Merge pull request #1656 from xianyi/develop

Update the 0.3 branch from develop

CMakeLists.txt

@@ -6,12 +6,15 @@ 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 0.dev)
+set(OpenBLAS_PATCH_VERSION 1.dev)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
 include(GNUInstallDirs)
 
+include(CMakePackageConfigHelpers)
+
+
 set(OpenBLAS_LIBNAME openblas)
 
 #######
@@ -20,6 +23,7 @@ option(BUILD_WITHOUT_LAPACK "Without LAPACK and LAPACKE (Only BLAS or CBLAS)" ON
 endif()
 option(BUILD_WITHOUT_CBLAS "Without CBLAS" OFF)
 option(DYNAMIC_ARCH "Build with DYNAMIC_ARCH" OFF)
+option(DYNAMIC_OLDER "Support older cpus with DYNAMIC_ARCH" OFF)
 option(BUILD_RELAPACK "Build with ReLAPACK (recursive LAPACK" OFF)
 #######
 if(BUILD_WITHOUT_LAPACK)
@@ -208,6 +212,7 @@ set_target_properties(${OpenBLAS_LIBNAME} PROPERTIES
 
 # Install libraries
 install(TARGETS ${OpenBLAS_LIBNAME}
+	EXPORT "OpenBLASTargets"
 	RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
 	ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
   LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR} )
@@ -267,3 +272,21 @@ if(PKG_CONFIG_FOUND)
 	configure_file(${PROJECT_SOURCE_DIR}/cmake/openblas.pc.in ${PROJECT_BINARY_DIR}/openblas.pc @ONLY)
 	install (FILES ${PROJECT_BINARY_DIR}/openblas.pc DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig/)
 endif()
+
+
+# GNUInstallDirs "DATADIR" wrong here; CMake search path wants "share".
+set(PN OpenBLAS)
+set(CMAKECONFIG_INSTALL_DIR "share/cmake/${PN}")
+configure_package_config_file(cmake/${PN}Config.cmake.in
+                              "${CMAKE_CURRENT_BINARY_DIR}/${PN}Config.cmake"
+                              INSTALL_DESTINATION ${CMAKECONFIG_INSTALL_DIR})
+write_basic_package_version_file(${CMAKE_CURRENT_BINARY_DIR}/${PN}ConfigVersion.cmake
+                                 VERSION ${${PN}_VERSION}
+                                 COMPATIBILITY AnyNewerVersion)
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${PN}Config.cmake
+              ${CMAKE_CURRENT_BINARY_DIR}/${PN}ConfigVersion.cmake
+        DESTINATION ${CMAKECONFIG_INSTALL_DIR})
+install(EXPORT "${PN}Targets"
+        NAMESPACE "${PN}::"
+        DESTINATION ${CMAKECONFIG_INSTALL_DIR})
+

Makefile

@@ -153,6 +153,9 @@ ifeq ($(DYNAMIC_ARCH), 1)
 	do  $(MAKE) GOTOBLAS_MAKEFILE= -C kernel TARGET_CORE=$$d kernel || exit 1 ;\
 	done
 	@echo DYNAMIC_ARCH=1 >> Makefile.conf_last
+ifeq ($(DYNAMIC_OLDER), 1)
+	@echo DYNAMIC_OLDER=1 >> Makefile.conf_last
+endif	
 endif
 ifdef USE_THREAD
 	@echo USE_THREAD=$(USE_THREAD) >>  Makefile.conf_last
@@ -294,9 +297,10 @@ endif
 
