Merge branch 'develop'

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]

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

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  0 > /proc/sys/vm/nr_hugepages		# need to be reset
-    $shell> echo 65 > /pros/sys/vm/nr_hugepages		# add 1 extra page
+    $shell> echo 65 > /proc/sys/vm/nr_hugepages		# add 1 extra page
-    $shell> echo 3355443200 > /pros/sys/kernel/shmmax   # just large number
+    $shell> echo 3355443200 > /proc/sys/kernel/shmmax   # just large number
-    $shell> echo 3355443200 > /pros/sys/kernel/shmall
+    $shell> echo 3355443200 > /proc/sys/kernel/shmall
 
     Also may add a few lines into /etc/security/limits.conf file.
 

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

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)
+	@echo Copying LAPACKE header files to $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)
-	@-install -pDm644 $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke.h $(DESTDIR)$(OPENBLAS_INCLUDE_DIR)/lapacke.h
+	@-install -pm644 $(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 -pm644 $(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 -pm644 $(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
+	@-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))

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
 #

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

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.

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

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
+#ACML=/opt/pb/acml-5-3-1-gfortran-64bit/gfortran64_fma4_mp/lib
-LIBACML = -fopenmp $(ACML)/libacml_mp.a -lgfortran -lm
+#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

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__")));

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__")));

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__")));

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__")));

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__")));

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__")));

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__")));

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__")));
 

benchmark/getri.c

@@ -137,7 +137,7 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *a,*work;
   FLOAT wkopt[4];
@@ -231,4 +231,4 @@ 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__")));

benchmark/hemm.c

@@ -107,7 +107,7 @@ 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};
@@ -189,4 +189,4 @@ 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__")));

benchmark/hemv.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};
@@ -205,4 +205,4 @@ 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__")));

benchmark/her2k.c

@@ -106,7 +106,7 @@ 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};
@@ -188,4 +188,4 @@ 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__")));

benchmark/herk.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, *c;
   FLOAT alpha[] = {1.0, 1.0};
@@ -186,4 +186,4 @@ 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__")));

benchmark/linpack.c

@@ -137,7 +137,7 @@ static void *huge_malloc(BLASLONG size){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
   FLOAT *a, *b;
   blasint *ipiv;
@@ -270,4 +270,4 @@ 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__")));

benchmark/potrf.c

@@ -114,7 +114,7 @@ int gettimeofday(struct timeval *tv, void *tz){
 
 #endif
 
-int MAIN__(int argc, char *argv[]){
+int main(int argc, char *argv[]){
 
 #ifndef COMPLEX
   char *trans[] = {"T", "N"};
@@ -278,5 +278,5 @@ 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__")));
 

benchmark/symm.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};
@@ -200,4 +200,4 @@ 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__")));

benchmark/symv.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};
@@ -215,4 +215,4 @@ 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__")));

benchmark/syr2k.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};
@@ -200,4 +200,4 @@ 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__")));

benchmark/syrk.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, *c;
   FLOAT alpha[] = {1.0, 1.0};
@@ -196,4 +196,4 @@ 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__")));

benchmark/trmm.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;
   FLOAT alpha[] = {1.0, 1.0};
@@ -199,4 +199,4 @@ 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__")));

benchmark/trsm.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;
   FLOAT alpha[] = {1.0, 1.0};
@@ -199,4 +199,4 @@ 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__")));

c_check

@@ -81,6 +81,10 @@ if (($architecture eq "mips32") || ($architecture eq "mips64")) {
     $defined = 1;
 }
 
+if (($architecture eq "arm") || ($architecture eq "arm64")) {
+    $defined = 1;
+}
+
 if ($architecture eq "alpha") {
     $defined = 1;
     $binary = 64;

cblas.h

@@ -13,6 +13,12 @@ extern "C" {
 void openblas_set_num_threads(int num_threads);
 void goto_set_num_threads(int num_threads);
 
+/*Get the number of threads on runtime.*/
+int openblas_get_num_threads(void);
+
+/*Get the number of physical processors (cores).*/
+int openblas_get_num_procs(void);
+
 /*Get the build configure on runtime.*/
 char* openblas_get_config(void);
 
@@ -341,6 +347,16 @@ void cblas_cimatcopy(OPENBLAS_CONST enum CBLAS_ORDER CORDER, OPENBLAS_CONST enum
 void cblas_zimatcopy(OPENBLAS_CONST enum CBLAS_ORDER CORDER, OPENBLAS_CONST enum CBLAS_TRANSPOSE CTRANS, OPENBLAS_CONST blasint crows, OPENBLAS_CONST blasint ccols, OPENBLAS_CONST double* calpha, double* a, 
 		     OPENBLAS_CONST blasint clda, OPENBLAS_CONST blasint cldb); 
 
+void cblas_sgeadd(OPENBLAS_CONST enum CBLAS_ORDER CORDER,OPENBLAS_CONST blasint crows, OPENBLAS_CONST blasint ccols, OPENBLAS_CONST float calpha, float *a, OPENBLAS_CONST blasint clda, OPENBLAS_CONST float cbeta, 
+		  float *c, OPENBLAS_CONST blasint cldc); 
+void cblas_dgeadd(OPENBLAS_CONST enum CBLAS_ORDER CORDER,OPENBLAS_CONST blasint crows, OPENBLAS_CONST blasint ccols, OPENBLAS_CONST double calpha, double *a, OPENBLAS_CONST blasint clda, OPENBLAS_CONST double cbeta, 
+		  double *c, OPENBLAS_CONST blasint cldc); 
+void cblas_cgeadd(OPENBLAS_CONST enum CBLAS_ORDER CORDER,OPENBLAS_CONST blasint crows, OPENBLAS_CONST blasint ccols, OPENBLAS_CONST float *calpha, float *a, OPENBLAS_CONST blasint clda, OPENBLAS_CONST float *cbeta, 
+		  float *c, OPENBLAS_CONST blasint cldc); 
+void cblas_zgeadd(OPENBLAS_CONST enum CBLAS_ORDER CORDER,OPENBLAS_CONST blasint crows, OPENBLAS_CONST blasint ccols, OPENBLAS_CONST double *calpha, double *a, OPENBLAS_CONST blasint clda, OPENBLAS_CONST double *cbeta, 
+		  double *c, OPENBLAS_CONST blasint cldc); 
+
+
 #ifdef __cplusplus
 }
 #endif  /* __cplusplus */

cblas_noconst.h

@@ -13,6 +13,12 @@ extern "C" {
 void openblas_set_num_threads(int num_threads);
 void goto_set_num_threads(int num_threads);
 
+/*Get the number of threads on runtime.*/
+int openblas_get_num_threads(void);
+
+/*Get the number of physical processors (cores).*/
+int openblas_get_num_procs(void);
+
 /*Get the build configure on runtime.*/
 char* openblas_get_config(void);
 
@@ -327,6 +333,16 @@ void cblas_cimatcopy( enum CBLAS_ORDER CORDER,  enum CBLAS_TRANSPOSE CTRANS,  bl
 		      blasint clda,  blasint cldb); 
 void cblas_zimatcopy( enum CBLAS_ORDER CORDER,  enum CBLAS_TRANSPOSE CTRANS,  blasint crows,  blasint ccols,  double* calpha, double* a, 
 		      blasint clda,  blasint cldb); 
+
+void cblas_sgeadd( enum CBLAS_ORDER CORDER, blasint crows,  blasint ccols,  float calpha, float *a,  blasint clda,  float cbeta, 
+		  float *c,  blasint cldc); 
+void cblas_dgeadd( enum CBLAS_ORDER CORDER, blasint crows,  blasint ccols,  double calpha, double *a,  blasint clda,  double cbeta, 
+		  double *c,  blasint cldc); 
+void cblas_cgeadd( enum CBLAS_ORDER CORDER, blasint crows,  blasint ccols,  float *calpha, float *a,  blasint clda,  float *cbeta, 
+		  float *c,  blasint cldc); 
+void cblas_zgeadd( enum CBLAS_ORDER CORDER, blasint crows,  blasint ccols,  double *calpha, double *a,  blasint clda,  double *cbeta, 
+		  double *c,  blasint cldc); 
+
 #ifdef __cplusplus
 }
 #endif  /* __cplusplus */

common.h

@@ -327,6 +327,14 @@ typedef int blasint;
 #endif
 #endif
 
+/*
+#ifdef STEAMROLLER
+#ifndef YIELDING
+#define YIELDING        __asm__ __volatile__ ("nop;nop;nop;nop;nop;nop;nop;nop;\n");
+#endif
+#endif
+*/
+
 #ifndef YIELDING
 #define YIELDING	sched_yield()
 #endif

common_c.h

@@ -220,6 +220,7 @@
 #define COMATCOPY_K_CTC         comatcopy_k_ctc
 #define COMATCOPY_K_RTC         comatcopy_k_rtc
 
+#define CGEADD_K                cgeadd_k 
 
 #else
 
@@ -402,6 +403,7 @@
 #define COMATCOPY_K_RNC         gotoblas -> comatcopy_k_rnc
 #define COMATCOPY_K_CTC         gotoblas -> comatcopy_k_ctc
 #define COMATCOPY_K_RTC         gotoblas -> comatcopy_k_rtc
+#define CGEADD_K                gotoblas -> cgeadd_k 
 
 #endif
 

common_d.h

@@ -149,6 +149,7 @@
 #define DOMATCOPY_K_RN		domatcopy_k_rn
 #define DOMATCOPY_K_CT		domatcopy_k_ct
 #define DOMATCOPY_K_RT		domatcopy_k_rt
+#define DGEADD_K                dgeadd_k 
 
 #else
 
@@ -267,6 +268,8 @@
 #define DOMATCOPY_K_CT		gotoblas -> domatcopy_k_ct
 #define DOMATCOPY_K_RT		gotoblas -> domatcopy_k_rt
 
+#define DGEADD_K                gotoblas -> dgeadd_k 
+
 #endif
 
 #define	DGEMM_NN		dgemm_nn

common_interface.h

@@ -754,6 +754,12 @@ void    BLASFUNC(dimatcopy) (char *, char *, blasint *, blasint *, double  *, do
 void    BLASFUNC(cimatcopy) (char *, char *, blasint *, blasint *, float  *, float  *, blasint *, blasint *);
 void    BLASFUNC(zimatcopy) (char *, char *, blasint *, blasint *, double  *, double  *, blasint *, blasint *);
 
+void    BLASFUNC(sgeadd) (blasint *, blasint *, float *, float *, blasint *, float *, float *, blasint*); 
+void    BLASFUNC(dgeadd) (blasint *, blasint *, double *, double *, blasint *, double *, double *, blasint*); 
+void    BLASFUNC(cgeadd) (blasint *, blasint *, float *, float *, blasint *, float *, float *, blasint*); 
+void    BLASFUNC(zgeadd) (blasint *, blasint *, double *, double *, blasint *, double *, double *, blasint*); 
+
+
 #ifdef __cplusplus
 }
 

common_level3.h

@@ -1762,6 +1762,11 @@ int zomatcopy_k_rnc(BLASLONG, BLASLONG,  double, double, double *, BLASLONG, dou
 int zomatcopy_k_ctc(BLASLONG, BLASLONG,  double, double, double *, BLASLONG, double  *, BLASLONG);
 int zomatcopy_k_rtc(BLASLONG, BLASLONG,  double, double, double *, BLASLONG, double  *, BLASLONG);
 
+int sgeadd_k(BLASLONG, BLASLONG, float, float*, BLASLONG, float, float *, BLASLONG); 
+int dgeadd_k(BLASLONG, BLASLONG, double, double*, BLASLONG, double, double *, BLASLONG); 
+int cgeadd_k(BLASLONG, BLASLONG, float, float, float*, BLASLONG, float, float, float *, BLASLONG); 
+int zgeadd_k(BLASLONG, BLASLONG, double,double, double*, BLASLONG, double, double, double *, BLASLONG); 
+
 
 #ifdef __CUDACC__
 }

common_macro.h

@@ -634,7 +634,7 @@
 #define OMATCOPY_K_RN		DOMATCOPY_K_RN
 #define OMATCOPY_K_CT		DOMATCOPY_K_CT
 #define OMATCOPY_K_RT		DOMATCOPY_K_RT
-
+#define GEADD_K                 DGEADD_K 
 #else
 
 #define	AMAX_K			SAMAX_K
@@ -932,6 +932,7 @@
 #define OMATCOPY_K_CT		SOMATCOPY_K_CT
 #define OMATCOPY_K_RT		SOMATCOPY_K_RT
 
+#define GEADD_K 		SGEADD_K 
 #endif
 #else
 #ifdef XDOUBLE
@@ -1746,6 +1747,7 @@
 #define OMATCOPY_K_RNC		ZOMATCOPY_K_RNC
 #define OMATCOPY_K_CTC		ZOMATCOPY_K_CTC
 #define OMATCOPY_K_RTC		ZOMATCOPY_K_RTC
+#define GEADD_K                 ZGEADD_K 
 
 #else
 
@@ -2159,6 +2161,8 @@
 #define OMATCOPY_K_CTC		COMATCOPY_K_CTC
 #define OMATCOPY_K_RTC		COMATCOPY_K_RTC
 
+#define GEADD_K                 CGEADD_K 
+
 #endif
 #endif
 

common_param.h

@@ -855,6 +855,10 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
   int    (*zomatcopy_k_rnc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
   int    (*zomatcopy_k_rtc)	(BLASLONG, BLASLONG, double, double, double*, BLASLONG, double*, BLASLONG);
 
+  int    (*sgeadd_k) (BLASLONG, BLASLONG, float, float *, BLASLONG, float, float *, BLASLONG); 
+  int    (*dgeadd_k) (BLASLONG, BLASLONG, double, double *, BLASLONG, double, double *, BLASLONG); 
+  int    (*cgeadd_k) (BLASLONG, BLASLONG, float, float,  float *,  BLASLONG, float, float, float *, BLASLONG); 
+  int    (*zgeadd_k) (BLASLONG, BLASLONG, float, double, double *, BLASLONG, double, double, double *, BLASLONG); 
 
 } gotoblas_t;
 

common_s.h

@@ -153,6 +153,7 @@
 #define SOMATCOPY_K_CT          somatcopy_k_ct
 #define SOMATCOPY_K_RT          somatcopy_k_rt
 
+#define SGEADD_K                sgeadd_k 
 
 #else
 
@@ -274,6 +275,7 @@
 #define SOMATCOPY_K_CT          gotoblas -> somatcopy_k_ct
 #define SOMATCOPY_K_RT          gotoblas -> somatcopy_k_rt
 
+#define SGEADD_K                gotoblas -> sgeadd_k 
 
 #endif
 

common_x86.h

@@ -171,7 +171,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #define MMXSTORE	movd
 #endif
 
-#if defined(PILEDRIVER) || defined(BULLDOZER)
+#if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER)
 //Enable some optimazation for barcelona.
 #define BARCELONA_OPTIMIZATION
 #endif

common_x86_64.h

@@ -226,7 +226,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 
 #ifdef ASSEMBLER
 
