Merge pull request #1264 from isuruf/dyn

Support DYNAMIC_ARCH with CMake

CMakeLists.txt

@@ -27,6 +27,7 @@ option(BUILD_WITHOUT_LAPACK "Without LAPACK and LAPACKE (Only BLAS or CBLAS)" ON
 endif()
 option(BUILD_WITHOUT_CBLAS "Without CBLAS" OFF)
 option(BUILD_DEBUG "Build Debug Version" OFF)
+option(DYNAMIC_ARCH "Build with DYNAMIC_ARCH" OFF)
 #######
 if(BUILD_WITHOUT_LAPACK)
 set(NO_LAPACK 1)
@@ -153,6 +154,22 @@ 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 ${DYNAMIC_CORE})
+    message("${TARGET_CORE}")
+    list(APPEND TARGET_OBJS "$<TARGET_OBJECTS:kernel_${TARGET_CORE}>")
+  endforeach()
+endif ()
+
+
 # add objects to the openblas lib
 add_library(${OpenBLAS_LIBNAME} SHARED ${LA_SOURCES} ${LAPACKE_SOURCES} ${TARGET_OBJS} ${OpenBLAS_DEF_FILE})
 
@@ -209,13 +226,7 @@ set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES
 #	@echo "#define EXPRECISION">> config_last.h
 #endif
 ###
-#ifeq ($(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

appveyor.yml

@@ -25,6 +25,8 @@ skip_commits:
 
 environment:
   matrix:
+    - COMPILER: clang-cl
+      DYNAMIC_ARCH: ON
     - COMPILER: clang-cl
     - COMPILER: cl
 
@@ -39,6 +41,7 @@ before_build:
   - cd c:\projects\OpenBLAS
   - if [%COMPILER%]==[cl] cmake -G "Visual Studio 12 Win64" .
   - if [%COMPILER%]==[clang-cl] cmake -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl .
+  - if [%DYNAMIC_ARCH%]==[ON] cmake -DDYNAMIC_ARCH=ON .
 
 build_script:
   - cmake --build .

cmake/arch.cmake

@@ -64,16 +64,16 @@ 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")
+    set(DYNAMIC_CORE KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO)
   endif ()
 
   if (${ARCH} STREQUAL "x86_64")
-    set(DYNAMIC_CORE "PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO")
+    set(DYNAMIC_CORE PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO)
     if (NOT NO_AVX)
-      set(DYNAMIC_CORE "${DYNAMIC_CORE} SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER")
+      set(DYNAMIC_CORE ${DYNAMIC_CORE} SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR)
     endif ()
     if (NOT NO_AVX2)
-      set(DYNAMIC_CORE "${DYNAMIC_CORE} HASWELL ZEN")
+      set(DYNAMIC_CORE ${DYNAMIC_CORE} HASWELL ZEN)
     endif ()
   endif ()
 

cmake/prebuild.cmake

@@ -46,9 +46,11 @@ if (DEFINED TARGET_CORE)
   set(GETARCH2_FLAGS "-DBUILD_KERNEL")
   set(TARGET_MAKE "Makefile_kernel.conf")
   set(TARGET_CONF "config_kernel.h")
+  set(TARGET_CONF_DIR ${PROJECT_BINARY_DIR}/kernel_config/${TARGET_CORE})
 else()
   set(TARGET_MAKE "Makefile.conf")
   set(TARGET_CONF "config.h")
+  set(TARGET_CONF_DIR ${PROJECT_BINARY_DIR})
 endif ()
 
 set(TARGET_CONF_TEMP "${PROJECT_BINARY_DIR}/${TARGET_CONF}.tmp")
@@ -128,7 +130,13 @@ execute_process(COMMAND ${PROJECT_BINARY_DIR}/${GETARCH2_BIN} 1 OUTPUT_VARIABLE
 
