Merge pull request #3394 from xianyi/develop

Merge from develop for 0.3.18
2021-10-02 19:35:27 +02:00 · 2021-10-02 19:35:27 +02:00 · c75759876c
parent 686e1f0052 9549167357
commit c75759876c
165 changed files with 35976 additions and 982 deletions
--- a/.travis.yml
+++ b/.travis.yml
@ -1,33 +1,38 @@
 # XXX: Precise is already deprecated, new default is Trusty.
 # https://blog.travis-ci.com/2017-07-11-trusty-as-default-linux-is-coming
-dist: precise
+dist: focal
 sudo: true
 language: c

 matrix:
  include:
    - &test-ubuntu
-      os: linux
+#      os: linux
      compiler: gcc
      addons:
        apt:
          packages:
            - gfortran
+#      before_script: &common-before
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
+#      script:
+#        - make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#        - make -C test $COMMON_FLAGS $BTYPE
+#        - make -C ctest $COMMON_FLAGS $BTYPE
+#        - make -C utest $COMMON_FLAGS $BTYPE
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64"
+#
+#    - <<: *test-ubuntu
+      os: linux-ppc64le
      before_script: &common-before
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
+        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=POWER8 NUM_THREADS=32"
      script:
        - make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
        - make -C test $COMMON_FLAGS $BTYPE
        - make -C ctest $COMMON_FLAGS $BTYPE
        - make -C utest $COMMON_FLAGS $BTYPE
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64"
-
-    - <<: *test-ubuntu
-      os: linux-ppc64le
-      before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=POWER8 NUM_THREADS=32"
      env:
        # for matrix annotation only
        - TARGET_BOX=PPC64LE_LINUX
@ -55,38 +60,38 @@ matrix:
        - TARGET_BOX=IBMZ_LINUX
        - BTYPE="BINARY=64 USE_OPENMP=0 CC=clang"

-    - <<: *test-ubuntu
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 USE_OPENMP=1"
-
-    - <<: *test-ubuntu
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 INTERFACE64=1"
-
-    - <<: *test-ubuntu
-      compiler: clang
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 CC=clang"
-
-    - <<: *test-ubuntu
-      compiler: clang
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 INTERFACE64=1 CC=clang"
-
-    - <<: *test-ubuntu
-      addons:
-        apt:
-          packages:
-            - gcc-multilib
-            - gfortran-multilib
-      env:
-        - TARGET_BOX=LINUX32
-        - BTYPE="BINARY=32"
-
+#    - <<: *test-ubuntu
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64 USE_OPENMP=1"
+#
+#    - <<: *test-ubuntu
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64 INTERFACE64=1"
+#
+#    - <<: *test-ubuntu
+#      compiler: clang
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64 CC=clang"
+#
+#    - <<: *test-ubuntu
+#      compiler: clang
+#      env:
+#        - TARGET_BOX=LINUX64
+#        - BTYPE="BINARY=64 INTERFACE64=1 CC=clang"
+#
+#    - <<: *test-ubuntu
+#      addons:
+#        apt:
+#          packages:
+#            - gcc-multilib
+#            - gfortran-multilib
+#      env:
+#        - TARGET_BOX=LINUX32
+#        - BTYPE="BINARY=32"
+#
    - os: linux
      arch: ppc64le
      dist: bionic
@ -121,47 +126,47 @@ matrix:
        # for matrix annotation only
        - TARGET_BOX=PPC64LE_LINUX_P9

-    - os: linux
-      compiler: gcc
-      addons:
-        apt:
-          packages:
-            - binutils-mingw-w64-x86-64
-            - gcc-mingw-w64-x86-64
-            - gfortran-mingw-w64-x86-64
-      before_script: *common-before
-      script:
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-      env:
-        - TARGET_BOX=WIN64
-        - BTYPE="BINARY=64 HOSTCC=gcc CC=x86_64-w64-mingw32-gcc FC=x86_64-w64-mingw32-gfortran"
-
+#    - os: linux
+#      compiler: gcc
+#      addons:
+#        apt:
+#          packages:
+#            - binutils-mingw-w64-x86-64
+#            - gcc-mingw-w64-x86-64
+#            - gfortran-mingw-w64-x86-64
+#      before_script: *common-before
+#      script:
+#        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#      env:
+#        - TARGET_BOX=WIN64
+#        - BTYPE="BINARY=64 HOSTCC=gcc CC=x86_64-w64-mingw32-gcc FC=x86_64-w64-mingw32-gfortran"
+#
    # Build & test on Alpine Linux inside chroot, i.e. on system with musl libc.
    # These jobs needs sudo, so Travis runs them on VM-based infrastructure
    # which is slower than container-based infrastructure used for jobs
    # that don't require sudo.
-    - &test-alpine
-      os: linux
-      dist: trusty
-      sudo: true
-      language: minimal
-      before_install:
-        - "wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.9.0/alpine-chroot-install' \
-          && echo 'e5dfbbdc0c4b3363b99334510976c86bfa6cb251  alpine-chroot-install' | sha1sum -c || exit 1"
-        - alpine() { /alpine/enter-chroot -u "$USER" "$@"; }
-      install:
-        - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
-      before_script: *common-before
-      script:
-        # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
-        - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-              CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
-        - alpine make -C test $COMMON_FLAGS $BTYPE
-        - alpine make -C ctest $COMMON_FLAGS $BTYPE
-        - alpine make -C utest $COMMON_FLAGS $BTYPE
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64"
+ #   - &test-alpine
+ #     os: linux
+ #     dist: trusty
+ #     sudo: true
+ #     language: minimal
+ #     before_install:
+ #       - "wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.9.0/alpine-chroot-install' \
+ #         && echo 'e5dfbbdc0c4b3363b99334510976c86bfa6cb251  alpine-chroot-install' | sha1sum -c || exit 1"
+ #       - alpine() { /alpine/enter-chroot -u "$USER" "$@"; }
+ #     install:
+ #       - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
+ #     before_script: *common-before
+ #     script:
+ #       # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
+ #       - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+ #             CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
+ #       - alpine make -C test $COMMON_FLAGS $BTYPE
+ #       - alpine make -C ctest $COMMON_FLAGS $BTYPE
+ #       - alpine make -C utest $COMMON_FLAGS $BTYPE
+ #     env:
+ #       - TARGET_BOX=LINUX64_MUSL
+ #       - BTYPE="BINARY=64"

    # XXX: This job segfaults in TESTS OF THE COMPLEX LEVEL 3 BLAS,
    # but only on Travis CI, cannot reproduce it elsewhere.
@ -171,98 +176,98 @@ matrix:
    #    - TARGET_BOX=LINUX64_MUSL
    #    - BTYPE="BINARY=64 USE_OPENMP=1"

-    - <<: *test-alpine
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64 INTERFACE64=1"
+#    - <<: *test-alpine
+#      env:
+#        - TARGET_BOX=LINUX64_MUSL
+#        - BTYPE="BINARY=64 INTERFACE64=1"
+#
+#    # Build with the same flags as Alpine do in OpenBLAS package.
+#    - <<: *test-alpine
+#      env:
+#        - TARGET_BOX=LINUX64_MUSL
+#        - BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=CORE2"

-    # Build with the same flags as Alpine do in OpenBLAS package.
-    - <<: *test-alpine
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=CORE2"
+#    - &test-cmake
+#      os: linux
+#      compiler: clang
+#      addons:
+#        apt:
+#          packages:
+#            - gfortran
+#            - cmake
+#      dist: trusty
+#      sudo: true
+#      before_script:
+#        - COMMON_ARGS="-DTARGET=NEHALEM -DNUM_THREADS=32"
+#      script:
+#        - mkdir build
+#        - CONFIG=Release
+#        - cmake -Bbuild -H. $CMAKE_ARGS $COMMON_ARGS -DCMAKE_BUILD_TYPE=$CONFIG
+#        - cmake --build build --config $CONFIG -- -j2
+#      env:
+#        - CMAKE=1
+#    - <<: *test-cmake
+#      env:
+#        - CMAKE=1 CMAKE_ARGS="-DNOFORTRAN=1"
+#    - <<: *test-cmake
+#      compiler: gcc
+#      env:
+#        - CMAKE=1

-    - &test-cmake
-      os: linux
-      compiler: clang
-      addons:
-        apt:
-          packages:
-            - gfortran
-            - cmake
-      dist: trusty
-      sudo: true
-      before_script:
-        - COMMON_ARGS="-DTARGET=NEHALEM -DNUM_THREADS=32"
-      script:
-        - mkdir build
-        - CONFIG=Release
-        - cmake -Bbuild -H. $CMAKE_ARGS $COMMON_ARGS -DCMAKE_BUILD_TYPE=$CONFIG
-        - cmake --build build --config $CONFIG -- -j2
-      env:
-        - CMAKE=1
-    - <<: *test-cmake
-      env:
-        - CMAKE=1 CMAKE_ARGS="-DNOFORTRAN=1"
-    - <<: *test-cmake
-      compiler: gcc
-      env:
-        - CMAKE=1
-
-    - &test-macos
-      os: osx
-      osx_image: xcode11.5
-      before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
-      script:
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-      env:
-        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-9"
-
-    - <<: *test-macos
-      osx_image: xcode12
-      before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
-        - brew update
-      script:
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-      env:
-        - BTYPE="TARGET=HASWELL USE_OPENMP=1 BINARY=64 INTERFACE64=1 CC=gcc-10 FC=gfortran-10"
-
-    - <<: *test-macos
-      osx_image: xcode12
-      before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
-        - brew update
-      script:
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-      env:
-        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-10"      
+#    - &test-macos
+#      os: osx
+#      osx_image: xcode11.5
+#      before_script:
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+#      script:
+#        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#      env:
+#        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-9"
+#
+#    - <<: *test-macos
+#      osx_image: xcode12
+#      before_script:
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+#        - brew update
+#      script:
+#        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#      env:
+#        - BTYPE="TARGET=HASWELL USE_OPENMP=1 BINARY=64 INTERFACE64=1 CC=gcc-10 FC=gfortran-10"
+#
+#    - <<: *test-macos
+#      osx_image: xcode12
+#      before_script:
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+#        - brew update
+#      script:
+#        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+#      env:
+#        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-10"      

  #  - <<: *test-macos
  #    osx_image: xcode10
  #    env:
  #      - BTYPE="TARGET=NEHALEM BINARY=32 NOFORTRAN=1"

-    - <<: *test-macos
-      osx_image: xcode11.5
-      before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
-        - brew update
-      env:
+#    - <<: *test-macos
+#      osx_image: xcode11.5
+#      before_script:
+#        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
+#        - brew update
+#      env:
 #        - CC="/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
 #        - CFLAGS="-O2 -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch arm64 -miphoneos-version-min=10.0"
-        - CC="/Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
-        - CFLAGS="-O2 -Wno-macro-redefined -isysroot /Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk -arch arm64 -miphoneos-version-min=10.0"
-        - BTYPE="TARGET=ARMV8 BINARY=64 HOSTCC=clang NOFORTRAN=1"
-    - <<: *test-macos
-      osx_image: xcode11.5
-      env:
-#        - CC="/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
-#        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch armv7 -miphoneos-version-min=5.1"
-        - CC="/Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
-        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk -arch armv7 -miphoneos-version-min=5.1"
-        - BTYPE="TARGET=ARMV7 HOSTCC=clang NOFORTRAN=1"
+#        - CC="/Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
+#        - CFLAGS="-O2 -Wno-macro-redefined -isysroot /Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk -arch arm64 -miphoneos-version-min=10.0"
+#        - BTYPE="TARGET=ARMV8 BINARY=64 HOSTCC=clang NOFORTRAN=1"
+#    - <<: *test-macos
+#      osx_image: xcode11.5
+#      env:
+##        - CC="/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
+##        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch armv7 -miphoneos-version-min=5.1"
+#        - CC="/Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
+#        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-11.5.GM.Seed.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk -arch armv7 -miphoneos-version-min=5.1"
+#        - BTYPE="TARGET=ARMV7 HOSTCC=clang NOFORTRAN=1"

    - &test-graviton2
      os: linux
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -132,7 +132,7 @@ endif ()

 if (BUILD_BFLOAT16)
  message(STATUS "Building Half Precision")
-  list(APPEND FLOAT_TYPES "BFLOAT16") # defines nothing
+  #  list(APPEND FLOAT_TYPES "BFLOAT16") # defines nothing
 endif ()

 if (NOT DEFINED CORE OR "${CORE}" STREQUAL "UNKNOWN")
--- a/Changelog.txt
+++ b/Changelog.txt
@ -1,4 +1,47 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.18
+ 02-Oct-2021
+
+general:
+ - when the build-time number of preconfigured threads is exceeded
+   at runtime (typically by an external program calling BLAS functions
+   from a larger number of threads in parallel), OpenBLAS will now 
+   allocate an auxiliary control structure for up to 512 additional
+   threads instead of aborting
+ - added support for Loongson's LoongArch64 cpu architecture
+ - fixed building OpenBLAS with CMAKE and -DBUILD_BFLOAT16=ON
+ - added support for building OpenBLAS as a CMAKE subproject
+ - added support for building for Windows/ARM64 targets with clang
+ - improved support for building with the IBM xlf compiler
+ - imported Reference-LAPACK PR 625 (out-of-bounds reads in ?LARRV)
+ - imported Reference-LAPACK PR 597 for testsuite compatibility with
+   LLVM's libomp
+
+x86_64:
+ - added SkylakeX S/DGEMM kernels for small problem sizes (M*N*K<=1000000)
+ - added optimized SBGEMM for Intel Cooper Lake
+ - reinstated the performance patch for AVX512 SGEMV_T with a proper fix
+ - added a workaround for a gcc11 tree-vectorizer bug that caused spurious
+   failures in the test programs for complex BLAS3 when compiling at -O3
+   (the default for cmake "release" builds)
+ - added support for runtime cpu count detection under Haiku OS
+ - worked around a long-standing miscompilation issue of the Haswell DGEMV_T
+   kernel with gcc that could produce NaN output in some corner cases
+
+POWER:
+ - improved performance of DASUM on POWER10
+
+ARMV8:
+ - fixed crashes (use of reserved register x18) on Apple M1 under OSX
+ - fixed building with gcc releases earlier than 5.1
+
+MIPS:
+ - fixed building under BSD
+
+MIPS64:
+ - fixed building under BSD
+
 ====================================================================
 Version 0.3.17
 15-Jul-2021
--- a/2
+++ b/2
@ -269,7 +269,7 @@ prof_lapack : lapack_prebuild
 lapack_prebuild :
 ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	-@echo "FC          = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "FFLAGS      = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "override FFLAGS      = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "FFLAGS_DRV  = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "POPTS       = $(LAPACK_FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "FFLAGS_NOOPT       = -O0 $(LAPACK_NOOPT)" >> $(NETLIB_LAPACK_DIR)/make.inc
--- a/Makefile.arm64
+++ b/Makefile.arm64
@ -1,4 +1,15 @@
 ifneq ($(C_COMPILER), PGI)
+
+ifneq ($(GCCVERSIONGT4), 1)
+CCOMMON_OPT += -march=armv8-a
+ifneq ($(F_COMPILER), NAG)
+FCOMMON_OPT += -march=armv8-a
+endif
+
+
+else 
+
+
 ifeq ($(CORE), ARMV8)
 CCOMMON_OPT += -march=armv8-a
 ifneq ($(F_COMPILER), NAG)
@ -138,4 +149,7 @@ FCOMMON_OPT += -march=armv8-a -mtune=emag
 endif
 endif
 endif
+
+endif
+
 endif
--- a/Makefile.loongarch64
+++ b/Makefile.loongarch64
@ -0,0 +1,3 @@
+ifdef BINARY64
+else
+endif
--- a/Makefile.power
+++ b/Makefile.power
@ -12,9 +12,13 @@ endif
 ifeq ($(CORE), POWER10)
 ifneq ($(C_COMPILER), PGI)
 CCOMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math
+ifeq ($(F_COMPILER), IBM)
+FCOMMON_OPT += -O2 -qrecur -qnosave
+else
 FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10  -fno-fast-math
 endif
 endif
+endif

 ifeq ($(CORE), POWER9)
 ifneq ($(C_COMPILER), PGI)
@ -33,7 +37,11 @@ else
 CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
 endif
 ifneq ($(F_COMPILER), PGI)
+ifeq ($(F_COMPILER), IBM)
+FCOMMON_OPT += -O2 -qrecur -qnosave
+else
 FCOMMON_OPT += -O2 -frecursive -fno-fast-math
+endif
 ifeq ($(C_COMPILER), GCC)
 ifneq ($(GCCVERSIONGT4), 1)
 $(warning your compiler is too old to fully support POWER9, getting a newer version of gcc is recommended)
@ -57,7 +65,11 @@ CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
 endif
 ifneq ($(F_COMPILER), PGI)
 ifeq ($(OSNAME), AIX)
+ifeq ($(F_COMPILER), IBM)
+FCOMMON_OPT += -O2 -qrecur -qnosave
+else
 FCOMMON_OPT += -O1 -frecursive -mcpu=power8 -mtune=power8  -fno-fast-math 
+endif
 else
 FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8  -fno-fast-math 
 endif
--- a/Makefile.rule
+++ b/Makefile.rule
@ -3,7 +3,7 @@
 #

 # This library's version
-VERSION = 0.3.17
+VERSION = 0.3.17.dev

 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
--- a/Makefile.system
+++ b/Makefile.system
@ -33,6 +33,10 @@ else ifeq ($(ARCH), armv7)
 override ARCH=arm
 else ifeq ($(ARCH), aarch64)
 override ARCH=arm64
+else ifeq ($(ARCH), mipsel)
+override ARCH=mips
+else ifeq ($(ARCH), mips64el)
+override ARCH=mips64
 else ifeq ($(ARCH), zarch)
 override ARCH=zarch
 endif
@ -244,6 +248,14 @@ else
 ONLY_CBLAS = 0
 endif

+#For small matrix optimization
+ifeq ($(ARCH), x86_64)
+SMALL_MATRIX_OPT = 1
+endif
+ifeq ($(SMALL_MATRIX_OPT), 1)
+CCOMMON_OPT += -DSMALL_MATRIX_OPT
+endif
+
 # This operation is expensive, so execution should be once.
 ifndef GOTOBLAS_MAKEFILE
 export GOTOBLAS_MAKEFILE = 1
@ -780,6 +792,11 @@ NO_BINARY_MODE  = 1
 BINARY_DEFINED  = 1
 endif

+ifeq ($(ARCH), loongarch64)
+NO_BINARY_MODE  = 1
+BINARY_DEFINED  = 1
+endif
+

 #
 #  C Compiler dependent settings
@ -850,6 +867,13 @@ ifeq ($(OSNAME), AIX)
 BINARY_DEFINED = 1
 endif

+ifeq ($(ARCH), loongarch64)
+ifeq ($(CORE), LOONGSON3R5)
+CCOMMON_OPT += -march=loongarch64 -mabi=lp64
+FCOMMON_OPT += -march=loongarch64 -mabi=lp64
+endif
+endif
+
 endif

 ifndef BINARY_DEFINED
--- a/README.md
+++ b/README.md
@ -2,7 +2,7 @@

 [![Join the chat at https://gitter.im/xianyi/OpenBLAS](https://badges.gitter.im/Join%20Chat.svg)](https://gitter.im/xianyi/OpenBLAS?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge)

-Travis CI: [![Build Status](https://travis-ci.org/xianyi/OpenBLAS.svg?branch=develop)](https://travis-ci.org/xianyi/OpenBLAS)
+Travis CI: [![Build Status](https://travis-ci.com/xianyi/OpenBLAS.svg?branch=develop)](https://travis-ci.com/xianyi/OpenBLAS)

 AppVeyor: [![Build status](https://ci.appveyor.com/api/projects/status/09sohd35n8nkkx64/branch/develop?svg=true)](https://ci.appveyor.com/project/xianyi/openblas/branch/develop)

@ -128,6 +128,7 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
 - **Intel Sandy Bridge**: Optimized Level-3 and Level-2 BLAS with AVX on x86-64.
 - **Intel Haswell**: Optimized Level-3 and Level-2 BLAS with AVX2 and FMA on x86-64.
 - **Intel Skylake-X**: Optimized Level-3 and Level-2 BLAS with AVX512 and FMA on x86-64.
+- **Intel Cooper Lake**: as Skylake-X with improved BFLOAT16 support.
 - **AMD Bobcat**: Used GotoBLAS2 Barcelona codes.
 - **AMD Bulldozer**: x86-64 ?GEMM FMA4 kernels. (Thanks to Werner Saar)
 - **AMD PILEDRIVER**: Uses Bulldozer codes with some optimizations.
@ -153,6 +154,7 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th

 - **ARMv8**: Basic ARMV8 with small caches, optimized Level-3 and Level-2 BLAS
 - **Cortex-A53**: same as ARMV8 (different cpu specifications)
+- **Cortex-A55**: same as ARMV8 (different cpu specifications)
 - **Cortex A57**: Optimized Level-3 and Level-2 functions
 - **Cortex A72**: same as A57 ( different cpu specifications)
 - **Cortex A73**: same as A57 (different cpu specifications)
@ -178,10 +180,11 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th

 #### RISC-V

- **C910V**: Optimized Leve-3 BLAS (real) and Level-1,2 by RISC-V Vector extension 0.7.1.
+- **C910V**: Optimized Level-3 BLAS (real) and Level-1,2 by RISC-V Vector extension 0.7.1.
  ```sh
  make HOSTCC=gcc TARGET=C910V CC=riscv64-unknown-linux-gnu-gcc FC=riscv64-unknown-linux-gnu-gfortran
  ```
+  (also known to work on C906)

 ### Support for multiple targets in a single library

--- a/TargetList.txt
+++ b/TargetList.txt
@ -110,3 +110,5 @@ Z14
 RISCV64_GENERIC
 C910V

+11.LOONGARCH64:
+LOONGSON3R5
--- a/azure-pipelines.yml
+++ b/azure-pipelines.yml
@ -19,7 +19,7 @@ jobs:
 # of gcc / glibc
 - job: manylinux1_gcc
  pool:
-    vmImage: 'ubuntu-16.04'
+    vmImage: 'ubuntu-latest'
  steps:
  - script: |
      echo "FROM quay.io/pypa/manylinux1_x86_64
@ -35,7 +35,7 @@ jobs:
    displayName: Run manylinux1 docker build
 - job: Intel_SDE_skx
  pool:
-    vmImage: 'ubuntu-16.04'
+    vmImage: 'ubuntu-latest'
  steps:
  - script: |
      # at the time of writing the available Azure Ubuntu vm image
@ -83,6 +83,8 @@ jobs:
  - script: |
      brew update
      make TARGET=CORE2 DYNAMIC_ARCH=1 USE_OPENMP=1 INTERFACE64=1 CC=gcc-10 FC=gfortran-10
+      make TARGET=CORE2 DYNAMIC_ARCH=1 USE_OPENMP=1 INTERFACE64=1 CC=gcc-10 FC=gfortran-10 PREFIX=../blasinst install
+      ls -lR ../blasinst
     
 - job: OSX_GCC_Nothreads
  pool:
@ -104,6 +106,38 @@ jobs:
      brew install llvm libomp
      make TARGET=CORE2 USE_OPENMP=1 INTERFACE64=1 DYNAMIC_ARCH=1 CC=/usr/local/opt/llvm/bin/clang FC=gfortran-10

+- job: OSX_OpenMP_Clang_cmake
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     LD_LIBRARY_PATH: /usr/local/opt/llvm/lib
+     LIBRARY_PATH: /usr/local/opt/llvm/lib
+  steps:   
+  - script: |
+      brew update
+      brew install llvm libomp
+      mkdir build
+      cd build
+      cmake -DTARGET=CORE2 -DUSE_OPENMP=1 -DINTERFACE64=1 -DDYNAMIC_ARCH=1 -DCMAKE_C_COMPILER=/usr/local/opt/llvm/bin/clang -DNOFORTRAN=1 -DNO_AVX512=1 ..
+      make
+      ctest
+      
+- job: OSX_OpenMP_Clang_gf_cmake
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     LD_LIBRARY_PATH: /usr/local/opt/llvm/lib
+     LIBRARY_PATH: /usr/local/opt/llvm/lib
+  steps:   
+  - script: |
+      brew update
+      brew install llvm libomp
+      mkdir build
+      cd build
+      cmake -DTARGET=CORE2 -DUSE_OPENMP=1 -DINTERFACE64=1 -DDYNAMIC_ARCH=1 -DCMAKE_C_COMPILER=/usr/local/opt/llvm/bin/clang -DNO_AVX512=1 ..
+      make
+      ctest
+        
 - job: OSX_Ifort_Clang
  pool:
     vmImage: 'macOS-10.15'
@ -147,13 +181,34 @@ jobs:
      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: OSX_IOS_ARMV8
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     CC: /Applications/Xcode_12.4.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+     CFLAGS: -O2 -Wno-macro-redefined -isysroot /Applications/Xcode_12.4.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS14.4.sdk -arch arm64 -miphoneos-version-min=10.0
+  steps:
+  - script: |
+     make TARGET=ARMV8 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1
+
+- job: OSX_IOS_ARMV7
+  pool:
+     vmImage: 'macOS-10.15'
+  variables:
+     CC: /Applications/Xcode_12.4.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+     CFLAGS: -O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode_12.4.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS14.4.sdk -arch armv7 -miphoneos-version-min=5.1
+  steps:
+  - script: |
+     make TARGET=ARMV7 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1
+
 - 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
+        wget https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.13.1/alpine-chroot-install \
+         && echo '7c7e3fa378e69aecc7f5f01bbc759e5f0a9d9b74  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
--- a/7
+++ b/7
@ -94,6 +94,7 @@ $architecture = arm    if ($data =~ /ARCH_ARM/);
 $architecture = arm64        if ($data =~ /ARCH_ARM64/);
 $architecture = zarch        if ($data =~ /ARCH_ZARCH/);
 $architecture = riscv64      if ($data =~ /ARCH_RISCV64/);
+$architecture = loongarch64  if ($data =~ /ARCH_LOONGARCH64/);

 $defined = 0;

@ -143,6 +144,11 @@ if ($architecture eq "riscv64") {
    $binary = 64;
 }

+if ($architecture eq "loongarch64") {
+    $defined = 1;
+    $binary = 64;
+}
+
 if ($compiler eq "PGI") {
    $compiler_name .= " -tp p7"    if ($binary eq "32");
    $compiler_name .= " -tp p7-64" if ($binary eq "64");
@ -226,6 +232,7 @@ $architecture = ia64   if ($data =~ /ARCH_IA64/);
 $architecture = arm          if ($data =~ /ARCH_ARM/);
 $architecture = arm64        if ($data =~ /ARCH_ARM64/);
 $architecture = zarch        if ($data =~ /ARCH_ZARCH/);
+$architecture = loongarch64  if ($data =~ /ARCH_LOONGARCH64/);

 $binformat    = bin32;
 $binformat    = bin64  if ($data =~ /BINARY_64/);
--- a/cblas.h
+++ b/cblas.h
@ -400,6 +400,8 @@ void   cblas_dbf16tod(OPENBLAS_CONST blasint n, OPENBLAS_CONST bfloat16 *in, OPE
 float  cblas_sbdot(OPENBLAS_CONST blasint n, OPENBLAS_CONST bfloat16 *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST bfloat16 *y, OPENBLAS_CONST blasint incy);
 void   cblas_sbgemv(OPENBLAS_CONST enum CBLAS_ORDER order,  OPENBLAS_CONST enum CBLAS_TRANSPOSE trans,  OPENBLAS_CONST blasint m, OPENBLAS_CONST blasint n, OPENBLAS_CONST float alpha, OPENBLAS_CONST bfloat16 *a, OPENBLAS_CONST blasint lda, OPENBLAS_CONST bfloat16 *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float beta, float *y, OPENBLAS_CONST blasint incy);

+void   cblas_sbgemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
+		    OPENBLAS_CONST float alpha, OPENBLAS_CONST bfloat16 *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST bfloat16 *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
 #ifdef __cplusplus
 }
 #endif  /* __cplusplus */
--- a/cmake/arch.cmake
+++ b/cmake/arch.cmake
@ -113,6 +113,10 @@ if (MIPS64)
  set(NO_BINARY_MODE 1)
 endif ()

+if (LOONGARCH64)
+  set(NO_BINARY_MODE 1)
+endif ()
+
 if (${ARCH} STREQUAL "alpha")
  set(NO_BINARY_MODE 1)
  set(BINARY_DEFINED 1)
--- a/cmake/cc.cmake
+++ b/cmake/cc.cmake
@ -29,6 +29,15 @@ if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU" OR ${CMAKE_C_COMPILER_ID} STREQUAL "LS
      set(FCOMMON_OPT "${FCOMMON_OPT} -march=mips64")
    endif ()

+    if (LOONGARCH64)
+      if (BINARY64)
+        set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=lp64")
+      else ()
+        set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=lp32")
+      endif ()
+      set(BINARY_DEFINED 1)
+    endif ()
+
    if (CMAKE_SYSTEM_NAME STREQUAL "AIX")
      set(BINARY_DEFINED 1)
    endif ()
@ -124,9 +133,9 @@ if (NOT DYNAMIC_ARCH)
 	if (HAVE_AVX)
        set (CCOMMON_OPT  "${CCOMMON_OPT} -mavx")
 	endif ()
-	if (HAVE_FMA3)
-	set (CCOMMON_OPT  "${CCOMMON_OPT} -mfma")
-	endif ()
+	#	if (HAVE_FMA3)
+	#set (CCOMMON_OPT  "${CCOMMON_OPT} -mfma")
+	#endif ()
 	if (HAVE_SSE)
 	set (CCOMMON_OPT  "${CCOMMON_OPT} -msse")
 	endif ()
--- a/cmake/fc.cmake
+++ b/cmake/fc.cmake
@ -61,6 +61,13 @@ if (${F_COMPILER} STREQUAL "GFORTRAN")
        set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=n32")
      endif ()
    endif ()
+    if (LOONGARCH64)
+      if (BINARY64)
+        set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=lp64")
+      else ()
+        set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=lp32")
+      endif ()
+    endif ()
  else ()
    if (BINARY64)
      set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
@ -97,7 +104,7 @@ endif ()

 if (${F_COMPILER} STREQUAL "IBM")
  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_IBM")
-  # FCOMMON_OPT	+= -qarch=440
+  set(FCOMMON_OPT "${FCOMMON_OPT} -qrecur")
  if (BINARY64)
    set(FCOMMON_OPT "${FCOMMON_OPT} -q64")
    if (INTERFACE64)
--- a/cmake/kernel.cmake
+++ b/cmake/kernel.cmake
@ -134,6 +134,8 @@ if (BUILD_BFLOAT16)
  set(SHSWAPKERNEL ../arm/swap.c)
  set(TOBF16KERNEL ../x86_64/tobf16.c)
  set(BF16TOKERNEL ../x86_64/bf16to.c)
+  set(SBGEMVNKERNEL ../x86_64/sbgemv_n.c)
+  set(SBGEMVTKERNEL ../x86_64/sbgemv_t.c)
 endif ()
 endmacro ()

--- a/cmake/system.cmake
+++ b/cmake/system.cmake
@ -186,11 +186,11 @@ if (DEFINED TARGET)
      	  set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
 	endif()
  endif()
-  if (DEFINED HAVE_FMA3)
-	if (NOT NO_AVX2)
-    set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mfma")
-	endif()
-  endif()
+  #  if (DEFINED HAVE_FMA3)
+  #	if (NOT NO_AVX2)
+  #  set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mfma")
+  #	endif()
+  #  endif()
    if (DEFINED HAVE_SSE)
    set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse")
  endif()
@ -258,6 +258,13 @@ if (NEED_PIC)
  endif()
 endif ()

+if (X86_64)
+  set(SMALL_MATRIX_OPT TRUE)
+endif ()
+if (SMALL_MATRIX_OPT)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DSMALL_MATRIX_OPT")
+endif ()
+
 if (DYNAMIC_ARCH)
  if (X86 OR X86_64 OR ARM64 OR PPC)
    set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_ARCH")
@ -462,6 +469,9 @@ endif()
 if (BUILD_COMPLEX16)
 	set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_COMPLEX16")
 endif()
+if (BUILD_BFLOAT16)
+       set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DBUILD_BFLOAT16")
+endif()
 if(NOT MSVC)
 set(CMAKE_ASM_FLAGS "${CMAKE_ASM_FLAGS} ${CCOMMON_OPT}")
 endif()
--- a/cmake/system_check.cmake
+++ b/cmake/system_check.cmake
@ -38,6 +38,8 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc.*|power.*|Power.*")
  set(PPC 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "mips64.*")
  set(MIPS64 1)
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "loongarch64.*")
+  set(LOONGARCH64 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
  if (NOT BINARY)
    if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
@ -95,7 +97,7 @@ else()
 endif ()

 if (NOT BINARY)
-  if (X86_64 OR ARM64 OR PPC OR MIPS64)
+  if (X86_64 OR ARM64 OR PPC OR MIPS64 OR LOONGARCH64)
    set(BINARY 64)
  else ()
    set(BINARY 32)
--- a/cmake/utils.cmake
+++ b/cmake/utils.cmake
@ -157,31 +157,31 @@ endfunction ()
 #                               STRING - compiles only the given type (e.g. DOUBLE)
 function(GenerateNamedObjects sources_in)

-  if (DEFINED ARGV1)
+  if (${ARGC} GREATER 1)
    set(defines_in ${ARGV1})
  endif ()

-  if (DEFINED ARGV2 AND NOT "${ARGV2}" STREQUAL "")
+  if (${ARGC} GREATER 2 AND NOT "${ARGV2}" STREQUAL "")
    set(name_in ${ARGV2})
    # strip off extension for kernel files that pass in the object name.
    get_filename_component(name_in ${name_in} NAME_WE)
  endif ()

-  if (DEFINED ARGV3)
+  if (${ARGC} GREATER 3)
    set(use_cblas ${ARGV3})
  else ()
    set(use_cblas false)
  endif ()

-  if (DEFINED ARGV4)
+  if (${ARGC} GREATER 4)
    set(replace_last_with ${ARGV4})
  endif ()

-  if (DEFINED ARGV5)
+  if (${ARGC} GREATER 5)
    set(append_with ${ARGV5})
  endif ()

-  if (DEFINED ARGV6)
+  if (${ARGC} GREATER 6)
    set(no_float_type ${ARGV6})
  else ()
    set(no_float_type false)
@ -196,7 +196,7 @@ function(GenerateNamedObjects sources_in)
  set(real_only false)
  set(complex_only false)
  set(mangle_complex_sources false)
-  if (DEFINED ARGV7 AND NOT "${ARGV7}" STREQUAL "")
+  if (${ARGC} GREATER 7 AND NOT "${ARGV7}" STREQUAL "")
    if (${ARGV7} EQUAL 1)
      set(real_only true)
    elseif (${ARGV7} EQUAL 2)
@ -311,7 +311,15 @@ function(GenerateNamedObjects sources_in)
      configure_file(${new_source_file}.tmp ${new_source_file} COPYONLY)
      file(REMOVE ${new_source_file}.tmp)
      list(APPEND SRC_LIST_OUT ${new_source_file})
-
+      message (STATUS ${new_source_file})
+      if (DEFINED HAVE_FMA3)
+        if ( ${new_source_file} MATCHES "(s|d?)rot_k.*c")
+		set_source_files_properties(${new_source_file} PROPERTIES COMPILE_OPTIONS "-mfma")
+        endif ()
+        if ( ${new_source_file} MATCHES "dgemv_t_k.*c")
+		set_source_files_properties(${new_source_file} PROPERTIES COMPILE_OPTIONS "-mfma")
+        endif ()
+      endif ()
    endforeach ()
  endforeach ()

@ -334,17 +342,17 @@ endfunction ()
 function(GenerateCombinationObjects sources_in defines_in absent_codes_in all_defines_in replace_scheme)

  set(alternate_name_in "")
-  if (DEFINED ARGV5)
+  if (${ARGC} GREATER 5)
    set(alternate_name_in ${ARGV5})
  endif ()

  set(no_float_type false)
-  if (DEFINED ARGV6)
+  if (${ARGC} GREATER 6)
    set(no_float_type ${ARGV6})
  endif ()

  set(complex_filename_scheme "")
-  if (DEFINED ARGV7)
+  if (${ARGC} GREATER 7)
    set(complex_filename_scheme ${ARGV7})
  endif ()

--- a/common.h
+++ b/common.h
@ -470,6 +470,10 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #include "common_zarch.h"
 #endif

+#ifdef ARCH_LOONGARCH64
+#include "common_loongarch64.h"
+#endif
+
 #ifndef ASSEMBLER
 #ifdef OS_WINDOWSSTORE
 typedef char env_var_t[MAX_PATH];
--- a/common_arm64.h
+++ b/common_arm64.h
@ -120,7 +120,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 	.text ;
 	.p2align 2 ;
 	.global	REALNAME ;
-#ifndef __APPLE__
+#if !defined(__APPLE__) && !defined(_WIN32)
 	.type	REALNAME, %function ;
 #endif
 REALNAME:
--- a/common_c.h
+++ b/common_c.h
@ -232,6 +232,8 @@

 #define CGEADD_K                cgeadd_k 

+#define CGEMM_SMALL_MATRIX_PERMIT	cgemm_small_matrix_permit
+
 #else

 #define	CAMAX_K			gotoblas -> camax_k
@ -426,8 +428,51 @@

 #define CGEADD_K                gotoblas -> cgeadd_k 

+#define CGEMM_SMALL_MATRIX_PERMIT	gotoblas -> cgemm_small_matrix_permit
+
 #endif

