Merge pull request #2398 from xianyi/develop

Update from develop in preparation of the 0.3.8 release

Author: martin-frbg

.travis.yml

@@ -17,7 +17,7 @@ matrix:
         - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
       script:
         - set -e
-        - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
+        - make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
         - make -C test $COMMON_FLAGS $BTYPE
         - make -C ctest $COMMON_FLAGS $BTYPE
         - make -C utest $COMMON_FLAGS $BTYPE
@@ -160,18 +160,25 @@ matrix:
       os: osx
       osx_image: xcode10.1
       before_script:
-        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
+        - COMMON_FLAGS="DYNAMIC_ARCH=1 NUM_THREADS=32"
         - brew update
-        - brew install gcc # for gfortran
+        - brew install gcc@8 # for gfortran
       script:
         - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
       env:
-        - BTYPE="BINARY=64 INTERFACE64=1"
+        - BTYPE="TARGET=NEHALEM BINARY=64 INTERFACE64=1 FC=gfortran-8"
 
     - <<: *test-macos
-      osx_image: xcode8.3
+      osx_image: xcode10.0
       env:
-        - BTYPE="BINARY=32"
+        - BTYPE="TARGET=NEHALEM BINARY=32 NOFORTRAN=1"
+
+    - <<: *test-macos
+      osx_image: xcode10.1
+      env:
+        - CC="/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk"
+        - CFLAGS="-O2 -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch arm64 -miphoneos-version-min=10.0"
+        - BTYPE="TARGET=ARMV8 BINARY=64 HOSTCC=clang NOFORTRAN=1"
 
 # whitelist
 branches:

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 7)
+set(OpenBLAS_PATCH_VERSION 8)
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions

CONTRIBUTORS.md

@@ -171,3 +171,12 @@ In chronological order:
   * [2019-02-01] added missing Blas Level-1,2 (single precision)  simd codes
   * [2019-03-14] power9 dgemm/dtrmm kernel
   * [2019-04-29] power9 sgemm/strmm kernel 
+
+* Jiachen Wang <https://github.com/wjc404>
+  * [2019-07-29] optimize AVX2 DGEMM
+  * [2019-10-20] AVX512 DGEMM kernel (4x8)
+  * [2019-11-06] optimize AVX512 SGEMM
+  * [2019-11-12] AVX512 CGEMM & ZGEMM kernels
+  * [2019-12-23] optimize AVX2 CGEMM and ZGEMM
+  * [2019-12-30] AVX2 CGEMM3M & ZGEMM3M kernels
+  * [2020-01-07] optimize AVX2 SGEMM and STRMM

Changelog.txt

@@ -1,46 +1,100 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.8
+ 9-Feb-2020
+
+common: 
+`	* LAPACK has been updated to 3.9.0 (plus patches up to 
+	  January 2nd, 2020)
+	* CMAKE support has been improved in several areas including
+	  cross-compilation
+	* a thread race condition in the GEMM3M kernels was resolved
+	* the "generic" (plain C) gemm beta kernel used by many targets
+	  has been sped up
+	* an optimized version of the LAPACK trtrs functions has been added
+	* an incompatibilty between the LAPACK tests and the OpenBLAS
+	  implementation of XERBLA was resolved, removing the numerous
+	  warnings about wrong error exits in the former 
+	* support for NetBSD has been added
+	* support for compilation with g95 and non-GNU versions of ld
+	  has been improved
+	* support for compilation with (upcoming) gcc 10 has been added
+
+POWER:
+	* worked around miscompilation of several POWER8 and POWER9 
+	  kernels by older versions of gcc
+	* added support for big-endian POWER8 and for compilation on AIX
+	* corrected bugs in the big-endian support for PPC440 and PPC970
+	* DYNAMIC_ARCH support is now available in CMAKE builds as well
+
+ARMV8:
+	* performance of DGEMM_BETA and SGEMM_NCOPY has been improved
+	* compilation for 32bit works again 
+	* performance of the RPCC function has been improved
+	* improved performance on small systems
+	* DYNAMIC_ARCH support is now available in CMAKE builds as well
+	* cross-compilation from OSX to IOS was simplified
+ 
+x86_64:
+	* a new AVX512 DGEMM kernel was added and the AVX512 SGEMM kernel
+	  was significantly improved
+	* optimized AVX512 kernels for CGEMM and ZGEMM have been added
+	* AVX2 kernels for STRMM, SGEMM, and CGEMM have been significantly
+	  sped up and optimized CGEMM3M and ZGEMM3M kernels have been added 
+	* added support for QEMU virtual cpus
+	* a compilation problem with PGI and SUN compilers was fixed
+	* Intel "Goldmont plus" is now autodetected
+	* a potential crash on program exit on MS Windows has been fixed
+
+x86:
+	* an unwanted case sensitivity in the implementation of LSAME
+  	  on older 32bit AMD cpus was fixed
+
+zarch:
+	* Z15 is now supported as Z14
+	* DYNAMIC_ARCH is now available on ZARCH as well
+
 ====================================================================
 Version 0.3.7
 11-Aug 2019
 
