Merge branch 'cmake' of https://github.com/hpanderson/OpenBLAS into hpanderson_cmake

.gitignore

@@ -65,3 +65,5 @@ test/sblat3
 test/zblat1
 test/zblat2
 test/zblat3
+build
+build.*

CMakeLists.txt

@@ -0,0 +1,125 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+##
+
+cmake_minimum_required(VERSION 2.8.4)
+project(OpenBLAS)
+set(OpenBLAS_MAJOR_VERSION 0)
+set(OpenBLAS_MINOR_VERSION 2)
+set(OpenBLAS_PATCH_VERSION 13)
+set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
+
+enable_language(ASM)
+
+message(WARNING "CMake support is experimental. This will not produce the same Makefiles that OpenBLAS ships with. Only x86 support is currently available.")
+
+include("${CMAKE_SOURCE_DIR}/cmake/utils.cmake")
+include("${CMAKE_SOURCE_DIR}/cmake/system.cmake")
+
+set(BLASDIRS interface driver/level2 driver/level3 driver/others)
+
+if (NOT DYNAMIC_ARCH)
+  list(APPEND BLASDIRS kernel)
+endif ()
+
+if (DEFINED UTEST_CHECK)
+  set(SANITY_CHECK 1)
+endif ()
+
+if (DEFINED SANITY_CHECK)
+  list(APPEND BLASDIRS reference)
+endif ()
+
+set(SUBDIRS	${BLASDIRS})
+if (NOT NO_LAPACK)
+  list(APPEND SUBDIRS lapack)
+endif ()
+
+# set which float types we want to build for
+if (NOT DEFINED BUILD_SINGLE AND NOT DEFINED BUILD_DOUBLE AND NOT DEFINED BUILD_COMPLEX AND NOT DEFINED BUILD_COMPLEX16)
+  # if none are defined, build for all
+  set(BUILD_SINGLE true)
+  set(BUILD_DOUBLE true)
+  set(BUILD_COMPLEX true)
+  set(BUILD_COMPLEX16 true)
+endif ()
+
+set(FLOAT_TYPES "")
+if (BUILD_SINGLE)
+  message(STATUS "Building Single Precision")
+  list(APPEND FLOAT_TYPES "SINGLE") # defines nothing
+endif ()
+
+if (BUILD_DOUBLE)
+  message(STATUS "Building Double Precision")
+  list(APPEND FLOAT_TYPES "DOUBLE") # defines DOUBLE
+endif ()
+
+if (BUILD_COMPLEX)
+  message(STATUS "Building Complex Precision")
+  list(APPEND FLOAT_TYPES "COMPLEX") # defines COMPLEX
+endif ()
+
+if (BUILD_COMPLEX16)
+  message(STATUS "Building Double Complex Precision")
+  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 ()
+
+# get obj vars into format that add_library likes: $<TARGET_OBJS:objlib> (see http://www.cmake.org/cmake/help/v3.0/command/add_library.html)
+set(TARGET_OBJS "")
+foreach (SUBDIR ${SUBDIRS})
+  add_subdirectory(${SUBDIR})
+  string(REPLACE "/" "_" subdir_obj ${SUBDIR})
+  list(APPEND TARGET_OBJS "$<TARGET_OBJECTS:${subdir_obj}>")
+endforeach ()
+
+# netlib:
+
+# Can't just use lapack-netlib's CMake files, since they are set up to search for BLAS, build and install a binary. We just want to build a couple of lib files out of lapack and lapacke.
+# Not using add_subdirectory here because lapack-netlib already has its own CMakeLists.txt. Instead include a cmake script with the sources we want.
+if (NOT NOFORTRAN)
+  include("${CMAKE_SOURCE_DIR}/cmake/lapack.cmake")
+endif ()
+if (NOT NO_LAPACKE)
+  include("${CMAKE_SOURCE_DIR}/cmake/lapacke.cmake")
+endif ()
+
+# add objects to the openblas lib
+add_library(openblas ${LA_SOURCES} ${LAPACKE_SOURCES} ${TARGET_OBJS})
+
+# TODO: Why is the config saved here? Is this necessary with CMake?
+#Save the config files for installation
+#	@cp Makefile.conf Makefile.conf_last
+#	@cp config.h config_last.h
+#ifdef QUAD_PRECISION
+#	@echo "#define QUAD_PRECISION">> config_last.h
+#endif
+#ifeq ($(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
+

cmake/arch.cmake

@@ -0,0 +1,115 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+## Description: Ported from portion of OpenBLAS/Makefile.system
+##              Sets various variables based on architecture.
+
+if (${ARCH} STREQUAL "x86" OR ${ARCH} STREQUAL "x86_64")
+
+  if (${ARCH} STREQUAL "x86")
+    if (NOT BINARY)
+      set(NO_BINARY_MODE 1)
+    endif ()
+  endif ()
+
+  if (NOT NO_EXPRECISION)
+    if (${F_COMPILER} MATCHES "GFORTRAN")
+      # N.B. I'm not sure if CMake differentiates between GCC and LSB -hpa
+      if (${CMAKE_C_COMPILER} STREQUAL "GNU" OR ${CMAKE_C_COMPILER} STREQUAL "LSB")
+        set(EXPRECISION	1)
+        set(CCOMMON_OPT "${CCOMMON_OPT} -DEXPRECISION -m128bit-long-double")
+        set(FCOMMON_OPT	"${FCOMMON_OPT} -m128bit-long-double")
+      endif ()
+      if (${CMAKE_C_COMPILER} STREQUAL "Clang")
+        set(EXPRECISION	1)
+        set(CCOMMON_OPT "${CCOMMON_OPT} -DEXPRECISION")
+        set(FCOMMON_OPT	"${FCOMMON_OPT} -m128bit-long-double")
+      endif ()
+    endif ()
+  endif ()
+endif ()
+
+if (${CMAKE_C_COMPILER} STREQUAL "Intel")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -wd981")
+endif ()
+
+if (USE_OPENMP)
+
+  if (${CMAKE_C_COMPILER} STREQUAL "GNU" OR ${CMAKE_C_COMPILER} STREQUAL "LSB")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -fopenmp")
+  endif ()
+
+  if (${CMAKE_C_COMPILER} STREQUAL "Clang")
+    message(WARNING "Clang doesn't support OpenMP yet.")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -fopenmp")
+  endif ()
+
+  if (${CMAKE_C_COMPILER} STREQUAL "Intel")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -openmp")
+  endif ()
+
+  if (${CMAKE_C_COMPILER} STREQUAL "PGI")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -mp")
+  endif ()
+
+  if (${CMAKE_C_COMPILER} STREQUAL "OPEN64")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -mp")
+    set(CEXTRALIB "${CEXTRALIB} -lstdc++")
+  endif ()
+
+  if (${CMAKE_C_COMPILER} STREQUAL "PATHSCALE")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -mp")
+  endif ()
+endif ()
+
+
+if (DYNAMIC_ARCH)
+  if (${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 (${ARCH} STREQUAL "x86_64")
+    set(DYNAMIC_CORE "PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO")
+    if (NOT NO_AVX)
+      set(DYNAMIC_CORE "${DYNAMIC_CORE} SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER")
+    endif ()
+    if (NOT NO_AVX2)
+      set(DYNAMIC_CORE "${DYNAMIC_CORE} HASWELL")
+    endif ()
+  endif ()
+
+  if (NOT DYNAMIC_CORE)
+    unset(DYNAMIC_ARCH)
+  endif ()
+endif ()
+
+if (${ARCH} STREQUAL "ia64")
+  set(NO_BINARY_MODE 1)
+  set(BINARY_DEFINED 1)
+
+  if (${F_COMPILER} MATCHES "GFORTRAN")
+    if (${CMAKE_C_COMPILER} STREQUAL "GNU")
+      # EXPRECISION	= 1
+      # CCOMMON_OPT	+= -DEXPRECISION
+    endif ()
+  endif ()
+endif ()
+
+if (${ARCH} STREQUAL "mips64")
+  set(NO_BINARY_MODE 1)
+endif ()
+
+if (${ARCH} STREQUAL "alpha")
+  set(NO_BINARY_MODE 1)
+  set(BINARY_DEFINED 1)
+endif ()
+
+if (${ARCH} STREQUAL "arm")
+  set(NO_BINARY_MODE 1)
+  set(BINARY_DEFINED 1)
+endif ()
+
+if (${ARCH} STREQUAL "arm64")
+  set(NO_BINARY_MODE 1)
+  set(BINARY_DEFINED 1)
+endif ()
+

cmake/c_check.cmake

@@ -0,0 +1,84 @@
+##
+## 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.
+
+# TODO: detect FU (front underscore) by compiling ctest1.c
+set(FU "_")
+
+# 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 ()
+
+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")
+

cmake/cc.cmake

@@ -0,0 +1,103 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+## Description: Ported from portion of OpenBLAS/Makefile.system
+##              Sets C related variables.
+
+if (${CMAKE_C_COMPILER} STREQUAL "GNU" OR ${CMAKE_C_COMPILER} STREQUAL "LSB" OR ${CMAKE_C_COMPILER} STREQUAL "Clang")
+
+  set(CCOMMON_OPT "${CCOMMON_OPT} -Wall")
+  set(COMMON_PROF "${COMMON_PROF} -fno-inline")
+  set(NO_UNINITIALIZED_WARN "-Wno-uninitialized")
+
+  if (QUIET_MAKE)
+    set(CCOMMON_OPT "${CCOMMON_OPT} ${NO_UNINITIALIZED_WARN} -Wno-unused")
+  endif ()
+
+  if (NO_BINARY_MODE)
+
+    if (${ARCH} STREQUAL "mips64")
+      if (BINARY64)
+        set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=64")
+      else ()
+        set(CCOMMON_OPT "${CCOMMON_OPT} -mabi=n32")
+      endif ()
+      set(BINARY_DEFINED 1)
+    endif ()
+
+    if (${CORE} STREQUAL "LOONGSON3A")
+      set(CCOMMON_OPT "${CCOMMON_OPT} -march=mips64")
+      set(FCOMMON_OPT "${FCOMMON_OPT} -march=mips64")
+    endif ()
+
+    if (${CORE} STREQUAL "LOONGSON3B")
+      set(CCOMMON_OPT "${CCOMMON_OPT} -march=mips64")
+      set(FCOMMON_OPT "${FCOMMON_OPT} -march=mips64")
+    endif ()
+
+    if (${OSNAME} STREQUAL "AIX")
+      set(BINARY_DEFINED 1)
+    endif ()
+  endif ()
+
+  if (NOT BINARY_DEFINED)
+    if (BINARY64)
+      set(CCOMMON_OPT "${CCOMMON_OPT} -m64")
+    else ()
+      set(CCOMMON_OPT "${CCOMMON_OPT} -m32")
+    endif ()
+  endif ()
+endif ()
+
+if (${CMAKE_C_COMPILER} STREQUAL "PGI")
+  if (BINARY64)
+    set(CCOMMON_OPT "${CCOMMON_OPT} -tp p7-64")
+  else ()
+    set(CCOMMON_OPT "${CCOMMON_OPT} -tp p7")
+  endif ()
+endif ()
+
+if (${CMAKE_C_COMPILER} STREQUAL "PATHSCALE")
+  if (BINARY64)
+    set(CCOMMON_OPT "${CCOMMON_OPT} -m64")
+  else ()
+    set(CCOMMON_OPT "${CCOMMON_OPT} -m32")
+  endif ()
+endif ()
+
+if (${CMAKE_C_COMPILER} STREQUAL "OPEN64")
+
+  if (${ARCH} STREQUAL "mips64")
+
+    if (NOT BINARY64)
+      set(CCOMMON_OPT "${CCOMMON_OPT} -n32")
+    else ()
+      set(CCOMMON_OPT "${CCOMMON_OPT} -n64")
+    endif ()
+
+    if (${CORE} STREQUAL "LOONGSON3A")
+      set(CCOMMON_OPT "${CCOMMON_OPT} -loongson3 -static")
+    endif ()
+
+    if (${CORE} STREQUAL "LOONGSON3B")
+      set(CCOMMON_OPT "${CCOMMON_OPT} -loongson3 -static")
+    endif ()
+
+  else ()
+
+    if (BINARY64)
+      set(CCOMMON_OPT "${CCOMMON_OPT} -m32")
+    else ()
+      set(CCOMMON_OPT "${CCOMMON_OPT} -m64")
+    endif ()
+  endif ()
+endif ()
+
+if (${CMAKE_C_COMPILER} STREQUAL "SUN")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -w")
+  if (${ARCH} STREQUAL "x86")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -m32")
+  else ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
+  endif ()
+endif ()
+

cmake/f_check.cmake

@@ -0,0 +1,60 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+## Copyright: (c) Stat-Ease, Inc.
+## Created: 12/29/14
+## Last Modified: 12/29/14
+## Description: Ported from the OpenBLAS/f_check perl script.
+##              This is triggered by prebuild.cmake and runs before any of the code is built.
+##              Appends Fortran information to config.h and Makefile.conf.
+
+# CMake vars set by this file:
+# F_COMPILER
+# FC
+# BU
+# NOFORTRAN
+# NEED2UNDERSCORES
+# FEXTRALIB
+
+# Defines set by this file:
+# BUNDERSCORE
+# 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 ()
+enable_language(Fortran)
+
+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}
+    "#define BUNDERSCORE _\n"
+    "#define NEEDBUNDERSCORE 1\n"
+    "#define NEED2UNDERSCORES 0\n")
+
+else ()
+
+  #When we only build CBLAS, we set NOFORTRAN=2
+  set(NOFORTRAN 2)
+  set(NO_FBLAS 1)
+  #set(F_COMPILER GFORTRAN) # CMake handles the fortran compiler
+  set(BU "_")
+  file(APPEND ${TARGET_CONF}
+    "#define BUNDERSCORE _\n"
+    "#define NEEDBUNDERSCORE 1\n")
+endif()
+
+get_filename_component(F_COMPILER ${CMAKE_Fortran_COMPILER} NAME_WE)
+string(TOUPPER ${F_COMPILER} F_COMPILER)
+

cmake/fc.cmake

@@ -0,0 +1,200 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+## Description: Ported from portion of OpenBLAS/Makefile.system
+##              Sets Fortran related variables.
+
+if (${F_COMPILER} STREQUAL "G77")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_G77")
+  set(FCOMMON_OPT "${FCOMMON_OPT} -Wall")
+  if (NOT NO_BINARY_MODE)
+    if (BINARY64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
+    else ()
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
+    endif ()
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "G95")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_G95")
+  set(FCOMMON_OPT "${FCOMMON_OPT} -Wall")
+  if (NOT NO_BINARY_MODE)
+    if (BINARY64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
+    else ()
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
+    endif ()
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "GFORTRAN")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_GFORT")
+  set(FCOMMON_OPT "${FCOMMON_OPT} -Wall")
+  #Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
+  if (NOT NO_LAPACK)
+    set(EXTRALIB "{EXTRALIB} -lgfortran")
+  endif ()
+  if (NO_BINARY_MODE)
+    if (${ARCH} STREQUAL "mips64")
+      if (BINARY64)
+        set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=64")
+      else ()
+        set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=n32")
+      endif ()
+    endif ()
+  else ()
+    if (BINARY64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
+      if (INTERFACE64)
+        set(FCOMMON_OPT "${FCOMMON_OPT} -fdefault-integer-8")
+      endif ()
+    else ()
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
+    endif ()
+  endif ()
+
+  if (USE_OPENMP)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -fopenmp")
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "INTEL")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_INTEL")
+  if (INTERFACE64)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -i8")
+  endif ()
+  if (USE_OPENMP)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -openmp")
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "FUJITSU")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_FUJITSU")
+  if (USE_OPENMP)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -openmp")
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "IBM")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_IBM")
+  # FCOMMON_OPT	+= -qarch=440
+  if (BINARY64)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -q64")
+    if (INTERFACE64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -qintsize=8")
+    endif ()
+  else ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -q32")
+  endif ()
+  if (USE_OPENMP)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -openmp")
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "PGI")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_PGI")
+  set(COMMON_PROF "${COMMON_PROF} -DPGICOMPILER")
+  if (BINARY64)
+    if (INTERFACE64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -i8")
+    endif ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -tp p7-64")
+  else ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -tp p7")
+  endif ()
+  if (USE_OPENMP)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -mp")
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "PATHSCALE")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_PATHSCALE")
+  if (BINARY64)
+    if (INTERFACE64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -i8")
+    endif ()
+  endif ()
+
+  if (NOT ${ARCH} STREQUAL "mips64")
+    if (NOT BINARY64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
+    else ()
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
+    endif ()
+  else ()
+    if (BINARY64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=64")
+    else ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=n32")
+    endif ()
+  endif ()
+
+  if (USE_OPENMP)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -mp")
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "OPEN64")
+
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_OPEN64")
+  if (BINARY64)
+    if (INTERFACE64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -i8")
+    endif ()
+  endif ()
+
+  if (${ARCH} STREQUAL "mips64")
+
+    if (NOT BINARY64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -n32")
+    else ()
+      set(FCOMMON_OPT "${FCOMMON_OPT} -n64")
+    endif ()
+
+    if (${CORE} STREQUAL "LOONGSON3A")
+      set(FCOMMON_OPT "${FCOMMON_OPT} -loongson3 -static")
+    endif ()
+
+    if (${CORE} STREQUAL "LOONGSON3B")
+    set(FCOMMON_OPT "${FCOMMON_OPT} -loongson3 -static")
+    endif ()
+  else ()
+    if (NOT BINARY64)
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
+    else ()
+      set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
+    endif ()
+  endif ()
+
+  if (USE_OPENMP)
+    set(FEXTRALIB "${FEXTRALIB} -lstdc++")
+    set(FCOMMON_OPT "${FCOMMON_OPT} -mp")
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "SUN")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_SUN")
+  if (${ARCH} STREQUAL "x86")
+    set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
+  else ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
+  endif ()
+  if (USE_OPENMP)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -xopenmp=parallel")
+  endif ()
+endif ()
+
+if (${F_COMPILER} STREQUAL "COMPAQ")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_COMPAQ")
+  if (USE_OPENMP)
+    set(FCOMMON_OPT "${FCOMMON_OPT} -openmp")
+  endif ()
+endif ()
+
+# from the root Makefile - this is for lapack-netlib to compile the correct secnd file.
+if (${F_COMPILER} STREQUAL "GFORTRAN")
+  set(TIMER "INT_ETIME")
+else ()
+  set(TIMER "NONE")
+endif ()
+