+#define CGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(cgemm_small_kernel_nn)
+#define CGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(cgemm_small_kernel_nt)
+#define CGEMM_SMALL_KERNEL_NR		FUNC_OFFSET(cgemm_small_kernel_nr)
+#define CGEMM_SMALL_KERNEL_NC		FUNC_OFFSET(cgemm_small_kernel_nc)
+
+#define CGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(cgemm_small_kernel_tn)
+#define CGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(cgemm_small_kernel_tt)
+#define CGEMM_SMALL_KERNEL_TR		FUNC_OFFSET(cgemm_small_kernel_tr)
+#define CGEMM_SMALL_KERNEL_TC		FUNC_OFFSET(cgemm_small_kernel_tc)
+
+#define CGEMM_SMALL_KERNEL_RN		FUNC_OFFSET(cgemm_small_kernel_rn)
+#define CGEMM_SMALL_KERNEL_RT		FUNC_OFFSET(cgemm_small_kernel_rt)
+#define CGEMM_SMALL_KERNEL_RR		FUNC_OFFSET(cgemm_small_kernel_rr)
+#define CGEMM_SMALL_KERNEL_RC		FUNC_OFFSET(cgemm_small_kernel_rc)
+
+#define CGEMM_SMALL_KERNEL_CN		FUNC_OFFSET(cgemm_small_kernel_cn)
+#define CGEMM_SMALL_KERNEL_CT		FUNC_OFFSET(cgemm_small_kernel_ct)
+#define CGEMM_SMALL_KERNEL_CR		FUNC_OFFSET(cgemm_small_kernel_cr)
+#define CGEMM_SMALL_KERNEL_CC		FUNC_OFFSET(cgemm_small_kernel_cc)
+
+#define CGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(cgemm_small_kernel_b0_nn)
+#define CGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(cgemm_small_kernel_b0_nt)
+#define CGEMM_SMALL_KERNEL_B0_NR	FUNC_OFFSET(cgemm_small_kernel_b0_nr)
+#define CGEMM_SMALL_KERNEL_B0_NC	FUNC_OFFSET(cgemm_small_kernel_b0_nc)
+
+#define CGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(cgemm_small_kernel_b0_tn)
+#define CGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(cgemm_small_kernel_b0_tt)
+#define CGEMM_SMALL_KERNEL_B0_TR	FUNC_OFFSET(cgemm_small_kernel_b0_tr)
+#define CGEMM_SMALL_KERNEL_B0_TC	FUNC_OFFSET(cgemm_small_kernel_b0_tc)
+
+#define CGEMM_SMALL_KERNEL_B0_RN	FUNC_OFFSET(cgemm_small_kernel_b0_rn)
+#define CGEMM_SMALL_KERNEL_B0_RT	FUNC_OFFSET(cgemm_small_kernel_b0_rt)
+#define CGEMM_SMALL_KERNEL_B0_RR	FUNC_OFFSET(cgemm_small_kernel_b0_rr)
+#define CGEMM_SMALL_KERNEL_B0_RC	FUNC_OFFSET(cgemm_small_kernel_b0_rc)
+
+#define CGEMM_SMALL_KERNEL_B0_CN	FUNC_OFFSET(cgemm_small_kernel_b0_cn)
+#define CGEMM_SMALL_KERNEL_B0_CT	FUNC_OFFSET(cgemm_small_kernel_b0_ct)
+#define CGEMM_SMALL_KERNEL_B0_CR	FUNC_OFFSET(cgemm_small_kernel_b0_cr)
+#define CGEMM_SMALL_KERNEL_B0_CC	FUNC_OFFSET(cgemm_small_kernel_b0_cc)
+
+
 #define	CGEMM_NN		cgemm_nn
 #define	CGEMM_CN		cgemm_cn
 #define	CGEMM_TN		cgemm_tn
--- a/common_d.h
+++ b/common_d.h
@ -157,6 +157,8 @@
 #define DIMATCOPY_K_RT      dimatcopy_k_rt
 #define DGEADD_K                dgeadd_k 

+#define DGEMM_SMALL_MATRIX_PERMIT	dgemm_small_matrix_permit
+
 #else

 #define	DAMAX_K			gotoblas -> damax_k
@ -281,8 +283,21 @@

 #define DGEADD_K                gotoblas -> dgeadd_k 

+#define DGEMM_SMALL_MATRIX_PERMIT	gotoblas -> dgemm_small_matrix_permit
+
 #endif

+#define DGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(dgemm_small_kernel_nn)
+#define DGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(dgemm_small_kernel_nt)
+#define DGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(dgemm_small_kernel_tn)
+#define DGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(dgemm_small_kernel_tt)
+
+#define DGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(dgemm_small_kernel_b0_nn)
+#define DGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(dgemm_small_kernel_b0_nt)
+#define DGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(dgemm_small_kernel_b0_tn)
+#define DGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(dgemm_small_kernel_b0_tt)
+
+
 #define	DGEMM_NN		dgemm_nn
 #define	DGEMM_CN		dgemm_tn
 #define	DGEMM_TN		dgemm_tn
--- a/common_level3.h
+++ b/common_level3.h
@ -515,6 +515,129 @@ int qgemm_kernel(BLASLONG, BLASLONG, BLASLONG, xidouble *, xidouble *, xidouble
 int qgemm_kernel(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
 #endif

+#ifdef SMALL_MATRIX_OPT
+int sbgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float beta);
+
+int sbgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+
+int sgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float beta);
+
+int sgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+int sgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+
+int dgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, double alpha, double beta);
+
+int dgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+int dgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+int dgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+int dgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+
+int sbgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sbgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+
+int sgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+int sgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+
+int dgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+int dgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+int dgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+int dgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+
+int cgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha0, float alpha1, float beta0, float beta1);
+
+int cgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_nr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_nc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+	
+int cgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_tr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_tc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+int cgemm_small_kernel_rn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_rt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_rr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_rc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+int cgemm_small_kernel_cn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_ct(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_cr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+int cgemm_small_kernel_cc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+int zgemm_small_matrix_permit(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, double alpha0, double alpha1, double beta0, double beta1);
+
+int zgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_nr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_nc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+	
+int zgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_tr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_tc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+int zgemm_small_kernel_rn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_rt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_rr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_rc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+int zgemm_small_kernel_cn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_ct(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_cr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+int zgemm_small_kernel_cc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+int cgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);	
+int cgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_nr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_nc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+	
+int cgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_tr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_tc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+int cgemm_small_kernel_b0_rn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_rt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_rr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_rc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+int cgemm_small_kernel_b0_cn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_ct(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_cr(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+int cgemm_small_kernel_b0_cc(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+int zgemm_small_kernel_b0_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);	
+int zgemm_small_kernel_b0_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_nr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_nc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+	
+int zgemm_small_kernel_b0_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_tr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_tc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+int zgemm_small_kernel_b0_rn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_rt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_rr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_rc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+int zgemm_small_kernel_b0_cn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_ct(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_cr(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+int zgemm_small_kernel_b0_cc(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+#endif
+
 int cgemm_kernel_n(BLASLONG, BLASLONG, BLASLONG, float,  float,  float  *, float  *, float  *, BLASLONG);
 int cgemm_kernel_l(BLASLONG, BLASLONG, BLASLONG, float,  float,  float  *, float  *, float  *, BLASLONG);
 int cgemm_kernel_r(BLASLONG, BLASLONG, BLASLONG, float,  float,  float  *, float  *, float  *, BLASLONG);
--- a/common_loongarch64.h
+++ b/common_loongarch64.h
@ -0,0 +1,199 @@
+/*****************************************************************************
+Copyright (c) 2011-2020, 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 COPYRIGHT OWNER 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.
+**********************************************************************************/
+
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* 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.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``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 UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN 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.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#ifndef COMMON_LOONGARCH64
+#define COMMON_LOONGARCH64
+
+#define MB  __sync_synchronize()
+#define WMB __sync_synchronize()
+#define RMB __sync_synchronize()
+
+#define INLINE inline
+
+#ifndef ASSEMBLER
+
+static inline int blas_quickdivide(blasint x, blasint y){
+  return x / y;
+}
+
+#ifdef DOUBLE
+#define GET_IMAGE(res)  __asm__ __volatile__("fmov.d %0, $f2" : "=f"(res)  : : "memory")
+#else
+#define GET_IMAGE(res)  __asm__ __volatile__("fmov.s %0, $f2" : "=f"(res)  : : "memory")
+#endif
+
+#define GET_IMAGE_CANCEL
+
+#else
+
+#ifdef DOUBLE
+#define LD      fld.d
+#define ST      fst.d
+#define MADD    fmadd.d
+#define NMADD   fnmadd.d
+#define MSUB    fmsub.d
+#define NMSUB   fnmsub.d
+#define ADD     fadd.d
+#define SUB     fsub.d
+#define MUL     fmul.d
+#define MOV     fmov.d
+#define CMOVT   fsel
+#define MTC     movgr2fr.d
+#define FABS    fabs.d
+#define CMPEQ   fcmp.ceq.d
+#define CMPLE   fcmp.cle.d
+#define CMPLT   fcmp.clt.d
+#define NEG     fneg.d
+#else
+#define LD      fld.s
+#define ST      fst.s
+#define MADD    fmadd.s
+#define NMADD   fnmadd.s
+#define MSUB    fmsub.s
+#define NMSUB   fnmsub.s
+#define ADD     fadd.s
+#define SUB     fsub.s
+#define MUL     fmul.s
+#define MOV     fmov.s
+#define CMOVT   fsel
+#define MTC     movgr2fr.w
+#define FABS    fabs.s
+#define CMPEQ   fcmp.ceq.s
+#define CMPLE   fcmp.cle.s
+#define CMPLT   fcmp.clt.s
+#define NEG     fneg.s
+#endif /* defined(DOUBLE) */
+
+#if defined(__64BIT__) && defined(USE64BITINT)
+#define LDINT   ld.d
+#define LDARG   ld.d
+#define SDARG   st.d
+#elif defined(__64BIT__) && !defined(USE64BITINT)
+#define LDINT   ld.w
+#define LDARG   ld.d
+#define SDARG   st.d
+#else
+#define LDINT   ld.w
+#define LDARG   ld.w
+#define SDARG   st.w
+#endif
+
+
+#ifndef F_INTERFACE
+#define REALNAME ASMNAME
+#else
+#define REALNAME ASMFNAME
+#endif /* defined(F_INTERFACE) */
+
+#if defined(ASSEMBLER) && !defined(NEEDPARAM)
+
+#define PROLOGUE \
+    .text ;\
+    .align 5 ;\
+    .globl  REALNAME ;\
+    .type   REALNAME, @function ;\
+REALNAME: ;\
+
+#if defined(__linux__) && defined(__ELF__)
+#define GNUSTACK .section .note.GNU-stack,"",@progbits
+#else
+#define GNUSTACK
+#endif /* defined(__linux__) && defined(__ELF__) */
+
+#define EPILOGUE      \
+    .end    REALNAME ;\
+    GNUSTACK
+
+#define PROFCODE
+
+#define MOVT(dst, src, cc)  \
+    bceqz  cc,   1f;        \
+    add.d dst,  src,  $r0;  \
+    1:
+
+#endif /* defined(ASSEMBLER) && !defined(NEEDPARAM) */
+
+#endif /* defined(ASSEMBLER) */
+
+#define SEEK_ADDRESS
+
+#define BUFFER_SIZE     ( 32 << 20)
+
+#define PAGESIZE        (16UL << 10)
+#define FIXED_PAGESIZE  (16UL << 10)
+#define HUGE_PAGESIZE   ( 2 << 20)
+
+#define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
+
+#ifndef MAP_ANONYMOUS
+#define MAP_ANONYMOUS MAP_ANON
+#endif
+
+#endif
--- a/common_macro.h
+++ b/common_macro.h
@ -644,6 +644,17 @@

 #define GEADD_K                 DGEADD_K

+#define GEMM_SMALL_MATRIX_PERMIT	DGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    DGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    DGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_TN    DGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    DGEMM_SMALL_KERNEL_TT
+#define GEMM_SMALL_KERNEL_B0_NN    DGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    DGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_TN    DGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    DGEMM_SMALL_KERNEL_B0_TT
+
 #elif defined(BFLOAT16)

 #define D_TO_BF16_K     SBDTOBF16_K
@ -931,6 +942,18 @@

 #define GEADD_K 		SGEADD_K

+#define GEMM_SMALL_MATRIX_PERMIT	SBGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    SBGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    SBGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_TN    SBGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    SBGEMM_SMALL_KERNEL_TT
+
+#define GEMM_SMALL_KERNEL_B0_NN    SBGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    SBGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_TN    SBGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    SBGEMM_SMALL_KERNEL_B0_TT
+
 #endif

 #else
@ -1236,6 +1259,19 @@
 #define IMATCOPY_K_RT		SIMATCOPY_K_RT

 #define GEADD_K 		SGEADD_K
+
+#define GEMM_SMALL_MATRIX_PERMIT	SGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    SGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    SGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_TN    SGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    SGEMM_SMALL_KERNEL_TT
+
+#define GEMM_SMALL_KERNEL_B0_NN    SGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    SGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_TN    SGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    SGEMM_SMALL_KERNEL_B0_TT
+
 #endif
 #else
 #ifdef XDOUBLE
@ -2063,6 +2099,48 @@

 #define GEADD_K                 ZGEADD_K

+#define GEMM_SMALL_MATRIX_PERMIT	ZGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    ZGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    ZGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_NR    ZGEMM_SMALL_KERNEL_NR
+#define GEMM_SMALL_KERNEL_NC    ZGEMM_SMALL_KERNEL_NC
+
+#define GEMM_SMALL_KERNEL_TN    ZGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    ZGEMM_SMALL_KERNEL_TT
+#define GEMM_SMALL_KERNEL_TR    ZGEMM_SMALL_KERNEL_TR
+#define GEMM_SMALL_KERNEL_TC    ZGEMM_SMALL_KERNEL_TC
+
+#define GEMM_SMALL_KERNEL_RN    ZGEMM_SMALL_KERNEL_RN
+#define GEMM_SMALL_KERNEL_RT    ZGEMM_SMALL_KERNEL_RT
+#define GEMM_SMALL_KERNEL_RR    ZGEMM_SMALL_KERNEL_RR
+#define GEMM_SMALL_KERNEL_RC    ZGEMM_SMALL_KERNEL_RC
+
+#define GEMM_SMALL_KERNEL_CN    ZGEMM_SMALL_KERNEL_CN
+#define GEMM_SMALL_KERNEL_CT    ZGEMM_SMALL_KERNEL_CT
+#define GEMM_SMALL_KERNEL_CR    ZGEMM_SMALL_KERNEL_CR
+#define GEMM_SMALL_KERNEL_CC    ZGEMM_SMALL_KERNEL_CC
+
+#define GEMM_SMALL_KERNEL_B0_NN    ZGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    ZGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_NR    ZGEMM_SMALL_KERNEL_B0_NR
+#define GEMM_SMALL_KERNEL_B0_NC    ZGEMM_SMALL_KERNEL_B0_NC
+
+#define GEMM_SMALL_KERNEL_B0_TN    ZGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    ZGEMM_SMALL_KERNEL_B0_TT
+#define GEMM_SMALL_KERNEL_B0_TR    ZGEMM_SMALL_KERNEL_B0_TR
+#define GEMM_SMALL_KERNEL_B0_TC    ZGEMM_SMALL_KERNEL_B0_TC
+
+#define GEMM_SMALL_KERNEL_B0_RN    ZGEMM_SMALL_KERNEL_B0_RN
+#define GEMM_SMALL_KERNEL_B0_RT    ZGEMM_SMALL_KERNEL_B0_RT
+#define GEMM_SMALL_KERNEL_B0_RR    ZGEMM_SMALL_KERNEL_B0_RR
+#define GEMM_SMALL_KERNEL_B0_RC    ZGEMM_SMALL_KERNEL_B0_RC
+
+#define GEMM_SMALL_KERNEL_B0_CN    ZGEMM_SMALL_KERNEL_B0_CN
+#define GEMM_SMALL_KERNEL_B0_CT    ZGEMM_SMALL_KERNEL_B0_CT
+#define GEMM_SMALL_KERNEL_B0_CR    ZGEMM_SMALL_KERNEL_B0_CR
+#define GEMM_SMALL_KERNEL_B0_CC    ZGEMM_SMALL_KERNEL_B0_CC
+
 #else

 #define	AMAX_K			CAMAX_K
@ -2486,11 +2564,54 @@

 #define GEADD_K                 CGEADD_K

+#define GEMM_SMALL_MATRIX_PERMIT	CGEMM_SMALL_MATRIX_PERMIT
+
+#define GEMM_SMALL_KERNEL_NN    CGEMM_SMALL_KERNEL_NN
+#define GEMM_SMALL_KERNEL_NT    CGEMM_SMALL_KERNEL_NT
+#define GEMM_SMALL_KERNEL_NR    CGEMM_SMALL_KERNEL_NR
+#define GEMM_SMALL_KERNEL_NC    CGEMM_SMALL_KERNEL_NC
+
+#define GEMM_SMALL_KERNEL_TN    CGEMM_SMALL_KERNEL_TN
+#define GEMM_SMALL_KERNEL_TT    CGEMM_SMALL_KERNEL_TT
+#define GEMM_SMALL_KERNEL_TR    CGEMM_SMALL_KERNEL_TR
+#define GEMM_SMALL_KERNEL_TC    CGEMM_SMALL_KERNEL_TC
+
+#define GEMM_SMALL_KERNEL_RN    CGEMM_SMALL_KERNEL_RN
+#define GEMM_SMALL_KERNEL_RT    CGEMM_SMALL_KERNEL_RT
+#define GEMM_SMALL_KERNEL_RR    CGEMM_SMALL_KERNEL_RR
+#define GEMM_SMALL_KERNEL_RC    CGEMM_SMALL_KERNEL_RC
+
+#define GEMM_SMALL_KERNEL_CN    CGEMM_SMALL_KERNEL_CN
+#define GEMM_SMALL_KERNEL_CT    CGEMM_SMALL_KERNEL_CT
+#define GEMM_SMALL_KERNEL_CR    CGEMM_SMALL_KERNEL_CR
+#define GEMM_SMALL_KERNEL_CC    CGEMM_SMALL_KERNEL_CC
+
+#define GEMM_SMALL_KERNEL_B0_NN    CGEMM_SMALL_KERNEL_B0_NN
+#define GEMM_SMALL_KERNEL_B0_NT    CGEMM_SMALL_KERNEL_B0_NT
+#define GEMM_SMALL_KERNEL_B0_NR    CGEMM_SMALL_KERNEL_B0_NR
+#define GEMM_SMALL_KERNEL_B0_NC    CGEMM_SMALL_KERNEL_B0_NC
+
+#define GEMM_SMALL_KERNEL_B0_TN    CGEMM_SMALL_KERNEL_B0_TN
+#define GEMM_SMALL_KERNEL_B0_TT    CGEMM_SMALL_KERNEL_B0_TT
+#define GEMM_SMALL_KERNEL_B0_TR    CGEMM_SMALL_KERNEL_B0_TR
+#define GEMM_SMALL_KERNEL_B0_TC    CGEMM_SMALL_KERNEL_B0_TC
+
+#define GEMM_SMALL_KERNEL_B0_RN    CGEMM_SMALL_KERNEL_B0_RN
+#define GEMM_SMALL_KERNEL_B0_RT    CGEMM_SMALL_KERNEL_B0_RT
+#define GEMM_SMALL_KERNEL_B0_RR    CGEMM_SMALL_KERNEL_B0_RR
+#define GEMM_SMALL_KERNEL_B0_RC    CGEMM_SMALL_KERNEL_B0_RC
+
+#define GEMM_SMALL_KERNEL_B0_CN    CGEMM_SMALL_KERNEL_B0_CN
+#define GEMM_SMALL_KERNEL_B0_CT    CGEMM_SMALL_KERNEL_B0_CT
+#define GEMM_SMALL_KERNEL_B0_CR    CGEMM_SMALL_KERNEL_B0_CR
+#define GEMM_SMALL_KERNEL_B0_CC    CGEMM_SMALL_KERNEL_B0_CC
+
 #endif
 #endif

 #ifndef ASSEMBLER
-#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64) || defined(ARCH_ARM64)
+#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64) || defined(ARCH_ARM64)\
+|| defined(ARCH_LOONGARCH64)
 extern BLASLONG gemm_offset_a;
 extern BLASLONG gemm_offset_b;
 extern BLASLONG sbgemm_p;
--- a/common_param.h
+++ b/common_param.h
@ -145,6 +145,19 @@ BLASLONG (*isbmin_k) (BLASLONG, float *, BLASLONG);
  int	 (*sbneg_tcopy)   (BLASLONG, BLASLONG, float *, BLASLONG, float *);
  int    (*sblaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);

+#ifdef SMALL_MATRIX_OPT
+  int    (*sbgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float beta);
+
+  int    (*sbgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+
+  int    (*sbgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sbgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, bfloat16 * A, BLASLONG lda, float alpha, bfloat16 * B, BLASLONG ldb, float * C, BLASLONG ldc);
+#endif
 #endif

 #if defined(BUILD_SINGLE) || defined(BUILD_COMPLEX)
@ -207,6 +220,20 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
  int    (*sgemm_otcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
 #endif
 #ifdef BUILD_SINGLE  
+#ifdef SMALL_MATRIX_OPT
+  int    (*sgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float beta);
+
+  int    (*sgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
+
+  int    (*sgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+  int    (*sgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float * C, BLASLONG ldc);
+#endif
+
  int    (*strsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
  int    (*strsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
  int    (*strsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
@ -314,6 +341,19 @@ BLASLONG (*idmin_k) (BLASLONG, double *, BLASLONG);
  int    (*dgemm_otcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
 #endif
 #ifdef BUILD_DOUBLE
+#ifdef SMALL_MATRIX_OPT
+  int    (*dgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, double alpha, double beta);
+
+  int    (*dgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
+
+  int    (*dgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+  int    (*dgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double * C, BLASLONG ldc);
+#endif
  int    (*dtrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
  int    (*dtrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
  int    (*dtrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG, BLASLONG);
@ -513,6 +553,50 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
  int    (*cgemm_oncopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
  int    (*cgemm_otcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);

+#ifdef SMALL_MATRIX_OPT
+  int    (*cgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, float alpha0, float alpha1, float beta0, float beta1);
+
+  int    (*cgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_nr    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_nc    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_tr    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_tc    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_rn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_rt    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_rr    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_rc    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_cn    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_ct    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_cr    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_cc    )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb, float beta0, float beta1, float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_nr )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_nc )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_tr )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_tc )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_b0_rn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_rt )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_rr )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_rc )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+
+  int    (*cgemm_small_kernel_b0_cn )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_ct )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_cr )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+  int    (*cgemm_small_kernel_b0_cc )(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha0, float alpha1, float * B, BLASLONG ldb,  float * C, BLASLONG ldc);
+#endif
+
  int    (*ctrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
  int    (*ctrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
  int    (*ctrsm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG, BLASLONG);
@ -679,6 +763,50 @@ BLASLONG (*izamin_k)(BLASLONG, double *, BLASLONG);
  int    (*zgemm_oncopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);
  int    (*zgemm_otcopy   )(BLASLONG, BLASLONG, double *, BLASLONG, double *);

+#ifdef SMALL_MATRIX_OPT
+  int    (*zgemm_small_matrix_permit)(int transa, int transb, BLASLONG m, BLASLONG n, BLASLONG k, double alpha0, double alpha1, double beta0, double beta1);
+
+  int    (*zgemm_small_kernel_nn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_nt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_nr    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_nc    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_tn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_tt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_tr    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_tc    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_rn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_rt    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_rr    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_rc    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_cn    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_ct    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_cr    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_cc    )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb, double beta0, double beta1, double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_b0_nn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_nt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_nr )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_nc )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_b0_tn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_tt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_tr )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_tc )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_b0_rn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_rt )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_rr )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_rc )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+
+  int    (*zgemm_small_kernel_b0_cn )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_ct )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_cr )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+  int    (*zgemm_small_kernel_b0_cc )(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha0, double alpha1, double * B, BLASLONG ldb,  double * C, BLASLONG ldc);
+#endif
+
  int    (*ztrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
  int    (*ztrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
  int    (*ztrsm_kernel_LR)(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG, BLASLONG);
@ -1069,6 +1197,8 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);

 extern gotoblas_t *gotoblas;

+#define FUNC_OFFSET(func)	(size_t)(&((gotoblas_t *)NULL)->func)
+
 #define DTB_ENTRIES  gotoblas -> dtb_entries
 #define GEMM_OFFSET_A	gotoblas -> offsetA
 #define GEMM_OFFSET_B	gotoblas -> offsetB
@ -1174,6 +1304,8 @@ extern gotoblas_t *gotoblas;

 #else

+#define FUNC_OFFSET(func)	(size_t)(func)
+
 #define DTB_ENTRIES  DTB_DEFAULT_ENTRIES

 #define GEMM_OFFSET_A	GEMM_DEFAULT_OFFSET_A
--- a/common_s.h
+++ b/common_s.h
@ -164,6 +164,8 @@

 #define SGEADD_K                sgeadd_k 

+#define SGEMM_SMALL_MATRIX_PERMIT	sgemm_small_matrix_permit
+
 #else

 #define	SAMAX_K			gotoblas -> samax_k
@ -299,8 +301,21 @@

 #define SGEADD_K                gotoblas -> sgeadd_k 

+#define SGEMM_SMALL_MATRIX_PERMIT	gotoblas -> sgemm_small_matrix_permit
+
 #endif

+#define SGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(sgemm_small_kernel_nn)
+#define SGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(sgemm_small_kernel_nt)
+#define SGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(sgemm_small_kernel_tn)
+#define SGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(sgemm_small_kernel_tt)
+
+#define SGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(sgemm_small_kernel_b0_nn)
+#define SGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(sgemm_small_kernel_b0_nt)
+#define SGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(sgemm_small_kernel_b0_tn)
+#define SGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(sgemm_small_kernel_b0_tt)
+
+
 #define	SGEMM_NN		sgemm_nn
 #define	SGEMM_CN		sgemm_tn
 #define	SGEMM_TN		sgemm_tn
--- a/common_sb.h
+++ b/common_sb.h
@ -24,6 +24,7 @@
 #define	SBGEMM_BETA		sbgemm_beta
 #define SBGEMM_KERNEL            sbgemm_kernel

+#define SBGEMM_SMALL_MATRIX_PERMIT	sbgemm_small_matrix_permit
 #else

 #define SBDOT_K             gotoblas -> sbdot_k
@ -41,8 +42,19 @@
 #define	SBGEMM_BETA		gotoblas -> sbgemm_beta
 #define	SBGEMM_KERNEL		gotoblas -> sbgemm_kernel

+#define SBGEMM_SMALL_MATRIX_PERMIT	gotoblas -> sbgemm_small_matrix_permit
 #endif

+#define SBGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(sbgemm_small_kernel_nn)
+#define SBGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(sbgemm_small_kernel_nt)
+#define SBGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(sbgemm_small_kernel_tn)
+#define SBGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(sbgemm_small_kernel_tt)
+
+#define SBGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(sbgemm_small_kernel_b0_nn)
+#define SBGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(sbgemm_small_kernel_b0_nt)
+#define SBGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(sbgemm_small_kernel_b0_tn)
+#define SBGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(sbgemm_small_kernel_b0_tt)
+
 #define	SBGEMM_NN		sbgemm_nn
 #define	SBGEMM_CN		sbgemm_tn
 #define	SBGEMM_TN		sbgemm_tn
--- a/common_z.h
+++ b/common_z.h
@ -232,6 +232,8 @@

 #define ZGEADD_K                zgeadd_k 

+#define ZGEMM_SMALL_MATRIX_PERMIT	zgemm_small_matrix_permit
+
 #else

 #define	ZAMAX_K			gotoblas -> zamax_k
@ -426,8 +428,51 @@

 #define ZGEADD_K                gotoblas -> zgeadd_k

+#define ZGEMM_SMALL_MATRIX_PERMIT	gotoblas -> zgemm_small_matrix_permit
+
 #endif

+#define ZGEMM_SMALL_KERNEL_NN		FUNC_OFFSET(zgemm_small_kernel_nn)
+#define ZGEMM_SMALL_KERNEL_NT		FUNC_OFFSET(zgemm_small_kernel_nt)
+#define ZGEMM_SMALL_KERNEL_NR		FUNC_OFFSET(zgemm_small_kernel_nr)
+#define ZGEMM_SMALL_KERNEL_NC		FUNC_OFFSET(zgemm_small_kernel_nc)
+
+#define ZGEMM_SMALL_KERNEL_TN		FUNC_OFFSET(zgemm_small_kernel_tn)
+#define ZGEMM_SMALL_KERNEL_TT		FUNC_OFFSET(zgemm_small_kernel_tt)
+#define ZGEMM_SMALL_KERNEL_TR		FUNC_OFFSET(zgemm_small_kernel_tr)
+#define ZGEMM_SMALL_KERNEL_TC		FUNC_OFFSET(zgemm_small_kernel_tc)
+
+#define ZGEMM_SMALL_KERNEL_RN		FUNC_OFFSET(zgemm_small_kernel_rn)
+#define ZGEMM_SMALL_KERNEL_RT		FUNC_OFFSET(zgemm_small_kernel_rt)
+#define ZGEMM_SMALL_KERNEL_RR		FUNC_OFFSET(zgemm_small_kernel_rr)
+#define ZGEMM_SMALL_KERNEL_RC		FUNC_OFFSET(zgemm_small_kernel_rc)
+
+#define ZGEMM_SMALL_KERNEL_CN		FUNC_OFFSET(zgemm_small_kernel_cn)
+#define ZGEMM_SMALL_KERNEL_CT		FUNC_OFFSET(zgemm_small_kernel_ct)
+#define ZGEMM_SMALL_KERNEL_CR		FUNC_OFFSET(zgemm_small_kernel_cr)
+#define ZGEMM_SMALL_KERNEL_CC		FUNC_OFFSET(zgemm_small_kernel_cc)
+
+#define ZGEMM_SMALL_KERNEL_B0_NN	FUNC_OFFSET(zgemm_small_kernel_b0_nn)
+#define ZGEMM_SMALL_KERNEL_B0_NT	FUNC_OFFSET(zgemm_small_kernel_b0_nt)
+#define ZGEMM_SMALL_KERNEL_B0_NR	FUNC_OFFSET(zgemm_small_kernel_b0_nr)
+#define ZGEMM_SMALL_KERNEL_B0_NC	FUNC_OFFSET(zgemm_small_kernel_b0_nc)
+
+#define ZGEMM_SMALL_KERNEL_B0_TN	FUNC_OFFSET(zgemm_small_kernel_b0_tn)
+#define ZGEMM_SMALL_KERNEL_B0_TT	FUNC_OFFSET(zgemm_small_kernel_b0_tt)
+#define ZGEMM_SMALL_KERNEL_B0_TR	FUNC_OFFSET(zgemm_small_kernel_b0_tr)
+#define ZGEMM_SMALL_KERNEL_B0_TC	FUNC_OFFSET(zgemm_small_kernel_b0_tc)
+
+#define ZGEMM_SMALL_KERNEL_B0_RN	FUNC_OFFSET(zgemm_small_kernel_b0_rn)
+#define ZGEMM_SMALL_KERNEL_B0_RT	FUNC_OFFSET(zgemm_small_kernel_b0_rt)
+#define ZGEMM_SMALL_KERNEL_B0_RR	FUNC_OFFSET(zgemm_small_kernel_b0_rr)
+#define ZGEMM_SMALL_KERNEL_B0_RC	FUNC_OFFSET(zgemm_small_kernel_b0_rc)
+
+#define ZGEMM_SMALL_KERNEL_B0_CN	FUNC_OFFSET(zgemm_small_kernel_b0_cn)
+#define ZGEMM_SMALL_KERNEL_B0_CT	FUNC_OFFSET(zgemm_small_kernel_b0_ct)
+#define ZGEMM_SMALL_KERNEL_B0_CR	FUNC_OFFSET(zgemm_small_kernel_b0_cr)
+#define ZGEMM_SMALL_KERNEL_B0_CC	FUNC_OFFSET(zgemm_small_kernel_b0_cc)
+
+
 #define	ZGEMM_NN		zgemm_nn
 #define	ZGEMM_CN		zgemm_cn
 #define	ZGEMM_TN		zgemm_tn
--- a/cpuid_loongarch64.c
+++ b/cpuid_loongarch64.c
@ -0,0 +1,110 @@
+/*****************************************************************************
+Copyright (c) 2011-2020, 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 COPYRIGHT OWNER 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 <stdint.h>
+
+#define CPU_UNKNOWN     0
+#define CPU_LOONGSON3R5 1
+
+#define LOONGARCH_CFG2  0x02
+#define LOONGARCH_LASX  1<<7
+
+static char *cpuname[] = {
+  "UNKNOWN",
+  "LOONGSON3R5"
+};
+
+int detect(void) {
+    uint32_t reg = 0;
+
+    __asm__ volatile (
+        "cpucfg %0, %1 \n\t"
+        : "+&r"(reg)
+        : "r"(LOONGARCH_CFG2)
+    );
+
+    if (reg & LOONGARCH_LASX)
+        return CPU_LOONGSON3R5;
+    else
+        return CPU_UNKNOWN;
+}
+
+char *get_corename(void) {
+  return cpuname[detect()];
+}
+
+void get_architecture(void) {
+  printf("LOONGARCH64");
+}
+
+void get_subarchitecture(void) {
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("LOONGSON3R5");
+  } else {
+    printf("UNKNOWN");
+  }
+}
+
+void get_subdirname(void) {
+  printf("loongarch64");
+}
+
+void get_cpuconfig(void) {
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("#define LOONGSON3R5\n");
+    printf("#define L1_DATA_SIZE 65536\n");
+    printf("#define L1_DATA_LINESIZE 64\n");
+    printf("#define L2_SIZE 1048576\n");
+    printf("#define L2_LINESIZE 64\n");
+    printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 16\n");
+  } else {
+    printf("#define LOONGSON3R5\n");
+    printf("#define L1_DATA_SIZE 65536\n");
+    printf("#define L1_DATA_LINESIZE 64\n");
+    printf("#define L2_SIZE 1048576\n");
+    printf("#define L2_LINESIZE 64\n");
+    printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 16\n");
+  }
+}
+
+void get_libname(void){
+  if (detect() == CPU_LOONGSON3R5) {
+    printf("loongson3r5\n");
+  } else {
+    printf("loongarch64\n");
+  }
+}
--- a/ctest.c
+++ b/ctest.c
@ -84,7 +84,7 @@ OS_AIX
 OS_OSF
 #endif

-#if defined(__WIN32) || defined(__WIN64) || defined(__WINNT)
+#if defined(__WIN32) || defined(__WIN64) || defined(_WIN32) || defined(_WIN64) || defined(__WINNT)
 OS_WINNT
 #endif

@ -141,7 +141,7 @@ ARCH_SPARC
 ARCH_IA64
 #endif

-#if defined(__LP64) || defined(__LP64__) || defined(__ptr64) || defined(__x86_64__) || defined(__amd64__) || defined(__64BIT__)
+#if defined(__LP64) || defined(__LP64__) || defined(__ptr64) || defined(__x86_64__) || defined(__amd64__) || defined(__64BIT__) || defined(__aarch64__)
 BINARY_64
 #endif

@ -157,6 +157,10 @@ ARCH_ARM64
 ARCH_RISCV64
 #endif

+#ifdef __loongarch64
+ARCH_LOONGARCH64
+#endif
+
 #if (defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L)
 HAVE_C11
 #endif
--- a/ctest/CMakeLists.txt
+++ b/ctest/CMakeLists.txt
@ -4,6 +4,9 @@ include_directories(${PROJECT_BINARY_DIR})
 enable_language(Fortran)

 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DADD${BU} -DCBLAS")
+if (CMAKE_Fortran_COMPILER_ID STREQUAL GNU)
+	set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fno-tree-vectorize")
+endif()

 if(WIN32)
 FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.ps1
--- a/ctest/Makefile
+++ b/ctest/Makefile
@ -6,6 +6,9 @@ TOPDIR = ..
 include $(TOPDIR)/Makefile.system

 override CFLAGS += -DADD$(BU) -DCBLAS
+ifeq ($(F_COMPILER),GFORTRAN)
+	override FFLAGS += -fno-tree-vectorize
+endif
 override TARGET_ARCH=
 override TARGET_MACH=

--- a/driver/level2/CMakeLists.txt
+++ b/driver/level2/CMakeLists.txt
@ -81,6 +81,7 @@ foreach (float_type ${FLOAT_TYPES})
    GenerateNamedObjects("gbmv_thread.c" "TRANSA" "gbmv_thread_t" false "" "" false ${float_type})
  endif ()

+# special defines for complex
  if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")

    foreach (u_source ${U_SOURCES})
@ -197,6 +198,13 @@ foreach (float_type ${FLOAT_TYPES})
  endif ()
 endforeach ()

+if (BUILD_BFLOAT16)
+  if (USE_THREAD)
+    GenerateNamedObjects("sbgemv_thread.c" "" "gemv_thread_n" false "" "" false "BFLOAT16")
+    GenerateNamedObjects("sbgemv_thread.c" "TRANSA" "gemv_thread_t" false "" "" false "BFLOAT16")
+  endif ()
+endif ()
+
 if ( BUILD_COMPLEX AND NOT  BUILD_SINGLE)
  if (USE_THREAD)
 	  GenerateNamedObjects("gemv_thread.c" "" "gemv_thread_n" false "" "" false "SINGLE")
--- a/driver/level3/CMakeLists.txt
+++ b/driver/level3/CMakeLists.txt
@ -12,6 +12,12 @@ foreach (GEMM_DEFINE ${GEMM_DEFINES})
  if (USE_THREAD AND NOT USE_SIMPLE_THREADED_LEVEL3)
    GenerateNamedObjects("gemm.c" "${GEMM_DEFINE};THREADED_LEVEL3" "gemm_thread_${GEMM_DEFINE_LC}" 0)
  endif ()
+  if (BUILD_BFLOAT16)
+    GenerateNamedObjects("gemm.c" "${GEMM_DEFINE}" "gemm_${GEMM_DEFINE_LC}" 0 "" "" false "BFLOAT16")
+    if (USE_THREAD AND NOT USE_SIMPLE_THREADED_LEVEL3)
+      GenerateNamedObjects("gemm.c" "${GEMM_DEFINE};THREADED_LEVEL3" "gemm_thread_${GEMM_DEFINE_LC}" 0 "" "" false "BFLOAT16")
+    endif ()
+  endif ()
 endforeach ()

 if ( BUILD_COMPLEX16 AND NOT  BUILD_DOUBLE)
--- a/driver/others/dynamic_power.c
+++ b/driver/others/dynamic_power.c
@ -6,10 +6,6 @@ extern gotoblas_t gotoblas_POWER8;
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 extern gotoblas_t gotoblas_POWER9;
 #endif
-//#if (!defined __GNUC__) || ( __GNUC__ >= 11) \
-//     || (__GNUC__ == 10 && __GNUC_MINOR__ >= 2)
-//#define HAVE_P10_SUPPORT 1
-//#endif
 #ifdef HAVE_P10_SUPPORT
 extern gotoblas_t gotoblas_POWER10;
 #endif
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@ -73,6 +73,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 #include "common.h"

+#ifndef likely
+#ifdef __GNUC__
+#define likely(x) __builtin_expect(!!(x), 1)
+#define unlikely(x) __builtin_expect(!!(x), 0)
+#else
+#define likely(x) (x)
+#define unlikely(x) (x)
+#endif
+#endif
+
 #if defined(USE_TLS) && defined(SMP)
 #define COMPILE_TLS