 lapack-test :
 	(cd $(NETLIB_LAPACK_DIR)/TESTING && rm -f x* *.out)
-	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING xeigtstc  xeigtstd  xeigtsts  xeigtstz  xlintstc  xlintstd  xlintstds  xlintstrfd  xlintstrfz  xlintsts  xlintstz  xlintstzc xlintstrfs xlintstrfc
+	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING/EIG xeigtstc  xeigtstd  xeigtsts  xeigtstz 
+	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING/LIN xlintstc  xlintstd  xlintstds  xlintstrfd  xlintstrfz  xlintsts  xlintstz  xlintstzc xlintstrfs xlintstrfc
 ifneq ($(CROSS), 1)
-	( cd $(NETLIB_LAPACK_DIR)/INSTALL; ./testlsame; ./testslamch; ./testdlamch; \
+	( cd $(NETLIB_LAPACK_DIR)/INSTALL; make all; ./testlsame; ./testslamch; ./testdlamch; \
         ./testsecond; ./testdsecnd; ./testieee; ./testversion )
 	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r )
 endif
@@ -308,9 +312,9 @@ lapack-runtest:
 
 
 blas-test:
-	(cd $(NETLIB_LAPACK_DIR)/BLAS && rm -f x* *.out)
+	(cd $(NETLIB_LAPACK_DIR)/BLAS/TESTING && rm -f x* *.out)
 	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR) blas_testing
-	(cd $(NETLIB_LAPACK_DIR)/BLAS && cat *.out)
+	(cd $(NETLIB_LAPACK_DIR)/BLAS/TESTING && cat *.out)
 
 
 dummy :

Makefile.install

@@ -98,7 +98,7 @@ endif
 	@echo Generating openblas.pc in "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)"
 	@echo 'libdir='$(OPENBLAS_LIBRARY_DIR) > "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@echo 'includedir='$(OPENBLAS_INCLUDE_DIR) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
-	@echo 'openblas_config= USE_64BITINT='$(USE_64BITINT) 'DYNAMIC_ARCH='$(DYNAMIC_ARCH) 'NO_CBLAS='$(NO_CBLAS) 'NO_LAPACK='$(NO_LAPACK) 'NO_LAPACKE='$(NO_LAPACKE) 'NO_AFFINITY='$(NO_AFFINITY) 'USE_OPENMP='$(USE_OPENMP) $(CORE) 'MAX_THREADS='$(NUM_THREADS)>> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
+	@echo 'openblas_config= USE_64BITINT='$(USE_64BITINT) 'DYNAMIC_ARCH='$(DYNAMIC_ARCH) 'DYNAMIC_OLDER='$(DYNAMIC_OLDER) 'NO_CBLAS='$(NO_CBLAS) 'NO_LAPACK='$(NO_LAPACK) 'NO_LAPACKE='$(NO_LAPACKE) 'NO_AFFINITY='$(NO_AFFINITY) 'USE_OPENMP='$(USE_OPENMP) $(CORE) 'MAX_THREADS='$(NUM_THREADS)>> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@echo 'version='$(VERSION) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@echo 'extralib='$(EXTRALIB) >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"
 	@cat openblas.pc.in >> "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/openblas.pc"

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.0.dev
+VERSION = 0.3.1.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
@@ -17,6 +17,11 @@ VERSION = 0.3.0.dev
 # If you want to support multiple architecture in one binary
 # DYNAMIC_ARCH = 1
 
+# If you want the full list of x86_64 architectures supported in DYNAMIC_ARCH
+# mode (including individual optimizied codes for PENRYN, DUNNINGTON, OPTERON,
+# OPTERON_SSE3, ATOM and NANO rather than fallbacks to older architectures)
+# DYNAMIC_OLDER = 1
+
 # C compiler including binary type(32bit / 64bit). Default is gcc.
 # Don't use Intel Compiler or PGI, it won't generate right codes as I expect.
 # CC = gcc
@@ -55,6 +60,14 @@ VERSION = 0.3.0.dev
 # This flag is always set for POWER8. Don't modify the flag 
 # USE_OPENMP = 1
 
+# The OpenMP scheduler to use - by default this is "static" and you
+# will normally not want to change this unless you know that your main
+# workload will involve tasks that have highly unbalanced running times
+# for individual threads. Changing away from "static" may also adversely
+# affect memory access locality in NUMA systems. Setting to "runtime" will
+# allow you to select the scheduler from the environment variable OMP_SCHEDULE
+# CCOMMON_OPT += -DOMP_SCHED=dynamic
+
 # You can define maximum number of threads. Basically it should be
 # less than actual number of cores. If you don't specify one, it's
 # automatically detected by the the script.
@@ -151,8 +164,11 @@ NO_AFFINITY = 1
 # CONSISTENT_FPCSR = 1
 