cmake/kernel.cmake

@@ -0,0 +1,158 @@
+# helper functions for the kernel CMakeLists.txt
+
+
+# Set the default filenames for L1 objects. Most of these will be overriden by the appropriate KERNEL file.
+macro(SetDefaultL1)
+  set(SAMAXKERNEL amax.S)
+  set(DAMAXKERNEL amax.S)
+  set(QAMAXKERNEL amax.S)
+  set(CAMAXKERNEL zamax.S)
+  set(ZAMAXKERNEL zamax.S)
+  set(XAMAXKERNEL zamax.S)
+  set(SAMINKERNEL amin.S)
+  set(DAMINKERNEL amin.S)
+  set(QAMINKERNEL amin.S)
+  set(CAMINKERNEL zamin.S)
+  set(ZAMINKERNEL zamin.S)
+  set(XAMINKERNEL zamin.S)
+  set(SMAXKERNEL max.S)
+  set(DMAXKERNEL max.S)
+  set(QMAXKERNEL max.S)
+  set(SMINKERNEL min.S)
+  set(DMINKERNEL min.S)
+  set(QMINKERNEL min.S)
+  set(ISAMAXKERNEL iamax.S)
+  set(IDAMAXKERNEL iamax.S)
+  set(IQAMAXKERNEL iamax.S)
+  set(ICAMAXKERNEL izamax.S)
+  set(IZAMAXKERNEL izamax.S)
+  set(IXAMAXKERNEL izamax.S)
+  set(ISAMINKERNEL iamin.S)
+  set(IDAMINKERNEL iamin.S)
+  set(IQAMINKERNEL iamin.S)
+  set(ICAMINKERNEL izamin.S)
+  set(IZAMINKERNEL izamin.S)
+  set(IXAMINKERNEL izamin.S)
+  set(ISMAXKERNEL iamax.S)
+  set(IDMAXKERNEL iamax.S)
+  set(IQMAXKERNEL iamax.S)
+  set(ISMINKERNEL iamin.S)
+  set(IDMINKERNEL iamin.S)
+  set(IQMINKERNEL iamin.S)
+  set(SASUMKERNEL asum.S)
+  set(DASUMKERNEL asum.S)
+  set(CASUMKERNEL zasum.S)
+  set(ZASUMKERNEL zasum.S)
+  set(QASUMKERNEL asum.S)
+  set(XASUMKERNEL zasum.S)
+  set(SAXPYKERNEL axpy.S)
+  set(DAXPYKERNEL axpy.S)
+  set(CAXPYKERNEL zaxpy.S)
+  set(ZAXPYKERNEL zaxpy.S)
+  set(QAXPYKERNEL axpy.S)
+  set(XAXPYKERNEL zaxpy.S)
+  set(SCOPYKERNEL copy.S)
+  set(DCOPYKERNEL copy.S)
+  set(CCOPYKERNEL zcopy.S)
+  set(ZCOPYKERNEL zcopy.S)
+  set(QCOPYKERNEL copy.S)
+  set(XCOPYKERNEL zcopy.S)
+  set(SDOTKERNEL dot.S)
+  set(DDOTKERNEL dot.S)
+  set(CDOTKERNEL zdot.S)
+  set(ZDOTKERNEL zdot.S)
+  set(QDOTKERNEL dot.S)
+  set(XDOTKERNEL zdot.S)
+  set(SNRM2KERNEL nrm2.S)
+  set(DNRM2KERNEL nrm2.S)
+  set(QNRM2KERNEL nrm2.S)
+  set(CNRM2KERNEL znrm2.S)
+  set(ZNRM2KERNEL znrm2.S)
+  set(XNRM2KERNEL znrm2.S)
+  set(SROTKERNEL rot.S)
+  set(DROTKERNEL rot.S)
+  set(QROTKERNEL rot.S)
+  set(CROTKERNEL zrot.S)
+  set(ZROTKERNEL zrot.S)
+  set(XROTKERNEL zrot.S)
+  set(SSCALKERNEL scal.S)
+  set(DSCALKERNEL scal.S)
+  set(CSCALKERNEL zscal.S)
+  set(ZSCALKERNEL zscal.S)
+  set(QSCALKERNEL scal.S)
+  set(XSCALKERNEL zscal.S)
+  set(SSWAPKERNEL swap.S)
+  set(DSWAPKERNEL swap.S)
+  set(CSWAPKERNEL zswap.S)
+  set(ZSWAPKERNEL zswap.S)
+  set(QSWAPKERNEL swap.S)
+  set(XSWAPKERNEL zswap.S)
+  set(SGEMVNKERNEL gemv_n.S)
+  set(SGEMVTKERNEL gemv_t.S)
+  set(DGEMVNKERNEL gemv_n.S)
+  set(DGEMVTKERNEL gemv_t.S)
+  set(CGEMVNKERNEL zgemv_n.S)
+  set(CGEMVTKERNEL zgemv_t.S)
+  set(ZGEMVNKERNEL zgemv_n.S)
+  set(ZGEMVTKERNEL zgemv_t.S)
+  set(QGEMVNKERNEL gemv_n.S)
+  set(QGEMVTKERNEL gemv_t.S)
+  set(XGEMVNKERNEL zgemv_n.S)
+  set(XGEMVTKERNEL zgemv_t.S)
+  set(SCABS_KERNEL cabs.S)
+  set(DCABS_KERNEL cabs.S)
+  set(QCABS_KERNEL cabs.S)
+  set(LSAME_KERNEL lsame.S)
+  set(SAXPBYKERNEL ../arm/axpby.c)
+  set(DAXPBYKERNEL ../arm/axpby.c)
+  set(CAXPBYKERNEL ../arm/zaxpby.c)
+  set(ZAXPBYKERNEL ../arm/zaxpby.c)
+endmacro ()
+
+macro(SetDefaultL2)
+  set(SGEMVNKERNEL gemv_n.S)
+  set(SGEMVTKERNEL gemv_t.S)
+  set(DGEMVNKERNEL gemv_n.S)
+  set(DGEMVTKERNEL gemv_t.S)
+  set(CGEMVNKERNEL zgemv_n.S)
+  set(CGEMVTKERNEL zgemv_t.S)
+  set(ZGEMVNKERNEL zgemv_n.S)
+  set(ZGEMVTKERNEL zgemv_t.S)
+  set(QGEMVNKERNEL gemv_n.S)
+  set(QGEMVTKERNEL gemv_t.S)
+  set(XGEMVNKERNEL zgemv_n.S)
+  set(XGEMVTKERNEL zgemv_t.S)
+  set(SGERKERNEL ../generic/ger.c)
+  set(DGERKERNEL ../generic/ger.c)
+  set(QGERKERNEL ../generic/ger.c)
+  set(CGERUKERNEL ../generic/zger.c)
+  set(CGERCKERNEL ../generic/zger.c)
+  set(ZGERUKERNEL ../generic/zger.c)
+  set(ZGERCKERNEL ../generic/zger.c)
+  set(XGERUKERNEL ../generic/zger.c)
+  set(XGERCKERNEL ../generic/zger.c)
+  set(SSYMV_U_KERNEL ../generic/symv_k.c)
+  set(SSYMV_L_KERNEL ../generic/symv_k.c)
+  set(DSYMV_U_KERNEL ../generic/symv_k.c)
+  set(DSYMV_L_KERNEL ../generic/symv_k.c)
+  set(QSYMV_U_KERNEL ../generic/symv_k.c)
+  set(QSYMV_L_KERNEL ../generic/symv_k.c)
+  set(CSYMV_U_KERNEL ../generic/zsymv_k.c)
+  set(CSYMV_L_KERNEL ../generic/zsymv_k.c)
+  set(ZSYMV_U_KERNEL ../generic/zsymv_k.c)
+  set(ZSYMV_L_KERNEL ../generic/zsymv_k.c)
+  set(XSYMV_U_KERNEL ../generic/zsymv_k.c)
+  set(XSYMV_L_KERNEL ../generic/zsymv_k.c)
+  set(CHEMV_U_KERNEL ../generic/zhemv_k.c)
+  set(CHEMV_L_KERNEL ../generic/zhemv_k.c)
+  set(CHEMV_V_KERNEL ../generic/zhemv_k.c)
+  set(CHEMV_M_KERNEL ../generic/zhemv_k.c)
+  set(ZHEMV_U_KERNEL ../generic/zhemv_k.c)
+  set(ZHEMV_L_KERNEL ../generic/zhemv_k.c)
+  set(ZHEMV_V_KERNEL ../generic/zhemv_k.c)
+  set(ZHEMV_M_KERNEL ../generic/zhemv_k.c)
+  set(XHEMV_U_KERNEL ../generic/zhemv_k.c)
+  set(XHEMV_L_KERNEL ../generic/zhemv_k.c)
+  set(XHEMV_V_KERNEL ../generic/zhemv_k.c)
+  set(XHEMV_M_KERNEL ../generic/zhemv_k.c)
+endmacro ()

cmake/lapack.cmake

@@ -0,0 +1,347 @@
+# Sources for compiling lapack-netlib. Can't use CMakeLists.txt because lapack-netlib already has its own cmake files.
+
+set(ALLAUX
+  ilaenv.f ieeeck.f lsamen.f xerbla_array.f iparmq.f
+  ilaprec.f ilatrans.f ilauplo.f iladiag.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_${TIMER}.f
+)
+
+set(DZLAUX
+  dbdsdc.f
+  dbdsqr.f ddisna.f dlabad.f dlacpy.f dladiv.f dlae2.f  dlaebz.f
+  dlaed0.f dlaed1.f dlaed2.f dlaed3.f dlaed4.f dlaed5.f dlaed6.f
+  dlaed7.f dlaed8.f dlaed9.f dlaeda.f dlaev2.f dlagtf.f
+  dlagts.f dlamrg.f dlanst.f
+  dlapy2.f dlapy3.f dlarnv.f
+  dlarra.f dlarrb.f dlarrc.f dlarrd.f dlarre.f dlarrf.f dlarrj.f
+  dlarrk.f dlarrr.f dlaneg.f
+  dlartg.f dlaruv.f dlas2.f  dlascl.f
+  dlasd0.f dlasd1.f dlasd2.f dlasd3.f dlasd4.f dlasd5.f dlasd6.f
+  dlasd7.f dlasd8.f dlasda.f dlasdq.f dlasdt.f
+  dlaset.f dlasq1.f dlasq2.f dlasq3.f dlasq4.f dlasq5.f dlasq6.f
+  dlasr.f  dlasrt.f dlassq.f dlasv2.f dpttrf.f dstebz.f dstedc.f
+  dsteqr.f dsterf.f dlaisnan.f disnan.f
+  dlartgp.f dlartgs.f
+  ../INSTALL/dlamch.f ../INSTALL/dsecnd_${TIMER}.f
+)
+
+set(SLASRC
+  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
+  sgegs.f  sgegv.f  sgehd2.f sgehrd.f sgelq2.f sgelqf.f
+  sgels.f  sgelsd.f sgelss.f sgelsx.f sgelsy.f sgeql2.f sgeqlf.f
+  sgeqp3.f sgeqpf.f sgeqr2.f sgeqr2p.f sgeqrf.f sgeqrfp.f sgerfs.f
+  sgerq2.f sgerqf.f sgesc2.f sgesdd.f  sgesvd.f sgesvx.f
+  sgetc2.f sgetri.f
+  sggbak.f sggbal.f sgges.f  sggesx.f sggev.f  sggevx.f
+  sggglm.f sgghrd.f sgglse.f sggqrf.f
+  sggrqf.f sggsvd.f sggsvp.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
+  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 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
+  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 strexc.f strrfs.f strsen.f strsna.f strsyl.f
+  strtrs.f 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
+)
+
+set(DSLASRC spotrs.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
+  cgegs.f  cgegv.f  cgehd2.f cgehrd.f cgelq2.f cgelqf.f
+  cgels.f  cgelsd.f cgelss.f cgelsx.f cgelsy.f cgeql2.f cgeqlf.f cgeqp3.f
+  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
+  cggsvd.f 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
+  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
+  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 ctrexc.f ctrrfs.f ctrsen.f ctrsna.f
+  ctrsyl.f ctrtrs.f 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
+)
+
+set(ZCLASRC cpotrs.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
+  dgegs.f  dgegv.f  dgehd2.f dgehrd.f dgelq2.f dgelqf.f
+  dgels.f  dgelsd.f dgelss.f dgelsx.f dgelsy.f dgeql2.f dgeqlf.f
+  dgeqp3.f dgeqpf.f dgeqr2.f dgeqr2p.f dgeqrf.f dgeqrfp.f dgerfs.f
+  dgerq2.f dgerqf.f dgesc2.f dgesdd.f  dgesvd.f dgesvx.f
+  dgetc2.f dgetri.f
+  dggbak.f dggbal.f dgges.f  dggesx.f dggev.f  dggevx.f
+  dggglm.f dgghrd.f dgglse.f dggqrf.f
+  dggrqf.f dggsvd.f 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
+  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 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
+  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 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 dtrexc.f dtrrfs.f dtrsen.f dtrsna.f dtrsyl.f
+  dtrtrs.f 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
+)
+
+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
+  zgegs.f  zgegv.f  zgehd2.f zgehrd.f zgelq2.f zgelqf.f
+  zgels.f  zgelsd.f zgelss.f zgelsx.f zgelsy.f zgeql2.f zgeqlf.f zgeqp3.f
+  zgeqpf.f zgeqr2.f zgeqr2p.f zgeqrf.f zgeqrfp.f zgerfs.f zgerq2.f zgerqf.f
+  zgesc2.f zgesdd.f zgesvd.f zgesvx.f zgetc2.f
+  zgetri.f
+  zggbak.f zggbal.f zgges.f  zggesx.f zggev.f  zggevx.f zggglm.f
+  zgghrd.f zgglse.f zggqrf.f zggrqf.f
+  zggsvd.f zggsvp.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
+  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
+  zlatbs.f zlatdf.f zlatps.f zlatrd.f zlatrs.f zlatrz.f 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 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
+  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 ztrexc.f ztrrfs.f ztrsen.f ztrsna.f
+  ztrsyl.f ztrtrs.f 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 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
+)
+
+set(LA_REL_SRC ${ALLAUX})
+if (BUILD_SINGLE)
+  list(APPEND LA_REL_SRC ${SLASRC} ${DSLASRC} ${SCLAUX})
+endif ()
+
+if (BUILD_DOUBLE)
+  list(APPEND LA_REL_SRC ${DLASRC} ${DSLASRC} ${DZLAUX})
+endif ()
+
+if (BUILD_COMPLEX)
+  list(APPEND LA_REL_SRC ${CLASRC} ${ZCLASRC} ${SCLAUX})
+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/${LA_FILE}")
+endforeach ()
+set_source_files_properties(${LA_SOURCES} PROPERTIES COMPILE_FLAGS "${LAPACK_FFLAGS}")