@ -428,7 +438,7 @@ extern int openblas_goto_num_threads_env();
 extern int openblas_omp_num_threads_env();

 int blas_get_cpu_number(void){
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
  int max_num;
 #endif
  int blas_goto_num   = 0;
@ -436,7 +446,7 @@ int blas_get_cpu_number(void){

  if (blas_num_threads) return blas_num_threads;

-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
  max_num = get_num_procs();
 #endif

@ -460,7 +470,7 @@ int blas_get_cpu_number(void){
  else if (blas_omp_num > 0) blas_num_threads = blas_omp_num;
  else blas_num_threads = MAX_CPU_NUMBER;

-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
  if (blas_num_threads > max_num) blas_num_threads = max_num;
 #endif

@ -1291,7 +1301,12 @@ UNLOCK_COMMAND(&alloc_lock);
  return (void *)(((char *)alloc_info) + sizeof(struct alloc_t));

 error:
-  printf("OpenBLAS : Program will terminate because you tried to allocate too many memory regions.\n");
+  printf("OpenBLAS : Program will terminate because you tried to allocate too many TLS memory regions.\n");
+  printf("This library was built to support a maximum of %d threads - either rebuild OpenBLAS\n", NUM_BUFFERS);
+  printf("with a larger NUM_THREADS value or set the environment variable OPENBLAS_NUM_THREADS to\n");
+  printf("a sufficiently small number. This error typically occurs when the software that relies on\n");
+  printf("OpenBLAS calls BLAS functions from many threads in parallel, or when your computer has more\n");
+  printf("cpu cores than what OpenBLAS was configured to handle.\n"); 

  return NULL;
 }
@ -1979,7 +1994,7 @@ extern int openblas_goto_num_threads_env();
 extern int openblas_omp_num_threads_env();

 int blas_get_cpu_number(void){
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
  int max_num;
 #endif
  int blas_goto_num   = 0;
@ -1987,7 +2002,7 @@ int blas_get_cpu_number(void){

  if (blas_num_threads) return blas_num_threads;

-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
  max_num = get_num_procs();
 #endif

@ -2011,7 +2026,7 @@ int blas_get_cpu_number(void){
  else if (blas_omp_num > 0) blas_num_threads = blas_omp_num;
  else blas_num_threads = MAX_CPU_NUMBER;

-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_NETBSD) || defined(OS_DRAGONFLY) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_HAIKU)
  if (blas_num_threads > max_num) blas_num_threads = max_num;
 #endif

@ -2055,6 +2070,7 @@ struct release_t {
 int hugetlb_allocated = 0;

 static struct release_t release_info[NUM_BUFFERS];
+static struct release_t *new_release_info;
 static int release_pos = 0;

 #if defined(OS_LINUX) && !defined(NO_WARMUP)
@ -2105,8 +2121,13 @@ static void *alloc_mmap(void *address){
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
    LOCK_COMMAND(&alloc_lock);
 #endif
+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].func    = alloc_mmap_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_mmap_free;
+    }
    release_pos ++;
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
    UNLOCK_COMMAND(&alloc_lock);
@ -2269,8 +2290,13 @@ static void *alloc_mmap(void *address){
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
    LOCK_COMMAND(&alloc_lock);
 #endif
+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].func    = alloc_mmap_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_mmap_free;
+    }
    release_pos ++;
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
    UNLOCK_COMMAND(&alloc_lock);
@ -2302,8 +2328,13 @@ static void *alloc_malloc(void *address){
  if (map_address == (void *)NULL) map_address = (void *)-1;

  if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].func    = alloc_malloc_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_malloc_free;
+    }
    release_pos ++;
  }

@ -2336,8 +2367,13 @@ static void *alloc_qalloc(void *address){
  if (map_address == (void *)NULL) map_address = (void *)-1;

  if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].func    = alloc_qalloc_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_qalloc_free;
+    }
    release_pos ++;
  }

@ -2365,8 +2401,13 @@ static void *alloc_windows(void *address){
  if (map_address == (void *)NULL) map_address = (void *)-1;

  if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].func    = alloc_windows_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_windows_free;
+    }
    release_pos ++;
  }

@ -2409,9 +2450,15 @@ static void *alloc_devicedirver(void *address){
                     fd, 0);

  if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].attr    = fd;
    release_info[release_pos].func    = alloc_devicedirver_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].attr    = fd;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_devicedirver_free;
+    }
    release_pos ++;
  }

@ -2445,9 +2492,15 @@ static void *alloc_shm(void *address){

    shmctl(shmid, IPC_RMID, 0);

+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].attr    = shmid;
    release_info[release_pos].func    = alloc_shm_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].attr    = shmid;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_shm_free;
+    }
    release_pos ++;
  }

@ -2551,8 +2604,13 @@ static void *alloc_hugetlb(void *address){
 #endif

  if (map_address != (void *)-1){
+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].func    = alloc_hugetlb_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_hugetlb_free;
+    }
    release_pos ++;
  }

@ -2599,9 +2657,15 @@ static void *alloc_hugetlbfile(void *address){
                     fd, 0);

  if (map_address != (void *)-1) {
+    if (likely(release_pos < NUM_BUFFERS)) {
    release_info[release_pos].address = map_address;
    release_info[release_pos].attr    = fd;
    release_info[release_pos].func    = alloc_hugetlbfile_free;
+    } else {
+    new_release_info[release_pos-NUM_BUFFERS].address = map_address;
+    new_release_info[release_pos-NUM_BUFFERS].attr    = fd;
+    new_release_info[release_pos-NUM_BUFFERS].func    = alloc_hugetlbfile_free;
+    }
    release_pos ++;
  }

@ -2631,8 +2695,25 @@ static volatile struct {

 } memory[NUM_BUFFERS];

-static int memory_initialized = 0;
+struct newmemstruct 
+{
+  BLASULONG lock;
+  void *addr;
+#if defined(WHEREAMI) && !defined(USE_OPENMP)
+  int   pos;
+#endif
+  int used;
+#ifndef __64BIT__
+  char dummy[48];
+#else
+  char dummy[40];
+#endif

+};
+static volatile struct newmemstruct *newmemory;
+
+static int memory_initialized = 0;
+static int memory_overflowed = 0;
 /*       Memory allocation routine           */
 /* procpos ... indicates where it comes from */
 /*                0 : Level 3 functions      */
@ -2641,6 +2722,8 @@ static int memory_initialized = 0;

 void *blas_memory_alloc(int procpos){

+  int i;
+  
  int position;
 #if defined(WHEREAMI) && !defined(USE_OPENMP)
  int mypos = 0;
@ -2774,6 +2857,29 @@ void *blas_memory_alloc(int procpos){
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
  UNLOCK_COMMAND(&alloc_lock);
 #endif
+  if (memory_overflowed) {
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+  LOCK_COMMAND(&alloc_lock);
+#endif
+  do {
+    RMB;
+#if defined(USE_OPENMP)
+    if (!newmemory[position-NUM_BUFFERS].used) {
+      blas_lock(&newmemory[position-NUM_BUFFERS].lock);
+#endif
+      if (!newmemory[position-NUM_BUFFERS].used) goto allocation2;
+
+#if defined(USE_OPENMP)
+      blas_unlock(&newmemory[position-NUM_BUFFERS].lock);
+    }
+#endif
+    position ++;
+
+  } while (position < 512+NUM_BUFFERS);
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+  UNLOCK_COMMAND(&alloc_lock);
+#endif
+} 
  goto error;

  allocation :
@ -2878,8 +2984,97 @@ void *blas_memory_alloc(int procpos){
  return (void *)memory[position].addr;

 error:
-  printf("BLAS : Program is Terminated. Because you tried to allocate too many memory regions.\n");
+ if (memory_overflowed) goto terminate;
+  fprintf(stderr,"OpenBLAS warning: precompiled NUM_THREADS exceeded, adding auxiliary array for thread metadata.\n");
+  memory_overflowed=1;
+  new_release_info = (struct release_t*) malloc(512*sizeof(struct release_t));
+  newmemory = (struct newmemstruct*) malloc(512*sizeof(struct newmemstruct));
+  for (i = 0; i < 512; i++) {
+  newmemory[i].addr   = (void *)0;
+#if defined(WHEREAMI) && !defined(USE_OPENMP)
+  newmemory[i].pos    = -1;
+#endif
+  newmemory[i].used   = 0;
+  newmemory[i].lock   = 0;
+}
+  newmemory[position-NUM_BUFFERS].used = 1;
  
+allocation2:
+  newmemory[position-NUM_BUFFERS].used = 1;
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+  UNLOCK_COMMAND(&alloc_lock);
+#else
+  blas_unlock(&newmemory[position-NUM_BUFFERS].lock);
+#endif
+    do {
+#ifdef DEBUG
+      printf("Allocation Start : %lx\n", base_address);
+#endif
+
+      map_address = (void *)-1;
+
+      func = &memoryalloc[0];
+
+      while ((func != NULL) && (map_address == (void *) -1)) {
+
+        map_address = (*func)((void *)base_address);
+
+#ifdef ALLOC_DEVICEDRIVER
+        if ((*func ==  alloc_devicedirver) && (map_address == (void *)-1)) {
+            fprintf(stderr, "OpenBLAS Warning ... Physically contiguous allocation was failed.\n");
+        }
+#endif
+
+#ifdef ALLOC_HUGETLBFILE
+        if ((*func == alloc_hugetlbfile) && (map_address == (void *)-1)) {
+#ifndef OS_WINDOWS
+            fprintf(stderr, "OpenBLAS Warning ... HugeTLB(File) allocation was failed.\n");
+#endif
+        }
+#endif
+
+#if (defined ALLOC_SHM) && (defined OS_LINUX  || defined OS_AIX  || defined __sun__  || defined OS_WINDOWS)
+        if ((*func == alloc_hugetlb) && (map_address != (void *)-1)) hugetlb_allocated = 1;
+#endif
+
+        func ++;
+      }
+
+#ifdef DEBUG
+      printf("  Success -> %08lx\n", map_address);
+#endif
+      if (((BLASLONG) map_address) == -1) base_address = 0UL;
+
+      if (base_address) base_address += BUFFER_SIZE + FIXED_PAGESIZE;
+
+    } while ((BLASLONG)map_address == -1);
+
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+    LOCK_COMMAND(&alloc_lock);
+#endif
+    newmemory[position-NUM_BUFFERS].addr = map_address;
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+    UNLOCK_COMMAND(&alloc_lock);
+#endif
+
+#ifdef DEBUG
+    printf("  Mapping Succeeded. %p(%d)\n", (void *)newmemory[position-NUM_BUFFERS].addr, position);
+#endif
+
+#if defined(WHEREAMI) && !defined(USE_OPENMP)
+
+  if (newmemory[position-NUM_BUFFERS].pos == -1) newmemory[position-NUM_BUFFERS].pos = mypos;
+
+#endif
+  return (void *)newmemory[position-NUM_BUFFERS].addr;
+
+terminate:
+  printf("OpenBLAS : Program is Terminated. Because you tried to allocate too many memory regions.\n");
+  printf("This library was built to support a maximum of %d threads - either rebuild OpenBLAS\n", NUM_BUFFERS);
+  printf("with a larger NUM_THREADS value or set the environment variable OPENBLAS_NUM_THREADS to\n");
+  printf("a sufficiently small number. This error typically occurs when the software that relies on\n");
+  printf("OpenBLAS calls BLAS functions from many threads in parallel, or when your computer has more\n");
+  printf("cpu cores than what OpenBLAS was configured to handle.\n"); 
  return NULL;
 }

@ -2898,13 +3093,28 @@ void blas_memory_free(void *free_area){
  while ((position < NUM_BUFFERS) && (memory[position].addr != free_area))
    position++;

-  if (position >= NUM_BUFFERS) goto error;
+  if (position >= NUM_BUFFERS && !memory_overflowed) goto error;

 #ifdef DEBUG
  if (memory[position].addr != free_area) goto error;
  printf("  Position : %d\n", position);
 #endif
+  if (unlikely(memory_overflowed && position >= NUM_BUFFERS)) {
+    while ((position < NUM_BUFFERS+512) && (newmemory[position-NUM_BUFFERS].addr != free_area))
+      position++;
+  // arm: ensure all writes are finished before other thread takes this memory
+  WMB;

+  newmemory[position].used = 0;
+#if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
+  UNLOCK_COMMAND(&alloc_lock);
+#endif
+
+#ifdef DEBUG
+  printf("Unmap from overflow area succeeded.\n\n");
+#endif
+  return;
+} else {
  // arm: ensure all writes are finished before other thread takes this memory
  WMB;

@ -2918,7 +3128,7 @@ void blas_memory_free(void *free_area){
 #endif

  return;
-
+}
 error:
  printf("BLAS : Bad memory unallocation! : %4d  %p\n", position,  free_area);

@ -2953,7 +3163,10 @@ void blas_shutdown(void){
  LOCK_COMMAND(&alloc_lock);

  for (pos = 0; pos < release_pos; pos ++) {
+    if (likely(pos < NUM_BUFFERS))
    release_info[pos].func(&release_info[pos]);
+    else
+    new_release_info[pos-NUM_BUFFERS].func(&new_release_info[pos-NUM_BUFFERS]);
  }

 #ifdef SEEK_ADDRESS
@ -2970,6 +3183,15 @@ void blas_shutdown(void){
 #endif
    memory[pos].lock   = 0;
  }
+  if (memory_overflowed)
+    for (pos = 0; pos < 512; pos ++){
+      newmemory[pos].addr   = (void *)0;
+      newmemory[pos].used   = 0;
+#if defined(WHEREAMI) && !defined(USE_OPENMP)
+      newmemory[pos].pos    = -1;
+#endif
+      newmemory[pos].lock   = 0;
+  }

  UNLOCK_COMMAND(&alloc_lock);

--- a/driver/others/parameter.c
+++ b/driver/others/parameter.c
@ -524,6 +524,9 @@ void blas_set_parameter(void){
  xgemm_p = ((xgemm_p + XGEMM_UNROLL_M - 1)/XGEMM_UNROLL_M) * XGEMM_UNROLL_M;
 #endif

+#ifdef BUILD_BFLOAT16
+  sbgemm_r = (((BUFFER_SIZE - ((SBGEMM_P * SBGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SBGEMM_Q *  4)) - 15) & ~15;
+#endif
  sgemm_r = (((BUFFER_SIZE - ((SGEMM_P * SGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SGEMM_Q *  4)) - 15) & ~15;
  dgemm_r = (((BUFFER_SIZE - ((DGEMM_P * DGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (DGEMM_Q *  8)) - 15) & ~15;
  cgemm_r = (((BUFFER_SIZE - ((CGEMM_P * CGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (CGEMM_Q *  8)) - 15) & ~15;
@ -629,7 +632,9 @@ void blas_set_parameter(void){
  xgemm_p =  16 * (size + 1);
 #endif

+#ifdef BUILD_BFLOAT16
  sbgemm_r = (((BUFFER_SIZE - ((SBGEMM_P * SBGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SBGEMM_Q *  4)) - 15) & ~15;
+#endif
  sgemm_r = (((BUFFER_SIZE - ((SGEMM_P * SGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SGEMM_Q *  4)) - 15) & ~15;
  dgemm_r = (((BUFFER_SIZE - ((DGEMM_P * DGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (DGEMM_Q *  8)) - 15) & ~15;
  cgemm_r = (((BUFFER_SIZE - ((CGEMM_P * CGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (CGEMM_Q *  8)) - 15) & ~15;
--- a/getarch.c
+++ b/getarch.c
@ -142,6 +142,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* #define FORCE_SICORTEX	*/
 /* #define FORCE_LOONGSON3R3	*/
 /* #define FORCE_LOONGSON3R4	*/
+/* #define FORCE_LOONGSON3R5	*/
 /* #define FORCE_I6400		*/
 /* #define FORCE_P6600		*/
 /* #define FORCE_P5600		*/
@ -312,6 +313,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define FORCE
 #define FORCE_INTEL
 #define ARCHITECTURE    "X86"
+#ifdef NO_AVX 
+#define SUBARCHITECTURE "NEHALEM"
+#define ARCHCONFIG   "-DNEHALEM " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2"
+#define LIBNAME   "nehalem"
+#define CORENAME  "NEHALEM"
+#else
 #define SUBARCHITECTURE "SANDYBRIDGE"
 #define ARCHCONFIG   "-DSANDYBRIDGE " \
 		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
@ -321,12 +332,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define LIBNAME   "sandybridge"
 #define CORENAME  "SANDYBRIDGE"
 #endif
+#endif

 #ifdef FORCE_HASWELL
 #define FORCE
 #define FORCE_INTEL
 #define ARCHITECTURE    "X86"
 #ifdef NO_AVX2
+#ifdef NO_AVX
+#define SUBARCHITECTURE "NEHALEM"
+#define ARCHCONFIG   "-DNEHALEM " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2"
+#define LIBNAME   "nehalem"
+#define CORENAME  "NEHALEM"
+#else
 #define SUBARCHITECTURE "SANDYBRIDGE"
 #define ARCHCONFIG   "-DSANDYBRIDGE " \
 		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
@ -335,6 +357,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 -DHAVE_AVX"
 #define LIBNAME   "sandybridge"
 #define CORENAME  "SANDYBRIDGE"
+#endif
 #else
 #define SUBARCHITECTURE "HASWELL"
 #define ARCHCONFIG   "-DHASWELL " \
@ -349,10 +372,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif

 #ifdef FORCE_SKYLAKEX
-#ifdef NO_AVX512
 #define FORCE
 #define FORCE_INTEL
 #define ARCHITECTURE    "X86"
+#ifdef NO_AVX512
+#ifdef NO_AVX2
+#ifdef NO_AVX
+#define SUBARCHITECTURE "NEHALEM"
+#define ARCHCONFIG   "-DNEHALEM " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2"
+#define LIBNAME   "nehalem"
+#define CORENAME  "NEHALEM"
+#else
+#define SUBARCHITECTURE "SANDYBRIDGE"
+#define ARCHCONFIG   "-DSANDYBRIDGE " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 -DHAVE_AVX"
+#define LIBNAME   "sandybridge"
+#define CORENAME  "SANDYBRIDGE"
+#endif
+#else
 #define SUBARCHITECTURE "HASWELL"
 #define ARCHCONFIG   "-DHASWELL " \
 		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
@ -362,10 +406,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
                     "-DHAVE_AVX2 -DHAVE_FMA3 -DFMA3"
 #define LIBNAME   "haswell"
 #define CORENAME  "HASWELL"
+#endif
 #else
-#define FORCE
-#define FORCE_INTEL
-#define ARCHITECTURE    "X86"
 #define SUBARCHITECTURE "SKYLAKEX"
 #define ARCHCONFIG   "-DSKYLAKEX " \
 		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
@ -379,10 +421,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif

 #ifdef FORCE_COOPERLAKE
-#ifdef NO_AVX512
 #define FORCE
 #define FORCE_INTEL
 #define ARCHITECTURE    "X86"
+#ifdef NO_AVX512
+#ifdef NO_AVX2
+#ifdef NO_AVX
+#define SUBARCHITECTURE "NEHALEM"
+#define ARCHCONFIG   "-DNEHALEM " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2"
+#define LIBNAME   "nehalem"
+#define CORENAME  "NEHALEM"
+#else
+#define SUBARCHITECTURE "SANDYBRIDGE"
+#define ARCHCONFIG   "-DSANDYBRIDGE " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 -DHAVE_AVX"
+#define LIBNAME   "sandybridge"
+#define CORENAME  "SANDYBRIDGE"
+#endif
+#else
 #define SUBARCHITECTURE "HASWELL"
 #define ARCHCONFIG   "-DHASWELL " \
                     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
@ -392,10 +455,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
                     "-DHAVE_AVX2 -DHAVE_FMA3 -DFMA3"
 #define LIBNAME   "haswell"
 #define CORENAME  "HASWELL"
+#endif
 #else
-#define FORCE
-#define FORCE_INTEL
-#define ARCHITECTURE    "X86"
 #define SUBARCHITECTURE "COOPERLAKE"
 #define ARCHCONFIG   "-DCOOPERLAKE " \
                     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
@ -563,6 +624,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define FORCE_INTEL
 #define ARCHITECTURE    "X86"
 #ifdef NO_AVX2
+#ifdef NO_AVX
+#define SUBARCHITECTURE "NEHALEM"
+#define ARCHCONFIG   "-DNEHALEM " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2"
+#define LIBNAME   "nehalem"
+#define CORENAME  "NEHALEM"
+#else
 #define SUBARCHITECTURE "SANDYBRIDGE"
 #define ARCHCONFIG   "-DSANDYBRIDGE " \
 		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
@ -571,6 +642,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 -DHAVE_AVX"
 #define LIBNAME   "sandybridge"
 #define CORENAME  "SANDYBRIDGE"
+#endif
 #else
 #define SUBARCHITECTURE "ZEN"
 #define ARCHCONFIG   "-DZEN " \
@ -842,6 +914,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif

+#ifdef FORCE_LOONGSON3R5
+#define FORCE
+#define ARCHITECTURE    "LOONGARCH"
+#define SUBARCHITECTURE "LOONGSON3R5"
+#define SUBDIRNAME      "loongarch64"
+#define ARCHCONFIG   "-DLOONGSON3R5 " \
+       "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=64 " \
+       "-DL2_SIZE=1048576 -DL2_LINESIZE=64 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=16 "
+#define LIBNAME   "loongson3r5"
+#define CORENAME  "LOONGSON3R5"
+#else
+#endif
+
 #ifdef FORCE_I6400
 #define FORCE
 #define ARCHITECTURE    "MIPS"
@ -1388,6 +1474,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define OPENBLAS_SUPPORTED
 #endif

+#ifdef __loongarch64
+#include "cpuid_loongarch64.c"
+#define OPENBLAS_SUPPORTED
+#endif
+
 #ifdef __riscv
 #include "cpuid_riscv64.c"
 #define OPENBLAS_SUPPORTED
@ -1463,7 +1554,7 @@ int main(int argc, char *argv[]){
 #ifdef FORCE
    printf("CORE=%s\n", CORENAME);
 #else
-#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc)
+#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc) || defined(__loongarch__)
    printf("CORE=%s\n", get_corename());
 #endif
 #endif
@ -1611,7 +1702,7 @@ printf("ELF_VERSION=2\n");
 #ifdef FORCE
    printf("#define CHAR_CORENAME \"%s\"\n", CORENAME);
 #else
-#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc)
+#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc) || defined(__loongarch__)
    printf("#define CHAR_CORENAME \"%s\"\n", get_corename());
 #endif
 #endif
--- a/interface/CMakeLists.txt
+++ b/interface/CMakeLists.txt
@ -82,6 +82,7 @@ foreach (CBLAS_FLAG ${CBLAS_FLAGS})
  GenerateNamedObjects("${BLAS3_SOURCES}" "" "" ${CBLAS_FLAG} "" "" false ${DISABLE_COMPLEX})
  GenerateNamedObjects("${BLAS3_MANGLED_SOURCES}" "" "" ${CBLAS_FLAG} "" "" false ${MANGLE_COMPLEX})

+  GenerateNamedObjects("xerbla.c" "" "xerbla" ${CBLAS_FLAG} "" "" true)
  #sdsdot, dsdot
  if (BUILD_SINGLE OR BUILD_DOUBLE)
  GenerateNamedObjects("sdsdot.c" "" "sdsdot" ${CBLAS_FLAG} "" "" true "SINGLE")
@ -104,6 +105,15 @@ endif ()
  GenerateNamedObjects("imax.c" "USE_ABS;USE_MIN" "i*amin" ${CBLAS_FLAG})
  GenerateNamedObjects("imax.c" "USE_MIN" "i*min" ${CBLAS_FLAG})

+if (BUILD_BFLOAT16)
+	GenerateNamedObjects("bf16dot.c" "" "sbdot" ${CBLAS_FLAG} "" "" true "BFLOAT16")
+	GenerateNamedObjects("gemm.c" "" "sbgemm" ${CBLAS_FLAG} "" "" true "BFLOAT16")
+	GenerateNamedObjects("sbgemv.c" "" "sbgemv" ${CBLAS_FLAG} "" "" true "BFLOAT16")
+	GenerateNamedObjects("tobf16.c" "SINGLE_PREC" "sbstobf16" ${CBLAS_FLAG} "" "" true "BFLOAT16")
+	GenerateNamedObjects("tobf16.c" "DOUBLE_PREC" "sbdtobf16" ${CBLAS_FLAG} "" "" true "BFLOAT16")
+	GenerateNamedObjects("bf16to.c" "SINGLE_PREC" "sbf16tos" ${CBLAS_FLAG} "" "" true "BFLOAT16")
+	GenerateNamedObjects("bf16to.c" "DOUBLE_PREC" "dbf16tod" ${CBLAS_FLAG} "" "" true "BFLOAT16")
+endif ()

 # complex-specific sources
 foreach (float_type ${FLOAT_TYPES})
--- a/interface/gemm.c
+++ b/interface/gemm.c
@ -105,6 +105,55 @@ static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, IFLOAT *, IFLOAT *, B
 #endif
 };

+#if defined(SMALL_MATRIX_OPT) && !defined(GEMM3M) && !defined(XDOUBLE)
+#define USE_SMALL_MATRIX_OPT 1
+#else
+#define USE_SMALL_MATRIX_OPT 0
+#endif
+
+#if USE_SMALL_MATRIX_OPT
+#ifndef DYNAMIC_ARCH
+#define SMALL_KERNEL_ADDR(table, idx) ((void *)(table[idx]))
+#else
+#define SMALL_KERNEL_ADDR(table, idx) ((void *)(*(uintptr_t *)((char *)gotoblas + (size_t)(table[idx]))))
+#endif
+
+
+#ifndef COMPLEX
+static size_t gemm_small_kernel[] = {
+	GEMM_SMALL_KERNEL_NN, GEMM_SMALL_KERNEL_TN, 0, 0,
+	GEMM_SMALL_KERNEL_NT, GEMM_SMALL_KERNEL_TT, 0, 0,
+};
+
+
+static size_t gemm_small_kernel_b0[] = {
+	GEMM_SMALL_KERNEL_B0_NN, GEMM_SMALL_KERNEL_B0_TN, 0, 0,
+	GEMM_SMALL_KERNEL_B0_NT, GEMM_SMALL_KERNEL_B0_TT, 0, 0,
+};
+
+#define GEMM_SMALL_KERNEL_B0(idx) (int (*)(BLASLONG, BLASLONG, BLASLONG, IFLOAT *, BLASLONG, FLOAT, IFLOAT *, BLASLONG, FLOAT *, BLASLONG)) SMALL_KERNEL_ADDR(gemm_small_kernel_b0, (idx))
+#define GEMM_SMALL_KERNEL(idx) (int (*)(BLASLONG, BLASLONG, BLASLONG, IFLOAT *, BLASLONG, FLOAT, IFLOAT *, BLASLONG, FLOAT, FLOAT *, BLASLONG)) SMALL_KERNEL_ADDR(gemm_small_kernel, (idx))
+#else
+
+static size_t zgemm_small_kernel[] = {
+	GEMM_SMALL_KERNEL_NN, GEMM_SMALL_KERNEL_TN, GEMM_SMALL_KERNEL_RN, GEMM_SMALL_KERNEL_CN,
+	GEMM_SMALL_KERNEL_NT, GEMM_SMALL_KERNEL_TT, GEMM_SMALL_KERNEL_RT, GEMM_SMALL_KERNEL_CT,
+	GEMM_SMALL_KERNEL_NR, GEMM_SMALL_KERNEL_TR, GEMM_SMALL_KERNEL_RR, GEMM_SMALL_KERNEL_CR,
+	GEMM_SMALL_KERNEL_NC, GEMM_SMALL_KERNEL_TC, GEMM_SMALL_KERNEL_RC, GEMM_SMALL_KERNEL_CC,
+};
+
+static size_t zgemm_small_kernel_b0[] = {
+	GEMM_SMALL_KERNEL_B0_NN, GEMM_SMALL_KERNEL_B0_TN, GEMM_SMALL_KERNEL_B0_RN, GEMM_SMALL_KERNEL_B0_CN,
+	GEMM_SMALL_KERNEL_B0_NT, GEMM_SMALL_KERNEL_B0_TT, GEMM_SMALL_KERNEL_B0_RT, GEMM_SMALL_KERNEL_B0_CT,
+	GEMM_SMALL_KERNEL_B0_NR, GEMM_SMALL_KERNEL_B0_TR, GEMM_SMALL_KERNEL_B0_RR, GEMM_SMALL_KERNEL_B0_CR,
+	GEMM_SMALL_KERNEL_B0_NC, GEMM_SMALL_KERNEL_B0_TC, GEMM_SMALL_KERNEL_B0_RC, GEMM_SMALL_KERNEL_B0_CC,
+};
+
+#define ZGEMM_SMALL_KERNEL(idx) (int (*)(BLASLONG, BLASLONG, BLASLONG, FLOAT *, BLASLONG, FLOAT , FLOAT, FLOAT *, BLASLONG, FLOAT , FLOAT, FLOAT *, BLASLONG)) SMALL_KERNEL_ADDR(zgemm_small_kernel, (idx))
+#define ZGEMM_SMALL_KERNEL_B0(idx) (int (*)(BLASLONG, BLASLONG, BLASLONG, FLOAT *, BLASLONG, FLOAT , FLOAT, FLOAT *, BLASLONG, FLOAT *, BLASLONG)) SMALL_KERNEL_ADDR(zgemm_small_kernel_b0, (idx))
+#endif
+#endif
+
 #ifndef CBLAS

 void NAME(char *TRANSA, char *TRANSB,
@ -224,8 +273,8 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
 	   blasint m, blasint n, blasint k,
 #ifndef COMPLEX
 	   FLOAT alpha,
-	   FLOAT *a, blasint lda,
-	   FLOAT *b, blasint ldb,
+	   IFLOAT *a, blasint lda,
+	   IFLOAT *b, blasint ldb,
 	   FLOAT beta,
 	   FLOAT *c, blasint ldc) {
 #else
@ -277,7 +326,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS

  PRINT_DEBUG_CNAME;

-#if !defined(COMPLEX) && !defined(DOUBLE) && defined(USE_SGEMM_KERNEL_DIRECT)
+#if !defined(COMPLEX) && !defined(DOUBLE) && !defined(BFLOAT16) && defined(USE_SGEMM_KERNEL_DIRECT)
 #ifdef DYNAMIC_ARCH
 if (support_avx512() )
 #endif  
@ -417,6 +466,28 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS

  FUNCTION_PROFILE_START();

+#if USE_SMALL_MATRIX_OPT
+#if !defined(COMPLEX)
+  if(GEMM_SMALL_MATRIX_PERMIT(transa, transb, args.m, args.n, args.k, *(FLOAT *)(args.alpha), *(FLOAT *)(args.beta))){
+	  if(*(FLOAT *)(args.beta) == 0.0){
+		(GEMM_SMALL_KERNEL_B0((transb << 2) | transa))(args.m, args.n, args.k, args.a, args.lda, *(FLOAT *)(args.alpha), args.b, args.ldb, args.c, args.ldc);
+	  }else{
+		(GEMM_SMALL_KERNEL((transb << 2) | transa))(args.m, args.n, args.k, args.a, args.lda, *(FLOAT *)(args.alpha), args.b, args.ldb, *(FLOAT *)(args.beta), args.c, args.ldc);
+	  }
+	  return;
+  }
+#else
+  if(GEMM_SMALL_MATRIX_PERMIT(transa, transb, args.m, args.n, args.k, alpha[0], alpha[1], beta[0], beta[1])){
+	  if(beta[0] == 0.0 && beta[1] == 0.0){
+		(ZGEMM_SMALL_KERNEL_B0((transb << 2) | transa))(args.m, args.n, args.k, args.a, args.lda, alpha[0], alpha[1], args.b, args.ldb, args.c, args.ldc);
+	  }else{
+		(ZGEMM_SMALL_KERNEL((transb << 2) | transa))(args.m, args.n, args.k, args.a, args.lda, alpha[0], alpha[1], args.b, args.ldb, beta[0], beta[1], args.c, args.ldc);
+	  }
+	  return;
+  }
+#endif
+#endif
+
  buffer = (XFLOAT *)blas_memory_alloc(0);

  sa = (XFLOAT *)((BLASLONG)buffer +GEMM_OFFSET_A);
--- a/interface/zsyr.c
+++ b/interface/zsyr.c
@ -119,7 +119,7 @@ void NAME(char *UPLO, blasint *N, FLOAT  *ALPHA,
 void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, int n, FLOAT alpha, FLOAT *x, int incx, FLOAT *a, int lda) {

  FLOAT *buffer;
-  int trans, uplo;
+  int uplo;
  blasint info;
  FLOAT * ALPHA = &alpha;
  FLOAT alpha_r	= ALPHA[0];
@ -130,7 +130,6 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, int n, FLOAT alpha, FLO

  PRINT_DEBUG_CNAME;

-  trans = -1;
  uplo  = -1;
  info  =  0;

--- a/kernel/CMakeLists.txt
+++ b/kernel/CMakeLists.txt
@ -91,6 +91,15 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
    GenerateNamedObjects("${KERNELDIR}/${DSDOTKERNEL}" "DSDOT" "d*dot_k" false "" "" false "SINGLE")
    GenerateNamedObjects("${KERNELDIR}/${DSDOTKERNEL}" "DSDOT" "dsdot_k" false "" "" false "SINGLE")

+    # sbdot
+    if (BUILD_BFLOAT16)
+	    GenerateNamedObjects("${KERNELDIR}/${SBDOTKERNEL}" "SBDOT" "dot_k" false "" "" false "BFLOAT16")
+	    GenerateNamedObjects("${KERNELDIR}/${BF16TOKERNEL}" "SINGLE" "f16tos_k" false "" "" false "BFLOAT16")
+	    GenerateNamedObjects("${KERNELDIR}/${BF16TOKERNEL}" "DOUBLE" "bf16tod_k" false "" "" false "DOUBLE")
+	    GenerateNamedObjects("${KERNELDIR}/${TOBF16KERNEL}" "SINGLE" "stobf16_k" false "" "" false "BFLOAT16")
+	    GenerateNamedObjects("${KERNELDIR}/${TOBF16KERNEL}" "DOUBLE" "dtobf16_k" false "" "" false "BFLOAT16")
+    endif()
+
    if ((BUILD_COMPLEX OR BUILD_DOUBLE)  AND NOT BUILD_SINGLE)
    GenerateNamedObjects("${KERNELDIR}/${SAMAXKERNEL}" "USE_ABS" "amax_k" false "" "" false "SINGLE")
    GenerateNamedObjects("${KERNELDIR}/${SAMINKERNEL}" "USE_ABS;USE_MIN" "amin_k" false "" "" false "SINGLE")
@ -149,9 +158,6 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
    GenerateNamedObjects("generic/ger.c" "" "ger_k" false "" "" "" 3)
    foreach (float_type ${FLOAT_TYPES})
      string(SUBSTRING ${float_type} 0 1 float_char)
-      if (${float_type} STREQUAL "BFLOAT16")
-	set (float_char "SB")
-      endif ()
      if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
        GenerateNamedObjects("${KERNELDIR}/${${float_char}GERUKERNEL}" "" "geru_k" false "" "" false ${float_type})
        GenerateNamedObjects("${KERNELDIR}/${${float_char}GERCKERNEL}" "CONJ" "gerc_k" false "" "" false ${float_type})
@ -185,6 +191,10 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
 	    GenerateNamedObjects("${KERNELDIR}/${SGEMVNKERNEL}" "" "gemv_n" false "" "" false "SINGLE")
 	    GenerateNamedObjects("${KERNELDIR}/${SGEMVTKERNEL}" "TRANS" "gemv_t" false "" "" false "SINGLE")
    endif ()
+    if (BUILD_BFLOAT16)
+	    GenerateNamedObjects("${KERNELDIR}/${SBGEMVNKERNEL}" "" "gemv_n" false "" "" false "BFLOAT16")
+	    GenerateNamedObjects("${KERNELDIR}/${SBGEMVTKERNEL}" "" "gemv_t" false "" "" false "BFLOAT16")
+    endif ()
    # Makefile.L3
    set(USE_TRMM false)
    string(TOUPPER ${TARGET_CORE} UC_TARGET_CORE)
@ -209,15 +219,8 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
 	  GenerateNamedObjects("${KERNELDIR}/${SGEMMDIRECTPERFORMANT}" "" "gemm_direct_performant" false "" "" false  SINGLE)
    endif()

-    foreach (float_type SINGLE DOUBLE BFLOAT16)
+    foreach (float_type SINGLE DOUBLE)
      string(SUBSTRING ${float_type} 0 1 float_char)
-      if (${float_type} STREQUAL "BFLOAT16")
-        if (NOT ${BUILD_BFLOAT16})
-	  continue ()
-        else ()
-	  set (float_char "SB")
-      endif ()
-      endif ()
      GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "" "gemm_kernel" false "" "" false ${float_type})
    endforeach()
    if (BUILD_COMPLEX16  AND NOT BUILD_DOUBLE)
@ -253,11 +256,24 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
 	    GenerateNamedObjects("${KERNELDIR}/${SGEMM_BETA}" "" "gemm_beta" false "" "" false "SINGLE")
    endif ()

+    if (BUILD_BFLOAT16)
+        if (SBGEMMINCOPY)
+		GenerateNamedObjects("${KERNELDIR}/${SBGEMMINCOPY}" "" "${SBGEMMINCOPYOBJ}" false "" "" true "BFLOAT16")
+        endif ()
+        if (SBGEMMITCOPY)
+		GenerateNamedObjects("${KERNELDIR}/${SBGEMMITCOPY}" "" "${SBGEMMITCOPYOBJ}" false "" "" true "BFLOAT16")
+        endif ()
+        if (SBGEMMONCOPY)
+		GenerateNamedObjects("${KERNELDIR}/${SBGEMMONCOPY}" "" "${SBGEMMONCOPYOBJ}" false "" "" true "BFLOAT16")
+        endif ()
+        if (SBGEMMOTCOPY)
+		GenerateNamedObjects("${KERNELDIR}/${SBGEMMOTCOPY}" "" "${SBGEMMOTCOPYOBJ}" false "" "" true "BFLOAT16")
+        endif ()
+	GenerateNamedObjects("${KERNELDIR}/${SBGEMMKERNEL}" "" "gemm_kernel" false "" "" false "BFLOAT16")
+	GenerateNamedObjects("${KERNELDIR}/${SBGEMM_BETA}" "" "gemm_beta" false "" "" false "BFLOAT16")
+    endif ()
    foreach (float_type ${FLOAT_TYPES})
      string(SUBSTRING ${float_type} 0 1 float_char)
-      if (${float_type} STREQUAL "BFLOAT16")
-	set (float_char "SB")
-      endif ()
      if (${float_char}GEMMINCOPY)
        GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMINCOPY}" "${float_type}" "${${float_char}GEMMINCOPYOBJ}" false "" "" true ${float_type})
      endif ()
@ -458,7 +474,155 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
      GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_RN}" "UPPER;RN;TRSMKERNEL" "trsm_kernel_RN" false "" "" false ${float_type})
      GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_RT}" "RT;TRSMKERNEL" "trsm_kernel_RT" false "" "" false ${float_type})

