Merge pull request #1247 from martin-frbg/revert-cgroups-pr

Revert #1246, "honor cgroup/cpuset limits" for now

.gitignore

@@ -5,7 +5,6 @@
 *.def
 *.o
 *.out
-*.tmp
 lapack-3.1.1
 lapack-3.1.1.tgz
 lapack-3.4.1

.travis.yml

@@ -1,179 +1,5 @@
-# XXX: Precise is already deprecated, new default is Trusty.
-# https://blog.travis-ci.com/2017-07-11-trusty-as-default-linux-is-coming
-dist: precise
-sudo: true
 language: c
 
-jobs:
-  include:
-    - &test-ubuntu
-      os: linux
-      stage: test
-      compiler: gcc
-      addons:
-        apt:
-          packages:
-            - gfortran
-      before_script: &common-before
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
-      script:
-        - set -e
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-        - make -C test $COMMON_FLAGS $BTYPE
-        - make -C ctest $COMMON_FLAGS $BTYPE
-        - make -C utest $COMMON_FLAGS $BTYPE
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64"
-
-    - <<: *test-ubuntu
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 USE_OPENMP=1"
-
-    - <<: *test-ubuntu
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 INTERFACE64=1"
-
-    - <<: *test-ubuntu
-      compiler: clang
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 CC=clang"
-
-    - <<: *test-ubuntu
-      compiler: clang
-      env:
-        - TARGET_BOX=LINUX64
-        - BTYPE="BINARY=64 INTERFACE64=1 CC=clang"
-
-    - <<: *test-ubuntu
-      addons:
-        apt:
-          packages:
-            - gcc-multilib
-            - gfortran-multilib
-      env:
-        - TARGET_BOX=LINUX32
-        - BTYPE="BINARY=32"
-
-    - os: linux
-      stage: test
-      compiler: gcc
-      addons:
-        apt:
-          packages:
-            - binutils-mingw-w64-x86-64
-            - gcc-mingw-w64-x86-64
-            - gfortran-mingw-w64-x86-64
-      before_script: *common-before
-      script:
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-      env:
-        - TARGET_BOX=WIN64
-        - BTYPE="BINARY=64 HOSTCC=gcc CC=x86_64-w64-mingw32-gcc FC=x86_64-w64-mingw32-gfortran"
-
-    # Build & test on Alpine Linux inside chroot, i.e. on system with musl libc.
-    # These jobs needs sudo, so Travis runs them on VM-based infrastructure
-    # which is slower than container-based infrastructure used for jobs
-    # that don't require sudo.
-    - &test-alpine
-      os: linux
-      stage: test
-      dist: trusty
-      sudo: true
-      language: minimal
-      before_install:
-        - "wget 'https://raw.githubusercontent.com/alpinelinux/alpine-chroot-install/v0.6.0/alpine-chroot-install' \
-              && echo 'a827a4ba3d0817e7c88bae17fe34e50204983d1e  alpine-chroot-install' | sha1sum -c || exit 1"
-        - alpine() { /alpine/enter-chroot -u "$USER" "$@"; }
-      install:
-        - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
-      before_script: *common-before
-      script:
-        - set -e
-        # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
-        - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
-              CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
-        - alpine make -C test $COMMON_FLAGS $BTYPE
-        - alpine make -C ctest $COMMON_FLAGS $BTYPE
-        - alpine make -C utest $COMMON_FLAGS $BTYPE
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64"
-
-    # XXX: This job segfaults in TESTS OF THE COMPLEX LEVEL 3 BLAS,
-    # but only on Travis CI, cannot reproduce it elsewhere.
-    #- &test-alpine-openmp
-    #  <<: *test-alpine
-    #  env:
-    #    - TARGET_BOX=LINUX64_MUSL
-    #    - BTYPE="BINARY=64 USE_OPENMP=1"
-
-    - <<: *test-alpine
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64 INTERFACE64=1"
-
-    # Build with the same flags as Alpine do in OpenBLAS package.
-    - <<: *test-alpine
-      env:
-        - TARGET_BOX=LINUX64_MUSL
-        - BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=core2"
-
-    - &test-cmake
-      os: linux
-      stage: test
-      compiler: clang
-      addons:
-        apt:
-          packages:
-            - gfortran
-            - cmake
-      dist: trusty
-      sudo: true
-      before_script:
-        - COMMON_ARGS="-DTARGET=NEHALEM -DNUM_THREADS=32"
-      script:
-        - set -e
-        - mkdir build
-        - CONFIG=Release
-        - cmake -Bbuild -H. $CMAKE_ARGS $COMMON_ARGS -DCMAKE_BUILD_TYPE=$CONFIG
-        - cmake --build build --config $CONFIG -- -j2
-      env:
-        - CMAKE=1
-    - <<: *test-cmake
-      env:
-        - CMAKE=1 CMAKE_ARGS="-DNOFORTRAN=1"
-    - <<: *test-cmake
-      compiler: gcc
-      env:
-        - CMAKE=1
-
-    - &test-macos
-      os: osx
-      stage: test
-      osx_image: xcode8
-      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 $BTYPE
-      env:
-        - BTYPE="BINARY=64 INTERFACE64=1"
-
-    - <<: *test-macos
-      env:
-        - BTYPE="BINARY=32"
-
-# whitelist
-branches:
-  only:
-    - master
-    - develop
-
 notifications:
   webhooks:
     urls:
@@ -181,3 +7,32 @@ notifications:
     on_success: change  # options: [always|never|change] default: always
     on_failure: always  # options: [always|never|change] default: always
     on_start: never     # options: [always|never|change] default: always
+
+compiler:
+  - gcc
+
+env:
+  - TARGET_BOX=LINUX64 BTYPE="BINARY=64"
+  - TARGET_BOX=LINUX64 BTYPE="BINARY=64 USE_OPENMP=1"
+  - TARGET_BOX=LINUX64 BTYPE="BINARY=64 INTERFACE64=1"
+  - TARGET_BOX=LINUX32 BTYPE="BINARY=32"
+  - TARGET_BOX=WIN64 BTYPE="BINARY=64 HOSTCC=gcc CC=x86_64-w64-mingw32-gcc FC=x86_64-w64-mingw32-gfortran"
+
+before_install:
+ - sudo apt-get update -qq
+ - sudo apt-get install -qq gfortran
+ - if [[ "$TARGET_BOX" == "WIN64" ]]; then sudo apt-get install -qq binutils-mingw-w64-x86-64 gcc-mingw-w64-x86-64 gfortran-mingw-w64-x86-64; fi
+ - if [[ "$TARGET_BOX" == "LINUX32" ]]; then sudo apt-get install -qq  gcc-multilib gfortran-multilib; fi
+
+script: 
+ - set -e
+ - make QUIET_MAKE=1 DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32 $BTYPE
+ - if [ "$TARGET_BOX" == "LINUX32" ] || [ "$TARGET_BOX" == "LINUX64" ]; then make -C test DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32 $BTYPE; fi
+ - if [ "$TARGET_BOX" == "LINUX32" ] || [ "$TARGET_BOX" == "LINUX64" ]; then make -C ctest DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32 $BTYPE; fi
+ - if [ "$TARGET_BOX" == "LINUX32" ] || [ "$TARGET_BOX" == "LINUX64" ]; then make -C utest DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32 $BTYPE; fi
+
+# whitelist
+branches:
+  only:
+    - master
+    - develop

CMakeLists.txt

@@ -3,34 +3,52 @@
 ##
 
 cmake_minimum_required(VERSION 2.8.5)
-project(OpenBLAS C ASM)
+project(OpenBLAS)
 set(OpenBLAS_MAJOR_VERSION 0)
-set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 2)
+set(OpenBLAS_MINOR_VERSION 2)
+set(OpenBLAS_PATCH_VERSION 20)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
+enable_language(ASM)
+enable_language(C)
+
 # Adhere to GNU filesystem layout conventions
 include(GNUInstallDirs)
 
-include(CMakePackageConfigHelpers)
-
-
+if(MSVC)
+set(OpenBLAS_LIBNAME libopenblas)
+else()
 set(OpenBLAS_LIBNAME openblas)
+endif()
 
 #######
 if(MSVC)
 option(BUILD_WITHOUT_LAPACK "Without LAPACK and LAPACKE (Only BLAS or CBLAS)" ON)
 endif()
 option(BUILD_WITHOUT_CBLAS "Without CBLAS" OFF)
-option(DYNAMIC_ARCH "Build with DYNAMIC_ARCH" OFF)
-option(DYNAMIC_OLDER "Support older cpus with DYNAMIC_ARCH" OFF)
-option(BUILD_RELAPACK "Build with ReLAPACK (recursive LAPACK" OFF)
+option(BUILD_DEBUG "Build Debug Version" OFF)
 #######
 if(BUILD_WITHOUT_LAPACK)
 set(NO_LAPACK 1)
 set(NO_LAPACKE 1)
 endif()
 
+if(CMAKE_CONFIGURATION_TYPES) # multiconfig generator?
+        set(CMAKE_CONFIGURATION_TYPES "Debug;Release" CACHE STRING "" FORCE)
+        set(CMAKE_BUILD_TYPE
+	        Debug Debug
+		Release Release
+	)
+else()
+	if( NOT CMAKE_BUILD_TYPE )
+		if(BUILD_DEBUG)
+			set(CMAKE_BUILD_TYPE Debug)
+		else()
+			set(CMAKE_BUILD_TYPE Release)
+		endif()
+	endif()
+endif()
+
 if(BUILD_WITHOUT_CBLAS)
 set(NO_CBLAS 1)
 endif()
@@ -56,9 +74,6 @@ endif ()
 set(SUBDIRS	${BLASDIRS})
 if (NOT NO_LAPACK)
   list(APPEND SUBDIRS lapack)
-  if(BUILD_RELAPACK)
-    list(APPEND SUBDIRS relapack/src)
-  endif()
 endif ()
 
 # set which float types we want to build for
@@ -70,10 +85,6 @@ if (NOT DEFINED BUILD_SINGLE AND NOT DEFINED BUILD_DOUBLE AND NOT DEFINED BUILD_
   set(BUILD_COMPLEX16 true)
 endif ()
 
-if (NOT DEFINED BUILD_MATGEN)
-  set(BUILD_MATGEN true)
-endif()
-
 set(FLOAT_TYPES "")
 if (BUILD_SINGLE)
   message(STATUS "Building Single Precision")
@@ -95,10 +106,19 @@ if (BUILD_COMPLEX16)
   list(APPEND FLOAT_TYPES "ZCOMPLEX") # defines COMPLEX and DOUBLE
 endif ()
 
+set(SUBDIRS_ALL ${SUBDIRS} test ctest utest exports benchmark ../laswp ../bench)
+
+# all :: libs netlib tests shared
+
+# libs :
 if (NOT DEFINED CORE OR "${CORE}" STREQUAL "UNKNOWN")
   message(FATAL_ERROR "Detecting CPU failed. Please set TARGET explicitly, e.g. make TARGET=your_cpu_target. Please read README for details.")
 endif ()
 
+if (${NO_STATIC} AND ${NO_SHARED})
+  message(FATAL_ERROR "Neither static nor shared are enabled.")
+endif ()
+
 #Set default output directory
 set( CMAKE_LIBRARY_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
 set( CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
@@ -120,51 +140,23 @@ endforeach ()
 # Not using add_subdirectory here because lapack-netlib already has its own CMakeLists.txt. Instead include a cmake script with the sources we want.
 if (NOT NOFORTRAN AND NOT NO_LAPACK)
   include("${PROJECT_SOURCE_DIR}/cmake/lapack.cmake")
-  if (NOT NO_LAPACKE)
-    include("${PROJECT_SOURCE_DIR}/cmake/lapacke.cmake")
-  endif ()
+if (NOT NO_LAPACKE)
+  include("${PROJECT_SOURCE_DIR}/cmake/lapacke.cmake")
+endif ()
 endif ()
 
 # Only generate .def for dll on MSVC and always produce pdb files for debug and release
 if(MSVC)
-  if (${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION} LESS 3.4)
-    set(OpenBLAS_DEF_FILE "${PROJECT_BINARY_DIR}/openblas.def")
-  endif()
-  set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} /Zi")
-  set(CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} /DEBUG /OPT:REF /OPT:ICF")
+set(OpenBLAS_DEF_FILE "${PROJECT_BINARY_DIR}/openblas.def")
+set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Zi")
+set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} /Zi")
+set(CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} /DEBUG /OPT:REF /OPT:ICF")
 endif()
 
-if (${DYNAMIC_ARCH})
-  add_subdirectory(kernel)
-  foreach(TARGET_CORE ${DYNAMIC_CORE})
-    message("${TARGET_CORE}")
-    list(APPEND TARGET_OBJS "$<TARGET_OBJECTS:kernel_${TARGET_CORE}>")
-  endforeach()
-endif ()
-
-# Only build shared libs for MSVC
-if (MSVC)
-  set(BUILD_SHARED_LIBS ON)
-endif()
-
-
 # add objects to the openblas lib
-add_library(${OpenBLAS_LIBNAME} ${LA_SOURCES} ${LAPACKE_SOURCES} ${RELA_SOURCES} ${TARGET_OBJS} ${OpenBLAS_DEF_FILE})
+add_library(${OpenBLAS_LIBNAME} SHARED ${LA_SOURCES} ${LAPACKE_SOURCES} ${TARGET_OBJS} ${OpenBLAS_DEF_FILE})
 
-# Android needs to explicitly link against libm
-if(ANDROID)
-  target_link_libraries(${OpenBLAS_LIBNAME} m)
-endif()
-
-# Handle MSVC exports
-if(MSVC AND BUILD_SHARED_LIBS)
-  if (${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION} LESS 3.4)
-    include("${PROJECT_SOURCE_DIR}/cmake/export.cmake")
-  else()
-    # Creates verbose .def file (51KB vs 18KB)
-    set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES WINDOWS_EXPORT_ALL_SYMBOLS true)
-  endif()
-endif()
+include("${PROJECT_SOURCE_DIR}/cmake/export.cmake")
 
 # Set output for libopenblas
 set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
@@ -172,87 +164,110 @@ set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES LIBRARY_OUTPUT_NAME_DEBUG
 
 foreach (OUTPUTCONFIG ${CMAKE_CONFIGURATION_TYPES})
   string( TOUPPER ${OUTPUTCONFIG} OUTPUTCONFIG )
-
+  
   set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
   set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES LIBRARY_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
   set_target_properties( ${OpenBLAS_LIBNAME} PROPERTIES ARCHIVE_OUTPUT_DIRECTORY_${OUTPUTCONFIG} ${PROJECT_BINARY_DIR}/lib/${OUTPUTCONFIG} )
 endforeach()
 
 enable_testing()
+add_subdirectory(utest)
 
-if (USE_THREAD)
-  # Add threading library to linker
-  find_package(Threads)
-  if (THREADS_HAVE_PTHREAD_ARG)
-    set_property(TARGET ${OpenBLAS_LIBNAME} PROPERTY COMPILE_OPTIONS "-pthread")
-    set_property(TARGET ${OpenBLAS_LIBNAME} PROPERTY INTERFACE_COMPILE_OPTIONS "-pthread")
-  endif()
-  target_link_libraries(${OpenBLAS_LIBNAME} ${CMAKE_THREAD_LIBS_INIT})
+if (NOT MSVC)
+	#only build shared library for MSVC
+
+	add_library(${OpenBLAS_LIBNAME}_static STATIC ${LA_SOURCES} ${LAPACKE_SOURCES} ${TARGET_OBJS})
+	set_target_properties(${OpenBLAS_LIBNAME}_static PROPERTIES OUTPUT_NAME ${OpenBLAS_LIBNAME})
+	set_target_properties(${OpenBLAS_LIBNAME}_static PROPERTIES CLEAN_DIRECT_OUTPUT 1)
+
+	if(SMP)
+	target_link_libraries(${OpenBLAS_LIBNAME} pthread)
+	target_link_libraries(${OpenBLAS_LIBNAME}_static pthread)
 endif()
 
-if (MSVC OR NOT NOFORTRAN)
-  # Broken without fortran on unix
-  add_subdirectory(utest)
+#build test and ctest
+add_subdirectory(test)
+if(NOT NO_CBLAS)
+add_subdirectory(ctest)
+endif()
 endif()
 
-if (NOT MSVC AND NOT NOFORTRAN)
-  # Build test and ctest
-  add_subdirectory(test)
-  if(NOT NO_CBLAS)
-    add_subdirectory(ctest)
-  endif()
-endif()
-
-set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES
+set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES 
   VERSION ${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}
   SOVERSION ${OpenBLAS_MAJOR_VERSION}
 )
 
+
+# TODO: Why is the config saved here? Is this necessary with CMake?
+#Save the config files for installation
+#	@cp Makefile.conf Makefile.conf_last
+#	@cp config.h config_last.h
+#ifdef QUAD_PRECISION
+#	@echo "#define QUAD_PRECISION">> config_last.h
+#endif
+#ifeq ($(EXPRECISION), 1)
+#	@echo "#define EXPRECISION">> config_last.h
+#endif
+###
+#ifeq ($(DYNAMIC_ARCH), 1)
+#	@$(MAKE) -C kernel commonlibs || exit 1
+#	@for d in $(DYNAMIC_CORE) ; \
+#	do  $(MAKE) GOTOBLAS_MAKEFILE= -C kernel TARGET_CORE=$$d kernel || exit 1 ;\
+#	done
+#	@echo DYNAMIC_ARCH=1 >> Makefile.conf_last
+#endif
+#ifdef USE_THREAD
+#	@echo USE_THREAD=$(USE_THREAD) >>  Makefile.conf_last
+#endif
+#	@touch lib.grd
+
 # Install project
 
 # Install libraries
 install(TARGETS ${OpenBLAS_LIBNAME}
-	EXPORT "OpenBLASTargets"
 	RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
 	ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
-  LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR} )
+	LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR} )
 
-message(STATUS "Generating openblas_config.h in ${CMAKE_INSTALL_INCLUDEDIR}")
+# Install include files
+	set (GENCONFIG_BIN ${CMAKE_BINARY_DIR}/gen_config_h${CMAKE_EXECUTABLE_SUFFIX})
+	ADD_CUSTOM_COMMAND(
+	OUTPUT ${CMAKE_BINARY_DIR}/openblas_config.h
+	DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/config.h
+	COMMAND ${GENCONFIG_BIN} ${CMAKE_CURRENT_SOURCE_DIR}/config.h ${CMAKE_CURRENT_SOURCE_DIR}/openblas_config_template.h > ${CMAKE_BINARY_DIR}/openblas_config.h
+	)
 
-set(OPENBLAS_CONFIG_H ${CMAKE_BINARY_DIR}/openblas_config.h)
-file(WRITE  ${OPENBLAS_CONFIG_H} "#ifndef OPENBLAS_CONFIG_H\n")
-file(APPEND ${OPENBLAS_CONFIG_H} "#define OPENBLAS_CONFIG_H\n")
-file(STRINGS ${PROJECT_BINARY_DIR}/config.h __lines)
-foreach(line ${__lines})
-  string(REPLACE "#define " "" line ${line})
-  file(APPEND ${OPENBLAS_CONFIG_H} "#define OPENBLAS_${line}\n")
-endforeach()
-file(APPEND ${OPENBLAS_CONFIG_H} "#define OPENBLAS_VERSION \"OpenBLAS ${OpenBLAS_VERSION}\"\n")
-file(READ   ${CMAKE_CURRENT_SOURCE_DIR}/openblas_config_template.h OPENBLAS_CONFIG_TEMPLATE_H_CONTENTS)
-file(APPEND ${OPENBLAS_CONFIG_H} "${OPENBLAS_CONFIG_TEMPLATE_H_CONTENTS}\n")
-file(APPEND ${OPENBLAS_CONFIG_H} "#endif /* OPENBLAS_CONFIG_H */\n")
-install (FILES ${OPENBLAS_CONFIG_H} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
+	ADD_CUSTOM_TARGET(genconfig
+	ALL
+	DEPENDS openblas_config.h
+	)
+	add_dependencies(genconfig ${OpenBLAS_LIBNAME})
 
-if(NOT NOFORTRAN)
-  message(STATUS "Generating f77blas.h in ${CMAKE_INSTALL_INCLUDEDIR}")
+	install (FILES ${CMAKE_BINARY_DIR}/openblas_config.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 
-  set(F77BLAS_H ${CMAKE_BINARY_DIR}/f77blas.h)
-  file(WRITE  ${F77BLAS_H} "#ifndef OPENBLAS_F77BLAS_H\n")
-  file(APPEND ${F77BLAS_H} "#define OPENBLAS_F77BLAS_H\n")
-  file(APPEND ${F77BLAS_H} "#include \"openblas_config.h\"\n")
-  file(READ ${CMAKE_CURRENT_SOURCE_DIR}/common_interface.h COMMON_INTERFACE_H_CONTENTS)
-  file(APPEND ${F77BLAS_H} "${COMMON_INTERFACE_H_CONTENTS}\n")
-  file(APPEND ${F77BLAS_H} "#endif")
-  install (FILES ${F77BLAS_H} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
-endif()
+	message(STATUS "Generating f77blas.h in ${CMAKE_INSTALL_INCLUDEDIR}")
+
+	ADD_CUSTOM_TARGET(genf77blas
+	ALL
+	COMMAND ${AWK} 'BEGIN{print \"\#ifndef OPENBLAS_F77BLAS_H\" \; print \"\#define OPENBLAS_F77BLAS_H\" \; print \"\#include \\"openblas_config.h\\" \"}; NF {print}; END{print \"\#endif\"}' ${CMAKE_CURRENT_SOURCE_DIR}/common_interface.h > ${CMAKE_BINARY_DIR}/f77blas.h
+	DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/config.h
+	)
+	add_dependencies(genf77blas ${OpenBLAS_LIBNAME})
+
+	install (FILES ${CMAKE_BINARY_DIR}/f77blas.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 
 if(NOT NO_CBLAS)
 	message (STATUS "Generating cblas.h in ${CMAKE_INSTALL_INCLUDEDIR}")
 
-	file(READ ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h CBLAS_H_CONTENTS)
-	string(REPLACE "common" "openblas_config" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
-	file(WRITE ${CMAKE_BINARY_DIR}/cblas.tmp "${CBLAS_H_CONTENTS_NEW}")
-	install (FILES ${CMAKE_BINARY_DIR}/cblas.tmp DESTINATION ${CMAKE_INSTALL_INCLUDEDIR} RENAME cblas.h)
+	ADD_CUSTOM_TARGET(gencblas
+	ALL
+	COMMAND ${SED} 's/common/openblas_config/g' ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h > "${CMAKE_BINARY_DIR}/cblas.tmp"
+	COMMAND cp "${CMAKE_BINARY_DIR}/cblas.tmp" "${CMAKE_BINARY_DIR}/cblas.h"
+	DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h
+	)
+	add_dependencies(gencblas ${OpenBLAS_LIBNAME})
+
+	install (FILES ${CMAKE_BINARY_DIR}/cblas.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 endif()
 
 if(NOT NO_LAPACKE)
@@ -260,33 +275,19 @@ if(NOT NO_LAPACKE)
 	add_dependencies( ${OpenBLAS_LIBNAME} genlapacke)
 	FILE(GLOB_RECURSE INCLUDE_FILES "${CMAKE_CURRENT_SOURCE_DIR}/lapack-netlib/LAPACKE/*.h")
 	install (FILES ${INCLUDE_FILES} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
-
+ 
 	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"
+	COMMAND cp ${CMAKE_CURRENT_SOURCE_DIR}/lapack-netlib/LAPACKE/include/lapacke_mangling_with_flags.h.in "${CMAKE_BINARY_DIR}/lapacke_mangling.h"
 	)
 	install (FILES ${CMAKE_BINARY_DIR}/lapacke_mangling.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 endif()
+ 
+if(NOT MSVC)
+	install (TARGETS ${OpenBLAS_LIBNAME}_static DESTINATION ${CMAKE_INSTALL_LIBDIR})
+endif()
 
 include(FindPkgConfig QUIET)
 if(PKG_CONFIG_FOUND)
 	configure_file(${PROJECT_SOURCE_DIR}/cmake/openblas.pc.in ${PROJECT_BINARY_DIR}/openblas.pc @ONLY)
 	install (FILES ${PROJECT_BINARY_DIR}/openblas.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}")
-configure_package_config_file(cmake/${PN}Config.cmake.in
-                              "${CMAKE_CURRENT_BINARY_DIR}/${PN}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}Config.cmake
-              ${CMAKE_CURRENT_BINARY_DIR}/${PN}ConfigVersion.cmake
-        DESTINATION ${CMAKECONFIG_INSTALL_DIR})
-install(EXPORT "${PN}Targets"
-        NAMESPACE "${PN}::"
-        DESTINATION ${CMAKECONFIG_INSTALL_DIR})
-

CONTRIBUTORS.md

@@ -166,5 +166,5 @@ In chronological order:
   * [2017-01-01] dgemm and dtrmm kernels for IBM z13
   * [2017-02-26] ztrmm kernel for IBM z13
   * [2017-03-13] strmm and ctrmm kernel for IBM z13
-  * [2017-09-01] initial Blas Level-1,2 (double precision) for IBM z13
+
 

Makefile

@@ -21,17 +21,6 @@ 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
@@ -58,7 +47,7 @@ endif
 endif
 
 	@echo "  C compiler       ... $(C_COMPILER)  (command line : $(CC))"
-ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
+ifndef NOFORTRAN
 	@echo "  Fortran compiler ... $(F_COMPILER)  (command line : $(FC))"
 endif
 ifneq ($(OSNAME), AIX)
@@ -102,7 +91,11 @@ ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android))
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
-ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
+ifeq ($(OSNAME), FreeBSD)
+	@$(MAKE) -C exports so
+	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
+endif
+ifeq ($(OSNAME), NetBSD)
 	@$(MAKE) -C exports so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
 endif
@@ -119,7 +112,7 @@ endif
 endif
 
 tests :
-ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
+ifndef NOFORTRAN
 	touch $(LIBNAME)
 ifndef NO_FBLAS
 	$(MAKE) -C test all
@@ -164,9 +157,6 @@ ifeq ($(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
@@ -221,7 +211,7 @@ netlib :
 
 else
 netlib : lapack_prebuild
-ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
+ifndef NOFORTRAN
 	@$(MAKE) -C $(NETLIB_LAPACK_DIR) lapacklib
 	@$(MAKE) -C $(NETLIB_LAPACK_DIR) tmglib
 endif
@@ -242,7 +232,7 @@ prof_lapack : lapack_prebuild
 	@$(MAKE) -C $(NETLIB_LAPACK_DIR) lapack_prof
 
 lapack_prebuild :
-ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
+ifndef NOFORTRAN
 	-@echo "FORTRAN     = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "OPTS        = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "POPTS       = $(LAPACK_FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
@@ -252,7 +242,7 @@ ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	-@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 "ARCHFLAGS   = $(ARFLAGS) -ru" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "ARCHFLAGS   = -ru" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "RANLIB      = $(RANLIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "LAPACKLIB   = ../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "TMGLIB      = ../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
@@ -280,26 +270,25 @@ endif
 ifeq ($(BUILD_LAPACK_DEPRECATED), 1)
 	-@echo "BUILD_DEPRECATED      = 1" >> $(NETLIB_LAPACK_DIR)/make.inc
 endif
-	-@echo "LAPACKE_WITH_TMG      = 1" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@cat  make.inc >> $(NETLIB_LAPACK_DIR)/make.inc
 endif
 
 large.tgz :
-ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
+ifndef NOFORTRAN
 	if [ ! -a $< ]; then
 	-wget http://www.netlib.org/lapack/timing/large.tgz;
 	fi
 endif
 
 timing.tgz :
-ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
+ifndef NOFORTRAN
 	if [ ! -a $< ]; then
 	-wget http://www.netlib.org/lapack/timing/timing.tgz;
 	fi
 endif
 
 lapack-timing : large.tgz timing.tgz
-ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
+ifndef 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
@@ -308,10 +297,9 @@ endif
 
 lapack-test :
 	(cd $(NETLIB_LAPACK_DIR)/TESTING && rm -f x* *.out)
-	$(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
+	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING xeigtstc  xeigtstd  xeigtsts  xeigtstz  xlintstc  xlintstd  xlintstds  xlintstrfd  xlintstrfz  xlintsts  xlintstz  xlintstzc xlintstrfs xlintstrfc
 ifneq ($(CROSS), 1)
-	( cd $(NETLIB_LAPACK_DIR)/INSTALL; make all; ./testlsame; ./testslamch; ./testdlamch; \
+	( cd $(NETLIB_LAPACK_DIR)/INSTALL; ./testlsame; ./testslamch; ./testdlamch; \
         ./testsecond; ./testdsecnd; ./testieee; ./testversion )
 	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r )
 endif
@@ -323,9 +311,9 @@ lapack-runtest:
 
 
 blas-test:
-	(cd $(NETLIB_LAPACK_DIR)/BLAS/TESTING && rm -f x* *.out)
+	(cd $(NETLIB_LAPACK_DIR)/BLAS && rm -f x* *.out)
 	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR) blas_testing
-	(cd $(NETLIB_LAPACK_DIR)/BLAS/TESTING && cat *.out)
+	(cd $(NETLIB_LAPACK_DIR)/BLAS && cat *.out)
 
 
 dummy :

Makefile.install

@@ -72,7 +72,12 @@ ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android))
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
-ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
+ifeq ($(OSNAME), FreeBSD)
+	@cp $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+	@cd "$(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) $(LIBPREFIX).so
@@ -85,23 +90,21 @@ ifeq ($(OSNAME), Darwin)
 endif
 ifeq ($(OSNAME), WINNT)
 	@-cp $(LIBDLLNAME) "$(DESTDIR)$(OPENBLAS_BINARY_DIR)"
-	@-cp $(IMPLIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
+	@-cp $(LIBDLLNAME).a "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 endif
 ifeq ($(OSNAME), CYGWIN_NT)
 	@-cp $(LIBDLLNAME) "$(DESTDIR)$(OPENBLAS_BINARY_DIR)"
-	@-cp $(IMPLIBNAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 endif
 endif
 
 
 #Generating openblas.pc
-	@echo Generating openblas.pc in "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)"
-	@echo 'libdir='$(OPENBLAS_LIBRARY_DIR) > "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
-	@echo 'includedir='$(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"
+	@echo Generating openblas.pc in $(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)
+	@echo 'libdir='$(OPENBLAS_LIBRARY_DIR) >> $(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc
+	@echo 'includedir='$(OPENBLAS_INCLUDE_DIR) >> $(DESTDIR)$(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
 
 
 #Generating OpenBLASConfig.cmake
@@ -111,7 +114,7 @@ endif
 
 ifndef NO_SHARED
 #ifeq logical or
-ifeq ($(OSNAME), $(filter $(OSNAME),Linux FreeBSD NetBSD OpenBSD DragonFly))
+ifeq ($(OSNAME), $(filter $(OSNAME),Linux FreeBSD NetBSD))
 	@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

@@ -17,10 +17,6 @@ ifdef CPUIDEMU
 EXFLAGS = -DCPUIDEMU -DVENDOR=99
 endif
 
-ifeq ($(TARGET), 1004K)
-TARGET_FLAGS = -mips32r2
-endif
-
 ifeq ($(TARGET), P5600)
 TARGET_FLAGS = -mips32r5
 endif
@@ -33,10 +29,6 @@ ifeq ($(TARGET), P6600)
 TARGET_FLAGS = -mips64r6
 endif
 
-ifeq ($(TARGET), I6500)
-TARGET_FLAGS = -mips64r6
-endif
-
 all: getarch_2nd
 	./getarch_2nd  0 >> $(TARGET_MAKE)
 	./getarch_2nd  1 >> $(TARGET_CONF)

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.2
+VERSION = 0.2.20
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
@@ -17,11 +17,6 @@ VERSION = 0.3.2
 # 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
@@ -60,26 +55,11 @@ VERSION = 0.3.2
 # 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
 
@@ -120,7 +100,7 @@ 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
@@ -164,11 +144,8 @@ NO_AFFINITY = 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. (Actually in recent versions this is a factor proportional to the
-# 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).
+# with single thread. You can use this flag to avoid the overhead of multi-threading
+# in small matrix sizes. The default value is 4.
 # GEMM_MULTITHREAD_THRESHOLD = 4
 
 # If you need santy check by comparing reference BLAS. It'll be very

Makefile.system

@@ -17,24 +17,15 @@ 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.
 
@@ -62,9 +53,6 @@ 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
@@ -98,9 +86,6 @@ 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
@@ -147,10 +132,6 @@ 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
@@ -194,17 +175,12 @@ endif
 
 endif
 
-ifndef NUM_PARALLEL
-NUM_PARALLEL = 1
-endif
-
 ifndef NUM_THREADS
 NUM_THREADS = $(NUM_CORES)
 endif
 
 ifeq ($(NUM_THREADS), 1)
 override USE_THREAD = 0
-override USE_OPENMP = 0
 endif
 
 ifdef USE_THREAD
@@ -248,12 +224,12 @@ endif
 
 ifeq ($(OSNAME), Darwin)
 ifndef MACOSX_DEPLOYMENT_TARGET
-export MACOSX_DEPLOYMENT_TARGET=10.8
+export MACOSX_DEPLOYMENT_TARGET=10.6
 endif
 MD5SUM = md5 -r
 endif
 
-ifneq (,$(findstring $(OSNAME), FreeBSD OpenBSD DragonFly))
+ifeq ($(OSNAME), FreeBSD)
 MD5SUM = md5 -r
 endif
 
@@ -327,7 +303,6 @@ endif
 ifeq ($(OSNAME), CYGWIN_NT)
 NEED_PIC = 0
 NO_EXPRECISION = 1
-OS_CYGWIN_NT = 1
 endif
 
 ifneq ($(OSNAME), WINNT)
@@ -447,7 +422,7 @@ CCOMMON_OPT    += -fopenmp
 endif
 
 ifeq ($(C_COMPILER), INTEL)
-CCOMMON_OPT    += -fopenmp
+CCOMMON_OPT    += -openmp
 endif
 
 ifeq ($(C_COMPILER), PGI)
@@ -472,37 +447,13 @@ DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
 endif
 
 ifeq ($(ARCH), x86_64)
-DYNAMIC_CORE = PRESCOTT CORE2 
-ifeq ($(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
+DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
 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
 
 # If DYNAMIC_CORE is not set, DYNAMIC_ARCH cannot do anything, so force it to empty
@@ -602,14 +553,9 @@ 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)
-FCOMMON_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)
 endif
 
 ifeq ($(CORE), I6400)
@@ -622,11 +568,6 @@ CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=p6600 $(MSA_FLAGS)
 FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=p6600 $(MSA_FLAGS)
 endif
 
-ifeq ($(CORE), I6500)
-CCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
-FCOMMON_OPT += -mips64r6 -mnan=2008 -mtune=i6400 $(MSA_FLAGS)
-endif
-
 ifeq ($(OSNAME), AIX)
 BINARY_DEFINED = 1
 endif
@@ -634,14 +575,12 @@ endif
 endif
 
 ifndef BINARY_DEFINED
-ifneq ($(OSNAME), AIX)
 ifdef BINARY64
 CCOMMON_OPT += -m64
 else
 CCOMMON_OPT += -m32
 endif
 endif
-endif
 
 endif
 
@@ -686,7 +625,6 @@ ifeq ($(F_COMPILER), G77)
 CCOMMON_OPT += -DF_INTERFACE_G77
 FCOMMON_OPT += -Wall
 ifndef NO_BINARY_MODE
-ifneq ($(OSNAME), AIX)
 ifdef BINARY64
 FCOMMON_OPT += -m64
 else
@@ -694,12 +632,10 @@ FCOMMON_OPT += -m32
 endif
 endif
 endif
-endif
 
 ifeq ($(F_COMPILER), G95)
 CCOMMON_OPT += -DF_INTERFACE_G95
 FCOMMON_OPT += -Wall
-ifneq ($(OSNAME), AIX)
 ifndef NO_BINARY_MODE
 ifdef BINARY64
 FCOMMON_OPT += -m64
@@ -708,7 +644,6 @@ FCOMMON_OPT += -m32
 endif
 endif
 endif
-endif
 
 ifeq ($(F_COMPILER), GFORTRAN)
 CCOMMON_OPT += -DF_INTERFACE_GFORT
@@ -729,20 +664,16 @@ FCOMMON_OPT += -mabi=32
 endif
 else
 ifdef BINARY64
-ifneq ($(OSNAME), AIX)
 FCOMMON_OPT += -m64
-endif
 ifdef INTERFACE64
 ifneq ($(INTERFACE64), 0)
 FCOMMON_OPT +=  -fdefault-integer-8
 endif
 endif
 else
-ifneq ($(OSNAME), AIX)
 FCOMMON_OPT += -m32
 endif
 endif
-endif
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -fopenmp
 endif
@@ -756,7 +687,7 @@ FCOMMON_OPT += -i8
 endif
 endif
 ifeq ($(USE_OPENMP), 1)
-FCOMMON_OPT += -fopenmp
+FCOMMON_OPT += -openmp
 endif
 endif
 
@@ -936,10 +867,6 @@ 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
@@ -962,10 +889,6 @@ ifeq ($(NO_AVX2), 1)
 CCOMMON_OPT	+= -DNO_AVX2
 endif
 
-ifeq ($(NO_AVX512), 1)
-CCOMMON_OPT	+= -DNO_AVX512
-endif
-
 ifdef SMP
 CCOMMON_OPT	+= -DSMP_SERVER
 
@@ -1012,8 +935,6 @@ 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
@@ -1027,15 +948,9 @@ SYMBOLSUFFIX =
 endif
 
 ifndef LIBNAMESUFFIX
-LIBNAMEBASE = $(SYMBOLPREFIX)openblas$(SYMBOLSUFFIX)
+LIBPREFIX = lib$(SYMBOLPREFIX)openblas$(SYMBOLSUFFIX)
 else
-LIBNAMEBASE = $(SYMBOLPREFIX)openblas$(SYMBOLSUFFIX)_$(LIBNAMESUFFIX)
-endif
-
-ifeq ($(OSNAME), CYGWIN_NT)
-LIBPREFIX = cyg$(LIBNAMEBASE)
-else
-LIBPREFIX = lib$(LIBNAMEBASE)
+LIBPREFIX = lib$(SYMBOLPREFIX)openblas$(SYMBOLSUFFIX)_$(LIBNAMESUFFIX)
 endif
 
 KERNELDIR	= $(TOPDIR)/kernel/$(ARCH)
@@ -1194,7 +1109,6 @@ endif
 
 
 LIBDLLNAME   = $(LIBPREFIX).dll
-IMPLIBNAME   = lib$(LIBNAMEBASE).dll.a
 LIBSONAME    = $(LIBNAME:.$(LIBSUFFIX)=.so)
 LIBDYNNAME   = $(LIBNAME:.$(LIBSUFFIX)=.dylib)
 LIBDEFNAME   = $(LIBNAME:.$(LIBSUFFIX)=.def)
@@ -1272,7 +1186,6 @@ 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

@@ -8,13 +8,6 @@ endif
 endif
 endif
 
-ifeq ($(CORE), SKYLAKEX)
-ifndef NO_AVX512
-CCOMMON_OPT += -march=skylake-avx512
-FCOMMON_OPT += -march=skylake-avx512
-endif
-endif
-
 ifeq ($(OSNAME), Interix)
 ARFLAGS		= -m x64
 endif

README.md

@@ -2,222 +2,175 @@
 
 [![Join the chat at https://gitter.im/xianyi/OpenBLAS](https://badges.gitter.im/Join%20Chat.svg)](https://gitter.im/xianyi/OpenBLAS?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge)
 
-Travis CI: [![Build Status](https://travis-ci.org/xianyi/OpenBLAS.svg?branch=develop)](https://travis-ci.org/xianyi/OpenBLAS)
+Travis CI: [![Build Status](https://travis-ci.org/xianyi/OpenBLAS.png?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 documentation on the OpenBLAS wiki pages: <http://github.com/xianyi/OpenBLAS/wiki>.
+Please read the documents on OpenBLAS wiki pages <http://github.com/xianyi/OpenBLAS/wiki>.
 
 ## Binary Packages
-
-We provide official binary packages for the following platform:
+We provide 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/
 
-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)
-
+Or, check out codes from git://github.com/xianyi/OpenBLAS.git
 ### Normal compile
-
-Simply invoking `make` (or `gmake` on BSD) will detect the CPU automatically.
-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`.
+  * type "make" to detect the CPU automatically.
+  or
+  * type "make TARGET=xxx" to set target CPU, e.g. "make TARGET=NEHALEM". The full target list is in file TargetList.txt.
 
 ### Cross compile
-
-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.
+Please set CC and FC with the cross toolchains. Then, set HOSTCC with your host C compiler. At last, set TARGET explicitly.
 