 # If any gemm arguement m, n or k is less or equal this threshold, gemm will be execute
-# with single thread. You can use this flag to avoid the overhead of multi-threading
-# in small matrix sizes. The default value is 4.
+# with single thread. (Actually in recent versions this is a factor proportional to the
+# number of floating point operations necessary for the given problem size, no longer
+# an individual dimension). You can use this setting to avoid the overhead of multi-
+# threading in small matrix sizes. The default value is 4, but values as high as 50 have 
+# been reported to be optimal for certain workloads (50 is the recommended value for Julia).
 # GEMM_MULTITHREAD_THRESHOLD = 4
 
 # If you need santy check by comparing reference BLAS. It'll be very

Makefile.system

@@ -62,6 +62,9 @@ ifeq ($(BINARY), 32)
 ifeq ($(TARGET), HASWELL)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
+ifeq ($(TARGET), SKYLAKEX)
+GETARCH_FLAGS := -DFORCE_NEHALEM
+endif
 ifeq ($(TARGET), SANDYBRIDGE)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
@@ -95,6 +98,9 @@ ifeq ($(BINARY), 32)
 ifeq ($(TARGET_CORE), HASWELL)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
+ifeq ($(TARGET_CORE), SKYLAKEX)
+GETARCH_FLAGS := -DFORCE_NEHALEM
+endif
 ifeq ($(TARGET_CORE), SANDYBRIDGE)
 GETARCH_FLAGS := -DFORCE_NEHALEM
 endif
@@ -141,6 +147,10 @@ ifeq ($(NO_AVX2), 1)
 GETARCH_FLAGS	+= -DNO_AVX2
 endif
 
+ifeq ($(NO_AVX512), 1)
+GETARCH_FLAGS	+= -DNO_AVX512
+endif
+
 ifeq ($(DEBUG), 1)
 GETARCH_FLAGS	+= -g
 endif
@@ -238,7 +248,7 @@ endif
 
 ifeq ($(OSNAME), Darwin)
 ifndef MACOSX_DEPLOYMENT_TARGET
-export MACOSX_DEPLOYMENT_TARGET=10.6
+export MACOSX_DEPLOYMENT_TARGET=10.8
 endif
 MD5SUM = md5 -r
 endif
@@ -462,13 +472,37 @@ DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
 endif
 
 ifeq ($(ARCH), x86_64)
-DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
+DYNAMIC_CORE = PRESCOTT CORE2 
+ifeq ($(DYNAMIC_OLDER), 1)
+DYNAMIC_CORE += PENRYN DUNNINGTON 
+endif
+DYNAMIC_CORE += NEHALEM 
+ifeq ($(DYNAMIC_OLDER), 1)
+DYNAMIC_CORE += OPTERON OPTERON_SSE3 
+endif
+DYNAMIC_CORE += BARCELONA 
+ifeq ($(DYNAMIC_OLDER), 1)
+DYNAMIC_CORE += BOBCAT ATOM NANO
+endif
 ifneq ($(NO_AVX), 1)
 DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER EXCAVATOR
 endif
 ifneq ($(NO_AVX2), 1)
 DYNAMIC_CORE += HASWELL ZEN
 endif
+ifneq ($(NO_AVX512), 1)
+ifneq ($(NO_AVX2), 1)
+DYNAMIC_CORE += SKYLAKEX
+endif
+endif
+endif
+
+ifdef DYNAMIC_LIST
+override DYNAMIC_CORE = PRESCOTT $(DYNAMIC_LIST)
+XCCOMMON_OPT = -DDYNAMIC_LIST -DDYN_PRESCOTT
+XCCOMMON_OPT += $(foreach dcore,$(DYNAMIC_LIST),-DDYN_$(dcore))
+CCOMMON_OPT += $(XCCOMMON_OPT)
+#CCOMMON_OPT	+= -DDYNAMIC_LIST='$(DYNAMIC_LIST)'
 endif
 