+      if (NOT DEFINED ${float_char}GEMM_SMALL_M_PERMIT)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_M_PERMIT ../generic/zgemm_small_matrix_permit.c)
+        else ()
+          set(${float_char}GEMM_SMALL_M_PERMIT ../generic/gemm_small_matrix_permit.c)
+        endif ()
+      endif ()
+      if (NOT DEFINED ${float_char}GEMM_SMALL_K_NN)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_K_NN ../generic/zgemm_small_matrix_kernel_nn.c)
+        else ()
+          set(${float_char}GEMM_SMALL_K_NN ../generic/gemm_small_matrix_kernel_nn.c)
+        endif ()
+      endif ()
+      if (NOT DEFINED ${float_char}GEMM_SMALL_K_NT)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_K_NT ../generic/zgemm_small_matrix_kernel_nt.c)
+        else ()
+          set(${float_char}GEMM_SMALL_K_NT ../generic/gemm_small_matrix_kernel_nt.c)
+        endif ()
+      endif ()
+      if (NOT DEFINED ${float_char}GEMM_SMALL_K_TN)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_K_TN ../generic/zgemm_small_matrix_kernel_tn.c)
+        else ()
+          set(${float_char}GEMM_SMALL_K_TN ../generic/gemm_small_matrix_kernel_tn.c)
+        endif ()
+      endif ()
+      if (NOT DEFINED ${float_char}GEMM_SMALL_K_TT)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_K_TT ../generic/zgemm_small_matrix_kernel_tt.c)
+        else ()
+          set(${float_char}GEMM_SMALL_K_TT ../generic/gemm_small_matrix_kernel_tt.c)
+        endif ()
+      endif ()
+      if (NOT DEFINED ${float_char}GEMM_SMALL_K_B0_NN)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_K_B0_NN ../generic/zgemm_small_matrix_kernel_nn.c)
+        else ()
+          set(${float_char}GEMM_SMALL_K_B0_NN ../generic/gemm_small_matrix_kernel_nn.c)
+        endif ()
+      endif ()
+      if (NOT DEFINED ${float_char}GEMM_SMALL_K_B0_NT)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_K_B0_NT ../generic/zgemm_small_matrix_kernel_nt.c)
+        else ()
+          set(${float_char}GEMM_SMALL_K_B0_NT ../generic/gemm_small_matrix_kernel_nt.c)
+        endif ()
+      endif ()
+      if (NOT DEFINED ${float_char}GEMM_SMALL_K_B0_TN)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_K_B0_TN ../generic/zgemm_small_matrix_kernel_tn.c)
+        else ()
+          set(${float_char}GEMM_SMALL_K_B0_TN ../generic/gemm_small_matrix_kernel_tn.c)
+        endif ()
+      endif ()
+      if (NOT DEFINED ${float_char}GEMM_SMALL_K_B0_TT)
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+          set(${float_char}GEMM_SMALL_K_B0_TT ../generic/zgemm_small_matrix_kernel_tt.c)
+        else ()
+          set(${float_char}GEMM_SMALL_K_B0_TT ../generic/gemm_small_matrix_kernel_tt.c)
+        endif ()
+      endif ()

+      if (SMALL_MATRIX_OPT)
+        GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_M_PERMIT}" "" "gemm_small_matrix_permit" false "" "" false ${float_type})
+        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NN}" "NN" "gemm_small_kernel_nn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NN}" "NR" "gemm_small_kernel_nr" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NN}" "RN" "gemm_small_kernel_rn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NN}" "RR" "gemm_small_kernel_rr" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NT}" "NT" "gemm_small_kernel_nt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NT}" "NC" "gemm_small_kernel_nc" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NT}" "RT" "gemm_small_kernel_rt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NT}" "RC" "gemm_small_kernel_rc" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TN}" "TN" "gemm_small_kernel_tn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TN}" "TR" "gemm_small_kernel_tr" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TN}" "CN" "gemm_small_kernel_cn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TN}" "CR" "gemm_small_kernel_cr" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TT}" "TT" "gemm_small_kernel_tt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TT}" "TC" "gemm_small_kernel_tc" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TT}" "CT" "gemm_small_kernel_ct" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TT}" "CC" "gemm_small_kernel_cc" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NN}" "NN;B0" "gemm_small_kernel_b0_nn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NN}" "NR;B0" "gemm_small_kernel_b0_nr" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NN}" "RN;B0" "gemm_small_kernel_b0_rn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NN}" "RR;B0" "gemm_small_kernel_b0_rr" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NT}" "NT;B0" "gemm_small_kernel_b0_nt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NT}" "NC;B0" "gemm_small_kernel_b0_nc" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NT}" "RT;B0" "gemm_small_kernel_b0_rt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NT}" "RC;B0" "gemm_small_kernel_b0_rc" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TN}" "TN;B0" "gemm_small_kernel_b0_tn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TN}" "TR;B0" "gemm_small_kernel_b0_tr" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TN}" "CN;B0" "gemm_small_kernel_b0_cn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TN}" "CR;B0" "gemm_small_kernel_b0_cr" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TT}" "TT;B0" "gemm_small_kernel_b0_tt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TT}" "TC;B0" "gemm_small_kernel_b0_tc" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TT}" "CT;B0" "gemm_small_kernel_b0_ct" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TT}" "CC;B0" "gemm_small_kernel_b0_cc" false "" "" false ${float_type})
+
+        else ()
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NN}" "" "gemm_small_kernel_nn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NT}" "" "gemm_small_kernel_nt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_TN}" "" "gemm_small_kernel_tn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_NT}" "" "gemm_small_kernel_tt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NN}" "B0" "gemm_small_kernel_b0_nn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NT}" "B0" "gemm_small_kernel_b0_nt" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_TN}" "B0" "gemm_small_kernel_b0_tn" false "" "" false ${float_type})
+            GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_SMALL_K_B0_NT}" "B0" "gemm_small_kernel_b0_tt" false "" "" false ${float_type})
+        endif ()
+	if (BUILD_BFLOAT16)
+      if (NOT DEFINED SBGEMM_SMALL_M_PERMIT)
+          set(SBGEMM_SMALL_M_PERMIT ../generic/gemm_small_matrix_permit.c)
+      endif ()
+      if (NOT DEFINED SBGEMM_SMALL_K_NN)
+          set(SBGEMM_SMALL_K_NN ../generic/gemm_small_matrix_kernel_nn.c)
+      endif ()
+      if (NOT DEFINED SBGEMM_SMALL_K_NT)
+          set(SBGEMM_SMALL_K_NT ../generic/gemm_small_matrix_kernel_nt.c)
+      endif ()
+      if (NOT DEFINED SBGEMM_SMALL_K_TN)
+          set(SBGEMM_SMALL_K_TN ../generic/gemm_small_matrix_kernel_tn.c)
+      endif ()
+      if (NOT DEFINED SBGEMM_SMALL_K_TT)
+          set(SBGEMM_SMALL_K_TT ../generic/gemm_small_matrix_kernel_tt.c)
+      endif ()
+      if (NOT DEFINED SBGEMM_SMALL_K_B0_NN)
+          set(SBGEMM_SMALL_K_B0_NN ../generic/gemm_small_matrix_kernel_nn.c)
+      endif ()
+      if (NOT DEFINED SBGEMM_SMALL_K_B0_NT)
+          set(SBGEMM_SMALL_K_B0_NT ../generic/gemm_small_matrix_kernel_nt.c)
+      endif ()
+      if (NOT DEFINED SBGEMM_SMALL_K_B0_TN)
+          set(SBGEMM_SMALL_K_B0_TN ../generic/gemm_small_matrix_kernel_tn.c)
+      endif ()
+      if (NOT DEFINED SBGEMM_SMALL_K_B0_TT)
+          set($SBGEMM_SMALL_K_B0_TT ../generic/gemm_small_matrix_kernel_tt.c)
+      endif ()
+	    GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_M_PERMIT}" "" "gemm_small_matrix_permit" false "" "" false "BFLOAT16")
+            GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_K_NN}" "" "gemm_small_kernel_nn" false "" "" false "BFLOAT16")
+            GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_K_NT}" "" "gemm_small_kernel_nt" false "" "" false "BFLOAT16")
+            GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_K_TN}" "" "gemm_small_kernel_tn" false "" "" false "BFLOAT16")
+            GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_K_NT}" "" "gemm_small_kernel_tt" false "" "" false "BFLOAT16")
+            GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_K_B0_NN}" "B0" "gemm_small_kernel_b0_nn" false "" "" false "BFLOAT16")
+            GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_K_B0_NT}" "B0" "gemm_small_kernel_b0_nt" false "" "" false "BFLOAT16")
+            GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_K_B0_TN}" "B0" "gemm_small_kernel_b0_tn" false "" "" false "BFLOAT16")
+            GenerateNamedObjects("${KERNELDIR}/${SBGEMM_SMALL_K_B0_NT}" "B0" "gemm_small_kernel_b0_tt" false "" "" false "BFLOAT16")
+        endif ()
+      endif ()

      if (NOT DEFINED ${float_char}OMATCOPY_CN)
        if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
@ -592,6 +756,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
      #geadd
      GenerateNamedObjects("${KERNELDIR}/${${float_char}GEADD_KERNEL}" "" "geadd_k" false "" "" false ${float_type})
    endforeach ()
+
    if (BUILD_DOUBLE AND NOT BUILD_SINGLE)
 	    GenerateNamedObjects("${KERNELDIR}/${STRSMKERNEL_LN}" "UPPER;LN;TRSMKERNEL" "trsm_kernel_LN" false "" "" false "SINGLE")
 	    GenerateNamedObjects("${KERNELDIR}/${STRSMKERNEL_LT}" "LT;TRSMKERNEL" "trsm_kernel_LT" false "" "" false "SINGLE")
--- a/kernel/Makefile.L2
+++ b/kernel/Makefile.L2
@ -1,3 +1,10 @@
+FMAFLAG=
+ifndef OLDGCC
+ifdef HAVE_FMA3
+FMAFLAG = -mfma
+endif
+endif
+
 ### GEMV ###

 ifndef SGEMVNKERNEL
@ -263,7 +270,7 @@ $(KDIR)dgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)dgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KER
 	$(CC) -c $(CFLAGS) -DDOUBLE -UCOMPLEX  -UTRANS $< -o $@

 $(KDIR)dgemv_t$(TSUFFIX).$(SUFFIX)  $(KDIR)dgemv_t$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -UCOMPLEX  -DTRANS $< -o $@
+	$(CC) -c $(CFLAGS) $(FMAFLAG) -DDOUBLE -UCOMPLEX  -DTRANS $< -o $@
 endif

 $(KDIR)qgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)qgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QGEMVNKERNEL)
--- a/kernel/Makefile.L3
+++ b/kernel/Makefile.L3
@ -447,6 +447,72 @@ XBLASOBJS += \

 endif

+######  BLAS small matrix optimization #####
+ifeq ($(SMALL_MATRIX_OPT), 1)
+
+ifeq ($(BUILD_BFLOAT16),1)
+SBBLASOBJS += \
+	sbgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) \
+	sbgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) sbgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) \
+	sbgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) sbgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) \
+	sbgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) sbgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) \
+	sbgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) sbgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX)
+endif
+
+SBLASOBJS += \
+	sgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) \
+	sgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) \
+	sgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) \
+	sgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) \
+	sgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) sgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX)
+
+DBLASOBJS += \
+	dgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) \
+	dgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) \
+	dgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) \
+	dgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) \
+	dgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) dgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX)
+
+CBLASOBJS += \
+	cgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_nr$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_nc$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_tr$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_tc$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_rn$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_rt$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_rr$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_rc$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_cn$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_ct$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_cr$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_cc$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_b0_nr$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_b0_nc$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_b0_tr$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_b0_tc$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_b0_rn$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_b0_rt$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_b0_rr$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_b0_rc$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_b0_cn$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_b0_ct$(TSUFFIX).$(SUFFIX) \
+	cgemm_small_kernel_b0_cr$(TSUFFIX).$(SUFFIX) cgemm_small_kernel_b0_cc$(TSUFFIX).$(SUFFIX)
+
+ZBLASOBJS += \
+	zgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_nr$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_nc$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_tr$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_tc$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_rn$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_rt$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_rr$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_rc$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_cn$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_ct$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_cr$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_cc$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_b0_nr$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_b0_nc$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_b0_tr$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_b0_tc$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_b0_rn$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_b0_rt$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_b0_rr$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_b0_rc$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_b0_cn$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_b0_ct$(TSUFFIX).$(SUFFIX) \
+	zgemm_small_kernel_b0_cr$(TSUFFIX).$(SUFFIX) zgemm_small_kernel_b0_cc$(TSUFFIX).$(SUFFIX)
+
+endif
+
 ######  BLAS extensions #####

 ifeq ($(BUILD_SINGLE),1)
@ -4237,3 +4303,469 @@ endif
 $(KDIR)zgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEADD_K)
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM $< -o $@

+
+
+######  BLAS small matrix optimization #####
+
+ifndef DGEMM_SMALL_M_PERMIT
+DGEMM_SMALL_M_PERMIT = ../generic/gemm_small_matrix_permit.c
+endif
+
+ifndef DGEMM_SMALL_K_NN
+DGEMM_SMALL_K_NN = ../generic/gemm_small_matrix_kernel_nn.c
+endif
+
+ifndef DGEMM_SMALL_K_NT
+DGEMM_SMALL_K_NT = ../generic/gemm_small_matrix_kernel_nt.c
+endif
+
+ifndef DGEMM_SMALL_K_TN
+DGEMM_SMALL_K_TN = ../generic/gemm_small_matrix_kernel_tn.c
+endif
+
+ifndef DGEMM_SMALL_K_TT
+DGEMM_SMALL_K_TT = ../generic/gemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)dgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_M_PERMIT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)dgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)dgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)dgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)dgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+
+ifndef DGEMM_SMALL_K_B0_NN
+DGEMM_SMALL_K_B0_NN = ../generic/gemm_small_matrix_kernel_nn.c
+endif
+
+ifndef DGEMM_SMALL_K_B0_NT
+DGEMM_SMALL_K_B0_NT = ../generic/gemm_small_matrix_kernel_nt.c
+endif
+
+ifndef DGEMM_SMALL_K_B0_TN
+DGEMM_SMALL_K_B0_TN = ../generic/gemm_small_matrix_kernel_tn.c
+endif
+
+ifndef DGEMM_SMALL_K_B0_TT
+DGEMM_SMALL_K_B0_TT = ../generic/gemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)dgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)dgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)dgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)dgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+ifndef SGEMM_SMALL_M_PERMIT
+SGEMM_SMALL_M_PERMIT = ../generic/gemm_small_matrix_permit.c
+endif
+
+ifndef SGEMM_SMALL_K_NN
+SGEMM_SMALL_K_NN = ../generic/gemm_small_matrix_kernel_nn.c
+endif
+
+ifndef SGEMM_SMALL_K_NT
+SGEMM_SMALL_K_NT = ../generic/gemm_small_matrix_kernel_nt.c
+endif
+
+ifndef SGEMM_SMALL_K_TN
+SGEMM_SMALL_K_TN = ../generic/gemm_small_matrix_kernel_tn.c
+endif
+
+ifndef SGEMM_SMALL_K_TT
+SGEMM_SMALL_K_TT = ../generic/gemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)sgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_M_PERMIT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)sgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)sgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)sgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)sgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+
+ifndef SGEMM_SMALL_K_B0_NN
+SGEMM_SMALL_K_B0_NN = ../generic/gemm_small_matrix_kernel_nn.c
+endif
+
+ifndef SGEMM_SMALL_K_B0_NT
+SGEMM_SMALL_K_B0_NT = ../generic/gemm_small_matrix_kernel_nt.c
+endif
+
+ifndef SGEMM_SMALL_K_B0_TN
+SGEMM_SMALL_K_B0_TN = ../generic/gemm_small_matrix_kernel_tn.c
+endif
+
+ifndef SGEMM_SMALL_K_B0_TT
+SGEMM_SMALL_K_B0_TT = ../generic/gemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)sgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)sgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)sgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)sgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+
+ifeq ($(BUILD_BFLOAT16), 1)
+ifndef SBGEMM_SMALL_M_PERMIT
+SBGEMM_SMALL_M_PERMIT = ../generic/gemm_small_matrix_permit.c
+endif
+
+ifndef SBGEMM_SMALL_K_NN
+SBGEMM_SMALL_K_NN = ../generic/gemm_small_matrix_kernel_nn.c
+endif
+
+ifndef SBGEMM_SMALL_K_NT
+SBGEMM_SMALL_K_NT = ../generic/gemm_small_matrix_kernel_nt.c
+endif
+
+ifndef SBGEMM_SMALL_K_TN
+SBGEMM_SMALL_K_TN = ../generic/gemm_small_matrix_kernel_tn.c
+endif
+
+ifndef SBGEMM_SMALL_K_TT
+SBGEMM_SMALL_K_TT = ../generic/gemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)sbgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_M_PERMIT)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)sbgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)sbgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)sbgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)sbgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+
+ifndef SBGEMM_SMALL_K_B0_NN
+SBGEMM_SMALL_K_B0_NN = ../generic/gemm_small_matrix_kernel_nn.c
+endif
+
+ifndef SBGEMM_SMALL_K_B0_NT
+SBGEMM_SMALL_K_B0_NT = ../generic/gemm_small_matrix_kernel_nt.c
+endif
+
+ifndef SBGEMM_SMALL_K_B0_TN
+SBGEMM_SMALL_K_B0_TN = ../generic/gemm_small_matrix_kernel_tn.c
+endif
+
+ifndef SBGEMM_SMALL_K_B0_TT
+SBGEMM_SMALL_K_B0_TT = ../generic/gemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)sbgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)sbgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)sbgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX -DB0 $< -o $@
+
+$(KDIR)sbgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX -DB0 $< -o $@
+endif
+
+ifndef CGEMM_SMALL_M_PERMIT
+CGEMM_SMALL_M_PERMIT = ../generic/zgemm_small_matrix_permit.c
+endif
+
+ifndef CGEMM_SMALL_K_NN
+CGEMM_SMALL_K_NN = ../generic/zgemm_small_matrix_kernel_nn.c
+endif
+
+ifndef CGEMM_SMALL_K_NT
+CGEMM_SMALL_K_NT = ../generic/zgemm_small_matrix_kernel_nt.c
+endif
+
+ifndef CGEMM_SMALL_K_TN
+CGEMM_SMALL_K_TN = ../generic/zgemm_small_matrix_kernel_tn.c
+endif
+
+ifndef CGEMM_SMALL_K_TT
+CGEMM_SMALL_K_TT = ../generic/zgemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)cgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_M_PERMIT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX $< -o $@
+
+$(KDIR)cgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
+	
+$(KDIR)cgemm_small_kernel_nr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNR $< -o $@
+
+$(KDIR)cgemm_small_kernel_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DRN $< -o $@
+
+$(KDIR)cgemm_small_kernel_rr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DRR $< -o $@
+
+$(KDIR)cgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNT $< -o $@
+
+$(KDIR)cgemm_small_kernel_nc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNC $< -o $@
+
+$(KDIR)cgemm_small_kernel_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DRT $< -o $@
+
+$(KDIR)cgemm_small_kernel_rc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DRC=RC $< -o $@
+
+$(KDIR)cgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DTN $< -o $@
+
+$(KDIR)cgemm_small_kernel_tr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DTR $< -o $@
+
+$(KDIR)cgemm_small_kernel_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $@
+
+$(KDIR)cgemm_small_kernel_cr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCR=CR $< -o $@
+
+$(KDIR)cgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DTT $< -o $@
+
+$(KDIR)cgemm_small_kernel_tc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DTC $< -o $@
+
+$(KDIR)cgemm_small_kernel_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCT $< -o $@
+
+$(KDIR)cgemm_small_kernel_cc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCC $< -o $@
+
+ifndef CGEMM_SMALL_K_B0_NN
+CGEMM_SMALL_K_B0_NN = ../generic/zgemm_small_matrix_kernel_nn.c
+endif
+
+ifndef CGEMM_SMALL_K_B0_NT
+CGEMM_SMALL_K_B0_NT = ../generic/zgemm_small_matrix_kernel_nt.c
+endif
+
+ifndef CGEMM_SMALL_K_B0_TN
+CGEMM_SMALL_K_B0_TN = ../generic/zgemm_small_matrix_kernel_tn.c
+endif
+
+ifndef CGEMM_SMALL_K_B0_TT
+CGEMM_SMALL_K_B0_TT = ../generic/zgemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)cgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNN -DB0 $< -o $@
+	
+$(KDIR)cgemm_small_kernel_b0_nr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNR -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DRN -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_rr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DRR -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNT -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_nc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNC -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DRT -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_rc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DRC=RC -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DTN -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_tr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DTR -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCN -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_cr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCR=CR -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DTT -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_tc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DTC -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCT -DB0 $< -o $@
+
+$(KDIR)cgemm_small_kernel_b0_cc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCC -DB0 $< -o $@
+
+ifndef ZGEMM_SMALL_M_PERMIT
+ZGEMM_SMALL_M_PERMIT = ../generic/zgemm_small_matrix_permit.c
+endif
+
+ifndef ZGEMM_SMALL_K_NN
+ZGEMM_SMALL_K_NN = ../generic/zgemm_small_matrix_kernel_nn.c
+endif
+
+ifndef ZGEMM_SMALL_K_NT
+ZGEMM_SMALL_K_NT = ../generic/zgemm_small_matrix_kernel_nt.c
+endif
+
+ifndef ZGEMM_SMALL_K_TN
+ZGEMM_SMALL_K_TN = ../generic/zgemm_small_matrix_kernel_tn.c
+endif
+
+ifndef ZGEMM_SMALL_K_TT
+ZGEMM_SMALL_K_TT = ../generic/zgemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)zgemm_small_matrix_permit$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_M_PERMIT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX $< -o $@
+
+
+$(KDIR)zgemm_small_kernel_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
+	
+$(KDIR)zgemm_small_kernel_nr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNR $< -o $@
+
+$(KDIR)zgemm_small_kernel_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DRN $< -o $@
+
+$(KDIR)zgemm_small_kernel_rr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DRR $< -o $@
+
+$(KDIR)zgemm_small_kernel_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNT $< -o $@
+
+$(KDIR)zgemm_small_kernel_nc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNC $< -o $@
+
+$(KDIR)zgemm_small_kernel_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DRT $< -o $@
+
+$(KDIR)zgemm_small_kernel_rc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DRC=RC $< -o $@
+
+$(KDIR)zgemm_small_kernel_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DTN $< -o $@
+
+$(KDIR)zgemm_small_kernel_tr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DTR $< -o $@
+
+$(KDIR)zgemm_small_kernel_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $@
+
+$(KDIR)zgemm_small_kernel_cr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCR=CR $< -o $@
+
+$(KDIR)zgemm_small_kernel_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DTT $< -o $@
+
+$(KDIR)zgemm_small_kernel_tc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DTC $< -o $@
+
+$(KDIR)zgemm_small_kernel_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCT $< -o $@
+
+$(KDIR)zgemm_small_kernel_cc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCC $< -o $@
+
+ifndef ZGEMM_SMALL_K_B0_NN
+ZGEMM_SMALL_K_B0_NN = ../generic/zgemm_small_matrix_kernel_nn.c
+endif
+
+ifndef ZGEMM_SMALL_K_B0_NT
+ZGEMM_SMALL_K_B0_NT = ../generic/zgemm_small_matrix_kernel_nt.c
+endif
+
+ifndef ZGEMM_SMALL_K_B0_TN
+ZGEMM_SMALL_K_B0_TN = ../generic/zgemm_small_matrix_kernel_tn.c
+endif
+
+ifndef ZGEMM_SMALL_K_B0_TT
+ZGEMM_SMALL_K_B0_TT = ../generic/zgemm_small_matrix_kernel_tt.c
+endif
+
+$(KDIR)zgemm_small_kernel_b0_nn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN -DB0 $< -o $@
+	
+$(KDIR)zgemm_small_kernel_b0_nr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNR -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DRN -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_rr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_NN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DRR -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_nt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNT -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_nc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNC -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DRT -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_rc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_NT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DRC=RC -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_tn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DTN -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_tr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DTR -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCN -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_cr$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_TN)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCR=CR -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_tt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DTT -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_tc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DTC -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCT -DB0 $< -o $@
+
+$(KDIR)zgemm_small_kernel_b0_cc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_SMALL_K_B0_TT)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCC -DB0 $< -o $@
--- a/kernel/arm64/dgemm_tcopy_8.S
+++ b/kernel/arm64/dgemm_tcopy_8.S
@ -50,11 +50,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define	B03		x16
 #define	B04		x17

-#define I		x18
-#define	J		x19
+#define I		x19
+#define	J		x20

-#define TEMP1		x20
-#define TEMP2		x21
+#define TEMP1		x21

 #define A_PREFETCH	2560
 #define B_PREFETCH	256
--- a/kernel/arm64/dtrmm_kernel_8x4.S
+++ b/kernel/arm64/dtrmm_kernel_8x4.S
@ -49,9 +49,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define pCRow3		x15
 #define pA		x16
 #define alpha		x17
-#define temp		x18
+//#define temp		x18
 #define tempOffset	x19
 #define tempK		x20
+#define temp		x21

 #define alpha0		d10
 #define alphaV0		v10.d[0]
--- a/kernel/arm64/sgemm_tcopy_16.S
+++ b/kernel/arm64/sgemm_tcopy_16.S
@ -30,7 +30,7 @@ All rights reserved.
 #define	B00		x22


-#define I		x18
+#define I		x21
 #define	J		x19

 #define TEMP1		x20
--- a/kernel/arm64/strmm_kernel_16x4.S
+++ b/kernel/arm64/strmm_kernel_16x4.S
@ -49,9 +49,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define pCRow3		x15
 #define pA		x16
 #define alpha		w17
-#define temp		x18
+//#define temp		x18
 #define tempOffset	x19
 #define tempK		x20
+#define temp            x21

 #define alpha0		s10
 #define alphaV0		v10.s[0]
--- a/kernel/arm64/zgemm_kernel_4x4.S
+++ b/kernel/arm64/zgemm_kernel_4x4.S
@ -48,8 +48,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define pCRow2		x14
 #define pCRow3		x15
 #define pA		x16
-#define alphaR		x17
-#define alphaI		x18
+#define alphaR		x19
+#define alphaI		x20

 #define alpha0_R	d10
 #define alphaV0_R	v10.d[0]
--- a/kernel/arm64/ztrmm_kernel_4x4.S
+++ b/kernel/arm64/ztrmm_kernel_4x4.S
@ -49,7 +49,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define pCRow3		x15
 #define pA		x16
 #define alphaR		x17
-#define alphaI		x18
+#define alphaI		x22
 #define temp		x19
 #define tempOffset	x20
 #define tempK		x21
--- a/kernel/generic/dot.c
+++ b/kernel/generic/dot.c
@ -47,7 +47,6 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)