-#if defined(PILEDRIVER) || defined(BULLDOZER)
+#if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER)
 //Enable some optimazation for barcelona.
 #define BARCELONA_OPTIMIZATION
 #endif

common_z.h

@@ -220,6 +220,7 @@
 #define ZOMATCOPY_K_CTC         zomatcopy_k_ctc
 #define ZOMATCOPY_K_RTC         zomatcopy_k_rtc
 
+#define ZGEADD_K                zgeadd_k 
 
 #else
 
@@ -403,6 +404,8 @@
 #define ZOMATCOPY_K_CTC         gotoblas -> zomatcopy_k_ctc
 #define ZOMATCOPY_K_RTC         gotoblas -> zomatcopy_k_rtc
 
+#define ZGEADD_K                gotoblas -> zgeadd_k
+
 #endif
 
 #define	ZGEMM_NN		zgemm_nn

cpuid.h

@@ -108,6 +108,7 @@
 #define CORE_BULLDOZER   22
 #define CORE_PILEDRIVER  23
 #define CORE_HASWELL     24
+#define CORE_STEAMROLLER 25
 
 #define HAVE_SSE      (1 <<  0)
 #define HAVE_SSE2     (1 <<  1)
@@ -201,5 +202,6 @@ typedef struct {
 #define CPUTYPE_BULLDOZER               46
 #define CPUTYPE_PILEDRIVER              47
 #define CPUTYPE_HASWELL 		48
+#define CPUTYPE_STEAMROLLER 		49
 
 #endif

cpuid_arm.c

@@ -30,16 +30,27 @@
 #define CPU_UNKNOWN     	0
 #define CPU_ARMV6       	1
 #define CPU_ARMV7       	2
-#define CPU_CORTEXA15       	3
+#define CPU_CORTEXA9       	3
+#define CPU_CORTEXA15       	4
 
 static char *cpuname[] = {
   "UNKOWN",
   "ARMV6",
   "ARMV7",
+  "CORTEXA9",
   "CORTEXA15"
 };
 
 
+static char *cpuname_lower[] = {
+  "unknown",
+  "armv6",
+  "armv7",
+  "cortexa9",
+  "cortexa15"
+};
+
+
 int get_feature(char *search)
 {
 
@@ -85,6 +96,29 @@ int detect(void)
   	char buffer[512], *p;
   	p = (char *) NULL ;
 
+  	infile = fopen("/proc/cpuinfo", "r");
+	while (fgets(buffer, sizeof(buffer), infile))
+	{
+
+		if (!strncmp("CPU part", buffer, 8))
+		{
+			p = strchr(buffer, ':') + 2;
+			break;
+      		}
+  	}
+
+  	fclose(infile);
+  	if(p != NULL) {
+	  if (strstr(p, "0xc09")) {
+	    return CPU_CORTEXA9;
+	  }
+	  if (strstr(p, "0xc0f")) {
+	    return CPU_CORTEXA15;
+	  }
+
+	}
+
+  	p = (char *) NULL ;
   	infile = fopen("/proc/cpuinfo", "r");
 
 	while (fgets(buffer, sizeof(buffer), infile))
@@ -142,21 +176,7 @@ void get_architecture(void)
 void get_subarchitecture(void)
 {
 	int d = detect();
-	switch (d)
+	printf("%s", cpuname[d]);
-	{
-
-		case CPU_ARMV7:
-			printf("ARMV7");
-			break;
-
-		case CPU_ARMV6:
-			printf("ARMV6");
-			break;
-
-		default:
-			printf("UNKNOWN");
-			break;
-	}
 }
 
 void get_subdirname(void)
@@ -170,6 +190,36 @@ void get_cpuconfig(void)
 	int d = detect();
 	switch (d)
 	{
+	       case CPU_CORTEXA9:
+    			printf("#define CORTEXA9\n");
+    			printf("#define HAVE_VFP\n");
+    			printf("#define HAVE_VFPV3\n");
+			if ( get_feature("neon"))	printf("#define HAVE_NEON\n");
+			if ( get_feature("vfpv4"))	printf("#define HAVE_VFPV4\n");
+    			printf("#define L1_DATA_SIZE 32768\n");
+    			printf("#define L1_DATA_LINESIZE 32\n");
+    			printf("#define L2_SIZE 1048576\n");
+    			printf("#define L2_LINESIZE 32\n");
+    			printf("#define DTB_DEFAULT_ENTRIES 128\n");
+    			printf("#define DTB_SIZE 4096\n");
+    			printf("#define L2_ASSOCIATIVE 4\n");
+			break;
+
+	       case CPU_CORTEXA15:
+    			printf("#define CORTEXA15\n");
+    			printf("#define HAVE_VFP\n");
+    			printf("#define HAVE_VFPV3\n");
+			if ( get_feature("neon"))	printf("#define HAVE_NEON\n");
+			if ( get_feature("vfpv4"))	printf("#define HAVE_VFPV4\n");
+    			printf("#define L1_DATA_SIZE 32768\n");
+    			printf("#define L1_DATA_LINESIZE 32\n");
+    			printf("#define L2_SIZE 1048576\n");
+    			printf("#define L2_LINESIZE 32\n");
+    			printf("#define DTB_DEFAULT_ENTRIES 128\n");
+    			printf("#define DTB_SIZE 4096\n");
+    			printf("#define L2_ASSOCIATIVE 4\n");
+			break;
+
 
 		case CPU_ARMV7:
     			printf("#define ARMV7\n");
@@ -206,18 +256,7 @@ void get_libname(void)
 {
 
 	int d = detect();
-	switch (d)
+	printf("%s", cpuname_lower[d]);
-	{
-
-		case CPU_ARMV7:
-    			printf("armv7\n");
-			break;
-
-		case CPU_ARMV6:
-    			printf("armv6\n");
-			break;
-
-	}
 }
 
 

cpuid_x86.c

@@ -1162,6 +1162,12 @@ int get_cpuname(void){
 	    return CPUTYPE_PILEDRIVER;
 	  else
 	    return CPUTYPE_BARCELONA; //OS don't support AVX.
+	case 0:
+	  if(support_avx())
+	    return CPUTYPE_STEAMROLLER;
+	  else
+	    return CPUTYPE_BARCELONA; //OS don't support AVX.
+
 	}
 	break;
       case  5:
@@ -1290,6 +1296,7 @@ static char *cpuname[] = {
   "BULLDOZER",
   "PILEDRIVER",
   "HASWELL",
+  "STEAMROLLER",
 };
 
 static char *lowercpuname[] = {
@@ -1341,6 +1348,7 @@ static char *lowercpuname[] = {
   "bulldozer",
   "piledriver",
   "haswell",
+  "steamroller",
 };
 
 static char *corename[] = {
@@ -1369,6 +1377,7 @@ static char *corename[] = {
   "BULLDOZER",
   "PILEDRIVER",
   "HASWELL",
+  "STEAMROLLER",
 };
 
 static char *corename_lower[] = {
@@ -1397,6 +1406,7 @@ static char *corename_lower[] = {
   "bulldozer",
   "piledriver",
   "haswell",
+  "steamroller",
 };
 
 
@@ -1562,7 +1572,15 @@ int get_coretype(void){
 	    return CORE_PILEDRIVER;
 	  else
 	    return CORE_BARCELONA; //OS don't support AVX.
+	
+	case 0:
+	  if(support_avx())
+	    return CORE_STEAMROLLER;
+	  else
+	    return CORE_BARCELONA; //OS don't support AVX.
 	}
+
+
       }else return CORE_BARCELONA;
     }
   }

driver/others/Makefile

@@ -1,7 +1,7 @@
 TOPDIR	= ../..
 include ../../Makefile.system
 
-COMMONOBJS	 = memory.$(SUFFIX) xerbla.$(SUFFIX) c_abs.$(SUFFIX) z_abs.$(SUFFIX) openblas_set_num_threads.$(SUFFIX) openblas_get_config.$(SUFFIX) openblas_get_parallel.$(SUFFIX) openblas_error_handle.$(SUFFIX)
+COMMONOBJS	 = memory.$(SUFFIX) xerbla.$(SUFFIX) c_abs.$(SUFFIX) z_abs.$(SUFFIX) openblas_set_num_threads.$(SUFFIX) openblas_get_num_threads.$(SUFFIX) openblas_get_num_procs.$(SUFFIX) openblas_get_config.$(SUFFIX) openblas_get_parallel.$(SUFFIX) openblas_error_handle.$(SUFFIX)
 
 #COMMONOBJS	+= slamch.$(SUFFIX) slamc3.$(SUFFIX) dlamch.$(SUFFIX)  dlamc3.$(SUFFIX)
 
@@ -103,6 +103,12 @@ blas_server.$(SUFFIX) : $(BLAS_SERVER) ../../common.h ../../common_thread.h ../.
 openblas_set_num_threads.$(SUFFIX) : openblas_set_num_threads.c
 	$(CC) $(CFLAGS) -c $< -o $(@F)
 
+openblas_get_num_threads.$(SUFFIX) : openblas_get_num_threads.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
+openblas_get_num_procs.$(SUFFIX) : openblas_get_num_procs.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
 openblas_get_config.$(SUFFIX) : openblas_get_config.c
 	$(CC) $(CFLAGS) -c $< -o $(@F)
 

driver/others/dynamic.c

@@ -66,6 +66,7 @@ extern gotoblas_t  gotoblas_BOBCAT;
 extern gotoblas_t  gotoblas_SANDYBRIDGE;
 extern gotoblas_t  gotoblas_BULLDOZER;
 extern gotoblas_t  gotoblas_PILEDRIVER;
+extern gotoblas_t  gotoblas_STEAMROLLER;
 #ifdef NO_AVX2
 #define gotoblas_HASWELL gotoblas_SANDYBRIDGE
 #else
@@ -77,6 +78,7 @@ extern gotoblas_t  gotoblas_HASWELL;
 #define gotoblas_HASWELL gotoblas_NEHALEM
 #define gotoblas_BULLDOZER gotoblas_BARCELONA
 #define gotoblas_PILEDRIVER gotoblas_BARCELONA
+#define gotoblas_STEAMROLLER gotoblas_BARCELONA
 #endif
 
 
@@ -275,7 +277,17 @@ static gotoblas_t *get_coretype(void){
 	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
 	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
 	  }
+	}else if(model == 0){
+	  //AMD STEAMROLLER
+	  if(support_avx())
+	    return &gotoblas_STEAMROLLER;
+	  else{
+	    openblas_warning(FALLBACK_VERBOSE, BARCELONA_FALLBACK);
+	    return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
 	  }
+	}
+
+
       } else {
 	return &gotoblas_BARCELONA;
       }
@@ -315,6 +327,7 @@ static char *corename[] = {
     "Bulldozer",
     "Piledriver",
     "Haswell",
+    "Steamroller",
 };
 
 char *gotoblas_corename(void) {
@@ -339,6 +352,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_BULLDOZER)    return corename[18];
   if (gotoblas == &gotoblas_PILEDRIVER)   return corename[19];
   if (gotoblas == &gotoblas_HASWELL)      return corename[20];
+  if (gotoblas == &gotoblas_STEAMROLLER)  return corename[21];
 
   return corename[0];
 }
@@ -349,9 +363,9 @@ static gotoblas_t *force_coretype(char *coretype){
 	int i ;
 	int found = -1;
 	char message[128];
-	char mname[20];
+	//char mname[20];
 
-	for ( i=1 ; i <= 20; i++)
+	for ( i=1 ; i <= 21; i++)
 	{
 		if (!strncasecmp(coretype,corename[i],20))
 		{
@@ -361,8 +375,8 @@ static gotoblas_t *force_coretype(char *coretype){
 	}
 	if (found < 0)
 	{
-		strncpy(mname,coretype,20);
+	        //strncpy(mname,coretype,20);
-    		sprintf(message, "Core not found: %s\n",mname);
+	        snprintf(message, 128, "Core not found: %s\n",coretype);
     		openblas_warning(1, message);
 		return(NULL);
 	}
@@ -370,6 +384,7 @@ static gotoblas_t *force_coretype(char *coretype){
 	switch (found)
 	{
 
+		case 21: return (&gotoblas_STEAMROLLER);
 		case 20: return (&gotoblas_HASWELL);
 		case 19: return (&gotoblas_PILEDRIVER);
 		case 18: return (&gotoblas_BULLDOZER);

driver/others/memory.c

@@ -241,6 +241,7 @@ void set_stack_limit(int limitMB){
 */
 #endif
 
+
 /*
 OpenBLAS uses the numbers of CPU cores in multithreading.
 It can be set by openblas_set_num_threads(int num_threads);
@@ -323,6 +324,23 @@ int blas_get_cpu_number(void){
 }
 #endif
 
+
+int openblas_get_num_procs(void) {
+#ifndef SMP
+  return 1;
+#else
+  return get_num_procs();
+#endif
+}
+
+int openblas_get_num_threads(void) {
+#ifndef SMP
+  return 1;
+#else
+  return blas_get_cpu_number();
+#endif
+}
+
 struct release_t {
   void *address;
   void (*func)(struct release_t *);
@@ -1335,6 +1353,8 @@ void DESTRUCTOR gotoblas_quit(void) {
 
   if (gotoblas_initialized == 0) return;
 
+  blas_shutdown();
+
 #ifdef PROFILE
    moncontrol (0);
 #endif
@@ -1356,8 +1376,6 @@ void DESTRUCTOR gotoblas_quit(void) {
 #ifdef PROFILE
    moncontrol (1);
 #endif
-
-   blas_shutdown();
 }
 
 #if (defined(C_PGI) || (!defined(C_SUN) && defined(F_INTERFACE_SUN))) && (defined(ARCH_X86) || defined(ARCH_X86_64))

driver/others/openblas_get_num_procs.c

@@ -0,0 +1,40 @@
+/*****************************************************************************
+Copyright (c) 2011-2014, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of 
+      its contributors may be used to endorse or promote products 
+      derived from this software without specific prior written 
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+**********************************************************************************/
+
+#include "common.h"
+
+extern int openblas_get_num_procs(void);
+
+int openblas_get_num_procs_(void) {
+  return openblas_get_num_procs();
+}

driver/others/openblas_get_num_threads.c

@@ -0,0 +1,40 @@
+/*****************************************************************************
+Copyright (c) 2011-2014, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of 
+      its contributors may be used to endorse or promote products 
+      derived from this software without specific prior written 
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+**********************************************************************************/
+
+#include "common.h"
+
+extern int openblas_get_num_threads(void);
+
+int openblas_get_num_threads_(void) {
+  return openblas_get_num_threads();
+}

driver/others/parameter.c

@@ -166,7 +166,7 @@ int get_L2_size(void){
 #if defined(ATHLON) || defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT) || defined(BULLDOZER) || \
     defined(CORE_PRESCOTT) || defined(CORE_CORE2) || defined(PENRYN) || defined(DUNNINGTON) || \
     defined(CORE_NEHALEM) || defined(CORE_SANDYBRIDGE) || defined(ATOM) || defined(GENERIC) || \
-    defined(PILEDRIVER) || defined(HASWELL)
+    defined(PILEDRIVER) || defined(HASWELL) || defined(STEAMROLLER)
 
   cpuid(0x80000006, &eax, &ebx, &ecx, &edx);
 
