Merge branch 'develop' into develop

.travis.yml

@@ -4,10 +4,10 @@ dist: precise
 sudo: true
 language: c
 
-jobs:
+matrix:
   include:
     - &test-ubuntu
-      stage: test
+      os: linux
       compiler: gcc
       addons:
         apt:
@@ -57,7 +57,7 @@ jobs:
         - TARGET_BOX=LINUX32
         - BTYPE="BINARY=32"
 
-    - stage: test
+    - os: linux
       compiler: gcc
       addons:
         apt:
@@ -77,13 +77,13 @@ jobs:
     # which is slower than container-based infrastructure used for jobs
     # that don't require sudo.
     - &test-alpine
-      stage: test
+      os: linux
       dist: trusty
       sudo: true
       language: minimal
       before_install:
-        - "wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.6.0/alpine-chroot-install' \
+        - "wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.9.0/alpine-chroot-install' \
-              && echo 'a827a4ba3d0817e7c88bae17fe34e50204983d1e  alpine-chroot-install' | sha1sum -c || exit 1"
+          && 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'
@@ -117,10 +117,10 @@ jobs:
     - <<: *test-alpine
       env:
         - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=core2"
+        - BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=CORE2"
 
     - &test-cmake
-      stage: test
+      os: linux
       compiler: clang
       addons:
         apt:
@@ -147,6 +147,58 @@ jobs:
       env:
         - CMAKE=1
 
+    - &test-macos
+      os: osx
+      osx_image: xcode8.3
+      before_script:
+        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
+        - brew update
+        - brew install gcc # for gfortran
+      script:
+        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+      env:
+        - BTYPE="BINARY=64 INTERFACE64=1"
+
+    - <<: *test-macos
+      env:
+        - BTYPE="BINARY=32"
+
+    - &emulated-arm
+      dist: trusty
+      sudo: required
+      services: docker
+      env: IMAGE_ARCH=arm32 TARGET_ARCH=ARMV6 COMPILER=gcc
+      name: "Emulated Build for ARMV6 with gcc"
+      before_install: sudo docker run --rm --privileged multiarch/qemu-user-static:register --reset
+      script: |
+        echo "FROM openblas/alpine:${IMAGE_ARCH}
+        COPY . /tmp/openblas
+        RUN mkdir /tmp/openblas/build                             &&  \
+            cd /tmp/openblas/build                                &&  \
+            CC=${COMPILER} cmake -D DYNAMIC_ARCH=OFF                  \
+                                 -D TARGET=${TARGET_ARCH}             \
+                                 -D BUILD_SHARED_LIBS=ON              \
+                                 -D BUILD_WITHOUT_LAPACK=ON           \
+                                 -D BUILD_WITHOUT_CBLAS=ON            \
+                                 -D CMAKE_BUILD_TYPE=Release ../  &&  \
+            cmake --build ." > Dockerfile
+        docker build .
+    - <<: *emulated-arm
+      env: IMAGE_ARCH=arm32 TARGET_ARCH=ARMV6 COMPILER=clang
+      name: "Emulated Build for ARMV6 with clang"
+    - <<: *emulated-arm
+      env: IMAGE_ARCH=arm64 TARGET_ARCH=ARMV8 COMPILER=gcc
+      name: "Emulated Build for ARMV8 with gcc"
+    - <<: *emulated-arm
+      env: IMAGE_ARCH=arm64 TARGET_ARCH=ARMV8 COMPILER=clang
+      name: "Emulated Build for ARMV8 with clang"
+
+  allow_failures:
+    - env: IMAGE_ARCH=arm32 TARGET_ARCH=ARMV6 COMPILER=gcc
+    - env: IMAGE_ARCH=arm32 TARGET_ARCH=ARMV6 COMPILER=clang
+    - env: IMAGE_ARCH=arm64 TARGET_ARCH=ARMV8 COMPILER=gcc
+    - env: IMAGE_ARCH=arm64 TARGET_ARCH=ARMV8 COMPILER=clang
+
 # whitelist
 branches:
   only:

CMakeLists.txt

@@ -6,21 +6,30 @@ cmake_minimum_required(VERSION 2.8.5)
 project(OpenBLAS C ASM)
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 0.dev)
+set(OpenBLAS_PATCH_VERSION 6.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
 include(GNUInstallDirs)
 
-set(OpenBLAS_LIBNAME openblas)
+include(CMakePackageConfigHelpers)
+
 
 #######
 if(MSVC)
-option(BUILD_WITHOUT_LAPACK "Without LAPACK and LAPACKE (Only BLAS or CBLAS)" ON)
+option(BUILD_WITHOUT_LAPACK "Do not build LAPACK and LAPACKE (Only BLAS or CBLAS)" ON)
 endif()
-option(BUILD_WITHOUT_CBLAS "Without CBLAS" OFF)
+option(BUILD_WITHOUT_CBLAS "Do not build the C interface (CBLAS) to the BLAS functions" OFF)
-option(DYNAMIC_ARCH "Build with DYNAMIC_ARCH" OFF)
+option(DYNAMIC_ARCH "Include support for multiple CPU targets, with automatic selection at runtime (x86/x86_64 only)" OFF)
-option(BUILD_RELAPACK "Build with ReLAPACK (recursive LAPACK" OFF)
+option(DYNAMIC_OLDER "Include specific support for older cpu models (Penryn,Dunnington,Atom,Nano,Opteron) with DYNAMIC_ARCH" OFF)
+option(BUILD_RELAPACK "Build with ReLAPACK (recursive implementation of several LAPACK functions on top of standard LAPACK)" OFF)
+
+# Add a prefix or suffix to all exported symbol names in the shared library.
+# Avoids conflicts with other BLAS libraries, especially when using
+# 64 bit integer interfaces in OpenBLAS.
+
+set(SYMBOLPREFIX "" CACHE STRING  "Add a prefix to all exported symbol names in the shared library to avoid conflicts with other BLAS libraries" )
+set(SYMBOLSUFFIX "" CACHE STRING  "Add a suffix to all exported symbol names in the shared library, e.g. _64 for INTERFACE64 builds" )
 #######
 if(BUILD_WITHOUT_LAPACK)
 set(NO_LAPACK 1)
@@ -34,11 +43,13 @@ endif()
 #######
 
 
-message(WARNING "CMake support is experimental. This will not produce the same Makefiles that OpenBLAS ships with. Only x86 support is currently available.")
+message(WARNING "CMake support is experimental. It does not yet support all build options and may not produce the same Makefiles that OpenBLAS ships with.")
 
 include("${PROJECT_SOURCE_DIR}/cmake/utils.cmake")
 include("${PROJECT_SOURCE_DIR}/cmake/system.cmake")
 
+set(OpenBLAS_LIBNAME openblas${SUFFIX64_UNDERSCORE})
+
 set(BLASDIRS interface driver/level2 driver/level3 driver/others)
 
 if (NOT DYNAMIC_ARCH)
@@ -146,6 +157,7 @@ endif()
 
 # add objects to the openblas lib
 add_library(${OpenBLAS_LIBNAME} ${LA_SOURCES} ${LAPACKE_SOURCES} ${RELA_SOURCES} ${TARGET_OBJS} ${OpenBLAS_DEF_FILE})
+target_include_directories(${OpenBLAS_LIBNAME} INTERFACE $<INSTALL_INTERFACE:include>)
 
 # Android needs to explicitly link against libm
 if(ANDROID)
@@ -165,6 +177,7 @@ endif()
 # Set output for libopenblas
 set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
 set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES LIBRARY_OUTPUT_NAME_DEBUG "${OpenBLAS_LIBNAME}_d")
+set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES EXPORT_NAME "OpenBLAS")
 
 foreach (OUTPUTCONFIG ${CMAKE_CONFIGURATION_TYPES})
   string( TOUPPER ${OUTPUTCONFIG} OUTPUTCONFIG )
@@ -204,14 +217,84 @@ set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES
   SOVERSION ${OpenBLAS_MAJOR_VERSION}
 )
 
+if (BUILD_SHARED_LIBS AND NOT ${SYMBOLPREFIX}${SYMBOLSUFIX} STREQUAL "")
+if (NOT DEFINED ARCH)
+  set(ARCH_IN "x86_64")
+else()
+  set(ARCH_IN ${ARCH})
+endif()
+
+if (${CORE} STREQUAL "generic")
+  set(ARCH_IN "GENERIC")
+endif ()
+
+if (NOT DEFINED EXPRECISION)
+  set(EXPRECISION_IN 0)
+else()
+  set(EXPRECISION_IN ${EXPRECISION})
+endif()
+
+if (NOT DEFINED NO_CBLAS)
+  set(NO_CBLAS_IN 0)
+else()
+  set(NO_CBLAS_IN ${NO_CBLAS})
+endif()
+
+if (NOT DEFINED NO_LAPACK)
+  set(NO_LAPACK_IN 0)
+else()
+  set(NO_LAPACK_IN ${NO_LAPACK})
+endif()
+
+if (NOT DEFINED NO_LAPACKE)
+  set(NO_LAPACKE_IN 0)
+else()
+  set(NO_LAPACKE_IN ${NO_LAPACKE})
+endif()
+
+if (NOT DEFINED NEED2UNDERSCORES)
+  set(NEED2UNDERSCORES_IN 0)
+else()
+  set(NEED2UNDERSCORES_IN ${NEED2UNDERSCORES})
+endif()
+
+if (NOT DEFINED ONLY_CBLAS)
+  set(ONLY_CBLAS_IN 0)
+else()
+  set(ONLY_CBLAS_IN ${ONLY_CBLAS})
+endif()
+
+if (NOT DEFINED BU)
+  set(BU _)
+endif()
+
+if (NOT ${SYMBOLPREFIX} STREQUAL "")
+message(STATUS "adding prefix ${SYMBOLPREFIX} to names of exported symbols in ${OpenBLAS_LIBNAME}")
+endif()
+if (NOT ${SYMBOLSUFFIX} STREQUAL "")
+message(STATUS "adding suffix ${SYMBOLSUFFIX} to names of exported symbols in ${OpenBLAS_LIBNAME}")
+endif()
+	add_custom_command(TARGET ${OpenBLAS_LIBNAME} POST_BUILD
+  	COMMAND perl  ${PROJECT_SOURCE_DIR}/exports/gensymbol "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BUILD_LAPACK_DEPRECATED}" > ${PROJECT_BINARY_DIR}/objcopy.def
+        COMMAND objcopy -v --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/lib${OpenBLAS_LIBNAME}.so
+        COMMENT "renaming symbols"
+        )
+endif()
+
+
 # Install project
 
 # Install libraries
 install(TARGETS ${OpenBLAS_LIBNAME}
+	EXPORT "OpenBLAS${SUFFIX64}Targets"
 	RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
 	ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
   LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR} )
 
+# Install headers
+set(CMAKE_INSTALL_INCLUDEDIR ${CMAKE_INSTALL_INCLUDEDIR}/openblas${SUFFIX64})
+set(CMAKE_INSTALL_FULL_INCLUDEDIR ${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_INCLUDEDIR})
+
 message(STATUS "Generating openblas_config.h in ${CMAKE_INSTALL_INCLUDEDIR}")
 
 set(OPENBLAS_CONFIG_H ${CMAKE_BINARY_DIR}/openblas_config.h)
@@ -259,11 +342,31 @@ if(NOT NO_LAPACKE)
 	ADD_CUSTOM_TARGET(genlapacke
 	COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/lapack-netlib/LAPACKE/include/lapacke_mangling_with_flags.h.in "${CMAKE_BINARY_DIR}/lapacke_mangling.h"
 	)
-	install (FILES ${CMAKE_BINARY_DIR}/lapacke_mangling.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
+	install (FILES ${CMAKE_BINARY_DIR}/lapacke_mangling.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/openblas${SUFFIX64})
 endif()
 
 include(FindPkgConfig QUIET)
 if(PKG_CONFIG_FOUND)
-	configure_file(${PROJECT_SOURCE_DIR}/cmake/openblas.pc.in ${PROJECT_BINARY_DIR}/openblas.pc @ONLY)
+	configure_file(${PROJECT_SOURCE_DIR}/cmake/openblas.pc.in ${PROJECT_BINARY_DIR}/openblas${SUFFIX64}.pc @ONLY)
-	install (FILES ${PROJECT_BINARY_DIR}/openblas.pc DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig/)
+	install (FILES ${PROJECT_BINARY_DIR}/openblas${SUFFIX64}.pc DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig/)
 endif()
+
+
+# GNUInstallDirs "DATADIR" wrong here; CMake search path wants "share".
+set(PN OpenBLAS)
+set(CMAKECONFIG_INSTALL_DIR "share/cmake/${PN}${SUFFIX64}")
+configure_package_config_file(cmake/${PN}Config.cmake.in
+                              "${CMAKE_CURRENT_BINARY_DIR}/${PN}${SUFFIX64}Config.cmake"
+                              INSTALL_DESTINATION ${CMAKECONFIG_INSTALL_DIR})
+write_basic_package_version_file(${CMAKE_CURRENT_BINARY_DIR}/${PN}ConfigVersion.cmake
+                                 VERSION ${${PN}_VERSION}
+                                 COMPATIBILITY AnyNewerVersion)
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${PN}${SUFFIX64}Config.cmake
+        DESTINATION ${CMAKECONFIG_INSTALL_DIR})
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${PN}ConfigVersion.cmake
+        RENAME ${PN}${SUFFIX64}ConfigVersion.cmake
+        DESTINATION ${CMAKECONFIG_INSTALL_DIR})
+install(EXPORT "${PN}${SUFFIX64}Targets"
+        NAMESPACE "${PN}${SUFFIX64}::"
+        DESTINATION ${CMAKECONFIG_INSTALL_DIR})
+

Changelog.txt

@@ -1,4 +1,247 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.5
+31-Dec-2018
+
+common:
+	* loop unrolling in TRMV has been enabled again.
+	* A domain error in the thread workload distribution for SYRK
+	  has been fixed.
+	* gmake builds will now automatically add -fPIC to the build
+	  options if the platform requires it.
+	* a pthreads key leakage (and associate crash on dlclose) in
+	  the USE_TLS codepath was fixed.
+	* building of the utest cases on systems that do not provide
+	  an implementation of complex.h was fixed.
+	  
+x86_64:
+	* the SkylakeX code was changed to compile on OSX.
+	* unwanted application of the -march=skylake-avx512 option
+	  to the common code parts of a DYNAMIC_ARCH build was fixed.
+	* improved performance of SGEMM for small workloads on Skylake X.
+	* performance of SGEMM and DGEMM was improved on Haswell.
+
+ARMV8:
+	* a configuration error that broke the CNRM2 kernel was corrected.
+	* compilation of the GEMM kernels with CMAKE was fixed.
+	* DYNAMIC_ARCH builds are now available with CMAKE as well.
+	* using CMAKE for cross-compilation to the new cpu TARGETs
+	  introduced in 0.3.4 now works.
+	  
+POWER:
+	* a problem in cpu autodetection for AIX has been corrected.
+	
+====================================================================
+Version 0.3.4
+02-Dec-2018
+
+common:
+	* the new, experimental thread-local memory allocation had 
+	  inadvertently been left enabled for gmake builds in 0.3.3
+	  despite the announcement. It is now disabled by default, and
+	  single-threaded builds will keep using the old allocator even
+	  if the USE_TLS option is turned on.
+	* OpenBLAS will now provide enough buffer space for at least 50
+	  threads by default.
+	* The output of openblas_get_config() now contains the version
+	  number.
+	* A serious thread safety bug in GEMV operation with small M and
+	  large N size has been fixed.
+	* The code will now automatically call blas_thread_init after a
+	  fork if needed before handling a call to openblas_set_num_threads
+	* Accesses to parallelized level3 functions from multiple callers
+	  are now serialized to avoid thread races (unless using OpenMP).
+	  This should provide better performance than the known-threadsafe
+	  (but non-default) USE_SIMPLE_THREADED_LEVEL3 option.
+	* When building LAPACK with gfortran, -frecursive is now (again)
+	  enabled by default to ensure correct behaviour.
+        * The OpenBLAS version cblas.h now supports both CBLAS_ORDER and
+	  CBLAS_LAYOUT as the name of the matrix row/column order option.
+	* Externally set LDFLAGS are now passed through to the final compile/link
+	  steps to facilitate setting platform-specific linker flags.
+	* A potential race condition during the build of LAPACK (that would 
+	  usually manifest itself as a failure to build TESTING/MATGEN) has been 
+	  fixed.
+	* xHEMV has been changed to stay single-threaded for small input sizes
+	  where the overhead of multithreading exceeds any possible gains
+	* CSWAP and ZSWAP have been limited to a single thread except on ARMV8 or
+	  ThunderX hardware with sizable input.
+	* Linker flags for the PGI compiler have been updated
+	* Behaviour of AXPY with zero increments is now handled in the C interface,
+	  correcting the result on at least Intel Atom.
+	* The result matrix from calling SGELSS with an all-zero input matrix is 
+	  now zeroed completely.
+	  
+x86_64:
+	* Autodetection of AMD Ryzen2 has been fixed (again).
+        * CMAKE builds now support labeling of an INTERFACE64=1 build of
+	  the library with the _64 suffix.
+	* AVX512 version of DGEMM has been added and the AVX512 SGEMM kernel
+	  has been sped up by rewriting with C intrinsics
+	* Fixed compilation on RHEL5/CENTOS5 (issue with typename __WAIT_STATUS)
+	
+POWER:
+	* added support for building on AIX (with gcc and GNU tools from AIX Toolbox).
+	* CPU type detection has been implemented for AIX.
+	* CPU type detection has been fixed for NETBSD.
+	
+MIPS64:
+	* AXPY on LOONGSON3A has been corrected to pass "zero increment" utest.
+	* DSDOT on LOONGSON3A has been fixed.
+	* the SGEMM microkernel has been hardened against potential data loss.
+	
+ARMV8:
+	* DYNAMic_ARCH support is now available for 64bit ARM
+	* cross-compiling for ARMV8 under iOS now works.
+	* cpu-specific code has been rearranged to make better use of both
+	  hardware commonalities and model-specific compiler optimizations.
+	* XGENE1 has been removed as a TARGET, superseded by the improved generic
+	  ARMV8 support.
+	
+ARMV7:
+	* Older assembly mnemonics have been converted to UAL form to allow
+	  building with clang 7.0
+	* Cross compiling LAPACKE for Android has been fixed again (broken by
+	  update to LAPACK 3.7.0 some while ago).  
+	  
+====================================================================
+Version 0.3.3
+31-Aug-2018
+
+common:
+	* thread memory allocation has been switched back to the method
+	  used before version 0.3.1 due to unexpected problems caused by
+	  the new code under some circumstances. A new compile-time option
+	  USE_TLS has been added to enable the new code, and it is hoped
+	  that this can become the default again in the next version.
+	* LAPAck PR272 has been integrated, which fixes spurious errors
+	  in DSYEVR and related functions caused by missing conversion 
+	  from ILAENV to ILAENV_2STAGE in several _2stage routines.
+	* the cmake-generated OpenBLASConfig.cmake now uses correct case
+	  for the name of the library
+	* added support for Haiku OS  
+
+x86_64:
+	* added AVX512 implementations of SDOT, DDOT, SAXPY, DAXPY,
+	  DSCAL, DGEMVN and DSYMVL 
+	* added a workaround for a cygwin issue that prevented compilation
+	  of AVX512 code
+	  
+IBM Z:
+	* added autodetection of Z14
+	* fixed TRMM errors in the generic target
+	
+====================================================================
+Version 0.3.2
+30-Jul-2018
+
+common:
+	* fixes for regressions caused by the rewrite of the thread
+	  initialization code in 0.3.1
+
+POWER:
+	* fixed cpu autodetection for the BSDs
+
+MIPS64:
+	* fixed utest errors in AXPY, DSDOT, ROT and SWAP
+
+x86_64:
+	* added autodetection of AMD Ryzen 2
+	* fixed build with older versions of MSVC
+
+====================================================================
+Version 0.3.1
+01-Jul-2018
+
+common:
+	* rewritten thread initialization code with significantly reduced overhead
+	* added CBLAS interfaces to the IxAMIN BLAS extension functions
+	* fixed the lapack-test target
+	* CMAKE builds now create an OpenBLASConfig.cmake file
+	* ZAXPY now uses a single thread for small input sizes
+	* the LAPACK code was updated from Reference-LAPACK/lapack#253
+          (fixing LAPACKE interfaces to Aasen's functions)
+
+POWER:
+	* corrected CROT and ZROT behaviour with zero INC_X
+
+ARMV7:
+	* corrected xDOT behaviour with zero INC_X or INC_Y
+
+x86_64:
+	* retired some older targets of DYNAMIC_ARCH builds to a new option DYNAMIC_OLDER,
+	  this affects PENRYN,DUNNINGTON,OPTERON,OPTERON_SSE3,BOBCAT,ATOM and NANO
+	  (which will still be supported via the slower PRESCOTT kernels when this option is not set)
+	* added an option DYNAMIC_LIST that (used in conjunction with DYNAMIC_ARCH) allows to 
+	  specify the list of x86_64 targets to include. Any target not on the list will be supported 
+	  by the Sandybridge or Nehalem kernels if available, or by Prescott.
+	* improved SWITCH_RATIO on Haswell for increased GEMM throughput
+	* added initial support for Intel Skylake X, including an AVX512 SGEMM kernel
+	* added autodetection of Intel Cannon Lake series as Skylake X
+	* added a default L2 cache size for hypervisors that return zero here (Chromebook)
+	* fixed a name clash with recent Windows10 headers that broke the build with (at least)
+	  recent mingw from MSYS2
+	* fixed a link error in mixed clang/gfortran builds with OpenMP
+	* updated the OSX deployment target to 10.8
+	* switched on parallel make for builds on MS Windows by default
+
+x86:
+	* fixed SSWAP and DSWAP behaviour with zero INC_X and INC_Y
+
+====================================================================
+Version 0.3.0
+23-May-2108
+
+common:
+	* fixed some more thread race and locking bugs
+	* added preliminary support for calling an OpenMP build of the library from multiple threads
+	* removed performance impact of thread locks added in 0.2.20 on OpenMP code
+	* general code cleanup 
+	* optimized DSDOT implementation
+	* improved thread distribution for GEMM
+	* corrected IMATCOPY/OMATCOPY implementation
+	* fixed out-of-bounds accesses in the multithreaded xBMV/xPMV and SYMV implementations
+	* cmake build improvements
+	* pkgconfig file now contains build options
+	* openblas_get_config() now reports USE_OPENMP and NUM_THREADS settings used for the build
+	* corrections and improvements for systems with more than 64 cpus
+	* LAPACK code updated to 3.8.0 including later fixes
+	* added ReLAPACK, a recursive implementation of several LAPACK functions
+	* Rewrote ROTMG to handle cases that the netlib code failed to address
+	* Disabled (broken) multithreading code for xTRMV
+	* corrected prototypes of complex CBLAS functions to make our cblas.h match the generally accepted standard
+	* shared memory access failures on startup are now handled more gracefully
+	* restored utests from earlier releases (and made them pass on all affected systems)
+
+SPARC:
+	* several fixes for cpu autodetection
+
+POWER:
+	* corrected vector register overwriting in several Power8 kernels
+	* optimized additional BLAS functions
+
+ARM:
+	* added support for CortexA53 and A72 
+	* added autodetection for ThunderX2T99
+	* made most optimized kernels the default for generic ARMv8 targets 
+
+x86_64:
+	* parallelized DDOT kernel for Haswell
+	* changed alignment directives in assembly kernels to boost performance on OSX
+	* fixed register handling in the GEMV microkernels (bug exposed by gcc7)
+	* added support for building on OpenBSD and Dragonfly 
+	* updated compiler options to work with Intel release 2018
+	* support fully optimized build with clang/flang on Microsoft Windows
+	* fixed building on AIX
+
+IBM Z:
+	* added optimized BLAS 1/2 functions
+
+MIPS:
+	* fixed cpu autodetection helper code
+	* added mips32 1004K cpu (Mediatek MT7621 and similar SoC)
+	* added mips64 I6500 cpu
+
 ====================================================================
 Version 0.2.20
 24-Jul-2017

Makefile

@@ -21,6 +21,17 @@ ifeq ($(BUILD_RELAPACK), 1)
 RELA = re_lapack
 endif
 
+ifeq ($(NO_FORTRAN), 1)
+define NOFORTRAN
+1
+endef
+define NO_LAPACK
+1
+endef
+export NOFORTRAN
+export NO_LAPACK
+endif
+
 LAPACK_NOOPT := $(filter-out -O0 -O1 -O2 -O3 -Ofast,$(LAPACK_FFLAGS))
 
 SUBDIRS_ALL = $(SUBDIRS) test ctest utest exports benchmark ../laswp ../bench
@@ -47,7 +58,7 @@ endif
 endif
 
 	@echo "  C compiler       ... $(C_COMPILER)  (command line : $(CC))"
-ifndef NOFORTRAN
+ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	@echo "  Fortran compiler ... $(F_COMPILER)  (command line : $(FC))"
 endif
 ifneq ($(OSNAME), AIX)
@@ -86,16 +97,12 @@ endif
 
 shared :
 ifndef NO_SHARED
-ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android))
+ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku))
 	@$(MAKE) -C exports so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
-ifeq ($(OSNAME), FreeBSD)
+ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
-	@$(MAKE) -C exports so
-	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
-endif
-ifeq ($(OSNAME), NetBSD)
 	@$(MAKE) -C exports so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
 endif
@@ -112,7 +119,7 @@ endif
 endif
 
 tests :
-ifndef NOFORTRAN
+ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	touch $(LIBNAME)
 ifndef NO_FBLAS
 	$(MAKE) -C test all
@@ -124,7 +131,7 @@ endif
 endif
 
 libs :
-ifeq ($(CORE), UNKOWN)
+ifeq ($(CORE), UNKNOWN)
 	$(error OpenBLAS: Detecting CPU failed. Please set TARGET explicitly, e.g. make TARGET=your_cpu_target. Please read README for the detail.)
 endif
 ifeq ($(NOFORTRAN), 1)
@@ -157,6 +164,9 @@ ifeq ($(DYNAMIC_ARCH), 1)
 	do  $(MAKE) GOTOBLAS_MAKEFILE= -C kernel TARGET_CORE=$$d kernel || exit 1 ;\
 	done
 	@echo DYNAMIC_ARCH=1 >> Makefile.conf_last
+ifeq ($(DYNAMIC_OLDER), 1)
+	@echo DYNAMIC_OLDER=1 >> Makefile.conf_last
+endif	
 endif
 ifdef USE_THREAD
 	@echo USE_THREAD=$(USE_THREAD) >>  Makefile.conf_last
@@ -211,7 +221,7 @@ netlib :
 
 else
 netlib : lapack_prebuild
-ifndef NOFORTRAN
+ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	@$(MAKE) -C $(NETLIB_LAPACK_DIR) lapacklib
 	@$(MAKE) -C $(NETLIB_LAPACK_DIR) tmglib
 endif
@@ -232,7 +242,7 @@ prof_lapack : lapack_prebuild
 	@$(MAKE) -C $(NETLIB_LAPACK_DIR) lapack_prof
 
 lapack_prebuild :
-ifndef NOFORTRAN
+ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	-@echo "FORTRAN     = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "OPTS        = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "POPTS       = $(LAPACK_FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
@@ -241,7 +251,7 @@ ifndef NOFORTRAN
 	-@echo "LOADOPTS    = $(FFLAGS) $(EXTRALIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "CC          = $(CC)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "override CFLAGS      = $(LAPACK_CFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "ARCH        = $(AR)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "override ARCH        = $(AR)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "ARCHFLAGS   = $(ARFLAGS) -ru" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "RANLIB      = $(RANLIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "LAPACKLIB   = ../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
@@ -257,6 +267,8 @@ ifeq ($(F_COMPILER), GFORTRAN)
 ifdef SMP
 ifeq ($(OSNAME), WINNT)
 	-@echo "LOADER      = $(FC)" >> $(NETLIB_LAPACK_DIR)/make.inc
+else ifeq ($(OSNAME), Haiku)
+	-@echo "LOADER      = $(FC)" >> $(NETLIB_LAPACK_DIR)/make.inc
 else
 	-@echo "LOADER      = $(FC) -pthread" >> $(NETLIB_LAPACK_DIR)/make.inc
 endif
@@ -275,21 +287,21 @@ endif
 endif
 
 large.tgz :
-ifndef NOFORTRAN
+ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	if [ ! -a $< ]; then
 	-wget http://www.netlib.org/lapack/timing/large.tgz;
 	fi
 endif
 
 timing.tgz :
-ifndef NOFORTRAN
+ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	if [ ! -a $< ]; then
 	-wget http://www.netlib.org/lapack/timing/timing.tgz;
 	fi
 endif
 
 lapack-timing : large.tgz timing.tgz
-ifndef NOFORTRAN
+ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	(cd $(NETLIB_LAPACK_DIR); $(TAR) zxf ../timing.tgz TIMING)
 	(cd $(NETLIB_LAPACK_DIR)/TIMING; $(TAR) zxf ../../large.tgz )
 	$(MAKE) -C $(NETLIB_LAPACK_DIR)/TIMING
@@ -298,9 +310,10 @@ endif
 
 lapack-test :
 	(cd $(NETLIB_LAPACK_DIR)/TESTING && rm -f x* *.out)
-	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING xeigtstc  xeigtstd  xeigtsts  xeigtstz  xlintstc  xlintstd  xlintstds  xlintstrfd  xlintstrfz  xlintsts  xlintstz  xlintstzc xlintstrfs xlintstrfc
+	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING/EIG xeigtstc  xeigtstd  xeigtsts  xeigtstz 
+	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING/LIN xlintstc  xlintstd  xlintstds  xlintstrfd  xlintstrfz  xlintsts  xlintstz  xlintstzc xlintstrfs xlintstrfc
 ifneq ($(CROSS), 1)
-	( cd $(NETLIB_LAPACK_DIR)/INSTALL; ./testlsame; ./testslamch; ./testdlamch; \
+	( cd $(NETLIB_LAPACK_DIR)/INSTALL; make all; ./testlsame; ./testslamch; ./testdlamch; \
         ./testsecond; ./testdsecnd; ./testieee; ./testversion )
 	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r )
 endif
@@ -312,9 +325,9 @@ lapack-runtest:
 
 
 blas-test:
-	(cd $(NETLIB_LAPACK_DIR)/BLAS && rm -f x* *.out)
+	(cd $(NETLIB_LAPACK_DIR)/BLAS/TESTING && rm -f x* *.out)
 	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR) blas_testing
-	(cd $(NETLIB_LAPACK_DIR)/BLAS && cat *.out)
+	(cd $(NETLIB_LAPACK_DIR)/BLAS/TESTING && cat *.out)
 
 
 dummy :

Makefile.arm64

@@ -4,22 +4,37 @@ CCOMMON_OPT += -march=armv8-a
 FCOMMON_OPT += -march=armv8-a
 endif
 
-ifeq ($(CORE), CORTEXA57)
+ifeq ($(CORE), CORTEXA53)
-CCOMMON_OPT += -march=armv8-a+crc+crypto+fp+simd -mtune=cortex-a57
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
-FCOMMON_OPT += -march=armv8-a+crc+crypto+fp+simd -mtune=cortex-a57
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
 endif
 
-ifeq ($(CORE), VULCAN)
+ifeq ($(CORE), CORTEXA57)
-CCOMMON_OPT += -mtune=vulcan -mcpu=vulcan
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a57
-FCOMMON_OPT += -mtune=vulcan -mcpu=vulcan
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a57
+endif
+
+ifeq ($(CORE), CORTEXA72)
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a72
+endif
+
+ifeq ($(CORE), CORTEXA73)
+CCOMMON_OPT += -march=armv8-a -mtune=cortex-a73
+FCOMMON_OPT += -march=armv8-a -mtune=cortex-a73
 endif
 
 ifeq ($(CORE), THUNDERX)
-CCOMMON_OPT += -mtune=thunderx -mcpu=thunderx
+CCOMMON_OPT += -march=armv8-a -mtune=thunderx
-FCOMMON_OPT += -mtune=thunderx -mcpu=thunderx
+FCOMMON_OPT += -march=armv8-a -mtune=thunderx
+endif
+
+ifeq ($(CORE), FALKOR)
+CCOMMON_OPT += -march=armv8-a -mtune=falkor
+FCOMMON_OPT += -march=armv8-a -mtune=falkor
 endif
 
 ifeq ($(CORE), THUNDERX2T99)
-CCOMMON_OPT += -mtune=thunderx2t99 -mcpu=thunderx2t99
+CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
-FCOMMON_OPT += -mtune=thunderx2t99 -mcpu=thunderx2t99
+FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 endif

Makefile.install

@@ -48,6 +48,7 @@ ifndef NO_CBLAS
 	@sed 's/common/openblas_config/g' cblas.h > "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/cblas.h"
 endif
 
+ifneq ($(OSNAME), AIX)
 ifndef NO_LAPACKE
 	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
 	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h"
@@ -66,18 +67,14 @@ endif
 #for install shared library
 ifndef NO_SHARED
 	@echo Copying the shared library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
-ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android))
+ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku))
 	@install -pm755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
-ifeq ($(OSNAME), FreeBSD)
+
-	@cp $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
-	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
-	ln -fs $(LIBSONAME) $(LIBPREFIX).so
-endif
-ifeq ($(OSNAME), NetBSD)
 	@cp $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so
@@ -98,11 +95,39 @@ ifeq ($(OSNAME), CYGWIN_NT)
 endif
 endif
 