 	if ( (inc_x == 1) && (inc_y == 1) )
 	{
-        int n1 = n & -4;
 #if V_SIMD && !defined(DSDOT)
        const int vstep = v_nlanes_f32;
        const int unrollx4 = n & (-vstep * 4);
@ -84,6 +83,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
        }
        dot = v_sum_f32(vsum0);
 #elif defined(DSDOT)
+        int n1 = n & -4;
 		for (; i < n1; i += 4)
 		{
 			dot += (double) y[i] * (double) x[i]
@ -92,6 +92,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
 			    + (double) y[i+3] * (double) x[i+3] ;
 		}
 #else
+        int n1 = n & -4;
 		for (; i < n1; i += 4)
 		{
 			dot += y[i] * x[i]
--- a/kernel/generic/gemm_small_matrix_kernel_nn.c
+++ b/kernel/generic/gemm_small_matrix_kernel_nn.c
@ -0,0 +1,56 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+#ifdef B0
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, IFLOAT * A, BLASLONG lda, FLOAT alpha, IFLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
+#else
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, IFLOAT * A, BLASLONG lda, FLOAT alpha, IFLOAT * B, BLASLONG ldb, FLOAT beta, FLOAT * C, BLASLONG ldc)
+#endif
+{
+	//naive implemtation
+	//Column major
+
+	BLASLONG i,j,k;
+	FLOAT result=0.0;
+
+	for(i=0; i<M; i++){
+		for(j=0; j<N; j++){
+			result=0.0;
+			for(k=0; k<K; k++){
+				result += A[i+k*lda] * B[k+j*ldb];
+			}
+#ifdef B0
+			C[i+j*ldc]=alpha * result;
+#else
+			C[i+j*ldc]=C[i+j*ldc] * beta + alpha * result;
+#endif
+		}
+	}
+	return 0;
+}
--- a/kernel/generic/gemm_small_matrix_kernel_nt.c
+++ b/kernel/generic/gemm_small_matrix_kernel_nt.c
@ -0,0 +1,56 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+#ifdef B0
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, IFLOAT * A, BLASLONG lda, FLOAT alpha, IFLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
+#else
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, IFLOAT * A, BLASLONG lda, FLOAT alpha, IFLOAT * B, BLASLONG ldb, FLOAT beta, FLOAT * C, BLASLONG ldc)
+#endif
+{
+	//naive implemtation
+	//Column major
+
+	BLASLONG i,j,k;
+	FLOAT result=0.0;
+
+	for(i=0; i<M; i++){
+		for(j=0; j<N; j++){
+			result=0.0;
+			for(k=0; k<K; k++){
+				result += A[i+k*lda] * B[k*ldb+j];
+			}
+#ifdef B0
+			C[i+j*ldc]=alpha * result;
+#else
+			C[i+j*ldc]=C[i+j*ldc] * beta + alpha * result;
+#endif
+		}
+	}
+	return 0;
+}
--- a/kernel/generic/gemm_small_matrix_kernel_tn.c
+++ b/kernel/generic/gemm_small_matrix_kernel_tn.c
@ -0,0 +1,57 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+#ifdef B0
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, IFLOAT * A, BLASLONG lda, FLOAT alpha, IFLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
+#else
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, IFLOAT * A, BLASLONG lda, FLOAT alpha, IFLOAT * B, BLASLONG ldb, FLOAT beta, FLOAT * C, BLASLONG ldc)
+#endif
+{
+	//naive implemtation
+	//Column major
+
+	BLASLONG i,j,k;
+	FLOAT result=0.0;
+
+	for(i=0; i<M; i++){
+		for(j=0; j<N; j++){
+			result=0.0;
+			for(k=0; k<K; k++){
+				result += A[i*lda+k] * B[k+j*ldb];
+			}
+#ifdef B0
+			C[i+j*ldc]=alpha * result;
+#else
+			C[i+j*ldc]=C[i+j*ldc] * beta + alpha * result;
+#endif
+		}
+	}
+
+	return 0;
+}
--- a/kernel/generic/gemm_small_matrix_kernel_tt.c
+++ b/kernel/generic/gemm_small_matrix_kernel_tt.c
@ -0,0 +1,57 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+#ifdef B0
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, IFLOAT * A, BLASLONG lda, FLOAT alpha, IFLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
+#else
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, IFLOAT * A, BLASLONG lda, FLOAT alpha, IFLOAT * B, BLASLONG ldb, FLOAT beta, FLOAT * C, BLASLONG ldc)
+#endif
+{
+	//naive implemtation
+	//Column major
+
+	BLASLONG i,j,k;
+	FLOAT result=0.0;
+
+	for(i=0; i<M; i++){
+		for(j=0; j<N; j++){
+			result=0.0;
+			for(k=0; k<K; k++){
+				result += A[i*lda+k] * B[k*ldb+j];
+			}
+#ifdef B0
+			C[i+j*ldc]=alpha * result;
+#else
+			C[i+j*ldc]=C[i+j*ldc] * beta + alpha * result;
+#endif
+		}
+	}
+
+	return 0;
+}
--- a/kernel/generic/gemm_small_matrix_permit.c
+++ b/kernel/generic/gemm_small_matrix_permit.c
@ -0,0 +1,40 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+int CNAME(int transa, int transb, BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, FLOAT beta)
+{
+	return 0;
+/*
+	double MNK = (double) M * (double) N * (double) K;
+	if (MNK <= 100.0*100.0*100.0)
+		return 1;
+	else
+		return 0;
+*/
+}
--- a/kernel/generic/zgemm_small_matrix_kernel_nn.c
+++ b/kernel/generic/zgemm_small_matrix_kernel_nn.c
@ -0,0 +1,89 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+#ifndef B0
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha0, FLOAT alpha1, FLOAT * B, BLASLONG ldb, FLOAT beta0, FLOAT beta1, FLOAT * C, BLASLONG ldc)
+#else
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha0, FLOAT alpha1, FLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
+#endif
+{
+	FLOAT real, imag;
+#ifndef B0
+	FLOAT tmp0, tmp1;
+#endif
+	int i, j, l;
+	for(i = 0; i < M; i++){
+		for(j = 0; j < N; j++){
+			real=0;
+			imag=0;
+
+			for(l = 0; l < K; l++){
+#if defined(NN)
+				real += (A[l*2*lda + 2*i]*B[j*2*ldb + 2*l]      
+					 -A[l*2*lda + 2*i + 1] * B[j*2*ldb + 2*l + 1]);
+
+				imag+=(A[l*2*lda + 2*i] * B[j*2*ldb + 2*l + 1]        
+				       + A[l*2*lda + 2*i + 1] * B[j*2*ldb + 2*l]);
+#elif defined(NR)
+				real += (A[l*2*lda + 2*i]*B[j*2*ldb + 2*l]      
+					 +A[l*2*lda + 2*i + 1] * B[j*2*ldb + 2*l + 1]);
+
+				imag+=(-A[l*2*lda + 2*i] * B[j*2*ldb + 2*l + 1]      
+				       + A[l*2*lda + 2*i + 1] * B[j*2*ldb + 2*l]);
+#elif defined(RN)
+				real += (A[l*2*lda + 2*i]*B[j*2*ldb + 2*l]
+					 +A[l*2*lda + 2*i + 1] * B[j*2*ldb + 2*l + 1]);
+
+				imag+=(A[l*2*lda + 2*i] * B[j*2*ldb + 2*l + 1]
+				       - A[l*2*lda + 2*i + 1] * B[j*2*ldb + 2*l]);
+#elif defined(RR)
+				real += (A[l*2*lda + 2*i]*B[j*2*ldb + 2*l]
+					 -A[l*2*lda + 2*i + 1] * B[j*2*ldb + 2*l + 1]);
+
+				imag+=(-A[l*2*lda + 2*i] * B[j*2*ldb + 2*l + 1]
+				       - A[l*2*lda + 2*i + 1] * B[j*2*ldb + 2*l]);
+#endif
+			}
+
+#ifndef B0
+			tmp0 = beta0*C[j*2*ldc + 2*i] - beta1*C[j*2*ldc+ 2*i + 1];
+			tmp1 = beta0*C[j*2*ldc+ 2*i + 1] + beta1*C[j*2*ldc + 2*i];
+
+
+			C[j*2*ldc + 2*i] =tmp0+ alpha0*real - alpha1*imag;
+			C[j*2*ldc+ 2*i + 1] = tmp1+ alpha0*imag + real*alpha1;
+#else
+			C[j*2*ldc + 2*i] = alpha0*real - alpha1*imag;
+			C[j*2*ldc+ 2*i + 1] = alpha0*imag + real*alpha1;
+#endif
+		}
+	}
+	
+	return 0;
+}
--- a/kernel/generic/zgemm_small_matrix_kernel_nt.c
+++ b/kernel/generic/zgemm_small_matrix_kernel_nt.c
@ -0,0 +1,93 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+#ifndef B0
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha0, FLOAT alpha1, FLOAT * B, BLASLONG ldb, FLOAT beta0, FLOAT beta1, FLOAT * C, BLASLONG ldc)
+#else
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha0, FLOAT alpha1, FLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
+#endif
+{
+	FLOAT real, imag;
+#ifndef B0
+	FLOAT tmp0, tmp1;
+#endif
+	int i, j, l;
+	for(i = 0; i < M; i++){
+		for(j = 0; j < N; j++){
+			real=0;
+			imag=0;
+
+			for(l = 0; l < K; l++){
+#if defined(NT)
+				real += (A[l*2*lda + 2*i]*B[l*2*ldb + 2*j]
+					 -A[l*2*lda + 2*i + 1] * B[l*2*ldb + 2*j + 1]);
+
+				imag+=(A[l*2*lda + 2*i] * B[l*2*ldb + 2*j + 1]    
+				       + A[l*2*lda + 2*i + 1] * B[l*2*ldb + 2*j]);
+
+#elif defined(NC)
+		                real += (A[l*2*lda + 2*i]*B[l*2*ldb + 2*j]          
+					 +A[l*2*lda + 2*i + 1] * B[l*2*ldb + 2*j + 1]);
+
+				imag+=(-A[l*2*lda + 2*i] * B[l*2*ldb + 2*j + 1]
+				       + A[l*2*lda + 2*i + 1] * B[l*2*ldb + 2*j]);
+
+#elif defined(RT)
+		                real += (A[l*2*lda + 2*i]*B[l*2*ldb + 2*j]
+					 +A[l*2*lda + 2*i + 1] * B[l*2*ldb + 2*j + 1]);
+
+				imag+=(A[l*2*lda + 2*i] * B[l*2*ldb + 2*j + 1]    
+				       - A[l*2*lda + 2*i + 1] * B[l*2*ldb + 2*j]);
+
+#elif defined(RC)
+		                real += (A[l*2*lda + 2*i]*B[l*2*ldb + 2*j]
+					 -A[l*2*lda + 2*i + 1] * B[l*2*ldb + 2*j + 1]);
+
+				imag+=(-A[l*2*lda + 2*i] * B[l*2*ldb + 2*j + 1]
+				       - A[l*2*lda + 2*i + 1] * B[l*2*ldb + 2*j]);
+
+#endif
+			}
+
+#ifndef B0
+			tmp0 = beta0*C[j*2*ldc + 2*i] - beta1*C[j*2*ldc+ 2*i + 1];
+			tmp1 = beta0*C[j*2*ldc+ 2*i + 1] + beta1*C[j*2*ldc + 2*i];
+
+
+			C[j*2*ldc + 2*i] =tmp0+ alpha0*real - alpha1*imag;
+			C[j*2*ldc+ 2*i + 1] = tmp1+ alpha0*imag + real*alpha1;
+#else
+			C[j*2*ldc + 2*i] = alpha0*real - alpha1*imag;
+			C[j*2*ldc+ 2*i + 1] = alpha0*imag + real*alpha1;
+#endif
+		}
+	}
+	
+	return 0;
+}
--- a/kernel/generic/zgemm_small_matrix_kernel_tn.c
+++ b/kernel/generic/zgemm_small_matrix_kernel_tn.c
@ -0,0 +1,93 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+#ifndef B0
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha0, FLOAT alpha1, FLOAT * B, BLASLONG ldb, FLOAT beta0, FLOAT beta1, FLOAT * C, BLASLONG ldc)
+#else
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha0, FLOAT alpha1, FLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
+#endif
+{
+	FLOAT real, imag;
+#ifndef B0
+	FLOAT tmp0, tmp1;
+#endif
+	int i, j, l;
+	for(i = 0; i < M; i++){
+		for(j = 0; j < N; j++){
+			real=0;
+			imag=0;
+
+			for(l = 0; l < K; l++){
+#if defined(TN)
+				real += (A[i*2*lda + 2*l]*B[j*2*ldb + 2*l]
+					 -A[i*2*lda + 2*l + 1] * B[j*2*ldb + 2*l + 1]);
+
+				imag+=(A[i*2*lda + 2*l] * B[j*2*ldb + 2*l + 1]
+				       + A[i*2*lda + 2*l + 1] * B[j*2*ldb + 2*l]);
+
+#elif defined(TR)
+				real += (A[i*2*lda + 2*l]*B[j*2*ldb + 2*l]
+					 +A[i*2*lda + 2*l + 1] * B[j*2*ldb + 2*l + 1]);
+
+				imag+=(-A[i*2*lda + 2*l] * B[j*2*ldb + 2*l + 1]
+				       + A[i*2*lda + 2*l + 1] * B[j*2*ldb + 2*l]);
+
+#elif defined(CN)
+				real += (A[i*2*lda + 2*l]*B[j*2*ldb + 2*l]
+					 +A[i*2*lda + 2*l + 1] * B[j*2*ldb + 2*l + 1]);
+
+				imag+=(A[i*2*lda + 2*l] * B[j*2*ldb + 2*l + 1]
+				       - A[i*2*lda + 2*l + 1] * B[j*2*ldb + 2*l]);
+
+#elif defined(CR)
+				real += (A[i*2*lda + 2*l]*B[j*2*ldb + 2*l]
+					 -A[i*2*lda + 2*l + 1] * B[j*2*ldb + 2*l + 1]);
+
+				imag+=(-A[i*2*lda + 2*l] * B[j*2*ldb + 2*l + 1]
+				       - A[i*2*lda + 2*l + 1] * B[j*2*ldb + 2*l]);
+
+#endif
+			}
+
+#ifndef B0
+			tmp0 = beta0*C[j*2*ldc + 2*i] - beta1*C[j*2*ldc+ 2*i + 1];
+			tmp1 = beta0*C[j*2*ldc+ 2*i + 1] + beta1*C[j*2*ldc + 2*i];
+
+
+			C[j*2*ldc + 2*i] =tmp0+ alpha0*real - alpha1*imag;
+			C[j*2*ldc+ 2*i + 1] = tmp1+ alpha0*imag + real*alpha1;
+#else
+			C[j*2*ldc + 2*i] = alpha0*real - alpha1*imag;
+			C[j*2*ldc+ 2*i + 1] = alpha0*imag + real*alpha1;
+#endif
+		}
+	}
+	
+	return 0;
+}
--- a/kernel/generic/zgemm_small_matrix_kernel_tt.c
+++ b/kernel/generic/zgemm_small_matrix_kernel_tt.c
@ -0,0 +1,93 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+#ifndef B0
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha0, FLOAT alpha1, FLOAT * B, BLASLONG ldb, FLOAT beta0, FLOAT beta1, FLOAT * C, BLASLONG ldc)
+#else
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alpha0, FLOAT alpha1, FLOAT * B, BLASLONG ldb, FLOAT * C, BLASLONG ldc)
+#endif
+{
+	FLOAT real, imag;
+#ifndef B0
+	FLOAT tmp0, tmp1;
+#endif
+	int i, j, l;
+	for(i = 0; i < M; i++){
+		for(j = 0; j < N; j++){
+			real=0;
+			imag=0;
+
+			for(l = 0; l < K; l++){
+#if defined(TT)
+				real += (A[i*2*lda + 2*l]*B[l*2*ldb + 2*j]
+					 -A[i*2*lda + 2*l + 1] * B[l*2*ldb + 2*j + 1]);
+
+				imag+=(A[i*2*lda + 2*l] * B[l*2*ldb + 2*j + 1]
+				       + A[i*2*lda + 2*l + 1] * B[l*2*ldb + 2*j]);
+
+#elif defined(TC)
+				real += (A[i*2*lda + 2*l]*B[l*2*ldb + 2*j]
+					 +A[i*2*lda + 2*l + 1] * B[l*2*ldb + 2*j + 1]);
+
+				imag+=(-A[i*2*lda + 2*l] * B[l*2*ldb + 2*j + 1]
+				       + A[i*2*lda + 2*l + 1] * B[l*2*ldb + 2*j]);
+
+#elif defined(CT)
+				real += (A[i*2*lda + 2*l]*B[l*2*ldb + 2*j]
+					 +A[i*2*lda + 2*l + 1] * B[l*2*ldb + 2*j + 1]);
+
+				imag+=(A[i*2*lda + 2*l] * B[l*2*ldb + 2*j + 1]
+				       - A[i*2*lda + 2*l + 1] * B[l*2*ldb + 2*j]);
+
+#elif defined(CC)
+				real += (A[i*2*lda + 2*l]*B[l*2*ldb + 2*j]
+					 -A[i*2*lda + 2*l + 1] * B[l*2*ldb + 2*j + 1]);
+
+				imag+=(-A[i*2*lda + 2*l] * B[l*2*ldb + 2*j + 1]
+				       - A[i*2*lda + 2*l + 1] * B[l*2*ldb + 2*j]);
+
+#endif
+			}
+
+#ifndef B0
+			tmp0 = beta0*C[j*2*ldc + 2*i] - beta1*C[j*2*ldc+ 2*i + 1];
+			tmp1 = beta0*C[j*2*ldc+ 2*i + 1] + beta1*C[j*2*ldc + 2*i];
+
+
+			C[j*2*ldc + 2*i] =tmp0+ alpha0*real - alpha1*imag;
+			C[j*2*ldc+ 2*i + 1] = tmp1+ alpha0*imag + real*alpha1;
+#else
+			C[j*2*ldc + 2*i] = alpha0*real - alpha1*imag;
+			C[j*2*ldc+ 2*i + 1] = alpha0*imag + real*alpha1;
+#endif
+		}
+	}
+	
+	return 0;
+}
--- a/kernel/generic/zgemm_small_matrix_permit.c
+++ b/kernel/generic/zgemm_small_matrix_permit.c
@ -0,0 +1,40 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+int CNAME(int transa, int transb, BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha0, FLOAT alpha1, FLOAT beta0, FLOAT beta1)
+{
+	return 0;
+/*
+	double MNK = (double) M * (double) N * (double) K;
+	if (MNK <= 100.0*100.0*100.0)
+		return 1;
+	else
+		return 0;
+*/
+}
--- a/kernel/loongarch64/KERNEL
+++ b/kernel/loongarch64/KERNEL
@ -0,0 +1,238 @@
+ifndef SAXPYKERNEL
+SAXPYKERNEL = ../arm/axpy.c
+endif
+
+ifndef DAXPYKERNEL
+DAXPYKERNEL = ../arm/axpy.c
+endif
+
+ifndef CAXPYKERNEL
+CAXPYKERNEL = ../arm/zaxpy.c
+endif
+
+ifndef ZAXPYKERNEL
+ZAXPYKERNEL = ../arm/zaxpy.c
+endif
+
+ifndef SROTKERNEL
+SROTKERNEL  = ../arm/rot.c
+endif
+
+ifndef DROTKERNEL
+DROTKERNEL  = ../arm/rot.c
+endif
+
+ifndef CROTKERNEL
+CROTKERNEL  = ../arm/zrot.c
+endif
+
+ifndef ZROTKERNEL
+ZROTKERNEL  = ../arm/zrot.c
+endif
+
+ifndef CSWAPKERNEL
+CSWAPKERNEL = ../arm/zswap.c
+endif
+
+ifndef ZSWAPKERNEL
+ZSWAPKERNEL = ../arm/zswap.c
+endif
+
+ifndef SSUMKERNEL
+SSUMKERNEL  = ../arm/sum.c
+endif
+
+ifndef DSUMKERNEL
+DSUMKERNEL  = ../arm/sum.c
+endif
+
+ifndef CSUMKERNEL
+CSUMKERNEL  = ../arm/zsum.c
+endif
+
+ifndef ZSUMKERNEL
+ZSUMKERNEL  = ../arm/zsum.c
+endif
+
+ifndef ISMAXKERNEL
+ISMAXKERNEL = ../arm/imax.c
+endif
+
+ifndef IDMAXKERNEL
+IDMAXKERNEL = ../arm/imax.c
+endif
+
+ifndef ISMINKERNEL
+ISMINKERNEL = ../arm/imin.c
+endif
+
+ifndef IDMINKERNEL
+IDMINKERNEL = ../arm/imin.c
+endif
+
+ifndef SNRM2KERNEL
+SNRM2KERNEL = snrm2.S
+endif
+
+ifndef DNRM2KERNEL
+DNRM2KERNEL = dnrm2.S
+endif
+
+ifndef CNRM2KERNEL
+CNRM2KERNEL = cnrm2.S
+endif
+
+ifndef ZNRM2KERNEL
+ZNRM2KERNEL = znrm2.S
+endif
+
+ifndef SCABS_KERNEL
+SCABS_KERNEL   = ../generic/cabs.c
+endif
+
+ifndef DCABS_KERNEL
+DCABS_KERNEL   = ../generic/cabs.c
+endif
+
+ifndef QCABS_KERNEL
+QCABS_KERNEL   = ../generic/cabs.c
+endif
+
+ifndef LSAME_KERNEL
+LSAME_KERNEL   = ../generic/lsame.c
+endif
+
+ifndef SGEMMKERNEL
+SGEMMKERNEL    =  gemm_kernel.S
+SGEMMINCOPY    = ../generic/gemm_ncopy_2.c
+SGEMMITCOPY    = ../generic/gemm_tcopy_2.c
+SGEMMONCOPY    = ../generic/gemm_ncopy_8.c
+SGEMMOTCOPY    = ../generic/gemm_tcopy_8.c
+SGEMMINCOPYOBJ =  sgemm_incopy.o
+SGEMMITCOPYOBJ =  sgemm_itcopy.o
+SGEMMONCOPYOBJ =  sgemm_oncopy.o
+SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+endif
+
+ifndef DGEMMKERNEL
+DGEMMKERNEL    =  gemm_kernel.S
+DGEMMINCOPY    = ../generic/gemm_ncopy_2.c
+DGEMMITCOPY    = ../generic/gemm_tcopy_2.c
+DGEMMONCOPY    = ../generic/gemm_ncopy_8.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_8.c
+DGEMMINCOPYOBJ =  dgemm_incopy.o
+DGEMMITCOPYOBJ =  dgemm_itcopy.o
+DGEMMONCOPYOBJ =  dgemm_oncopy.o
+DGEMMOTCOPYOBJ =  dgemm_otcopy.o
+endif
+
+ifndef CGEMMKERNEL
+CGEMMKERNEL    =  zgemm_kernel.S
+CGEMMINCOPY    = ../generic/zgemm_ncopy_1.c
+CGEMMITCOPY    = ../generic/zgemm_tcopy_1.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_4.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_4.c
+CGEMMINCOPYOBJ =  cgemm_incopy.o
+CGEMMITCOPYOBJ =  cgemm_itcopy.o
+CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+endif
+
+ifndef ZGEMMKERNEL
+ZGEMMKERNEL    =  zgemm_kernel.S
+ZGEMMINCOPY    = ../generic/zgemm_ncopy_1.c
+ZGEMMITCOPY    = ../generic/zgemm_tcopy_1.c
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_4.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_4.c
+ZGEMMINCOPYOBJ =  zgemm_incopy.o
+ZGEMMITCOPYOBJ =  zgemm_itcopy.o
+ZGEMMONCOPYOBJ =  zgemm_oncopy.o
+ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+endif
+
+ifndef SGEMM_BETA
+SGEMM_BETA = ../generic/gemm_beta.c
+endif
+ifndef DGEMM_BETA
+DGEMM_BETA = ../generic/gemm_beta.c
+endif
+ifndef CGEMM_BETA
+CGEMM_BETA = ../generic/zgemm_beta.c
+endif
+ifndef ZGEMM_BETA
+ZGEMM_BETA = ../generic/zgemm_beta.c
+endif
+
+ifndef STRSMKERNEL_LN
+STRSMKERNEL_LN =  trsm_kernel_LN.S
+endif
+
+ifndef STRSMKERNEL_LT
+STRSMKERNEL_LT =  trsm_kernel_LT.S
+endif
+
+ifndef STRSMKERNEL_RN
+STRSMKERNEL_RN =  trsm_kernel_LT.S
+endif
+
+ifndef STRSMKERNEL_RT
+STRSMKERNEL_RT =  trsm_kernel_RT.S
+endif
+
+ifndef DTRSMKERNEL_LN
+DTRSMKERNEL_LN =  trsm_kernel_LN.S
+endif
+
+ifndef DTRSMKERNEL_LT
+DTRSMKERNEL_LT =  trsm_kernel_LT.S
+endif
+
+ifndef DTRSMKERNEL_RN
+DTRSMKERNEL_RN =  trsm_kernel_LT.S
+endif
+
+ifndef DTRSMKERNEL_RT
+DTRSMKERNEL_RT =  trsm_kernel_RT.S
+endif
+
+ifndef CTRSMKERNEL_LN
+CTRSMKERNEL_LN =  ztrsm_kernel_LT.S
+endif
+
+ifndef CTRSMKERNEL_LT
+CTRSMKERNEL_LT =  ztrsm_kernel_LT.S
+endif
+
+ifndef CTRSMKERNEL_RN
+CTRSMKERNEL_RN =  ztrsm_kernel_LT.S
+endif
+
+ifndef CTRSMKERNEL_RT
+CTRSMKERNEL_RT =  ztrsm_kernel_RT.S
+endif
+
+ifndef ZTRSMKERNEL_LN
+ZTRSMKERNEL_LN =  ztrsm_kernel_LT.S
+endif
+
+ifndef ZTRSMKERNEL_LT
+ZTRSMKERNEL_LT =  ztrsm_kernel_LT.S
+endif
+
+ifndef ZTRSMKERNEL_RN
+ZTRSMKERNEL_RN =  ztrsm_kernel_LT.S
+endif
+
+ifndef ZTRSMKERNEL_RT
+ZTRSMKERNEL_RT =  ztrsm_kernel_RT.S
+endif
+
+ifndef CGEMM3MKERNEL
+CGEMM3MKERNEL    =  zgemm3m_kernel.S
+endif
+
+ifndef ZGEMM3MKERNEL
+ZGEMM3MKERNEL    =  zgemm3m_kernel.S
+endif
+
+DSDOTKERNEL  = dot.S
--- a/kernel/loongarch64/KERNEL.LOONGSON3R5
+++ b/kernel/loongarch64/KERNEL.LOONGSON3R5
@ -0,0 +1 @@
+#TODO: Add loongarch64 SIMD optimizations
--- a/kernel/loongarch64/KERNEL.generic
+++ b/kernel/loongarch64/KERNEL.generic
@ -0,0 +1,167 @@
+SGEMM_BETA = ../generic/gemm_beta.c
+DGEMM_BETA = ../generic/gemm_beta.c
+CGEMM_BETA = ../generic/zgemm_beta.c
+ZGEMM_BETA = ../generic/zgemm_beta.c
+
+STRMMKERNEL    = ../generic/trmmkernel_2x2.c
+DTRMMKERNEL    = ../generic/trmmkernel_2x2.c
+CTRMMKERNEL    = ../generic/ztrmmkernel_2x2.c
+ZTRMMKERNEL    = ../generic/ztrmmkernel_2x2.c
+
+SGEMMKERNEL    =  ../generic/gemmkernel_2x2.c
+SGEMMONCOPY    =  ../generic/gemm_ncopy_2.c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_2.c
+SGEMMONCOPYOBJ =  sgemm_oncopy.o
+SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+
+DGEMMKERNEL    =  ../generic/gemmkernel_2x2.c
+DGEMMONCOPY    = ../generic/gemm_ncopy_2.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy.o
+
+CGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+
+ZGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+ZGEMMONCOPYOBJ =  zgemm_oncopy.o
+ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+
+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
+
+#Pure C for other kernels
+SAMAXKERNEL  = ../arm/amax.c
+DAMAXKERNEL  = ../arm/amax.c
+CAMAXKERNEL  = ../arm/zamax.c
+ZAMAXKERNEL  = ../arm/zamax.c
+
+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
+
+ISAMAXKERNEL = ../arm/iamax.c
+IDAMAXKERNEL = ../arm/iamax.c
+ICAMAXKERNEL = ../arm/izamax.c
+IZAMAXKERNEL = ../arm/izamax.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
+
+SASUMKERNEL  = ../arm/asum.c
+DASUMKERNEL  = ../arm/asum.c
+CASUMKERNEL  = ../arm/zasum.c
+ZASUMKERNEL  = ../arm/zasum.c
+
+SSUMKERNEL   = ../arm/sum.c
+DSUMKERNEL   = ../arm/sum.c
+CSUMKERNEL   = ../arm/zsum.c
+ZSUMKERNEL   = ../arm/zsum.c
+
+
+SAXPYKERNEL  = ../arm/axpy.c
+DAXPYKERNEL  = ../arm/axpy.c
+CAXPYKERNEL  = ../arm/zaxpy.c
+ZAXPYKERNEL  = ../arm/zaxpy.c
+
+SCOPYKERNEL  = ../arm/copy.c
+DCOPYKERNEL  = ../arm/copy.c
+CCOPYKERNEL  = ../arm/zcopy.c
+ZCOPYKERNEL  = ../arm/zcopy.c
+
+SDOTKERNEL   = ../generic/dot.c
+DDOTKERNEL   = ../arm/dot.c
+CDOTKERNEL   = ../arm/zdot.c
+ZDOTKERNEL   = ../arm/zdot.c
+
+SNRM2KERNEL  = ../arm/nrm2.c
+DNRM2KERNEL  = ../arm/nrm2.c
+CNRM2KERNEL  = ../arm/znrm2.c
+ZNRM2KERNEL  = ../arm/znrm2.c
+
+SROTKERNEL   = ../arm/rot.c
+DROTKERNEL   = ../arm/rot.c
+CROTKERNEL   = ../arm/zrot.c
+ZROTKERNEL   = ../arm/zrot.c
+
+SSCALKERNEL  = ../arm/scal.c
+DSCALKERNEL  = ../arm/scal.c
+CSCALKERNEL  = ../arm/zscal.c
+ZSCALKERNEL  = ../arm/zscal.c
+
+SSWAPKERNEL  = ../arm/swap.c
+DSWAPKERNEL  = ../arm/swap.c
+CSWAPKERNEL  = ../arm/zswap.c
+ZSWAPKERNEL  = ../arm/zswap.c
+
+SGEMVNKERNEL = ../arm/gemv_n.c
+DGEMVNKERNEL = ../arm/gemv_n.c
+CGEMVNKERNEL = ../arm/zgemv_n.c
+ZGEMVNKERNEL = ../arm/zgemv_n.c
+
+SGEMVTKERNEL = ../arm/gemv_t.c
+DGEMVTKERNEL = ../arm/gemv_t.c
+CGEMVTKERNEL = ../arm/zgemv_t.c
+ZGEMVTKERNEL = ../arm/zgemv_t.c
+
+SSYMV_U_KERNEL =  ../generic/symv_k.c
+SSYMV_L_KERNEL =  ../generic/symv_k.c
+DSYMV_U_KERNEL =  ../generic/symv_k.c
+DSYMV_L_KERNEL =  ../generic/symv_k.c
+QSYMV_U_KERNEL =  ../generic/symv_k.c
+QSYMV_L_KERNEL =  ../generic/symv_k.c
+CSYMV_U_KERNEL =  ../generic/zsymv_k.c
+CSYMV_L_KERNEL =  ../generic/zsymv_k.c
+ZSYMV_U_KERNEL =  ../generic/zsymv_k.c
+ZSYMV_L_KERNEL =  ../generic/zsymv_k.c
+XSYMV_U_KERNEL =  ../generic/zsymv_k.c
+XSYMV_L_KERNEL =  ../generic/zsymv_k.c
+
+ZHEMV_U_KERNEL =  ../generic/zhemv_k.c
+ZHEMV_L_KERNEL =  ../generic/zhemv_k.c
+
+LSAME_KERNEL   = ../generic/lsame.c
+SCABS_KERNEL   = ../generic/cabs.c
+DCABS_KERNEL   = ../generic/cabs.c
+QCABS_KERNEL   = ../generic/cabs.c
+
+#Dump kernel
+CGEMM3MKERNEL  = ../generic/zgemm3mkernel_dump.c
+ZGEMM3MKERNEL  = ../generic/zgemm3mkernel_dump.c
--- a/kernel/loongarch64/Makefile
+++ b/kernel/loongarch64/Makefile
@ -0,0 +1 @@
+clean ::
--- a/kernel/loongarch64/amax.S
+++ b/kernel/loongarch64/amax.S
@ -0,0 +1,230 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+
+#define I      $r17
+#define TEMP   $r18
+
+#define a1 $f10
+#define a2 $f11
+#define a3 $f12
+#define a4 $f13
+#define a5 $f14
+#define a6 $f15
+#define a7 $f16
+#define a8 $f17
+
+#define t1 $f0
+#define t2 $f1
+#define t3 $f2
+#define t4 $f3
+
+#define s1 $f22
+#define s2 $f8
+#define s3 $f23
+#define s4 $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+
+   add.d   X, X, INCX
+   FABS    s1, a1
+
+   FABS   s2, a1
+   bge $r0,    N, .L999
+
+   FABS    s3, a1
+   srai.d  I, N, 3
+
+   FABS   s4, a1
+   bge $r0,    I, .L15
+
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+
+   CMPLT   $fcc0, s1, t1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+
+   CMPLT   $fcc2, s3, t3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+
+   CMPLT   $fcc2, s3, t3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+
+   CMOVT  s4,  s4,  t4,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+
+   FABS    t1, a5
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+
+   FABS    t1, a1
+
+   CMPLT   $fcc0, s1, t1
+
+   CMOVT  s1,  s1,  t1,  $fcc0
+
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+  move $r4, $r17
+  fmov.d $f0, $f22
+  jirl    $r0, $r1, 0x0
+
+  EPILOGUE
--- a/kernel/loongarch64/amin.S
+++ b/kernel/loongarch64/amin.S
@ -0,0 +1,186 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   add.d   X, X, INCX
+   FABS    s1, a1
+   FABS    s2, a1
+   bge     $r0,    N, .L999
+   FABS    s3, a1
+   srai.d  I, N, 3
+   FABS   s4, a1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, t1, s1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, t2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, t3, s3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, t4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, t1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, t2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, t3, s3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, t4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t2, s2
+   CMPLT   $fcc2, t3, s3
+   CMPLT   $fcc3, t4, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t2, s2
+   CMPLT   $fcc2, t3, s3
+   CMPLT   $fcc3, t4, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   .align 3
+.L15:
+   andi    I,  N, 7
+NOP
+   bge $r0,    I, .L998
+   .align  3
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   CMPLT   $fcc0, t1, s1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   EPILOGUE
--- a/kernel/loongarch64/asum.S
+++ b/kernel/loongarch64/asum.S
@ -0,0 +1,232 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1 $f23
+#define a2 $f9
+#define a3 $f10
+#define a4 $f11
+#define a5 $f12
+#define a6 $f13
+#define a7 $f14
+#define a8 $f15
+#define t1 $f16
+#define t2 $f17
+#define t3 $f0
+#define t4 $f1
+#define s1 $f22
+#define s2 $f8
+   PROLOGUE
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+   MTC  s1, $r0
+   MTC  s2, $r0
+   slli.d  INCX, INCX, BASE_SHIFT
+   li.d TEMP, SIZE
+   bge $r0,    N, .L999
+   srai.d  I, N, 3
+   bne INCX, TEMP, .L20
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   FABS    t1, a1
+   LD a6,  X,   5 * SIZE
+   FABS    t2, a2
+   LD a7,  X,   6 * SIZE
+   FABS    t3, a3
+   FABS    t4, a4
+   addi.d  I, I, -1
+   LD a8,  X,   7 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+.L12:
+   ADD s1, s1, t1
+   LD a1,  X,   8 * SIZE
+   FABS    t1, a5
+   addi.d  I, I, -1
+   ADD s2, s2, t2
+   LD a2,  X,   9 * SIZE
+   FABS    t2, a6
+   NOP
+   ADD s1, s1, t3
+   LD a3,  X,  10 * SIZE
+   FABS    t3, a7
+   NOP
+   ADD s2, s2, t4
+   LD a4,  X,  11 * SIZE
+   FABS    t4, a8
+   addi.d  X, X, 8 * SIZE
+   ADD s1, s1, t1
+   LD a5,  X,   4 * SIZE
+   FABS    t1, a1
+   NOP
+   ADD s2, s2, t2
+   LD a6,  X,   5 * SIZE
+   FABS    t2, a2
+   NOP
+   ADD s1, s1, t3
+   LD a7,  X,   6 * SIZE
+   FABS    t3, a3
+   NOP
+   ADD s2, s2, t4
+   LD a8,  X,   7 * SIZE
+   FABS   t4, a4
+   blt $r0,    I, .L12
+   .align 3
+.L13:
+   ADD s1, s1, t1
+   addi.d  X, X, 8 * SIZE
+   FABS    t1, a5
+   NOP
+   ADD s2, s2, t2
+   FABS    t2, a6
+   ADD s1, s1, t3
+   FABS    t3, a7
+   ADD s2, s2, t4
+   FABS    t4, a8
+   ADD s1, s1, t1
+   ADD s2, s2, t2
+   ADD s1, s1, t3
+   ADD s2, s2, t4
+   .align 3
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   ADD s1, s1, t1
+   addi.d X, X, SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+.L20:
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   LD a7,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a8,  X,   0 * SIZE
+   FABS    t4, a4
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L24
+   .align 3
+.L23:
+   ADD s1, s1, t1
+   LD a1,  X,   0 * SIZE
+   FABS    t1, a5
+   add.d   X, X, INCX
+   ADD s2, s2, t2
+   LD a2,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   ADD s1, s1, t3
+   LD a3,  X,   0 * SIZE
+   FABS    t3, a7
+   add.d   X, X, INCX
+   ADD s2, s2, t4
+   LD a4,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   ADD s1, s1, t1
+   LD a5,  X,   0 * SIZE
+   FABS    t1, a1
+   add.d   X, X, INCX
+   ADD s2, s2, t2
+   LD a6,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   ADD s1, s1, t3
+   LD a7,  X,   0 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   ADD s2, s2, t4
+   LD a8,  X,   0 * SIZE
+   FABS    t4, a4
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L23
+   .align 3
+.L24:
+   ADD s1, s1, t1
+   FABS    t1, a5
+   ADD s2, s2, t2
+   FABS    t2, a6
+   ADD s1, s1, t3
+   FABS    t3, a7
+   ADD s2, s2, t4
+   FABS    t4, a8
+   ADD s1, s1, t1
+   ADD s2, s2, t2
+   ADD s1, s1, t3
+   ADD s2, s2, t4
+   .align 3
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   add.d   X, X, INCX
+   ADD    s1, s1, t1
+   blt $r0,    I, .L26
+   .align 3
+.L999:
+   ADD s1, s1, s2
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   EPILOGUE
--- a/kernel/loongarch64/cnrm2.S
+++ b/kernel/loongarch64/cnrm2.S
@ -0,0 +1,159 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f12
+#define a2     $f13
+#define a3     $f14
+#define a4     $f15
+#define a5     $f16
+#define a6     $f17
+#define a7     $f0
+#define a8     $f1
+#define s1     $f22
+#define s2     $f8
+#define t1     $f23
+#define t2     $f9
+#define t3     $f10
+#define t4     $f11
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   movgr2fr.d  s1,  $r0
+   li.d  TEMP, 2 * SIZE
+   fmov.d s2, s1
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   srai.d  I, N, 2
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   fcvt.d.s    t1, a1
+   LD a7,  X,   0 * SIZE
+   fcvt.d.s    t2, a2
+   LD a8,  X,   1 * SIZE
+   fcvt.d.s    t3, a3
+   addi.d  I, I, -1
+   fcvt.d.s    t4, a4
+   add.d  X, X, INCX
+   bge $r0,    I, .L24
+   .align 3
+
+.L23:
+   fmadd.d  s1, t1, t1, s1
+   LD a1,  X,   0 * SIZE
+   fcvt.d.s    t1, a5
+   fmadd.d  s2, t2, t2, s2
+   LD a2,  X,   1 * SIZE
+   fcvt.d.s    t2, a6
+   add.d   X, X, INCX
+   fmadd.d  s1, t3, t3, s1
+   LD a3,  X,   0 * SIZE
+   fcvt.d.s    t3, a7
+   fmadd.d  s2, t4, t4, s2
+   LD a4,  X,   1 * SIZE
+   fcvt.d.s    t4, a8
+   add.d   X, X, INCX
+   fmadd.d  s1, t1, t1, s1
+   LD a5,  X,   0 * SIZE
+   fcvt.d.s    t1, a1
+   addi.d  I, I, -1
+   fmadd.d  s2, t2, t2, s2
+   LD a6,  X,   1 * SIZE
+   fcvt.d.s    t2, a2
+   add.d   X, X, INCX
+   fmadd.d  s1, t3, t3, s1
+   LD a7,  X,   0 * SIZE
+   fcvt.d.s    t3, a3
+   LD a8,  X,   1 * SIZE
+   fmadd.d  s2, t4, t4, s2
+   add.d   X, X, INCX
+   fcvt.d.s   t4, a4
+   blt $r0,    I, .L23
+   .align 3
+
+.L24:
+   fmadd.d  s1, t1, t1, s1
+   fcvt.d.s    t1, a5
+   fmadd.d  s2, t2, t2, s2
+   fcvt.d.s    t2, a6
+   fmadd.d  s1, t3, t3, s1
+   fcvt.d.s    t3, a7
+   fmadd.d  s2, t4, t4, s2
+   fcvt.d.s    t4, a8
+   fmadd.d  s1, t1, t1, s1
+   fmadd.d  s2, t2, t2, s2
+   fmadd.d  s1, t3, t3, s1
+   fmadd.d  s2, t4, t4, s2
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   fcvt.d.s    t2, a2
+   fmadd.d  s1, t1, t1, s1
+   add.d   X, X, INCX
+   fmadd.d  s2, t2, t2, s2
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   fadd.d  s1, s1, s2
+   fsqrt.d s1, s1
+   move $r4, $r17
+   fcvt.s.d    $f0, s1
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/copy.S
+++ b/kernel/loongarch64/copy.S
@ -0,0 +1,225 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define Y      $r7
+#define INCY   $r8
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+   LDINT   INCY,  0(INCY)
+#endif
+
+   li.d  TEMP, SIZE
+   NOP
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    N, .L999
+   slli.d INCY, INCY, BASE_SHIFT
+   bne INCX, TEMP, .L20
+   srai.d I, N, 3
+   bne INCY, TEMP, .L20
+   addi.d I, I, -1
+   blt     I, $r0,     .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   LD a7,  X,   6 * SIZE
+   LD a8,  X,   7 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   ST a1,  Y,   0 * SIZE
+   LD a1,  X,   8 * SIZE
+   ST a2,  Y,   1 * SIZE
+   LD a2,  X,   9 * SIZE
+   ST a3,  Y,   2 * SIZE
+   LD a3,  X,  10 * SIZE
+   ST a4,  Y,   3 * SIZE
+   LD a4,  X,  11 * SIZE
+   ST a5,  Y,   4 * SIZE
+   LD a5,  X,  12 * SIZE
+   ST a6,  Y,   5 * SIZE
+   LD a6,  X,  13 * SIZE
+   ST a7,  Y,   6 * SIZE
+   LD a7,  X,  14 * SIZE
+   ST a8,  Y,   7 * SIZE
+   LD a8,  X,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+   addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   ST a1,  Y,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   ST a3,  Y,   2 * SIZE
+   ST a4,  Y,   3 * SIZE
+   ST a5,  Y,   4 * SIZE
+   ST a6,  Y,   5 * SIZE
+   ST a7,  Y,   6 * SIZE
+   ST a8,  Y,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   addi.d  Y, Y, 8 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  X, X, SIZE
+   addi.d  I, I, -1
+   addi.d  Y, Y, SIZE
+   ST a1,  Y,  -1 * SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+
+.L20:
+   srai.d  I, N, 3
+   addi.d I, I, -1
+   blt I,  $r0, .L25
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   bge $r0,    I, .L23
+   .align 3
+
+.L22:
+   ST a1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a2,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a3,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a4,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a5,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a6,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a7,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a8,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   ST a1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a2,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a4,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a5,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a6,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a7,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a8,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   ST a1,  Y,   0 * SIZE
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/dnrm2.S
+++ b/kernel/loongarch64/dnrm2.S
@ -0,0 +1,314 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define XX     $r7
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define ALPHA  $f4
+#define max    $f5
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   move    XX, X
+   NOP
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   add.d   X, X, INCX
+   FABS    s1, a1
+   FABS   s2, a1
+   bge $r0,    N, .L999
+   FABS    s3, a1
+   srai.d  I, N, 3
+   FABS   s4, a1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, t3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, t3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   FABS    t1, a5
