed157431 100644
--- a/lapack-netlib/TESTING/EIG/cbdt05.f
+++ b/lapack-netlib/TESTING/EIG/cbdt05.f
@@ -158,9 +158,8 @@
 *     .. External Functions ..
       LOGICAL            LSAME
       INTEGER            ISAMAX
-      REAL               SASUM, SLAMCH, CLANGE
-      EXTERNAL           LSAME, ISAMAX, SASUM, SLAMCH, CLANGE
-      REAL               SCASUM
+      REAL               SASUM, SCASUM, SLAMCH, CLANGE
+      EXTERNAL           LSAME, ISAMAX, SASUM, SCASUM, SLAMCH, CLANGE
 *     ..
 *     .. External Subroutines ..
       EXTERNAL           CGEMM
diff --git a/lapack-netlib/TESTING/EIG/cckcsd.f b/lapack-netlib/TESTING/EIG/cckcsd.f
index 9783f0361..9524cb30b 100644
--- a/lapack-netlib/TESTING/EIG/cckcsd.f
+++ b/lapack-netlib/TESTING/EIG/cckcsd.f
@@ -228,7 +228,7 @@
 *     ..
 *     .. External Subroutines ..
       EXTERNAL           ALAHDG, ALAREQ, ALASUM, CCSDTS, CLACSG, CLAROR,
-     $                   CLASET
+     $                   CLASET, CSROT
 *     ..
 *     .. Intrinsic Functions ..
       INTRINSIC          ABS, MIN
diff --git a/lapack-netlib/TESTING/EIG/dckcsd.f b/lapack-netlib/TESTING/EIG/dckcsd.f
index 50db6baa0..063a5ef5c 100644
--- a/lapack-netlib/TESTING/EIG/dckcsd.f
+++ b/lapack-netlib/TESTING/EIG/dckcsd.f
@@ -226,7 +226,7 @@
 *     ..
 *     .. External Subroutines ..
       EXTERNAL           ALAHDG, ALAREQ, ALASUM, DCSDTS, DLACSG, DLAROR,
-     $                   DLASET
+     $                   DLASET, DROT
 *     ..
 *     .. Intrinsic Functions ..
       INTRINSIC          ABS, MIN
diff --git a/lapack-netlib/TESTING/EIG/sckcsd.f b/lapack-netlib/TESTING/EIG/sckcsd.f
index 5a6e4a099..be91eed51 100644
--- a/lapack-netlib/TESTING/EIG/sckcsd.f
+++ b/lapack-netlib/TESTING/EIG/sckcsd.f
@@ -226,7 +226,7 @@
 *     ..
 *     .. External Subroutines ..
       EXTERNAL           ALAHDG, ALAREQ, ALASUM, SCSDTS, SLACSG, SLAROR,
-     $                   SLASET
+     $                   SLASET, SROT
 *     ..
 *     .. Intrinsic Functions ..
       INTRINSIC          ABS, MIN
diff --git a/lapack-netlib/TESTING/EIG/zbdt05.f b/lapack-netlib/TESTING/EIG/zbdt05.f
index bbf0208b7..f262351e4 100644
--- a/lapack-netlib/TESTING/EIG/zbdt05.f
+++ b/lapack-netlib/TESTING/EIG/zbdt05.f
@@ -158,9 +158,8 @@
 *     .. External Functions ..
       LOGICAL            LSAME
       INTEGER            IDAMAX
-      DOUBLE PRECISION   DASUM, DLAMCH, ZLANGE
-      EXTERNAL           LSAME, IDAMAX, DASUM, DLAMCH, ZLANGE
-      DOUBLE PRECISION   DZASUM
+      DOUBLE PRECISION   DASUM, DZASUM, DLAMCH, ZLANGE
+      EXTERNAL           LSAME, IDAMAX, DASUM, DZASUM, DLAMCH, ZLANGE
 *     ..
 *     .. External Subroutines ..
       EXTERNAL           ZGEMM
diff --git a/lapack-netlib/TESTING/EIG/zckcsd.f b/lapack-netlib/TESTING/EIG/zckcsd.f
index f77b111a4..92760337c 100644
--- a/lapack-netlib/TESTING/EIG/zckcsd.f
+++ b/lapack-netlib/TESTING/EIG/zckcsd.f
@@ -228,7 +228,7 @@
 *     ..
 *     .. External Subroutines ..
       EXTERNAL           ALAHDG, ALAREQ, ALASUM, ZCSDTS, ZLACSG, ZLAROR,
-     $                   ZLASET
+     $                   ZLASET, ZDROT
 *     ..
 *     .. Intrinsic Functions ..
       INTRINSIC          ABS, MIN
diff --git a/lapack-netlib/TESTING/LIN/CMakeLists.txt b/lapack-netlib/TESTING/LIN/CMakeLists.txt
index 309ed7e77..fc55b8a96 100644
--- a/lapack-netlib/TESTING/LIN/CMakeLists.txt
+++ b/lapack-netlib/TESTING/LIN/CMakeLists.txt
@@ -6,7 +6,7 @@ set(SCLNTST slaord.f)
 
 set(DZLNTST dlaord.f)
 
-set(SLINTST schkaa.f
+set(SLINTST schkaa.F
    schkeq.f schkgb.f schkge.f schkgt.f
    schklq.f schkpb.f schkpo.f schkps.f schkpp.f
    schkpt.f schkq3.f schkql.f schkqr.f schkrq.f
@@ -51,7 +51,7 @@ else()
                       serrvx.f serrge.f serrsy.f serrpo.f)
 endif()
 
-set(CLINTST cchkaa.f
+set(CLINTST cchkaa.F
    cchkeq.f cchkgb.f cchkge.f cchkgt.f
    cchkhe.f cchkhe_rook.f cchkhe_rk.f 
    cchkhe_aa.f cchkhe_aa_2stage.f
@@ -107,7 +107,7 @@ else()
                       cerrvx.f cerrge.f cerrhe.f cerrsy.f cerrpo.f)
 endif()
 
-set(DLINTST dchkaa.f
+set(DLINTST dchkaa.F
    dchkeq.f dchkgb.f dchkge.f dchkgt.f
    dchklq.f dchkpb.f dchkpo.f dchkps.f dchkpp.f
    dchkpt.f dchkq3.f dchkql.f dchkqr.f dchkrq.f
@@ -153,7 +153,7 @@ else()
                       derrvx.f derrge.f derrsy.f derrpo.f)
 endif()
 