+else
+#install on AIX has different options syntax
+ifndef NO_LAPACKE
+	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
+	@-installbsd -c -m 644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h"
+	@-installbsd -c -m 644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_config.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_config.h"
+	@-installbsd -c -m 644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_mangling_with_flags.h.in "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h"
+	@-installbsd -c -m 644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_utils.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_utils.h"
+endif
+
+#for install static library
+ifndef NO_STATIC
+	@echo Copying the static library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
+	@installbsd -c -m 644 $(LIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
+	ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
+endif
+#for install shared library
+ifndef NO_SHARED
+	@echo Copying the shared library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
+	@installbsd -c -m 755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
+	ln -fs $(LIBSONAME) $(LIBPREFIX).so ; \
+	ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
+endif
+
+endif
 
 #Generating openblas.pc
 	@echo Generating openblas.pc in "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)"
-	@echo 'libdir='$(OPENBLAS_LIBRARY_DIR) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
+	@echo 'libdir='$(OPENBLAS_LIBRARY_DIR) > "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@echo 'includedir='$(OPENBLAS_INCLUDE_DIR) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
+	@echo 'openblas_config= USE_64BITINT='$(USE_64BITINT) 'DYNAMIC_ARCH='$(DYNAMIC_ARCH) 'DYNAMIC_OLDER='$(DYNAMIC_OLDER) 'NO_CBLAS='$(NO_CBLAS) 'NO_LAPACK='$(NO_LAPACK) 'NO_LAPACKE='$(NO_LAPACKE) 'NO_AFFINITY='$(NO_AFFINITY) 'USE_OPENMP='$(USE_OPENMP) $(CORE) 'MAX_THREADS='$(NUM_THREADS)>> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@echo 'version='$(VERSION) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@echo 'extralib='$(EXTRALIB) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@cat openblas.pc.in >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
@@ -115,7 +140,7 @@ endif
 
 ifndef NO_SHARED
 #ifeq logical or
-ifeq ($(OSNAME), $(filter $(OSNAME),Linux FreeBSD NetBSD))
+ifeq ($(OSNAME), $(filter $(OSNAME),Linux FreeBSD NetBSD OpenBSD DragonFly))
 	@echo "SET(OpenBLAS_LIBRARIES ${OPENBLAS_LIBRARY_DIR}/$(LIBPREFIX).so)" >> "$(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG)"
 endif
 ifeq ($(OSNAME), $(filter $(OSNAME),WINNT CYGWIN_NT))

Makefile.prebuild

@@ -17,6 +17,10 @@ ifdef CPUIDEMU
 EXFLAGS = -DCPUIDEMU -DVENDOR=99
 endif
 
+ifeq ($(TARGET), 1004K)
+TARGET_FLAGS = -mips32r2
+endif
+
 ifeq ($(TARGET), P5600)
 TARGET_FLAGS = -mips32r5
 endif

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.0.dev
+VERSION = 0.3.6.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
@@ -17,6 +17,11 @@ VERSION = 0.3.0.dev
 # If you want to support multiple architecture in one binary
 # DYNAMIC_ARCH = 1
 
+# If you want the full list of x86_64 architectures supported in DYNAMIC_ARCH
+# mode (including individual optimizied codes for PENRYN, DUNNINGTON, OPTERON,
+# OPTERON_SSE3, ATOM and NANO rather than fallbacks to older architectures)
+# DYNAMIC_OLDER = 1
+
 # C compiler including binary type(32bit / 64bit). Default is gcc.
 # Don't use Intel Compiler or PGI, it won't generate right codes as I expect.
 # CC = gcc
@@ -55,11 +60,26 @@ VERSION = 0.3.0.dev
 # This flag is always set for POWER8. Don't modify the flag 
 # USE_OPENMP = 1
 
+# The OpenMP scheduler to use - by default this is "static" and you
+# will normally not want to change this unless you know that your main
+# workload will involve tasks that have highly unbalanced running times
+# for individual threads. Changing away from "static" may also adversely
+# affect memory access locality in NUMA systems. Setting to "runtime" will
+# allow you to select the scheduler from the environment variable OMP_SCHEDULE
+# CCOMMON_OPT += -DOMP_SCHED=dynamic
+
 # You can define maximum number of threads. Basically it should be
 # less than actual number of cores. If you don't specify one, it's
 # automatically detected by the the script.
 # NUM_THREADS = 24
 
+# If you have enabled USE_OPENMP and your application would call
+# OpenBLAS's calculation API from multi threads, please comment it in.
+# This flag defines how many instances of OpenBLAS's calculation API can 
+# actually run in parallel. If more threads call OpenBLAS's calculation API,
+# they need to wait for the preceding API calls to finish or risk data corruption.
+# NUM_PARALLEL = 2
+
 # if you don't need to install the static library, please comment it in.
 # NO_STATIC = 1
 
@@ -89,6 +109,12 @@ BUILD_LAPACK_DEPRECATED = 1
 # If you want to use legacy threaded Level 3 implementation.
 # USE_SIMPLE_THREADED_LEVEL3 = 1
 
+# If you want to use the new, still somewhat experimental code that uses
+# thread-local storage instead of a central memory buffer in memory.c
+# Note that if your system uses GLIBC, it needs to have at least glibc 2.21
+# for this to work.
+# USE_TLS = 1
+
 # If you want to drive whole 64bit region by BLAS. Not all Fortran
 # compiler supports this. It's safe to keep comment it out if you
 # are not sure(equivalent to "-i8" option).
@@ -100,7 +126,7 @@ BUILD_LAPACK_DEPRECATED = 1
 NO_WARMUP = 1
 
 # If you want to disable CPU/Memory affinity on Linux.
-#NO_AFFINITY = 1
+NO_AFFINITY = 1
 
 # if you are compiling for Linux and you have more than 16 numa nodes or more than 256 cpus
 # BIGNUMA = 1
@@ -126,6 +152,9 @@ NO_WARMUP = 1
 # FUNCTION_PROFILE = 1
 
 # Support for IEEE quad precision(it's *real* REAL*16)( under testing)
+# This option should not be used - it is a holdover from unfinished code present
+# in the original GotoBLAS2 library that may be usable as a starting point but
+# is not even expected to compile in its present form.
 # QUAD_PRECISION = 1
 
 # Theads are still working for a while after finishing BLAS operation
@@ -144,8 +173,11 @@ NO_WARMUP = 1
 # CONSISTENT_FPCSR = 1
 
 # If any gemm arguement m, n or k is less or equal this threshold, gemm will be execute
-# with single thread. You can use this flag to avoid the overhead of multi-threading
+# with single thread. (Actually in recent versions this is a factor proportional to the
-# in small matrix sizes. The default value is 4.
+# number of floating point operations necessary for the given problem size, no longer
+# an individual dimension). You can use this setting to avoid the overhead of multi-
+# threading in small matrix sizes. The default value is 4, but values as high as 50 have 
+# been reported to be optimal for certain workloads (50 is the recommended value for Julia).
 # GEMM_MULTITHREAD_THRESHOLD = 4
 
 # If you need santy check by comparing reference BLAS. It'll be very
@@ -160,8 +192,8 @@ NO_WARMUP = 1
 # Flags for POWER8 are defined in Makefile.power. Don't modify COMMON_OPT
 # COMMON_OPT = -O2
 
-# gfortran option for LAPACK
+# gfortran option for LAPACK to improve thread-safety
-# enable this flag only on 64bit Linux and if you need a thread safe lapack library
+# It is enabled by default in Makefile.system for gfortran
 # Flags for POWER8 are defined in Makefile.power. Don't modify FCOMMON_OPT
 # FCOMMON_OPT = -frecursive
 

Makefile.system

@@ -9,6 +9,17 @@ ifndef TOPDIR
 TOPDIR = .
 endif
 
+# Catch conflicting usage of ARCH in some BSD environments
+ifeq ($(ARCH), amd64)
+override ARCH=x86_64
+else ifeq ($(ARCH), powerpc64)
+override ARCH=power
+else ifeq ($(ARCH), i386)
+override ARCH=x86
+else ifeq ($(ARCH), aarch64)
+override ARCH=arm64
+endif
+
 NETLIB_LAPACK_DIR = $(TOPDIR)/lapack-netlib
 
 # Default C compiler
@@ -17,15 +28,24 @@ NETLIB_LAPACK_DIR = $(TOPDIR)/lapack-netlib
 #   http://stackoverflow.com/questions/4029274/mingw-and-make-variables
 # - Default value is 'cc' which is not always a valid command (e.g. MinGW).
 ifeq ($(origin CC),default)
+
+# Check if $(CC) refers to a valid command and set the value to gcc if not
+ifneq ($(findstring cmd.exe,$(SHELL)),)
+ifeq ($(shell where $(CC) 2>NUL),)
 CC = gcc
-# Change the default compile to clang on Mac OSX.
-# http://stackoverflow.com/questions/714100/os-detecting-makefile
-UNAME_S := $(shell uname -s)
-ifeq ($(UNAME_S),Darwin)
-     CC = clang
-#     EXTRALIB += -Wl,-no_compact_unwind
-endif
 endif
+else # POSIX-ish
+ifeq ($(shell command -v $(CC) 2>/dev/null),)
+ifeq ($(shell uname -s),Darwin)
+CC = clang
+# EXTRALIB += -Wl,-no_compact_unwind
+else
+CC = gcc
+endif # Darwin
+endif # CC exists
+endif # Shell is sane
+
+endif # CC is set to default
 
 # Default Fortran compiler (FC) is selected by f_check.
 
@@ -45,6 +65,7 @@ endif
 
 ifdef TARGET
 GETARCH_FLAGS := -DFORCE_$(TARGET)
+GETARCH_FLAGS += -DUSER_TARGET
 endif
 
 # Force fallbacks for 32bit
@@ -53,6 +74,9 @@ ifeq ($(BINARY), 32)
 ifeq ($(TARGET), HASWELL)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
+ifeq ($(TARGET), SKYLAKEX)
+GETARCH_FLAGS := -DFORCE_NEHALEM
+endif
 ifeq ($(TARGET), SANDYBRIDGE)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
@@ -86,6 +110,9 @@ ifeq ($(BINARY), 32)
 ifeq ($(TARGET_CORE), HASWELL)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
+ifeq ($(TARGET_CORE), SKYLAKEX)
+GETARCH_FLAGS := -DFORCE_NEHALEM
+endif
 ifeq ($(TARGET_CORE), SANDYBRIDGE)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
@@ -132,6 +159,10 @@ ifeq ($(NO_AVX2), 1)
 GETARCH_FLAGS	+= -DNO_AVX2
 endif
 
+ifeq ($(NO_AVX512), 1)
+GETARCH_FLAGS	+= -DNO_AVX512
+endif
+
 ifeq ($(DEBUG), 1)
 GETARCH_FLAGS	+= -g
 endif
@@ -175,6 +206,10 @@ endif
 
 endif
 
+ifndef NUM_PARALLEL
+NUM_PARALLEL = 1
+endif
+
 ifndef NUM_THREADS
 NUM_THREADS = $(NUM_CORES)
 endif
@@ -225,12 +260,12 @@ endif
 
 ifeq ($(OSNAME), Darwin)
 ifndef MACOSX_DEPLOYMENT_TARGET
-export MACOSX_DEPLOYMENT_TARGET=10.6
+export MACOSX_DEPLOYMENT_TARGET=10.8
 endif
 MD5SUM = md5 -r
 endif
 
-ifeq ($(OSNAME), FreeBSD)
+ifneq (,$(findstring $(OSNAME), FreeBSD OpenBSD DragonFly))
 MD5SUM = md5 -r
 endif
 
@@ -304,6 +339,7 @@ endif
 ifeq ($(OSNAME), CYGWIN_NT)
 NEED_PIC = 0
 NO_EXPRECISION = 1
+OS_CYGWIN_NT = 1
 endif
 
 ifneq ($(OSNAME), WINNT)
@@ -423,7 +459,7 @@ CCOMMON_OPT    += -fopenmp
 endif
 
 ifeq ($(C_COMPILER), INTEL)
-CCOMMON_OPT    += -openmp
+CCOMMON_OPT    += -fopenmp
 endif
 
 ifeq ($(C_COMPILER), PGI)
@@ -448,13 +484,44 @@ DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
 endif
 
 ifeq ($(ARCH), x86_64)
-DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
+DYNAMIC_CORE = PRESCOTT CORE2 
+ifeq ($(DYNAMIC_OLDER), 1)
+DYNAMIC_CORE += PENRYN DUNNINGTON 
+endif
+DYNAMIC_CORE += NEHALEM 
+ifeq ($(DYNAMIC_OLDER), 1)
+DYNAMIC_CORE += OPTERON OPTERON_SSE3 
+endif
+DYNAMIC_CORE += BARCELONA 
+ifeq ($(DYNAMIC_OLDER), 1)
+DYNAMIC_CORE += BOBCAT ATOM NANO
+endif
 ifneq ($(NO_AVX), 1)
 DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR
 endif
 ifneq ($(NO_AVX2), 1)
 DYNAMIC_CORE += HASWELL ZEN
 endif
+ifneq ($(NO_AVX512), 1)
+ifneq ($(NO_AVX2), 1)
+DYNAMIC_CORE += SKYLAKEX
+endif
+endif
+endif
+
+ifdef DYNAMIC_LIST
+override DYNAMIC_CORE = PRESCOTT $(DYNAMIC_LIST)
+XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_PRESCOTT
+XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
+CCOMMON_OPT += $(XCCOMMON_OPT)
+#CCOMMON_OPT	+= -DDYNAMIC_LIST='$(DYNAMIC_LIST)'
+endif
+
+ifeq ($(ARCH), arm64)
+DYNAMIC_CORE =  ARMV8
+DYNAMIC_CORE += CORTEXA57
+DYNAMIC_CORE += THUNDERX
+DYNAMIC_CORE += THUNDERX2T99
 endif
 
 # If DYNAMIC_CORE is not set, DYNAMIC_ARCH cannot do anything, so force it to empty
@@ -554,9 +621,14 @@ CCOMMON_OPT += -march=mips64
 FCOMMON_OPT += -march=mips64
 endif
 
+ifeq ($(CORE), 1004K)
+CCOMMON_OPT += -mips32r2   $(MSA_FLAGS)
+FCOMMON_OPT += -mips32r2   $(MSA_FLAGS)
+endif
+
 ifeq ($(CORE), P5600)
-CCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600 $(MSA_FLAGS)
+CCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600  $(MSA_FLAGS)
-FCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600 $(MSA_FLAGS)
+FCOMMON_OPT += -mips32r5 -mnan=2008 -mtune=p5600  $(MSA_FLAGS)
 endif
 
 ifeq ($(CORE), I6400)
@@ -660,6 +732,8 @@ endif
 ifeq ($(F_COMPILER), GFORTRAN)
 CCOMMON_OPT += -DF_INTERFACE_GFORT
 FCOMMON_OPT += -Wall
+# make single-threaded LAPACK calls thread-safe #1847
+FCOMMON_OPT += -frecursive
 #Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
 ifneq ($(NO_LAPACK), 1)
 EXTRALIB += -lgfortran
@@ -703,7 +777,7 @@ FCOMMON_OPT += -i8
 endif
 endif
 ifeq ($(USE_OPENMP), 1)
-FCOMMON_OPT += -openmp
+FCOMMON_OPT += -fopenmp
 endif
 endif
 
@@ -883,6 +957,10 @@ ifeq ($(DYNAMIC_ARCH), 1)
 CCOMMON_OPT	+= -DDYNAMIC_ARCH
 endif
 
+ifeq ($(DYNAMIC_OLDER), 1)
+CCOMMON_OPT	+= -DDYNAMIC_OLDER
+endif
+
 ifeq ($(NO_LAPACK), 1)
 CCOMMON_OPT	+= -DNO_LAPACK
 #Disable LAPACK C interface
@@ -905,6 +983,10 @@ ifeq ($(NO_AVX2), 1)
 CCOMMON_OPT	+= -DNO_AVX2
 endif
 
+ifeq ($(NO_AVX512), 1)
+CCOMMON_OPT	+= -DNO_AVX512
+endif
+
 ifdef SMP
 CCOMMON_OPT	+= -DSMP_SERVER
 
@@ -951,10 +1033,18 @@ endif
 
 CCOMMON_OPT	+= -DMAX_CPU_NUMBER=$(NUM_THREADS)
 
+CCOMMON_OPT	+= -DMAX_PARALLEL_NUMBER=$(NUM_PARALLEL)
+
 ifdef USE_SIMPLE_THREADED_LEVEL3
 CCOMMON_OPT	+= -DUSE_SIMPLE_THREADED_LEVEL3
 endif
 
+ifdef USE_TLS
+CCOMMON_OPT += -DUSE_TLS
+endif
+
+CCOMMON_OPT += -DVERSION=\"$(VERSION)\"
+
 ifndef SYMBOLPREFIX
 SYMBOLPREFIX =
 endif
@@ -1065,8 +1155,6 @@ ifndef FCOMMON_OPT
 FCOMMON_OPT = -O2 -frecursive
 endif
 
-
-
 override CFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR)
 override PFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR) -DPROFILE $(COMMON_PROF)
 
@@ -1074,6 +1162,12 @@ override FFLAGS     += $(COMMON_OPT) $(FCOMMON_OPT)
 override FPFLAGS    += $(FCOMMON_OPT) $(COMMON_PROF)
 #MAKEOVERRIDES =
 
+ifdef NEED_PIC
+ifeq (,$(findstring PIC,$(FFLAGS)))
+override FFLAGS += -fPIC
+endif
+endif
+
 #For LAPACK Fortran codes.
 #Disable -fopenmp for LAPACK Fortran codes on Windows.
 ifdef OS_WINDOWS
@@ -1132,7 +1226,11 @@ endif
 
 LIBDLLNAME   = $(LIBPREFIX).dll
 IMPLIBNAME   = lib$(LIBNAMEBASE).dll.a
+ifneq ($(OSNAME), AIX)
 LIBSONAME    = $(LIBNAME:.$(LIBSUFFIX)=.so)
+else
+LIBSONAME    = $(LIBNAME:.$(LIBSUFFIX)=.a)
+endif
 LIBDYNNAME   = $(LIBNAME:.$(LIBSUFFIX)=.dylib)
 LIBDEFNAME   = $(LIBNAME:.$(LIBSUFFIX)=.def)
 LIBEXPNAME   = $(LIBNAME:.$(LIBSUFFIX)=.exp)
@@ -1209,6 +1307,7 @@ export MSA_FLAGS
 export KERNELDIR
 export FUNCTION_PROFILE
 export TARGET_CORE
+export NO_AVX512
 
 export SGEMM_UNROLL_M
 export SGEMM_UNROLL_N

Makefile.x86_64

@@ -8,6 +8,34 @@ endif
 endif
 endif
 
+ifeq ($(CORE), SKYLAKEX)
+ifndef DYNAMIC_ARCH
+ifndef NO_AVX512
+CCOMMON_OPT += -march=skylake-avx512
+FCOMMON_OPT += -march=skylake-avx512
+ifeq ($(OSNAME), CYGWIN_NT)
+CCOMMON_OPT += -fno-asynchronous-unwind-tables
+endif
+ifeq ($(OSNAME), WINNT)
+ifeq ($(C_COMPILER), GCC)
+CCOMMON_OPT += -fno-asynchronous-unwind-tables
+endif
+endif
+endif
+endif
+endif
+
+ifeq ($(CORE), HASWELL)
+ifndef DYNAMIC_ARCH
+ifndef NO_AVX2
+CCOMMON_OPT += -mavx2
+FCOMMON_OPT += -mavx2
+endif
+endif
+endif
+
+
+
 ifeq ($(OSNAME), Interix)
 ARFLAGS		= -m x64
 endif

README.md

@@ -5,175 +5,221 @@
 Travis CI: [![Build Status](https://travis-ci.org/xianyi/OpenBLAS.svg?branch=develop)](https://travis-ci.org/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)
+
 ## Introduction
+
 OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
 
-Please read the documents on OpenBLAS wiki pages <http://github.com/xianyi/OpenBLAS/wiki>.
+Please read the documentation on the OpenBLAS wiki pages: <http://github.com/xianyi/OpenBLAS/wiki>.
 
 ## Binary Packages
-We provide binary packages for the following platform.
+
+We provide official binary packages for the following platform:
 
   * Windows x86/x86_64
 
 You can download them from [file hosting on sourceforge.net](https://sourceforge.net/projects/openblas/files/).
 
 ## Installation from Source
-Download from project homepage. http://xianyi.github.com/OpenBLAS/
 
-Or, check out codes from git://github.com/xianyi/OpenBLAS.git
+Download from project homepage, http://xianyi.github.com/OpenBLAS/, or check out the code
+using Git from https://github.com/xianyi/OpenBLAS.git.
+
+### Dependencies
+
+Building OpenBLAS requires the following to be installed:
+
+* GNU Make
+* A C compiler, e.g. GCC or Clang
+* A Fortran compiler (optional, for LAPACK)
+* IBM MASS (optional, see below)
+
 ### Normal compile
-  * type "make" to detect the CPU automatically.
+
-  or
+Simply invoking `make` (or `gmake` on BSD) will detect the CPU automatically.
-  * type "make TARGET=xxx" to set target CPU, e.g. "make TARGET=NEHALEM". The full target list is in file TargetList.txt.
+To set a specific target CPU, use `make TARGET=xxx`, e.g. `make TARGET=NEHALEM`.
+The full target list is in the file `TargetList.txt`.
 
 ### Cross compile
-Please set CC and FC with the cross toolchains. Then, set HOSTCC with your host C compiler. At last, set TARGET explicitly.
+
+Set `CC` and `FC` to point to the cross toolchains, and set `HOSTCC` to your host C compiler.
+The target must be specified explicitly when cross compiling.
 
 Examples:
 
-On X86 box, compile this library for loongson3a CPU.
+* On an x86 box, compile this library for a loongson3a CPU:
+  ```sh
+  make BINARY=64 CC=mips64el-unknown-linux-gnu-gcc FC=mips64el-unknown-linux-gnu-gfortran HOSTCC=gcc TARGET=LOONGSON3A
+  ```
 
-    make BINARY=64 CC=mips64el-unknown-linux-gnu-gcc FC=mips64el-unknown-linux-gnu-gfortran HOSTCC=gcc TARGET=LOONGSON3A
+* On an x86 box, compile this library for a loongson3a CPU with loongcc (based on Open64) compiler:
-
+  ```sh
-On X86 box, compile this library for loongson3a CPU with loongcc (based on Open64) compiler.
+  make CC=loongcc FC=loongf95 HOSTCC=gcc TARGET=LOONGSON3A CROSS=1 CROSS_SUFFIX=mips64el-st-linux-gnu-   NO_LAPACKE=1 NO_SHARED=1 BINARY=32
-
+  ```
-    make CC=loongcc FC=loongf95 HOSTCC=gcc TARGET=LOONGSON3A CROSS=1 CROSS_SUFFIX=mips64el-st-linux-gnu-   NO_LAPACKE=1 NO_SHARED=1 BINARY=32
 
 ### Debug version
 
-    make DEBUG=1
+A debug version can be built using `make DEBUG=1`.
 
-### Compile with MASS Support on Power CPU (Optional dependency)
+### Compile with MASS support on Power CPU (optional)
 
-[IBM MASS](http://www-01.ibm.com/software/awdtools/mass/linux/mass-linux.html) library consists of a set of mathematical functions for C, C++, and
+The [IBM MASS](http://www-01.ibm.com/software/awdtools/mass/linux/mass-linux.html) library
-Fortran-language applications that are tuned for optimum performance on POWER architectures. OpenBLAS with MASS requires 64-bit, little-endian OS on POWER.
+consists of a set of mathematical functions for C, C++, and Fortran applications that are
-The library can be installed as below -
+are tuned for optimum performance on POWER architectures.
+OpenBLAS with MASS requires a 64-bit, little-endian OS on POWER.
+The library can be installed as shown:
 
- * On Ubuntu:
+* On Ubuntu:
+  ```sh
+  wget -q http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/ubuntu/public.gpg -O- | sudo apt-key add -
+  echo "deb http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/ubuntu/ trusty main" | sudo tee /etc/apt/sources.list.d/ibm-xl-compiler-eval.list
+  sudo apt-get update
+  sudo apt-get install libxlmass-devel.8.1.5
+  ```
 
-    wget -q http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/ubuntu/public.gpg -O- | sudo apt-key add -</br>
+* On RHEL/CentOS:
-    echo "deb http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/ubuntu/ trusty main" | sudo tee /etc/apt/sources.list.d/ibm-xl-compiler-eval.list</br>
+  ```sh
-    sudo apt-get update</br>
+  wget http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/rhel7/repodata/repomd.xml.key
-    sudo apt-get install libxlmass-devel.8.1.5</br>
+  sudo rpm --import repomd.xml.key
+  wget http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/rhel7/ibm-xl-compiler-eval.repo
+  sudo cp ibm-xl-compiler-eval.repo /etc/yum.repos.d/
+  sudo yum install libxlmass-devel.8.1.5
+  ```
 
- * On RHEL/CentOS:
+After installing the MASS library, compile OpenBLAS with `USE_MASS=1`.
+For example, to compile on Power8 with MASS support: `make USE_MASS=1 TARGET=POWER8`.
 
-    wget http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/rhel7/repodata/repomd.xml.key</br>
+### Install to a specific directory (optional)
-    sudo rpm --import repomd.xml.key</br>
-    wget http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/rhel7/ibm-xl-compiler-eval.repo</br>
-    sudo cp ibm-xl-compiler-eval.repo /etc/yum.repos.d/</br>
-    sudo yum install libxlmass-devel.8.1.5</br>
 
-After installing MASS library, compile openblas with USE_MASS=1.
+Use `PREFIX=` when invoking `make`, for example
 
-Example:
+```sh
+make install PREFIX=your_installation_directory
+```
 
-Compiling on Power8 with MASS support -
+The default installation directory is `/opt/OpenBLAS`.
 
-    make USE_MASS=1 TARGET=POWER8
+## Supported CPUs and Operating Systems
 
-### Install to the directory (optional)
+Please read `GotoBLAS_01Readme.txt`.
 
-Example:
+### Additional supported CPUs
 
-    make install PREFIX=your_installation_directory
+#### x86/x86-64
 
-The default directory is /opt/OpenBLAS
-
-## Support CPU & OS
-Please read GotoBLAS_01Readme.txt
-
-### Additional support CPU:
-
-#### x86/x86-64:
 - **Intel Xeon 56xx (Westmere)**: Used GotoBLAS2 Nehalem codes.
 - **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**: Optimized Level-3 and Level-2 BLAS with AVX512 and FMA  on x86-64.
 - **AMD Bobcat**: Used GotoBLAS2 Barcelona codes.
-- **AMD Bulldozer**: x86-64 ?GEMM FMA4 kernels. (Thank Werner Saar)
+- **AMD Bulldozer**: x86-64 ?GEMM FMA4 kernels. (Thanks to Werner Saar)
 - **AMD PILEDRIVER**: Uses Bulldozer codes with some optimizations.
 - **AMD STEAMROLLER**: Uses Bulldozer codes with some optimizations.
 
-#### MIPS64:
+#### MIPS64
+
 - **ICT Loongson 3A**: Optimized Level-3 BLAS and the part of Level-1,2.
 - **ICT Loongson 3B**: Experimental
 
-#### ARM:
+#### ARM
-- **ARMV6**: Optimized BLAS for vfpv2 and vfpv3-d16 ( e.g. BCM2835, Cortex M0+ )
-- **ARMV7**: Optimized BLAS for vfpv3-d32 ( e.g. Cortex A8, A9 and A15 )
 
-#### ARM64:
+- **ARMv6**: Optimized BLAS for vfpv2 and vfpv3-d16 (e.g. BCM2835, Cortex M0+)
-- **ARMV8**: Experimental
+- **ARMv7**: Optimized BLAS for vfpv3-d32 (e.g. Cortex A8, A9 and A15)
+
+#### ARM64
+
+- **ARMv8**: Experimental
 - **ARM Cortex-A57**: Experimental
 
 #### PPC/PPC64
-- **POWER8**: Optmized Level-3 BLAS and some Level-1, only with USE_OPENMP=1
 
-#### IBM zEnterprise System:
+- **POWER8**: Optmized Level-3 BLAS and some Level-1, only with `USE_OPENMP=1`
+
+#### IBM zEnterprise System
+
 - **Z13**: Optimized Level-3 BLAS and Level-1,2 (double precision)
 
+### Supported OS
 
-### Support OS:
 - **GNU/Linux**
-- **MingWin or Visual Studio(CMake)/Windows**: Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-use-OpenBLAS-in-Microsoft-Visual-Studio>.
+- **MinGW or Visual Studio (CMake)/Windows**: Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-use-OpenBLAS-in-Microsoft-Visual-Studio>.
-- **Darwin/Mac OS X**: Experimental. Although GotoBLAS2 supports Darwin, we are the beginner on Mac OS X.
+- **Darwin/macOS**: Experimental. Although GotoBLAS2 supports Darwin, we are not macOS experts.
-- **FreeBSD**: Supported by community. We didn't test the library on this OS.
+- **FreeBSD**: Supported by the community. We don't actively test the library on this OS.
-- **Android**: Supported by community. Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-build-OpenBLAS-for-Android>.
+- **OpenBSD**: Supported by the community. We don't actively test the library on this OS.
+- **DragonFly BSD**: Supported by the community. We don't actively test the library on this OS.
+- **Android**: Supported by the community. Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-build-OpenBLAS-for-Android>.
 
-## Usages
+## Usage
-Link with libopenblas.a or -lopenblas for shared library.
 
-### Set the number of threads with environment variables.
+Statically link with `libopenblas.a` or dynamically link with `-lopenblas` if OpenBLAS was
+compiled as a shared library.
 
-Examples:
+### Setting the number of threads using environment variables
 
-    export OPENBLAS_NUM_THREADS=4
+Environment variables are used to specify a maximum number of threads.
+For example,
 
- or
+```sh
+export OPENBLAS_NUM_THREADS=4
+export GOTO_NUM_THREADS=4
+export OMP_NUM_THREADS=4
+```
 
-    export GOTO_NUM_THREADS=4
+The priorities are `OPENBLAS_NUM_THREADS` > `GOTO_NUM_THREADS` > `OMP_NUM_THREADS`.
 
- or
+If you compile this library with `USE_OPENMP=1`, you should set the `OMP_NUM_THREADS`
+environment variable; OpenBLAS ignores `OPENBLAS_NUM_THREADS` and `GOTO_NUM_THREADS` when
+compiled with `USE_OPENMP=1`.
 
-    export OMP_NUM_THREADS=4
+### Setting the number of threads at runtime
 
-The priorities are OPENBLAS_NUM_THREADS > GOTO_NUM_THREADS > OMP_NUM_THREADS.
+We provide the following functions to control the number of threads at runtime:
 
-If you compile this lib with USE_OPENMP=1, you should set OMP_NUM_THREADS environment variable. OpenBLAS ignores OPENBLAS_NUM_THREADS and GOTO_NUM_THREADS with USE_OPENMP=1.
+```c
+void goto_set_num_threads(int num_threads);
+void openblas_set_num_threads(int num_threads);
+```
 
-### Set the number of threads on runtime.
+If you compile this library with `USE_OPENMP=1`, you should use the above functions too.
 
-We provided the below functions to control the number of threads on runtime.
+## Reporting bugs
 
-    void goto_set_num_threads(int num_threads);
+Please submit an issue in https://github.com/xianyi/OpenBLAS/issues.
-
-    void openblas_set_num_threads(int num_threads);
-
-If you compile this lib with USE_OPENMP=1, you should use the above functions, too.
-
-## Report Bugs
-Please add a issue in https://github.com/xianyi/OpenBLAS/issues
 
 ## Contact
+
 * OpenBLAS users mailing list: https://groups.google.com/forum/#!forum/openblas-users
 * OpenBLAS developers mailing list: https://groups.google.com/forum/#!forum/openblas-dev
 
-## ChangeLog
+## Change log
-Please see Changelog.txt to obtain the differences between GotoBLAS2 1.13 BSD version.
+
+Please see Changelog.txt to view the differences between OpenBLAS and GotoBLAS2 1.13 BSD version.
 
 ## Troubleshooting
-* Please read [Faq](https://github.com/xianyi/OpenBLAS/wiki/Faq) at first.
+
-* Please use gcc version 4.6 and above to compile Sandy Bridge AVX kernels on Linux/MingW/BSD.
+* Please read the [FAQ](https://github.com/xianyi/OpenBLAS/wiki/Faq) first.
-* Please use Clang version 3.1 and above to compile the library on Sandy Bridge microarchitecture. The Clang 3.0 will generate the wrong AVX binary code.
+* Please use GCC version 4.6 and above to compile Sandy Bridge AVX kernels on Linux/MinGW/BSD.
-* The number of CPUs/Cores should less than or equal to 256. On Linux x86_64(amd64), there is experimental support for up to 1024 CPUs/Cores and 128 numa nodes if you build the library with BIGNUMA=1.
+* Please use Clang version 3.1 and above to compile the library on Sandy Bridge microarchitecture.
-* OpenBLAS does not set processor affinity by default. On Linux, you can enable processor affinity by commenting the line NO_AFFINITY=1 in Makefile.rule. But this may cause [the conflict with R parallel](https://stat.ethz.ch/pipermail/r-sig-hpc/2012-April/001348.html).
+  Clang 3.0 will generate the wrong AVX binary code.
-* On Loongson 3A. make test would be failed because of pthread_create error. The error code is EAGAIN. However, it will be OK when you run the same testcase on shell.
+* Please use GCC version 6 or LLVM version 6 and above to compile Skylake AVX512 kernels.
+* The number of CPUs/cores should less than or equal to 256. On Linux `x86_64` (`amd64`),
+  there is experimental support for up to 1024 CPUs/cores and 128 numa nodes if you build
+  the library with `BIGNUMA=1`.
+* OpenBLAS does not set processor affinity by default.
+  On Linux, you can enable processor affinity by commenting out the line `NO_AFFINITY=1` in
+  Makefile.rule. However, note that this may cause
+  [a conflict with R parallel](https://stat.ethz.ch/pipermail/r-sig-hpc/2012-April/001348.html).
+* On Loongson 3A, `make test` may fail with a `pthread_create` error (`EAGAIN`).
+  However, it will be okay when you run the same test case on the shell.
 