 Examples:
 
-* 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
-  ```
+On X86 box, compile this library for loongson3a CPU.
 
-* On an x86 box, compile this library for a loongson3a CPU with loongcc (based on Open64) compiler:
-  ```sh
-  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 BINARY=64 CC=mips64el-unknown-linux-gnu-gcc FC=mips64el-unknown-linux-gnu-gfortran HOSTCC=gcc TARGET=LOONGSON3A
+
+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
 
 ### Debug version
 
-A debug version can be built using `make DEBUG=1`.
+    make DEBUG=1
 
-### Compile with MASS support on Power CPU (optional)
+### Compile with MASS Support on Power CPU (Optional dependency)
 
-The [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 Fortran applications that are
-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:
+[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
+Fortran-language applications that are tuned for optimum performance on POWER architectures. OpenBLAS with MASS requires 64-bit, little-endian OS on POWER.
+The library can be installed as below -
 
-* 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
-  ```
+ * On Ubuntu:
 
-* On RHEL/CentOS:
-  ```sh
-  wget http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/rhel7/repodata/repomd.xml.key
-  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
-  ```
+    wget -q http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/ubuntu/public.gpg -O- | sudo apt-key add -</br>
+    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>
+    sudo apt-get update</br>
+    sudo apt-get install libxlmass-devel.8.1.5</br>
 
-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`.
+ * On RHEL/CentOS:
 
-### Install to a specific directory (optional)
+    wget http://public.dhe.ibm.com/software/server/POWER/Linux/xl-compiler/eval/ppc64le/rhel7/repodata/repomd.xml.key</br>
+    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>
 
-Use `PREFIX=` when invoking `make`, for example
+After installing MASS library, compile openblas with USE_MASS=1.
 
-```sh
-make install PREFIX=your_installation_directory
-```
+Example:
 
-The default installation directory is `/opt/OpenBLAS`.
+Compiling on Power8 with MASS support -
 
-## Supported CPUs and Operating Systems
+    make USE_MASS=1 TARGET=POWER8
 
-Please read `GotoBLAS_01Readme.txt`.
+### Install to the directory (optional)
 
-### Additional supported CPUs
+Example:
 
-#### x86/x86-64
+    make install PREFIX=your_installation_directory
 
+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.
 - **AMD Bobcat**: Used GotoBLAS2 Barcelona codes.
-- **AMD Bulldozer**: x86-64 ?GEMM FMA4 kernels. (Thanks to Werner Saar)
+- **AMD Bulldozer**: x86-64 ?GEMM FMA4 kernels. (Thank 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 )
 
-- **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
-
-- **ARMv8**: Experimental
+#### 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
-
-- **Z13**: Optimized Level-3 BLAS and Level-1,2 (double precision)
-
-### Supported OS
+#### IBM zEnterprise System:
+- **Z13**: Optimized Level-3 BLAS 
+     
 
+### Support OS:
 - **GNU/Linux**
-- **MinGW or Visual Studio (CMake)/Windows**: Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-use-OpenBLAS-in-Microsoft-Visual-Studio>.
-- **Darwin/macOS**: Experimental. Although GotoBLAS2 supports Darwin, we are not macOS experts.
-- **FreeBSD**: Supported by the community. We don't actively test the library on this OS.
-- **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>.
+- **MingWin 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.
+- **FreeBSD**: Supported by community. We didn't test the library on this OS.
+- **Android**: Supported by community. Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-build-OpenBLAS-for-Android>.
 
-## Usage
+## Usages
+Link with libopenblas.a or -lopenblas for shared library.
 
-Statically link with `libopenblas.a` or dynamically link with `-lopenblas` if OpenBLAS was
-compiled as a shared library.
+### Set the number of threads with environment variables.
 
-### Setting the number of threads using environment variables
+Examples:
 
-Environment variables are used to specify a maximum number of threads.
-For example,
+    export OPENBLAS_NUM_THREADS=4
 
-```sh
-export OPENBLAS_NUM_THREADS=4
-export GOTO_NUM_THREADS=4
-export OMP_NUM_THREADS=4
-```
+ or
 
-The priorities are `OPENBLAS_NUM_THREADS` > `GOTO_NUM_THREADS` > `OMP_NUM_THREADS`.
+    export GOTO_NUM_THREADS=4
 
-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`.
+ or
 
-### Setting the number of threads at runtime
+    export OMP_NUM_THREADS=4
 
-We provide the following functions to control the number of threads at runtime:
+The priorities are OPENBLAS_NUM_THREADS > GOTO_NUM_THREADS > OMP_NUM_THREADS.
 
-```c
-void goto_set_num_threads(int num_threads);
-void openblas_set_num_threads(int num_threads);
-```
+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.
 
-If you compile this library with `USE_OPENMP=1`, you should use the above functions too.
+### Set the number of threads on runtime.
 
-## Reporting bugs
+We provided the below functions to control the number of threads on runtime.
 
-Please submit an issue in https://github.com/xianyi/OpenBLAS/issues.
+    void goto_set_num_threads(int num_threads);
+
+    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
 
-## Change log
-
-Please see Changelog.txt to view the differences between OpenBLAS and GotoBLAS2 1.13 BSD version.
+## ChangeLog
+Please see Changelog.txt to obtain the differences between GotoBLAS2 1.13 BSD version.
 
 ## Troubleshooting
-
-* Please read the [FAQ](https://github.com/xianyi/OpenBLAS/wiki/Faq) first.
-* Please use GCC version 4.6 and above to compile Sandy Bridge AVX kernels on Linux/MinGW/BSD.
-* Please use Clang version 3.1 and above to compile the library on Sandy Bridge microarchitecture.
-  Clang 3.0 will generate the wrong AVX binary code.
-* 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.
+* 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 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.
+* 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 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).
+* 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.
 
 ## 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.
-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`.
+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. Write a test which shows that the bug was fixed or that the feature works as expected.
+1. 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

@@ -20,7 +20,6 @@ DUNNINGTON
 NEHALEM
 SANDYBRIDGE
 HASWELL
-SKYLAKEX
 ATOM
 
 b)AMD CPU:
@@ -57,7 +56,6 @@ CELL
 
 3.MIPS CPU:
 P5600
-1004K
 
 4.MIPS64 CPU:
 SICORTEX
@@ -65,7 +63,6 @@ LOONGSON3A
 LOONGSON3B
 I6400
 P6600
-I6500
 
 5.IA64 CPU:
 ITANIUM2
@@ -88,6 +85,3 @@ VULCAN
 THUNDERX
 THUNDERX2T99
 
-9.System Z:
-ZARCH_GENERIC
-Z13

USAGE.md

@@ -14,20 +14,6 @@ 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

appveyor.yml

@@ -5,8 +5,6 @@ version: 0.2.19.{build}
 platform: 
   - x64
 
-os: Visual Studio 2017
-
 configuration: Release
 
 clone_folder: c:\projects\OpenBLAS
@@ -14,53 +12,33 @@ clone_folder: c:\projects\OpenBLAS
 init:
   - git config --global core.autocrlf input
 
+build:
+  project: OpenBLAS.sln
+
 clone_depth: 5
 
+#branches to build
+branches:
+  only:
+    - master
+    - develop
+    - cmake
+
 skip_tags: true
 
 matrix:
-  fast_finish: false
+  fast_finish: true
 
 skip_commits:
 # Add [av skip] to commit messages
   message: /\[av skip\]/
 
-environment:
-  global:
-    CONDA_INSTALL_LOCN: C:\\Miniconda36-x64
-  matrix:
-    - COMPILER: clang-cl
-      WITH_FORTRAN: yes
-    - COMPILER: clang-cl
-      DYNAMIC_ARCH: ON
-      WITH_FORTRAN: no
-    - COMPILER: cl
-
-install:
-  - if [%COMPILER%]==[clang-cl] call %CONDA_INSTALL_LOCN%\Scripts\activate.bat
-  - if [%COMPILER%]==[clang-cl] conda config --add channels conda-forge --force
-  - if [%COMPILER%]==[clang-cl] conda install --yes --quiet clangdev cmake
-
-  - if [%WITH_FORTRAN%]==[no] conda install --yes --quiet ninja
-  - if [%WITH_FORTRAN%]==[yes] conda install --yes --quiet -c isuruf kitware-ninja
-  - if [%WITH_FORTRAN%]==[yes] conda install --yes --quiet flang
-
-  - if [%COMPILER%]==[clang-cl] call "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\vcvarsall.bat" x64
-  - if [%COMPILER%]==[clang-cl] set "LIB=%CONDA_INSTALL_LOCN%\Library\lib;%LIB%"
-  - if [%COMPILER%]==[clang-cl] set "CPATH=%CONDA_INSTALL_LOCN%\Library\include;%CPATH%"
-
 before_build:
-  - ps: if (-Not (Test-Path .\build)) { mkdir build }
-  - cd build
-  - if [%COMPILER%]==[cl] cmake -G "Visual Studio 15 2017 Win64" ..
-  - if [%WITH_FORTRAN%]==[no] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl ..
-  - if [%WITH_FORTRAN%]==[yes] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DBUILD_WITHOUT_LAPACK=no -DNOFORTRAN=0 ..
-  - if [%DYNAMIC_ARCH%]==[ON] cmake -DDYNAMIC_ARCH=ON ..
-
-build_script:
-  - cmake --build .
+  - echo Running cmake...
+  - cd c:\projects\OpenBLAS
+  - cmake -G "Visual Studio 12 Win64" .
 
 test_script:
   - echo Running Test
-  - cd utest
+  - cd c:\projects\OpenBLAS\utest
   - openblas_utest

benchmark/gemm.c

@@ -121,15 +121,13 @@ static void *huge_malloc(BLASLONG size){
 int main(int argc, char *argv[]){
 
   FLOAT *a, *b, *c;
-  FLOAT alpha[] = {1.0, 0.0};
+  FLOAT alpha[] = {1.0, 1.0};
   FLOAT beta [] = {0.0, 0.0};
-  char transa = 'N';
-  char transb = 'N';
-  blasint m, n, k, i, j, lda, ldb, ldc;
+  char trans='N';
+  blasint m, n, i, j;
   int loops = 1;
-  int has_param_m = 0;
-  int has_param_n = 0;
-  int has_param_k = 0;
+  int has_param_n=0;
+  int l;
   char *p;
 
   int from =   1;
@@ -137,108 +135,86 @@ int main(int argc, char *argv[]){
   int step =   1;
 
   struct timeval start, stop;
-  double time1, timeg;
+  double time1,timeg;
 
   argc--;argv++;
 
-  if (argc > 0) { from = atol(*argv);            argc--; argv++; }
-  if (argc > 0) { to   = MAX(atol(*argv), from); argc--; argv++; }
-  if (argc > 0) { step = atol(*argv);            argc--; argv++; }
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
 
-  if ((p = getenv("OPENBLAS_TRANS"))) {
-    transa=*p;
-    transb=*p;
-  }
-  if ((p = getenv("OPENBLAS_TRANSA"))) {
-    transa=*p;
-  }
-  if ((p = getenv("OPENBLAS_TRANSB"))) {
-    transb=*p;
-  }
-  TOUPPER(transa);
-  TOUPPER(transb);
+  if ((p = getenv("OPENBLAS_TRANS")))  trans=*p;
 
-  fprintf(stderr, "From : %3d  To : %3d Step=%d : Transa=%c : Transb=%c\n", from, to, step, transa, transb);
+  fprintf(stderr, "From : %3d  To : %3d Step=%d : Trans=%c\n", from, to, step, trans);
+
+  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( b = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( c = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
 
   p = getenv("OPENBLAS_LOOPS");
-  if ( p != NULL ) {
-    loops = atoi(p);
-  }
+  if ( p != NULL )
+	loops = atoi(p);
 
-  if ((p = getenv("OPENBLAS_PARAM_M"))) {
-    m = atoi(p);
-    has_param_m=1;
-  } else {
-    m = to;
-  }
   if ((p = getenv("OPENBLAS_PARAM_N"))) {
-    n = atoi(p);
-    has_param_n=1;
-  } else {
-    n = to;
-  }
-  if ((p = getenv("OPENBLAS_PARAM_K"))) {
-    k = atoi(p);
-    has_param_k=1;
-  } else {
-    k = to;
-  }
-
-  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * m * k * COMPSIZE)) == NULL) {
-    fprintf(stderr,"Out of Memory!!\n");exit(1);
-  }
-  if (( b = (FLOAT *)malloc(sizeof(FLOAT) * k * n * COMPSIZE)) == NULL) {
-    fprintf(stderr,"Out of Memory!!\n");exit(1);
-  }
-  if (( c = (FLOAT *)malloc(sizeof(FLOAT) * m * n * COMPSIZE)) == NULL) {
-    fprintf(stderr,"Out of Memory!!\n");exit(1);
+          n = atoi(p);
+	  has_param_n=1;	  
   }
 
 #ifdef linux
   srandom(getpid());
 #endif
-
-  for (i = 0; i < m * k * COMPSIZE; i++) {
-    a[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-  }
-  for (i = 0; i < k * n * COMPSIZE; i++) {
-    b[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-  }
-  for (i = 0; i < m * n * COMPSIZE; i++) {
-    c[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-  }
  
-  fprintf(stderr, "          SIZE                   Flops             Time\n");
+	for(j = 0; j < to; j++){
+      		for(i = 0; i < to * COMPSIZE; i++){
+			a[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+			b[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+			c[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+      		}
+    	}
+
+
+
+  fprintf(stderr, "   SIZE          Flops          Time\n");
+
+  for(m = from; m <= to; m += step)
+  {
 
-  for (i = from; i <= to; i += step) {
-    
     timeg=0;
 
-    if (!has_param_m) { m = i; }
-    if (!has_param_n) { n = i; }
-    if (!has_param_k) { k = i; }
+    if ( has_param_n == 1 && n <= m )
+    	n=n;
+    else
+    	n=m;
 
-    if (transa == 'N') { lda = m; }
-    else { lda = k; }
-    if (transb == 'N') { ldb = k; }
-    else { ldb = n; }
-    ldc = m;
 
-    fprintf(stderr, " M=%4d, N=%4d, K=%4d : ", (int)m, (int)n, (int)k);
+
+    fprintf(stderr, " %6dx%d : ", (int)m, (int)n);
     gettimeofday( &start, (struct timezone *)0);
 
-    for (j=0; j<loops; j++) {
-      GEMM (&transa, &transb, &m, &n, &k, alpha, a, &lda, b, &ldb, beta, c, &ldc);
-    }
+    for (l=0; l<loops; l++)
+    {
 
-    gettimeofday( &stop, (struct timezone *)0);
-    time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+    	GEMM (&trans, &trans, &m, &n, &m, alpha, a, &m, b, &m, beta, c, &m );
+
+
+
+    }
+   gettimeofday( &stop, (struct timezone *)0);
+   time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
     timeg = time1/loops;
     fprintf(stderr,
 	    " %10.2f MFlops %10.6f sec\n",
-	    COMPSIZE * COMPSIZE * 2. * (double)k * (double)m * (double)n / timeg * 1.e-6, time1);
-    
+	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)n / timeg * 1.e-6, time1);
+
   }
 
   return 0;

c_check

@@ -54,8 +54,6 @@ $compiler = GCC       if ($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/);
@@ -201,21 +199,6 @@ $architecture = zarch  if ($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 $tmpf "int main(void){ __asm__ volatile($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([_\.]*)(.*)/;
@@ -303,7 +286,6 @@ 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

@@ -57,50 +57,45 @@ double cblas_dsdot (OPENBLAS_CONST blasint n, OPENBLAS_CONST float *x, OPENBLAS_
 float  cblas_sdot(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float  *y, OPENBLAS_CONST blasint incy);
 double cblas_ddot(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST double *y, OPENBLAS_CONST blasint incy);
 
-openblas_complex_float  cblas_cdotu(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST void  *y, OPENBLAS_CONST blasint incy);
-openblas_complex_float  cblas_cdotc(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST void  *y, OPENBLAS_CONST blasint incy);
-openblas_complex_double cblas_zdotu(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST void *y, OPENBLAS_CONST blasint incy);
-openblas_complex_double cblas_zdotc(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST void *y, OPENBLAS_CONST blasint incy);
+openblas_complex_float  cblas_cdotu(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float  *y, OPENBLAS_CONST blasint incy);
+openblas_complex_float  cblas_cdotc(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float  *y, OPENBLAS_CONST blasint incy);
+openblas_complex_double cblas_zdotu(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST double *y, OPENBLAS_CONST blasint incy);
+openblas_complex_double cblas_zdotc(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST double *y, OPENBLAS_CONST blasint incy);
 
-void  cblas_cdotu_sub(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST void  *y, OPENBLAS_CONST blasint incy, void  *ret);
-void  cblas_cdotc_sub(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST void  *y, OPENBLAS_CONST blasint incy, void  *ret);
-void  cblas_zdotu_sub(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST void *y, OPENBLAS_CONST blasint incy, void *ret);
-void  cblas_zdotc_sub(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST void *y, OPENBLAS_CONST blasint incy, void *ret);
+void  cblas_cdotu_sub(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float  *y, OPENBLAS_CONST blasint incy, openblas_complex_float  *ret);
+void  cblas_cdotc_sub(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST float  *y, OPENBLAS_CONST blasint incy, openblas_complex_float  *ret);
+void  cblas_zdotu_sub(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST double *y, OPENBLAS_CONST blasint incy, openblas_complex_double *ret);
+void  cblas_zdotc_sub(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, OPENBLAS_CONST double *y, OPENBLAS_CONST blasint incy, openblas_complex_double *ret);
 
 float  cblas_sasum (OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx);
 double cblas_dasum (OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx);
-float  cblas_scasum(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx);
-double cblas_dzasum(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx);
+float  cblas_scasum(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx);
+double cblas_dzasum(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx);
 
 float  cblas_snrm2 (OPENBLAS_CONST blasint N, OPENBLAS_CONST float  *X, OPENBLAS_CONST blasint incX);
 double cblas_dnrm2 (OPENBLAS_CONST blasint N, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX);
-float  cblas_scnrm2(OPENBLAS_CONST blasint N, OPENBLAS_CONST void  *X, OPENBLAS_CONST blasint incX);
-double cblas_dznrm2(OPENBLAS_CONST blasint N, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX);
+float  cblas_scnrm2(OPENBLAS_CONST blasint N, OPENBLAS_CONST float  *X, OPENBLAS_CONST blasint incX);
+double cblas_dznrm2(OPENBLAS_CONST blasint N, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX);
 
 CBLAS_INDEX cblas_isamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx);
 CBLAS_INDEX cblas_idamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx);
-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);
+CBLAS_INDEX cblas_icamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx);
+CBLAS_INDEX cblas_izamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *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);
-void cblas_zaxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incy);
+void cblas_caxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *x, OPENBLAS_CONST blasint incx, float *y, OPENBLAS_CONST blasint incy);
+void cblas_zaxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, double *y, OPENBLAS_CONST blasint incy);
 
 void cblas_scopy(OPENBLAS_CONST blasint n, OPENBLAS_CONST float *x, OPENBLAS_CONST blasint incx, float *y, OPENBLAS_CONST blasint incy);
 void cblas_dcopy(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, double *y, OPENBLAS_CONST blasint incy);
-void cblas_ccopy(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incy);
-void cblas_zcopy(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incy);
+void cblas_ccopy(OPENBLAS_CONST blasint n, OPENBLAS_CONST float *x, OPENBLAS_CONST blasint incx, float *y, OPENBLAS_CONST blasint incy);
+void cblas_zcopy(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, double *y, OPENBLAS_CONST blasint incy);
 
 void cblas_sswap(OPENBLAS_CONST blasint n, float *x, OPENBLAS_CONST blasint incx, float *y, OPENBLAS_CONST blasint incy);
 void cblas_dswap(OPENBLAS_CONST blasint n, double *x, OPENBLAS_CONST blasint incx, double *y, OPENBLAS_CONST blasint incy);
-void cblas_cswap(OPENBLAS_CONST blasint n, void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incy);
-void cblas_zswap(OPENBLAS_CONST blasint n, void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incy);
+void cblas_cswap(OPENBLAS_CONST blasint n, float *x, OPENBLAS_CONST blasint incx, float *y, OPENBLAS_CONST blasint incy);
+void cblas_zswap(OPENBLAS_CONST blasint n, double *x, OPENBLAS_CONST blasint incx, double *y, OPENBLAS_CONST blasint incy);
 
 void cblas_srot(OPENBLAS_CONST blasint N, float *X, OPENBLAS_CONST blasint incX, float *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST float c, OPENBLAS_CONST float s);
 void cblas_drot(OPENBLAS_CONST blasint N, double *X, OPENBLAS_CONST blasint incX, double *Y, OPENBLAS_CONST blasint incY, OPENBLAS_CONST double c, OPENBLAS_CONST double  s);
@@ -116,59 +111,59 @@ void cblas_drotmg(double *d1, double *d2, double *b1, OPENBLAS_CONST double b2,
 
 void cblas_sscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, float *X, OPENBLAS_CONST blasint incX);
 void cblas_dscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, double *X, OPENBLAS_CONST blasint incX);
-void cblas_cscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, void *X, OPENBLAS_CONST blasint incX);
-void cblas_zscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, void *X, OPENBLAS_CONST blasint incX);
-void cblas_csscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, void *X, OPENBLAS_CONST blasint incX);
-void cblas_zdscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, void *X, OPENBLAS_CONST blasint incX);
+void cblas_cscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST float *alpha, float *X, OPENBLAS_CONST blasint incX);
+void cblas_zscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST double *alpha, double *X, OPENBLAS_CONST blasint incX);
+void cblas_csscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, float *X, OPENBLAS_CONST blasint incX);
+void cblas_zdscal(OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, double *X, OPENBLAS_CONST blasint incX);
 
 void cblas_sgemv(OPENBLAS_CONST enum CBLAS_ORDER order,  OPENBLAS_CONST enum CBLAS_TRANSPOSE trans,  OPENBLAS_CONST blasint m, OPENBLAS_CONST blasint n,
 		 OPENBLAS_CONST float alpha, OPENBLAS_CONST float  *a, OPENBLAS_CONST blasint lda,  OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx,  OPENBLAS_CONST float beta,  float  *y, OPENBLAS_CONST blasint incy);
 void cblas_dgemv(OPENBLAS_CONST enum CBLAS_ORDER order,  OPENBLAS_CONST enum CBLAS_TRANSPOSE trans,  OPENBLAS_CONST blasint m, OPENBLAS_CONST blasint n,
 		 OPENBLAS_CONST double alpha, OPENBLAS_CONST double  *a, OPENBLAS_CONST blasint lda,  OPENBLAS_CONST double  *x, OPENBLAS_CONST blasint incx,  OPENBLAS_CONST double beta,  double  *y, OPENBLAS_CONST blasint incy);
 void cblas_cgemv(OPENBLAS_CONST enum CBLAS_ORDER order,  OPENBLAS_CONST enum CBLAS_TRANSPOSE trans,  OPENBLAS_CONST blasint m, OPENBLAS_CONST blasint n,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void  *a, OPENBLAS_CONST blasint lda,  OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx,  OPENBLAS_CONST void *beta,  void  *y, OPENBLAS_CONST blasint incy);
+		 OPENBLAS_CONST float *alpha, OPENBLAS_CONST float  *a, OPENBLAS_CONST blasint lda,  OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx,  OPENBLAS_CONST float *beta,  float  *y, OPENBLAS_CONST blasint incy);
 void cblas_zgemv(OPENBLAS_CONST enum CBLAS_ORDER order,  OPENBLAS_CONST enum CBLAS_TRANSPOSE trans,  OPENBLAS_CONST blasint m, OPENBLAS_CONST blasint n,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void  *a, OPENBLAS_CONST blasint lda,  OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx,  OPENBLAS_CONST void *beta,  void  *y, OPENBLAS_CONST blasint incy);
+		 OPENBLAS_CONST double *alpha, OPENBLAS_CONST double  *a, OPENBLAS_CONST blasint lda,  OPENBLAS_CONST double  *x, OPENBLAS_CONST blasint incx,  OPENBLAS_CONST double *beta,  double  *y, OPENBLAS_CONST blasint incy);
 
 void cblas_sger (OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST float   alpha, OPENBLAS_CONST float  *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float  *Y, OPENBLAS_CONST blasint incY, float  *A, OPENBLAS_CONST blasint lda);
 void cblas_dger (OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST double  alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *Y, OPENBLAS_CONST blasint incY, double *A, OPENBLAS_CONST blasint lda);
-void cblas_cgeru(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST void  *alpha, OPENBLAS_CONST void  *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void  *Y, OPENBLAS_CONST blasint incY, void  *A, OPENBLAS_CONST blasint lda);
-void cblas_cgerc(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST void  *alpha, OPENBLAS_CONST void  *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void  *Y, OPENBLAS_CONST blasint incY, void  *A, OPENBLAS_CONST blasint lda);
-void cblas_zgeru(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *Y, OPENBLAS_CONST blasint incY, void *A, OPENBLAS_CONST blasint lda);
-void cblas_zgerc(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *Y, OPENBLAS_CONST blasint incY, void *A, OPENBLAS_CONST blasint lda);
+void cblas_cgeru(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST float  *alpha, OPENBLAS_CONST float  *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float  *Y, OPENBLAS_CONST blasint incY, float  *A, OPENBLAS_CONST blasint lda);
+void cblas_cgerc(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST float  *alpha, OPENBLAS_CONST float  *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float  *Y, OPENBLAS_CONST blasint incY, float  *A, OPENBLAS_CONST blasint lda);
+void cblas_zgeru(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *Y, OPENBLAS_CONST blasint incY, double *A, OPENBLAS_CONST blasint lda);
+void cblas_zgerc(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *Y, OPENBLAS_CONST blasint incY, double *A, OPENBLAS_CONST blasint lda);
 
 void cblas_strsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *X, OPENBLAS_CONST blasint incX);
 void cblas_dtrsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *X, OPENBLAS_CONST blasint incX);
-void cblas_ctrsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *X, OPENBLAS_CONST blasint incX);
-void cblas_ztrsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *X, OPENBLAS_CONST blasint incX);
+void cblas_ctrsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *X, OPENBLAS_CONST blasint incX);
+void cblas_ztrsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *X, OPENBLAS_CONST blasint incX);
 
 void cblas_strmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *X, OPENBLAS_CONST blasint incX);
 void cblas_dtrmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *X, OPENBLAS_CONST blasint incX);
-void cblas_ctrmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *X, OPENBLAS_CONST blasint incX);
-void cblas_ztrmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *X, OPENBLAS_CONST blasint incX);
+void cblas_ctrmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *X, OPENBLAS_CONST blasint incX);
+void cblas_ztrmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *X, OPENBLAS_CONST blasint incX);
 
 void cblas_ssyr(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, float *A, OPENBLAS_CONST blasint lda);
 void cblas_dsyr(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, double *A, OPENBLAS_CONST blasint lda);
-void cblas_cher(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, void *A, OPENBLAS_CONST blasint lda);
-void cblas_zher(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, void *A, OPENBLAS_CONST blasint lda);
+void cblas_cher(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, float *A, OPENBLAS_CONST blasint lda);
+void cblas_zher(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, double *A, OPENBLAS_CONST blasint lda);
 
 void cblas_ssyr2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo,OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *X,
                 OPENBLAS_CONST blasint incX, OPENBLAS_CONST float *Y, OPENBLAS_CONST blasint incY, float *A, OPENBLAS_CONST blasint lda);
 void cblas_dsyr2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *X,
                 OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *Y, OPENBLAS_CONST blasint incY, double *A, OPENBLAS_CONST blasint lda);
-void cblas_cher2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX,
-                OPENBLAS_CONST void *Y, OPENBLAS_CONST blasint incY, void *A, OPENBLAS_CONST blasint lda);
-void cblas_zher2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX,
-                OPENBLAS_CONST void *Y, OPENBLAS_CONST blasint incY, void *A, OPENBLAS_CONST blasint lda);
+void cblas_cher2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX,
+                OPENBLAS_CONST float *Y, OPENBLAS_CONST blasint incY, float *A, OPENBLAS_CONST blasint lda);
+void cblas_zher2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX,
+                OPENBLAS_CONST double *Y, OPENBLAS_CONST blasint incY, double *A, OPENBLAS_CONST blasint lda);
 
 void cblas_sgbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
                  OPENBLAS_CONST blasint KL, OPENBLAS_CONST blasint KU, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float beta, float *Y, OPENBLAS_CONST blasint incY);
 void cblas_dgbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
                  OPENBLAS_CONST blasint KL, OPENBLAS_CONST blasint KU, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double beta, double *Y, OPENBLAS_CONST blasint incY);
 void cblas_cgbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
-                 OPENBLAS_CONST blasint KL, OPENBLAS_CONST blasint KU, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *beta, void *Y, OPENBLAS_CONST blasint incY);
+                 OPENBLAS_CONST blasint KL, OPENBLAS_CONST blasint KU, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float *beta, float *Y, OPENBLAS_CONST blasint incY);
 void cblas_zgbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
-                 OPENBLAS_CONST blasint KL, OPENBLAS_CONST blasint KU, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *beta, void *Y, OPENBLAS_CONST blasint incY);
+                 OPENBLAS_CONST blasint KL, OPENBLAS_CONST blasint KU, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *beta, double *Y, OPENBLAS_CONST blasint incY);
 
 void cblas_ssbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A,
                  OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float beta, float *Y, OPENBLAS_CONST blasint incY);
@@ -181,45 +176,45 @@ void cblas_stbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLA
 void cblas_dtbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
                  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *X, OPENBLAS_CONST blasint incX);
 void cblas_ctbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
-                 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *X, OPENBLAS_CONST blasint incX);
+                 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *X, OPENBLAS_CONST blasint incX);
 void cblas_ztbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
-                 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *X, OPENBLAS_CONST blasint incX);
+                 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *X, OPENBLAS_CONST blasint incX);
 
 void cblas_stbsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
                  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *X, OPENBLAS_CONST blasint incX);
 void cblas_dtbsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
                  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *X, OPENBLAS_CONST blasint incX);
 void cblas_ctbsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
-                 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *X, OPENBLAS_CONST blasint incX);
+                 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *X, OPENBLAS_CONST blasint incX);
 void cblas_ztbsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
-                 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *X, OPENBLAS_CONST blasint incX);
+                 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *X, OPENBLAS_CONST blasint incX);
 
 void cblas_stpmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
                  OPENBLAS_CONST blasint N, OPENBLAS_CONST float *Ap, float *X, OPENBLAS_CONST blasint incX);
 void cblas_dtpmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
                  OPENBLAS_CONST blasint N, OPENBLAS_CONST double *Ap, double *X, OPENBLAS_CONST blasint incX);
 void cblas_ctpmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
-                 OPENBLAS_CONST blasint N, OPENBLAS_CONST void *Ap, void *X, OPENBLAS_CONST blasint incX);
+                 OPENBLAS_CONST blasint N, OPENBLAS_CONST float *Ap, float *X, OPENBLAS_CONST blasint incX);
 void cblas_ztpmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
-                 OPENBLAS_CONST blasint N, OPENBLAS_CONST void *Ap, void *X, OPENBLAS_CONST blasint incX);
+                 OPENBLAS_CONST blasint N, OPENBLAS_CONST double *Ap, double *X, OPENBLAS_CONST blasint incX);
 
 void cblas_stpsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
                  OPENBLAS_CONST blasint N, OPENBLAS_CONST float *Ap, float *X, OPENBLAS_CONST blasint incX);
 void cblas_dtpsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
                  OPENBLAS_CONST blasint N, OPENBLAS_CONST double *Ap, double *X, OPENBLAS_CONST blasint incX);
 void cblas_ctpsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
-                 OPENBLAS_CONST blasint N, OPENBLAS_CONST void *Ap, void *X, OPENBLAS_CONST blasint incX);
+                 OPENBLAS_CONST blasint N, OPENBLAS_CONST float *Ap, float *X, OPENBLAS_CONST blasint incX);
 void cblas_ztpsv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_DIAG Diag,
-                 OPENBLAS_CONST blasint N, OPENBLAS_CONST void *Ap, void *X, OPENBLAS_CONST blasint incX);
+                 OPENBLAS_CONST blasint N, OPENBLAS_CONST double *Ap, double *X, OPENBLAS_CONST blasint incX);
 
 void cblas_ssymv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A,
                  OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float beta, float *Y, OPENBLAS_CONST blasint incY);
 void cblas_dsymv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A,
                  OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double beta, double *Y, OPENBLAS_CONST blasint incY);
-void cblas_chemv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A,
-                 OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *beta, void *Y, OPENBLAS_CONST blasint incY);
-void cblas_zhemv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A,
-                 OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *beta, void *Y, OPENBLAS_CONST blasint incY);
+void cblas_chemv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A,
+                 OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float *beta, float *Y, OPENBLAS_CONST blasint incY);
+void cblas_zhemv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A,
+                 OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *beta, double *Y, OPENBLAS_CONST blasint incY);
 
 
 void cblas_sspmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *Ap,
@@ -230,36 +225,36 @@ void cblas_dspmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLA
 void cblas_sspr(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, float *Ap);
 void cblas_dspr(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, double *Ap);
 
-void cblas_chpr(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, void *A);
-void cblas_zhpr(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST void *X,OPENBLAS_CONST blasint incX, void *A);
+void cblas_chpr(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, float *A);
+void cblas_zhpr(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *X,OPENBLAS_CONST blasint incX, double *A);
 
 void cblas_sspr2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float *Y, OPENBLAS_CONST blasint incY, float *A);
 void cblas_dspr2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *Y, OPENBLAS_CONST blasint incY, double *A);
-void cblas_chpr2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *Y, OPENBLAS_CONST blasint incY, void *Ap);