cmake/os.cmake

@@ -0,0 +1,104 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+## 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 "${EXTRALIB} -lm")
+  set(NO_EXPRECISION 1)
+endif ()
+
+if (${CMAKE_SYSTEM_NAME} STREQUAL "AIX")
+  set(EXTRALIB "${EXTRALIB} -lm")
+endif ()
+
+# TODO: this is probably meant for mingw, not other windows compilers
+if (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
+
+  set(NEED_PIC 0)
+  set(NO_EXPRECISION 1)
+
+  set(EXTRALIB "${EXTRALIB} -defaultlib:advapi32")
+
+  # probably not going to use these
+  set(SUFFIX "obj")
+  set(PSUFFIX "pobj")
+  set(LIBSUFFIX "a")
+
+  if (${CMAKE_C_COMPILER_ID} STREQUAL "Clang")
+    set(CCOMMON_OPT	"${CCOMMON_OPT} -DMS_ABI")
+  endif ()
+
+  if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
+
+    # Test for supporting MS_ABI
+    # removed string parsing in favor of CMake's version comparison -hpa
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 4.7 OR ${GCC_VERSION} VERSION_EQUAL 4.7)
+      # GCC Version >=4.7
+      # It is compatible with MSVC ABI.
+      set(CCOMMON_OPT "${CCOMMON_OPT} -DMS_ABI")
+    endif ()
+  endif ()
+
+  # Ensure the correct stack alignment on Win32
+  # http://permalink.gmane.org/gmane.comp.lib.openblas.general/97
+  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 ()
+    set(FCOMMON_OPT "${FCOMMON_OPT} -mincoming-stack-boundary=2")
+  endif ()
+  
+endif ()
+
+if (${CMAKE_SYSTEM_NAME} STREQUAL "Interix")
+  set(NEED_PIC 0)
+  set(NO_EXPRECISION 1)
+  
+  set(INTERIX_TOOL_DIR STREQUAL "/opt/gcc.3.3/i586-pc-interix3/bin")
+endif ()
+
+if (CYGWIN)
+  set(NEED_PIC 0)
+  set(NO_EXPRECISION 1)
+endif ()
+
+if (NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Windows" AND NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Interix")
+  if (SMP)
+    set(EXTRALIB "${EXTRALIB} -lpthread")
+  endif ()
+endif ()
+
+if (QUAD_PRECISION)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DQUAD_PRECISION")
+  set(NO_EXPRECISION 1)
+endif ()
+
+if (${ARCH} STREQUAL "x86")
+  set(NO_EXPRECISION 1)
+endif ()
+
+if (UTEST_CHECK)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DUTEST_CHECK")
+  set(SANITY_CHECK 1)
+endif ()
+
+if (SANITY_CHECK)
+  # TODO: need some way to get $(*F) (target filename)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DSANITY_CHECK -DREFNAME=$(*F)f${BU}")
+endif ()
+

cmake/prebuild.cmake

@@ -0,0 +1,108 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+## Description: Ported from OpenBLAS/Makefile.prebuild
+##              This is triggered by system.cmake and runs before any of the code is built.
+##              Creates config.h and Makefile.conf by first running the c_check perl script (which creates those files).
+##              Next it runs f_check and appends some fortran information to the files.
+##              Finally it runs getarch and getarch_2nd for even more environment information.
+
+# CMake vars set by this file:
+# CORE
+# LIBCORE
+# NUM_CORES
+# HAVE_MMX
+# HAVE_SSE
+# HAVE_SSE2
+# HAVE_SSE3
+# MAKE
+# SGEMM_UNROLL_M
+# SGEMM_UNROLL_N
+# DGEMM_UNROLL_M
+# DGEMM_UNROLL_M
+# QGEMM_UNROLL_N
+# QGEMM_UNROLL_N
+# CGEMM_UNROLL_M
+# CGEMM_UNROLL_M
+# ZGEMM_UNROLL_N
+# ZGEMM_UNROLL_N
+# XGEMM_UNROLL_M
+# XGEMM_UNROLL_N
+# CGEMM3M_UNROLL_M
+# CGEMM3M_UNROLL_N
+# ZGEMM3M_UNROLL_M
+# ZGEMM3M_UNROLL_M
+# XGEMM3M_UNROLL_N
+# XGEMM3M_UNROLL_N
+
+# CPUIDEMU = ../../cpuid/table.o
+
+if (DEFINED CPUIDEMU)
+  set(EXFLAGS "-DCPUIDEMU -DVENDOR=99")
+endif ()
+
+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")
+else()
+  set(TARGET_MAKE "Makefile.conf")
+  set(TARGET_CONF "config.h")
+endif ()
+
+include("${CMAKE_SOURCE_DIR}/cmake/c_check.cmake")
+
+if (NOT NOFORTRAN)
+  include("${CMAKE_SOURCE_DIR}/cmake/f_check.cmake")
+endif ()
+
+# compile getarch
+set(GETARCH_SRC
+  ${CMAKE_SOURCE_DIR}/getarch.c
+  ${CPUIDEMO}
+)
+
+if (NOT MSVC)
+  list(APPEND GETARCH_SRC ${CMAKE_SOURCE_DIR}/cpuid.S)
+endif ()
+
+set(GETARCH_DIR "${PROJECT_BINARY_DIR}/getarch_build")
+set(GETARCH_BIN "getarch${CMAKE_EXECUTABLE_SUFFIX}")
+file(MAKE_DIRECTORY ${GETARCH_DIR})
+try_compile(GETARCH_RESULT ${GETARCH_DIR}
+  SOURCES ${GETARCH_SRC}
+  COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I${CMAKE_SOURCE_DIR}
+  OUTPUT_VARIABLE GETARCH_LOG
+  COPY_FILE ${GETARCH_BIN}
+)
+
+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 ${GETARCH_BIN} 0 OUTPUT_VARIABLE GETARCH_MAKE_OUT)
+execute_process(COMMAND ${GETARCH_BIN} 1 OUTPUT_VARIABLE GETARCH_CONF_OUT)
+
+#message(STATUS "GETARCH results:\n${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})
+try_compile(GETARCH2_RESULT ${GETARCH2_DIR}
+  SOURCES ${CMAKE_SOURCE_DIR}/getarch_2nd.c
+  COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I${CMAKE_SOURCE_DIR}
+  OUTPUT_VARIABLE GETARCH2_LOG
+  COPY_FILE ${GETARCH2_BIN}
+)
+
+# use the cmake binary w/ the -E param to run a shell command in a cross-platform way
+execute_process(COMMAND ${GETARCH2_BIN} 0 OUTPUT_VARIABLE GETARCH2_MAKE_OUT)
+execute_process(COMMAND ${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})
+

cmake/system.cmake

@@ -0,0 +1,527 @@
+##
+## Author: Hank Anderson <hank@statease.com>
+## Description: Ported from OpenBLAS/Makefile.system
+##
+
+set(NETLIB_LAPACK_DIR "${CMAKE_SOURCE_DIR}/lapack-netlib")
+
+# TODO: Makefile.system detects Darwin (mac) and switches to clang here -hpa
+# http://stackoverflow.com/questions/714100/os-detecting-makefile
+
+# 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)
+  set(TARGET ${TARGET_CORE})
+endif ()
+
+# Force fallbacks for 32bit
+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 ${TARGET} STREQUAL "SANDYBRIDGE")
+    set(TARGET "NEHALEM")
+  endif ()
+  if (${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "PILEDRIVER")
+    set(TARGET "BARCELONA")
+  endif ()
+endif ()
+
+if (DEFINED TARGET)
+  message(STATUS "Targetting the ${TARGET} architecture.")
+  set(GETARCH_FLAGS "-DFORCE_${TARGET}")
+endif ()
+
+if (INTERFACE64)
+  message(STATUS "Using 64-bit integers.")
+  set(GETARCH_FLAGS	"${GETARCH_FLAGS} -DUSE64BITINT")
+endif ()
+
+if (NOT DEFINED GEMM_MULTITHREAD_THRESHOLD)
+  set(GEMM_MULTITHREAD_THRESHOLD 4)
+endif ()
+message(STATUS "GEMM multithread threshold set to ${GEMM_MULTITHREAD_THRESHOLD}.")
+set(GETARCH_FLAGS	"${GETARCH_FLAGS} -DGEMM_MULTITHREAD_THRESHOLD=${GEMM_MULTITHREAD_THRESHOLD}")
+
+if (NO_AVX)
+  message(STATUS "Disabling Advanced Vector Extensions (AVX).")
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX")
+endif ()
+
+if (NO_AVX2)
+  message(STATUS "Disabling Advanced Vector Extensions 2 (AVX2).")
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX2")
+endif ()
+
+if (CMAKE_BUILD_TYPE STREQUAL Debug)
+  set(GETARCH_FLAGS "${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_CXX_COMPILER STREQUAL loongcc)
+  set(GETARCH_FLAGS	"${GETARCH_FLAGS} -static")
+endif ()
+
+#if don't use Fortran, it will only compile CBLAS.
+if (ONLY_CBLAS)
+  set(NO_LAPACK 1)
+else ()
+  set(ONLY_CBLAS 0)
+endif ()
+
+include("${CMAKE_SOURCE_DIR}/cmake/prebuild.cmake")
+
+if (NOT DEFINED NUM_THREADS)
+  set(NUM_THREADS ${NUM_CORES})
+endif ()
+
+if (${NUM_THREADS} EQUAL 1)
+  set(USE_THREAD 0)
+endif ()
+
+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 ()
+
+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("${CMAKE_SOURCE_DIR}/cmake/os.cmake")
+
+# Architecture dependent settings
+include("${CMAKE_SOURCE_DIR}/cmake/arch.cmake")
+
+# C Compiler dependent settings
+include("${CMAKE_SOURCE_DIR}/cmake/cc.cmake")
+
+if (NOT NOFORTRAN)
+  # Fortran Compiler dependent settings
+  include("${CMAKE_SOURCE_DIR}/cmake/fc.cmake")
+endif ()
+
+if (BINARY64)
+  if (INTERFACE64)
+    # CCOMMON_OPT += -DUSE64BITINT
+  endif ()
+endif ()
+
+if (NEED_PIC)
+  if (${CMAKE_C_COMPILER} STREQUAL "IBM")
+    set(CCOMMON_OPT "${CCOMMON_OPT} -qpic=large")
+  else ()
+    set(CCOMMON_OPT "${CCOMMON_OPT} -fPIC")
+  endif ()
+
+  if (${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 "${CCOMMON_OPT} -DDYNAMIC_ARCH")
+endif ()
+
+if (NO_LAPACK)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_LAPACK")
+  #Disable LAPACK C interface
+  set(NO_LAPACKE 1)
+endif ()
+
+if (NO_LAPACKE)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_LAPACKE")
+endif ()
+
+if (NO_AVX)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX")
+endif ()
+
+if (${ARCH} STREQUAL "x86")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX")
+endif ()
+
+if (NO_AVX2)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX2")
+endif ()
+
+if (SMP)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DSMP_SERVER")
+
+  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 "${CCOMMON_OPT} -DBIGNUMA")
+  endif ()
+
+endif ()
+
+if (NO_WARMUP)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_WARMUP")
+endif ()
+
+if (CONSISTENT_FPCSR)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DCONSISTENT_FPCSR")
+endif ()
+
+# Only for development
+# set(CCOMMON_OPT "${CCOMMON_OPT} -DPARAMTEST")
+# set(CCOMMON_OPT "${CCOMMON_OPT} -DPREFETCHTEST")
+# set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_SWITCHING")
+# set(USE_PAPI 1)
+
+if (USE_PAPI)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_PAPI")
+  set(EXTRALIB "${EXTRALIB} -lpapi -lperfctr")
+endif ()
+
+if (DYNAMIC_THREADS)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_THREADS")
+endif ()
+
+set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_CPU_NUMBER=${NUM_THREADS}")
+
+if (USE_SIMPLE_THREADED_LEVEL3)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_SIMPLE_THREADED_LEVEL3")
+endif ()
+
+if (DEFINED LIBNAMESUFFIX)
+  set(LIBPREFIX "libopenblas_${LIBNAMESUFFIX}")
+else ()
+  set(LIBPREFIX "libopenblas")
+endif ()
+
+if (NOT DEFINED SYMBOLPREFIX)
+  set(SYMBOLPREFIX "")
+endif ()
+
+if (NOT DEFINED SYMBOLSUFFIX)
+  set(SYMBOLSUFFIX "")
+endif ()
+
+set(KERNELDIR	"${CMAKE_SOURCE_DIR}/kernel/${ARCH}")
+
+# TODO: nead to convert these Makefiles
+# include ${CMAKE_SOURCE_DIR}/cmake/${ARCH}.cmake
+
+if (${CORE} STREQUAL "PPC440")
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DALLOC_QALLOC")
+endif ()
+
+if (${CORE} STREQUAL "PPC440FP2")
+  set(STATIC_ALLOCATION 1)
+endif ()
+
+if (NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
+  set(NO_AFFINITY 1)
+endif ()
+
+if (NOT ${ARCH} STREQUAL "x86_64" AND NOT ${ARCH} STREQUAL "x86" AND NOT ${CORE} STREQUAL "LOONGSON3B")
+  set(NO_AFFINITY 1)
+endif ()
+
+if (NO_AFFINITY)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AFFINITY")
+endif ()
+
+if (FUNCTION_PROFILE)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DFUNCTION_PROFILE")
+endif ()
+
+if (HUGETLB_ALLOCATION)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DALLOC_HUGETLB")
+endif ()
+
+if (DEFINED HUGETLBFILE_ALLOCATION)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DALLOC_HUGETLBFILE -DHUGETLB_FILE_NAME=${HUGETLBFILE_ALLOCATION})")
+endif ()
+
+if (STATIC_ALLOCATION)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DALLOC_STATIC")
+endif ()
+
+if (DEVICEDRIVER_ALLOCATION)
+  set(CCOMMON_OPT "${CCOMMON_OPT} -DALLOC_DEVICEDRIVER -DDEVICEDRIVER_NAME=\"/dev/mapper\"")
+endif ()
+
+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(REVISION "-r${OpenBLAS_VERSION}")
+set(MAJOR_VERSION ${OpenBLAS_MAJOR_VERSION})
+
+if (DEBUG)
+  set(COMMON_OPT "${COMMON_OPT} -g")
+endif ()
+
+if (NOT DEFINED COMMON_OPT)
+  set(COMMON_OPT "-O2")
+endif ()
+
+set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${COMMON_OPT} ${CCOMMON_OPT}")
+# TODO: not sure what PFLAGS is -hpa
+set(PFLAGS "${PFLAGS} ${COMMON_OPT} ${CCOMMON_OPT} -I${TOPDIR} -DPROFILE ${COMMON_PROF}")
+
+set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} ${COMMON_OPT} ${FCOMMON_OPT}")
+# TODO: not sure what FPFLAGS is -hpa
+set(FPFLAGS "${FPFLAGS} ${COMMON_OPT} ${FCOMMON_OPT} ${COMMON_PROF}")
+
+#For LAPACK Fortran codes.
+set(LAPACK_FFLAGS "${LAPACK_FFLAGS} ${CMAKE_Fortran_FLAGS}")
+set(LAPACK_FPFLAGS "${LAPACK_FPFLAGS} ${FPFLAGS}")
+
+#Disable -fopenmp for LAPACK Fortran codes on Windows.
+if (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
+  set(FILTER_FLAGS "-fopenmp;-mp;-openmp;-xopenmp=parralel")
+  foreach (FILTER_FLAG ${FILTER_FLAGS})
+    string(REPLACE ${FILTER_FLAG} "" LAPACK_FFLAGS ${LAPACK_FFLAGS})
+    string(REPLACE ${FILTER_FLAG} "" LAPACK_FPFLAGS ${LAPACK_FPFLAGS})
+  endforeach ()
+endif ()
+
+if ("${F_COMPILER}" STREQUAL "GFORTRAN")
+  # lapack-netlib is rife with uninitialized warnings -hpa
+  set(LAPACK_FFLAGS "${LAPACK_FFLAGS} -Wno-maybe-uninitialized")
+endif ()
+
+set(LAPACK_CFLAGS "${CMAKE_C_CFLAGS} -DHAVE_LAPACK_CONFIG_H")
+if (INTERFACE64)
+  set(LAPACK_CFLAGS "${LAPACK_CFLAGS} -DLAPACK_ILP64")
+endif ()
+
+if (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
+  set(LAPACK_CFLAGS "${LAPACK_CFLAGS} -DOPENBLAS_OS_WINDOWS")
+endif ()
+
+if (${CMAKE_C_COMPILER} STREQUAL "LSB" OR ${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
+  set(LAPACK_CFLAGS "${LAPACK_CFLAGS} -DLAPACK_COMPLEX_STRUCTURE")
+endif ()
+
+if (NOT DEFINED SUFFIX)
+  set(SUFFIX o)
+endif ()
+
+if (NOT DEFINED PSUFFIX)
+  set(PSUFFIX po)
+endif ()
+
+if (NOT DEFINED LIBSUFFIX)
+  set(LIBSUFFIX a)
+endif ()
+
+if (DYNAMIC_ARCH)
+  if (DEFINED SMP)
+    set(LIBNAME "${LIBPREFIX}p${REVISION}.${LIBSUFFIX}")
+    set(LIBNAME_P	"${LIBPREFIX}p${REVISION}_p.${LIBSUFFIX}")
+  else ()
+    set(LIBNAME "${LIBPREFIX}${REVISION}.${LIBSUFFIX}")
+    set(LIBNAME_P	"${LIBPREFIX}${REVISION}_p.${LIBSUFFIX}")
+  endif ()
+else ()
+  if (DEFINED SMP)
+    set(LIBNAME "${LIBPREFIX}_${LIBCORE}p${REVISION}.${LIBSUFFIX}")
+    set(LIBNAME_P	"${LIBPREFIX}_${LIBCORE}p${REVISION}_p.${LIBSUFFIX}")
+  else ()
+    set(LIBNAME	"${LIBPREFIX}_${LIBCORE}${REVISION}.${LIBSUFFIX}")
+    set(LIBNAME_P	"${LIBPREFIX}_${LIBCORE}${REVISION}_p.${LIBSUFFIX}")
+  endif ()
+endif ()
+
+
+set(LIBDLLNAME "${LIBPREFIX}.dll")
+set(LIBSONAME "${LIBNAME}.${LIBSUFFIX}.so")
+set(LIBDYNNAME "${LIBNAME}.${LIBSUFFIX}.dylib")
+set(LIBDEFNAME "${LIBNAME}.${LIBSUFFIX}.def")
+set(LIBEXPNAME "${LIBNAME}.${LIBSUFFIX}.exp")
+set(LIBZIPNAME "${LIBNAME}.${LIBSUFFIX}.zip")
+
+set(LIBS "${CMAKE_SOURCE_DIR}/${LIBNAME}")
+set(LIBS_P "${CMAKE_SOURCE_DIR}/${LIBNAME_P}")
+
+
+set(LIB_COMPONENTS BLAS)
+if (NOT NO_CBLAS)
+  set(LIB_COMPONENTS "${LIB_COMPONENTS} CBLAS")
+endif ()
+
+if (NOT NO_LAPACK)
+  set(LIB_COMPONENTS "${LIB_COMPONENTS} LAPACK")
+  if (NOT NO_LAPACKE)
+    set(LIB_COMPONENTS "${LIB_COMPONENTS} LAPACKE")
+  endif ()
+endif ()
+
+if (ONLY_CBLAS)
+  set(LIB_COMPONENTS CBLAS)
+endif ()
+
+#export OSNAME
+#export ARCH
+#export CORE
+#export LIBCORE
+#export PGCPATH
+#export CONFIG
+#export CC
+#export FC
+#export BU
+#export FU
+#export NEED2UNDERSCORES
+#export USE_THREAD
+#export NUM_THREADS
+#export NUM_CORES
+#export SMP
+#export MAKEFILE_RULE
+#export NEED_PIC
+#export BINARY
+#export BINARY32
+#export BINARY64
+#export F_COMPILER
+#export C_COMPILER
+#export USE_OPENMP
+#export CROSS
+#export CROSS_SUFFIX
+#export NOFORTRAN
+#export NO_FBLAS
+#export EXTRALIB
+#export CEXTRALIB
+#export FEXTRALIB
+#export HAVE_SSE
+#export HAVE_SSE2
+#export HAVE_SSE3
+#export HAVE_SSSE3
+#export HAVE_SSE4_1
+#export HAVE_SSE4_2
+#export HAVE_SSE4A
+#export HAVE_SSE5
+#export HAVE_AVX
+#export HAVE_VFP
+#export HAVE_VFPV3
+#export HAVE_VFPV4
+#export HAVE_NEON
+#export KERNELDIR
+#export FUNCTION_PROFILE
+#export TARGET_CORE
+#
+#export SGEMM_UNROLL_M
+#export SGEMM_UNROLL_N
+#export DGEMM_UNROLL_M
+#export DGEMM_UNROLL_N
+#export QGEMM_UNROLL_M
+#export QGEMM_UNROLL_N
+#export CGEMM_UNROLL_M
+#export CGEMM_UNROLL_N
+#export ZGEMM_UNROLL_M
+#export ZGEMM_UNROLL_N
+#export XGEMM_UNROLL_M
+#export XGEMM_UNROLL_N
+#export CGEMM3M_UNROLL_M
+#export CGEMM3M_UNROLL_N
+#export ZGEMM3M_UNROLL_M
+#export ZGEMM3M_UNROLL_N
+#export XGEMM3M_UNROLL_M
+#export XGEMM3M_UNROLL_N
+
+
+#if (USE_CUDA)
+#  export CUDADIR
+#  export CUCC
+#  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/utils.cmake