 ## Contributing
-1. [Check for open issues](https://github.com/xianyi/OpenBLAS/issues) or open a fresh issue to start a discussion around a feature idea or a bug.
+
-1. Fork the [OpenBLAS](https://github.com/xianyi/OpenBLAS) repository to start making your changes.
+1. [Check for open issues](https://github.com/xianyi/OpenBLAS/issues) or open a fresh issue
-1. Write a test which shows that the bug was fixed or that the feature works as expected.
+   to start a discussion around a feature idea or a bug.
-1. Send a pull request. Make sure to add yourself to `CONTRIBUTORS.md`.
+2. Fork the [OpenBLAS](https://github.com/xianyi/OpenBLAS) repository to start making your changes.
+3. Write a test which shows that the bug was fixed or that the feature works as expected.
+4. Send a pull request. Make sure to add yourself to `CONTRIBUTORS.md`.
 
 ## Donation
+
 Please read [this wiki page](https://github.com/xianyi/OpenBLAS/wiki/Donation).

TargetList.txt

@@ -20,6 +20,7 @@ DUNNINGTON
 NEHALEM
 SANDYBRIDGE
 HASWELL
+SKYLAKEX
 ATOM
 
 b)AMD CPU:
@@ -56,6 +57,7 @@ CELL
 
 3.MIPS CPU:
 P5600
+1004K
 
 4.MIPS64 CPU:
 SICORTEX
@@ -81,8 +83,11 @@ ARMV5
 
 8.ARM 64-bit CPU:
 ARMV8
+CORTEXA53
 CORTEXA57
-VULCAN
+CORTEXA72
+CORTEXA73
+FALKOR
 THUNDERX
 THUNDERX2T99
 

USAGE.md

@@ -14,6 +14,20 @@ Please build OpenBLAS with larger `NUM_THREADS`. For example, `make
 NUM_THREADS=32` or `make NUM_THREADS=64`.  In `Makefile.system`, we will set
 `MAX_CPU_NUMBER=NUM_THREADS`.
 
+Despite its name, and due to the use of memory buffers in functions like SGEMM,
+the setting of NUM_THREADS can be relevant even for a single-threaded build 
+of OpenBLAS, if such functions get called by multiple threads of a program
+that uses OpenBLAS. In some cases, the affected code may simply crash or throw 
+a segmentation fault without displaying the above warning first.
+
+Note that the number of threads used at runtime can be altered to differ from the
+value NUM_THREADS was set to at build time. At runtime, the actual number of
+threads can be set anywhere from 1 to the build's NUM_THREADS (note however,
+that this does not change the number of memory buffers that will be allocated,
+which is set at build time). The number of threads for a process can be set by
+using the mechanisms described below.
+
+
 #### How can I use OpenBLAS in multi-threaded applications?
 
 If your application is already multi-threaded, it will conflict with OpenBLAS

benchmark/gemm.c

@@ -237,7 +237,7 @@ int main(int argc, char *argv[]){
     timeg = time1/loops;
     fprintf(stderr,
 	    " %10.2f MFlops %10.6f sec\n",
-	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)n / timeg * 1.e-6, time1);
+	    COMPSIZE * COMPSIZE * 2. * (double)k * (double)m * (double)n / timeg * 1.e-6, time1);
     
   }
 

benchmark/gemv.c

@@ -122,7 +122,7 @@ int main(int argc, char *argv[]){
 
   FLOAT *a, *x, *y;
   FLOAT alpha[] = {1.0, 1.0};
-  FLOAT beta [] = {1.0, 1.0};
+  FLOAT beta [] = {1.0, 0.0};
   char trans='N';
   blasint m, i, j;
   blasint inc_x=1,inc_y=1;

c_check

@@ -54,6 +54,8 @@ $compiler = GCC       if ($compiler eq "");
 $os = Linux           if ($data =~ /OS_LINUX/);
 $os = FreeBSD         if ($data =~ /OS_FREEBSD/);
 $os = NetBSD          if ($data =~ /OS_NETBSD/);
+$os = OpenBSD         if ($data =~ /OS_OPENBSD/);
+$os = DragonFly       if ($data =~ /OS_DRAGONFLY/);
 $os = Darwin          if ($data =~ /OS_DARWIN/);
 $os = SunOS           if ($data =~ /OS_SUNOS/);
 $os = AIX             if ($data =~ /OS_AIX/);
@@ -62,6 +64,7 @@ $os = WINNT           if ($data =~ /OS_WINNT/);
 $os = CYGWIN_NT       if ($data =~ /OS_CYGWIN_NT/);
 $os = Interix         if ($data =~ /OS_INTERIX/);
 $os = Android         if ($data =~ /OS_ANDROID/);
+$os = Haiku           if ($data =~ /OS_HAIKU/);
 
 $architecture = x86    if ($data =~ /ARCH_X86/);
 $architecture = x86_64 if ($data =~ /ARCH_X86_64/);
@@ -199,6 +202,21 @@ $architecture = zarch  if ($data =~ /ARCH_ZARCH/);
 $binformat    = bin32;
 $binformat    = bin64  if ($data =~ /BINARY_64/);
 
+$no_avx512= 0;
+if (($architecture eq "x86") || ($architecture eq "x86_64")) {
+    $code = '"vbroadcastss -4 * 4(%rsi), %zmm2"';
+    print $tmpf "#include <immintrin.h>\n\nint main(void){ __asm__ volatile($code); }\n";
+    $args = " -march=skylake-avx512 -o $tmpf.o -x c $tmpf";
+    my @cmd = ("$compiler_name $args >/dev/null 2>/dev/null");
+    system(@cmd) == 0;
+    if ($? != 0) {
+	$no_avx512 = 1;
+    } else {
+	$no_avx512 = 0;
+    }
+    unlink("tmpf.o");
+}
+
 $data = `$compiler_name -S ctest1.c && grep globl ctest1.s | head -n 1 && rm -f ctest1.s`;
 
 $data =~ /globl\s([_\.]*)(.*)/;
@@ -206,7 +224,6 @@ $data =~ /globl\s([_\.]*)(.*)/;
 $need_fu      = $1;
 
 $cross = 0;
-$cross = 1 if ($os ne $hostos);
 
 if ($architecture ne $hostarch) {
     $cross = 1;
@@ -214,6 +231,8 @@ if ($architecture ne $hostarch) {
     $cross = 0 if (($hostarch eq "mips64") && ($architecture eq "mips"));
 }
 
+$cross = 1 if ($os ne $hostos);
+
 $openmp = "" if $ENV{USE_OPENMP} != 1;
 
 $linker_L = "";
@@ -286,6 +305,7 @@ print MAKEFILE "CROSS=1\n" if $cross != 0;
 print MAKEFILE "CEXTRALIB=$linker_L $linker_l $linker_a\n";
 print MAKEFILE "HAVE_MSA=1\n" if $have_msa eq 1;
 print MAKEFILE "MSA_FLAGS=$msa_flags\n" if $have_msa eq 1;
+print MAKEFILE "NO_AVX512=1\n" if $no_avx512 eq 1;
 
 $os           =~ tr/[a-z]/[A-Z]/;
 $architecture =~ tr/[a-z]/[A-Z]/;

cblas.h

@@ -51,6 +51,7 @@ typedef enum CBLAS_TRANSPOSE {CblasNoTrans=111, CblasTrans=112, CblasConjTrans=1
 typedef enum CBLAS_UPLO      {CblasUpper=121, CblasLower=122} CBLAS_UPLO;
 typedef enum CBLAS_DIAG      {CblasNonUnit=131, CblasUnit=132} CBLAS_DIAG;
 typedef enum CBLAS_SIDE      {CblasLeft=141, CblasRight=142} CBLAS_SIDE;
+typedef CBLAS_ORDER CBLAS_LAYOUT;
 	
 float  cblas_sdsdot(OPENBLAS_CONST blasint n, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float *y, OPENBLAS_CONST blasint incy);
 double cblas_dsdot (OPENBLAS_CONST blasint n, OPENBLAS_CONST float *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float *y, OPENBLAS_CONST blasint incy);
@@ -82,6 +83,11 @@ CBLAS_INDEX cblas_idamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPE
 CBLAS_INDEX cblas_icamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx);
 CBLAS_INDEX cblas_izamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx);
 
+CBLAS_INDEX cblas_isamin(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx);
+CBLAS_INDEX cblas_idamin(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx);
+CBLAS_INDEX cblas_icamin(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx);
+CBLAS_INDEX cblas_izamin(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx);
+
 void cblas_saxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *x, OPENBLAS_CONST blasint incx, float *y, OPENBLAS_CONST blasint incy);
 void cblas_daxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, double *y, OPENBLAS_CONST blasint incy);
 void cblas_caxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incy);

cmake/OpenBLASConfig.cmake.in

@@ -0,0 +1,79 @@
+# OpenBLASConfig.cmake
+# --------------------
+#
+# OpenBLAS cmake module.
+# This module sets the following variables in your project::
+#
+#   OpenBLAS_FOUND - true if OpenBLAS and all required components found on the system
+#   OpenBLAS_VERSION - OpenBLAS version in format Major.Minor.Release
+#   OpenBLAS_INCLUDE_DIRS - Directory where OpenBLAS header is located.
+#   OpenBLAS_INCLUDE_DIR - same as DIRS
+#   OpenBLAS_LIBRARIES - OpenBLAS library to link against.
+#   OpenBLAS_LIBRARY - same as LIBRARIES
+#
+#
+# Available components::
+#
+##   shared - search for only shared library
+##   static - search for only static library
+#   serial - search for unthreaded library
+#   pthread - search for native pthread threaded library
+#   openmp - search for OpenMP threaded library
+#
+#
+# Exported targets::
+#
+# If OpenBLAS is found, this module defines the following :prop_tgt:`IMPORTED`
+## target. Target is shared _or_ static, so, for both, use separate, not
+## overlapping, installations. ::
+#
+#   OpenBLAS::OpenBLAS - the main OpenBLAS library #with header & defs attached.
+#
+#
+# Suggested usage::
+#
+#   find_package(OpenBLAS)
+#   find_package(OpenBLAS 0.2.20 EXACT CONFIG REQUIRED COMPONENTS pthread)
+#
+#
+# The following variables can be set to guide the search for this package::
+#
+#   OpenBLAS_DIR - CMake variable, set to directory containing this Config file
+#   CMAKE_PREFIX_PATH - CMake variable, set to root directory of this package
+#   PATH - environment variable, set to bin directory of this package
+#   CMAKE_DISABLE_FIND_PACKAGE_OpenBLAS - CMake variable, disables
+#     find_package(OpenBLAS) when not REQUIRED, perhaps to force internal build
+
+@PACKAGE_INIT@
+
+set(PN OpenBLAS)
+
+# need to check that the @USE_*@ evaluate to something cmake can perform boolean logic upon
+if(@USE_OPENMP@)
+    set(${PN}_openmp_FOUND 1)
+elseif(@USE_THREAD@)
+    set(${PN}_pthread_FOUND 1)
+else()
+    set(${PN}_serial_FOUND 1)
+endif()
+
+check_required_components(${PN})
+
+#-----------------------------------------------------------------------------
+# Don't include targets if this file is being picked up by another
+# project which has already built this as a subproject
+#-----------------------------------------------------------------------------
+if(NOT TARGET ${PN}::OpenBLAS)
+    include("${CMAKE_CURRENT_LIST_DIR}/${PN}Targets.cmake")
+
+    get_property(_loc TARGET ${PN}::OpenBLAS PROPERTY LOCATION)
+    set(${PN}_LIBRARY ${_loc})
+    get_property(_ill TARGET ${PN}::OpenBLAS PROPERTY INTERFACE_LINK_LIBRARIES)
+    set(${PN}_LIBRARIES ${_ill})
+
+    get_property(_id TARGET ${PN}::OpenBLAS PROPERTY INCLUDE_DIRECTORIES)
+    set(${PN}_INCLUDE_DIR ${_id})
+    get_property(_iid TARGET ${PN}::OpenBLAS PROPERTY INTERFACE_INCLUDE_DIRECTORIES)
+    set(${PN}_INCLUDE_DIRS ${_iid})
+endif()
+

cmake/arch.cmake

@@ -44,18 +44,36 @@ endif ()
 
 
 if (DYNAMIC_ARCH)
+  if (ARM64)
+    set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99)
+  endif ()
+  
   if (X86)
     set(DYNAMIC_CORE KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO)
   endif ()
 
   if (X86_64)
-    set(DYNAMIC_CORE PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO)
+    set(DYNAMIC_CORE PRESCOTT CORE2)
+    if (DYNAMIC_OLDER)
+	set (DYNAMIC_CORE ${DYNAMIC_CORE} PENRYN DUNNINGTON)
+    endif ()
+    set (DYNAMIC_CORE ${DYNAMIC_CORE} NEHALEM)
+    if (DYNAMIC_OLDER)
+	set (DYNAMIC_CORE ${DYNAMIC_CORE} OPTERON OPTERON_SSE3)
+    endif ()
+    set (DYNAMIC_CORE ${DYNAMIC_CORE} BARCELONA) 
+    if (DYNAMIC_OLDER)
+	set (DYNAMIC_CORE ${DYNAMIC_CORE} BOBCAT ATOM NANO)
+    endif ()
     if (NOT NO_AVX)
       set(DYNAMIC_CORE ${DYNAMIC_CORE} SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR)
     endif ()
     if (NOT NO_AVX2)
       set(DYNAMIC_CORE ${DYNAMIC_CORE} HASWELL ZEN)
     endif ()
+    if (NOT NO_AVX512)
+      set(DYNAMIC_CORE ${DYNAMIC_CORE} SKYLAKEX)
+    endif ()
   endif ()
 
   if (NOT DYNAMIC_CORE)

cmake/fc.cmake

@@ -3,6 +3,11 @@
 ## Description: Ported from portion of OpenBLAS/Makefile.system
 ##              Sets Fortran related variables.
 
+if (INTERFACE64)
+  set(SUFFIX64 64)
+  set(SUFFIX64_UNDERSCORE _64)
+endif()
+
 if (${F_COMPILER} STREQUAL "FLANG")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_FLANG")
   if (BINARY64 AND INTERFACE64)
@@ -39,7 +44,7 @@ endif ()
 
 if (${F_COMPILER} STREQUAL "GFORTRAN")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_GFORT")
-  set(FCOMMON_OPT "${FCOMMON_OPT} -Wall")
+  set(FCOMMON_OPT "${FCOMMON_OPT} -Wall -frecursive")
   #Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
   if (NOT NO_LAPACK)
     set(EXTRALIB "{EXTRALIB} -lgfortran")

cmake/openblas.pc.in

@@ -1,9 +1,11 @@
 libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+libsuffix=@SUFFIX64_UNDERSCORE@
 includedir=@CMAKE_INSTALL_FULL_INCLUDEDIR@
 
+openblas_config=USE_64BITINT=@USE_64BITINT@ NO_CBLAS=@NO_CBLAS@ NO_LAPACK=@NO_LAPACK@ NO_LAPACKE=@NO_LAPACKE@ DYNAMIC_ARCH=@DYNAMIC_ARCH@ DYNAMIC_OLDER=@DYNAMIC_OLDER@ NO_AFFINITY=@NO_AFFINITY@ USE_OPENMP=@USE_OPENMP@ @CORE@ MAX_THREADS=@NUM_THREADS@ 
 Name: OpenBLAS
 Description: OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version
 Version: @OPENBLAS_VERSION@
 URL: https://github.com/xianyi/OpenBLAS
-Libs: -L${libdir} -lopenblas
+Libs: -L${libdir} -lopenblas${libsuffix}
 Cflags: -I${includedir}

cmake/prebuild.cmake

@@ -85,15 +85,20 @@ if (NOT NOFORTRAN)
 endif ()
 
 # Cannot run getarch on target if we are cross-compiling
-if (DEFINED CORE AND CMAKE_CROSSCOMPILING)
+if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSSTORE"))
   # Write to config as getarch would
+  if (DEFINED TARGET_CORE)
+  set(TCORE ${TARGET_CORE})
+  else()
+  set(TCORE ${CORE})
+  endif()
 
   # TODO: Set up defines that getarch sets up based on every other target
   # Perhaps this should be inside a different file as it grows larger
   file(APPEND ${TARGET_CONF_TEMP}
-    "#define ${CORE}\n"
+    "#define ${TCORE}\n"
-    "#define CHAR_CORENAME \"${CORE}\"\n")
+    "#define CHAR_CORENAME \"${TCORE}\"\n")
-  if ("${CORE}" STREQUAL "ARMV7")
+  if ("${TCORE}" STREQUAL "ARMV7")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_DATA_SIZE\t65536\n"
       "#define L1_DATA_LINESIZE\t32\n"
@@ -108,7 +113,7 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING)
     set(SGEMM_UNROLL_N 4)
     set(DGEMM_UNROLL_M 4)
     set(DGEMM_UNROLL_N 4)
-  elseif ("${CORE}" STREQUAL "ARMV8")
+  elseif ("${TCORE}" STREQUAL "ARMV8")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_DATA_SIZE\t32768\n"
       "#define L1_DATA_LINESIZE\t64\n"
@@ -116,18 +121,26 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING)
       "#define L2_LINESIZE\t64\n"
       "#define DTB_DEFAULT_ENTRIES\t64\n"
       "#define DTB_SIZE\t4096\n"
-      "#define L2_ASSOCIATIVE\t32\n")
+      "#define L2_ASSOCIATIVE\t32\n"
-    set(SGEMM_UNROLL_M 4)
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 16)
     set(SGEMM_UNROLL_N 4)
-  elseif ("${CORE}" STREQUAL "CORTEXA57")
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "CORTEXA57" OR "${TCORE}" STREQUAL "CORTEXA53")
     file(APPEND ${TARGET_CONF_TEMP}
-      "#define L1_CODE_SIZE\t49152\n"
+      "#define L1_CODE_SIZE\t32768\n"
       "#define L1_CODE_LINESIZE\t64\n"
       "#define L1_CODE_ASSOCIATIVE\t3\n"
       "#define L1_DATA_SIZE\t32768\n"
       "#define L1_DATA_LINESIZE\t64\n"
       "#define L1_DATA_ASSOCIATIVE\t2\n"
-      "#define L2_SIZE\t2097152\n"
+      "#define L2_SIZE\t262144\n"
       "#define L2_LINESIZE\t64\n"
       "#define L2_ASSOCIATIVE\t16\n"
       "#define DTB_DEFAULT_ENTRIES\t64\n"
@@ -135,15 +148,124 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING)
       "#define HAVE_VFPV4\n"
       "#define HAVE_VFPV3\n"
       "#define HAVE_VFP\n"
-      "#define HAVE_NEON\n")
+      "#define HAVE_NEON\n"
+      "#define ARMV8\n")
     set(SGEMM_UNROLL_M 16)
     set(SGEMM_UNROLL_N 4)
     set(DGEMM_UNROLL_M 8)
     set(DGEMM_UNROLL_N 4)
     set(CGEMM_UNROLL_M 8)
     set(CGEMM_UNROLL_N 4)
-    set(ZGEMM_UNROLL_M 8)
+    set(ZGEMM_UNROLL_M 4)
     set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "CORTEXA72" OR "${TCORE}" STREQUAL "CORTEXA73")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L1_CODE_SIZE\t49152\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t3\n"
+      "#define L1_DATA_SIZE\t32768\n"
+      "#define L1_DATA_LINESIZE\t64\n"
+      "#define L1_DATA_ASSOCIATIVE\t2\n"
+      "#define L2_SIZE\t524288\n"
+      "#define L2_LINESIZE\t64\n"
+      "#define L2_ASSOCIATIVE\t16\n"
+      "#define DTB_DEFAULT_ENTRIES\t64\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define HAVE_VFPV4\n"
+      "#define HAVE_VFPV3\n"
+      "#define HAVE_VFP\n"
+      "#define HAVE_NEON\n"
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 16)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "FALKOR")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L1_CODE_SIZE\t65536\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t3\n"
+      "#define L1_DATA_SIZE\t32768\n"
+      "#define L1_DATA_LINESIZE\t128\n"
+      "#define L1_DATA_ASSOCIATIVE\t2\n"
+      "#define L2_SIZE\t524288\n"
+      "#define L2_LINESIZE\t64\n"
+      "#define L2_ASSOCIATIVE\t16\n"
+      "#define DTB_DEFAULT_ENTRIES\t64\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define HAVE_VFPV4\n"
+      "#define HAVE_VFPV3\n"
+      "#define HAVE_VFP\n"
+      "#define HAVE_NEON\n"
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 16)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "THUNDERX")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L1_CODE_SIZE\t32768\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t3\n"
+      "#define L1_DATA_SIZE\t32768\n"
+      "#define L1_DATA_LINESIZE\t128\n"
+      "#define L1_DATA_ASSOCIATIVE\t2\n"
+      "#define L2_SIZE\t167772164\n"
+      "#define L2_LINESIZE\t128\n"
+      "#define L2_ASSOCIATIVE\t16\n"
+      "#define DTB_DEFAULT_ENTRIES\t64\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define HAVE_VFPV4\n"
+      "#define HAVE_VFPV3\n"
+      "#define HAVE_VFP\n"
+      "#define HAVE_NEON\n"
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 4)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 2)
+    set(DGEMM_UNROLL_N 2)
+    set(CGEMM_UNROLL_M 2)
+    set(CGEMM_UNROLL_N 2)
+    set(ZGEMM_UNROLL_M 2)
+    set(ZGEMM_UNROLL_N 2)
+    set(SYMV_P 16)
+  elseif ("${TCORE}" STREQUAL "THUNDERX2T99")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L1_CODE_SIZE\t32768\n"
+      "#define L1_CODE_LINESIZE\t64\n"
+      "#define L1_CODE_ASSOCIATIVE\t8\n"
+      "#define L1_DATA_SIZE\t32768\n"
+      "#define L1_DATA_LINESIZE\t64\n"
+      "#define L1_DATA_ASSOCIATIVE\t8\n"
+      "#define L2_SIZE\t262144\n"
+      "#define L2_LINESIZE\t64\n"
+      "#define L2_ASSOCIATIVE\t8\n"
+      "#define L3_SIZE\t33554432\n"
+      "#define L3_LINESIZE\t64\n"
+      "#define L3_ASSOCIATIVE\t32\n"
+      "#define DTB_DEFAULT_ENTRIES\t64\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define ARMV8\n")
+    set(SGEMM_UNROLL_M 16)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 8)
+    set(DGEMM_UNROLL_N 4)
+    set(CGEMM_UNROLL_M 8)
+    set(CGEMM_UNROLL_N 4)
+    set(ZGEMM_UNROLL_M 4)
+    set(ZGEMM_UNROLL_N 4)
+    set(SYMV_P 16)
   endif()
 
   # Or should this actually be NUM_CORES?
@@ -163,6 +285,7 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING)
   file(APPEND ${TARGET_CONF_TEMP}
     "#define GEMM_MULTITHREAD_THRESHOLD\t${GEMM_MULTITHREAD_THRESHOLD}\n")
   # Move to where gen_config_h would place it
+  file(MAKE_DIRECTORY ${TARGET_CONF_DIR})
   file(RENAME ${TARGET_CONF_TEMP} "${TARGET_CONF_DIR}/${TARGET_CONF}")  
 
 else(NOT CMAKE_CROSSCOMPILING)

cmake/system.cmake

@@ -33,7 +33,7 @@ endif ()
 if (DEFINED BINARY AND DEFINED TARGET AND BINARY EQUAL 32)
   message(STATUS "Compiling a ${BINARY}-bit binary.")
   set(NO_AVX 1)
-  if (${TARGET} STREQUAL "HASWELL" OR ${TARGET} STREQUAL "SANDYBRIDGE")
+  if (${TARGET} STREQUAL "HASWELL" OR ${TARGET} STREQUAL "SANDYBRIDGE" OR ${TARGET} STREQUAL "SKYLAKEX")
     set(TARGET "NEHALEM")
   endif ()
   if (${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "PILEDRIVER" OR ${TARGET} STREQUAL "ZEN")
@@ -41,6 +41,22 @@ if (DEFINED BINARY AND DEFINED TARGET AND BINARY EQUAL 32)
   endif ()
 endif ()
 
+if (DEFINED TARGET)
+  if (${TARGET} STREQUAL "SKYLAKEX" AND NOT NO_AVX512)
+    set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=skylake-avx512")
+  endif()
+  if (${TARGET} STREQUAL "HASWELL" AND NOT NO_AVX2)
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
+      execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+      if (${GCC_VERSION} VERSION_GREATER 4.7 OR ${GCC_VERSION} VERSION_EQUAL 4.7)
+        set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
+      endif()
+    elseif (${CMAKE_C_COMPILER_ID} STREQUAL "CLANG")
+      set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
+    endif()
+  endif()
+endif()
+
 if (DEFINED TARGET)
   message(STATUS "Targeting the ${TARGET} architecture.")
   set(GETARCH_FLAGS "-DFORCE_${TARGET}")
@@ -96,8 +112,12 @@ if (NOT CMAKE_CROSSCOMPILING)
 
 endif()
 
+if (NOT DEFINED NUM_PARALLEL)
+  set(NUM_PARALLEL 1)
+endif()
+
 if (NOT DEFINED NUM_THREADS)
-  if (NOT NUM_CORES EQUAL 0)
+  if (DEFINED NUM_CORES AND NOT NUM_CORES EQUAL 0)
     # HT?
     set(NUM_THREADS ${NUM_CORES})
   else ()
@@ -159,6 +179,9 @@ endif ()
 
 if (DYNAMIC_ARCH)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_ARCH")
+  if (DYNAMIC_OLDER)
+    set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_OLDER")
+  endif ()
 endif ()
 
 if (NO_LAPACK)
@@ -207,6 +230,10 @@ if (CONSISTENT_FPCSR)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DCONSISTENT_FPCSR")
 endif ()
 
+if (USE_TLS)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_TLS")
+endif ()
+
 # Only for development
 # set(CCOMMON_OPT "${CCOMMON_OPT} -DPARAMTEST")
 # set(CCOMMON_OPT "${CCOMMON_OPT} -DPREFETCHTEST")
@@ -224,6 +251,8 @@ endif ()
 
 set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_CPU_NUMBER=${NUM_THREADS}")
 
+set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_PARALLEL_NUMBER=${NUM_PARALLEL}")
+
 if (USE_SIMPLE_THREADED_LEVEL3)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_SIMPLE_THREADED_LEVEL3")
 endif ()
@@ -291,6 +320,8 @@ if (MIXED_MEMORY_ALLOCATION)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DMIXED_MEMORY_ALLOCATION")
 endif ()
 
+set(CCOMMON_OPT "${CCOMMON_OPT} -DVERSION=\"\\\"${OpenBLAS_VERSION}\\\"\"")
+
 set(REVISION "-r${OpenBLAS_VERSION}")
 set(MAJOR_VERSION ${OpenBLAS_MAJOR_VERSION})
 

cmake/system_check.cmake

@@ -10,6 +10,16 @@ if (${HOST_OS} STREQUAL "WINDOWS")
   set(HOST_OS WINNT)
 endif ()
 
+if (${HOST_OS} STREQUAL "LINUX")
+# check if we're building natively on Android (TERMUX)
+    EXECUTE_PROCESS( COMMAND uname -o COMMAND tr -d '\n' OUTPUT_VARIABLE OPERATING_SYSTEM)
+      if(${OPERATING_SYSTEM} MATCHES "Android")
+        set(HOST_OS ANDROID)
+      endif(${OPERATING_SYSTEM} MATCHES "Android")
+endif()
+
+
+
 if(CMAKE_COMPILER_IS_GNUCC AND WIN32)
     execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpmachine
               OUTPUT_VARIABLE OPENBLAS_GCC_TARGET_MACHINE
@@ -66,3 +76,12 @@ else()
   set(BINARY32 1)
 endif()
 
+if (X86_64 OR X86)
+  file(WRITE ${PROJECT_BINARY_DIR}/avx512.tmp "#include <immintrin.h>\n\nint main(void){ __asm__ volatile(\"vbroadcastss -4 * 4(%rsi), %zmm2\"); }")
+execute_process(COMMAND ${CMAKE_C_COMPILER} -march=skylake-avx512 -v -o ${PROJECT_BINARY_DIR}/avx512.o -x c ${PROJECT_BINARY_DIR}/avx512.tmp OUTPUT_QUIET ERROR_QUIET RESULT_VARIABLE NO_AVX512)
+if (NO_AVX512 EQUAL 1)
+set (CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX512")
+endif()
+  file(REMOVE "avx512.tmp" "avx512.o")
+endif()
+

common.h

@@ -93,7 +93,7 @@ extern "C" {
 #include <sched.h>
 #endif
 
-#if defined(OS_DARWIN) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_ANDROID)
+#if defined(OS_DARWIN) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(OS_ANDROID)
 #include <sched.h>
 #endif
 
@@ -105,6 +105,10 @@ extern "C" {
 #endif
 #endif
 
+#ifdef OS_HAIKU
+#define NO_SYSV_IPC
+#endif
+
 #ifdef OS_WINDOWS
 #ifdef  ATOM
 #define GOTO_ATOM ATOM
@@ -179,7 +183,7 @@ extern "C" {
 
 #define ALLOCA_ALIGN 63UL
 
-#define NUM_BUFFERS (MAX_CPU_NUMBER * 2)
+#define NUM_BUFFERS MAX(50,(MAX_CPU_NUMBER * 2 * MAX_PARALLEL_NUMBER))
 
 #ifdef NEEDBUNDERSCORE
 #define BLASFUNC(FUNC) FUNC##_
@@ -253,8 +257,14 @@ typedef unsigned long BLASULONG;
 
 #ifdef USE64BITINT
 typedef BLASLONG blasint;
+#if defined(OS_WINDOWS) && defined(__64BIT__)
+#define blasabs(x) llabs(x)
+#else
+#define blasabs(x) labs(x)
+#endif
 #else
 typedef int blasint;
+#define blasabs(x) abs(x)
 #endif
 #else
 #ifdef USE64BITINT
@@ -642,6 +652,7 @@ void gotoblas_profile_init(void);
 void gotoblas_profile_quit(void);
 
 #ifdef USE_OPENMP
+
 #ifndef C_MSVC
 int omp_in_parallel(void);
 int omp_get_num_procs(void);
@@ -649,6 +660,21 @@ int omp_get_num_procs(void);
 __declspec(dllimport) int __cdecl omp_in_parallel(void);
 __declspec(dllimport) int __cdecl omp_get_num_procs(void);
 #endif
+
+#if (__STDC_VERSION__ >= 201112L)
+#if defined(C_GCC) && ( __GNUC__ < 7) 
+// workaround for GCC bug 65467
+#ifndef _Atomic
+#define _Atomic volatile
+#endif
+#endif
+#include <stdatomic.h>
+#else
+#ifndef _Atomic
+#define _Atomic volatile
+#endif
+#endif
+
 #else
 #ifdef __ELF__
 int omp_in_parallel  (void) __attribute__ ((weak));

common_level3.h

@@ -47,6 +47,14 @@ __global__ void cuda_dgemm_kernel(int, int, int, double *, double *, double *);
 extern "C" {
 #endif
 
+extern void sgemm_kernel_direct(BLASLONG M, BLASLONG N, BLASLONG K,
+	float * A, BLASLONG strideA,
+	float * B, BLASLONG strideB,
+	float * R, BLASLONG strideR);
+
+extern int sgemm_kernel_direct_performant(BLASLONG M, BLASLONG N, BLASLONG K);
+
+
 int sgemm_beta(BLASLONG, BLASLONG, BLASLONG, float,
 	       float  *, BLASLONG, float   *, BLASLONG, float  *, BLASLONG);
 int dgemm_beta(BLASLONG, BLASLONG, BLASLONG, double,

common_mips64.h

@@ -94,7 +94,7 @@ static inline unsigned int rpcc(void){
 #define RPCC_DEFINED
 
 #ifndef NO_AFFINITY
-#define WHEREAMI
+//#define WHEREAMI
 static inline int WhereAmI(void){
   int ret=0;
   __asm__ __volatile__(".set push \n"

common_stackalloc.h

@@ -47,14 +47,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  * - large enough to support all architectures and kernel
  * Chosing a too small SIZE will lead to a stack smashing.
  */
-#define STACK_ALLOC(SIZE, TYPE, BUFFER)                                    \
+#define STACK_ALLOC(SIZE, TYPE, BUFFER)                                        \
-  /* make it volatile because some function (ex: dgemv_n.S) */             \
+  /* make it volatile because some function (ex: dgemv_n.S) */                 \
-  /* do not restore all register */                                        \
+  /* do not restore all register */                                            \
-  volatile int stack_alloc_size = SIZE;                                    \
+  volatile int stack_alloc_size = SIZE;                                        \
-  if(stack_alloc_size > MAX_STACK_ALLOC / sizeof(TYPE))                    \
+  if (stack_alloc_size > MAX_STACK_ALLOC / sizeof(TYPE)) stack_alloc_size = 0; \
-    stack_alloc_size = 0;                                                  \
+  STACK_ALLOC_PROTECT_SET                                                      \
-  STACK_ALLOC_PROTECT_SET                                                  \
+  /* Avoid declaring an array of length 0 */                                   \
-  TYPE stack_buffer[stack_alloc_size] __attribute__((aligned(0x20)));    \
+  TYPE stack_buffer[stack_alloc_size ? stack_alloc_size : 1]                   \
+      __attribute__((aligned(0x20)));                                          \
   BUFFER = stack_alloc_size ? stack_buffer : (TYPE *)blas_memory_alloc(1);
 #else
   //Original OpenBLAS/GotoBLAS codes.