 # If DYNAMIC_CORE is not set, DYNAMIC_ARCH cannot do anything, so force it to empty
@@ -902,6 +936,10 @@ ifeq ($(DYNAMIC_ARCH), 1)
 CCOMMON_OPT	+= -DDYNAMIC_ARCH
 endif
 
+ifeq ($(DYNAMIC_OLDER), 1)
+CCOMMON_OPT	+= -DDYNAMIC_OLDER
+endif
+
 ifeq ($(NO_LAPACK), 1)
 CCOMMON_OPT	+= -DNO_LAPACK
 #Disable LAPACK C interface
@@ -924,6 +962,10 @@ ifeq ($(NO_AVX2), 1)
 CCOMMON_OPT	+= -DNO_AVX2
 endif
 
+ifeq ($(NO_AVX512), 1)
+CCOMMON_OPT	+= -DNO_AVX512
+endif
+
 ifdef SMP
 CCOMMON_OPT	+= -DSMP_SERVER
 
@@ -1230,6 +1272,7 @@ export MSA_FLAGS
 export KERNELDIR
 export FUNCTION_PROFILE
 export TARGET_CORE
+export NO_AVX512
 
 export SGEMM_UNROLL_M
 export SGEMM_UNROLL_N

Makefile.x86_64

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

TargetList.txt

@@ -20,6 +20,7 @@ DUNNINGTON
 NEHALEM
 SANDYBRIDGE
 HASWELL
+SKYLAKEX
 ATOM
 
 b)AMD CPU:

c_check

@@ -201,6 +201,21 @@ $architecture = zarch  if ($data =~ /ARCH_ZARCH/);
 $binformat    = bin32;
 $binformat    = bin64  if ($data =~ /BINARY_64/);
 
+$no_avx512= 0;
+if (($architecture eq "x86") || ($architecture eq "x86_64")) {
+    $code = '"vbroadcastss -4 * 4(%rsi), %zmm2"';
+    print $tmpf "int main(void){ __asm__ volatile($code); }\n";
+    $args = " -march=skylake-avx512 -o $tmpf.o -x c $tmpf";
+    my @cmd = ("$compiler_name $args >/dev/null 2>/dev/null");
+    system(@cmd) == 0;
+    if ($? != 0) {
+	$no_avx512 = 1;
+    } else {
+	$no_avx512 = 0;
+    }
+    unlink("tmpf.o");
+}
+
 $data = `$compiler_name -S ctest1.c && grep globl ctest1.s | head -n 1 && rm -f ctest1.s`;
 
 $data =~ /globl\s([_\.]*)(.*)/;
@@ -288,6 +303,7 @@ print MAKEFILE "CROSS=1\n" if $cross != 0;
 print MAKEFILE "CEXTRALIB=$linker_L $linker_l $linker_a\n";
 print MAKEFILE "HAVE_MSA=1\n" if $have_msa eq 1;
 print MAKEFILE "MSA_FLAGS=$msa_flags\n" if $have_msa eq 1;
+print MAKEFILE "NO_AVX512=1\n" if $no_avx512 eq 1;
 
 $os           =~ tr/[a-z]/[A-Z]/;
 $architecture =~ tr/[a-z]/[A-Z]/;

cblas.h

@@ -82,6 +82,11 @@ CBLAS_INDEX cblas_idamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPE
 CBLAS_INDEX cblas_icamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx);
 CBLAS_INDEX cblas_izamax(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx);
 
+CBLAS_INDEX cblas_isamin(OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx);
+CBLAS_INDEX cblas_idamin(OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx);
+CBLAS_INDEX cblas_icamin(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx);
+CBLAS_INDEX cblas_izamin(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx);
+
 void cblas_saxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST float alpha, OPENBLAS_CONST float *x, OPENBLAS_CONST blasint incx, float *y, OPENBLAS_CONST blasint incy);
 void cblas_daxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST double alpha, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx, double *y, OPENBLAS_CONST blasint incy);
 void cblas_caxpy(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx, void *y, OPENBLAS_CONST blasint incy);