Merge branch 'develop'

It overflowed the internal buffer. Thus, we split vector x into blocks when m is very large.

Thank @wangqian for this patch.

.gitignore

@@ -1,16 +1,25 @@
 *.obj
 *.lib
 *.dll
+*.dylib
 *.def
 *.o
 lapack-3.1.1
 lapack-3.1.1.tgz
+lapack-3.4.1
+lapack-3.4.1.tgz
+lapack-3.4.2
+lapack-3.4.2.tgz
 *.so
 *.a
 .svn
 *~
+lib.grd
+nohup.out
 config.h
 Makefile.conf
+Makefile.conf_last
+config_last.h
 getarch
 getarch_2nd
 utest/openblas_utest

Changelog.txt

@@ -1,4 +1,115 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.2.5
+26-Nov-2012
+common:
+	* Added NO_SHARED flag to disable generating the shared library.
+	* Compile LAPACKE with ILP64 modle when INTERFACE64=1 (#158)
+	* Export LAPACK 3.4.2 symbols in shared library. (#147)
+	* Only detect the number of physical CPU cores on Mac OSX. (#157)
+	* Fixed NetBSD build. (#155)
+	* Fixed compilation with TARGET=GENERIC. (#160)
+x86/x86-64:
+	* Restore the original CPU affinity when calling 
+	  openblas_set_num_threads(1) (#153)
+	* Fixed a SEGFAULT bug in dgemv_t when m is very large.(#154)
+MIPS64:
+
+====================================================================
+Version 0.2.4
+8-Oct-2012
+common:
+	* Upgraded LAPACK to 3.4.2 version. (#145)
+	* Provided support for passing CFLAGS, FFLAGS, PFLAGS, 
+	  FPFLAGS to make. (#137)
+	* f77blas.h:compatibility for compilers without C99 complex 
+	  number support. (#141)
+x86/x86-64:
+	* Added NO_AVX flag. Check OS supporting AVX on runtime. (#139)
+	* Fixed zdot incompatibility ABI issue with GCC 4.7 on 
+	  Windows 32-bit. (#140)
+MIPS64:
+	* Fixed the generation of shared library bug.
+	* Fixed the detection bug on the Loongson 3A server.
+====================================================================
+Version 0.2.3
+20-Aug-2012
+common:
+	* Fixed LAPACK unstable bug about ?laswp. (#130)
+	* Fixed the shared library bug about unloading the library on 
+	  Linux (#132).
+	* Fixed the compilation failure on BlueGene/P (TARGET=PPC440FP2)
+	  Please use gcc and IBM xlf. (#134)
+x86/x86-64:
+	* Supported goto_set_num_threads and openblas_set_num_threads 
+	  APIs in Windows. They can set the number of threads on runtime.
+	  
+====================================================================
+Version 0.2.2
+6-July-2012
+common:
+	* Fixed exporting DLL functions bug on Windows/MingW
+	* Support GNU Hurd (Thank Sylvestre Ledru)
+	* Support kfreebsd kernel (Thank Sylvestre Ledru)
+x86/x86-64:
+	* Support Intel Sandy Bridge 22nm desktop/mobile CPU
+SPARC:
+	* Improve the detection of SPARC (Thank Sylvestre Ledru)
+
+====================================================================
+Version 0.2.1
+30-Jun-2012
+common:
+x86/x86-64:
+	* Fixed the SEGFAULT bug about hyper-theading
+	* Support AMD Bulldozer by using GotoBLAS2 AMD Barcelona codes
+
+====================================================================
+Version 0.2.0
+26-Jun-2012
+common:
+	* Removed the limitation (64) of numbers of CPU cores.
+	  Now, it supports 256 cores at max.
+	* Supported clang compiler.
+	* Fixed some build bugs on FreeBSD
+x86/x86-64:
+	* Optimized Level-3 BLAS on Intel Sandy Bridge x86-64 by AVX instructions.
+	  Please use gcc >= 4.6 or clang >=3.1.
+	* Support AMD Bobcat by using GotoBLAS2 AMD Barcelona codes.
+
+====================================================================
+Version 0.1.1
+29-Apr-2012
+common:
+	* Upgraded LAPACK to 3.4.1 version. (Thank Zaheer Chothia)
+	* Supported LAPACKE, a C interface to LAPACKE. (Thank Zaheer Chothia)
+	* Fixed the build bug (MD5 and download) on Mac OSX.
+	* Auto download CUnit 2.1.2-2 from SF.net with UTEST_CHECK=1.
+	* Fxied the compatibility issue for compilers without C99 complex number
+	  (e.g. Visual Studio)
+x86/x86_64:
+	* Auto-detect Intel Sandy Bridge Core i7-3xxx & Xeon E7 Westmere-EX.
+	* Test alpha=Nan in dscale.
+	* Fixed a SEGFAULT bug in samax on x86 windows.
+	
+====================================================================
+Version 0.1.0
+23-Mar-2012
+common:
+	* Set soname of shared library on Linux.
+	* Added LIBNAMESUFFIX flag in Makefile.rule. The user can use 
+	  this flag to control the library name, e.g. libopenblas.a, 
+	  libopenblas_ifort.a or libopenblas_omp.a.
+	* Added GEMM_MULTITHREAD_THRESHOLD flag in Makefile.rule.
+	  The lib use single thread in GEMM function with small matrices.
+x86/x86_64:
+	* Used GEMV SSE/SSE2 kernels on x86 32-bit.
+	* Exported CBLAS functions in Windows DLL.
+MIPS64:
+	* Completed Level-3 BLAS optimization on Loongson 3A CPU.
+	* Improved GEMV performance on Loongson 3A CPU.
+	* Improved Level-3 BLAS performance on Loongson 3B CPU. (EXPERIMENT)
+
 ====================================================================
 Version 0.1 alpha2.5
 19-Feb-2012

GotoBLAS_03FAQ.txt

@@ -90,6 +90,15 @@
        number of threads will consume extra resource. I recommend you to
        specify minimum number of threads.
 
+1.9  Q I have segfaults when I compile with USE_OPENMP=1. What's wrong?
+
+     A This may be related to a bug in the Linux kernel 2.6.32. Try applying
+     the patch segaults.patch using
+
+     patch < segfaults.patch
+
+     and see if the crashes persist. Note that this patch will lead to many
+     compiler warnings.
 
 2. Architecture Specific issue or Implementation
 

LICENSE

@@ -1,4 +1,4 @@
-Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without

Makefile

@@ -3,7 +3,7 @@ include ./Makefile.system
 
 BLASDIRS = interface driver/level2 driver/level3 driver/others
 
-ifndef DYNAMIC_ARCH
+ifneq ($(DYNAMIC_ARCH), 1)
 BLASDIRS += kernel 
 endif
 
@@ -26,7 +26,7 @@ endif
 
 SUBDIRS_ALL = $(SUBDIRS) test ctest utest exports benchmark ../laswp ../bench
 
-.PHONY : all libs netlib test ctest shared install 
+.PHONY : all libs netlib test ctest shared install
 .NOTPARALLEL : all libs prof lapack-test install
 
 all :: libs netlib tests shared
@@ -80,29 +80,30 @@ endif
 	@echo
 
 shared :
+ifndef NO_SHARED
 ifeq ($(OSNAME), Linux)
 	$(MAKE) -C exports so
-	-ln -fs $(LIBSONAME) libopenblas.so
+	-ln -fs $(LIBSONAME) $(LIBPREFIX).so
+	-ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
 ifeq ($(OSNAME), FreeBSD)
 	$(MAKE) -C exports so
-	-ln -fs $(LIBSONAME) libopenblas.so
+	-ln -fs $(LIBSONAME) $(LIBPREFIX).so
 endif
 ifeq ($(OSNAME), NetBSD)
 	$(MAKE) -C exports so
-	-ln -fs $(LIBSONAME) libopenblas.so
+	-ln -fs $(LIBSONAME) $(LIBPREFIX).so
 endif
 ifeq ($(OSNAME), Darwin)
 	$(MAKE) -C exports dyn
-	-ln -fs $(LIBDYNNAME) libopenblas.dylib
+	-ln -fs $(LIBDYNNAME) $(LIBPREFIX).dylib
 endif
 ifeq ($(OSNAME), WINNT)
 	$(MAKE) -C exports dll
-	-ln -fs $(LIBDLLNAME) libopenblas.dll
 endif
 ifeq ($(OSNAME), CYGWIN_NT)
 	$(MAKE) -C exports dll
-	-ln -fs $(LIBDLLNAME) libopenblas.dll
+endif
 endif
 
 tests :
@@ -130,7 +131,7 @@ endif
 ifeq ($(NOFORTRAN), 1)
 	$(error OpenBLAS: Detecting fortran compiler failed. Please install fortran compiler, e.g. gfortran, ifort, openf90.)
 endif
-	-ln -fs $(LIBNAME) libopenblas.$(LIBSUFFIX)
+	-ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 	for d in $(SUBDIRS) ; \
 	do if test -d $$d; then \
 	  $(MAKE) -C $$d $(@F) || exit 1 ; \
@@ -146,7 +147,7 @@ ifeq ($(EXPRECISION), 1)
 	echo "#define EXPRECISION">> config_last.h
 endif
 ## 
-ifdef DYNAMIC_ARCH
+ifeq ($(DYNAMIC_ARCH), 1)
 	  $(MAKE) -C kernel commonlibs || exit 1
 	for d in $(DYNAMIC_CORE) ; \
 	do  $(MAKE) GOTOBLAS_MAKEFILE= -C kernel TARGET_CORE=$$d kernel || exit 1 ;\
@@ -158,18 +159,18 @@ endif
 prof : prof_blas prof_lapack
 
 prof_blas :
-	ln -fs $(LIBNAME_P) libopenblas_p.$(LIBSUFFIX)
+	ln -fs $(LIBNAME_P) $(LIBPREFIX)_p.$(LIBSUFFIX)
 	for d in $(SUBDIRS) ; \
 	do if test -d $$d; then \
 	  $(MAKE) -C $$d prof || exit 1 ; \
 	fi; \
 	done
-ifdef DYNAMIC_ARCH
+ifeq ($(DYNAMIC_ARCH), 1)
 	  $(MAKE) -C kernel commonprof || exit 1
 endif
 
 blas :
-	ln -fs $(LIBNAME) libopenblas.$(LIBSUFFIX)
+	ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 	for d in $(BLASDIRS) ; \
 	do if test -d $$d; then \
 	  $(MAKE) -C $$d libs || exit 1 ; \
@@ -177,13 +178,13 @@ blas :
 	done
 
 hpl : 
-	ln -fs $(LIBNAME) libopenblas.$(LIBSUFFIX)
+	ln -fs $(LIBNAME) $(LIBPREFIX).$(LIBSUFFIX)
 	for d in $(BLASDIRS) ../laswp exports ; \
 	do if test -d $$d; then \
 	  $(MAKE) -C $$d $(@F) || exit 1 ; \
 	fi; \
 	done
-ifdef DYNAMIC_ARCH
+ifeq ($(DYNAMIC_ARCH), 1)
 	  $(MAKE) -C kernel commonlibs || exit 1
 	for d in $(DYNAMIC_CORE) ; \
 	do  $(MAKE) GOTOBLAS_MAKEFILE= -C kernel TARGET_CORE=$$d kernel || exit 1 ;\
@@ -191,7 +192,7 @@ ifdef DYNAMIC_ARCH
 endif
 
 hpl_p :
-	ln -fs $(LIBNAME_P) libopenblas_p.$(LIBSUFFIX)
+	ln -fs $(LIBNAME_P) $(LIBPREFIX)_p.$(LIBSUFFIX)
 	for d in $(SUBDIRS) ../laswp exports ; \
 	do if test -d $$d; then \
 	  $(MAKE) -C $$d $(@F) || exit 1 ; \
@@ -202,47 +203,73 @@ ifeq ($(NO_LAPACK), 1)
 netlib : 
 
 else
-netlib : lapack-3.4.0 patch.for_lapack-3.4.0 lapack-3.4.0/make.inc
+netlib : lapack-3.4.2 patch.for_lapack-3.4.2 $(NETLIB_LAPACK_DIR)/make.inc
 ifndef NOFORTRAN
-	-@$(MAKE) -C lapack-3.4.0 lapacklib
+	-@$(MAKE) -C $(NETLIB_LAPACK_DIR) lapacklib
+endif
+ifndef NO_LAPACKE
+	-@$(MAKE) -C $(NETLIB_LAPACK_DIR) lapackelib
 endif
 endif
 