-set(ZLINTST zchkaa.f
+set(ZLINTST zchkaa.F
    zchkeq.f zchkgb.f zchkge.f zchkgt.f
    zchkhe.f zchkhe_rook.f zchkhe_rk.f 
    zchkhe_aa.f zchkhe_aa_2stage.f
diff --git a/lapack-netlib/TESTING/LIN/Makefile b/lapack-netlib/TESTING/LIN/Makefile
index 674265816..54b26455e 100644
--- a/lapack-netlib/TESTING/LIN/Makefile
+++ b/lapack-netlib/TESTING/LIN/Makefile
@@ -317,13 +317,13 @@ cleanobj:
 cleanexe:
 	rm -f xlintst*
 
-schkaa.o: schkaa.f
+schkaa.o: schkaa.F
 	$(FC) $(FFLAGS_DRV) -c -o $@ $<
-dchkaa.o: dchkaa.f
+dchkaa.o: dchkaa.F
 	$(FC) $(FFLAGS_DRV) -c -o $@ $<
-cchkaa.o: cchkaa.f
+cchkaa.o: cchkaa.F
 	$(FC) $(FFLAGS_DRV) -c -o $@ $<
-zchkaa.o: zchkaa.f
+zchkaa.o: zchkaa.F
 	$(FC) $(FFLAGS_DRV) -c -o $@ $<
 