@@ -251,7 +251,7 @@ void blas_set_parameter(void){
 
   env_var_t p;
   int factor;
-#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(SANDYBRIDGE) || defined(NEHALEM) || defined(HASWELL)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(SANDYBRIDGE) || defined(NEHALEM) || defined(HASWELL) || defined(STEAMROLLER)
   int size = 16;
 #else
   int size = get_L2_size();

exports/Makefile

@@ -100,7 +100,12 @@ else
 	$(OBJCONV) @objconv.def ../$(LIBNAME) ../$(LIBNAME).renamed
 $(LIBDYNNAME) : ../$(LIBNAME).renamed osx.def
 endif
+ifeq ($(NOFORTRAN), 2)
+#only build cblas without Fortran
+	$(CC) $(CFLAGS) -all_load -headerpad_max_install_names -install_name $(CURDIR)/../$(LIBDYNNAME) -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+else
 	$(FC) $(FFLAGS) -all_load -headerpad_max_install_names -install_name $(CURDIR)/../$(LIBDYNNAME) -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+endif
 
 dllinit.$(SUFFIX) : dllinit.c
 	$(CC) $(CFLAGS) -c -o $(@F) -s $<

exports/gensymbol

@@ -23,7 +23,8 @@
 	       zhpr,zrotg,zscal,zswap,zsymm,zsyr2k,zsyrk,ztbmv,
 	       ztbsv,ztpmv,ztpsv,ztrmm,ztrmv,ztrsm,ztrsv, zsymv,
 	       xerbla,
-	       saxpby,daxpby,caxpby,zaxpby
+	       saxpby,daxpby,caxpby,zaxpby, 
+	       sgeadd,dgeadd,cgeadd,zgeadd, 
 	       );
 
 @cblasobjs  = (
@@ -55,6 +56,7 @@
                cblas_saxpby,cblas_daxpby,cblas_caxpby,cblas_zaxpby,
 	       cblas_somatcopy, cblas_domatcopy, cblas_comatcopy, cblas_zomatcopy, 
 	       cblas_simatcopy, cblas_dimatcopy, cblas_cimatcopy, cblas_zimatcopy,
+	       cblas_sgeadd, cblas_dgeadd,cblas_cgeadd, cblas_zgeadd 
                );
 
 @exblasobjs = (
@@ -81,7 +83,10 @@
 
 #both underscore and no underscore
 @misc_common_objs = (
-                     openblas_set_num_threads, openblas_get_parallel,
+                     openblas_get_parallel,
+                     openblas_get_num_procs,
+                     openblas_set_num_threads,
+                     openblas_get_num_threads,
                     );
 
 @misc_no_underscore_objs = (

getarch.c

@@ -432,6 +432,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "PILEDRIVER"
 #endif
 
+#if defined (FORCE_STEAMROLLER)
+#define FORCE
+#define FORCE_INTEL
+#define ARCHITECTURE    "X86"
+#define SUBARCHITECTURE "STEAMROLLER"
+#define ARCHCONFIG   "-DSTEAMROLLER " \
+		     "-DL1_DATA_SIZE=16384 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=2097152 -DL2_LINESIZE=64  -DL3_SIZE=12582912 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 " \
+		     "-DHAVE_SSE4A -DHAVE_MISALIGNSSE -DHAVE_128BITFPU -DHAVE_FASTMOVU -DHAVE_CFLUSH " \
+                     "-DHAVE_AVX -DHAVE_FMA4 -DHAVE_FMA3"
+#define LIBNAME   "steamroller"
+#define CORENAME  "STEAMROLLER"
+#endif
+
+
 #ifdef FORCE_SSE_GENERIC
 #define FORCE
 #define FORCE_INTEL
@@ -710,6 +727,36 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_CORTEXA9
+#define FORCE
+#define ARCHITECTURE    "ARM"
+#define SUBARCHITECTURE "CORTEXA9"
+#define SUBDIRNAME      "arm"
+#define ARCHCONFIG   "-DCORTEXA9 " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=32 " \
+       "-DL2_SIZE=1048576 -DL2_LINESIZE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=128 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=4 " \
+       "-DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON"
+#define LIBNAME   "cortexa9"
+#define CORENAME  "CORTEXA9"
+#else
+#endif
+
+#ifdef FORCE_CORTEXA15
+#define FORCE
+#define ARCHITECTURE    "ARM"
+#define SUBARCHITECTURE "CORTEXA15"
+#define SUBDIRNAME      "arm"
+#define ARCHCONFIG   "-DCORTEXA15 " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=32 " \
+       "-DL2_SIZE=1048576 -DL2_LINESIZE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=128 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=4 " \
+       "-DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON"
+#define LIBNAME   "cortexa15"
+#define CORENAME  "CORTEXA15"
+#else
+#endif
+
 #ifdef FORCE_ARMV6
 #define FORCE
 #define ARCHITECTURE    "ARM"

interface/Makefile

@@ -43,7 +43,8 @@ SBLAS2OBJS    = \
 SBLAS3OBJS    = \
 		sgemm.$(SUFFIX) ssymm.$(SUFFIX) strmm.$(SUFFIX) \
 		strsm.$(SUFFIX) ssyrk.$(SUFFIX) ssyr2k.$(SUFFIX) \
-		somatcopy.$(SUFFIX) simatcopy.$(SUFFIX)
+		somatcopy.$(SUFFIX) simatcopy.$(SUFFIX)\
+		sgeadd.$(SUFFIX)
 
 
 DBLAS1OBJS    = \
@@ -68,7 +69,8 @@ DBLAS2OBJS    = \
 DBLAS3OBJS    = \
 		dgemm.$(SUFFIX) dsymm.$(SUFFIX) dtrmm.$(SUFFIX) \
 		dtrsm.$(SUFFIX) dsyrk.$(SUFFIX) dsyr2k.$(SUFFIX) \
-		domatcopy.$(SUFFIX) dimatcopy.$(SUFFIX)
+		domatcopy.$(SUFFIX) dimatcopy.$(SUFFIX)\
+		dgeadd.$(SUFFIX) 
 
 CBLAS1OBJS    = \
 		caxpy.$(SUFFIX) caxpyc.$(SUFFIX) cswap.$(SUFFIX) \
@@ -96,7 +98,8 @@ CBLAS3OBJS    = \
 		cgemm.$(SUFFIX) csymm.$(SUFFIX) ctrmm.$(SUFFIX) \
 		ctrsm.$(SUFFIX) csyrk.$(SUFFIX) csyr2k.$(SUFFIX) \
 	       	chemm.$(SUFFIX) cherk.$(SUFFIX) cher2k.$(SUFFIX) \
-		comatcopy.$(SUFFIX) cimatcopy.$(SUFFIX)
+		comatcopy.$(SUFFIX) cimatcopy.$(SUFFIX)\
+		cgeadd.$(SUFFIX) 
 
 ZBLAS1OBJS    = \
 		zaxpy.$(SUFFIX) zaxpyc.$(SUFFIX) zswap.$(SUFFIX) \
@@ -124,7 +127,8 @@ ZBLAS3OBJS    = \
 		zgemm.$(SUFFIX) zsymm.$(SUFFIX) ztrmm.$(SUFFIX) \
 		ztrsm.$(SUFFIX) zsyrk.$(SUFFIX) zsyr2k.$(SUFFIX) \
 	       	zhemm.$(SUFFIX) zherk.$(SUFFIX) zher2k.$(SUFFIX) \
-		zomatcopy.$(SUFFIX) zimatcopy.$(SUFFIX)
+		zomatcopy.$(SUFFIX) zimatcopy.$(SUFFIX)\
+		zgeadd.$(SUFFIX) 
 
 ifeq ($(SUPPORT_GEMM3M), 1)
 
@@ -269,7 +273,8 @@ CSBLAS2OBJS   = \
 
 CSBLAS3OBJS   = \
 	cblas_sgemm.$(SUFFIX) cblas_ssymm.$(SUFFIX) cblas_strmm.$(SUFFIX) cblas_strsm.$(SUFFIX) \
-	cblas_ssyrk.$(SUFFIX) cblas_ssyr2k.$(SUFFIX) cblas_somatcopy.$(SUFFIX)  cblas_simatcopy.$(SUFFIX) 
+	cblas_ssyrk.$(SUFFIX) cblas_ssyr2k.$(SUFFIX) cblas_somatcopy.$(SUFFIX)  cblas_simatcopy.$(SUFFIX)\
+	cblas_sgeadd.$(SUFFIX)
 
 CDBLAS1OBJS   = \
 	cblas_idamax.$(SUFFIX) cblas_dasum.$(SUFFIX) cblas_daxpy.$(SUFFIX) \
@@ -285,7 +290,8 @@ CDBLAS2OBJS   = \
 
 CDBLAS3OBJS   += \
 	cblas_dgemm.$(SUFFIX) cblas_dsymm.$(SUFFIX) cblas_dtrmm.$(SUFFIX) cblas_dtrsm.$(SUFFIX) \
-	cblas_dsyrk.$(SUFFIX) cblas_dsyr2k.$(SUFFIX) cblas_domatcopy.$(SUFFIX)  cblas_dimatcopy.$(SUFFIX) 
+	cblas_dsyrk.$(SUFFIX) cblas_dsyr2k.$(SUFFIX) cblas_domatcopy.$(SUFFIX)  cblas_dimatcopy.$(SUFFIX) \
+        cblas_dgeadd.$(SUFFIX) 
 
 CCBLAS1OBJS   = \
 	cblas_icamax.$(SUFFIX) cblas_scasum.$(SUFFIX)  cblas_caxpy.$(SUFFIX) \
@@ -308,7 +314,9 @@ CCBLAS3OBJS   = \
 	cblas_cgemm.$(SUFFIX) cblas_csymm.$(SUFFIX) cblas_ctrmm.$(SUFFIX) cblas_ctrsm.$(SUFFIX) \
 	cblas_csyrk.$(SUFFIX) cblas_csyr2k.$(SUFFIX) \
 	cblas_chemm.$(SUFFIX) cblas_cherk.$(SUFFIX) cblas_cher2k.$(SUFFIX) \
-	cblas_comatcopy.$(SUFFIX) cblas_cimatcopy.$(SUFFIX) 
+	cblas_comatcopy.$(SUFFIX) cblas_cimatcopy.$(SUFFIX)\
+	cblas_cgeadd.$(SUFFIX)
+
 
 
 CZBLAS1OBJS   = \
@@ -332,7 +340,9 @@ CZBLAS3OBJS   = \
 	cblas_zgemm.$(SUFFIX) cblas_zsymm.$(SUFFIX) cblas_ztrmm.$(SUFFIX) cblas_ztrsm.$(SUFFIX) \
 	cblas_zsyrk.$(SUFFIX) cblas_zsyr2k.$(SUFFIX) \
 	cblas_zhemm.$(SUFFIX) cblas_zherk.$(SUFFIX) cblas_zher2k.$(SUFFIX)\
-	cblas_zomatcopy.$(SUFFIX) cblas_zimatcopy.$(SUFFIX) 
+	cblas_zomatcopy.$(SUFFIX) cblas_zimatcopy.$(SUFFIX) \
+	cblas_zgeadd.$(SUFFIX)
+
 
 ifeq ($(SUPPORT_GEMM3M), 1)
 
@@ -2103,4 +2113,27 @@ zimatcopy.$(SUFFIX) zimatcopy.$(PSUFFIX) : zimatcopy.c
 cblas_zimatcopy.$(SUFFIX) cblas_zimatcopy.$(PSUFFIX) : zimatcopy.c
 	$(CC) -c $(CFLAGS) -DCBLAS $< -o $(@F)
 
+sgeadd.$(SUFFIX) sgeadd.$(PSUFFIX) : geadd.c
+	$(CC) -c $(CFLAGS) $< -o $(@F)
+
+dgeadd.$(SUFFIX) dgeadd.$(PSUFFIX) : geadd.c
+	$(CC) -c $(CFLAGS) $< -o $(@F)
+
+cgeadd.$(SUFFIX) cgeadd.$(PSUFFIX) : zgeadd.c
+	$(CC) -c $(CFLAGS) $< -o $(@F)
+
+zgeadd.$(SUFFIX) zgeadd.$(PSUFFIX) : zgeadd.c
+	$(CC) -c $(CFLAGS) $< -o $(@F)
+
+cblas_sgeadd.$(SUFFIX) cblas_sgeadd.$(PSUFFIX) : geadd.c
+	$(CC) -c $(CFLAGS) -DCBLAS $< -o $(@F)
+
+cblas_dgeadd.$(SUFFIX) cblas_dgeadd.$(PSUFFIX) : geadd.c
+	$(CC) -c $(CFLAGS) -DCBLAS $< -o $(@F)
+
+cblas_cgeadd.$(SUFFIX) cblas_cgeadd.$(PSUFFIX) : zgeadd.c
+	$(CC) -c $(CFLAGS) -DCBLAS $< -o $(@F)
+
+cblas_zgeadd.$(SUFFIX) cblas_zgeadd.$(PSUFFIX) : zgeadd.c
+	$(CC) -c $(CFLAGS) -DCBLAS $< -o $(@F)
 

interface/geadd.c

@@ -0,0 +1,148 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#include <stdio.h>
+#include "common.h"
+#ifdef FUNCTION_PROFILE
+#include "functable.h"
+#endif
+
+#if defined(DOUBLE)
+#define ERROR_NAME "DGEADD "
+#else
+#define ERROR_NAME "SGEADD "
+#endif
+
+#ifndef CBLAS
+
+void NAME(blasint *M, blasint *N, FLOAT *ALPHA, FLOAT *a, blasint *LDA,
+		                  FLOAT *BETA,  FLOAT *c, blasint *LDC)
+{
+
+  blasint m = *M;
+  blasint n = *N;
+  blasint lda = *LDA;
+  blasint ldc = *LDC; 
+  FLOAT alpha = *ALPHA;
+  FLOAT beta  = *BETA;
+
+  blasint info;
+
+  PRINT_DEBUG_NAME;
+
+  info = 0;
+
+
+  if (lda < MAX(1, m))	info = 6;
+  if (ldc < MAX(1, m))	info = 8;
+
+  if (n < 0)		info = 2;
+  if (m < 0)		info = 1;
+
+  if (info != 0){
+    BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
+    return;
+  }
+
+#else
+void CNAME( enum CBLAS_ORDER order, blasint m,  blasint n,  FLOAT alpha, FLOAT *a,  blasint lda, FLOAT beta, 
+		  FLOAT *c,  blasint ldc)
+{
+/* 
+void CNAME(enum CBLAS_ORDER order,
+	   blasint m, blasint n,
+	   FLOAT alpha,
+	   FLOAT  *a, blasint lda,
+	   FLOAT beta,
+	   FLOAT  *c, blasint ldc){ */
+
+  blasint info, t;
+
+  PRINT_DEBUG_CNAME;
+
+  info  =  0;
+
+  if (order == CblasColMajor) {
+
+    info = -1;
+
+    if (ldc < MAX(1, m))  info = 8;
+    if (lda < MAX(1, m))  info = 5;
+    if (n < 0)		  info = 2;
+    if (m < 0)		  info = 1;
+
+  }
+
+  if (order == CblasRowMajor) {
+    info = -1;
+
+    t = n;
+    n = m;
+    m = t;
+
+    if (ldc < MAX(1, m))  info = 8;
+    if (lda < MAX(1, m))  info = 5;
+    if (n < 0)		  info = 2;
+    if (m < 0)		  info = 1;
+  }
+
+  if (info >= 0) {
+    BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
+    return;
+  }
+
+#endif
+
+  if ((m==0) || (n==0)) return;
+
+
+  IDEBUG_START;
+
+  FUNCTION_PROFILE_START();
+
+
+  GEADD_K(m,n,alpha, a, lda, beta, c, ldc); 
+
+
+  FUNCTION_PROFILE_END(1, 2* m * n ,  2 * m * n);
+
+  IDEBUG_END;
+
+  return;
+
+}

interface/gemv.c

@@ -208,7 +208,20 @@ void CNAME(enum CBLAS_ORDER order,
   if (incx < 0) x -= (lenx - 1) * incx;
   if (incy < 0) y -= (leny - 1) * incy;
 