common_x86.h

@@ -178,6 +178,12 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
   result = x/y;
   return result;
 #else
+#if (MAX_CPU_NUMBER > 64)
+  if ( y > 64) {
+	  result = x/y;
+	  return result;
+  }
+#endif
 	
   y = blas_quick_divide_table[y];
 
@@ -327,7 +333,7 @@ REALNAME:
 #endif
 #endif
 
-#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__)
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_OPENBSD) || defined(__ELF__)
 #define PROLOGUE \
 	.text; \
 	.align 16; \

common_x86_64.h

@@ -60,8 +60,13 @@
 #endif
 */
 
-#define MB
+#ifdef __GNUC__
-#define WMB
+#define MB do { __asm__ __volatile__("": : :"memory"); } while (0)
+#define WMB do { __asm__ __volatile__("": : :"memory"); } while (0)
+#else
+#define MB do {} while (0)
+#define WMB do {} while (0)
+#endif
 
 static void __inline blas_lock(volatile BLASULONG *address){
 
@@ -129,7 +134,7 @@ static __inline void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
 			     "=b" (*ebx),
 			     "=c" (*ecx),
 			     "=d" (*edx)
-			     : "0" (op));
+			     : "0" (op), "c"(0));
 #endif
 }
 
@@ -196,6 +201,13 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 
   if (y <= 1) return x;
 
+#if (MAX_CPU_NUMBER > 64)  
+  if (y > 64) { 
+	  result = x / y;
+	  return result;
+  }
+#endif
+	
   y = blas_quick_divide_table[y];
 
   __asm__ __volatile__  ("mull %0" :"=d" (result) :"a"(x), "0" (y));
@@ -403,7 +415,7 @@ REALNAME:
 #define EPILOGUE .end
 #endif
 
-#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__) || defined(C_PGI)
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(__ELF__) || defined(C_PGI)
 #define PROLOGUE \
 	.text; \
 	.align 512; \

cpuid.h

@@ -53,6 +53,7 @@
 #define VENDOR_SIS	  8
 #define VENDOR_TRANSMETA  9
 #define VENDOR_NSC	 10
+#define VENDOR_HYGON	 11
 #define VENDOR_UNKNOWN   99
 
 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
@@ -115,6 +116,8 @@
 #define CORE_STEAMROLLER 25
 #define CORE_EXCAVATOR   26
 #define CORE_ZEN         27
+#define CORE_SKYLAKEX    28
+#define CORE_DHYANA	 29
 
 #define HAVE_SSE      (1 <<  0)
 #define HAVE_SSE2     (1 <<  1)
@@ -137,6 +140,8 @@
 #define HAVE_AVX      (1 <<  18)
 #define HAVE_FMA4     (1 <<  19)
 #define HAVE_FMA3     (1 <<  20)
+#define HAVE_AVX512VL (1 <<  21)
+#define HAVE_AVX2     (1 <<  22)
 
 #define CACHE_INFO_L1_I     1
 #define CACHE_INFO_L1_D     2
@@ -211,5 +216,9 @@ typedef struct {
 #define CPUTYPE_STEAMROLLER 		49
 #define CPUTYPE_EXCAVATOR 		50
 #define CPUTYPE_ZEN 			51
+#define CPUTYPE_SKYLAKEX		52
+#define CPUTYPE_DHYANA			53
+
+#define CPUTYPE_HYGON_UNKNOWN		54
 
 #endif

cpuid_arm.c

@@ -34,7 +34,7 @@
 #define CPU_CORTEXA15       	4
 
 static char *cpuname[] = {
-  "UNKOWN",
+  "UNKNOWN",
   "ARMV6",
   "ARMV7",
   "CORTEXA9",

cpuid_arm64.c

@@ -29,25 +29,37 @@
 
 #define CPU_UNKNOWN     	0
 #define CPU_ARMV8       	1
-#define CPU_CORTEXA57       	2
+// Arm
-#define CPU_VULCAN       	3
+#define CPU_CORTEXA53     2
-#define CPU_THUNDERX    	4
+#define CPU_CORTEXA57     3
-#define CPU_THUNDERX2T99   	5
+#define CPU_CORTEXA72     4
+#define CPU_CORTEXA73     5
+// Qualcomm
+#define CPU_FALKOR        6
+// Cavium
+#define CPU_THUNDERX      7
+#define CPU_THUNDERX2T99  8
 
 static char *cpuname[] = {
   "UNKNOWN",
   "ARMV8" ,
+  "CORTEXA53",
   "CORTEXA57",
-  "VULCAN",
+  "CORTEXA72",
+  "CORTEXA73",
+  "FALKOR",
   "THUNDERX",
   "THUNDERX2T99"
 };
 
 static char *cpuname_lower[] = {
   "unknown",
-  "armv8" ,
+  "armv8",
+  "cortexa53",
   "cortexa57",
-  "vulcan",
+  "cortexa72",
+  "cortexa73",
+  "falkor",
   "thunderx",
   "thunderx2t99"
 };
@@ -114,13 +126,24 @@ int detect(void)
 
 	fclose(infile);
 	if(cpu_part != NULL && cpu_implementer != NULL) {
-		if (strstr(cpu_part, "0xd07") && strstr(cpu_implementer, "0x41"))
+    // Arm
-			return CPU_CORTEXA57;
+    if (strstr(cpu_implementer, "0x41")) {
-		else if (strstr(cpu_part, "0x516") && strstr(cpu_implementer, "0x42"))
+      if (strstr(cpu_part, "0xd03"))
-			return CPU_VULCAN;
+        return CPU_CORTEXA53;
-		else if (strstr(cpu_part, "0x0a1") && strstr(cpu_implementer, "0x43"))
+      else if (strstr(cpu_part, "0xd07"))
+        return CPU_CORTEXA57;
+      else if (strstr(cpu_part, "0xd08"))
+        return CPU_CORTEXA72;
+      else if (strstr(cpu_part, "0xd09"))
+        return CPU_CORTEXA73;
+    }
+    // Qualcomm
+    else if (strstr(cpu_implementer, "0x51") && strstr(cpu_part, "0xc00"))
+      return CPU_FALKOR;
+    // Cavium
+    else if (strstr(cpu_implementer, "0x43") && strstr(cpu_part, "0x0a1"))
 			return CPU_THUNDERX;
-		else if (strstr(cpu_part, "0xFFF") && strstr(cpu_implementer, "0x43")) /* TODO */
+    else if (strstr(cpu_implementer, "0x43") && strstr(cpu_part, "0x0af"))
 			return CPU_THUNDERX2T99;
 	}
 
@@ -179,64 +202,63 @@ void get_subdirname(void)
 void get_cpuconfig(void)
 {
 
+  // All arches should define ARMv8
+  printf("#define ARMV8\n");
+  printf("#define HAVE_NEON\n"); // This shouldn't be necessary
+  printf("#define HAVE_VFPV4\n"); // This shouldn't be necessary
+
 	int d = detect();
 	switch (d)
 	{
 
+    case CPU_CORTEXA53:
+      printf("#define %s\n", cpuname[d]);
+      // Fall-through
 		case CPU_ARMV8:
-    			printf("#define ARMV8\n");
+      // Minimum parameters for ARMv8 (based on A53)
-    			printf("#define L1_DATA_SIZE 32768\n");
+    	printf("#define L1_DATA_SIZE 32768\n");
-    			printf("#define L1_DATA_LINESIZE 64\n");
+    	printf("#define L1_DATA_LINESIZE 64\n");
-    			printf("#define L2_SIZE 262144\n");
+    	printf("#define L2_SIZE 262144\n");
-    			printf("#define L2_LINESIZE 64\n");
+    	printf("#define L2_LINESIZE 64\n");
-    			printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    	printf("#define DTB_DEFAULT_ENTRIES 64\n");
-    			printf("#define DTB_SIZE 4096\n");
+    	printf("#define DTB_SIZE 4096\n");
-    			printf("#define L2_ASSOCIATIVE 4\n");
+    	printf("#define L2_ASSOCIATIVE 4\n");
-			break;
-
-		case CPU_VULCAN:
-			printf("#define VULCAN                        \n");
-			printf("#define HAVE_VFP                      \n");
-			printf("#define HAVE_VFPV3                    \n");
-			printf("#define HAVE_NEON                     \n");
-			printf("#define HAVE_VFPV4                    \n");
-			printf("#define L1_CODE_SIZE         32768    \n");
-			printf("#define L1_CODE_LINESIZE     64       \n");
-			printf("#define L1_CODE_ASSOCIATIVE  8        \n");
-			printf("#define L1_DATA_SIZE         32768    \n");
-			printf("#define L1_DATA_LINESIZE     64       \n");
-			printf("#define L1_DATA_ASSOCIATIVE  8        \n");
-			printf("#define L2_SIZE              262144   \n");
-			printf("#define L2_LINESIZE          64       \n");
-			printf("#define L2_ASSOCIATIVE       8        \n");
-			printf("#define L3_SIZE              33554432 \n");
-			printf("#define L3_LINESIZE          64       \n");
-			printf("#define L3_ASSOCIATIVE       32       \n");
-			printf("#define DTB_DEFAULT_ENTRIES  64       \n");
-			printf("#define DTB_SIZE             4096     \n");
 			break;
 
 		case CPU_CORTEXA57:
-			printf("#define CORTEXA57\n");
+		case CPU_CORTEXA72:
-			printf("#define HAVE_VFP\n");
+		case CPU_CORTEXA73:
-			printf("#define HAVE_VFPV3\n");
+      // Common minimum settings for these Arm cores
-			printf("#define HAVE_NEON\n");
+      // Can change a lot, but we need to be conservative
-			printf("#define HAVE_VFPV4\n");
+      // TODO: detect info from /sys if possible
+      printf("#define %s\n", cpuname[d]);
 			printf("#define L1_CODE_SIZE 49152\n");
 			printf("#define L1_CODE_LINESIZE 64\n");
 			printf("#define L1_CODE_ASSOCIATIVE 3\n");
 			printf("#define L1_DATA_SIZE 32768\n");
 			printf("#define L1_DATA_LINESIZE 64\n");
 			printf("#define L1_DATA_ASSOCIATIVE 2\n");
-			printf("#define L2_SIZE 2097152\n");
+      printf("#define L2_SIZE 524288\n");
 			printf("#define L2_LINESIZE 64\n");
 			printf("#define L2_ASSOCIATIVE 16\n");
 			printf("#define DTB_DEFAULT_ENTRIES 64\n");
 			printf("#define DTB_SIZE 4096\n");
 			break;
 
+    case CPU_FALKOR:
+      printf("#define FALKOR\n");
+      printf("#define L1_CODE_SIZE 65536\n");
+      printf("#define L1_CODE_LINESIZE 64\n");
+      printf("#define L1_DATA_SIZE 32768\n");
+      printf("#define L1_DATA_LINESIZE 128\n");
+      printf("#define L2_SIZE 524288\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");
+      break;
+
 		case CPU_THUNDERX:
-			printf("#define ARMV8\n");
 			printf("#define THUNDERX\n");
 			printf("#define L1_DATA_SIZE 32768\n");
 			printf("#define L1_DATA_LINESIZE 128\n");
@@ -248,11 +270,7 @@ void get_cpuconfig(void)
 			break;
 
 		case CPU_THUNDERX2T99:
-			printf("#define VULCAN                        \n");
+			printf("#define THUNDERX2T99                  \n");
-			printf("#define HAVE_VFP                      \n");
-			printf("#define HAVE_VFPV3                    \n");
-			printf("#define HAVE_NEON                     \n");
-			printf("#define HAVE_VFPV4                    \n");
 			printf("#define L1_CODE_SIZE         32768    \n");
 			printf("#define L1_CODE_LINESIZE     64       \n");
 			printf("#define L1_CODE_ASSOCIATIVE  8        \n");

cpuid_mips.c

@@ -72,10 +72,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define CPU_UNKNOWN     0
 #define CPU_P5600       1
+#define CPU_1004K	2
 
 static char *cpuname[] = {
-  "UNKOWN",
+  "UNKNOWN",
-  "P5600"
+  "P5600",
+  "1004K"
 };
 
 int detect(void){
@@ -90,7 +92,7 @@ int detect(void){
     if (!strncmp("cpu", buffer, 3)){
 	p = strchr(buffer, ':') + 2;
 #if 0
-	fprintf(stderr, "%s\n", p);
+	fprintf(stderr, "%s \n", p);
 #endif
 	break;
       }
@@ -99,43 +101,13 @@ int detect(void){
   fclose(infile);
 
   if(p != NULL){
-  if (strstr(p, "Loongson-3A")){
+  if (strstr(p, "5600")) {
-    return CPU_LOONGSON3A;
+    return CPU_P5600;
-  }else if(strstr(p, "Loongson-3B")){
+  } else if (strstr(p, "1004K")) {
-    return CPU_LOONGSON3B;
+    return CPU_1004K;
-  }else if (strstr(p, "Loongson-3")){
+  } else  
-    infile = fopen("/proc/cpuinfo", "r");
-    p = (char *)NULL;
-    while (fgets(buffer, sizeof(buffer), infile)){
-      if (!strncmp("system type", buffer, 11)){
-	p = strchr(buffer, ':') + 2;
-	break;
-      }
-    }
-    fclose(infile);
-    if (strstr(p, "loongson3a"))
-      return CPU_LOONGSON3A;
-  }else{
     return CPU_UNKNOWN;
   }
-  }
-  //Check model name for Loongson3
-  infile = fopen("/proc/cpuinfo", "r");
-  p = (char *)NULL;
-  while (fgets(buffer, sizeof(buffer), infile)){
-    if (!strncmp("model name", buffer, 10)){
-      p = strchr(buffer, ':') + 2;
-      break;
-    }
-  }
-  fclose(infile);
-  if(p != NULL){
-  if (strstr(p, "Loongson-3A")){
-    return CPU_LOONGSON3A;
-  }else if(strstr(p, "Loongson-3B")){
-    return CPU_LOONGSON3B;
-  }
-  }
 #endif
     return CPU_UNKNOWN;
 }
@@ -149,7 +121,7 @@ void get_architecture(void){
 }
 
 void get_subarchitecture(void){
-  if(detect()==CPU_P5600){
+  if(detect()==CPU_P5600|| detect()==CPU_1004K){
     printf("P5600");
   }else{
     printf("UNKNOWN");
@@ -170,6 +142,14 @@ void get_cpuconfig(void){
     printf("#define DTB_DEFAULT_ENTRIES 64\n");
     printf("#define DTB_SIZE 4096\n");
     printf("#define L2_ASSOCIATIVE 8\n");
+  } else if (detect()==CPU_1004K) {
+    printf("#define MIPS1004K\n");
+    printf("#define L1_DATA_SIZE 32768\n");
+    printf("#define L1_DATA_LINESIZE 32\n");
+    printf("#define L2_SIZE 26144\n");
+    printf("#define DTB_DEFAULT_ENTRIES 8\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 4\n");
   }else{
     printf("#define UNKNOWN\n");
   }
@@ -178,6 +158,8 @@ void get_cpuconfig(void){
 void get_libname(void){
   if(detect()==CPU_P5600) {
     printf("p5600\n");
+  } else if (detect()==CPU_1004K) {
+    printf("1004K\n");
   }else{
     printf("mips\n");
   }

cpuid_mips64.c

@@ -79,7 +79,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CPU_I6500       6
 
 static char *cpuname[] = {
-  "UNKOWN",
+  "UNKNOWN",
   "SICORTEX",
   "LOONGSON3A",
   "LOONGSON3B",

cpuid_power.c

@@ -56,6 +56,7 @@
 #define CPUTYPE_CELL       6
 #define CPUTYPE_PPCG4	   7
 #define CPUTYPE_POWER8     8
+#define CPUTYPE_POWER9     9
 
 char *cpuname[] = {
   "UNKNOWN",
@@ -66,7 +67,8 @@ char *cpuname[] = {
   "POWER6",
   "CELL",
   "PPCG4",
-  "POWER8"
+  "POWER8",
+  "POWER9"
 };
 
 char *lowercpuname[] = {
@@ -78,7 +80,8 @@ char *lowercpuname[] = {
   "power6",
   "cell",
   "ppcg4",
-  "power8"
+  "power8",
+  "power9"	
 };
 
 char *corename[] = {
@@ -90,6 +93,7 @@ char *corename[] = {
   "POWER6",
   "CELL",
   "PPCG4",
+  "POWER8",
   "POWER8"   	
 };
 
@@ -120,6 +124,7 @@ int detect(void){
   if (!strncasecmp(p, "POWER6", 6)) return CPUTYPE_POWER6;
   if (!strncasecmp(p, "POWER7", 6)) return CPUTYPE_POWER6;
   if (!strncasecmp(p, "POWER8", 6)) return CPUTYPE_POWER8;
+  if (!strncasecmp(p, "POWER9", 6)) return CPUTYPE_POWER8;
   if (!strncasecmp(p, "Cell",   4)) return CPUTYPE_CELL;
   if (!strncasecmp(p, "7447",   4)) return CPUTYPE_PPCG4;
 
@@ -127,6 +132,33 @@ int detect(void){
 #endif
 
 #ifdef _AIX
+  FILE *infile;
+  char buffer[512], *p;
+
+  p = (char *)NULL;
+  infile = popen("prtconf|grep 'Processor Type'", "r");
+  while (fgets(buffer, sizeof(buffer), infile)){
+    if (!strncmp("Pro", buffer, 3)){
+	p = strchr(buffer, ':') + 2;
+#if 0
+	fprintf(stderr, "%s\n", p);
+#endif
+	break;
+      }
+  }
+
+  pclose(infile);
+
+  if (!strncasecmp(p, "POWER3", 6)) return CPUTYPE_POWER3;
+  if (!strncasecmp(p, "POWER4", 6)) return CPUTYPE_POWER4;
+  if (!strncasecmp(p, "PPC970", 6)) return CPUTYPE_PPC970;
+  if (!strncasecmp(p, "POWER5", 6)) return CPUTYPE_POWER5;
+  if (!strncasecmp(p, "POWER6", 6)) return CPUTYPE_POWER6;
+  if (!strncasecmp(p, "POWER7", 6)) return CPUTYPE_POWER6;
+  if (!strncasecmp(p, "POWER8", 6)) return CPUTYPE_POWER8;
+  if (!strncasecmp(p, "POWER9", 6)) return CPUTYPE_POWER8;
+  if (!strncasecmp(p, "Cell",   4)) return CPUTYPE_CELL;
+  if (!strncasecmp(p, "7447",   4)) return CPUTYPE_PPCG4;
   return CPUTYPE_POWER5;
 #endif
 
@@ -142,6 +174,52 @@ int detect(void){
 
   return  CPUTYPE_PPC970;
 #endif
+
+#if defined(__FreeBSD__) || defined(__OpenBSD__) || defined(__NetBSD__)
+int id;
+__asm __volatile("mfpvr %0" : "=r"(id));
+switch ( id >> 16 ) {
+  case 0x4e: // POWER9
+    return CPUTYPE_POWER8;
+    break;
+  case 0x4d:
+  case 0x4b: // POWER8/8E 
+    return CPUTYPE_POWER8;
+    break;
+  case 0x4a:
+  case 0x3f:  // POWER7/7E
+    return CPUTYPE_POWER6; 
+    break;
+  case 0x3e:
+    return CPUTYPE_POWER6;
+    break;
+  case 0x3a:
+    return CPUTYPE_POWER5;
+    break;
+  case 0x35:
+  case 0x38: // POWER4 /4+ 
+    return CPUTYPE_POWER4;
+    break;
+  case 0x40:
+  case 0x41: // POWER3 /3+ 
+    return CPUTYPE_POWER3;
+    break;
+  case 0x39:
+  case 0x3c:
+  case 0x44:
+  case 0x45:
+    return CPUTYPE_PPC970;
+    break;
+  case 0x70: 
+    return CPUTYPE_CELL;
+    break;
+  case 0x8003: 
+    return CPUTYPE_PPCG4;
+    break;
+  default:  
+    return  CPUTYPE_UNKNOWN;
+  }
+#endif
 }
 
 void get_architecture(void){

cpuid_sparc.c

@@ -57,3 +57,8 @@ void get_cpuconfig(void){
 void get_libname(void){
   printf("v9\n");
 }
+
+char *get_corename(void){
+  return "sparc";
+}
+

cpuid_x86.c

@@ -50,6 +50,8 @@
 #ifdef NO_AVX
 #define CPUTYPE_HASWELL CPUTYPE_NEHALEM
 #define CORE_HASWELL CORE_NEHALEM
+#define CPUTYPE_SKYLAKEX CPUTYPE_NEHALEM
+#define CORE_SKYLAKEX CORE_NEHALEM
 #define CPUTYPE_SANDYBRIDGE CPUTYPE_NEHALEM
 #define CORE_SANDYBRIDGE CORE_NEHALEM
 #define CPUTYPE_BULLDOZER CPUTYPE_BARCELONA
@@ -95,10 +97,10 @@ static C_INLINE void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
     ("mov %%ebx, %%edi;"
      "cpuid;"
      "xchgl %%ebx, %%edi;"
-     : "=a" (*eax), "=D" (*ebx), "=c" (*ecx), "=d" (*edx) : "a" (op) : "cc");
+     : "=a" (*eax), "=D" (*ebx), "=c" (*ecx), "=d" (*edx) : "a" (op), "c" (0) : "cc");
 #else
   __asm__ __volatile__
-    ("cpuid": "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx) : "a" (op) : "cc");
+    ("cpuid": "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx) : "a" (op) , "c" (0) : "cc");
 #endif
 }
 
@@ -209,6 +211,44 @@ int support_avx(){
 #endif
 }
 
+int support_avx2(){
+#ifndef NO_AVX2
+  int eax, ebx, ecx=0, edx;
+  int ret=0;
+
+  if (!support_avx()) 
+    return 0;
+  cpuid(7, &eax, &ebx, &ecx, &edx);
+  if((ebx & (1<<7)) != 0)
+      ret=1;  //OS supports AVX2
+  return ret;
+#else
+  return 0;
+#endif
+}
+
+int support_avx512(){
+#ifndef NO_AVX512
+  int eax, ebx, ecx, edx;
+  int ret=0;
+
+  if (!support_avx()) 
+    return 0;
+  cpuid(7, &eax, &ebx, &ecx, &edx);
+  if((ebx & 32) != 32){
+      ret=0;  //OS does not even support AVX2
+  }
+  if((ebx & (1<<31)) != 0){
+    xgetbv(0, &eax, &edx); 
+    if((eax & 0xe0) == 0xe0)
+      ret=1;  //OS supports AVX512VL
+  }
+  return ret;
+#else
+  return 0;
+#endif
+}
+
 
 int get_vendor(void){
   int eax, ebx, ecx, edx;
@@ -231,6 +271,7 @@ int get_vendor(void){
   if (!strcmp(vendor, " SiS SiS SiS")) return VENDOR_SIS;
   if (!strcmp(vendor, "GenuineTMx86")) return VENDOR_TRANSMETA;
   if (!strcmp(vendor, "Geode by NSC")) return VENDOR_NSC;
+  if (!strcmp(vendor, "HygonGenuine")) return VENDOR_HYGON;
 
   if ((eax == 0) || ((eax & 0x500) != 0)) return VENDOR_INTEL;
 
@@ -292,6 +333,8 @@ int get_cputype(int gettype){
     if ((ecx & (1 << 20)) != 0) feature |= HAVE_SSE4_2;
 #ifndef NO_AVX
     if (support_avx()) feature |= HAVE_AVX;
+    if (support_avx2()) feature |= HAVE_AVX2;
+    if (support_avx512()) feature |= HAVE_AVX512VL;
     if ((ecx & (1 << 12)) != 0) feature |= HAVE_FMA3;
 #endif
 
@@ -1004,7 +1047,9 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
     }
   }
 
-  if ((get_vendor() == VENDOR_AMD) || (get_vendor() == VENDOR_CENTAUR)) {
+  if ((get_vendor() == VENDOR_AMD) ||
+      (get_vendor() == VENDOR_HYGON) ||
+      (get_vendor() == VENDOR_CENTAUR)) {
     cpuid(0x80000005, &eax, &ebx, &ecx, &edx);
 
     LDTB.size        = 4096;
@@ -1226,22 +1271,18 @@ int get_cpuname(void){
 	    return CPUTYPE_NEHALEM;
         case 12:
 	case 15:
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	case 13:
 	  //Broadwell
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	}
@@ -1250,33 +1291,27 @@ int get_cpuname(void){
         switch (model) {
         case 5:
 	case 6:
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	case 7:
 	case 15:
 	  //Broadwell
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	case 14:
 	  //Skylake
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	case 12:
@@ -1290,33 +1325,36 @@ int get_cpuname(void){
         switch (model) {
 	case 6:
 	  //Broadwell
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	case 5:
+	  // Skylake X
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	  return CPUTYPE_NEHALEM;
         case 14:
 	  // Skylake
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	case 7:
 	    // Xeon Phi Knights Landing
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	case 12:
@@ -1324,16 +1362,27 @@ int get_cpuname(void){
 	    return CPUTYPE_NEHALEM;
 	}
 	break;
+      case 6:
+        switch (model) {
+        case 6: // Cannon Lake
+          if(support_avx512())
+            return CPUTYPE_SKYLAKEX;
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	  return CPUTYPE_NEHALEM;
+        }
+      break;  
       case 9:
       case 8: 
         switch (model) {
 	case 14: // Kaby Lake
-          if(support_avx())
+          if(support_avx2())
-#ifndef NO_AVX2
             return CPUTYPE_HASWELL;
-#else
+          if(support_avx())
 	    return CPUTYPE_SANDYBRIDGE;
-#endif
           else
 	    return CPUTYPE_NEHALEM;
 	}
@@ -1420,6 +1469,8 @@ int get_cpuname(void){
 	switch (model) {
 	case 1:
 	  // AMD Ryzen
+	case 8:
+	  // AMD Ryzen2
 	  if(support_avx())
 #ifndef NO_AVX2
 	    return CPUTYPE_ZEN;
@@ -1435,6 +1486,26 @@ int get_cpuname(void){
     return CPUTYPE_AMD_UNKNOWN;
   }
 
+  if (vendor == VENDOR_HYGON){
+    switch (family) {
+    case 0xf:
+      switch (exfamily) {
+      case 9:
+          //Hygon Dhyana
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CPUTYPE_ZEN;
+#else
+	    return CPUTYPE_SANDYBRIDGE; // closer in architecture to Sandy Bridge than to Excavator
+#endif
+	  else
+	    return CPUTYPE_BARCELONA;
+        }
+      break;
+    }
+    return CPUTYPE_HYGON_UNKNOWN;
+  }
+
   if (vendor == VENDOR_CYRIX){
     switch (family) {
     case 0x4:
@@ -1556,6 +1627,8 @@ static char *cpuname[] = {
   "STEAMROLLER",
   "EXCAVATOR",
   "ZEN",
+  "SKYLAKEX",
+  "DHYANA"
 };
 
 static char *lowercpuname[] = {
@@ -1610,10 +1683,12 @@ static char *lowercpuname[] = {
   "steamroller",
   "excavator",
   "zen",
+  "skylakex",
+  "dhyana"
 };
 
 static char *corename[] = {
-  "UNKOWN",
+  "UNKNOWN",
   "80486",
   "P5",
   "P6",
@@ -1641,6 +1716,8 @@ static char *corename[] = {
   "STEAMROLLER",
   "EXCAVATOR",
   "ZEN",
+  "SKYLAKEX",
+  "DHYANA"
 };
 
 static char *corename_lower[] = {
@@ -1672,6 +1749,8 @@ static char *corename_lower[] = {
   "steamroller",
   "excavator",
   "zen",
+  "skylakex",
+  "dhyana"
 };
 
 
@@ -1860,6 +1939,19 @@ int get_coretype(void){
           else
 	    return CORE_NEHALEM;
 	case 5:
+	 // Skylake X
+#ifndef NO_AVX512
+	    return CORE_SKYLAKEX;
+#else
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CORE_HASWELL;
+#else
+	    return CORE_SANDYBRIDGE;
+#endif
+	  else
+	    return CORE_NEHALEM;
+#endif			
 	case 14:
 	  // Skylake
           if(support_avx())
@@ -1958,6 +2050,8 @@ int get_coretype(void){
 	switch (model) {
 	case 1:
 	  // AMD Ryzen
+	case 8:
+	  // Ryzen 2		
 	  if(support_avx())
 #ifndef NO_AVX2
 	    return CORE_ZEN;
@@ -1973,6 +2067,23 @@ int get_coretype(void){
     }
   }
 
+  if (vendor == VENDOR_HYGON){
+    if (family == 0xf){
+        if (exfamily == 9) {
+	  if(support_avx())
+#ifndef NO_AVX2
+	    return CORE_ZEN;
+#else
+	    return CORE_SANDYBRIDGE; // closer in architecture to Sandy Bridge than to Excavator
+#endif
+	  else
+	    return CORE_BARCELONA;
+	} else {
+		return CORE_BARCELONA;
+	}
+    }
+  }
+
   if (vendor == VENDOR_CENTAUR) {
     switch (family) {
     case 0x6:
@@ -2059,6 +2170,8 @@ void get_cpuconfig(void){
     if (features & HAVE_SSE4A)   printf("#define HAVE_SSE4A\n");
     if (features & HAVE_SSE5 )   printf("#define HAVE_SSSE5\n");
     if (features & HAVE_AVX )    printf("#define HAVE_AVX\n");
+    if (features & HAVE_AVX2 )    printf("#define HAVE_AVX2\n");
+    if (features & HAVE_AVX512VL )    printf("#define HAVE_AVX512VL\n");
     if (features & HAVE_3DNOWEX) printf("#define HAVE_3DNOWEX\n");
     if (features & HAVE_3DNOW)   printf("#define HAVE_3DNOW\n");
     if (features & HAVE_FMA4 )    printf("#define HAVE_FMA4\n");
@@ -2127,6 +2240,8 @@ void get_sse(void){
   if (features & HAVE_SSE4A)   printf("HAVE_SSE4A=1\n");
   if (features & HAVE_SSE5 )   printf("HAVE_SSSE5=1\n");
   if (features & HAVE_AVX )    printf("HAVE_AVX=1\n");
+  if (features & HAVE_AVX2 )    printf("HAVE_AVX2=1\n");
+  if (features & HAVE_AVX512VL )    printf("HAVE_AVX512VL=1\n");
   if (features & HAVE_3DNOWEX) printf("HAVE_3DNOWEX=1\n");
   if (features & HAVE_3DNOW)   printf("HAVE_3DNOW=1\n");
   if (features & HAVE_FMA4 )    printf("HAVE_FMA4=1\n");

cpuid_zarch.c

@@ -29,15 +29,18 @@
 
 #define CPU_GENERIC    	0
 #define CPU_Z13       	1
+#define CPU_Z14       	2
 
 static char *cpuname[] = {
   "ZARCH_GENERIC",
-  "Z13"
+  "Z13",
+  "Z14"
 };
 
 static char *cpuname_lower[] = {
   "zarch_generic",
-  "z13"
+  "z13",
+  "z14"
 };
 
 int detect(void)
@@ -62,6 +65,10 @@ int detect(void)
   if (strstr(p, "2964")) return CPU_Z13;
   if (strstr(p, "2965")) return CPU_Z13;
 
+  /* detect z14, but fall back to z13 */
+  if (strstr(p, "3906")) return CPU_Z13;
+  if (strstr(p, "3907")) return CPU_Z13;
+
   return CPU_GENERIC;
 }
 
@@ -107,5 +114,9 @@ void get_cpuconfig(void)
 	  printf("#define Z13\n");
 	  printf("#define DTB_DEFAULT_ENTRIES 64\n");
 	  break;
+	case CPU_Z14:
+	  printf("#define Z14\n");
+	  printf("#define DTB_DEFAULT_ENTRIES 64\n");
+	  break;
 	}
 }

ctest.c

@@ -60,6 +60,14 @@ OS_FREEBSD
 OS_NETBSD
 #endif
 
+#if defined(__OpenBSD__)
+OS_OPENBSD
+#endif
+
+#if defined(__DragonFly__)
+OS_DRAGONFLY
+#endif
+
 #if defined(__sun)
 OS_SUNOS
 #endif
@@ -93,6 +101,10 @@ OS_INTERIX
 OS_LINUX
 #endif
 
+#if defined(__HAIKU__)
+OS_HAIKU
+#endif
+
 #if defined(__i386) || defined(_X86)
 ARCH_X86
 #endif

ctest/Makefile

@@ -102,7 +102,13 @@ clean ::
 	rm -f x*
 
 FLDFLAGS = $(FFLAGS:-fPIC=) $(LDFLAGS)
-CEXTRALIB =
+ifeq ($(USE_OPENMP), 1)
+ifeq ($(F_COMPILER), GFORTRAN)
+ifeq ($(C_COMPILER), CLANG)
+CEXTRALIB = -lomp
+endif
+endif
+endif
 
 # Single real
 xscblat1: $(stestl1o) c_sblat1.o $(TOPDIR)/$(LIBNAME)

driver/level2/gemv_thread.c

@@ -62,9 +62,36 @@
 #endif
 #endif
 
-#ifndef TRANSA
+#ifndef thread_local
+# if __STDC_VERSION__ >= 201112 && !defined __STDC_NO_THREADS__
+#  define thread_local _Thread_local
+# elif defined _WIN32 && ( \
+       defined _MSC_VER || \
+       defined __ICL || \
+       defined __DMC__ || \
+       defined __BORLANDC__ )
+#  define thread_local __declspec(thread) 
+/* note that ICC (linux) and Clang are covered by __GNUC__ */
+# elif defined __GNUC__ || \
+       defined __SUNPRO_C || \
+       defined __xlC__
+#  define thread_local __thread
+# else
+# define UNSAFE
+#endif
+#endif
+#if defined USE_OPENMP
+#undef UNSAFE
+#endif
+
+#if !defined(TRANSA) && !defined(UNSAFE)
 #define Y_DUMMY_NUM 1024
+#if defined(USE_OPENMP)
 static FLOAT y_dummy[Y_DUMMY_NUM];
+#pragma omp threadprivate(y_dummy)
+# else
+static thread_local FLOAT y_dummy[Y_DUMMY_NUM];
+# endif
 #endif
 
 static int gemv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *dummy1, FLOAT *buffer, BLASLONG pos){
@@ -105,10 +132,12 @@ static int gemv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
 #ifdef TRANSA
     y += n_from * incy * COMPSIZE;
 #else
+# ifndef UNSAFE
     //for split matrix row (n) direction and vector x of gemv_n
     x += n_from * incx * COMPSIZE;
     //store partial result for every thread
     y += (m_to - m_from) * 1 * COMPSIZE * pos;
+# endif
 #endif
   }
 