 .NOTPARALLEL:
diff --git a/lapack-netlib/TESTING/LIN/cchkaa.f b/lapack-netlib/TESTING/LIN/cchkaa.F
similarity index 97%
rename from lapack-netlib/TESTING/LIN/cchkaa.f
rename to lapack-netlib/TESTING/LIN/cchkaa.F
index d36770be7..ec1534ed4 100644
--- a/lapack-netlib/TESTING/LIN/cchkaa.f
+++ b/lapack-netlib/TESTING/LIN/cchkaa.F
@@ -110,17 +110,14 @@
 *> \author Univ. of Colorado Denver
 *> \author NAG Ltd.
 *
-*> \date November 2019
-*
 *> \ingroup complex_lin
 *
 *  =====================================================================
       PROGRAM CCHKAA
 *
-*  -- LAPACK test routine (version 3.9.0) --
+*  -- LAPACK test routine --
 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-*     November 2017
 *
 *  =====================================================================
 *
@@ -156,9 +153,13 @@
      $                   NBVAL( MAXIN ), NBVAL2( MAXIN ),
      $                   NSVAL( MAXIN ), NVAL( MAXIN ), NXVAL( MAXIN ),
      $                   RANKVAL( MAXIN ), PIV( NMAX )
-      REAL               RWORK( 150*NMAX+2*MAXRHS ), S( 2*NMAX )
-      COMPLEX            A( ( KDMAX+1 )*NMAX, 7 ), B( NMAX*MAXRHS, 4 ),
-     $                   E( NMAX ), WORK( NMAX, NMAX+MAXRHS+10 )
+      REAL               S( 2*NMAX )
+      COMPLEX            E( NMAX )
+*     ..
+*     .. Allocatable Arrays ..
+      INTEGER AllocateStatus
+      REAL, DIMENSION(:), ALLOCATABLE :: RWORK
+      COMPLEX, DIMENSION(:,:), ALLOCATABLE :: A, B, WORK
 *     ..
 *     .. External Functions ..
       LOGICAL            LSAME, LSAMEN
@@ -194,6 +195,17 @@
 *     .. Data statements ..
       DATA               THREQ / 2.0 / , INTSTR / '0123456789' /
 *     ..
+*     .. Allocate memory dynamically ..
+*
+      ALLOCATE ( A( ( KDMAX+1 )*NMAX, 7 ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE ( B( NMAX*MAXRHS, 4 ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE ( WORK( NMAX, NMAX+MAXRHS+10 ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE ( RWORK( 150*NMAX+2*MAXRHS ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+*     ..
 *     .. Executable Statements ..
 *
       S1 = SECOND( )
@@ -1196,6 +1208,11 @@
       S2 = SECOND( )
       WRITE( NOUT, FMT = 9998 )
       WRITE( NOUT, FMT = 9997 )S2 - S1
+*
+      DEALLOCATE (A, STAT = AllocateStatus)
+      DEALLOCATE (B, STAT = AllocateStatus)
+      DEALLOCATE (WORK, STAT = AllocateStatus)
+      DEALLOCATE (RWORK,  STAT = AllocateStatus)
 *
  9999 FORMAT( / ' Execution not attempted due to input errors' )
  9998 FORMAT( / ' End of tests' )
diff --git a/lapack-netlib/TESTING/LIN/cchktsqr.f b/lapack-netlib/TESTING/LIN/cchktsqr.f
index 8288916db..62b6ce434 100644
--- a/lapack-netlib/TESTING/LIN/cchktsqr.f
+++ b/lapack-netlib/TESTING/LIN/cchktsqr.f
@@ -159,6 +159,8 @@
 *
 *     Test the error exits
 *
+      CALL XLAENV( 1, 0 )
+      CALL XLAENV( 2, 0 ) 
       IF( TSTERR ) CALL CERRTSQR( PATH, NOUT )
       INFOT = 0
 *
diff --git a/lapack-netlib/TESTING/LIN/dchkaa.f b/lapack-netlib/TESTING/LIN/dchkaa.F
similarity index 96%
rename from lapack-netlib/TESTING/LIN/dchkaa.f
rename to lapack-netlib/TESTING/LIN/dchkaa.F
index 03575c4d1..ef9d7808c 100644
--- a/lapack-netlib/TESTING/LIN/dchkaa.f
+++ b/lapack-netlib/TESTING/LIN/dchkaa.F
@@ -106,17 +106,14 @@
 *> \author Univ. of Colorado Denver
 *> \author NAG Ltd.
 *
-*> \date November 2019
-*
 *> \ingroup double_lin
 *
 *  =====================================================================
       PROGRAM DCHKAA
 *
-*  -- LAPACK test routine (version 3.9.0) --
+*  -- LAPACK test routine --
 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-*     Novemebr 2019
 *
 *  =====================================================================
 *
@@ -152,9 +149,12 @@
      $                   NBVAL( MAXIN ), NBVAL2( MAXIN ),
      $                   NSVAL( MAXIN ), NVAL( MAXIN ), NXVAL( MAXIN ),
      $                   RANKVAL( MAXIN ), PIV( NMAX )
-      DOUBLE PRECISION   A( ( KDMAX+1 )*NMAX, 7 ), B( NMAX*MAXRHS, 4 ),
-     $                   E( NMAX ), RWORK( 5*NMAX+2*MAXRHS ),
-     $                   S( 2*NMAX ), WORK( NMAX, 3*NMAX+MAXRHS+30 )
+      DOUBLE PRECISION   E( NMAX ), S( 2*NMAX )
+*     ..
+*     .. Allocatable Arrays ..
+      INTEGER AllocateStatus
+      DOUBLE PRECISION, DIMENSION(:), ALLOCATABLE :: RWORK
+      DOUBLE PRECISION, DIMENSION(:,:), ALLOCATABLE :: A, B, WORK      
 *     ..
 *     .. External Functions ..
       LOGICAL            LSAME, LSAMEN
@@ -188,6 +188,18 @@
 *     .. Data statements ..
       DATA               THREQ / 2.0D0 / , INTSTR / '0123456789' /
 *     ..
+*     ..
+*     .. Allocate memory dynamically ..
+*
+      ALLOCATE ( A( ( KDMAX+1 )*NMAX, 7 ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE ( B( NMAX*MAXRHS, 4 ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE ( WORK( NMAX, 3*NMAX+MAXRHS+30 ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE ( RWORK( 5*NMAX+2*MAXRHS ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+*
 *     .. Executable Statements ..
 *
       S1 = DSECND( )
@@ -677,7 +689,7 @@
 *
 *        SK:  symmetric indefinite matrices,
 *             with bounded Bunch-Kaufman (rook) pivoting algorithm,
-*             differnet matrix storage format than SR path version.
+*             different matrix storage format than SR path version.
 *
          NTYPES = 10
          CALL ALAREQ( PATH, NMATS, DOTYPE, NTYPES, NIN, NOUT )
@@ -1039,6 +1051,11 @@
       S2 = DSECND( )
       WRITE( NOUT, FMT = 9998 )
       WRITE( NOUT, FMT = 9997 )S2 - S1
+*
+      DEALLOCATE (A, STAT = AllocateStatus)
+      DEALLOCATE (B, STAT = AllocateStatus)
+      DEALLOCATE (WORK, STAT = AllocateStatus)
+      DEALLOCATE (RWORK,  STAT = AllocateStatus)
 *
  9999 FORMAT( / ' Execution not attempted due to input errors' )
  9998 FORMAT( / ' End of tests' )
diff --git a/lapack-netlib/TESTING/LIN/dchktsqr.f b/lapack-netlib/TESTING/LIN/dchktsqr.f
index c4b1f01bd..14119e6e5 100644
--- a/lapack-netlib/TESTING/LIN/dchktsqr.f
+++ b/lapack-netlib/TESTING/LIN/dchktsqr.f
@@ -159,6 +159,8 @@
 *
 *     Test the error exits
 *
+      CALL XLAENV( 1, 0 )
+      CALL XLAENV( 2, 0 )            
       IF( TSTERR ) CALL DERRTSQR( PATH, NOUT )
       INFOT = 0
 *
diff --git a/lapack-netlib/TESTING/LIN/schkaa.f b/lapack-netlib/TESTING/LIN/schkaa.F
similarity index 97%
rename from lapack-netlib/TESTING/LIN/schkaa.f
rename to lapack-netlib/TESTING/LIN/schkaa.F
index a9c13e442..a5b826d06 100644
--- a/lapack-netlib/TESTING/LIN/schkaa.f
+++ b/lapack-netlib/TESTING/LIN/schkaa.F
@@ -104,17 +104,14 @@
 *> \author Univ. of Colorado Denver
 *> \author NAG Ltd.
 *
-*> \date November 2019
-*
 *> \ingroup single_lin
 *
 *  =====================================================================
       PROGRAM SCHKAA
 *
-*  -- LAPACK test routine (version 3.9.0) --
+*  -- LAPACK test routine --
 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-*     November 2019
 *
 *  =====================================================================
 *
@@ -150,9 +147,12 @@
      $                   NBVAL( MAXIN ), NBVAL2( MAXIN ),
      $                   NSVAL( MAXIN ), NVAL( MAXIN ), NXVAL( MAXIN ),
      $                   RANKVAL( MAXIN ), PIV( NMAX )
-      REAL               A( ( KDMAX+1 )*NMAX, 7 ), B( NMAX*MAXRHS, 4 ),
-     $                   E( NMAX ), RWORK( 5*NMAX+2*MAXRHS ),
-     $                   S( 2*NMAX ), WORK( NMAX, NMAX+MAXRHS+30 )
+      REAL               E( NMAX ), S( 2*NMAX )
+*     ..
+*     .. Allocatable Arrays ..
+      INTEGER AllocateStatus
+      REAL, DIMENSION(:), ALLOCATABLE :: RWORK
+      REAL, DIMENSION(:,:), ALLOCATABLE :: A, B, WORK
 *     ..
 *     .. External Functions ..
       LOGICAL            LSAME, LSAMEN
@@ -186,6 +186,17 @@
 *     .. Data statements ..
       DATA               THREQ / 2.0E0 / , INTSTR / '0123456789' /
 *     ..
+*     .. Allocate memory dynamically ..
+*
+      ALLOCATE (A( ( KDMAX+1 )*NMAX, 7 ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE (B( NMAX*MAXRHS, 4 ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE (WORK( NMAX, NMAX+MAXRHS+30 ) , STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE (RWORK( 5*NMAX+2*MAXRHS ), STAT = AllocateStatus )
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***" 
+*     ..
 *     .. Executable Statements ..
 *
       S1 = SECOND( )
@@ -1034,6 +1045,11 @@
       S2 = SECOND( )
       WRITE( NOUT, FMT = 9998 )
       WRITE( NOUT, FMT = 9997 )S2 - S1
+*
+      DEALLOCATE (A, STAT = AllocateStatus)
+      DEALLOCATE (B, STAT = AllocateStatus)
+      DEALLOCATE (WORK, STAT = AllocateStatus)
+      DEALLOCATE (RWORK,  STAT = AllocateStatus)
 *
  9999 FORMAT( / ' Execution not attempted due to input errors' )
  9998 FORMAT( / ' End of tests' )
diff --git a/lapack-netlib/TESTING/LIN/schktsqr.f b/lapack-netlib/TESTING/LIN/schktsqr.f
index 2bed434a8..aa4d6f9c4 100644
--- a/lapack-netlib/TESTING/LIN/schktsqr.f
+++ b/lapack-netlib/TESTING/LIN/schktsqr.f
@@ -159,6 +159,8 @@
 *
 *     Test the error exits
 *
+      CALL XLAENV( 1, 0 )
+      CALL XLAENV( 2, 0 ) 
       IF( TSTERR ) CALL SERRTSQR( PATH, NOUT )
       INFOT = 0
 *
diff --git a/lapack-netlib/TESTING/LIN/zchkaa.f b/lapack-netlib/TESTING/LIN/zchkaa.F
similarity index 97%
rename from lapack-netlib/TESTING/LIN/zchkaa.f
rename to lapack-netlib/TESTING/LIN/zchkaa.F
index 30d2a084a..a118515a5 100644
--- a/lapack-netlib/TESTING/LIN/zchkaa.f
+++ b/lapack-netlib/TESTING/LIN/zchkaa.F
@@ -110,17 +110,14 @@
 *> \author Univ. of Colorado Denver
 *> \author NAG Ltd.
 *
-*> \date November 2019
-*
 *> \ingroup complex16_lin
 *
 *  =====================================================================
       PROGRAM ZCHKAA
 *
-*  -- LAPACK test routine (version 3.9.0) --
+*  -- LAPACK test routine --
 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
-*     November 2019
 *
 *  =====================================================================
 *
@@ -156,9 +153,13 @@
      $                   NBVAL( MAXIN ), NBVAL2( MAXIN ),
      $                   NSVAL( MAXIN ), NVAL( MAXIN ), NXVAL( MAXIN ),
      $                   RANKVAL( MAXIN ), PIV( NMAX )
-      DOUBLE PRECISION   RWORK( 150*NMAX+2*MAXRHS ), S( 2*NMAX )
-      COMPLEX*16         A( ( KDMAX+1 )*NMAX, 7 ), B( NMAX*MAXRHS, 4 ),
-     $                   E( NMAX ),  WORK( NMAX, NMAX+MAXRHS+10 )
+      DOUBLE PRECISION   S( 2*NMAX )
+      COMPLEX*16         E( NMAX )
+*
+*    .. Allocatable Arrays ..
+      INTEGER AllocateStatus
+      DOUBLE PRECISION, DIMENSION(:), ALLOCATABLE::   RWORK
+      COMPLEX*16, DIMENSION(:,:), ALLOCATABLE::  A, B, WORK
 *     ..
 *     .. External Functions ..
       LOGICAL            LSAME, LSAMEN
@@ -194,6 +195,16 @@
 *     ..
 *     .. Data statements ..
       DATA               THREQ / 2.0D0 / , INTSTR / '0123456789' /
+*
+*     .. Allocate memory dynamically ..
+      ALLOCATE (RWORK( 150*NMAX+2*MAXRHS ), STAT = AllocateStatus)
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE (A ((KDMAX+1) * NMAX, 7), STAT = AllocateStatus)
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE (B (NMAX * MAXRHS, 4), STAT = AllocateStatus)
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
+      ALLOCATE (WORK (NMAX, NMAX+MAXRHS+10), STAT = AllocateStatus)
+      IF (AllocateStatus /= 0) STOP "*** Not enough memory ***"
 *     ..
 *     .. Executable Statements ..
 *
@@ -1231,6 +1242,11 @@
       S2 = DSECND( )
       WRITE( NOUT, FMT = 9998 )
       WRITE( NOUT, FMT = 9997 )S2 - S1
+*
+      DEALLOCATE (A, STAT = AllocateStatus)
+      DEALLOCATE (B, STAT = AllocateStatus)
+      DEALLOCATE (RWORK, STAT = AllocateStatus)
+      DEALLOCATE (WORK,  STAT = AllocateStatus)
 *
  9999 FORMAT( / ' Execution not attempted due to input errors' )
  9998 FORMAT( / ' End of tests' )
diff --git a/lapack-netlib/TESTING/LIN/zchktsqr.f b/lapack-netlib/TESTING/LIN/zchktsqr.f
index e6e6ac556..678b1772f 100644
--- a/lapack-netlib/TESTING/LIN/zchktsqr.f
+++ b/lapack-netlib/TESTING/LIN/zchktsqr.f
@@ -159,6 +159,8 @@
 *
 *     Test the error exits
 *
+      CALL XLAENV( 1, 0 )
+      CALL XLAENV( 2, 0 ) 
       IF( TSTERR ) CALL ZERRTSQR( PATH, NOUT )
       INFOT = 0
 *
diff --git a/openblas_config_template.h b/openblas_config_template.h
index 858b8c5cb..1e17c9a16 100644
--- a/openblas_config_template.h
+++ b/openblas_config_template.h
@@ -99,5 +99,6 @@ typedef int blasint;
 
