Merge pull request #75 from xianyi/develop

rebase

Author: martin-frbg

CMakeLists.txt

@@ -6,7 +6,7 @@ cmake_minimum_required(VERSION 2.8.5)
 project(OpenBLAS C ASM)
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 9.dev)
+set(OpenBLAS_PATCH_VERSION 10.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
@@ -249,7 +249,7 @@ if (BUILD_SHARED_LIBS AND BUILD_RELAPACK)
   endif()
 endif()
 
-if (BUILD_SHARED_LIBS AND NOT ${SYMBOLPREFIX}${SYMBOLSUFIX} STREQUAL "")
+if (BUILD_SHARED_LIBS AND NOT ${SYMBOLPREFIX}${SYMBOLSUFFIX} STREQUAL "")
 if (NOT DEFINED ARCH)
   set(ARCH_IN "x86_64")
 else()
@@ -358,10 +358,21 @@ endif()
 
 if(NOT NO_CBLAS)
 	message (STATUS "Generating cblas.h in ${CMAKE_INSTALL_INCLUDEDIR}")
-
 	set(CBLAS_H ${CMAKE_BINARY_DIR}/generated/cblas.h)
 	file(READ ${CMAKE_CURRENT_SOURCE_DIR}/cblas.h CBLAS_H_CONTENTS)
 	string(REPLACE "common" "openblas_config" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
+	if (NOT ${SYMBOLPREFIX} STREQUAL "")
+	string(REPLACE " cblas" " ${SYMBOLPREFIX}cblas" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
+	string(REPLACE " openblas" " ${SYMBOLPREFIX}openblas" CBLAS_H_CONTENTS_NEW	"${CBLAS_H_CONTENTS}")
+	string (REPLACE " ${SYMBOLPREFIX}openblas_complex" " openblas_complex" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
+	string(REPLACE " goto" " ${SYMBOLPREFIX}goto" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
+	endif()
+	if (NOT ${SYMBOLSUFFIX} STREQUAL "")
+	string(REGEX REPLACE "(cblas[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
+	string(REGEX REPLACE "(openblas[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS_NEW "${CBLAS_H_CONTENTS}")
+	string(REGEX REPLACE "(openblas_complex[^ ]*)${SYMBOLSUFFIX}" "\\1" CBLAS_H_CONTENTS	"${CBLAS_H_CONTENTS_NEW}")
+	string(REGEX REPLACE "(goto[^ (]*)" "\\1${SYMBOLSUFFIX}" CBLAS_H_CONTENTS_NEW	"${CBLAS_H_CONTENTS}")
+	endif()
 	file(WRITE ${CBLAS_H} "${CBLAS_H_CONTENTS_NEW}")
 	install (FILES ${CBLAS_H} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 endif()

Makefile

@@ -365,11 +365,12 @@ clean ::
 	@$(MAKE) -C kernel clean
 #endif
 	@$(MAKE) -C reference clean
-	@rm -f *.$(LIBSUFFIX) *.so *~ *.exe getarch getarch_2nd *.dll *.lib *.$(SUFFIX) *.dwf $(LIBPREFIX).$(LIBSUFFIX) $(LIBPREFIX)_p.$(LIBSUFFIX) $(LIBPREFIX).so.$(MAJOR_VERSION) *.lnk myconfig.h
+	@rm -f *.$(LIBSUFFIX) *.so *~ *.exe getarch getarch_2nd *.dll *.lib *.$(SUFFIX) *.dwf $(LIBPREFIX).$(LIBSUFFIX) $(LIBPREFIX)_p.$(LIBSUFFIX) $(LIBPREFIX).so.$(MAJOR_VERSION) *.lnk myconfig.h *.so.renamed *.a.renamed *.so.0
 ifeq ($(OSNAME), Darwin)
 	@rm -rf getarch.dSYM getarch_2nd.dSYM
 endif
 	@rm -f Makefile.conf config.h Makefile_kernel.conf config_kernel.h st* *.dylib
+	@rm -f cblas.tmp cblas.tmp2
 	@touch $(NETLIB_LAPACK_DIR)/make.inc
 	@$(MAKE) -C $(NETLIB_LAPACK_DIR) clean
 	@rm -f $(NETLIB_LAPACK_DIR)/make.inc $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke_mangling.h