+#ifdef MAX_STACK_ALLOC
+  // make it volatile because some gemv implementation (ex: dgemv_n.S)
+  // do not restore all register
+  volatile int stack_alloc_size = m + n;
+  if(stack_alloc_size < 128)
+      //dgemv_n.S require a 128 bytes buffer
+      stack_alloc_size = 128;
+  if(stack_alloc_size > MAX_STACK_ALLOC / sizeof(FLOAT))
+      stack_alloc_size = 0;
+  FLOAT stack_buffer[stack_alloc_size];
+  buffer = stack_alloc_size ? stack_buffer : (FLOAT *)blas_memory_alloc(1);
+#else
   buffer = (FLOAT *)blas_memory_alloc(1);
+#endif
 
 #ifdef SMP
 
@@ -237,6 +250,9 @@ void CNAME(enum CBLAS_ORDER order,
   }
 #endif
 
+#ifdef MAX_STACK_ALLOC
+  if(!stack_alloc_size)
+#endif
     blas_memory_free(buffer);
 
   FUNCTION_PROFILE_END(1, m * n + m + n,  2 * m * n);

interface/ger.c

@@ -171,7 +171,15 @@ void CNAME(enum CBLAS_ORDER order,
   if (incy < 0) y -= (n - 1) * incy;
   if (incx < 0) x -= (m - 1) * incx;
 
+#ifdef MAX_STACK_ALLOC
+  volatile int stack_alloc_size = m;
+  if(stack_alloc_size > MAX_STACK_ALLOC / sizeof(FLOAT))
+      stack_alloc_size = 0;
+  FLOAT stack_buffer[stack_alloc_size];
+  buffer = stack_alloc_size ? stack_buffer : (FLOAT *)blas_memory_alloc(1);
+#else
   buffer = (FLOAT *)blas_memory_alloc(1);
+#endif
 
 #ifdef SMPTEST
   nthreads = num_cpu_avail(2);
@@ -190,6 +198,9 @@ void CNAME(enum CBLAS_ORDER order,
   }
 #endif
 
+#ifdef MAX_STACK_ALLOC
+  if(!stack_alloc_size)
+#endif
     blas_memory_free(buffer);
 
   FUNCTION_PROFILE_END(1, m * n + m + n, 2 * m * n);

interface/zgeadd.c

@@ -0,0 +1,146 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#include <stdio.h>
+#include "common.h"
+#ifdef FUNCTION_PROFILE
+#include "functable.h"
+#endif
+
+#if defined(DOUBLE)
+#define ERROR_NAME "ZGEADD "
+#else
+#define ERROR_NAME "CGEADD "
+#endif
+
+#ifndef CBLAS
+
+void NAME(blasint *M, blasint *N, FLOAT *ALPHA, FLOAT *a, blasint *LDA,
+		                  FLOAT *BETA,  FLOAT *c, blasint *LDC)
+{
+
+  blasint m = *M;
+  blasint n = *N;
+  blasint lda = *LDA;
+  blasint ldc = *LDC; 
+
+  blasint info;
+
+  PRINT_DEBUG_NAME;
+
+  info = 0;
+
+
+  if (lda < MAX(1, m))	info = 6;
+  if (ldc < MAX(1, m))	info = 8;
+
+  if (n < 0)		info = 2;
+  if (m < 0)		info = 1;
+
+  if (info != 0){
+    BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
+    return;
+  }
+
+#else
+void CNAME( enum CBLAS_ORDER order, blasint m,  blasint n,  FLOAT *ALPHA, FLOAT *a,  blasint lda, FLOAT *BETA, 
+		  FLOAT *c,  blasint ldc)
+{
+/* 
+void CNAME(enum CBLAS_ORDER order,
+	   blasint m, blasint n,
+	   FLOAT alpha,
+	   FLOAT  *a, blasint lda,
+	   FLOAT beta,
+	   FLOAT  *c, blasint ldc){ */
+
+  blasint info, t;
+
+  PRINT_DEBUG_CNAME;
+
+  info  =  0;
+
+  if (order == CblasColMajor) {
+
+    info = -1;
+
+    if (ldc < MAX(1, m))  info = 8;
+    if (lda < MAX(1, m))  info = 5;
+    if (n < 0)		  info = 2;
+    if (m < 0)		  info = 1;
+
+  }
+
+  if (order == CblasRowMajor) {
+    info = -1;
+
+    t = n;
+    n = m;
+    m = t;
+
+    if (ldc < MAX(1, m))  info = 8;
+    if (lda < MAX(1, m))  info = 5;
+    if (n < 0)		  info = 2;
+    if (m < 0)		  info = 1;
+  }
+
+  if (info >= 0) {
+    BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
+    return;
+  }
+
+#endif
+
+  if ((m==0) || (n==0)) return;
+
+
+  IDEBUG_START;
+
+  FUNCTION_PROFILE_START();
+
+
+  GEADD_K(m,n,ALPHA[0],ALPHA[1], a, lda, BETA[0], BETA[1], c, ldc); 
+
+
+  FUNCTION_PROFILE_END(1, 2* m * n ,  2 * m * n);
+
+  IDEBUG_END;
+
+  return;
+
+}

interface/zsyr.c

@@ -173,7 +173,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, int n, FLOAT alpha, FLO
 
   FUNCTION_PROFILE_START();
 
-  if (incx < 0 ) x -= (n - 1) * incx;
+  if (incx < 0 ) x -= (n - 1) * incx * 2;
 
   buffer = (FLOAT *)blas_memory_alloc(1);
 

kernel/Makefile.L3

@@ -329,23 +329,27 @@ endif
 ######  BLAS extensions #####
 SBLASOBJS += \
 	somatcopy_k_cn$(TSUFFIX).$(SUFFIX) somatcopy_k_rn$(TSUFFIX).$(SUFFIX) \
-	somatcopy_k_ct$(TSUFFIX).$(SUFFIX) somatcopy_k_rt$(TSUFFIX).$(SUFFIX)
+	somatcopy_k_ct$(TSUFFIX).$(SUFFIX) somatcopy_k_rt$(TSUFFIX).$(SUFFIX) \
+	sgeadd_k$(TSUFFIX).$(SUFFIX) 
 
 DBLASOBJS += \
 	domatcopy_k_cn$(TSUFFIX).$(SUFFIX) domatcopy_k_rn$(TSUFFIX).$(SUFFIX) \
-	domatcopy_k_ct$(TSUFFIX).$(SUFFIX) domatcopy_k_rt$(TSUFFIX).$(SUFFIX)
+	domatcopy_k_ct$(TSUFFIX).$(SUFFIX) domatcopy_k_rt$(TSUFFIX).$(SUFFIX) \
+	dgeadd_k$(TSUFFIX).$(SUFFIX) 
 
 CBLASOBJS += \
 	comatcopy_k_cn$(TSUFFIX).$(SUFFIX) comatcopy_k_rn$(TSUFFIX).$(SUFFIX) \
 	comatcopy_k_ct$(TSUFFIX).$(SUFFIX) comatcopy_k_rt$(TSUFFIX).$(SUFFIX) \
 	comatcopy_k_cnc$(TSUFFIX).$(SUFFIX) comatcopy_k_rnc$(TSUFFIX).$(SUFFIX) \
-	comatcopy_k_ctc$(TSUFFIX).$(SUFFIX) comatcopy_k_rtc$(TSUFFIX).$(SUFFIX)
+	comatcopy_k_ctc$(TSUFFIX).$(SUFFIX) comatcopy_k_rtc$(TSUFFIX).$(SUFFIX) \
+	cgeadd_k$(TSUFFIX).$(SUFFIX) 
 
 ZBLASOBJS += \
 	zomatcopy_k_cn$(TSUFFIX).$(SUFFIX) zomatcopy_k_rn$(TSUFFIX).$(SUFFIX) \
 	zomatcopy_k_ct$(TSUFFIX).$(SUFFIX) zomatcopy_k_rt$(TSUFFIX).$(SUFFIX) \
 	zomatcopy_k_cnc$(TSUFFIX).$(SUFFIX) zomatcopy_k_rnc$(TSUFFIX).$(SUFFIX) \
-	zomatcopy_k_ctc$(TSUFFIX).$(SUFFIX) zomatcopy_k_rtc$(TSUFFIX).$(SUFFIX)
+	zomatcopy_k_ctc$(TSUFFIX).$(SUFFIX) zomatcopy_k_rtc$(TSUFFIX).$(SUFFIX) \
+	zgeadd_k$(TSUFFIX).$(SUFFIX) 
 
 
 SGEMMINCOPYOBJ_P = $(SGEMMINCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
@@ -3440,3 +3444,31 @@ $(KDIR)zomatcopy_k_rtc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_RTC)
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 
+ifndef SGEADD_K
+SGEADD_K = ../generic/geadd.c
+endif
+
+$(KDIR)sgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEADD_K)
+	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
+
+ifndef DGEADD_K
+DGEADD_K = ../generic/geadd.c
+endif
+
+$(KDIR)dgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEADD_K)
+	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
+
+ifndef CGEADD_K
+CGEADD_K = ../generic/zgeadd.c
+endif
+
+$(KDIR)cgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEADD_K)
+	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM $< -o $@
+
+ifndef ZGEADD_K
+ZGEADD_K = ../generic/zgeadd.c
+endif
+
+$(KDIR)zgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEADD_K)
+	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM $< -o $@
+

kernel/arm/KERNEL.CORTEXA15

@@ -0,0 +1 @@
+include $(KERNELDIR)/KERNEL.ARMV7

kernel/arm/KERNEL.CORTEXA9

@@ -0,0 +1 @@
+include $(KERNELDIR)/KERNEL.ARMV7

kernel/generic/geadd.c

@@ -0,0 +1,64 @@
+/***************************************************************************
+Copyright (c) 2013, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+
+int CNAME(BLASLONG rows, BLASLONG cols, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT beta, FLOAT *b, BLASLONG ldb)
+{
+	BLASLONG i;
+	FLOAT *aptr,*bptr;
+
+	if ( rows <= 0     )  return(0);
+	if ( cols <= 0     )  return(0);
+
+	
+	aptr = a;
+	bptr = b;
+
+	if ( alpha == 0.0 )
+	{
+		for ( i=0; i<cols ; i++ )
+		{
+			SCAL_K(rows, 0,0, beta, bptr, 1,  NULL, 0,NULL,0); 
+			bptr+=ldb; 
+		}
+
+		return(0);
+	}
+
+	for (i = 0; i < cols; i++) {
+		AXPBY_K(rows, alpha, aptr, 1, beta, bptr, 1); 
+		aptr += lda; 
+		bptr += ldb; 
+	}
+
+	return(0);
+
+}
+
+

kernel/generic/zgeadd.c

@@ -0,0 +1,65 @@
+/***************************************************************************
+Copyright (c) 2013, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+
+
+int CNAME(BLASLONG rows, BLASLONG cols, FLOAT alphar, FLOAT alphai, FLOAT *a, BLASLONG lda, FLOAT betar, FLOAT betai , FLOAT *b, BLASLONG ldb)
+{
+	BLASLONG i;
+	FLOAT *aptr,*bptr;
+
+	if ( rows <= 0     )  return(0);
+	if ( cols <= 0     )  return(0);
+
+	
+	aptr = a;
+	bptr = b;
+	lda *= 2; 
+	ldb *= 2; 
+
+	if ( alphar == 0.0 && alphai == 0.0 )
+	{
+		for ( i=0; i<cols ; i++ )
+		{
+			SCAL_K(rows, 0,0, betar, betai, bptr, 1,  NULL, 0,NULL,0); 
+			bptr+=ldb; 
+		}
+
+		return(0);
+	}
+
+	for (i = 0; i < cols; i++) {
+		AXPBY_K(rows, alphar, alphai, aptr, 1, betar, betai,  bptr, 1); 
+		aptr += lda; 
+		bptr += ldb; 
+	}
+	return(0);
+
+}
+
+

kernel/setparam-ref.c

@@ -548,8 +548,9 @@ gotoblas_t TABLE_NAME = {
   comatcopy_k_cnTS, comatcopy_k_ctTS, comatcopy_k_rnTS, comatcopy_k_rtTS,
   comatcopy_k_cncTS, comatcopy_k_ctcTS, comatcopy_k_rncTS, comatcopy_k_rtcTS,
   zomatcopy_k_cnTS, zomatcopy_k_ctTS, zomatcopy_k_rnTS, zomatcopy_k_rtTS,
-  zomatcopy_k_cncTS, zomatcopy_k_ctcTS, zomatcopy_k_rncTS, zomatcopy_k_rtcTS
+  zomatcopy_k_cncTS, zomatcopy_k_ctcTS, zomatcopy_k_rncTS, zomatcopy_k_rtcTS,
 
+  sgeadd_kTS, dgeadd_kTS, cgeadd_kTS, zgeadd_kTS
 
 };
 
@@ -941,6 +942,23 @@ static void init_parameter(void) {
 #endif
 #endif
 
+#ifdef STEAMROLLER
+
+#ifdef DEBUG
+  fprintf(stderr, "Steamroller\n");
+#endif
+
+  TABLE_NAME.sgemm_p = SGEMM_DEFAULT_P;
+  TABLE_NAME.dgemm_p = DGEMM_DEFAULT_P;
+  TABLE_NAME.cgemm_p = CGEMM_DEFAULT_P;
+  TABLE_NAME.zgemm_p = ZGEMM_DEFAULT_P;
+#ifdef EXPRECISION
+  TABLE_NAME.qgemm_p = QGEMM_DEFAULT_P;
+  TABLE_NAME.xgemm_p = XGEMM_DEFAULT_P;
+#endif
+#endif
+
+
 #ifdef NANO
 