 /* Inclusion of Linux-specific header is needed for definition of cpu_set_t. */
 #ifdef OPENBLAS_OS_LINUX
+#define _GNU_SOURCE
 #include <sched.h>
 #endif
diff --git a/param.h b/param.h
index a35ce69bd..01048023f 100644
--- a/param.h
+++ b/param.h
@@ -72,13 +72,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifndef PARAM_H
 #define PARAM_H
 
-#define LONGCAST (BLASLONG)
-#if defined(__BYTE_ORDER__)
-#if __GNUC__ < 9
-#undef LONGCAST
-#define LONGCAST
-#endif
-#endif
 
 #define SBGEMM_DEFAULT_UNROLL_N 4
 #define SBGEMM_DEFAULT_UNROLL_M 8
@@ -2096,7 +2089,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef PPCG4
 #define GEMM_DEFAULT_OFFSET_A    0
 #define GEMM_DEFAULT_OFFSET_B 1024
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x0ffffUL
+#define GEMM_DEFAULT_ALIGN 0x0ffffUL
 
 #define SGEMM_DEFAULT_UNROLL_M 16
 #define SGEMM_DEFAULT_UNROLL_N 4
@@ -2127,7 +2120,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define GEMM_DEFAULT_OFFSET_A 2688
 #define GEMM_DEFAULT_OFFSET_B 3072
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x03fffUL
+#define GEMM_DEFAULT_ALIGN 0x03fffUL
 
