Merge branch 'develop'

OpenMathLib · Mar 24, 2015 · d0c51c4 · d0c51c4
2 parents 51ce5ef + a3491e1
commit d0c51c4
Show file tree

Hide file tree

Showing 137 changed files with 1,947 additions and 444 deletions.
diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
@@ -121,5 +121,11 @@ In chronological order:
   * [2014-10-10] trmm and sgemm kernels (optimized for APM's X-Gene 1).
                  ARMv8 support.
 
+* Dan Kortschak
+  * [2015-01-07] Added test for drotmg bug #484.
+
+* Ton van den Heuvel <https://github.com/ton>
+  * [2015-03-18] Fix race condition during shutdown causing a crash in gotoblas_set_affinity().
+
 * [Your name or handle] <[email or website]>
   * [Date] [Brief summary of your changes]
diff --git a/Changelog.txt b/Changelog.txt
@@ -1,4 +1,24 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.2.14
+24-Mar-2015
+common:
+	* Improve OpenBLASConfig.cmake. (#474, #475. Thanks, xantares.)
+	* Improve ger and gemv for small matrices by stack allocation.
+	  e.g. make -DMAX_STACK_ALLOC=2048 (#482. Thanks, Jerome Robert.)
+	* Introduce openblas_get_num_threads and openblas_get_num_procs.
+	  (#497. Thanks, Erik Schnetter.)
+	* Add ATLAS-style ?geadd function. (#509. Thanks, Martin Köhler.)
+	* Fix c/zsyr bug with negative incx. (#492.)
+	* Fix race condition during shutdown causing a crash in
+	  gotoblas_set_affinity(). (#508. Thanks, Ton van den Heuvel.)
+
+x86/x86-64:
+	* Support AMD Streamroller.
+
+ARM:
+	* Add Cortex-A9 and Cortex-A15 targets.
+
 ====================================================================
 Version 0.2.13
 3-Dec-2014

diff --git a/GotoBLAS_05LargePage.txt b/GotoBLAS_05LargePage.txt
@@ -9,10 +9,10 @@
 
     If you want to allocate 64 large pages,
 
-    $shell> echo  0 > /pros/sys/vm/nr_hugepages		# need to be reset
-    $shell> echo 65 > /pros/sys/vm/nr_hugepages		# add 1 extra page
-    $shell> echo 3355443200 > /pros/sys/kernel/shmmax   # just large number
-    $shell> echo 3355443200 > /pros/sys/kernel/shmall
+    $shell> echo  0 > /proc/sys/vm/nr_hugepages		# need to be reset
+    $shell> echo 65 > /proc/sys/vm/nr_hugepages		# add 1 extra page
+    $shell> echo 3355443200 > /proc/sys/kernel/shmmax   # just large number
+    $shell> echo 3355443200 > /proc/sys/kernel/shmall
 
     Also may add a few lines into /etc/security/limits.conf file.
 

diff --git a/Makefile.arm b/Makefile.arm
@@ -1,3 +1,8 @@
+# ifeq logical or
+ifeq ($(CORE), $(filter $(CORE),CORTEXA9 CORTEXA15))
+CCOMMON_OPT += -marm -mfpu=vfpv3  -mfloat-abi=hard -march=armv7-a
+FCOMMON_OPT += -marm -mfpu=vfpv3  -mfloat-abi=hard -march=armv7-a
+endif
 
 ifeq ($(CORE), ARMV7)
 CCOMMON_OPT += -marm -mfpu=vfpv3  -mfloat-abi=hard -march=armv7-a

diff --git a/Makefile.install b/Makefile.install
@@ -9,7 +9,7 @@ OPENBLAS_INCLUDE_DIR := $(PREFIX)/include
 OPENBLAS_LIBRARY_DIR := $(PREFIX)/lib
 OPENBLAS_BINARY_DIR := $(PREFIX)/bin
 OPENBLAS_BUILD_DIR := $(CURDIR)
-OPENBLAS_CMAKE_DIR := $(PREFIX)/cmake
+OPENBLAS_CMAKE_DIR := $(OPENBLAS_LIBRARY_DIR)/cmake/openblas
 OPENBLAS_CMAKE_CONFIG := OpenBLASConfig.cmake
 
 .PHONY : install
@@ -46,11 +46,11 @@ ifndef NO_CBLAS
 endif
 
 ifndef NO_LAPACKE
-	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
-	@-install -pDm644 $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke.h $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h
-	@-install -pDm644 $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke_config.h $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_config.h
-	@-install -pDm644 $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke_mangling_with_flags.h $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h
-	@-install -pDm644 $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke_utils.h $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_utils.h
+	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
+	@-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 $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h
+	@-install -pm644 $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke_utils.h $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke_utils.h
 endif
 