@@ -136,7 +165,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
 
   BLASLONG width, i, num_cpu;
 
-#ifndef TRANSA
+#if !defined(TRANSA) && !defined(UNSAFE)
   int split_x=0;
 #endif
 
@@ -212,7 +241,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
     i -= width;
   }
 
-#ifndef TRANSA
+#if !defined(TRANSA) && !defined(UNSAFE) 
   //try to split matrix on row direction and x.
   //Then, reduction.
   if (num_cpu < nthreads) {
@@ -272,7 +301,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
     exec_blas(num_cpu, queue);
   }
 
-#ifndef TRANSA
+#if !defined(TRANSA) && !defined(UNSAFE)
   if(split_x==1){
     //reduction
     for(i=0; i<num_cpu; i++){

driver/level2/trmv_U.c

@@ -54,16 +54,12 @@ int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, FLOAT *bu
     COPY_K(m, b, incb, buffer, 1);
   }
 
-/*FIXME the GEMV unrolling performed here was found to be broken, see issue 1332 */
+  for (is = 0; is < m; is += DTB_ENTRIES){
-/* Multiplying DTB size by 100 is just a quick-and-dirty hack to disable it for now[B */
 
-  for (is = 0; is < m; is += DTB_ENTRIES * 100){
+    min_i = MIN(m - is, DTB_ENTRIES);
-
-    min_i = MIN(m - is, DTB_ENTRIES * 100);
 
 #ifndef TRANSA
     if (is > 0){ 
-fprintf(stderr,"WARNING unrolling of the trmv_U loop may give wrong results\n");    
       GEMV_N(is, min_i, 0, dp1,
 	     a + is * lda,  lda,
 	     B + is, 1,

driver/level3/Makefile

@@ -362,7 +362,7 @@ cgemm_ct.$(SUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 cgemm_cr.$(SUFFIX) : gemm.c level3.c  ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 cgemm_cc.$(SUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -410,7 +410,7 @@ zgemm_ct.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 zgemm_cr.$(SUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 zgemm_cc.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -458,7 +458,7 @@ xgemm_ct.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 xgemm_cr.$(SUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 xgemm_cc.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -558,7 +558,7 @@ cgemm_thread_ct.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 cgemm_thread_cr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 cgemm_thread_cc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -606,7 +606,7 @@ zgemm_thread_ct.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 zgemm_thread_cr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 zgemm_thread_cc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -654,7 +654,7 @@ xgemm_thread_ct.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 xgemm_thread_cr.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 xgemm_thread_cc.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -1821,7 +1821,7 @@ cgemm3m_ct.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 cgemm3m_cr.$(SUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 cgemm3m_cc.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -1869,7 +1869,7 @@ zgemm3m_ct.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 zgemm3m_cr.$(SUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 zgemm3m_cc.$(SUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -1917,7 +1917,7 @@ xgemm3m_ct.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 xgemm3m_cr.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 xgemm3m_cc.$(SUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -1974,7 +1974,7 @@ cgemm3m_thread_ct.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 cgemm3m_thread_cr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 cgemm3m_thread_cc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -2022,7 +2022,7 @@ zgemm3m_thread_ct.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 zgemm3m_thread_cr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 zgemm3m_thread_cc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -2070,7 +2070,7 @@ xgemm3m_thread_ct.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 xgemm3m_thread_cr.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 xgemm3m_thread_cc.$(SUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -2731,7 +2731,7 @@ cgemm_ct.$(PSUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 cgemm_cr.$(PSUFFIX) : gemm.c level3.c  ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 cgemm_cc.$(PSUFFIX) : gemm.c level3.c  ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -2779,7 +2779,7 @@ zgemm_ct.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 zgemm_cr.$(PSUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 zgemm_cc.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -2827,7 +2827,7 @@ xgemm_ct.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 xgemm_cr.$(PSUFFIX) : gemm.c level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 xgemm_cc.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -2927,7 +2927,7 @@ cgemm_thread_ct.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 cgemm_thread_cr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 cgemm_thread_cc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -2975,7 +2975,7 @@ zgemm_thread_ct.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 zgemm_thread_cr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 zgemm_thread_cc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -3023,7 +3023,7 @@ xgemm_thread_ct.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 xgemm_thread_cr.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 xgemm_thread_cc.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -4190,7 +4190,7 @@ cgemm3m_ct.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 cgemm3m_cr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 cgemm3m_cc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -4238,7 +4238,7 @@ zgemm3m_ct.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 zgemm3m_cr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 zgemm3m_cc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -4286,7 +4286,7 @@ xgemm3m_ct.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 xgemm3m_cr.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 xgemm3m_cc.$(PSUFFIX) : gemm3m.c gemm3m_level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -4343,7 +4343,7 @@ cgemm3m_thread_ct.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 cgemm3m_thread_cr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 cgemm3m_thread_cc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -4391,7 +4391,7 @@ zgemm3m_thread_ct.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 zgemm3m_thread_cr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 zgemm3m_thread_cc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCC $< -o $(@F)
@@ -4439,7 +4439,7 @@ xgemm3m_thread_ct.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCT $< -o $(@F)
 
 xgemm3m_thread_cr.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
-	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR=CR $< -o $(@F)
 
 xgemm3m_thread_cc.$(PSUFFIX) : gemm3m.c level3_gemm3m_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCC $< -o $(@F)

driver/level3/level3_gemm3m_thread.c

@@ -91,7 +91,12 @@
 #endif
 
 typedef struct {
-  volatile BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
+#if __STDC_VERSION__ >= 201112L
+  _Atomic
+#else
+  volatile
+#endif  
+   BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
 } job_t;
 
 

driver/level3/level3_syrk_threaded.c

@@ -67,7 +67,12 @@
 #endif
 
 typedef struct {
-  volatile BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
+#if __STDC_VERSION__ >= 201112L
+_Atomic
+#else 
+  volatile
+#endif
+   BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
 } job_t;
 
 

driver/level3/level3_thread.c

@@ -48,6 +48,10 @@
 #define SWITCH_RATIO 2
 #endif
 
+#ifndef GEMM_PREFERED_SIZE
+#define GEMM_PREFERED_SIZE 1
+#endif
+
 //The array of job_t may overflow the stack.
 //Instead, use malloc to alloc job_t.
 #if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD
@@ -91,7 +95,8 @@
 #endif
 
 typedef struct {
-  volatile BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
+  volatile
+   BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
 } job_t;
 
 
@@ -346,7 +351,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
       /* Make sure if no one is using workspace */
       START_RPCC();
       for (i = 0; i < args -> nthreads; i++)
-	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;};
+	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;MB;};
       STOP_RPCC(waiting1);
 
 #if defined(FUSED_GEMM) && !defined(TIMING)
@@ -408,7 +413,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
 	  /* Wait until other region of B is initialized */
 	  START_RPCC();
-	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;};
+	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;MB;};
 	  STOP_RPCC(waiting2);
 
           /* Apply kernel with local region of A and part of other region of B */
@@ -426,6 +431,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
         /* Clear synchronization flag if this thread is done with other region of B */
 	if (m_to - m_from == min_i) {
 	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
+	  WMB;
 	}
       }
     } while (current != mypos);
@@ -487,7 +493,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
   START_RPCC();
   for (i = 0; i < args -> nthreads; i++) {
     for (js = 0; js < DIVIDE_RATE; js++) {
-      while (job[mypos].working[i][CACHE_LINE_SIZE * js] ) {YIELDING;};
+      while (job[mypos].working[i][CACHE_LINE_SIZE * js] ) {YIELDING;MB;};
     }
   }
   STOP_RPCC(waiting3);
@@ -508,10 +514,29 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
   return 0;
 }
 
+static int round_up(int remainder, int width, int multiple)
+{
+	if (multiple > remainder || width <= multiple)
+		return width;
+	width = (width + multiple - 1) / multiple;
+	width = width * multiple;
+	return width;
+}
+
+
 static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
 		       *range_n, FLOAT *sa, FLOAT *sb,
                        BLASLONG nthreads_m, BLASLONG nthreads_n) {
 
+#ifndef USE_OPENMP
+#ifndef OS_WINDOWS
+static pthread_mutex_t  level3_lock    = PTHREAD_MUTEX_INITIALIZER;
+#else
+CRITICAL_SECTION level3_lock;
+InitializeCriticalSection((PCRITICAL_SECTION)&level3_lock);
+#endif
+#endif
+
   blas_arg_t newarg;
 
 #ifndef USE_ALLOC_HEAP
@@ -552,6 +577,14 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
 #endif
 #endif
 
+#ifndef USE_OPENMP
+#ifndef OS_WINDOWS
+pthread_mutex_lock(&level3_lock);
+#else
+EnterCriticalSection((PCRITICAL_SECTION)&level3_lock);
+#endif
+#endif
+
 #ifdef USE_ALLOC_HEAP
   /* Dynamically allocate workspace */
   job = (job_t*)malloc(MAX_CPU_NUMBER * sizeof(job_t));
@@ -599,9 +632,14 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
   num_parts = 0;
   while (m > 0){
     width = blas_quickdivide(m + nthreads_m - num_parts - 1, nthreads_m - num_parts);
+
+    width = round_up(m, width, GEMM_PREFERED_SIZE);
+
     m -= width;
+
     if (m < 0) width = width + m;
     range_M[num_parts + 1] = range_M[num_parts] + width;
+
     num_parts ++;
   }
   for (i = num_parts; i < MAX_CPU_NUMBER; i++) {
@@ -643,9 +681,12 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
       if (width < SWITCH_RATIO) {
         width = SWITCH_RATIO;
       }
+      width = round_up(n, width, GEMM_PREFERED_SIZE);
+
       n -= width;
       if (n < 0) width = width + n;
       range_N[num_parts + 1] = range_N[num_parts] + width;
+
       num_parts ++;
     }
     for (j = num_parts; j < MAX_CPU_NUMBER; j++) {
@@ -653,8 +694,8 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
     }
 
     /* Clear synchronization flags */
-    for (i = 0; i < MAX_CPU_NUMBER; i++) {
+    for (i = 0; i < nthreads; i++) {
-      for (j = 0; j < MAX_CPU_NUMBER; j++) {
+      for (j = 0; j < nthreads; j++) {
 	for (k = 0; k < DIVIDE_RATE; k++) {
 	  job[i].working[j][CACHE_LINE_SIZE * k] = 0;
 	}
@@ -669,6 +710,14 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
   free(job);
 #endif
 
+#ifndef USE_OPENMP
+#ifndef OS_WINDOWS
+  pthread_mutex_unlock(&level3_lock);
+#else
+  LeaveCriticalSection((PCRITICAL_SECTION)&level3_lock);
+#endif
+#endif
+
   return 0;
 }
 

driver/level3/syrk_thread.c

@@ -48,7 +48,7 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
 
   BLASLONG width, i;
   BLASLONG n_from, n_to;
-  double dnum, nf, nt, di;
+  double dnum, nf, nt, di, dinum;
 
   int num_cpu;
   int mask = 0;
@@ -109,7 +109,11 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
       if (nthreads - num_cpu > 1) {
 
 	di = (double)i;
-	width = (BLASLONG)(( sqrt(di * di + dnum) - di + mask)/(mask+1)) * (mask+1);
+	dinum = di * di +dnum;
+	if (dinum <0)
+	  width = (BLASLONG)(( - di + mask)/(mask+1)) * (mask+1);
+	else
+	  width = (BLASLONG)(( sqrt(dinum) - di + mask)/(mask+1)) * (mask+1);
 
 	if ((width <= 0) || (width > n_to - i)) width = n_to - i;
 
@@ -136,9 +140,7 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
 
     nf = (double)(arg -> n - n_from);
     nt = (double)(arg -> n - n_to);
-
     dnum = (nt * nt - nf * nf) / (double)nthreads;
-
     num_cpu  = 0;
 
     range[0] = n_from;
@@ -149,8 +151,11 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
       if (nthreads - num_cpu > 1) {
 
 	di = (double)(arg -> n - i);
-	width = ((BLASLONG)((-sqrt(di * di + dnum) + di) + mask)/(mask+1)) * (mask+1);
+	dinum = di * di + dnum;
-
+	if (dinum<0)
+	  width = ((BLASLONG)(di + mask)/(mask+1)) * (mask+1);
+	else
+	  width = ((BLASLONG)((-sqrt(dinum) + di) + mask)/(mask+1)) * (mask+1);
 	if ((width <= 0) || (width > n_to - i)) width = n_to - i;
 
       } else {

driver/others/CMakeLists.txt

@@ -47,7 +47,11 @@ GenerateNamedObjects("abs.c" "DOUBLE" "z_abs" 0 "" "" 1)
 GenerateNamedObjects("openblas_get_config.c;openblas_get_parallel.c" "" "" 0 "" "" 1)
 
 if (DYNAMIC_ARCH)
-  list(APPEND COMMON_SOURCES dynamic.c)
+  if (ARM64)
+    list(APPEND COMMON_SOURCES dynamic_arm64.c)
+  else ()  
+    list(APPEND COMMON_SOURCES dynamic.c)
+  endif ()  
 else ()
   list(APPEND COMMON_SOURCES parameter.c)
 endif ()

driver/others/Makefile

@@ -15,7 +15,11 @@ endif
 # COMMONOBJS	+=  info.$(SUFFIX)
 
 ifeq ($(DYNAMIC_ARCH), 1)
+ifeq ($(ARCH),arm64)
+COMMONOBJS	+=  dynamic_arm64.$(SUFFIX)
+else
 COMMONOBJS	+=  dynamic.$(SUFFIX)
+endif
 else
 COMMONOBJS	+=  parameter.$(SUFFIX)
 endif
@@ -71,7 +75,11 @@ BLAS_SERVER = blas_server.c
 endif
 
 ifeq ($(DYNAMIC_ARCH), 1)
+ifeq ($(ARCH),arm64)
+HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic_arm64.$(SUFFIX)
+else
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic.$(SUFFIX)
+endif
 else
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) parameter.$(SUFFIX)
 endif

driver/others/blas_server.c

@@ -70,7 +70,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /*********************************************************************/
 
 #include "common.h"
-#if defined(OS_LINUX) || defined(OS_NETBSD) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_SUNOS) || defined(OS_FREEBSD)
+#if defined(OS_LINUX) || defined(OS_NETBSD) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_SUNOS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(OS_HAIKU)
 #include <dlfcn.h>
 #include <signal.h>
 #include <sys/resource.h>
@@ -582,7 +582,7 @@ int blas_thread_init(void){
       if(ret!=0){
 	struct rlimit rlim;
         const char *msg = strerror(ret);
-        fprintf(STDERR, "OpenBLAS blas_thread_init: pthread_create: %s\n", msg);
+        fprintf(STDERR, "OpenBLAS blas_thread_init: pthread_create failed for thread %ld of %ld: %s\n", i+1,blas_num_threads,msg);
 #ifdef RLIMIT_NPROC
         if(0 == getrlimit(RLIMIT_NPROC, &rlim)) {
           fprintf(STDERR, "OpenBLAS blas_thread_init: RLIMIT_NPROC "
@@ -850,6 +850,11 @@ void goto_set_num_threads(int num_threads) {
 
   long i;
 
+#ifdef SMP_SERVER
+  // Handle lazy re-init of the thread-pool after a POSIX fork
+  if (unlikely(blas_server_avail == 0)) blas_thread_init();
+#endif
+
   if (num_threads < 1) num_threads = blas_num_threads;
 
 #ifndef NO_AFFINITY

driver/others/blas_server_omp.c

@@ -36,6 +36,7 @@
 /* or implied, of The University of Texas at Austin.                 */
 /*********************************************************************/
 
+#include <stdbool.h>
 #include <stdio.h>
 #include <stdlib.h>
 //#include <sys/mman.h>
@@ -47,13 +48,22 @@
 
 #else
 
+#ifndef OMP_SCHED
+#define OMP_SCHED static
+#endif
+
 int blas_server_avail = 0;
 
-static void * blas_thread_buffer[MAX_CPU_NUMBER];
+static void * blas_thread_buffer[MAX_PARALLEL_NUMBER][MAX_CPU_NUMBER];
+#if __STDC_VERSION__ >= 201112L
+static atomic_bool blas_buffer_inuse[MAX_PARALLEL_NUMBER];
+#else
+static _Bool blas_buffer_inuse[MAX_PARALLEL_NUMBER];
+#endif
 
 void goto_set_num_threads(int num_threads) {
 
-  int i=0;
+  int i=0, j=0;
 
   if (num_threads < 1) num_threads = blas_num_threads;
 
@@ -68,15 +78,17 @@ void goto_set_num_threads(int num_threads) {
   omp_set_num_threads(blas_cpu_number);
 
   //adjust buffer for each thread
-  for(i=0; i<blas_cpu_number; i++){
+  for(i=0; i<MAX_PARALLEL_NUMBER; i++) {
-    if(blas_thread_buffer[i]==NULL){
+    for(j=0; j<blas_cpu_number; j++){
-      blas_thread_buffer[i]=blas_memory_alloc(2);
+      if(blas_thread_buffer[i][j]==NULL){
+        blas_thread_buffer[i][j]=blas_memory_alloc(2);
+      }
     }
-  }
+    for(; j<MAX_CPU_NUMBER; j++){
-  for(; i<MAX_CPU_NUMBER; i++){
+      if(blas_thread_buffer[i][j]!=NULL){
-    if(blas_thread_buffer[i]!=NULL){
+        blas_memory_free(blas_thread_buffer[i][j]);
-      blas_memory_free(blas_thread_buffer[i]);
+        blas_thread_buffer[i][j]=NULL;
-      blas_thread_buffer[i]=NULL;
+      }
     }
   }
 #if defined(ARCH_MIPS64)
@@ -92,30 +104,34 @@ void openblas_set_num_threads(int num_threads) {
 
 int blas_thread_init(void){
 
-  int i=0;
+  int i=0, j=0;
 
   blas_get_cpu_number();
 
   blas_server_avail = 1;
 
-  for(i=0; i<blas_num_threads; i++){
+  for(i=0; i<MAX_PARALLEL_NUMBER; i++) {
-    blas_thread_buffer[i]=blas_memory_alloc(2);
+    for(j=0; j<blas_num_threads; j++){
-  }
+      blas_thread_buffer[i][j]=blas_memory_alloc(2);
-  for(; i<MAX_CPU_NUMBER; i++){
+    }
-      blas_thread_buffer[i]=NULL;
+    for(; j<MAX_CPU_NUMBER; j++){
+      blas_thread_buffer[i][j]=NULL;
+    }
   }
 
   return 0;
 }
 
 int BLASFUNC(blas_thread_shutdown)(void){
-  int i=0;
+  int i=0, j=0;
   blas_server_avail = 0;
 
-  for(i=0; i<MAX_CPU_NUMBER; i++){
+  for(i=0; i<MAX_PARALLEL_NUMBER; i++) {
-    if(blas_thread_buffer[i]!=NULL){
+    for(j=0; j<MAX_CPU_NUMBER; j++){
-      blas_memory_free(blas_thread_buffer[i]);
+      if(blas_thread_buffer[i][j]!=NULL){
-      blas_thread_buffer[i]=NULL;
+        blas_memory_free(blas_thread_buffer[i][j]);
+        blas_thread_buffer[i][j]=NULL;
+      }
     }
   }
 
@@ -206,7 +222,7 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
       }
 }
 
-static void exec_threads(blas_queue_t *queue){
+static void exec_threads(blas_queue_t *queue, int buf_index){
 
   void *buffer, *sa, *sb;
   int pos=0, release_flag=0;
@@ -223,7 +239,7 @@ static void exec_threads(blas_queue_t *queue){
   if ((sa == NULL) && (sb == NULL) && ((queue -> mode & BLAS_PTHREAD) == 0)) {
 
     pos = omp_get_thread_num();
-    buffer = blas_thread_buffer[pos];
+    buffer = blas_thread_buffer[buf_index][pos];
 
     //fallback
     if(buffer==NULL) {
@@ -291,7 +307,7 @@ static void exec_threads(blas_queue_t *queue){
 
 int exec_blas(BLASLONG num, blas_queue_t *queue){
 
-  BLASLONG i;
+  BLASLONG i, buf_index;
 
   if ((num <= 0) || (queue == NULL)) return 0;
 
@@ -302,16 +318,39 @@ int exec_blas(BLASLONG num, blas_queue_t *queue){
   }
 #endif
 
-#pragma omp parallel for schedule(static)
+  while(true) {
+    for(i=0; i < MAX_PARALLEL_NUMBER; i++) {
+#if __STDC_VERSION__ >= 201112L
+      _Bool inuse = false;
+      if(atomic_compare_exchange_weak(&blas_buffer_inuse[i], &inuse, true)) {
+#else
+      if(blas_buffer_inuse[i] == false) {
+        blas_buffer_inuse[i] = true;
+#endif
+        buf_index = i;
+        break;
+      }
+    }
+    if(i != MAX_PARALLEL_NUMBER)
+      break;
+  }
+
+#pragma omp parallel for schedule(OMP_SCHED)
   for (i = 0; i < num; i ++) {
 
 #ifndef USE_SIMPLE_THREADED_LEVEL3
     queue[i].position = i;
 #endif
 
-    exec_threads(&queue[i]);
+    exec_threads(&queue[i], buf_index);
   }
 
+#if __STDC_VERSION__ >= 201112L
+  atomic_store(&blas_buffer_inuse[buf_index], false);
+#else
+  blas_buffer_inuse[buf_index] = false;
+#endif
+
   return 0;
 }
 

driver/others/blas_server_win32.c

@@ -40,6 +40,14 @@
 #include <stdlib.h>
 #include "common.h"
 
+#if defined(OS_CYGWIN_NT) && !defined(unlikely)
+#ifdef __GNUC__
+#define unlikely(x) __builtin_expect(!!(x), 0)
+#else
+#define unlikely(x) (x)
+#endif
+#endif
+
 /* This is a thread implementation for Win32 lazy implementation */
 
 /* Thread server common infomation */
@@ -53,7 +61,7 @@ typedef struct{
 
 } blas_pool_t;
 
-/* We need this grobal for cheking if initialization is finished.   */
+/* We need this global for cheking if initialization is finished.   */
 int blas_server_avail = 0;
 
 /* Local Variables */
@@ -340,6 +348,11 @@ int blas_thread_init(void){
 
 int exec_blas_async(BLASLONG pos, blas_queue_t *queue){
 
+#if defined(SMP_SERVER) && defined(OS_CYGWIN_NT)
+  // Handle lazy re-init of the thread-pool after a POSIX fork
+  if (unlikely(blas_server_avail == 0)) blas_thread_init();
+#endif
+
   blas_queue_t *current;
 
   current = queue;
@@ -405,6 +418,11 @@ int exec_blas_async_wait(BLASLONG num, blas_queue_t *queue){
 /* Execute Threads */
 int exec_blas(BLASLONG num, blas_queue_t *queue){
 
+#if defined(SMP_SERVER) && defined(OS_CYGWIN_NT)
+  // Handle lazy re-init of the thread-pool after a POSIX fork
+  if (unlikely(blas_server_avail == 0)) blas_thread_init();
+#endif
+
 #ifndef ALL_THREADED
    int (*routine)(blas_arg_t *, void *, void *, double *, double *, BLASLONG);
 #endif
@@ -460,7 +478,12 @@ int BLASFUNC(blas_thread_shutdown)(void){
 
 void goto_set_num_threads(int num_threads)
 {
-	 long i;
+	long i;
+
+#if defined(SMP_SERVER) && defined(OS_CYGWIN_NT)
+	// Handle lazy re-init of the thread-pool after a POSIX fork
+	if (unlikely(blas_server_avail == 0)) blas_thread_init();
+#endif
 
 	if (num_threads < 1) num_threads = blas_cpu_number;
 

driver/others/dynamic.c

@@ -49,6 +49,167 @@
 #define EXTERN
 #endif
 
+#ifdef DYNAMIC_LIST
+extern gotoblas_t gotoblas_PRESCOTT;
+
+#ifdef DYN_ATHLON
+extern gotoblas_t gotoblas_ATHLON;
+#else
+#define gotoblas_ATHLON gotoblas_PRESCOTT
+#endif
+#ifdef DYN_KATMAI
+extern gotoblas_t gotoblas_KATMAI;
+#else
+#define gotoblas_KATMAI gotoblas_PRESCOTT
+#endif
+#ifdef DYN_BANIAS
+extern gotoblas_t gotoblas_BANIAS;
+#else
+#define gotoblas_BANIAS gotoblas_PRESCOTT
+#endif
+#ifdef DYN_COPPERMINE
+extern gotoblas_t gotoblas_COPPERMINE;
+#else
+#define gotoblas_COPPERMINE gotoblas_PRESCOTT
+#endif
+#ifdef DYN_NORTHWOOD
+extern gotoblas_t gotoblas_NORTHWOOD;
+#else
+#define gotoblas_NORTHWOOD gotoblas_PRESCOTT
+#endif
+#ifdef DYN_CORE2
+extern gotoblas_t gotoblas_CORE2;
+#else
+#define gotoblas_CORE2 gotoblas_PRESCOTT
+#endif
+#ifdef DYN_NEHALEM
+extern gotoblas_t gotoblas_NEHALEM;
+#else
+#define gotoblas_NEHALEM gotoblas_PRESCOTT
+#endif
+#ifdef DYN_BARCELONA
+extern gotoblas_t gotoblas_BARCELONA;
+#elif defined(DYN_NEHALEM)
+#define gotoblas_BARCELONA gotoblas_NEHALEM
+#else
+#define gotoblas_BARCELONA gotoblas_PRESCOTT
+#endif
+#ifdef DYN_ATOM
+extern gotoblas_t gotoblas_ATOM;
+elif defined(DYN_NEHALEM)
+#define gotoblas_ATOM gotoblas_NEHALEM
+#else
+#define gotoblas_ATOM gotoblas_PRESCOTT
+#endif
+#ifdef DYN_NANO
+extern gotoblas_t gotoblas_NANO;
+#else
+#define gotoblas_NANO gotoblas_PRESCOTT
+#endif
+#ifdef DYN_PENRYN
+extern gotoblas_t gotoblas_PENRYN;
+#else
+#define gotoblas_PENRYN gotoblas_PRESCOTT
+#endif
+#ifdef DYN_DUNNINGTON
+extern gotoblas_t gotoblas_DUNNINGTON;
+#else
+#define gotoblas_DUNNINGTON gotoblas_PRESCOTT
+#endif
+#ifdef DYN_OPTERON
+extern gotoblas_t gotoblas_OPTERON;
+#else
+#define gotoblas_OPTERON gotoblas_PRESCOTT
+#endif
+#ifdef DYN_OPTERON_SSE3
+extern gotoblas_t gotoblas_OPTERON_SSE3;
+#else
+#define gotoblas_OPTERON_SSE3 gotoblas_PRESCOTT
+#endif
+#ifdef DYN_BOBCAT
+extern gotoblas_t gotoblas_BOBCAT;
+#elif defined(DYN_NEHALEM)
+#define gotoblas_BOBCAT gotoblas_NEHALEM
+#else
+#define gotoblas_BOBCAT gotoblas_PRESCOTT
+#endif
+#ifdef DYN_SANDYBRIDGE
+extern gotoblas_t gotoblas_SANDYBRIDGE;
+#elif defined(DYN_NEHALEM)
+#define gotoblas_SANDYBRIDGE gotoblas_NEHALEM
+#else
+#define gotoblas_SANDYBRIDGE gotoblas_PRESCOTT
+#endif
+#ifdef DYN_BULLDOZER
+extern gotoblas_t gotoblas_BULLDOZER;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_BULLDOZER gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_BULLDOZER gotoblas_NEHALEM
+#else
+#define gotoblas_BULLDOZER gotoblas_PRESCOTT
+#endif
+#ifdef DYN_PILEDRIVER
+extern gotoblas_t gotoblas_PILEDRIVER;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_PILEDRIVER gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_PILEDRIVER gotoblas_NEHALEM
+#else
+#define gotoblas_PILEDRIVER gotoblas_PRESCOTT
+#endif
+#ifdef DYN_STEAMROLLER
+extern gotoblas_t gotoblas_STEAMROLLER;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_STEAMROLLER gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_STEAMROLLER gotoblas_NEHALEM
+#else
+#define gotoblas_STEAMROLLER gotoblas_PRESCOTT
+#endif
+#ifdef DYN_EXCAVATOR
+extern gotoblas_t gotoblas_EXCAVATOR;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_EXCAVATOR gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_EXCAVATOR gotoblas_NEHALEM
+#else
+#define gotoblas_EXCAVATOR gotoblas_PRESCOTT
+#endif
+#ifdef DYN_HASWELL
+extern gotoblas_t gotoblas_HASWELL;
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_HASWELL gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_HASWELL gotoblas_NEHALEM
+#else
+#define gotoblas_HASWELL gotoblas_PRESCOTT
+#endif
+#ifdef DYN_ZEN
+extern gotoblas_t gotoblas_ZEN;
+#elif defined(DYN_HASWELL)
+#define gotoblas_ZEN gotoblas_HASWELL
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_ZEN gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_ZEN gotoblas_NEHALEM
+#else
+#define gotoblas_ZEN gotoblas_PRESCOTT
+#endif
+#ifdef DYN_SKYLAKEX
+extern gotoblas_t gotoblas_SKYLAKEX;
+#elif defined(DYN_HASWELL)
+#define gotoblas_SKYLAKEX gotoblas_HASWELL
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_SKYLAKEX gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_SKYLAKEX gotoblas_NEHALEM
+#else
+#define gotoblas_SKYLAKEX gotoblas_PRESCOTT
+#endif
+
+
+#else // not DYNAMIC_LIST
 EXTERN gotoblas_t  gotoblas_KATMAI;
 EXTERN gotoblas_t  gotoblas_COPPERMINE;
 EXTERN gotoblas_t  gotoblas_NORTHWOOD;
@@ -56,16 +217,27 @@ EXTERN gotoblas_t  gotoblas_BANIAS;
 EXTERN gotoblas_t  gotoblas_ATHLON;
 
 extern gotoblas_t  gotoblas_PRESCOTT;
+extern gotoblas_t  gotoblas_CORE2;
+extern gotoblas_t  gotoblas_NEHALEM;
+extern gotoblas_t  gotoblas_BARCELONA;
+#ifdef DYNAMIC_OLDER
 extern gotoblas_t  gotoblas_ATOM;
 extern gotoblas_t  gotoblas_NANO;
-extern gotoblas_t  gotoblas_CORE2;
 extern gotoblas_t  gotoblas_PENRYN;
 extern gotoblas_t  gotoblas_DUNNINGTON;
-extern gotoblas_t  gotoblas_NEHALEM;
 extern gotoblas_t  gotoblas_OPTERON;
 extern gotoblas_t  gotoblas_OPTERON_SSE3;
-extern gotoblas_t  gotoblas_BARCELONA;
 extern gotoblas_t  gotoblas_BOBCAT;
+#else
+#define gotoblas_ATOM gotoblas_NEHALEM
+#define gotoblas_NANO gotoblas_NEHALEM
+#define gotoblas_PENRYN gotoblas_CORE2
+#define gotoblas_DUNNINGTON gotoblas_CORE2
+#define gotoblas_OPTERON gotoblas_CORE2
+#define gotoblas_OPTERON_SSE3 gotoblas_CORE2
+#define gotoblas_BOBCAT gotoblas_CORE2
+#endif
+
 #ifndef NO_AVX
 extern gotoblas_t  gotoblas_SANDYBRIDGE;
 extern gotoblas_t  gotoblas_BULLDOZER;
@@ -74,15 +246,22 @@ extern gotoblas_t  gotoblas_STEAMROLLER;
 extern gotoblas_t  gotoblas_EXCAVATOR;
 #ifdef NO_AVX2
 #define gotoblas_HASWELL gotoblas_SANDYBRIDGE
+#define gotoblas_SKYLAKEX gotoblas_SANDYBRIDGE
 #define gotoblas_ZEN gotoblas_SANDYBRIDGE
 #else
 extern gotoblas_t  gotoblas_HASWELL;
 extern gotoblas_t  gotoblas_ZEN;
+#ifndef NO_AVX512
+extern gotoblas_t  gotoblas_SKYLAKEX;
+#else
+#define gotoblas_SKYLAKEX gotoblas_HASWELL
+#endif
 #endif
 #else
 //Use NEHALEM kernels for sandy bridge
 #define gotoblas_SANDYBRIDGE gotoblas_NEHALEM
 #define gotoblas_HASWELL gotoblas_NEHALEM
+#define gotoblas_SKYLAKEX gotoblas_NEHALEM
 #define gotoblas_BULLDOZER gotoblas_BARCELONA
 #define gotoblas_PILEDRIVER gotoblas_BARCELONA
 #define gotoblas_STEAMROLLER gotoblas_BARCELONA
@@ -90,10 +269,12 @@ extern gotoblas_t  gotoblas_ZEN;
 #define gotoblas_ZEN gotoblas_BARCELONA
 #endif
 
+#endif // DYNAMIC_LIST
 
 #define VENDOR_INTEL      1
 #define VENDOR_AMD        2
 #define VENDOR_CENTAUR    3
+#define VENDOR_HYGON	  4
 #define VENDOR_UNKNOWN   99
 