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t2,  $fcc1
+   CMOVT  s3,  s3,  t3,  $fcc2
+   CMOVT  s4,  s4,  t4,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L100
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   CMPLT   $fcc0, s1, t1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L100:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   addi.d  N, N, 1
+   lu12i.w TEMP, 0x3f800
+   movgr2fr.d      a1,   $r0
+   movgr2fr.w  ALPHA, TEMP
+   CMPEQ   $fcc0, s1, a1
+   fcvt.d.s   ALPHA, ALPHA
+   bcnez   $fcc0, .L999
+   fdiv.d  ALPHA, ALPHA, s1
+   MOV max, s1
+   MOV s1, a1
+   MOV s2, a1
+   MOV s3, a1
+   MOV s4, a1
+   srai.d  I, N, 3
+   bge $r0,    I, .L105
+   LD a1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a5,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a6,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a7,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD a8,  XX,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  XX, XX, INCX
+   bge $r0,    I, .L104
+   .align 3
+
+.L103:
+   MUL t1, ALPHA, a1
+   LD a1,  XX,   0 * SIZE
+   MUL t2, ALPHA, a2
+   add.d   XX, XX, INCX
+   MUL t3, ALPHA, a3
+   LD a2,  XX,   0 * SIZE
+   MUL t4, ALPHA, a4
+   add.d   XX, XX, INCX
+   MADD  s1, t1, t1, s1
+   LD a3,  XX,   0 * SIZE
+   MADD  s2, t2, t2, s2
+   add.d   XX, XX, INCX
+   MADD  s3, t3, t3, s3
+   LD a4,  XX,   0 * SIZE
+   MADD  s4, t4, t4, s4
+   add.d   XX, XX, INCX
+   MUL t1, ALPHA, a5
+   LD a5,  XX,   0 * SIZE
+   MUL t2, ALPHA, a6
+   add.d   XX, XX, INCX
+   MUL t3, ALPHA, a7
+   LD a6,  XX,   0 * SIZE
+   MUL t4, ALPHA, a8
+   add.d   XX, XX, INCX
+   MADD  s1, t1, t1, s1
+   LD a7,  XX,   0 * SIZE
+   MADD  s2, t2, t2, s2
+   add.d   XX, XX, INCX
+   MADD  s3, t3, t3, s3
+   LD a8,  XX,   0 * SIZE
+   MADD  s4, t4, t4, s4
+   addi.d  I, I, -1
+   add.d  XX, XX, INCX
+   blt $r0,    I, .L103
+   .align 3
+
+.L104:
+   MUL t1, ALPHA, a1
+   MUL t2, ALPHA, a2
+   MUL t3, ALPHA, a3
+   MUL t4, ALPHA, a4
+   MADD  s1, t1, t1, s1
+   MADD  s2, t2, t2, s2
+   MADD  s3, t3, t3, s3
+   MADD  s4, t4, t4, s4
+   MUL t1, ALPHA, a5
+   MUL t2, ALPHA, a6
+   MUL t3, ALPHA, a7
+   MUL t4, ALPHA, a8
+   MADD  s1, t1, t1, s1
+   MADD  s2, t2, t2, s2
+   MADD  s3, t3, t3, s3
+   MADD  s4, t4, t4, s4
+   .align 3
+
+.L105:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L106:
+   LD a1,  XX,   0 * SIZE
+   addi.d  I, I, -1
+   MUL t1, ALPHA, a1
+   add.d   XX, XX, INCX
+   MADD  s1, t1, t1, s1
+   blt $r0,    I, .L106
+   .align 3
+
+.L998:
+   ADD s1, s1, s2
+   ADD s3, s3, s4
+   ADD s1, s1, s3
+   fsqrt.d s1, s1
+   move $r4, $r17
+   MUL $f0, max, s1
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/dot.S
+++ b/kernel/loongarch64/dot.S
@ -0,0 +1,391 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define Y      $r7
+#define INCY   $r8
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f23
+#define a2     $f9
+#define a3     $f10
+#define a4     $f11
+#define b1     $f12
+#define b2     $f13
+#define b3     $f14
+#define b4     $f15
+#define s1     $f22
+#define s2     $f8
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+   LDINT   INCY,  0(INCY)
+#endif
+
+   MTC  s1, $r0
+   MTC  s2, $r0
+   slli.d  INCX, INCX, BASE_SHIFT
+   li.d  TEMP, SIZE
+   slli.d INCY, INCY, BASE_SHIFT
+   bge $r0,    N, .L999
+   srai.d I, N, 3
+   bne INCX, TEMP, .L20
+   bne INCY, TEMP, .L20
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b2,  Y,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD b3,  Y,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   addi.d  I, I, -1
+   LD b4,  Y,   3 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   4 * SIZE
+   LD b1,  Y,   4 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a2, a2
+   fcvt.d.s  b2, b2
+   fmadd.d  s2, b2, a2, s2
+#else
+   MADD  s2, b2, a2, s2
+#endif
+   LD a2,  X,   5 * SIZE
+   LD b2,  Y,   5 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a3, a3
+   fcvt.d.s  b3, b3
+   fmadd.d  s1, b3, a3, s1
+#else
+   MADD  s1, b3, a3, s1
+#endif
+   LD a3,  X,   6 * SIZE
+   LD b3,  Y,   6 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a4, a4
+   fcvt.d.s  b4, b4
+   fmadd.d  s2, b4, a4, s2
+#else
+   MADD  s2, b4, a4, s2
+#endif
+   LD a4,  X,   7 * SIZE
+   LD b4,  Y,   7 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   8 * SIZE
+   LD b1,  Y,   8 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a2, a2
+   fcvt.d.s  b2, b2
+   fmadd.d  s2, b2, a2, s2
+#else
+   MADD  s2, b2, a2, s2
+#endif
+   LD a2,  X,   9 * SIZE
+   LD b2,  Y,   9 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a3, a3
+   fcvt.d.s  b3, b3
+   fmadd.d  s1, b3, a3, s1
+#else
+   MADD  s1, b3, a3, s1
+#endif
+   LD a3,  X,  10 * SIZE
+   LD b3,  Y,  10 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a4, a4
+   fcvt.d.s  b4, b4
+   fmadd.d  s2, b4, a4, s2
+#else
+   MADD  s2, b4, a4, s2
+#endif
+   LD a4,  X,  11 * SIZE
+   LD b4,  Y,  11 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+.L13:
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   4 * SIZE
+   LD b1,  Y,   4 * SIZE
+#ifdef DSDOT
+   fcvt.d.s a2, a2
+   fcvt.d.s b2, b2
+   fmadd.d  s2, b2, a2, s2
+#else
+   MADD  s2, b2, a2, s2
+#endif
+   LD a2,  X,   5 * SIZE
+   LD b2,  Y,   5 * SIZE
+#ifdef DSDOT
+   fcvt.d.s a3, a3
+   fcvt.d.s b3, b3
+   fmadd.d  s1, b3, a3, s1
+#else
+   MADD  s1, b3, a3, s1
+#endif
+   LD a3,  X,   6 * SIZE
+   LD b3,  Y,   6 * SIZE
+#ifdef DSDOT
+   fcvt.d.s a4, a4
+   fcvt.d.s b4, b4
+   fmadd.d  s2, b4, a4, s2
+#else
+   MADD  s2, b4, a4, s2
+#endif
+   LD a4,  X,   7 * SIZE
+   LD b4,  Y,   7 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   addi.d  X, X, 8 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a2, a2
+   fcvt.d.s  b2, b2
+   fmadd.d  s2, b2, a2, s2
+#else
+   MADD  s2, b2, a2, s2
+#endif
+   addi.d  Y, Y, 8 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a3, a3
+   fcvt.d.s  b3, b3
+   fmadd.d  s1, b3, a3, s1
+#else
+   MADD  s1, b3, a3, s1
+#endif
+#ifdef DSDOT
+   fcvt.d.s  a4, a4
+   fcvt.d.s  b4, b4
+   fmadd.d  s2, b4, a4, s2
+#else
+   MADD  s2, b4, a4, s2
+#endif
+   .align 3
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   addi.d  I, I, -1
+   addi.d  X, X, SIZE
+   addi.d  Y, Y, SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+
+.L20:
+#ifdef F_INTERFACE
+   bgez    INCX, .L21
+   addi.d  TEMP, N, -1
+   mult    TEMP, INCX
+   mflo    TEMP
+   dsub    X, X, TEMP
+   .align 3
+
+.L21:
+   bgez    INCY, .L22
+   addi.d  TEMP, N, -1
+   mult    TEMP, INCY
+   mflo    TEMP
+   dsub    Y, Y, TEMP
+   .align 3
+
+.L22:
+#endif
+   bge $r0,    I, .L25
+   .align 3
+
+.L23:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s2, b1, a1, s2
+#else
+   MADD  s2, b1, a1, s2
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s2, b1, a1, s2
+#else
+   MADD  s2, b1, a1, s2
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s2, b1, a1, s2
+#else
+   MADD  s2, b1, a1, s2
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   addi.d  I, I, -1
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s2, b1, a1, s2
+#else
+   MADD  s2, b1, a1, s2
+#endif
+   blt $r0,    I, .L23
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   addi.d  I, I, -1
+#ifdef DSDOT
+   fcvt.d.s  a1, a1
+   fcvt.d.s  b1, b1
+   fmadd.d  s1, b1, a1, s1
+#else
+   MADD  s1, b1, a1, s1
+#endif
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+#ifdef DSDOT
+   fadd.d $f0, s1, s2
+#else
+   ADD    $f0, s1, s2
+#endif
+   move $r4, $r17
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/gemm_kernel.S
+++ b/kernel/loongarch64/gemm_kernel.S
--- a/kernel/loongarch64/gemv_n.S
+++ b/kernel/loongarch64/gemv_n.S
@ -0,0 +1,531 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+/* Unused param dummy1 */
+#define M      $r4
+#define N      $r5
+#define A      $r7
+#define LDA    $r8
+#define X      $r9
+#define INCX   $r10
+#define Y      $r11
+#define INCY   $r6
+#define BUFFER $r16
+#define YORIG  $r18
+#define XX     $r12
+#define YY     $r13
+#define I      $r14
+#define J      $r15
+#define AO1    $r23
+#define AO2    $r24
+#define ALPHA  $f0
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+#define x1     $f14
+#define x2     $f15
+#define y1     $f16
+#define y2     $f17
+#define y3     $f3
+#define y4     $f1
+#define y5     $f2
+#define y6     $f4
+#define y7     $f5
+#define y8     $f6
+#define t1     $f7
+#define t2     $f18
+#define t3     $f19
+#define t4     $f20
+
+   PROLOGUE
+
+   LDARG  INCY,    $sp,  0
+   LDARG  BUFFER,  $sp,  8
+#ifdef __64BIT__
+   addi.d  $sp, $sp, -16
+#else
+   addi.d  $sp, $sp, -48
+#endif
+   SDARG  $r23,  $sp,    0
+   SDARG  $r24,  $sp,    8
+   slli.d     LDA,     LDA,  BASE_SHIFT
+#ifndef __64BIT__
+   fst.d  $f18,  $sp,  16
+   fst.d  $f19,  $sp,  24
+   fst.d  $f20,  $sp,  32
+#endif
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    M, .L999
+   slli.d INCY, INCY, BASE_SHIFT
+   bge $r0,    N, .L999
+   li.d  I, SIZE
+   move   YORIG, Y
+   beq INCY, I, .L10
+   srai.d  I,  M, 2
+   move    YORIG, BUFFER
+   move    XX, Y
+   move   YY, BUFFER
+   bge $r0,    I, .L05
+   .align 3
+
+.L02:
+   LD a1,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   LD a2,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   LD a3,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   LD a4,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   ST a1,  YY,  0 * SIZE
+   ST a2,  YY,  1 * SIZE
+   ST a3,  YY,  2 * SIZE
+   ST a4,  YY,  3 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 4 * SIZE
+   blt $r0,    I, .L02
+   .align 3
+
+.L05:
+   andi    I,  M, 3
+   bge $r0,    I, .L10
+   .align 3
+
+.L06:
+   LD a1,  XX,  0 * SIZE
+   add.d   XX, XX, INCY
+   ST a1,  YY,  0 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 1 * SIZE
+   blt $r0,    I, .L06
+   .align 3
+
+.L10:
+   srai.d  J,  N, 1
+   bge $r0,    J, .L20
+   .align 3
+
+.L11:
+   LD x1,  X,  0 * SIZE
+   add.d   X, X, INCX
+   LD x2,  X,  0 * SIZE
+   add.d   X, X, INCX
+   move    AO1, A
+   add.d   AO2, A,      LDA
+   add.d   A,   AO2,    LDA
+   move    YY, YORIG
+   MUL x1, ALPHA, x1
+   srai.d  I,  M, 3
+   MUL    x2, ALPHA, x2
+   bge $r0,    I, .L15
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   LD y5,  YY,  4 * SIZE
+   LD a6,  AO2,  1 * SIZE
+   LD y6,  YY,  5 * SIZE
+   LD a7,  AO2,  2 * SIZE
+   LD y7,  YY,  6 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   addi.d  I, I, -1
+   LD y8,  YY,  7 * SIZE
+   bge $r0,    I, .L13
+   .align  3
+.L12:
+   MADD  t1, a1, x1, y1
+   LD a1,  AO1,   4 * SIZE
+   MADD  t2, a2, x1, y2
+   LD a2,  AO1,   5 * SIZE
+   LD y1,  YY,   8 * SIZE
+   LD y2,  YY,   9 * SIZE
+   MADD  t3, a3, x1, y3
+   LD a3,  AO1,   6 * SIZE
+   MADD  t4, a4, x1, y4
+   LD a4,  AO1,   7 * SIZE
+   LD y3,  YY,  10 * SIZE
+   LD y4,  YY,  11 * SIZE
+   MADD  t1, a5, x2, t1
+   LD a5,  AO2,   4 * SIZE
+   MADD  t2, a6, x2, t2
+   LD a6,  AO2,   5 * SIZE
+   MADD  t3, a7, x2, t3
+   LD a7,  AO2,   6 * SIZE
+   MADD  t4, a8, x2, t4
+   LD a8,  AO2,   7 * SIZE
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   ST t3,  YY,   2 * SIZE
+   ST t4,  YY,   3 * SIZE
+   MADD  t1, a1, x1, y5
+   LD a1,  AO1,   8 * SIZE
+   MADD  t2, a2, x1, y6
+   LD a2,  AO1,   9 * SIZE
+   LD y5,  YY,  12 * SIZE
+   LD y6,  YY,  13 * SIZE
+   MADD  t3, a3, x1, y7
+   LD a3,  AO1,  10 * SIZE
+   MADD  t4, a4, x1, y8
+   LD a4,  AO1,  11 * SIZE
+   LD y7,  YY,  14 * SIZE
+   LD y8,  YY,  15 * SIZE
+   MADD  t1, a5, x2, t1
+   LD a5,  AO2,   8 * SIZE
+   MADD  t2, a6, x2, t2
+   LD a6,  AO2,   9 * SIZE
+   MADD  t3, a7, x2, t3
+   LD a7,  AO2,  10 * SIZE
+   MADD  t4, a8, x2, t4
+   LD a8,  AO2,  11 * SIZE
+   ST t1,  YY,  4 * SIZE
+   ST t2,  YY,  5 * SIZE
+   ST t3,  YY,  6 * SIZE
+   ST t4,  YY,  7 * SIZE
+   addi.d  I, I, -1
+   addi.d  YY,  YY,   8 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   addi.d AO2, AO2,  8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   MADD  t1, a1, x1, y1
+   LD a1,  AO1,   4 * SIZE
+   MADD  t2, a2, x1, y2
+   LD a2,  AO1,   5 * SIZE
+   MADD  t3, a3, x1, y3
+   LD a3,  AO1,   6 * SIZE
+   MADD  t4, a4, x1, y4
+   LD a4,  AO1,   7 * SIZE
+   MADD  t1, a5, x2, t1
+   LD a5,  AO2,   4 * SIZE
+   MADD  t2, a6, x2, t2
+   LD a6,  AO2,   5 * SIZE
+   MADD  t3, a7, x2, t3
+   LD a7,  AO2,   6 * SIZE
+   MADD  t4, a8, x2, t4
+   LD a8,  AO2,   7 * SIZE
+   ST t1,  YY,   0 * SIZE
+   MADD  t1, a1, x1, y5
+   ST t2,  YY,   1 * SIZE
+   MADD  t2, a2, x1, y6
+   ST t3,  YY,   2 * SIZE
+   MADD  t3, a3, x1, y7
+   ST t4,  YY,   3 * SIZE
+   MADD  t4, a4, x1, y8
+   MADD  t1, a5, x2, t1
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD  t2, a6, x2, t2
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD  t3, a7, x2, t3
+   addi.d  YY,  YY,   8 * SIZE
+   MADD  t4, a8, x2, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  M, 4
+   bge $r0,    I, .L16
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   MADD  y1, a1, x1, y1
+   LD a6,  AO2,  1 * SIZE
+   MADD  y2, a2, x1, y2
+   LD a7,  AO2,  2 * SIZE
+   MADD  y3, a3, x1, y3
+   LD a8,  AO2,  3 * SIZE
+   MADD  y4, a4, x1, y4
+   MADD  y1, a5, x2, y1
+   addi.d  YY,  YY,   4 * SIZE
+   MADD  y2, a6, x2, y2
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD  y3, a7, x2, y3
+   addi.d  AO2, AO2,  4 * SIZE
+   MADD  y4, a8, x2, y4
+   ST y1,  YY,  -4 * SIZE
+   ST y2,  YY,  -3 * SIZE
+   ST y3,  YY,  -2 * SIZE
+   ST y4,  YY,  -1 * SIZE
+   .align 3
+
+.L16:
+   andi    I,  M, 2
+   bge $r0,    I, .L17
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   LD a6,  AO2,  1 * SIZE
+   MADD  y1, a1, x1, y1
+   MADD  y2, a2, x1, y2
+   addi.d  YY,  YY,   2 * SIZE
+   MADD  y1, a5, x2, y1
+   addi.d  AO1, AO1,  2 * SIZE
+   MADD  y2, a6, x2, y2
+   addi.d  AO2, AO2,  2 * SIZE
+   ST y1,  YY,  -2 * SIZE
+   ST y2,  YY,  -1 * SIZE
+   .align 3
+
+.L17:
+   andi    I,  M, 1
+   bge $r0,    I, .L19
+   LD y1,  YY,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   MADD  y1, a1, x1, y1
+   MADD  y1, a5, x2, y1
+   ST y1,  YY,  0 * SIZE
+   .align 3
+
+.L19:
+   addi.d  J, J, -1
+   blt $r0,    J, .L11
+   .align 3
+
+.L20:
+   andi    J,  N, 1
+   bge $r0,    J, .L900
+   .align 3
+
+.L21:
+   LD x1,  X,  0 * SIZE
+   add.d   X, X, INCX
+   move    YY, YORIG
+   move    AO1, A
+   srai.d  I,  M, 3
+   MUL    x1, ALPHA, x1
+   bge $r0,    I, .L25
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD y5,  YY,  4 * SIZE
+   LD y6,  YY,  5 * SIZE
+   LD y7,  YY,  6 * SIZE
+   addi.d  I, I, -1
+   LD y8,  YY,  7 * SIZE
+   bge $r0,    I, .L23
+   .align  3
+.L22:
+   MADD  t1, a1, x1, y1
+   LD a1,  AO1,   4 * SIZE
+   MADD  t2, a2, x1, y2
+   LD a2,  AO1,   5 * SIZE
+   LD y1,  YY,   8 * SIZE
+   LD y2,  YY,   9 * SIZE
+   MADD  t3, a3, x1, y3
+   LD a3,  AO1,   6 * SIZE
+   MADD  t4, a4, x1, y4
+   LD a4,  AO1,   7 * SIZE
+   LD y3,  YY,  10 * SIZE
+   LD y4,  YY,  11 * SIZE
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   ST t3,  YY,   2 * SIZE
+   ST t4,  YY,   3 * SIZE
+   MADD  t1, a1, x1, y5
+   LD a1,  AO1,   8 * SIZE
+   MADD  t2, a2, x1, y6
+   LD a2,  AO1,   9 * SIZE
+   LD y5,  YY,  12 * SIZE
+   LD y6,  YY,  13 * SIZE
+   MADD  t3, a3, x1, y7
+   LD a3,  AO1,  10 * SIZE
+   MADD  t4, a4, x1, y8
+   LD a4,  AO1,  11 * SIZE
+   LD y7,  YY,  14 * SIZE
+   LD y8,  YY,  15 * SIZE
+   ST t1,  YY,  4 * SIZE
+   ST t2,  YY,  5 * SIZE
+   ST t3,  YY,  6 * SIZE
+   ST t4,  YY,  7 * SIZE
+   addi.d  I, I, -1
+   addi.d  YY,  YY,   8 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   MADD  t1, a1, x1, y1
+   LD a1,  AO1,   4 * SIZE
+   MADD  t2, a2, x1, y2
+   LD a2,  AO1,   5 * SIZE
+   MADD  t3, a3, x1, y3
+   LD a3,  AO1,   6 * SIZE
+   MADD  t4, a4, x1, y4
+   LD a4,  AO1,   7 * SIZE
+   ST t1,  YY,   0 * SIZE
+   MADD  t1, a1, x1, y5
+   ST t2,  YY,   1 * SIZE
+   MADD  t2, a2, x1, y6
+   ST t3,  YY,   2 * SIZE
+   MADD  t3, a3, x1, y7
+   ST t4,  YY,   3 * SIZE
+   MADD  t4, a4, x1, y8
+   ST t1,  YY,   4 * SIZE
+   ST t2,  YY,   5 * SIZE
+   ST t3,  YY,   6 * SIZE
+   ST t4,  YY,   7 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   addi.d  YY,  YY,   8 * SIZE
+   .align 3
+
+.L25:
+   andi    I,  M, 4
+   bge $r0,    I, .L26
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   MADD  y1, a1, x1, y1
+   MADD  y2, a2, x1, y2
+   MADD  y3, a3, x1, y3
+   addi.d  YY,  YY,   4 * SIZE
+   MADD  y4, a4, x1, y4
+   addi.d  AO1, AO1,  4 * SIZE
+   ST y1,  YY,  -4 * SIZE
+   ST y2,  YY,  -3 * SIZE
+   ST y3,  YY,  -2 * SIZE
+   ST y4,  YY,  -1 * SIZE
+   .align 3
+
+.L26:
+   andi    I,  M, 2
+   bge $r0,    I, .L27
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   MADD  y1, a1, x1, y1
+   addi.d  YY,  YY,   2 * SIZE
+   MADD  y2, a2, x1, y2
+   addi.d  AO1, AO1,  2 * SIZE
+   ST y1,  YY,  -2 * SIZE
+   ST y2,  YY,  -1 * SIZE
+   .align 3
+
+.L27:
+   andi    I,  M, 1
+   bge $r0,    I, .L900
+   LD y1,  YY,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   MADD  y1, a1, x1, y1
+   ST y1,  YY,  0 * SIZE
+   .align 3
+
+.L900:
+   li.d  YORIG, SIZE
+   srai.d I,  M, 2
+   beq INCY, YORIG, .L999
+   move   XX, BUFFER
+   bge $r0,    I, .L905
+   .align 3
+
+.L902:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   LD a3,  XX,  2 * SIZE
+   LD a4,  XX,  3 * SIZE
+   ST a1,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   ST a2,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   ST a4,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   addi.d  I, I, -1
+   addi.d XX, XX, 4 * SIZE
+   blt $r0,    I, .L902
+   .align 3
+
+.L905:
+   andi    I,  M, 3
+   bge $r0,    I, .L999
+   .align 3
+
+.L906:
+   LD a1,  XX,  0 * SIZE
+   addi.d  XX, XX, 1 * SIZE
+   ST a1,  Y,  0 * SIZE
+   addi.d  I, I, -1
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L906
+   .align 3
+
+.L999:
+   LDARG  $r23,  $sp,    0
+   LDARG  $r24,  $sp,    8
+#ifndef __64BIT__
+   fld.d  $f18,  $sp,  16
+   fld.d  $f19,  $sp,  24
+   fld.d  $f20,  $sp,  32
+#endif
+#ifdef __64BIT__
+   addi.d  $sp, $sp, 16
+#else
+   addi.d  $sp, $sp, 48
+#endif
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/gemv_t.S
+++ b/kernel/loongarch64/gemv_t.S
@ -0,0 +1,436 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+/* Unused param dummy1 */
+#define M      $r4
+#define N      $r5
+#define A      $r7
+#define LDA    $r8
+#define X      $r9
+#define INCX   $r10
+#define Y      $r11
+#define INCY   $r6
+#define BUFFER $r16
+#define XORIG  $r18
+#define XX     $r12
+#define YY     $r13
+#define I      $r14
+#define J      $r15
+#define AO1    $r23
+#define AO2    $r24
+#define ALPHA  $f0
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+#define y1     $f14
+#define y2     $f15
+#define y3     $f16
+#define y4     $f17
+#define x1     $f3
+#define x2     $f1
+#define x3     $f2
+#define x4     $f4
+#define x5     $f5
+#define x6     $f6
+#define x7     $f7
+#define x8     $f18
+
+   PROLOGUE
+
+   LDARG  INCY,    $sp,  0
+   LDARG  BUFFER,  $sp,  8
+#ifdef __64BIT__
+   addi.d  $sp, $sp, -16
+#else
+   addi.d  $sp, $sp, -32
+#endif
+   MTC  y1, $r0
+   SDARG  $r23,  $sp,    0
+   SDARG  $r24,  $sp,    8
+   slli.d     LDA,     LDA,  BASE_SHIFT
+#ifndef __64BIT__
+   fst.d  $f18,  $sp,  16
+#endif
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    M, .L999
+   slli.d INCY, INCY, BASE_SHIFT
+   bge $r0,    N, .L999
+   li.d  I, SIZE
+   move   XORIG, X
+   beq INCX, I, .L10
+   srai.d  I,  M, 2
+   move    XORIG, BUFFER
+   move   YY, BUFFER
+   bge $r0,    I, .L05
+   .align 3
+
+.L02:
+   LD a1,  X,  0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,  0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,  0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,  0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  YY,  0 * SIZE
+   ST a2,  YY,  1 * SIZE
+   ST a3,  YY,  2 * SIZE
+   ST a4,  YY,  3 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 4 * SIZE
+   blt $r0,    I, .L02
+   .align 3
+
+.L05:
+   andi    I,  M, 3
+   bge $r0,    I, .L10
+   .align 3
+
+.L06:
+   LD a1,  X,  0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  YY,  0 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 1 * SIZE
+   blt $r0,    I, .L06
+   .align 3
+
+.L10:
+   srai.d  J,  N, 1
+   move   YY, Y
+   bge $r0,    J, .L20
+   .align 3
+
+.L11:
+   move    AO1, A
+   MOV y2, y1
+   add.d   AO2, A,      LDA
+   MOV y3, y1
+   add.d   A,   AO2,    LDA
+   MOV y4, y1
+   srai.d  I,  M, 3
+   move   XX, XORIG
+   bge $r0,    I, .L15
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO2,  0 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a3,  AO1,  1 * SIZE
+   LD x3,  XX,  2 * SIZE
+   LD a4,  AO2,  1 * SIZE
+   LD x4,  XX,  3 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x5,  XX,  4 * SIZE
+   LD a6,  AO2,  2 * SIZE
+   LD x6,  XX,  5 * SIZE
+   LD a7,  AO1,  3 * SIZE
+   LD x7,  XX,  6 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   addi.d  I, I, -1
+   LD x8,  XX,  7 * SIZE
+   bge $r0,    I, .L13
+   .align  3
+.L12:
+   MADD  y1, a1, x1, y1
+   LD a1,  AO1,  4 * SIZE
+   MADD  y2, a2, x1, y2
+   LD a2,  AO2,  4 * SIZE
+   MADD  y3, a3, x2, y3
+   LD a3,  AO1,  5 * SIZE
+   MADD  y4, a4, x2, y4
+   LD a4,  AO2,  5 * SIZE
+   LD x1,  XX,   8 * SIZE
+   LD x2,  XX,   9 * SIZE
+   MADD  y1, a5, x3, y1
+   LD a5,  AO1,  6 * SIZE
+   MADD  y2, a6, x3, y2
+   LD a6,  AO2,  6 * SIZE
+   MADD  y3, a7, x4, y3
+   LD a7,  AO1,  7 * SIZE
+   MADD  y4, a8, x4, y4
+   LD a8,  AO2,  7 * SIZE
+   LD x3,  XX,  10 * SIZE
+   LD x4,  XX,  11 * SIZE
+   MADD  y1, a1, x5, y1
+   LD a1,  AO1,  8 * SIZE
+   MADD  y2, a2, x5, y2
+   LD a2,  AO2,  8 * SIZE
+   MADD  y3, a3, x6, y3
+   LD a3,  AO1,  9 * SIZE
+   MADD  y4, a4, x6, y4
+   LD a4,  AO2,  9 * SIZE
+   LD x5,  XX,  12 * SIZE
+   LD x6,  XX,  13 * SIZE
+   MADD  y1, a5, x7, y1
+   LD a5,  AO1, 10 * SIZE
+   MADD  y2, a6, x7, y2
+   LD a6,  AO2, 10 * SIZE
+   MADD  y3, a7, x8, y3
+   LD a7,  AO1, 11 * SIZE
+   MADD  y4, a8, x8, y4
+   LD a8,  AO2, 11 * SIZE
+   LD x7,  XX,  14 * SIZE
+   LD x8,  XX,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX,  XX,   8 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   addi.d AO2, AO2,  8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   MADD  y1, a1, x1, y1
+   LD a1,  AO1,  4 * SIZE
+   MADD  y2, a2, x1, y2
+   LD a2,  AO2,  4 * SIZE
+   MADD  y3, a3, x2, y3
+   LD a3,  AO1,  5 * SIZE
+   MADD  y4, a4, x2, y4
+   LD a4,  AO2,  5 * SIZE
+   MADD  y1, a5, x3, y1
+   LD a5,  AO1,  6 * SIZE
+   MADD  y2, a6, x3, y2
+   LD a6,  AO2,  6 * SIZE
+   MADD  y3, a7, x4, y3
+   LD a7,  AO1,  7 * SIZE
+   MADD  y4, a8, x4, y4
+   LD a8,  AO2,  7 * SIZE
+   MADD  y1, a1, x5, y1
+   MADD  y2, a2, x5, y2
+   MADD  y3, a3, x6, y3
+   MADD  y4, a4, x6, y4
+   MADD  y1, a5, x7, y1
+   addi.d  XX,  XX,   8 * SIZE
+   MADD  y2, a6, x7, y2
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD  y3, a7, x8, y3
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD  y4, a8, x8, y4
+   .align 3
+
+.L15:
+   andi    I,  M, 4
+   bge $r0,    I, .L17
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO2,  0 * SIZE
+   LD a3,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a4,  AO2,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x3,  XX,  2 * SIZE
+   MADD  y1, a1, x1, y1
+   LD a6,  AO2,  2 * SIZE
+   MADD  y2, a2, x1, y2
+   LD a7,  AO1,  3 * SIZE
+   MADD  y3, a3, x2, y3
+   LD x4,  XX,  3 * SIZE
+   MADD  y4, a4, x2, y4
+   LD a8,  AO2,  3 * SIZE
+   MADD  y1, a5, x3, y1
+   MADD  y2, a6, x3, y2
+   addi.d  XX,  XX,   4 * SIZE
+   MADD  y3, a7, x4, y3
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD  y4, a8, x4, y4
+   addi.d  AO2, AO2,  4 * SIZE
+   .align 3
+
+.L17:
+   andi    I,  M, 3
+   ADD y1, y1, y3
+   ADD    y2, y2, y4
+   bge $r0,    I, .L19
+   .align  3
+.L18:
+   LD x1,  XX,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a2,  AO2,  0 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX, XX, 1 * SIZE
+   addi.d  AO1, AO1,  1 * SIZE
+   addi.d  AO2, AO2,  1 * SIZE
+   MADD  y1, a1, x1, y1
+   MADD  y2, a2, x1, y2
+   blt $r0,    I, .L18
+   .align 3
+
+.L19:
+   LD a1,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   LD a2,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   MADD  a1, y1, ALPHA, a1
+   addi.d  J, J, -1
+   MADD  a2, y2, ALPHA, a2
+   MTC  y1, $r0
+   ST a1,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST a2,  YY,   0 * SIZE
+   add.d  YY, YY, INCY
+   blt $r0,    J, .L11
+   .align 3
+
+.L20:
+   andi    J,  N, 1
+   MOV y3, y1
+   move   AO1, A
+   bge $r0,    J, .L999
+   srai.d  I,  M, 3
+   move   XX, XORIG
+   bge $r0,    I, .L25
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a3,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x3,  XX,  2 * SIZE
+   LD a7,  AO1,  3 * SIZE
+   LD x4,  XX,  3 * SIZE
+   LD x5,  XX,  4 * SIZE
+   LD x6,  XX,  5 * SIZE
+   LD x7,  XX,  6 * SIZE
+   addi.d  I, I, -1
+   LD x8,  XX,  7 * SIZE
+   bge $r0,    I, .L23
+   .align  3
+.L22:
+   MADD  y1, a1, x1, y1
+   LD a1,  AO1,  4 * SIZE
+   MADD  y3, a3, x2, y3
+   LD a3,  AO1,  5 * SIZE
+   LD x1,  XX,   8 * SIZE
+   LD x2,  XX,   9 * SIZE
+   MADD  y1, a5, x3, y1
+   LD a5,  AO1,  6 * SIZE
+   MADD  y3, a7, x4, y3
+   LD a7,  AO1,  7 * SIZE
+   LD x3,  XX,  10 * SIZE
+   LD x4,  XX,  11 * SIZE
+   MADD  y1, a1, x5, y1
+   LD a1,  AO1,   8 * SIZE
+   MADD  y3, a3, x6, y3
+   LD a3,  AO1,   9 * SIZE
+   LD x5,  XX,  12 * SIZE
+   LD x6,  XX,  13 * SIZE
+   MADD  y1, a5, x7, y1
+   LD a5,  AO1,  10 * SIZE
+   MADD  y3, a7, x8, y3
+   LD a7,  AO1,  11 * SIZE
+   LD x7,  XX,  14 * SIZE
+   LD x8,  XX,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX,  XX,   8 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   MADD  y1, a1, x1, y1
+   LD a1,  AO1,  4 * SIZE
+   MADD  y3, a3, x2, y3
+   LD a3,  AO1,  5 * SIZE
+   MADD  y1, a5, x3, y1
+   LD a5,  AO1,  6 * SIZE
+   MADD  y3, a7, x4, y3
+   LD a7,  AO1,  7 * SIZE
+   MADD  y1, a1, x5, y1
+   MADD  y3, a3, x6, y3
+   MADD  y1, a5, x7, y1
+   MADD  y3, a7, x8, y3
+   addi.d  XX,  XX,   8 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   .align 3
+
+.L25:
+   andi    I,  M, 4
+   bge $r0,    I, .L27
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a3,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x3,  XX,  2 * SIZE
+   MADD  y1, a1, x1, y1
+   LD a7,  AO1,  3 * SIZE
+   MADD  y3, a3, x2, y3
+   LD x4,  XX,  3 * SIZE
+   MADD  y1, a5, x3, y1
+   addi.d  XX,  XX,   4 * SIZE
+   MADD  y3, a7, x4, y3
+   addi.d  AO1, AO1,  4 * SIZE
+   .align 3
+
+.L27:
+   andi    I,  M, 3
+   ADD y1, y1, y3
+   bge $r0,    I, .L29
+   .align  3
+.L28:
+   LD x1,  XX,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX, XX, 1 * SIZE
+   addi.d  AO1, AO1,  1 * SIZE
+   MADD  y1, a1, x1, y1
+   blt $r0,    I, .L28
+   .align 3
+
+.L29:
+   LD a1,  Y,  0 * SIZE
+   add.d   Y, Y, INCY
+   MADD  a1, y1, ALPHA, a1
+   ST a1,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   .align 3
+
+.L999:
+   LDARG  $r23,  $sp,    0
+   LDARG  $r24,  $sp,    8
+#ifndef __64BIT__
+   fld.d  $f18,  $sp,  16
+#endif
+#ifdef __64BIT__
+   addi.d  $sp, $sp, 16
+#else
+   addi.d  $sp, $sp, 32
+#endif
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/iamax.S
+++ b/kernel/loongarch64/iamax.S
@ -0,0 +1,233 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r18
+#define TEMP   $r7
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define x1     $r17
+#define x2     $r8
+#define x3     $r9
+#define x4     $r10
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   li.d x1, 0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   li.d x1, 1
+   bge $r0,    N, .L999
+   FABS    s1, a1
+   add.d   X, X, INCX
+   FABS    s2, a1
+   li.d  x2, 1
+   FABS    s3, a1
+   srai.d  I, N, 3
+   FABS    s4, a1
+   li.d  x3, 1
+   li.d  TEMP, 2
+   li.d x4, 1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, t3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  I, I, -1
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, t3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, t4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d TEMP, TEMP, 4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   FABS    t1, a5
+   addi.d  TEMP, TEMP, 4
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t2
+   CMPLT   $fcc2, s3, t3
+   CMPLT   $fcc3, s4, t4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  x2, x2, 1
+   addi.d  x3, x3, 2
+   addi.d  x4, x4, 3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   addi.d  I, I, -1
+   CMPLT   $fcc0, s1, t1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   addi.d TEMP, TEMP, 1
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   MOVT(x1, x2, $fcc0)
+   CMOVT  s3,  s3,  s4,  $fcc1
+   MOVT(x3, x4, $fcc1)
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   MOVT(x1, x3, $fcc0)
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/iamin.S
+++ b/kernel/loongarch64/iamin.S
@ -0,0 +1,233 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r18
+#define TEMP   $r7
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define x1     $r17
+#define x2     $r8
+#define x3     $r9
+#define x4     $r10
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   li.d x1, 0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   li.d x1, 1
+   bge $r0,    N, .L999
+   FABS    s1, a1
+   add.d   X, X, INCX
+   FABS    s2, a1
+   li.d  x2, 1
+   FABS    s3, a1
+   srai.d  I, N, 3
+   FABS    s4, a1
+   li.d  x3, 1
+   li.d  TEMP, 2
+   li.d x4, 1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   FABS    t3, a3
+   LD a2,  X,   0 * SIZE
+   FABS    t4, a4
+   add.d   X, X, INCX
+   CMPLT   $fcc0, t1, s1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, t2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, t3, s3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, t4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  I, I, -1
+   FABS    t1, a5
+   LD a5,  X,   0 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   FABS    t3, a7
+   LD a6,  X,   0 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   CMPLT   $fcc0, t1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, t2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, t3, s3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, t4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d TEMP, TEMP, 4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t2, s2
+   CMPLT   $fcc2, t3, s3
+   CMPLT   $fcc3, t4, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   FABS    t1, a5
+   addi.d  TEMP, TEMP, 4
+   FABS    t2, a6
+   FABS    t3, a7
+   FABS    t4, a8
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t2, s2
+   CMPLT   $fcc2, t3, s3
+   CMPLT   $fcc3, t4, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t2,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t3,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t4,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  x2, x2, 1
+   addi.d  x3, x3, 2
+   addi.d  x4, x4, 3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   addi.d  I, I, -1
+   CMPLT   $fcc0, t1, s1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   addi.d TEMP, TEMP, 1
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   MOVT(x1, x2, $fcc0)
+   CMOVT  s3,  s3,  s4,  $fcc1
+   MOVT(x3, x4, $fcc1)
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   MOVT(x1, x3, $fcc0)
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/izamax.S
+++ b/kernel/loongarch64/izamax.S
@ -0,0 +1,217 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r18
+#define TEMP   $r7
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define t5     $f4
+#define t6     $f5
+#define t7     $f6
+#define t8     $f7
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define x1     $r17
+#define x2     $r8
+#define x3     $r9
+#define x4     $r10
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+      LDINT   N,     0(N)
+      LDINT   INCX,  0(INCX)
+#endif
+
+   li.d x1, 0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD s1, t1, t2
+   ADD s2, t1, t2
+   ADD s3, t1, t2
+   ADD s4, t1, t2
+   addi.d  N, N, -1
+   li.d x1, 1
+   bge $r0,    N, .L999
+   add.d   X, X, INCX
+   li.d  x2, 1
+   srai.d  I, N, 2
+   li.d  x3, 1
+   li.d  TEMP, 2
+   li.d x4, 1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   LD a2,  X,   1 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   FABS    t4, a4
+   FABS    t5, a5
+   LD a3,  X,   0 * SIZE
+   FABS    t6, a6
+   LD a4,  X,   1 * SIZE
+   FABS    t7, a7
+   add.d   X, X, INCX
+   FABS    t8, a8
+   ADD t1, t1, t2
+   LD a5,  X,   0 * SIZE
+   ADD t3, t3, t4
+   LD a6,  X,   1 * SIZE
+   ADD t5, t5, t6
+   add.d   X, X, INCX
+   ADD t7, t7, t8
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t3
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc2, s3, t5
+   add.d   X, X, INCX
+   CMPLT   $fcc3, s4, t7
+   addi.d  I, I, -1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t3,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t5,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t7,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d TEMP, TEMP, 4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   FABS    t5, a5
+   FABS    t6, a6
+   FABS    t7, a7