 #if defined(__BYTE_ORDER__)&&(__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
 #define SGEMM_DEFAULT_UNROLL_M 4
@@ -2176,7 +2169,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define GEMM_DEFAULT_OFFSET_A (32 * 0)
 #define GEMM_DEFAULT_OFFSET_B (32 * 0)
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x0ffffUL
+#define GEMM_DEFAULT_ALIGN 0x0ffffUL
 
 #define SGEMM_DEFAULT_UNROLL_M 4
 #define SGEMM_DEFAULT_UNROLL_N 4
@@ -2212,7 +2205,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define GEMM_DEFAULT_OFFSET_A (32 * 0)
 #define GEMM_DEFAULT_OFFSET_B (32 * 0)
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x0ffffUL
+#define GEMM_DEFAULT_ALIGN 0x0ffffUL
 
 #define SGEMM_DEFAULT_UNROLL_M 8
 #define SGEMM_DEFAULT_UNROLL_N 4
@@ -2247,7 +2240,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(POWER3) || defined(POWER4) || defined(POWER5)
 #define GEMM_DEFAULT_OFFSET_A 0
 #define GEMM_DEFAULT_OFFSET_B 2048
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x0ffffUL
+#define GEMM_DEFAULT_ALIGN 0x0ffffUL
 
 #define SGEMM_DEFAULT_UNROLL_M 4
 #define SGEMM_DEFAULT_UNROLL_N 4
@@ -2271,6 +2264,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define DGEMM_DEFAULT_Q 216
 #define DGEMM_DEFAULT_R 1012
 
+#define CGEMM_DEFAULT_P 256
+#define CGEMM_DEFAULT_Q 104
+#define CGEMM_DEFAULT_R 1012
+   
 #define ZGEMM_DEFAULT_P 256
 #define ZGEMM_DEFAULT_Q 104
 #define ZGEMM_DEFAULT_R 1012
@@ -2288,6 +2285,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_DEFAULT_P 144
 #define ZGEMM_DEFAULT_P 144
 #endif
+
+#define SGEMM_DEFAULT_Q 256
+#define CGEMM_DEFAULT_Q 256
+#define DGEMM_DEFAULT_Q 256
+#define ZGEMM_DEFAULT_Q 256
 #endif
 
 #if defined(POWER5)
@@ -2320,7 +2322,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define GEMM_DEFAULT_OFFSET_A  384
 #define GEMM_DEFAULT_OFFSET_B 1024
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x03fffUL
+#define GEMM_DEFAULT_ALIGN 0x03fffUL
 
 #define SGEMM_DEFAULT_UNROLL_M 4
 #define SGEMM_DEFAULT_UNROLL_N 4
@@ -2353,7 +2355,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define GEMM_DEFAULT_OFFSET_A 0 
 #define GEMM_DEFAULT_OFFSET_B 65536
 
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x0ffffUL
+#define GEMM_DEFAULT_ALIGN 0x0ffffUL
 #if defined(__32BIT__)
 #warning using BINARY32==POWER6
 #define SGEMM_DEFAULT_UNROLL_M 4
@@ -2406,7 +2408,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define GEMM_DEFAULT_OFFSET_A 0 
 #define GEMM_DEFAULT_OFFSET_B 65536
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x0ffffUL
+#define GEMM_DEFAULT_ALIGN 0x0ffffUL
 
 #define SWITCH_RATIO            16
 #define GEMM_PREFERED_SIZE      16
@@ -2445,7 +2447,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define GEMM_DEFAULT_OFFSET_A 0
 #define GEMM_DEFAULT_OFFSET_B 65536
-#define GEMM_DEFAULT_ALIGN LONGCAST 0x0ffffUL
+#define GEMM_DEFAULT_ALIGN 0x0ffffUL
 
 #define SWITCH_RATIO            16
 #define GEMM_PREFERED_SIZE      16
@@ -2957,7 +2959,7 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define CGEMM_DEFAULT_R 4096
 #define ZGEMM_DEFAULT_R 2048
 
-#elif defined(CORTEXA53)
+#elif defined(CORTEXA53) || defined(CORTEXA55)
 