 #for install static library
@@ -95,7 +95,8 @@ endif
 endif
 #Generating OpenBLASConfig.cmake
 	@echo Generating $(OPENBLAS_CMAKE_CONFIG) in $(DESTDIR)$(OPENBLAS_CMAKE_DIR)
-	@echo "SET(OpenBLAS_INCLUDE_DIRS ${OPENBLAS_INCLUDE_DIR})" > $(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG)
+	@echo "SET(OpenBLAS_VERSION \"${VERSION}\")" > $(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG)	
+	@echo "SET(OpenBLAS_INCLUDE_DIRS ${OPENBLAS_INCLUDE_DIR})" >> $(DESTDIR)$(OPENBLAS_CMAKE_DIR)/$(OPENBLAS_CMAKE_CONFIG)
 ifndef NO_SHARED
 #ifeq logical or
 ifeq ($(OSNAME), $(filter $(OSNAME),Linux FreeBSD NetBSD))

diff --git a/Makefile.rule b/Makefile.rule
@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.2.13
+VERSION = 0.2.14
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
@@ -159,6 +159,19 @@ COMMON_PROF = -pg
 # Build Debug version
 # DEBUG = 1
 
+# Improve GEMV and GER for small matrices by stack allocation.
+# For details, https://github.com/xianyi/OpenBLAS/pull/482
+#
+# MAX_STACK_ALLOC=2048
+
+# Add a prefix or suffix to all exported symbol names in the shared library.
+# Avoid conflicts with other BLAS libraries, especially when using
+# 64 bit integer interfaces in OpenBLAS.
+# For details, https://github.com/xianyi/OpenBLAS/pull/459
+#
+# SYMBOLPREFIX=
+# SYMBOLSUFFIX=
+
 #
 #  End of user configuration
 #
diff --git a/Makefile.system b/Makefile.system
@@ -61,6 +61,9 @@ endif
 ifeq ($(TARGET), PILEDRIVER)
 GETARCH_FLAGS := -DFORCE_BARCELONA
 endif
+ifeq ($(TARGET), STEAMROLLER)
+GETARCH_FLAGS := -DFORCE_BARCELONA
+endif
 endif
 
 
@@ -85,6 +88,9 @@ endif
 ifeq ($(TARGET_CORE), PILEDRIVER)
 GETARCH_FLAGS := -DFORCE_BARCELONA
 endif
+ifeq ($(TARGET_CORE), STEAMROLLER)
+GETARCH_FLAGS := -DFORCE_BARCELONA
+endif
 endif
 
 
@@ -305,6 +311,10 @@ ifdef SANITY_CHECK
 CCOMMON_OPT	+= -DSANITY_CHECK -DREFNAME=$(*F)f$(BU)
 endif
 
+ifdef MAX_STACK_ALLOC
+CCOMMON_OPT	+= -DMAX_STACK_ALLOC=$(MAX_STACK_ALLOC)
+endif
+
 #
 #  Architecture dependent settings
 #
@@ -354,6 +364,12 @@ endif
 
 
 ifeq ($(USE_OPENMP), 1)
+
+#check 
+ifeq ($(USE_THREAD), 0)
+$(error OpenBLAS: Cannot set both USE_OPENMP=1 and USE_THREAD=0. The USE_THREAD=0 is only for building single thread version.)
+endif
+
 # ifeq logical or. GCC or LSB
 ifeq ($(C_COMPILER), $(filter $(C_COMPILER),GCC LSB))
 CCOMMON_OPT    += -fopenmp
@@ -392,7 +408,7 @@ endif
 ifeq ($(ARCH), x86_64)
 DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
 ifneq ($(NO_AVX), 1)
-DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER
+DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER STEAMROLLER
 endif
 ifneq ($(NO_AVX2), 1)
 DYNAMIC_CORE += HASWELL

diff --git a/README.md b/README.md
@@ -60,6 +60,7 @@ Please read GotoBLAS_01Readme.txt
 - **AMD Bobcat**: Used GotoBLAS2 Barcelona codes.
 - **AMD Bulldozer**: x86-64 ?GEMM FMA4 kernels. (Thank Werner Saar)
 - **AMD PILEDRIVER**: Uses Bulldozer codes with some optimizations.
+- **AMD STEAMROLLER**: Uses Bulldozer codes with some optimizations.
 
 #### MIPS64:
 - **ICT Loongson 3A**: Optimized Level-3 BLAS and the part of Level-1,2.

diff --git a/TargetList.txt b/TargetList.txt
@@ -32,6 +32,7 @@ ISTANBUL
 BOBCAT
 BULLDOZER
 PILEDRIVER
+STEAMROLLER
 
 c)VIA CPU:
 SSE_GENERIC
@@ -62,6 +63,11 @@ SPARC
 SPARCV7
 
 6.ARM CPU:
+CORTEXA15
+CORTEXA9
 ARMV7
 ARMV6
 ARMV5
+
+7.ARM 64-bit CPU:
+ARMV8
diff --git a/benchmark/Makefile b/benchmark/Makefile
@@ -6,8 +6,13 @@ include $(TOPDIR)/Makefile.system
 #LIBACML = -fopenmp $(ACML)/libacml_mp.a -lgfortran -lm
 