-void cblas_zhpr2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *Y, OPENBLAS_CONST blasint incY, void *Ap);
+void cblas_chpr2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float *Y, OPENBLAS_CONST blasint incY, float *Ap);
+void cblas_zhpr2(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *Y, OPENBLAS_CONST blasint incY, double *Ap);
 
 void cblas_chbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *beta, void *Y, OPENBLAS_CONST blasint incY);
+		 OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float *beta, float *Y, OPENBLAS_CONST blasint incY);
 void cblas_zhbmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *beta, void *Y, OPENBLAS_CONST blasint incY);
+		 OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *beta, double *Y, OPENBLAS_CONST blasint incY);
 
 void cblas_chpmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *Ap, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *beta, void *Y, OPENBLAS_CONST blasint incY);
+		 OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *Ap, OPENBLAS_CONST float *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST float *beta, float *Y, OPENBLAS_CONST blasint incY);
 void cblas_zhpmv(OPENBLAS_CONST enum CBLAS_ORDER order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint N,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *Ap, OPENBLAS_CONST void *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST void *beta, void *Y, OPENBLAS_CONST blasint incY);
+		 OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *Ap, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX, OPENBLAS_CONST double *beta, double *Y, OPENBLAS_CONST blasint incY);
 
 void cblas_sgemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
 		 OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_dgemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
 		 OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double beta, double *C, OPENBLAS_CONST blasint ldc);
 void cblas_cgemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+		 OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float *beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_cgemm3m(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+		 OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float *beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_zgemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+		 OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double *beta, double *C, OPENBLAS_CONST blasint ldc);
 void cblas_zgemm3m(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransB, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-		 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+		 OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double *beta, double *C, OPENBLAS_CONST blasint ldc);
 
 
 void cblas_ssymm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
@@ -267,60 +262,60 @@ void cblas_ssymm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLA
 void cblas_dsymm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
                  OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double beta, double *C, OPENBLAS_CONST blasint ldc);
 void cblas_csymm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
-                 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+                 OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float *beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_zsymm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
-                 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+                 OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double *beta, double *C, OPENBLAS_CONST blasint ldc);
 
 void cblas_ssyrk(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans,
 		 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_dsyrk(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans,
 		 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double beta, double *C, OPENBLAS_CONST blasint ldc);
 void cblas_csyrk(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans,
-		 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+		 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_zsyrk(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans,
-		 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+		 OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *beta, double *C, OPENBLAS_CONST blasint ldc);
 
 void cblas_ssyr2k(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans,
 		  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_dsyr2k(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans,
 		  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double beta, double *C, OPENBLAS_CONST blasint ldc);
 void cblas_csyr2k(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans,
-		  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+		  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float *beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_zsyr2k(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans,
-		  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+		  OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double *beta, double *C, OPENBLAS_CONST blasint ldc);
 
 void cblas_strmm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA,
                  OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *B, OPENBLAS_CONST blasint ldb);
 void cblas_dtrmm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA,
                  OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *B, OPENBLAS_CONST blasint ldb);
 void cblas_ctrmm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA,
-                 OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *B, OPENBLAS_CONST blasint ldb);
+                 OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *B, OPENBLAS_CONST blasint ldb);
 void cblas_ztrmm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA,
-                 OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *B, OPENBLAS_CONST blasint ldb);
+                 OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *B, OPENBLAS_CONST blasint ldb);
 
 void cblas_strsm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA,
                  OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *B, OPENBLAS_CONST blasint ldb);
 void cblas_dtrsm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA,
                  OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *B, OPENBLAS_CONST blasint ldb);
 void cblas_ctrsm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA,
-                 OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *B, OPENBLAS_CONST blasint ldb);
+                 OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, float *B, OPENBLAS_CONST blasint ldb);
 void cblas_ztrsm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE TransA,
-                 OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, void *B, OPENBLAS_CONST blasint ldb);
+                 OPENBLAS_CONST enum CBLAS_DIAG Diag, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, double *B, OPENBLAS_CONST blasint ldb);
 
 void cblas_chemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
-                 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+                 OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float *beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_zhemm(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_SIDE Side, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST blasint M, OPENBLAS_CONST blasint N,
-                 OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST void *beta, void *C, OPENBLAS_CONST blasint ldc);
+                 OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double *beta, double *C, OPENBLAS_CONST blasint ldc);
 
 void cblas_cherk(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-                 OPENBLAS_CONST float alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float beta, void *C, OPENBLAS_CONST blasint ldc);
+                 OPENBLAS_CONST float alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_zherk(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-                 OPENBLAS_CONST double alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double beta, void *C, OPENBLAS_CONST blasint ldc);
+                 OPENBLAS_CONST double alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double beta, double *C, OPENBLAS_CONST blasint ldc);
 
 void cblas_cher2k(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-                  OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float beta, void *C, OPENBLAS_CONST blasint ldc);
+                  OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST float *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST float beta, float *C, OPENBLAS_CONST blasint ldc);
 void cblas_zher2k(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
-                  OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double beta, void *C, OPENBLAS_CONST blasint ldc);
+                  OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST double *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double beta, double *C, OPENBLAS_CONST blasint ldc);
 
 void cblas_xerbla(blasint p, char *rout, char *form, ...);
 
@@ -330,9 +325,9 @@ void cblas_saxpby(OPENBLAS_CONST blasint n, OPENBLAS_CONST float alpha, OPENBLAS
 
 void cblas_daxpby(OPENBLAS_CONST blasint n, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx,OPENBLAS_CONST double beta, double *y, OPENBLAS_CONST blasint incy);
 
-void cblas_caxpby(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx,OPENBLAS_CONST void *beta, void *y, OPENBLAS_CONST blasint incy);
+void cblas_caxpby(OPENBLAS_CONST blasint n, OPENBLAS_CONST float *alpha, OPENBLAS_CONST float *x, OPENBLAS_CONST blasint incx,OPENBLAS_CONST float *beta, float *y, OPENBLAS_CONST blasint incy);
 
-void cblas_zaxpby(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx,OPENBLAS_CONST void *beta, void *y, OPENBLAS_CONST blasint incy);
+void cblas_zaxpby(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *alpha, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx,OPENBLAS_CONST double *beta, double *y, OPENBLAS_CONST blasint incy);
 
 void cblas_somatcopy(OPENBLAS_CONST enum CBLAS_ORDER CORDER, OPENBLAS_CONST enum CBLAS_TRANSPOSE CTRANS, OPENBLAS_CONST blasint crows, OPENBLAS_CONST blasint ccols, OPENBLAS_CONST float calpha, OPENBLAS_CONST float *a, 
 		     OPENBLAS_CONST blasint clda, float *b, OPENBLAS_CONST blasint cldb); 

cmake/OpenBLASConfig.cmake.in

@@ -1,79 +0,0 @@
-# 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

@@ -3,9 +3,9 @@
 ## Description: Ported from portion of OpenBLAS/Makefile.system
 ##              Sets various variables based on architecture.
 
-if (X86 OR X86_64)
+if (${ARCH} STREQUAL "x86" OR ${ARCH} STREQUAL "x86_64")
 
-  if (X86)
+  if (${ARCH} STREQUAL "x86")
     if (NOT BINARY)
       set(NO_BINARY_MODE 1)
     endif ()
@@ -33,42 +33,47 @@ if (${CMAKE_C_COMPILER_ID} STREQUAL "Intel")
 endif ()
 
 if (USE_OPENMP)
-  # USE_SIMPLE_THREADED_LEVEL3 = 1
-  # NO_AFFINITY = 1
-  find_package(OpenMP REQUIRED)
-  if (OpenMP_FOUND)
-    set(CCOMMON_OPT "${CCOMMON_OPT} ${OpenMP_C_FLAGS} -DUSE_OPENMP")
-    set(FCOMMON_OPT "${FCOMMON_OPT} ${OpenMP_Fortran_FLAGS}")
-  endif()
+
+  if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU" OR ${CMAKE_C_COMPILER_ID} STREQUAL "LSB")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -fopenmp")
+  endif ()
+
+  if (${CMAKE_C_COMPILER_ID} STREQUAL "Clang")
+    message(WARNING "Clang doesn't support OpenMP yet.")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -fopenmp")
+  endif ()
+
+  if (${CMAKE_C_COMPILER_ID} STREQUAL "Intel")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -openmp")
+  endif ()
+
+  if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -mp")
+  endif ()
+
+  if (${CMAKE_C_COMPILER_ID} STREQUAL "OPEN64")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -mp")
+    set(CEXTRALIB "${CEXTRALIB} -lstdc++")
+  endif ()
+
+  if (${CMAKE_C_COMPILER_ID} STREQUAL "PATHSCALE")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -mp")
+  endif ()
 endif ()
 
 
 if (DYNAMIC_ARCH)
-  if (X86)
-    set(DYNAMIC_CORE KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO)
+  if (${ARCH} STREQUAL "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)
-    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 ${DYNAMIC_CORE} BOBCAT ATOM NANO)
-    endif ()
+  if (${ARCH} STREQUAL "x86_64")
+    set(DYNAMIC_CORE "PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO")
     if (NOT NO_AVX)
-      set(DYNAMIC_CORE ${DYNAMIC_CORE} SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR)
+      set(DYNAMIC_CORE "${DYNAMIC_CORE} SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER")
     endif ()
     if (NOT NO_AVX2)
-      set(DYNAMIC_CORE ${DYNAMIC_CORE} HASWELL ZEN)
-    endif ()
-    if (NOT NO_AVX512)
-      set(DYNAMIC_CORE ${DYNAMIC_CORE} SKYLAKEX)
+      set(DYNAMIC_CORE "${DYNAMIC_CORE} HASWELL ZEN")
     endif ()
   endif ()
 
@@ -89,7 +94,7 @@ if (${ARCH} STREQUAL "ia64")
   endif ()
 endif ()
 
-if (MIPS64)
+if (${ARCH} STREQUAL "mips64")
   set(NO_BINARY_MODE 1)
 endif ()
 
@@ -98,12 +103,12 @@ if (${ARCH} STREQUAL "alpha")
   set(BINARY_DEFINED 1)
 endif ()
 
-if (ARM)
+if (${ARCH} STREQUAL "arm")
   set(NO_BINARY_MODE 1)
   set(BINARY_DEFINED 1)
 endif ()
 
-if (ARM64)
+if (${ARCH} STREQUAL "arm64")
   set(NO_BINARY_MODE 1)
   set(BINARY_DEFINED 1)
 endif ()

cmake/c_check.cmake

@@ -0,0 +1,98 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+## Description: Ported from the OpenBLAS/c_check perl script.
+##              This is triggered by prebuild.cmake and runs before any of the code is built.
+##              Creates config.h and Makefile.conf.
+
+# CMake vars set by this file:
+# OSNAME (use CMAKE_SYSTEM_NAME)
+# ARCH
+# C_COMPILER (use CMAKE_C_COMPILER)
+# BINARY32
+# BINARY64
+# FU
+# CROSS_SUFFIX
+# CROSS
+# CEXTRALIB
+
+# Defines set by this file:
+# OS_
+# ARCH_
+# C_
+# __32BIT__
+# __64BIT__
+# FUNDERSCORE
+# PTHREAD_CREATE_FUNC
+
+# N.B. c_check (and ctest.c) is not cross-platform, so instead try to use CMake variables.
+set(FU "")
+if(APPLE)
+set(FU "_")
+elseif(MSVC)
+set(FU "_")
+elseif(UNIX)
+set(FU "")
+endif()
+
+# Convert CMake vars into the format that OpenBLAS expects
+string(TOUPPER ${CMAKE_SYSTEM_NAME} HOST_OS)
+if (${HOST_OS} STREQUAL "WINDOWS")
+  set(HOST_OS WINNT)
+endif ()
+
+# added by hpa - check size of void ptr to detect 64-bit compile
+if (NOT DEFINED BINARY)
+  set(BINARY 32)
+  if (CMAKE_SIZEOF_VOID_P EQUAL 8)
+    set(BINARY 64)
+  endif ()
+endif ()
+
+if (BINARY EQUAL 64)
+  set(BINARY64 1)
+else ()
+  set(BINARY32 1)
+endif ()
+
+# CMake docs define these:
+# CMAKE_SYSTEM_PROCESSOR - The name of the CPU CMake is building for.
+# CMAKE_HOST_SYSTEM_PROCESSOR - The name of the CPU CMake is running on.
+#
+# TODO: CMAKE_SYSTEM_PROCESSOR doesn't seem to be correct - instead get it from the compiler a la c_check
+set(ARCH ${CMAKE_SYSTEM_PROCESSOR})
+if (${ARCH} STREQUAL "AMD64")
+  set(ARCH "x86_64")
+endif ()
+
+# If you are using a 32-bit compiler on a 64-bit system CMAKE_SYSTEM_PROCESSOR will be wrong
+if (${ARCH} STREQUAL "x86_64" AND BINARY EQUAL 32)
+  set(ARCH x86)
+endif ()
+
+if (${ARCH} STREQUAL "X86")
+  set(ARCH x86)
+endif ()
+
+if (${ARCH} MATCHES "ppc")
+  set(ARCH power)
+endif ()
+
+set(COMPILER_ID ${CMAKE_CXX_COMPILER_ID})
+if (${COMPILER_ID} STREQUAL "GNU")
+  set(COMPILER_ID "GCC")
+endif ()
+
+string(TOUPPER ${ARCH} UC_ARCH)
+
+file(WRITE ${TARGET_CONF}
+  "#define OS_${HOST_OS}\t1\n"
+  "#define ARCH_${UC_ARCH}\t1\n"
+  "#define C_${COMPILER_ID}\t1\n"
+  "#define __${BINARY}BIT__\t1\n"
+  "#define FUNDERSCORE\t${FU}\n")
+
+if (${HOST_OS} STREQUAL "WINDOWSSTORE")
+  file(APPEND ${TARGET_CONF}
+    "#define OS_WINNT\t1\n")
+endif ()
+

cmake/cc.cmake

@@ -15,7 +15,7 @@ if (${CMAKE_C_COMPILER} STREQUAL "GNU" OR ${CMAKE_C_COMPILER} STREQUAL "LSB" OR
 
   if (NO_BINARY_MODE)
 
-    if (MIPS64)
+    if (${ARCH} STREQUAL "mips64")
       if (BINARY64)
         set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=64")
       else ()
@@ -24,12 +24,17 @@ if (${CMAKE_C_COMPILER} STREQUAL "GNU" OR ${CMAKE_C_COMPILER} STREQUAL "LSB" OR
       set(BINARY_DEFINED 1)
     endif ()
 
-    if (${CORE} STREQUAL "LOONGSON3A" OR ${CORE} STREQUAL "LOONGSON3B")
+    if (${CORE} STREQUAL "LOONGSON3A")
       set(CCOMMON_OPT "${CCOMMON_OPT} -march=mips64")
       set(FCOMMON_OPT "${FCOMMON_OPT} -march=mips64")
     endif ()
 
-    if (CMAKE_SYSTEM_NAME STREQUAL "AIX")
+    if (${CORE} STREQUAL "LOONGSON3B")
+      set(CCOMMON_OPT "${CCOMMON_OPT} -march=mips64")
+      set(FCOMMON_OPT "${FCOMMON_OPT} -march=mips64")
+    endif ()
+
+    if (${OSNAME} STREQUAL "AIX")
       set(BINARY_DEFINED 1)
     endif ()
   endif ()
@@ -61,7 +66,7 @@ endif ()
 
 if (${CMAKE_C_COMPILER} STREQUAL "OPEN64")
 
-  if (MIPS64)
+  if (${ARCH} STREQUAL "mips64")
 
     if (NOT BINARY64)
       set(CCOMMON_OPT "${CCOMMON_OPT} -n32")
@@ -89,10 +94,10 @@ endif ()
 
 if (${CMAKE_C_COMPILER} STREQUAL "SUN")
   set(CCOMMON_OPT "${CCOMMON_OPT} -w")
-  if (X86)
+  if (${ARCH} STREQUAL "x86")
     set(CCOMMON_OPT "${CCOMMON_OPT} -m32")
   else ()
-    set(CCOMMON_OPT "${CCOMMON_OPT} -m64")
+    set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
   endif ()
 endif ()
 

cmake/export.cmake

@@ -51,8 +51,7 @@ else()
 endif()
 
 add_custom_command(
-  OUTPUT ${PROJECT_BINARY_DIR}/openblas.def
-  #TARGET ${OpenBLAS_LIBNAME} PRE_LINK
+  TARGET ${OpenBLAS_LIBNAME} PRE_LINK 
   COMMAND perl 
   ARGS "${PROJECT_SOURCE_DIR}/exports/gensymbol" "win2k" "${ARCH_IN}" "dummy" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" "${SYMBOLPREFIX}" "${SYMBOLSUFFIX}" > "${PROJECT_BINARY_DIR}/openblas.def"
   COMMENT "Create openblas.def file"

cmake/f_check.cmake

@@ -20,6 +20,12 @@
 # NEEDBUNDERSCORE
 # NEED2UNDERSCORES
 
+if (MSVC)
+  # had to do this for MSVC, else CMake automatically assumes I have ifort... -hpa
+  include(CMakeForceCompiler)
+  CMAKE_FORCE_Fortran_COMPILER(gfortran GNU)
+endif ()
+
 if (NOT NO_LAPACK)
   enable_language(Fortran)
 else()
@@ -28,13 +34,17 @@ else()
 endif()
 
 if (NOT ONLY_CBLAS)
+  # N.B. f_check is not cross-platform, so instead try to use CMake variables
   # run f_check (appends to TARGET files)
+#  message(STATUS "Running f_check...")
+#  execute_process(COMMAND perl f_check ${TARGET_MAKE} ${TARGET_CONF} ${CMAKE_Fortran_COMPILER}
+#    WORKING_DIRECTORY ${PROJECT_SOURCE_DIR})
 
   # TODO: detect whether underscore needed, set #defines and BU appropriately - use try_compile
   # TODO: set FEXTRALIB flags a la f_check?
 
   set(BU "_")
-  file(APPEND ${TARGET_CONF_TEMP}
+  file(APPEND ${TARGET_CONF}
     "#define BUNDERSCORE _\n"
     "#define NEEDBUNDERSCORE 1\n"
     "#define NEED2UNDERSCORES 0\n")
@@ -46,7 +56,7 @@ else ()
   set(NO_FBLAS 1)
   #set(F_COMPILER GFORTRAN) # CMake handles the fortran compiler
   set(BU "_")
-  file(APPEND ${TARGET_CONF_TEMP}
+  file(APPEND ${TARGET_CONF}
     "#define BUNDERSCORE _\n"
     "#define NEEDBUNDERSCORE 1\n")
 endif()

cmake/fc.cmake

@@ -5,8 +5,13 @@
 
 if (${F_COMPILER} STREQUAL "FLANG")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_FLANG")
-  if (BINARY64 AND INTERFACE64)
-    set(FCOMMON_OPT "${FCOMMON_OPT} -i8")
+  if (BINARY64)
+    if (INTERFACE64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -i8")
+    endif ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -Wall")
+  else ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -Wall")
   endif ()
   if (USE_OPENMP)
     set(FCOMMON_OPT "${FCOMMON_OPT} -fopenmp")
@@ -45,7 +50,7 @@ if (${F_COMPILER} STREQUAL "GFORTRAN")
     set(EXTRALIB "{EXTRALIB} -lgfortran")
   endif ()
   if (NO_BINARY_MODE)
-    if (MIPS64)
+    if (${ARCH} STREQUAL "mips64")
       if (BINARY64)
         set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=64")
       else ()
@@ -125,7 +130,7 @@ if (${F_COMPILER} STREQUAL "PATHSCALE")
     endif ()
   endif ()
 
-  if (NOT MIPS64)
+  if (NOT ${ARCH} STREQUAL "mips64")
     if (NOT BINARY64)
       set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
     else ()
@@ -153,7 +158,7 @@ if (${F_COMPILER} STREQUAL "OPEN64")
     endif ()
   endif ()
 
-  if (MIPS64)
+  if (${ARCH} STREQUAL "mips64")
 
     if (NOT BINARY64)
       set(FCOMMON_OPT "${FCOMMON_OPT} -n32")
@@ -184,7 +189,7 @@ endif ()
 
 if (${F_COMPILER} STREQUAL "SUN")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_SUN")
-  if (X86)
+  if (${ARCH} STREQUAL "x86")
     set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
   else ()
     set(FCOMMON_OPT "${FCOMMON_OPT} -m64")

cmake/lapack.cmake

@@ -1,485 +1,387 @@
 # Sources for compiling lapack-netlib. Can't use CMakeLists.txt because lapack-netlib already has its own cmake files.
 
-set(ALLAUX ilaenv.f ilaenv2stage.f ieeeck.f lsamen.f iparmq.f iparam2stage.F
-   ilaprec.f ilatrans.f ilauplo.f iladiag.f chla_transtype.f
-   ../INSTALL/ilaver.f xerbla_array.f
-   ../INSTALL/slamch.f)
+set(ALLAUX
+  ilaenv.f ieeeck.f lsamen.f xerbla_array.f iparmq.f
+  ilaprec.f ilatrans.f ilauplo.f iladiag.f iparam2stage.F chla_transtype.f
+  ../INSTALL/ilaver.f ../INSTALL/slamch.f
+)
 
 set(SCLAUX
-   sbdsdc.f
-   sbdsqr.f sdisna.f slabad.f slacpy.f sladiv.f slae2.f  slaebz.f
-   slaed0.f slaed1.f slaed2.f slaed3.f slaed4.f slaed5.f slaed6.f
-   slaed7.f slaed8.f slaed9.f slaeda.f slaev2.f slagtf.f
-   slagts.f slamrg.f slanst.f
-   slapy2.f slapy3.f slarnv.f
-   slarra.f slarrb.f slarrc.f slarrd.f slarre.f slarrf.f slarrj.f
-   slarrk.f slarrr.f slaneg.f
-   slartg.f slaruv.f slas2.f  slascl.f
-   slasd0.f slasd1.f slasd2.f slasd3.f slasd4.f slasd5.f slasd6.f
-   slasd7.f slasd8.f slasda.f slasdq.f slasdt.f
-   slaset.f slasq1.f slasq2.f slasq3.f slasq4.f slasq5.f slasq6.f
-   slasr.f  slasrt.f slassq.f slasv2.f spttrf.f sstebz.f sstedc.f
-   ssteqr.f ssterf.f slaisnan.f sisnan.f
-   slartgp.f slartgs.f
-   ../INSTALL/second_${TIMER}.f)
+  sbdsdc.f
+  sbdsqr.f sdisna.f slabad.f slacpy.f sladiv.f slae2.f  slaebz.f
+  slaed0.f slaed1.f slaed2.f slaed3.f slaed4.f slaed5.f slaed6.f
+  slaed7.f slaed8.f slaed9.f slaeda.f slaev2.f slagtf.f
+  slagts.f slamrg.f slanst.f
+  slapy2.f slapy3.f slarnv.f
+  slarra.f slarrb.f slarrc.f slarrd.f slarre.f slarrf.f slarrj.f
+  slarrk.f slarrr.f slaneg.f
+  slartg.f slaruv.f slas2.f  slascl.f
+  slasd0.f slasd1.f slasd2.f slasd3.f slasd4.f slasd5.f slasd6.f
+  slasd7.f slasd8.f slasda.f slasdq.f slasdt.f
+  slaset.f slasq1.f slasq2.f slasq3.f slasq4.f slasq5.f slasq6.f
+  slasr.f  slasrt.f slassq.f slasv2.f spttrf.f sstebz.f sstedc.f
+  ssteqr.f ssterf.f slaisnan.f sisnan.f
+  slartgp.f slartgs.f
+  ../INSTALL/second_${TIMER}.f
+)
 
 set(DZLAUX
-   dbdsdc.f
-   dbdsqr.f ddisna.f dlabad.f dlacpy.f dladiv.f dlae2.f  dlaebz.f
-   dlaed0.f dlaed1.f dlaed2.f dlaed3.f dlaed4.f dlaed5.f dlaed6.f
-   dlaed7.f dlaed8.f dlaed9.f dlaeda.f dlaev2.f dlagtf.f
-   dlagts.f dlamrg.f dlanst.f
-   dlapy2.f dlapy3.f dlarnv.f
-   dlarra.f dlarrb.f dlarrc.f dlarrd.f dlarre.f dlarrf.f dlarrj.f
-   dlarrk.f dlarrr.f dlaneg.f
-   dlartg.f dlaruv.f dlas2.f  dlascl.f
-   dlasd0.f dlasd1.f dlasd2.f dlasd3.f dlasd4.f dlasd5.f dlasd6.f
-   dlasd7.f dlasd8.f dlasda.f dlasdq.f dlasdt.f
-   dlaset.f dlasq1.f dlasq2.f dlasq3.f dlasq4.f dlasq5.f dlasq6.f
-   dlasr.f  dlasrt.f dlassq.f dlasv2.f dpttrf.f dstebz.f dstedc.f
-   dsteqr.f dsterf.f dlaisnan.f disnan.f
-   dlartgp.f dlartgs.f
-   ../INSTALL/dlamch.f ../INSTALL/dsecnd_${TIMER}.f)
+  dbdsdc.f dbdsvdx.f 
+  dbdsqr.f ddisna.f dlabad.f dlacpy.f dladiv.f dlae2.f  dlaebz.f
+  dlaed0.f dlaed1.f dlaed2.f dlaed3.f dlaed4.f dlaed5.f dlaed6.f
+  dlaed7.f dlaed8.f dlaed9.f dlaeda.f dlaev2.f dlagtf.f
+  dlagts.f dlamrg.f dlanst.f
+  dlapy2.f dlapy3.f dlarnv.f
+  dlarra.f dlarrb.f dlarrc.f dlarrd.f dlarre.f dlarrf.f dlarrj.f
+  dlarrk.f dlarrr.f dlaneg.f
+  dlartg.f dlaruv.f dlas2.f  dlascl.f
+  dlasd0.f dlasd1.f dlasd2.f dlasd3.f dlasd4.f dlasd5.f dlasd6.f
+  dlasd7.f dlasd8.f dlasda.f dlasdq.f dlasdt.f
+  dlaset.f dlasq1.f dlasq2.f dlasq3.f dlasq4.f dlasq5.f dlasq6.f
+  dlasr.f  dlasrt.f dlassq.f dlasv2.f dpttrf.f dstebz.f dstedc.f
+  dsteqr.f dsterf.f dlaisnan.f disnan.f
+  dlartgp.f dlartgs.f
+  ../INSTALL/dlamch.f ../INSTALL/dsecnd_${TIMER}.f
+  dgelq.f dgelqt.f dgelqt3.f dgemlq.f dgemlqt.f dgemqr.f dgeqr.f   
+  dgetsls.f dlamswlq.f dlamtsqr.f dlaswlq.f dlatsqr.f dtplqt.f  
+  dtplqt2.f dtpmlqt.f dsysv_aa.f dsytrf_aa.f dsytrs_aa.f dlasyf_aa.f  
+  dsytf2_rk.f dlasyf_rk.f dsytrf_rk.f dsytrs_3.f dsycon_3.f dsytri_3.f   
+  dsytri_3x.f dsysv_rk.f dsb2st_kernels.f dsbev_2stage.f dsbevd_2stage.f  
+  dsbevx_2stage.f dsyev_2stage.f dsyevd_2stage.f dsyevr_2stage.f  
+  dsyevx_2stage.f dsygv_2stage.f dsytrd_2stage.f dsytrd_sb2st.F 
+  dsytrd_sy2sb.f dlarfy.f         
+)
 
 set(SLASRC
-   sbdsvdx.f sgbbrd.f sgbcon.f sgbequ.f sgbrfs.f sgbsv.f
-   sgbsvx.f sgbtf2.f sgbtrf.f sgbtrs.f sgebak.f sgebal.f sgebd2.f
-   sgebrd.f sgecon.f sgeequ.f sgees.f  sgeesx.f sgeev.f  sgeevx.f
-   sgehd2.f sgehrd.f sgelq2.f sgelqf.f
-   sgels.f  sgelsd.f sgelss.f sgelsy.f sgeql2.f sgeqlf.f
-   sgeqp3.f sgeqr2.f sgeqr2p.f sgeqrf.f sgeqrfp.f sgerfs.f sgerq2.f sgerqf.f
-   sgesc2.f sgesdd.f sgesvd.f sgesvdx.f sgesvx.f sgetc2.f
-   sgetrf2.f sgetri.f
-   sggbak.f sggbal.f
-   sgges.f  sgges3.f sggesx.f sggev.f  sggev3.f sggevx.f
-   sggglm.f sgghrd.f sgghd3.f sgglse.f sggqrf.f
-   sggrqf.f sggsvd3.f sggsvp3.f sgtcon.f sgtrfs.f sgtsv.f
-   sgtsvx.f sgttrf.f sgttrs.f sgtts2.f shgeqz.f
-   shsein.f shseqr.f slabrd.f slacon.f slacn2.f
-   slaein.f slaexc.f slag2.f  slags2.f slagtm.f slagv2.f slahqr.f
-   slahr2.f slaic1.f slaln2.f slals0.f slalsa.f slalsd.f
-   slangb.f slange.f slangt.f slanhs.f slansb.f slansp.f
-   slansy.f slantb.f slantp.f slantr.f slanv2.f
-   slapll.f slapmt.f
-   slaqgb.f slaqge.f slaqp2.f slaqps.f slaqsb.f slaqsp.f slaqsy.f
-   slaqr0.f slaqr1.f slaqr2.f slaqr3.f slaqr4.f slaqr5.f
-   slaqtr.f slar1v.f slar2v.f ilaslr.f ilaslc.f
-   slarf.f  slarfb.f slarfg.f slarfgp.f slarft.f slarfx.f slarfy.f slargv.f
-   slarrv.f slartv.f
-   slarz.f  slarzb.f slarzt.f slasy2.f
-   slasyf.f slasyf_rook.f slasyf_rk.f slasyf_aa.f
-   slatbs.f slatdf.f slatps.f slatrd.f slatrs.f slatrz.f
-   sopgtr.f sopmtr.f sorg2l.f sorg2r.f
-   sorgbr.f sorghr.f sorgl2.f sorglq.f sorgql.f sorgqr.f sorgr2.f
-   sorgrq.f sorgtr.f sorm2l.f sorm2r.f sorm22.f
-   sormbr.f sormhr.f sorml2.f sormlq.f sormql.f sormqr.f sormr2.f
-   sormr3.f sormrq.f sormrz.f sormtr.f spbcon.f spbequ.f spbrfs.f
-   spbstf.f spbsv.f  spbsvx.f
-   spbtf2.f spbtrf.f spbtrs.f spocon.f spoequ.f sporfs.f sposv.f
-   sposvx.f spotrf2.f spotri.f spstrf.f spstf2.f
-   sppcon.f sppequ.f
-   spprfs.f sppsv.f  sppsvx.f spptrf.f spptri.f spptrs.f sptcon.f
-   spteqr.f sptrfs.f sptsv.f  sptsvx.f spttrs.f sptts2.f srscl.f
-   ssbev.f  ssbevd.f ssbevx.f ssbgst.f ssbgv.f  ssbgvd.f ssbgvx.f
-   ssbtrd.f sspcon.f sspev.f  sspevd.f sspevx.f sspgst.f
-   sspgv.f  sspgvd.f sspgvx.f ssprfs.f sspsv.f  sspsvx.f ssptrd.f
-   ssptrf.f ssptri.f ssptrs.f sstegr.f sstein.f sstev.f  sstevd.f sstevr.f
-   sstevx.f ssycon.f ssyev.f  ssyevd.f ssyevr.f ssyevx.f ssygs2.f
-   ssygst.f ssygv.f  ssygvd.f ssygvx.f ssyrfs.f ssysv.f  ssysvx.f
-   ssytd2.f ssytf2.f ssytrd.f ssytrf.f ssytri.f ssytri2.f ssytri2x.f
-   ssyswapr.f ssytrs.f ssytrs2.f
-   ssyconv.f ssyconvf.f ssyconvf_rook.f
-   ssysv_aa.f ssysv_aa_2stage.f ssytrf_aa.f ssytrf_aa_2stage.f ssytrs_aa.f ssytrs_aa_2stage.f
-   ssytf2_rook.f ssytrf_rook.f ssytrs_rook.f
-   ssytri_rook.f ssycon_rook.f ssysv_rook.f
-   ssytf2_rk.f ssytrf_rk.f ssytrs_3.f
-   ssytri_3.f ssytri_3x.f ssycon_3.f ssysv_rk.f
-   ssysv_aa.f ssytrf_aa.f ssytrs_aa.f
-   stbcon.f
-   stbrfs.f stbtrs.f stgevc.f stgex2.f stgexc.f stgsen.f
-   stgsja.f stgsna.f stgsy2.f stgsyl.f stpcon.f stprfs.f stptri.f
-   stptrs.f
-   strcon.f strevc.f strevc3.f strexc.f strrfs.f strsen.f strsna.f strsyl.f
-   strtrs.f stzrzf.f sstemr.f
-   slansf.f spftrf.f spftri.f spftrs.f ssfrk.f stfsm.f stftri.f stfttp.f
-   stfttr.f stpttf.f stpttr.f strttf.f strttp.f
-   sgejsv.f sgesvj.f sgsvj0.f sgsvj1.f
-   sgeequb.f ssyequb.f spoequb.f sgbequb.f
-   sbbcsd.f slapmr.f sorbdb.f sorbdb1.f sorbdb2.f sorbdb3.f sorbdb4.f
-   sorbdb5.f sorbdb6.f sorcsd.f sorcsd2by1.f
-   sgeqrt.f sgeqrt2.f sgeqrt3.f sgemqrt.f
-   stpqrt.f stpqrt2.f stpmqrt.f stprfb.f
-   sgelqt.f sgelqt3.f sgemlqt.f
-   sgetsls.f sgeqr.f slatsqr.f slamtsqr.f sgemqr.f
-   sgelq.f slaswlq.f slamswlq.f sgemlq.f
-   stplqt.f stplqt2.f stpmlqt.f
-   ssytrd_2stage.f ssytrd_sy2sb.f ssytrd_sb2st.F ssb2st_kernels.f
-   ssyevd_2stage.f ssyev_2stage.f ssyevx_2stage.f ssyevr_2stage.f
-   ssbev_2stage.f ssbevx_2stage.f ssbevd_2stage.f ssygv_2stage.f)
+  sbdsvdx.f sgbbrd.f sgbcon.f sgbequ.f sgbrfs.f sgbsv.f
+  sgbsvx.f sgbtf2.f sgbtrf.f sgbtrs.f sgebak.f sgebal.f sgebd2.f
+  sgebrd.f sgecon.f sgeequ.f sgees.f  sgeesx.f sgeev.f  sgeevx.f
+  DEPRECATED/sgegs.f  DEPRECATED/sgegv.f  sgehd2.f sgehrd.f sgelq2.f sgelqf.f
+  sgels.f  sgelsd.f sgelss.f DEPRECATED/sgelsx.f sgelsy.f sgeql2.f sgeqlf.f
+  sgeqp3.f DEPRECATED/sgeqpf.f sgeqr2.f sgeqr2p.f sgeqrf.f sgeqrfp.f sgerfs.f
+  sgerq2.f sgerqf.f sgesc2.f sgesdd.f  sgesvd.f sgesvdx.f sgesvx.f
+  sgetc2.f sgetri.f sgetrf2.f
+  sggbak.f sggbal.f sgghd3.f  sgges.f  sgges3.f sggesx.f sggev.f sggev3.f  sggevx.f
+  sggglm.f sgghrd.f sgglse.f sggqrf.f
+  sggrqf.f DEPRECATED/sggsvd.f sggsvd3.f DEPRECATED/sggsvp.f sggsvp3.f sgtcon.f sgtrfs.f sgtsv.f
+  sgtsvx.f sgttrf.f sgttrs.f sgtts2.f shgeqz.f
+  shsein.f shseqr.f slabrd.f slacon.f slacn2.f
+  slaein.f slaexc.f slag2.f  slags2.f slagtm.f slagv2.f slahqr.f
+  DEPRECATED/slahrd.f slahr2.f slaic1.f slaln2.f slals0.f slalsa.f slalsd.f
+  slangb.f slange.f slangt.f slanhs.f slansb.f slansp.f
+  slansy.f slantb.f slantp.f slantr.f slanv2.f
+  slapll.f slapmt.f
+  slaqgb.f slaqge.f slaqp2.f slaqps.f slaqsb.f slaqsp.f slaqsy.f
+  slaqr0.f slaqr1.f slaqr2.f slaqr3.f slaqr4.f slaqr5.f
+  slaqtr.f slar1v.f slar2v.f ilaslr.f ilaslc.f
+  slarf.f  slarfb.f slarfg.f slarfgp.f slarft.f slarfx.f slargv.f
+  slarrv.f slartv.f
+  slarz.f  slarzb.f slarzt.f slasy2.f slasyf.f slasyf_rook.f
+  slatbs.f slatdf.f slatps.f slatrd.f slatrs.f slatrz.f DEPRECATED/slatzm.f
+  sopgtr.f sopmtr.f sorg2l.f sorg2r.f
+  sorgbr.f sorghr.f sorgl2.f sorglq.f sorgql.f sorgqr.f sorgr2.f
+  sorgrq.f sorgtr.f sorm2l.f sorm2r.f sorm22.f
+  sormbr.f sormhr.f sorml2.f sormlq.f sormql.f sormqr.f sormr2.f
+  sormr3.f sormrq.f sormrz.f sormtr.f spbcon.f spbequ.f spbrfs.f
+  spbstf.f spbsv.f  spbsvx.f
+  spbtf2.f spbtrf.f spbtrs.f spocon.f spoequ.f sporfs.f sposv.f
+  sposvx.f spstrf.f spstf2.f
+  sppcon.f sppequ.f
+  spprfs.f sppsv.f  sppsvx.f spptrf.f spptri.f spptrs.f sptcon.f
+  spteqr.f sptrfs.f sptsv.f  sptsvx.f spttrs.f sptts2.f srscl.f
+  ssbev.f  ssbevd.f ssbevx.f ssbgst.f ssbgv.f  ssbgvd.f ssbgvx.f
+  ssbtrd.f sspcon.f sspev.f  sspevd.f sspevx.f sspgst.f
+  sspgv.f  sspgvd.f sspgvx.f ssprfs.f sspsv.f  sspsvx.f ssptrd.f
+  ssptrf.f ssptri.f ssptrs.f sstegr.f sstein.f sstev.f  sstevd.f sstevr.f
+  sstevx.f
+  ssycon.f ssyev.f  ssyevd.f ssyevr.f ssyevx.f ssygs2.f
+  ssygst.f ssygv.f  ssygvd.f ssygvx.f ssyrfs.f ssysv.f  ssysvx.f
+  ssytd2.f ssytf2.f ssytrd.f ssytrf.f ssytri.f ssytri2.f ssytri2x.f
+  ssyswapr.f ssytrs.f ssytrs2.f ssyconv.f
+  ssytf2_rook.f ssytrf_rook.f ssytrs_rook.f
+  ssytri_rook.f ssycon_rook.f ssysv_rook.f
+  stbcon.f
+  stbrfs.f stbtrs.f stgevc.f stgex2.f stgexc.f stgsen.f
+  stgsja.f stgsna.f stgsy2.f stgsyl.f stpcon.f stprfs.f stptri.f
+  stptrs.f
+  strcon.f strevc.f strevc3.f strexc.f strrfs.f strsen.f strsna.f strsyl.f
+  strtrs.f DEPRECATED/stzrqf.f stzrzf.f sstemr.f
+  slansf.f spftrf.f spftri.f spftrs.f ssfrk.f stfsm.f stftri.f stfttp.f
+  stfttr.f stpttf.f stpttr.f strttf.f strttp.f
+  sgejsv.f  sgesvj.f  sgsvj0.f  sgsvj1.f
+  sgeequb.f ssyequb.f spoequb.f sgbequb.f
+  sbbcsd.f slapmr.f sorbdb.f sorbdb1.f sorbdb2.f sorbdb3.f sorbdb4.f
+  sorbdb5.f sorbdb6.f sorcsd.f sorcsd2by1.f
+  sgeqrt.f sgeqrt2.f sgeqrt3.f sgemqrt.f
+  stpqrt.f stpqrt2.f stpmqrt.f stprfb.f spotri.f
+  sgelq.f  sgelqt.f sgelqt3.f sgemlq.f sgemlqt.f sgemqr.f sgeqr.f sgetsls.f    
+  slamswlq.f slamtsqr.f slaswlq.f slatsqr.f stplqt.f stplqt2.f stpmlqt.f    
+  ssysv_aa.f ssytrf_aa.f ssytrs_aa.f slasyf_aa.f ssytf2_rk.f slasyf_rk.f       
+  ssytrf_rk.f ssytrs_3.f ssycon_3.f ssytri_3.f ssytri_3x.f ssysv_rk.f        
+  ssb2st_kernels.f ssbev_2stage.f ssbevd_2stage.f ssbevx_2stage.f  
+  ssyev_2stage.f ssyevd_2stage.f ssyevr_2stage.f ssyevx_2stage.f 
+  ssygv_2stage.f ssytrd_2stage.f ssytrd_sb2st.F ssytrd_sy2sb.f slarfy.f         
+)
 
-set(SXLASRC sgesvxx.f sgerfsx.f sla_gerfsx_extended.f sla_geamv.f
-   sla_gercond.f sla_gerpvgrw.f ssysvxx.f ssyrfsx.f
-   sla_syrfsx_extended.f sla_syamv.f sla_syrcond.f sla_syrpvgrw.f
-   sposvxx.f sporfsx.f sla_porfsx_extended.f sla_porcond.f
-   sla_porpvgrw.f sgbsvxx.f sgbrfsx.f sla_gbrfsx_extended.f
-   sla_gbamv.f sla_gbrcond.f sla_gbrpvgrw.f sla_lin_berr.f slarscl2.f
-   slascl2.f sla_wwaddw.f)
+set(DSLASRC spotrs.f spotrf2.f)
 
 set(CLASRC
-   cbdsqr.f cgbbrd.f cgbcon.f cgbequ.f cgbrfs.f cgbsv.f  cgbsvx.f
-   cgbtf2.f cgbtrf.f cgbtrs.f cgebak.f cgebal.f cgebd2.f cgebrd.f
-   cgecon.f cgeequ.f cgees.f  cgeesx.f cgeev.f  cgeevx.f
-   cgehd2.f cgehrd.f cgelq2.f cgelqf.f
-   cgels.f  cgelsd.f cgelss.f cgelsy.f cgeql2.f cgeqlf.f cgeqp3.f
-   cgeqr2.f cgeqr2p.f cgeqrf.f cgeqrfp.f cgerfs.f cgerq2.f cgerqf.f
-   cgesc2.f cgesdd.f cgesvd.f cgesvdx.f
-   cgesvj.f cgejsv.f cgsvj0.f cgsvj1.f
-   cgesvx.f cgetc2.f cgetrf2.f
-   cgetri.f
-   cggbak.f cggbal.f
-   cgges.f  cgges3.f cggesx.f cggev.f  cggev3.f cggevx.f
-   cggglm.f cgghrd.f cgghd3.f cgglse.f cggqrf.f cggrqf.f
-   cggsvd3.f cggsvp3.f
-   cgtcon.f cgtrfs.f cgtsv.f  cgtsvx.f cgttrf.f cgttrs.f cgtts2.f chbev.f
-   chbevd.f chbevx.f chbgst.f chbgv.f  chbgvd.f chbgvx.f chbtrd.f
-   checon.f cheev.f  cheevd.f cheevr.f cheevx.f chegs2.f chegst.f
-   chegv.f  chegvd.f chegvx.f cherfs.f chesv.f  chesvx.f chetd2.f
-   chetf2.f chetrd.f
-   chetrf.f chetri.f chetri2.f chetri2x.f cheswapr.f
-   chetrs.f chetrs2.f
-   chetf2_rook.f chetrf_rook.f chetri_rook.f
-   chetrs_rook.f checon_rook.f chesv_rook.f
-   chetf2_rk.f chetrf_rk.f chetri_3.f chetri_3x.f
-   chetrs_3.f checon_3.f chesv_rk.f
-   chesv_aa.f chesv_aa_2stage.f chetrf_aa.f chetrf_aa_2stage.f chetrs_aa.f chetrs_aa_2stage.f
-   chgeqz.f chpcon.f chpev.f  chpevd.f
-   chpevx.f chpgst.f chpgv.f  chpgvd.f chpgvx.f chprfs.f chpsv.f
-   chpsvx.f
-   chptrd.f chptrf.f chptri.f chptrs.f chsein.f chseqr.f clabrd.f
-   clacgv.f clacon.f clacn2.f clacp2.f clacpy.f clacrm.f clacrt.f cladiv.f
-   claed0.f claed7.f claed8.f
-   claein.f claesy.f claev2.f clags2.f clagtm.f
-   clahef.f clahef_rook.f clahef_rk.f clahef_aa.f clahqr.f
-   clahr2.f claic1.f clals0.f clalsa.f clalsd.f clangb.f clange.f clangt.f
-   clanhb.f clanhe.f
-   clanhp.f clanhs.f clanht.f clansb.f clansp.f clansy.f clantb.f
-   clantp.f clantr.f clapll.f clapmt.f clarcm.f claqgb.f claqge.f
-   claqhb.f claqhe.f claqhp.f claqp2.f claqps.f claqsb.f
-   claqr0.f claqr1.f claqr2.f claqr3.f claqr4.f claqr5.f
-   claqsp.f claqsy.f clar1v.f clar2v.f ilaclr.f ilaclc.f
-   clarf.f  clarfb.f clarfg.f clarfgp.f clarft.f
-   clarfx.f clarfy.f clargv.f clarnv.f clarrv.f clartg.f clartv.f
-   clarz.f  clarzb.f clarzt.f clascl.f claset.f clasr.f  classq.f
-   clasyf.f clasyf_rook.f clasyf_rk.f clasyf_aa.f
-   clatbs.f clatdf.f clatps.f clatrd.f clatrs.f clatrz.f
-   cpbcon.f cpbequ.f cpbrfs.f cpbstf.f cpbsv.f
-   cpbsvx.f cpbtf2.f cpbtrf.f cpbtrs.f cpocon.f cpoequ.f cporfs.f
-   cposv.f  cposvx.f cpotrf2.f cpotri.f cpstrf.f cpstf2.f
-   cppcon.f cppequ.f cpprfs.f cppsv.f  cppsvx.f cpptrf.f cpptri.f cpptrs.f
-   cptcon.f cpteqr.f cptrfs.f cptsv.f  cptsvx.f cpttrf.f cpttrs.f cptts2.f
-   crot.f   cspcon.f csprfs.f cspsv.f
-   cspsvx.f csptrf.f csptri.f csptrs.f csrscl.f cstedc.f