@@ -0,0 +1,342 @@
+# Functions to help with the OpenBLAS build
+
+# Reads string from getarch into CMake vars. Format of getarch vars is VARNAME=VALUE
+function(ParseGetArchVars GETARCH_IN)
+  string(REGEX MATCHALL "[0-9_a-zA-Z]+=[0-9_a-zA-Z]+" GETARCH_RESULT_LIST "${GETARCH_IN}")
+  foreach (GETARCH_LINE ${GETARCH_RESULT_LIST})
+    # split the line into var and value, then assign the value to a CMake var
+    string(REGEX MATCHALL "[0-9_a-zA-Z]+" SPLIT_VAR "${GETARCH_LINE}")
+    list(GET SPLIT_VAR 0 VAR_NAME)
+    list(GET SPLIT_VAR 1 VAR_VALUE)
+    set(${VAR_NAME} ${VAR_VALUE} PARENT_SCOPE)
+  endforeach ()
+endfunction ()
+
+# Reads a Makefile into CMake vars.
+macro(ParseMakefileVars MAKEFILE_IN)
+  message(STATUS "Reading vars from ${MAKEFILE_IN}...")
+  file(STRINGS ${MAKEFILE_IN} makefile_contents)
+  foreach (makefile_line ${makefile_contents})
+    string(REGEX MATCH "([0-9_a-zA-Z]+)[ \t]*=[ \t]*(.+)$" line_match "${makefile_line}")
+    if (NOT "${line_match}" STREQUAL "")
+      set(var_name ${CMAKE_MATCH_1})
+      set(var_value ${CMAKE_MATCH_2})
+      # check for Makefile variables in the string, e.g. $(TSUFFIX)
+      string(REGEX MATCHALL "\\$\\(([0-9_a-zA-Z]+)\\)" make_var_matches ${var_value})
+      foreach (make_var ${make_var_matches})
+        # strip out Makefile $() markup
+        string(REGEX REPLACE "\\$\\(([0-9_a-zA-Z]+)\\)" "\\1" make_var ${make_var})
+        # now replace the instance of the Makefile variable with the value of the CMake variable (note the double quote)
+        string(REPLACE "$(${make_var})" "${${make_var}}" var_value ${var_value})
+      endforeach ()
+      set(${var_name} ${var_value})
+    else ()
+      string(REGEX MATCH "include \\$\\(KERNELDIR\\)/(.+)$" line_match "${makefile_line}")
+      if (NOT "${line_match}" STREQUAL "")
+        ParseMakefileVars(${KERNELDIR}/${CMAKE_MATCH_1})
+      endif ()
+    endif ()
+  endforeach ()
+endmacro ()
+
+# Returns all combinations of the input list, as a list with colon-separated combinations
+# E.g. input of A B C returns A B C A:B A:C B:C
+# N.B. The input is meant to be a list, and to past a list to a function in CMake you must quote it (e.g. AllCombinations("${LIST_VAR}")).
+# #param absent_codes codes to use when an element is absent from a combination. For example, if you have TRANS;UNIT;UPPER you may want the code to be NNL when nothing is present.
+# @returns LIST_OUT a list of combinations
+#          CODES_OUT a list of codes corresponding to each combination, with N meaning the item is not present, and the first letter of the list item meaning it is presen
+function(AllCombinations list_in absent_codes_in)
+  list(LENGTH list_in list_count)
+  set(num_combos 1)
+  # subtract 1 since we will iterate from 0 to num_combos
+  math(EXPR num_combos "(${num_combos} << ${list_count}) - 1")
+  set(LIST_OUT "")
+  set(CODES_OUT "")
+  foreach (c RANGE 0 ${num_combos})
+
+    set(current_combo "")
+    set(current_code "")
+
+    # this is a little ridiculous just to iterate through a list w/ indices
+    math(EXPR last_list_index "${list_count} - 1")
+    foreach (list_index RANGE 0 ${last_list_index})
+      math(EXPR bit "1 << ${list_index}")
+      math(EXPR combo_has_bit "${c} & ${bit}")
+      list(GET list_in ${list_index} list_elem)
+      if (combo_has_bit)
+        if (current_combo)
+          set(current_combo "${current_combo}:${list_elem}")
+        else ()
+          set(current_combo ${list_elem})
+        endif ()
+        string(SUBSTRING ${list_elem} 0 1 code_char)
+      else ()
+        list(GET absent_codes_in ${list_index} code_char)
+      endif ()
+      set(current_code "${current_code}${code_char}")
+    endforeach ()
+
+    if (current_combo STREQUAL "")
+      list(APPEND LIST_OUT " ") # Empty set is a valid combination, but CMake isn't appending the empty string for some reason, use a space
+    else ()
+      list(APPEND LIST_OUT ${current_combo})
+    endif ()
+    list(APPEND CODES_OUT ${current_code})
+
+  endforeach ()
+
+  set(LIST_OUT ${LIST_OUT} PARENT_SCOPE)
+  set(CODES_OUT ${CODES_OUT} PARENT_SCOPE)
+endfunction ()
+
+# generates object files for each of the sources, using the BLAS naming scheme to pass the funciton name as a preprocessor definition
+# @param sources_in the source files to build from
+# @param defines_in (optional) preprocessor definitions that will be applied to all objects
+# @param name_in (optional) if this is set this name will be used instead of the filename. Use a * to indicate where the float character should go, if no star the character will be prepended.
+#                           e.g. with DOUBLE set, "i*max" will generate the name "idmax", and "max" will be "dmax"
+# @param replace_last_with replaces the last character in the filename with this string (e.g. symm_k should be symm_TU)
+# @param append_with appends the filename with this string (e.g. trmm_R should be trmm_RTUU or some other combination of characters)
+# @param no_float_type turns off the float type define for this build (e.g. SINGLE/DOUBLE/etc)
+# @param complex_filename_scheme some routines have separate source files for complex and non-complex float types.
+#                               0 - compiles for all types
+#                               1 - compiles the sources for non-complex types only (SINGLE/DOUBLE)
+#                               2 - compiles for complex types only (COMPLEX/DOUBLE COMPLEX)
+#                               3 - compiles for all types, but changes source names for complex by prepending z (e.g. axpy.c becomes zaxpy.c)
+#                               STRING - compiles only the given type (e.g. DOUBLE)
+function(GenerateNamedObjects sources_in)
+
+  if (DEFINED ARGV1)
+    set(defines_in ${ARGV1})
+  endif ()
+
+  if (DEFINED ARGV2 AND NOT "${ARGV2}" STREQUAL "")
+    set(name_in ${ARGV2})
+    # strip off extension for kernel files that pass in the object name.
+    get_filename_component(name_in ${name_in} NAME_WE)
+  endif ()
+
+  if (DEFINED ARGV3)
+    set(use_cblas ${ARGV3})
+  else ()
+    set(use_cblas false)
+  endif ()
+
+  if (DEFINED ARGV4)
+    set(replace_last_with ${ARGV4})
+  endif ()
+
+  if (DEFINED ARGV5)
+    set(append_with ${ARGV5})
+  endif ()
+
+  if (DEFINED ARGV6)
+    set(no_float_type ${ARGV6})
+  else ()
+    set(no_float_type false)
+  endif ()
+
+  if (no_float_type)
+    set(float_list "DUMMY") # still need to loop once
+  else ()
+    set(float_list "${FLOAT_TYPES}")
+  endif ()
+
+  set(real_only false)
+  set(complex_only false)
+  set(mangle_complex_sources false)
+  if (DEFINED ARGV7 AND NOT "${ARGV7}" STREQUAL "")
+    if (${ARGV7} EQUAL 1)
+      set(real_only true)
+    elseif (${ARGV7} EQUAL 2)
+      set(complex_only true)
+    elseif (${ARGV7} EQUAL 3)
+      set(mangle_complex_sources true)
+    elseif (NOT ${ARGV7} EQUAL 0)
+      set(float_list ${ARGV7})
+    endif ()
+  endif ()
+
+  if (complex_only)
+    list(REMOVE_ITEM float_list "SINGLE")
+    list(REMOVE_ITEM float_list "DOUBLE")
+  elseif (real_only)
+    list(REMOVE_ITEM float_list "COMPLEX")
+    list(REMOVE_ITEM float_list "ZCOMPLEX")
+  endif ()
+
+  set(float_char "")
+  set(OBJ_LIST_OUT "")
+  foreach (float_type ${float_list})
+    foreach (source_file ${sources_in})
+
+      if (NOT no_float_type)
+        string(SUBSTRING ${float_type} 0 1 float_char)
+        string(TOLOWER ${float_char} float_char)
+      endif ()
+
+      if (NOT name_in)
+        get_filename_component(source_name ${source_file} NAME_WE)
+        set(obj_name "${float_char}${source_name}")
+      else ()
+        # replace * with float_char
+        if (${name_in} MATCHES "\\*")
+          string(REPLACE "*" ${float_char} obj_name ${name_in})
+        else ()
+          set(obj_name "${float_char}${name_in}")
+        endif ()
+      endif ()
+
+      if (replace_last_with)
+        string(REGEX REPLACE ".$" ${replace_last_with} obj_name ${obj_name})
+      else ()
+        set(obj_name "${obj_name}${append_with}")
+      endif ()
+
+      # now add the object and set the defines
+      set(obj_defines ${defines_in})
+
+      if (use_cblas)
+        set(obj_name "cblas_${obj_name}")
+        list(APPEND obj_defines "CBLAS")
+      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=\"${obj_name}\"")
+      if (${float_type} STREQUAL "DOUBLE" OR ${float_type} STREQUAL "ZCOMPLEX")
+        list(APPEND obj_defines "DOUBLE")
+      endif ()
+      if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
+        list(APPEND obj_defines "COMPLEX")
+        if (mangle_complex_sources)
+          # add a z to the filename
+          get_filename_component(source_name ${source_file} NAME)
+          get_filename_component(source_dir ${source_file} DIRECTORY)
+          string(REPLACE ${source_name} "z${source_name}" source_file ${source_file})
+        endif ()
+      endif ()
+
+      if (VERBOSE_GEN)
+        message(STATUS "${obj_name}:${source_file}")
+        message(STATUS "${obj_defines}")
+      endif ()
+
+      # create a copy of the source to avoid duplicate obj filename problem with ar.exe
+      get_filename_component(source_extension ${source_file} EXT)
+      set(new_source_file "${CMAKE_CURRENT_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/${obj_name}${source_extension}")
+      if (IS_ABSOLUTE ${source_file})
+        set(old_source_file ${source_file})
+      else ()
+        set(old_source_file "${CMAKE_CURRENT_LIST_DIR}/${source_file}")
+      endif ()
+
+      string(REPLACE ";" "\n#define " define_source "${obj_defines}")
+      string(REPLACE "=" " " define_source "${define_source}")
+      file(WRITE ${new_source_file} "#define ${define_source}\n#include \"${old_source_file}\"")
+      list(APPEND SRC_LIST_OUT ${new_source_file})
+
+    endforeach ()
+  endforeach ()
+
+  list(APPEND OPENBLAS_SRC ${SRC_LIST_OUT})
+  set(OPENBLAS_SRC ${OPENBLAS_SRC} PARENT_SCOPE)
+endfunction ()
+
+# generates object files for each of the sources for each of the combinations of the preprocessor definitions passed in
+# @param sources_in the source files to build from
+# @param defines_in the preprocessor definitions that will be combined to create the object files
+# @param all_defines_in (optional) preprocessor definitions that will be applied to all objects
+# @param replace_scheme If 1, replace the "k" in the filename with the define combo letters. E.g. symm_k.c with TRANS and UNIT defined will be symm_TU.
+#                  If 0, it will simply append the code, e.g. symm_L.c with TRANS and UNIT will be symm_LTU.
+#                  If 2, it will append the code with an underscore, e.g. symm.c with TRANS and UNIT will be symm_TU.
+#                  If 3, it will insert the code *around* the last character with an underscore, e.g. symm_L.c with TRANS and UNIT will be symm_TLU (required by BLAS level2 objects).
+#                  If 4, it will insert the code before the last underscore. E.g. trtri_U_parallel with TRANS will be trtri_UT_parallel
+# @param alternate_name replaces the source name as the object name (define codes are still appended)
+# @param no_float_type turns off the float type define for this build (e.g. SINGLE/DOUBLE/etc)
+# @param complex_filename_scheme see GenerateNamedObjects
+function(GenerateCombinationObjects sources_in defines_in absent_codes_in all_defines_in replace_scheme)
+
+  set(alternate_name_in "")
+  if (DEFINED ARGV5)
+    set(alternate_name_in ${ARGV5})
+  endif ()
+
+  set(no_float_type false)
+  if (DEFINED ARGV6)
+    set(no_float_type ${ARGV6})
+  endif ()
+
+  set(complex_filename_scheme "")
+  if (DEFINED ARGV7)
+    set(complex_filename_scheme ${ARGV7})
+  endif ()
+
+  AllCombinations("${defines_in}" "${absent_codes_in}")
+  set(define_combos ${LIST_OUT})
+  set(define_codes ${CODES_OUT})
+
+  list(LENGTH define_combos num_combos)
+  math(EXPR num_combos "${num_combos} - 1")
+
+  foreach (c RANGE 0 ${num_combos})
+
+    list(GET define_combos ${c} define_combo)
+    list(GET define_codes ${c} define_code)
+
+    foreach (source_file ${sources_in})
+
+      set(alternate_name ${alternate_name_in})
+
+      # replace colon separated list with semicolons, this turns it into a CMake list that we can use foreach with
+      string(REPLACE ":" ";" define_combo ${define_combo})
+
+      # now add the object and set the defines
+      set(cur_defines ${define_combo})
+      if ("${cur_defines}" STREQUAL " ")
+        set(cur_defines ${all_defines_in})
+      else ()
+        list(APPEND cur_defines ${all_defines_in})
+      endif ()
+
+      set(replace_code "")
+      set(append_code "")
+      if (replace_scheme EQUAL 1)
+        set(replace_code ${define_code})
+      else ()
+        if (replace_scheme EQUAL 2)
+          set(append_code "_${define_code}")
+        elseif (replace_scheme EQUAL 3)
+          if ("${alternate_name}" STREQUAL "")
+            string(REGEX MATCH "[a-zA-Z]\\." last_letter ${source_file})
+          else ()
+            string(REGEX MATCH "[a-zA-Z]$" last_letter ${alternate_name})
+          endif ()
+          # first extract the last letter
+          string(SUBSTRING ${last_letter} 0 1 last_letter) # remove period from match
+          # break the code up into the first letter and the remaining (should only be 2 anyway)
+          string(SUBSTRING ${define_code} 0 1 define_code_first)
+          string(SUBSTRING ${define_code} 1 -1 define_code_second)
+          set(replace_code "${define_code_first}${last_letter}${define_code_second}")
+        elseif (replace_scheme EQUAL 4)
+          # insert code before the last underscore and pass that in as the alternate_name
+          if ("${alternate_name}" STREQUAL "")
+            get_filename_component(alternate_name ${source_file} NAME_WE)
+          endif ()
+          set(extra_underscore "")
+          # check if filename has two underscores, insert another if not (e.g. getrs_parallel needs to become getrs_U_parallel not getrsU_parallel)
+          string(REGEX MATCH "_[a-zA-Z]+_" underscores ${alternate_name})
+          string(LENGTH "${underscores}" underscores)
+          if (underscores EQUAL 0)
+            set(extra_underscore "_")
+          endif ()
+          string(REGEX REPLACE "(.+)(_[^_]+)$" "\\1${extra_underscore}${define_code}\\2" alternate_name ${alternate_name})
+        else()
+          set(append_code ${define_code}) # replace_scheme should be 0
+        endif ()
+      endif ()
+
+      GenerateNamedObjects("${source_file}" "${cur_defines}" "${alternate_name}" false "${replace_code}" "${append_code}" "${no_float_type}" "${complex_filename_scheme}")
+    endforeach ()
+  endforeach ()
+
+  set(OPENBLAS_SRC ${OPENBLAS_SRC} PARENT_SCOPE)
+endfunction ()
+