 #define SGEMM_DEFAULT_UNROLL_M  8
 #define SGEMM_DEFAULT_UNROLL_N  8
diff --git a/relapack/src/lapack_wrappers.c.orig b/relapack/src/lapack_wrappers.c.orig
deleted file mode 100644
index d89d2fe2f..000000000
--- a/relapack/src/lapack_wrappers.c.orig
+++ /dev/null
@@ -1,607 +0,0 @@
-#include "relapack.h"
-
-////////////
-// XLAUUM //
-////////////
-
-#if INCLUDE_SLAUUM
-void LAPACK(slauum)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_slauum(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_DLAUUM
-void LAPACK(dlauum)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_dlauum(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_CLAUUM
-void LAPACK(clauum)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_clauum(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_ZLAUUM
-void LAPACK(zlauum)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_zlauum(uplo, n, A, ldA, info);
-}
-#endif
-
-
-////////////
-// XSYGST //
-////////////
-
-#if INCLUDE_SSYGST
-void LAPACK(ssygst)(
-    const int *itype, const char *uplo, const int *n,
-    float *A, const int *ldA, const float *B, const int *ldB,
-    int *info
-) {
-    RELAPACK_ssygst(itype, uplo, n, A, ldA, B, ldB, info);
-}
-#endif
-
-#if INCLUDE_DSYGST
-void LAPACK(dsygst)(
-    const int *itype, const char *uplo, const int *n,
-    double *A, const int *ldA, const double *B, const int *ldB,
-    int *info
-) {
-    RELAPACK_dsygst(itype, uplo, n, A, ldA, B, ldB, info);
-}
-#endif
-
-#if INCLUDE_CSYGST
-void LAPACK(csygst)(
-    const int *itype, const char *uplo, const int *n,
-    float *A, const int *ldA, const float *B, const int *ldB,
-    int *info
-) {
-    RELAPACK_csygst(itype, uplo, n, A, ldA, B, ldB, info);
-}
-#endif
-
-#if INCLUDE_ZSYGST
-void LAPACK(zsygst)(
-    const int *itype, const char *uplo, const int *n,
-    double *A, const int *ldA, const double *B, const int *ldB,
-    int *info
-) {
-    RELAPACK_zsygst(itype, uplo, n, A, ldA, B, ldB, info);
-}
-#endif
-
-
-////////////
-// XTRTRI //
-////////////
-
-#if INCLUDE_STRTRI
-void LAPACK(strtri)(
-    const char *uplo, const char *diag, const int *n,
-    float *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_strtri(uplo, diag, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_DTRTRI
-void LAPACK(dtrtri)(
-    const char *uplo, const char *diag, const int *n,
-    double *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_dtrtri(uplo, diag, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_CTRTRI
-void LAPACK(ctrtri)(
-    const char *uplo, const char *diag, const int *n,
-    float *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_ctrtri(uplo, diag, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_ZTRTRI
-void LAPACK(ztrtri)(
-    const char *uplo, const char *diag, const int *n,
-    double *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_ztrtri(uplo, diag, n, A, ldA, info);
-}
-#endif
-
-
-////////////
-// XPOTRF //
-////////////
-
-#if INCLUDE_SPOTRF
-void LAPACK(spotrf)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_spotrf(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_DPOTRF
-void LAPACK(dpotrf)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_dpotrf(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_CPOTRF
-void LAPACK(cpotrf)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_cpotrf(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_ZPOTRF
-void LAPACK(zpotrf)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA,
-    int *info
-) {
-    RELAPACK_zpotrf(uplo, n, A, ldA, info);
-}
-#endif
-
-
-////////////
-// XPBTRF //
-////////////
-
-#if INCLUDE_SPBTRF
-void LAPACK(spbtrf)(
-    const char *uplo, const int *n, const int *kd,
-    float *Ab, const int *ldAb,
-    int *info
-) {
-    RELAPACK_spbtrf(uplo, n, kd, Ab, ldAb, info);
-}
-#endif
-
-#if INCLUDE_DPBTRF
-void LAPACK(dpbtrf)(
-    const char *uplo, const int *n, const int *kd,
-    double *Ab, const int *ldAb,
-    int *info
-) {
-    RELAPACK_dpbtrf(uplo, n, kd, Ab, ldAb, info);
-}
-#endif
-
-#if INCLUDE_CPBTRF
-void LAPACK(cpbtrf)(
-    const char *uplo, const int *n, const int *kd,
-    float *Ab, const int *ldAb,
-    int *info
-) {
-    RELAPACK_cpbtrf(uplo, n, kd, Ab, ldAb, info);
-}
-#endif
-
-#if INCLUDE_ZPBTRF
-void LAPACK(zpbtrf)(
-    const char *uplo, const int *n, const int *kd,
-    double *Ab, const int *ldAb,
-    int *info
-) {
-    RELAPACK_zpbtrf(uplo, n, kd, Ab, ldAb, info);
-}
-#endif
-
-
-////////////
-// XSYTRF //
-////////////
-
-#if INCLUDE_SSYTRF
-void LAPACK(ssytrf)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA, int *ipiv,
-    float *Work, const int *lWork, int *info
-) {
-    RELAPACK_ssytrf(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_DSYTRF
-void LAPACK(dsytrf)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA, int *ipiv,
-    double *Work, const int *lWork, int *info
-) {
-    RELAPACK_dsytrf(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_CSYTRF
-void LAPACK(csytrf)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA, int *ipiv,
-    float *Work, const int *lWork, int *info
-) {
-    RELAPACK_csytrf(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_ZSYTRF
-void LAPACK(zsytrf)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA, int *ipiv,
-    double *Work, const int *lWork, int *info
-) {
-    RELAPACK_zsytrf(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_CHETRF
-void LAPACK(chetrf)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA, int *ipiv,
-    float *Work, const int *lWork, int *info
-) {
-    RELAPACK_chetrf(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_ZHETRF
-void LAPACK(zhetrf)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA, int *ipiv,
-    double *Work, const int *lWork, int *info
-) {
-    RELAPACK_zhetrf(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_SSYTRF_ROOK
-void LAPACK(ssytrf_rook)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA, int *ipiv,
-    float *Work, const int *lWork, int *info
-) {
-    RELAPACK_ssytrf_rook(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_DSYTRF_ROOK
-void LAPACK(dsytrf_rook)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA, int *ipiv,
-    double *Work, const int *lWork, int *info
-) {
-    RELAPACK_dsytrf_rook(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_CSYTRF_ROOK
-void LAPACK(csytrf_rook)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA, int *ipiv,
-    float *Work, const int *lWork, int *info
-) {
-    RELAPACK_csytrf_rook(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_ZSYTRF_ROOK
-void LAPACK(zsytrf_rook)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA, int *ipiv,
-    double *Work, const int *lWork, int *info
-) {
-    RELAPACK_zsytrf_rook(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_CHETRF_ROOK
-void LAPACK(chetrf_rook)(
-    const char *uplo, const int *n,
-    float *A, const int *ldA, int *ipiv,
-    float *Work, const int *lWork, int *info
-) {
-    RELAPACK_chetrf_rook(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-#if INCLUDE_ZHETRF_ROOK
-void LAPACK(zhetrf_rook)(
-    const char *uplo, const int *n,
-    double *A, const int *ldA, int *ipiv,
-    double *Work, const int *lWork, int *info
-) {
-    RELAPACK_zhetrf_rook(uplo, n, A, ldA, ipiv, Work, lWork, info);