-   cstegr.f cstein.f csteqr.f csycon.f
-   csyrfs.f csysv.f  csysvx.f csytf2.f csytrf.f csytri.f
-   csytri2.f csytri2x.f csyswapr.f
-   csytrs.f csytrs2.f
-   csyconv.f csyconvf.f csyconvf_rook.f
-   csytf2_rook.f csytrf_rook.f csytrs_rook.f
-   csytri_rook.f csycon_rook.f csysv_rook.f
-   csytf2_rk.f csytrf_rk.f csytrf_aa.f csytrf_aa_2stage.f csytrs_3.f csytrs_aa.f csytrs_aa_2stage.f
-   csytri_3.f csytri_3x.f csycon_3.f csysv_rk.f csysv_aa.f csysv_aa_2stage.f
-   ctbcon.f ctbrfs.f ctbtrs.f ctgevc.f ctgex2.f
-   ctgexc.f ctgsen.f ctgsja.f ctgsna.f ctgsy2.f ctgsyl.f ctpcon.f
-   ctprfs.f ctptri.f
-   ctptrs.f ctrcon.f ctrevc.f ctrevc3.f ctrexc.f ctrrfs.f ctrsen.f ctrsna.f
-   ctrsyl.f ctrtrs.f ctzrzf.f cung2l.f cung2r.f
-   cungbr.f cunghr.f cungl2.f cunglq.f cungql.f cungqr.f cungr2.f
-   cungrq.f cungtr.f cunm2l.f cunm2r.f cunmbr.f cunmhr.f cunml2.f cunm22.f
-   cunmlq.f cunmql.f cunmqr.f cunmr2.f cunmr3.f cunmrq.f cunmrz.f
-   cunmtr.f cupgtr.f cupmtr.f icmax1.f scsum1.f cstemr.f
-   chfrk.f ctfttp.f clanhf.f cpftrf.f cpftri.f cpftrs.f ctfsm.f ctftri.f
-   ctfttr.f ctpttf.f ctpttr.f ctrttf.f ctrttp.f
-   cgeequb.f cgbequb.f csyequb.f cpoequb.f cheequb.f
-   cbbcsd.f clapmr.f cunbdb.f cunbdb1.f cunbdb2.f cunbdb3.f cunbdb4.f
-   cunbdb5.f cunbdb6.f cuncsd.f cuncsd2by1.f
-   cgeqrt.f cgeqrt2.f cgeqrt3.f cgemqrt.f
-   ctpqrt.f ctpqrt2.f ctpmqrt.f ctprfb.f
-   cgelqt.f cgelqt3.f cgemlqt.f
-   cgetsls.f cgeqr.f clatsqr.f clamtsqr.f cgemqr.f
-   cgelq.f claswlq.f clamswlq.f cgemlq.f
-   ctplqt.f ctplqt2.f ctpmlqt.f
-   chetrd_2stage.f chetrd_he2hb.f chetrd_hb2st.F chb2st_kernels.f
-   cheevd_2stage.f cheev_2stage.f cheevx_2stage.f cheevr_2stage.f
-   chbev_2stage.f chbevx_2stage.f chbevd_2stage.f chegv_2stage.f)
-
-set(CXLASRC cgesvxx.f cgerfsx.f cla_gerfsx_extended.f cla_geamv.f
-   cla_gercond_c.f cla_gercond_x.f cla_gerpvgrw.f
-   csysvxx.f csyrfsx.f cla_syrfsx_extended.f cla_syamv.f
-   cla_syrcond_c.f cla_syrcond_x.f cla_syrpvgrw.f
-   cposvxx.f cporfsx.f cla_porfsx_extended.f
-   cla_porcond_c.f cla_porcond_x.f cla_porpvgrw.f
-   cgbsvxx.f cgbrfsx.f cla_gbrfsx_extended.f cla_gbamv.f
-   cla_gbrcond_c.f cla_gbrcond_x.f cla_gbrpvgrw.f
-   chesvxx.f cherfsx.f cla_herfsx_extended.f cla_heamv.f
-   cla_hercond_c.f cla_hercond_x.f cla_herpvgrw.f
-   cla_lin_berr.f clarscl2.f clascl2.f cla_wwaddw.f)
-
-set(DLASRC
-   dbdsvdx.f dgbbrd.f dgbcon.f dgbequ.f dgbrfs.f dgbsv.f
-   dgbsvx.f dgbtf2.f dgbtrf.f dgbtrs.f dgebak.f dgebal.f dgebd2.f
-   dgebrd.f dgecon.f dgeequ.f dgees.f  dgeesx.f dgeev.f  dgeevx.f
-   dgehd2.f dgehrd.f dgelq2.f dgelqf.f
-   dgels.f  dgelsd.f dgelss.f dgelsy.f dgeql2.f dgeqlf.f
-   dgeqp3.f dgeqr2.f dgeqr2p.f dgeqrf.f dgeqrfp.f dgerfs.f dgerq2.f dgerqf.f
-   dgesc2.f dgesdd.f dgesvd.f dgesvdx.f dgesvx.f dgetc2.f
-   dgetrf2.f dgetri.f
-   dggbak.f dggbal.f
-   dgges.f  dgges3.f dggesx.f dggev.f  dggev3.f dggevx.f
-   dggglm.f dgghrd.f dgghd3.f dgglse.f dggqrf.f
-   dggrqf.f dggsvd3.f dggsvp3.f dgtcon.f dgtrfs.f dgtsv.f
-   dgtsvx.f dgttrf.f dgttrs.f dgtts2.f dhgeqz.f
-   dhsein.f dhseqr.f dlabrd.f dlacon.f dlacn2.f
-   dlaein.f dlaexc.f dlag2.f  dlags2.f dlagtm.f dlagv2.f dlahqr.f
-   dlahr2.f dlaic1.f dlaln2.f dlals0.f dlalsa.f dlalsd.f
-   dlangb.f dlange.f dlangt.f dlanhs.f dlansb.f dlansp.f
-   dlansy.f dlantb.f dlantp.f dlantr.f dlanv2.f
-   dlapll.f dlapmt.f
-   dlaqgb.f dlaqge.f dlaqp2.f dlaqps.f dlaqsb.f dlaqsp.f dlaqsy.f
-   dlaqr0.f dlaqr1.f dlaqr2.f dlaqr3.f dlaqr4.f dlaqr5.f
-   dlaqtr.f dlar1v.f dlar2v.f iladlr.f iladlc.f
-   dlarf.f  dlarfb.f dlarfg.f dlarfgp.f dlarft.f dlarfx.f dlarfy.f
-   dlargv.f dlarrv.f dlartv.f
-   dlarz.f  dlarzb.f dlarzt.f dlasy2.f
-   dlasyf.f dlasyf_rook.f dlasyf_rk.f dlasyf_aa.f
-   dlatbs.f dlatdf.f dlatps.f dlatrd.f dlatrs.f dlatrz.f
-   dopgtr.f dopmtr.f dorg2l.f dorg2r.f
-   dorgbr.f dorghr.f dorgl2.f dorglq.f dorgql.f dorgqr.f dorgr2.f
-   dorgrq.f dorgtr.f dorm2l.f dorm2r.f dorm22.f
-   dormbr.f dormhr.f dorml2.f dormlq.f dormql.f dormqr.f dormr2.f
-   dormr3.f dormrq.f dormrz.f dormtr.f dpbcon.f dpbequ.f dpbrfs.f
-   dpbstf.f dpbsv.f  dpbsvx.f
-   dpbtf2.f dpbtrf.f dpbtrs.f dpocon.f dpoequ.f dporfs.f dposv.f
-   dposvx.f dpotrf2.f dpotri.f dpotrs.f dpstrf.f dpstf2.f
-   dppcon.f dppequ.f
-   dpprfs.f dppsv.f  dppsvx.f dpptrf.f dpptri.f dpptrs.f dptcon.f
-   dpteqr.f dptrfs.f dptsv.f  dptsvx.f dpttrs.f dptts2.f drscl.f
-   dsbev.f  dsbevd.f dsbevx.f dsbgst.f dsbgv.f  dsbgvd.f dsbgvx.f
-   dsbtrd.f dspcon.f dspev.f  dspevd.f dspevx.f dspgst.f
-   dspgv.f  dspgvd.f dspgvx.f dsprfs.f dspsv.f  dspsvx.f dsptrd.f
-   dsptrf.f dsptri.f dsptrs.f dstegr.f dstein.f dstev.f  dstevd.f dstevr.f
-   dstevx.f dsycon.f dsyev.f  dsyevd.f dsyevr.f
-   dsyevx.f dsygs2.f dsygst.f dsygv.f  dsygvd.f dsygvx.f dsyrfs.f
-   dsysv.f  dsysvx.f
-   dsytd2.f dsytf2.f dsytrd.f dsytrf.f dsytri.f dsytrs.f dsytrs2.f
-   dsytri2.f dsytri2x.f dsyswapr.f
-   dsyconv.f dsyconvf.f dsyconvf_rook.f
-   dsytf2_rook.f dsytrf_rook.f dsytrs_rook.f
-   dsytri_rook.f dsycon_rook.f dsysv_rook.f
-   dsytf2_rk.f dsytrf_rk.f dsytrs_3.f
-   dsytri_3.f dsytri_3x.f dsycon_3.f dsysv_rk.f
-   dsysv_aa.f dsysv_aa_2stage.f dsytrf_aa.f dsytrf_aa_2stage.f dsytrs_aa.f dsytrs_aa_2stage.f
-   dtbcon.f
-   dtbrfs.f dtbtrs.f dtgevc.f dtgex2.f dtgexc.f dtgsen.f
-   dtgsja.f dtgsna.f dtgsy2.f dtgsyl.f dtpcon.f dtprfs.f dtptri.f
-   dtptrs.f
-   dtrcon.f dtrevc.f dtrevc3.f dtrexc.f dtrrfs.f dtrsen.f dtrsna.f dtrsyl.f
-   dtrtrs.f dtzrzf.f dstemr.f
-   dsgesv.f dsposv.f dlag2s.f slag2d.f dlat2s.f
-   dlansf.f dpftrf.f dpftri.f dpftrs.f dsfrk.f dtfsm.f dtftri.f dtfttp.f
-   dtfttr.f dtpttf.f dtpttr.f dtrttf.f dtrttp.f
-   dgejsv.f dgesvj.f dgsvj0.f dgsvj1.f
-   dgeequb.f dsyequb.f dpoequb.f dgbequb.f
-   dbbcsd.f dlapmr.f dorbdb.f dorbdb1.f dorbdb2.f dorbdb3.f dorbdb4.f
-   dorbdb5.f dorbdb6.f dorcsd.f dorcsd2by1.f
-   dgeqrt.f dgeqrt2.f dgeqrt3.f dgemqrt.f
-   dtpqrt.f dtpqrt2.f dtpmqrt.f dtprfb.f
-   dgelqt.f dgelqt3.f dgemlqt.f
-   dgetsls.f dgeqr.f dlatsqr.f dlamtsqr.f dgemqr.f
-   dgelq.f dlaswlq.f dlamswlq.f dgemlq.f
-   dtplqt.f dtplqt2.f dtpmlqt.f
-   dsytrd_2stage.f dsytrd_sy2sb.f dsytrd_sb2st.F dsb2st_kernels.f
-   dsyevd_2stage.f dsyev_2stage.f dsyevx_2stage.f dsyevr_2stage.f
-   dsbev_2stage.f dsbevx_2stage.f dsbevd_2stage.f dsygv_2stage.f)
-
-set(DXLASRC dgesvxx.f dgerfsx.f dla_gerfsx_extended.f dla_geamv.f
-   dla_gercond.f dla_gerpvgrw.f dsysvxx.f dsyrfsx.f
-   dla_syrfsx_extended.f dla_syamv.f dla_syrcond.f dla_syrpvgrw.f
-   dposvxx.f dporfsx.f dla_porfsx_extended.f dla_porcond.f
-   dla_porpvgrw.f dgbsvxx.f dgbrfsx.f dla_gbrfsx_extended.f
-   dla_gbamv.f dla_gbrcond.f dla_gbrpvgrw.f dla_lin_berr.f dlarscl2.f
-   dlascl2.f dla_wwaddw.f)
-
-set(ZLASRC
-   zbdsqr.f zgbbrd.f zgbcon.f zgbequ.f zgbrfs.f zgbsv.f  zgbsvx.f
-   zgbtf2.f zgbtrf.f zgbtrs.f zgebak.f zgebal.f zgebd2.f zgebrd.f
-   zgecon.f zgeequ.f zgees.f  zgeesx.f zgeev.f  zgeevx.f
-   zgehd2.f zgehrd.f zgelq2.f zgelqf.f
-   zgels.f  zgelsd.f zgelss.f zgelsy.f zgeql2.f zgeqlf.f zgeqp3.f
-   zgeqr2.f zgeqr2p.f zgeqrf.f zgeqrfp.f zgerfs.f zgerq2.f zgerqf.f
-   zgesc2.f zgesdd.f zgesvd.f zgesvdx.f zgesvx.f
-   zgesvj.f zgejsv.f zgsvj0.f zgsvj1.f
-   zgetc2.f zgetrf2.f
-   zgetri.f
-   zggbak.f zggbal.f
-   zgges.f  zgges3.f zggesx.f zggev.f  zggev3.f zggevx.f
-   zggglm.f zgghrd.f zgghd3.f zgglse.f zggqrf.f zggrqf.f
-   zggsvd3.f zggsvp3.f
-   zgtcon.f zgtrfs.f zgtsv.f  zgtsvx.f zgttrf.f zgttrs.f zgtts2.f zhbev.f
-   zhbevd.f zhbevx.f zhbgst.f zhbgv.f  zhbgvd.f zhbgvx.f zhbtrd.f
-   zhecon.f zheev.f  zheevd.f zheevr.f zheevx.f zhegs2.f zhegst.f
-   zhegv.f  zhegvd.f zhegvx.f zherfs.f zhesv.f  zhesvx.f zhetd2.f
-   zhetf2.f zhetrd.f
-   zhetrf.f zhetri.f zhetri2.f zhetri2x.f zheswapr.f
-   zhetrs.f zhetrs2.f
-   zhetf2_rook.f zhetrf_rook.f zhetri_rook.f
-   zhetrs_rook.f zhecon_rook.f zhesv_rook.f
-   zhetf2_rk.f zhetrf_rk.f zhetri_3.f zhetri_3x.f
-   zhetrs_3.f zhecon_3.f zhesv_rk.f
-   zhesv_aa.f zhesv_aa_2stage.f zhetrf_aa.f zhetrf_aa_2stage.f zhetrs_aa.f zhetrs_aa_2stage.f
-   zhgeqz.f zhpcon.f zhpev.f  zhpevd.f
-   zhpevx.f zhpgst.f zhpgv.f  zhpgvd.f zhpgvx.f zhprfs.f zhpsv.f
-   zhpsvx.f
-   zhptrd.f zhptrf.f zhptri.f zhptrs.f zhsein.f zhseqr.f zlabrd.f
-   zlacgv.f zlacon.f zlacn2.f zlacp2.f zlacpy.f zlacrm.f zlacrt.f zladiv.f
-   zlaed0.f zlaed7.f zlaed8.f
-   zlaein.f zlaesy.f zlaev2.f zlags2.f zlagtm.f
-   zlahef.f zlahef_rook.f zlahef_rk.f zlahef_aa.f zlahqr.f
-   zlahr2.f zlaic1.f zlals0.f zlalsa.f zlalsd.f zlangb.f zlange.f
-   zlangt.f zlanhb.f
-   zlanhe.f
-   zlanhp.f zlanhs.f zlanht.f zlansb.f zlansp.f zlansy.f zlantb.f
-   zlantp.f zlantr.f zlapll.f zlapmt.f zlaqgb.f zlaqge.f
-   zlaqhb.f zlaqhe.f zlaqhp.f zlaqp2.f zlaqps.f zlaqsb.f
-   zlaqr0.f zlaqr1.f zlaqr2.f zlaqr3.f zlaqr4.f zlaqr5.f
-   zlaqsp.f zlaqsy.f zlar1v.f zlar2v.f ilazlr.f ilazlc.f
-   zlarcm.f zlarf.f  zlarfb.f
-   zlarfg.f zlarfgp.f zlarft.f
-   zlarfx.f zlarfy.f zlargv.f zlarnv.f zlarrv.f zlartg.f zlartv.f
-   zlarz.f  zlarzb.f zlarzt.f zlascl.f zlaset.f zlasr.f
-   zlassq.f zlasyf.f zlasyf_rook.f zlasyf_rk.f zlasyf_aa.f
-   zlatbs.f zlatdf.f zlatps.f zlatrd.f zlatrs.f zlatrz.f
-   zpbcon.f zpbequ.f zpbrfs.f zpbstf.f zpbsv.f
-   zpbsvx.f zpbtf2.f zpbtrf.f zpbtrs.f zpocon.f zpoequ.f zporfs.f
-   zposv.f  zposvx.f zpotrf2.f zpotri.f zpotrs.f zpstrf.f zpstf2.f
-   zppcon.f zppequ.f zpprfs.f zppsv.f  zppsvx.f zpptrf.f zpptri.f zpptrs.f
-   zptcon.f zpteqr.f zptrfs.f zptsv.f  zptsvx.f zpttrf.f zpttrs.f zptts2.f
-   zrot.f   zspcon.f zsprfs.f zspsv.f
-   zspsvx.f zsptrf.f zsptri.f zsptrs.f zdrscl.f zstedc.f
-   zstegr.f zstein.f zsteqr.f zsycon.f
-   zsyrfs.f zsysv.f  zsysvx.f zsytf2.f zsytrf.f zsytri.f
-   zsytri2.f zsytri2x.f zsyswapr.f
-   zsytrs.f zsytrs2.f
-   zsyconv.f zsyconvf.f zsyconvf_rook.f
-   zsytf2_rook.f zsytrf_rook.f zsytrs_rook.f zsytrs_aa.f zsytrs_aa_2stage.f
-   zsytri_rook.f zsycon_rook.f zsysv_rook.f
-   zsytf2_rk.f zsytrf_rk.f zsytrf_aa.f zsytrf_aa_2stage.f zsytrs_3.f
-   zsytri_3.f zsytri_3x.f zsycon_3.f zsysv_rk.f zsysv_aa.f zsysv_aa_2stage.f
-   ztbcon.f ztbrfs.f ztbtrs.f ztgevc.f ztgex2.f
-   ztgexc.f ztgsen.f ztgsja.f ztgsna.f ztgsy2.f ztgsyl.f ztpcon.f
-   ztprfs.f ztptri.f
-   ztptrs.f ztrcon.f ztrevc.f ztrevc3.f ztrexc.f ztrrfs.f ztrsen.f ztrsna.f
-   ztrsyl.f ztrtrs.f ztzrzf.f zung2l.f
-   zung2r.f zungbr.f zunghr.f zungl2.f zunglq.f zungql.f zungqr.f zungr2.f
-   zungrq.f zungtr.f zunm2l.f zunm2r.f zunmbr.f zunmhr.f zunml2.f zunm22.f
-   zunmlq.f zunmql.f zunmqr.f zunmr2.f zunmr3.f zunmrq.f zunmrz.f
-   zunmtr.f zupgtr.f
-   zupmtr.f izmax1.f dzsum1.f zstemr.f
-   zcgesv.f zcposv.f zlag2c.f clag2z.f zlat2c.f
-   zhfrk.f ztfttp.f zlanhf.f zpftrf.f zpftri.f zpftrs.f ztfsm.f ztftri.f
-   ztfttr.f ztpttf.f ztpttr.f ztrttf.f ztrttp.f
-   zgeequb.f zgbequb.f zsyequb.f zpoequb.f zheequb.f
-   zbbcsd.f zlapmr.f zunbdb.f zunbdb1.f zunbdb2.f zunbdb3.f zunbdb4.f
-   zunbdb5.f zunbdb6.f zuncsd.f zuncsd2by1.f
-   zgeqrt.f zgeqrt2.f zgeqrt3.f zgemqrt.f
-   ztpqrt.f ztpqrt2.f ztpmqrt.f ztprfb.f
-   ztplqt.f ztplqt2.f ztpmlqt.f
-   zgelqt.f zgelqt3.f zgemlqt.f
-   zgetsls.f zgeqr.f zlatsqr.f zlamtsqr.f zgemqr.f
-   zgelq.f zlaswlq.f zlamswlq.f zgemlq.f
-   zhetrd_2stage.f zhetrd_he2hb.f zhetrd_hb2st.F zhb2st_kernels.f
-   zheevd_2stage.f zheev_2stage.f zheevx_2stage.f zheevr_2stage.f
-   zhbev_2stage.f zhbevx_2stage.f zhbevd_2stage.f zhegv_2stage.f)
-
-set(ZXLASRC zgesvxx.f zgerfsx.f zla_gerfsx_extended.f zla_geamv.f
-   zla_gercond_c.f zla_gercond_x.f zla_gerpvgrw.f zsysvxx.f zsyrfsx.f
-   zla_syrfsx_extended.f zla_syamv.f zla_syrcond_c.f zla_syrcond_x.f
-   zla_syrpvgrw.f zposvxx.f zporfsx.f zla_porfsx_extended.f
-   zla_porcond_c.f zla_porcond_x.f zla_porpvgrw.f zgbsvxx.f zgbrfsx.f
-   zla_gbrfsx_extended.f zla_gbamv.f zla_gbrcond_c.f zla_gbrcond_x.f
-   zla_gbrpvgrw.f zhesvxx.f zherfsx.f zla_herfsx_extended.f
-   zla_heamv.f zla_hercond_c.f zla_hercond_x.f zla_herpvgrw.f
-   zla_lin_berr.f zlarscl2.f zlascl2.f zla_wwaddw.f)
-
-
-if(USE_XBLAS)
-  set(ALLXOBJ ${SXLASRC} ${DXLASRC} ${CXLASRC} ${ZXLASRC})
-endif()
-
-list(APPEND SLASRC DEPRECATED/sgegs.f DEPRECATED/sgegv.f
-  DEPRECATED/sgeqpf.f DEPRECATED/sgelsx.f DEPRECATED/sggsvd.f
-  DEPRECATED/sggsvp.f DEPRECATED/slahrd.f DEPRECATED/slatzm.f DEPRECATED/stzrqf.f)
-list(APPEND DLASRC DEPRECATED/dgegs.f DEPRECATED/dgegv.f
-  DEPRECATED/dgeqpf.f DEPRECATED/dgelsx.f DEPRECATED/dggsvd.f
-  DEPRECATED/dggsvp.f DEPRECATED/dlahrd.f DEPRECATED/dlatzm.f DEPRECATED/dtzrqf.f)
-list(APPEND CLASRC DEPRECATED/cgegs.f DEPRECATED/cgegv.f
-  DEPRECATED/cgeqpf.f DEPRECATED/cgelsx.f DEPRECATED/cggsvd.f
-  DEPRECATED/cggsvp.f DEPRECATED/clahrd.f DEPRECATED/clatzm.f DEPRECATED/ctzrqf.f)
-list(APPEND ZLASRC DEPRECATED/zgegs.f DEPRECATED/zgegv.f
-  DEPRECATED/zgeqpf.f DEPRECATED/zgelsx.f DEPRECATED/zggsvd.f
-  DEPRECATED/zggsvp.f DEPRECATED/zlahrd.f DEPRECATED/zlatzm.f DEPRECATED/ztzrqf.f)
-message(STATUS "Building deprecated routines")
-
-set(DSLASRC spotrs.f)
+  cbdsqr.f cgbbrd.f cgbcon.f cgbequ.f cgbrfs.f cgbsv.f  cgbsvx.f
+  cgbtf2.f cgbtrf.f cgbtrs.f cgebak.f cgebal.f cgebd2.f cgebrd.f
+  cgecon.f cgeequ.f cgees.f  cgeesx.f cgeev.f  cgeevx.f
+  DEPRECATED/cgegs.f  DEPRECATED/cgegv.f  cgehd2.f cgehrd.f cgelq2.f cgelqf.f
+  cgels.f  cgelsd.f cgelss.f DEPRECATED/cgelsx.f cgelsy.f cgeql2.f cgeqlf.f cgeqp3.f
+  DEPRECATED/cgeqpf.f cgeqr2.f cgeqr2p.f cgeqrf.f cgeqrfp.f cgerfs.f
+  cgerq2.f cgerqf.f cgesc2.f cgesdd.f  cgesvd.f
+  cgesvx.f cgetc2.f cgetri.f
+  cggbak.f cggbal.f cgges.f  cggesx.f cggev.f  cggevx.f cggglm.f
+  cgghrd.f cgglse.f cggqrf.f cggrqf.f
+  DEPRECATED/cggsvd.f DEPRECATED/cggsvp.f
+  cgtcon.f cgtrfs.f cgtsv.f  cgtsvx.f cgttrf.f cgttrs.f cgtts2.f chbev.f
+  chbevd.f chbevx.f chbgst.f chbgv.f  chbgvd.f chbgvx.f chbtrd.f
+  checon.f cheev.f  cheevd.f cheevr.f cheevx.f chegs2.f chegst.f
+  chegv.f  chegvd.f chegvx.f cherfs.f chesv.f  chesvx.f chetd2.f
+  chetf2.f chetrd.f
+  chetrf.f chetri.f chetri2.f chetri2x.f cheswapr.f
+  chetrs.f chetrs2.f
+  chetf2_rook.f chetrf_rook.f chetri_rook.f chetrs_rook.f checon_rook.f chesv_rook.f
+  chgeqz.f chpcon.f chpev.f  chpevd.f
+  chpevx.f chpgst.f chpgv.f  chpgvd.f chpgvx.f chprfs.f chpsv.f
+  chpsvx.f
+  chptrd.f chptrf.f chptri.f chptrs.f chsein.f chseqr.f clabrd.f
+  clacgv.f clacon.f clacn2.f clacp2.f clacpy.f clacrm.f clacrt.f cladiv.f
+  claed0.f claed7.f claed8.f
+  claein.f claesy.f claev2.f clags2.f clagtm.f
+  clahef.f clahef_rook.f clahqr.f
+  DEPRECATED/clahrd.f clahr2.f claic1.f clals0.f clalsa.f clalsd.f clangb.f clange.f clangt.f
+  clanhb.f clanhe.f
+  clanhp.f clanhs.f clanht.f clansb.f clansp.f clansy.f clantb.f
+  clantp.f clantr.f clapll.f clapmt.f clarcm.f claqgb.f claqge.f
+  claqhb.f claqhe.f claqhp.f claqp2.f claqps.f claqsb.f
+  claqr0.f claqr1.f claqr2.f claqr3.f claqr4.f claqr5.f
+  claqsp.f claqsy.f clar1v.f clar2v.f ilaclr.f ilaclc.f
+  clarf.f  clarfb.f clarfg.f clarft.f clarfgp.f
+  clarfx.f clargv.f clarnv.f clarrv.f clartg.f clartv.f
+  clarz.f  clarzb.f clarzt.f clascl.f claset.f clasr.f  classq.f
+  clasyf.f clasyf_rook.f clatbs.f clatdf.f clatps.f clatrd.f clatrs.f clatrz.f
+  DEPRECATED/clatzm.f cpbcon.f cpbequ.f cpbrfs.f cpbstf.f cpbsv.f
+  cpbsvx.f cpbtf2.f cpbtrf.f cpbtrs.f cpocon.f cpoequ.f cporfs.f
+  cposv.f  cposvx.f cpstrf.f cpstf2.f
+  cppcon.f cppequ.f cpprfs.f cppsv.f  cppsvx.f cpptrf.f cpptri.f cpptrs.f
+  cptcon.f cpteqr.f cptrfs.f cptsv.f  cptsvx.f cpttrf.f cpttrs.f cptts2.f
+  crot.f   cspcon.f csprfs.f cspsv.f
+  cspsvx.f csptrf.f csptri.f csptrs.f csrscl.f cstedc.f
+  cstegr.f cstein.f csteqr.f
+  csycon.f
+  csyrfs.f csysv.f  csysvx.f csytf2.f csytrf.f csytri.f csytri2.f csytri2x.f
+  csyswapr.f csytrs.f csytrs2.f csyconv.f
+  csytf2_rook.f csytrf_rook.f csytrs_rook.f
+  csytri_rook.f csycon_rook.f csysv_rook.f
+  ctbcon.f ctbrfs.f ctbtrs.f ctgevc.f ctgex2.f
+  ctgexc.f ctgsen.f ctgsja.f ctgsna.f ctgsy2.f ctgsyl.f ctpcon.f
+  ctprfs.f ctptri.f
+  ctptrs.f ctrcon.f ctrevc.f ctrevc3.f ctrexc.f ctrrfs.f ctrsen.f ctrsna.f
+  ctrsyl.f ctrtrs.f DEPRECATED/ctzrqf.f ctzrzf.f cung2l.f cung2r.f
+  cungbr.f cunghr.f cungl2.f cunglq.f cungql.f cungqr.f cungr2.f
+  cungrq.f cungtr.f cunm2l.f cunm2r.f cunmbr.f cunmhr.f cunml2.f
+  cunmlq.f cunmql.f cunmqr.f cunmr2.f cunmr3.f cunmrq.f cunmrz.f
+  cunmtr.f cupgtr.f cupmtr.f icmax1.f scsum1.f cstemr.f
+  chfrk.f ctfttp.f clanhf.f cpftrf.f cpftri.f cpftrs.f ctfsm.f ctftri.f
+  ctfttr.f ctpttf.f ctpttr.f ctrttf.f ctrttp.f
+  cgeequb.f cgbequb.f csyequb.f cpoequb.f cheequb.f
+  cbbcsd.f clapmr.f cunbdb.f cunbdb1.f cunbdb2.f cunbdb3.f cunbdb4.f
+  cunbdb5.f cunbdb6.f cuncsd.f cuncsd2by1.f
+  cgeqrt.f cgeqrt2.f cgeqrt3.f cgemqrt.f
+  ctpqrt.f ctpqrt2.f ctpmqrt.f ctprfb.f cpotri.f
+  cgelq.f cgelqt.f cgelqt3.f cgemlq.f cgemlqt.f cgemqr.f cgeqr.f cgetsls.f
+  clamswlq.f clamtsqr.f claswlq.f clatsqr.f ctplqt.f ctplqt2.f ctpmlqt.f
+  chesv_aa.f chetrf_aa.f chetrs_aa.f clahef_aa.f csytf2_rk.f clasyf_rk.f
+  csytrf_rk.f csytrs_3.f csycon_3.f csytri_3.f csytri_3x.f csysv_rk.f
+  chetf2_rk.f clahef_rk.f chetrf_rk.f chetrs_3.f checon_3.f chetri_3.f
+  chetri_3x.f chesv_rk.f chb2st_kernels.f chbev_2stage.f chbevd_2stage.f
+  chbevx_2stage.f cheev_2stage.f cheevd_2stage.f cheevr_2stage.f cheevx_2stage.f
+  chegv_2stage.f chetrd_2stage.f chetrd_hb2st.F chetrd_he2hb.f clarfy.f
+)
 
 set(ZCLASRC cpotrs.f)
 
-set(SCATGEN slatm1.f slaran.f slarnd.f)
+set(DLASRC
+  dgbbrd.f dgbcon.f dgbequ.f dgbrfs.f dgbsv.f
+  dgbsvx.f dgbtf2.f dgbtrf.f dgbtrs.f dgebak.f dgebal.f dgebd2.f
+  dgebrd.f dgecon.f dgeequ.f dgees.f  dgeesx.f dgeev.f  dgeevx.f
+  DEPRECATED/dgegs.f  DEPRECATED/dgegv.f  dgehd2.f dgehrd.f dgelq2.f dgelqf.f
+  dgels.f  dgelsd.f dgelss.f DEPRECATED/dgelsx.f dgelsy.f dgeql2.f dgeqlf.f
+  dgeqp3.f DEPRECATED/dgeqpf.f dgeqr2.f dgeqr2p.f dgeqrf.f dgeqrfp.f dgerfs.f
+  dgerq2.f dgerqf.f dgesc2.f dgesdd.f  dgesvd.f dgesvdx.f dgesvx.f
+  dgetc2.f dgetri.f dgetrf2.f
+  dggbak.f dggbal.f dgges.f  dgges3.f dggesx.f dggev.f dggev3.f  dggevx.f
+  dggglm.f dgghd3.f dgghrd.f dgglse.f dggqrf.f
+  dggrqf.f dggsvd3.f dggsvp3.f DEPRECATED/dggsvd.f DEPRECATED/dggsvp.f dgtcon.f dgtrfs.f dgtsv.f
+  dgtsvx.f dgttrf.f dgttrs.f dgtts2.f dhgeqz.f
+  dhsein.f dhseqr.f dlabrd.f dlacon.f dlacn2.f
+  dlaein.f dlaexc.f dlag2.f  dlags2.f dlagtm.f dlagv2.f dlahqr.f
+  DEPRECATED/dlahrd.f dlahr2.f dlaic1.f dlaln2.f dlals0.f dlalsa.f dlalsd.f
+  dlangb.f dlange.f dlangt.f dlanhs.f dlansb.f dlansp.f
+  dlansy.f dlantb.f dlantp.f dlantr.f dlanv2.f
+  dlapll.f dlapmt.f
+  dlaqgb.f dlaqge.f dlaqp2.f dlaqps.f dlaqsb.f dlaqsp.f dlaqsy.f
+  dlaqr0.f dlaqr1.f dlaqr2.f dlaqr3.f dlaqr4.f dlaqr5.f
+  dlaqtr.f dlar1v.f dlar2v.f iladlr.f iladlc.f
+  dlarf.f  dlarfb.f dlarfg.f dlarfgp.f dlarft.f dlarfx.f
+  dlargv.f dlarrv.f dlartv.f
+  dlarz.f  dlarzb.f dlarzt.f dlasy2.f dlasyf.f dlasyf_rook.f
+  dlatbs.f dlatdf.f dlatps.f dlatrd.f dlatrs.f dlatrz.f DEPRECATED/dlatzm.f
+  dopgtr.f dopmtr.f dorg2l.f dorg2r.f
+  dorgbr.f dorghr.f dorgl2.f dorglq.f dorgql.f dorgqr.f dorgr2.f
+  dorgrq.f dorgtr.f dorm2l.f dorm2r.f dorm22.f
+  dormbr.f dormhr.f dorml2.f dormlq.f dormql.f dormqr.f dormr2.f
+  dormr3.f dormrq.f dormrz.f dormtr.f dpbcon.f dpbequ.f dpbrfs.f
+  dpbstf.f dpbsv.f  dpbsvx.f
+  dpbtf2.f dpbtrf.f dpbtrs.f dpocon.f dpoequ.f dporfs.f dposv.f
+  dposvx.f dpotrf2.f dpotrs.f dpstrf.f dpstf2.f
+  dppcon.f dppequ.f
+  dpprfs.f dppsv.f  dppsvx.f dpptrf.f dpptri.f dpptrs.f dptcon.f
+  dpteqr.f dptrfs.f dptsv.f  dptsvx.f dpttrs.f dptts2.f drscl.f
+  dsbev.f  dsbevd.f dsbevx.f dsbgst.f dsbgv.f  dsbgvd.f dsbgvx.f
+  dsbtrd.f  dspcon.f dspev.f  dspevd.f dspevx.f dspgst.f
+  dspgv.f  dspgvd.f dspgvx.f dsprfs.f dspsv.f  dspsvx.f dsptrd.f
+  dsptrf.f dsptri.f dsptrs.f dstegr.f dstein.f dstev.f  dstevd.f dstevr.f
+  dstevx.f
+  dsycon.f dsyev.f  dsyevd.f dsyevr.f
+  dsyevx.f dsygs2.f dsygst.f dsygv.f  dsygvd.f dsygvx.f dsyrfs.f
+  dsysv.f  dsysvx.f
+  dsytd2.f dsytf2.f dsytrd.f dsytrf.f dsytri.f dsytri2.f dsytri2x.f
+  dsyswapr.f dsytrs.f dsytrs2.f dsyconv.f
+  dsytf2_rook.f dsytrf_rook.f dsytrs_rook.f
+  dsytri_rook.f dsycon_rook.f dsysv_rook.f
+  dtbcon.f dtbrfs.f dtbtrs.f dtgevc.f dtgex2.f dtgexc.f dtgsen.f
+  dtgsja.f dtgsna.f dtgsy2.f dtgsyl.f dtpcon.f dtprfs.f dtptri.f
+  dtptrs.f
+  dtrcon.f dtrevc.f dtrevc3.f dtrexc.f dtrrfs.f dtrsen.f dtrsna.f dtrsyl.f
+  dtrtrs.f DEPRECATED/dtzrqf.f dtzrzf.f dstemr.f
+  dsgesv.f dsposv.f dlag2s.f slag2d.f dlat2s.f
+  dlansf.f dpftrf.f dpftri.f dpftrs.f dsfrk.f dtfsm.f dtftri.f dtfttp.f
+  dtfttr.f dtpttf.f dtpttr.f dtrttf.f dtrttp.f
+  dgejsv.f  dgesvj.f  dgsvj0.f  dgsvj1.f
+  dgeequb.f dsyequb.f dpoequb.f dgbequb.f
+  dbbcsd.f dlapmr.f dorbdb.f dorbdb1.f dorbdb2.f dorbdb3.f dorbdb4.f
+  dorbdb5.f dorbdb6.f dorcsd.f dorcsd2by1.f
+  dgeqrt.f dgeqrt2.f dgeqrt3.f dgemqrt.f
+  dtpqrt.f dtpqrt2.f dtpmqrt.f dtprfb.f dpotri.f
+  dgelq.f dgelqt.f dgelqt3.f dgemlq.f dgemlqt.f dgemqr.f dgeqr.f dgetsls.f 
+  dlamswlq.f dlamtsqr.f dlaswlq.f dlatsqr.f dtplqt.f dtplqt2.f dtpmlqt.f 
+  dsysv_aa.f dsytrf_aa.f dsytrs_aa.f dlasyf_aa.f dsytf2_rk.f dlasyf_rk.f  
+  dsytrf_rk.f dsytrs_3.f dsycon_3.f dsytri_3.f dsytri_3x.f dsysv_rk.f   
+  dsb2st_kernels.f dsbev_2stage.f dsbevd_2stage.f dsbevx_2stage.f  
+  dsyev_2stage.f dsyevd_2stage.f dsyevr_2stage.f dsyevx_2stage.f  
+  dsygv_2stage.f dsytrd_2stage.f dsytrd_sb2st.F dsytrd_sy2sb.f dlarfy.f         
+)
 
-set(SMATGEN slatms.f slatme.f slatmr.f slatmt.f
-   slagge.f slagsy.f slakf2.f slarge.f slaror.f slarot.f slatm2.f
-   slatm3.f slatm5.f slatm6.f slatm7.f slahilb.f)
+set(ZLASRC
+  zbdsqr.f zgbbrd.f zgbcon.f zgbequ.f zgbrfs.f zgbsv.f  zgbsvx.f
+  zgbtf2.f zgbtrf.f zgbtrs.f zgebak.f zgebal.f zgebd2.f zgebrd.f
+  zgecon.f zgeequ.f zgees.f  zgeesx.f zgeev.f  zgeevx.f
+  DEPRECATED/zgegs.f  DEPRECATED/zgegv.f  zgehd2.f zgehrd.f zgejsv.f zgelq2.f zgelqf.f
+  zgels.f  zgelsd.f zgelss.f DEPRECATED/zgelsx.f zgelsy.f zgeql2.f zgeqlf.f zgeqp3.f
+  DEPRECATED/zgeqpf.f zgeqr2.f zgeqr2p.f zgeqrf.f zgeqrfp.f zgerfs.f zgerq2.f zgerqf.f
+  zgesc2.f zgesdd.f zgesvd.f zgesvdx.f zgesvj.f zgesvx.f zgetc2.f
+  zgetri.f zgetrf2.f
+  zggbak.f zggbal.f zgges.f zgges3.f zggesx.f zggev.f zggev3.f zggevx.f zggglm.f
+  zgghd3.f zgghrd.f zgglse.f zggqrf.f zggrqf.f
+  DEPRECATED/zggsvd.f zggsvd3.f DEPRECATED/zggsvp.f zggsvp3.f
+  zgsvj0.f zgsvj1.f
+  zgtcon.f zgtrfs.f zgtsv.f  zgtsvx.f zgttrf.f zgttrs.f zgtts2.f zhbev.f
+  zhbevd.f zhbevx.f zhbgst.f zhbgv.f  zhbgvd.f zhbgvx.f zhbtrd.f
+  zhecon.f zheev.f  zheevd.f zheevr.f zheevx.f zhegs2.f zhegst.f
+  zhegv.f  zhegvd.f zhegvx.f zherfs.f zhesv.f  zhesvx.f zhetd2.f
+  zhetf2.f zhetrd.f
+  zhetrf.f zhetri.f zhetri2.f zhetri2x.f zheswapr.f
+  zhetrs.f zhetrs2.f
+  zhetf2_rook.f zhetrf_rook.f zhetri_rook.f zhetrs_rook.f zhecon_rook.f zhesv_rook.f
+  zhgeqz.f zhpcon.f zhpev.f  zhpevd.f
+  zhpevx.f zhpgst.f zhpgv.f  zhpgvd.f zhpgvx.f zhprfs.f zhpsv.f
+  zhpsvx.f
+  zhptrd.f zhptrf.f zhptri.f zhptrs.f zhsein.f zhseqr.f zlabrd.f
+  zlacgv.f zlacon.f zlacn2.f zlacp2.f zlacpy.f zlacrm.f zlacrt.f zladiv.f
+  zlaed0.f zlaed7.f zlaed8.f
+  zlaein.f zlaesy.f zlaev2.f zlags2.f zlagtm.f
+  zlahef.f zlahef_rook.f zlahqr.f
+  DEPRECATED/zlahrd.f zlahr2.f zlaic1.f zlals0.f zlalsa.f zlalsd.f zlangb.f zlange.f
+  zlangt.f zlanhb.f
+  zlanhe.f
+  zlanhp.f zlanhs.f zlanht.f zlansb.f zlansp.f zlansy.f zlantb.f
+  zlantp.f zlantr.f zlapll.f zlapmt.f zlaqgb.f zlaqge.f
+  zlaqhb.f zlaqhe.f zlaqhp.f zlaqp2.f zlaqps.f zlaqsb.f
+  zlaqr0.f zlaqr1.f zlaqr2.f zlaqr3.f zlaqr4.f zlaqr5.f
+  zlaqsp.f zlaqsy.f zlar1v.f zlar2v.f ilazlr.f ilazlc.f
+  zlarcm.f zlarf.f  zlarfb.f
+  zlarfg.f zlarft.f zlarfgp.f
+  zlarfx.f zlargv.f zlarnv.f zlarrv.f zlartg.f zlartv.f
+  zlarz.f  zlarzb.f zlarzt.f zlascl.f zlaset.f zlasr.f
+  zlassq.f zlasyf.f zlasyf_rook.f zlasyf_aa.f
+  zlatbs.f zlatdf.f zlatps.f zlatrd.f zlatrs.f zlatrz.f DEPRECATED/zlatzm.f
+  zpbcon.f zpbequ.f zpbrfs.f zpbstf.f zpbsv.f
+  zpbsvx.f zpbtf2.f zpbtrf.f zpbtrs.f zpocon.f zpoequ.f zporfs.f
+  zposv.f  zposvx.f zpotrf2.f zpotrs.f zpstrf.f zpstf2.f
+  zppcon.f zppequ.f zpprfs.f zppsv.f  zppsvx.f zpptrf.f zpptri.f zpptrs.f
+  zptcon.f zpteqr.f zptrfs.f zptsv.f  zptsvx.f zpttrf.f zpttrs.f zptts2.f
+  zrot.f   zspcon.f zsprfs.f zspsv.f
+  zspsvx.f zsptrf.f zsptri.f zsptrs.f zdrscl.f zstedc.f
+  zstegr.f zstein.f zsteqr.f
+  zsycon.f zsysv_aa.f
+  zsyrfs.f zsysv.f  zsysvx.f zsytf2.f zsytrf.f zsytri.f zsytri2.f zsytri2x.f
+  zsyswapr.f zsytrs.f zsytrs_aa.f zsytrs2.f zsyconv.f
+  zsytf2_rook.f zsytrf_rook.f zsytrs_rook.f
+  zsytri_rook.f zsycon_rook.f zsysv_rook.f
+  ztbcon.f ztbrfs.f ztbtrs.f ztgevc.f ztgex2.f
+  ztgexc.f ztgsen.f ztgsja.f ztgsna.f ztgsy2.f ztgsyl.f ztpcon.f
+  ztprfs.f ztptri.f
+  ztptrs.f ztrcon.f ztrevc.f ztrevc3.f ztrexc.f ztrrfs.f ztrsen.f ztrsna.f
+  ztrsyl.f ztrtrs.f DEPRECATED/ztzrqf.f ztzrzf.f zung2l.f
+  zung2r.f zungbr.f zunghr.f zungl2.f zunglq.f zungql.f zungqr.f zungr2.f
+  zungrq.f zungtr.f zunm2l.f zunm2r.f zunmbr.f zunmhr.f zunm22.f zunml2.f
+  zunmlq.f zunmql.f zunmqr.f zunmr2.f zunmr3.f zunmrq.f zunmrz.f
+  zunmtr.f zupgtr.f
+  zupmtr.f izmax1.f dzsum1.f zstemr.f
+  zcgesv.f zcposv.f zlag2c.f clag2z.f zlat2c.f
+  zhfrk.f ztfttp.f zlanhf.f zpftrf.f zpftri.f zpftrs.f ztfsm.f ztftri.f
+  ztfttr.f ztpttf.f ztpttr.f ztrttf.f ztrttp.f
+  zgeequb.f zgbequb.f zsyequb.f zpoequb.f zheequb.f
+  zbbcsd.f zlapmr.f zunbdb.f zunbdb1.f zunbdb2.f zunbdb3.f zunbdb4.f
+  zunbdb5.f zunbdb6.f zuncsd.f zuncsd2by1.f
+  zgeqrt.f zgeqrt2.f zgeqrt3.f zgemqrt.f
+  ztpqrt.f ztpqrt2.f ztpmqrt.f ztprfb.f zpotri.f
+  zgelq.f zgelqt.f zgelqt3.f zgemlq.f zgemlqt.f zgemqr.f zgeqr.f zgetsls.f
+  zlamswlq.f zlamtsqr.f zlaswlq.f zlatsqr.f ztplqt.f ztplqt2.f ztpmlqt.f
+  zhesv_aa.f zhetrf_aa.f zhetrs_aa.f zlahef_aa.f zsytf2_rk.f zlasyf_rk.f
+  zsytrf_aa.f zsytrf_rk.f zsytrs_3.f zsycon_3.f zsytri_3.f zsytri_3x.f zsysv_rk.f
+  zhetf2_rk.f zlahef_rk.f zhetrf_rk.f zhetrs_3.f zhecon_3.f zhetri_3.f
+  zhetri_3x.f zhesv_rk.f zhb2st_kernels.f zhbev_2stage.f zhbevd_2stage.f
+  zhbevx_2stage.f zheev_2stage.f zheevd_2stage.f zheevr_2stage.f
+  zheevx_2stage.f zhegv_2stage.f zhetrd_2stage.f zhetrd_hb2st.F zhetrd_he2hb.f
+  zlarfy.f
+)
 
-set(CMATGEN clatms.f clatme.f clatmr.f clatmt.f
-   clagge.f claghe.f clagsy.f clakf2.f clarge.f claror.f clarot.f
-   clatm1.f clarnd.f clatm2.f clatm3.f clatm5.f clatm6.f clahilb.f slatm7.f)
+set(LA_REL_SRC ${ALLAUX})
+if (BUILD_SINGLE)
+  list(APPEND LA_REL_SRC ${SLASRC} ${DSLASRC} ${SCLAUX})
+endif ()
 
-set(DZATGEN dlatm1.f dlaran.f dlarnd.f)
+if (BUILD_DOUBLE)
+  list(APPEND LA_REL_SRC ${DLASRC} ${DSLASRC} ${DZLAUX})
+endif ()
 
-set(DMATGEN dlatms.f dlatme.f dlatmr.f dlatmt.f
-   dlagge.f dlagsy.f dlakf2.f dlarge.f dlaror.f dlarot.f dlatm2.f
-   dlatm3.f dlatm5.f dlatm6.f dlatm7.f dlahilb.f)
+if (BUILD_COMPLEX)
+  list(APPEND LA_REL_SRC ${CLASRC} ${ZCLASRC} ${SCLAUX})
+endif ()
 
-set(ZMATGEN zlatms.f zlatme.f zlatmr.f zlatmt.f
-  zlagge.f zlaghe.f zlagsy.f zlakf2.f zlarge.f zlaror.f zlarot.f
-  zlatm1.f zlarnd.f zlatm2.f zlatm3.f zlatm5.f zlatm6.f zlahilb.f dlatm7.f)
-
-if(BUILD_SINGLE)
-  set(LA_REL_SRC ${SLASRC} ${DSLASRC} ${ALLAUX} ${SCLAUX})
-  set(LA_GEN_SRC ${SMATGEN} ${SCATGEN})
-  message(STATUS "Building Single Precision")
-endif()
-if(BUILD_DOUBLE)
-  set(LA_REL_SRC ${LA_REL_SRC} ${DLASRC} ${DSLASRC} ${ALLAUX} ${DZLAUX})
-  set(LA_GEN_SRC ${LA_GEN_SRC} ${DMATGEN} ${DZATGEN})
-  message(STATUS "Building Double Precision")
-endif()
-if(BUILD_COMPLEX)
-  set(LA_REL_SRC ${LA_REL_SRC} ${CLASRC} ${ZCLASRC} ${ALLAUX} ${SCLAUX})
-  SET(LA_GEN_SRC ${LA_GEN_SRC} ${CMATGEN} ${SCATGEN})
-  message(STATUS "Building Complex Precision")
-endif()
-if(BUILD_COMPLEX16)
-  set(LA_REL_SRC ${LA_REL_SRC} ${ZLASRC} ${ZCLASRC} ${ALLAUX} ${DZLAUX})
-  SET(LA_GEN_SRC ${LA_GEN_SRC} ${ZMATGEN} ${DZATGEN})
-  message(STATUS "Building Double Complex Precision")
-endif()
+if (BUILD_COMPLEX16)
+  list(APPEND LA_REL_SRC ${ZLASRC} ${ZCLASRC} ${DZLAUX})
+endif ()
 
 # add lapack-netlib folder to the sources
 set(LA_SOURCES "")
 foreach (LA_FILE ${LA_REL_SRC})
   list(APPEND LA_SOURCES "${NETLIB_LAPACK_DIR}/SRC/${LA_FILE}")
 endforeach ()
-foreach (LA_FILE ${LA_GEN_SRC})
-  list(APPEND LA_SOURCES "${NETLIB_LAPACK_DIR}/TESTING/MATGEN/${LA_FILE}")
-endforeach ()
 set_source_files_properties(${LA_SOURCES} PROPERTIES COMPILE_FLAGS "${LAPACK_FFLAGS}")

cmake/lapacke.cmake

@@ -1,5 +1,5 @@
 
-set(CSRC
+set(C_SRC
   lapacke_cbbcsd.c
   lapacke_cbbcsd_work.c
   lapacke_cbdsqr.c
@@ -46,8 +46,6 @@ set(CSRC
   lapacke_cgehrd_work.c
   lapacke_cgejsv.c
   lapacke_cgejsv_work.c
-  lapacke_cgelq.c
-  lapacke_cgelq_work.c
   lapacke_cgelq2.c
   lapacke_cgelq2_work.c
   lapacke_cgelqf.c
@@ -62,18 +60,12 @@ set(CSRC
   lapacke_cgelsy_work.c
   lapacke_cgemqr.c
   lapacke_cgemqr_work.c
-  lapacke_cgemlq.c
-  lapacke_cgemlq_work.c
   lapacke_cgemqrt.c
   lapacke_cgemqrt_work.c
   lapacke_cgeqlf.c
   lapacke_cgeqlf_work.c
   lapacke_cgeqp3.c
   lapacke_cgeqp3_work.c
-  lapacke_cgeqpf.c
-  lapacke_cgeqpf_work.c
-  lapacke_cgeqr.c
-  lapacke_cgeqr_work.c
   lapacke_cgeqr2.c
   lapacke_cgeqr2_work.c
   lapacke_cgeqrf.c
@@ -142,12 +134,8 @@ set(CSRC
   lapacke_cggqrf_work.c
   lapacke_cggrqf.c
   lapacke_cggrqf_work.c
-  lapacke_cggsvd.c
-  lapacke_cggsvd_work.c
   lapacke_cggsvd3.c
   lapacke_cggsvd3_work.c
-  lapacke_cggsvp.c
-  lapacke_cggsvp_work.c
   lapacke_cggsvp3.c
   lapacke_cggsvp3_work.c
   lapacke_cgtcon.c
@@ -222,8 +210,6 @@ set(CSRC
   lapacke_chesv_work.c
   lapacke_chesv_aa.c
   lapacke_chesv_aa_work.c
-  lapacke_chesv_aa_2stage.c
-  lapacke_chesv_aa_2stage_work.c
   lapacke_chesv_rk.c
   lapacke_chesv_rk_work.c
   lapacke_chesvx.c
@@ -238,8 +224,6 @@ set(CSRC
   lapacke_chetrf_rook_work.c
   lapacke_chetrf_aa.c
   lapacke_chetrf_aa_work.c
-  lapacke_chetrf_aa_2stage.c
-  lapacke_chetrf_aa_2stage_work.c
   lapacke_chetrf_rk.c
   lapacke_chetrf_rk_work.c
   lapacke_chetri.c
@@ -258,9 +242,6 @@ set(CSRC
   lapacke_chetrs_rook_work.c
   lapacke_chetrs_aa.c
   lapacke_chetrs_aa_work.c
-  lapacke_chetrs_aa_2stage.c
-  lapacke_chetrs_aa_2stage_work.c
-  lapacke_chetrf_rk.c
   lapacke_chetrs_3.c
   lapacke_chetrs_3_work.c
   lapacke_chfrk.c
@@ -309,11 +290,6 @@ set(CSRC
   lapacke_clacp2_work.c
   lapacke_clacpy.c
   lapacke_clacpy_work.c
-  lapacke_clacrm.c
-  lapacke_clacrm_work.c
-  lapacke_clarcm.c
-  lapacke_clarcm_work.c
-  lapacke_clacn2.c
   lapacke_clag2z.c
   lapacke_clag2z_work.c
   lapacke_clange.c
@@ -342,8 +318,6 @@ set(CSRC
   lapacke_clascl_work.c
   lapacke_claset.c
   lapacke_claset_work.c
-  lapacke_classq.c
-  lapacke_classq_work.c
   lapacke_claswp.c
   lapacke_claswp_work.c
   lapacke_clauum.c
@@ -462,8 +436,6 @@ set(CSRC
   lapacke_csysv_work.c
   lapacke_csysv_aa.c
   lapacke_csysv_aa_work.c
-  lapacke_csysv_aa_2stage.c
-  lapacke_csysv_aa_2stage_work.c
   lapacke_csysv_rk.c
   lapacke_csysv_rk_work.c
   lapacke_csysvx.c
@@ -476,8 +448,6 @@ set(CSRC
   lapacke_csytrf_rook_work.c
   lapacke_csytrf_aa.c
   lapacke_csytrf_aa_work.c
-  lapacke_csytrf_aa_2stage.c
-  lapacke_csytrf_aa_2stage_work.c
   lapacke_csytrf_rk.c
   lapacke_csytrf_rk_work.c
   lapacke_csytri.c
@@ -496,8 +466,6 @@ set(CSRC
   lapacke_csytrs_rook_work.c
   lapacke_csytrs_aa.c
   lapacke_csytrs_aa_work.c
-  lapacke_csytrs_aa_2stage.c
-  lapacke_csytrs_aa_2stage_work.c
   lapacke_csytrs_3.c
   lapacke_csytrs_3_work.c
   lapacke_ctbcon.c
@@ -665,8 +633,6 @@ set(DSRC
   lapacke_dgehrd_work.c
   lapacke_dgejsv.c
   lapacke_dgejsv_work.c
-  lapacke_dgelq.c
-  lapacke_dgelq_work.c
   lapacke_dgelq2.c
   lapacke_dgelq2_work.c
   lapacke_dgelqf.c
@@ -679,8 +645,6 @@ set(DSRC
   lapacke_dgelss_work.c
   lapacke_dgelsy.c
   lapacke_dgelsy_work.c
-  lapacke_dgemlq.c
-  lapacke_dgemlq_work.c
   lapacke_dgemqr.c
   lapacke_dgemqr_work.c
   lapacke_dgemqrt.c
@@ -689,10 +653,6 @@ set(DSRC
   lapacke_dgeqlf_work.c
   lapacke_dgeqp3.c
   lapacke_dgeqp3_work.c
-  lapacke_dgeqpf.c
-  lapacke_dgeqpf_work.c
-  lapacke_dgeqr.c
-  lapacke_dgeqr_work.c
   lapacke_dgeqr2.c
   lapacke_dgeqr2_work.c
   lapacke_dgeqrf.c
@@ -761,12 +721,8 @@ set(DSRC
   lapacke_dggqrf_work.c
   lapacke_dggrqf.c
   lapacke_dggrqf_work.c
-  lapacke_dggsvd.c
-  lapacke_dggsvd_work.c
   lapacke_dggsvd3.c
   lapacke_dggsvd3_work.c
-  lapacke_dggsvp.c
-  lapacke_dggsvp_work.c
   lapacke_dggsvp3.c
   lapacke_dggsvp3_work.c
   lapacke_dgtcon.c
@@ -829,9 +785,6 @@ set(DSRC
   lapacke_dlaset_work.c
   lapacke_dlasrt.c
   lapacke_dlasrt_work.c
-  lapacke_dlassq.c
-  lapacke_dlassq_work.c
-  lapacke_dlaswp.c
   lapacke_dlaswp.c
   lapacke_dlaswp_work.c
   lapacke_dlauum.c
@@ -1072,8 +1025,6 @@ set(DSRC
   lapacke_dsysv_work.c
   lapacke_dsysv_aa.c
   lapacke_dsysv_aa_work.c
-  lapacke_dsysv_aa_2stage.c
-  lapacke_dsysv_aa_2stage_work.c
   lapacke_dsysv_rk.c
   lapacke_dsysv_rk_work.c
   lapacke_dsysvx.c
@@ -1088,8 +1039,6 @@ set(DSRC
   lapacke_dsytrf_rook_work.c
   lapacke_dsytrf_aa.c
   lapacke_dsytrf_aa_work.c
-  lapacke_dsytrf_aa_2stage.c
-  lapacke_dsytrf_aa_2stage_work.c
   lapacke_dsytrf_rk.c
   lapacke_dsytrf_rk_work.c
   lapacke_dsytri.c
@@ -1106,8 +1055,6 @@ set(DSRC
   lapacke_dsytrs2_work.c
   lapacke_dsytrs_aa.c
   lapacke_dsytrs_aa_work.c
-  lapacke_dsytrs_aa_2stage.c
-  lapacke_dsytrs_aa_2stage_work.c
   lapacke_dsytrs_3.c
   lapacke_dsytrs_3_work.c
   lapacke_dsytrs_work.c
@@ -1237,8 +1184,6 @@ set(SSRC
   lapacke_sgehrd_work.c
   lapacke_sgejsv.c
   lapacke_sgejsv_work.c
-  lapacke_sgelq.c
-  lapacke_sgelq_work.c
   lapacke_sgelq2.c
   lapacke_sgelq2_work.c
   lapacke_sgelqf.c
@@ -1251,8 +1196,6 @@ set(SSRC
   lapacke_sgelss_work.c
   lapacke_sgelsy.c
   lapacke_sgelsy_work.c
-  lapacke_sgemlq.c
-  lapacke_sgemlq_work.c
   lapacke_sgemqr.c
   lapacke_sgemqr_work.c
   lapacke_sgemqrt.c
@@ -1261,10 +1204,6 @@ set(SSRC
   lapacke_sgeqlf_work.c
   lapacke_sgeqp3.c
   lapacke_sgeqp3_work.c
-  lapacke_sgeqpf.c
-  lapacke_sgeqpf_work.c
-  lapacke_sgeqr.c
-  lapacke_sgeqr_work.c
   lapacke_sgeqr2.c
   lapacke_sgeqr2_work.c
   lapacke_sgeqrf.c
@@ -1333,12 +1272,8 @@ set(SSRC
   lapacke_sggqrf_work.c
   lapacke_sggrqf.c
   lapacke_sggrqf_work.c
-  lapacke_sggsvd.c
-  lapacke_sggsvd_work.c
   lapacke_sggsvd3.c
   lapacke_sggsvd3_work.c
-  lapacke_sggsvp.c
-  lapacke_sggsvp_work.c
   lapacke_sggsvp3.c
   lapacke_sggsvp3_work.c
   lapacke_sgtcon.c
@@ -1401,8 +1336,6 @@ set(SSRC
   lapacke_slaset_work.c
   lapacke_slasrt.c
   lapacke_slasrt_work.c
-  lapacke_slassq.c
-  lapacke_slassq_work.c
   lapacke_slaswp.c
   lapacke_slaswp_work.c
   lapacke_slauum.c
@@ -1639,8 +1572,6 @@ set(SSRC
   lapacke_ssysv_work.c
   lapacke_ssysv_aa.c
   lapacke_ssysv_aa_work.c
-  lapacke_ssysv_aa_2stage.c
-  lapacke_ssysv_aa_2stage_work.c
   lapacke_ssysv_rk.c
   lapacke_ssysv_rk_work.c
   lapacke_ssysvx.c
@@ -1655,9 +1586,6 @@ set(SSRC
   lapacke_ssytrf_rook_work.c
   lapacke_ssytrf_aa.c
   lapacke_ssytrf_aa_work.c
-  lapacke_ssytrf_aa_2stage.c
-  lapacke_ssytrf_aa_2stage_work.c
-  lapacke_ssytrf_rook.c
   lapacke_ssytrf_rk.c
   lapacke_ssytrf_rk_work.c
   lapacke_ssytri.c
@@ -1674,8 +1602,6 @@ set(SSRC
   lapacke_ssytrs2_work.c
   lapacke_ssytrs_aa.c
   lapacke_ssytrs_aa_work.c
-  lapacke_ssytrs_aa_2stage.c
-  lapacke_ssytrs_aa_2stage_work.c
   lapacke_ssytrs_3.c
   lapacke_ssytrs_3_work.c
   lapacke_ssytrs_work.c
@@ -1803,8 +1729,6 @@ set(ZSRC
   lapacke_zgehrd_work.c
   lapacke_zgejsv.c
   lapacke_zgejsv_work.c
-  lapacke_zgelq.c
-  lapacke_zgelq_work.c
   lapacke_zgelq2.c
   lapacke_zgelq2_work.c
   lapacke_zgelqf.c
@@ -1817,8 +1741,6 @@ set(ZSRC
   lapacke_zgelss_work.c
   lapacke_zgelsy.c
   lapacke_zgelsy_work.c
-  lapacke_zgemlq.c
-  lapacke_zgemlq_work.c
   lapacke_zgemqr.c
   lapacke_zgemqr_work.c
   lapacke_zgemqrt.c
@@ -1827,10 +1749,6 @@ set(ZSRC
   lapacke_zgeqlf_work.c
   lapacke_zgeqp3.c
   lapacke_zgeqp3_work.c
-  lapacke_zgeqpf.c
-  lapacke_zgeqpf_work.c
-  lapacke_zgeqr.c
-  lapacke_zgeqr_work.c
   lapacke_zgeqr2.c
   lapacke_zgeqr2_work.c
   lapacke_zgeqrf.c
@@ -1899,12 +1817,8 @@ set(ZSRC
   lapacke_zggqrf_work.c
   lapacke_zggrqf.c
   lapacke_zggrqf_work.c
-  lapacke_zggsvd.c
-  lapacke_zggsvd_work.c
   lapacke_zggsvd3.c
   lapacke_zggsvd3_work.c
-  lapacke_zggsvp.c
-  lapacke_zggsvp_work.c
   lapacke_zggsvp3.c
   lapacke_zggsvp3_work.c
   lapacke_zgtcon.c
@@ -1925,12 +1839,6 @@ set(ZSRC