 # append config data from getarch_2nd to the TARGET file and read in CMake vars
 file(APPEND ${TARGET_CONF_TEMP} ${GETARCH2_CONF_OUT})
+
+if (${BUILD_KERNEL})
+    configure_file(${TARGET_CONF_TEMP} ${PROJECT_BINARY_DIR}/kernel_config/${TARGET_CORE}/${TARGET_CONF} COPYONLY)
+else ()
     configure_file(${TARGET_CONF_TEMP} ${PROJECT_BINARY_DIR}/${TARGET_CONF} COPYONLY)
+endif ()
+
 ParseGetArchVars(${GETARCH2_MAKE_OUT})
 
 # compile get_config_h

cmake/system.cmake

@@ -2,7 +2,6 @@
 ## Author: Hank Anderson <hank@statease.com>
 ## Description: Ported from OpenBLAS/Makefile.system
 ##
-
 set(NETLIB_LAPACK_DIR "${PROJECT_SOURCE_DIR}/lapack-netlib")
 
 # TODO: Makefile.system detects Darwin (mac) and switches to clang here -hpa

cmake/utils.cmake

@@ -202,6 +202,8 @@ function(GenerateNamedObjects sources_in)
       if (use_cblas)
         set(obj_name "cblas_${obj_name}")
         list(APPEND obj_defines "CBLAS")
+      elseif (NOT "${obj_name}" MATCHES "${ARCH_SUFFIX}")
+        set(obj_name "${obj_name}${ARCH_SUFFIX}")
       endif ()
 
       list(APPEND obj_defines "ASMNAME=${FU}${obj_name};ASMFNAME=${FU}${obj_name}${BU};NAME=${obj_name}${BU};CNAME=${obj_name};CHAR_NAME=\"${obj_name}${BU}\";CHAR_CNAME=\"${obj_name}\"")

driver/others/dynamic.c

@@ -38,6 +38,10 @@
 
 #include "common.h"
 
+#ifdef _MSC_VER
+#define strncasecmp _strnicmp
+#define strcasecmp _stricmp
+#endif
 
 #ifdef ARCH_X86
 #define EXTERN extern

kernel/CMakeLists.txt

@@ -1,27 +1,26 @@
 
 include_directories(${PROJECT_SOURCE_DIR})
-include_directories(${PROJECT_BINARY_DIR})
-include("${PROJECT_SOURCE_DIR}/cmake/kernel.cmake")
 
 # Makefile
-
+function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
-if (DEFINED TARGET_CORE)
+    set (OPENBLAS_SRC "")
-  #override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
+    set (ARCH_SUFFIX "")
-  set(BUILD_KERNEL 1)
+    include("${PROJECT_SOURCE_DIR}/cmake/kernel.cmake")
-  set(KDIR "")
+    if (${DYNAMIC_ARCH})
-  set(TSUFFIX "_${TARGET_CORE}")
+        include("${PROJECT_SOURCE_DIR}/cmake/system.cmake")
-else ()
-  set(TARGET_CORE ${CORE})
-  set(KDIR "")
-  set(TSUFFIX "")
     endif ()
-
     SetDefaultL1()
     SetDefaultL2()
     SetDefaultL3()
     ParseMakefileVars("${KERNELDIR}/KERNEL")
     ParseMakefileVars("${KERNELDIR}/KERNEL.${TARGET_CORE}")
 
+    set(KERNEL_INTERFACE common_level1.h common_level2.h common_level3.h)
+    if(NOT ${NO_LAPACK})
+        set(KERNEL_INTERFACE ${KERNEL_INTERFACE} common_lapack.h)
+    endif ()
+
+    if (${ADD_COMMONOBJS})
         if (${ARCH} STREQUAL "x86")
         if (NOT "${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
           GenerateNamedObjects("${KERNELDIR}/cpuid.S" "" "" false "" "" true)
@@ -31,10 +30,13 @@ endif()
         endif ()
 