+   FABS    t8, a8
+   ADD t1, t1, t2
+   ADD t3, t3, t4
+   ADD t5, t5, t6
+   ADD t7, t7, t8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t3
+   CMPLT   $fcc2, s3, t5
+   CMPLT   $fcc3, s4, t7
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t3,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t5,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t7,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  x2, x2, 1
+   addi.d  x3, x3, 2
+   addi.d  x4, x4, 3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD t1, t1, t2
+   addi.d  I, I, -1
+   CMPLT   $fcc0, s1, t1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   addi.d TEMP, TEMP, 1
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   MOVT(x1, x2, $fcc0)
+   CMOVT  s3,  s3,  s4,  $fcc1
+   MOVT(x3, x4, $fcc1)
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   MOVT(x1, x3, $fcc0)
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/izamin.S
+++ b/kernel/loongarch64/izamin.S
@ -0,0 +1,217 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r18
+#define TEMP   $r7
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define t5     $f4
+#define t6     $f5
+#define t7     $f6
+#define t8     $f7
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+#define x1     $r17
+#define x2     $r8
+#define x3     $r9
+#define x4     $r10
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   li.d x1, 0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD s1, t1, t2
+   ADD s2, t1, t2
+   ADD s3, t1, t2
+   ADD s4, t1, t2
+   addi.d  N, N, -1
+   li.d x1, 1
+   bge $r0,    N, .L999
+   add.d   X, X, INCX
+   li.d  x2, 1
+   srai.d  I, N, 2
+   li.d  x3, 1
+   li.d  TEMP, 2
+   li.d x4, 1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   LD a2,  X,   1 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   FABS    t4, a4
+   FABS    t5, a5
+   LD a3,  X,   0 * SIZE
+   FABS    t6, a6
+   LD a4,  X,   1 * SIZE
+   FABS    t7, a7
+   add.d   X, X, INCX
+   FABS    t8, a8
+   ADD t1, t1, t2
+   LD a5,  X,   0 * SIZE
+   ADD t3, t3, t4
+   LD a6,  X,   1 * SIZE
+   ADD t5, t5, t6
+   add.d   X, X, INCX
+   ADD t7, t7, t8
+   CMPLT   $fcc0, t1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, t3, s2
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc2, t5, s3
+   add.d   X, X, INCX
+   CMPLT   $fcc3, t7, s4
+   addi.d  I, I, -1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t3,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t5,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t7,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d TEMP, TEMP, 4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   FABS    t5, a5
+   FABS    t6, a6
+   FABS    t7, a7
+   FABS    t8, a8
+   ADD t1, t1, t2
+   ADD t3, t3, t4
+   ADD t5, t5, t6
+   ADD t7, t7, t8
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t3, s2
+   CMPLT   $fcc2, t5, s3
+   CMPLT   $fcc3, t7, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   CMOVT  s2,  s2,  t3,  $fcc1
+   MOVT(x2, TEMP, $fcc1)
+   CMOVT  s3,  s3,  t5,  $fcc2
+   MOVT(x3, TEMP, $fcc2)
+   CMOVT  s4,  s4,  t7,  $fcc3
+   MOVT(x4, TEMP, $fcc3)
+   addi.d  TEMP, TEMP, 4
+   addi.d  x2, x2, 1
+   addi.d  x3, x3, 2
+   addi.d  x4, x4, 3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD t1, t1, t2
+   addi.d  I, I, -1
+   CMPLT   $fcc0, t1, s1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   MOVT(x1, TEMP, $fcc0)
+   addi.d TEMP, TEMP, 1
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   MOVT(x1, x2, $fcc0)
+   CMOVT  s3,  s3,  s4,  $fcc1
+   MOVT(x3, x4, $fcc1)
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   MOVT(x1, x3, $fcc0)
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/max.S
+++ b/kernel/loongarch64/max.S
@ -0,0 +1,174 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+      LDINT   N,     0(N)
+      LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD s1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   add.d   X, X, INCX
+   MOV    s2, s1
+   bge $r0,    N, .L999
+   MOV s3, s1
+   srai.d  I, N, 3
+   MOV    s4, s1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   CMPLT   $fcc0, s1, a1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, a2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, a3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, a4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a1,  $fcc0
+   LD a1,  X,   0 * SIZE
+   CMOVT  s2,  s2,  a2,  $fcc1
+   add.d   X, X, INCX
+   CMOVT  s3,  s3,  a3,  $fcc2
+   LD a2,  X,   0 * SIZE
+   CMOVT  s4,  s4,  a4,  $fcc3
+   add.d   X, X, INCX
+   CMPLT   $fcc0, s1, a5
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, a6
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, a7
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, a8
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a5,  $fcc0
+   LD a5,  X,   0 * SIZE
+   CMOVT  s2,  s2,  a6,  $fcc1
+   add.d   X, X, INCX
+   CMOVT  s3,  s3,  a7,  $fcc2
+   LD a6,  X,   0 * SIZE
+   CMOVT  s4,  s4,  a8,  $fcc3
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   CMPLT   $fcc0, s1, a1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, a2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, s3, a3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, s4, a4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a1,  $fcc0
+   CMOVT  s2,  s2,  a2,  $fcc1
+   CMOVT  s3,  s3,  a3,  $fcc2
+   CMOVT  s4,  s4,  a4,  $fcc3
+   CMPLT   $fcc0, s1, a5
+   CMPLT   $fcc1, s2, a6
+   CMPLT   $fcc2, s3, a7
+   CMPLT   $fcc3, s4, a8
+   CMOVT  s1,  s1,  a5,  $fcc0
+   CMOVT  s2,  s2,  a6,  $fcc1
+   CMOVT  s3,  s3,  a7,  $fcc2
+   CMOVT  s4,  s4,  a8,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   CMPLT   $fcc0, s1, a1
+   CMOVT  s1,  s1,  a1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/min.S
+++ b/kernel/loongarch64/min.S
@ -0,0 +1,174 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD s1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   add.d   X, X, INCX
+   MOV    s2, s1
+   bge $r0,    N, .L999
+   MOV s3, s1
+   srai.d  I, N, 3
+   MOV    s4, s1
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   CMPLT   $fcc0, a1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, a2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, a3, s3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, a4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a1,  $fcc0
+   LD a1,  X,   0 * SIZE
+   CMOVT  s2,  s2,  a2,  $fcc1
+   add.d   X, X, INCX
+   CMOVT  s3,  s3,  a3,  $fcc2
+   LD a2,  X,   0 * SIZE
+   CMOVT  s4,  s4,  a4,  $fcc3
+   add.d   X, X, INCX
+   CMPLT   $fcc0, a5, s1
+   LD a3,  X,   0 * SIZE
+   CMPLT   $fcc1, a6, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, a7, s3
+   LD a4,  X,   0 * SIZE
+   CMPLT   $fcc3, a8, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a5,  $fcc0
+   LD a5,  X,   0 * SIZE
+   CMOVT  s2,  s2,  a6,  $fcc1
+   add.d   X, X, INCX
+   CMOVT  s3,  s3,  a7,  $fcc2
+   LD a6,  X,   0 * SIZE
+   CMOVT  s4,  s4,  a8,  $fcc3
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   CMPLT   $fcc0, a1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, a2, s2
+   add.d   X, X, INCX
+   CMPLT   $fcc2, a3, s3
+   LD a8,  X,   0 * SIZE
+   CMPLT   $fcc3, a4, s4
+   add.d   X, X, INCX
+   CMOVT  s1,  s1,  a1,  $fcc0
+   CMOVT  s2,  s2,  a2,  $fcc1
+   CMOVT  s3,  s3,  a3,  $fcc2
+   CMOVT  s4,  s4,  a4,  $fcc3
+   CMPLT   $fcc0, a5, s1
+   CMPLT   $fcc1, a6, s2
+   CMPLT   $fcc2, a7, s3
+   CMPLT   $fcc3, a8, s4
+   CMOVT  s1,  s1,  a5,  $fcc0
+   CMOVT  s2,  s2,  a6,  $fcc1
+   CMOVT  s3,  s3,  a7,  $fcc2
+   CMOVT  s4,  s4,  a8,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   CMPLT   $fcc0, a1, s1
+   CMOVT  s1,  s1,  a1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/scal.S
+++ b/kernel/loongarch64/scal.S
@ -0,0 +1,330 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r7
+#define INCX   $r8
+
+#define I      $r17
+#define TEMP   $r18
+#define XX     $r5
+#define ALPHA  $f0
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+#define t1     $f14
+#define t2     $f15
+#define t3     $f16
+#define t4     $f17
+
+   PROLOGUE
+
+   li.d  TEMP, SIZE
+   MTC  a1, $r0
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    N, .L999
+   CMPEQ   $fcc0, ALPHA, a1
+   bceqz   $fcc0, .L50
+   srai.d I, N, 3
+   bne INCX, TEMP, .L20
+   bge $r0,    I, .L15
+   .align 3
+
+.L12:
+   ST a1,  X,   0 * SIZE
+   ST a1,  X,   1 * SIZE
+   ST a1,  X,   2 * SIZE
+   ST a1,  X,   3 * SIZE
+   ST a1,  X,   4 * SIZE
+   ST a1,  X,   5 * SIZE
+   ST a1,  X,   6 * SIZE
+   ST a1,  X,   7 * SIZE
+   addi.w  I, I, -1
+   addi.d X, X, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   ST a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   addi.d X, X, SIZE
+   blt $r0,    I, .L16
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L20:
+   srai.d  I, N, 3
+   bge $r0,    I, .L25
+   .align 3
+
+.L22:
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L22
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   addi.d  I, I, -1
+   ST a1,  X,   0 * SIZE
+   add.d  X, X, INCX
+   blt $r0,    I, .L26
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L50:
+   srai.d I, N, 3
+   bne INCX, TEMP, .L60
+   addi.d I, I, -1
+   blt I,  $r0, .L55
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   LD a7,  X,   6 * SIZE
+   LD a8,  X,   7 * SIZE
+   bge $r0,    I, .L53
+   .align 3
+
+.L52:
+   MUL t1, ALPHA, a1
+   LD a1,  X,   8 * SIZE
+   MUL t2, ALPHA, a2
+   LD a2,  X,   9 * SIZE
+   MUL t3, ALPHA, a3
+   LD a3,  X,  10 * SIZE
+   MUL t4, ALPHA, a4
+   LD a4,  X,  11 * SIZE
+   ST t1,  X,   0 * SIZE
+   MUL t1, ALPHA, a5
+   LD a5,  X,  12 * SIZE
+   ST t2,  X,   1 * SIZE
+   MUL t2, ALPHA, a6
+   LD a6,  X,  13 * SIZE
+   ST t3,  X,   2 * SIZE
+   MUL t3, ALPHA, a7
+   LD a7,  X,  14 * SIZE
+   ST t4,  X,   3 * SIZE
+   MUL t4, ALPHA, a8
+   LD a8,  X,  15 * SIZE
+   addi.d  I, I, -1
+   ST t1,  X,   4 * SIZE
+   ST t2,  X,   5 * SIZE
+   ST t3,  X,   6 * SIZE
+   ST t4,  X,   7 * SIZE
+   addi.d X, X, 8 * SIZE
+   blt $r0,    I, .L52
+   .align 3
+
+.L53:
+   MUL t1, ALPHA, a1
+   MUL t2, ALPHA, a2
+   MUL t3, ALPHA, a3
+   MUL t4, ALPHA, a4
+   ST t1,  X,   0 * SIZE
+   MUL t1, ALPHA, a5
+   ST t2,  X,   1 * SIZE
+   MUL t2, ALPHA, a6
+   ST t3,  X,   2 * SIZE
+   MUL t3, ALPHA, a7
+   ST t4,  X,   3 * SIZE
+   MUL t4, ALPHA, a8
+   ST t1,  X,   4 * SIZE
+   ST t2,  X,   5 * SIZE
+   ST t3,  X,   6 * SIZE
+   ST t4,  X,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   .align 3
+
+.L55:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L56:
+   LD a1,  X,   0 * SIZE
+   MUL t1, ALPHA, a1
+   addi.d  X, X, SIZE
+   addi.d  I, I, -1
+   ST t1,  X,  -1 * SIZE
+   blt $r0,    I, .L56
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   .align 3
+
+.L60:
+   srai.d  I, N, 3
+   move    XX, X
+   addi.d I, I, -1
+   blt I,  $r0, .L65
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   bge $r0,    I, .L63
+   .align 3
+
+.L62:
+   MUL t1, ALPHA, a1
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t2, ALPHA, a2
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t3, ALPHA, a3
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t4, ALPHA, a4
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST t1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   MUL t1, ALPHA, a5
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t2, ALPHA, a6
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t3, ALPHA, a7
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   MUL t4, ALPHA, a8
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST t1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t4,  XX,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  XX, XX, INCX
+   blt $r0,    I, .L62
+   .align 3
+
+.L63:
+   MUL t1, ALPHA, a1
+   MUL t2, ALPHA, a2
+   MUL t3, ALPHA, a3
+   MUL t4, ALPHA, a4
+   ST t1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   MUL t1, ALPHA, a5
+   MUL t2, ALPHA, a6
+   MUL t3, ALPHA, a7
+   MUL t4, ALPHA, a8
+   ST t1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST t4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   .align 3
+
+.L65:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L66:
+   LD a1,  X,   0 * SIZE
+   MUL t1, ALPHA, a1
+   addi.d  I, I, -1
+   ST t1,  X,   0 * SIZE
+   add.d  X, X, INCX
+   blt $r0,    I, .L66
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/snrm2.S
+++ b/kernel/loongarch64/snrm2.S
@ -0,0 +1,249 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f12
+#define a2     $f13
+#define a3     $f14
+#define a4     $f15
+#define a5     $f16
+#define a6     $f17
+#define a7     $f0
+#define a8     $f1
+#define s1     $f22
+#define s2     $f8
+#define t1     $f23
+#define t2     $f9
+#define t3     $f10
+#define t4     $f11
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   movgr2fr.d  s1, $r0
+   li.d  TEMP, SIZE
+   fmov.d s2, s1
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    INCX, .L999
+   srai.d I, N, 3
+   bne INCX, TEMP, .L20
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   LD a6,  X,   5 * SIZE
+   fcvt.d.s    t2, a2
+   LD a7,  X,   6 * SIZE
+   fcvt.d.s    t3, a3
+   LD a8,  X,   7 * SIZE
+   fcvt.d.s   t4, a4
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   fmadd.d  s1, t1, t1, s1
+   LD a1,  X,   8 * SIZE
+   fcvt.d.s    t1, a5
+   NOP
+   fmadd.d  s2, t2, t2, s2
+   LD a2,  X,   9 * SIZE
+   fcvt.d.s    t2, a6
+   NOP
+   fmadd.d  s1, t3, t3, s1
+   LD a3,  X,  10 * SIZE
+   fcvt.d.s    t3, a7
+   NOP
+   fmadd.d  s2, t4, t4, s2
+   LD a4,  X,  11 * SIZE
+   fcvt.d.s    t4, a8
+   NOP
+   fmadd.d  s1, t1, t1, s1
+   LD a5,  X,  12 * SIZE
+   fcvt.d.s    t1, a1
+   NOP
+   fmadd.d  s2, t2, t2, s2
+   LD a6,  X,  13 * SIZE
+   fcvt.d.s    t2, a2
+   addi.d  I, I, -1
+   fmadd.d  s1, t3, t3, s1
+   LD a7,  X,  14 * SIZE
+   fcvt.d.s    t3, a3
+   addi.d  X, X, 8 * SIZE
+   fmadd.d  s2, t4, t4, s2
+   LD a8,  X,   7 * SIZE
+   fcvt.d.s   t4, a4
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   fmadd.d  s1, t1, t1, s1
+   fcvt.d.s    t1, a5
+   fmadd.d  s2, t2, t2, s2
+   fcvt.d.s    t2, a6
+   fmadd.d  s1, t3, t3, s1
+   fcvt.d.s    t3, a7
+   fmadd.d  s2, t4, t4, s2
+   fcvt.d.s    t4, a8
+   fmadd.d  s1, t1, t1, s1
+   fmadd.d  s2, t2, t2, s2
+   fmadd.d  s1, t3, t3, s1
+   fmadd.d  s2, t4, t4, s2
+   addi.d  X, X, 8 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   fmadd.d  s1, t1, t1, s1
+   addi.d X, X, SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+
+.L20:
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD a8,  X,   0 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   fcvt.d.s    t2, a2
+   fcvt.d.s    t3, a3
+   fcvt.d.s    t4, a4
+   add.d  X, X, INCX
+   bge $r0,    I, .L24
+   .align 3
+
+.L23:
+   fmadd.d  s1, t1, t1, s1
+   LD a1,  X,   0 * SIZE
+   fcvt.d.s    t1, a5
+   add.d   X, X, INCX
+   fmadd.d  s2, t2, t2, s2
+   LD a2,  X,   0 * SIZE
+   fcvt.d.s    t2, a6
+   add.d   X, X, INCX
+   fmadd.d  s1, t3, t3, s1
+   LD a3,  X,   0 * SIZE
+   fcvt.d.s    t3, a7
+   add.d   X, X, INCX
+   fmadd.d  s2, t4, t4, s2
+   LD a4,  X,   0 * SIZE
+   fcvt.d.s    t4, a8
+   add.d   X, X, INCX
+   fmadd.d  s1, t1, t1, s1
+   LD a5,  X,   0 * SIZE
+   fcvt.d.s    t1, a1
+   add.d   X, X, INCX
+   fmadd.d  s2, t2, t2, s2
+   LD a6,  X,   0 * SIZE
+   fcvt.d.s    t2, a2
+   add.d   X, X, INCX
+   fmadd.d  s1, t3, t3, s1
+   LD a7,  X,   0 * SIZE
+   fcvt.d.s    t3, a3
+   add.d   X, X, INCX
+   fmadd.d  s2, t4, t4, s2
+   LD a8,  X,   0 * SIZE
+   fcvt.d.s    t4, a4
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L23
+   .align 3
+
+.L24:
+   fmadd.d  s1, t1, t1, s1
+   fcvt.d.s    t1, a5
+   fmadd.d  s2, t2, t2, s2
+   fcvt.d.s    t2, a6
+   fmadd.d  s1, t3, t3, s1
+   fcvt.d.s    t3, a7
+   fmadd.d  s2, t4, t4, s2
+   fcvt.d.s    t4, a8
+   fmadd.d  s1, t1, t1, s1
+   fmadd.d  s2, t2, t2, s2
+   fmadd.d  s1, t3, t3, s1
+   fmadd.d  s2, t4, t4, s2
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   addi.d  I, I, -1
+   fcvt.d.s    t1, a1
+   add.d   X, X, INCX
+   fmadd.d  s1, t1, t1, s1
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   fadd.d  s1, s1, s2
+   fsqrt.d s1, s1
+   move $r4, $r17
+   fcvt.s.d    $f0, s1
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/swap.S
+++ b/kernel/loongarch64/swap.S
@ -0,0 +1,330 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r7
+#define INCX   $r8
+#define Y      $r9
+#define INCY   $r10
+
+#define I      $r17
+#define TEMP   $r18
+#define XX     $r5
+#define YY     $r6
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+#define b1     $f14
+#define b2     $f15
+#define b3     $f16
+#define b4     $f17
+#define b5     $f0
+#define b6     $f1
+#define b7     $f2
+#define b8     $f3
+
+   PROLOGUE
+
+   li.d  TEMP, SIZE
+   slli.d INCX, INCX, BASE_SHIFT
+   bge $r0,    N, .L999
+   slli.d INCY, INCY, BASE_SHIFT
+   bne INCX, TEMP, .L20
+   srai.d I, N, 3
+   bne INCY, TEMP, .L20
+   addi.d I, I, -1
+   blt I,  $r0, .L15
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b2,  Y,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD b3,  Y,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD b4,  Y,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD b5,  Y,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   LD b6,  Y,   5 * SIZE
+   LD a7,  X,   6 * SIZE
+   LD b7,  Y,   6 * SIZE
+   LD a8,  X,   7 * SIZE
+   LD b8,  Y,   7 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   ST a1,  Y,   0 * SIZE
+   LD a1,  X,   8 * SIZE
+   ST b1,  X,   0 * SIZE
+   LD b1,  Y,   8 * SIZE
+   ST a2,  Y,   1 * SIZE
+   LD a2,  X,   9 * SIZE
+   ST b2,  X,   1 * SIZE
+   LD b2,  Y,   9 * SIZE
+   ST a3,  Y,   2 * SIZE
+   LD a3,  X,  10 * SIZE
+   ST b3,  X,   2 * SIZE
+   LD b3,  Y,  10 * SIZE
+   ST a4,  Y,   3 * SIZE
+   LD a4,  X,  11 * SIZE
+   ST b4,  X,   3 * SIZE
+   LD b4,  Y,  11 * SIZE
+   ST a5,  Y,   4 * SIZE
+   LD a5,  X,  12 * SIZE
+   ST b5,  X,   4 * SIZE
+   LD b5,  Y,  12 * SIZE
+   ST a6,  Y,   5 * SIZE
+   LD a6,  X,  13 * SIZE
+   ST b6,  X,   5 * SIZE
+   LD b6,  Y,  13 * SIZE
+   ST a7,  Y,   6 * SIZE
+   LD a7,  X,  14 * SIZE
+   ST b7,  X,   6 * SIZE
+   LD b7,  Y,  14 * SIZE
+   ST a8,  Y,   7 * SIZE
+   LD a8,  X,  15 * SIZE
+   ST b8,  X,   7 * SIZE
+   LD b8,  Y,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+   addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   ST a1,  Y,   0 * SIZE
+   ST b1,  X,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   ST b2,  X,   1 * SIZE
+   ST a3,  Y,   2 * SIZE
+   ST b3,  X,   2 * SIZE
+   ST a4,  Y,   3 * SIZE
+   ST b4,  X,   3 * SIZE
+   ST a5,  Y,   4 * SIZE
+   ST b5,  X,   4 * SIZE
+   ST a6,  Y,   5 * SIZE
+   ST b6,  X,   5 * SIZE
+   ST a7,  Y,   6 * SIZE
+   ST b7,  X,   6 * SIZE
+   ST a8,  Y,   7 * SIZE
+   ST b8,  X,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   addi.d  Y, Y, 8 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+   addi.d  X, X, SIZE
+   addi.d  I, I, -1
+   addi.d  Y, Y, SIZE
+   ST b1,  X,  -1 * SIZE
+   ST a1,  Y,  -1 * SIZE
+   blt $r0,    I, .L16
+   b   .L999
+   .align 3
+
+.L20:
+   srai.d  I, N, 3
+   move    XX, X
+   move    YY, Y
+   addi.d I, I, -1
+   blt I,  $r0, .L25
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b2,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b3,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b4,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b5,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b6,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b7,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   LD b8,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   bge $r0,    I, .L23
+   .align 3
+
+.L22:
+   ST a1,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b1,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a2,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a2,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b2,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a3,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b3,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a4,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a4,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b4,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a5,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a5,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b5,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b5,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a6,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a6,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b6,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b6,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a7,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a7,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b7,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b7,  Y,   0 * SIZE
+   add.d   Y, Y, INCY
+   ST a8,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   LD a8,  X,   0 * SIZE
+   add.d   X, X, INCX
+   ST b8,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   LD b8,  Y,   0 * SIZE
+   addi.d  I, I, -1
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   ST a1,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b1,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a2,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b2,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a3,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b3,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a4,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b4,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a5,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b5,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a6,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b6,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a7,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b7,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   ST a8,  YY,   0 * SIZE
+   add.d   YY, YY, INCY
+   ST b8,  XX,   0 * SIZE
+   add.d   XX, XX, INCX
+   .align 3
+
+.L25:
+   andi    I,  N, 7
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD b1,  Y,   0 * SIZE
+   addi.d  I, I, -1
+   ST a1,  Y,   0 * SIZE
+   ST b1,  X,   0 * SIZE
+   add.d   X, X, INCX
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/trsm_kernel_LN.S
+++ b/kernel/loongarch64/trsm_kernel_LN.S
--- a/kernel/loongarch64/trsm_kernel_LT.S
+++ b/kernel/loongarch64/trsm_kernel_LT.S
--- a/kernel/loongarch64/trsm_kernel_RT.S
+++ b/kernel/loongarch64/trsm_kernel_RT.S
--- a/kernel/loongarch64/zamax.S
+++ b/kernel/loongarch64/zamax.S
@ -0,0 +1,190 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define a5     $f14
+#define a6     $f15
+#define a7     $f16
+#define a8     $f17
+#define t1     $f0
+#define t2     $f1
+#define t3     $f2
+#define t4     $f3
+#define t5     $f4
+#define t6     $f5
+#define t7     $f6
+#define t8     $f7
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD    s1, t1, t2
+   bge $r0,    N, .L999
+   ADD s2, t1, t2
+   srai.d  I, N, 2
+   ADD s3, t1, t2
+   ADD    s4, t1, t2
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   LD a2,  X,   1 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   FABS    t4, a4
+   FABS    t5, a5
+   LD a3,  X,   0 * SIZE
+   FABS    t6, a6
+   LD a4,  X,   1 * SIZE
+   FABS    t7, a7
+   add.d   X, X, INCX
+   FABS    t8, a8
+   ADD t1, t1, t2
+   LD a5,  X,   0 * SIZE
+   ADD t3, t3, t4
+   LD a6,  X,   1 * SIZE
+   ADD t5, t5, t6
+   add.d   X, X, INCX
+   ADD t7, t7, t8
+   CMPLT   $fcc0, s1, t1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, s2, t3
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc2, s3, t5
+   add.d   X, X, INCX
+   CMPLT   $fcc3, s4, t7
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t3,  $fcc1
+   CMOVT  s3,  s3,  t5,  $fcc2
+   CMOVT  s4,  s4,  t7,  $fcc3
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   FABS    t5, a5
+   FABS    t6, a6
+   FABS    t7, a7
+   FABS    t8, a8
+   ADD t1, t1, t2
+   ADD t3, t3, t4
+   ADD t5, t5, t6
+   ADD t7, t7, t8
+   CMPLT   $fcc0, s1, t1
+   CMPLT   $fcc1, s2, t3
+   CMPLT   $fcc2, s3, t5
+   CMPLT   $fcc3, s4, t7
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t3,  $fcc1
+   CMOVT  s3,  s3,  t5,  $fcc2
+   CMOVT  s4,  s4,  t7,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD t1, t1, t2
+   CMPLT   $fcc0, s1, t1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s1, s2
+   CMPLT   $fcc1, s3, s4
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s1, s3
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/zamin.S
+++ b/kernel/loongarch64/zamin.S
@ -0,0 +1,198 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N     $r4
+#define X     $r5
+#define INCX  $r6
+#define I     $r17
+#define TEMP  $r18
+#define a1    $f10
+#define a2    $f11
+#define a3    $f12
+#define a4    $f13
+#define a5    $f14
+#define a6    $f15
+#define a7    $f16
+#define a8    $f17
+#define t1    $f0
+#define t2    $f1
+#define t3    $f2
+#define t4    $f3
+#define t5    $f4
+#define t6    $f5
+#define t7    $f6
+#define t8    $f7
+#define s1    $f22
+#define s2    $f8
+#define s3    $f23
+#define s4    $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   bge $r0,    N, .L999
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    INCX, .L999
+   LD a1,  X,   0 * SIZE
+   addi.d  N, N, -1
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD    s1, t1, t2
+   bge $r0,    N, .L999
+   NOP
+   ADD s2, t1, t2
+   srai.d  I, N, 2
+   ADD s3, t1, t2
+   ADD    s4, t1, t2
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   FABS    t1, a1
+   LD a1,  X,   0 * SIZE
+   FABS    t2, a2
+   LD a2,  X,   1 * SIZE
+   FABS    t3, a3
+   add.d   X, X, INCX
+   FABS    t4, a4
+   NOP
+   FABS    t5, a5
+   LD a3,  X,   0 * SIZE
+   FABS    t6, a6
+   LD a4,  X,   1 * SIZE
+   FABS    t7, a7
+   add.d   X, X, INCX
+   FABS    t8, a8
+   NOP
+   ADD t1, t1, t2
+   LD a5,  X,   0 * SIZE
+   ADD t3, t3, t4
+   LD a6,  X,   1 * SIZE
+   ADD t5, t5, t6
+   add.d   X, X, INCX
+   ADD t7, t7, t8
+   NOP
+   CMPLT   $fcc0, t1, s1
+   LD a7,  X,   0 * SIZE
+   CMPLT   $fcc1, t3, s2
+   LD a8,  X,   1 * SIZE
+   CMPLT   $fcc2, t5, s3
+   add.d   X, X, INCX
+   CMPLT   $fcc3, t7, s4
+   NOP
+   CMOVT  s1,  s1,  t1,  $fcc0
+   addi.d  I, I, -1
+   CMOVT  s2,  s2,  t3,  $fcc1
+   NOP
+   CMOVT  s3,  s3,  t5,  $fcc2
+   CMOVT  s4,  s4,  t7,  $fcc3
+   blt $r0,    I, .L12
+   NOP
+   .align 3
+
+.L13:
+   FABS    t1, a1
+   FABS    t2, a2
+   FABS    t3, a3
+   FABS    t4, a4
+   FABS    t5, a5
+   FABS    t6, a6
+   FABS    t7, a7
+   FABS    t8, a8
+   ADD t1, t1, t2
+   ADD t3, t3, t4
+   ADD t5, t5, t6
+   ADD t7, t7, t8
+   CMPLT   $fcc0, t1, s1
+   CMPLT   $fcc1, t3, s2
+   CMPLT   $fcc2, t5, s3
+   CMPLT   $fcc3, t7, s4
+   CMOVT  s1,  s1,  t1,  $fcc0
+   CMOVT  s2,  s2,  t3,  $fcc1
+   CMOVT  s3,  s3,  t5,  $fcc2
+   CMOVT  s4,  s4,  t7,  $fcc3
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L998
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  I, I, -1
+   FABS    t1, a1
+   FABS    t2, a2
+   ADD t1, t1, t2
+   CMPLT   $fcc0, t1, s1
+   CMOVT  s1,  s1,  t1,  $fcc0
+   add.d  X, X, INCX
+   blt $r0,    I, .L16
+   .align 3
+
+.L998:
+   CMPLT   $fcc0, s2, s1
+   CMPLT   $fcc1, s4, s3
+   CMOVT  s1,  s1,  s2,  $fcc0
+   CMOVT  s3,  s3,  s4,  $fcc1
+   CMPLT   $fcc0, s3, s1
+   CMOVT  s1,  s1,  s3,  $fcc0
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   NOP
+
+   EPILOGUE
--- a/kernel/loongarch64/zasum.S
+++ b/kernel/loongarch64/zasum.S
@ -0,0 +1,158 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f23
+#define a2     $f9
+#define a3     $f10
+#define a4     $f11
+#define a5     $f12
+#define a6     $f13
+#define a7     $f14
+#define a8     $f15
+#define t1     $f16
+#define t2     $f17
+#define t3     $f0
+#define t4     $f1
+#define s1     $f22
+#define s2     $f8
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+#endif
+
+   MTC  s1, $r0
+   MTC  s2, $r0
+   slli.d  INCX, INCX, ZBASE_SHIFT
+   srai.d I, N, 2
+   bge $r0,    N, .L999
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   FABS    t1, a1
+   FABS    t2, a2
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   FABS    t3, a3
+   FABS    t4, a4
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   bge $r0,    I, .L24
+   .align 3
+
+.L23:
+   ADD s1, s1, t1
+   LD a1,  X,   0 * SIZE
+   FABS    t1, a5
+   addi.d  I, I, -1
+   ADD s2, s2, t2
+   LD a2,  X,   1 * SIZE
+   FABS    t2, a6
+   add.d   X, X, INCX
+   ADD s1, s1, t3
+   LD a3,  X,   0 * SIZE
+   FABS    t3, a7
+   NOP
+   ADD s2, s2, t4
+   LD a4,  X,   1 * SIZE
+   FABS    t4, a8
+   add.d   X, X, INCX
+   ADD s1, s1, t1
+   LD a5,  X,   0 * SIZE
+   FABS    t1, a1
+   NOP
+   ADD s2, s2, t2
+   LD a6,  X,   1 * SIZE
+   FABS    t2, a2
+   add.d   X, X, INCX
+   ADD s1, s1, t3
+   LD a7,  X,   0 * SIZE
+   FABS    t3, a3
+   LD a8,  X,   1 * SIZE
+   ADD s2, s2, t4
+   add.d   X, X, INCX
+   FABS   t4, a4
+   blt $r0,    I, .L23
+   .align 3
+
+.L24:
+   ADD s1, s1, t1
+   FABS    t1, a5
+   ADD s2, s2, t2
+   FABS    t2, a6
+   ADD s1, s1, t3
+   FABS    t3, a7
+   ADD s2, s2, t4
+   FABS    t4, a8
+   ADD s1, s1, t1
+   ADD s2, s2, t2
+   ADD s1, s1, t3
+   ADD s2, s2, t4
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   FABS    t1, a1
+   addi.d  I, I, -1
+   FABS    t2, a2
+   add.d   X, X, INCX
+   ADD s1, s1, t1
+   ADD    s2, s2, t2
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   ADD s1, s1, s2
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/zcopy.S
+++ b/kernel/loongarch64/zcopy.S
@ -0,0 +1,217 @@
+/***************************************************************************
+Copyright (c) 2021, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define Y      $r7
+#define INCY   $r8
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f22
+#define a2     $f8
+#define a3     $f23
+#define a4     $f9
+#define a5     $f10
+#define a6     $f11
+#define a7     $f12
+#define a8     $f13
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+   LDINT   INCY,  0(INCY)
+#endif
+
+   li.d  TEMP, 2 * SIZE
+   NOP
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   slli.d INCY, INCY, ZBASE_SHIFT
+   bne INCX, TEMP, .L20
+   srai.d I, N, 2
+   bne INCY, TEMP, .L20
+   addi.d I, I, -1
+   blt     I, $r0,     .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD a3,  X,   2 * SIZE
+   LD a4,  X,   3 * SIZE
+   LD a5,  X,   4 * SIZE
+   LD a6,  X,   5 * SIZE
+   LD a7,  X,   6 * SIZE
+   LD a8,  X,   7 * SIZE
+   bge $r0,    I, .L13
+   .align 3
+
+.L12:
+   ST a1,  Y,   0 * SIZE
+   LD a1,  X,   8 * SIZE
+   ST a2,  Y,   1 * SIZE
+   LD a2,  X,   9 * SIZE
+   ST a3,  Y,   2 * SIZE
+   LD a3,  X,  10 * SIZE
+   ST a4,  Y,   3 * SIZE
+   LD a4,  X,  11 * SIZE
+   ST a5,  Y,   4 * SIZE
+   LD a5,  X,  12 * SIZE
+   ST a6,  Y,   5 * SIZE
+   LD a6,  X,  13 * SIZE
+   ST a7,  Y,   6 * SIZE
+   LD a7,  X,  14 * SIZE
+   ST a8,  Y,   7 * SIZE
+   LD a8,  X,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+   addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   ST a1,  Y,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   ST a3,  Y,   2 * SIZE
+   ST a4,  Y,   3 * SIZE
+   ST a5,  Y,   4 * SIZE
+   ST a6,  Y,   5 * SIZE
+   ST a7,  Y,   6 * SIZE
+   ST a8,  Y,   7 * SIZE
+   addi.d  X, X, 8 * SIZE
+   addi.d  Y, Y, 8 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+
+.L16:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   addi.d  X, X, 2 * SIZE
+   addi.d  Y, Y, 2 * SIZE
+   ST a1,  Y,  -2 * SIZE
+   addi.d  I, I, -1
+   ST a2,  Y,  -1 * SIZE
+   blt $r0,    I, .L16
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+   NOP
+   .align 3
+
+.L20:
+   srai.d  I, N, 2
+   addi.d I, I, -1
+   blt I,  $r0, .L25
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,   0 * SIZE
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,   0 * SIZE
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,   0 * SIZE
+   LD a8,  X,   1 * SIZE
+   add.d  X, X, INCX
+   bge $r0,    I, .L23
+   .align 3
+
+.L22:
+   ST a1,  Y,   0 * SIZE
+   LD a1,  X,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a3,  Y,   0 * SIZE
+   LD a3,  X,   0 * SIZE
+   ST a4,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   LD a4,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a5,  Y,   0 * SIZE
+   LD a5,  X,   0 * SIZE
+   ST a6,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   LD a6,  X,   1 * SIZE
+   add.d   X, X, INCX
+   ST a7,  Y,   0 * SIZE
+   LD a7,  X,   0 * SIZE
+   ST a8,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   LD a8,  X,   1 * SIZE
+   addi.d  I, I, -1
+   add.d  X, X, INCX
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   ST a1,  Y,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  Y,   0 * SIZE
+   ST a4,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   ST a5,  Y,   0 * SIZE
+   ST a6,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   ST a7,  Y,   0 * SIZE
+   ST a8,  Y,   1 * SIZE
+   add.d   Y, Y, INCY
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   ST a1,  Y,   0 * SIZE
+   ST a2,  Y,   1 * SIZE