   lapacke_zhbevd_work.c
   lapacke_zhbevx.c
   lapacke_zhbevx_work.c
-  lapacke_zhbev_2stage.c
-  lapacke_zhbev_2stage_work.c
-  lapacke_zhbevd_2stage.c
-  lapacke_zhbevd_2stage_work.c
-  lapacke_zhbevx_2stage.c
-  lapacke_zhbevx_2stage_work.c
   lapacke_zhbgst.c
   lapacke_zhbgst_work.c
   lapacke_zhbgv.c
@@ -1979,8 +1887,6 @@ set(ZSRC
   lapacke_zhesv_work.c
   lapacke_zhesv_aa.c
   lapacke_zhesv_aa_work.c
-  lapacke_zhesv_aa_2stage.c
-  lapacke_zhesv_aa_2stage_work.c
   lapacke_zhesv_rk.c
   lapacke_zhesv_rk_work.c
   lapacke_zhesvx.c
@@ -1995,8 +1901,6 @@ set(ZSRC
   lapacke_zhetrf_rook_work.c
   lapacke_zhetrf_aa.c
   lapacke_zhetrf_aa_work.c
-  lapacke_zhetrf_aa_2stage.c
-  lapacke_zhetrf_aa_2stage_work.c
   lapacke_zhetrf_rk.c
   lapacke_zhetrf_rk_work.c
   lapacke_zhetri.c
@@ -2014,8 +1918,6 @@ set(ZSRC
   lapacke_zhetrs_work.c
   lapacke_zhetrs_aa.c
   lapacke_zhetrs_aa_work.c
-  lapacke_zhetrs_aa_2stage.c
-  lapacke_zhetrs_aa_2stage_work.c
   lapacke_zhetrs_3.c
   lapacke_zhetrs_3_work.c
   lapacke_zhetrs_rook_work.c
@@ -2065,8 +1967,6 @@ set(ZSRC
   lapacke_zlacp2_work.c
   lapacke_zlacpy.c
   lapacke_zlacpy_work.c
-  lapacke_zlacrm.c
-  lapacke_zlacrm_work.c
   lapacke_zlag2c.c
   lapacke_zlag2c_work.c
   lapacke_zlange.c
@@ -2081,8 +1981,6 @@ set(ZSRC
   lapacke_zlapmr_work.c
   lapacke_zlapmt.c
   lapacke_zlapmt_work.c
-  lapacke_zlarcm.c
-  lapacke_zlarcm_work.c
   lapacke_zlarfb.c
   lapacke_zlarfb_work.c
   lapacke_zlarfg.c
@@ -2097,8 +1995,6 @@ set(ZSRC
   lapacke_zlascl_work.c
   lapacke_zlaset.c
   lapacke_zlaset_work.c
-  lapacke_zlassq.c
-  lapacke_zlassq_work.c
   lapacke_zlaswp.c
   lapacke_zlaswp_work.c
   lapacke_zlauum.c
@@ -2217,8 +2113,6 @@ set(ZSRC
   lapacke_zsysv_work.c
   lapacke_zsysv_aa.c
   lapacke_zsysv_aa_work.c
-  lapacke_zsysv_aa_2stage.c
-  lapacke_zsysv_aa_2stage_work.c
   lapacke_zsysv_rk.c
   lapacke_zsysv_rk_work.c
   lapacke_zsysvx.c
@@ -2231,8 +2125,6 @@ set(ZSRC
   lapacke_zsytrf_rook_work.c
   lapacke_zsytrf_aa.c
   lapacke_zsytrf_aa_work.c
-  lapacke_zsytrf_aa_2stage.c
-  lapacke_zsytrf_aa_2stage_work.c
   lapacke_zsytrf_rk.c
   lapacke_zsytrf_rk_work.c
   lapacke_zsytri.c
@@ -2251,8 +2143,6 @@ set(ZSRC
   lapacke_zsytrs_rook_work.c
   lapacke_zsytrs_aa.c
   lapacke_zsytrs_aa_work.c
-  lapacke_zsytrs_aa_2stage.c
-  lapacke_zsytrs_aa_2stage_work.c
   lapacke_zsytrs_3.c
   lapacke_zsytrs_3_work.c
   lapacke_ztbcon.c
@@ -2373,92 +2263,104 @@ set(ZSRC
 )
 
 set(SRCX
-  lapacke_cgbrfsx.c       lapacke_cporfsx.c       lapacke_dgerfsx.c       lapacke_sgbrfsx.c       lapacke_ssyrfsx.c       lapacke_zherfsx.c
-  lapacke_cgbrfsx_work.c  lapacke_cporfsx_work.c  lapacke_dgerfsx_work.c  lapacke_sgbrfsx_work.c  lapacke_ssyrfsx_work.c  lapacke_zherfsx_work.c
-  lapacke_cgerfsx.c       lapacke_csyrfsx.c       lapacke_dporfsx.c       lapacke_sgerfsx.c       lapacke_zgbrfsx.c       lapacke_zporfsx.c
-  lapacke_cgerfsx_work.c  lapacke_csyrfsx_work.c  lapacke_dporfsx_work.c  lapacke_sgerfsx_work.c  lapacke_zgbrfsx_work.c  lapacke_zporfsx_work.c
-  lapacke_cherfsx.c       lapacke_dgbrfsx.c       lapacke_dsyrfsx.c       lapacke_sporfsx.c       lapacke_zgerfsx.c       lapacke_zsyrfsx.c
-  lapacke_cherfsx_work.c  lapacke_dgbrfsx_work.c  lapacke_dsyrfsx_work.c  lapacke_sporfsx_work.c  lapacke_zgerfsx_work.c  lapacke_zsyrfsx_work.c
-  lapacke_cgbsvxx.c       lapacke_cposvxx.c       lapacke_dgesvxx.c       lapacke_sgbsvxx.c       lapacke_ssysvxx.c       lapacke_zhesvxx.c
-  lapacke_cgbsvxx_work.c  lapacke_cposvxx_work.c  lapacke_dgesvxx_work.c  lapacke_sgbsvxx_work.c  lapacke_ssysvxx_work.c  lapacke_zhesvxx_work.c
-  lapacke_cgesvxx.c       lapacke_csysvxx.c       lapacke_dposvxx.c       lapacke_sgesvxx.c       lapacke_zgbsvxx.c       lapacke_zposvxx.c
-  lapacke_cgesvxx_work.c  lapacke_csysvxx_work.c  lapacke_dposvxx_work.c  lapacke_sgesvxx_work.c  lapacke_zgbsvxx_work.c  lapacke_zposvxx_work.c
-  lapacke_chesvxx.c       lapacke_dgbsvxx.c       lapacke_dsysvxx.c       lapacke_sposvxx.c       lapacke_zgesvxx.c       lapacke_zsysvxx.c
+  lapacke_cgbrfsx.c       lapacke_cporfsx.c       lapacke_dgerfsx.c       lapacke_sgbrfsx.c       lapacke_ssyrfsx.c       lapacke_zherfsx.c 
+  lapacke_cgbrfsx_work.c  lapacke_cporfsx_work.c  lapacke_dgerfsx_work.c  lapacke_sgbrfsx_work.c  lapacke_ssyrfsx_work.c  lapacke_zherfsx_work.c 
+  lapacke_cgerfsx.c       lapacke_csyrfsx.c       lapacke_dporfsx.c       lapacke_sgerfsx.c       lapacke_zgbrfsx.c       lapacke_zporfsx.c 
+  lapacke_cgerfsx_work.c  lapacke_csyrfsx_work.c  lapacke_dporfsx_work.c  lapacke_sgerfsx_work.c  lapacke_zgbrfsx_work.c  lapacke_zporfsx_work.c 
+  lapacke_cherfsx.c       lapacke_dgbrfsx.c       lapacke_dsyrfsx.c       lapacke_sporfsx.c       lapacke_zgerfsx.c       lapacke_zsyrfsx.c 
+  lapacke_cherfsx_work.c  lapacke_dgbrfsx_work.c  lapacke_dsyrfsx_work.c  lapacke_sporfsx_work.c  lapacke_zgerfsx_work.c  lapacke_zsyrfsx_work.c 
+  lapacke_cgbsvxx.c       lapacke_cposvxx.c       lapacke_dgesvxx.c       lapacke_sgbsvxx.c       lapacke_ssysvxx.c       lapacke_zhesvxx.c 
+  lapacke_cgbsvxx_work.c  lapacke_cposvxx_work.c  lapacke_dgesvxx_work.c  lapacke_sgbsvxx_work.c  lapacke_ssysvxx_work.c  lapacke_zhesvxx_work.c 
+  lapacke_cgesvxx.c       lapacke_csysvxx.c       lapacke_dposvxx.c       lapacke_sgesvxx.c       lapacke_zgbsvxx.c       lapacke_zposvxx.c 
+  lapacke_cgesvxx_work.c  lapacke_csysvxx_work.c  lapacke_dposvxx_work.c  lapacke_sgesvxx_work.c  lapacke_zgbsvxx_work.c  lapacke_zposvxx_work.c 
+  lapacke_chesvxx.c       lapacke_dgbsvxx.c       lapacke_dsysvxx.c       lapacke_sposvxx.c       lapacke_zgesvxx.c       lapacke_zsysvxx.c 
   lapacke_chesvxx_work.c  lapacke_dgbsvxx_work.c  lapacke_dsysvxx_work.c  lapacke_sposvxx_work.c  lapacke_zgesvxx_work.c  lapacke_zsysvxx_work.c
 )
 
 
-# FILE PARTS OF TMGLIB
+# FILE PARTS OF TMGLIB 
 set(MATGEN
-  lapacke_clatms.c
-  lapacke_clatms_work.c
-  lapacke_dlatms.c
-  lapacke_dlatms_work.c
-  lapacke_slatms.c
-  lapacke_slatms_work.c
-  lapacke_zlatms.c
-  lapacke_zlatms_work.c
-  lapacke_clagge.c
-  lapacke_clagge_work.c
-  lapacke_dlagge.c
-  lapacke_dlagge_work.c
-  lapacke_slagge.c
-  lapacke_slagge_work.c
-  lapacke_zlagge.c
-  lapacke_zlagge_work.c
-  lapacke_claghe.c
-  lapacke_claghe_work.c
-  lapacke_zlaghe.c
-  lapacke_zlaghe_work.c
-  lapacke_clagsy.c
-  lapacke_clagsy_work.c
-  lapacke_dlagsy.c
-  lapacke_dlagsy_work.c
-  lapacke_slagsy.c
-  lapacke_slagsy_work.c
-  lapacke_zlagsy.c
+  lapacke_clatms.c 
+  lapacke_clatms_work.c 
+  lapacke_dlatms.c 
+  lapacke_dlatms_work.c 
+  lapacke_slatms.c 
+  lapacke_slatms_work.c 
+  lapacke_zlatms.c 
+  lapacke_zlatms_work.c 
+  lapacke_clagge.c 
+  lapacke_clagge_work.c 
+  lapacke_dlagge.c 
+  lapacke_dlagge_work.c 
+  lapacke_slagge.c 
+  lapacke_slagge_work.c 
+  lapacke_zlagge.c 
+  lapacke_zlagge_work.c 
+  lapacke_claghe.c 
+  lapacke_claghe_work.c 
+  lapacke_zlaghe.c 
+  lapacke_zlaghe_work.c 
+  lapacke_clagsy.c 
+  lapacke_clagsy_work.c 
+  lapacke_dlagsy.c 
+  lapacke_dlagsy_work.c 
+  lapacke_slagsy.c 
+  lapacke_slagsy_work.c 
+  lapacke_zlagsy.c 
   lapacke_zlagsy_work.c
-  lapacke_nancheck.c
 )
 
 set(Utils_SRC
-  lapacke_c_nancheck.c            lapacke_ctr_trans.c             lapacke_make_complex_float.c    lapacke_zgb_nancheck.c
-  lapacke_cgb_nancheck.c          lapacke_d_nancheck.c            lapacke_s_nancheck.c            lapacke_zgb_trans.c
-  lapacke_cgb_trans.c             lapacke_dgb_nancheck.c          lapacke_sgb_nancheck.c          lapacke_zge_nancheck.c
-  lapacke_cge_nancheck.c          lapacke_dgb_trans.c             lapacke_sgb_trans.c             lapacke_zge_trans.c
-  lapacke_cge_trans.c             lapacke_dge_nancheck.c          lapacke_sge_nancheck.c          lapacke_zgg_nancheck.c
-  lapacke_cgg_nancheck.c          lapacke_dge_trans.c             lapacke_sge_trans.c             lapacke_zgg_trans.c
-  lapacke_cgg_trans.c             lapacke_dgg_nancheck.c          lapacke_sgg_nancheck.c          lapacke_zgt_nancheck.c
-  lapacke_cgt_nancheck.c          lapacke_dgg_trans.c             lapacke_sgg_trans.c             lapacke_zhb_nancheck.c
-  lapacke_chb_nancheck.c          lapacke_dgt_nancheck.c          lapacke_sgt_nancheck.c          lapacke_zhb_trans.c
-  lapacke_chb_trans.c             lapacke_dhs_nancheck.c          lapacke_shs_nancheck.c          lapacke_zhe_nancheck.c
-  lapacke_che_nancheck.c          lapacke_dhs_trans.c             lapacke_shs_trans.c             lapacke_zhe_trans.c
-  lapacke_che_trans.c             lapacke_dpb_nancheck.c          lapacke_spb_nancheck.c          lapacke_zhp_nancheck.c
-  lapacke_chp_nancheck.c          lapacke_dpb_trans.c             lapacke_spb_trans.c             lapacke_zhp_trans.c
-  lapacke_chp_trans.c             lapacke_dpf_nancheck.c          lapacke_spf_nancheck.c          lapacke_zhs_nancheck.c
-  lapacke_chs_nancheck.c          lapacke_dpf_trans.c             lapacke_spf_trans.c             lapacke_zhs_trans.c
-  lapacke_chs_trans.c             lapacke_dpo_nancheck.c          lapacke_spo_nancheck.c          lapacke_zpb_nancheck.c
-  lapacke_cpb_nancheck.c          lapacke_dpo_trans.c             lapacke_spo_trans.c             lapacke_zpb_trans.c
-  lapacke_cpb_trans.c             lapacke_dpp_nancheck.c          lapacke_spp_nancheck.c          lapacke_zpf_nancheck.c
-  lapacke_cpf_nancheck.c          lapacke_dpp_trans.c             lapacke_spp_trans.c             lapacke_zpf_trans.c
-  lapacke_cpf_trans.c             lapacke_dpt_nancheck.c          lapacke_spt_nancheck.c          lapacke_zpo_nancheck.c
-  lapacke_cpo_nancheck.c          lapacke_dsb_nancheck.c          lapacke_ssb_nancheck.c          lapacke_zpo_trans.c
-  lapacke_cpo_trans.c             lapacke_dsb_trans.c             lapacke_ssb_trans.c             lapacke_zpp_nancheck.c
-  lapacke_cpp_nancheck.c          lapacke_dsp_nancheck.c          lapacke_ssp_nancheck.c          lapacke_zpp_trans.c
-  lapacke_cpp_trans.c             lapacke_dsp_trans.c             lapacke_ssp_trans.c             lapacke_zpt_nancheck.c
-  lapacke_cpt_nancheck.c          lapacke_dst_nancheck.c          lapacke_sst_nancheck.c          lapacke_zsp_nancheck.c
-  lapacke_csp_nancheck.c          lapacke_dsy_nancheck.c          lapacke_ssy_nancheck.c          lapacke_zsp_trans.c
-  lapacke_csp_trans.c             lapacke_dsy_trans.c             lapacke_ssy_trans.c             lapacke_zst_nancheck.c
-  lapacke_cst_nancheck.c          lapacke_dtb_nancheck.c          lapacke_stb_nancheck.c          lapacke_zsy_nancheck.c
-  lapacke_csy_nancheck.c          lapacke_dtb_trans.c             lapacke_stb_trans.c             lapacke_zsy_trans.c
-  lapacke_csy_trans.c             lapacke_dtf_nancheck.c          lapacke_stf_nancheck.c          lapacke_ztb_nancheck.c
-  lapacke_ctb_nancheck.c          lapacke_dtf_trans.c             lapacke_stf_trans.c             lapacke_ztb_trans.c
-  lapacke_ctb_trans.c             lapacke_dtp_nancheck.c          lapacke_stp_nancheck.c          lapacke_ztf_nancheck.c
-  lapacke_ctf_nancheck.c          lapacke_dtp_trans.c             lapacke_stp_trans.c             lapacke_ztf_trans.c
-  lapacke_ctf_trans.c             lapacke_dtr_nancheck.c          lapacke_str_nancheck.c          lapacke_ztp_nancheck.c
-  lapacke_ctp_nancheck.c          lapacke_dtr_trans.c             lapacke_str_trans.c             lapacke_ztp_trans.c
-  lapacke_ctp_trans.c             lapacke_lsame.c                 lapacke_xerbla.c                lapacke_ztr_nancheck.c
-  lapacke_ctr_nancheck.c          lapacke_make_complex_double.c   lapacke_z_nancheck.c            lapacke_ztr_trans.c
+lapacke_cgb_nancheck.c  lapacke_dpf_nancheck.c         lapacke_ssy_trans.c
+lapacke_cgb_trans.c     lapacke_dpf_trans.c            lapacke_stb_nancheck.c
+lapacke_cge_nancheck.c  lapacke_dpo_nancheck.c         lapacke_stb_trans.c
+lapacke_cge_trans.c     lapacke_dpo_trans.c            lapacke_stf_nancheck.c
+lapacke_cgg_nancheck.c  lapacke_dpp_nancheck.c         lapacke_stf_trans.c
+lapacke_cgg_trans.c     lapacke_dpp_trans.c            lapacke_stp_nancheck.c
+lapacke_cgt_nancheck.c  lapacke_dpt_nancheck.c         lapacke_stp_trans.c
+lapacke_chb_nancheck.c  lapacke_dsb_nancheck.c         lapacke_str_nancheck.c
+lapacke_chb_trans.c     lapacke_dsb_trans.c            lapacke_str_trans.c
+lapacke_che_nancheck.c  lapacke_dsp_nancheck.c         lapacke_xerbla.c
+lapacke_che_trans.c     lapacke_dsp_trans.c            lapacke_zgb_nancheck.c
+lapacke_chp_nancheck.c  lapacke_dst_nancheck.c         lapacke_zgb_trans.c
+lapacke_chp_trans.c     lapacke_dsy_nancheck.c         lapacke_zge_nancheck.c
+lapacke_chs_nancheck.c  lapacke_dsy_trans.c            lapacke_zge_trans.c
+lapacke_chs_trans.c     lapacke_dtb_nancheck.c         lapacke_zgg_nancheck.c
+lapacke_c_nancheck.c    lapacke_dtb_trans.c            lapacke_zgg_trans.c
+lapacke_cpb_nancheck.c  lapacke_dtf_nancheck.c         lapacke_zgt_nancheck.c
+lapacke_cpb_trans.c     lapacke_dtf_trans.c            lapacke_zhb_nancheck.c
+lapacke_cpf_nancheck.c  lapacke_dtp_nancheck.c         lapacke_zhb_trans.c
+lapacke_cpf_trans.c     lapacke_dtp_trans.c            lapacke_zhe_nancheck.c
+lapacke_cpo_nancheck.c  lapacke_dtr_nancheck.c         lapacke_zhe_trans.c
+lapacke_cpo_trans.c     lapacke_dtr_trans.c            lapacke_zhp_nancheck.c
+lapacke_cpp_nancheck.c  lapacke_lsame.c                lapacke_zhp_trans.c
+lapacke_cpp_trans.c     lapacke_make_complex_double.c  lapacke_zhs_nancheck.c
+lapacke_cpt_nancheck.c  lapacke_make_complex_float.c   lapacke_zhs_trans.c
+lapacke_csp_nancheck.c  lapacke_sgb_nancheck.c         lapacke_z_nancheck.c
+lapacke_csp_trans.c     lapacke_sgb_trans.c            lapacke_zpb_nancheck.c
+lapacke_cst_nancheck.c  lapacke_sge_nancheck.c         lapacke_zpb_trans.c
+lapacke_csy_nancheck.c  lapacke_sge_trans.c            lapacke_zpf_nancheck.c
+lapacke_csy_trans.c     lapacke_sgg_nancheck.c         lapacke_zpf_trans.c
+lapacke_ctb_nancheck.c  lapacke_sgg_trans.c            lapacke_zpo_nancheck.c
+lapacke_ctb_trans.c     lapacke_sgt_nancheck.c         lapacke_zpo_trans.c
+lapacke_ctf_nancheck.c  lapacke_shs_nancheck.c         lapacke_zpp_nancheck.c
+lapacke_ctf_trans.c     lapacke_shs_trans.c            lapacke_zpp_trans.c
+lapacke_ctp_nancheck.c  lapacke_s_nancheck.c           lapacke_zpt_nancheck.c
+lapacke_ctp_trans.c     lapacke_spb_nancheck.c         lapacke_zsp_nancheck.c
+lapacke_ctr_nancheck.c  lapacke_spb_trans.c            lapacke_zsp_trans.c
+lapacke_ctr_trans.c     lapacke_spf_nancheck.c         lapacke_zst_nancheck.c
+lapacke_dgb_nancheck.c  lapacke_spf_trans.c            lapacke_zsy_nancheck.c
+lapacke_dgb_trans.c     lapacke_spo_nancheck.c         lapacke_zsy_trans.c
+lapacke_dge_nancheck.c  lapacke_spo_trans.c            lapacke_ztb_nancheck.c
+lapacke_dge_trans.c     lapacke_spp_nancheck.c         lapacke_ztb_trans.c
+lapacke_dgg_nancheck.c  lapacke_spp_trans.c            lapacke_ztf_nancheck.c
+lapacke_dgg_trans.c     lapacke_spt_nancheck.c         lapacke_ztf_trans.c
+lapacke_dgt_nancheck.c  lapacke_ssb_nancheck.c         lapacke_ztp_nancheck.c
+lapacke_dhs_nancheck.c  lapacke_ssb_trans.c            lapacke_ztp_trans.c
+lapacke_dhs_trans.c     lapacke_ssp_nancheck.c         lapacke_ztr_nancheck.c
+lapacke_d_nancheck.c    lapacke_ssp_trans.c            lapacke_ztr_trans.c
+lapacke_dpb_nancheck.c  lapacke_sst_nancheck.c
+lapacke_dpb_trans.c     lapacke_ssy_nancheck.c
 )
 
 set(LAPACKE_REL_SRC "")
@@ -2478,10 +2380,6 @@ if (BUILD_COMPLEX16)
   list(APPEND LAPACKE_REL_SRC ${ZSRC})
 endif ()
 
-if (BUILD_MATGEN)
-  list(APPEND LAPACKE_REL_SRC ${MATGEN})
-endif ()
-
 # add lapack-netlib folder to the sources
 set(LAPACKE_SOURCES "")
 foreach (LAE_FILE ${LAPACKE_REL_SRC})
@@ -2493,6 +2391,6 @@ foreach (Utils_FILE ${Utils_SRC})
 endforeach ()
 
 set(lapacke_include_dir "${NETLIB_LAPACK_DIR}/LAPACKE/include")
-configure_file("${lapacke_include_dir}/lapacke_mangling_with_flags.h.in" "${lapacke_include_dir}/lapacke_mangling.h" COPYONLY)
+execute_process(COMMAND ${CMAKE_COMMAND} -E copy "${lapacke_include_dir}/lapacke_mangling_with_flags.h.in" "${lapacke_include_dir}/lapacke_mangling.h")
 include_directories(${lapacke_include_dir})
 set_source_files_properties(${LAPACKE_SOURCES} PROPERTIES COMPILE_FLAGS "${LAPACK_CFLAGS}")

cmake/openblas.pc.in

@@ -1,7 +1,6 @@
 libdir=@CMAKE_INSTALL_FULL_LIBDIR@
 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@

cmake/os.cmake

@@ -3,6 +3,19 @@
 ## Description: Ported from portion of OpenBLAS/Makefile.system
 ##              Detects the OS and sets appropriate variables.
 
+if (${CMAKE_SYSTEM_NAME} STREQUAL "Darwin")
+  set(ENV{MACOSX_DEPLOYMENT_TARGET} "10.2") # TODO: should be exported as an env var
+  set(MD5SUM "md5 -r")
+endif ()
+
+if (${CMAKE_SYSTEM_NAME} STREQUAL "FreeBSD")
+  set(MD5SUM "md5 -r")
+endif ()
+
+if (${CMAKE_SYSTEM_NAME} STREQUAL "NetBSD")
+  set(MD5SUM "md5 -n")
+endif ()
+
 if (${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
   set(EXTRALIB "${EXTRALIB} -lm")
   set(NO_EXPRECISION 1)
@@ -43,7 +56,7 @@ if (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
 
   # Ensure the correct stack alignment on Win32
   # http://permalink.gmane.org/gmane.comp.lib.openblas.general/97
-  if (X86)
+  if (${ARCH} STREQUAL "x86")
     if (NOT MSVC AND NOT ${CMAKE_C_COMPILER_ID} STREQUAL "Clang")
       set(CCOMMON_OPT "${CCOMMON_OPT} -mincoming-stack-boundary=2")
     endif ()
@@ -65,7 +78,7 @@ if (CYGWIN)
 endif ()
 
 if (NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Windows" AND NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Interix" AND NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Android")
-  if (USE_THREAD)
+  if (SMP)
     set(EXTRALIB "${EXTRALIB} -lpthread")
   endif ()
 endif ()
@@ -75,7 +88,7 @@ if (QUAD_PRECISION)
   set(NO_EXPRECISION 1)
 endif ()
 
-if (X86)
+if (${ARCH} STREQUAL "x86")
   set(NO_EXPRECISION 1)
 endif ()
 

cmake/prebuild.cmake

@@ -37,208 +37,112 @@
 
 # CPUIDEMU = ../../cpuid/table.o
 
-
 if (DEFINED CPUIDEMU)
   set(EXFLAGS "-DCPUIDEMU -DVENDOR=99")
 endif ()
 
-if (BUILD_KERNEL)
+if (DEFINED TARGET_CORE)
   # set the C flags for just this file
   set(GETARCH2_FLAGS "-DBUILD_KERNEL")
+  set(TARGET_MAKE "Makefile_kernel.conf")
   set(TARGET_CONF "config_kernel.h")
-  set(TARGET_CONF_DIR ${PROJECT_BINARY_DIR}/kernel_config/${TARGET_CORE})
 else()
+  set(TARGET_MAKE "Makefile.conf")
   set(TARGET_CONF "config.h")
-  set(TARGET_CONF_DIR ${PROJECT_BINARY_DIR})
 endif ()
 
-set(TARGET_CONF_TEMP "${PROJECT_BINARY_DIR}/${TARGET_CONF}.tmp")
+include("${PROJECT_SOURCE_DIR}/cmake/c_check.cmake")
 
-# c_check
-set(FU "")
-if (APPLE OR (MSVC AND NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang"))
-  set(FU "_")
-endif()
-
-set(COMPILER_ID ${CMAKE_C_COMPILER_ID})
-if (${COMPILER_ID} STREQUAL "GNU")
-  set(COMPILER_ID "GCC")
-endif ()
-
-string(TOUPPER ${ARCH} UC_ARCH)
-
-file(WRITE ${TARGET_CONF_TEMP}
-  "#define OS_${HOST_OS}\t1\n"
-  "#define ARCH_${UC_ARCH}\t1\n"
-  "#define C_${COMPILER_ID}\t1\n"
-  "#define __${BINARY}BIT__\t1\n"
-  "#define FUNDERSCORE\t${FU}\n")
-
-if (${HOST_OS} STREQUAL "WINDOWSSTORE")
-  file(APPEND ${TARGET_CONF_TEMP}
-    "#define OS_WINNT\t1\n")
-endif ()
-
-# f_check
 if (NOT NOFORTRAN)
   include("${PROJECT_SOURCE_DIR}/cmake/f_check.cmake")
 endif ()
 
-# Cannot run getarch on target if we are cross-compiling
-if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSSTORE"))
-  # Write to config as getarch would
+# compile getarch
+set(GETARCH_SRC
+  ${PROJECT_SOURCE_DIR}/getarch.c
+  ${CPUIDEMO}
+)
 
-  # 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 ${CORE}\n"
-    "#define CHAR_CORENAME \"${CORE}\"\n")
-  if ("${CORE}" STREQUAL "ARMV7")
-    file(APPEND ${TARGET_CONF_TEMP}
-      "#define L1_DATA_SIZE\t65536\n"
-      "#define L1_DATA_LINESIZE\t32\n"
-      "#define L2_SIZE\t512488\n"
-      "#define L2_LINESIZE\t32\n"
-      "#define DTB_DEFAULT_ENTRIES\t64\n"
-      "#define DTB_SIZE\t4096\n"
-      "#define L2_ASSOCIATIVE\t4\n"
-      "#define HAVE_VFPV3\n"
-      "#define HAVE_VFP\n")
-    set(SGEMM_UNROLL_M 4)
-    set(SGEMM_UNROLL_N 4)
-    set(DGEMM_UNROLL_M 4)
-    set(DGEMM_UNROLL_N 4)
-  elseif ("${CORE}" STREQUAL "ARMV8")
-    file(APPEND ${TARGET_CONF_TEMP}
-      "#define L1_DATA_SIZE\t32768\n"
-      "#define L1_DATA_LINESIZE\t64\n"
-      "#define L2_SIZE\t262144\n"
-      "#define L2_LINESIZE\t64\n"
-      "#define DTB_DEFAULT_ENTRIES\t64\n"
-      "#define DTB_SIZE\t4096\n"
-      "#define L2_ASSOCIATIVE\t32\n")
-    set(SGEMM_UNROLL_M 4)
-    set(SGEMM_UNROLL_N 4)
-  elseif ("${CORE}" STREQUAL "CORTEXA57")
-    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\t2097152\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")
-    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_N 4)
-  endif()
+if (NOT MSVC)
+  list(APPEND GETARCH_SRC ${PROJECT_SOURCE_DIR}/cpuid.S)
+endif ()
 
-  # Or should this actually be NUM_CORES?
-  if (${NUM_THREADS} GREATER 0)
-    file(APPEND ${TARGET_CONF_TEMP} "#define NUM_CORES\t${NUM_THREADS}\n")
-  endif()
+if (MSVC)
+#Use generic for MSVC now
+set(GETARCH_FLAGS ${GETARCH_FLAGS} -DFORCE_GENERIC)
+endif()
 
-  # GetArch_2nd
-  foreach(float_char S;D;Q;C;Z;X)
-    if (NOT DEFINED ${float_char}GEMM_UNROLL_M)
-      set(${float_char}GEMM_UNROLL_M 2)
-    endif()
-    if (NOT DEFINED ${float_char}GEMM_UNROLL_N)
-      set(${float_char}GEMM_UNROLL_N 2)
-    endif()
-  endforeach()
-  file(APPEND ${TARGET_CONF_TEMP}
-    "#define GEMM_MULTITHREAD_THRESHOLD\t${GEMM_MULTITHREAD_THRESHOLD}\n")
-  # Move to where gen_config_h would place it
-  file(RENAME ${TARGET_CONF_TEMP} "${TARGET_CONF_DIR}/${TARGET_CONF}")  
+if ("${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
+  # disable WindowsStore strict CRT checks
+  set(GETARCH_FLAGS ${GETARCH_FLAGS} -D_CRT_SECURE_NO_WARNINGS)
+endif ()
 
-else(NOT CMAKE_CROSSCOMPILING)
-  # compile getarch
-  set(GETARCH_SRC
-    ${PROJECT_SOURCE_DIR}/getarch.c
-    ${CPUIDEMU}
+set(GETARCH_DIR "${PROJECT_BINARY_DIR}/getarch_build")
+set(GETARCH_BIN "getarch${CMAKE_EXECUTABLE_SUFFIX}")
+file(MAKE_DIRECTORY ${GETARCH_DIR})
+if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
+  try_compile(GETARCH_RESULT ${GETARCH_DIR}
+    SOURCES ${GETARCH_SRC}
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I${PROJECT_SOURCE_DIR}
+    OUTPUT_VARIABLE GETARCH_LOG
+    COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH_BIN}
   )
 
-  if ("${CMAKE_C_COMPILER_ID}" STREQUAL "MSVC")
-    #Use generic for MSVC now
-    message("MSVC")
-    set(GETARCH_FLAGS ${GETARCH_FLAGS} -DFORCE_GENERIC)
-  else()
-    list(APPEND GETARCH_SRC ${PROJECT_SOURCE_DIR}/cpuid.S)
+  if (NOT ${GETARCH_RESULT})
+    MESSAGE(FATAL_ERROR "Compiling getarch failed ${GETARCH_LOG}")
   endif ()
+endif ()
+message(STATUS "Running getarch")
 
-  if ("${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
-    # disable WindowsStore strict CRT checks
-    set(GETARCH_FLAGS ${GETARCH_FLAGS} -D_CRT_SECURE_NO_WARNINGS)
+# use the cmake binary w/ the -E param to run a shell command in a cross-platform way
+execute_process(COMMAND ${PROJECT_BINARY_DIR}/${GETARCH_BIN} 0 OUTPUT_VARIABLE GETARCH_MAKE_OUT)
+execute_process(COMMAND ${PROJECT_BINARY_DIR}/${GETARCH_BIN} 1 OUTPUT_VARIABLE GETARCH_CONF_OUT)
+
+message(STATUS "GETARCH results:\n${GETARCH_MAKE_OUT}")
+
+# append config data from getarch to the TARGET file and read in CMake vars
+file(APPEND ${TARGET_CONF} ${GETARCH_CONF_OUT})
+ParseGetArchVars(${GETARCH_MAKE_OUT})
+
+set(GETARCH2_DIR "${PROJECT_BINARY_DIR}/getarch2_build")
+set(GETARCH2_BIN "getarch_2nd${CMAKE_EXECUTABLE_SUFFIX}")
+file(MAKE_DIRECTORY ${GETARCH2_DIR})
+if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
+  try_compile(GETARCH2_RESULT ${GETARCH2_DIR}
+    SOURCES ${PROJECT_SOURCE_DIR}/getarch_2nd.c
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I${PROJECT_SOURCE_DIR}
+    OUTPUT_VARIABLE GETARCH2_LOG
+    COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH2_BIN}
+  )
+
+  if (NOT ${GETARCH2_RESULT})
+    MESSAGE(FATAL_ERROR "Compiling getarch_2nd failed ${GETARCH2_LOG}")
   endif ()
+endif ()
 
-  set(GETARCH_DIR "${PROJECT_BINARY_DIR}/getarch_build")
-  set(GETARCH_BIN "getarch${CMAKE_EXECUTABLE_SUFFIX}")
-  file(MAKE_DIRECTORY ${GETARCH_DIR})
-  configure_file(${TARGET_CONF_TEMP} ${GETARCH_DIR}/${TARGET_CONF} COPYONLY)
-  if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
-    try_compile(GETARCH_RESULT ${GETARCH_DIR}
-      SOURCES ${GETARCH_SRC}
-    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I${GETARCH_DIR} -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
-      OUTPUT_VARIABLE GETARCH_LOG
-      COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH_BIN}
-    )
-  
-    if (NOT ${GETARCH_RESULT})
-      MESSAGE(FATAL_ERROR "Compiling getarch failed ${GETARCH_LOG}")
-    endif ()
+# use the cmake binary w/ the -E param to run a shell command in a cross-platform way
+execute_process(COMMAND ${PROJECT_BINARY_DIR}/${GETARCH2_BIN} 0 OUTPUT_VARIABLE GETARCH2_MAKE_OUT)
+execute_process(COMMAND ${PROJECT_BINARY_DIR}/${GETARCH2_BIN} 1 OUTPUT_VARIABLE GETARCH2_CONF_OUT)
+
+# append config data from getarch_2nd to the TARGET file and read in CMake vars
+file(APPEND ${TARGET_CONF} ${GETARCH2_CONF_OUT})
+ParseGetArchVars(${GETARCH2_MAKE_OUT})
+
+# compile get_config_h
+set(GEN_CONFIG_H_DIR "${PROJECT_BINARY_DIR}/genconfig_h_build")
+set(GEN_CONFIG_H_BIN "gen_config_h${CMAKE_EXECUTABLE_SUFFIX}")
+set(GEN_CONFIG_H_FLAGS "-DVERSION=\"${OpenBLAS_VERSION}\"")
+file(MAKE_DIRECTORY ${GEN_CONFIG_H_DIR})
+
+if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
+  try_compile(GEN_CONFIG_H_RESULT ${GEN_CONFIG_H_DIR}
+    SOURCES ${PROJECT_SOURCE_DIR}/gen_config_h.c
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GEN_CONFIG_H_FLAGS} -I${PROJECT_SOURCE_DIR}
+    OUTPUT_VARIABLE GEN_CONFIG_H_LOG
+    COPY_FILE ${PROJECT_BINARY_DIR}/${GEN_CONFIG_H_BIN}
+  )
+
+  if (NOT ${GEN_CONFIG_H_RESULT})
+    MESSAGE(FATAL_ERROR "Compiling gen_config_h failed ${GEN_CONFIG_H_LOG}")
   endif ()
-  message(STATUS "Running getarch")
-
-  # use the cmake binary w/ the -E param to run a shell command in a cross-platform way
-execute_process(COMMAND "${PROJECT_BINARY_DIR}/${GETARCH_BIN}" 0 OUTPUT_VARIABLE GETARCH_MAKE_OUT)
-execute_process(COMMAND "${PROJECT_BINARY_DIR}/${GETARCH_BIN}" 1 OUTPUT_VARIABLE GETARCH_CONF_OUT)
-
-  message(STATUS "GETARCH results:\n${GETARCH_MAKE_OUT}")
-
-  # append config data from getarch to the TARGET file and read in CMake vars
-  file(APPEND ${TARGET_CONF_TEMP} ${GETARCH_CONF_OUT})
-  ParseGetArchVars(${GETARCH_MAKE_OUT})
-
-  set(GETARCH2_DIR "${PROJECT_BINARY_DIR}/getarch2_build")
-  set(GETARCH2_BIN "getarch_2nd${CMAKE_EXECUTABLE_SUFFIX}")
-  file(MAKE_DIRECTORY ${GETARCH2_DIR})
-  configure_file(${TARGET_CONF_TEMP} ${GETARCH2_DIR}/${TARGET_CONF} COPYONLY)
-  if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
-    try_compile(GETARCH2_RESULT ${GETARCH2_DIR}
-      SOURCES ${PROJECT_SOURCE_DIR}/getarch_2nd.c
-    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I${GETARCH2_DIR} -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
-      OUTPUT_VARIABLE GETARCH2_LOG
-      COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH2_BIN}
-    )
-
-    if (NOT ${GETARCH2_RESULT})
-      MESSAGE(FATAL_ERROR "Compiling getarch_2nd failed ${GETARCH2_LOG}")
-    endif ()
-  endif ()
-
-  # use the cmake binary w/ the -E param to run a shell command in a cross-platform way
-execute_process(COMMAND "${PROJECT_BINARY_DIR}/${GETARCH2_BIN}" 0 OUTPUT_VARIABLE GETARCH2_MAKE_OUT)
-execute_process(COMMAND "${PROJECT_BINARY_DIR}/${GETARCH2_BIN}" 1 OUTPUT_VARIABLE GETARCH2_CONF_OUT)
-
-  # append config data from getarch_2nd to the TARGET file and read in CMake vars
-  file(APPEND ${TARGET_CONF_TEMP} ${GETARCH2_CONF_OUT})
-
-  configure_file(${TARGET_CONF_TEMP} ${TARGET_CONF_DIR}/${TARGET_CONF} COPYONLY)
-
-  ParseGetArchVars(${GETARCH2_MAKE_OUT})
-
-endif()
+endif ()

cmake/system.cmake

@@ -2,27 +2,13 @@
 ## Author: Hank Anderson <hank@statease.com>
 ## Description: Ported from OpenBLAS/Makefile.system
 ##
+
 set(NETLIB_LAPACK_DIR "${PROJECT_SOURCE_DIR}/lapack-netlib")
 
-# System detection, via CMake.
-include("${PROJECT_SOURCE_DIR}/cmake/system_check.cmake")
+# TODO: Makefile.system detects Darwin (mac) and switches to clang here -hpa
+# http://stackoverflow.com/questions/714100/os-detecting-makefile
 
-if(CMAKE_CROSSCOMPILING AND NOT DEFINED TARGET)
-  # Detect target without running getarch
-  if (ARM64)
-    set(TARGET "ARMV8")
-  elseif(ARM)
-    set(TARGET "ARMV7") # TODO: Ask compiler which arch this is
-  else()
-    message(FATAL_ERROR "When cross compiling, a TARGET is required.")
-  endif()
-endif()
-
-# Other files expect CORE, which is actually TARGET and will become TARGET_CORE for kernel build. Confused yet?
-# It seems we are meant to use TARGET as input and CORE internally as kernel.
-if(NOT DEFINED CORE AND DEFINED TARGET)
-  set(CORE ${TARGET})
-endif()
+# TODO: Makefile.system sets HOSTCC = $(CC) here if not already set -hpa
 
 # TARGET_CORE will override TARGET which is used in DYNAMIC_ARCH=1.
 if (DEFINED TARGET_CORE)
@@ -33,7 +19,7 @@ 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" OR ${TARGET} STREQUAL "SKYLAKEX")
+  if (${TARGET} STREQUAL "HASWELL" OR ${TARGET} STREQUAL "SANDYBRIDGE")
     set(TARGET "NEHALEM")
   endif ()
   if (${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "PILEDRIVER" OR ${TARGET} STREQUAL "ZEN")
@@ -42,7 +28,7 @@ if (DEFINED BINARY AND DEFINED TARGET AND BINARY EQUAL 32)
 endif ()
 
 if (DEFINED TARGET)
-  message(STATUS "Targeting the ${TARGET} architecture.")
+  message(STATUS "Targetting the ${TARGET} architecture.")
   set(GETARCH_FLAGS "-DFORCE_${TARGET}")
 endif ()
 
@@ -67,16 +53,21 @@ if (NO_AVX2)
   set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX2")
 endif ()
 
-if (CMAKE_BUILD_TYPE STREQUAL "Debug")
-  set(GETARCH_FLAGS "${GETARCH_FLAGS} ${CMAKE_C_FLAGS_DEBUG}")
+if (CMAKE_BUILD_TYPE STREQUAL Debug)
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -g")
 endif ()
 
+# TODO: let CMake handle this? -hpa
+#if (${QUIET_MAKE})
+#  set(MAKE "${MAKE} -s")
+#endif()
+
 if (NOT DEFINED NO_PARALLEL_MAKE)
   set(NO_PARALLEL_MAKE 0)
 endif ()
 set(GETARCH_FLAGS	"${GETARCH_FLAGS} -DNO_PARALLEL_MAKE=${NO_PARALLEL_MAKE}")
 
-if (CMAKE_C_COMPILER STREQUAL loongcc)
+if (CMAKE_CXX_COMPILER STREQUAL loongcc)
   set(GETARCH_FLAGS	"${GETARCH_FLAGS} -static")
 endif ()
 
@@ -87,44 +78,51 @@ else ()
   set(ONLY_CBLAS 0)
 endif ()
 
-# N.B. this is NUM_THREAD in Makefile.system which is probably a bug -hpa
-if (NOT CMAKE_CROSSCOMPILING)
-  if (NOT DEFINED NUM_CORES)
-    include(ProcessorCount)
-    ProcessorCount(NUM_CORES)
-  endif()
-
-endif()
-
-if (NOT DEFINED NUM_PARALLEL)
-  set(NUM_PARALLEL 1)
-endif()
+include("${PROJECT_SOURCE_DIR}/cmake/prebuild.cmake")
 
 if (NOT DEFINED NUM_THREADS)
-  if (DEFINED NUM_CORES AND NOT NUM_CORES EQUAL 0)
-    # HT?
-    set(NUM_THREADS ${NUM_CORES})
-  else ()
-    set(NUM_THREADS 0)
-  endif ()
-endif()
+  set(NUM_THREADS ${NUM_CORES})
+endif ()
 
-if (${NUM_THREADS} LESS 2)
+if (${NUM_THREADS} EQUAL 1)
   set(USE_THREAD 0)
-elseif(NOT DEFINED USE_THREAD)
-  set(USE_THREAD 1)
 endif ()
 
-if (USE_THREAD)
-  message(STATUS "Multi-threading enabled with ${NUM_THREADS} threads.")
+if (DEFINED USE_THREAD)
+  if (NOT ${USE_THREAD})
+    unset(SMP)
+  else ()
+    set(SMP 1)
+  endif ()
+else ()
+  # N.B. this is NUM_THREAD in Makefile.system which is probably a bug -hpa
+  if (${NUM_THREADS} EQUAL 1)
+    unset(SMP)
+  else ()
+    set(SMP 1)
+  endif ()
 endif ()
 
-include("${PROJECT_SOURCE_DIR}/cmake/prebuild.cmake")
+if (${SMP})
+  message(STATUS "SMP enabled.")
+endif ()
 
 if (NOT DEFINED NEED_PIC)
   set(NEED_PIC 1)
 endif ()
 
+# TODO: I think CMake should be handling all this stuff -hpa
+unset(ARFLAGS)
+set(CPP "${COMPILER} -E")
+set(AR "${CROSS_SUFFIX}ar")
+set(AS "${CROSS_SUFFIX}as")
+set(LD "${CROSS_SUFFIX}ld")
+set(RANLIB "${CROSS_SUFFIX}ranlib")
+set(NM "${CROSS_SUFFIX}nm")
+set(DLLWRAP "${CROSS_SUFFIX}dllwrap")
+set(OBJCOPY "${CROSS_SUFFIX}objcopy")
+set(OBJCONV "${CROSS_SUFFIX}objconv")
+
 # OS dependent settings
 include("${PROJECT_SOURCE_DIR}/cmake/os.cmake")
 
@@ -152,20 +150,15 @@ if (NEED_PIC)
     set(CCOMMON_OPT "${CCOMMON_OPT} -fPIC")
   endif ()
 
-  if (NOT NOFORTRAN)
-    if (${F_COMPILER} STREQUAL "SUN")
-      set(FCOMMON_OPT "${FCOMMON_OPT} -pic")
-    else ()
-      set(FCOMMON_OPT "${FCOMMON_OPT} -fPIC")
-    endif ()
-  endif()
+  if (${F_COMPILER} STREQUAL "SUN")
+    set(FCOMMON_OPT "${FCOMMON_OPT} -pic")
+  else ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -fPIC")
+  endif ()
 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)
@@ -182,7 +175,7 @@ if (NO_AVX)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX")
 endif ()
 
-if (X86)
+if (${ARCH} STREQUAL "x86")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX")
 endif ()
 
@@ -190,20 +183,25 @@ if (NO_AVX2)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX2")
 endif ()
 
-if (USE_THREAD)
-  # USE_SIMPLE_THREADED_LEVEL3 = 1
-  # NO_AFFINITY = 1
+if (SMP)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DSMP_SERVER")
 
-  if (MIPS64)
+  if (${ARCH} STREQUAL "mips64")
     if (NOT ${CORE} STREQUAL "LOONGSON3B")
       set(USE_SIMPLE_THREADED_LEVEL3 1)
     endif ()
   endif ()
 
+  if (USE_OPENMP)
+    # USE_SIMPLE_THREADED_LEVEL3 = 1
+    # NO_AFFINITY = 1
+    set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_OPENMP")
+  endif ()
+
   if (BIGNUMA)
     set(CCOMMON_OPT "${CCOMMON_OPT} -DBIGNUMA")
   endif ()
+
 endif ()
 
 if (NO_WARMUP)
@@ -231,8 +229,6 @@ 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 ()
@@ -268,7 +264,7 @@ if (NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
   set(NO_AFFINITY 1)
 endif ()
 
-if (NOT X86_64 AND NOT X86 AND NOT ${CORE} STREQUAL "LOONGSON3B")
+if (NOT ${ARCH} STREQUAL "x86_64" AND NOT ${ARCH} STREQUAL "x86" AND NOT ${CORE} STREQUAL "LOONGSON3B")
   set(NO_AFFINITY 1)
 endif ()
 
@@ -300,19 +296,52 @@ if (MIXED_MEMORY_ALLOCATION)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DMIXED_MEMORY_ALLOCATION")
 endif ()
 
+if (${CMAKE_SYSTEM_NAME} STREQUAL "SunOS")
+  set(TAR	gtar)
+  set(PATCH	gpatch)
+  set(GREP ggrep)
+else ()
+  set(TAR tar)
+  set(PATCH patch)
+  set(GREP grep)
+endif ()
+
+if (NOT DEFINED MD5SUM)
+  set(MD5SUM md5sum)
+endif ()
+
+set(AWK awk)
+
+set(SED sed)
+
 set(REVISION "-r${OpenBLAS_VERSION}")
 set(MAJOR_VERSION ${OpenBLAS_MAJOR_VERSION})
 
-set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${CCOMMON_OPT}")
+if (DEBUG)
+  set(COMMON_OPT "${COMMON_OPT} -g")
+endif ()
+
+if (NOT DEFINED COMMON_OPT)
+  set(COMMON_OPT "-O2")
+endif ()
+
+#For x86 32-bit
+if (DEFINED BINARY AND BINARY EQUAL 32)
+if (NOT MSVC)
+  set(COMMON_OPT "${COMMON_OPT} -m32")
+endif()
+endif()
+
+set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${COMMON_OPT} ${CCOMMON_OPT}")
 if(NOT MSVC)
-set(CMAKE_ASM_FLAGS "${CMAKE_ASM_FLAGS} ${CCOMMON_OPT}")
+set(CMAKE_ASM_FLAGS "${CMAKE_ASM_FLAGS} ${COMMON_OPT} ${CCOMMON_OPT}")
 endif()
 # TODO: not sure what PFLAGS is -hpa
-set(PFLAGS "${PFLAGS} ${CCOMMON_OPT} -I${TOPDIR} -DPROFILE ${COMMON_PROF}")
+set(PFLAGS "${PFLAGS} ${COMMON_OPT} ${CCOMMON_OPT} -I${TOPDIR} -DPROFILE ${COMMON_PROF}")
 
-set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} ${FCOMMON_OPT}")
+set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} ${COMMON_OPT} ${FCOMMON_OPT}")
 # TODO: not sure what FPFLAGS is -hpa
-set(FPFLAGS "${FPFLAGS} ${FCOMMON_OPT} ${COMMON_PROF}")
+set(FPFLAGS "${FPFLAGS} ${COMMON_OPT} ${FCOMMON_OPT} ${COMMON_PROF}")
 
 #For LAPACK Fortran codes.
 set(LAPACK_FFLAGS "${LAPACK_FFLAGS} ${CMAKE_Fortran_FLAGS}")
@@ -320,7 +349,7 @@ set(LAPACK_FPFLAGS "${LAPACK_FPFLAGS} ${FPFLAGS}")
 
 #Disable -fopenmp for LAPACK Fortran codes on Windows.
 if (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
-  set(FILTER_FLAGS "-fopenmp;-mp;-openmp;-xopenmp=parallel")
+  set(FILTER_FLAGS "-fopenmp;-mp;-openmp;-xopenmp=parralel")