-prof_lapack : lapack-3.4.0 lapack-3.4.0/make.inc
-	-@$(MAKE) -C lapack-3.4.0 lapack_prof
+prof_lapack : lapack-3.4.2 $(NETLIB_LAPACK_DIR)/make.inc
+	-@$(MAKE) -C $(NETLIB_LAPACK_DIR) lapack_prof
 
-lapack-3.4.0/make.inc :
+$(NETLIB_LAPACK_DIR)/make.inc :
 ifndef NOFORTRAN
-	-@echo "FORTRAN   = $(FC)" > lapack-3.4.0/make.inc
-	-@echo "OPTS      = $(FFLAGS)" >> lapack-3.4.0/make.inc
-	-@echo "POPTS     = $(FPFLAGS)" >> lapack-3.4.0/make.inc
-	-@echo "NOOPT     = $(FFLAGS) -O0" >> lapack-3.4.0/make.inc
-	-@echo "PNOOPT     = $(FPFLAGS) -O0" >> lapack-3.4.0/make.inc
-	-@echo "LOADOPTS  = $(FFLAGS) $(EXTRALIB)" >> lapack-3.4.0/make.inc
-	-@echo "ARCH      = $(AR)" >> lapack-3.4.0/make.inc
-	-@echo "RANLIB    = $(RANLIB)" >> lapack-3.4.0/make.inc
-	-@echo "LAPACKLIB = ../$(LIBNAME)" >> lapack-3.4.0/make.inc
-	-@echo "LAPACKLIB_P = ../$(LIBNAME_P)" >> lapack-3.4.0/make.inc
-	-@echo "SUFFIX     = $(SUFFIX)" >> lapack-3.4.0/make.inc
-	-@echo "PSUFFIX    = $(PSUFFIX)" >> lapack-3.4.0/make.inc
-#	-@echo "CEXTRALIB  = $(CEXTRALIB)" >> lapack-3.4.0/make.inc
-	-@cat  make.inc >> lapack-3.4.0/make.inc
+	-@echo "FORTRAN     = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "OPTS        = $(FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "POPTS       = $(FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "NOOPT       = $(FFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "PNOOPT      = $(FPFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "LOADOPTS    = $(FFLAGS) $(EXTRALIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "CC          = $(CC)" >> $(NETLIB_LAPACK_DIR)/make.inc
+ifdef INTERFACE64
+	-@echo "CFLAGS      = $(CFLAGS) -DHAVE_LAPACK_CONFIG_H  -DLAPACK_ILP64" >> $(NETLIB_LAPACK_DIR)/make.inc
+else
+	-@echo "CFLAGS      = $(CFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+endif
+	-@echo "ARCH        = $(AR)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "ARCHFLAGS   = -ru" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "RANLIB      = $(RANLIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "LAPACKLIB   = ../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "LAPACKELIB  = ../$(LIBNAME)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "LAPACKLIB_P = ../$(LIBNAME_P)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "SUFFIX      = $(SUFFIX)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "PSUFFIX     = $(PSUFFIX)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "CEXTRALIB   = $(EXTRALIB)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@cat  make.inc >> $(NETLIB_LAPACK_DIR)/make.inc
 endif
 
-lapack-3.4.0 : lapack-3.4.0.tgz
+lapack-3.4.2 : lapack-3.4.2.tgz
 ifndef NOFORTRAN
-	@if test `$(MD5SUM) lapack-3.4.0.tgz | $(AWK) '{print $$1}'` = 02d5706ec03ba885fc246e5fa10d8c70; then \
+ifndef NO_LAPACK
+	@if test `$(MD5SUM) lapack-3.4.2.tgz | $(AWK) '{print $$1}'` = 61bf1a8a4469d4bdb7604f5897179478; then \
 		echo $(TAR) zxf $< ;\
-		$(TAR) zxf $< && (cd lapack-3.4.0; $(PATCH) -p1 < ../patch.for_lapack-3.4.0) ;\
+		$(TAR) zxf $< && (cd $(NETLIB_LAPACK_DIR); $(PATCH) -p1 < ../patch.for_lapack-3.4.2) ;\
+		rm -f $(NETLIB_LAPACK_DIR)/lapacke/make.inc ;\
 	else \
-		echo "	lapack-3.4.0.tgz check sum is wrong (Please use orignal)." ;\
-		rm -rf lapack-3.4.0 ;\
+		rm -rf $(NETLIB_LAPACK_DIR) ;\
+		echo "	Cannot download lapack-3.4.2.tgz or the MD5 check sum is wrong (Please use orignal)."; \
+		exit 1; \
 	fi
 endif
+endif
 
-lapack-3.4.0.tgz :
+LAPACK_URL=http://www.netlib.org/lapack/lapack-3.4.2.tgz
+
+lapack-3.4.2.tgz :
 ifndef NOFORTRAN
-	-wget http://www.netlib.org/lapack/lapack-3.4.0.tgz
+#http://stackoverflow.com/questions/7656425/makefile-ifeq-logical-or
+ifeq ($(OSNAME), $(filter $(OSNAME),Darwin NetBSD))
+	curl -O $(LAPACK_URL)
+else
+ifeq ($(OSNAME), FreeBSD)
+	fetch $(LAPACK_URL)
+else
+	wget $(LAPACK_URL)
+endif
+endif
 endif
 
 large.tgz : 
@@ -255,21 +282,21 @@ ifndef NOFORTRAN
 	-wget http://www.netlib.org/lapack/timing/timing.tgz
 endif
 
-lapack-timing : lapack-3.4.0 large.tgz timing.tgz
+lapack-timing : lapack-3.4.2 large.tgz timing.tgz
 ifndef NOFORTRAN
-	(cd lapack-3.4.0; $(TAR) zxf ../timing.tgz TIMING)
-	(cd lapack-3.4.0/TIMING; $(TAR) zxf ../../large.tgz )
-	make -C lapack-3.4.0 tmglib
-	make -C lapack-3.4.0/TIMING
+	(cd $(NETLIB_LAPACK_DIR); $(TAR) zxf ../timing.tgz TIMING)
+	(cd $(NETLIB_LAPACK_DIR)/TIMING; $(TAR) zxf ../../large.tgz )
+	make -C $(NETLIB_LAPACK_DIR) tmglib
+	make -C $(NETLIB_LAPACK_DIR)/TIMING
 endif
 
 
 lapack-test :
-	$(MAKE) -C lapack-3.4.0 tmglib
-	$(MAKE) -C lapack-3.4.0/TESTING xeigtstc xeigtstd xeigtsts xeigtstz xlintstc xlintstd xlintstds xlintsts xlintstz xlintstzc
-	@rm	-f lapack-3.4.0/TESTING/*.out
-	$(MAKE) -j 1 -C lapack-3.4.0/TESTING
-	$(GREP) failed lapack-3.4.0/TESTING/*.out
+	$(MAKE) -C $(NETLIB_LAPACK_DIR) tmglib
+	$(MAKE) -C $(NETLIB_LAPACK_DIR)/TESTING xeigtstc xeigtstd xeigtsts xeigtstz xlintstc xlintstd xlintstds xlintsts xlintstz xlintstzc
+	@rm	-f $(NETLIB_LAPACK_DIR)/TESTING/*.out
+	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING
+	$(GREP) failed $(NETLIB_LAPACK_DIR)/TESTING/*.out
 
 dummy :
 
@@ -285,11 +312,12 @@ clean ::
 #ifdef DYNAMIC_ARCH
 	@$(MAKE) -C kernel clean
 #endif
-	@rm -f *.$(LIBSUFFIX) *.so *~ *.exe getarch getarch_2nd *.dll *.lib *.$(SUFFIX) *.dwf libopenblas.$(LIBSUFFIX) libopenblas_p.$(LIBSUFFIX) *.lnk myconfig.h
+	@$(MAKE) -C reference clean
+	@rm -f *.$(LIBSUFFIX) *.so *~ *.exe getarch getarch_2nd *.dll *.lib *.$(SUFFIX) *.dwf $(LIBPREFIX).$(LIBSUFFIX) $(LIBPREFIX)_p.$(LIBSUFFIX) $(LIBPREFIX).so.$(MAJOR_VERSION) *.lnk myconfig.h
 	@rm -f Makefile.conf config.h Makefile_kernel.conf config_kernel.h st* *.dylib
-	@if test -d lapack-3.4.0; then \
-	echo deleting lapack-3.4.0; \
-	rm -rf lapack-3.4.0 ;\
+	@if test -d $(NETLIB_LAPACK_DIR); then \
+	echo deleting $(NETLIB_LAPACK_DIR); \
+	rm -rf $(NETLIB_LAPACK_DIR) ;\
 	fi
 	@rm -f *.grd Makefile.conf_last config_last.h
 	@echo Done.

Makefile.install

@@ -23,7 +23,7 @@ install : 	lib.grd
 	@cat config_last.h >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
 	@echo \#define VERSION \" OpenBLAS $(VERSION) \" >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
 	@cat openblas_config_template.h >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
-	@echo \#endif >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
+	@echo \#endif  \/\* OPENBLAS_CONFIG_H \*\/ >> $(OPENBLAS_INCLUDE_DIR)/openblas_config.h
 
 	@echo Generating f77blas.h in $(OPENBLAS_INCLUDE_DIR)
 	@echo \#ifndef OPENBLAS_F77BLAS_H > $(OPENBLAS_INCLUDE_DIR)/f77blas.h
@@ -32,39 +32,48 @@ install : 	lib.grd
 	@cat common_interface.h >> $(OPENBLAS_INCLUDE_DIR)/f77blas.h
 	@echo \#endif >> $(OPENBLAS_INCLUDE_DIR)/f77blas.h
 
+ifndef NO_CBLAS
 	@echo Generating cblas.h in $(OPENBLAS_INCLUDE_DIR)
 	@sed 's/common/openblas_config/g' cblas.h > $(OPENBLAS_INCLUDE_DIR)/cblas.h
+endif
+
+ifndef NO_LAPACKE
+	@echo Copying LAPACKE header files to $(OPENBLAS_LIBRARY_DIR)
+	@-cp $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke.h $(OPENBLAS_INCLUDE_DIR)/lapacke.h
+	@-cp $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke_config.h $(OPENBLAS_INCLUDE_DIR)/lapacke_config.h
+	@-cp $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke_mangling_with_flags.h $(OPENBLAS_INCLUDE_DIR)/lapacke_mangling.h
+	@-cp $(NETLIB_LAPACK_DIR)/lapacke/include/lapacke_utils.h $(OPENBLAS_INCLUDE_DIR)/lapacke_utils.h
+endif
 
 #for install static library 
 	@echo Copy the static library to $(OPENBLAS_LIBRARY_DIR)
 	@cp $(LIBNAME) $(OPENBLAS_LIBRARY_DIR)
-	@-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBNAME) $(OPENBLAS_LIBRARY_DIR)/libopenblas.$(LIBSUFFIX)
+	@-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBNAME) $(OPENBLAS_LIBRARY_DIR)/$(LIBPREFIX).$(LIBSUFFIX)
 #for install shared library 
 	@echo Copy the shared library to $(OPENBLAS_LIBRARY_DIR)
 ifeq ($(OSNAME), Linux)
 	-cp $(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)
-	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)/libopenblas.so
+	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)/$(LIBPREFIX).so
+	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)/$(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
 ifeq ($(OSNAME), FreeBSD)
 	-cp $(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)
-	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)/libopenblas.so
+	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)/$(LIBPREFIX).so
 endif
 ifeq ($(OSNAME), NetBSD)
 	-cp $(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)
-	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)/libopenblas.so
+	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBSONAME) $(OPENBLAS_LIBRARY_DIR)/$(LIBPREFIX).so
 endif
 ifeq ($(OSNAME), Darwin)     
 	-cp $(LIBDYNNAME) $(OPENBLAS_LIBRARY_DIR)
 	-install_name_tool -id $(OPENBLAS_LIBRARY_DIR)/$(LIBDYNNAME) $(OPENBLAS_LIBRARY_DIR)/$(LIBDYNNAME)
-	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBDYNNAME) $(OPENBLAS_LIBRARY_DIR)/libopenblas.dylib
+	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBDYNNAME) $(OPENBLAS_LIBRARY_DIR)/$(LIBPREFIX).dylib
 endif
 ifeq ($(OSNAME), WINNT)
 	-cp $(LIBDLLNAME) $(OPENBLAS_LIBRARY_DIR)
-	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBDLLNAME) $(OPENBLAS_LIBRARY_DIR)/libopenblas.dll
 endif
 ifeq ($(OSNAME), CYGWIN_NT)
 	-cp $(LIBDLLNAME) $(OPENBLAS_LIBRARY_DIR)
-	-ln -fs $(OPENBLAS_LIBRARY_DIR)/$(LIBDLLNAME) $(OPENBLAS_LIBRARY_DIR)/libopenblas.dll
 endif
 
 	@echo Install OK!