 #ifdef DEBUG

kernel/x86_64/KERNEL.SANDYBRIDGE

@@ -34,17 +34,17 @@ CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
 CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
 CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-
+ZGEMMKERNEL    =  zgemm_kernel_1x4_nehalem.S
-ZGEMMKERNEL    =  zgemm_kernel_4x4_sandy.S
+ZGEMMINCOPY    =  zgemm_ncopy_1.S
-ZGEMMINCOPY    =
+ZGEMMITCOPY    =  zgemm_tcopy_1.S
-ZGEMMITCOPY    =
 ZGEMMONCOPY    =  ../generic/zgemm_ncopy_4.c
 ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_4.c
-ZGEMMINCOPYOBJ =
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
-ZGEMMITCOPYOBJ =
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
 ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
 ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
+
 #STRSMKERNEL_LN	=  trsm_kernel_LN_4x8_nehalem.S
 #STRSMKERNEL_LT	=  trsm_kernel_LT_4x8_nehalem.S
 #STRSMKERNEL_RN	=  trsm_kernel_LT_4x8_nehalem.S

kernel/x86_64/KERNEL.STEAMROLLER

@@ -0,0 +1,88 @@
+DAXPYKERNEL = daxpy.c
+CAXPYKERNEL = caxpy.c
+ZAXPYKERNEL = zaxpy.c
+
+SDOTKERNEL     = sdot.c
+DDOTKERNEL     = ddot.c
+
+DSYMV_U_KERNEL = dsymv_U.c
+DSYMV_L_KERNEL = dsymv_L.c
+SSYMV_U_KERNEL = ssymv_U.c
+SSYMV_L_KERNEL = ssymv_L.c
+
+SGEMVNKERNEL = sgemv_n_4.c
+SGEMVTKERNEL = sgemv_t_4.c
+
+DGEMVNKERNEL = dgemv_n_4.c
+DGEMVTKERNEL = dgemv_t_4.c
+
+ZGEMVNKERNEL = zgemv_n_dup.S
+ZGEMVTKERNEL = zgemv_t_4.c
+
+DCOPYKERNEL  = dcopy_bulldozer.S
+
+
+SGEMMKERNEL    =  sgemm_kernel_16x2_piledriver.S
+SGEMMINCOPY    =  ../generic/gemm_ncopy_16.c
+SGEMMITCOPY    =  ../generic/gemm_tcopy_16.c
+SGEMMONCOPY    =  gemm_ncopy_2_bulldozer.S
+SGEMMOTCOPY    =  gemm_tcopy_2_bulldozer.S
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+DGEMMKERNEL    =  dgemm_kernel_8x2_piledriver.S
+DGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_8.c
+DGEMMONCOPY    =  gemm_ncopy_2_bulldozer.S
+DGEMMOTCOPY    =  gemm_tcopy_2_bulldozer.S
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CGEMMKERNEL    =  cgemm_kernel_4x2_piledriver.S
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_4.c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_4.c
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMKERNEL    =  zgemm_kernel_2x2_piledriver.S
+ZGEMMINCOPY    =
+ZGEMMITCOPY    =
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
+ZGEMMINCOPYOBJ =
+ZGEMMITCOPYOBJ =
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CGEMM3MKERNEL    =  zgemm3m_kernel_8x4_barcelona.S
+ZGEMM3MKERNEL    =  zgemm3m_kernel_4x4_barcelona.S
+
+STRSMKERNEL_LN  =  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT  =  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN  =  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT  =  ../generic/trsm_kernel_RT.c
+
+
+DTRSMKERNEL_LN  = ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT  = dtrsm_kernel_LT_8x2_bulldozer.S
+DTRSMKERNEL_RN  = dtrsm_kernel_RN_8x2_bulldozer.S
+DTRSMKERNEL_RT  = ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN  = ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT  = ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN  = ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT  = ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN  = ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT  = ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN  = ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT  = ../generic/trsm_kernel_RT.c
+
+

kernel/x86_64/caxpy.c

@@ -29,7 +29,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(BULLDOZER)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER)
 #include "caxpy_microk_bulldozer-2.c"
 #endif
 

kernel/x86_64/caxpy_microk_bulldozer-2.c

@@ -40,7 +40,7 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vbroadcastss	       4(%4), %%xmm1		    \n\t"  // imag part of alpha
 
 	".align 16				            \n\t"
-	".L01LOOP%=:				            \n\t"
+	"1:				            \n\t"
 
 	"prefetcht0  768(%2,%0,4)                           \n\t"
 	"vmovups        (%2,%0,4), %%xmm5                   \n\t" // 2 complex values from x
@@ -113,7 +113,7 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 
 	"addq		$16, %0	  	 	             \n\t"
 	"subq	        $8 , %1			             \n\t"		
-	"jnz		.L01LOOP%=		             \n\t"
+	"jnz		1b		             \n\t"
 
 	:
         : 

kernel/x86_64/cgemv_n_microk_haswell-4.c

@@ -49,10 +49,10 @@ static void cgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
 	"vbroadcastss	28(%2), %%ymm7                  \n\t"  // imag part x3
 
 	"cmpq		$0 , %1				\n\t"
-	"je		.L01END%=		        \n\t"
+	"je		2f			        \n\t"
 
 	".align 16				        \n\t"
-	".L01LOOP%=:				        \n\t"
+	"1:				        \n\t"
 	"prefetcht0      320(%4,%0,4)			\n\t"
 	"vmovups	(%4,%0,4), %%ymm8	        \n\t" // 4 complex values form a0
 	"vmovups      32(%4,%0,4), %%ymm9	        \n\t" // 4 complex values form a0
@@ -115,12 +115,12 @@ static void cgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
 
         "addq		$16, %0	  	 	        \n\t"
 	"subq	        $8 , %1			        \n\t"		
-	"jnz		.L01LOOP%=		        \n\t"
+	"jnz		1b		        \n\t"
 
-	".L01END%=:				        \n\t"
+	"2:				        \n\t"
 
 	"cmpq		$4, %8				\n\t"
-	"jne		.L02END%=			\n\t"
+	"jne		3f				\n\t"
 
 	"vmovups	(%4,%0,4), %%ymm8	        \n\t" // 4 complex values form a0
 	"vmovups	(%5,%0,4), %%ymm10              \n\t" // 4 complex values form a1
@@ -155,7 +155,7 @@ static void cgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
 
 	"vmovups  %%ymm12,   (%3,%0,4)		        \n\t" // 4 complex values to y	
 
-	".L02END%=:				        \n\t"
+	"3:				        \n\t"
 	"vzeroupper			 \n\t"
 
 	:
@@ -200,10 +200,10 @@ static void cgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
 	"vbroadcastss	12(%2), %%ymm3                  \n\t"  // imag part x1
 
 	"cmpq		$0 , %1				\n\t"
-	"je		.L01END%=		        \n\t"
+	"je		2f			        \n\t"
 
 	".align 16				        \n\t"
-	".L01LOOP%=:				        \n\t"
+	"1:				        \n\t"
 	"prefetcht0      320(%4,%0,4)			\n\t"
 	"vmovups	(%4,%0,4), %%ymm8	        \n\t" // 4 complex values form a0
 	"vmovups      32(%4,%0,4), %%ymm9	        \n\t" // 4 complex values form a0
@@ -248,12 +248,12 @@ static void cgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
 
         "addq		$16, %0	  	 	        \n\t"
 	"subq	        $8 , %1			        \n\t"		
-	"jnz		.L01LOOP%=		        \n\t"
+	"jnz		1b		        \n\t"
 
-	".L01END%=:				        \n\t"
+	"2:				        \n\t"
 
 	"cmpq		$4, %6				\n\t"
-	"jne		.L02END%=			\n\t"
+	"jne		3f				\n\t"
 
 	"vmovups	(%4,%0,4), %%ymm8	        \n\t" // 4 complex values form a0
 	"vmovups	(%5,%0,4), %%ymm10              \n\t" // 4 complex values form a1
@@ -279,7 +279,7 @@ static void cgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
 
 	"vmovups  %%ymm12,   (%3,%0,4)		        \n\t" // 4 complex values to y	
 
-	".L02END%=:				        \n\t"
+	"3:				        \n\t"
 	"vzeroupper			 \n\t"
 
 	:
@@ -320,10 +320,10 @@ static void cgemv_kernel_4x1( BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
 	"vbroadcastss	 4(%2), %%ymm1                  \n\t"  // imag part x0
 
 	"cmpq		$0 , %1				\n\t"
-	"je		.L01END%=		        \n\t"
+	"je		2f			        \n\t"
 
 	".align 16				        \n\t"
-	".L01LOOP%=:				        \n\t"
+	"1:				        \n\t"
 	"prefetcht0      320(%4,%0,4)			\n\t"
 	"vmovups	(%4,%0,4), %%ymm8	        \n\t" // 4 complex values form a0
 	"vmovups      32(%4,%0,4), %%ymm9	        \n\t" // 4 complex values form a0
@@ -359,12 +359,12 @@ static void cgemv_kernel_4x1( BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
 	"vmovups  %%ymm12,-64(%3,%0,4)		        \n\t" // 4 complex values to y	
 	"vmovups  %%ymm13,-32(%3,%0,4)		        \n\t"	
 
-	"jnz		.L01LOOP%=		        \n\t"
+	"jnz		1b		        \n\t"
 
-	".L01END%=:				        \n\t"
+	"2:				        \n\t"
 
 	"cmpq		$4, %5				\n\t"
-	"jne		.L02END%=			\n\t"
+	"jne		3f				\n\t"
 
 	"vmovups	(%4,%0,4), %%ymm8	        \n\t" // 4 complex values form a0
 
@@ -386,7 +386,7 @@ static void cgemv_kernel_4x1( BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
 
 	"vmovups  %%ymm12,   (%3,%0,4)		        \n\t" // 4 complex values to y	
 
-	".L02END%=:				        \n\t"
+	"3:				        \n\t"
 	"vzeroupper			 \n\t"
 
 	:
@@ -452,10 +452,10 @@ static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest,FLOAT a
 	"vbroadcastss	  (%5), %%ymm1                  \n\t"  // alpha_i
 
 	"cmpq		$0 , %1				\n\t"
-	"je		.L01END%=		        \n\t"
+	"je		2f			        \n\t"
 
 	".align 16				        \n\t"
-	".L01LOOP%=:				        \n\t"
+	"1:				        \n\t"
 	"vmovups	(%2,%0,4), %%ymm8	        \n\t" // 4 complex values from src
 	"vmovups      32(%2,%0,4), %%ymm9	        \n\t" 
 
@@ -489,12 +489,12 @@ static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest,FLOAT a
 	"vmovups  %%ymm12,-64(%3,%0,4)		        \n\t" // 4 complex values to y	
 	"vmovups  %%ymm13,-32(%3,%0,4)		        \n\t"	
 
-	"jnz		.L01LOOP%=		        \n\t"
+	"jnz		1b		        \n\t"
 
-	".L01END%=:				        \n\t"
+	"2:				        \n\t"
 
 	"cmpq		$4, %6				\n\t"
-	"jne		.L02END%=			\n\t"
+	"jne		3f				\n\t"
 
 	"vmovups	(%2,%0,4), %%ymm8	        \n\t" // 4 complex values src
 
@@ -516,7 +516,7 @@ static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest,FLOAT a
 
 	"vmovups  %%ymm12,   (%3,%0,4)		        \n\t" // 4 complex values to y	
 
-	".L02END%=:				        \n\t"
+	"3:				        \n\t"
 	"vzeroupper			 \n\t"
 
 	:

kernel/x86_64/cgemv_t_microk_haswell-4.c

@@ -47,7 +47,7 @@ static void cgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"vxorps		%%ymm15, %%ymm15, %%ymm15	\n\t"
 
         "testq          $0x04, %1                      \n\t"
-        "jz             .L08LABEL%=                    \n\t"
+        "jz             2f                      \n\t"
 
 	"vmovups	(%4,%0,4), %%ymm4	        \n\t" // 4 complex values from a0
 	"vmovups	(%5,%0,4), %%ymm5               \n\t" // 4 complex values from a1
@@ -72,12 +72,12 @@ static void cgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
         "addq		$8  , %0	  	 	        \n\t"
 	"subq	        $4  , %1			        \n\t"		
 
-        ".L08LABEL%=:                                  \n\t"
+        "2:                                  \n\t"
 	"cmpq           $0, %1                         \n\t"
-        "je             .L08END%=                      \n\t"
+        "je             3f                      \n\t"
 
 	".align 16				        \n\t"
-	".L01LOOP%=:				        \n\t"
+	"1:				        \n\t"
         "prefetcht0      192(%4,%0,4)                   \n\t"
 	"vmovups	(%4,%0,4), %%ymm4	        \n\t" // 4 complex values from a0
         "prefetcht0      192(%5,%0,4)                   \n\t"
@@ -125,9 +125,9 @@ static void cgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 
         "addq		$16 , %0	  	 	        \n\t"
 	"subq	        $8  , %1			        \n\t"		
-	"jnz		.L01LOOP%=		        \n\t"
+	"jnz		1b		        \n\t"
 
-        ".L08END%=:                                   \n\t"
+        "3:                                   \n\t"
 
         "vbroadcastss    (%8)  , %%xmm0                \n\t"  // value from alpha
         "vbroadcastss   4(%8)  , %%xmm1                \n\t"  // value from alpha
@@ -269,7 +269,7 @@ static void cgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"vxorps		%%ymm11, %%ymm11, %%ymm11	\n\t" // temp
 
         "testq          $0x04, %1                      \n\t"
-        "jz             .L08LABEL%=                    \n\t"
+        "jz             2f                    \n\t"
 
 	"vmovups	(%4,%0,4), %%ymm4	        \n\t" // 4 complex values from a0
 	"vmovups	(%5,%0,4), %%ymm5               \n\t" // 4 complex values from a1
@@ -288,12 +288,12 @@ static void cgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
         "addq		$8  , %0	  	 	        \n\t"
 	"subq	        $4  , %1			        \n\t"		
 
-        ".L08LABEL%=:                                  \n\t"
+        "2:                                  \n\t"
 	"cmpq           $0, %1                         \n\t"
-        "je             .L08END%=                      \n\t"
+        "je             3f                      \n\t"
 
 	".align 16				        \n\t"
-	".L01LOOP%=:				        \n\t"
+	"1:				        \n\t"
         "prefetcht0      192(%4,%0,4)                   \n\t"
 	"vmovups	(%4,%0,4), %%ymm4	        \n\t" // 4 complex values from a0
         "prefetcht0      192(%5,%0,4)                   \n\t"
@@ -325,9 +325,9 @@ static void cgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 
         "addq		$16 , %0	  	 	        \n\t"
 	"subq	        $8  , %1			        \n\t"		
-	"jnz		.L01LOOP%=		        \n\t"
+	"jnz		1b		        \n\t"
 
-        ".L08END%=:                                   \n\t"
+        "3:                                   \n\t"
 