   foreach (FILTER_FLAG ${FILTER_FLAGS})
     string(REPLACE ${FILTER_FLAG} "" LAPACK_FFLAGS ${LAPACK_FFLAGS})
     string(REPLACE ${FILTER_FLAG} "" LAPACK_FPFLAGS ${LAPACK_FPFLAGS})
@@ -358,7 +387,7 @@ if (NOT DEFINED LIBSUFFIX)
 endif ()
 
 if (DYNAMIC_ARCH)
-  if (USE_THREAD)
+  if (DEFINED SMP)
     set(LIBNAME "${LIBPREFIX}p${REVISION}.${LIBSUFFIX}")
     set(LIBNAME_P	"${LIBPREFIX}p${REVISION}_p.${LIBSUFFIX}")
   else ()
@@ -366,7 +395,7 @@ if (DYNAMIC_ARCH)
     set(LIBNAME_P	"${LIBPREFIX}${REVISION}_p.${LIBSUFFIX}")
   endif ()
 else ()
-  if (USE_THREAD)
+  if (DEFINED SMP)
     set(LIBNAME "${LIBPREFIX}_${LIBCORE}p${REVISION}.${LIBSUFFIX}")
     set(LIBNAME_P	"${LIBPREFIX}_${LIBCORE}p${REVISION}_p.${LIBSUFFIX}")
   else ()
@@ -397,9 +426,6 @@ if (NOT NO_LAPACK)
   if (NOT NO_LAPACKE)
     set(LIB_COMPONENTS "${LIB_COMPONENTS} LAPACKE")
   endif ()
-  if (BUILD_RELAPACK)
-    set(LIB_COMPONENTS "${LIB_COMPONENTS} ReLAPACK")
-  endif ()
 endif ()
 
 if (ONLY_CBLAS)
@@ -411,7 +437,7 @@ endif ()
 set(USE_GEMM3M 0)
 
 if (DEFINED ARCH)
-  if (X86 OR X86_64 OR ${ARCH} STREQUAL "ia64" OR MIPS64)
+  if (${ARCH} STREQUAL "x86" OR ${ARCH} STREQUAL "x86_64" OR ${ARCH} STREQUAL "ia64" OR ${ARCH} STREQUAL "MIPS")
     set(USE_GEMM3M 1)
   endif ()
 
@@ -494,3 +520,35 @@ endif ()
 #  export CUFLAGS
 #  export CULIB
 #endif
+
+#.SUFFIXES: .$(PSUFFIX) .$(SUFFIX) .f
+#
+#.f.$(SUFFIX):
+#	$(FC) $(FFLAGS) -c $<  -o $(@F)
+#
+#.f.$(PSUFFIX):
+#	$(FC) $(FPFLAGS) -pg -c $<  -o $(@F)
+
+# these are not cross-platform
+#ifdef BINARY64
+#PATHSCALEPATH	= /opt/pathscale/lib/3.1
+#PGIPATH		= /opt/pgi/linux86-64/7.1-5/lib
+#else
+#PATHSCALEPATH	= /opt/pathscale/lib/3.1/32
+#PGIPATH		= /opt/pgi/linux86/7.1-5/lib
+#endif
+
+#ACMLPATH	= /opt/acml/4.3.0
+#ifneq ($(OSNAME), Darwin)
+#MKLPATH         = /opt/intel/mkl/10.2.2.025/lib
+#else
+#MKLPATH         = /Library/Frameworks/Intel_MKL.framework/Versions/10.0.1.014/lib
+#endif
+#ATLASPATH	= /opt/atlas/3.9.17/opteron
+#FLAMEPATH	= $(HOME)/flame/lib
+#ifneq ($(OSNAME), SunOS)
+#SUNPATH		= /opt/sunstudio12.1
+#else
+#SUNPATH		= /opt/SUNWspro
+#endif
+

cmake/system_check.cmake

@@ -1,77 +0,0 @@
-##
-## Author: Hank Anderson <hank@statease.com>
-## Description: Ported from the OpenBLAS/c_check perl script.
-##              This is triggered by prebuild.cmake and runs before any of the code is built.
-##              Creates config.h and Makefile.conf.
-
-# Convert CMake vars into the format that OpenBLAS expects
-string(TOUPPER ${CMAKE_SYSTEM_NAME} HOST_OS)
-if (${HOST_OS} STREQUAL "WINDOWS")
-  set(HOST_OS WINNT)
-endif ()
-
-if(CMAKE_COMPILER_IS_GNUCC AND WIN32)
-    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpmachine
-              OUTPUT_VARIABLE OPENBLAS_GCC_TARGET_MACHINE
-              OUTPUT_STRIP_TRAILING_WHITESPACE)
-    if(OPENBLAS_GCC_TARGET_MACHINE MATCHES "amd64|x86_64|AMD64")
-      set(MINGW64 1)
-    endif()
-endif()
-
-# Pretty thorough determination of arch. Add more if needed
-if(CMAKE_CL_64 OR MINGW64)
-  set(X86_64 1)
-elseif(MINGW OR (MSVC AND NOT CMAKE_CROSSCOMPILING))
-  set(X86 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc.*|power.*|Power.*")
-  set(PPC 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "mips64.*")
-  set(MIPS64 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
-  set(X86_64 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "i686.*|i386.*|x86.*|amd64.*|AMD64.*")
-  set(X86 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm.*|ARM.*)")
-  set(ARM 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64.*|AARCH64.*)")
-  set(ARM64 1)
-endif()
-
-if (X86_64)
-  set(ARCH "x86_64")
-elseif(X86)
-  set(ARCH "x86")
-elseif(PPC)
-  set(ARCH "power")
-elseif(ARM)
-  set(ARCH "arm")
-elseif(ARM64)
-  set(ARCH "arm64")
-else()
-  set(ARCH ${CMAKE_SYSTEM_PROCESSOR} CACHE STRING "Target Architecture")
-endif ()
-
-if (NOT BINARY)
-  if (X86_64 OR ARM64 OR PPC OR MIPS64)
-    set(BINARY 64)
-  else ()
-    set(BINARY 32)
-  endif ()
-endif()
-
-if(BINARY EQUAL 64)
-  set(BINARY64 1)
-else()
-  set(BINARY32 1)
-endif()
-
-if (X86_64 OR X86)
-  file(WRITE ${PROJECT_BINARY_DIR}/avx512.tmp "int 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()
-

cmake/utils.cmake

@@ -202,8 +202,6 @@ function(GenerateNamedObjects sources_in)
       if (use_cblas)
         set(obj_name "cblas_${obj_name}")
         list(APPEND obj_defines "CBLAS")
-      elseif (NOT "${obj_name}" MATCHES "${ARCH_SUFFIX}")
-        set(obj_name "${obj_name}${ARCH_SUFFIX}")
       endif ()
 
       list(APPEND obj_defines "ASMNAME=${FU}${obj_name};ASMFNAME=${FU}${obj_name}${BU};NAME=${obj_name}${BU};CNAME=${obj_name};CHAR_NAME=\"${obj_name}${BU}\";CHAR_CNAME=\"${obj_name}\"")
@@ -236,9 +234,7 @@ function(GenerateNamedObjects sources_in)
 
       string(REPLACE ";" "\n#define " define_source "${obj_defines}")
       string(REPLACE "=" " " define_source "${define_source}")
-      file(WRITE ${new_source_file}.tmp "#define ${define_source}\n#include \"${old_source_file}\"")
-      configure_file(${new_source_file}.tmp ${new_source_file} COPYONLY)
-      file(REMOVE ${new_source_file}.tmp)
+      file(WRITE ${new_source_file} "#define ${define_source}\n#include \"${old_source_file}\"")
       list(APPEND SRC_LIST_OUT ${new_source_file})
 
     endforeach ()

common.h

@@ -93,7 +93,7 @@ extern "C" {
 #include <sched.h>
 #endif
 
-#if defined(OS_DARWIN) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(OS_ANDROID)
+#if defined(OS_DARWIN) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_ANDROID)
 #include <sched.h>
 #endif
 
@@ -179,7 +179,7 @@ extern "C" {
 
 #define ALLOCA_ALIGN 63UL
 
-#define NUM_BUFFERS (MAX_CPU_NUMBER * 2 * MAX_PARALLEL_NUMBER)
+#define NUM_BUFFERS (MAX_CPU_NUMBER * 2)
 
 #ifdef NEEDBUNDERSCORE
 #define BLASFUNC(FUNC) FUNC##_
@@ -495,33 +495,6 @@ static void __inline blas_lock(volatile BLASULONG *address){
 #define MMAP_POLICY (MAP_PRIVATE | MAP_ANONYMOUS)
 #endif
 
-#ifndef ASSEMBLER
-/* C99 supports complex floating numbers natively, which GCC also offers as an
-   extension since version 3.0.  If neither are available, use a compatible
-   structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
-#if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \
-      (__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT))) && !defined(_MSC_VER)
-  #define OPENBLAS_COMPLEX_C99
-  #ifndef __cplusplus
-    #include <complex.h>
-  #endif
-  typedef float _Complex openblas_complex_float;
-  typedef double _Complex openblas_complex_double;
-  typedef xdouble _Complex openblas_complex_xdouble;
-  #define openblas_make_complex_float(real, imag)    ((real) + ((imag) * _Complex_I))
-  #define openblas_make_complex_double(real, imag)   ((real) + ((imag) * _Complex_I))
-  #define openblas_make_complex_xdouble(real, imag)  ((real) + ((imag) * _Complex_I))
-#else
-  #define OPENBLAS_COMPLEX_STRUCT
-  typedef struct { float real, imag; } openblas_complex_float;
-  typedef struct { double real, imag; } openblas_complex_double;
-  typedef struct { xdouble real, imag; } openblas_complex_xdouble;
-  #define openblas_make_complex_float(real, imag)    {(real), (imag)}
-  #define openblas_make_complex_double(real, imag)   {(real), (imag)}
-  #define openblas_make_complex_xdouble(real, imag)  {(real), (imag)}
-#endif
-#endif
-
 #include "param.h"
 #include "common_param.h"
 
@@ -551,6 +524,31 @@ static void __inline blas_lock(volatile BLASULONG *address){
 #include <stdio.h>
 #endif  // NOINCLUDE
 
+/* C99 supports complex floating numbers natively, which GCC also offers as an
+   extension since version 3.0.  If neither are available, use a compatible
+   structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
+#if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \
+      (__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT)))
+  #define OPENBLAS_COMPLEX_C99
+  #ifndef __cplusplus
+    #include <complex.h>
+  #endif
+  typedef float _Complex openblas_complex_float;
+  typedef double _Complex openblas_complex_double;
+  typedef xdouble _Complex openblas_complex_xdouble;
+  #define openblas_make_complex_float(real, imag)    ((real) + ((imag) * _Complex_I))
+  #define openblas_make_complex_double(real, imag)   ((real) + ((imag) * _Complex_I))
+  #define openblas_make_complex_xdouble(real, imag)  ((real) + ((imag) * _Complex_I))
+#else
+  #define OPENBLAS_COMPLEX_STRUCT
+  typedef struct { float real, imag; } openblas_complex_float;
+  typedef struct { double real, imag; } openblas_complex_double;
+  typedef struct { xdouble real, imag; } openblas_complex_xdouble;
+  #define openblas_make_complex_float(real, imag)    {(real), (imag)}
+  #define openblas_make_complex_double(real, imag)   {(real), (imag)}
+  #define openblas_make_complex_xdouble(real, imag)  {(real), (imag)}
+#endif
+
 #ifdef XDOUBLE
 #define OPENBLAS_COMPLEX_FLOAT openblas_complex_xdouble
 #define OPENBLAS_MAKE_COMPLEX_FLOAT(r,i) openblas_make_complex_xdouble(r,i)
@@ -642,7 +640,6 @@ 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);
@@ -650,21 +647,6 @@ 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_param.h

@@ -333,8 +333,8 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
   float (*cnrm2_k) (BLASLONG, float *, BLASLONG);
   float (*casum_k) (BLASLONG, float *, BLASLONG);
   int    (*ccopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  openblas_complex_float (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
-  openblas_complex_float (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  float _Complex (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  float _Complex (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   int    (*csrot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
 
   int    (*caxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
@@ -496,8 +496,8 @@ BLASLONG (*izamin_k)(BLASLONG, double *, BLASLONG);
   double (*znrm2_k) (BLASLONG, double *, BLASLONG);
   double (*zasum_k) (BLASLONG, double *, BLASLONG);
   int    (*zcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  openblas_complex_double (*zdotu_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-  openblas_complex_double (*zdotc_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
+  double _Complex (*zdotu_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
+  double _Complex (*zdotc_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
   int    (*zdrot_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);
 
   int    (*zaxpy_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
@@ -661,8 +661,8 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   xdouble (*xnrm2_k) (BLASLONG, xdouble *, BLASLONG);
   xdouble (*xasum_k) (BLASLONG, xdouble *, BLASLONG);
   int    (*xcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  openblas_complex_xdouble (*xdotu_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-  openblas_complex_xdouble (*xdotc_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
+  xdouble _Complex (*xdotu_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
+  xdouble _Complex (*xdotc_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
   int    (*xqrot_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);
 
   int    (*xaxpy_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
@@ -888,7 +888,7 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   int    (*sgeadd_k) (BLASLONG, BLASLONG, float, float *, BLASLONG, float, float *, BLASLONG); 
   int    (*dgeadd_k) (BLASLONG, BLASLONG, double, double *, BLASLONG, double, double *, BLASLONG); 
   int    (*cgeadd_k) (BLASLONG, BLASLONG, float, float,  float *,  BLASLONG, float, float, float *, BLASLONG); 
-  int    (*zgeadd_k) (BLASLONG, BLASLONG, double, double, double *, BLASLONG, double, double, double *, BLASLONG); 
+  int    (*zgeadd_k) (BLASLONG, BLASLONG, float, double, double *, BLASLONG, double, double, double *, BLASLONG); 
 
 } gotoblas_t;
 

common_stackalloc.h

@@ -47,15 +47,14 @@ 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)                                        \
-  /* make it volatile because some function (ex: dgemv_n.S) */                 \
-  /* do not restore all register */                                            \
-  volatile int stack_alloc_size = SIZE;                                        \
-  if (stack_alloc_size > MAX_STACK_ALLOC / sizeof(TYPE)) stack_alloc_size = 0; \
-  STACK_ALLOC_PROTECT_SET                                                      \
-  /* Avoid declaring an array of length 0 */                                   \
-  TYPE stack_buffer[stack_alloc_size ? stack_alloc_size : 1]                   \
-      __attribute__((aligned(0x20)));                                          \
+#define STACK_ALLOC(SIZE, TYPE, BUFFER)                                    \
+  /* make it volatile because some function (ex: dgemv_n.S) */             \
+  /* do not restore all register */                                        \
+  volatile int stack_alloc_size = SIZE;                                    \
+  if(stack_alloc_size > MAX_STACK_ALLOC / sizeof(TYPE))                    \
+    stack_alloc_size = 0;                                                  \
+  STACK_ALLOC_PROTECT_SET                                                  \
+  TYPE stack_buffer[stack_alloc_size] __attribute__((aligned(0x20)));    \
   BUFFER = stack_alloc_size ? stack_buffer : (TYPE *)blas_memory_alloc(1);
 #else
   //Original OpenBLAS/GotoBLAS codes.

common_x86.h

@@ -178,13 +178,7 @@ 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];
 
   __asm__ __volatile__  ("mull %0" :"=d" (result) :"a"(x), "0" (y));
@@ -333,7 +327,7 @@ REALNAME:
 #endif
 #endif
 
-#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_OPENBSD) || defined(__ELF__)
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__)
 #define PROLOGUE \
 	.text; \
 	.align 16; \

common_x86_64.h

@@ -60,13 +60,8 @@
 #endif
 */
 
-#ifdef __GNUC__
-#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
+#define MB
+#define WMB
 
 static void __inline blas_lock(volatile BLASULONG *address){
 
@@ -201,13 +196,6 @@ 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));
@@ -415,7 +403,7 @@ REALNAME:
 #define EPILOGUE .end
 #endif
 
-#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(__ELF__) || defined(C_PGI)
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__) || defined(C_PGI)
 #define PROLOGUE \
 	.text; \
 	.align 512; \

cpuid.h

@@ -115,7 +115,6 @@
 #define CORE_STEAMROLLER 25
 #define CORE_EXCAVATOR   26
 #define CORE_ZEN         27
-#define CORE_SKYLAKEX    28
 
 #define HAVE_SSE      (1 <<  0)
 #define HAVE_SSE2     (1 <<  1)
@@ -138,7 +137,6 @@
 #define HAVE_AVX      (1 <<  18)
 #define HAVE_FMA4     (1 <<  19)
 #define HAVE_FMA3     (1 <<  20)
-#define HAVE_AVX512VL (1 <<  21)
 
 #define CACHE_INFO_L1_I     1
 #define CACHE_INFO_L1_D     2
@@ -213,6 +211,5 @@ typedef struct {
 #define CPUTYPE_STEAMROLLER 		49
 #define CPUTYPE_EXCAVATOR 		50
 #define CPUTYPE_ZEN 			51
-#define CPUTYPE_SKYLAKEX		52
 
 #endif

cpuid_arm64.c

@@ -114,14 +114,13 @@ int detect(void)
 
 	fclose(infile);
 	if(cpu_part != NULL && cpu_implementer != NULL) {
-		if (strstr(cpu_implementer, "0x41") && 
-		(strstr(cpu_part, "0xd07") || strstr(cpu_part,"0xd08") || strstr(cpu_part,"0xd03") ))
-			return CPU_CORTEXA57; //or compatible A53, A72
+		if (strstr(cpu_part, "0xd07") && strstr(cpu_implementer, "0x41"))
+			return CPU_CORTEXA57;
 		else if (strstr(cpu_part, "0x516") && strstr(cpu_implementer, "0x42"))
 			return CPU_VULCAN;
 		else if (strstr(cpu_part, "0x0a1") && strstr(cpu_implementer, "0x43"))
 			return CPU_THUNDERX;
-		else if (strstr(cpu_part, "0x0af") && strstr(cpu_implementer, "0x43"))
+		else if (strstr(cpu_part, "0xFFF") && strstr(cpu_implementer, "0x43")) /* TODO */
 			return CPU_THUNDERX2T99;
 	}
 
@@ -143,7 +142,7 @@ int detect(void)
   	if(p != NULL)
 	{
 
-		if ((strstr(p, "AArch64")) || (strstr(p, "8")))
+		if (strstr(p, "AArch64"))
 		{
 			return CPU_ARMV8;
 

cpuid_mips.c

@@ -72,12 +72,10 @@ 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",
-  "P5600",
-  "1004K"
+  "P5600"
 };
 
 int detect(void){
@@ -92,7 +90,7 @@ 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;
       }
@@ -101,13 +99,43 @@ int detect(void){
   fclose(infile);
 
   if(p != NULL){
-  if (strstr(p, "5600")) {
-    return CPU_P5600;
-  } else if (strstr(p, "1004K")) {
-    return CPU_1004K;
-  } else  
+  if (strstr(p, "Loongson-3A")){
+    return CPU_LOONGSON3A;
+  }else if(strstr(p, "Loongson-3B")){
+    return CPU_LOONGSON3B;
+  }else if (strstr(p, "Loongson-3")){
+    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;
 }
@@ -121,7 +149,7 @@ void get_architecture(void){
 }
 
 void get_subarchitecture(void){
-  if(detect()==CPU_P5600|| detect()==CPU_1004K){
+  if(detect()==CPU_P5600){
     printf("P5600");
   }else{
     printf("UNKNOWN");
@@ -142,14 +170,6 @@ 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");
   }
@@ -158,8 +178,6 @@ 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

@@ -76,7 +76,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CPU_LOONGSON3B  3
 #define CPU_I6400       4
 #define CPU_P6600       5
-#define CPU_I6500       6
 
 static char *cpuname[] = {
   "UNKOWN",
@@ -84,8 +83,7 @@ static char *cpuname[] = {
   "LOONGSON3A",
   "LOONGSON3B",
   "I6400",
-  "P6600",
-  "I6500"
+  "P6600"
 };
 
 int detect(void){
@@ -167,8 +165,6 @@ void get_subarchitecture(void){
     printf("I6400");
   }else if(detect()==CPU_P6600){
     printf("P6600");
-  }else if(detect()==CPU_I6500){
-    printf("I6500");
   }else{
     printf("SICORTEX");
   }
@@ -215,15 +211,6 @@ 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_I6500){
-    printf("#define I6500\n");
-    printf("#define L1_DATA_SIZE 65536\n");
-    printf("#define L1_DATA_LINESIZE 32\n");
-    printf("#define L2_SIZE 1048576\n");
-    printf("#define L2_LINESIZE 32\n");
-    printf("#define DTB_DEFAULT_ENTRIES 64\n");
-    printf("#define DTB_SIZE 4096\n");
-    printf("#define L2_ASSOCIATIVE 8\n");
   }else{
     printf("#define SICORTEX\n");
     printf("#define L1_DATA_SIZE 32768\n");
@@ -245,8 +232,6 @@ void get_libname(void){
     printf("i6400\n");
   }else if(detect()==CPU_P6600) {
     printf("p6600\n");
-  }else if(detect()==CPU_I6500) {
-    printf("i6500\n");
   }else{
     printf("mips64\n");
   }

cpuid_power.c

@@ -142,52 +142,6 @@ int detect(void){
 
   return  CPUTYPE_PPC970;
 #endif
-
-#if defined(__FreeBSD__) || defined(__OpenBSD__) || defined(__DragonFly__)
-int id;
-id = __asm __volatile("mfpvr %0" : "=r"(id));
-switch ( id >> 16 ) {
-  case 0x4e: // POWER9
-    return  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

@@ -49,7 +49,6 @@ void get_subdirname(void){
 }
 
 void get_cpuconfig(void){
-  printf("#define SPARC\n");
   printf("#define V9\n");
   printf("#define DTB_DEFAULT_ENTRIES 32\n");
 }
@@ -57,8 +56,3 @@ void get_cpuconfig(void){
 void get_libname(void){
   printf("v9\n");
 }
-
-char *get_corename(void){
-  return "sparc";
-}
-

cpuid_x86.c

@@ -50,8 +50,6 @@
 #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
@@ -73,23 +71,12 @@ void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx)
   *edx = cpuInfo[3];
 }
 
-void cpuid_count(int op, int count, int *eax, int *ebx, int *ecx, int *edx)
-{
-  int cpuInfo[4] = {-1};
-  __cpuidex(cpuInfo, op, count);
-  *eax = cpuInfo[0];
-  *ebx = cpuInfo[1];
-  *ecx = cpuInfo[2];
-  *edx = cpuInfo[3];
-}
-
 #else
 
 #ifndef CPUIDEMU
 
 #if defined(__APPLE__) && defined(__i386__)
 void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx);
-void cpuid_count(int op, int count, int *eax, int *ebx, int *ecx, int *edx);
 #else
 static C_INLINE void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
 #if defined(__i386__) && defined(__PIC__)
@@ -103,19 +90,6 @@ static C_INLINE void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
     ("cpuid": "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx) : "a" (op) : "cc");
 #endif
 }
-
-static C_INLINE void cpuid_count(int op, int count ,int *eax, int *ebx, int *ecx, int *edx){
-#if defined(__i386__) && defined(__PIC__)
-  __asm__ __volatile__
-    ("mov %%ebx, %%edi;"
-     "cpuid;"
-     "xchgl %%ebx, %%edi;"
-     : "=a" (*eax), "=D" (*ebx), "=c" (*ecx), "=d" (*edx) : "0" (op), "2" (count) : "cc");
-#else
-  __asm__ __volatile__
-    ("cpuid": "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx) : "0" (op), "2" (count) : "cc");
-#endif
-}
 #endif
 
 #else
@@ -159,10 +133,6 @@ void cpuid(unsigned int op, unsigned int *eax, unsigned int *ebx, unsigned int *
   *edx = idlist[current].d;
 }
 
-void cpuid_count (unsigned int op, unsigned int count, unsigned int *eax, unsigned int *ebx, unsigned int *ecx, unsigned int *edx) {
-  return cpuid (op, eax, ebx, ecx, edx);
-}
-
 #endif
 
 #endif // _MSC_VER
@@ -342,9 +312,9 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
   cpuid(0, &cpuid_level, &ebx, &ecx, &edx);
 
   if (cpuid_level > 1) {
-    int numcalls =0 ;
+
     cpuid(2, &eax, &ebx, &ecx, &edx);
-    numcalls = BITMASK(eax, 0, 0xff); //FIXME some systems may require repeated calls to read all entries
+
     info[ 0] = BITMASK(eax,  8, 0xff);
     info[ 1] = BITMASK(eax, 16, 0xff);
     info[ 2] = BITMASK(eax, 24, 0xff);
@@ -365,6 +335,7 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
     info[14] = BITMASK(edx, 24, 0xff);
 
     for (i = 0; i < 15; i++){
+
       switch (info[i]){
 
 	/* This table is from http://www.sandpile.org/ia32/cpuid.htm */
@@ -666,13 +637,12 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
 	LD1.linesize    = 64;
 	break;
       case 0x63 :
-  	DTB.size        = 2048;
-  	DTB.associative = 4;
-  	DTB.linesize    = 32;
-  	LDTB.size       = 4096;
-  	LDTB.associative= 4;
-  	LDTB.linesize   = 32;
-	break;
+  DTB.size        = 2048;
+  DTB.associative = 4;
+  DTB.linesize    = 32;
+  LDTB.size       = 4096;
+  LDTB.associative= 4;
+  LDTB.linesize   = 32;
       case 0x66 :
 	LD1.size        = 8;
 	LD1.associative = 4;
@@ -705,13 +675,12 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
 	LC1.associative = 8;
 	break;
       case 0x76 :
-  	ITB.size        = 2048;
-  	ITB.associative = 0;
-  	ITB.linesize    = 8;
-  	LITB.size       = 4096;
-  	LITB.associative= 0;
-  	LITB.linesize   = 8;
-	break;
+  ITB.size        = 2048;
+  ITB.associative = 0;
+  ITB.linesize    = 8;
+  LITB.size       = 4096;
+  LITB.associative= 0;
+  LITB.linesize   = 8;
       case 0x77 :
 	LC1.size        = 16;
 	LC1.associative = 4;
@@ -922,67 +891,6 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
   }
 
   if (get_vendor() == VENDOR_INTEL) {
-      if(LD1.size<=0 || LC1.size<=0){
-	//If we didn't detect L1 correctly before,
-	int count;
-	for (count=0;count <4;count++) {
-	cpuid_count(4, count, &eax, &ebx, &ecx, &edx);
-        switch (eax &0x1f) {
-        case 0:
-          continue;
-          case 1:
-          case 3:
-          {
-            switch ((eax >>5) &0x07)
-            {
-            case 1:
-            {
-//            fprintf(stderr,"L1 data cache...\n");
-            int sets = ecx+1;
-            int lines = (ebx & 0x0fff) +1;
-            ebx>>=12;
-            int part = (ebx&0x03ff)+1;
-            ebx >>=10;
-            int assoc = (ebx&0x03ff)+1;
-            LD1.size = (assoc*part*lines*sets)/1024;
-            LD1.associative = assoc;
-            LD1.linesize= lines;
-            break;
-            }
-            default: 
-              break;
-           }
-          break;
-          }
-         case 2:
-          {
-            switch ((eax >>5) &0x07)
-            {
-            case 1:
-            {
-//            fprintf(stderr,"L1 instruction cache...\n");
-            int sets = ecx+1;
-            int lines = (ebx & 0x0fff) +1;
-            ebx>>=12;
-            int part = (ebx&0x03ff)+1;
-            ebx >>=10;
-            int assoc = (ebx&0x03ff)+1;
-            LC1.size = (assoc*part*lines*sets)/1024;
-            LC1.associative = assoc;
-            LC1.linesize= lines;
-            break;
-            }
-            default: 
-              break;
-           }
-          break;
-          
-          }
-          default:
-          break;
-        }
-      }
-    }
     cpuid(0x80000000, &cpuid_level, &ebx, &ecx, &edx);
     if (cpuid_level >= 0x80000006) {
       if(L2.size<=0){
@@ -1301,19 +1209,6 @@ int get_cpuname(void){
           else
 	    return CPUTYPE_NEHALEM;
 	case 5:
-	  // Skylake X
-#ifndef NO_AVX512
-	  return CPUTYPE_SKYLAKEX;
-#else
-	  if(support_avx())
-#ifndef NO_AVX2
-	  return CPUTYPE_HASWELL;
-#else
-	  return CPUTYPE_SANDYBRIDGE;
-#endif
-	  else
-	  return CPUTYPE_NEHALEM;
-#endif			
         case 14:
 	  // Skylake
           if(support_avx())
@@ -1339,23 +1234,6 @@ int get_cpuname(void){
 	    return CPUTYPE_NEHALEM;
 	}
 	break;
-      case 6:
-        switch (model) {
-        case 6: // Cannon Lake
-#ifndef NO_AVX512
-	  return CPUTYPE_SKYLAKEX;
-#else
-	  if(support_avx())
-#ifndef NO_AVX2
-	  return CPUTYPE_HASWELL;
-#else
-	  return CPUTYPE_SANDYBRIDGE;
-#endif
-	  else
-	  return CPUTYPE_NEHALEM;
-#endif			
-        }
-      break;  
       case 9:
       case 8: 
         switch (model) {
@@ -1452,8 +1330,6 @@ int get_cpuname(void){
 	switch (model) {
 	case 1:
 	  // AMD Ryzen
-	case 8:
-	  // AMD Ryzen2
 	  if(support_avx())
 #ifndef NO_AVX2
 	    return CPUTYPE_ZEN;
@@ -1590,7 +1466,6 @@ static char *cpuname[] = {
   "STEAMROLLER",
   "EXCAVATOR",
   "ZEN",
-  "SKYLAKEX"	
 };
 
 static char *lowercpuname[] = {
@@ -1645,7 +1520,6 @@ static char *lowercpuname[] = {
   "steamroller",
   "excavator",
   "zen",
-  "skylakex"
 };
 
 static char *corename[] = {
@@ -1677,7 +1551,6 @@ static char *corename[] = {
   "STEAMROLLER",
   "EXCAVATOR",
   "ZEN",
-  "SKYLAKEX"	
 };
 
 static char *corename_lower[] = {
@@ -1709,7 +1582,6 @@ static char *corename_lower[] = {
   "steamroller",
   "excavator",
   "zen",
-  "skylakex"	
 };
 
 
@@ -1815,8 +1687,6 @@ int get_coretype(void){
 	break;
       case 3:
 	switch (model) {
-	case 7:
-	  return CORE_ATOM;		
 	case 10:
 	case 14:
 	  if(support_avx())
@@ -1898,19 +1768,6 @@ 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())

ctest.c

@@ -60,14 +60,6 @@ OS_FREEBSD
 OS_NETBSD
 #endif
 
-#if defined(__OpenBSD__)
-OS_OPENBSD
-#endif
-
-#if defined(__DragonFly__)
-OS_DRAGONFLY
-#endif
-
 #if defined(__sun)
 OS_SUNOS
 #endif

ctest/CMakeLists.txt

@@ -1,5 +1,4 @@
 include_directories(${PROJECT_SOURCE_DIR})
-include_directories(${PROJECT_BINARY_DIR})
 
 enable_language(Fortran)
 
@@ -16,7 +15,7 @@ foreach(float_type ${FLOAT_TYPES})
   add_executable(x${float_char}cblat1
     c_${float_char}blat1.f
     c_${float_char}blas1.c)
-  target_link_libraries(x${float_char}cblat1 ${OpenBLAS_LIBNAME})
+  target_link_libraries(x${float_char}cblat1 ${OpenBLAS_LIBNAME}_static)
   add_test(NAME "x${float_char}cblat1"
     COMMAND "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat1")
 
@@ -28,7 +27,7 @@ foreach(float_type ${FLOAT_TYPES})
     auxiliary.c
     c_xerbla.c
     constant.c)
-  target_link_libraries(x${float_char}cblat2 ${OpenBLAS_LIBNAME})
+  target_link_libraries(x${float_char}cblat2 ${OpenBLAS_LIBNAME}_static)
   add_test(NAME "x${float_char}cblat2"
     COMMAND sh "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh" "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat2" "${PROJECT_SOURCE_DIR}/ctest/${float_char}in2")
 
@@ -40,7 +39,7 @@ foreach(float_type ${FLOAT_TYPES})
     auxiliary.c
     c_xerbla.c
     constant.c)
-  target_link_libraries(x${float_char}cblat3 ${OpenBLAS_LIBNAME})
+  target_link_libraries(x${float_char}cblat3 ${OpenBLAS_LIBNAME}_static)
   add_test(NAME "x${float_char}cblat3"
     COMMAND sh "${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh" "${CMAKE_CURRENT_BINARY_DIR}/x${float_char}cblat3" "${PROJECT_SOURCE_DIR}/ctest/${float_char}in3")
 

ctest/Makefile

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

ctest/c_cblas1.c

@@ -9,7 +9,7 @@
 #include "common.h"
 #include "cblas_test.h"
 
-void F77_caxpy(const int *N, OPENBLAS_CONST void *alpha, void *X,
+void F77_caxpy(const int *N, const void *alpha, void *X,
                     const int *incX, void *Y, const int *incY)
 {
    cblas_caxpy(*N, alpha, X, *incX, Y, *incY);
@@ -58,13 +58,13 @@ void F77_cswap( const int *N, void *X, const int *incX,
    return;
 }
 
-int F77_icamax(const int *N, OPENBLAS_CONST void *X, const int *incX)
+int F77_icamax(const int *N, const void *X, const int *incX)
 {
    if (*N < 1 || *incX < 1) return(0);
    return (cblas_icamax(*N, X, *incX)+1);
 }
 
-float F77_scnrm2(const int *N, OPENBLAS_CONST void *X, const int *incX)
+float F77_scnrm2(const int *N, const void *X, const int *incX)
 {
    return cblas_scnrm2(*N, X, *incX);
 }

ctest/c_cblas2.c

@@ -9,9 +9,9 @@
 #include "cblas_test.h"
 
 void F77_cgemv(int *order, char *transp, int *m, int *n,
-          OPENBLAS_CONST void *alpha,
-          CBLAS_TEST_COMPLEX *a, int *lda, OPENBLAS_CONST void *x, int *incx,
-          OPENBLAS_CONST void *beta, void *y, int *incy) {
+          const void *alpha,
+          CBLAS_TEST_COMPLEX *a, int *lda, const void *x, int *incx,
+          const void *beta, void *y, int *incy) {
 
   CBLAS_TEST_COMPLEX *A;
   int i,j,LDA;

ctest/c_cblat2.f

@@ -349,13 +349,13 @@
            CALL CCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NKB, KB, NINC, INC,
      $                  NMAX, INCMAX, A, AA, AS, Y, YY, YS, YT, G, Z,
-     $                  0 )
+     $			0 )
            END IF
            IF (RORDER) THEN
            CALL CCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NKB, KB, NINC, INC,
      $                  NMAX, INCMAX, A, AA, AS, Y, YY, YS, YT, G, Z,
-     $                  1 )
+     $			1 )
            END IF
             GO TO 200
 *           Test CGERC, 12, CGERU, 13.
@@ -2660,7 +2660,7 @@
    50    CONTINUE
       END IF
 *
-C   60 CONTINUE
+   60 CONTINUE
       LCERES = .TRUE.
       GO TO 80
    70 CONTINUE

ctest/c_cblat3.f

@@ -329,13 +329,13 @@
             CALL CCHK3(SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                 REWI, FATAL, NIDIM, IDIM, NALF, ALF, NMAX, AB,
      $                 AA, AS, AB( 1, NMAX + 1 ), BB, BS, CT, G, C,
-     $          0 )
+     $		0 )
             END IF
             IF (RORDER) THEN
             CALL CCHK3(SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                 REWI, FATAL, NIDIM, IDIM, NALF, ALF, NMAX, AB,
      $                 AA, AS, AB( 1, NMAX + 1 ), BB, BS, CT, G, C,
-     $          1 )
+     $		1 )
             END IF
             GO TO 190
 *           Test CHERK, 06, CSYRK, 07.
@@ -357,13 +357,13 @@
             CALL CCHK5(SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                 REWI, FATAL, NIDIM, IDIM, NALF, ALF, NBET, BET,
      $                 NMAX, AB, AA, AS, BB, BS, C, CC, CS, CT, G, W,
-     $          0 )
+     $		0 )
             END IF
             IF (RORDER) THEN
             CALL CCHK5(SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                 REWI, FATAL, NIDIM, IDIM, NALF, ALF, NBET, BET,
      $                 NMAX, AB, AA, AS, BB, BS, C, CC, CS, CT, G, W,
-     $          1 )
+     $		1 )
             END IF
             GO TO 190
 *
@@ -707,9 +707,9 @@
  9998 FORMAT(' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT( 1X, I6, ': ', A12,'(''', A1, ''',''', A1, ''',',
-C     $     3( I3, ',' ), '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3,
-C     $     ',(', F4.1, ',', F4.1, '), C,', I3, ').' )
+ 9995 FORMAT( 1X, I6, ': ', A12,'(''', A1, ''',''', A1, ''',',
+     $     3( I3, ',' ), '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3,
+     $     ',(', F4.1, ',', F4.1, '), C,', I3, ').' )
  9994 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1033,9 +1033,9 @@ C     $     ',(', F4.1, ',', F4.1, '), C,', I3, ').' )
  9998 FORMAT(' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',(', F4.1,
-C     $      ',', F4.1, '), C,', I3, ')    .' )
+ 9995 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',(', F4.1,
+     $      ',', F4.1, '), C,', I3, ')    .' )
  9994 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1385,9 +1385,9 @@ C     $      ',', F4.1, '), C,', I3, ')    .' )
  9998 FORMAT(' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT(' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT(1X, I6, ': ', A12,'(', 4( '''', A1, ''',' ), 2( I3, ',' ),
-C     $     '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ')         ',
-C     $      '      .' )
+ 9995 FORMAT(1X, I6, ': ', A12,'(', 4( '''', A1, ''',' ), 2( I3, ',' ),
+     $     '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ')         ',
+     $      '      .' )
  9994 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1768,12 +1768,12 @@ C     $      '      .' )
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
-C 9994 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $     F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')               ',
-C     $      '          .' )
-C 9993 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      '(', F4.1, ',', F4.1, ') , A,', I3, ',(', F4.1, ',', F4.1,
-C     $      '), C,', I3, ')          .' )
+ 9994 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $     F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')               ',
+     $      '          .' )
+ 9993 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      '(', F4.1, ',', F4.1, ') , A,', I3, ',(', F4.1, ',', F4.1,
+     $      '), C,', I3, ')          .' )
  9992 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -2221,12 +2221,12 @@ C     $      '), C,', I3, ')          .' )
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
-C 9994 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',', F4.1,
-C     $      ', C,', I3, ')           .' )
-C 9993 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',(', F4.1,
-C     $      ',', F4.1, '), C,', I3, ')    .' )
+ 9994 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',', F4.1,
+     $      ', C,', I3, ')           .' )
+ 9993 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',(', F4.1,
+     $      ',', F4.1, '), C,', I3, ')    .' )
  9992 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -2702,7 +2702,7 @@ C     $      ',', F4.1, '), C,', I3, ')    .' )
    50    CONTINUE
       END IF
 *
-C   60 CONTINUE
+   60 CONTINUE
       LCERES = .TRUE.
       GO TO 80
    70 CONTINUE

ctest/c_dblas1.c

@@ -14,7 +14,7 @@ double F77_dasum(const int *N, double *X, const int *incX)
    return cblas_dasum(*N, X, *incX);
 }
 
-void F77_daxpy(const int *N, const double *alpha, OPENBLAS_CONST double *X,
+void F77_daxpy(const int *N, const double *alpha, const double *X,
                     const int *incX, double *Y, const int *incY)
 {
    cblas_daxpy(*N, *alpha, X, *incX, Y, *incY);
@@ -28,13 +28,13 @@ void F77_dcopy(const int *N, double *X, const int *incX,
    return;
 }
 
-double F77_ddot(const int *N, OPENBLAS_CONST double *X, const int *incX,
-                OPENBLAS_CONST double *Y, const int *incY)
+double F77_ddot(const int *N, const double *X, const int *incX,
+                const double *Y, const int *incY)
 {
    return cblas_ddot(*N, X, *incX, Y, *incY);
 }
 
-double F77_dnrm2(const int *N, OPENBLAS_CONST double *X, const int *incX)
+double F77_dnrm2(const int *N, const double *X, const int *incX)
 {
    return cblas_dnrm2(*N, X, *incX);
 }
@@ -72,12 +72,12 @@ double F77_dzasum(const int *N, void *X, const int *incX)
    return cblas_dzasum(*N, X, *incX);
 }
 
-double F77_dznrm2(const int *N, OPENBLAS_CONST void *X, const int *incX)
+double F77_dznrm2(const int *N, const void *X, const int *incX)
 {
    return cblas_dznrm2(*N, X, *incX);
 }
 
-int F77_idamax(const int *N, OPENBLAS_CONST double *X, const int *incX)
+int F77_idamax(const int *N, const double *X, const int *incX)
 {
    if (*N < 1 || *incX < 1) return(0);
    return (cblas_idamax(*N, X, *incX)+1);

ctest/c_dblat1.f

@@ -211,11 +211,11 @@
             IF (ICASE.EQ.7) THEN