Makefile.install

@@ -45,7 +45,22 @@ install : 	lib.grd
 
 ifndef NO_CBLAS
 	@echo Generating cblas.h in $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
-	@sed 's/common/openblas_config/g' cblas.h > "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/cblas.h"
+	@cp cblas.h cblas.tmp
+ifdef SYMBOLPREFIX
+	@sed 's/cblas[^( ]*/$(SYMBOLPREFIX)&/g' cblas.tmp > cblas.tmp2
+	@sed 's/openblas[^( ]*/$(SYMBOLPREFIX)&/g'  cblas.tmp2 > cblas.tmp
+	#change back any openblas_complex_float and double that got hit
+	@sed 's/$(SYMBOLPREFIX)openblas_complex_/openblas_complex_/g'  cblas.tmp > cblas.tmp2
+	@sed 's/goto[^( ]*/$(SYMBOLPREFIX)&/g'  cblas.tmp2 > cblas.tmp
+endif
+ifdef SYMBOLSUFFIX
+	@sed 's/cblas[^( ]*/&$(SYMBOLSUFFIX)/g' cblas.tmp > cblas.tmp2
+	@sed 's/openblas[^( ]*/&$(SYMBOLSUFFIX)/g'  cblas.tmp2 > cblas.tmp
+	#change back any openblas_complex_float and double that got hit
+	@sed 's/\(openblas_complex_\)\([^ ]*\)$(SYMBOLSUFFIX)/\1\2 /g'  cblas.tmp > cblas.tmp2
+	@sed 's/goto[^( ]*/&$(SYMBOLSUFFIX)/g'  cblas.tmp2 > cblas.tmp
+endif
+	@sed 's/common/openblas_config/g' cblas.tmp > "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/cblas.h"
 endif
 
 ifneq ($(OSNAME), AIX)
@@ -168,4 +183,3 @@ endif
 	@echo "  endif ()" >> "$(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG_VERSION)"
 	@echo "endif ()" >> "$(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG_VERSION)"
 	@echo Install OK!
-

Makefile.power

@@ -10,54 +10,36 @@ USE_OPENMP = 1
 endif
 
 ifeq ($(CORE), POWER10)
-ifeq ($(USE_OPENMP), 1)
-COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx  -DUSE_OPENMP -fno-fast-math -fopenmp
-FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10  -DUSE_OPENMP -fno-fast-math -fopenmp
-else
 COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx  -fno-fast-math
 FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10  -fno-fast-math
 endif
-endif
 
 ifeq ($(CORE), POWER9)
-ifeq ($(USE_OPENMP), 1)
 ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx  -DUSE_OPENMP -fno-fast-math -fopenmp
-else
-CCOMMON_OPT += -fast -Mvect=simd -Mcache_align -DUSE_OPENMP -mp
-endif
-ifneq ($(F_COMPILER), PGI)
-FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9  -DUSE_OPENMP -fno-fast-math -fopenmp
+CCOMMON_OPT += -Ofast -mvsx -fno-fast-math
+ifneq ($(GCCVERSIONGT4), 1)
+$(warning your compiler is too old to fully support POWER9, getting a newer version of gcc is recommended)
+CCOMMON_OPT += -mcpu=power8 -mtune=power8 
 else
-FCOMMON_OPT += -O2 -Mrecursive -DUSE_OPENMP -mp
+CCOMMON_OPT += -mcpu=power9 -mtune=power9 
 endif
 else
-ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx  -fno-fast-math
-else
 CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
 endif
 ifneq ($(F_COMPILER), PGI)
-FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9  -fno-fast-math
+FCOMMON_OPT += -O2 -frecursive -fno-fast-math
+ifneq ($(GCCVERSIONGT4), 1)
+$(warning your compiler is too old to fully support POWER9, getting a newer version of gcc is recommended)
+FCOMMON_OPT += -mcpu=power8 -mtune=power8 
 else
-FCOMMON_OPT += -O2 -Mrecursive
+FCOMMON_OPT += -mcpu=power9 -mtune=power9 
 endif
+else
+FCOMMON_OPT += -O2 -Mrecursive
 endif
 endif
 
 ifeq ($(CORE), POWER8)
-ifeq ($(USE_OPENMP), 1)
-ifneq ($(C_COMPILER), PGI)
-CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx  -DUSE_OPENMP -fno-fast-math -fopenmp
-else
-CCOMMON_OPT += -fast -Mvect=simd -Mcache_align -DUSE_OPENMP -mp
-endif
-ifneq ($(F_COMPILER), PGI)
-FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8  -DUSE_OPENMP -fno-fast-math -fopenmp
-else
-FCOMMON_OPT += -O2 -Mrecursive -DUSE_OPENMP -mp
-endif
-else
 ifneq ($(C_COMPILER), PGI)
 CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx  -fno-fast-math
 else
@@ -73,6 +55,18 @@ else
 FCOMMON_OPT += -O2 -Mrecursive
 endif
 endif
+
+ifeq ($(USE_OPENMP), 1)
+ifneq ($(C_COMPILER), PGI)
+CCOMMON_OPT += -DUSE_OPENMP -fopenmp
+else
+CCOMMON_OPT += -DUSE_OPENMP -mp
+endif
+ifneq ($(F_COMPILER), PGI)
+FCOMMON_OPT += -DUSE_OPENMP -fopenmp
+else
+FCOMMON_OPT += -DUSE_OPENMP -mp
+endif
 endif
 