Makefile.rule

@@ -3,7 +3,12 @@
 #
 
 # This library's version
-VERSION = 0.1alpha2.5
+VERSION = 0.2.5
+
+# If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
+# and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library 
+# is libopenblas_$(LIBNAMESUFFIX).so.0.
+# LIBNAMESUFFIX = omp
 
 # You can specify the target architecture, otherwise it's
 # automatically detected.
@@ -19,10 +24,13 @@ VERSION = 0.1alpha2.5
 # Fortran compiler. Default is g77.
 # FC = gfortran
 
-# Even you can specify cross compiler
+# Even you can specify cross compiler. Meanwhile, please set HOSTCC.
 # CC = x86_64-w64-mingw32-gcc
 # FC = x86_64-w64-mingw32-gfortran
 
+# If you use the cross compiler, please set this host compiler.
+# HOSTCC = gcc
+
 # If you need 32bit binary, define BINARY=32, otherwise define BINARY=64
 # BINARY=64
 
@@ -40,12 +48,19 @@ VERSION = 0.1alpha2.5
 # automatically detected by the the script.
 # NUM_THREADS = 24
 
+# if you don't need generate the shared library, please comment it in.
+# NO_SHARED = 1
+
 # If you don't need CBLAS interface, please comment it in.
 # NO_CBLAS = 1
 
-# If you don't need LAPACK, please comment it in.
+# If you don't need LAPACK, please comment it in. 
+# If you set NO_LAPACK=1, the library automatically sets NO_LAPACKE=1.
 # NO_LAPACK = 1
 
+# If you don't need LAPACKE (C Interface to LAPACK), please comment it in.
+# NO_LAPACKE = 1
+
 # If you want to use legacy threaded Level 3 implementation.
 # USE_SIMPLE_THREADED_LEVEL3 = 1
 
@@ -62,6 +77,10 @@ VERSION = 0.1alpha2.5
 # If you want to disable CPU/Memory affinity on Linux.
 # NO_AFFINITY = 1
 
+# Don't use AVX kernel on Sandy Bridge. It is compatible with old compilers 
+# and OS. However, the performance is low.
+# NO_AVX = 1
+
 # If you would like to know minute performance report of GotoBLAS.
 # FUNCTION_PROFILE = 1
 
@@ -83,6 +102,11 @@ VERSION = 0.1alpha2.5
 # If you need to synchronize FP CSR between threads (for x86/x86_64 only).
 # CONSISTENT_FPCSR = 1
 
+# If any gemm arguement m, n or k is less or equal this threshold, gemm will be execute 
+# with single thread. You can use this flag to avoid the overhead of multi-threading 
+# in small matrix sizes. The default value is 50.
+# GEMM_MULTITHREAD_THRESHOLD = 50
+
 # If you need santy check by comparing reference BLAS. It'll be very
 # slow (Not implemented yet).
 # SANITY_CHECK = 1
@@ -94,19 +118,16 @@ VERSION = 0.1alpha2.5
 # The installation directory.
 # PREFIX = /opt/OpenBLAS
 
-# Common Optimization Flag; -O2 is enough.
-# DEBUG = 1
-
-ifeq ($(DEBUG), 1)
-COMMON_OPT += -g
-# -DDEBUG
-else
-COMMON_OPT += -O2
-endif
+# Common Optimization Flag; 
+# The default -O2 is enough.
+# COMMON_OPT = -O2
 
 # Profiling flags
 COMMON_PROF = -pg
 
+# Build Debug version
+# DEBUG = 1
+
 #
 #  End of user configuration 
 #

Makefile.system

@@ -9,8 +9,20 @@ ifndef TOPDIR
 TOPDIR = .
 endif
 
+ifndef NETLIB_LAPACK_DIR
+NETLIB_LAPACK_DIR = $(TOPDIR)/lapack-3.4.2
+endif
+
 # Default C compiler
+# - Only set if not specified on the command line or inherited from the environment.
+# - CC is an implicit variable so neither '?=' or 'ifndef' can be used.
+#   http://stackoverflow.com/questions/4029274/mingw-and-make-variables
+# - Default value is 'cc' which is not always a valid command (e.g. MinGW).
+ifeq ($(origin CC),default)
 CC = gcc
+endif
+
+# Default Fortran compiler (FC) is selected by f_check.
 
 ifndef MAKEFILE_RULE
 include $(TOPDIR)/Makefile.rule
@@ -40,6 +52,19 @@ ifdef INTERFACE64
 GETARCH_FLAGS	+= -DUSE64BITINT
 endif
 
+ifndef GEMM_MULTITHREAD_THRESHOLD
+GEMM_MULTITHREAD_THRESHOLD=50
+endif
+GETARCH_FLAGS	+= -DGEMM_MULTITHREAD_THRESHOLD=$(GEMM_MULTITHREAD_THRESHOLD) 
+
+ifeq ($(NO_AVX), 1)
+GETARCH_FLAGS	+= -DNO_AVX
+endif
+
+ifeq ($(DEBUG), 1)
+GETARCH_FLAGS	+= -g
+endif
+
 # This operation is expensive, so execution should be once.
 ifndef GOTOBLAS_MAKEFILE
 export GOTOBLAS_MAKEFILE = 1
@@ -96,6 +121,15 @@ DLLWRAP = $(CROSS_SUFFIX)dllwrap
 
 ifeq ($(OSNAME), Darwin)
 export MACOSX_DEPLOYMENT_TARGET=10.2
+MD5SUM = md5 -r
+endif
+
+ifeq ($(OSNAME), FreeBSD)
+MD5SUM = md5 -r
+endif
+
+ifeq ($(OSNAME), NetBSD)
+MD5SUM = md5 -n
 endif
 
 ifeq ($(OSNAME), Linux)
@@ -115,6 +149,26 @@ EXTRALIB        += -defaultlib:advapi32
 SUFFIX  = obj
 PSUFFIX = pobj
 LIBSUFFIX = lib
+ifeq ($(C_COMPILER), GCC)
+#Test for supporting MS_ABI
+GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
+GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
+GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
+ifeq ($(GCCVERSIONGT4), 1)
+# GCC Majar version > 4
+# It is compatible with MSVC ABI. 
+CCOMMON_OPT	+= -DMS_ABI
+endif
+
+ifeq ($(GCCVERSIONGTEQ4), 1)
+ifeq ($(GCCMINORVERSIONGTEQ7), 1)
+# GCC Version >=4.7
+# It is compatible with MSVC ABI. 
+CCOMMON_OPT	+= -DMS_ABI
+endif
+endif
+
+endif
 endif
 
 ifeq ($(OSNAME), Interix)
@@ -218,14 +272,20 @@ endif
 endif
 
 
-ifdef DYNAMIC_ARCH
+ifeq ($(DYNAMIC_ARCH), 1)
 ifeq ($(ARCH), x86)
 DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
-	       CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA ATOM NANO
+	       CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
+ifneq ($(NO_AVX), 1)
+DYNAMIC_CORE += SANDYBRIDGE 
+endif
 endif
 
 ifeq ($(ARCH), x86_64)
-DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA ATOM NANO
+DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
+ifneq ($(NO_AVX), 1)
+DYNAMIC_CORE += SANDYBRIDGE 
+endif
 endif
 
 ifndef DYNAMIC_CORE
@@ -274,7 +334,12 @@ endif
 BINARY_DEFINED = 1
 endif
 
-ifeq ($(CORE), LOONGSON3A)
+ifeq ($(CORE), LOONGSON3A) 
+CCOMMON_OPT += -march=mips64
+FCOMMON_OPT += -march=mips64
+endif
+
+ifeq ($(CORE), LOONGSON3B) 
 CCOMMON_OPT += -march=mips64
 FCOMMON_OPT += -march=mips64
 endif
@@ -341,7 +406,8 @@ endif
 
 ifeq ($(F_COMPILER), GFORTRAN)
 CCOMMON_OPT += -DF_INTERFACE_GFORT
-FCOMMON_OPT += -Wall 
+FCOMMON_OPT += -Wall
+EXTRALIB += -lgfortran 
 ifdef NO_BINARY_MODE
 ifeq ($(ARCH), mips64)
 ifdef BINARY64
@@ -522,14 +588,26 @@ endif
 
 ifeq ($(NO_LAPACK), 1)
 CCOMMON_OPT	+= -DNO_LAPACK
+#Disable LAPACK C interface
+NO_LAPACKE = 1
+endif
+
+ifeq ($(NO_LAPACKE), 1)
+CCOMMON_OPT	+= -DNO_LAPACKE
+endif
+
+ifeq ($(NO_AVX), 1)
+CCOMMON_OPT	+= -DNO_AVX
 endif
 
 ifdef SMP
 CCOMMON_OPT	+= -DSMP_SERVER
 
 ifeq ($(ARCH), mips64)
+ifneq ($(CORE), LOONGSON3B)
 USE_SIMPLE_THREADED_LEVEL3 = 1
 endif
+endif
 
 ifeq ($(USE_OPENMP), 1)
 # USE_SIMPLE_THREADED_LEVEL3 = 1
@@ -568,7 +646,11 @@ ifdef USE_SIMPLE_THREADED_LEVEL3
 CCOMMON_OPT	+= -DUSE_SIMPLE_THREADED_LEVEL3
 endif
 
+ifndef LIBNAMESUFFIX
 LIBPREFIX = libopenblas
+else
+LIBPREFIX = libopenblas_$(LIBNAMESUFFIX)
+endif
 
 KERNELDIR	= $(TOPDIR)/kernel/$(ARCH)
 
@@ -590,9 +672,11 @@ endif
 
 ifneq ($(ARCH), x86_64)
 ifneq ($(ARCH), x86)
+ifneq ($(CORE), LOONGSON3B)
 NO_AFFINITY = 1
 endif
 endif
+endif
 
 ifdef NO_AFFINITY
 CCOMMON_OPT	+= -DNO_AFFINITY
@@ -632,16 +716,30 @@ PATCH	= patch
 GREP	= grep
 endif
 
+ifndef MD5SUM
 MD5SUM	= md5sum
+endif
+
 AWK	= awk
 
 REVISION = -r$(VERSION)
+MAJOR_VERSION = $(word 1,$(subst ., ,$(VERSION)))
 
-CFLAGS     = $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR)
-PFLAGS     = $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR) -DPROFILE $(COMMON_PROF)
+ifeq ($(DEBUG), 1)
+COMMON_OPT += -g
+endif
 
-FFLAGS     = $(COMMON_OPT) $(FCOMMON_OPT)
-FPFLAGS    = $(COMMON_OPT) $(FCOMMON_OPT) $(COMMON_PROF)
+ifndef COMMON_OPT
+COMMON_OPT = -O2
+endif
+
+
+override CFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR)
+override PFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR) -DPROFILE $(COMMON_PROF)
+
+override FFLAGS     += $(COMMON_OPT) $(FCOMMON_OPT)
+override FPFLAGS    += $(COMMON_OPT) $(FCOMMON_OPT) $(COMMON_PROF)
+#MAKEOVERRIDES =
 
 ifndef SUFFIX
 SUFFIX  = o
@@ -655,7 +753,7 @@ ifndef LIBSUFFIX
 LIBSUFFIX = a
 endif
 
-ifndef DYNAMIC_ARCH
+ifneq ($(DYNAMIC_ARCH), 1)
 ifndef SMP
 LIBNAME		= $(LIBPREFIX)_$(LIBCORE)$(REVISION).$(LIBSUFFIX)
 LIBNAME_P	= $(LIBPREFIX)_$(LIBCORE)$(REVISION)_p.$(LIBSUFFIX)
@@ -674,8 +772,8 @@ endif
 endif
 
 
+LIBDLLNAME   = $(LIBPREFIX).dll
 LIBSONAME    = $(LIBNAME:.$(LIBSUFFIX)=.so)
-LIBDLLNAME   = $(LIBNAME:.$(LIBSUFFIX)=.dll)
 LIBDYNNAME   = $(LIBNAME:.$(LIBSUFFIX)=.dylib)
 LIBDEFNAME   = $(LIBNAME:.$(LIBSUFFIX)=.def)
 LIBEXPNAME   = $(LIBNAME:.$(LIBSUFFIX)=.exp)
@@ -720,6 +818,7 @@ export HAVE_SSE4_1
 export HAVE_SSE4_2
 export HAVE_SSE4A
 export HAVE_SSE5
+export HAVE_AVX
 export KERNELDIR
 export FUNCTION_PROFILE
 export TARGET_CORE