 *              .. DNRM2TEST ..
                STEMP(1) = DTRUE1(NP1)
-               CALL STEST1(DNRM2TEST(N,SX,INCX),STEMP(1),STEMP,SFAC)
+               CALL STEST1(DNRM2TEST(N,SX,INCX),STEMP,STEMP,SFAC)
             ELSE IF (ICASE.EQ.8) THEN
 *              .. DASUMTEST ..
                STEMP(1) = DTRUE3(NP1)
-               CALL STEST1(DASUMTEST(N,SX,INCX),STEMP(1),STEMP,SFAC)
+               CALL STEST1(DASUMTEST(N,SX,INCX),STEMP,STEMP,SFAC)
             ELSE IF (ICASE.EQ.9) THEN
 *              .. DSCALTEST ..
                CALL DSCALTEST(N,SA((INCX-1)*5+NP1),SX,INCX)

ctest/c_dblat2.f

@@ -345,13 +345,13 @@
             CALL DCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NKB, KB, NINC, INC,
      $                  NMAX, INCMAX, A, AA, AS, Y, YY, YS, YT, G, Z,
-     $                  0 )
+     $			0 )
             END IF
             IF (RORDER) THEN
             CALL DCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NKB, KB, NINC, INC,
      $                  NMAX, INCMAX, A, AA, AS, Y, YY, YS, YT, G, Z,
-     $                  1 )
+     $			1 )
             END IF
             GO TO 200
 *           Test DGER, 12.
@@ -797,9 +797,9 @@
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( 1X, I6, ': ',A12, '(', A14, ',', 4( I3, ',' ), F4.1,
      $      ', A,', I3, ',',/ 10x,'X,', I2, ',', F4.1, ', Y,',
@@ -1004,7 +1004,7 @@ C     $      ' - SUSPECT *******' )
      $                           REWIND NTRA
                               CALL CDSBMV( IORDER, UPLO, N, K, ALPHA,
      $                                    AA, LDA, XX, INCX, BETA, YY,
-     $                                    INCY )
+     $					  INCY )
                            ELSE IF( PACKED )THEN
                               IF( TRACE )
      $                           WRITE( NTRA, FMT = 9995 )NC, SNAME,
@@ -1156,9 +1156,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( 1X, I6, ': ',A12, '(', A14, ',', I3, ',', F4.1, ', AP',
      $      ', X,', I2, ',', F4.1, ', Y,', I2, ') .' )
@@ -1191,7 +1191,7 @@ C     $      ' - SUSPECT *******' )
 *     .. Scalar Arguments ..
       DOUBLE PRECISION   EPS, THRESH
       INTEGER            INCMAX, NIDIM, NINC, NKB, NMAX, NOUT, NTRA,
-     $                   IORDER
+     $			 IORDER
       LOGICAL            FATAL, REWI, TRACE
       CHARACTER*12       SNAME
 *     .. Array Arguments ..
@@ -1216,7 +1216,7 @@ C     $      ' - SUSPECT *******' )
       EXTERNAL           LDE, LDERES
 *     .. External Subroutines ..
       EXTERNAL           DMAKE, DMVCH, CDTBMV, CDTBSV, CDTPMV,
-     $                   CDTPSV, CDTRMV,  CDTRSV
+     $			 CDTPSV, CDTRMV,  CDTRSV
 *     .. Intrinsic Functions ..
       INTRINSIC          ABS, MAX
 *     .. Scalars in Common ..
@@ -1544,9 +1544,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( 1X, I6, ': ',A12, '(', 3( A14,',' ),/ 10x, I3, ', AP, ',
      $      'X,', I2, ') .' )
@@ -1579,7 +1579,7 @@ C     $      ' - SUSPECT *******' )
 *     .. Scalar Arguments ..
       DOUBLE PRECISION   EPS, THRESH
       INTEGER            INCMAX, NALF, NIDIM, NINC, NMAX, NOUT, NTRA,
-     $                   IORDER
+     $			 IORDER
       LOGICAL            FATAL, REWI, TRACE
       CHARACTER*12       SNAME
 *     .. Array Arguments ..
@@ -1819,9 +1819,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
  9994 FORMAT( 1X, I6, ': ',A12, '(', 2( I3, ',' ), F4.1, ', X,', I2,
@@ -1851,7 +1851,7 @@ C     $      ' - SUSPECT *******' )
 *     .. Scalar Arguments ..
       DOUBLE PRECISION   EPS, THRESH
       INTEGER            INCMAX, NALF, NIDIM, NINC, NMAX, NOUT, NTRA,
-     $                   IORDER
+     $			 IORDER
       LOGICAL            FATAL, REWI, TRACE
       CHARACTER*12       SNAME
 *     .. Array Arguments ..
@@ -1973,7 +1973,7 @@ C     $      ' - SUSPECT *******' )
                      IF( REWI )
      $                  REWIND NTRA
                      CALL CDSYR( IORDER, UPLO, N, ALPHA, XX, INCX,
-     $                            AA, LDA )
+     $				  AA, LDA )
                   ELSE IF( PACKED )THEN
                      IF( TRACE )
      $                  WRITE( NTRA, FMT = 9994 )NC, SNAME, CUPLO, N,
@@ -2113,9 +2113,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
  9994 FORMAT( 1X, I6, ': ',A12, '(', A14, ',', I3, ',', F4.1, ', X,',
@@ -2147,7 +2147,7 @@ C     $      ' - SUSPECT *******' )
 *     .. Scalar Arguments ..
       DOUBLE PRECISION   EPS, THRESH
       INTEGER            INCMAX, NALF, NIDIM, NINC, NMAX, NOUT, NTRA,
-     $                   IORDER
+     $			 IORDER
       LOGICAL            FATAL, REWI, TRACE
       CHARACTER*12       SNAME
 *     .. Array Arguments ..
@@ -2445,9 +2445,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
  9994 FORMAT( 1X, I6, ': ',A12, '(', A14, ',', I3, ',', F4.1, ', X,',
@@ -2833,7 +2833,7 @@ C     $      ' - SUSPECT *******' )
    50    CONTINUE
       END IF
 *
-C   60 CONTINUE
+   60 CONTINUE
       LDERES = .TRUE.
       GO TO 80
    70 CONTINUE

ctest/c_dblat3.f

@@ -56,7 +56,7 @@
 *     .. Local Scalars ..
       DOUBLE PRECISION   EPS, ERR, THRESH
       INTEGER            I, ISNUM, J, N, NALF, NBET, NIDIM, NTRA,
-     $                            LAYOUT
+     $			          LAYOUT
       LOGICAL            FATAL, LTESTT, REWI, SAME, SFATAL, TRACE,
      $                   TSTERR, CORDER, RORDER
       CHARACTER*1        TRANSA, TRANSB
@@ -78,7 +78,7 @@
       EXTERNAL           DDIFF, LDE
 *     .. External Subroutines ..
       EXTERNAL           DCHK1, DCHK2, DCHK3, DCHK4, DCHK5, CD3CHKE,
-     $                   DMMCH
+     $			 DMMCH
 *     .. Intrinsic Functions ..
       INTRINSIC          MAX, MIN
 *     .. Scalars in Common ..
@@ -323,13 +323,13 @@
             CALL DCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NALF, ALF, NMAX, AB,
      $                  AA, AS, AB( 1, NMAX + 1 ), BB, BS, CT, G, C,
-     $                  0 )
+     $			0 )
             END IF
             IF (RORDER) THEN
             CALL DCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NALF, ALF, NMAX, AB,
      $                  AA, AS, AB( 1, NMAX + 1 ), BB, BS, CT, G, C,
-     $                  1 )
+     $			1 )
             END IF
             GO TO 190
 *           Test DSYRK, 05.
@@ -351,13 +351,13 @@
             CALL DCHK5( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NALF, ALF, NBET, BET,
      $                  NMAX, AB, AA, AS, BB, BS, C, CC, CS, CT, G, W,
-     $                  0 )
+     $			0 )
             END IF
             IF (RORDER) THEN
             CALL DCHK5( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NALF, ALF, NBET, BET,
      $                  NMAX, AB, AA, AS, BB, BS, C, CC, CS, CT, G, W,
-     $                  1 )
+     $			1 )
             END IF
             GO TO 190
 *
@@ -588,7 +588,7 @@
      $                        REWIND NTRA
                            CALL CDGEMM( IORDER, TRANSA, TRANSB, M, N,
      $                                   K, ALPHA, AA, LDA, BB, LDB,
-     $                                   BETA, CC, LDC )
+     $					 BETA, CC, LDC )
 *
 *                          Check if error-exit was taken incorrectly.
 *
@@ -694,9 +694,9 @@
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT( 1X, I6, ': ', A12,'(''', A1, ''',''', A1, ''',',
-C     $      3( I3, ',' ), F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', ',
-C     $      'C,', I3, ').' )
+ 9995 FORMAT( 1X, I6, ': ', A12,'(''', A1, ''',''', A1, ''',',
+     $      3( I3, ',' ), F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', ',
+     $      'C,', I3, ').' )
  9994 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1007,9 +1007,9 @@ C     $      'C,', I3, ').' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', C,', I3, ')   ',
-C     $      ' .' )
+ 9995 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', C,', I3, ')   ',
+     $      ' .' )
  9994 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1201,7 +1201,7 @@ C     $      ' .' )
      $                           REWIND NTRA
                               CALL CDTRMM( IORDER, SIDE, UPLO, TRANSA,
      $                                    DIAG, M, N, ALPHA, AA, LDA,
-     $                                    BB, LDB )
+     $					  BB, LDB )
                            ELSE IF( SNAME( 10: 11 ).EQ.'sm' )THEN
                               IF( TRACE )
      $                           CALL DPRCN3( NTRA, NC, SNAME, IORDER,
@@ -1211,7 +1211,7 @@ C     $      ' .' )
      $                           REWIND NTRA
                               CALL CDTRSM( IORDER, SIDE, UPLO, TRANSA,
      $                                    DIAG, M, N, ALPHA, AA, LDA,
-     $                                    BB, LDB )
+     $					  BB, LDB )
                            END IF
 *
 *                          Check if error-exit was taken incorrectly.
@@ -1355,8 +1355,8 @@ C     $      ' .' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT( 1X, I6, ': ', A12,'(', 4( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      F4.1, ', A,', I3, ', B,', I3, ')        .' )
+ 9995 FORMAT( 1X, I6, ': ', A12,'(', 4( '''', A1, ''',' ), 2( I3, ',' ),
+     $      F4.1, ', A,', I3, ', B,', I3, ')        .' )
  9994 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1681,8 +1681,8 @@ C     $      F4.1, ', A,', I3, ', B,', I3, ')        .' )
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
-C 9994 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')           .' )
+ 9994 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')           .' )
  9993 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1726,7 +1726,7 @@ C     $      F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')           .' )
       SUBROUTINE DCHK5( SNAME, EPS, THRESH, NOUT, NTRA, TRACE, REWI,
      $                  FATAL, NIDIM, IDIM, NALF, ALF, NBET, BET, NMAX,
      $                  AB, AA, AS, BB, BS, C, CC, CS, CT, G, W,
-     $                  IORDER )
+     $			IORDER )
 *
 *  Tests DSYR2K.
 *
@@ -1888,7 +1888,7 @@ C     $      F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')           .' )
      $                     REWIND NTRA
                         CALL CDSYR2K( IORDER, UPLO, TRANS, N, K,
      $                               ALPHA, AA, LDA, BB, LDB, BETA,
-     $                               CC, LDC )
+     $				     CC, LDC )
 *
 *                       Check if error-exit was taken incorrectly.
 *
@@ -2037,9 +2037,9 @@ C     $      F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')           .' )
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
-C 9994 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', C,', I3, ')   ',
-C     $      ' .' )
+ 9994 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', C,', I3, ')   ',
+     $      ' .' )
  9993 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -2399,7 +2399,7 @@ C     $      ' .' )
    50    CONTINUE
       END IF
 *
-C   60 CONTINUE
+   60 CONTINUE
       LDERES = .TRUE.
       GO TO 80
    70 CONTINUE

ctest/c_sblas1.c

@@ -14,7 +14,7 @@ float F77_sasum(blasint *N, float *X, blasint *incX)
    return cblas_sasum(*N, X, *incX);
 }
 
-void F77_saxpy(blasint *N, const float *alpha, OPENBLAS_CONST float *X,
+void F77_saxpy(blasint *N, const float *alpha, const float *X,
                     blasint *incX, float *Y, blasint *incY)
 {
    cblas_saxpy(*N, *alpha, X, *incX, Y, *incY);
@@ -26,25 +26,25 @@ float F77_scasum(blasint *N, float *X, blasint *incX)
    return cblas_scasum(*N, X, *incX);
 }
 
-float F77_scnrm2(blasint *N, OPENBLAS_CONST float *X, blasint *incX)
+float F77_scnrm2(blasint *N, const float *X, blasint *incX)
 {
    return cblas_scnrm2(*N, X, *incX);
 }
 
-void F77_scopy(blasint *N, OPENBLAS_CONST float *X, blasint *incX,
+void F77_scopy(blasint *N, const float *X, blasint *incX,
                     float *Y, blasint *incY)
 {
    cblas_scopy(*N, X, *incX, Y, *incY);
    return;
 }
 
-float F77_sdot(blasint *N, OPENBLAS_CONST float *X, blasint *incX,
-                        OPENBLAS_CONST float *Y, blasint *incY)
+float F77_sdot(blasint *N, const float *X, blasint *incX,
+                        const float *Y, blasint *incY)
 {
    return cblas_sdot(*N, X, *incX, Y, *incY);
 }
 
-float F77_snrm2(blasint *N, OPENBLAS_CONST float *X, blasint *incX)
+float F77_snrm2(blasint *N, const float *X, blasint *incX)
 {
    return cblas_snrm2(*N, X, *incX);
 }
@@ -76,7 +76,7 @@ void F77_sswap( blasint *N, float *X, blasint *incX,
    return;
 }
 
-int F77_isamax(blasint *N, OPENBLAS_CONST float *X, blasint *incX)
+int F77_isamax(blasint *N, const float *X, blasint *incX)
 {
    if (*N < 1 || *incX < 1) return(0);
    return (cblas_isamax(*N, X, *incX)+1);

ctest/c_sblat1.f

@@ -211,11 +211,11 @@
             IF (ICASE.EQ.7) THEN
 *              .. SNRM2TEST ..
                STEMP(1) = DTRUE1(NP1)
-               CALL STEST1(SNRM2TEST(N,SX,INCX),STEMP(1),STEMP,SFAC)
+               CALL STEST1(SNRM2TEST(N,SX,INCX),STEMP,STEMP,SFAC)
             ELSE IF (ICASE.EQ.8) THEN
 *              .. SASUMTEST ..
                STEMP(1) = DTRUE3(NP1)
-               CALL STEST1(SASUMTEST(N,SX,INCX),STEMP(1),STEMP,SFAC)
+               CALL STEST1(SASUMTEST(N,SX,INCX),STEMP,STEMP,SFAC)
             ELSE IF (ICASE.EQ.9) THEN
 *              .. SSCALTEST ..
                CALL SSCALTEST(N,SA((INCX-1)*5+NP1),SX,INCX)

ctest/c_sblat2.f

@@ -345,13 +345,13 @@
             CALL SCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NKB, KB, NINC, INC,
      $                  NMAX, INCMAX, A, AA, AS, Y, YY, YS, YT, G, Z,
-     $                  0 )
+     $			0 )
             END IF
             IF (RORDER) THEN
             CALL SCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NKB, KB, NINC, INC,
      $                  NMAX, INCMAX, A, AA, AS, Y, YY, YS, YT, G, Z,
-     $                  1 )
+     $			1 )
             END IF
             GO TO 200
 *           Test SGER, 12.
@@ -797,9 +797,9 @@
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( 1X, I6, ': ',A12, '(', A14, ',', 4( I3, ',' ), F4.1,
      $      ', A,', I3, ',',/ 10x, 'X,', I2, ',', F4.1, ', Y,',
@@ -1004,7 +1004,7 @@ C     $      ' - SUSPECT *******' )
      $                           REWIND NTRA
                               CALL CSSBMV( IORDER, UPLO, N, K, ALPHA,
      $                                    AA, LDA, XX, INCX, BETA, YY,
-     $                                    INCY )
+     $					  INCY )
                            ELSE IF( PACKED )THEN
                               IF( TRACE )
      $                           WRITE( NTRA, FMT = 9995 )NC, SNAME,
@@ -1156,9 +1156,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( 1X, I6, ': ',A12, '(', A14, ',', I3, ',', F4.1, ', AP',
      $      ', X,', I2, ',', F4.1, ', Y,', I2, ') .' )
@@ -1191,7 +1191,7 @@ C     $      ' - SUSPECT *******' )
 *     .. Scalar Arguments ..
       REAL               EPS, THRESH
       INTEGER            INCMAX, NIDIM, NINC, NKB, NMAX, NOUT, NTRA,
-     $                   IORDER
+     $			 IORDER
       LOGICAL            FATAL, REWI, TRACE
       CHARACTER*12       SNAME
 *     .. Array Arguments ..
@@ -1216,7 +1216,7 @@ C     $      ' - SUSPECT *******' )
       EXTERNAL           LSE, LSERES
 *     .. External Subroutines ..
       EXTERNAL           SMAKE, SMVCH, CSTBMV, CSTBSV, CSTPMV,
-     $                   CSTPSV, CSTRMV,  CSTRSV
+     $			 CSTPSV, CSTRMV,  CSTRSV
 *     .. Intrinsic Functions ..
       INTRINSIC          ABS, MAX
 *     .. Scalars in Common ..
@@ -1544,9 +1544,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( 1X, I6, ': ',A12, '(', 3( A14,',' ),/ 10x, I3, ', AP, ',
      $      'X,', I2, ') .' )
@@ -1579,7 +1579,7 @@ C     $      ' - SUSPECT *******' )
 *     .. Scalar Arguments ..
       REAL               EPS, THRESH
       INTEGER            INCMAX, NALF, NIDIM, NINC, NMAX, NOUT, NTRA,
-     $                   IORDER
+     $			 IORDER
       LOGICAL            FATAL, REWI, TRACE
       CHARACTER*12       SNAME
 *     .. Array Arguments ..
@@ -1819,9 +1819,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
  9994 FORMAT( 1X, I6, ': ',A12, '(', 2( I3, ',' ), F4.1, ', X,', I2,
@@ -1851,7 +1851,7 @@ C     $      ' - SUSPECT *******' )
 *     .. Scalar Arguments ..
       REAL               EPS, THRESH
       INTEGER            INCMAX, NALF, NIDIM, NINC, NMAX, NOUT, NTRA,
-     $                   IORDER
+     $			 IORDER
       LOGICAL            FATAL, REWI, TRACE
       CHARACTER*12       SNAME
 *     .. Array Arguments ..
@@ -1973,7 +1973,7 @@ C     $      ' - SUSPECT *******' )
                      IF( REWI )
      $                  REWIND NTRA
                      CALL CSSYR( IORDER, UPLO, N, ALPHA, XX, INCX,
-     $                            AA, LDA )
+     $				  AA, LDA )
                   ELSE IF( PACKED )THEN
                      IF( TRACE )
      $                  WRITE( NTRA, FMT = 9994 )NC, SNAME, CUPLO, N,
@@ -2113,9 +2113,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
  9994 FORMAT( 1X, I6, ': ',A12, '(', A14, ',', I3, ',', F4.1, ', X,',
@@ -2147,7 +2147,7 @@ C     $      ' - SUSPECT *******' )
 *     .. Scalar Arguments ..
       REAL               EPS, THRESH
       INTEGER            INCMAX, NALF, NIDIM, NINC, NMAX, NOUT, NTRA,
-     $                   IORDER
+     $			 IORDER
       LOGICAL            FATAL, REWI, TRACE
       CHARACTER*12       SNAME
 *     .. Array Arguments ..
@@ -2445,9 +2445,9 @@ C     $      ' - SUSPECT *******' )
      $ ' (', I6, ' CALL', 'S)' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
-C 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
-C     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
-C     $      ' - SUSPECT *******' )
+ 9997 FORMAT( ' ',A12, ' COMPLETED THE COMPUTATIONAL TESTS (', I6, ' C',
+     $      'ALLS)', /' ******* BUT WITH MAXIMUM TEST RATIO', F8.2,
+     $      ' - SUSPECT *******' )
  9996 FORMAT( ' ******* ',A12, ' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
  9994 FORMAT( 1X, I6, ': ',A12, '(', A14, ',', I3, ',', F4.1, ', X,',
@@ -2833,7 +2833,7 @@ C     $      ' - SUSPECT *******' )
    50    CONTINUE
       END IF
 *
-C   60 CONTINUE
+   60 CONTINUE
       LSERES = .TRUE.
       GO TO 80
    70 CONTINUE

ctest/c_sblat3.f

@@ -694,9 +694,9 @@
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT( 1X, I6, ': ', A12,'(''', A1, ''',''', A1, ''',',
-C     $      3( I3, ',' ), F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', ',
-C     $      'C,', I3, ').' )
+ 9995 FORMAT( 1X, I6, ': ', A12,'(''', A1, ''',''', A1, ''',',
+     $      3( I3, ',' ), F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', ',
+     $      'C,', I3, ').' )
  9994 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1011,9 +1011,9 @@ C     $      'C,', I3, ').' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', C,', I3, ')   ',
-C     $      ' .' )
+ 9995 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', C,', I3, ')   ',
+     $      ' .' )
  9994 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1359,8 +1359,8 @@ C     $      ' .' )
  9998 FORMAT( ' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT( 1X, I6, ': ', A12,'(', 4( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      F4.1, ', A,', I3, ', B,', I3, ')        .' )
+ 9995 FORMAT( 1X, I6, ': ', A12,'(', 4( '''', A1, ''',' ), 2( I3, ',' ),
+     $      F4.1, ', A,', I3, ', B,', I3, ')        .' )
  9994 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1686,8 +1686,8 @@ C     $      F4.1, ', A,', I3, ', B,', I3, ')        .' )
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
-C 9994 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')           .' )
+ 9994 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')           .' )
  9993 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -2041,9 +2041,9 @@ C     $      F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')           .' )
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
-C 9994 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', C,', I3, ')   ',
-C     $      ' .' )
+ 9994 FORMAT( 1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      F4.1, ', A,', I3, ', B,', I3, ',', F4.1, ', C,', I3, ')   ',
+     $      ' .' )
  9993 FORMAT( ' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -2403,7 +2403,7 @@ C     $      ' .' )
    50    CONTINUE
       END IF
 *
-C   60 CONTINUE
+   60 CONTINUE
       LSERES = .TRUE.
       GO TO 80
    70 CONTINUE

ctest/c_xerbla.c

@@ -131,7 +131,7 @@ void F77_xerbla(char *srname, void *vinfo)
 int    BLASFUNC(xerbla)(char *name, blasint *info, blasint length) {
 
   F77_xerbla(name, info);
-  return 0;
+
 };
 
 

ctest/c_zblas1.c

@@ -9,7 +9,7 @@
 #include "common.h"
 #include "cblas_test.h"
 
-void F77_zaxpy(const int *N, OPENBLAS_CONST void *alpha, void *X,
+void F77_zaxpy(const int *N, const void *alpha, void *X,
                     const int *incX, void *Y, const int *incY)
 {
    cblas_zaxpy(*N, alpha, X, *incX, Y, *incY);
@@ -23,8 +23,8 @@ void F77_zcopy(const int *N, void *X, const int *incX,
    return;
 }
 
-void F77_zdotc(const int *N, OPENBLAS_CONST void *X, const int *incX,
-                     OPENBLAS_CONST void *Y, const int *incY,void *dotc)
+void F77_zdotc(const int *N, const void *X, const int *incX,
+                     const void *Y, const int *incY,void *dotc)
 {
    cblas_zdotc_sub(*N, X, *incX, Y, *incY, dotc);
    return;
@@ -58,13 +58,13 @@ void F77_zswap( const int *N, void *X, const int *incX,
    return;
 }
 
-int F77_izamax(const int *N, OPENBLAS_CONST void *X, const int *incX)
+int F77_izamax(const int *N, const void *X, const int *incX)
 {
    if (*N < 1 || *incX < 1) return(0);
    return(cblas_izamax(*N, X, *incX)+1);
 }
 
-double F77_dznrm2(const int *N, OPENBLAS_CONST void *X, const int *incX)
+double F77_dznrm2(const int *N, const void *X, const int *incX)
 {
    return cblas_dznrm2(*N, X, *incX);
 }

ctest/c_zblas2.c

@@ -9,9 +9,9 @@
 #include "cblas_test.h"
 
 void F77_zgemv(int *order, char *transp, int *m, int *n,
-          OPENBLAS_CONST void *alpha,
-          CBLAS_TEST_ZOMPLEX *a, int *lda, OPENBLAS_CONST void *x, int *incx,
-          OPENBLAS_CONST void *beta, void *y, int *incy) {
+          const void *alpha,
+          CBLAS_TEST_ZOMPLEX *a, int *lda, const void *x, int *incx,
+          const void *beta, void *y, int *incy) {
 
   CBLAS_TEST_ZOMPLEX *A;
   int i,j,LDA;

ctest/c_zblat2.f

@@ -349,13 +349,13 @@
            CALL ZCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NKB, KB, NINC, INC,
      $                  NMAX, INCMAX, A, AA, AS, Y, YY, YS, YT, G, Z,
-     $                  0 )
+     $			0 )
            END IF
            IF (RORDER) THEN
            CALL ZCHK3( SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                  REWI, FATAL, NIDIM, IDIM, NKB, KB, NINC, INC,
      $                  NMAX, INCMAX, A, AA, AS, Y, YY, YS, YT, G, Z,
-     $                  1 )
+     $			1 )
            END IF
             GO TO 200
 *           Test ZGERC, 12, ZGERU, 13.

ctest/c_zblat3.f

@@ -330,13 +330,13 @@
             CALL ZCHK3(SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                 REWI, FATAL, NIDIM, IDIM, NALF, ALF, NMAX, AB,
      $                 AA, AS, AB( 1, NMAX + 1 ), BB, BS, CT, G, C,
-     $                 0 )
+     $		       0 )
             END IF
             IF (RORDER) THEN
             CALL ZCHK3(SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                 REWI, FATAL, NIDIM, IDIM, NALF, ALF, NMAX, AB,
      $                 AA, AS, AB( 1, NMAX + 1 ), BB, BS, CT, G, C,
-     $                 1 )
+     $		       1 )
             END IF
             GO TO 190
 *           Test ZHERK, 06, ZSYRK, 07.
@@ -358,13 +358,13 @@
             CALL ZCHK5(SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                 REWI, FATAL, NIDIM, IDIM, NALF, ALF, NBET, BET,
      $                 NMAX, AB, AA, AS, BB, BS, C, CC, CS, CT, G, W,
-     $                 0 )
+     $		       0 )
             END IF
             IF (RORDER) THEN
             CALL ZCHK5(SNAMES( ISNUM ), EPS, THRESH, NOUT, NTRA, TRACE,
      $                 REWI, FATAL, NIDIM, IDIM, NALF, ALF, NBET, BET,
      $                 NMAX, AB, AA, AS, BB, BS, C, CC, CS, CT, G, W,
-     $                 1 )
+     $		       1 )
             END IF
             GO TO 190
 *
@@ -708,9 +708,9 @@
  9998 FORMAT(' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT( 1X, I6, ': ', A12,'(''', A1, ''',''', A1, ''',',
-C     $     3( I3, ',' ), '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3,
-C     $     ',(', F4.1, ',', F4.1, '), C,', I3, ').' )
+ 9995 FORMAT( 1X, I6, ': ', A12,'(''', A1, ''',''', A1, ''',',
+     $     3( I3, ',' ), '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3,
+     $     ',(', F4.1, ',', F4.1, '), C,', I3, ').' )
  9994 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1034,9 +1034,9 @@ C     $     ',(', F4.1, ',', F4.1, '), C,', I3, ').' )
  9998 FORMAT(' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',(', F4.1,
-C     $      ',', F4.1, '), C,', I3, ')    .' )
+ 9995 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',(', F4.1,
+     $      ',', F4.1, '), C,', I3, ')    .' )
  9994 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1386,9 +1386,9 @@ C     $      ',', F4.1, '), C,', I3, ')    .' )
  9998 FORMAT(' ******* FATAL ERROR - PARAMETER NUMBER ', I2, ' WAS CH',
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT(' ******* ', A12,' FAILED ON CALL NUMBER:' )
-C 9995 FORMAT(1X, I6, ': ', A12,'(', 4( '''', A1, ''',' ), 2( I3, ',' ),
-C     $     '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ')         ',
-C     $      '      .' )
+ 9995 FORMAT(1X, I6, ': ', A12,'(', 4( '''', A1, ''',' ), 2( I3, ',' ),
+     $     '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ')         ',
+     $      '      .' )
  9994 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -1769,12 +1769,12 @@ C     $      '      .' )
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
-C 9994 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $     F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')               ',
-C     $      '          .' )
-C 9993 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      '(', F4.1, ',', F4.1, ') , A,', I3, ',(', F4.1, ',', F4.1,
-C     $      '), C,', I3, ')          .' )
+ 9994 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $     F4.1, ', A,', I3, ',', F4.1, ', C,', I3, ')               ',
+     $      '          .' )
+ 9993 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      '(', F4.1, ',', F4.1, ') , A,', I3, ',(', F4.1, ',', F4.1,
+     $      '), C,', I3, ')          .' )
  9992 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -2222,12 +2222,12 @@ C     $      '), C,', I3, ')          .' )
      $      'ANGED INCORRECTLY *******' )
  9996 FORMAT( ' ******* ', A12,' FAILED ON CALL NUMBER:' )
  9995 FORMAT( '      THESE ARE THE RESULTS FOR COLUMN ', I3 )
-C 9994 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',', F4.1,
-C     $      ', C,', I3, ')           .' )
-C 9993 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
-C     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',(', F4.1,
-C     $      ',', F4.1, '), C,', I3, ')    .' )
+ 9994 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',', F4.1,
+     $      ', C,', I3, ')           .' )
+ 9993 FORMAT(1X, I6, ': ', A12,'(', 2( '''', A1, ''',' ), 2( I3, ',' ),
+     $      '(', F4.1, ',', F4.1, '), A,', I3, ', B,', I3, ',(', F4.1,
+     $      ',', F4.1, '), C,', I3, ')    .' )
  9992 FORMAT(' ******* FATAL ERROR - ERROR-EXIT TAKEN ON VALID CALL *',
      $      '******' )
 *
@@ -2706,7 +2706,7 @@ C     $      ',', F4.1, '), C,', I3, ')    .' )
    50    CONTINUE
       END IF
 *
-C   60 CONTINUE
+   60 CONTINUE
       LZERES = .TRUE.
       GO TO 80
    70 CONTINUE

driver/level2/CMakeLists.txt

@@ -1,6 +1,5 @@
 
 include_directories(${PROJECT_SOURCE_DIR})
-include_directories(${PROJECT_BINARY_DIR})
 
 # sources that need to be compiled twice, once with no flags and once with LOWER
 set(UL_SOURCES
@@ -73,7 +72,7 @@ GenerateNamedObjects("zgbmv_k.c" "TRANS;CONJ;XCONJ" "gbmv_d" false "" "" "" 2)
 # special defines for complex
 foreach (float_type ${FLOAT_TYPES})
 
-  if (USE_THREAD)
+  if (SMP)
     GenerateNamedObjects("gemv_thread.c" "" "gemv_thread_n" false "" "" false ${float_type})
     GenerateNamedObjects("gemv_thread.c" "TRANSA" "gemv_thread_t" false "" "" false ${float_type})
 
@@ -107,7 +106,7 @@ foreach (float_type ${FLOAT_TYPES})
       GenerateNamedObjects("z${ulvm_source}" "LOWER;HEMVREV" "${op_name}_M" false "" "" false ${float_type})
     endforeach()
 
-    if (USE_THREAD)
+    if (SMP)
 
       GenerateNamedObjects("gemv_thread.c" "CONJ" "gemv_thread_r" false "" "" false ${float_type})
       GenerateNamedObjects("gemv_thread.c" "CONJ;TRANSA" "gemv_thread_c" false "" "" false ${float_type})
@@ -186,7 +185,7 @@ foreach (float_type ${FLOAT_TYPES})
       GenerateCombinationObjects("${l_source}" "UNIT" "N" "TRANSA" 0 "${op_name}_TU" false ${float_type})
     endforeach ()
 
-    if (USE_THREAD)
+    if (SMP)
       GenerateNamedObjects("ger_thread.c" "" "" false "" "" false ${float_type})
       foreach(nu_smp_source ${NU_SMP_SOURCES})
         string(REGEX MATCH "[a-z]+_[a-z]+" op_name ${nu_smp_source})
@@ -197,7 +196,7 @@ foreach (float_type ${FLOAT_TYPES})
   endif ()
 endforeach ()
 
-if (USE_THREAD)
+if (SMP)
   GenerateCombinationObjects("${UL_SMP_SOURCES}" "LOWER" "U" "" 2)
 endif ()
 

driver/level2/gbmv_k.c

@@ -62,13 +62,13 @@ void CNAME(BLASLONG m, BLASLONG n, BLASLONG ku, BLASLONG kl, FLOAT alpha,
   if (incy != 1) {
     Y = bufferY;
     bufferX    = (FLOAT *)(((BLASLONG)bufferY + M * sizeof(FLOAT) + 4095) & ~4095);
-    // gemvbuffer = bufferX;
+    gemvbuffer = bufferX;
     COPY_K(M, y, incy, Y, 1);
   }
 
   if (incx != 1) {
     X = bufferX;
-    // gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + N * sizeof(FLOAT) + 4095) & ~4095);
+    gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + N * sizeof(FLOAT) + 4095) & ~4095);
     COPY_K(N, x, incx, X, 1);
   }
 

driver/level2/gbmv_thread.c

@@ -96,7 +96,7 @@ static int gbmv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
     COPY_K(args -> m, x, incx, buffer, 1);
 
     x = buffer;
-    // buffer += ((COMPSIZE * args -> m  + 1023) & ~1023);
+    buffer += ((COMPSIZE * args -> m  + 1023) & ~1023);
   }
 #endif
 
@@ -230,10 +230,8 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG ku, BLASLONG kl, FLOAT *alpha, FLOAT
 
 #ifndef TRANSA
     range_m[num_cpu] = num_cpu * ((m + 15) & ~15);
-    if (range_m[num_cpu] > m * num_cpu) range_m[num_cpu] = m * num_cpu;
 #else
     range_m[num_cpu] = num_cpu * ((n + 15) & ~15);
-    if (range_m[num_cpu] > n * num_cpu) range_m[num_cpu] = n * num_cpu;
 #endif
 
     queue[num_cpu].mode    = mode;

driver/level2/sbmv_k.c

@@ -55,13 +55,13 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT alpha,
   if (incy != 1) {
     Y = bufferY;
     bufferX    = (FLOAT *)(((BLASLONG)bufferY + n * sizeof(FLOAT) + 4095) & ~4095);
-    // sbmvbuffer = bufferX;
+    sbmvbuffer = bufferX;
     COPY_K(n, y, incy, Y, 1);
   }
 
   if (incx != 1) {
     X = bufferX;
-    // sbmvbuffer = (FLOAT *)(((BLASLONG)bufferX + n * sizeof(FLOAT) + 4095) & ~4095);
+    sbmvbuffer = (FLOAT *)(((BLASLONG)bufferX + n * sizeof(FLOAT) + 4095) & ~4095);
     COPY_K(n, x, incx, X, 1);
   }
 

driver/level2/sbmv_thread.c

@@ -91,7 +91,7 @@ static int sbmv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
     COPY_K(n, x, incx, buffer, 1);
 
     x = buffer;
-    // buffer += ((COMPSIZE * n  + 1023) & ~1023);
+    buffer += ((COMPSIZE * n  + 1023) & ~1023);
   }
 
   SCAL_K(n, 0, 0, ZERO,
@@ -246,7 +246,6 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
 
       range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
-      if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = sbmv_kernel;
@@ -286,7 +285,6 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
 
       range_m[num_cpu + 1] = range_m[num_cpu] + width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
-      if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = sbmv_kernel;
@@ -318,7 +316,6 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
       range_m[num_cpu + 1] = range_m[num_cpu] + width;
 
       range_n[num_cpu] = num_cpu * ((n + 15) & ~15);
-      if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = sbmv_kernel;

driver/level2/spmv_k.c

@@ -53,13 +53,13 @@ int CNAME(BLASLONG m, FLOAT alpha, FLOAT *a,
   if (incy != 1) {
     Y = bufferY;
     bufferX    = (FLOAT *)(((BLASLONG)bufferY + m * sizeof(FLOAT) + 4095) & ~4095);
-    // gemvbuffer = bufferX;
+    gemvbuffer = bufferX;
     COPY_K(m, y, incy, Y, 1);
   }
 
   if (incx != 1) {
     X = bufferX;
-    // gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + m * sizeof(FLOAT) + 4095) & ~4095);
+    gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + m * sizeof(FLOAT) + 4095) & ~4095);
     COPY_K(m, x, incx, X, 1);
   }
 

driver/level2/spmv_thread.c

@@ -246,7 +246,6 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *y,
 
     range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m * num_cpu) range_n[num_cpu] = m * num_cpu;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = spmv_kernel;
@@ -286,7 +285,6 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *y,
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m * num_cpu) range_n[num_cpu] = m * num_cpu;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = spmv_kernel;

driver/level2/symv_thread.c

@@ -177,8 +177,7 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG i
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m * num_cpu) range_n[num_cpu] = m * num_cpu;
-    
+
     queue[MAX_CPU_NUMBER - num_cpu - 1].mode    = mode;
     queue[MAX_CPU_NUMBER - num_cpu - 1].routine = symv_kernel;
     queue[MAX_CPU_NUMBER - num_cpu - 1].args    = &args;
@@ -226,7 +225,6 @@ int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG i
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m * num_cpu) range_n[num_cpu] = m * num_cpu;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = symv_kernel;

driver/level2/tbmv_L.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/tbmv_U.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/tbmv_thread.c

@@ -107,7 +107,7 @@ static int trmv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
     COPY_K(args -> n, x, incx, buffer, 1);
 
     x = buffer;
-    // buffer += ((args -> n * COMPSIZE + 1023) & ~1023);
+    buffer += ((args -> n * COMPSIZE + 1023) & ~1023);
   }
 
   if (range_n) y += *range_n * COMPSIZE;
@@ -288,7 +288,6 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc
 
       range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
-      if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = trmv_kernel;
@@ -328,7 +327,6 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc
 
       range_m[num_cpu + 1] = range_m[num_cpu] + width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
-      if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = trmv_kernel;
@@ -358,7 +356,6 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc
 
       range_m[num_cpu + 1] = range_m[num_cpu] + width;
       range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
-      if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
 
       queue[num_cpu].mode    = mode;
       queue[num_cpu].routine = trmv_kernel;

driver/level2/tbsv_L.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/tbsv_U.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/tpmv_L.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG m, FLOAT *a, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/tpmv_U.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG m, FLOAT *a, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/tpmv_thread.c

@@ -112,7 +112,7 @@ static int tpmv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