 # workaround for C->FORTRAN ABI violation in LAPACKE

cpuid_power.c

@@ -38,6 +38,7 @@
 
 #include  <sys/utsname.h>
 #ifdef _AIX
+#include <sys/systemcfg.h>
 #include <sys/vminfo.h>
 #endif
 #ifdef __APPLE__
@@ -137,35 +138,19 @@ int detect(void){
 #endif
 
 #ifdef _AIX
-  FILE *infile;
-  char buffer[512], *p;
-
-  p = (char *)NULL;
-  infile = popen("prtconf|grep 'Processor Type'", "r");
-  while (fgets(buffer, sizeof(buffer), infile)){
-    if (!strncmp("Pro", buffer, 3)){
-	p = strchr(buffer, ':') + 2;
-#if 0
-	fprintf(stderr, "%s\n", p);
-#endif
-	break;
-      }
-  }
-
-  pclose(infile);
+  // Cast from int to unsigned to ensure comparisons work for all bits in
+  // the bit mask, even the top bit
+  unsigned implementation = (unsigned) _system_configuration.implementation;
 
-  if (strstr(p, "POWER3")) return CPUTYPE_POWER3;
-  if (strstr(p, "POWER4")) return CPUTYPE_POWER4;
-  if (strstr(p, "PPC970")) return CPUTYPE_PPC970;
-  if (strstr(p, "POWER5")) return CPUTYPE_POWER5;
-  if (strstr(p, "POWER6")) return CPUTYPE_POWER6;
-  if (strstr(p, "POWER7")) return CPUTYPE_POWER6;
-  if (strstr(p, "POWER8")) return CPUTYPE_POWER8;
-  if (strstr(p, "POWER9")) return CPUTYPE_POWER9;
-  if (strstr(p, "POWER10")) return CPUTYPE_POWER10;
-  if (strstr(p, "Cell")) return CPUTYPE_CELL;
-  if (strstr(p, "7447")) return CPUTYPE_PPCG4;
-  return CPUTYPE_POWER5;
+  if (implementation >= 0x40000u) return CPUTYPE_POWER10;
+  else if (implementation & 0x20000) return CPUTYPE_POWER9;
+  else if (implementation & 0x10000) return CPUTYPE_POWER8;
+  else if (implementation & 0x08000) return CPUTYPE_POWER7; // POWER 7
+  else if (implementation & 0x04000) return CPUTYPE_POWER6;
+  else if (implementation & 0x02000) return CPUTYPE_POWER5;
+  else if (implementation & 0x01000) return CPUTYPE_POWER4; // MPC7450
+  else if (implementation & 0x00800) return CPUTYPE_POWER4;
+  else return CPUTYPE_POWER3;
 #endif
 
 #ifdef __APPLE__

driver/others/Makefile

@@ -47,8 +47,10 @@ endif
 endif
 
 ifdef USE_CUDA
+ifeq ($(USE_CUDA), 1)
 COMMONOBJS	+= cuda_init.$(SUFFIX)
 endif
+endif
 
 ifdef FUNCTION_PROFILE
 COMMONOBJS	+= profile.$(SUFFIX)

kernel/power/KERNEL.POWER10

@@ -187,12 +187,12 @@ ZSWAPKERNEL  = zswap.c
 #
 
 SGEMVNKERNEL = sgemv_n.c
-DGEMVNKERNEL = dgemv_n.c
+DGEMVNKERNEL = dgemv_n_power10.c
 CGEMVNKERNEL = cgemv_n.c
 ZGEMVNKERNEL = zgemv_n_4.c
 #
 SGEMVTKERNEL = sgemv_t.c
-DGEMVTKERNEL = dgemv_t.c
+DGEMVTKERNEL = dgemv_t_power10.c
 CGEMVTKERNEL = cgemv_t.c
 ZGEMVTKERNEL = zgemv_t_4.c