Makefile.tail

@@ -22,19 +22,19 @@ BLASOBJS   += $(QBLASOBJS)   $(XBLASOBJS)
 BLASOBJS_P += $(QBLASOBJS_P) $(XBLASOBJS_P)
 endif
 
-$(SBLASOBJS) $(SBLASOBJS_P) : CFLAGS += -UDOUBLE  -UCOMPLEX
-$(DBLASOBJS) $(DBLASOBJS_P) : CFLAGS += -DDOUBLE  -UCOMPLEX
-$(QBLASOBJS) $(QBLASOBJS_P) : CFLAGS += -DXDOUBLE -UCOMPLEX
-$(CBLASOBJS) $(CBLASOBJS_P) : CFLAGS += -UDOUBLE  -DCOMPLEX
-$(ZBLASOBJS) $(ZBLASOBJS_P) : CFLAGS += -DDOUBLE  -DCOMPLEX
-$(XBLASOBJS) $(XBLASOBJS_P) : CFLAGS += -DXDOUBLE -DCOMPLEX
+$(SBLASOBJS) $(SBLASOBJS_P) : override CFLAGS += -UDOUBLE  -UCOMPLEX
+$(DBLASOBJS) $(DBLASOBJS_P) : override CFLAGS += -DDOUBLE  -UCOMPLEX
+$(QBLASOBJS) $(QBLASOBJS_P) : override CFLAGS += -DXDOUBLE -UCOMPLEX
+$(CBLASOBJS) $(CBLASOBJS_P) : override CFLAGS += -UDOUBLE  -DCOMPLEX
+$(ZBLASOBJS) $(ZBLASOBJS_P) : override CFLAGS += -DDOUBLE  -DCOMPLEX
+$(XBLASOBJS) $(XBLASOBJS_P) : override CFLAGS += -DXDOUBLE -DCOMPLEX
 
-$(SBLASOBJS_P) : CFLAGS += -DPROFILE $(COMMON_PROF)
-$(DBLASOBJS_P) : CFLAGS += -DPROFILE $(COMMON_PROF)
-$(QBLASOBJS_P) : CFLAGS += -DPROFILE $(COMMON_PROF)
-$(CBLASOBJS_P) : CFLAGS += -DPROFILE $(COMMON_PROF)
-$(ZBLASOBJS_P) : CFLAGS += -DPROFILE $(COMMON_PROF)
-$(XBLASOBJS_P) : CFLAGS += -DPROFILE $(COMMON_PROF)
+$(SBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
+$(DBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
+$(QBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
+$(CBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
+$(ZBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
+$(XBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
 
 libs    :: $(BLASOBJS) $(COMMONOBJS)
 	$(AR) $(ARFLAGS) -ru $(TOPDIR)/$(LIBNAME) $^

README

@@ -1,82 +0,0 @@
-OpenBLAS Readme
-
-1.Introduction
-OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version. OpenBLAS is an open source project supported by Lab of Parallel Software and Computational Science, ISCAS.(http://www.rdcps.ac.cn)
-
-2.Intallation
-Download from project homepage. http://xianyi.github.com/OpenBLAS/
-Or,
-check out codes from git://github.com/xianyi/OpenBLAS.git
-1)Normal compile
-  (a) type "make" to detect the CPU automatically.
-  or
-  (b) type "make TARGET=xxx" to set target CPU, e.g. "make TARGET=NEHALEM". The full target list is in file TargetList.txt.
-
-2)Cross compile
-Please set CC and FC with the cross toolchains. Then, set HOSTCC with your host C compiler. At last, set TARGET explicitly.
-
-examples:
-On X86 box, compile this library for loongson3a CPU.
-make BINARY=64 CC=mips64el-unknown-linux-gnu-gcc FC=mips64el-unknown-linux-gnu-gfortran HOSTCC=gcc TARGET=LOONGSON3A
-
-3)Debug version
-make DEBUG=1
-
-4)Intall to the directory (Optional)
-e.g.
-make install PREFIX=your_installation_directory
-The default directory is /opt/OpenBLAS
-
-3.Support CPU & OS
-Please read GotoBLAS_01Readme.txt
-
-Additional support CPU:
-x86_64:
-	Intel Xeon 56xx (Westmere) //Used GotoBLAS2 Nehalem codes.
-MIPS64:
-	ICT Loongson 3A //The initial version used GotoBLAS2 MIPS64 kernels. Thus, the performance is not good.
-
-4.Usages
-Link with libopenblas.a or -lopenblas for shared library.
-
-4.1 Set the number of threads with environment variables. for example,
-export OPENBLAS_NUM_THREADS=4
- or
-export GOTO_NUM_THREADS=4
- or 
-export OMP_NUM_THREADS=4
-
-The priorities are OPENBLAS_NUM_THREADS > GOTO_NUM_THREADS > OMP_NUM_THREADS.
-
-If you compile this lib with USE_OPENMP=1, you should set OMP_NUM_THREADS environment variable. OpenBLAS ignores OPENBLAS_NUM_THREADS and GOTO_NUM_THREADS with USE_OPENMP=1.
-
-4.2 Set the number of threads with calling functions. for example,
-void goto_set_num_threads(int num_threads);
-or
-void openblas_set_num_threads(int num_threads);
-
-If you compile this lib with USE_OPENMP=1, you should use the above functions, too.
-
-5.Report Bugs
-Please add a issue in https://github.com/xianyi/OpenBLAS/issues
-
-6.To-Do List:
-Optimization on ICT Loongson 3A CPU
-
-7.Contact
-OpenBLAS users mailing list: http://list.rdcps.ac.cn/mailman/listinfo/openblas
-
-8.ChangeLog
-Please see Changelog.txt to obtain the differences between GotoBLAS2 1.13 BSD version.
-
-9.Known Issues
-* The number of CPUs/Cores should less than or equal to 8*sizeof(unsigned long). On 64 bits, the limit 
-  is 64. On 32 bits, it is 32.
-
-10. Specification of Git Branches
-We used the git branching model in this article (http://nvie.com/posts/a-successful-git-branching-model/). 
-Now, there are 4 branches in github.com.
-  * The master branch. This a main branch to reflect a production-ready state.
-  * The develop branch. This a main branch to reflect a state with the latest delivered development changes for the next release.
-  * The loongson3a branch. This is a feature branch. We develop Loongson3A codes on this branch. We will merge this feature to develop branch in future.
-  * The gh-pages branch. This is for web pages

README.md

@@ -0,0 +1,113 @@
+# OpenBLAS
+
+## Introduction
+OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version. OpenBLAS is an open source project supported by Lab of Parallel Software and Computational Science, ISCAS <http://www.rdcps.ac.cn>.
+
+Please read the documents on OpenBLAS wiki pages <http://github.com/xianyi/OpenBLAS/wiki>.
+
+## Installation
+Download from project homepage. http://xianyi.github.com/OpenBLAS/
+
+Or, check out codes from git://github.com/xianyi/OpenBLAS.git
+### Normal compile
+  * type "make" to detect the CPU automatically.
+  or
+  * type "make TARGET=xxx" to set target CPU, e.g. "make TARGET=NEHALEM". The full target list is in file TargetList.txt.
+
+### Cross compile
+Please set CC and FC with the cross toolchains. Then, set HOSTCC with your host C compiler. At last, set TARGET explicitly.
+
+Examples:
+
+On X86 box, compile this library for loongson3a CPU.
+
+    make BINARY=64 CC=mips64el-unknown-linux-gnu-gcc FC=mips64el-unknown-linux-gnu-gfortran HOSTCC=gcc TARGET=LOONGSON3A
+
+### Debug version
+
+    make DEBUG=1
+
+### Intall to the directory (Optional)
+
+Example:
+
+    make install PREFIX=your_installation_directory
+
+The default directory is /opt/OpenBLAS
+
+## Support CPU & OS
+Please read GotoBLAS_01Readme.txt
+
+### Additional support CPU:
+
+#### x86/x86-64:
+- **Intel Xeon 56xx (Westmere)**: Used GotoBLAS2 Nehalem codes.
+- **Intel Sandy Bridge**: Optimized Level-3 BLAS with AVX on x86-64.
+- **AMD Bobcat**: Used GotoBLAS2 Barcelona codes.
+- **AMD Bulldozer**: Used GotoBLAS2 Barcelona codes.
+
+#### MIPS64:
+- **ICT Loongson 3A**: Optimized Level-3 BLAS and the part of Level-1,2.
+- **ICT Loongson 3B**: Experimental
+
+### Support OS:
+- **GNU/Linux**
+- **MingWin/Windows**: Please read <https://github.com/xianyi/OpenBLAS/wiki/How-to-use-OpenBLAS-in-Microsoft-Visual-Studio>.
+- **Darwin/Mac OS X**: Experimental. Although GotoBLAS2 supports Darwin, we are the beginner on Mac OS X.
+- **FreeBSD**: Supportted by community. We didn't test the library on this OS.
+
+## Usages
+Link with libopenblas.a or -lopenblas for shared library.
+
+### Set the number of threads with environment variables. 
+
+Examples:
+
+    export OPENBLAS_NUM_THREADS=4
+
+ or
+
+    export GOTO_NUM_THREADS=4
+
+ or 
+
+    export OMP_NUM_THREADS=4
+
+The priorities are OPENBLAS_NUM_THREADS > GOTO_NUM_THREADS > OMP_NUM_THREADS.
+
+If you compile this lib with USE_OPENMP=1, you should set OMP_NUM_THREADS environment variable. OpenBLAS ignores OPENBLAS_NUM_THREADS and GOTO_NUM_THREADS with USE_OPENMP=1.
+
+### Set the number of threads on runtime. 
+
+We provided the below functions to controll the number of threads on runtime.
+
+    void goto_set_num_threads(int num_threads);
+
+    void openblas_set_num_threads(int num_threads);
+
+If you compile this lib with USE_OPENMP=1, you should use the above functions, too.
+
+## Report Bugs
+Please add a issue in https://github.com/xianyi/OpenBLAS/issues
+
+## Contact
+OpenBLAS users mailing list: http://list.rdcps.ac.cn/mailman/listinfo/openblas
+
+## ChangeLog
+Please see Changelog.txt to obtain the differences between GotoBLAS2 1.13 BSD version.
+
+## Troubleshooting
+* Please read [Faq](https://github.com/xianyi/OpenBLAS/wiki/Faq) at first.
+* Please use gcc version 4.6 and above to compile Sandy Bridge AVX kernels on Linux/MingW/BSD.
+* Please use Clang version 3.1 and above to compile the library on Sandy Bridge microarchitecture. The Clang 3.0 will generate the wrong AVX binary code.
+* The number of CPUs/Cores should less than or equal to 256. 
+* On Linux, OpenBLAS sets the processor affinity by default. This may cause [the conflict with R parallel](https://stat.ethz.ch/pipermail/r-sig-hpc/2012-April/001348.html). You can build the library with NO_AFFINITY=1.
+* On Loongson 3A. make test would be failed because of pthread_create error. The error code is EAGAIN. However, it will be OK when you run the same testcase on shell. 
+
+## Specification of Git Branches
+We used the git branching model in this article (http://nvie.com/posts/a-successful-git-branching-model/). 
+Now, there are 4 branches in github.com.
+  * The master branch. This a main branch to reflect a production-ready state.
+  * The develop branch. This a main branch to reflect a state with the latest delivered development changes for the next release.
+  * The loongson3a branch. This is a feature branch. We develop Loongson3A codes on this branch. We will merge this feature to develop branch in future.
+  * The gh-pages branch. This is for web pages

TargetList.txt

@@ -18,6 +18,7 @@ CORE2
 PENRYN
 DUNNINGTON
 NEHALEM
+SANDYBRIDGE
 ATOM
 
 b)AMD CPU:
@@ -27,6 +28,7 @@ OPTERON_SSE3
 BARCELONA
 SHANGHAI
 ISTANBUL
+BOBCAT
 
 c)VIA CPU:
 SSE_GENERIC
@@ -47,6 +49,7 @@ CELL
 3.MIPS64 CPU:
 SICORTEX
 LOONGSON3A
+LOONGSON3B
 
 4.IA64 CPU:
 ITANIUM2

c_check

@@ -43,14 +43,14 @@ $compiler = DEC       if ($data =~ /COMPILER_DEC/);
 $compiler = GCC       if ($compiler eq "");
 