 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
@@ -124,9 +305,49 @@ int support_avx(){
 #endif
 }
 
+int support_avx2(){
+#ifndef NO_AVX2
+  int eax, ebx, ecx=0, edx;
+  int ret=0;
+
+  if (!support_avx())
+    return 0;
+  cpuid(7, &eax, &ebx, &ecx, &edx);
+  if((ebx & (1<<7)) != 0)
+      ret=1;  //OS supports AVX2
+  return ret;
+#else
+  return 0;
+#endif
+}
+
+int support_avx512(){
+#ifndef NO_AVX512
+  int eax, ebx, ecx, edx;
+  int ret=0;
+
+  if (!support_avx())
+    return 0;
+  cpuid(7, &eax, &ebx, &ecx, &edx);
+  if((ebx & (1<<7)) != 1){
+      ret=0;  //OS does not even support AVX2
+  }
+  if((ebx & (1<<31)) != 0){
+    xgetbv(0, &eax, &edx);
+    if((eax & 0xe0) == 0xe0)
+      ret=1;  //OS supports AVX512VL
+  }
+  return ret;
+#else
+  return 0;
+#endif
+}
+
 extern void openblas_warning(int verbose, const char * msg);
 #define FALLBACK_VERBOSE 1
 #define NEHALEM_FALLBACK "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.\n"
+#define SANDYBRIDGE_FALLBACK "OpenBLAS : Your OS does not support AVX2 instructions. OpenBLAS is using Sandybridge kernels as a fallback, which may give poorer performance.\n"
+#define HASWELL_FALLBACK "OpenBLAS : Your OS does not support AVX512VL instructions. OpenBLAS is using Haswell kernels as a fallback, which may give poorer performance.\n"
 #define BARCELONA_FALLBACK "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Barcelona kernels as a fallback, which may give poorer performance.\n"
 
 static int get_vendor(void){
@@ -149,6 +370,7 @@ static int get_vendor(void){
   if (!strcmp(vendor.vchar, "GenuineIntel")) return VENDOR_INTEL;
   if (!strcmp(vendor.vchar, "AuthenticAMD")) return VENDOR_AMD;
   if (!strcmp(vendor.vchar, "CentaurHauls")) return VENDOR_CENTAUR;
+  if (!strcmp(vendor.vchar, "HygonGenuine")) return VENDOR_HYGON;
 
   if ((eax == 0) || ((eax & 0x500) != 0)) return VENDOR_INTEL;
 
@@ -223,18 +445,24 @@ static gotoblas_t *get_coretype(void){
 	}
 	//Intel Haswell
 	if (model == 12 || model == 15) {
-	  if(support_avx())
+	  if(support_avx2())
 	    return &gotoblas_HASWELL;
-	  else{
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
 	//Intel Broadwell
 	if (model == 13) {
-	  if(support_avx())
+	  if(support_avx2())
 	    return &gotoblas_HASWELL;
-	  else{
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
@@ -244,27 +472,36 @@ static gotoblas_t *get_coretype(void){
       case 4:
 		//Intel Haswell
 	if (model == 5 || model == 6) {
-	  if(support_avx())
+	  if(support_avx2())
 	    return &gotoblas_HASWELL;
-	  else{
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
 	//Intel Broadwell
 	if (model == 7 || model == 15) {
-	  if(support_avx())
+	  if(support_avx2())
 	    return &gotoblas_HASWELL;
-	  else{
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
 	//Intel Skylake
 	if (model == 14) {
-	  if(support_avx())
+	  if(support_avx2())
 	    return &gotoblas_HASWELL;
-	  else{
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
@@ -277,27 +514,54 @@ static gotoblas_t *get_coretype(void){
       case 5:
 	//Intel Broadwell
 	if (model == 6) {
-	  if(support_avx())
+	  if(support_avx2())
 	    return &gotoblas_HASWELL;
-	  else{
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
-	//Intel Skylake
+	if (model == 5) {	
-	if (model == 14 || model == 5) {
+	// Intel Skylake X
-	  if(support_avx())
+          if (support_avx512()) 
+	    return &gotoblas_SKYLAKEX;
+	  if(support_avx2()){
+	    openblas_warning(FALLBACK_VERBOSE, HASWELL_FALLBACK);
 	    return &gotoblas_HASWELL;
-	  else{
+          }
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+          openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+          return &gotoblas_NEHALEM;
+          }
+	}
+	//Intel Skylake
+	if (model == 14) {
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
 	//Intel Phi Knights Landing
 	if (model == 7) {
-	  if(support_avx())
+	  if(support_avx2()){
+	    openblas_warning(FALLBACK_VERBOSE, HASWELL_FALLBACK);
 	    return &gotoblas_HASWELL;
-	  else{
+	  }  
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
@@ -307,12 +571,29 @@ static gotoblas_t *get_coretype(void){
 	  return &gotoblas_NEHALEM;
 	}	
 	return NULL;
+      case 6:
+        if (model == 6) {
+          // Cannon Lake
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM;
+	  }
+        }
+        return NULL;  
       case 9:
       case 8:
 	if (model == 14 ) { // Kaby Lake
-	  if(support_avx())
+	  if(support_avx2())
 	    return &gotoblas_HASWELL;
-	  else{
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
 	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
@@ -325,7 +606,7 @@ static gotoblas_t *get_coretype(void){
     }
   }
 
-  if (vendor == VENDOR_AMD){
+  if (vendor == VENDOR_AMD || vendor == VENDOR_HYGON){
     if (family <= 0xe) {
         // Verify that CPU has 3dnow and 3dnowext before claiming it is Athlon
         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
@@ -397,7 +678,7 @@ static gotoblas_t *get_coretype(void){
 	  }
 	}
       } else if (exfamily == 8) {
-	if (model == 1) {
+	if (model == 1 || model == 8) {
 	  if(support_avx())
 	    return &gotoblas_ZEN;
 	  else{
@@ -405,6 +686,13 @@ static gotoblas_t *get_coretype(void){
 	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
+      } else if (exfamily == 9) {
+	  if(support_avx())
+	    return &gotoblas_ZEN;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
+        }
       }else {
 	return &gotoblas_BARCELONA;
       }
@@ -445,7 +733,8 @@ static char *corename[] = {
     "Haswell",
     "Steamroller",
     "Excavator",
-    "Zen"
+    "Zen",
+    "SkylakeX"	
 };
 
 char *gotoblas_corename(void) {
@@ -473,7 +762,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_STEAMROLLER)  return corename[21];
   if (gotoblas == &gotoblas_EXCAVATOR)    return corename[22];
   if (gotoblas == &gotoblas_ZEN)          return corename[23];
-
+  if (gotoblas == &gotoblas_SKYLAKEX)     return corename[24];
   return corename[0];
 }
 
@@ -485,7 +774,7 @@ static gotoblas_t *force_coretype(char *coretype){
 	char message[128];
 	//char mname[20];
 
-	for ( i=1 ; i <= 23; i++)
+	for ( i=1 ; i <= 24; i++)
 	{
 		if (!strncasecmp(coretype,corename[i],20))
 		{
@@ -503,6 +792,7 @@ static gotoblas_t *force_coretype(char *coretype){
 
 	switch (found)
 	{
+		case 24: return (&gotoblas_SKYLAKEX);	
 		case 23: return (&gotoblas_ZEN);
 		case 22: return (&gotoblas_EXCAVATOR);
 		case 21: return (&gotoblas_STEAMROLLER);

driver/others/dynamic_arm64.c

@@ -0,0 +1,198 @@
+/*********************************************************************/
+/* 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.                 */
+/*********************************************************************/
+
+#include "common.h"
+#include <asm/hwcap.h>
+#include <sys/auxv.h>
+
+extern gotoblas_t  gotoblas_ARMV8;
+extern gotoblas_t  gotoblas_CORTEXA57;
+extern gotoblas_t  gotoblas_THUNDERX;
+extern gotoblas_t  gotoblas_THUNDERX2T99;
+
+extern void openblas_warning(int verbose, const char * msg);
+
+#define NUM_CORETYPES    4
+
+/*
+ * In case asm/hwcap.h is outdated on the build system, make sure
+ * that HWCAP_CPUID is defined 
+ */
+#ifndef HWCAP_CPUID
+#define HWCAP_CPUID (1 << 11)
+#endif
+
+#define get_cpu_ftr(id, var) ({					\
+		asm("mrs %0, "#id : "=r" (var));		\
+	})
+
+static char *corename[] = {
+  "armv8",
+  "cortexa57",
+  "thunderx",
+  "thunderx2t99",
+  "unknown"
+};
+
+char *gotoblas_corename(void) {
+  if (gotoblas == &gotoblas_ARMV8)        return corename[ 0];
+  if (gotoblas == &gotoblas_CORTEXA57)    return corename[ 1];
+  if (gotoblas == &gotoblas_THUNDERX)     return corename[ 2];
+  if (gotoblas == &gotoblas_THUNDERX2T99) return corename[ 3];
+  return corename[NUM_CORETYPES];
+}
+
+static gotoblas_t *force_coretype(char *coretype) {
+  int i ;
+  int found = -1;
+  char message[128];
+
+  for ( i=0 ; i < NUM_CORETYPES; i++)
+  {
+    if (!strncasecmp(coretype, corename[i], 20))
+    {
+        found = i;
+        break;
+    }
+  }
+
+  switch (found)
+  {
+    case  0: return (&gotoblas_ARMV8);
+    case  1: return (&gotoblas_CORTEXA57);
+    case  2: return (&gotoblas_THUNDERX);
+    case  3: return (&gotoblas_THUNDERX2T99);
+  }
+  snprintf(message, 128, "Core not found: %s\n", coretype);
+  openblas_warning(1, message);
+  return NULL;
+}
+
+static gotoblas_t *get_coretype(void) {
+  int implementer, variant, part, arch, revision, midr_el1;
+  
+  if (!(getauxval(AT_HWCAP) & HWCAP_CPUID)) {
+    char coremsg[128];
+    snprintf(coremsg, 128, "Kernel lacks cpuid feature support. Auto detection of core type failed !!!\n");
+    openblas_warning(1, coremsg);
+    return NULL;
+  }
+
+  get_cpu_ftr(MIDR_EL1, midr_el1);
+  /*
+   * MIDR_EL1
+   *
+   * 31          24 23     20 19          16 15          4 3        0
+   * -----------------------------------------------------------------
+   * | Implementer | Variant | Architecture | Part Number | Revision |
+   * -----------------------------------------------------------------
+   */
+  implementer = (midr_el1 >> 24) & 0xFF;
+  part        = (midr_el1 >> 4)  & 0xFFF;
+
+  switch(implementer)
+  {
+    case 0x41: // ARM
+      switch (part)
+      {
+        case 0xd07: // Cortex A57
+        case 0xd08: // Cortex A72
+        case 0xd03: // Cortex A53
+          return &gotoblas_CORTEXA57;
+      }
+      break;
+    case 0x42: // Broadcom
+      switch (part)
+      {
+        case 0x516: // Vulcan
+          return &gotoblas_THUNDERX2T99;
+      }
+      break;
+    case 0x43: // Cavium
+      switch (part)
+      {
+        case 0x0a1: // ThunderX
+          return &gotoblas_THUNDERX;
+        case 0x0af: // ThunderX2
+          return &gotoblas_THUNDERX2T99;
+      }
+      break;
+  }
+  return NULL;
+}
+
+void gotoblas_dynamic_init(void) {
+
+  char coremsg[128];
+  char coren[22];
+  char *p;
+
+  if (gotoblas) return;
+
+  p = getenv("OPENBLAS_CORETYPE");
+  if ( p )
+  {
+    gotoblas = force_coretype(p);
+  }
+  else
+  {
+    gotoblas = get_coretype();
+  }
+
+  if (gotoblas == NULL)
+  {
+    snprintf(coremsg, 128, "Falling back to generic ARMV8 core\n");
+    openblas_warning(1, coremsg);
+    gotoblas = &gotoblas_ARMV8;
+  }
+
+  if (gotoblas && gotoblas->init) {
+    strncpy(coren, gotoblas_corename(), 20);
+    sprintf(coremsg, "Core: %s\n", coren);
+    openblas_warning(2, coremsg);
+    gotoblas -> init();
+  } else {
+    openblas_warning(0, "OpenBLAS : Architecture Initialization failed. No initialization function found.\n");
+    exit(1);
+  }
+
+}
+
+void gotoblas_dynamic_quit(void) {
+  gotoblas = NULL;
+}

driver/others/openblas_get_config.c

@@ -35,9 +35,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include <string.h>
 
+#if defined(_WIN32) && defined(_MSC_VER)
+#if _MSC_VER < 1900
+#define snprintf _snprintf
+#endif
+#endif
+
 static char* openblas_config_str=""
+"OpenBLAS "
+ VERSION
+" "
 #ifdef USE64BITINT
-  "USE64BITINT "
+  " USE64BITINT "
 #endif
 #ifdef NO_CBLAS
   "NO_CBLAS "
@@ -54,6 +63,9 @@ static char* openblas_config_str=""
 #ifdef NO_AFFINITY
   "NO_AFFINITY "
 #endif
+#ifdef USE_OPENMP
+  "USE_OPENMP "
+#endif
 #ifndef DYNAMIC_ARCH
   CHAR_CORENAME
 #endif
@@ -61,18 +73,23 @@ static char* openblas_config_str=""
 
 #ifdef DYNAMIC_ARCH
 char *gotoblas_corename();
-static char tmp_config_str[256];
 #endif
 
+static char tmp_config_str[256];
+int openblas_get_parallel();
 
 char* CNAME() {
-#ifndef DYNAMIC_ARCH
+char tmpstr[20];
-  return openblas_config_str;
-#else
   strcpy(tmp_config_str, openblas_config_str);
+#ifdef DYNAMIC_ARCH
   strcat(tmp_config_str, gotoblas_corename());
-  return tmp_config_str;
 #endif
+if (openblas_get_parallel() == 0)
+  sprintf(tmpstr, " SINGLE_THREADED");
+else 
+  snprintf(tmpstr,19," MAX_THREADS=%d",MAX_CPU_NUMBER);
+  strcat(tmp_config_str, tmpstr);
+  return tmp_config_str;
 }
 
 
@@ -83,3 +100,4 @@ char* openblas_get_corename() {
   return gotoblas_corename();
 #endif
 }
+

driver/others/parameter.c

@@ -167,7 +167,7 @@ int get_L2_size(void){
 #if defined(ATHLON) || defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT) || defined(BULLDOZER) || \
     defined(CORE_PRESCOTT) || defined(CORE_CORE2) || defined(PENRYN) || defined(DUNNINGTON) || \
     defined(CORE_NEHALEM) || defined(CORE_SANDYBRIDGE) || defined(ATOM) || defined(GENERIC) || \
-    defined(PILEDRIVER) || defined(HASWELL) || defined(STEAMROLLER) || defined(EXCAVATOR) || defined(ZEN)
+    defined(PILEDRIVER) || defined(HASWELL) || defined(STEAMROLLER) || defined(EXCAVATOR) || defined(ZEN) || defined(SKYLAKEX)
 
   cpuid(0x80000006, &eax, &ebx, &ecx, &edx);
 
@@ -251,7 +251,7 @@ int get_L2_size(void){
 void blas_set_parameter(void){
 
   int factor;
-#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(SANDYBRIDGE) || defined(NEHALEM) || defined(HASWELL) || defined(STEAMROLLER) || defined(EXCAVATOR) || defined(ZEN)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(SANDYBRIDGE) || defined(NEHALEM) || defined(HASWELL) || defined(STEAMROLLER) || defined(EXCAVATOR) || defined(ZEN) || defined(SKYLAKEX)
   int size = 16;
 #else
   int size = get_L2_size();
@@ -730,35 +730,8 @@ void blas_set_parameter(void){
 
 #if defined(ARCH_ARM64)
 
-#if defined(VULCAN) || defined(THUNDERX2T99)
-unsigned long dgemm_prefetch_size_a;
-unsigned long dgemm_prefetch_size_b;
-unsigned long dgemm_prefetch_size_c;
-#endif
-
 void blas_set_parameter(void)
 {
-#if defined(VULCAN) || defined(THUNDERX2T99)
-  dgemm_p = 160;
-  dgemm_q = 128;
-  dgemm_r = 4096;
-
-  sgemm_p = 128;
-  sgemm_q = 352;
-  sgemm_r = 4096;
-
-  cgemm_p = 128;
-  cgemm_q = 224;
-  cgemm_r = 4096;
-
-  zgemm_p = 128;
-  zgemm_q = 112;
-  zgemm_r = 4096;
-
-  dgemm_prefetch_size_a = 3584;
-  dgemm_prefetch_size_b = 512;
-  dgemm_prefetch_size_c = 128;
-#endif
 }
 
 #endif

exports/Makefile

@@ -114,20 +114,22 @@ $(LIBDYNNAME) : ../$(LIBNAME).osx.renamed osx.def
 endif
 ifneq (,$(filter 1 2,$(NOFORTRAN)))
 #only build without Fortran
-	$(CC) $(CFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(LIBDYNNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+	$(CC) $(CFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(LIBDYNNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 else
-	$(FC) $(FFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(LIBDYNNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+	$(FC) $(FFLAGS) $(LDFLAGS) -all_load -headerpad_max_install_names -install_name "$(CURDIR)/../$(LIBDYNNAME)" -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 endif
 
 dllinit.$(SUFFIX) : dllinit.c
 	$(CC) $(CFLAGS) -c -o $(@F) -s $<
 
-ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android))
+ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku))
 
 so : ../$(LIBSONAME)
 
 ifeq ($(OSNAME), Android)
 INTERNALNAME = $(LIBPREFIX).so
+FEXTRALIB += -lm
+EXTRALIB += -lm
 else
 INTERNALNAME = $(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
@@ -156,7 +158,7 @@ endif
 endif
 
 #http://stackoverflow.com/questions/7656425/makefile-ifeq-logical-or
-ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD NetBSD))
+ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
 
 so : ../$(LIBSONAME)
 

f_check

@@ -97,7 +97,7 @@ if ($compiler eq "") {
 
 	if ($data =~ /Intel/) {
 	    $vendor = INTEL;
-	    $openmp = "-openmp";
+	    $openmp = "-fopenmp";
 	}
 
         if ($data =~ /Sun Fortran/) {
@@ -127,7 +127,7 @@ if ($compiler eq "") {
 
 	# for embeded underscore name, e.g. zho_ge, it may append 2 underscores.
 	$data = `$compiler -O2 -S ftest3.f > /dev/null 2>&1 && cat ftest3.s && rm -f ftest3.s`;
-	if ($data =~ /zho_ge__/) {
+	if ($data =~ / zho_ge__/) {
 	    $need2bu       = 1;
 	}
     }
@@ -155,7 +155,7 @@ if ($compiler eq "") {
 	if ($compiler =~ /ifort/) {
 	    $vendor = INTEL;
 	    $bu       = "_";
-	    $openmp = "-openmp";
+	    $openmp = "-fopenmp";
 	}
 
 	if ($compiler =~ /pathf/) {
@@ -292,9 +292,6 @@ if ($link ne "") {
 	    && ($flags !~ /^-LIST:/)
 	    && ($flags !~ /^-LANG:/)
 	    ) {
-	    if ($vendor eq "PGI") {
-		$flags =~ s/lib$/libso/;
-	    }
 	    $linker_L .= $flags . " ";
 	}
 
@@ -311,17 +308,11 @@ if ($link ne "") {
 
 	if ($flags =~ /^\-rpath\@/) {
 	    $flags =~ s/\@/\,/g;
-	    if ($vendor eq "PGI") {
-		$flags =~ s/lib$/libso/;
-	    }
 	    $linker_L .= "-Wl,". $flags . " " ;
 	}
 
 	if ($flags =~ /^\-rpath-link\@/) {
 	    $flags =~ s/\@/\,/g;
-	    if ($vendor eq "PGI") {
-		$flags =~ s/lib$/libso/;
-	    }
 	    $linker_L .= "-Wl,". $flags . " " ;
 	}
 
@@ -330,7 +321,6 @@ if ($link ne "") {
 	    && ($flags !~ /gfortranbegin/)
 	    && ($flags !~ /frtbegin/)
 	    && ($flags !~ /pathfstart/)
-	    && ($flags !~ /numa/)
 	    && ($flags !~ /crt[0-9]/)
 	    && ($flags !~ /gcc/)
 	    && ($flags !~ /user32/)

getarch.c

@@ -82,7 +82,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef OS_WINDOWS
 #include <windows.h>
 #endif
-#if defined(__FreeBSD__) || defined(__APPLE__)
+#if defined(__FreeBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__APPLE__)
 #include <sys/types.h>
 #include <sys/sysctl.h>
 #endif
@@ -326,6 +326,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "HASWELL"
 #endif
 
+#ifdef FORCE_SKYLAKEX
+#define FORCE
+#define FORCE_INTEL
+#define ARCHITECTURE    "X86"
+#define SUBARCHITECTURE "SKYLAKEX"
+#define ARCHCONFIG   "-DSKYLAKEX " \
+		     "-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 " \
+                     "-DFMA3 -DHAVE_AVX512VL -march=skylake-avx512"
+#define LIBNAME   "skylakex"
+#define CORENAME  "SKYLAKEX"
+#endif
+
 #ifdef FORCE_ATOM
 #define FORCE
 #define FORCE_INTEL
@@ -912,11 +927,28 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ARCHCONFIG   "-DARMV8 " \
        "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
        "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
-       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=32 " 
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=32 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
 #define LIBNAME   "armv8"
 #define CORENAME  "ARMV8"
 #endif
 
+#ifdef FORCE_CORTEXA53
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "CORTEXA53"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DCORTEXA53 " \
+       "-DL1_CODE_SIZE=32768 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=3 " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=2 " \
+       "-DL2_SIZE=262144 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
+#define LIBNAME   "cortexa53"
+#define CORENAME  "CORTEXA53"
+#else
+#endif
+
 #ifdef FORCE_CORTEXA57
 #define FORCE
 #define ARCHITECTURE    "ARM64"
@@ -927,26 +959,57 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
        "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=2 " \
        "-DL2_SIZE=2097152 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
        "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
-       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON"
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
 #define LIBNAME   "cortexa57"
 #define CORENAME  "CORTEXA57"
 #else
 #endif
 
-#ifdef FORCE_VULCAN
+#ifdef FORCE_CORTEXA72
 #define FORCE
 #define ARCHITECTURE    "ARM64"
-#define SUBARCHITECTURE "VULCAN"
+#define SUBARCHITECTURE "CORTEXA72"
 #define SUBDIRNAME      "arm64"
-#define ARCHCONFIG   "-DVULCAN " \
+#define ARCHCONFIG   "-DCORTEXA72 " \
-       "-DL1_CODE_SIZE=32768 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=8 " \
+       "-DL1_CODE_SIZE=49152 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=3 " \
-       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=8 " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=2 " \
-       "-DL2_SIZE=262144 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=8 " \
+       "-DL2_SIZE=2097152 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
-       "-DL3_SIZE=33554432 -DL3_LINESIZE=64 -DL3_ASSOCIATIVE=32 " \
        "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
-       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON"
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
-#define LIBNAME   "vulcan"
+#define LIBNAME   "cortexa72"
-#define CORENAME  "VULCAN"
+#define CORENAME  "CORTEXA72"
+#else
+#endif
+
+#ifdef FORCE_CORTEXA73
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "CORTEXA73"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DCORTEXA73 " \
+       "-DL1_CODE_SIZE=49152 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=3 " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=2 " \
+       "-DL2_SIZE=2097152 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
+#define LIBNAME   "cortexa73"
+#define CORENAME  "CORTEXA73"
+#else
+#endif
+
+#ifdef FORCE_FALKOR
+#define FORCE
+#define ARCHITECTURE    "ARM64"
+#define SUBARCHITECTURE "FALKOR"
+#define SUBDIRNAME      "arm64"
+#define ARCHCONFIG   "-DFALKOR " \
+       "-DL1_CODE_SIZE=49152 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=3 " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=2 " \
+       "-DL2_SIZE=2097152 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
+#define LIBNAME   "falkor"
+#define CORENAME  "FALKOR"
 #else
 #endif
 
@@ -958,13 +1021,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ARCHCONFIG   "-DTHUNDERX " \
        "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=128 " \
        "-DL2_SIZE=16777216 -DL2_LINESIZE=128 -DL2_ASSOCIATIVE=16 " \
-       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 "
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
 #define LIBNAME   "thunderx"
 #define CORENAME  "THUNDERX"
 #else
 #endif
 
 #ifdef FORCE_THUNDERX2T99
+#define ARMV8
 #define FORCE
 #define ARCHITECTURE    "ARM64"
 #define SUBARCHITECTURE "THUNDERX2T99"
@@ -975,7 +1040,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
        "-DL2_SIZE=262144 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=8 " \
        "-DL3_SIZE=33554432 -DL3_LINESIZE=64 -DL3_ASSOCIATIVE=32 " \
        "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
-       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON"
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
 #define LIBNAME   "thunderx2t99"
 #define CORENAME  "THUNDERX2T99"
 #else
@@ -1003,6 +1068,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #ifndef FORCE
 
+#ifdef USER_TARGET
+#error "The TARGET specified on the command line or in Makefile.rule is not supported. Please choose a target from TargetList.txt"
+#endif
+
 #if defined(__powerpc__) || defined(__powerpc) || defined(powerpc) || \
     defined(__PPC__) || defined(PPC) || defined(_POWER) || defined(__POWERPC__)
 #ifndef POWER
@@ -1074,7 +1143,7 @@ static int get_num_cores(void) {
 
 #ifdef OS_WINDOWS
   SYSTEM_INFO sysinfo;
-#elif defined(__FreeBSD__) || defined(__APPLE__)
+#elif defined(__FreeBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__APPLE__)
   int m[2], count;
   size_t len;
 #endif
@@ -1088,7 +1157,7 @@ static int get_num_cores(void) {
   GetSystemInfo(&sysinfo);
   return sysinfo.dwNumberOfProcessors;
 
-#elif defined(__FreeBSD__) || defined(__APPLE__)
+#elif defined(__FreeBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__APPLE__)
   m[0] = CTL_HW;
   m[1] = HW_NCPU;
   len = sizeof(int);
@@ -1116,7 +1185,7 @@ 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)
+#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc)
     printf("CORE=%s\n", get_corename());
 #endif
 #endif
@@ -1181,9 +1250,7 @@ int main(int argc, char *argv[]){
 #elif NO_PARALLEL_MAKE==1
     printf("MAKE += -j 1\n");
 #else
-#ifndef OS_WINDOWS
     printf("MAKE += -j %d\n", get_num_cores());
-#endif
 #endif
 
     break;
@@ -1224,7 +1291,7 @@ int main(int argc, char *argv[]){
 #ifdef FORCE
     printf("#define CHAR_CORENAME \"%s\"\n", CORENAME);
 #else
-#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH)
+#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__) || defined(ZARCH) || defined(sparc)
     printf("#define CHAR_CORENAME \"%s\"\n", get_corename());
 #endif
 #endif

interface/Makefile

@@ -260,7 +260,7 @@ HPLOBJS = dgemm.$(SUFFIX) dtrsm.$(SUFFIX) \
 	  idamax.$(SUFFIX) daxpy.$(SUFFIX) dcopy.$(SUFFIX) dscal.$(SUFFIX)
 
 CSBLAS1OBJS   = \
-	cblas_isamax.$(SUFFIX) cblas_sasum.$(SUFFIX) cblas_saxpy.$(SUFFIX) \
+	cblas_isamax.$(SUFFIX) cblas_isamin.$(SUFFIX) cblas_sasum.$(SUFFIX) cblas_saxpy.$(SUFFIX) \
 	cblas_scopy.$(SUFFIX) cblas_sdot.$(SUFFIX) cblas_sdsdot.$(SUFFIX) cblas_dsdot.$(SUFFIX) \
 	cblas_srot.$(SUFFIX) cblas_srotg.$(SUFFIX) cblas_srotm.$(SUFFIX) cblas_srotmg.$(SUFFIX) \
 	cblas_sscal.$(SUFFIX) cblas_sswap.$(SUFFIX) cblas_snrm2.$(SUFFIX) cblas_saxpby.$(SUFFIX)
@@ -277,7 +277,7 @@ CSBLAS3OBJS   = \
 	cblas_sgeadd.$(SUFFIX)
 
 CDBLAS1OBJS   = \
-	cblas_idamax.$(SUFFIX) cblas_dasum.$(SUFFIX) cblas_daxpy.$(SUFFIX) \
+	cblas_idamax.$(SUFFIX) cblas_idamin.$(SUFFIX) cblas_dasum.$(SUFFIX) cblas_daxpy.$(SUFFIX) \
 	cblas_dcopy.$(SUFFIX) cblas_ddot.$(SUFFIX) \
 	cblas_drot.$(SUFFIX) cblas_drotg.$(SUFFIX) cblas_drotm.$(SUFFIX) cblas_drotmg.$(SUFFIX) \
 	cblas_dscal.$(SUFFIX) cblas_dswap.$(SUFFIX) cblas_dnrm2.$(SUFFIX) cblas_daxpby.$(SUFFIX)
@@ -294,7 +294,7 @@ CDBLAS3OBJS   += \
         cblas_dgeadd.$(SUFFIX) 
 
 CCBLAS1OBJS   = \
-	cblas_icamax.$(SUFFIX) cblas_scasum.$(SUFFIX)  cblas_caxpy.$(SUFFIX) \
+	cblas_icamax.$(SUFFIX) cblas_icamin.$(SUFFIX) cblas_scasum.$(SUFFIX)  cblas_caxpy.$(SUFFIX) \
 	cblas_ccopy.$(SUFFIX) \
 	cblas_cdotc.$(SUFFIX) cblas_cdotu.$(SUFFIX) \
 	cblas_cdotc_sub.$(SUFFIX) cblas_cdotu_sub.$(SUFFIX) \
@@ -320,7 +320,7 @@ CCBLAS3OBJS   = \
 
 
 CZBLAS1OBJS   = \
-	cblas_izamax.$(SUFFIX) cblas_dzasum.$(SUFFIX)  cblas_zaxpy.$(SUFFIX) \
+	cblas_izamax.$(SUFFIX) cblas_izamin.$(SUFFIX) cblas_dzasum.$(SUFFIX)  cblas_zaxpy.$(SUFFIX) \
 	cblas_zcopy.$(SUFFIX) \
 	cblas_zdotc.$(SUFFIX) cblas_zdotu.$(SUFFIX) \
 	cblas_zdotc_sub.$(SUFFIX) cblas_zdotu_sub.$(SUFFIX) \
@@ -1359,6 +1359,18 @@ cblas_icamax.$(SUFFIX) cblas_icamax.$(PSUFFIX) : imax.c
 cblas_izamax.$(SUFFIX) cblas_izamax.$(PSUFFIX) : imax.c
 	$(CC) $(CFLAGS) -DCBLAS -c -DUSE_ABS -UUSE_MIN $< -o $(@F)
 
+cblas_isamin.$(SUFFIX) cblas_isamin.$(PSUFFIX) : imax.c
+	$(CC) $(CFLAGS) -DCBLAS -c -DUSE_ABS -DUSE_MIN $< -o $(@F)
+
+cblas_idamin.$(SUFFIX) cblas_idamin.$(PSUFFIX) : imax.c
+	$(CC) $(CFLAGS) -DCBLAS -c -DUSE_ABS -DUSE_MIN $< -o $(@F)
+
+cblas_icamin.$(SUFFIX) cblas_icamin.$(PSUFFIX) : imax.c
+	$(CC) $(CFLAGS) -DCBLAS -c -DUSE_ABS -DUSE_MIN $< -o $(@F)
+
+cblas_izamin.$(SUFFIX) cblas_izamin.$(PSUFFIX) : imax.c
+	$(CC) $(CFLAGS) -DCBLAS -c -DUSE_ABS -DUSE_MIN $< -o $(@F)
+
 cblas_ismax.$(SUFFIX) cblas_ismax.$(PSUFFIX) : imax.c
 	$(CC) $(CFLAGS) -DCBLAS -c -UUSE_ABS -UUSE_MIN $< -o $(@F)
 

interface/axpy.c

@@ -75,6 +75,11 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx, FLOAT *y, blasint inc
 
   if (alpha == ZERO) return;
 
+  if (incx == 0 && incy == 0) {
+    *y += n * alpha *(*x);
+    return;
+  }
+    
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
@@ -83,17 +88,15 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx, FLOAT *y, blasint inc
   if (incy < 0) y -= (n - 1) * incy;
 
 #ifdef SMP
-  nthreads = num_cpu_avail(1);
-
   //disable multi-thread when incx==0 or incy==0
   //In that case, the threads would be dependent.
-  if (incx == 0 || incy == 0)
+  //
-	  nthreads = 1;
-
   //Temporarily work-around the low performance issue with small imput size &
   //multithreads.
-  if (n <= MULTI_THREAD_MINIMAL)
+  if (incx == 0 || incy == 0 || n <= MULTI_THREAD_MINIMAL)
 	  nthreads = 1;
+  else
+	  nthreads = num_cpu_avail(1);
 
   if (nthreads == 1) {
 #endif

interface/gbmv.c

@@ -213,7 +213,7 @@ void CNAME(enum CBLAS_ORDER order,
   if (trans) lenx = m;
   if (trans) leny = n;
 
-  if (beta != ONE) SCAL_K(leny, 0, 0, beta, y, abs(incy), NULL, 0, NULL, 0);
+  if (beta != ONE) SCAL_K(leny, 0, 0, beta, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if (alpha == ZERO) return;
 

interface/gemm.c

@@ -44,6 +44,7 @@
 #endif
 
 #ifndef COMPLEX
+#define SMP_THRESHOLD_MIN 65536.0
 #ifdef XDOUBLE
 #define ERROR_NAME "QGEMM "
 #elif defined(DOUBLE)
@@ -52,6 +53,7 @@
 #define ERROR_NAME "SGEMM "
 #endif
 #else
+#define SMP_THRESHOLD_MIN 8192.0
 #ifndef GEMM3M
 #ifdef XDOUBLE
 #define ERROR_NAME "XGEMM "
@@ -121,8 +123,6 @@ void NAME(char *TRANSA, char *TRANSB,
   FLOAT *sa, *sb;
 