         # don't use float type name mangling here
-GenerateNamedObjects("${KERNELDIR}/${LSAME_KERNEL}" "F_INTERFACE" "lsame" false "" "" true)
+        GenerateNamedObjects("${KERNELDIR}/${LSAME_KERNEL}" "F_INTERFACE" "lsame" false "" "" true "")
-GenerateNamedObjects("${KERNELDIR}/${SCABS_KERNEL}" "COMPLEX;F_INTERFACE" "scabs1" false "" "" true)
+        GenerateNamedObjects("${KERNELDIR}/${SCABS_KERNEL}" "COMPLEX;F_INTERFACE" "scabs1" false "" "" true "")
-GenerateNamedObjects("${KERNELDIR}/${DCABS_KERNEL}" "DOUBLE;COMPLEX;F_INTERFACE" "dcabs1" false "" "" true)
+        GenerateNamedObjects("${KERNELDIR}/${DCABS_KERNEL}" "DOUBLE;COMPLEX;F_INTERFACE" "dcabs1" false "" "" true "")
+    endif ()
 
+    # Run with no ARCH_SUFFIX for above
+    set (ARCH_SUFFIX "${TSUFFIX}")
     # Makefile.L1
     foreach (float_type ${FLOAT_TYPES})
       # a bit of metaprogramming here to pull out the appropriate KERNEL var
@@ -248,6 +250,23 @@ foreach (float_type ${FLOAT_TYPES})
         GenerateNamedObjects("generic/zgemm3m_tcopy_${${float_char}GEMM3M_UNROLL_M}.c" "ICOPY;REAL_ONLY" "gemm3m_itcopyr" false "" "" false ${float_type})
         GenerateNamedObjects("generic/zgemm3m_tcopy_${${float_char}GEMM3M_UNROLL_M}.c" "ICOPY;IMAGE_ONLY" "gemm3m_itcopyi" false "" "" false ${float_type})
 
+#hemm3m and symm3m
+        foreach(name symm3m hemm3m)
+          GenerateNamedObjects("generic/z${name}_ucopy_${${float_char}GEMM3M_UNROLL_M}.c" "USE_ALPHA" "${name}_oucopyb.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_ucopy_${${float_char}GEMM3M_UNROLL_M}.c" "USE_ALPHA;REAL_ONLY" "${name}_oucopyr.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_ucopy_${${float_char}GEMM3M_UNROLL_M}.c" "USE_ALPHA;IMAGE_ONLY" "${name}_oucopyi.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_lcopy_${${float_char}GEMM3M_UNROLL_M}.c" "USE_ALPHA" "${name}_olcopyb.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_lcopy_${${float_char}GEMM3M_UNROLL_M}.c" "USE_ALPHA;REAL_ONLY" "${name}_olcopyr.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_lcopy_${${float_char}GEMM3M_UNROLL_M}.c" "USE_ALPHA;IMAGE_ONLY" "${name}_olcopyi.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_ucopy_${${float_char}GEMM3M_UNROLL_M}.c" "" "${name}_iucopyb.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_ucopy_${${float_char}GEMM3M_UNROLL_M}.c" "REAL_ONLY" "${name}_iucopyr.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_ucopy_${${float_char}GEMM3M_UNROLL_M}.c" "IMAGE_ONLY" "${name}_iucopyi.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_lcopy_${${float_char}GEMM3M_UNROLL_M}.c" "" "${name}_ilcopyb.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_lcopy_${${float_char}GEMM3M_UNROLL_M}.c" "REAL_ONLY" "${name}_ilcopyr.c" false "" "" false ${float_type})
+          GenerateNamedObjects("generic/z${name}_lcopy_${${float_char}GEMM3M_UNROLL_M}.c" "IMAGE_ONLY" "${name}_ilcopyi.c" false "" "" false ${float_type})
+
+        endforeach ()
+
       endif()
 
       else ()  #For real
@@ -445,7 +464,52 @@ foreach (float_type ${FLOAT_TYPES})
       GenerateNamedObjects("${KERNELDIR}/${${float_char}GEADD_KERNEL}" "" "geadd_k" false "" "" false ${float_type})
     endforeach ()
 