-}
-#endif
-
-
-////////////
-// XGETRF //
-////////////
-
-#if INCLUDE_SGETRF
-void LAPACK(sgetrf)(
-    const int *m, const int *n,
-    float *A, const int *ldA, int *ipiv,
-    int *info
-) {
-    RELAPACK_sgetrf(m, n, A, ldA, ipiv, info);
-}
-#endif
-
-#if INCLUDE_DGETRF
-void LAPACK(dgetrf)(
-    const int *m, const int *n,
-    double *A, const int *ldA, int *ipiv,
-    int *info
-) {
-    RELAPACK_dgetrf(m, n, A, ldA, ipiv, info);
-}
-#endif
-
-#if INCLUDE_CGETRF
-void LAPACK(cgetrf)(
-    const int *m, const int *n,
-    float *A, const int *ldA, int *ipiv,
-    int *info
-) {
-    RELAPACK_cgetrf(m, n, A, ldA, ipiv, info);
-}
-#endif
-
-#if INCLUDE_ZGETRF
-void LAPACK(zgetrf)(
-    const int *m, const int *n,
-    double *A, const int *ldA, int *ipiv,
-    int *info
-) {
-    RELAPACK_zgetrf(m, n, A, ldA, ipiv, info);
-}
-#endif
-
-
-////////////
-// XGBTRF //
-////////////
-
-#if INCLUDE_SGBTRF
-void LAPACK(sgbtrf)(
-    const int *m, const int *n, const int *kl, const int *ku,
-    float *Ab, const int *ldAb, int *ipiv,
-    int *info
-) {
-    RELAPACK_sgbtrf(m, n, kl, ku, Ab, ldAb, ipiv, info);
-}
-#endif
-
-#if INCLUDE_DGBTRF
-void LAPACK(dgbtrf)(
-    const int *m, const int *n, const int *kl, const int *ku,
-    double *Ab, const int *ldAb, int *ipiv,
-    int *info
-) {
-    RELAPACK_dgbtrf(m, n, kl, ku, Ab, ldAb, ipiv, info);
-}
-#endif
-
-#if INCLUDE_CGBTRF
-void LAPACK(cgbtrf)(
-    const int *m, const int *n, const int *kl, const int *ku,
-    float *Ab, const int *ldAb, int *ipiv,
-    int *info
-) {
-    RELAPACK_cgbtrf(m, n, kl, ku, Ab, ldAb, ipiv, info);
-}
-#endif
-
-#if INCLUDE_ZGBTRF
-void LAPACK(zgbtrf)(
-    const int *m, const int *n, const int *kl, const int *ku,
-    double *Ab, const int *ldAb, int *ipiv,
-    int *info
-) {
-    RELAPACK_zgbtrf(m, n, kl, ku, Ab, ldAb, ipiv, info);
-}
-#endif
-
-
-////////////
-// XTRSYL //
-////////////
-
-#if INCLUDE_STRSYL
-void LAPACK(strsyl)(
-    const char *tranA, const char *tranB, const int *isgn,
-    const int *m, const int *n,
-    const float *A, const int *ldA, const float *B, const int *ldB,
-    float *C, const int *ldC, float *scale,
-    int *info
-) {
-    RELAPACK_strsyl(tranA, tranB, isgn, m, n, A, ldA, B, ldB, C, ldC, scale, info);
-}
-#endif
-
-#if INCLUDE_DTRSYL
-void LAPACK(dtrsyl)(
-    const char *tranA, const char *tranB, const int *isgn,
-    const int *m, const int *n,
-    const double *A, const int *ldA, const double *B, const int *ldB,
-    double *C, const int *ldC, double *scale,
-    int *info
-) {
-    RELAPACK_dtrsyl(tranA, tranB, isgn, m, n, A, ldA, B, ldB, C, ldC, scale, info);
-}
-#endif
-
-#if INCLUDE_CTRSYL
-void LAPACK(ctrsyl)(
-    const char *tranA, const char *tranB, const int *isgn,
-    const int *m, const int *n,
-    const float *A, const int *ldA, const float *B, const int *ldB,
-    float *C, const int *ldC, float *scale,
-    int *info
-) {
-    RELAPACK_ctrsyl(tranA, tranB, isgn, m, n, A, ldA, B, ldB, C, ldC, scale, info);
-}
-#endif
-
-#if INCLUDE_ZTRSYL
-void LAPACK(ztrsyl)(
-    const char *tranA, const char *tranB, const int *isgn,
-    const int *m, const int *n,
-    const double *A, const int *ldA, const double *B, const int *ldB,
-    double *C, const int *ldC, double *scale,
-    int *info
-) {
-    RELAPACK_ztrsyl(tranA, tranB, isgn, m, n, A, ldA, B, ldB, C, ldC, scale, info);
-}
-#endif
-
-
-////////////
-// XTGSYL //
-////////////
-
-#if INCLUDE_STGSYL
-void LAPACK(stgsyl)(
-    const char *trans, const int *ijob, const int *m, const int *n,
-    const float *A, const int *ldA, const float *B, const int *ldB,
-    float *C, const int *ldC,
-    const float *D, const int *ldD, const float *E, const int *ldE,
-    float *F, const int *ldF,
-    float *scale, float *dif,
-    float *Work, const int *lWork, int *iWork, int *info
-) {
-    RELAPACK_stgsyl(trans, ijob, m, n, A, ldA, B, ldB, C, ldC, D, ldD, E, ldE, F, ldF, scale, dif, Work, lWork, iWork, info);
-}
-#endif
-
-#if INCLUDE_DTGSYL
-void LAPACK(dtgsyl)(
-    const char *trans, const int *ijob, const int *m, const int *n,
-    const double *A, const int *ldA, const double *B, const int *ldB,
-    double *C, const int *ldC,
-    const double *D, const int *ldD, const double *E, const int *ldE,
-    double *F, const int *ldF,
-    double *scale, double *dif,
-    double *Work, const int *lWork, int *iWork, int *info
-) {
-    RELAPACK_dtgsyl(trans, ijob, m, n, A, ldA, B, ldB, C, ldC, D, ldD, E, ldE, F, ldF, scale, dif, Work, lWork, iWork, info);
-}
-#endif
-
-#if INCLUDE_CTGSYL
-void LAPACK(ctgsyl)(
-    const char *trans, const int *ijob, const int *m, const int *n,
-    const float *A, const int *ldA, const float *B, const int *ldB,
-    float *C, const int *ldC,
-    const float *D, const int *ldD, const float *E, const int *ldE,
-    float *F, const int *ldF,
-    float *scale, float *dif,
-    float *Work, const int *lWork, int *iWork, int *info
-) {
-    RELAPACK_ctgsyl(trans, ijob, m, n, A, ldA, B, ldB, C, ldC, D, ldD, E, ldE, F, ldF, scale, dif, Work, lWork, iWork, info);
-}
-#endif
-
-#if INCLUDE_ZTGSYL
-void LAPACK(ztgsyl)(
-    const char *trans, const int *ijob, const int *m, const int *n,
-    const double *A, const int *ldA, const double *B, const int *ldB,
-    double *C, const int *ldC,
-    const double *D, const int *ldD, const double *E, const int *ldE,
-    double *F, const int *ldF,
-    double *scale, double *dif,
-    double *Work, const int *lWork, int *iWork, int *info
-) {
-    RELAPACK_ztgsyl(trans, ijob, m, n, A, ldA, B, ldB, C, ldC, D, ldD, E, ldE, F, ldF, scale, dif, Work, lWork, iWork, info);
-}
-#endif
-
-
-////////////
-// XGEMMT //
-////////////
-
-#if INCLUDE_SGEMMT
-void LAPACK(sgemmt)(
-    const char *uplo, const char *transA, const char *transB,
-    const int *n, const int *k,
-    const float *alpha, const float *A, const int *ldA,
-    const float *B, const int *ldB,
-    const float *beta, float *C, const int *ldC
-) {
-    RELAPACK_sgemmt(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_DGEMMT
-void LAPACK(dgemmt)(
-    const char *uplo, const char *transA, const char *transB,
-    const int *n, const int *k,
-    const double *alpha, const double *A, const int *ldA,
-    const double *B, const int *ldB,
-    const double *beta, double *C, const int *ldC
-) {
-    RELAPACK_dgemmt(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_CGEMMT
-void LAPACK(cgemmt)(
-    const char *uplo, const char *transA, const char *transB,
-    const int *n, const int *k,
-    const float *alpha, const float *A, const int *ldA,
-    const float *B, const int *ldB,
-    const float *beta, float *C, const int *ldC
-) {
-    RELAPACK_cgemmt(uplo, n, A, ldA, info);
-}
-#endif
-
-#if INCLUDE_ZGEMMT
-void LAPACK(zgemmt)(
-    const char *uplo, const char *transA, const char *transB,
-    const int *n, const int *k,
-    const double *alpha, const double *A, const int *ldA,
-    const double *B, const int *ldB,
-    const double *beta, double *C, const int *ldC
-) {
-    RELAPACK_zgemmt(uplo, n, A, ldA, info);
-}
-#endif
diff --git a/test/Makefile b/test/Makefile
index 54fa60533..6c5f041c2 100644
--- a/test/Makefile
+++ b/test/Makefile
@@ -259,10 +259,6 @@ endif
 
