Clean up, fix old typos. Simplify arch usages. Move system arch check to earlier position.

CMakeLists.txt

@@ -87,11 +87,6 @@ if (BUILD_COMPLEX16)
   list(APPEND FLOAT_TYPES "ZCOMPLEX") # defines COMPLEX and DOUBLE
 endif ()
 
-set(SUBDIRS_ALL ${SUBDIRS} test ctest utest exports benchmark ../laswp ../bench)
-
-# all :: libs netlib tests shared
-
-# libs :
 if (NOT DEFINED CORE OR "${CORE}" STREQUAL "UNKNOWN")
   message(FATAL_ERROR "Detecting CPU failed. Please set TARGET explicitly, e.g. make TARGET=your_cpu_target. Please read README for details.")
 endif ()
@@ -125,18 +120,10 @@ endif ()
 # Only generate .def for dll on MSVC and always produce pdb files for debug and release
 if(MSVC)
 set(OpenBLAS_DEF_FILE "${PROJECT_BINARY_DIR}/openblas.def")
-set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Zi")
 set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} /Zi")
 set(CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} /DEBUG /OPT:REF /OPT:ICF")
 endif()
 
-#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
 if (${DYNAMIC_ARCH})
   add_subdirectory(kernel)
   foreach(TARGET_CORE ${DYNAMIC_CORE})
@@ -197,24 +184,6 @@ set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES
   SOVERSION ${OpenBLAS_MAJOR_VERSION}
 )
 
-
-# TODO: Why is the config saved here? Is this necessary with CMake?
-#Save the config files for installation
-#	@cp Makefile.conf Makefile.conf_last
-#	@cp config.h config_last.h
-#ifdef QUAD_PRECISION
-#	@echo "#define QUAD_PRECISION">> config_last.h
-#endif
-#ifeq ($(EXPRECISION), 1)
-#	@echo "#define EXPRECISION">> config_last.h
-#endif
-###
-
-#ifdef USE_THREAD
-#	@echo USE_THREAD=$(USE_THREAD) >>  Makefile.conf_last
-#endif
-#	@touch lib.grd
-
 # Install project
 
 # Install libraries

cmake/arch.cmake

@@ -3,9 +3,9 @@
 ## Description: Ported from portion of OpenBLAS/Makefile.system
 ##              Sets various variables based on architecture.
 
-if (${ARCH} STREQUAL "x86" OR ${ARCH} STREQUAL "x86_64")
+if (X86 OR X86_64)
 
-  if (${ARCH} STREQUAL "x86")
+  if (X86)
     if (NOT BINARY)
       set(NO_BINARY_MODE 1)
     endif ()
@@ -46,11 +46,11 @@ endif ()
 
 
 if (DYNAMIC_ARCH)
-  if (${ARCH} STREQUAL "x86")
+  if (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")
+  if (X86_64)
     set(DYNAMIC_CORE PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO)
     if (NOT NO_AVX)
       set(DYNAMIC_CORE ${DYNAMIC_CORE} SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR)
@@ -77,7 +77,7 @@ if (${ARCH} STREQUAL "ia64")
   endif ()
 endif ()
 
-if (${ARCH} STREQUAL "mips64")
+if (MIPS64)
   set(NO_BINARY_MODE 1)
 endif ()
 
@@ -86,12 +86,12 @@ if (${ARCH} STREQUAL "alpha")
   set(BINARY_DEFINED 1)
 endif ()
 
-if (${ARCH} STREQUAL "arm")
+if (ARM)
   set(NO_BINARY_MODE 1)
   set(BINARY_DEFINED 1)
 endif ()
 
-if (${ARCH} STREQUAL "arm64")
+if (ARM64)
   set(NO_BINARY_MODE 1)
   set(BINARY_DEFINED 1)
 endif ()

cmake/cc.cmake

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

cmake/fc.cmake

@@ -50,7 +50,7 @@ if (${F_COMPILER} STREQUAL "GFORTRAN")
     set(EXTRALIB "{EXTRALIB} -lgfortran")
   endif ()
   if (NO_BINARY_MODE)
-    if (${ARCH} STREQUAL "mips64")
+    if (MIPS64)
       if (BINARY64)
         set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=64")
       else ()
@@ -130,7 +130,7 @@ if (${F_COMPILER} STREQUAL "PATHSCALE")
     endif ()
   endif ()
 
-  if (NOT ${ARCH} STREQUAL "mips64")
+  if (NOT MIPS64)
     if (NOT BINARY64)
       set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
     else ()
@@ -158,7 +158,7 @@ if (${F_COMPILER} STREQUAL "OPEN64")
     endif ()
   endif ()
 