common.h

@@ -82,7 +82,10 @@ extern "C" {
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+
+#if !defined(_MSC_VER)
 #include <unistd.h>
+#endif
 
 #ifdef OS_LINUX
 #include <malloc.h>
@@ -313,8 +316,12 @@ typedef int blasint;
 #endif
 
 #if defined(OS_WINDOWS)
+#if defined(_MSC_VER) && !defined(__clang__)
+#define YIELDING    YieldProcessor()
+#else
 #define YIELDING	SwitchToThread()
 #endif
+#endif
 
 #if defined(ARMV7) || defined(ARMV6) || defined(ARMV8) || defined(ARMV5)
 #define YIELDING        asm volatile ("nop;nop;nop;nop;nop;nop;nop;nop; \n");

common_level1.h

@@ -47,12 +47,12 @@ double dsdot_k(BLASLONG, float   *, BLASLONG, float *, BLASLONG);
 double  ddot_k(BLASLONG, double  *, BLASLONG, double  *, BLASLONG);
 xdouble qdot_k(BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
 
-float   _Complex cdotc_k (BLASLONG, float  *, BLASLONG, float  *, BLASLONG);
-float   _Complex cdotu_k (BLASLONG, float  *, BLASLONG, float  *, BLASLONG);
-double  _Complex zdotc_k (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-double  _Complex zdotu_k (BLASLONG, double *, BLASLONG, double *, BLASLONG);
-xdouble _Complex xdotc_k (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
-xdouble _Complex xdotu_k (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
+openblas_complex_float cdotc_k (BLASLONG, float  *, BLASLONG, float  *, BLASLONG);
+openblas_complex_float cdotu_k (BLASLONG, float  *, BLASLONG, float  *, BLASLONG);
+openblas_complex_double zdotc_k (BLASLONG, double *, BLASLONG, double *, BLASLONG);
+openblas_complex_double zdotu_k (BLASLONG, double *, BLASLONG, double *, BLASLONG);
+openblas_complex_xdouble xdotc_k (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
+openblas_complex_xdouble xdotu_k (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
 
 int    saxpy_k (BLASLONG, BLASLONG, BLASLONG, float,
 	       float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG);

common_x86.h

@@ -56,41 +56,65 @@ static void __inline blas_lock(volatile BLASULONG *address){
   do {
     while (*address) {YIELDING;};
 
+#if defined(_MSC_VER) && !defined(__clang__)
+	// use intrinsic instead of inline assembly
+	ret = _InterlockedExchange(address, 1);
+	// inline assembly
+	/*__asm {
+		mov eax, address
+		mov ebx, 1
+		xchg [eax], ebx
+		mov ret, ebx
+	}*/
+#else
     __asm__ __volatile__(
 			 "xchgl %0, %1\n"
 			 : "=r"(ret), "=m"(*address)
 			 : "0"(1), "m"(*address)
 			 : "memory");
+#endif
 
   } while (ret);
 
 }
 
 static __inline unsigned long long rpcc(void){
+#if defined(_MSC_VER) && !defined(__clang__)
+  return __rdtsc(); // use MSVC intrinsic
+#else
   unsigned int a, d;
 
   __asm__ __volatile__ ("rdtsc" : "=a" (a), "=d" (d));
 
   return ((unsigned long long)a + ((unsigned long long)d << 32));
+#endif
 };
 
 static __inline unsigned long getstackaddr(void){
+#if defined(_MSC_VER) && !defined(__clang__)
+  return (unsigned long)_ReturnAddress(); // use MSVC intrinsic
+#else
   unsigned long addr;
 
   __asm__ __volatile__ ("mov %%esp, %0"
 			 : "=r"(addr) : : "memory");
 
   return addr;
+#endif
 };
 
 
 static __inline long double sqrt_long(long double val) {
+#if defined(_MSC_VER) && !defined(__clang__)
+  return sqrt(val); // not sure if this will use fsqrt
+#else
   long double result;
 
   __asm__ __volatile__ ("fldt %1\n"
 		    "fsqrt\n"
 		    "fstpt %0\n" : "=m" (result) : "m"(val));
   return result;
+#endif
 }
 
 #define SQRT(a)  sqrt_long(a)
@@ -100,7 +124,7 @@ void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx);
 
 #define WHEREAMI
 
-static inline int WhereAmI(void){
+static __inline int WhereAmI(void){
   int eax, ebx, ecx, edx;
   int apicid;
 
@@ -146,9 +170,14 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 
   y = blas_quick_divide_table[y];
 
+#if defined(_MSC_VER) && !defined(__clang__)
+  (void*)result;
+  return x*y;
+#else
   __asm__ __volatile__  ("mull %0" :"=d" (result) :"a"(x), "0" (y));
 
   return result;
+#endif
 }
 #endif
 
@@ -284,8 +313,12 @@ REALNAME:
 
 #define PROFCODE
 
+#ifdef __clang__
+#define EPILOGUE .end
+#else
 #define EPILOGUE .end	 REALNAME
 #endif
+#endif
 
 #if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__)
 #define PROLOGUE \

cpuid.h

@@ -39,6 +39,10 @@
 #ifndef CPUID_H
 #define CPUID_H
 
+#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86) || defined(_M_X64)
+#define INTEL_AMD
+#endif
+
 #define VENDOR_INTEL      1
 #define VENDOR_UMC        2
 #define VENDOR_AMD        3
@@ -59,7 +63,7 @@
 #define FAMILY_PM     7
 #define FAMILY_IA64   8
 
-#if defined(__i386__) || defined(__x86_64__)
+#ifdef INTEL_AMD
 #define GET_EXFAMILY  1
 #define GET_EXMODEL   2
 #define GET_TYPE      3

cpuid_x86.c

@@ -40,6 +40,12 @@
 #include <string.h>
 #include "cpuid.h"
 
+#if defined(_MSC_VER) && !defined(__clang__)
+#define C_INLINE __inline
+#else
+#define C_INLINE inline
+#endif
+
 /*
 #ifdef NO_AVX
 #define CPUTYPE_HASWELL CPUTYPE_NEHALEM
@@ -53,12 +59,26 @@
 #endif
 */
 
+#if defined(_MSC_VER) && !defined(__clang__)
+
+void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx)
+{
+  int cpuInfo[4] = {-1};
+  __cpuid(cpuInfo, op);
+  *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);
 #else
-static inline void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
+static C_INLINE void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
 #if defined(__i386__) && defined(__PIC__)
   __asm__ __volatile__
     ("mov %%ebx, %%edi;"
@@ -115,14 +135,16 @@ void cpuid(unsigned int op, unsigned int *eax, unsigned int *ebx, unsigned int *
 
 #endif
 
-static inline int have_cpuid(void){
+#endif // _MSC_VER
+
+static C_INLINE int have_cpuid(void){
   int eax, ebx, ecx, edx;
 
   cpuid(0, &eax, &ebx, &ecx, &edx);
   return eax;
 }
 
-static inline int have_excpuid(void){
+static C_INLINE int have_excpuid(void){
   int eax, ebx, ecx, edx;
 
   cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
@@ -130,10 +152,14 @@ static inline int have_excpuid(void){
 }
 
 #ifndef NO_AVX
-static inline void xgetbv(int op, int * eax, int * edx){
+static C_INLINE void xgetbv(int op, int * eax, int * edx){
   //Use binary code for xgetbv
+#if defined(_MSC_VER) && !defined(__clang__)
+  *eax = __xgetbv(op);
+#else
   __asm__ __volatile__
     (".byte 0x0f, 0x01, 0xd0": "=a" (*eax), "=d" (*edx) : "c" (op) : "cc");
+#endif
 }
 #endif
 

driver/level2/CMakeLists.txt

@@ -0,0 +1,129 @@
+
+include_directories(${CMAKE_SOURCE_DIR})
+
+# sources that need to be compiled twice, once with no flags and once with LOWER
+set(UL_SOURCES
+  sbmv_k.c
+  spmv_k.c
+  spr_k.c
+  spr2_k.c
+  syr_k.c
+  syr2_k.c
+)
+
+# sources that need to be compiled several times, for UNIT, TRANSA
+set(U_SOURCES
+  trmv_U.c
+  tbmv_U.c
+  tbsv_U.c
+  tpmv_U.c
+  tpsv_U.c
+  trsv_U.c
+)
+
+set(L_SOURCES
+  trmv_L.c
+  tbmv_L.c
+  tbsv_L.c
+  tpmv_L.c
+  tpsv_L.c
+  trsv_L.c
+)
+
+set(UL_SMP_SOURCES
+  symv_thread.c
+  syr_thread.c
+  syr2_thread.c
+  spr_thread.c
+  spr2_thread.c
+  spmv_thread.c
+  sbmv_thread.c
+)
+
+set(NU_SMP_SOURCES
+  trmv_thread.c
+  tpmv_thread.c
+  tbmv_thread.c
+)
+
+# objects that need LOWER set
+GenerateCombinationObjects("${UL_SOURCES}" "LOWER" "U" "" 1 "" "" 3)
+
+# gbmv uses a lowercase n and t
+GenerateNamedObjects("gbmv_k.c" "" "gbmv_n" false "" "" "" 3)
+GenerateNamedObjects("gbmv_k.c" "TRANS" "gbmv_t" false "" "" "" 3)
+
+# special defines for complex
+foreach (float_type ${FLOAT_TYPES})
+
+  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})
+
+    GenerateNamedObjects("gbmv_thread.c" "" "gbmv_thread_n" false "" "" false ${float_type})
+    GenerateNamedObjects("gbmv_thread.c" "TRANSA" "gbmv_thread_t" false "" "" false ${float_type})
+  endif ()
+
+  if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
+
+    foreach (u_source ${U_SOURCES})
+      string(REGEX MATCH "[a-z]+" op_name ${u_source})
+      GenerateCombinationObjects("z${u_source}" "UNIT" "N" "TRANSA=1" 0 "${op_name}_NU" false ${float_type})
+      GenerateCombinationObjects("z${u_source}" "UNIT" "N" "TRANSA=2" 0 "${op_name}_TL" false ${float_type})
+      GenerateCombinationObjects("z${u_source}" "UNIT" "N" "TRANSA=3" 0 "${op_name}_RU" false ${float_type})
+      GenerateCombinationObjects("z${u_source}" "UNIT" "N" "TRANSA=4" 0 "${op_name}_CL" false ${float_type})
+    endforeach ()
+
+    foreach (l_source ${L_SOURCES})
+      string(REGEX MATCH "[a-z]+" op_name ${l_source})
+      GenerateCombinationObjects("z${l_source}" "UNIT" "N" "TRANSA=1" 0 "${op_name}_NL" false ${float_type})
+      GenerateCombinationObjects("z${l_source}" "UNIT" "N" "TRANSA=2" 0 "${op_name}_TU" false ${float_type})
+      GenerateCombinationObjects("z${l_source}" "UNIT" "N" "TRANSA=3" 0 "${op_name}_RL" false ${float_type})
+      GenerateCombinationObjects("z${l_source}" "UNIT" "N" "TRANSA=4" 0 "${op_name}_CU" false ${float_type})
+    endforeach ()
+
+    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})
+      GenerateNamedObjects("gemv_thread.c" "XCONJ" "gemv_thread_o" false "" "" false ${float_type})
+      GenerateNamedObjects("gemv_thread.c" "XCONJ;TRANSA" "gemv_thread_u" false "" "" false ${float_type})
+      GenerateNamedObjects("gemv_thread.c" "XCONJ;CONJ" "gemv_thread_s" false "" "" false ${float_type})
+      GenerateNamedObjects("gemv_thread.c" "XCONJ;CONJ;TRANSA" "gemv_thread_d" false "" "" false ${float_type})
+
+      GenerateNamedObjects("gbmv_thread.c" "CONJ" "gbmv_thread_r" false "" "" false ${float_type})
+      GenerateNamedObjects("gbmv_thread.c" "CONJ;TRANSA" "gbmv_thread_c" false "" "" false ${float_type})
+      GenerateNamedObjects("gbmv_thread.c" "XCONJ" "gbmv_thread_o" false "" "" false ${float_type})
+      GenerateNamedObjects("gbmv_thread.c" "XCONJ;TRANSA" "gbmv_thread_u" false "" "" false ${float_type})
+      GenerateNamedObjects("gbmv_thread.c" "XCONJ;CONJ" "gbmv_thread_s" false "" "" false ${float_type})
+      GenerateNamedObjects("gbmv_thread.c" "XCONJ;CONJ;TRANSA" "gbmv_thread_d" false "" "" false ${float_type})
+
+      GenerateNamedObjects("ger_thread.c" "" "ger_thread_U" false "" "" false ${float_type})
+      GenerateNamedObjects("ger_thread.c" "CONJ" "ger_thread_C" false "" "" false ${float_type})
+      GenerateNamedObjects("ger_thread.c" "XCONJ" "ger_thread_V" false "" "" false ${float_type})
+      GenerateNamedObjects("ger_thread.c" "XCONJ;CONJ" "ger_thread_D" false "" "" false ${float_type})
+
+      foreach (nu_smp_src ${NU_SMP_SOURCES})
+        string(REGEX MATCH "[a-z]+_[a-z]+" op_name ${nu_smp_src})
+        GenerateCombinationObjects("${nu_smp_src}" "LOWER;UNIT" "U;N" "TRANSA=1" 0 "${op_name}_N" false ${float_type})
+        GenerateCombinationObjects("${nu_smp_src}" "LOWER;UNIT" "U;N" "TRANSA=2" 0 "${op_name}_T" false ${float_type})
+        GenerateCombinationObjects("${nu_smp_src}" "LOWER;UNIT" "U;N" "TRANSA=3" 0 "${op_name}_R" false ${float_type})
+        GenerateCombinationObjects("${nu_smp_src}" "LOWER;UNIT" "U;N" "TRANSA=4" 0 "${op_name}_C" false ${float_type})
+      endforeach ()
+    endif ()
+
+  else ()
+    # N.B. BLAS wants to put the U/L from the filename in the *MIDDLE*
+    GenerateCombinationObjects("${U_SOURCES};${L_SOURCES}" "TRANSA;UNIT" "N;N" "" 3 "" false ${float_type})
+    if (SMP)
+      GenerateNamedObjects("ger_thread.c" "" "" false "" "" false ${float_type})
+      GenerateCombinationObjects("${NU_SMP_SOURCES}" "TRANSA;LOWER;UNIT" "N;U;N" "" 2 "" false ${float_type})
+    endif ()
+  endif ()
+endforeach ()
+
+if (SMP)
+  GenerateCombinationObjects("${UL_SMP_SOURCES}" "LOWER" "U" "" 2)
+endif ()
+
+add_library(driver_level2 OBJECT ${OPENBLAS_SRC})