 common:
-	* having the gmake special variables TARGET_ARCH or TARGET_MACH
-	  defined no longer causes build failures in ctest or utest
-	* defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
-	  has the same effect as setting them to 1
-	* a new test program was added to allow checking the library for
-	  thread safety
-	* a new option USE_LOCKING was added to ensure thread safety when
-	  OpenBLAS itself is built without multithreading but will be 
-	  called from multiple threads.
-	* a build failure on Linux with glibc versions earlier than 2.5
-	  was fixed
-	* a runtime error with CPU enumeration (and NO_AFFINITY not set)
-	  on glibc 2.6 was fixed
-	* NO_AFFINITY was added to the CMAKE options (and defaults to being
-	  active on Linux, as in the gmake builds)
+  * having the gmake special variables TARGET_ARCH or TARGET_MACH
+    defined no longer causes build failures in ctest or utest
+  * defining NO_AFFINITY or USE_TLS to 0 in gmake builds no longer
+    has the same effect as setting them to 1
+  * a new test program was added to allow checking the library for
+    thread safety
+  * a new option USE_LOCKING was added to ensure thread safety when
+    OpenBLAS itself is built without multithreading but will be 
+    called from multiple threads.
+  * a build failure on Linux with glibc versions earlier than 2.5
+    was fixed
+  * a runtime error with CPU enumeration (and NO_AFFINITY not set)
+    on glibc 2.6 was fixed
+  * NO_AFFINITY was added to the CMAKE options (and defaults to being
+    active on Linux, as in the gmake builds)
 
 x86_64:
-	* the build-time logic for detection of AVX512 availability in
-	  the processor and compiler was fixed
-	* gmake builds on OSX now set the internal name of the library to 
-	  libopenblas.0.dylib (consistent with CMAKE)
- 	* the Haswell DGEMM kernel received a significant speedup through
-	  improved prefetch and load instructions
-	* performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
-	  increased by avoiding vpermpd instructions
- 	* the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
-	  to fix remaining errors in DGEMM, DSYMM and DTRMM
-
-## POWER:
-	* added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
-	* added optimized kernels for POWER9 single and double precision complex BLAS3
-	* added optimized kernels for POWER9 SGEMM and STRMM
-
-## ARMV7:
-	* fixed the softfp implementations of xAMAX and IxAMAX
-	* removed the predefined -march= flags on both ARMV5 and ARMV6 as
-	  they were appropriate for only a subset of platforms
+  * the build-time logic for detection of AVX512 availability in
+    the processor and compiler was fixed
+  * gmake builds on OSX now set the internal name of the library to 
+    libopenblas.0.dylib (consistent with CMAKE)
+  * the Haswell DGEMM kernel received a significant speedup through
+    improved prefetch and load instructions
+  * performance of DGEMM, DTRMM, DTRSM and ZDOT on Zen/Zen2 was markedly
+    increased by avoiding vpermpd instructions
+  * the SKYLAKEX (AVX512) DGEMM helper functions have now been disabled
+    to fix remaining errors in DGEMM, DSYMM and DTRMM
+
+POWER:
+  * added support for building on FreeBSD/powerpc64 and FreeBSD/ppc970
+  * added optimized kernels for POWER9 SGEMM and STRMM
+
+ARMV7:
+  * fixed the softfp implementations of xAMAX and IxAMAX
+  * removed the predefined -march= flags on both ARMV5 and ARMV6 as
+    they were appropriate for only a subset of platforms
 