         "vbroadcastss    (%6)  , %%xmm0                \n\t"  // value from alpha
         "vbroadcastss   4(%6)  , %%xmm1                \n\t"  // value from alpha
@@ -426,7 +426,7 @@ static void cgemv_kernel_4x1( BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *
 	"vxorps		%%ymm9 , %%ymm9 , %%ymm9 	\n\t" // temp
 
         "testq          $0x04, %1                      \n\t"
-        "jz             .L08LABEL%=                    \n\t"
+        "jz             2f                    \n\t"
 
 	"vmovups	(%4,%0,4), %%ymm4	        \n\t" // 4 complex values from a0
 
@@ -442,12 +442,12 @@ static void cgemv_kernel_4x1( BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *
         "addq		$8  , %0	  	 	        \n\t"
 	"subq	        $4  , %1			        \n\t"		
 
-        ".L08LABEL%=:                                  \n\t"
+        "2:                                  \n\t"
 	"cmpq           $0, %1                         \n\t"
-        "je             .L08END%=                      \n\t"
+        "je             3f                      \n\t"
 
 	".align 16				        \n\t"
-	".L01LOOP%=:				        \n\t"
+	"1:				        \n\t"
         "prefetcht0      192(%4,%0,4)                   \n\t"
 	"vmovups	(%4,%0,4), %%ymm4	        \n\t" // 4 complex values from a0
 
@@ -472,9 +472,9 @@ static void cgemv_kernel_4x1( BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *
 
         "addq		$16 , %0	  	 	        \n\t"
 	"subq	        $8  , %1			        \n\t"		
-	"jnz		.L01LOOP%=		        \n\t"
+	"jnz		1b		        \n\t"
 
-        ".L08END%=:                                   \n\t"
+        "3:                                   \n\t"
 
         "vbroadcastss    (%5)  , %%xmm0                \n\t"  // value from alpha
         "vbroadcastss   4(%5)  , %%xmm1                \n\t"  // value from alpha

kernel/x86_64/daxpy.c

@@ -31,7 +31,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(NEHALEM)
 #include "daxpy_microk_nehalem-2.c"
-#elif defined(BULLDOZER)
+#elif defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER)
 #include "daxpy_microk_bulldozer-2.c"
 #endif
 

kernel/x86_64/daxpy_microk_bulldozer-2.c

@@ -39,7 +39,7 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"vmovddup		(%4), %%xmm0		    \n\t"  // alpha	
 
 	".align 16				            \n\t"
-	".L01LOOP%=:				            \n\t"
+	"1:				            \n\t"
 
         "prefetcht0      768(%3,%0,8)                       \n\t"
         "vmovups                  (%2,%0,8), %%xmm12         \n\t"  // 2 * x
@@ -61,7 +61,7 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 
 	"addq		$8 , %0	  	 	             \n\t"
 	"subq	        $8 , %1			             \n\t"		
-	"jnz		.L01LOOP%=		             \n\t"
+	"jnz		1b		             \n\t"
 
 	:
         : 

kernel/x86_64/daxpy_microk_nehalem-2.c

@@ -40,7 +40,7 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"shufpd          $0,  %%xmm0, %%xmm0                \n\t"
 
 	".align 16				            \n\t"
-	".L01LOOP%=:				            \n\t"
+	"1:				            \n\t"
         // "prefetcht0      192(%2,%0,8)                       \n\t"
         // "prefetcht0      192(%3,%0,8)                       \n\t"
 
@@ -70,7 +70,7 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 
 	"addq		$8 , %0	  	 	             \n\t"
 	"subq	        $8 , %1			             \n\t"		
-	"jnz		.L01LOOP%=		             \n\t"
+	"jnz		1b		             \n\t"
 
 	:
         : 

kernel/x86_64/ddot.c

@@ -29,7 +29,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(BULLDOZER) || defined(PILEDRIVER)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER)
 #include "ddot_microk_bulldozer-2.c"
 #elif defined(NEHALEM)
 #include "ddot_microk_nehalem-2.c"

kernel/x86_64/ddot_microk_bulldozer-2.c

@@ -42,7 +42,7 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vxorpd		%%xmm7, %%xmm7, %%xmm7	             \n\t"
 
 	".align 16				             \n\t"
-	".L01LOOP%=:				             \n\t"
+	"1:				             \n\t"
         "vmovups                  (%2,%0,8), %%xmm12         \n\t"  // 2 * x
         "vmovups                16(%2,%0,8), %%xmm13         \n\t"  // 2 * x
         "vmovups                32(%2,%0,8), %%xmm14         \n\t"  // 2 * x
@@ -55,7 +55,7 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 
 	"addq		$8 , %0	  	 	             \n\t"
 	"subq	        $8 , %1			             \n\t"		
-	"jnz		.L01LOOP%=		             \n\t"
+	"jnz		1b		             \n\t"
 
 	"vaddpd        %%xmm4, %%xmm5, %%xmm4	\n\t"
 	"vaddpd        %%xmm6, %%xmm7, %%xmm6	\n\t"

kernel/x86_64/ddot_microk_nehalem-2.c

@@ -42,7 +42,7 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"xorpd		%%xmm7, %%xmm7	             \n\t"
 
 	".align 16				            \n\t"
-	".L01LOOP%=:				            \n\t"
+	"1:				            \n\t"
 
         "movups                  (%2,%0,8), %%xmm12         \n\t"  // 2 * x
         "movups                  (%3,%0,8), %%xmm8          \n\t"  // 2 * y
@@ -65,7 +65,7 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 
 	"addq		$8 , %0	  	 	             \n\t"
 	"subq	        $8 , %1			             \n\t"		
-	"jnz		.L01LOOP%=		             \n\t"
+	"jnz		1b		             \n\t"
 
 	"addpd        %%xmm5, %%xmm4	\n\t"
 	"addpd        %%xmm7, %%xmm6	\n\t"

kernel/x86_64/dgemm_kernel_4x4_haswell.S

@@ -1092,18 +1092,48 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro INIT4x1
 
-	vxorpd		%xmm4 , %xmm4 , %xmm4
+	vxorpd		%ymm4 , %ymm4 , %ymm4
-	vxorpd		%xmm5 , %xmm5 , %xmm5
+	vxorpd		%ymm5 , %ymm5 , %ymm5
+	vxorpd		%ymm6 , %ymm6 , %ymm6
+	vxorpd		%ymm7 , %ymm7 , %ymm7
+
+.endm
+
+
+.macro KERNEL4x1
+
+	vbroadcastsd	-12 * SIZE(BO), %ymm0
+	vbroadcastsd	-11 * SIZE(BO), %ymm1
+	vbroadcastsd	-10 * SIZE(BO), %ymm2
+	vbroadcastsd	-9  * SIZE(BO), %ymm3
+
+	vfmadd231pd  	-16 * SIZE(AO) ,%ymm0  , %ymm4
+	vfmadd231pd  	-12 * SIZE(AO) ,%ymm1  , %ymm5
+
+	vbroadcastsd	-8  * SIZE(BO), %ymm0
+	vbroadcastsd	-7  * SIZE(BO), %ymm1
+
+	vfmadd231pd  	-8  * SIZE(AO) ,%ymm2  , %ymm6
+	vfmadd231pd  	-4  * SIZE(AO) ,%ymm3  , %ymm7
+
+	vbroadcastsd	-6  * SIZE(BO), %ymm2
+	vbroadcastsd	-5  * SIZE(BO), %ymm3
+
+	vfmadd231pd  	 0  * SIZE(AO) ,%ymm0  , %ymm4
+	vfmadd231pd  	 4  * SIZE(AO) ,%ymm1  , %ymm5
+	vfmadd231pd  	 8  * SIZE(AO) ,%ymm2  , %ymm6
+	vfmadd231pd  	 12 * SIZE(AO) ,%ymm3  , %ymm7
+
+	addq		$ 8 *SIZE, BO
+	addq		$ 32*SIZE, AO
 
 .endm
 
 
 .macro KERNEL4x1_SUB
-	vmovddup	-12 * SIZE(BO), %xmm2
+	vbroadcastsd	-12 * SIZE(BO), %ymm2
-	vmovups 	-16 * SIZE(AO), %xmm0
+	vmovups 	-16 * SIZE(AO), %ymm0
-	vmovups 	-14 * SIZE(AO), %xmm1
+	vfmadd231pd  	%ymm0 ,%ymm2  , %ymm4
-	vfmadd231pd  	%xmm0 ,%xmm2  , %xmm4
-	vfmadd231pd  	%xmm1 ,%xmm2  , %xmm5
 	addq		$ 1*SIZE, BO
 	addq		$ 4*SIZE, AO
 
@@ -1112,21 +1142,22 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .macro SAVE4x1
 
-	vmovddup	ALPHA, %xmm0
+	vbroadcastsd	ALPHA, %ymm0
 
-	vmulpd	%xmm0 , %xmm4 , %xmm4
+	vaddpd	%ymm4,%ymm5, %ymm4 
-	vmulpd	%xmm0 , %xmm5 , %xmm5
+	vaddpd	%ymm6,%ymm7, %ymm6 
+	vaddpd	%ymm4,%ymm6, %ymm4 
+
+	vmulpd	%ymm0 , %ymm4 , %ymm4
 
 
 #if !defined(TRMMKERNEL)
 
-	vaddpd 	                (CO1)     , %xmm4, %xmm4
+	vaddpd 	                (CO1)     , %ymm4, %ymm4
-	vaddpd 	        2 * SIZE(CO1)     , %xmm5, %xmm5
 
 #endif
 
-	vmovups	%xmm4 ,  	(CO1)
+	vmovups	%ymm4 ,  	(CO1)
-	vmovups	%xmm5 , 2 * SIZE(CO1)
 
 	addq	$ 4*SIZE, CO1
 .endm
@@ -2112,15 +2143,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .L1_12:
 
-	KERNEL4x1_SUB
+	KERNEL4x1
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
-
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
 
 	dec	%rax
 	jne	.L1_12
@@ -3180,15 +3203,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 .L1_12:
 
-	KERNEL4x1_SUB
+	KERNEL4x1
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
-
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
-	KERNEL4x1_SUB
 
 	dec	%rax
 	jne	.L1_12

kernel/x86_64/dgemv_n_4.c

@@ -31,7 +31,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(NEHALEM)
 #include "dgemv_n_microk_nehalem-4.c"
-#elif defined(HASWELL)
+#elif defined(HASWELL) || defined(STEAMROLLER)
 #include "dgemv_n_microk_haswell-4.c"
 #endif
 
@@ -125,7 +125,7 @@ static void dgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"shufpd $0,  %%xmm13, %%xmm13    \n\t"	
 
 	".align 16				       \n\t"
-	".L01LOOP%=:				       \n\t"
+	"1:				       \n\t"
 	"movups	       (%3,%0,8), %%xmm4	       \n\t"	// 2 * y
 	"movups	     16(%3,%0,8), %%xmm5	       \n\t"	// 2 * y
 
@@ -148,7 +148,7 @@ static void dgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 
         "addq		$4 , %0	  	 	       \n\t"
 	"subq	        $4 , %1			       \n\t"		
-	"jnz		.L01LOOP%=		       \n\t"
+	"jnz		1b		       \n\t"
 
 	:
         : 
@@ -187,7 +187,7 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a
         "shufpd $0,  %%xmm12, %%xmm12            \n\t"
 
         ".align 16                               \n\t"
-        ".L01LOOP%=:                             \n\t"
+        "1:                             \n\t"
         "movups       (%4,%0,8), %%xmm8          \n\t"  // 2 * a
         "movups     16(%4,%0,8), %%xmm9          \n\t"  // 2 * a
         "movups       (%3,%0,8), %%xmm4          \n\t"  // 2 * y
@@ -203,7 +203,7 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a
         "addq           $4 , %0                  \n\t"
         "subq           $4 , %1                  \n\t"
 
-        "jnz            .L01LOOP%=               \n\t"
+        "jnz            1b               \n\t"
 
         :
         :

kernel/x86_64/dgemv_n_microk_haswell-4.c

@@ -50,7 +50,7 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vbroadcastsd    (%9), %%ymm6 	 \n\t"	// alpha 
 
         "testq          $0x04, %1                      \n\t"
-        "jz             .L8LABEL%=                     \n\t"
+        "jz             2f                     \n\t"
 
 	"vmovupd	(%3,%0,8), %%ymm7	       \n\t"	// 4 * y
 	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
@@ -77,14 +77,14 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
         "addq		$4 , %0	  	 	       \n\t"
 	"subq	        $4 , %1			       \n\t"		
 
-        ".L8LABEL%=:                                   \n\t"
+        "2:                                   \n\t"
 
         "cmpq           $0, %1                         \n\t"
-        "je             .L16END%=                      \n\t"
+        "je             3f                      \n\t"
 
 
 	".align 16				 \n\t"
-	".L01LOOP%=:				 \n\t"
+	"1:				 \n\t"
 
 	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
 	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"
@@ -118,9 +118,9 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"subq	        $8 , %1			      \n\t"		
 	"vmovupd  %%ymm9,-32(%3,%0,8)		      \n\t"	// 4 * y
 
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      \n\t"
 
-        ".L16END%=:                             \n\t"
+        "3:                             \n\t"
 	"vzeroupper			        \n\t"
 
 	:
@@ -168,7 +168,7 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"vbroadcastsd    (%8), %%ymm6 	 \n\t"	// alpha 
 
         "testq          $0x04, %1                      \n\t"
-        "jz             .L8LABEL%=                     \n\t"
+        "jz             2f                     \n\t"
 
 	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
 	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"
@@ -188,14 +188,14 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
         "addq		$4 , %0	  	 	       \n\t"
 	"subq	        $4 , %1			       \n\t"		
 
-        ".L8LABEL%=:                                   \n\t"
+        "2:                                   \n\t"
 
         "cmpq           $0, %1                         \n\t"
-        "je             .L8END%=                       \n\t"
+        "je             3f                       \n\t"
 
 
 	".align 16				 \n\t"
-	".L01LOOP%=:				 \n\t"
+	"1:				 \n\t"
 	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
 	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"
 	"vmovupd	(%3,%0,8), %%ymm8	       \n\t"	// 4 * y
@@ -218,9 +218,9 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 
         "addq		$8 , %0	  	 	      \n\t"
 	"subq	        $8 , %1			      \n\t"		
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      \n\t"
 
-        ".L8END%=:                                    \n\t"
+        "3:                                    \n\t"
 	"vzeroupper			              \n\t"
 
 	:

kernel/x86_64/dgemv_n_microk_nehalem-4.c

@@ -60,7 +60,7 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 
 
 	".align 16				 \n\t"
-	".L01LOOP%=:				 \n\t"
+	"1:				 \n\t"
 	"xorpd           %%xmm4 , %%xmm4	 \n\t"
 	"xorpd           %%xmm5 , %%xmm5	 \n\t"
 	"movups             (%3,%0,8), %%xmm7          \n\t" // 2 * y
@@ -142,7 +142,7 @@ static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 
         "addq		$4 , %0	  	 	       \n\t"
 	"subq	        $4 , %1			       \n\t"		
-	"jnz		.L01LOOP%=		       \n\t"
+	"jnz		1b		       \n\t"
 