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/zdot.S
+++ b/kernel/loongarch64/zdot.S
@ -0,0 +1,330 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define Y      $r7
+#define INCY   $r8
+#define I      $r17
+#define TEMP   $r18
+#define a1     $f10
+#define a2     $f11
+#define a3     $f12
+#define a4     $f13
+#define b1     $f14
+#define b2     $f15
+#define b3     $f16
+#define b4     $f17
+#define s1     $f22
+#define s2     $f8
+#define s3     $f23
+#define s4     $f9
+
+   PROLOGUE
+
+#ifdef F_INTERFACE
+   LDINT   N,     0(N)
+   LDINT   INCX,  0(INCX)
+   LDINT   INCY,  0(INCY)
+#endif
+
+   MTC  s1, $r0
+   MOV s2, s1
+   MOV s3, s2
+   MOV s4, s3
+   slli.d  INCX, INCX, ZBASE_SHIFT
+   li.d  TEMP, 2 * SIZE
+   slli.d INCY, INCY, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   srai.d I, N, 2
+   bne INCX, TEMP, .L20
+   bne INCY, TEMP, .L20
+   bge $r0,    I, .L15
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b1,  Y,   0 * SIZE
+   addi.d  I, I, -1
+   LD b2,  Y,   1 * SIZE
+   bge $r0,    I, .L14
+   .align 3
+
+.L13:
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   2 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   3 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   2 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   3 * SIZE
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   4 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   5 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   4 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   5 * SIZE
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   6 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   7 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   6 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   7 * SIZE
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   8 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   9 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   8 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   9 * SIZE
+   addi.d  I, I, -1
+   addi.d  X, X, 8 * SIZE
+   addi.d Y, Y, 8 * SIZE
+   blt $r0,    I, .L13
+   .align 3
+
+.L14:
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   2 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   3 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   2 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   3 * SIZE
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   4 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   5 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   4 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   5 * SIZE
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   6 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   7 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   6 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   7 * SIZE
+   MADD  s1, b3, a3, s1
+   addi.d  X, X, 8 * SIZE
+   MADD  s2, b3, a4, s2
+   addi.d  Y, Y, 8 * SIZE
+   MADD  s3, b4, a3, s3
+   MADD  s4, b4, a4, s4
+   .align 3
+
+.L15:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b1,  Y,   0 * SIZE
+   addi.d  I, I, -1
+   LD b2,  Y,   1 * SIZE
+   bge $r0,    I, .L17
+   .align 3
+
+.L16:
+   MADD  s1, b1, a1, s1
+   addi.d  I, I, -1
+   MADD  s2, b1, a2, s2
+   LD b1,  Y,   2 * SIZE
+   MADD  s3, b2, a1, s3
+   LD a1,  X,   2 * SIZE
+   MADD  s4, b2, a2, s4
+   LD a2,  X,   3 * SIZE
+   LD b2,  Y,   3 * SIZE
+   addi.d  X, X, 2 * SIZE
+   addi.d Y, Y, 2 * SIZE
+   blt $r0,    I, .L16
+   .align 3
+
+.L17:
+   MADD  s1, b1, a1, s1
+   MADD  s2, b1, a2, s2
+   MADD  s3, b2, a1, s3
+   MADD  s4, b2, a2, s4
+   b   .L999
+   .align 3
+
+.L20:
+#ifdef F_INTERFACE
+   bgez    INCX, .L21
+   addi.d  TEMP, N, -1
+   mult    TEMP, INCX
+   mflo    TEMP
+   dsub    X, X, TEMP
+   .align 3
+
+.L21:
+   bgez    INCY, .L22
+   addi.d  TEMP, N, -1
+   mult    TEMP, INCY
+   mflo    TEMP
+   dsub    Y, Y, TEMP
+   .align 3
+
+.L22:
+#endif
+   bge $r0,    I, .L25
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b1,  Y,   0 * SIZE
+   LD b2,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   add.d  Y, Y, INCY
+   bge $r0,    I, .L24
+   .align 3
+
+.L23:
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   0 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   1 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   0 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   0 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   1 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   0 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   0 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   1 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   0 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   0 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   1 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   0 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L23
+   .align 3
+
+.L24:
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   0 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   1 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   0 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b3, a3, s1
+   LD a1,  X,   0 * SIZE
+   MADD  s2, b3, a4, s2
+   LD a2,  X,   1 * SIZE
+   MADD  s3, b4, a3, s3
+   LD b1,  Y,   0 * SIZE
+   MADD  s4, b4, a4, s4
+   LD b2,  Y,   1 * SIZE
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   MADD  s1, b1, a1, s1
+   LD a3,  X,   0 * SIZE
+   MADD  s2, b1, a2, s2
+   LD a4,  X,   1 * SIZE
+   MADD  s3, b2, a1, s3
+   LD b3,  Y,   0 * SIZE
+   MADD  s4, b2, a2, s4
+   LD b4,  Y,   1 * SIZE
+   MADD  s1, b3, a3, s1
+   add.d   X, X, INCX
+   MADD  s2, b3, a4, s2
+   add.d   Y, Y, INCY
+   MADD  s3, b4, a3, s3
+   MADD  s4, b4, a4, s4
+   .align 3
+
+.L25:
+   andi    I,  N, 3
+   bge $r0,    I, .L999
+   .align  3
+.L26:
+   LD a1,  X,   0 * SIZE
+   LD a2,  X,   1 * SIZE
+   LD b1,  Y,   0 * SIZE
+   LD b2,  Y,   1 * SIZE
+   MADD  s1, b1, a1, s1
+   MADD  s2, b1, a2, s2
+   MADD  s3, b2, a1, s3
+   MADD  s4, b2, a2, s4
+   add.d   X, X, INCX
+   add.d   Y, Y, INCY
+   addi.d  I, I, -1
+   blt $r0,    I, .L26
+   .align 3
+
+.L999:
+#ifndef CONJ
+   SUB $f0, s1, s4
+#else
+   ADD $f0, s1, s4
+#endif
+#ifndef CONJ
+   ADD $f1, s3, s2
+#else
+   SUB $f1, s3, s2
+#endif
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/zgemm3m_kernel.S
+++ b/kernel/loongarch64/zgemm3m_kernel.S
--- a/kernel/loongarch64/zgemm_kernel.S
+++ b/kernel/loongarch64/zgemm_kernel.S
--- a/kernel/loongarch64/zgemv_n.S
+++ b/kernel/loongarch64/zgemv_n.S
@ -0,0 +1,648 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define M          $r4
+#define N          $r5
+#define A          $r7
+#define LDA        $r8
+#define X          $r9
+#define INCX       $r10
+#define Y          $r11
+#define INCY       $r6
+#define BUFFER     $r17
+
+#define YORIG      $r18
+#define XX         $r12
+#define YY         $r13
+#define I          $r14
+#define J          $r15
+#define AO1        $r23
+#define AO2        $r24
+
+#define ALPHA_R    $f0
+#define ALPHA_I    $f1
+#define a1         $f22
+#define a2         $f8
+#define a3         $f23
+#define a4         $f9
+#define a5         $f10
+#define a6         $f11
+#define a7         $f12
+#define a8         $f13
+#define x1         $f14
+#define x2         $f15
+#define x3         $f16
+#define x4         $f17
+#define y1         $f3
+#define y2         $f4
+#define y3         $f2
+#define y4         $f5
+#define t1         $f6
+#define t2         $f7
+#define t3         $f18
+#define t4         $f19
+#define t5         $f20
+#define t6         $f21
+#define t7         $f24
+#define t8         $f25
+
+#if !defined(CONJ) && !defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        MADD
+#define    MADD3        NMSUB
+#define    MADD4        MADD
+#endif
+#if  defined(CONJ) && !defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        MADD
+#define    MADD3        MADD
+#define    MADD4        NMSUB
+#endif
+#if  !defined(CONJ) && defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        NMSUB
+#define    MADD3        MADD
+#define    MADD4        MADD
+#endif
+#if  defined(CONJ) && defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        NMSUB
+#define    MADD3        NMSUB
+#define    MADD4        NMSUB
+#endif
+
+   PROLOGUE
+
+   LDARG  INCY,    $sp,  0
+   LDARG  BUFFER,  $sp,  8
+#ifndef __64BIT__
+   addi.d  $sp, $sp, -64
+#else
+   addi.d  $sp, $sp, -32
+#endif
+   SDARG  $r23,  $sp,    0
+   SDARG  $r24,  $sp,    8
+   fst.d  $f24,  $sp,  16
+   fst.d  $f25,  $sp,  24
+#ifndef __64BIT__
+   fst.d  $f18,  $sp,  32
+   fst.d  $f19,  $sp,  40
+   fst.d  $f20,  $sp,  48
+   fst.d  $f21,  $sp,  56
+#endif
+   slli.d     LDA,     LDA,  ZBASE_SHIFT
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    M, .L999
+   slli.d INCY, INCY, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   li.d  I, 2 * SIZE
+   move   YORIG, Y
+   beq INCY, I, .L10
+   srai.d  I,  M, 2
+   move    YORIG, BUFFER
+   move    XX, Y
+   move   YY, BUFFER
+   bge $r0,    I, .L05
+   .align 3
+
+.L02:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   LD a3,  XX,  0 * SIZE
+   LD a4,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   LD a5,  XX,  0 * SIZE
+   LD a6,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   LD a7,  XX,  0 * SIZE
+   LD a8,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   addi.d  I, I, -1
+   addi.d  YY, YY, 8 * SIZE
+   ST a1,  YY,  -8 * SIZE
+   ST a2,  YY,  -7 * SIZE
+   ST a3,  YY,  -6 * SIZE
+   ST a4,  YY,  -5 * SIZE
+   ST a5,  YY,  -4 * SIZE
+   ST a6,  YY,  -3 * SIZE
+   ST a7,  YY,  -2 * SIZE
+   ST a8,  YY,  -1 * SIZE
+   blt $r0,    I, .L02
+   .align 3
+
+.L05:
+   andi    I,  M, 3
+   bge $r0,    I, .L10
+   .align 3
+
+.L06:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   add.d   XX, XX, INCY
+   addi.d  I, I, -1
+   ST a1,  YY,  0 * SIZE
+   ST a2,  YY,  1 * SIZE
+   addi.d YY, YY, 2 * SIZE
+   blt $r0,    I, .L06
+   .align 3
+
+.L10:
+   srai.d  J,  N, 1
+   bge $r0,    J, .L20
+   .align 3
+
+.L11:
+   LD x1,  X,  0 * SIZE
+   LD x2,  X,  1 * SIZE
+   add.d   X, X, INCX
+   LD x3,  X,  0 * SIZE
+   LD x4,  X,  1 * SIZE
+   add.d   X, X, INCX
+   MUL a1, ALPHA_R, x1
+   move    AO1, A
+   MUL a2, ALPHA_I, x1
+   add.d   AO2, A,      LDA
+   MUL a3, ALPHA_R, x3
+   add.d   A,   AO2,    LDA
+   MUL a4, ALPHA_I, x3
+#ifndef XCONJ
+   NMSUB  x1, x2, ALPHA_I, a1
+   MADD  x2, x2, ALPHA_R, a2
+   NMSUB  x3, x4, ALPHA_I, a3
+   MADD  x4, x4, ALPHA_R, a4
+#else
+   MADD  x1, x2, ALPHA_I, a1
+   MSUB  x2, x2, ALPHA_R, a2
+   MADD  x3, x4, ALPHA_I, a3
+   MSUB  x4, x4, ALPHA_R, a4
+#endif
+   srai.d  I,  M, 2
+   move   YY, YORIG
+   bge $r0,    I, .L15
+   LD y1,  YY,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD a5,  AO2,  0 * SIZE
+   LD a6,  AO2,  1 * SIZE
+   LD a7,  AO2,  2 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD y1,  YY,   4 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a1,  AO1,   4 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD y2,  YY,   5 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a3,  AO1,   6 * SIZE
+   MADD3  t1, a2, x2, t1
+   LD y3,  YY,   6 * SIZE
+   MADD4  t2, a2, x1, t2
+   LD a2,  AO1,   5 * SIZE
+   MADD3  t3, a4, x2, t3
+   LD y4,  YY,   7 * SIZE
+   MADD4  t4, a4, x1, t4
+   LD a4,  AO1,   7 * SIZE
+   MADD1  t1, a5, x3, t1
+   MADD2  t2, a5, x4, t2
+   LD a5,  AO2,   4 * SIZE
+   MADD1  t3, a7, x3, t3
+   MADD2  t4, a7, x4, t4
+   LD a7,  AO2,   6 * SIZE
+   MADD3  t1, a6, x4, t1
+   MADD4  t2, a6, x3, t2
+   LD a6,  AO2,   5 * SIZE
+   MADD3  t3, a8, x4, t3
+   addi.d  I, I, -1
+   MADD4  t4, a8, x3, t4
+   LD a8,  AO2,   7 * SIZE
+   bge $r0,    I, .L13
+   .align  3
+.L12:
+   MADD1  t5, a1, x1, y1
+   LD y1,  YY,   8 * SIZE
+   MADD2  t6, a1, x2, y2
+   LD a1,  AO1,   8 * SIZE
+   MADD1  t7, a3, x1, y3
+   LD y2,  YY,   9 * SIZE
+   MADD2  t8, a3, x2, y4
+   LD a3,  AO1,  10 * SIZE
+   MADD3  t5, a2, x2, t5
+   LD y3,  YY,  10 * SIZE
+   MADD4  t6, a2, x1, t6
+   LD a2,  AO1,   9 * SIZE
+   MADD3  t7, a4, x2, t7
+   LD y4,  YY,  11 * SIZE
+   MADD4  t8, a4, x1, t8
+   LD a4,  AO1,  11 * SIZE
+   MADD1  t5, a5, x3, t5
+   ST t1,  YY,   0 * SIZE
+   MADD2  t6, a5, x4, t6
+   LD a5,  AO2,   8 * SIZE
+   MADD1  t7, a7, x3, t7
+   ST t2,  YY,   1 * SIZE
+   MADD2  t8, a7, x4, t8
+   LD a7,  AO2,  10 * SIZE
+   MADD3  t5, a6, x4, t5
+   ST t3,  YY,   2 * SIZE
+   MADD4  t6, a6, x3, t6
+   LD a6,  AO2,   9 * SIZE
+   MADD3  t7, a8, x4, t7
+   ST t4,  YY,   3 * SIZE
+   MADD4  t8, a8, x3, t8
+   LD a8,  AO2,  11 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD y1,  YY,  12 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a1,  AO1,  12 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD y2,  YY,  13 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a3,  AO1,  14 * SIZE
+   MADD3  t1, a2, x2, t1
+   LD y3,  YY,  14 * SIZE
+   MADD4  t2, a2, x1, t2
+   LD a2,  AO1,  13 * SIZE
+   MADD3  t3, a4, x2, t3
+   LD y4,  YY,  15 * SIZE
+   MADD4  t4, a4, x1, t4
+   LD a4,  AO1,  15 * SIZE
+   MADD1  t1, a5, x3, t1
+   ST t5,  YY,   4 * SIZE
+   MADD2  t2, a5, x4, t2
+   LD a5,  AO2,  12 * SIZE
+   MADD1  t3, a7, x3, t3
+   ST t6,  YY,   5 * SIZE
+   MADD2  t4, a7, x4, t4
+   LD a7,  AO2,  14 * SIZE
+   MADD3  t1, a6, x4, t1
+   ST t7,  YY,   6 * SIZE
+   MADD4  t2, a6, x3, t2
+   LD a6,  AO2,  13 * SIZE
+   MADD3  t3, a8, x4, t3
+   ST t8,  YY,   7 * SIZE
+   MADD4  t4, a8, x3, t4
+   LD a8,  AO2,  15 * SIZE
+   addi.d  I, I, -1
+   addi.d  YY,  YY,   8 * SIZE
+   addi.d  AO1, AO1,  8 * SIZE
+   addi.d AO2, AO2,  8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   ST t1,  YY,   0 * SIZE
+   MADD1  t1, a1, x1, y1
+   ST t2,  YY,   1 * SIZE
+   MADD2  t2, a1, x2, y2
+   ST t3,  YY,   2 * SIZE
+   MADD1  t3, a3, x1, y3
+   ST t4,  YY,   3 * SIZE
+   MADD2  t4, a3, x2, y4
+   MADD3  t1, a2, x2, t1
+   MADD4  t2, a2, x1, t2
+   MADD3  t3, a4, x2, t3
+   MADD4  t4, a4, x1, t4
+   MADD1  t1, a5, x3, t1
+   MADD2  t2, a5, x4, t2
+   MADD1  t3, a7, x3, t3
+   MADD2  t4, a7, x4, t4
+   MADD3  t1, a6, x4, t1
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD4  t2, a6, x3, t2
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD3  t3, a8, x4, t3
+   addi.d  YY,  YY,   8 * SIZE
+   MADD4  t4, a8, x3, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L15:
+   andi    I,  M, 2
+   bge $r0,    I, .L16
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD a5,  AO2,  0 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a6,  AO2,  1 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD a7,  AO2,  2 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a8,  AO2,  3 * SIZE
+   MADD3  t1, a2, x2, t1
+   MADD4  t2, a2, x1, t2
+   MADD3  t3, a4, x2, t3
+   MADD4  t4, a4, x1, t4
+   MADD1  t1, a5, x3, t1
+   MADD2  t2, a5, x4, t2
+   MADD1  t3, a7, x3, t3
+   MADD2  t4, a7, x4, t4
+   MADD3  t1, a6, x4, t1
+   addi.d  YY,  YY,   4 * SIZE
+   MADD4  t2, a6, x3, t2
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD3  t3, a8, x4, t3
+   addi.d  AO2, AO2,  4 * SIZE
+   MADD4  t4, a8, x3, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L16:
+   andi    I,  M, 1
+   bge $r0,    I, .L19
+   LD y1,  YY,  0 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD a5,  AO2,  0 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a6,  AO2,  1 * SIZE
+   MADD3  t1, a2, x2, t1
+   MADD4  t2, a2, x1, t2
+   MADD1  t1, a5, x3, t1
+   MADD2  t2, a5, x4, t2
+   MADD3  t1, a6, x4, t1
+   MADD4  t2, a6, x3, t2
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   .align 3
+
+.L19:
+   addi.d  J, J, -1
+   blt $r0,    J, .L11
+   .align 3
+
+.L20:
+   andi    J,  N, 1
+   bge $r0,    J, .L900
+   LD x1,  X,  0 * SIZE
+   LD x2,  X,  1 * SIZE
+   add.d   X, X, INCX
+   MUL a1, ALPHA_R, x1
+   move    AO1, A
+   MUL a2, ALPHA_I, x1
+#ifndef XCONJ
+   NMSUB  x1, x2, ALPHA_I, a1
+   MADD  x2, x2, ALPHA_R, a2
+#else
+   MADD  x1, x2, ALPHA_I, a1
+   MSUB  x2, x2, ALPHA_R, a2
+#endif
+   srai.d  I,  M, 2
+   move   YY, YORIG
+   bge $r0,    I, .L25
+   LD y1,  YY,  0 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y4,  YY,  3 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD y1,  YY,   4 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a1,  AO1,   4 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD y2,  YY,   5 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a3,  AO1,   6 * SIZE
+   MADD3  t1, a2, x2, t1
+   LD y3,  YY,   6 * SIZE
+   MADD4  t2, a2, x1, t2
+   LD a2,  AO1,   5 * SIZE
+   MADD3  t3, a4, x2, t3
+   LD y4,  YY,   7 * SIZE
+   MADD4  t4, a4, x1, t4
+   addi.d  I, I, -1
+   LD a4,  AO1,   7 * SIZE
+   bge $r0,    I, .L23
+   .align  3
+.L22:
+   MADD1  t5, a1, x1, y1
+   LD y1,  YY,   8 * SIZE
+   MADD2  t6, a1, x2, y2
+   LD a1,  AO1,   8 * SIZE
+   MADD1  t7, a3, x1, y3
+   LD y2,  YY,   9 * SIZE
+   MADD2  t8, a3, x2, y4
+   LD a3,  AO1,  10 * SIZE
+   MADD3  t5, a2, x2, t5
+   LD y3,  YY,  10 * SIZE
+   MADD4  t6, a2, x1, t6
+   LD a2,  AO1,   9 * SIZE
+   MADD3  t7, a4, x2, t7
+   LD y4,  YY,  11 * SIZE
+   MADD4  t8, a4, x1, t8
+   LD a4,  AO1,  11 * SIZE
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   ST t3,  YY,   2 * SIZE
+   ST t4,  YY,   3 * SIZE
+   MADD1  t1, a1, x1, y1
+   LD y1,  YY,  12 * SIZE
+   MADD2  t2, a1, x2, y2
+   LD a1,  AO1,  12 * SIZE
+   MADD1  t3, a3, x1, y3
+   LD y2,  YY,  13 * SIZE
+   MADD2  t4, a3, x2, y4
+   LD a3,  AO1,  14 * SIZE
+   MADD3  t1, a2, x2, t1
+   LD y3,  YY,  14 * SIZE
+   MADD4  t2, a2, x1, t2
+   LD a2,  AO1,  13 * SIZE
+   MADD3  t3, a4, x2, t3
+   LD y4,  YY,  15 * SIZE
+   MADD4  t4, a4, x1, t4
+   LD a4,  AO1,  15 * SIZE
+   ST t5,  YY,   4 * SIZE
+   ST t6,  YY,   5 * SIZE
+   ST t7,  YY,   6 * SIZE
+   ST t8,  YY,   7 * SIZE
+   addi.d  I, I, -1
+   addi.d  YY,  YY,   8 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   ST t1,  YY,   0 * SIZE
+   MADD1  t1, a1, x1, y1
+   ST t2,  YY,   1 * SIZE
+   MADD2  t2, a1, x2, y2
+   ST t3,  YY,   2 * SIZE
+   MADD1  t3, a3, x1, y3
+   ST t4,  YY,   3 * SIZE
+   MADD2  t4, a3, x2, y4
+   MADD3  t1, a2, x2, t1
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD4  t2, a2, x1, t2
+   addi.d  YY,  YY,   8 * SIZE
+   MADD3  t3, a4, x2, t3
+   MADD4  t4, a4, x1, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L25:
+   andi    I,  M, 2
+   bge $r0,    I, .L26
+   LD a1,  AO1,  0 * SIZE
+   LD y1,  YY,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a3,  AO1,  2 * SIZE
+   LD y3,  YY,  2 * SIZE
+   LD a4,  AO1,  3 * SIZE
+   LD y4,  YY,  3 * SIZE
+   MADD1  t1, a1, x1, y1
+   MADD2  t2, a1, x2, y2
+   MADD1  t3, a3, x1, y3
+   MADD2  t4, a3, x2, y4
+   MADD3  t1, a2, x2, t1
+   addi.d  YY,  YY,   4 * SIZE
+   MADD4  t2, a2, x1, t2
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD3  t3, a4, x2, t3
+   MADD4  t4, a4, x1, t4
+   ST t1,  YY,  -4 * SIZE
+   ST t2,  YY,  -3 * SIZE
+   ST t3,  YY,  -2 * SIZE
+   ST t4,  YY,  -1 * SIZE
+   .align 3
+
+.L26:
+   andi    I,  M, 1
+   bge $r0,    I, .L900
+   LD y1,  YY,  0 * SIZE
+   LD y2,  YY,  1 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   MADD1  t1, a1, x1, y1
+   MADD2  t2, a1, x2, y2
+   MADD3  t1, a2, x2, t1
+   MADD4  t2, a2, x1, t2
+   ST t1,  YY,   0 * SIZE
+   ST t2,  YY,   1 * SIZE
+   .align 3
+
+.L900:
+   li.d  YORIG, 2 * SIZE
+   srai.d I,  M, 2
+   beq INCY, YORIG, .L999
+   move   XX, BUFFER
+   bge $r0,    I, .L905
+   .align 3
+
+.L902:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   LD a3,  XX,  2 * SIZE
+   LD a4,  XX,  3 * SIZE
+   LD a5,  XX,  4 * SIZE
+   LD a6,  XX,  5 * SIZE
+   LD a7,  XX,  6 * SIZE
+   LD a8,  XX,  7 * SIZE
+   addi.d  I, I, -1
+   ST a1,  Y,  0 * SIZE
+   ST a2,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   ST a3,  Y,  0 * SIZE
+   ST a4,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   ST a5,  Y,  0 * SIZE
+   ST a6,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   ST a7,  Y,  0 * SIZE
+   ST a8,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   addi.d XX, XX, 8 * SIZE
+   blt $r0,    I, .L902
+   .align 3
+
+.L905:
+   andi    I,  M, 3
+   bge $r0,    I, .L999
+   .align 3
+
+.L906:
+   LD a1,  XX,  0 * SIZE
+   LD a2,  XX,  1 * SIZE
+   addi.d  XX, XX, 2 * SIZE
+   addi.d  I, I, -1
+   ST a1,  Y,  0 * SIZE
+   ST a2,  Y,  1 * SIZE
+   add.d  Y, Y, INCY
+   blt $r0,    I, .L906
+   .align 3
+
+.L999:
+   LDARG  $r23,  $sp,    0
+   LDARG  $r24,  $sp,    8
+   fld.d  $f24,  $sp,  16
+   fld.d  $f25,  $sp,  24
+#ifndef __64BIT__
+   fld.d  $f18,  $sp,  32
+   fld.d  $f19,  $sp,  40
+   fld.d  $f20,  $sp,  48
+   fld.d  $f21,  $sp,  56
+#endif
+#ifdef __64BIT__
+   addi.d  $sp, $sp, 32
+#else
+   addi.d  $sp, $sp, 64
+#endif
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/kernel/loongarch64/zgemv_t.S
+++ b/kernel/loongarch64/zgemv_t.S
@ -0,0 +1,556 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+
+#define M          $r4
+#define N          $r5
+#define A          $r7
+#define LDA        $r8
+#define X          $r9
+#define INCX       $r10
+#define Y          $r11
+#define INCY       $r6
+#define BUFFER     $r17
+
+#define XORIG      $r18
+#define XX         $r12
+#define YY         $r13
+#define I          $r14
+#define J          $r15
+#define AO1        $r23
+#define AO2        $r24
+
+#define ALPHA_R    $f0
+#define ALPHA_I    $f1
+#define a1         $f22
+#define a2         $f8
+#define a3         $f23
+#define a4         $f9
+#define a5         $f10
+#define a6         $f11
+#define a7         $f12
+#define a8         $f13
+#define y1         $f14
+#define y2         $f15
+#define y3         $f16
+#define y4         $f17
+#define x1         $f3
+#define x2         $f4
+#define x3         $f2
+#define x4         $f5
+#define x5         $f6
+#define x6         $f7
+#define x7         $f18
+#define x8         $f19
+
+#if !defined(CONJ) && !defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        MADD
+#define    MADD3        NMSUB
+#define    MADD4        MADD
+#endif
+#if  defined(CONJ) && !defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        MADD
+#define    MADD3        MADD
+#define    MADD4        NMSUB
+#endif
+#if  !defined(CONJ) && defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        NMSUB
+#define    MADD3        MADD
+#define    MADD4        MADD
+#endif
+#if  defined(CONJ) && defined(XCONJ)
+#define    MADD1        MADD
+#define    MADD2        NMSUB
+#define    MADD3        NMSUB
+#define    MADD4        NMSUB
+#endif
+
+   PROLOGUE
+
+   LDARG  INCY,    $sp,  0
+   LDARG  BUFFER,  $sp,  8
+#ifdef __64BIT__
+   addi.d  $sp, $sp, -16
+#else
+   addi.d  $sp, $sp, -32
+#endif
+   MTC  y1, $r0
+   SDARG  $r23,  $sp,    0
+   SDARG  $r24,  $sp,    8
+   slli.d     LDA,     LDA,  ZBASE_SHIFT
+#ifndef __64BIT__
+   fst.d  $f18,  $sp,  16
+   fst.d  $f19,  $sp,  24
+#endif
+   slli.d INCX, INCX, ZBASE_SHIFT
+   bge $r0,    M, .L999
+   slli.d INCY, INCY, ZBASE_SHIFT
+   bge $r0,    N, .L999
+   li.d  I, 2 * SIZE
+   move   XORIG, X
+   beq INCX, I, .L10
+   srai.d  I,  M, 2
+   move    XORIG, BUFFER
+   move   YY, BUFFER
+   bge $r0,    I, .L05
+   .align 3
+
+.L02:
+   LD a1,  X,  0 * SIZE
+   LD a2,  X,  1 * SIZE
+   add.d   X, X, INCX
+   LD a3,  X,  0 * SIZE
+   LD a4,  X,  1 * SIZE
+   add.d   X, X, INCX
+   LD a5,  X,  0 * SIZE
+   LD a6,  X,  1 * SIZE
+   add.d   X, X, INCX
+   LD a7,  X,  0 * SIZE
+   LD a8,  X,  1 * SIZE
+   add.d   X, X, INCX
+   addi.d  I, I, -1
+   addi.d  YY, YY, 8 * SIZE
+   ST a1,  YY,  -8 * SIZE
+   ST a2,  YY,  -7 * SIZE
+   ST a3,  YY,  -6 * SIZE
+   ST a4,  YY,  -5 * SIZE
+   ST a5,  YY,  -4 * SIZE
+   ST a6,  YY,  -3 * SIZE
+   ST a7,  YY,  -2 * SIZE
+   ST a8,  YY,  -1 * SIZE
+   blt $r0,    I, .L02
+   .align 3
+
+.L05:
+   andi    I,  M, 3
+   bge $r0,    I, .L10
+   .align 3
+
+.L06:
+   LD a1,  X,  0 * SIZE
+   LD a2,  X,  1 * SIZE
+   add.d   X, X, INCX
+   ST a1,  YY,  0 * SIZE
+   ST a2,  YY,  1 * SIZE
+   addi.d  I, I, -1
+   addi.d YY, YY, 2 * SIZE
+   blt $r0,    I, .L06
+   .align 3
+
+.L10:
+   srai.d  J,  N, 1
+   move   YY, Y
+   bge $r0,    J, .L20
+   .align 3
+
+.L11:
+   move    AO1, A
+   MOV y2, y1
+   add.d   AO2, A,      LDA
+   MOV y3, y1
+   add.d   A,   AO2,    LDA
+   MOV y4, y1
+   srai.d  I,  M, 2
+   move   XX, XORIG
+   bge $r0,    I, .L15
+   LD x1,  XX,  0 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD x4,  XX,  3 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a3,  AO2,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD a4,  AO2,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD a7,  AO2,  2 * SIZE
+   LD a6,  AO1,  3 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   addi.d  I, I, -1
+   bge $r0,    I, .L13
+   .align  3
+.L12:
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,  4 * SIZE
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   LD a3,  AO2,  4 * SIZE
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   LD a2,  AO1,  5 * SIZE
+   MADD3  y3, a4, x2, y3
+   LD x2,  XX,  5 * SIZE
+   MADD4  y4, a4, x1, y4
+   LD a4,  AO2,  5 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,  4 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  6 * SIZE
+   MADD1  y3, a7, x3, y3
+   MADD2  y4, a7, x4, y4
+   LD a7,  AO2,  6 * SIZE
+   MADD3  y1, a6, x4, y1
+   addi.d  I, I, -1
+   MADD4  y2, a6, x3, y2
+   LD a6,  AO1,  7 * SIZE
+   MADD3  y3, a8, x4, y3
+   LD x4,  XX,  7 * SIZE
+   MADD4  y4, a8, x3, y4
+   LD a8,  AO2,  7 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  6 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,   8 * SIZE
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   LD a3,  AO2,   8 * SIZE
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   LD a2,  AO1,   9 * SIZE
+   MADD3  y3, a4, x2, y3
+   LD x2,  XX,   9 * SIZE
+   MADD4  y4, a4, x1, y4
+   LD a4,  AO2,   9 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,   8 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  10 * SIZE
+   MADD1  y3, a7, x3, y3
+   addi.d  XX,  XX,   8 * SIZE
+   MADD2  y4, a7, x4, y4
+   LD a7,  AO2,  10 * SIZE
+   MADD3  y1, a6, x4, y1
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD4  y2, a6, x3, y2
+   LD a6,  AO1,  11 * SIZE
+   MADD3  y3, a8, x4, y3
+   LD x4,  XX,   3 * SIZE
+   MADD4  y4, a8, x3, y4
+   LD a8,  AO2,   3 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L12
+   .align 3
+
+.L13:
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,   2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,  4 * SIZE
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   LD a3,  AO2,  4 * SIZE
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   LD a2,  AO1,  5 * SIZE
+   MADD3  y3, a4, x2, y3
+   LD x2,  XX,  5 * SIZE
+   MADD4  y4, a4, x1, y4
+   LD a4,  AO2,  5 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,  4 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  6 * SIZE
+   MADD1  y3, a7, x3, y3
+   MADD2  y4, a7, x4, y4
+   LD a7,  AO2,  6 * SIZE
+   MADD3  y1, a6, x4, y1
+   MADD4  y2, a6, x3, y2
+   LD a6,  AO1,  7 * SIZE
+   MADD3  y3, a8, x4, y3
+   LD x4,  XX,  7 * SIZE
+   MADD4  y4, a8, x3, y4
+   LD a8,  AO2,  7 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  6 * SIZE
+   MADD2  y2, a1, x2, y2
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   MADD3  y3, a4, x2, y3
+   MADD4  y4, a4, x1, y4
+   MADD1  y1, a5, x3, y1
+   MADD2  y2, a5, x4, y2
+   MADD1  y3, a7, x3, y3
+   MADD2  y4, a7, x4, y4
+   MADD3  y1, a6, x4, y1
+   addi.d  XX,  XX,   8 * SIZE
+   MADD4  y2, a6, x3, y2
+   addi.d  AO1, AO1,  8 * SIZE
+   MADD3  y3, a8, x4, y3
+   addi.d  AO2, AO2,  8 * SIZE
+   MADD4  y4, a8, x3, y4
+   .align 3
+
+.L15:
+   andi    I,  M, 2
+   bge $r0,    I, .L17
+   LD x1,  XX,  0 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD x3,  XX,  2 * SIZE
+   LD x4,  XX,  3 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a3,  AO2,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD a4,  AO2,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD a7,  AO2,  2 * SIZE
+   LD a6,  AO1,  3 * SIZE
+   LD a8,  AO2,  3 * SIZE
+   MADD1  y1, a1, x1, y1
+   MADD2  y2, a1, x2, y2
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   MADD3  y3, a4, x2, y3
+   MADD4  y4, a4, x1, y4
+   MADD1  y1, a5, x3, y1
+   MADD2  y2, a5, x4, y2
+   MADD1  y3, a7, x3, y3
+   MADD2  y4, a7, x4, y4
+   MADD3  y1, a6, x4, y1
+   addi.d  XX,  XX,   4 * SIZE
+   MADD4  y2, a6, x3, y2
+   addi.d  AO1, AO1,  4 * SIZE
+   MADD3  y3, a8, x4, y3
+   addi.d  AO2, AO2,  4 * SIZE
+   MADD4  y4, a8, x3, y4
+   .align 3
+
+.L17:
+   andi    I,  M, 1
+.align 3
+
+   bge $r0,    I, .L19
+.L18:
+   LD x1,  XX,  0 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a1,  AO1,  0 * SIZE
+   LD a3,  AO2,  0 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD a2,  AO1,  1 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a4,  AO2,  1 * SIZE
+   MADD1  y3, a3, x1, y3
+   MADD2  y4, a3, x2, y4
+   MADD3  y1, a2, x2, y1
+   MADD4  y2, a2, x1, y2
+   MADD3  y3, a4, x2, y3
+   MADD4  y4, a4, x1, y4
+   .align 3
+
+.L19:
+   LD a1,  Y,  0 * SIZE
+   LD a2,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   LD a3,  Y,  0 * SIZE
+   LD a4,  Y,  1 * SIZE
+   add.d   Y, Y, INCY
+   MADD  a1, y1, ALPHA_R, a1
+   MADD  a2, y1, ALPHA_I, a2
+   MADD  a3, y3, ALPHA_R, a3
+   MADD  a4, y3, ALPHA_I, a4
+   NMSUB  a1, y2, ALPHA_I, a1
+   MADD  a2, y2, ALPHA_R, a2
+   NMSUB  a3, y4, ALPHA_I, a3
+   MTC  y1, $r0
+   MADD  a4, y4, ALPHA_R, a4
+   addi.d  J, J, -1
+   ST a1,  YY,   0 * SIZE
+   ST a2,  YY,   1 * SIZE
+   add.d   YY, YY, INCY
+   ST a3,  YY,   0 * SIZE
+   ST a4,  YY,   1 * SIZE
+   add.d  YY, YY, INCY
+   blt $r0,    J, .L11
+   .align 3
+
+.L20:
+   andi    J,  N, 1
+   MOV y2, y1
+   srai.d I,  M, 2
+   bge $r0,    J, .L999
+   MOV y3, y1
+   move    AO1, A
+   MOV y4, y1
+   move   XX, XORIG
+   bge $r0,    I, .L25
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   LD x4,  XX,  3 * SIZE
+   addi.d  I, I, -1
+   LD a6,  AO1,  3 * SIZE
+   bge $r0,    I, .L23
+   .align  3
+.L22:
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,  4 * SIZE
+   MADD3  y3, a2, x2, y3
+   LD x2,  XX,  5 * SIZE
+   MADD4  y4, a2, x1, y4
+   LD a2,  AO1,  5 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,  4 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  6 * SIZE
+   MADD3  y3, a6, x4, y3
+   LD x4,  XX,  7 * SIZE
+   MADD4  y4, a6, x3, y4
+   LD a6,  AO1,  7 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  6 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,   8 * SIZE
+   MADD3  y3, a2, x2, y3
+   LD x2,  XX,   9 * SIZE
+   MADD4  y4, a2, x1, y4
+   LD a2,  AO1,   9 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,   8 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  10 * SIZE
+   MADD3  y3, a6, x4, y3
+   LD x4,  XX,  11 * SIZE
+   MADD4  y4, a6, x3, y4
+   LD a6,  AO1,  11 * SIZE
+   addi.d  I, I, -1
+   addi.d  XX,  XX,   8 * SIZE
+   addi.d AO1, AO1,  8 * SIZE
+   blt $r0,    I, .L22
+   .align 3
+
+.L23:
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,   2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a1,  AO1,  4 * SIZE
+   MADD3  y3, a2, x2, y3
+   LD x2,  XX,  5 * SIZE
+   MADD4  y4, a2, x1, y4
+   LD a2,  AO1,  5 * SIZE
+   MADD1  y1, a5, x3, y1
+   LD x1,  XX,  4 * SIZE
+   MADD2  y2, a5, x4, y2
+   LD a5,  AO1,  6 * SIZE
+   MADD3  y3, a6, x4, y3
+   LD x4,  XX,  7 * SIZE
+   MADD4  y4, a6, x3, y4
+   LD a6,  AO1,  7 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  6 * SIZE
+   MADD2  y2, a1, x2, y2
+   MADD3  y3, a2, x2, y3
+   MADD4  y4, a2, x1, y4
+   MADD1  y1, a5, x3, y1
+   MADD2  y2, a5, x4, y2
+   MADD3  y3, a6, x4, y3
+   addi.d  XX,  XX,   8 * SIZE
+   MADD4  y4, a6, x3, y4
+   addi.d  AO1, AO1,  8 * SIZE
+   .align 3
+
+.L25:
+   andi    I,  M, 2
+   bge $r0,    I, .L27
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   LD a5,  AO1,  2 * SIZE
+   MADD1  y1, a1, x1, y1
+   LD x3,  XX,  2 * SIZE
+   MADD2  y2, a1, x2, y2
+   LD a6,  AO1,  3 * SIZE
+   MADD3  y3, a2, x2, y3
+   LD x4,  XX,  3 * SIZE
+   MADD4  y4, a2, x1, y4
+   MADD1  y1, a5, x3, y1
+   MADD2  y2, a5, x4, y2
+   MADD3  y3, a6, x4, y3
+   addi.d  XX,  XX,   4 * SIZE
+   MADD4  y4, a6, x3, y4
+   addi.d  AO1, AO1,  4 * SIZE
+   .align 3
+
+.L27:
+   andi    I,  M, 1
+.align 3
+
+   bge $r0,    I, .L29
+.L28:
+   LD a1,  AO1,  0 * SIZE
+   LD x1,  XX,  0 * SIZE
+   LD a2,  AO1,  1 * SIZE
+   LD x2,  XX,  1 * SIZE
+   MADD1  y1, a1, x1, y1
+   MADD2  y2, a1, x2, y2
+   MADD3  y3, a2, x2, y3
+   MADD4  y4, a2, x1, y4
+   .align 3
+
+.L29:
+   LD a1,  Y,  0 * SIZE
+   LD a2,  Y,  1 * SIZE
+   ADD y1, y1, y3
+   ADD y2, y2, y4
+   MADD  a1, y1, ALPHA_R, a1
+   MADD  a2, y1, ALPHA_I, a2
+   NMSUB  a1, y2, ALPHA_I, a1
+   MADD  a2, y2, ALPHA_R, a2
+   ST a1,  YY,   0 * SIZE
+   ST a2,  YY,   1 * SIZE
+   .align 3
+
+.L999:
+   LDARG  $r23,  $sp,    0
+   LDARG  $r24,  $sp,    8
+#ifndef __64BIT__
+   fld.d  $f18,  $sp,  16
+   fld.d  $f19,  $sp,  24
+#endif
+#ifdef __64BIT__
+   addi.d  $sp, $sp, 16
+#else
+   addi.d  $sp, $sp, 32
+#endif
+   move $r4, $r17
+   fmov.d $f0, $f22
+   jirl    $r0, $r1, 0x0
+
+   EPILOGUE
--- a/Show More
+++ b/Show More