 ====================================================================
 Version 0.3.6

Makefile

@@ -247,21 +247,21 @@ prof_lapack : lapack_prebuild
 
 lapack_prebuild :
 ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
-	-@echo "FORTRAN     = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "OPTS        = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "FC          = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "FFLAGS      = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "POPTS       = $(LAPACK_FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "NOOPT       = -O0 $(LAPACK_NOOPT)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "FFLAGS_NOOPT       = -O0 $(LAPACK_NOOPT)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "PNOOPT      = $(LAPACK_FPFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "LOADOPTS    = $(FFLAGS) $(EXTRALIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "LDFLAGS     = $(FFLAGS) $(EXTRALIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "CC          = $(CC)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "override CFLAGS      = $(LAPACK_CFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "override ARCH        = $(AR)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "ARCHFLAGS   = $(ARFLAGS) -ru" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "AR          = $(AR)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "ARFLAGS     = $(ARFLAGS) -ru" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "RANLIB      = $(RANLIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "LAPACKLIB   = ../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "TMGLIB      = ../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "LAPACKLIB   = ../../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "TMGLIB      = ../../../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "BLASLIB     = ../../../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
-	-@echo "LAPACKELIB  = ../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "LAPACKELIB  = ../../../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "LAPACKLIB_P = ../$(LIBNAME_P)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "SUFFIX      = $(SUFFIX)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "PSUFFIX     = $(PSUFFIX)" >> $(NETLIB_LAPACK_DIR)/make.inc
@@ -319,7 +319,7 @@ lapack-test :
 ifneq ($(CROSS), 1)
 	( cd $(NETLIB_LAPACK_DIR)/INSTALL; make all; ./testlsame; ./testslamch; ./testdlamch; \
         ./testsecond; ./testdsecnd; ./testieee; ./testversion )
-	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r )
+	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r -b TESTING)
 endif
 
 lapack-runtest:

Makefile.arm64

@@ -39,7 +39,10 @@ CCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 FCOMMON_OPT += -march=armv8.1-a -mtune=thunderx2t99
 endif
 
+ifeq ($(GCCVERSIONGTEQ9), 1)
 ifeq ($(CORE), TSV110)
 CCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
 FCOMMON_OPT += -march=armv8.2-a -mtune=tsv110
 endif
+endif
+

Makefile.install

@@ -51,6 +51,7 @@ endif
 ifneq ($(OSNAME), AIX)
 ifndef NO_LAPACKE
 	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
+	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapack.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapack.h"
 	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h"
 	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_config.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_config.h"
 	@-install -pm644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_mangling_with_flags.h.in "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h"
@@ -100,6 +101,7 @@ else
 #install on AIX has different options syntax
 ifndef NO_LAPACKE
 	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
+	@-installbsd -c -m 644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapack.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapack.h"
 	@-installbsd -c -m 644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h"
 	@-installbsd -c -m 644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_config.h "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_config.h"
 	@-installbsd -c -m 644 $(NETLIB_LAPACK_DIR)/LAPACKE/include/lapacke_mangling_with_flags.h.in "$(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h"