-  if (${ARCH} STREQUAL "mips64")
+  if (MIPS64)
 
     if (NOT BINARY64)
       set(FCOMMON_OPT "${FCOMMON_OPT} -n32")
@@ -189,7 +189,7 @@ endif ()
 
 if (${F_COMPILER} STREQUAL "SUN")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_SUN")
-  if (${ARCH} STREQUAL "x86")
+  if (X86)
     set(FCOMMON_OPT "${FCOMMON_OPT} -m32")
   else ()
     set(FCOMMON_OPT "${FCOMMON_OPT} -m64")

cmake/os.cmake

@@ -43,7 +43,7 @@ if (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
 
   # Ensure the correct stack alignment on Win32
   # http://permalink.gmane.org/gmane.comp.lib.openblas.general/97
-  if (${ARCH} STREQUAL "x86")
+  if (X86)
     if (NOT MSVC AND NOT ${CMAKE_C_COMPILER_ID} STREQUAL "Clang")
       set(CCOMMON_OPT "${CCOMMON_OPT} -mincoming-stack-boundary=2")
     endif ()
@@ -75,7 +75,7 @@ if (QUAD_PRECISION)
   set(NO_EXPRECISION 1)
 endif ()
 
-if (${ARCH} STREQUAL "x86")
+if (X86)
   set(NO_EXPRECISION 1)
 endif ()
 

cmake/prebuild.cmake

@@ -53,22 +53,37 @@ else()
 endif ()
 
 set(TARGET_CONF_TEMP "${PROJECT_BINARY_DIR}/${TARGET_CONF}.tmp")
-include("${PROJECT_SOURCE_DIR}/cmake/c_check.cmake")
 
+# c_check
+set(FU "")
+if (APPLE OR (MSVC AND NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang"))
+  set(FU "_")
+endif()
+
+set(COMPILER_ID ${CMAKE_C_COMPILER_ID})
+if (${COMPILER_ID} STREQUAL "GNU")
+  set(COMPILER_ID "GCC")
+endif ()
+
+string(TOUPPER ${ARCH} UC_ARCH)
+
+file(WRITE ${TARGET_CONF_TEMP}
+  "#define OS_${HOST_OS}\t1\n"
+  "#define ARCH_${UC_ARCH}\t1\n"
+  "#define C_${COMPILER_ID}\t1\n"
+  "#define __${BINARY}BIT__\t1\n"
+  "#define FUNDERSCORE\t${FU}\n")
+
+if (${HOST_OS} STREQUAL "WINDOWSSTORE")
+  file(APPEND ${TARGET_CONF_TEMP}
+    "#define OS_WINNT\t1\n")
+endif ()
+
+# f_check
 if (NOT NOFORTRAN)
   include("${PROJECT_SOURCE_DIR}/cmake/f_check.cmake")
 endif ()
 
-# This check requires c_check for arch but it should probably be done earlier
-if(CMAKE_CROSSCOMPILING AND NOT DEFINED CORE)
-  # Detect target without running getarch
-  if(ARM64)
-    set(CORE "ARMV8")
-  else()
-    message(FATAL_ERROR "When cross compiling, a CORE is required.")
-  endif()
-endif()
-
 # Cannot run getarch on target if we are cross-compiling
 if (DEFINED CORE AND CMAKE_CROSSCOMPILING)
   # Write to config as getarch would
@@ -78,7 +93,22 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING)
   file(APPEND ${TARGET_CONF_TEMP}
     "#define ${CORE}\n"
     "#define CHAR_CORENAME \"${CORE}\"\n")
-  if ("${CORE}" STREQUAL "ARMV8")
+  if ("${CORE}" STREQUAL "ARMV7")
+    file(APPEND ${TARGET_CONF_TEMP}
+      "#define L1_DATA_SIZE\t65536\n"
+      "#define L1_DATA_LINESIZE\t32\n"
+      "#define L2_SIZE\t512488\n"
+      "#define L2_LINESIZE\t32\n"
+      "#define DTB_DEFAULT_ENTRIES\t64\n"
+      "#define DTB_SIZE\t4096\n"
+      "#define L2_ASSOCIATIVE\t4\n"
+      "#define HAVE_VFPV3\n"
+      "#define HAVE_VFP\n")
+    set(SGEMM_UNROLL_M 4)
+    set(SGEMM_UNROLL_N 4)
+    set(DGEMM_UNROLL_M 4)
+    set(DGEMM_UNROLL_N 4)
+  elseif ("${CORE}" STREQUAL "ARMV8")
     file(APPEND ${TARGET_CONF_TEMP}
       "#define L1_DATA_SIZE\t32768\n"
       "#define L1_DATA_LINESIZE\t64\n"

cmake/system.cmake

@@ -4,6 +4,26 @@
 ##
 set(NETLIB_LAPACK_DIR "${PROJECT_SOURCE_DIR}/lapack-netlib")
 