 FLDFLAGS = $(FFLAGS:-fPIC=) $(LDFLAGS)
 
-ifeq ($(CORE), C910V)
-EXTRALIB =
-CEXTRALIB =
-endif
 
 ifeq ($(USE_OPENMP), 1)
 ifeq ($(F_COMPILER), GFORTRAN)
diff --git a/utest/ctest.h b/utest/ctest.h
index d316b1494..037f7f28d 100644
--- a/utest/ctest.h
+++ b/utest/ctest.h
@@ -28,7 +28,10 @@
 #define WEAK
 #endif
 
+#ifndef __MSC_VER
 #include <inttypes.h> /* intmax_t, uintmax_t, PRI* */
+#endif
+
 #include <stddef.h> /* size_t */
 
 typedef void (*SetupFunc)(void*);
@@ -72,6 +75,13 @@ struct ctest {
 #define __CTEST_NO_TIME
 #define CTEST_NO_COLORS
 
+#if __MSC_VER >= 1500
+#include <inttypes.h>
+#else
+#include <stdint.h>
+#define CTEST_NO_INTTYPES
+#endif
+
 #ifndef CTEST_ADD_TESTS_MANUALLY
 #pragma section(".ctest$a")
 #pragma section(".ctest$u")
@@ -480,11 +490,19 @@ void assert_data(const unsigned char* exp, size_t expsize,
                  const char* caller, int line) {
     size_t i;
     if (expsize != realsize) {
+#ifndef CTEST_NO_INTTYPES
         CTEST_ERR("%s:%d  expected %" PRIuMAX " bytes, got %" PRIuMAX, caller, line, (uintmax_t) expsize, (uintmax_t) realsize);
+#else
+        CTEST_ERR("%s:%d  expected %u bytes, got %u", caller, line, (uintmax_t) expsize, (uintmax_t) realsize);
+#endif
     }
     for (i=0; i<expsize; i++) {
         if (exp[i] != real[i]) {
+#ifndef CTEST_NO_INTTYPES
             CTEST_ERR("%s:%d expected 0x%02x at offset %" PRIuMAX " got 0x%02x",
+#else
+            CTEST_ERR("%s:%d expected 0x%02x at offset %u got 0x%02x",
+#endif
                 caller, line, exp[i], (uintmax_t) i, real[i]);
         }
     }
@@ -492,31 +510,51 @@ void assert_data(const unsigned char* exp, size_t expsize,
 
 void assert_equal(intmax_t exp, intmax_t real, const char* caller, int line) {
     if (exp != real) {
+#ifndef CTEST_NO_INTTYPES
         CTEST_ERR("%s:%d  expected %" PRIdMAX ", got %" PRIdMAX, caller, line, exp, real);
+#else
+        CTEST_ERR("%s:%d  expected %d, got %d", caller, line, exp, real);
+#endif
     }
 }
 
 void assert_equal_u(uintmax_t exp, uintmax_t real, const char* caller, int line) {
     if (exp != real) {
+#ifndef CTEST_NO_INTTYPES
         CTEST_ERR("%s:%d  expected %" PRIuMAX ", got %" PRIuMAX, caller, line, exp, real);
+#else
+        CTEST_ERR("%s:%d  expected %u, got %u", caller, line, exp, real);
+#endif
     }
 }
 
 void assert_not_equal(intmax_t exp, intmax_t real, const char* caller, int line) {
     if ((exp) == (real)) {
+#ifndef CTEST_NO_INTTYPES
         CTEST_ERR("%s:%d  should not be %" PRIdMAX, caller, line, real);
+#else
+        CTEST_ERR("%s:%d  should not be %d", caller, line, real);
+#endif
     }
 }
 
 void assert_not_equal_u(uintmax_t exp, uintmax_t real, const char* caller, int line) {
     if ((exp) == (real)) {
+#ifndef CTEST_NO_INTTYPES
         CTEST_ERR("%s:%d  should not be %" PRIuMAX, caller, line, real);
+#else
+        CTEST_ERR("%s:%d  should not be %u", caller, line, real);
+#endif
     }
 }
 
 void assert_interval(intmax_t exp1, intmax_t exp2, intmax_t real, const char* caller, int line) {
     if (real < exp1 || real > exp2) {
+#ifndef CTEST_NO_INTTYPES
         CTEST_ERR("%s:%d  expected %" PRIdMAX "-%" PRIdMAX ", got %" PRIdMAX, caller, line, exp1, exp2, real);
+#else
+        CTEST_ERR("%s:%d  expected %d-%d, got %d", caller, line, exp1, exp2, real);
+#endif
     }
 }