 	:
         : 
@@ -194,7 +194,7 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"shufpd $0,  %%xmm6 , %%xmm6 \n\t"	
 
 	".align 16				 \n\t"
-	".L01LOOP%=:				 \n\t"
+	"1:				 \n\t"
 	"xorpd           %%xmm4 , %%xmm4	 \n\t"
 	"xorpd           %%xmm5 , %%xmm5	 \n\t"
 	"movups	       (%3,%0,8), %%xmm7	 \n\t"	// 2 * y
@@ -239,7 +239,7 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 
         "addq		$4 , %0	  	 	       \n\t"
 	"subq	        $4 , %1			       \n\t"		
-	"jnz		.L01LOOP%=		       \n\t"
+	"jnz		1b		       \n\t"
 
 	:
         : 

kernel/x86_64/dgemv_n_microk_piledriver-4.c

@@ -0,0 +1,247 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+
+
+#define HAVE_KERNEL_4x8 1
+static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLONG lda4, FLOAT *alpha) __attribute__ ((noinline));
+
+static void dgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLONG lda4, FLOAT *alpha)
+{
+
+	BLASLONG register i = 0;
+
+	__asm__  __volatile__
+	(
+	"vzeroupper			 \n\t"
+	"vbroadcastsd    (%2), %%ymm12	 \n\t"	// x0 
+	"vbroadcastsd   8(%2), %%ymm13	 \n\t"	// x1 
+	"vbroadcastsd  16(%2), %%ymm14	 \n\t"	// x2 
+	"vbroadcastsd  24(%2), %%ymm15	 \n\t"	// x3 
+	"vbroadcastsd  32(%2), %%ymm0 	 \n\t"	// x4 
+	"vbroadcastsd  40(%2), %%ymm1 	 \n\t"	// x5 
+	"vbroadcastsd  48(%2), %%ymm2 	 \n\t"	// x6 
+	"vbroadcastsd  56(%2), %%ymm3 	 \n\t"	// x7 
+
+	"vbroadcastsd    (%9), %%ymm6 	 \n\t"	// alpha 
+
+        "testq          $0x04, %1                      \n\t"
+        "jz             2f                     \n\t"
+
+	"vmovupd	(%3,%0,8), %%ymm7	       \n\t"	// 4 * y
+	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
+	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"
+
+	"vfmadd231pd   (%4,%0,8), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231pd   (%5,%0,8), %%ymm13, %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%0,8), %%ymm14, %%ymm4      \n\t" 
+	"vfmadd231pd   (%7,%0,8), %%ymm15, %%ymm5      \n\t" 
+
+	"vfmadd231pd   (%4,%8,8), %%ymm0 , %%ymm4      \n\t" 
+	"vfmadd231pd   (%5,%8,8), %%ymm1 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%8,8), %%ymm2 , %%ymm4      \n\t" 
+	"vfmadd231pd   (%7,%8,8), %%ymm3 , %%ymm5      \n\t" 
+
+	"vaddpd		%%ymm4 , %%ymm5 , %%ymm5       \n\t"
+	"vmulpd		%%ymm6 , %%ymm5 , %%ymm5       \n\t"
+	"vaddpd		%%ymm7 , %%ymm5 , %%ymm5       \n\t"
+
+
+	"vmovupd  %%ymm5,   (%3,%0,8)		       \n\t"	// 4 * y
+
+        "addq		$4 , %8	  	 	       \n\t"
+        "addq		$4 , %0	  	 	       \n\t"
+	"subq	        $4 , %1			       \n\t"		
+
+        "2:                                   \n\t"
+
+        "cmpq           $0, %1                         \n\t"
+        "je             3f                      \n\t"
+
+
+	".align 16				 \n\t"
+	"1:				 \n\t"
+
+	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
+	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"
+	"vmovupd	(%3,%0,8), %%ymm8	       \n\t"	// 4 * y
+	"vmovupd      32(%3,%0,8), %%ymm9	       \n\t"	// 4 * y
+
+	"vfmadd231pd   (%4,%0,8), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%4,%0,8), %%ymm12, %%ymm5      \n\t" 
+	"vfmadd231pd   (%5,%0,8), %%ymm13, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%5,%0,8), %%ymm13, %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%0,8), %%ymm14, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%6,%0,8), %%ymm14, %%ymm5      \n\t" 
+	"vfmadd231pd   (%7,%0,8), %%ymm15, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%7,%0,8), %%ymm15, %%ymm5      \n\t" 
+
+	"vfmadd231pd   (%4,%8,8), %%ymm0 , %%ymm4      \n\t" 
+        "addq		$8 , %0	  	 	       \n\t"
+	"vfmadd231pd 32(%4,%8,8), %%ymm0 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%5,%8,8), %%ymm1 , %%ymm4      \n\t" 
+	"vfmadd231pd 32(%5,%8,8), %%ymm1 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%8,8), %%ymm2 , %%ymm4      \n\t" 
+	"vfmadd231pd 32(%6,%8,8), %%ymm2 , %%ymm5      \n\t" 
+	"vfmadd231pd   (%7,%8,8), %%ymm3 , %%ymm4      \n\t" 
+	"vfmadd231pd 32(%7,%8,8), %%ymm3 , %%ymm5      \n\t" 
+
+	"vfmadd231pd     %%ymm6 , %%ymm4 , %%ymm8      \n\t"
+	"vfmadd231pd     %%ymm6 , %%ymm5 , %%ymm9      \n\t"
+
+        "addq		$8 , %8	  	 	      \n\t"
+	"vmovupd  %%ymm8,-64(%3,%0,8)		      \n\t"	// 4 * y
+	"subq	        $8 , %1			      \n\t"		
+	"vmovupd  %%ymm9,-32(%3,%0,8)		      \n\t"	// 4 * y
+
+	"jnz		1b		      \n\t"
+
+        "3:                             \n\t"
+	"vzeroupper			        \n\t"
+
+	:
+        : 
+          "r" (i),	// 0	
+	  "r" (n),  	// 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (ap[0]),  // 4
+          "r" (ap[1]),  // 5
+          "r" (ap[2]),  // 6
+          "r" (ap[3]),  // 7
+          "r" (lda4),   // 8
+          "r" (alpha)   // 9
+	: "cc", 
+	  "%xmm0", "%xmm1", 
+	  "%xmm2", "%xmm3", 
+	  "%xmm4", "%xmm5", 
+	  "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+
+
+#define HAVE_KERNEL_4x4 1
+static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha) __attribute__ ((noinline));
+
+static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
+{
+
+	BLASLONG register i = 0;
+
+	__asm__  __volatile__
+	(
+	"vzeroupper			 \n\t"
+	"vbroadcastsd    (%2), %%ymm12	 \n\t"	// x0 
+	"vbroadcastsd   8(%2), %%ymm13	 \n\t"	// x1 
+	"vbroadcastsd  16(%2), %%ymm14	 \n\t"	// x2 
+	"vbroadcastsd  24(%2), %%ymm15	 \n\t"	// x3 
+
+	"vbroadcastsd    (%8), %%ymm6 	 \n\t"	// alpha 
+
+        "testq          $0x04, %1                      \n\t"
+        "jz             2f                     \n\t"
+
+	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
+	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"
+	"vmovupd	(%3,%0,8), %%ymm7	       \n\t"	// 4 * y
+
+	"vfmadd231pd   (%4,%0,8), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231pd   (%5,%0,8), %%ymm13, %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%0,8), %%ymm14, %%ymm4      \n\t" 
+	"vfmadd231pd   (%7,%0,8), %%ymm15, %%ymm5      \n\t" 
+
+	"vaddpd		%%ymm4 , %%ymm5 , %%ymm5       \n\t"
+	"vmulpd		%%ymm6 , %%ymm5 , %%ymm5       \n\t"
+	"vaddpd		%%ymm7 , %%ymm5 , %%ymm5       \n\t"
+
+	"vmovupd  %%ymm5,   (%3,%0,8)		       \n\t"	// 4 * y
+
+        "addq		$4 , %0	  	 	       \n\t"
+	"subq	        $4 , %1			       \n\t"		
+
+        "2:                                   \n\t"
+
+        "cmpq           $0, %1                         \n\t"
+        "je             3f                       \n\t"
+
+
+	".align 16				 \n\t"
+	"1:				 \n\t"
+	"vxorpd		%%ymm4 , %%ymm4, %%ymm4        \n\t"
+	"vxorpd		%%ymm5 , %%ymm5, %%ymm5        \n\t"
+	"vmovupd	(%3,%0,8), %%ymm8	       \n\t"	// 4 * y
+	"vmovupd      32(%3,%0,8), %%ymm9	       \n\t"	// 4 * y
+
+	"vfmadd231pd   (%4,%0,8), %%ymm12, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%4,%0,8), %%ymm12, %%ymm5      \n\t" 
+	"vfmadd231pd   (%5,%0,8), %%ymm13, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%5,%0,8), %%ymm13, %%ymm5      \n\t" 
+	"vfmadd231pd   (%6,%0,8), %%ymm14, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%6,%0,8), %%ymm14, %%ymm5      \n\t" 
+	"vfmadd231pd   (%7,%0,8), %%ymm15, %%ymm4      \n\t" 
+	"vfmadd231pd 32(%7,%0,8), %%ymm15, %%ymm5      \n\t" 
+
+	"vfmadd231pd     %%ymm6 , %%ymm4 , %%ymm8      \n\t"
+	"vfmadd231pd     %%ymm6 , %%ymm5 , %%ymm9      \n\t"
+
+	"vmovupd  %%ymm8,   (%3,%0,8)		      \n\t"	// 4 * y
+	"vmovupd  %%ymm9, 32(%3,%0,8)		      \n\t"	// 4 * y
+
+        "addq		$8 , %0	  	 	      \n\t"
+	"subq	        $8 , %1			      \n\t"		
+	"jnz		1b		      \n\t"
+
+        "3:                                    \n\t"
+	"vzeroupper			              \n\t"
+
+	:
+        : 
+          "r" (i),	// 0	
+	  "r" (n),  	// 1
+          "r" (x),      // 2
+          "r" (y),      // 3
+          "r" (ap[0]),  // 4
+          "r" (ap[1]),  // 5
+          "r" (ap[2]),  // 6
+          "r" (ap[3]),  // 7
+          "r" (alpha)   // 8
+	: "cc", 
+	  "%xmm4", "%xmm5", 
+	  "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
+	  "memory"
+	);
+
+} 
+
+

kernel/x86_64/dgemv_t_4.c

@@ -28,7 +28,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(HASWELL)
+#if defined(HASWELL) || defined(STEAMROLLER)
 #include "dgemv_t_microk_haswell-4.c"
 #endif
 
@@ -78,7 +78,7 @@ static void dgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT
 	"xorpd %%xmm11 , %%xmm11		\n\t"
 		
 	"testq	$2 , %1				\n\t"
-	"jz	.L01LABEL%=			\n\t"
+	"jz	2f			\n\t"
 
 	"movups  (%5,%0,8) , %%xmm14		\n\t" // x
 	"movups  (%3,%0,8) , %%xmm12		\n\t" // ap0
@@ -90,13 +90,13 @@ static void dgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT
         "subq           $2 , %1                 \n\t"
 	"addpd   %%xmm13   , %%xmm11		\n\t"
 
-        ".L01LABEL%=:                           \n\t"
+        "2:                           \n\t"
 
 	"cmpq	$0, %1				\n\t"
-	"je	.L01END%=			\n\t"
+	"je	3f			\n\t"
 
         ".align 16                              \n\t"
-        ".L01LOOP%=:                            \n\t"
+        "1:                            \n\t"
 
 	"movups  (%5,%0,8) , %%xmm14		\n\t" // x
 	"movups  (%3,%0,8) , %%xmm12		\n\t" // ap0
@@ -116,9 +116,9 @@ static void dgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT
 
         "addq           $4 , %0                 \n\t"
         "subq           $4 , %1                 \n\t"
-        "jnz            .L01LOOP%=              \n\t"
+        "jnz            1b              \n\t"
 
-        ".L01END%=:                             \n\t"
+        "3:                             \n\t"
 
 	"haddpd        %%xmm10, %%xmm10         \n\t"
 	"haddpd        %%xmm11, %%xmm11         \n\t"
@@ -157,7 +157,7 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
 	"xorpd %%xmm10 , %%xmm10		\n\t"
 	
 	"testq	$2 , %1				\n\t"
-	"jz	.L01LABEL%=			\n\t"
+	"jz	2f			\n\t"
 
 	"movups  (%3,%0,8) , %%xmm12		\n\t"
 	"movups  (%4,%0,8) , %%xmm11		\n\t"
@@ -166,13 +166,13 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
 	"addpd   %%xmm12   , %%xmm10		\n\t"
         "subq           $2 , %1                 \n\t"
 
-        ".L01LABEL%=:                           \n\t"
+        "2:                           \n\t"
 
 	"cmpq	$0, %1				\n\t"
-	"je	.L01END%=			\n\t"
+	"je	3f			\n\t"
 
         ".align 16                              \n\t"
-        ".L01LOOP%=:                            \n\t"
+        "1:                            \n\t"
 
 	"movups    (%3,%0,8) , %%xmm12		\n\t"
 	"movups  16(%3,%0,8) , %%xmm14		\n\t"
@@ -185,9 +185,9 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
         "subq           $4 , %1                 \n\t"
 	"addpd   %%xmm14   , %%xmm9 		\n\t"
 
-        "jnz            .L01LOOP%=              \n\t"
+        "jnz            1b              \n\t"
 
-        ".L01END%=:                             \n\t"
+        "3:                             \n\t"
 
 	"addpd	       %%xmm9 , %%xmm10         \n\t"
 	"haddpd        %%xmm10, %%xmm10         \n\t"
@@ -246,7 +246,7 @@ static void add_y(BLASLONG n, FLOAT da , FLOAT *src, FLOAT *dest, BLASLONG inc_d
 	"shufpd  $0 , %%xmm10 , %%xmm10		\n\t"
 
         ".align 16                              \n\t"
-        ".L01LOOP%=:                            \n\t"
+        "1:                            \n\t"
 
 	"movups  (%3,%0,8) , %%xmm12		\n\t"
 	"movups  (%4,%0,8) , %%xmm11		\n\t"
@@ -256,7 +256,7 @@ static void add_y(BLASLONG n, FLOAT da , FLOAT *src, FLOAT *dest, BLASLONG inc_d
         "subq           $2 , %1                 \n\t"
 	"movups  %%xmm11, -16(%4,%0,8)		\n\t"
 
-        "jnz            .L01LOOP%=              \n\t"
+        "jnz            1b              \n\t"
 
         :
    	:

kernel/x86_64/dgemv_t_microk_haswell-4.c

@@ -42,7 +42,7 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
 	"vxorpd		%%ymm7 , %%ymm7, %%ymm7  \n\t"
 
         "testq          $0x04, %1                      \n\t"
-        "jz             .L08LABEL%=                    \n\t"
+        "jz             2f                    \n\t"
 