driver/level3/CMakeLists.txt

@@ -0,0 +1,81 @@
+include_directories(${CMAKE_SOURCE_DIR})
+
+set(USE_GEMM3M 0)
+
+if (DEFINED ARCH)
+  if (${ARCH} STREQUAL "x86" OR ${ARCH} STREQUAL "x86_64" OR ${ARCH} STREQUAL "ia64" OR ${ARCH} STREQUAL "MIPS")
+    set(USE_GEMM3M 1)
+  endif ()
+endif ()
+
+# 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
+
+# loop through gemm.c defines
+set(GEMM_DEFINES NN NT TN TT)
+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_${GEMM_DEFINE_LC}" 0)
+  if (SMP AND NOT USE_SIMPLE_THREADED_LEVEL3)
+    GenerateNamedObjects("gemm.c" "${GEMM_DEFINE};THREADED_LEVEL3" "gemm_thread_${GEMM_DEFINE_LC}" 0)
+  endif ()
+endforeach ()
+
+GenerateCombinationObjects("trmm_L.c;trmm_R.c;trsm_L.c;trsm_R.c" "TRANS;UPPER;UNIT" "N;L;N" "" 0)
+GenerateCombinationObjects("symm_k.c" "RSIDE;LOWER" "L;U" "NN" 1)
+GenerateCombinationObjects("syrk_k.c;syr2k_k.c" "LOWER;TRANS" "U;N" "" 1)
+GenerateCombinationObjects("syrk_kernel.c;syr2k_kernel.c" "LOWER" "U" "" 2)
+
+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)
+
+  if (NOT USE_SIMPLE_THREADED_LEVEL3)
+    GenerateCombinationObjects("syrk_k.c" "LOWER;TRANS" "U;N" "THREADED_LEVEL3" 2 "syrk_thread")
+    GenerateCombinationObjects("symm_k.c" "RSIDE;LOWER" "L;U" "THREADED_LEVEL3;NN" 2 "symm_thread")
+  endif ()
+endif ()
+
+foreach (float_type ${FLOAT_TYPES})
+  if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
+    GenerateCombinationObjects("zherk_kernel.c" "LOWER;CONJ" "U;N" "HERK" 2 "herk_kernel" false ${float_type})
+    # TRANS needs to be set/unset when CONJ is set/unset, so can't use it as a combination
+    GenerateCombinationObjects("zherk_k.c" "LOWER" "U" "HERK" 3 "herk_N" false ${float_type})
+    GenerateCombinationObjects("zherk_k.c" "LOWER" "U" "HERK;TRANS;CONJ" 3 "herk_C" false ${float_type})
+    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})
+    # Need to set CONJ for trmm and trsm
+    GenerateCombinationObjects("trmm_L.c" "UPPER;UNIT" "L;N" "CONJ" 0 "trmm_LR" false ${float_type})
+    GenerateCombinationObjects("trmm_L.c" "UPPER;UNIT" "L;N" "TRANS;CONJ" 0 "trmm_LC" false ${float_type})
+    GenerateCombinationObjects("trmm_R.c" "UPPER;UNIT" "L;N" "CONJ" 0 "trmm_RR" false ${float_type})
+    GenerateCombinationObjects("trmm_R.c" "UPPER;UNIT" "L;N" "TRANS;CONJ" 0 "trmm_RC" false ${float_type})
+    GenerateCombinationObjects("trsm_L.c" "UPPER;UNIT" "L;N" "CONJ" 0 "trsm_LR" false ${float_type})
+    GenerateCombinationObjects("trsm_L.c" "UPPER;UNIT" "L;N" "TRANS;CONJ" 0 "trsm_LC" false ${float_type})
+    GenerateCombinationObjects("trsm_R.c" "UPPER;UNIT" "L;N" "CONJ" 0 "trsm_RR" false ${float_type})
+    GenerateCombinationObjects("trsm_R.c" "UPPER;UNIT" "L;N" "TRANS;CONJ" 0 "trsm_RC" false ${float_type})
+    # special gemm defines for complex
+    foreach (gemm_define ${GEMM_COMPLEX_DEFINES})
+      string(TOLOWER ${gemm_define} gemm_define_LC)
+      GenerateNamedObjects("gemm.c" "${gemm_define}" "gemm_${gemm_define_LC}" false "" "" false ${float_type})
+      if (SMP AND NOT USE_SIMPLE_THREADED_LEVEL3)
+        GenerateNamedObjects("gemm.c" "${gemm_define};THREADED_LEVEL3" "gemm_thread_${gemm_define_LC}" false "" "" false ${float_type})
+      endif ()
+    endforeach ()
+  endif ()
+endforeach ()
+
+#HPLOBJS =
+#	dgemm_nn.c dgemm_nt.c dgemm_tn.c dgemm_tt.c
+#	dtrsm_LNUU.c dtrsm_LNUN.c dtrsm_LNLU.c dtrsm_LNLN.c
+#	dtrsm_LTUU.c dtrsm_LTUN.c dtrsm_LTLU.c dtrsm_LTLN.c
+#	dtrsm_RNUU.c dtrsm_RNUN.c dtrsm_RNLU.c dtrsm_RNLN.c
+#	dtrsm_RTUU.c dtrsm_RTUN.c dtrsm_RTLU.c dtrsm_RTLN.c
+#
+#if (USE_SIMPLE_THREADED_LEVEL3)
+#  HPLOBJS += dgemm_thread_nn.c dgemm_thread_nt.c
+#	   dgemm_thread_tn.c dgemm_thread_tt.c
+#endif
+#
+
+add_library(driver_level3 OBJECT ${OPENBLAS_SRC})

driver/level3/syr2k_k.c

@@ -47,7 +47,7 @@
 #endif
 #endif
 
-static inline int syrk_beta(BLASLONG m_from, BLASLONG m_to, BLASLONG n_from, BLASLONG n_to, FLOAT *alpha, FLOAT *c, BLASLONG ldc) {
+static __inline int syrk_beta(BLASLONG m_from, BLASLONG m_to, BLASLONG n_from, BLASLONG n_to, FLOAT *alpha, FLOAT *c, BLASLONG ldc) {
 
   BLASLONG i;
 

driver/level3/syrk_k.c

@@ -49,7 +49,7 @@
 #endif
 #endif
 
-static inline int syrk_beta(BLASLONG m_from, BLASLONG m_to, BLASLONG n_from, BLASLONG n_to, FLOAT *alpha, FLOAT *c, BLASLONG ldc) {
+static __inline int syrk_beta(BLASLONG m_from, BLASLONG m_to, BLASLONG n_from, BLASLONG n_to, FLOAT *alpha, FLOAT *c, BLASLONG ldc) {
 
   BLASLONG i;
 

driver/level3/trmm_R.c

@@ -70,6 +70,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   BLASLONG ls, is, js;
   BLASLONG min_l, min_i, min_j;
   BLASLONG jjs, min_jj;
+#if !((!defined(UPPER) && !defined(TRANSA)) || (defined(UPPER) && defined(TRANSA)))
+  BLASLONG start_ls;
+#endif
 
   m = args -> m;
   n = args -> n;
@@ -226,8 +229,6 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   }
 
 #else
-  BLASLONG start_ls;
-
   for(js = n; js > 0; js -= GEMM_R){
     min_j = js;
     if (min_j > GEMM_R) min_j = GEMM_R;

driver/level3/trsm_L.c

@@ -76,6 +76,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   BLASLONG ls, is, js;
   BLASLONG min_l, min_i, min_j;
   BLASLONG jjs, min_jj;
+#if !((!defined(UPPER) && !defined(TRANSA)) || (defined(UPPER) && defined(TRANSA)))
+  BLASLONG start_is;
+#endif
 
   m = args -> m;
   n = args -> n;
@@ -178,8 +181,6 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
       }
     }
 #else
-    BLASLONG start_is;
-
     for(ls = m; ls > 0; ls -= GEMM_Q){
       min_l = ls;
       if (min_l > GEMM_Q) min_l = GEMM_Q;

driver/level3/trsm_R.c

@@ -75,6 +75,9 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   BLASLONG ls, is, js;
   BLASLONG min_l, min_i, min_j;
   BLASLONG jjs, min_jj;
+#if !((defined(UPPER) && !defined(TRANSA)) || (!defined(UPPER) && defined(TRANSA)))
+  BLASLONG start_ls;
+#endif
 
   m = args -> m;
   n = args -> n;
@@ -226,8 +229,6 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   }
 
 #else
-  BLASLONG start_ls;
-
   for(js = n; js > 0; js -= GEMM_R){
     min_j = js;
     if (min_j > GEMM_R) min_j = GEMM_R;

driver/others/CMakeLists.txt

@@ -0,0 +1,73 @@
+include_directories(${CMAKE_SOURCE_DIR})
+
+if (${CORE} STREQUAL "PPC440")
+  set(MEMORY memory_qalloc.c)
+else ()
+  set(MEMORY memory.c)
+endif ()
+
+if (SMP)
+
+  if (USE_OPENMP)
+    set(BLAS_SERVER blas_server_omp.c)
+  elseif (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
+    set(BLAS_SERVER blas_server_win32.c)
+  endif ()
+
+  if (NOT DEFINED BLAS_SERVER)
+    set(BLAS_SERVER blas_server.c)
+  endif ()
+
+  set(SMP_SOURCES
+    ${BLAS_SERVER}
+    divtable.c # TODO: Makefile has -UDOUBLE
+    blas_l1_thread.c
+  )
+
+  if (NOT NO_AFFINITY)
+    list(APPEND SMP_SOURCES init.c)
+  endif ()
+endif ()
+
+set(COMMON_SOURCES
+  xerbla.c
+  openblas_set_num_threads.c
+  openblas_error_handle.c
+)
+
+# these need to have NAME/CNAME set, so use GenerateNamedObjects, but don't use standard name mangling
+GenerateNamedObjects("abs.c" "" "c_abs" 0 "" "" 1 )
+GenerateNamedObjects("abs.c" "DOUBLE" "z_abs" 0 "" "" 1)
+GenerateNamedObjects("openblas_get_config.c;openblas_get_parallel.c" "" "" 0 "" "" 1)
+
+if (DYNAMIC_ARCH)
+  list(APPEND COMMON_SOURCES dynamic.c)
+else ()
+  list(APPEND COMMON_SOURCES parameter.c)
+endif ()
+
+#ifdef EXPRECISION
+#COMMONOBJS	+= x_abs.$(SUFFIX) qlamch.$(SUFFIX) qlamc3.$(SUFFIX)
+#endif
+#
+#ifdef QUAD_PRECISION
+#COMMONOBJS	+= addx.$(SUFFIX) mulx.$(SUFFIX)
+#endif
+#
+#ifdef USE_CUDA
+#COMMONOBJS	+= cuda_init.$(SUFFIX)
+#endif
+#
+#ifdef FUNCTION_PROFILE
+#COMMONOBJS	+= profile.$(SUFFIX)
+#endif
+
+#LIBOTHERS = libothers.$(LIBSUFFIX)
+
+#ifeq ($(DYNAMIC_ARCH), 1)
+#HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic.$(SUFFIX)
+#else
+#HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) parameter.$(SUFFIX)
+#endif
+
+add_library(driver_others OBJECT ${OPENBLAS_SRC} ${MEMORY} ${SMP_SOURCES} ${COMMON_SOURCES})

driver/others/memory.c

@@ -139,8 +139,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
 
+#if defined(_MSC_VER) && !defined(__clang__)
+#define CONSTRUCTOR __cdecl
+#define DESTRUCTOR __cdecl
+#else
 #define CONSTRUCTOR	__attribute__ ((constructor))
 #define DESTRUCTOR	__attribute__ ((destructor))
+#endif
 
 #ifdef DYNAMIC_ARCH
 gotoblas_t *gotoblas = NULL;
@@ -795,12 +800,12 @@ static void *alloc_hugetlb(void *address){
   
   if (LookupPrivilegeValue(NULL, SE_LOCK_MEMORY_NAME, &tp.Privileges[0].Luid) != TRUE) {
       CloseHandle(hToken);
-      return -1;
+      return (void*)-1;
   }
 
   if (AdjustTokenPrivileges(hToken, FALSE, &tp, 0, NULL, NULL) != TRUE) {
       CloseHandle(hToken);
-      return -1;
+      return (void*)-1;
   }
 
   map_address  = (void *)VirtualAlloc(address,
@@ -1399,6 +1404,28 @@ void DESTRUCTOR gotoblas_quit(void) {
 #endif
 }
 
+#if defined(_MSC_VER) && !defined(__clang__)
+BOOL APIENTRY DllMain(HMODULE hModule, DWORD  ul_reason_for_call, LPVOID lpReserved)
+{
+  switch (ul_reason_for_call)
+  {
+    case DLL_PROCESS_ATTACH:
+      gotoblas_init();
+      break;
+    case DLL_THREAD_ATTACH:
+      break;
+    case DLL_THREAD_DETACH:
+      break;
+    case DLL_PROCESS_DETACH:
+      gotoblas_quit();
+      break;
+    default:
+      break;
+  }
+  return TRUE;
+}
+#endif
+
 #if (defined(C_PGI) || (!defined(C_SUN) && defined(F_INTERFACE_SUN))) && (defined(ARCH_X86) || defined(ARCH_X86_64))
 /* Don't call me; this is just work around for PGI / Sun bug */
 void gotoblas_dummy_for_PGI(void) {

getarch.c

@@ -69,10 +69,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* or implied, of The University of Texas at Austin.                 */
 /*********************************************************************/
 
-#if defined(__WIN32__) || defined(__WIN64__) || defined(__CYGWIN32__) || defined(__CYGWIN64__)
+#if defined(__WIN32__) || defined(__WIN64__) || defined(__CYGWIN32__) || defined(__CYGWIN64__) || defined(_WIN32) || defined(_WIN64)
 #define OS_WINDOWS
 #endif
 
+#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86) || defined(_M_X64)
+#define INTEL_AMD
+#endif
+
 #include <stdio.h>
 #include <string.h>
 #ifdef OS_WINDOWS
@@ -829,7 +833,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define OPENBLAS_SUPPORTED
 #endif
 