+    if (${DYNAMIC_ARCH})
+        set(SETPARAM_TARGET_DIR ${CMAKE_CURRENT_BINARY_DIR}${CMAKE_FILES_DIRECTORY})
+        file(READ ${CMAKE_CURRENT_SOURCE_DIR}/setparam-ref.c SETPARAM_REF_CONTENTS)
+        string(REPLACE "TS" "${TSUFFIX}" SETPARAM_REF_CONTENTS_NEW "${SETPARAM_REF_CONTENTS}")
+        file(WRITE ${SETPARAM_TARGET_DIR}/setparam${TSUFFIX}.tmp "${SETPARAM_REF_CONTENTS_NEW}")
+        configure_file(${SETPARAM_TARGET_DIR}/setparam${TSUFFIX}.tmp ${SETPARAM_TARGET_DIR}/setparam${TSUFFIX}.c COPYONLY)
+        set(OPENBLAS_SRC ${OPENBLAS_SRC} ${SETPARAM_TARGET_DIR}/setparam${TSUFFIX}.c)
+        file(REMOVE ${SETPARAM_TARGET_DIR}/setparam${TSUFFIX}.tmp)
+
+        set(KERNEL_TSUFFIX_CONTENTS "")
+        foreach (KERNEL_INTERFACE_H ${KERNEL_INTERFACE})
+            file(READ ${PROJECT_SOURCE_DIR}/${KERNEL_INTERFACE_H} KERNEL_INTERFACE_H_CONTENTS)
+            string(REGEX REPLACE "[ ]*\\(" "${TSUFFIX}(" KERNEL_INTERFACE_H_CONTENTS_NEW "${KERNEL_INTERFACE_H_CONTENTS}")
+            set(KERNEL_TSUFFIX_CONTENTS "${KERNEL_TSUFFIX_CONTENTS}\n${KERNEL_INTERFACE_H_CONTENTS_NEW}")
+        endforeach()
+	    file(WRITE ${SETPARAM_TARGET_DIR}/kernel${TSUFFIX}.tmp "${KERNEL_TSUFFIX_CONTENTS}")
+        configure_file(${SETPARAM_TARGET_DIR}/kernel${TSUFFIX}.tmp ${SETPARAM_TARGET_DIR}/kernel${TSUFFIX}.h COPYONLY)
+        file(REMOVE ${SETPARAM_TARGET_DIR}/kernel${TSUFFIX}.tmp)
+    endif ()
+
     # Makefile.LA
     #DBLASOBJS	+= dneg_tcopy$(TSUFFIX).$(SUFFIX) dlaswp_ncopy$(TSUFFIX).$(SUFFIX)
 
-add_library(kernel OBJECT ${OPENBLAS_SRC})
+    add_library(kernel${TSUFFIX} OBJECT ${OPENBLAS_SRC})
+    set_target_properties(kernel${TSUFFIX} PROPERTIES COMPILE_FLAGS "${KERNEL_DEFINITIONS}")
+    get_target_property(KERNEL_INCLUDE_DIRECTORIES kernel${TSUFFIX} INCLUDE_DIRECTORIES)
+    set_target_properties(kernel${TSUFFIX} PROPERTIES INCLUDE_DIRECTORIES "${KERNEL_INCLUDE_DIRECTORIES};${TARGET_CONF_DIR}")
+endfunction ()
+
+
+set(ADD_COMMONOBJS 1)
+if (${DYNAMIC_ARCH})
+  foreach(TARGET_CORE ${DYNAMIC_CORE})
+      set(BUILD_KERNEL 1)
+      set(KDIR "")
+      set(TSUFFIX "_${TARGET_CORE}")
+      set(KERNEL_DEFINITIONS "-DBUILD_KERNEL -DTABLE_NAME=gotoblas_${TARGET_CORE} -DTS=${TSUFFIX}")
+      build_core("${TARGET_CORE}" "${KDIR}" "${TSUFFIX}" "${KERNEL_DEFINITIONS}")
+      set(ADD_COMMONOBJS 0)
+  endforeach()
+else ()
+  set(TARGET_CONF_DIR ${PROJECT_BINARY_DIR})
+  set(TARGET_CORE ${CORE})
+  set(KDIR "")
+  set(TSUFFIX "")
+  set(KERNEL_DEFINITIONS "")
+  build_core("${TARGET_CORE}" "${KDIR}" "${TSUFFIX}" "${KERNEL_DEFINITIONS}")
+endif ()
+