 #endif
 
     x = buffer;
-    // buffer += ((COMPSIZE * args -> m + 1023) & ~1023);
+    buffer += ((COMPSIZE * args -> m + 1023) & ~1023);
   }
 
 #ifndef TRANS
@@ -234,7 +234,11 @@ static int tpmv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
   return 0;
 }
 
+#ifndef COMPLEX
 int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthreads){
+#else
+int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthreads){
+#endif
 
   blas_arg_t args;
   blas_queue_t queue[MAX_CPU_NUMBER];
@@ -303,8 +307,7 @@ int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthr
 
     range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m * num_cpu) range_n[num_cpu] = m * num_cpu;
-    
+
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = tpmv_kernel;
     queue[num_cpu].args    = &args;
@@ -343,7 +346,6 @@ int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthr
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m * num_cpu) range_n[num_cpu] = m * num_cpu;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = tpmv_kernel;

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 */
-/* 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){
 
-  for (is = 0; is < m; is += DTB_ENTRIES * 100){
-
-    min_i = MIN(m - is, DTB_ENTRIES * 100);
+    min_i = MIN(m - is, DTB_ENTRIES);
 
 #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/level2/trmv_thread.c

@@ -346,7 +346,6 @@ int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *bu
 
     range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = trmv_kernel;
@@ -386,7 +385,6 @@ int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *bu
 
     range_m[num_cpu + 1] = range_m[num_cpu] + width;
     range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
-    if (range_n[num_cpu] > m) range_n[num_cpu] = m;
 
     queue[num_cpu].mode    = mode;
     queue[num_cpu].routine = trmv_kernel;

driver/level2/zgbmv_k.c

@@ -83,13 +83,13 @@ void CNAME(BLASLONG m, BLASLONG n, BLASLONG ku, BLASLONG kl, FLOAT alpha_r, FLOA
   if (incy != 1) {
     Y = bufferY;
     bufferX    = (FLOAT *)(((BLASLONG)bufferY + M * sizeof(FLOAT) * 2 + 4095) & ~4095);
-    // gemvbuffer = bufferX;
+    gemvbuffer = bufferX;
     COPY_K(M, y, incy, Y, 1);
   }
 
   if (incx != 1) {
     X = bufferX;
-    // gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + N * sizeof(FLOAT) * 2 + 4095) & ~4095);
+    gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + N * sizeof(FLOAT) * 2 + 4095) & ~4095);
     COPY_K(N, x, incx, X, 1);
   }
 

driver/level2/zhbmv_k.c

@@ -61,13 +61,13 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT alpha_r, FLOAT alpha_i,
   if (incy != 1) {
     Y = bufferY;
     bufferX    = (FLOAT *)(((BLASLONG)bufferY + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
-    // sbmvbuffer = bufferX;
+    sbmvbuffer = bufferX;
     COPY_K(n, y, incy, Y, 1);
   }
 
   if (incx != 1) {
     X = bufferX;
-    // sbmvbuffer = (FLOAT *)(((BLASLONG)bufferX + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
+    sbmvbuffer = (FLOAT *)(((BLASLONG)bufferX + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
     COPY_K(n, x, incx, X, 1);
   }
 

driver/level2/zhpmv_k.c

@@ -56,13 +56,13 @@ int CNAME(BLASLONG m, FLOAT alpha_r, FLOAT alpha_i,
   if (incy != 1) {
     Y = bufferY;
     bufferX    = (FLOAT *)(((BLASLONG)bufferY + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
-    // gemvbuffer = bufferX;
+    gemvbuffer = bufferX;
     COPY_K(m, y, incy, Y, 1);
   }
 
   if (incx != 1) {
     X = bufferX;
-    // gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
+    gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
     COPY_K(m, x, incx, X, 1);
   }
 

driver/level2/zsbmv_k.c

@@ -60,13 +60,13 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT alpha_r, FLOAT alpha_i,
   if (incy != 1) {
     Y = bufferY;
     bufferX    = (FLOAT *)(((BLASLONG)bufferY + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
-    // sbmvbuffer = bufferX;
+    sbmvbuffer = bufferX;
     COPY_K(n, y, incy, Y, 1);
   }
 
   if (incx != 1) {
     X = bufferX;
-    // sbmvbuffer = (FLOAT *)(((BLASLONG)bufferX + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
+    sbmvbuffer = (FLOAT *)(((BLASLONG)bufferX + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
     COPY_K(n, x, incx, X, 1);
   }
 

driver/level2/zspmv_k.c

@@ -55,13 +55,13 @@ int CNAME(BLASLONG m, FLOAT alpha_r, FLOAT alpha_i,
   if (incy != 1) {
     Y = bufferY;
     bufferX    = (FLOAT *)(((BLASLONG)bufferY + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
-    // gemvbuffer = bufferX;
+    gemvbuffer = bufferX;
     COPY_K(m, y, incy, Y, 1);
   }
 
   if (incx != 1) {
     X = bufferX;
-    // gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
+    gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
     COPY_K(m, x, incx, X, 1);
   }
 

driver/level2/ztbmv_L.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/ztbmv_U.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/ztbsv_L.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/ztbsv_U.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dp1 = 1.;
+const static FLOAT dp1 = 1.;
 
 int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level2/ztpsv_L.c

@@ -40,7 +40,7 @@
 #include <ctype.h>
 #include "common.h"
 
-// const static FLOAT dm1 = -1.;
+const static FLOAT dm1 = -1.;
 
 int CNAME(BLASLONG m, FLOAT *a, FLOAT *b, BLASLONG incb, void *buffer){
 

driver/level3/CMakeLists.txt

@@ -1,5 +1,4 @@
 include_directories(${PROJECT_SOURCE_DIR})
-include_directories(${PROJECT_BINARY_DIR})
 
 # N.B. In the original makefile there was a BLOCKS define used in the compilation of these files but I don't see any evidence of it being set anywhere. -hpa
 
@@ -9,7 +8,7 @@ set(GEMM_COMPLEX_DEFINES RN CN RT CT NR TR RR CR NC TC RC CC)
 foreach (GEMM_DEFINE ${GEMM_DEFINES})
   string(TOLOWER ${GEMM_DEFINE} GEMM_DEFINE_LC)
   GenerateNamedObjects("gemm.c" "${GEMM_DEFINE}" "gemm_${GEMM_DEFINE_LC}" 0)
-  if (USE_THREAD AND NOT USE_SIMPLE_THREADED_LEVEL3)
+  if (SMP AND NOT USE_SIMPLE_THREADED_LEVEL3)
     GenerateNamedObjects("gemm.c" "${GEMM_DEFINE};THREADED_LEVEL3" "gemm_thread_${GEMM_DEFINE_LC}" 0)
   endif ()
 endforeach ()
@@ -32,7 +31,7 @@ GenerateCombinationObjects("syrk_k.c" "LOWER;TRANS" "U;N" "" 1)
 GenerateCombinationObjects("syr2k_k.c" "LOWER;TRANS" "U;N" "" 1)
 GenerateCombinationObjects("syrk_kernel.c" "LOWER" "U" "" 2)
 GenerateCombinationObjects("syr2k_kernel.c" "LOWER" "U" "" 2)
-if (USE_THREAD)
+if (SMP)
 
   # N.B. these do NOT have a float type (e.g. DOUBLE) defined!
   GenerateNamedObjects("gemm_thread_m.c;gemm_thread_n.c;gemm_thread_mn.c;gemm_thread_variable.c;syrk_thread.c" "" "" 0 "" "" 1)
@@ -71,7 +70,7 @@ foreach (float_type ${FLOAT_TYPES})
     GenerateNamedObjects("zher2k_k.c" "HER2K;LOWER" "her2k_LN" false "" "" false ${float_type})
     GenerateNamedObjects("zher2k_k.c" "HER2K;LOWER;TRANS;CONJ" "her2k_LC" false "" "" false ${float_type})
 
-    if (USE_THREAD AND NOT USE_SIMPLE_THREADED_LEVEL3)
+    if (SMP AND NOT USE_SIMPLE_THREADED_LEVEL3)
       #herk
       GenerateCombinationObjects("zherk_k.c" "LOWER" "U" "HERK;THREADED_LEVEL3" 3 "herk_thread_N" false ${float_type})
       GenerateCombinationObjects("zherk_k.c" "LOWER" "U" "HERK;THREADED_LEVEL3;TRANS;CONJ" 3 "herk_thread_C" false ${float_type})
@@ -93,7 +92,7 @@ foreach (float_type ${FLOAT_TYPES})
       if(USE_GEMM3M)
 	GenerateNamedObjects("gemm3m.c" "${gemm_define}" "gemm3m_${gemm_define_LC}" false "" "" false ${float_type})
       endif()
-      if (USE_THREAD AND NOT USE_SIMPLE_THREADED_LEVEL3)
+      if (SMP AND NOT USE_SIMPLE_THREADED_LEVEL3)
         GenerateNamedObjects("gemm.c" "${gemm_define};THREADED_LEVEL3" "gemm_thread_${gemm_define_LC}" false "" "" false ${float_type})
 	if(USE_GEMM3M)
 	  GenerateNamedObjects("gemm3m.c" "${gemm_define};THREADED_LEVEL3" "gemm3m_thread_${gemm_define_LC}" false "" "" false ${float_type})
@@ -106,7 +105,7 @@ foreach (float_type ${FLOAT_TYPES})
       foreach (GEMM_DEFINE ${GEMM_DEFINES})
 	string(TOLOWER ${GEMM_DEFINE} GEMM_DEFINE_LC)
 	GenerateNamedObjects("gemm3m.c" "${GEMM_DEFINE}" "gemm3m_${GEMM_DEFINE_LC}" false "" "" false ${float_type})
-	if (USE_THREAD AND NOT USE_SIMPLE_THREADED_LEVEL3)
+	if (SMP AND NOT USE_SIMPLE_THREADED_LEVEL3)
 	  GenerateNamedObjects("gemm3m.c" "${GEMM_DEFINE};THREADED_LEVEL3" "gemm3m_thread_${GEMM_DEFINE_LC}" false "" "" false ${float_type})
 	endif ()
       endforeach ()

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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DXDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DDOUBLE -DCOMPLEX -DCR $< -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=CR $< -o $(@F)
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DXDOUBLE -DCOMPLEX -DCR $< -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.c

@@ -251,11 +251,11 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
   if ((k == 0) || (alpha == NULL)) return 0;
 
 #if !defined(XDOUBLE) || !defined(QUAD_PRECISION)
-  if ( alpha[0] == ZERO
+  if ((alpha[0] == ZERO)
 #ifdef COMPLEX
-      && alpha[1] == ZERO
+      && (alpha[1] == ZERO)
 #endif
-	 ) return 0; 
+      ) return 0;
 #else
   if (((alpha[0].x[0] | alpha[0].x[1]
 #ifdef COMPLEX
@@ -293,7 +293,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
       min_l = k - ls;
 
       if (min_l >= GEMM_Q * 2) {
-	// gemm_p = GEMM_P;
+	gemm_p = GEMM_P;
 	min_l  = GEMM_Q;
       } else {
 	if (min_l > GEMM_Q) {

driver/level3/level3_gemm3m_thread.c

@@ -91,12 +91,7 @@
 #endif
 
 typedef struct {
-#if __STDC_VERSION__ >= 201112L
-  _Atomic
-#else
-  volatile
-#endif  
-   BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
+  volatile BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
 } job_t;
 
 
@@ -979,7 +974,7 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
 int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){
 
   BLASLONG m = args -> m;
-  // BLASLONG n = args -> n;
+  BLASLONG n = args -> n;
   BLASLONG nthreads = args -> nthreads;
   BLASLONG divN, divT;
   int mode;
@@ -990,14 +985,13 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
     m = m_to - m_from;
   }
-/*
+
   if (range_n) {
     BLASLONG n_from = *(((BLASLONG *)range_n) + 0);
     BLASLONG n_to   = *(((BLASLONG *)range_n) + 1);
 
     n = n_to - n_from;
   }
-*/
 
   if ((args -> m < nthreads * SWITCH_RATIO) || (args -> n < nthreads * SWITCH_RATIO)) {
     GEMM3M_LOCAL(args, range_m, range_n, sa, sb, 0);

driver/level3/level3_syr2k.c

@@ -154,9 +154,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
   if ((k == 0) || (alpha == NULL)) return 0;
 
-  if (alpha[0] == ZERO
+  if ((alpha[0] == ZERO)
 #ifdef COMPLEX
-      && alpha[1] == ZERO
+      && (alpha[1] == ZERO)
 #endif
       ) return 0;
 

driver/level3/level3_syrk.c

@@ -158,9 +158,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
   if ((k == 0) || (alpha == NULL)) return 0;
 
-  if (alpha[0] == ZERO
+  if ((alpha[0] == ZERO)
 #if defined(COMPLEX) && !defined(HERK)
-      && alpha[1] == ZERO
+      && (alpha[1] == ZERO)
 #endif
       ) return 0;
 

driver/level3/level3_syrk_threaded.c

@@ -67,12 +67,7 @@
 #endif
 
 typedef struct {
-#if __STDC_VERSION__ >= 201112L
-_Atomic
-#else 
-  volatile
-#endif
-   BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
+  volatile BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
 } job_t;
 
 
@@ -205,9 +200,9 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
   if ((k == 0) || (alpha == NULL)) return 0;
 
-  if (alpha[0] == ZERO
+  if ((alpha[0] == ZERO)
 #if defined(COMPLEX) && !defined(HERK)
-      && alpha[1] == ZERO
+      && (alpha[1] == ZERO)
 #endif
       ) return 0;
 

driver/level3/level3_thread.c

@@ -91,28 +91,27 @@
 #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;
 
 
 #ifndef BETA_OPERATION
 #ifndef COMPLEX
-#define BETA_OPERATION(M_FROM, M_TO, N_FROM, N_TO, BETA, C, LDC)        \
-  GEMM_BETA((M_TO) - (M_FROM), (N_TO - N_FROM), 0,                      \
-            BETA[0], NULL, 0, NULL, 0,                                  \
-            (FLOAT *)(C) + ((M_FROM) + (N_FROM) * (LDC)) * COMPSIZE, LDC)
+#define BETA_OPERATION(M_FROM, M_TO, N_FROM, N_TO, BETA, C, LDC) \
+	GEMM_BETA((M_TO) - (M_FROM), (N_TO - N_FROM), 0, \
+		  BETA[0], NULL, 0, NULL, 0, \
+		  (FLOAT *)(C) + ((M_FROM) + (N_FROM) * (LDC)) * COMPSIZE, LDC)
 #else
-#define BETA_OPERATION(M_FROM, M_TO, N_FROM, N_TO, BETA, C, LDC)        \
-  GEMM_BETA((M_TO) - (M_FROM), (N_TO - N_FROM), 0,                      \
-            BETA[0], BETA[1], NULL, 0, NULL, 0,                         \
-            (FLOAT *)(C) + ((M_FROM) + (N_FROM) * (LDC)) * COMPSIZE, LDC)
+#define BETA_OPERATION(M_FROM, M_TO, N_FROM, N_TO, BETA, C, LDC) \
+	GEMM_BETA((M_TO) - (M_FROM), (N_TO - N_FROM), 0, \
+		  BETA[0], BETA[1], NULL, 0, NULL, 0, \
+		  (FLOAT *)(C) + ((M_FROM) + (N_FROM) * (LDC)) * COMPSIZE, LDC)
 #endif
 #endif
 
 #ifndef ICOPY_OPERATION
 #if defined(NN) || defined(NT) || defined(NC) || defined(NR) || \
-  defined(RN) || defined(RT) || defined(RC) || defined(RR)
+    defined(RN) || defined(RT) || defined(RC) || defined(RR)
 #define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
 #define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
@@ -121,7 +120,7 @@ typedef struct {
 
 #ifndef OCOPY_OPERATION
 #if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
-  defined(NR) || defined(TR) || defined(CR) || defined(RR)
+    defined(NR) || defined(TR) || defined(CR) || defined(RR)
 #define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
 #define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
@@ -145,36 +144,36 @@ typedef struct {
 
 #ifndef KERNEL_OPERATION
 #ifndef COMPLEX
-#define KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, C, LDC, X, Y)          \
-  KERNEL_FUNC(M, N, K, ALPHA[0], SA, SB, (FLOAT *)(C) + ((X) + (Y) * LDC) * COMPSIZE, LDC)
+#define KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, C, LDC, X, Y) \
+	KERNEL_FUNC(M, N, K, ALPHA[0], SA, SB, (FLOAT *)(C) + ((X) + (Y) * LDC) * COMPSIZE, LDC)
 #else
-#define KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, C, LDC, X, Y)          \
-  KERNEL_FUNC(M, N, K, ALPHA[0], ALPHA[1], SA, SB, (FLOAT *)(C) + ((X) + (Y) * LDC) * COMPSIZE, LDC)
+#define KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, C, LDC, X, Y) \
+	KERNEL_FUNC(M, N, K, ALPHA[0], ALPHA[1], SA, SB, (FLOAT *)(C) + ((X) + (Y) * LDC) * COMPSIZE, LDC)
 #endif
 #endif
 
 #ifndef FUSED_KERNEL_OPERATION
 #if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
-  defined(NR) || defined(TR) || defined(CR) || defined(RR)
+    defined(NR) || defined(TR) || defined(CR) || defined(RR)
 #ifndef COMPLEX
 #define FUSED_KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, B, LDB, C, LDC, I, J, L) \
-  FUSED_GEMM_KERNEL_N(M, N, K, ALPHA[0], SA, SB,                        \
-                      (FLOAT *)(B) + ((L) + (J) * LDB) * COMPSIZE, LDB, (FLOAT *)(C) + ((I) + (J) * LDC) * COMPSIZE, LDC)
+	FUSED_GEMM_KERNEL_N(M, N, K, ALPHA[0], SA, SB, \
+	(FLOAT *)(B) + ((L) + (J) * LDB) * COMPSIZE, LDB, (FLOAT *)(C) + ((I) + (J) * LDC) * COMPSIZE, LDC)
 #else
 #define FUSED_KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, B, LDB, C, LDC, I, J, L) \
-  FUSED_GEMM_KERNEL_N(M, N, K, ALPHA[0], ALPHA[1], SA, SB,              \
-                      (FLOAT *)(B) + ((L) + (J) * LDB) * COMPSIZE, LDB, (FLOAT *)(C) + ((I) + (J) * LDC) * COMPSIZE, LDC)
+	FUSED_GEMM_KERNEL_N(M, N, K, ALPHA[0], ALPHA[1], SA, SB, \
+	(FLOAT *)(B) + ((L) + (J) * LDB) * COMPSIZE, LDB, (FLOAT *)(C) + ((I) + (J) * LDC) * COMPSIZE, LDC)
 
 #endif
 #else
 #ifndef COMPLEX
 #define FUSED_KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, B, LDB, C, LDC, I, J, L) \
-  FUSED_GEMM_KERNEL_T(M, N, K, ALPHA[0], SA, SB,                        \
-                      (FLOAT *)(B) + ((J) + (L) * LDB) * COMPSIZE, LDB, (FLOAT *)(C) + ((I) + (J) * LDC) * COMPSIZE, LDC)
+	FUSED_GEMM_KERNEL_T(M, N, K, ALPHA[0], SA, SB, \
+	(FLOAT *)(B) + ((J) + (L) * LDB) * COMPSIZE, LDB, (FLOAT *)(C) + ((I) + (J) * LDC) * COMPSIZE, LDC)
 #else
 #define FUSED_KERNEL_OPERATION(M, N, K, ALPHA, SA, SB, B, LDB, C, LDC, I, J, L) \
-  FUSED_GEMM_KERNEL_T(M, N, K, ALPHA[0], ALPHA[1], SA, SB,              \
-                      (FLOAT *)(B) + ((J) + (L) * LDB) * COMPSIZE, LDB, (FLOAT *)(C) + ((I) + (J) * LDC) * COMPSIZE, LDC)
+	FUSED_GEMM_KERNEL_T(M, N, K, ALPHA[0], ALPHA[1], SA, SB, \
+	(FLOAT *)(B) + ((J) + (L) * LDB) * COMPSIZE, LDB, (FLOAT *)(C) + ((I) + (J) * LDC) * COMPSIZE, LDC)
 #endif
 #endif
 #endif
@@ -220,17 +219,15 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
   FLOAT *buffer[DIVIDE_RATE];
 
   BLASLONG k, lda, ldb, ldc;
-  BLASLONG m_from, m_to, n_from, n_to;
+  BLASLONG m_from, m_to, n_from, n_to, N_from, N_to;
 
   FLOAT *alpha, *beta;
   FLOAT *a, *b, *c;
   job_t *job = (job_t *)args -> common;
+  BLASLONG xxx, bufferside;
 
-  BLASLONG nthreads_m;
-  BLASLONG mypos_m, mypos_n;
-
-  BLASLONG is, js, ls, bufferside, jjs;
-  BLASLONG min_i, min_l, div_n, min_jj;
+  BLASLONG ls, min_l, jjs, min_jj;
+  BLASLONG is, min_i, div_n;
 
   BLASLONG i, current;
   BLASLONG l1stride;
@@ -262,69 +259,74 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
   alpha = (FLOAT *)args -> alpha;
   beta  = (FLOAT *)args -> beta;
 
-  /* Initialize 2D CPU distribution */
-  nthreads_m = args -> nthreads;
-  if (range_m) {
-    nthreads_m = range_m[-1];
-  }
-  mypos_n = blas_quickdivide(mypos, nthreads_m);  /* mypos_n = mypos / nthreads_m */
-  mypos_m = mypos - mypos_n * nthreads_m;         /* mypos_m = mypos % nthreads_m */
-
-  /* Initialize m and n */
   m_from = 0;
   m_to   = M;
+
   if (range_m) {
-    m_from = range_m[mypos_m + 0];
-    m_to   = range_m[mypos_m + 1];
+    m_from = range_m[0];
+    m_to   = range_m[1];
   }
+
   n_from = 0;
   n_to   = N;
+
+  N_from = 0;
+  N_to   = N;
+
   if (range_n) {
     n_from = range_n[mypos + 0];
     n_to   = range_n[mypos + 1];
+
+    N_from = range_n[0];
+    N_to   = range_n[args -> nthreads];
   }
 
-  /* Multiply C by beta if needed */
   if (beta) {
 #ifndef COMPLEX
     if (beta[0] != ONE)
 #else
     if ((beta[0] != ONE) || (beta[1] != ZERO))
 #endif
-      BETA_OPERATION(m_from, m_to, range_n[mypos_n * nthreads_m], range_n[(mypos_n + 1) * nthreads_m], beta, c, ldc);
+      BETA_OPERATION(m_from, m_to, N_from, N_to, beta, c, ldc);
   }
 
-  /* Return early if no more computation is needed */
   if ((k == 0) || (alpha == NULL)) return 0;
-  if (alpha[0] == ZERO
+
+  if ((alpha[0] == ZERO)
 #ifdef COMPLEX
-      && alpha[1] == ZERO
+      && (alpha[1] == ZERO)
 #endif
       ) return 0;
 
-  /* Initialize workspace for local region of B */
+#if 0
+  fprintf(stderr, "Thread[%ld]  m_from : %ld m_to : %ld n_from : %ld n_to : %ld N_from : %ld N_to : %ld\n",
+	  mypos, m_from, m_to, n_from, n_to, N_from, N_to);
+
+  fprintf(stderr, "GEMM: P = %4ld  Q = %4ld  R = %4ld\n", (BLASLONG)GEMM_P, (BLASLONG)GEMM_Q, (BLASLONG)GEMM_R);
+
+#endif
+
   div_n = (n_to - n_from + DIVIDE_RATE - 1) / DIVIDE_RATE;
+
   buffer[0] = sb;
   for (i = 1; i < DIVIDE_RATE; i++) {
     buffer[i] = buffer[i - 1] + GEMM_Q * ((div_n + GEMM_UNROLL_N - 1)/GEMM_UNROLL_N) * GEMM_UNROLL_N * COMPSIZE;
   }
 
-  /* Iterate through steps of k */
+
   for(ls = 0; ls < k; ls += min_l){
 
-    /* Determine step size in k */
     min_l = k - ls;
+
     if (min_l >= GEMM_Q * 2) {
       min_l  = GEMM_Q;
     } else {
       if (min_l > GEMM_Q) min_l = (min_l + 1) / 2;
     }
 
-    /* Determine step size in m
-     * Note: We are currently on the first step in m
-     */
     l1stride = 1;
     min_i = m_to - m_from;
+
     if (min_i >= GEMM_P * 2) {
       min_i = GEMM_P;
     } else {
@@ -335,107 +337,108 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
       }
     }
 
-    /* Copy local region of A into workspace */
     START_RPCC();
+
     ICOPY_OPERATION(min_l, min_i, a, lda, ls, m_from, sa);
+
     STOP_RPCC(copy_A);
 
-    /* Copy local region of B into workspace and apply kernel */
     div_n = (n_to - n_from + DIVIDE_RATE - 1) / DIVIDE_RATE;
-    for (js = n_from, bufferside = 0; js < n_to; js += div_n, bufferside ++) {
 
-      /* Make sure if no one is using workspace */
+    for (xxx = n_from, bufferside = 0; xxx < n_to; xxx += div_n, bufferside ++) {
+
       START_RPCC();
+
+      /* Make sure if no one is using buffer */
       for (i = 0; i < args -> nthreads; i++)
-	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;MB;};
+	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;};
+
       STOP_RPCC(waiting1);
 
 #if defined(FUSED_GEMM) && !defined(TIMING)
 
-      /* Fused operation to copy region of B into workspace and apply kernel */
-      FUSED_KERNEL_OPERATION(min_i, MIN(n_to, js + div_n) - js, min_l, alpha,
-			     sa, buffer[bufferside], b, ldb, c, ldc, m_from, js, ls);
+      FUSED_KERNEL_OPERATION(min_i, MIN(n_to, xxx + div_n) - xxx, min_l, alpha,
+			     sa, buffer[bufferside], b, ldb, c, ldc, m_from, xxx, ls);
 
 #else
 
-      /* Split local region of B into parts */
-      for(jjs = js; jjs < MIN(n_to, js + div_n); jjs += min_jj){
-	min_jj = MIN(n_to, js + div_n) - jjs;
+      for(jjs = xxx; jjs < MIN(n_to, xxx + div_n); jjs += min_jj){
+	min_jj = MIN(n_to, xxx + div_n) - jjs;
+
 	if (min_jj >= 3*GEMM_UNROLL_N) min_jj = 3*GEMM_UNROLL_N;
 	else
-          if (min_jj >= 2*GEMM_UNROLL_N) min_jj = 2*GEMM_UNROLL_N;
-          else
-            if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+		if (min_jj >= 2*GEMM_UNROLL_N) min_jj = 2*GEMM_UNROLL_N;
+		else
+			if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
+
 
-        /* Copy part of local region of B into workspace */
 	START_RPCC();
+
 	OCOPY_OPERATION(min_l, min_jj, b, ldb, ls, jjs,
-			buffer[bufferside] + min_l * (jjs - js) * COMPSIZE * l1stride);
+			buffer[bufferside] + min_l * (jjs - xxx) * COMPSIZE * l1stride);
+
 	STOP_RPCC(copy_B);
 
-        /* Apply kernel with local region of A and part of local region of B */
 	START_RPCC();
+
 	KERNEL_OPERATION(min_i, min_jj, min_l, alpha,
-			 sa, buffer[bufferside] + min_l * (jjs - js) * COMPSIZE * l1stride,
+			 sa, buffer[bufferside] + min_l * (jjs - xxx) * COMPSIZE * l1stride,
 			 c, ldc, m_from, jjs);
+
 	STOP_RPCC(kernel);
 
 #ifdef TIMING
-        ops += 2 * min_i * min_jj * min_l;
+	  ops += 2 * min_i * min_jj * min_l;
 #endif
 
       }
 #endif
 
-      /* Set flag so other threads can access local region of B */
-      for (i = mypos_n * nthreads_m; i < (mypos_n + 1) * nthreads_m; i++)
-        job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
+      for (i = 0; i < args -> nthreads; i++) job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
       WMB;
     }
 
-    /* Get regions of B from other threads and apply kernel */
     current = mypos;
+
     do {
-
-      /* This thread accesses regions of B from threads in the range
-       * [ mypos_n * nthreads_m, (mypos_n+1) * nthreads_m ) */
       current ++;
-      if (current >= (mypos_n + 1) * nthreads_m) current = mypos_n * nthreads_m;
+      if (current >= args -> nthreads) current = 0;
 
-      /* Split other region of B into parts */
       div_n = (range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE;
-      for (js = range_n[current], bufferside = 0; js < range_n[current + 1]; js += div_n, bufferside ++) {
-        if (current != mypos) {
 
-	  /* Wait until other region of B is initialized */
+      for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {
+
+	if (current != mypos) {
+
 	  START_RPCC();
-	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;MB;};
+
+	  /* thread has to wait */
+	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;};
+
 	  STOP_RPCC(waiting2);
 
-          /* Apply kernel with local region of A and part of other region of B */
 	  START_RPCC();
-	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1]  - js,  div_n), min_l, alpha,
-			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
-			   c, ldc, m_from, js);
-          STOP_RPCC(kernel);
 
+	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1]  - xxx,  div_n), min_l, alpha,
+			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
+			   c, ldc, m_from, xxx);
+
+	STOP_RPCC(kernel);
 #ifdef TIMING
-	  ops += 2 * min_i * MIN(range_n[current + 1]  - js,  div_n) * min_l;
+	  ops += 2 * min_i * MIN(range_n[current + 1]  - xxx,  div_n) * min_l;
 #endif
 	}
 
-        /* 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);
 
-    /* Iterate through steps of m 
-     * Note: First step has already been finished */
+
     for(is = m_from + min_i; is < m_to; is += min_i){
       min_i = m_to - is;
+
       if (min_i >= GEMM_P * 2) {
 	min_i = GEMM_P;
       } else
@@ -443,41 +446,40 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 	  min_i = (((min_i + 1) / 2 + GEMM_UNROLL_M - 1)/GEMM_UNROLL_M) * GEMM_UNROLL_M;
 	}
 
-      /* Copy local region of A into workspace */
       START_RPCC();
+
       ICOPY_OPERATION(min_l, min_i, a, lda, ls, is, sa);
+
       STOP_RPCC(copy_A);
 
-      /* Get regions of B and apply kernel */
       current = mypos;
       do {
 
-        /* Split region of B into parts and apply kernel */
 	div_n = (range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE;
-	for (js = range_n[current], bufferside = 0; js < range_n[current + 1]; js += div_n, bufferside ++) {
 
-          /* Apply kernel with local region of A and part of region of B */
+	for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {
+
 	  START_RPCC();
-	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - js, div_n), min_l, alpha,
+
+	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - xxx, div_n), min_l, alpha,
 			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
-			   c, ldc, is, js);
-          STOP_RPCC(kernel);
-          
+			   c, ldc, is, xxx);
+
+	STOP_RPCC(kernel);
+
 #ifdef TIMING
-          ops += 2 * min_i * MIN(range_n[current + 1]  - js, div_n) * min_l;
+	ops += 2 * min_i * MIN(range_n[current + 1]  - xxx, div_n) * min_l;
 #endif
-          
-          /* Clear synchronization flag if this thread is done with region of B */
-          if (is + min_i >= m_to) {
-            job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
-            WMB;
-          }
+
+	if (is + min_i >= m_to) {
+	  /* Thread doesn't need this buffer any more */
+	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
+	  WMB;
+	}
 	}
 
-        /* This thread accesses regions of B from threads in the range
-         * [ mypos_n * nthreads_m, (mypos_n+1) * nthreads_m ) */
 	current ++;
-	if (current >= (mypos_n + 1) * nthreads_m) current = mypos_n * nthreads_m;
+	if (current >= args -> nthreads) current = 0;
 
       } while (current != mypos);
 
@@ -485,13 +487,14 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
   }
 
-  /* Wait until all other threads are done with local region of B */
   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;MB;};
+    for (xxx = 0; xxx < DIVIDE_RATE; xxx++) {
+      while (job[mypos].working[i][CACHE_LINE_SIZE * xxx] ) {YIELDING;};
     }
   }
+
   STOP_RPCC(waiting3);
 
 #ifdef TIMING
@@ -504,6 +507,17 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 	  (double)waiting2 /(double)total * 100.,
 	  (double)waiting3 /(double)total * 100.,
 	  (double)ops/(double)kernel / 4. * 100.);
+
+#if 0
+  fprintf(stderr, "GEMM   [%2ld] Copy_A : %6.2ld  Copy_B : %6.2ld  Wait : %6.2ld\n",
+	  mypos, copy_A, copy_B, waiting);
+
+  fprintf(stderr, "Waiting[%2ld] %6.2f %6.2f %6.2f\n",
+	  mypos,
+	  (double)waiting1/(double)waiting * 100.,
+	  (double)waiting2/(double)waiting * 100.,
+	  (double)waiting3/(double)waiting * 100.);
+#endif
   fprintf(stderr, "\n");
 #endif
 
@@ -511,8 +525,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 }
 
 static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
-		       *range_n, FLOAT *sa, FLOAT *sb,
-                       BLASLONG nthreads_m, BLASLONG nthreads_n) {
+		       *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){
 
   blas_arg_t newarg;
 
@@ -524,18 +537,17 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
 
   blas_queue_t queue[MAX_CPU_NUMBER];
 
-  BLASLONG range_M_buffer[MAX_CPU_NUMBER + 2];
-  BLASLONG range_N_buffer[MAX_CPU_NUMBER + 2];
-  BLASLONG *range_M, *range_N;
-  BLASLONG num_parts;
+  BLASLONG range_M[MAX_CPU_NUMBER + 1];
+  BLASLONG range_N[MAX_CPU_NUMBER + 1];
+
+  BLASLONG num_cpu_m, num_cpu_n;
 
   BLASLONG nthreads = args -> nthreads;
 
   BLASLONG width, i, j, k, js;
   BLASLONG m, n, n_from, n_to;
-  int mode;
+  int  mode;
 
-  /* Get execution mode */
 #ifndef COMPLEX
 #ifdef XDOUBLE
   mode  =  BLAS_XDOUBLE | BLAS_REAL | BLAS_NODE;
@@ -554,16 +566,6 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
 #endif
 #endif
 
-#ifdef USE_ALLOC_HEAP
-  /* Dynamically allocate workspace */
-  job = (job_t*)malloc(MAX_CPU_NUMBER * sizeof(job_t));
-  if(job==NULL){
-    fprintf(stderr, "OpenBLAS: malloc failed in %s\n", __func__);
-    exit(1);
-  }
-#endif
-
-  /* Initialize struct for arguments */
   newarg.m        = args -> m;
   newarg.n        = args -> n;
   newarg.k        = args -> k;
@@ -576,19 +578,23 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
   newarg.alpha    = args -> alpha;
   newarg.beta     = args -> beta;
   newarg.nthreads = args -> nthreads;
-  newarg.common   = (void *)job;
-#ifdef PARAMTEST
-  newarg.gemm_p   = args -> gemm_p;
-  newarg.gemm_q   = args -> gemm_q;
-  newarg.gemm_r   = args -> gemm_r;
+
+#ifdef USE_ALLOC_HEAP
+  job = (job_t*)malloc(MAX_CPU_NUMBER * sizeof(job_t));
+  if(job==NULL){
+    fprintf(stderr, "OpenBLAS: malloc failed in %s\n", __func__);
+    exit(1);
+  }
+#endif
+
+  newarg.common   = (void *)job;
+
+#ifdef PARAMTEST
+  newarg.gemm_p  = args -> gemm_p;
+  newarg.gemm_q  = args -> gemm_q;
+  newarg.gemm_r  = args -> gemm_r;
 #endif
 
-  /* Initialize partitions in m and n
-   * Note: The number of CPU partitions is stored in the -1 entry */
-  range_M = &range_M_buffer[1];
-  range_N = &range_N_buffer[1];
-  range_M[-1] = nthreads_m;
-  range_N[-1] = nthreads_n;
   if (!range_m) {
     range_M[0] = 0;
     m          = args -> m;
@@ -597,35 +603,34 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
     m          = range_m[1] - range_m[0];
   }
 
-  /* Partition m into nthreads_m regions */
-  num_parts = 0;
+  num_cpu_m  = 0;
+
   while (m > 0){
-    width = blas_quickdivide(m + nthreads_m - num_parts - 1, nthreads_m - num_parts);
+
+    width  = blas_quickdivide(m + nthreads - num_cpu_m - 1, nthreads - num_cpu_m);
+
     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++) {
-    range_M[i + 1] = range_M[num_parts];
+
+    range_M[num_cpu_m + 1] = range_M[num_cpu_m] + width;
+
+    num_cpu_m ++;
   }
 
-  /* Initialize parameters for parallel execution */
-  for (i = 0; i < nthreads; i++) {
+  for (i = 0; i < num_cpu_m; i++) {
     queue[i].mode    = mode;
     queue[i].routine = inner_thread;
     queue[i].args    = &newarg;
-    queue[i].range_m = range_M;
-    queue[i].range_n = range_N;
+    queue[i].range_m = &range_M[i];
+    queue[i].range_n = &range_N[0];
     queue[i].sa      = NULL;
     queue[i].sb      = NULL;
     queue[i].next    = &queue[i + 1];
   }
+
   queue[0].sa = sa;
   queue[0].sb = sb;
-  queue[nthreads - 1].next = NULL;
 
-  /* Iterate through steps of n */
   if (!range_n) {
     n_from = 0;
     n_to   = args -> n;
@@ -633,38 +638,38 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
     n_from = range_n[0];
     n_to   = range_n[1];
   }
+
   for(js = n_from; js < n_to; js += GEMM_R * nthreads){
     n = n_to - js;
     if (n > GEMM_R * nthreads) n = GEMM_R * nthreads;
 
-    /* Partition (a step of) n into nthreads regions */
     range_N[0] = js;
-    num_parts  = 0;
+
+    num_cpu_n  = 0;
+
     while (n > 0){
-      width = blas_quickdivide(n + nthreads - num_parts - 1, nthreads - num_parts);
-      if (width < SWITCH_RATIO) {
-        width = SWITCH_RATIO;
-      }
+
+      width  = blas_quickdivide(n + nthreads - num_cpu_n - 1, nthreads - num_cpu_n);
+
       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++) {
-      range_N[j + 1] = range_N[num_parts];
+
+      range_N[num_cpu_n + 1] = range_N[num_cpu_n] + width;
+
+      num_cpu_n ++;
     }
 
-    /* Clear synchronization flags */
-    for (i = 0; i < nthreads; i++) {
-      for (j = 0; j < nthreads; j++) {
+    for (j = 0; j < num_cpu_m; j++) {
+      for (i = 0; i < num_cpu_m; i++) {
 	for (k = 0; k < DIVIDE_RATE; k++) {
-	  job[i].working[j][CACHE_LINE_SIZE * k] = 0;
+	  job[j].working[i][CACHE_LINE_SIZE * k] = 0;
 	}
       }
     }
 
-    /* Execute parallel computation */
-    exec_blas(nthreads, queue);
+    queue[num_cpu_m - 1].next = NULL;
+
+    exec_blas(num_cpu_m, queue);
   }
 
 #ifdef USE_ALLOC_HEAP
@@ -678,42 +683,88 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
 
   BLASLONG m = args -> m;
   BLASLONG n = args -> n;
-  BLASLONG nthreads_m, nthreads_n;
+  BLASLONG nthreads = args -> nthreads;
+  BLASLONG divN, divT;
+  int mode;
 
-  /* Get dimensions from index ranges if available */
-  if (range_m) {
-    m = range_m[1] - range_m[0];
-  }
-  if (range_n) {
-    n = range_n[1] - range_n[0];
-  }
-
-  /* Partitions in m should have at least SWITCH_RATIO rows */
-  if (m < 2 * SWITCH_RATIO) {
-    nthreads_m = 1;
-  } else {
-    nthreads_m = args -> nthreads;
-    while (m < nthreads_m * SWITCH_RATIO) {
-      nthreads_m = nthreads_m / 2;
-    }
-  }
-
-  /* Partitions in n should have at most SWITCH_RATIO * nthreads_m columns */
-  if (n < SWITCH_RATIO * nthreads_m) {
-    nthreads_n = 1;
-  } else {
-    nthreads_n = (n + SWITCH_RATIO * nthreads_m - 1) / (SWITCH_RATIO * nthreads_m);
-    if (nthreads_m * nthreads_n > args -> nthreads) {
-      nthreads_n = blas_quickdivide(args -> nthreads, nthreads_m);
-    }
-  }
-
-  /* Execute serial or parallel computation */
-  if (nthreads_m * nthreads_n <= 1) {
+  if (nthreads  == 1) {
     GEMM_LOCAL(args, range_m, range_n, sa, sb, 0);
+    return 0;
+  }
+
+  if (range_m) {
+    BLASLONG m_from = *(((BLASLONG *)range_m) + 0);
+    BLASLONG m_to   = *(((BLASLONG *)range_m) + 1);
+
+    m = m_to - m_from;
+  }
+
+  if (range_n) {
+    BLASLONG n_from = *(((BLASLONG *)range_n) + 0);
+    BLASLONG n_to   = *(((BLASLONG *)range_n) + 1);
+
+    n = n_to - n_from;
+  }
+
+  if ((m < nthreads * SWITCH_RATIO) || (n < nthreads * SWITCH_RATIO)) {
+    GEMM_LOCAL(args, range_m, range_n, sa, sb, 0);
+    return 0;
+  }
+
+  divT = nthreads;
+  divN = 1;
+
+#if 0
+  while ((GEMM_P * divT > m * SWITCH_RATIO) && (divT > 1)) {
+    do {
+      divT --;
+      divN = 1;
+      while (divT * divN < nthreads) divN ++;
+    } while ((divT * divN != nthreads) && (divT > 1));
+  }
+#endif
+
+  // fprintf(stderr, "divN = %4ld  divT = %4ld\n", divN, divT);
+
+  args -> nthreads = divT;
+
+  if (divN == 1){
+
+    gemm_driver(args, range_m, range_n, sa, sb, 0);
   } else {
-    args -> nthreads = nthreads_m * nthreads_n;
-    gemm_driver(args, range_m, range_n, sa, sb, nthreads_m, nthreads_n);
+#ifndef COMPLEX
+#ifdef XDOUBLE
+    mode  =  BLAS_XDOUBLE | BLAS_REAL;
+#elif defined(DOUBLE)
+    mode  =  BLAS_DOUBLE  | BLAS_REAL;
+#else
+    mode  =  BLAS_SINGLE  | BLAS_REAL;
+#endif
+#else
+#ifdef XDOUBLE
+    mode  =  BLAS_XDOUBLE | BLAS_COMPLEX;
+#elif defined(DOUBLE)
+    mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
+#else
+    mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
+#endif
+#endif
+
+#if defined(TN) || defined(TT) || defined(TR) || defined(TC) || \
+    defined(CN) || defined(CT) || defined(CR) || defined(CC)
+    mode |= (BLAS_TRANSA_T);
+#endif
+#if defined(NT) || defined(TT) || defined(RT) || defined(CT) || \
+    defined(NC) || defined(TC) || defined(RC) || defined(CC)
+    mode |= (BLAS_TRANSB_T);
+#endif
+
+#ifdef OS_WINDOWS
+    gemm_thread_n(mode, args, range_m, range_n, GEMM_LOCAL,  sa, sb, divN);
+#else
+    gemm_thread_n(mode, args, range_m, range_n, gemm_driver, sa, sb, divN);
+#endif
+
   }
 
   return 0;