-#if defined(__i386__) || (__x86_64__)
+#ifdef INTEL_AMD
 #include "cpuid_x86.c"
 #define OPENBLAS_SUPPORTED
 #endif
@@ -924,7 +928,7 @@ int main(int argc, char *argv[]){
 #ifdef FORCE
     printf("CORE=%s\n", CORENAME);
 #else
-#if defined(__i386__) || defined(__x86_64__) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__)
+#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__)
     printf("CORE=%s\n", get_corename());
 #endif
 #endif
@@ -944,7 +948,7 @@ int main(int argc, char *argv[]){
 #endif
 
 
-#if defined(__i386__) || defined(__x86_64__)
+#ifdef INTEL_AMD
 #ifndef FORCE
     get_sse();
 #else
@@ -1024,7 +1028,7 @@ int main(int argc, char *argv[]){
 #ifdef FORCE
     printf("#define CHAR_CORENAME \"%s\"\n", CORENAME);
 #else
-#if defined(__i386__) || defined(__x86_64__) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__)
+#if defined(INTEL_AMD) || defined(POWER) || defined(__mips__) || defined(__arm__) || defined(__aarch64__)
     printf("#define CHAR_CORENAME \"%s\"\n", get_corename());
 #endif
 #endif

interface/CMakeLists.txt

@@ -0,0 +1,120 @@
+
+include_directories(${CMAKE_SOURCE_DIR})
+
+set(BLAS1_SOURCES
+  copy.c
+  asum.c nrm2.c
+)
+
+set(BLAS1_REAL_ONLY_SOURCES
+  rotm.c rotmg.c # N.B. these do not have complex counterparts
+)
+
+# these will have 'z' prepended for the complex version
+set(BLAS1_MANGLED_SOURCES
+  axpy.c swap.c
+  scal.c
+  dot.c
+  rot.c rotg.c
+  axpby.c
+)
+
+# TODO: USE_NETLIB_GEMV shoudl switch gemv.c to netlib/*gemv.f
+# these all have 'z' sources for complex versions
+set(BLAS2_SOURCES
+  gemv.c ger.c
+  trsv.c trmv.c symv.c
+  syr.c syr2.c gbmv.c
+  sbmv.c spmv.c
+  spr.c spr2.c
+  tbsv.c tbmv.c
+  tpsv.c tpmv.c
+)
+
+# these do not have separate 'z' sources
+set(BLAS3_SOURCES
+  gemm.c symm.c
+  trsm.c syrk.c syr2k.c
+)
+
+set(BLAS3_MANGLED_SOURCES
+  omatcopy.c imatcopy.c
+)
+
+# generate the BLAS objs once with and once without cblas
+set (CBLAS_FLAGS "")
+
+if (NOT DEFINED NO_FBLAS)
+  list(APPEND CBLAS_FLAGS 0)
+endif ()
+
+if (NOT DEFINED NO_CBLAS)
+  list(APPEND CBLAS_FLAGS 1)
+endif ()
+
+foreach (CBLAS_FLAG ${CBLAS_FLAGS})
+
+  # TODO: don't compile complex sources with cblas for now, the naming schemes are all different and they will have to be handled separately from SINGLE/DOUBLE
+  set(DISABLE_COMPLEX 0)
+  set(MANGLE_COMPLEX 3)
+  if (CBLAS_FLAG EQUAL 1)
+    set(DISABLE_COMPLEX 1)
+    set(MANGLE_COMPLEX 1)
+  endif ()
+  GenerateNamedObjects("${BLAS1_SOURCES}" "" "" ${CBLAS_FLAG} "" "" false ${DISABLE_COMPLEX})
+  GenerateNamedObjects("${BLAS1_REAL_ONLY_SOURCES}" "" "" ${CBLAS_FLAG} "" "" false 1)
+  GenerateNamedObjects("${BLAS1_MANGLED_SOURCES}" "" "" ${CBLAS_FLAG} "" "" false ${MANGLE_COMPLEX})
+  GenerateNamedObjects("${BLAS2_SOURCES}" "" "" ${CBLAS_FLAG} "" "" false ${MANGLE_COMPLEX})
+  GenerateNamedObjects("${BLAS3_SOURCES}" "" "" ${CBLAS_FLAG} "" "" false ${DISABLE_COMPLEX})
+  GenerateNamedObjects("${BLAS3_MANGLED_SOURCES}" "" "" ${CBLAS_FLAG} "" "" false ${MANGLE_COMPLEX})
+
+  # trmm is trsm with a compiler flag set
+  GenerateNamedObjects("trsm.c" "TRMM" "trmm" ${CBLAS_FLAG})
+
+  # max and imax are compiled 4 times
+  GenerateNamedObjects("max.c" "" "" ${CBLAS_FLAG})
+  GenerateNamedObjects("max.c" "USE_ABS" "amax" ${CBLAS_FLAG})
+  GenerateNamedObjects("max.c" "USE_ABS;USE_MIN" "amin" ${CBLAS_FLAG})
+  GenerateNamedObjects("max.c" "USE_MIN" "min" ${CBLAS_FLAG})
+
+  GenerateNamedObjects("imax.c" "" "i*max" ${CBLAS_FLAG})
+  GenerateNamedObjects("imax.c" "USE_ABS" "i*amax" ${CBLAS_FLAG})
+  GenerateNamedObjects("imax.c" "USE_ABS;USE_MIN" "i*amin" ${CBLAS_FLAG})
+  GenerateNamedObjects("imax.c" "USE_MIN" "i*min" ${CBLAS_FLAG})
+
+endforeach ()
+
+# complex-specific sources
+foreach (float_type ${FLOAT_TYPES})
+  if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
+    GenerateNamedObjects("zger.c" "" "geru" false "" "" false ${float_type})
+    GenerateNamedObjects("zger.c" "CONJ" "gerc" false "" "" false ${float_type})
+  endif ()
+  if (${float_type} STREQUAL "COMPLEX")
+    GenerateNamedObjects("zscal.c" "SSCAL" "sscal" false "" "" false "COMPLEX")
+    GenerateNamedObjects("nrm2.c" "" "scnrm2" false "" "" true "COMPLEX")
+  endif ()
+  if (${float_type} STREQUAL "ZCOMPLEX")
+    GenerateNamedObjects("zscal.c" "SSCAL" "dscal" false "" "" false "ZCOMPLEX")
+    GenerateNamedObjects("nrm2.c" "" "dznrm2" false "" "" true "ZCOMPLEX")
+  endif ()
+endforeach ()
+
+
+if (NOT DEFINED NO_LAPACK)
+  set(LAPACK_SOURCES
+    lapack/gesv.c
+  )
+
+  # prepend z for complex versions
+  set(LAPACK_MANGLED_SOURCES
+    lapack/getrf.c lapack/getrs.c lapack/potrf.c lapack/getf2.c
+    lapack/potf2.c lapack/laswp.c lapack/lauu2.c
+    lapack/lauum.c lapack/trti2.c lapack/trtri.c
+  )
+
+  GenerateNamedObjects("${LAPACK_SOURCES}")
+  GenerateNamedObjects("${LAPACK_MANGLED_SOURCES}" "" "" 0 "" "" 0 3)
+endif ()
+
+add_library(interface OBJECT ${OPENBLAS_SRC})

interface/gemm.c

@@ -121,6 +121,9 @@ void NAME(char *TRANSA, char *TRANSB,
   FLOAT *sa, *sb;
 
 #ifdef SMP
+  int nthreads_max;
+  int nthreads_avail;
+  double MNK;
 #ifndef COMPLEX
 #ifdef XDOUBLE
   int mode  =  BLAS_XDOUBLE | BLAS_REAL;
@@ -237,6 +240,9 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
   XFLOAT *sa, *sb;
 
 #ifdef SMP
+  int nthreads_max;
+  int nthreads_avail;
+  double MNK;
 #ifndef COMPLEX
 #ifdef XDOUBLE
   int mode  =  BLAS_XDOUBLE | BLAS_REAL;
@@ -400,15 +406,15 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
   mode |= (transa << BLAS_TRANSA_SHIFT);
   mode |= (transb << BLAS_TRANSB_SHIFT);
 
-  int nthreads_max = num_cpu_avail(3);
-  int nthreads_avail = nthreads_max;
+  nthreads_max = num_cpu_avail(3);
+  nthreads_avail = nthreads_max;
 
 #ifndef COMPLEX
-  double MNK = (double) args.m * (double) args.n * (double) args.k;
+  MNK = (double) args.m * (double) args.n * (double) args.k;
   if ( MNK <= (65536.0  * (double) GEMM_MULTITHREAD_THRESHOLD)  )
 	nthreads_max = 1;
 #else
-  double MNK = (double) args.m * (double) args.n * (double) args.k;
+  MNK = (double) args.m * (double) args.n * (double) args.k;
   if ( MNK <= (8192.0  * (double) GEMM_MULTITHREAD_THRESHOLD)  )
 	nthreads_max = 1;
 #endif

interface/gemv.c

@@ -81,6 +81,9 @@ void NAME(char *TRANS, blasint *M, blasint *N,
   FLOAT *buffer;
 #ifdef SMP
   int nthreads;
+  int nthreads_max;
+  int nthreads_avail;
+  double MNK;
 #endif
 
   int (*gemv[])(BLASLONG, BLASLONG, BLASLONG, FLOAT, FLOAT *, BLASLONG,  FLOAT * , BLASLONG, FLOAT *, BLASLONG, FLOAT *) = {
@@ -135,6 +138,9 @@ void CNAME(enum CBLAS_ORDER order,
   blasint info, t;
 #ifdef SMP
   int nthreads;
+  int nthreads_max;
+  int nthreads_avail;
+  double MNK;
 #endif
 
   int (*gemv[])(BLASLONG, BLASLONG, BLASLONG, FLOAT, FLOAT *, BLASLONG,  FLOAT * , BLASLONG, FLOAT *, BLASLONG, FLOAT *) = {
@@ -235,10 +241,10 @@ void CNAME(enum CBLAS_ORDER order,
 
 #ifdef SMP
 
-  int  nthreads_max = num_cpu_avail(2);
-  int  nthreads_avail = nthreads_max;
+  nthreads_max = num_cpu_avail(2);
+  nthreads_avail = nthreads_max;
 
-  double MNK = (double) m * (double) n;
+  MNK = (double) m * (double) n;
   if ( MNK <= (24.0 * 24.0  * (double) (GEMM_MULTITHREAD_THRESHOLD*GEMM_MULTITHREAD_THRESHOLD) )  )
         nthreads_max = 1;
 