+# System detection, via CMake.
+include("${PROJECT_SOURCE_DIR}/cmake/system_check.cmake")
+
+if(CMAKE_CROSSCOMPILING AND NOT DEFINED TARGET)
+  # Detect target without running getarch
+  if (ARM64)
+    set(TARGET "ARMV8")
+  elseif(ARM)
+    set(TARGET "ARMV7") # TODO: Ask compiler which arch this is
+  else()
+    message(FATAL_ERROR "When cross compiling, a TARGET is required.")
+  endif()
+endif()
+
+# Other files expect CORE, which is actually TARGET and will become TARGET_CORE for kernel build. Confused yet?
+# It seems we are meant to use TARGET as input and CORE internally as kernel.
+if(NOT DEFINED CORE AND DEFINED TARGET)
+  set(CORE ${TARGET})
+endif()
+
 # TARGET_CORE will override TARGET which is used in DYNAMIC_ARCH=1.
 if (DEFINED TARGET_CORE)
   set(TARGET ${TARGET_CORE})
@@ -56,7 +76,7 @@ if (NOT DEFINED NO_PARALLEL_MAKE)
 endif ()
 set(GETARCH_FLAGS	"${GETARCH_FLAGS} -DNO_PARALLEL_MAKE=${NO_PARALLEL_MAKE}")
 
-if (CMAKE_CXX_COMPILER STREQUAL loongcc)
+if (CMAKE_C_COMPILER STREQUAL loongcc)
   set(GETARCH_FLAGS	"${GETARCH_FLAGS} -static")
 endif ()
 
@@ -75,6 +95,7 @@ if (NOT CMAKE_CROSSCOMPILING)
   endif()
 
   if (NOT NUM_CORES EQUAL 0)
+    # HT?
     set(NUM_THREADS ${NUM_CORES})
   endif ()
 endif()
@@ -153,7 +174,7 @@ if (NO_AVX)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX")
 endif ()
 
-if (${ARCH} STREQUAL "x86")
+if (X86)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX")
 endif ()
 
@@ -166,7 +187,7 @@ if (USE_THREAD)
   # NO_AFFINITY = 1
   set(CCOMMON_OPT "${CCOMMON_OPT} -DSMP_SERVER")
 
-  if (${ARCH} STREQUAL "mips64")
+  if (MIPS64)
     if (NOT ${CORE} STREQUAL "LOONGSON3B")
       set(USE_SIMPLE_THREADED_LEVEL3 1)
     endif ()
@@ -237,7 +258,7 @@ 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")
+if (NOT X86_64 AND NOT X86 AND NOT ${CORE} STREQUAL "LOONGSON3B")
   set(NO_AFFINITY 1)
 endif ()
 
@@ -377,7 +398,7 @@ endif ()
 set(USE_GEMM3M 0)
 
 if (DEFINED ARCH)
-  if (${ARCH} STREQUAL "x86" OR ${ARCH} STREQUAL "x86_64" OR ${ARCH} STREQUAL "ia64" OR ${ARCH} STREQUAL "MIPS")
+  if (X86 OR X86_64 OR ${ARCH} STREQUAL "ia64" OR MIPS64)
     set(USE_GEMM3M 1)
   endif ()
 
@@ -460,35 +481,3 @@ endif ()
 #  export CUFLAGS
 #  export CULIB
 #endif
-
-#.SUFFIXES: .$(PSUFFIX) .$(SUFFIX) .f
-#
-#.f.$(SUFFIX):
-#	$(FC) $(FFLAGS) -c $<  -o $(@F)
-#
-#.f.$(PSUFFIX):
-#	$(FC) $(FPFLAGS) -pg -c $<  -o $(@F)
-
-# these are not cross-platform
-#ifdef BINARY64
-#PATHSCALEPATH	= /opt/pathscale/lib/3.1
-#PGIPATH		= /opt/pgi/linux86-64/7.1-5/lib
-#else
-#PATHSCALEPATH	= /opt/pathscale/lib/3.1/32
-#PGIPATH		= /opt/pgi/linux86/7.1-5/lib
-#endif
-
-#ACMLPATH	= /opt/acml/4.3.0
-#ifneq ($(OSNAME), Darwin)
-#MKLPATH         = /opt/intel/mkl/10.2.2.025/lib
-#else
-#MKLPATH         = /Library/Frameworks/Intel_MKL.framework/Versions/10.0.1.014/lib
-#endif
-#ATLASPATH	= /opt/atlas/3.9.17/opteron
-#FLAMEPATH	= $(HOME)/flame/lib
-#ifneq ($(OSNAME), SunOS)
-#SUNPATH		= /opt/sunstudio12.1
-#else
-#SUNPATH		= /opt/SUNWspro
-#endif
-

cmake/system_check.cmake

@@ -4,32 +4,6 @@
 ##              This is triggered by prebuild.cmake and runs before any of the code is built.
 ##              Creates config.h and Makefile.conf.
 