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.7
+VERSION = 0.3.8
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library

Makefile.system

@@ -9,9 +9,11 @@ ifndef TOPDIR
 TOPDIR = .
 endif
 
-# If ARCH is not set, we use the host system's architecture.
+# If ARCH is not set, we use the host system's architecture for getarch compile options.
 ifndef ARCH
-ARCH := $(shell uname -m)
+HOSTARCH := $(shell uname -m)
+else
+HOSTARCH = $(ARCH)
 endif
 
 # Catch conflicting usage of ARCH in some BSD environments
@@ -23,6 +25,8 @@ else ifeq ($(ARCH), i386)
 override ARCH=x86
 else ifeq ($(ARCH), aarch64)
 override ARCH=arm64
+else ifeq ($(ARCH), zarch)
+override ARCH=zarch
 endif
 
 NETLIB_LAPACK_DIR = $(TOPDIR)/lapack-netlib
@@ -142,9 +146,9 @@ endif
 endif
 
 
-# On x86_64 build getarch with march=native. This is required to detect AVX512 support in getarch.
-ifeq ($(ARCH), x86_64)
-ifneq ($(C_COMPILER), PGI)
+# On x86_64 build getarch with march=native unless the compiler is PGI. This is required to detect AVX512 support in getarch.
+ifeq ($(HOSTARCH), x86_64)
+ifeq ($(findstring pgcc,$(HOSTCC)),)
 GETARCH_FLAGS += -march=native
 endif
 endif
@@ -320,12 +324,14 @@ CCOMMON_OPT	+= -DMS_ABI
 endif
 
 ifeq ($(C_COMPILER), GCC)
-#Test for supporting MS_ABI
+#Version tests for supporting specific features (MS_ABI, POWER9 intrinsics)
 GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
 GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
+GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
+GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
 GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
 ifeq ($(GCCVERSIONGT4), 1)
-# GCC Majar version > 4
+# GCC Major version > 4
 # It is compatible with MSVC ABI.
 CCOMMON_OPT	+= -DMS_ABI
 endif
@@ -544,15 +550,34 @@ endif
 
 ifeq ($(ARCH), arm64)
 DYNAMIC_CORE =  ARMV8
+DYNAMIC_CORE += CORTEXA53
 DYNAMIC_CORE += CORTEXA57
+DYNAMIC_CORE += CORTEXA72
+DYNAMIC_CORE += CORTEXA73
+DYNAMIC_CORE += FALKOR
 DYNAMIC_CORE += THUNDERX
 DYNAMIC_CORE += THUNDERX2T99
+DYNAMIC_CORE += TSV110
+endif
+
+ifeq ($(ARCH), zarch)
+DYNAMIC_CORE = Z13
+DYNAMIC_CORE += Z14
 endif
 
 ifeq ($(ARCH), power)
 DYNAMIC_CORE = POWER6
 DYNAMIC_CORE += POWER8
+ifneq ($(C_COMPILER), GCC)
+DYNAMIC_CORE += POWER9
+endif
+ifeq ($(C_COMPILER), GCC)
+ifeq ($(GCCVERSIONGT5), 1)
 DYNAMIC_CORE += POWER9
+else
+$(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
+endif
+endif
 endif
 
 # If DYNAMIC_CORE is not set, DYNAMIC_ARCH cannot do anything, so force it to empty
@@ -697,7 +722,7 @@ endif
 
 ifeq ($(C_COMPILER), PGI)
 ifdef BINARY64
-CCOMMON_OPT += -tp p7-64
+CCOMMON_OPT += -tp p7-64 -D__MMX__ -Mnollvm
 else
 CCOMMON_OPT += -tp p7
 endif
@@ -757,6 +782,9 @@ else
 FCOMMON_OPT += -m32
 endif
 endif
+ifneq ($(NO_LAPACKE), 1)
+FCOMMON_OPT += -fno-second-underscore
+endif
 endif
 endif