kernel/CMakeLists.txt

@@ -0,0 +1,254 @@
+
+include_directories(${CMAKE_SOURCE_DIR})
+include("${CMAKE_SOURCE_DIR}/cmake/kernel.cmake")
+
+# Makefile
+
+if (DEFINED TARGET_CORE)
+  #override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
+  set(BUILD_KERNEL 1)
+  set(KDIR "")
+  set(TSUFFIX "_${TARGET_CORE}")
+else ()
+  set(TARGET_CORE ${CORE})
+  set(KDIR "")
+  set(TSUFFIX "")
+endif ()
+
+SetDefaultL1()
+ParseMakefileVars("${KERNELDIR}/KERNEL.${TARGET_CORE}")
+ParseMakefileVars("${KERNELDIR}/KERNEL")
+
+if (${ARCH} STREQUAL "x86")
+  GenerateNamedObjects("${KERNELDIR}/cpuid.S" "" "" false "" "" true)
+endif ()
+
+# don't use float type name mangling here
+GenerateNamedObjects("${KERNELDIR}/${LSAME_KERNEL}" "F_INTERFACE" "lsame" false "" "" true)
+GenerateNamedObjects("${KERNELDIR}/${SCABS_KERNEL}" "COMPLEX;F_INTERFACE" "scabs1" false "" "" true)
+GenerateNamedObjects("${KERNELDIR}/${DCABS_KERNEL}" "DOUBLE;COMPLEX;F_INTERFACE" "dcabs1" false "" "" true)
+
+# Makefile.L1
+foreach (float_type ${FLOAT_TYPES})
+  # a bit of metaprogramming here to pull out the appropriate KERNEL var
+  string(SUBSTRING ${float_type} 0 1 float_char)
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}AMAXKERNEL}" "USE_ABS" "amax_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}AMINKERNEL}" "USE_ABS;USE_MIN" "amin_k" false "" "" false ${float_type})
+  if (DEFINED ${float_char}MAXKERNEL)
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}MAXKERNEL}" "" "max_k" false "" "" false ${float_type})
+  endif ()
+  if (DEFINED ${float_char}MINKERNEL)
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}MINKERNEL}" "" "min_k" false "" "" false ${float_type})
+  endif ()
+  GenerateNamedObjects("${KERNELDIR}/${I${float_char}AMAXKERNEL}" "USE_ABS" "i*amax_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${I${float_char}AMINKERNEL}" "USE_ABS;USE_MIN" "i*amin_k" false "" "" false ${float_type})
+  if (DEFINED I${float_char}MAXKERNEL)
+    GenerateNamedObjects("${KERNELDIR}/${I${float_char}MAXKERNEL}" "" "i*max_k" false "" "" false ${float_type})
+  endif ()
+  if (DEFINED I${float_char}MINKERNEL)
+    GenerateNamedObjects("${KERNELDIR}/${I${float_char}MINKERNEL}" "" "i*min_k" false "" "" false ${float_type})
+  endif ()
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}ASUMKERNEL}" "" "asum_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}AXPYKERNEL}" "" "axpy_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}COPYKERNEL}" "C_INTERFACE" "copy_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}NRM2KERNEL}" "" "nrm2_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}ROTKERNEL}" "" "rot_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}SCALKERNEL}" "" "scal_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}SWAPKERNEL}" "" "swap_k" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}AXPBYKERNEL}" "" "axpby_k" false "" "" false ${float_type})
+
+  if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}AXPYKERNEL}" "CONJ" "axpyc_k" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}DOTKERNEL}" "" "dotu_k" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}DOTKERNEL}" "CONJ" "dotc_k" false "" "" false ${float_type})
+  else ()
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}DOTKERNEL}" "" "dot_k" false "" "" false ${float_type})
+  endif ()
+endforeach ()
+
+# Makefile.L2
+SetDefaultL2()
+GenerateCombinationObjects("generic/symv_k.c" "LOWER" "U" "" 1 "" "" 3)
+GenerateNamedObjects("generic/ger.c" "" "ger_k" false "" "" "" 3)
+foreach (float_type ${FLOAT_TYPES})
+  string(SUBSTRING ${float_type} 0 1 float_char)
+  if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GERUKERNEL}" "" "geru_k" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GERCKERNEL}" "CONJ" "gerc_k" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GERUKERNEL}" "XCONJ" "gerv_k" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GERCKERNEL}" "CONJ;XCONJ" "gerd_k" false "" "" false ${float_type})
+
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVNKERNEL}" "" "gemv_n" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVTKERNEL}" "TRANSA" "gemv_t" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVNKERNEL}" "CONJ" "gemv_r" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVTKERNEL}" "CONJ;TRANSA" "gemv_c" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVNKERNEL}" "XCONJ" "gemv_o" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVTKERNEL}" "XCONJ;TRANSA" "gemv_u" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVNKERNEL}" "XCONJ;CONJ" "gemv_s" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVTKERNEL}" "XCONJ;CONJ;TRANSA" "gemv_d" false "" "" false ${float_type})
+  else ()
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVNKERNEL}" "" "gemv_n" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMVTKERNEL}" "TRANS" "gemv_t" false "" "" false ${float_type})
+  endif ()
+endforeach ()
+
+# Makefile.L3
+set(USE_GEMM3M false)
+set(USE_TRMM false)
+
+if (${ARCH} STREQUAL "x86" OR ${ARCH} STREQUAL "x86_64" OR ${ARCH} STREQUAL "ia64" OR ${ARCH} STREQUAL "MIPS")
+  set(USE_GEMM3M true)
+endif ()
+
+if (${ARCH} STREQUAL "arm" OR ${ARCH} STREQUAL "arm64" OR "${TARGET}" STREQUAL "LONGSOON3B" OR "${TARGET}" STREQUAL "GENERIC")
+  set(USE_TRMM true)
+endif ()
+
+foreach (float_type ${FLOAT_TYPES})
+  string(SUBSTRING ${float_type} 0 1 float_char)
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "" "gemm_kernel" false "" "" false ${float_type})
+
+  if (${float_char}GEMMINCOPY)
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMINCOPY}" "${float_type}" "${${float_char}GEMMINCOPYOBJ}" false "" "" true ${float_type})
+  endif ()
+
+  if (${float_char}GEMMITCOPY)
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMITCOPY}" "${float_type}" "${${float_char}GEMMITCOPYOBJ}" false "" "" true ${float_type})
+  endif ()
+
+  if (${float_char}GEMMONCOPY)
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMONCOPY}" "${float_type}" "${${float_char}GEMMONCOPYOBJ}" false "" "" true ${float_type})
+  endif ()
+
+  if (${float_char}GEMMOTCOPY)
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMOTCOPY}" "${float_type}" "${${float_char}GEMMOTCOPYOBJ}" false "" "" true ${float_type})
+  endif ()
+
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMM_BETA}" "" "gemm_beta" false "" "" false ${float_type})
+
+  if (USE_TRMM)
+    set(TRMM_KERNEL "${${float_char}TRMMKERNEL}")
+  else ()
+    set(TRMM_KERNEL "${${float_char}GEMMKERNEL}")
+  endif ()
+
+  if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
+
+    # just enumerate all these. there is an extra define for these indicating which side is a conjugate (e.g. CN NC NN) that I don't really want to work into GenerateCombinationObjects
+
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "NN" "gemm_kernel_n" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "CN" "gemm_kernel_l" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "NC" "gemm_kernel_r" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "CC" "gemm_kernel_b" false "" "" false ${float_type})
+
+    GenerateNamedObjects("${KERNELDIR}/${TRMM_KERNEL}" "LEFT;NN" "trmm_kernel_LN" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${TRMM_KERNEL}" "LEFT;TRANSA;NN" "trmm_kernel_LT" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${TRMM_KERNEL}" "LEFT;CONJ;CN" "trmm_kernel_LR" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${TRMM_KERNEL}" "LEFT;TRANSA;CONJ;CN" "trmm_kernel_LC" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${TRMM_KERNEL}" "NN" "trmm_kernel_RN" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${TRMM_KERNEL}" "TRANSA;NN" "trmm_kernel_RT" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${TRMM_KERNEL}" "CONJ;NC" "trmm_kernel_RR" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${TRMM_KERNEL}" "TRANSA;CONJ;NC" "trmm_kernel_RC" false "" "" false ${float_type})
+
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_LN}" "UPPER;LN;TRSMKERNEL;CONJ" "trsm_kernel_LR" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_LT}" "UPPER;LT;TRSMKERNEL;CONJ" "trsm_kernel_LC" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_RN}" "UPPER;RN;TRSMKERNEL;CONJ" "trsm_kernel_RR" false "" "" false ${float_type})
+    GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_RT}" "UPPER;RN;TRSMKERNEL;CONJ" "trsm_kernel_RC" false "" "" false ${float_type})
+
+  else ()
+    GenerateCombinationObjects("${KERNELDIR}/${TRMM_KERNEL}" "LEFT;TRANSA" "R;N" "TRMMKERNEL" 2 "trmm_kernel" false ${float_type})
+  endif ()
+
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_LN}" "UPPER;LN;TRSMKERNEL" "trsm_kernel_LN" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_LT}" "LT;TRSMKERNEL" "trsm_kernel_LT" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_RN}" "UPPER;RN;TRSMKERNEL" "trsm_kernel_RN" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}TRSMKERNEL_RT}" "RT;TRSMKERNEL" "trsm_kernel_RT" false "" "" false ${float_type})
+
+  # These don't use a scheme that is easy to iterate over - the filenames have part of the DEFINE codes in them, for UPPER/TRANS but not for UNIT/OUTER. Also TRANS is not passed in as a define.
+  # Could simplify it a bit by pairing up by -UUNIT/-DUNIT.
+  GenerateNamedObjects("generic/trmm_uncopy_${${float_char}GEMM_UNROLL_M}.c" "UNIT" "trmm_iunucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_uncopy_${${float_char}GEMM_UNROLL_M}.c" "" "trmm_iunncopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_uncopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;UNIT" "trmm_ounucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_uncopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER" "trmm_ounncopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/trmm_lncopy_${${float_char}GEMM_UNROLL_M}.c" "LOWER;UNIT" "trmm_ilnucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_lncopy_${${float_char}GEMM_UNROLL_M}.c" "LOWER" "trmm_ilnncopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_lncopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;LOWER;UNIT" "trmm_olnucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_lncopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;LOWER" "trmm_olnncopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/trmm_utcopy_${${float_char}GEMM_UNROLL_M}.c" "UNIT" "trmm_iutucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_utcopy_${${float_char}GEMM_UNROLL_M}.c" "" "trmm_iutncopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_utcopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;UNIT" "trmm_outucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_utcopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER" "trmm_outncopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/trmm_ltcopy_${${float_char}GEMM_UNROLL_M}.c" "LOWER;UNIT" "trmm_iltucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_ltcopy_${${float_char}GEMM_UNROLL_M}.c" "LOWER" "trmm_iltncopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_ltcopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;LOWER;UNIT" "trmm_oltucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trmm_ltcopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;LOWER" "trmm_oltncopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/trsm_uncopy_${${float_char}GEMM_UNROLL_M}.c" "UNIT" "trsm_iunucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_uncopy_${${float_char}GEMM_UNROLL_M}.c" "" "trsm_iunncopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_uncopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;UNIT" "trsm_ounucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_uncopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER" "trsm_ounncopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/trsm_lncopy_${${float_char}GEMM_UNROLL_M}.c" "LOWER;UNIT" "trsm_ilnucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_lncopy_${${float_char}GEMM_UNROLL_M}.c" "LOWER" "trsm_ilnncopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_lncopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;LOWER;UNIT" "trsm_olnucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_lncopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;LOWER" "trsm_olnncopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/trsm_utcopy_${${float_char}GEMM_UNROLL_M}.c" "UNIT" "trsm_iutucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_utcopy_${${float_char}GEMM_UNROLL_M}.c" "" "trsm_iutncopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_utcopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;UNIT" "trsm_outucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_utcopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER" "trsm_outncopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/trsm_ltcopy_${${float_char}GEMM_UNROLL_M}.c" "LOWER;UNIT" "trsm_iltucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_ltcopy_${${float_char}GEMM_UNROLL_M}.c" "LOWER" "trsm_iltncopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_ltcopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;LOWER;UNIT" "trsm_oltucopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/trsm_ltcopy_${${float_char}GEMM_UNROLL_M}.c" "OUTER;LOWER" "trsm_oltncopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/symm_ucopy_${${float_char}GEMM_UNROLL_N}.c" "OUTER" "symm_outcopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/symm_ucopy_${${float_char}GEMM_UNROLL_N}.c" "" "symm_iutcopy" false "" "" false ${float_type})
+
+  GenerateNamedObjects("generic/symm_lcopy_${${float_char}GEMM_UNROLL_N}.c" "LOWER;OUTER" "symm_oltcopy" false "" "" false ${float_type})
+  GenerateNamedObjects("generic/symm_lcopy_${${float_char}GEMM_UNROLL_N}.c" "LOWER" "symm_iltcopy" false "" "" false ${float_type})
+
+  if (NOT DEFINED ${float_char}OMATCOPY_CN)
+    if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+      set(${float_char}OMATCOPY_CN ../arm/zomatcopy_cn.c)
+    else ()
+      set(${float_char}OMATCOPY_CN ../arm/omatcopy_cn.c)
+    endif ()
+  endif ()
+  if (NOT DEFINED ${float_char}OMATCOPY_RN)
+    if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+      set(${float_char}OMATCOPY_RN ../arm/zomatcopy_rn.c)
+    else ()
+      set(${float_char}OMATCOPY_RN ../arm/omatcopy_rn.c)
+    endif  ()
+  endif ()
+  if (NOT DEFINED ${float_char}OMATCOPY_CT)
+    if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+      set(${float_char}OMATCOPY_CT ../arm/zomatcopy_ct.c)
+    else ()
+      set(${float_char}OMATCOPY_CT ../arm/omatcopy_ct.c)
+    endif ()
+  endif ()
+  if (NOT DEFINED ${float_char}OMATCOPY_RT)
+    if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C")
+      set(${float_char}OMATCOPY_RT ../arm/zomatcopy_rt.c)
+    else ()
+      set(${float_char}OMATCOPY_RT ../arm/omatcopy_rt.c)
+    endif ()
+  endif ()
+
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}OMATCOPY_CN}" "" "domatcopy_k_cn" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}OMATCOPY_RN}" "ROWM" "domatcopy_k_rn" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}OMATCOPY_CT}" "" "domatcopy_k_ct" false "" "" false ${float_type})
+  GenerateNamedObjects("${KERNELDIR}/${${float_char}OMATCOPY_RT}" "ROWM" "domatcopy_k_rt" false "" "" false ${float_type})
+
+endforeach ()
+
+# Makefile.LA
+#DBLASOBJS	+= dneg_tcopy$(TSUFFIX).$(SUFFIX) dlaswp_ncopy$(TSUFFIX).$(SUFFIX)
+
+add_library(kernel OBJECT ${OPENBLAS_SRC})

kernel/arm/zdot.c

@@ -35,9 +35,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 **************************************************************************************/
 
 #include "common.h"
-#include <complex.h>
 
+#ifndef _MSC_VER
+#include <complex.h>
 FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
+#else
+openblas_complex_double CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
+#endif
 {
 	BLASLONG i=0;
 	BLASLONG ix=0,iy=0;

lapack/CMakeLists.txt

@@ -0,0 +1,98 @@
+
+include_directories(${CMAKE_SOURCE_DIR})
+
+
+set(LAPACK_SOURCES
+  getrf/getrf_single.c
+  potrf/potrf_U_single.c
+  potrf/potrf_L_single.c
+  lauum/lauum_U_single.c
+  lauum/lauum_L_single.c
+)
+
+# add a 'z' to filename for complex version
+set(LAPACK_MANGLED_SOURCES
+  getf2/getf2_k.c
+  lauu2/lauu2_U.c
+  lauu2/lauu2_L.c
+  potf2/potf2_U.c
+  potf2/potf2_L.c
+)
+
+# sources that need TRANS set
+# this has a 'z' version
+set(TRANS_SOURCES
+  getrs/getrs_single.c
+)
+
+# sources that need UNIT set
+# these do NOT have a z version
+set(UNIT_SOURCES
+  trtri/trtri_U_single.c
+  trtri/trtri_L_single.c
+)
+
+# these have a 'z' version
+set(UNIT_SOURCES2
+  trti2/trti2_U.c
+  trti2/trti2_L.c
+)
+
+GenerateNamedObjects("${LAPACK_SOURCES}")
+GenerateNamedObjects("${LAPACK_MANGLED_SOURCES}" "" "" false "" "" false 3)
+
+# TODO: laswp needs arch specific code
+GenerateNamedObjects("laswp/generic/laswp_k.c" "" "laswp_plus" false "" "" false 3)
+GenerateNamedObjects("laswp/generic/laswp_k.c" "MINUS" "laswp_minus" false "" "" false 3)
+
+if (SMP)
+
+  if (USE_OPENMP)
+    set(GETRF_SRC getrf/getrf_parallel_omp.c)
+  else ()
+    set(GETRF_SRC getrf/getrf_parallel.c)
+  endif ()
+
+  # these do not have 'z' versions
+  set(PARALLEL_SOURCES
+    ${GETRF_SRC}
+    lauum/lauum_U_parallel.c
+    lauum/lauum_L_parallel.c
+    potrf/potrf_U_parallel.c
+    potrf/potrf_L_parallel.c
+  )
+
+  # this has a z version
+  list(APPEND TRANS_SOURCES
+    getrs/getrs_parallel.c
+  )
+
+  # these do NOT have a z version
+  list(APPEND UNIT_SOURCES
+    trtri/trtri_U_parallel.c
+    trtri/trtri_L_parallel.c
+  )
+
+  GenerateNamedObjects("${PARALLEL_SOURCES}")
+endif ()
+
+foreach (float_type ${FLOAT_TYPES})
+  if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
+    foreach (trans_src ${TRANS_SOURCES})
+      string(REGEX MATCH "[a-z]/([a-z]+_)([a-z]+)" op_name ${trans_src})
+      string(REPLACE "/" "/z" ztrans_src ${trans_src})
+      GenerateNamedObjects("${ztrans_src}" "TRANS=1" "${CMAKE_MATCH_1}N_${CMAKE_MATCH_2}" false "" "" false ${float_type})
+      GenerateNamedObjects("${ztrans_src}" "TRANS=2" "${CMAKE_MATCH_1}T_${CMAKE_MATCH_2}" false "" "" false ${float_type})
+      GenerateNamedObjects("${ztrans_src}" "TRANS=3" "${CMAKE_MATCH_1}R_${CMAKE_MATCH_2}" false "" "" false ${float_type})
+      GenerateNamedObjects("${ztrans_src}" "TRANS=4" "${CMAKE_MATCH_1}C_${CMAKE_MATCH_2}" false "" "" false ${float_type})
+    endforeach ()
+  else ()
+    GenerateCombinationObjects("${TRANS_SOURCES}" "TRANS" "N" "" 4 "" false ${float_type})
+  endif ()
+endforeach ()
+
+GenerateCombinationObjects("${UNIT_SOURCES}" "UNIT" "N" "" 4)
+GenerateCombinationObjects("${UNIT_SOURCES2}" "UNIT" "N" "" 0 "" "" 3)
+
+add_library(lapack OBJECT ${OPENBLAS_SRC})
+

lapack/getrf/getrf_parallel.c

@@ -67,7 +67,7 @@ double sqrt(double);
 #undef  GETRF_FACTOR
 #define GETRF_FACTOR 1.00
 
-static inline BLASLONG FORMULA1(BLASLONG M, BLASLONG N, BLASLONG IS, BLASLONG BK, BLASLONG T) {
+static __inline BLASLONG FORMULA1(BLASLONG M, BLASLONG N, BLASLONG IS, BLASLONG BK, BLASLONG T) {
 
   double m = (double)(M - IS - BK);
   double n = (double)(N - IS - BK);
@@ -373,7 +373,11 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
 
   BLASLONG num_cpu;
 
+#ifdef _MSC_VER
+  BLASLONG flag[MAX_CPU_NUMBER * CACHE_LINE_SIZE];
+#else
   volatile BLASLONG flag[MAX_CPU_NUMBER * CACHE_LINE_SIZE] __attribute__((aligned(128)));
+#endif
 
 #ifndef COMPLEX
 #ifdef XDOUBLE

symcopy.h

@@ -43,7 +43,7 @@
 
 #if !defined(XDOUBLE) || !defined(QUAD_PRECISION)
 
-static inline void SYMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void SYMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -141,7 +141,7 @@ static inline void SYMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void SYMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void SYMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -232,7 +232,7 @@ static inline void SYMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
 }
 
 
-static inline void ZSYMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZSYMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -362,7 +362,7 @@ static inline void ZSYMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void ZSYMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZSYMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -486,7 +486,7 @@ static inline void ZSYMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void ZHEMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZHEMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -613,7 +613,7 @@ static inline void ZHEMCOPY_L(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void ZHEMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZHEMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -735,7 +735,7 @@ static inline void ZHEMCOPY_U(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
 }
 
 
-static inline void ZHEMCOPY_M(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZHEMCOPY_M(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -862,7 +862,7 @@ static inline void ZHEMCOPY_M(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void ZHEMCOPY_V(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZHEMCOPY_V(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -984,7 +984,7 @@ static inline void ZHEMCOPY_V(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
 }
 
 
-static inline void TRMCOPY_NL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void TRMCOPY_NL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -1082,7 +1082,7 @@ static inline void TRMCOPY_NL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void TRMCOPY_TL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void TRMCOPY_TL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -1180,7 +1180,7 @@ static inline void TRMCOPY_TL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void TRMCOPY_NU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void TRMCOPY_NU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -1270,7 +1270,7 @@ static inline void TRMCOPY_NU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void TRMCOPY_TU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void TRMCOPY_TU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -1360,7 +1360,7 @@ static inline void TRMCOPY_TU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void ZTRMCOPY_NL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZTRMCOPY_NL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -1490,7 +1490,7 @@ static inline void ZTRMCOPY_NL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void ZTRMCOPY_TL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZTRMCOPY_TL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -1620,7 +1620,7 @@ static inline void ZTRMCOPY_TL(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void ZTRMCOPY_NU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZTRMCOPY_NU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;
@@ -1744,7 +1744,7 @@ static inline void ZTRMCOPY_NU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   }
 }
 
-static inline void ZTRMCOPY_TU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
+static __inline void ZTRMCOPY_TU(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b){
   BLASLONG is, js;
 
   FLOAT *aa1, *aa2;