 $os = Linux           if ($data =~ /OS_LINUX/);
-$os = FreeBSD         if ($data =~ /OS_FreeBSD/);
-$os = NetBSD          if ($data =~ /OS_NetBSD/);
-$os = Darwin          if ($data =~ /OS_Darwin/);
-$os = SunOS           if ($data =~ /OS_SunOS/);
+$os = FreeBSD         if ($data =~ /OS_FREEBSD/);
+$os = NetBSD          if ($data =~ /OS_NETBSD/);
+$os = Darwin          if ($data =~ /OS_DARWIN/);
+$os = SunOS           if ($data =~ /OS_SUNOS/);
 $os = AIX             if ($data =~ /OS_AIX/);
 $os = osf             if ($data =~ /OS_OSF/);
 $os = WINNT           if ($data =~ /OS_WINNT/);
-$os = CYGWIN_NT       if ($data =~ /OS_CYGWIN/);
+$os = CYGWIN_NT       if ($data =~ /OS_CYGWIN_NT/);
 $os = Interix         if ($data =~ /OS_INTERIX/);
 
 $architecture = x86    if ($data =~ /ARCH_X86/);
@@ -174,6 +174,8 @@ $linker_a = "";
     $link =~ s/\-Y\sP\,/\-Y/g;
     
     @flags = split(/[\s\,\n]/, $link);
+    # remove leading and trailing quotes from each flag.
+    @flags = map {s/^['"]|['"]$//g; $_} @flags;
 
     foreach $flags (@flags) {
 	if (

cblas.h

@@ -9,6 +9,10 @@ extern "C" {
 #include <stddef.h>
 #include "common.h"
 
+/*Set the number of threads on runtime.*/
+void openblas_set_num_threads(int num_threads);
+void goto_set_num_threads(int num_threads);
+
 #define CBLAS_INDEX size_t
 
 enum CBLAS_ORDER     {CblasRowMajor=101, CblasColMajor=102};
@@ -22,15 +26,15 @@ double cblas_dsdot (blasint n, float *x, blasint incx, float *y, blasint incy);
 float  cblas_sdot(blasint n, float  *x, blasint incx, float  *y, blasint incy);
 double cblas_ddot(blasint n, double *x, blasint incx, double *y, blasint incy);
 
-float  _Complex cblas_cdotu(blasint n, float  *x, blasint incx, float  *y, blasint incy);
-float  _Complex cblas_cdotc(blasint n, float  *x, blasint incx, float  *y, blasint incy);
-double _Complex cblas_zdotu(blasint n, double *x, blasint incx, double *y, blasint incy);
-double _Complex cblas_zdotc(blasint n, double *x, blasint incx, double *y, blasint incy);
+openblas_complex_float  cblas_cdotu(blasint n, float  *x, blasint incx, float  *y, blasint incy);
+openblas_complex_float  cblas_cdotc(blasint n, float  *x, blasint incx, float  *y, blasint incy);
+openblas_complex_double cblas_zdotu(blasint n, double *x, blasint incx, double *y, blasint incy);
+openblas_complex_double cblas_zdotc(blasint n, double *x, blasint incx, double *y, blasint incy);
 
-void  cblas_cdotu_sub(blasint n, float  *x, blasint incx, float  *y, blasint incy, float  _Complex *ret);
-void  cblas_cdotc_sub(blasint n, float  *x, blasint incx, float  *y, blasint incy, float  _Complex *ret);
-void  cblas_zdotu_sub(blasint n, double *x, blasint incx, double *y, blasint incy, double _Complex *ret);
-void  cblas_zdotc_sub(blasint n, double *x, blasint incx, double *y, blasint incy, double _Complex *ret);
+void  cblas_cdotu_sub(blasint n, float  *x, blasint incx, float  *y, blasint incy, openblas_complex_float  *ret);
+void  cblas_cdotc_sub(blasint n, float  *x, blasint incx, float  *y, blasint incy, openblas_complex_float  *ret);
+void  cblas_zdotu_sub(blasint n, double *x, blasint incx, double *y, blasint incy, openblas_complex_double *ret);
+void  cblas_zdotc_sub(blasint n, double *x, blasint incx, double *y, blasint incy, openblas_complex_double *ret);
 
 float  cblas_sasum (blasint n, float  *x, blasint incx);
 double cblas_dasum (blasint n, double *x, blasint incx);

common.h

@@ -68,7 +68,7 @@ extern "C" {
 #define SMP
 #endif
 
-#if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_Interix)
+#if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
 #define WINDOWS_ABI
 #define OS_WINDOWS
 
@@ -89,7 +89,7 @@ extern "C" {
 #include <sched.h>
 #endif
 
-#ifdef OS_DARWIN
+#if defined(OS_DARWIN) || defined(OS_FREEBSD) || defined(OS_NETBSD)
 #include <sched.h>
 #endif
 
@@ -351,7 +351,12 @@ typedef int blasint;
 #endif
 
 #define MMAP_ACCESS (PROT_READ | PROT_WRITE)
+
+#ifdef __NetBSD__
+#define MMAP_POLICY (MAP_PRIVATE | MAP_ANON)
+#else
 #define MMAP_POLICY (MAP_PRIVATE | MAP_ANONYMOUS)
+#endif
 
 #include "param.h"
 #include "common_param.h"
@@ -374,6 +379,30 @@ typedef int blasint;
 #endif
 #endif
 
+#ifndef ASSEMBLER
+#ifndef NOINCLUDE
+/* Inclusion of a standard header file is needed for definition of __STDC_*
+   predefined macros with some compilers (e.g. GCC 4.7 on Linux).  This occurs
+   as a side effect of including either <features.h> or <stdc-predef.h>. */
+#include <stdio.h>
+#endif  // NOINCLUDE
+
+/* C99 supports complex floating numbers natively, which GCC also offers as an
+   extension since version 3.0.  If neither are available, use a compatible
+   structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
+#if defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || __GNUC__ >= 3
+  #define OPENBLAS_COMPLEX_C99
+  typedef float _Complex openblas_complex_float;
+  typedef double _Complex openblas_complex_double;
+  typedef xdouble _Complex openblas_complex_xdouble;
+#else
+  #define OPENBLAS_COMPLEX_STRUCT
+  typedef struct { float real, imag; } openblas_complex_float;
+  typedef struct { double real, imag; } openblas_complex_double;
+  typedef struct { xdouble real, imag; } openblas_complex_xdouble;
+#endif
+#endif  // ASSEMBLER
+
 #ifndef IFLUSH
 #define IFLUSH
 #endif

common_interface.h

@@ -45,6 +45,8 @@ extern "C" {
 
 int    BLASFUNC(xerbla)(char *, blasint *info, blasint);
 
+void    openblas_set_num_threads_(int *);
+
 FLOATRET  BLASFUNC(sdot)  (blasint *, float  *, blasint *, float  *, blasint *);
 FLOATRET  BLASFUNC(sdsdot)(blasint *, float  *,        float  *, blasint *, float  *, blasint *);
 
@@ -74,19 +76,19 @@ myxcomplex_t    BLASFUNC(xdotu)  (blasint *, xdouble  *, blasint *, xdouble  *,
 myxcomplex_t    BLASFUNC(xdotc)  (blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
 
 #elif defined RETURN_BY_STACK
-void  BLASFUNC(cdotu)  (float   _Complex *,  blasint *, float  * , blasint *, float  *,  blasint *);
-void  BLASFUNC(cdotc)  (float   _Complex *,  blasint *, float  *,  blasint *, float  *,  blasint *);
-void  BLASFUNC(zdotu)  (double  _Complex *, blasint *, double  *, blasint *, double  *, blasint *);
-void  BLASFUNC(zdotc)  (double  _Complex *, blasint *, double  *, blasint *, double  *, blasint *);
-void  BLASFUNC(xdotu)  (xdouble _Complex *, blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
-void  BLASFUNC(xdotc)  (xdouble _Complex *, blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
+void  BLASFUNC(cdotu)  (openblas_complex_float   *,  blasint *, float  * , blasint *, float  *,  blasint *);
+void  BLASFUNC(cdotc)  (openblas_complex_float   *,  blasint *, float  *,  blasint *, float  *,  blasint *);
+void  BLASFUNC(zdotu)  (openblas_complex_double  *, blasint *, double  *, blasint *, double  *, blasint *);
+void  BLASFUNC(zdotc)  (openblas_complex_double  *, blasint *, double  *, blasint *, double  *, blasint *);
+void  BLASFUNC(xdotu)  (openblas_complex_xdouble *, blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
+void  BLASFUNC(xdotc)  (openblas_complex_xdouble *, blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
 #else
-float   _Complex BLASFUNC(cdotu)  (blasint *, float  *, blasint *, float  *, blasint *);
-float   _Complex BLASFUNC(cdotc)  (blasint *, float  *, blasint *, float  *, blasint *);
-double  _Complex BLASFUNC(zdotu)  (blasint *, double  *, blasint *, double  *, blasint *);
-double  _Complex BLASFUNC(zdotc)  (blasint *, double  *, blasint *, double  *, blasint *);
-xdouble _Complex BLASFUNC(xdotu)  (blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
-xdouble _Complex BLASFUNC(xdotc)  (blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
+openblas_complex_float   BLASFUNC(cdotu)  (blasint *, float  *, blasint *, float  *, blasint *);
+openblas_complex_float   BLASFUNC(cdotc)  (blasint *, float  *, blasint *, float  *, blasint *);
+openblas_complex_double  BLASFUNC(zdotu)  (blasint *, double  *, blasint *, double  *, blasint *);
+openblas_complex_double  BLASFUNC(zdotc)  (blasint *, double  *, blasint *, double  *, blasint *);
+openblas_complex_xdouble BLASFUNC(xdotu)  (blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
+openblas_complex_xdouble BLASFUNC(xdotc)  (blasint *, xdouble  *, blasint *, xdouble  *, blasint *);
 #endif
 
 void    BLASFUNC(saxpy) (blasint *, float  *, float  *, blasint *, float  *, blasint *);
@@ -640,6 +642,8 @@ int BLASFUNC(zgemc)(char *, char *, blasint *, blasint *, blasint *, double *,
 int BLASFUNC(xgemc)(char *, char *, blasint *, blasint *, blasint *, xdouble *,
 	   xdouble *, blasint *, xdouble *, blasint *, xdouble *, blasint *, xdouble *, xdouble *, blasint *);
 
+/* Lapack routines */
+
 int BLASFUNC(sgetf2)(blasint *, blasint *, float  *, blasint *, blasint *, blasint *);
 int BLASFUNC(dgetf2)(blasint *, blasint *, double *, blasint *, blasint *, blasint *);
 int BLASFUNC(qgetf2)(blasint *, blasint *, xdouble *, blasint *, blasint *, blasint *);
@@ -675,6 +679,13 @@ int BLASFUNC(cgesv)(blasint *, blasint *, float  *, blasint *, blasint *, float
 int BLASFUNC(zgesv)(blasint *, blasint *, double *, blasint *, blasint *, double*, blasint *, blasint *);
 int BLASFUNC(xgesv)(blasint *, blasint *, xdouble *, blasint *, blasint *, xdouble*, blasint *, blasint *);
 
+int BLASFUNC(sgesvd)(char *, char *, blasint *, blasint *, float   *, blasint *, float   *, float   *, blasint *, float   *, blasint *, float   *, blasint *, blasint *);
+int BLASFUNC(dgesvd)(char *, char *, blasint *, blasint *, double  *, blasint *, double  *, double  *, blasint *, double  *, blasint *, double  *, blasint *, blasint *);
+int BLASFUNC(qgesvd)(char *, char *, blasint *, blasint *, xdouble *, blasint *, xdouble *, xdouble *, blasint *, xdouble *, blasint *, xdouble *, blasint *, blasint *);
+int BLASFUNC(cgesvd)(char *, char *, blasint *, blasint *, float   *, blasint *, float   *, float   *, blasint *, float   *, blasint *, float   *, blasint *, blasint *);
+int BLASFUNC(zgesvd)(char *, char *, blasint *, blasint *, double  *, blasint *, double  *, double  *, blasint *, double  *, blasint *, double  *, blasint *, blasint *);
+int BLASFUNC(xgesvd)(char *, char *, blasint *, blasint *, xdouble *, blasint *, xdouble *, xdouble *, blasint *, xdouble *, blasint *, xdouble *, blasint *, blasint *);
+
 int BLASFUNC(spotf2)(char *, blasint *, float  *, blasint *, blasint *);
 int BLASFUNC(dpotf2)(char *, blasint *, double *, blasint *, blasint *);
 int BLASFUNC(qpotf2)(char *, blasint *, xdouble *, blasint *, blasint *);
@@ -689,6 +700,13 @@ int BLASFUNC(cpotrf)(char *, blasint *, float  *, blasint *, blasint *);
 int BLASFUNC(zpotrf)(char *, blasint *, double *, blasint *, blasint *);
 int BLASFUNC(xpotrf)(char *, blasint *, xdouble *, blasint *, blasint *);
 