-# CMake vars set by this file:
-# OSNAME (use CMAKE_SYSTEM_NAME)
-# ARCH
-# C_COMPILER (use CMAKE_C_COMPILER)
-# BINARY32
-# BINARY64
-# FU
-# CROSS_SUFFIX
-# CROSS
-# CEXTRALIB
-
-# Defines set by this file:
-# OS_
-# ARCH_
-# C_
-# __32BIT__
-# __64BIT__
-# FUNDERSCORE
-# PTHREAD_CREATE_FUNC
-
-# N.B. c_check (and ctest.c) is not cross-platform, so instead try to use CMake variables.
-set(FU "")
-if (APPLE OR (MSVC AND NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang"))
-  set(FU "_")
-endif()
-
 # Convert CMake vars into the format that OpenBLAS expects
 string(TOUPPER ${CMAKE_SYSTEM_NAME} HOST_OS)
 if (${HOST_OS} STREQUAL "WINDOWS")
@@ -37,7 +11,7 @@ if (${HOST_OS} STREQUAL "WINDOWS")
 endif ()
 
 if(CMAKE_COMPILER_IS_GNUCC AND WIN32)
-    execute_process(COMMAND ${CMAKE_CXX_COMPILER} -dumpmachine
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpmachine
               OUTPUT_VARIABLE OPENBLAS_GCC_TARGET_MACHINE
               OUTPUT_STRIP_TRAILING_WHITESPACE)
     if(OPENBLAS_GCC_TARGET_MACHINE MATCHES "amd64|x86_64|AMD64")
@@ -50,8 +24,10 @@ if(CMAKE_CL_64 OR MINGW64)
   set(X86_64 1)
 elseif(MINGW OR (MSVC AND NOT CMAKE_CROSSCOMPILING))
   set(X86 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc.*|power.*|Power.*")
   set(PPC 1)
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "mips64.*")
+  set(MIPS64 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
   set(X86_64 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "i686.*|i386.*|x86.*|amd64.*|AMD64.*")
@@ -77,7 +53,7 @@ else()
 endif ()
 
 if (NOT BINARY)
-  if (X86_64 OR ARM64 OR PPC OR ARCH STREQUAL "mips64")
+  if (X86_64 OR ARM64 OR PPC OR MIPS64)
     set(BINARY 64)
   else ()
     set(BINARY 32)
@@ -90,22 +66,3 @@ else()
   set(BINARY32 1)
 endif()
 
-set(COMPILER_ID ${CMAKE_C_COMPILER_ID})
-if (${COMPILER_ID} STREQUAL "GNU")
-  set(COMPILER_ID "GCC")
-endif ()
-
-string(TOUPPER ${ARCH} UC_ARCH)
-
-file(WRITE ${TARGET_CONF_TEMP}
-  "#define OS_${HOST_OS}\t1\n"
-  "#define ARCH_${UC_ARCH}\t1\n"
-  "#define C_${COMPILER_ID}\t1\n"
-  "#define __${BINARY}BIT__\t1\n"
-  "#define FUNDERSCORE\t${FU}\n")
-
-if (${HOST_OS} STREQUAL "WINDOWSSTORE")
-  file(APPEND ${TARGET_CONF_TEMP}
-    "#define OS_WINNT\t1\n")
-endif ()
-

kernel/CMakeLists.txt

@@ -21,7 +21,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     endif ()
 
     if (${ADD_COMMONOBJS})
-        if (${ARCH} STREQUAL "x86")
+        if (X86)
         if (NOT "${CMAKE_C_COMPILER_ID}" STREQUAL "MSVC")
           GenerateNamedObjects("${KERNELDIR}/cpuid.S" "" "" false "" "" true)
         else()
@@ -121,7 +121,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     # Makefile.L3
     set(USE_TRMM false)
 
-    if (${ARCH} STREQUAL "arm" OR ${ARCH} STREQUAL "arm64" OR "${TARGET}" STREQUAL "LONGSOON3B" OR "${TARGET}" STREQUAL "GENERIC" OR "${CORE}" STREQUAL "generic" OR "${TARGET}" STREQUAL "HASWELL" OR "${CORE}" STREQUAL "haswell" OR "{CORE}" STREQUAL "zen")
+    if (ARM OR ARM64 OR "${TARGET}" STREQUAL "LONGSOON3B" OR "${TARGET}" STREQUAL "GENERIC" OR "${CORE}" STREQUAL "generic" OR "${TARGET}" STREQUAL "HASWELL" OR "${CORE}" STREQUAL "haswell" OR "{CORE}" STREQUAL "zen")
       set(USE_TRMM true)
     endif ()