 # ACML custom
-ACML=/opt/pb/acml-5-3-1-gfortran-64bit/gfortran64_fma4_mp/lib
-LIBACML = -fopenmp $(ACML)/libacml_mp.a -lgfortran -lm
+#ACML=/opt/pb/acml-5-3-1-gfortran-64bit/gfortran64_fma4_mp/lib
+#LIBACML = -fopenmp $(ACML)/libacml_mp.a -lgfortran -lm
+
+# ACML 6.1 custom
+ACML=/home/werner/project/acml6.1/gfortran64_mp/lib
+LIBACML = -fopenmp $(ACML)/libacml_mp.so -lgfortran -lm
+
 
 # Atlas Ubuntu 
 #ATLAS=/usr/lib/atlas-base

diff --git a/benchmark/axpy.c b/benchmark/axpy.c
@@ -114,7 +114,7 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *x, *y;
   FLOAT alpha[2] = { 2.0, 2.0 };
@@ -198,4 +198,4 @@ int MAIN__(int argc, char *argv[]){
   return 0;
 }
 
-void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/cholesky.c b/benchmark/cholesky.c
@@ -117,7 +117,7 @@ static __inline double getmflops(int ratio, int m, double secs){
 }
 
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
 #ifndef COMPLEX
   char *trans[] = {"T", "N"};
@@ -273,4 +273,4 @@ int MAIN__(int argc, char *argv[]){
   return 0;
 }
 
-void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/dot.c b/benchmark/dot.c
@@ -108,7 +108,7 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *x, *y;
   FLOAT result;
@@ -192,4 +192,4 @@ int MAIN__(int argc, char *argv[]){
   return 0;
 }
 
-void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/geev.c b/benchmark/geev.c
@@ -139,7 +139,7 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *a,*vl,*vr,*wi,*wr,*work,*rwork;
   FLOAT wkopt[4];
@@ -257,4 +257,4 @@ int MAIN__(int argc, char *argv[]){
   return 0;
 }
 
-void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/gemm.c b/benchmark/gemm.c
@@ -118,14 +118,15 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *a, *b, *c;
   FLOAT alpha[] = {1.0, 1.0};
   FLOAT beta [] = {1.0, 1.0};
   char trans='N';
-  blasint m, i, j;
+  blasint m, n, i, j;
   int loops = 1;
+  int has_param_n=0;
   int l;
   char *p;
 
@@ -162,6 +163,11 @@ int MAIN__(int argc, char *argv[]){
   if ( p != NULL )
 	loops = atoi(p);
 
+  if ((p = getenv("OPENBLAS_PARAM_N"))) {
+          n = atoi(p);
+	  has_param_n=1;	  
+  }
+
 
 #ifdef linux
   srandom(getpid());
@@ -174,7 +180,14 @@ int MAIN__(int argc, char *argv[]){
 
     timeg=0;
 
-    fprintf(stderr, " %6d : ", (int)m);
+    if ( has_param_n == 1 && n <= m )
+    	n=n;
+    else
+    	n=m;
+
+
+
+    fprintf(stderr, " %6dx%d : ", (int)m, (int)n);
 
     for (l=0; l<loops; l++)
     {
@@ -189,7 +202,7 @@ int MAIN__(int argc, char *argv[]){
 
     	gettimeofday( &start, (struct timezone *)0);
 
-    	GEMM (&trans, &trans, &m, &m, &m, alpha, a, &m, b, &m, beta, c, &m );
+    	GEMM (&trans, &trans, &m, &n, &m, alpha, a, &m, b, &m, beta, c, &m );
 
     	gettimeofday( &stop, (struct timezone *)0);
 
@@ -202,11 +215,11 @@ int MAIN__(int argc, char *argv[]){
     timeg /= loops;
     fprintf(stderr,
 	    " %10.2f MFlops\n",
-	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)m / timeg * 1.e-6);
+	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)n / timeg * 1.e-6);
 
   }
 
   return 0;
 }
 
-void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/gemm3m.c b/benchmark/gemm3m.c
@@ -118,7 +118,7 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *a, *b, *c;
   FLOAT alpha[] = {1.0, 1.0};
@@ -209,4 +209,4 @@ int MAIN__(int argc, char *argv[]){
   return 0;
 }
 
-void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/gemv.c b/benchmark/gemv.c
@@ -118,7 +118,7 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *a, *x, *y;
   FLOAT alpha[] = {1.0, 1.0};
@@ -266,4 +266,4 @@ int MAIN__(int argc, char *argv[]){
   return 0;
 }
 
-void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/ger.c b/benchmark/ger.c
@@ -108,7 +108,7 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *a, *x, *y;
   FLOAT alpha[] = {1.0, 1.0};
@@ -214,5 +214,5 @@ int MAIN__(int argc, char *argv[]){
   return 0;
 }
 
-void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));