+int BLASFUNC(spotrs)(char *, blasint *, blasint *, float   *, blasint *, float   *, blasint *, blasint *);
+int BLASFUNC(dpotrs)(char *, blasint *, blasint *, double  *, blasint *, double  *, blasint *, blasint *);
+int BLASFUNC(qpotrs)(char *, blasint *, blasint *, xdouble *, blasint *, xdouble *, blasint *, blasint *);
+int BLASFUNC(cpotrs)(char *, blasint *, blasint *, float   *, blasint *, float   *, blasint *, blasint *);
+int BLASFUNC(zpotrs)(char *, blasint *, blasint *, double  *, blasint *, double  *, blasint *, blasint *);
+int BLASFUNC(xpotrs)(char *, blasint *, blasint *, xdouble *, blasint *, xdouble *, blasint *, blasint *);
+
 int BLASFUNC(slauu2)(char *, blasint *, float  *, blasint *, blasint *);
 int BLASFUNC(dlauu2)(char *, blasint *, double *, blasint *, blasint *);
 int BLASFUNC(qlauu2)(char *, blasint *, xdouble *, blasint *, blasint *);

common_linux.h

@@ -68,9 +68,17 @@ extern long int syscall (long int __sysno, ...);
 static inline int my_mbind(void *addr, unsigned long len, int mode,
 			   unsigned long *nodemask, unsigned long maxnode,
 			   unsigned flags) {
+#if defined (LOONGSON3B) 
+#if defined (__64BIT__)
+	return syscall(SYS_mbind, addr, len, mode, nodemask, maxnode, flags);
+#else
+	return 0; //NULL Implementation on Loongson 3B 32bit.
+#endif
+#else
 //Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34
-	unsigned long null_nodemask=0;
-	return syscall(SYS_mbind, addr, len, mode, &null_nodemask, maxnode, flags);
+//	unsigned long null_nodemask=0;
+	return syscall(SYS_mbind, addr, len, mode, nodemask, maxnode, flags);
+#endif
 }
 
 static inline int my_set_mempolicy(int mode, const unsigned long *addr, unsigned long flag) {
@@ -78,7 +86,13 @@ static inline int my_set_mempolicy(int mode, const unsigned long *addr, unsigned
   return syscall(SYS_set_mempolicy, mode, addr, flag);
 }
 
-static inline int my_gettid(void) { return syscall(SYS_gettid); }
+static inline int my_gettid(void) { 
+#ifdef SYS_gettid
+return syscall(SYS_gettid); 
+#else
+return getpid();
+#endif
+}
 
 #endif
 #endif

common_macro.h

@@ -2127,7 +2127,9 @@
 #endif
 
 #ifndef ASSEMBLER
-#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64)
+#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64)
+extern BLASLONG gemm_offset_a;
+extern BLASLONG gemm_offset_b;
 extern BLASLONG sgemm_p;
 extern BLASLONG sgemm_q;
 extern BLASLONG sgemm_r;

common_mips64.h

@@ -101,10 +101,15 @@ static void INLINE blas_lock(volatile unsigned long *address){
 
 static inline unsigned int rpcc(void){
   unsigned long ret;
-#if defined(LOONGSON3A)
-  unsigned long long tmp;
-  __asm__ __volatile__("dmfc0 %0, $25, 1": "=r"(tmp):: "memory");
-  ret=tmp;
+#if defined(LOONGSON3A) || defined(LOONGSON3B)
+  //  unsigned long long tmp;
+  //__asm__ __volatile__("dmfc0 %0, $25, 1": "=r"(tmp):: "memory");
+  //ret=tmp;
+  __asm__ __volatile__(".set push \n"
+                       ".set mips32r2\n"
+                       "rdhwr %0, $2\n"
+                       ".set pop": "=r"(ret):: "memory");
+
 #else
   __asm__ __volatile__(".set   push    \n"                                     
           ".set   mips32r2\n"                                                  
@@ -114,6 +119,21 @@ static inline unsigned int rpcc(void){
   return ret;
 }
 
+#if defined(LOONGSON3A) || defined(LOONGSON3B)
+#ifndef NO_AFFINITY
+#define WHEREAMI
+static inline int WhereAmI(void){
+  int ret=0;
+  __asm__ __volatile__(".set push \n"
+                       ".set mips32r2\n"
+                       "rdhwr %0, $0\n"
+                       ".set pop": "=r"(ret):: "memory");
+  return ret;
+
+}
+#endif
+#endif
+
 static inline int blas_quickdivide(blasint x, blasint y){
   return x / y;
 }
@@ -152,6 +172,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define CMPEQ	c.eq.d
 #define CMPLE	c.le.d
 #define CMPLT	c.lt.d
+#define	NEG	neg.d
 #else
 #define LD	lwc1
 #define ST	swc1
@@ -170,6 +191,14 @@ static inline int blas_quickdivide(blasint x, blasint y){
 #define CMPEQ	c.eq.s
 #define CMPLE	c.le.s
 #define CMPLT	c.lt.s
+#define PLU     plu.ps                                                    
+#define PLL     pll.ps   
+#define PUU     puu.ps    
+#define PUL     pul.ps   
+#define MADPS   madd.ps   
+#define CVTU    cvt.s.pu    
+#define CVTL    cvt.s.pl 
+#define	NEG	neg.s
 #endif
 
 #if   defined(__64BIT__) &&  defined(USE64BITINT)
@@ -218,13 +247,18 @@ REALNAME: ;\
 
 #define SEEK_ADDRESS
 
-#define BUFFER_SIZE     ( 8 << 20)
+#define BUFFER_SIZE     ( 32 << 20)
 
 #if defined(LOONGSON3A)
 #define PAGESIZE	(16UL << 10)
 #define FIXED_PAGESIZE	(16UL << 10)
 #endif
 
+#if defined(LOONGSON3B)
+#define PAGESIZE	(32UL << 10)
+#define FIXED_PAGESIZE	(32UL << 10)
+#endif
+
 #ifndef PAGESIZE
 #define PAGESIZE	(64UL << 10)
 #endif
@@ -236,7 +270,7 @@ REALNAME: ;\
 #define MAP_ANONYMOUS MAP_ANON
 #endif
 
-#if defined(LOONGSON3A)
+#if defined(LOONGSON3A) || defined(LOONGSON3B)
 #define PREFETCHD_(x) ld $0, x
 #define PREFETCHD(x)  PREFETCHD_(x)  
 #else

common_reference.h

@@ -63,5 +63,7 @@ double  _Complex BLASFUNC_REF(zdotc)  (blasint *, double  *, blasint *, double
 void BLASFUNC_REF(drotmg)(double *, double *, double *, double *, double *);
 
 double BLASFUNC_REF(dsdot)(blasint *, float *, blasint *, float *, blasint*);
+
+FLOATRET  BLASFUNC_REF(samax) (blasint *, float  *, blasint *);
 												   
 #endif

common_thread.h

@@ -135,7 +135,7 @@ static __inline int num_cpu_avail(int level) {
 	int openmp_nthreads=0;
 #endif
 
-  if ((blas_cpu_number == 1) 
+  if (blas_cpu_number == 1
 
 #ifdef USE_OPENMP
       || omp_in_parallel()

common_x86.h

@@ -254,7 +254,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #define PROFCODE
 #endif
 
-#if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INERIX)
+#if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
 #define SAVEREGISTERS \
 	subl	$32, %esp;\
 	movups	%xmm6,    0(%esp);\
@@ -269,7 +269,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #define RESTOREREGISTERS
 #endif
 
-#if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INERIX)
+#if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
 #define PROLOGUE \
 	.text; \
 	.align 16; \
@@ -282,7 +282,7 @@ REALNAME:
 #define EPILOGUE .end	 REALNAME
 #endif
 
-#if defined(OS_LINUX) || defined(OS_FreeBSD) || defined(OS_NetBSD) || defined(__ELF__)
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__)
 #define PROLOGUE \
 	.text; \
 	.align 16; \
@@ -356,4 +356,11 @@ REALNAME:
 
 #ifndef ALIGN_6
 #define ALIGN_6 .align 64
+
+// ffreep %st(0). 
+// Because Clang didn't support ffreep, we directly use the opcode.
+// Please check out http://www.sandpile.org/x86/opc_fpu.htm 
+#ifndef ffreep
+#define ffreep .byte 0xdf, 0xc0 #
+#endif
 #endif

common_x86_64.h

@@ -353,7 +353,7 @@ REALNAME:
 #define EPILOGUE .end	 REALNAME
 #endif
 
-#if defined(OS_LINUX) || defined(OS_FreeBSD) || defined(OS_NetBSD) || defined(__ELF__) || defined(C_PGI)
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__) || defined(C_PGI)
 #define PROLOGUE \
 	.text; \
 	.align 512; \
@@ -425,6 +425,7 @@ REALNAME:
 #define ALIGN_2 .align 2
 #define ALIGN_3 .align 3
 #define ALIGN_4 .align 4
+#define ALIGN_5 .align 5
 #define ffreep	fstp
 #endif
 
@@ -448,4 +449,10 @@ REALNAME:
 #define ALIGN_6 .align 64
 #endif
 
+// ffreep %st(0). 
+// Because Clang didn't support ffreep, we directly use the opcode.
+// Please check out http://www.sandpile.org/x86/opc_fpu.htm 
+#ifndef ffreep
+#define ffreep .byte 0xdf, 0xc0 #
+#endif
 #endif

cpuid.h

@@ -103,6 +103,9 @@
 #define CORE_NEHALEM	17
 #define CORE_ATOM	18
 #define CORE_NANO	19
+#define CORE_SANDYBRIDGE 20
+#define CORE_BOBCAT     21
+#define CORE_BULLDOZER  22
 
 #define HAVE_SSE      (1 <<  0)
 #define HAVE_SSE2     (1 <<  1)
@@ -122,6 +125,7 @@
 #define HAVE_MISALIGNSSE (1 << 15)
 #define HAVE_128BITFPU   (1 << 16)
 #define HAVE_FASTMOVU    (1 << 17)
+#define HAVE_AVX     (1 <<  18)
 
 #define CACHE_INFO_L1_I     1
 #define CACHE_INFO_L1_D     2
@@ -188,4 +192,7 @@ typedef struct {
 #define CPUTYPE_NSGEODE			41
 #define CPUTYPE_VIAC3			42
 #define CPUTYPE_NANO			43
+#define CPUTYPE_SANDYBRIDGE             44
+#define CPUTYPE_BOBCAT                  45
+#define CPUTYPE_BULLDOZER               46
 #endif

cpuid_mips.c

@@ -1,5 +1,5 @@
 /*****************************************************************************
-Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -72,11 +72,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CPU_UNKNOWN     0
 #define CPU_SICORTEX    1
 #define CPU_LOONGSON3A  2
+#define CPU_LOONGSON3B  3
 
 static char *cpuname[] = {
   "UNKOWN",
   "SICORTEX",
-  "LOONGSON3A"
+  "LOONGSON3A",
+  "LOONGSON3B"
 };
 
 int detect(void){
@@ -99,10 +101,14 @@ int detect(void){
 
   fclose(infile);
 
+  if(p != NULL){
   if (strstr(p, "Loongson-3A")){
     return CPU_LOONGSON3A;
+  }else if(strstr(p, "Loongson-3B")){
+    return CPU_LOONGSON3B;
   }else if (strstr(p, "Loongson-3")){
     infile = fopen("/proc/cpuinfo", "r");
+    p = (char *)NULL;
     while (fgets(buffer, sizeof(buffer), infile)){
       if (!strncmp("system type", buffer, 11)){
 	p = strchr(buffer, ':') + 2;
@@ -115,6 +121,24 @@ int detect(void){
   }else{
     return CPU_SICORTEX;
   }
+  }
+  //Check model name for Loongson3
+  infile = fopen("/proc/cpuinfo", "r");
+  p = (char *)NULL;
+  while (fgets(buffer, sizeof(buffer), infile)){
+    if (!strncmp("model name", buffer, 10)){
+      p = strchr(buffer, ':') + 2;
+      break;
+    }
+  }
+  fclose(infile);
+  if(p != NULL){
+  if (strstr(p, "Loongson-3A")){
+    return CPU_LOONGSON3A;
+  }else if(strstr(p, "Loongson-3B")){
+    return CPU_LOONGSON3B;
+  }
+  }
 #endif
     return CPU_UNKNOWN;
 }
@@ -130,6 +154,8 @@ void get_architecture(void){
 void get_subarchitecture(void){
   if(detect()==CPU_LOONGSON3A) {
     printf("LOONGSON3A");
+  }else if(detect()==CPU_LOONGSON3B){
+    printf("LOONGSON3B");
   }else{
     printf("SICORTEX");
   }
@@ -149,6 +175,15 @@ void get_cpuconfig(void){
     printf("#define DTB_DEFAULT_ENTRIES 64\n");
     printf("#define DTB_SIZE 4096\n");
     printf("#define L2_ASSOCIATIVE 4\n");
+  }else if(detect()==CPU_LOONGSON3B){
+    printf("#define LOONGSON3B\n");
+    printf("#define L1_DATA_SIZE 65536\n");
+    printf("#define L1_DATA_LINESIZE 32\n");
+    printf("#define L2_SIZE 512488\n");
+    printf("#define L2_LINESIZE 32\n");
+    printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 4\n");
   }else{
     printf("#define SICORTEX\n");
     printf("#define L1_DATA_SIZE 32768\n");
@@ -164,6 +199,8 @@ void get_cpuconfig(void){
 void get_libname(void){
   if(detect()==CPU_LOONGSON3A) {
     printf("loongson3a\n");
+  }else if(detect()==CPU_LOONGSON3B) {
+    printf("loongson3b\n");
   }else{
 #ifdef __mips64
   printf("mips64\n");

cpuid_x86.c

@@ -40,6 +40,11 @@
 #include <string.h>
 #include "cpuid.h"
 