 #ifdef SMP
-  int nthreads_max;
-  int nthreads_avail;
   double MNK;
 #ifndef COMPLEX
 #ifdef XDOUBLE
@@ -245,8 +245,6 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
   XFLOAT *sa, *sb;
 
 #ifdef SMP
-  int nthreads_max;
-  int nthreads_avail;
   double MNK;
 #ifndef COMPLEX
 #ifdef XDOUBLE
@@ -273,6 +271,14 @@ 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 (beta == 0 && alpha == 1.0 && order == CblasRowMajor && TransA == CblasNoTrans && TransB == CblasNoTrans && sgemm_kernel_direct_performant(m,n,k)) {
+	sgemm_kernel_direct(m, n, k, a, lda, b, ldb, c, ldc);
+	return;
+  }
+
+#endif
+
 #ifndef COMPLEX
   args.alpha = (void *)α
   args.beta  = (void *)β
@@ -411,25 +417,12 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
   mode |= (transa << BLAS_TRANSA_SHIFT);
   mode |= (transb << BLAS_TRANSB_SHIFT);
 
-  nthreads_max = num_cpu_avail(3);
-  nthreads_avail = nthreads_max;
-
-#ifndef COMPLEX
   MNK = (double) args.m * (double) args.n * (double) args.k;
-  if ( MNK <= (65536.0  * (double) GEMM_MULTITHREAD_THRESHOLD)  )
+  if ( MNK <= (SMP_THRESHOLD_MIN  * (double) GEMM_MULTITHREAD_THRESHOLD)  )
-	nthreads_max = 1;
+	args.nthreads = 1;
-#else
-  MNK = (double) args.m * (double) args.n * (double) args.k;
-  if ( MNK <= (8192.0  * (double) GEMM_MULTITHREAD_THRESHOLD)  )
-	nthreads_max = 1;
-#endif
-  args.common = NULL;
-
-  if ( nthreads_max > nthreads_avail )
-  	args.nthreads = nthreads_avail;
   else
-  	args.nthreads = nthreads_max;
+	args.nthreads = num_cpu_avail(3);
-
+  args.common = NULL;
 
  if (args.nthreads == 1) {
 #endif

interface/gemv.c

@@ -199,7 +199,7 @@ void CNAME(enum CBLAS_ORDER order,
   if (trans) lenx = m;
   if (trans) leny = n;
 
-  if (beta != ONE) SCAL_K(leny, 0, 0, beta, y, abs(incy), NULL, 0, NULL, 0);
+  if (beta != ONE) SCAL_K(leny, 0, 0, beta, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if (alpha == ZERO) return;
 

interface/lapack/laswp.c

@@ -97,7 +97,7 @@ int NAME(blasint *N, FLOAT *a, blasint *LDA, blasint *K1, blasint *K2, blasint *
 
   blas_level1_thread(mode, n, k1, k2, dummyalpha,
 		     a, lda, NULL, 0, ipiv, incx,
-		     laswp[flag], nthreads);
+		     (int(*)())laswp[flag], nthreads);
   }
 #endif
 

interface/lapack/zlaswp.c

@@ -96,7 +96,7 @@ int NAME(blasint *N, FLOAT *a, blasint *LDA, blasint *K1, blasint *K2, blasint *
   mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
 #endif
 
-  blas_level1_thread(mode, n, k1, k2, dummyalpha, a, lda, NULL, 0, ipiv, incx, laswp[flag], nthreads);
+  blas_level1_thread(mode, n, k1, k2, dummyalpha, a, lda, NULL, 0, ipiv, incx, (int(*)())laswp[flag], nthreads);
   }
 #endif
 

interface/rotg.c

@@ -22,8 +22,8 @@ void CNAME(FLOAT *DA, FLOAT *DB, FLOAT *C, FLOAT *S){
   long double s;
   long double r, roe, z;
 
-  long double ada = fabs(da);
+  long double ada = fabsl(da);
-  long double adb = fabs(db);
+  long double adb = fabsl(db);
   long double scale = ada + adb;
 
 #ifndef CBLAS

interface/rotmg.c

@@ -64,6 +64,13 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
 
 	FLOAT du, dp1, dp2, dq2, dq1, dh11=ZERO, dh21=ZERO, dh12=ZERO, dh22=ZERO, dflag=-ONE, dtemp;
 
+	if (*dd2 == ZERO || dy1 == ZERO)
+	{
+		dflag = -TWO;
+		dparam[0] = dflag;
+		return;
+	}
+		
 	if(*dd1 < ZERO)
 	{
 		dflag = -ONE;
@@ -76,6 +83,16 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
 		*dd2  = ZERO;
 		*dx1  = ZERO;
 	}
+	else if ((*dd1 == ZERO || *dx1 == ZERO) && *dd2 > ZERO)
+	{
+		dflag = ONE;
+		dh12 = 1;
+		dh21 = -1;
+		*dx1 = dy1;
+		dtemp = *dd1;
+		*dd1 = *dd2;
+		*dd2 = dtemp;
+	} 
 	else
 	{
 		dp2 = *dd2 * dy1;
@@ -90,6 +107,9 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
 		dq1 =  dp1 * *dx1;
 		if(ABS(dq1) > ABS(dq2))
 		{
+			dflag = ZERO;
+			dh11  =  ONE;
+			dh22  =  ONE;
 			dh21 = -  dy1 / *dx1;
 			dh12 =    dp2 /  dp1;
 
@@ -100,8 +120,19 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
 				*dd1  = *dd1 / du;
 				*dd2  = *dd2 / du;
 				*dx1  = *dx1 * du;
+			} else {
+				dflag = -ONE;
 
+				dh11  = ZERO;
+				dh12  = ZERO;
+				dh21  = ZERO;
+				dh22  = ZERO;
+
+				*dd1  = ZERO;
+				*dd2  = ZERO;
+				*dx1  = ZERO;
 			}
+			
 		}
 		else
 		{
@@ -120,7 +151,9 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
 			}
 			else
 			{
-				dflag = ONE;
+				dflag =  ONE;
+				dh21  = -ONE;
+				dh12  =  ONE;
 
 				dh11  =  dp1 /  dp2;
 				dh22  = *dx1 /  dy1;
@@ -134,74 +167,33 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
 		}
 
 
-		if(*dd1 != ZERO)
+		while ( *dd1 <= RGAMSQ && *dd1 != ZERO)
 		{
-			while( (*dd1 <= RGAMSQ) || (*dd1 >= GAMSQ) )
+			dflag = -ONE;
-			{
+			*dd1  = *dd1 * (GAM * GAM);
-				if(dflag == ZERO)
+			*dx1  = *dx1 / GAM;
-				{
+			dh11  = dh11 / GAM;
-					dh11  =  ONE;
+			dh12  = dh12 / GAM;
-					dh22  =  ONE;
+		}
-					dflag = -ONE;
+		while (ABS(*dd1) > GAMSQ) {
-				}
+			dflag = -ONE;
-				else
+			*dd1  = *dd1 / (GAM * GAM);
-				{
+			*dx1  = *dx1 * GAM;
-					if(dflag == ONE)
+			dh11  = dh11 * GAM;
-					{
+			dh12  = dh12 * GAM;
-						dh21  = -ONE;
-						dh12  =  ONE;
-						dflag = -ONE;
-					}
-				}
-				if( *dd1 <= RGAMSQ )
-				{
-					*dd1  = *dd1 * (GAM * GAM);
-					*dx1  = *dx1 / GAM;
-					dh11  = dh11 / GAM;
-					dh12  = dh12 / GAM;
-				}
-				else
-				{
-					*dd1  = *dd1 / (GAM * GAM);
-					*dx1  = *dx1 * GAM;
-					dh11  = dh11 * GAM;
-					dh12  = dh12 * GAM;
-				}
-			}
 		}
 
-		if(*dd2 != ZERO)
+		while (ABS(*dd2) <= RGAMSQ && *dd2 != ZERO) {
-		{
+			dflag = -ONE;
-			while( (ABS(*dd2) <= RGAMSQ) || (ABS(*dd2) >= GAMSQ) )
+			*dd2  = *dd2 * (GAM * GAM);
-			{
+			dh21  = dh21 / GAM;
-				if(dflag == ZERO)
+			dh22  = dh22 / GAM;
-				{
+		}
-					dh11  =  ONE;
+		while (ABS(*dd2) > GAMSQ) {
-					dh22  =  ONE;
+			dflag = -ONE;
-					dflag = -ONE;
+			*dd2  = *dd2 / (GAM * GAM);
-				}
+			dh21  = dh21 * GAM;
-				else
+			dh22  = dh22 * GAM;
-				{
-					if(dflag == ONE)
-					{
-						dh21  = -ONE;
-						dh12  =  ONE;
-						dflag = -ONE;
-					}
-				}
-				if( ABS(*dd2) <= RGAMSQ )
-				{
-					*dd2  = *dd2 * (GAM * GAM);
-					dh21  = dh21 / GAM;
-					dh22  = dh22 / GAM;
-				}
-				else
-				{
-					*dd2  = *dd2 / (GAM * GAM);
-					dh21  = dh21 * GAM;
-					dh22  = dh22 * GAM;
-				}
-			}
 		}
 
 	}

interface/sbmv.c

@@ -184,7 +184,7 @@ void CNAME(enum CBLAS_ORDER order,
 
   if (n == 0) return;
 
-  if (beta != ONE) SCAL_K(n, 0, 0, beta, y, abs(incy), NULL, 0, NULL, 0);
+  if (beta != ONE) SCAL_K(n, 0, 0, beta, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if (alpha == ZERO) return;
 

interface/scal.c

@@ -76,10 +76,11 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx){
 
 
 #ifdef SMP
-  nthreads = num_cpu_avail(1);
-
   if (n <= 1048576 )
 	nthreads = 1;
+  else
+	nthreads = num_cpu_avail(1);
+
 
   if (nthreads == 1) {
 #endif

interface/spmv.c

@@ -168,7 +168,7 @@ void CNAME(enum CBLAS_ORDER order,
 
   if (n == 0) return;
 
-  if (beta != ONE) SCAL_K(n, 0, 0, beta, y, abs(incy), NULL, 0, NULL, 0);
+  if (beta != ONE) SCAL_K(n, 0, 0, beta, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if (alpha == ZERO) return;
 

interface/swap.c

@@ -42,7 +42,7 @@
 #include "functable.h"
 #endif
 
-#if defined(THUNDERX2T99) || defined(VULCAN)
+#if defined(THUNDERX2T99) || defined(VULCAN) || defined(ARMV8)
 // Multithreaded swap gives performance benefits in ThunderX2T99
 #else
 // Disable multi-threading as it does not show any performance

interface/symv.c

@@ -166,7 +166,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, blasint n, FLOAT alpha,
 
   if (n == 0) return;
 
-  if (beta != ONE) SCAL_K(n, 0, 0, beta, y, abs(incy), NULL, 0, NULL, 0);
+  if (beta != ONE) SCAL_K(n, 0, 0, beta, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if (alpha == ZERO) return;
 

interface/trsm.c

@@ -366,12 +366,13 @@ void CNAME(enum CBLAS_ORDER order,
   mode |= (trans << BLAS_TRANSA_SHIFT);
   mode |= (side  << BLAS_RSIDE_SHIFT);
 
-  args.nthreads = num_cpu_avail(3);
   if ( args.m < 2*GEMM_MULTITHREAD_THRESHOLD )
 	args.nthreads = 1;
   else
 	if ( args.n < 2*GEMM_MULTITHREAD_THRESHOLD )
 		args.nthreads = 1;
+  else
+	args.nthreads = num_cpu_avail(3);
 		
 
   if (args.nthreads == 1) {

interface/zaxpy.c

@@ -41,7 +41,11 @@
 #ifdef FUNCTION_PROFILE
 #include "functable.h"
 #endif
-
+#if  defined(Z13)
+#define MULTI_THREAD_MINIMAL  200000
+#else
+#define MULTI_THREAD_MINIMAL  10000
+#endif
 #ifndef CBLAS
 
 void NAME(blasint *N, FLOAT *ALPHA, FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY){
@@ -69,7 +73,7 @@ void CNAME(blasint n, FLOAT *ALPHA, FLOAT *x, blasint incx, FLOAT *y, blasint in
 #endif
 
 #ifndef CBLAS
-  PRINT_DEBUG_CNAME;
+  PRINT_DEBUG_NAME;
 #else
   PRINT_DEBUG_CNAME;
 #endif
@@ -78,6 +82,12 @@ void CNAME(blasint n, FLOAT *ALPHA, FLOAT *x, blasint incx, FLOAT *y, blasint in
 
   if ((alpha_r == ZERO) && (alpha_i == ZERO)) return;
 
+  if (incx == 0 && incy == 0) {
+  *y += n * (alpha_r * (*x) - alpha_i* (*(x+1)) );
+  *(y+1) += n * (alpha_i * (*x) + alpha_r * (*(x +1)) );
+  return;
+  }
+  
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
@@ -86,12 +96,15 @@ void CNAME(blasint n, FLOAT *ALPHA, FLOAT *x, blasint incx, FLOAT *y, blasint in
   if (incy < 0) y -= (n - 1) * incy * 2;
 
 #ifdef SMP
-  nthreads = num_cpu_avail(1);
-
   //disable multi-thread when incx==0 or incy==0
   //In that case, the threads would be dependent.
-  if (incx == 0 || incy == 0)
+  //
+  //Temporarily work-around the low performance issue with small imput size &
+  //multithreads.
+  if (incx == 0 || incy == 0 || n <= MULTI_THREAD_MINIMAL)
 	  nthreads = 1;
+  else
+	  nthreads = num_cpu_avail(1);
 
   if (nthreads == 1) {
 #endif

interface/zgbmv.c

@@ -237,7 +237,7 @@ void CNAME(enum CBLAS_ORDER order,
   if (trans & 1) lenx = m;
   if (trans & 1) leny = n;
 
-  if (beta_r != ONE || beta_i != ZERO) SCAL_K(leny, 0, 0, beta_r, beta_i, y, abs(incy), NULL, 0, NULL, 0);
+  if (beta_r != ONE || beta_i != ZERO) SCAL_K(leny, 0, 0, beta_r, beta_i, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if (alpha_r == ZERO && alpha_i == ZERO) return;
 

interface/zgemv.c

@@ -225,7 +225,7 @@ void CNAME(enum CBLAS_ORDER order,
   if (trans & 1) lenx = m;
   if (trans & 1) leny = n;
 
-  if (beta_r != ONE || beta_i != ZERO) SCAL_K(leny, 0, 0, beta_r, beta_i, y, abs(incy), NULL, 0, NULL, 0);
+  if (beta_r != ONE || beta_i != ZERO) SCAL_K(leny, 0, 0, beta_r, beta_i, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if (alpha_r == ZERO && alpha_i == ZERO) return;
 

interface/zhbmv.c

@@ -190,7 +190,7 @@ void CNAME(enum CBLAS_ORDER order,
 
   if (n == 0) return;
 
-  if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, y, abs(incy), NULL, 0, NULL, 0);
+  if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if ((alpha_r == ZERO) && (alpha_i == ZERO)) return;
 

interface/zhemv.c

@@ -43,6 +43,10 @@
 #include "functable.h"
 #endif
 
+// this is smallest dimension N of square input a to permit threading
+// see graph in issue #1820 for explanation
+#define MULTI_THREAD_MINIMAL 362
+
 #ifdef XDOUBLE
 #define ERROR_NAME "XHEMV "
 #elif defined(DOUBLE)
@@ -181,7 +185,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, blasint n, void *VALPHA
 
   if (n == 0) return;
 
-  if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, y, abs(incy), NULL, 0, NULL, 0);
+  if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if ((alpha_r == ZERO) && (alpha_i == ZERO)) return;
 
@@ -195,7 +199,11 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, blasint n, void *VALPHA
   buffer = (FLOAT *)blas_memory_alloc(1);
 
 #ifdef SMP
-  nthreads = num_cpu_avail(2);
+  if (n<MULTI_THREAD_MINIMAL) {
+	  nthreads = 1 ;
+  } else {
+  	  nthreads = num_cpu_avail(2);
+  };
 
   if (nthreads == 1) {
 #endif

interface/zhpmv.c

@@ -180,7 +180,7 @@ void CNAME(enum CBLAS_ORDER order,
 
   if (n == 0) return;
 
-  if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, y, abs(incy), NULL, 0, NULL, 0);
+  if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, y, blasabs(incy), NULL, 0, NULL, 0);
 
   if ((alpha_r == ZERO) && (alpha_i == ZERO)) return;
 

interface/zrotg.c

@@ -14,7 +14,7 @@ void NAME(FLOAT *DA, FLOAT *DB, FLOAT *C, FLOAT *S){
   long double db_i = *(DB + 1);
   long double r;
 
-  long double ada = fabs(da_r) + fabs(da_i);
+  long double ada = fabsl(da_r) + fabsl(da_i);
 
   PRINT_DEBUG_NAME;
 

interface/zsbmv.c

@@ -126,7 +126,7 @@ void NAME(char *UPLO, blasint *N, blasint *K, FLOAT  *ALPHA, FLOAT *a, blasint *
 
   if (n == 0) return;
 
-  if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, c, abs(incy), NULL, 0, NULL, 0);
+  if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, c, blasabs(incy), NULL, 0, NULL, 0);
 
   if ((alpha_r == ZERO) && (alpha_i == ZERO)) return;
 

interface/zscal.c

@@ -90,10 +90,10 @@ void CNAME(blasint n, FLOAT alpha_r, void *vx, blasint incx){
   FUNCTION_PROFILE_START();
 
 #ifdef SMP
-  nthreads = num_cpu_avail(1);
-
   if ( n <= 1048576 )
 	nthreads = 1;
+  else
+	nthreads = num_cpu_avail(1);
 
   if (nthreads == 1) {
 #endif

interface/zswap.c

@@ -42,6 +42,14 @@
 #include "functable.h"
 #endif
 
+#if defined(THUNDERX2T99) || defined(VULCAN) || defined(ARMV8)
+// Multithreaded swap gives performance benefits in ThunderX2T99
+#else
+// Disable multi-threading as it does not show any performance
+// benefits. Keep the multi-threading code for the record.
+#undef SMP
+#endif
+
 #ifndef CBLAS
 
 void NAME(blasint *N, FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY){
@@ -79,12 +87,12 @@ FLOAT *y = (FLOAT*)vy;
   if (incy < 0) y -= (n - 1) * incy * 2;
 
 #ifdef SMP
-  nthreads = num_cpu_avail(1);
-
   //disable multi-thread when incx==0 or incy==0
   //In that case, the threads would be dependent.
-  if (incx == 0 || incy == 0)
+  if (incx == 0 || incy == 0 || n < 1048576 * GEMM_MULTITHREAD_THRESHOLD / sizeof(FLOAT))
 	  nthreads = 1;
+  else
+	  nthreads = num_cpu_avail(1);
 
   if (nthreads == 1) {
 #endif

interface/ztrmv.c

@@ -239,6 +239,9 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
   } else
       nthreads = 1;
 
+/* FIXME TRMV multithreading appears to be broken, see issue 1332*/
+  nthreads = 1;
+
   if(nthreads > 1) {
     buffer_size = n > 16 ? 0 : n * 4 + 40;
   }

kernel/CMakeLists.txt

@@ -121,14 +121,17 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     # Makefile.L3
     set(USE_TRMM false)
 
-    if (ARM OR ARM64 OR "${TARGET_CORE}" STREQUAL "LONGSOON3B" OR "${TARGET_CORE}" STREQUAL "GENERIC" OR "${CORE}" STREQUAL "generic" OR "${TARGET_CORE}" STREQUAL "HASWELL" OR "${CORE}" STREQUAL "haswell" OR "${CORE}" STREQUAL "zen")
+    if (ARM OR ARM64 OR "${TARGET_CORE}" STREQUAL "LONGSOON3B" OR "${TARGET_CORE}" STREQUAL "GENERIC" OR "${CORE}" STREQUAL "generic" OR "${TARGET_CORE}" STREQUAL "HASWELL" OR "${CORE}" STREQUAL "haswell" OR "${CORE}" STREQUAL "zen" OR "${TARGET_CORE}" STREQUAL "SKYLAKEX" OR "${CORE}" STREQUAL "skylakex")
       set(USE_TRMM true)
     endif ()
 
-    foreach (float_type ${FLOAT_TYPES})
+    foreach (float_type SINGLE DOUBLE)
       string(SUBSTRING ${float_type} 0 1 float_char)
       GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "" "gemm_kernel" false "" "" false ${float_type})
+    endforeach()
 
+    foreach (float_type ${FLOAT_TYPES})
+      string(SUBSTRING ${float_type} 0 1 float_char)
       if (${float_char}GEMMINCOPY)
         GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMINCOPY}" "${float_type}" "${${float_char}GEMMINCOPYOBJ}" false "" "" true ${float_type})
       endif ()

kernel/Makefile

@@ -5,8 +5,43 @@ endif
 TOPDIR	= ..
 include $(TOPDIR)/Makefile.system
 
+AVX2OPT = 
+ifeq ($(C_COMPILER), GCC)
+# AVX2 support was added in 4.7.0
+  GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
+  GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
+  ifeq ($(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7), 11)
+   AVX2OPT = -mavx2
+  endif
+endif
+ifeq ($(C_COMPILER), CLANG)
+# Any clang posing as gcc 4.2 should be new enough (3.4 or later)
+  GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
+  GCCMINORVERSIONGTEQ2 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 2)
+  ifeq ($(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ2), 11)
+   AVX2OPT = -mavx2
+  endif
+endif
+ifdef NO_AVX2
+ AVX2OPT=
+endif
+
 ifdef TARGET_CORE
-override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
+ifeq ($(TARGET_CORE), SKYLAKEX)
+ override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -march=skylake-avx512
+ ifeq ($(OSNAME), CYGWIN_NT)
+  override CFLAGS += -fno-asynchronous-unwind-tables
+ endif
+ ifeq ($(OSNAME), WINNT)
+  ifeq ($(C_COMPILER), GCC)
+   override CFLAGS += -fno-asynchronous-unwind-tables
+  endif
+ endif
+else ifeq ($(TARGET_CORE), HASWELL)
+ override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(AVX2OPT)
+else
+ override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
+endif
 BUILD_KERNEL = 1
 KDIR =
 TSUFFIX = _$(TARGET_CORE)
@@ -88,7 +123,11 @@ lsame.$(SUFFIX): $(KERNELDIR)/$(LSAME_KERNEL)
 	$(CC) -c $(CFLAGS) -DF_INTERFACE $< -o $(@F)
 
 setparam$(TSUFFIX).$(SUFFIX): setparam$(TSUFFIX).c kernel$(TSUFFIX).h
+ifeq ($(USE_GEMM3M), 1)
+	$(CC) -c $(CFLAGS) -DUSE_GEMM3M $< -o $@
+else
 	$(CC) -c $(CFLAGS) $< -o $@
+endif
 
 setparam$(TSUFFIX).c : setparam-ref.c
 	sed 's/TS/$(TSUFFIX)/g' $< > $(@F)

kernel/Makefile.L3

@@ -29,9 +29,11 @@ USE_TRMM = 1
 endif
 
 ifeq ($(CORE), HASWELL)
-ifeq ($(ARCH), x86_64)
 USE_TRMM = 1
 endif
+
+ifeq ($(CORE), SKYLAKEX)
+USE_TRMM = 1
 endif
 
 ifeq ($(CORE), ZEN)
@@ -42,7 +44,7 @@ ifeq ($(CORE), POWER8)
 USE_TRMM = 1
 endif
 
-ifeq ($(CORE), Z13)
+ifeq ($(ARCH), zarch)
 USE_TRMM = 1
 endif
 

kernel/arm/KERNEL.ARMV5

@@ -49,6 +49,7 @@ SDOTKERNEL   = ../arm/dot.c
 DDOTKERNEL   = ../arm/dot.c
 CDOTKERNEL   = ../arm/zdot.c
 ZDOTKERNEL   = ../arm/zdot.c
+DSDOTKERNEL  = ../generic/dot.c
 
 SNRM2KERNEL  = ../arm/nrm2.c
 DNRM2KERNEL  = ../arm/nrm2.c

kernel/arm/asum_vfp.S

@@ -58,11 +58,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL_F4
 
 	pld	[ X, #X_PRE  ]
-	fldmiad	X!, { d4 - d5 }
+	vldmia.f64	X!, { d4 - d5 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	vabs.f64   d5,  d5
-	fldmiad	X!, { d6 - d7 }
+	vldmia.f64	X!, { d6 - d7 }
 	vabs.f64   d6,  d6
 	vadd.f64   d1  , d1,  d5
 	vabs.f64   d7,  d7
@@ -73,7 +73,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmiad	X!, { d4 }
+	vldmia.f64	X!, { d4 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 
@@ -82,22 +82,22 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S4
 
-	fldmiad	X, { d4 }
+	vldmia.f64	X, { d4 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	add	X, X, INC_X
 
-	fldmiad	X, { d4 }
+	vldmia.f64	X, { d4 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	add	X, X, INC_X
 
-	fldmiad	X, { d4 }
+	vldmia.f64	X, { d4 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	add	X, X, INC_X
 
-	fldmiad	X, { d4 }
+	vldmia.f64	X, { d4 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	add	X, X, INC_X
@@ -107,7 +107,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S1
 
-	fldmiad	X, { d4 }
+	vldmia.f64	X, { d4 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	add	X, X, INC_X
@@ -118,11 +118,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F4
 
-	fldmias	X!, { s4 - s5 }
+	vldmia.f32	X!, { s4 - s5 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	vabs.f32   s5,  s5
-	fldmias	X!, { s6 - s7 }
+	vldmia.f32	X!, { s6 - s7 }
 	vabs.f32   s6,  s6
 	vadd.f32   s1  , s1,  s5
 	vabs.f32   s7,  s7
@@ -133,7 +133,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmias	X!, { s4 }
+	vldmia.f32	X!, { s4 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 
@@ -142,22 +142,22 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S4
 
-	fldmias	X, { s4 }
+	vldmia.f32	X, { s4 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	add	X, X, INC_X
 
-	fldmias	X, { s4 }
+	vldmia.f32	X, { s4 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	add	X, X, INC_X
 
-	fldmias	X, { s4 }
+	vldmia.f32	X, { s4 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	add	X, X, INC_X
 
-	fldmias	X, { s4 }
+	vldmia.f32	X, { s4 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	add	X, X, INC_X
@@ -167,7 +167,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S1
 
-	fldmias	X, { s4 }
+	vldmia.f32	X, { s4 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	add	X, X, INC_X
@@ -184,11 +184,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL_F4
 
 	pld	[ X, #X_PRE  ]
-	fldmiad	X!, { d4 - d5 }
+	vldmia.f64	X!, { d4 - d5 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	vabs.f64   d5,  d5
-	fldmiad	X!, { d6 - d7 }
+	vldmia.f64	X!, { d6 - d7 }
 	vabs.f64   d6,  d6
 	vadd.f64   d1  , d1,  d5
 	vabs.f64   d7,  d7
@@ -196,11 +196,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vadd.f64   d1  , d1,  d7
 
 	pld	[ X, #X_PRE  ]
-	fldmiad	X!, { d4 - d5 }
+	vldmia.f64	X!, { d4 - d5 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	vabs.f64   d5,  d5
-	fldmiad	X!, { d6 - d7 }
+	vldmia.f64	X!, { d6 - d7 }
 	vabs.f64   d6,  d6
 	vadd.f64   d1  , d1,  d5
 	vabs.f64   d7,  d7
@@ -212,11 +212,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmiad	X!, { d4 }
+	vldmia.f64	X!, { d4 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 
-	fldmiad	X!, { d4 }
+	vldmia.f64	X!, { d4 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 
@@ -226,28 +226,28 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S4
 
-	fldmiad	X, { d4 -d5 }
+	vldmia.f64	X, { d4 -d5 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	vabs.f64   d5,  d5
 	vadd.f64   d0  , d0,  d5
 	add	X, X, INC_X
 
-	fldmiad	X, { d4 -d5 }
+	vldmia.f64	X, { d4 -d5 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	vabs.f64   d5,  d5
 	vadd.f64   d0  , d0,  d5
 	add	X, X, INC_X
 
-	fldmiad	X, { d4 -d5 }
+	vldmia.f64	X, { d4 -d5 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	vabs.f64   d5,  d5
 	vadd.f64   d0  , d0,  d5
 	add	X, X, INC_X
 
-	fldmiad	X, { d4 -d5 }
+	vldmia.f64	X, { d4 -d5 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	vabs.f64   d5,  d5
@@ -259,7 +259,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S1
 
-	fldmiad	X, { d4 -d5 }
+	vldmia.f64	X, { d4 -d5 }
 	vabs.f64   d4,  d4
 	vadd.f64   d0  , d0,  d4
 	vabs.f64   d5,  d5
@@ -273,22 +273,22 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL_F4
 
 	pld	[ X, #X_PRE  ]
-	fldmias	X!, { s4 - s5 }
+	vldmia.f32	X!, { s4 - s5 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	vabs.f32   s5,  s5
-	fldmias	X!, { s6 - s7 }
+	vldmia.f32	X!, { s6 - s7 }
 	vabs.f32   s6,  s6
 	vadd.f32   s1  , s1,  s5
 	vabs.f32   s7,  s7
 	vadd.f32   s0  , s0,  s6
 	vadd.f32   s1  , s1,  s7
 
-	fldmias	X!, { s4 - s5 }
+	vldmia.f32	X!, { s4 - s5 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	vabs.f32   s5,  s5
-	fldmias	X!, { s6 - s7 }
+	vldmia.f32	X!, { s6 - s7 }
 	vabs.f32   s6,  s6
 	vadd.f32   s1  , s1,  s5
 	vabs.f32   s7,  s7
@@ -300,11 +300,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmias	X!, { s4 }
+	vldmia.f32	X!, { s4 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 
-	fldmias	X!, { s4 }
+	vldmia.f32	X!, { s4 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 
@@ -313,28 +313,28 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S4
 
-	fldmias	X, { s4 -s5 }
+	vldmia.f32	X, { s4 -s5 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	vabs.f32   s5,  s5
 	vadd.f32   s0  , s0,  s5
 	add	X, X, INC_X
 
-	fldmias	X, { s4 -s5 }
+	vldmia.f32	X, { s4 -s5 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	vabs.f32   s5,  s5
 	vadd.f32   s0  , s0,  s5
 	add	X, X, INC_X
 
-	fldmias	X, { s4 -s5 }
+	vldmia.f32	X, { s4 -s5 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	vabs.f32   s5,  s5
 	vadd.f32   s0  , s0,  s5
 	add	X, X, INC_X
 
-	fldmias	X, { s4 -s5 }
+	vldmia.f32	X, { s4 -s5 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	vabs.f32   s5,  s5
@@ -346,7 +346,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S1
 
-	fldmias	X, { s4 -s5 }
+	vldmia.f32	X, { s4 -s5 }
 	vabs.f32   s4,  s4
 	vadd.f32   s0  , s0,  s4
 	vabs.f32   s5,  s5

kernel/arm/axpy_vfp.S

@@ -146,17 +146,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL_F4
 
 	pld	[ X, #X_PRE ]
-	fldmiad		X!,  { d4 - d7  }
+	vldmia.f64		X!,  { d4 - d7  }
 	pld	[ Y, #X_PRE ]
-	fldmiad		Y ,  { d8 - d11 }
+	vldmia.f64		Y ,  { d8 - d11 }
 	fmacd   	d8 , d0, d4
-	fstmiad		Y!, { d8 }
+	vstmia.f64		Y!, { d8 }
 	fmacd   	d9 , d0, d5
-	fstmiad		Y!, { d9 }
+	vstmia.f64		Y!, { d9 }
 	fmacd   	d10, d0, d6
-	fstmiad		Y!, { d10 }
+	vstmia.f64		Y!, { d10 }
 	fmacd   	d11, d0, d7
-	fstmiad		Y!, { d11 }
+	vstmia.f64		Y!, { d11 }
 
 
 .endm
@@ -164,19 +164,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmiad		X!,  { d4 }
+	vldmia.f64		X!,  { d4 }
-	fldmiad		Y ,  { d8 }
+	vldmia.f64		Y ,  { d8 }
 	fmacd   	d8 , d0, d4
-	fstmiad		Y!, { d8 }
+	vstmia.f64		Y!, { d8 }
 
 .endm
 
 .macro KERNEL_S1
 
-	fldmiad		X ,  { d4 }
+	vldmia.f64		X ,  { d4 }
-	fldmiad		Y ,  { d8 }
+	vldmia.f64		Y ,  { d8 }
 	fmacd   	d8 , d0, d4
-	fstmiad		Y , { d8 }
+	vstmia.f64		Y , { d8 }
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
@@ -186,16 +186,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F4
 
-	fldmias		X!,  { s4 - s7  }
+	vldmia.f32		X!,  { s4 - s7  }
-	fldmias		Y ,  { s8 - s11 }
+	vldmia.f32		Y ,  { s8 - s11 }
 	fmacs   	s8 , s0, s4
-	fstmias		Y!, { s8 }
+	vstmia.f32		Y!, { s8 }
 	fmacs   	s9 , s0, s5
-	fstmias		Y!, { s9 }
+	vstmia.f32		Y!, { s9 }
 	fmacs   	s10, s0, s6
-	fstmias		Y!, { s10 }
+	vstmia.f32		Y!, { s10 }
 	fmacs   	s11, s0, s7
-	fstmias		Y!, { s11 }
+	vstmia.f32		Y!, { s11 }
 