 	"vmovups	(%2,%0,8), %%ymm12       \n\t"	// 4 * x
 
@@ -54,13 +54,13 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
         "addq		$4 , %0	  	 	      \n\t"
 	"subq	        $4 , %1			      \n\t"		
 
-        ".L08LABEL%=:                                  \n\t"
+        "2:                                  \n\t"
 
         "cmpq           $0, %1                         \n\t"
-        "je             .L16END%=                      \n\t"
+        "je             3f                      \n\t"
 
 	".align 16				 \n\t"
-	".L01LOOP%=:				 \n\t"
+	"1:				 \n\t"
 	// "prefetcht0	 384(%2,%0,8)		 \n\t"
 	"vmovups	(%2,%0,8), %%ymm12       \n\t"	// 4 * x
 	"vmovups      32(%2,%0,8), %%ymm13       \n\t"	// 4 * x
@@ -80,9 +80,9 @@ static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y)
 	"subq	        $8 , %1			       \n\t"		
 	"vfmadd231pd -32(%7,%0,8), %%ymm13, %%ymm7     \n\t" 
 
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      \n\t"
 
-        ".L16END%=:                                   \n\t"
+        "3:                                   \n\t"
 
 	"vextractf128   $1 , %%ymm4, %%xmm12	      \n\t"
 	"vextractf128   $1 , %%ymm5, %%xmm13	      \n\t"

kernel/x86_64/dsymv_L.c

@@ -28,7 +28,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(BULLDOZER)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER)
 #include "dsymv_L_microk_bulldozer-2.c"
 #elif defined(NEHALEM)
 #include "dsymv_L_microk_nehalem-2.c"

kernel/x86_64/dsymv_L_microk_bulldozer-2.c

@@ -44,7 +44,7 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	"vmovddup 24(%8),    %%xmm7	             \n\t"	// temp1[1]
 
 	".align 16				     \n\t"
-	".L01LOOP%=:				     \n\t"
+	"1:				     \n\t"
 
 	"vmovups	(%4,%0,8), %%xmm12	           \n\t"  // 2 * a
 	"vmovups	(%2,%0,8), %%xmm8	           \n\t"  // 2 * x
@@ -90,7 +90,7 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	"vmovups	%%xmm11 , -16(%3,%0,8)		   \n\t"
 
 	"cmpq		%0 , %1			      \n\t"
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      \n\t"
 
 	"vmovsd		  (%9), %%xmm4		      \n\t"
 	"vmovsd		 8(%9), %%xmm5		      \n\t"

kernel/x86_64/dsymv_L_microk_nehalem-2.c

@@ -48,7 +48,7 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	"shufpd $0,  %%xmm7, %%xmm7                  \n\t"
 
 	".align 16		  		       \n\t"
-	".L01LOOP%=:				       \n\t"
+	"1:				       \n\t"
 	"movups	            (%4,%0,8), %%xmm12	       \n\t"	// 2 * a
 	"movups	            (%2,%0,8), %%xmm8	       \n\t"	// 2 * x
 	"movups		    %%xmm12  , %%xmm11	       \n\t"
@@ -85,7 +85,7 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	"movups             %%xmm9,-16(%3,%0,8)       \n\t"    // 2 * y
 
 	"cmpq	        %0 , %1			      \n\t"		
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      \n\t"
 
 	"movsd	       (%9),    %%xmm4	             \n\t"	// temp1[0]
 	"movsd        8(%9),    %%xmm5	             \n\t"	// temp1[1]

kernel/x86_64/dsymv_U.c

@@ -29,7 +29,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(BULLDOZER)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER)
 #include "dsymv_U_microk_bulldozer-2.c"
 #elif defined(NEHALEM)
 #include "dsymv_U_microk_nehalem-2.c"

kernel/x86_64/dsymv_U_microk_bulldozer-2.c

@@ -47,7 +47,7 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
 	"xorq		%0,%0			     \n\t"
 
 	".align 16				     \n\t"
-	".L01LOOP%=:				     \n\t"
+	"1:				     \n\t"
 
 	"vmovups	(%4,%0,8), %%xmm12	           \n\t"  // 2 * a
 	"vmovups	(%2,%0,8), %%xmm8	           \n\t"  // 2 * x
@@ -93,7 +93,7 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
 	"vmovups	%%xmm9 ,  -32(%3,%0,8)		   \n\t"
 	"vmovups	%%xmm11 , -16(%3,%0,8)		   \n\t"
 
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      \n\t"
 
 	"vhaddpd        %%xmm0, %%xmm0, %%xmm0  \n\t"
 	"vhaddpd        %%xmm1, %%xmm1, %%xmm1  \n\t"

kernel/x86_64/dsymv_U_microk_nehalem-2.c

@@ -51,7 +51,7 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
 	"xorq		%0,%0			     \n\t"
 
 	".align 16		  		       \n\t"
-	".L01LOOP%=:				       \n\t"
+	"1:				       \n\t"
 	"movups	            (%4,%0,8), %%xmm12	       \n\t"	// 2 * a
 	"movups	            (%2,%0,8), %%xmm8	       \n\t"	// 2 * x
 	"movups		    %%xmm12  , %%xmm11	       \n\t"
@@ -88,7 +88,7 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
 	"movups             %%xmm9,-16(%3,%0,8)       \n\t"    // 2 * y
 
 	"subq	        $2 , %1			      \n\t"		
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      \n\t"
 
 	"haddpd        %%xmm0, %%xmm0  \n\t"
 	"haddpd        %%xmm1, %%xmm1  \n\t"

kernel/x86_64/saxpy_microk_nehalem-2.c

@@ -40,7 +40,7 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 	"shufps          $0,  %%xmm0, %%xmm0                \n\t"
 
 	".align 16				            \n\t"
-	".L01LOOP%=:				            \n\t"
+	"1:				            \n\t"
         // "prefetcht0      192(%2,%0,4)                       \n\t"
         // "prefetcht0      192(%3,%0,4)                       \n\t"
 
@@ -70,7 +70,7 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
 
 	"addq		$16, %0	  	 	             \n\t"
 	"subq	        $16, %1			             \n\t"		
-	"jnz		.L01LOOP%=		             \n\t"
+	"jnz		1b		             \n\t"
 
 	:
         : 

kernel/x86_64/sdot.c

@@ -28,7 +28,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "common.h"
 
-#if defined(BULLDOZER) || defined(PILEDRIVER)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER)
 #include "sdot_microk_bulldozer-2.c"
 #elif defined(NEHALEM)
 #include "sdot_microk_nehalem-2.c"

kernel/x86_64/sdot_microk_bulldozer-2.c

@@ -42,7 +42,7 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"vxorps		%%xmm7, %%xmm7, %%xmm7	             \n\t"
 
 	".align 16				             \n\t"
-	".L01LOOP%=:				             \n\t"
+	"1:				             \n\t"
         "vmovups                  (%2,%0,4), %%xmm12         \n\t"  // 4 * x
         "vmovups                16(%2,%0,4), %%xmm13         \n\t"  // 4 * x
         "vmovups                32(%2,%0,4), %%xmm14         \n\t"  // 4 * x
@@ -55,7 +55,7 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 
 	"addq		$16, %0	  	 	             \n\t"
 	"subq	        $16, %1			             \n\t"		
-	"jnz		.L01LOOP%=		             \n\t"
+	"jnz		1b		             \n\t"
 
 	"vaddps        %%xmm4, %%xmm5, %%xmm4	\n\t"
 	"vaddps        %%xmm6, %%xmm7, %%xmm6	\n\t"

kernel/x86_64/sdot_microk_nehalem-2.c

@@ -42,7 +42,7 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	"xorps		%%xmm7, %%xmm7	             \n\t"
 
 	".align 16				            \n\t"
-	".L01LOOP%=:				            \n\t"
+	"1:				            \n\t"
         "movups                  (%2,%0,4), %%xmm12         \n\t"  // 4 * x
         "movups                  (%3,%0,4), %%xmm8          \n\t"  // 4 * x
         "movups                16(%2,%0,4), %%xmm13         \n\t"  // 4 * x
@@ -64,7 +64,7 @@ static void sdot_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 
 	"addq		$16, %0	  	 	             \n\t"
 	"subq	        $16, %1			             \n\t"		
-	"jnz		.L01LOOP%=		             \n\t"
+	"jnz		1b		             \n\t"
 
 	"addps        %%xmm5, %%xmm4	\n\t"
 	"addps        %%xmm7, %%xmm6	\n\t"

kernel/x86_64/sgemv_n_4.c

@@ -29,7 +29,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 
-#if defined(BULLDOZER) || defined(PILEDRIVER)
+#if defined(BULLDOZER) || defined(PILEDRIVER) || defined(STEAMROLLER)
 #include "sgemv_n_microk_bulldozer-4.c"
 #elif defined(NEHALEM)
 #include "sgemv_n_microk_nehalem-4.c"
@@ -39,8 +39,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "sgemv_n_microk_haswell-4.c"
 #endif
 
-
+#if defined(STEAMROLLER)
+#define NBMAX 2048
+#else
 #define NBMAX 4096
+#endif
 
 #ifndef HAVE_KERNEL_4x8
 
@@ -129,7 +132,7 @@ static void sgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"shufps $0,  %%xmm13, %%xmm13    \n\t"	
 
 	".align 16				       \n\t"
-	".L01LOOP%=:				       \n\t"
+	"1:				       \n\t"
 	"movups	       (%3,%0,4), %%xmm4	       \n\t"	// 4 * y
 
 	"movups             (%4,%0,4), %%xmm8          \n\t" 
@@ -143,7 +146,7 @@ static void sgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"movups  %%xmm4 , -16(%3,%0,4)		       \n\t"	// 4 * y
 
 	"subq	        $4 , %1			       \n\t"		
-	"jnz		.L01LOOP%=		       \n\t"
+	"jnz		1b		       \n\t"
 
 	:
         : 
@@ -166,7 +169,7 @@ static void sgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 
 #endif
 
-#ifndef HAVE_KERNEL_4x2
+#ifndef HAVE_KERNEL_4x1
 
 static void sgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *alpha) __attribute__ ((noinline));
 
@@ -184,10 +187,10 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a
         "shufps $0,  %%xmm12, %%xmm12            \n\t"
 
         "cmpq           $0, %1                   \n\t"
-        "je             .L16END%=                \n\t"
+        "je             2f                \n\t"
 
         ".align 16                               \n\t"
-        ".L01LOOP%=:                             \n\t"
+        "1:                             \n\t"
         "movups       (%3,%0,4), %%xmm4          \n\t"  // 4 * y
         "movups     16(%3,%0,4), %%xmm5          \n\t"  // 4 * y
         "movups       (%4,%0,4), %%xmm8          \n\t"  // 4 * a
@@ -203,12 +206,12 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a
 
         "subq           $8 , %1                  \n\t"
 
-        "jnz            .L01LOOP%=               \n\t"
+        "jnz            1b               \n\t"
 
-        ".L16END%=:                              \n\t"
+        "2:                              \n\t"
 
         "testq          $0x04, %5                \n\t"
-        "jz             .L08LABEL%=              \n\t"
+        "jz             3f              \n\t"
 
         "movups       (%3,%0,4), %%xmm4          \n\t"  // 4 * y
         "movups       (%4,%0,4), %%xmm8          \n\t"  // 4 * a
@@ -218,7 +221,7 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a
         "addq           $4 , %0                  \n\t"
         "subq           $4 , %1                  \n\t"
 
-        ".L08LABEL%=:      			 \n\t" 
+        "3:      			 \n\t" 
         :
         :
           "r" (i),      // 0    
@@ -262,7 +265,7 @@ static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest)
         (
 
         ".align 16                              \n\t"
-        ".L01LOOP%=:                            \n\t"
+        "1:                            \n\t"
 
         "movups  (%2,%0,4) , %%xmm12            \n\t"
         "movups  (%3,%0,4) , %%xmm11            \n\t"
@@ -271,7 +274,7 @@ static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest)
         "movups  %%xmm11, -16(%3,%0,4)          \n\t"
 
         "subq           $4 , %1                 \n\t"
-        "jnz            .L01LOOP%=              \n\t"
+        "jnz            1b              \n\t"
 
         :
         :

kernel/x86_64/sgemv_n_microk_bulldozer-4.c

@@ -49,7 +49,7 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vbroadcastss    (%9), %%xmm8 	 \n\t"	// alpha 
 
         "testq          $0x04, %1                      \n\t"
-        "jz             .L08LABEL%=                    \n\t"
+        "jz             2f                    \n\t"
 
 	"vxorps		%%xmm4, %%xmm4 , %%xmm4  \n\t"
 	"vxorps		%%xmm5, %%xmm5 , %%xmm5  \n\t"
@@ -71,10 +71,10 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"subq	        $4 , %1			       \n\t"		
 	"vmovups  %%xmm6, -16(%3,%0,4)		       \n\t"	// 4 * y
 
-	".L08LABEL%=:                                  \n\t"
+	"2:                                  \n\t"
 
         "testq          $0x08, %1                      \n\t"
-        "jz             .L16LABEL%=                    \n\t"
+        "jz             3f                    \n\t"
 
 	"vxorps		%%xmm4, %%xmm4 , %%xmm4  \n\t"
 	"vxorps		%%xmm5, %%xmm5 , %%xmm5  \n\t"
@@ -107,13 +107,13 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"subq	        $8 , %1			      \n\t"		
 
 
-        ".L16LABEL%=:                                  \n\t"
+        "3:                                  \n\t"
 
         "cmpq           $0, %1                         \n\t"
-        "je             .L16END%=                      \n\t"
+        "je             4f                      \n\t"
 
 	".align 16				 \n\t"
-	".L01LOOP%=:				 \n\t"
+	"1:				 \n\t"
 
 	"vxorps		%%xmm4, %%xmm4 , %%xmm4  \n\t"
 	"vxorps		%%xmm5, %%xmm5 , %%xmm5  \n\t"
@@ -178,9 +178,9 @@ static void sgemv_kernel_4x8( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, BLASLO
 	"vmovups  %%xmm7,-16(%3,%0,4)		      \n\t"	// 4 * y
 
 	"subq	        $16, %1			      \n\t"		
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      \n\t"
 
-	".L16END%=:                             \n\t"
+	"4:                             \n\t"
 
 	:
         : 
@@ -227,7 +227,7 @@ static void sgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 	"vbroadcastss    (%8), %%xmm8 	 \n\t"	// alpha 
 
 	".align 16				 \n\t"
-	".L01LOOP%=:				 \n\t"
+	"1:				 \n\t"
 	"vxorps		%%xmm4, %%xmm4 , %%xmm4  \n\t"
 	"vxorps		%%xmm5, %%xmm5 , %%xmm5  \n\t"
 
@@ -243,7 +243,7 @@ static void sgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
 
         "addq		$4 , %0	  	 	      \n\t"
 	"subq	        $4 , %1			      \n\t"		
-	"jnz		.L01LOOP%=		      \n\t"
+	"jnz		1b		      	\n\t"
 
 	:
         :