+#ifdef NO_AVX
+#define CPUTYPE_SANDYBRIDGE CPUTYPE_NEHALEM
+#define CORE_SANDYBRIDGE CORE_NEHALEM
+#endif
+
 #ifndef CPUIDEMU
 
 #if defined(__APPLE__) && defined(__i386__)
@@ -109,6 +114,32 @@ static inline int have_excpuid(void){
   return eax & 0xffff;
 }
 
+#ifndef NO_AVX
+static inline void xgetbv(int op, int * eax, int * edx){
+  __asm__ __volatile__
+    ("xgetbv": "=a" (*eax), "=d" (*edx) : "c" (op) : "cc");
+}
+#endif
+
+int support_avx(){
+#ifndef NO_AVX
+  int eax, ebx, ecx, edx;
+  int ret=0;
+  
+  cpuid(1, &eax, &ebx, &ecx, &edx);
+  if ((ecx & (1 << 28)) != 0 && (ecx & (1 << 27)) != 0){
+    xgetbv(0, &eax, &edx);
+    if((eax & 6) == 6){
+      ret=1;  //OS support AVX
+    }
+  }
+  return ret;
+#else
+  return 0;
+#endif
+}
+
+
 int get_vendor(void){
   int eax, ebx, ecx, edx;
   char vendor[13];
@@ -189,6 +220,9 @@ int get_cputype(int gettype){
     if ((ecx & (1 <<  9)) != 0) feature |= HAVE_SSSE3;
     if ((ecx & (1 << 19)) != 0) feature |= HAVE_SSE4_1;
     if ((ecx & (1 << 20)) != 0) feature |= HAVE_SSE4_2;
+#ifndef NO_AVX
+    if (support_avx()) feature |= HAVE_AVX;
+#endif
 
     if (have_excpuid() >= 0x01) {
       cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
@@ -974,21 +1008,42 @@ int get_cpuname(void){
 	  return CPUTYPE_DUNNINGTON;
 	}
 	break;
-	  case  2:
-		  switch (model) {
-		  case 5:
-			  //Intel Core (Clarkdale) / Core (Arrandale) 
-			  // Pentium (Clarkdale) / Pentium Mobile (Arrandale) 
-			  // Xeon (Clarkdale), 32nm
-			  return CPUTYPE_NEHALEM;
-		  case 10:
-                          //Intel Core i5-2000 /i7-2000 (Sandy Bridge)
-                          return CPUTYPE_NEHALEM;
-		  case 12:
-			  //Xeon Processor 5600 (Westmere-EP)
-			  return CPUTYPE_NEHALEM;
-		  }
-		  break;
+      case  2:
+	switch (model) {
+	case 5:
+	  //Intel Core (Clarkdale) / Core (Arrandale) 
+	  // Pentium (Clarkdale) / Pentium Mobile (Arrandale) 
+	  // Xeon (Clarkdale), 32nm
+	  return CPUTYPE_NEHALEM;
+	case 10:
+	  //Intel Core i5-2000 /i7-2000 (Sandy Bridge)
+	  if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	    return CPUTYPE_NEHALEM; //OS doesn't support AVX
+	case 12:
+	  //Xeon Processor 5600 (Westmere-EP)
+	  return CPUTYPE_NEHALEM;
+	case 13:
+	  //Intel Core i7-3000 / Xeon E5 (Sandy Bridge)
+	  if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	    return CPUTYPE_NEHALEM;
+	case 15:
+	  //Xeon Processor E7 (Westmere-EX)
+	  return CPUTYPE_NEHALEM;
+	}
+	break;
+      case 3:
+	switch (model) {
+	case 10:
+	  if(support_avx())
+	    return CPUTYPE_SANDYBRIDGE;
+	  else
+	    return CPUTYPE_NEHALEM;
+	}
+	break;
       }
       break;
     case 0x7:
@@ -1020,7 +1075,10 @@ int get_cpuname(void){
 	return CPUTYPE_OPTERON;
       case  1:
       case 10:
+      case  6:   //AMD Bulldozer Opteron 6200 / Opteron 4200 / AMD FX-Series
 	return CPUTYPE_BARCELONA;
+      case  5:
+	return CPUTYPE_BOBCAT;
       }
       break;
     }
@@ -1140,6 +1198,9 @@ static char *cpuname[] = {
   "NSGEODE",
   "VIAC3",
   "NANO",
+  "SANDYBRIDGE",
+  "BOBCAT",
+  "BULLDOZER",
 };
 
 static char *lowercpuname[] = {
@@ -1186,6 +1247,9 @@ static char *lowercpuname[] = {
   "tms3x00",
   "nsgeode",
   "nano",
+  "sandybridge",
+  "bobcat",
+  "bulldozer",
 };
 
 static char *corename[] = {
@@ -1209,6 +1273,9 @@ static char *corename[] = {
   "NEHALEM",
   "ATOM",
   "NANO",
+  "SANDYBRIDGE",
+  "BOBCAT",
+  "BULLDOZER",
 };
 
 static char *corename_lower[] = {
@@ -1232,6 +1299,9 @@ static char *corename_lower[] = {
   "nehalem",
   "atom",
   "nano",
+  "sandybridge",
+  "bobcat",
+  "bulldozer",
 };
 
 
@@ -1315,10 +1385,31 @@ int get_coretype(void){
 	  return CORE_NEHALEM;
 	case 10:
           //Intel Core i5-2000 /i7-2000 (Sandy Bridge)
-          return CORE_NEHALEM;
+	  if(support_avx())
+	    return CORE_SANDYBRIDGE;
+	  else
+	    return CORE_NEHALEM; //OS doesn't support AVX
 	case 12:
 	  //Xeon Processor 5600 (Westmere-EP)
 	  return CORE_NEHALEM;
+	case 13:
+          //Intel Core i7-3000 / Xeon E5 (Sandy Bridge)
+	  if(support_avx())
+	    return CORE_SANDYBRIDGE;
+	  else
+	    return CORE_NEHALEM; //OS doesn't support AVX
+	case 15:
+	  //Xeon Processor E7 (Westmere-EX)
+	  return CORE_NEHALEM;
+	}
+	break;
+      case 3:
+	switch (model) {
+	case 10:
+	  if(support_avx())
+	    return CORE_SANDYBRIDGE;
+	  else
+	    return CORE_NEHALEM; //OS doesn't support AVX
 	}
 	break;
       }
@@ -1334,7 +1425,10 @@ int get_coretype(void){
     if (family <= 0x5) return CORE_80486;
     if (family <= 0xe) return CORE_ATHLON;
     if (family == 0xf){
-      if ((exfamily == 0) || (exfamily == 2)) return CORE_OPTERON; else return CORE_BARCELONA;
+      if ((exfamily == 0) || (exfamily == 2)) return CORE_OPTERON; 
+      else if (exfamily == 5) return CORE_BOBCAT; 
+      else if (exfamily == 6) return CORE_BARCELONA;  //AMD Bulldozer Opteron 6200 / Opteron 4200 / AMD FX-Series
+      else return CORE_BARCELONA;
     }
   }
 
@@ -1414,6 +1508,7 @@ void get_cpuconfig(void){
     if (features & HAVE_SSE4_2)   printf("#define HAVE_SSE4_2\n");
     if (features & HAVE_SSE4A)   printf("#define HAVE_SSE4A\n");
     if (features & HAVE_SSE5 )   printf("#define HAVE_SSSE5\n");
+    if (features & HAVE_AVX )    printf("#define HAVE_AVX\n");
     if (features & HAVE_3DNOWEX) printf("#define HAVE_3DNOWEX\n");
     if (features & HAVE_3DNOW)   printf("#define HAVE_3DNOW\n");
     if (features & HAVE_CFLUSH)  printf("#define HAVE_CFLUSH\n");
@@ -1479,6 +1574,7 @@ void get_sse(void){
   if (features & HAVE_SSE4_2)   printf("HAVE_SSE4_2=1\n");
   if (features & HAVE_SSE4A)   printf("HAVE_SSE4A=1\n");
   if (features & HAVE_SSE5 )   printf("HAVE_SSSE5=1\n");
+  if (features & HAVE_AVX )    printf("HAVE_AVX=1\n");
   if (features & HAVE_3DNOWEX) printf("HAVE_3DNOWEX=1\n");
   if (features & HAVE_3DNOW)   printf("HAVE_3DNOW=1\n");
 

ctest.c

@@ -34,20 +34,20 @@ COMPILER_GNU
 OS_LINUX
 #endif
 
-#if defined(__FreeBSD__)
-OS_FreeBSD
+#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
+OS_FREEBSD
 #endif
 
 #if defined(__NetBSD__)
-OS_NetBSD
+OS_NETBSD
 #endif
 
 #if defined(__sun)
-OS_SunOS
+OS_SUNOS
 #endif
 
 #if defined(__APPLE__)
-OS_Darwin
+OS_DARWIN
 #endif
 
 #if defined(_AIX)
@@ -63,13 +63,18 @@ OS_WINNT
 #endif
 
 #if defined(__CYGWIN__)
-OS_CYGWIN
+OS_CYGWIN_NT
 #endif
 
 #if defined(__INTERIX)
 OS_INTERIX
 #endif
 
+#if defined(__gnu_hurd__)
+/* Hurd is very similar to GNU/Linux, it should work out of the box */
+OS_LINUX
+#endif
+
 #if defined(__i386) || defined(_X86)
 ARCH_X86
 #endif

ctest/Makefile

@@ -5,7 +5,7 @@
 TOPDIR = ..
 include $(TOPDIR)/Makefile.system
 
-CFLAGS += -DADD$(BU) -DCBLAS
+override CFLAGS += -DADD$(BU) -DCBLAS
 
 LIB = $(TOPDIR)/$(LIBNAME)
 

driver/level3/gemm_thread_mn.c

@@ -77,8 +77,8 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
     range_M[0] = 0;
     i          = arg -> m;
   } else {
-    range_M[0] = range_M[0];
-    i          = range_M[1] - range_M[0];
+    range_M[0] = range_m[0];
+    i          = range_m[1] - range_m[0];
   }
 
   num_cpu_m  = 0;

driver/level3/gemm_thread_n.c

@@ -71,16 +71,25 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
     queue[num_cpu].args    = arg;
     queue[num_cpu].range_m = range_m;
     queue[num_cpu].range_n = &range[num_cpu];
-    queue[num_cpu].sa      = NULL;
+#if defined(LOONGSON3A)
+    queue[num_cpu].sa      = sa	+ GEMM_OFFSET_A1 * num_cpu;
+    queue[num_cpu].sb      = queue[num_cpu].sa + GEMM_OFFSET_A1 * 5;
+#else
+	queue[num_cpu].sa      = NULL;
     queue[num_cpu].sb      = NULL;
+#endif
     queue[num_cpu].next    = &queue[num_cpu + 1];
     num_cpu ++;
   }
   
   if (num_cpu) {
+#if defined(LOONGSON3A)
     queue[0].sa = sa;
-    queue[0].sb = sb;
-
+    queue[0].sb = sa + GEMM_OFFSET_A1 * 5;
+#else
+	queue[0].sa = sa;
+	queue[0].sb = sb;
+#endif
     queue[num_cpu - 1].next = NULL;
     
     exec_blas(num_cpu,

driver/level3/gemm_thread_variable.c

@@ -55,8 +55,8 @@ int CNAME(int mode,
     range_M[0] = 0;
     i          = arg -> m;
   } else {
-    range_M[0] = range_M[0];
-    i          = range_M[1] - range_M[0];
+    range_M[0] = range_m[0];
+    i          = range_m[1] - range_m[0];
   }
 
   num_cpu_m  = 0;

driver/others/Makefile

@@ -1,12 +1,12 @@
 TOPDIR	= ../..
 include ../../Makefile.system
 
-COMMONOBJS	 = memory.$(SUFFIX) xerbla.$(SUFFIX) c_abs.$(SUFFIX) z_abs.$(SUFFIX)
+COMMONOBJS	 = memory.$(SUFFIX) xerbla.$(SUFFIX) c_abs.$(SUFFIX) z_abs.$(SUFFIX) openblas_set_num_threads.$(SUFFIX)
 