 
 .endm
@@ -203,19 +203,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmias		X!,  { s4 }
+	vldmia.f32		X!,  { s4 }
-	fldmias		Y ,  { s8 }
+	vldmia.f32		Y ,  { s8 }
 	fmacs   	s8 , s0, s4
-	fstmias		Y!, { s8 }
+	vstmia.f32		Y!, { s8 }
 
 .endm
 
 .macro KERNEL_S1
 
-	fldmias		X ,  { s4 }
+	vldmia.f32		X ,  { s4 }
-	fldmias		Y ,  { s8 }
+	vldmia.f32		Y ,  { s8 }
 	fmacs   	s8 , s0, s4
-	fstmias		Y , { s8 }
+	vstmia.f32		Y , { s8 }
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
@@ -231,42 +231,42 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL_F4
 
 	pld	[ X, #X_PRE ]
-	fldmiad		X!,  { d4 - d7  }
+	vldmia.f64		X!,  { d4 - d7  }
 	pld	[ Y, #X_PRE ]
-	fldmiad		Y ,  { d8 - d11 }
+	vldmia.f64		Y ,  { d8 - d11 }
 
 	FMAC_R1		d8 , d0, d4
 	FMAC_R2		d8 , d1, d5
 	FMAC_I1		d9 , d0, d5
 	FMAC_I2		d9 , d1, d4
-	fstmiad		Y!, { d8 }
+	vstmia.f64		Y!, { d8 }
-	fstmiad		Y!, { d9 }
+	vstmia.f64		Y!, { d9 }
 
 	FMAC_R1		d10, d0, d6
 	FMAC_R2		d10, d1, d7
 	FMAC_I1		d11, d0, d7
 	FMAC_I2		d11, d1, d6
-	fstmiad		Y!, { d10 }
+	vstmia.f64		Y!, { d10 }
-	fstmiad		Y!, { d11 }
+	vstmia.f64		Y!, { d11 }
 
 	pld	[ X, #X_PRE ]
-	fldmiad		X!,  { d4 - d7  }
+	vldmia.f64		X!,  { d4 - d7  }
 	pld	[ Y, #X_PRE ]
-	fldmiad		Y ,  { d8 - d11 }
+	vldmia.f64		Y ,  { d8 - d11 }
 
 	FMAC_R1		d8 , d0, d4
 	FMAC_R2		d8 , d1, d5
 	FMAC_I1		d9 , d0, d5
 	FMAC_I2		d9 , d1, d4
-	fstmiad		Y!, { d8 }
+	vstmia.f64		Y!, { d8 }
-	fstmiad		Y!, { d9 }
+	vstmia.f64		Y!, { d9 }
 
 	FMAC_R1		d10, d0, d6
 	FMAC_R2		d10, d1, d7
 	FMAC_I1		d11, d0, d7
 	FMAC_I2		d11, d1, d6
-	fstmiad		Y!, { d10 }
+	vstmia.f64		Y!, { d10 }
-	fstmiad		Y!, { d11 }
+	vstmia.f64		Y!, { d11 }
 
 
 
@@ -277,15 +277,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmiad		X!,  { d4 - d5  }
+	vldmia.f64		X!,  { d4 - d5  }
-	fldmiad		Y ,  { d8 - d9 }
+	vldmia.f64		Y ,  { d8 - d9 }
 
 	FMAC_R1		d8 , d0, d4
 	FMAC_R2		d8 , d1, d5
 	FMAC_I1		d9 , d0, d5
 	FMAC_I2		d9 , d1, d4
-	fstmiad		Y!, { d8 }
+	vstmia.f64		Y!, { d8 }
-	fstmiad		Y!, { d9 }
+	vstmia.f64		Y!, { d9 }
 
 
 
@@ -293,14 +293,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S1
 
-	fldmiad		X ,  { d4 - d5 }
+	vldmia.f64		X ,  { d4 - d5 }
-	fldmiad		Y ,  { d8 - d9 }
+	vldmia.f64		Y ,  { d8 - d9 }
 
 	FMAC_R1		d8 , d0, d4
 	FMAC_R2		d8 , d1, d5
 	FMAC_I1		d9 , d0, d5
 	FMAC_I2		d9 , d1, d4
-	fstmiad		Y  , { d8 - d9 }
+	vstmia.f64		Y  , { d8 - d9 }
 
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
@@ -314,40 +314,40 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL_F4
 
 	pld	[ X, #X_PRE ]
-	fldmias		X!,  { s4 - s7  }
+	vldmia.f32		X!,  { s4 - s7  }
 	pld	[ Y, #X_PRE ]
-	fldmias		Y ,  { s8 - s11 }
+	vldmia.f32		Y ,  { s8 - s11 }
 
 	FMAC_R1		s8 , s0, s4
 	FMAC_R2		s8 , s1, s5
 	FMAC_I1		s9 , s0, s5
 	FMAC_I2		s9 , s1, s4
-	fstmias		Y!, { s8 }
+	vstmia.f32		Y!, { s8 }
-	fstmias		Y!, { s9 }
+	vstmia.f32		Y!, { s9 }
 
 	FMAC_R1		s10, s0, s6
 	FMAC_R2		s10, s1, s7
 	FMAC_I1		s11, s0, s7
 	FMAC_I2		s11, s1, s6
-	fstmias		Y!, { s10 }
+	vstmia.f32		Y!, { s10 }
-	fstmias		Y!, { s11 }
+	vstmia.f32		Y!, { s11 }
 
-	fldmias		X!,  { s4 - s7  }
+	vldmia.f32		X!,  { s4 - s7  }
-	fldmias		Y ,  { s8 - s11 }
+	vldmia.f32		Y ,  { s8 - s11 }
 
 	FMAC_R1		s8 , s0, s4
 	FMAC_R2		s8 , s1, s5
 	FMAC_I1		s9 , s0, s5
 	FMAC_I2		s9 , s1, s4
-	fstmias		Y!, { s8 }
+	vstmia.f32		Y!, { s8 }
-	fstmias		Y!, { s9 }
+	vstmia.f32		Y!, { s9 }
 
 	FMAC_R1		s10, s0, s6
 	FMAC_R2		s10, s1, s7
 	FMAC_I1		s11, s0, s7
 	FMAC_I2		s11, s1, s6
-	fstmias		Y!, { s10 }
+	vstmia.f32		Y!, { s10 }
-	fstmias		Y!, { s11 }
+	vstmia.f32		Y!, { s11 }
 
 
 
@@ -358,15 +358,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmias		X!,  { s4 - s5  }
+	vldmia.f32		X!,  { s4 - s5  }
-	fldmias		Y ,  { s8 - s9 }
+	vldmia.f32		Y ,  { s8 - s9 }
 
 	FMAC_R1		s8 , s0, s4
 	FMAC_R2		s8 , s1, s5
 	FMAC_I1		s9 , s0, s5
 	FMAC_I2		s9 , s1, s4
-	fstmias		Y!, { s8 }
+	vstmia.f32		Y!, { s8 }
-	fstmias		Y!, { s9 }
+	vstmia.f32		Y!, { s9 }
 
 
 
@@ -374,14 +374,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S1
 
-	fldmias		X ,  { s4 - s5 }
+	vldmia.f32		X ,  { s4 - s5 }
-	fldmias		Y ,  { s8 - s9 }
+	vldmia.f32		Y ,  { s8 - s9 }
 
 	FMAC_R1		s8 , s0, s4
 	FMAC_R2		s8 , s1, s5
 	FMAC_I1		s9 , s0, s5
 	FMAC_I2		s9 , s1, s4
-	fstmias		Y  , { s8 - s9 }
+	vstmia.f32		Y  , { s8 - s9 }
 
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
@@ -440,13 +440,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	cmp	N, #0
 	ble	axpy_kernel_L999
-
+/*
 	cmp	INC_X, #0
 	beq	axpy_kernel_L999
 
 	cmp	INC_Y, #0
 	beq	axpy_kernel_L999
-
+*/
 	cmp	INC_X, #1
 	bne	axpy_kernel_S_BEGIN
 

kernel/arm/ccopy_vfp.S

@@ -65,15 +65,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro COPY_F4
 
 	pld	[ X, #X_PRE  ]
-	fldmias	X!, { s0 - s7 }
+	vldmia.f32	X!, { s0 - s7 }
-	fstmias	Y!, { s0 - s7 }
+	vstmia.f32	Y!, { s0 - s7 }
 
 .endm
 
 .macro COPY_F1
 
-	fldmias	X!, { s0 - s1 }
+	vldmia.f32	X!, { s0 - s1 }
-	fstmias	Y!, { s0 - s1 }
+	vstmia.f32	Y!, { s0 - s1 }
 
 .endm
 
@@ -83,23 +83,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro COPY_S4
 
 	nop
-	fldmias	X, { s0 - s1 }
+	vldmia.f32	X, { s0 - s1 }
-	fstmias	Y, { s0 - s1 }
+	vstmia.f32	Y, { s0 - s1 }
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
-	fldmias	X, { s2 - s3 }
+	vldmia.f32	X, { s2 - s3 }
-	fstmias	Y, { s2 - s3 }
+	vstmia.f32	Y, { s2 - s3 }
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
-	fldmias	X, { s0 - s1 }
+	vldmia.f32	X, { s0 - s1 }
-	fstmias	Y, { s0 - s1 }
+	vstmia.f32	Y, { s0 - s1 }
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
-	fldmias	X, { s2 - s3 }
+	vldmia.f32	X, { s2 - s3 }
-	fstmias	Y, { s2 - s3 }
+	vstmia.f32	Y, { s2 - s3 }
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
@@ -108,8 +108,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro COPY_S1
 
-	fldmias	X, { s0 - s1 }
+	vldmia.f32	X, { s0 - s1 }
-	fstmias	Y, { s0 - s1 }
+	vstmia.f32	Y, { s0 - s1 }
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 

kernel/arm/cdot_vfp.S

@@ -76,30 +76,30 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	pld	[ X, #X_PRE  ]
 	pld	[ Y, #X_PRE  ]
 
-	fldmias	X!, { s4 - s5 }
+	vldmia.f32	X!, { s4 - s5 }
-	fldmias	Y!, { s8 - s9 }
+	vldmia.f32	Y!, { s8 - s9 }
 	fmacs   s0  , s4,  s8
 	fmacs   s1  , s4,  s9
-	fldmias	X!, { s6 - s7 }
+	vldmia.f32	X!, { s6 - s7 }
 	fmacs   s2  , s5,  s9
 	fmacs   s3  , s5,  s8
 
-	fldmias	Y!, { s10 - s11 }
+	vldmia.f32	Y!, { s10 - s11 }
 	fmacs   s0  , s6,  s10
 	fmacs   s1  , s6,  s11
 	fmacs   s2  , s7,  s11
 	fmacs   s3  , s7,  s10
 
 
-	fldmias	X!, { s4 - s5 }
+	vldmia.f32	X!, { s4 - s5 }
-	fldmias	Y!, { s8 - s9 }
+	vldmia.f32	Y!, { s8 - s9 }
 	fmacs   s0  , s4,  s8
 	fmacs   s1  , s4,  s9
-	fldmias	X!, { s6 - s7 }
+	vldmia.f32	X!, { s6 - s7 }
 	fmacs   s2  , s5,  s9
 	fmacs   s3  , s5,  s8
 
-	fldmias	Y!, { s10 - s11 }
+	vldmia.f32	Y!, { s10 - s11 }
 	fmacs   s0  , s6,  s10
 	fmacs   s1  , s6,  s11
 	fmacs   s2  , s7,  s11
@@ -109,8 +109,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_F1
 
-	fldmias	X!, { s4 - s5 }
+	vldmia.f32	X!, { s4 - s5 }
-	fldmias	Y!, { s8 - s9 }
+	vldmia.f32	Y!, { s8 - s9 }
 	fmacs   s0  , s4,  s8
 	fmacs   s1  , s4,  s9
 	fmacs   s2  , s5,  s9
@@ -125,8 +125,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	nop
 
-	fldmias	X, { s4 - s5 }
+	vldmia.f32	X, { s4 - s5 }
-	fldmias	Y, { s8 - s9 }
+	vldmia.f32	Y, { s8 - s9 }
 	fmacs   s0  , s4,  s8
 	fmacs   s1  , s4,  s9
 	fmacs   s2  , s5,  s9
@@ -134,8 +134,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
-	fldmias	X, { s4 - s5 }
+	vldmia.f32	X, { s4 - s5 }
-	fldmias	Y, { s8 - s9 }
+	vldmia.f32	Y, { s8 - s9 }
 	fmacs   s0  , s4,  s8
 	fmacs   s1  , s4,  s9
 	fmacs   s2  , s5,  s9
@@ -143,8 +143,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
-	fldmias	X, { s4 - s5 }
+	vldmia.f32	X, { s4 - s5 }
-	fldmias	Y, { s8 - s9 }
+	vldmia.f32	Y, { s8 - s9 }
 	fmacs   s0  , s4,  s8
 	fmacs   s1  , s4,  s9
 	fmacs   s2  , s5,  s9
@@ -152,8 +152,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	X, X, INC_X
 	add	Y, Y, INC_Y
 
-	fldmias	X, { s4 - s5 }
+	vldmia.f32	X, { s4 - s5 }
-	fldmias	Y, { s8 - s9 }
+	vldmia.f32	Y, { s8 - s9 }
 	fmacs   s0  , s4,  s8
 	fmacs   s1  , s4,  s9
 	fmacs   s2  , s5,  s9
@@ -166,8 +166,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL_S1
 
-	fldmias	X, { s4 - s5 }
+	vldmia.f32	X, { s4 - s5 }
-	fldmias	Y, { s8 - s9 }
+	vldmia.f32	Y, { s8 - s9 }
 	fmacs   s0  , s4,  s8
 	fmacs   s1  , s4,  s9
 	fmacs   s2  , s5,  s9
@@ -215,11 +215,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	N, #0
 	ble	cdot_kernel_L999
 
-	cmp	INC_X, #0
+#	cmp	INC_X, #0
-	beq	cdot_kernel_L999
+#	beq	cdot_kernel_L999
 
-	cmp	INC_Y, #0
+#	cmp	INC_Y, #0
-	beq	cdot_kernel_L999
+#	beq	cdot_kernel_L999
 
 	cmp	INC_X, #1
 	bne	cdot_kernel_S_BEGIN

kernel/arm/cgemm_kernel_2x2_vfp.S

@@ -165,9 +165,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL2x2_I
 
 	pld	[ AO, #A_PRE ]
-	fldmias	AO!, { s0 - s3 }
+	vldmia.f32	AO!, { s0 - s3 }
 	pld	[ BO, #B_PRE ]
-	fldmias	BO!, { s4 - s7 }
+	vldmia.f32	BO!, { s4 - s7 }
 
 
 	fmuls	s8  , s0,  s4
@@ -197,9 +197,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL2x2_M1
 
 	pld	[ AO, #A_PRE ]
-	fldmias	AO!, { s0 - s3 }
+	vldmia.f32	AO!, { s0 - s3 }
 	pld	[ BO, #B_PRE ]
-	fldmias	BO!, { s4 - s7 }
+	vldmia.f32	BO!, { s4 - s7 }
 
 	fmacs	s8  , s0,  s4
 	fmacs	s9  , s0,  s5
@@ -225,8 +225,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL2x2_M2
 
-	fldmias	AO!, { s0 - s3 }
+	vldmia.f32	AO!, { s0 - s3 }
-	fldmias	BO!, { s4 - s7 }
+	vldmia.f32	BO!, { s4 - s7 }
 
 	fmacs	s8  , s0,  s4
 	fmacs	s9  , s0,  s5
@@ -254,8 +254,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL2x2_E
 
-	fldmias	AO!, { s0 - s3 }
+	vldmia.f32	AO!, { s0 - s3 }
-	fldmias	BO!, { s4 - s7 }
+	vldmia.f32	BO!, { s4 - s7 }
 
 	fmacs	s8  , s0,  s4
 	fmacs	s9  , s0,  s5
@@ -282,8 +282,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL2x2_SUB
 
-	fldmias	AO!, { s0 - s3 }
+	vldmia.f32	AO!, { s0 - s3 }
-	fldmias	BO!, { s4 - s7 }
+	vldmia.f32	BO!, { s4 - s7 }
 
 	fmacs	s8  , s0,  s4
 	fmacs	s9  , s0,  s5
@@ -317,7 +317,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds		s0, ALPHA_R
 	flds		s1, ALPHA_I
 
-	fldmias CO1, { s4 - s7 }
+	vldmia.f32 CO1, { s4 - s7 }
 
 	FMAC_R1 s4 , s0 , s8
 	FMAC_I1 s5 , s0 , s9
@@ -329,9 +329,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	FMAC_R2 s6 , s1 , s11
 	FMAC_I2	s7 , s1 , s10
 
-	fstmias CO1, { s4 - s7 }
+	vstmia.f32 CO1, { s4 - s7 }
 
-	fldmias CO2, { s4 - s7 }
+	vldmia.f32 CO2, { s4 - s7 }
 
 	FMAC_R1 s4 , s0 , s12
 	FMAC_I1 s5 , s0 , s13
@@ -343,7 +343,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	FMAC_R2 s6 , s1 , s15
 	FMAC_I2	s7 , s1 , s14
 
-	fstmias CO2, { s4 - s7 }
+	vstmia.f32 CO2, { s4 - s7 }
 
 	add	CO1, CO1, #16
 
@@ -500,23 +500,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds		s0, ALPHA_R
 	flds		s1, ALPHA_I
 
-	fldmias CO1, { s4 - s5 }
+	vldmia.f32 CO1, { s4 - s5 }
 
 	FMAC_R1 s4 , s0 , s8
 	FMAC_I1 s5 , s0 , s9
 	FMAC_R2 s4 , s1 , s9
 	FMAC_I2	s5 , s1 , s8
 
-	fstmias CO1, { s4 - s5 }
+	vstmia.f32 CO1, { s4 - s5 }
 
-	fldmias CO2, { s4 - s5 }
+	vldmia.f32 CO2, { s4 - s5 }
 
 	FMAC_R1 s4 , s0 , s12
 	FMAC_I1 s5 , s0 , s13
 	FMAC_R2 s4 , s1 , s13
 	FMAC_I2	s5 , s1 , s12
 
-	fstmias CO2, { s4 - s5 }
+	vstmia.f32 CO2, { s4 - s5 }
 
 	add	CO1, CO1, #8
 
@@ -671,7 +671,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds		s0, ALPHA_R
 	flds		s1, ALPHA_I
 
-	fldmias CO1, { s4 - s7 }
+	vldmia.f32 CO1, { s4 - s7 }
 
 	FMAC_R1 s4 , s0 , s8
 	FMAC_I1 s5 , s0 , s9
@@ -683,7 +683,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	FMAC_R2 s6 , s1 , s11
 	FMAC_I2	s7 , s1 , s10
 
-	fstmias CO1, { s4 - s7 }
+	vstmia.f32 CO1, { s4 - s7 }
 
 	add	CO1, CO1, #16
 
@@ -800,14 +800,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds		s0, ALPHA_R
 	flds		s1, ALPHA_I
 
-	fldmias CO1, { s4 - s5 }
+	vldmia.f32 CO1, { s4 - s5 }
 
 	FMAC_R1 s4 , s0 , s8
 	FMAC_I1 s5 , s0 , s9
 	FMAC_R2 s4 , s1 , s9
 	FMAC_I2	s5 , s1 , s8
 
-	fstmias CO1, { s4 - s5 }
+	vstmia.f32 CO1, { s4 - s5 }
 
 	add	CO1, CO1, #8
 

kernel/arm/cgemm_kernel_2x2_vfpv3.S

@@ -182,30 +182,30 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL2x2_I
 	pld	[ AO , #A_PRE ]
 	pld	[ BO , #B_PRE ]
-	fldmias AO!, { s0 - s1 }
+	vldmia.f32 AO!, { s0 - s1 }
-	fldmias BO!, { s8 - s9 }
+	vldmia.f32 BO!, { s8 - s9 }
 
 	fmuls	s16  , s0,  s8
 	fmuls	s24  , s1,  s9
-	fldmias AO!, { s2 - s3 }
+	vldmia.f32 AO!, { s2 - s3 }
 	fmuls	s17  , s0,  s9
 	fmuls	s25  , s1,  s8
 
-	fldmias BO!, { s10 - s11 }
+	vldmia.f32 BO!, { s10 - s11 }
 	fmuls	s18  , s2,  s8
 	fmuls	s26  , s3,  s9
-	fldmias AO!, { s4 - s5 }
+	vldmia.f32 AO!, { s4 - s5 }
 	fmuls	s19  , s2,  s9
 	fmuls	s27  , s3,  s8
 
-	fldmias BO!, { s12 - s13 }
+	vldmia.f32 BO!, { s12 - s13 }
 	fmuls	s20  , s0,  s10
 	fmuls	s28  , s1,  s11
-	fldmias AO!, { s6 - s7 }
+	vldmia.f32 AO!, { s6 - s7 }
 	fmuls	s21  , s0,  s11
 	fmuls	s29  , s1,  s10
 
-	fldmias BO!, { s14 - s15 }
+	vldmia.f32 BO!, { s14 - s15 }
 	fmuls	s22  , s2,  s10
 	fmuls	s30  , s3,  s11
 	fmuls	s23  , s2,  s11
@@ -218,17 +218,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro KERNEL2x2_M1
 
 	fmacs	s16  , s0,  s8
-	fldmias AO!, { s4 - s5 }
+	vldmia.f32 AO!, { s4 - s5 }
 	fmacs	s24  , s1,  s9
 	fmacs	s17  , s0,  s9
-	fldmias BO!, { s12 - s13 }
+	vldmia.f32 BO!, { s12 - s13 }
 	fmacs	s25  , s1,  s8
 
 	fmacs	s18  , s2,  s8
-	fldmias AO!, { s6 - s7 }
+	vldmia.f32 AO!, { s6 - s7 }
 	fmacs	s26  , s3,  s9
 	fmacs	s19  , s2,  s9
-	fldmias BO!, { s14 - s15 }
+	vldmia.f32 BO!, { s14 - s15 }
 	fmacs	s27  , s3,  s8
 
 	fmacs	s20  , s0,  s10
@@ -250,19 +250,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	pld	[ BO , #B_PRE ]
 	fmacs	s24  , s5,  s13
 	fmacs	s17  , s4,  s13
-	fldmias AO!, { s0 - s1 }
+	vldmia.f32 AO!, { s0 - s1 }
 	fmacs	s25  , s5,  s12
 
 	fmacs	s18  , s6,  s12
 	fmacs	s26  , s7,  s13
-	fldmias BO!, { s8 - s9 }
+	vldmia.f32 BO!, { s8 - s9 }
 	fmacs	s19  , s6,  s13
 	fmacs	s27  , s7,  s12
 
-	fldmias AO!, { s2 - s3 }
+	vldmia.f32 AO!, { s2 - s3 }
 	fmacs	s20  , s4,  s14
 	fmacs	s28  , s5,  s15
-	fldmias BO!, { s10 - s11 }
+	vldmia.f32 BO!, { s10 - s11 }
 	fmacs	s21  , s4,  s15
 	fmacs	s29  , s5,  s14
 
@@ -300,16 +300,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro KERNEL2x2_SUB
 
-	fldmias AO!, { s0 - s1 }
+	vldmia.f32 AO!, { s0 - s1 }
-	fldmias BO!, { s8 - s9 }
+	vldmia.f32 BO!, { s8 - s9 }
 
 	fmacs	s16  , s0,  s8
 	fmacs	s24  , s1,  s9
-	fldmias AO!, { s2 - s3 }
+	vldmia.f32 AO!, { s2 - s3 }
 	fmacs	s17  , s0,  s9
 	fmacs	s25  , s1,  s8
 
-	fldmias BO!, { s10 - s11 }
+	vldmia.f32 BO!, { s10 - s11 }
 	fmacs	s18  , s2,  s8
 	fmacs	s26  , s3,  s9
 	fmacs	s19  , s2,  s9
@@ -338,8 +338,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds		s0, ALPHA_R
 	flds		s1, ALPHA_I
 
-	fldmias CO1, { s4 - s7 }
+	vldmia.f32 CO1, { s4 - s7 }
-	fldmias CO2, { s8 - s11 }
+	vldmia.f32 CO2, { s8 - s11 }
 
 	FADD_R	s16, s24 , s16
 	FADD_I  s17, s25 , s17
@@ -370,8 +370,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	FMAC_R2 s10, s1 , s23
 	FMAC_I2	s11, s1 , s22
 
-	fstmias CO1, { s4 - s7 }
+	vstmia.f32 CO1, { s4 - s7 }
-	fstmias CO2, { s8 - s11 }
+	vstmia.f32 CO2, { s8 - s11 }
 
 	add	CO1, CO1, #16
 
@@ -534,8 +534,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds		s0, ALPHA_R
 	flds		s1, ALPHA_I
 
-	fldmias CO1, { s4 - s5 }
+	vldmia.f32 CO1, { s4 - s5 }
-	fldmias CO2, { s8 - s9  }
+	vldmia.f32 CO2, { s8 - s9  }
 
 	FADD_R	s16, s24 , s16
 	FADD_I  s17, s25 , s17
@@ -552,8 +552,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	FMAC_R2 s8 , s1 , s21
 	FMAC_I2	s9 , s1 , s20
 
-	fstmias CO1, { s4 - s5 }
+	vstmia.f32 CO1, { s4 - s5 }
-	fstmias CO2, { s8 - s9  }
+	vstmia.f32 CO2, { s8 - s9  }
 
 	add	CO1, CO1, #8
 
@@ -716,7 +716,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds		s0, ALPHA_R
 	flds		s1, ALPHA_I
 
-	fldmias CO1, { s4 - s7 }
+	vldmia.f32 CO1, { s4 - s7 }
 
 	FADD_R	s16, s24 , s16
 	FADD_I  s17, s25 , s17
@@ -733,7 +733,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	FMAC_R2 s6 , s1 , s19
 	FMAC_I2	s7 , s1 , s18
 
-	fstmias CO1, { s4 - s7 }
+	vstmia.f32 CO1, { s4 - s7 }
 
 	add	CO1, CO1, #16
 
@@ -851,7 +851,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds		s0, ALPHA_R
 	flds		s1, ALPHA_I
 
-	fldmias CO1, { s4 - s5 }
+	vldmia.f32 CO1, { s4 - s5 }
 
 	FADD_R	s16, s24 , s16
 	FADD_I  s17, s25 , s17
@@ -861,7 +861,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	FMAC_R2 s4 , s1 , s17
 	FMAC_I2	s5 , s1 , s16
 
-	fstmias CO1, { s4 - s5 }
+	vstmia.f32 CO1, { s4 - s5 }
 
 	add	CO1, CO1, #8
 

kernel/arm/cgemm_ncopy_2_vfp.S

@@ -85,7 +85,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds	s6 , [ AO2, #8 ]
 	flds	s7 , [ AO2, #12 ]
 
-	fstmias	BO!, { s0 - s7 }
+	vstmia.f32	BO!, { s0 - s7 }
 	add	AO2, AO2, #16
 
 .endm
@@ -99,7 +99,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds	s3 , [ AO2, #4  ]
 
 	add	AO1, AO1, #8
-	fstmias	BO!, { s0 - s3 }
+	vstmia.f32	BO!, { s0 - s3 }
 	add	AO2, AO2, #8
 
 .endm
@@ -111,7 +111,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds	s2 , [ AO1, #8 ]
 	flds	s3 , [ AO1, #12 ]
 
-	fstmias	BO!, { s0 - s3 }
+	vstmia.f32	BO!, { s0 - s3 }
 	add	AO1, AO1, #16
 
 .endm
@@ -122,7 +122,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	flds	s0 , [ AO1, #0  ]
 	flds	s1 , [ AO1, #4  ]
 
-	fstmias	BO!, { s0 - s1 }
+	vstmia.f32	BO!, { s0 - s1 }
 	add	AO1, AO1, #8
 
 .endm

kernel/arm/cgemm_tcopy_2_vfp.S

@@ -73,12 +73,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 **************************************************************************************/
 .macro COPY2x2
 
-	fldmias	AO1, { s0 - s3 }
+	vldmia.f32	AO1, { s0 - s3 }
 
 	add	r3, AO1, LDA
-	fldmias	r3, { s4 - s7 }
+	vldmia.f32	r3, { s4 - s7 }
 
-	fstmias	BO1, { s0 - s7 }
+	vstmia.f32	BO1, { s0 - s7 }
 	add	AO1, AO1, #16
 	add	BO1, BO1, M4
 
@@ -86,12 +86,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro COPY1x2
 
-	fldmias	AO1, { s0 -s1 }
+	vldmia.f32	AO1, { s0 -s1 }
 
 	add	r3, AO1, LDA
-	fldmias	r3, { s2 - s3 }
+	vldmia.f32	r3, { s2 - s3 }
 
-	fstmias	BO2, { s0 - s3 }
+	vstmia.f32	BO2, { s0 - s3 }
 	add	AO1, AO1, #8
 	add	BO2, BO2, #16
 
@@ -100,9 +100,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /*************************************************************************************************************************/
 .macro COPY2x1
 
-	fldmias	AO1, { s0 - s3 }
+	vldmia.f32	AO1, { s0 - s3 }
 
-	fstmias	BO1, { s0 - s3 }
+	vstmia.f32	BO1, { s0 - s3 }
 	add	AO1, AO1, #16
 	add	BO1, BO1, M4
 
@@ -110,9 +110,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro COPY1x1
 
-	fldmias	AO1, { s0 - s1 }
+	vldmia.f32	AO1, { s0 - s1 }
 
-	fstmias	BO2, { s0 - s1 }
+	vstmia.f32	BO2, { s0 - s1 }
 	add	AO1, AO1, #8
 	add	BO2, BO2, #8
 

kernel/arm/cgemv_n_vfp.S

@@ -201,7 +201,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         flds            s0, ALPHA_R
         flds            s1, ALPHA_I
 
-        fldmias YO, { s4 - s7 }
+        vldmia.f32 YO, { s4 - s7 }
 
         FMAC_R1 s4 , s0 , s8
         FMAC_I1 s5 , s0 , s9
@@ -213,9 +213,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         FMAC_R2 s6 , s1 , s11
         FMAC_I2 s7 , s1 , s10
 
-        fstmias YO!, { s4 - s7 }
+        vstmia.f32 YO!, { s4 - s7 }
 
-        fldmias YO, { s4 - s7 }
+        vldmia.f32 YO, { s4 - s7 }
 
         FMAC_R1 s4 , s0 , s12
         FMAC_I1 s5 , s0 , s13
@@ -227,7 +227,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         FMAC_R2 s6 , s1 , s15
         FMAC_I2 s7 , s1 , s14
 
-        fstmias YO!, { s4 - s7 }
+        vstmia.f32 YO!, { s4 - s7 }
 
 .endm
 
@@ -266,14 +266,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         flds            s0, ALPHA_R
         flds            s1, ALPHA_I
 
-        fldmias YO, { s4 - s5 }
+        vldmia.f32 YO, { s4 - s5 }
 
         FMAC_R1 s4 , s0 , s8
         FMAC_I1 s5 , s0 , s9
         FMAC_R2 s4 , s1 , s9
         FMAC_I2 s5 , s1 , s8
 
-        fstmias YO, { s4 - s5 }
+        vstmia.f32 YO, { s4 - s5 }
 
         add     YO, YO, #8
 
@@ -349,47 +349,47 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         flds            s0, ALPHA_R
         flds            s1, ALPHA_I
 
-        fldmias YO, { s4 - s5 }
+        vldmia.f32 YO, { s4 - s5 }
 
         FMAC_R1 s4 , s0 , s8
         FMAC_I1 s5 , s0 , s9
         FMAC_R2 s4 , s1 , s9
         FMAC_I2 s5 , s1 , s8
 
-        fstmias YO, { s4 - s5 }
+        vstmia.f32 YO, { s4 - s5 }
 
 	add	YO, YO, INC_Y
 
-        fldmias YO, { s6 - s7 }
+        vldmia.f32 YO, { s6 - s7 }
 
         FMAC_R1 s6 , s0 , s10
         FMAC_I1 s7 , s0 , s11
         FMAC_R2 s6 , s1 , s11
         FMAC_I2 s7 , s1 , s10
 
-        fstmias YO, { s6 - s7 }
+        vstmia.f32 YO, { s6 - s7 }
 
 	add	YO, YO, INC_Y
 
-        fldmias YO, { s4 - s5 }
+        vldmia.f32 YO, { s4 - s5 }
 
         FMAC_R1 s4 , s0 , s12
         FMAC_I1 s5 , s0 , s13
         FMAC_R2 s4 , s1 , s13
         FMAC_I2 s5 , s1 , s12
 
-        fstmias YO, { s4 - s5 }
+        vstmia.f32 YO, { s4 - s5 }
 
 	add	YO, YO, INC_Y
 
-        fldmias YO, { s6 - s7 }
+        vldmia.f32 YO, { s6 - s7 }
 
         FMAC_R1 s6 , s0 , s14
         FMAC_I1 s7 , s0 , s15
         FMAC_R2 s6 , s1 , s15
         FMAC_I2 s7 , s1 , s14
 
-        fstmias YO, { s6 - s7 }
+        vstmia.f32 YO, { s6 - s7 }
 
 	add	YO, YO, INC_Y
 
@@ -430,14 +430,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         flds            s0, ALPHA_R
         flds            s1, ALPHA_I
 
-        fldmias YO, { s4 - s5 }
+        vldmia.f32 YO, { s4 - s5 }
 
         FMAC_R1 s4 , s0 , s8
         FMAC_I1 s5 , s0 , s9
         FMAC_R2 s4 , s1 , s9
         FMAC_I2 s5 , s1 , s8
 
-        fstmias YO, { s4 - s5 }
+        vstmia.f32 YO, { s4 - s5 }
 
         add     YO, YO, INC_Y