 COMMONOBJS	+= slamch.$(SUFFIX) slamc3.$(SUFFIX) dlamch.$(SUFFIX)  dlamc3.$(SUFFIX)
 
 ifdef SMP
-COMMONOBJS	+= blas_server.$(SUFFIX) divtable.$(SUFFIX) blasL1thread.$(SUFFIX) openblas_set_num_threads.$(SUFFIX)
+COMMONOBJS	+= blas_server.$(SUFFIX) divtable.$(SUFFIX) blasL1thread.$(SUFFIX) 
 ifndef NO_AFFINITY
 COMMONOBJS	+= init.$(SUFFIX) 
 endif
@@ -14,7 +14,7 @@ endif
 
 # COMMONOBJS	+=  info.$(SUFFIX)
 
-ifdef DYNAMIC_ARCH
+ifeq ($(DYNAMIC_ARCH), 1)
 COMMONOBJS	+=  dynamic.$(SUFFIX)
 else
 COMMONOBJS	+=  parameter.$(SUFFIX)
@@ -70,7 +70,7 @@ ifndef BLAS_SERVER
 BLAS_SERVER = blas_server.c
 endif
 
-ifdef DYNAMIC_ARCH
+ifeq ($(DYNAMIC_ARCH), 1)
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) dynamic.$(SUFFIX)
 else
 HPLOBJS = memory.$(SUFFIX) xerbla.$(SUFFIX) parameter.$(SUFFIX)
@@ -215,7 +215,7 @@ info.$(SUFFIX) : info.c info.h ../../common.h ../../param.h
 	$(CC) $(CFLAGS) -c $< -o $(@F)
 
 
-hpl   : CFLAGS += -DHPL
-hpl_p : CFLAGS += -DHPL
+hpl   : override CFLAGS += -DHPL
+hpl_p : override CFLAGS += -DHPL
 
 include $(TOPDIR)/Makefile.tail

driver/others/blas_server.c

@@ -435,7 +435,7 @@ static int blas_thread_server(void *arg){
 
   blas_memory_free(buffer);
 
-  pthread_exit(NULL);
+  //pthread_exit(NULL);
 
   return 0;
 }
@@ -500,6 +500,7 @@ static int blas_monitor(void *arg){
 /* Initializing routine */
 int blas_thread_init(void){
   BLASLONG i;
+  int ret;
 #ifdef NEED_STACKATTR
   pthread_attr_t attr;
 #endif
@@ -545,12 +546,16 @@ int blas_thread_init(void){
       pthread_cond_init (&thread_status[i].wakeup, NULL);
       
 #ifdef NEED_STACKATTR
-      pthread_create(&blas_threads[i], &attr, 
+      ret=pthread_create(&blas_threads[i], &attr, 
 		     (void *)&blas_thread_server, (void *)i);
 #else
-      pthread_create(&blas_threads[i], NULL, 
+      ret=pthread_create(&blas_threads[i], NULL, 
 		     (void *)&blas_thread_server, (void *)i);
 #endif
+      if(ret!=0){
+	fprintf(STDERR,"OpenBLAS: pthread_creat error in blas_thread_init function. Error code:%d\n",ret);
+	exit(1);
+      }
     }
 
 #ifdef MONITOR
@@ -765,6 +770,19 @@ void goto_set_num_threads(int num_threads) {
 
   if (num_threads < 1) num_threads = blas_num_threads;
 
+#ifndef NO_AFFINITY
+  if (num_threads == 1) {
+    if (blas_cpu_number == 1){
+      //OpenBLAS is already single thread.
+      return; 
+    }else{
+      //From multi-threads to single thread
+      //Restore the original affinity mask
+      gotoblas_set_affinity(-1);
+    }
+  }
+#endif
+
   if (num_threads > MAX_CPU_NUMBER) num_threads = MAX_CPU_NUMBER;
 
   if (num_threads > blas_num_threads) {
@@ -795,8 +813,20 @@ void goto_set_num_threads(int num_threads) {
     UNLOCK_COMMAND(&server_lock);
   }
 
+#ifndef NO_AFFINITY
+  if(blas_cpu_number == 1 && num_threads > 1){
+    //Restore the thread 0 affinity.
+    gotoblas_set_affinity(0);
+  }
+#endif
+
   blas_cpu_number  = num_threads;
 
+#if defined(ARCH_MIPS64) 
+  //set parameters for different number of threads.
+  blas_set_parameter();
+#endif
+
 }
 
 void openblas_set_num_threads(int num_threads) {

driver/others/blas_server_omp.c

@@ -63,6 +63,11 @@ void goto_set_num_threads(int num_threads) {
 
   omp_set_num_threads(blas_cpu_number);
   
+#if defined(ARCH_MIPS64) 
+  //set parameters for different number of threads.
+  blas_set_parameter();
+#endif
+
 }
 void openblas_set_num_threads(int num_threads) {
 

driver/others/blas_server_win32.c

@@ -63,6 +63,8 @@ static blas_pool_t   pool;
 static HANDLE	    blas_threads   [MAX_CPU_NUMBER];
 static DWORD	    blas_threads_id[MAX_CPU_NUMBER];
 
+
+
 static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
 
       if (!(mode & BLAS_COMPLEX)){
@@ -179,7 +181,7 @@ static DWORD WINAPI blas_thread_server(void *arg){
     
     do {
       action = WaitForMultipleObjects(2, handles, FALSE, INFINITE);
-    } while ((action != WAIT_OBJECT_0) && (action == WAIT_OBJECT_0 + 1));
+    } while ((action != WAIT_OBJECT_0) && (action != WAIT_OBJECT_0 + 1));
     
     if (action == WAIT_OBJECT_0 + 1) break;
 
@@ -263,7 +265,9 @@ static DWORD WINAPI blas_thread_server(void *arg){
       } else {
 	legacy_exec(routine, queue -> mode, queue -> args, sb);
       }
-    }
+    }else{
+		continue; //if queue == NULL
+	}
     
 #ifdef SMP_DEBUG
     fprintf(STDERR, "Server[%2ld] Finished!\n", cpu);
@@ -425,7 +429,7 @@ int exec_blas(BLASLONG num, blas_queue_t *queue){
 /* Shutdown procedure, but user don't have to call this routine. The */
 /* kernel automatically kill threads.                                */
 
-int blas_thread_shutdown_(void){
+int BLASFUNC(blas_thread_shutdown)(void){
 
   int i;
 
@@ -437,7 +441,7 @@ int blas_thread_shutdown_(void){
 
     SetEvent(pool.killed);
     
-    for(i = 0; i < blas_cpu_number - 1; i++){
+    for(i = 0; i < blas_num_threads - 1; i++){
       WaitForSingleObject(blas_threads[i], INFINITE);
     }
     
@@ -448,3 +452,47 @@ int blas_thread_shutdown_(void){
   
   return 0;
 }
+
+void goto_set_num_threads(int num_threads)
+{
+	 long i;
+
+	if (num_threads < 1) num_threads = blas_cpu_number;
+
+	if (num_threads > MAX_CPU_NUMBER) num_threads = MAX_CPU_NUMBER;
+
+	if (num_threads > blas_num_threads) {
+
+		LOCK_COMMAND(&server_lock);
+		
+		//increased_threads = 1;
+	    if (!blas_server_avail){
+
+			InitializeCriticalSection(&pool.lock);
+			pool.filled = CreateEvent(NULL, FALSE, FALSE, NULL);
+			pool.killed = CreateEvent(NULL, TRUE,  FALSE, NULL);
+
+			pool.shutdown = 0;
+			pool.queue    = NULL;
+			blas_server_avail = 1;
+		}
+		
+		for(i = blas_num_threads - 1; i < num_threads - 1; i++){	  
+		  
+			blas_threads[i] = CreateThread(NULL, 0, 
+				     blas_thread_server, (void *)i,
+				     0, &blas_threads_id[i]);
+		}
+	
+		blas_num_threads = num_threads;
+
+		UNLOCK_COMMAND(&server_lock);
+	}
+
+	blas_cpu_number  = num_threads;
+}
+
+void openblas_set_num_threads(int num)
+{
+	goto_set_num_threads(num);
+}

driver/others/dynamic.c

@@ -60,6 +60,14 @@ extern gotoblas_t  gotoblas_NEHALEM;
 extern gotoblas_t  gotoblas_OPTERON;
 extern gotoblas_t  gotoblas_OPTERON_SSE3;
 extern gotoblas_t  gotoblas_BARCELONA;
+extern gotoblas_t  gotoblas_BOBCAT;
+#ifndef NO_AVX
+extern gotoblas_t  gotoblas_SANDYBRIDGE;
+#else
+//Use NEHALEM kernels for sandy bridge
+#define gotoblas_SANDYBRIDGE gotoblas_NEHALEM
+#endif
+
 
 #define VENDOR_INTEL      1
 #define VENDOR_AMD        2
@@ -68,6 +76,31 @@ extern gotoblas_t  gotoblas_BARCELONA;
 
 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
 
+#ifndef NO_AVX
+static inline void xgetbv(int op, int * eax, int * edx){
+  __asm__ __volatile__
+    ("xgetbv": "=a" (*eax), "=d" (*edx) : "c" (op) : "cc");
+}
+#endif
+
+int support_avx(){
+#ifndef NO_AVX
+  int eax, ebx, ecx, edx;
+  int ret=0;
+  
+  cpuid(1, &eax, &ebx, &ecx, &edx);
+  if ((ecx & (1 << 28)) != 0 && (ecx & (1 << 27)) != 0){
+    xgetbv(0, &eax, &edx);
+    if((eax & 6) == 6){
+      ret=1;  //OS support AVX
+    }
+  }
+  return ret;
+#else
+  return 0;
+#endif
+}
+
 static int get_vendor(void){
   int eax, ebx, ecx, edx;
   char vendor[13];
@@ -122,15 +155,38 @@ static gotoblas_t *get_coretype(void){
 	if (model == 12) return &gotoblas_ATOM;
 	return NULL;
 
-	  case 2:
-		  //Intel Core (Clarkdale) / Core (Arrandale)
-		  // Pentium (Clarkdale) / Pentium Mobile (Arrandale)
-		  // Xeon (Clarkdale), 32nm
-		  if (model ==  5) return &gotoblas_NEHALEM;
+      case 2:
+	//Intel Core (Clarkdale) / Core (Arrandale)
+	// Pentium (Clarkdale) / Pentium Mobile (Arrandale)
+	// Xeon (Clarkdale), 32nm
+	if (model ==  5) return &gotoblas_NEHALEM;
 		  
-		  //Intel Xeon Processor 5600 (Westmere-EP)
-		  if (model == 12) return &gotoblas_NEHALEM;
-		  return NULL;
+	//Intel Xeon Processor 5600 (Westmere-EP)
+	//Xeon Processor E7 (Westmere-EX)
+	if (model == 12 || model == 15) return &gotoblas_NEHALEM;
+
+	//Intel Core i5-2000 /i7-2000 (Sandy Bridge)
+	//Intel Core i7-3000 / Xeon E5
+	if (model == 10 || model == 13) {
+	  if(support_avx())
+	    return &gotoblas_SANDYBRIDGE;
+	  else{
+	    fprintf(stderr, "OpenBLAS : Your OS doesn't support AVX. Use Nehalem kernels.\n");
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+	return NULL;
+      case 3:
+	//Intel Sandy Bridge 22nm (Ivy Bridge?)
+	if (model == 10) {
+	  if(support_avx())
+	    return &gotoblas_SANDYBRIDGE;
+	  else{
+	    fprintf(stderr, "OpenBLAS : Your OS doesn't support AVX. Use Nehalem kernels.\n");
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+	}
+	return NULL;
       }
       case 0xf:
       if (model <= 0x2) return &gotoblas_NORTHWOOD;
@@ -144,7 +200,9 @@ static gotoblas_t *get_coretype(void){
       if ((exfamily == 0) || (exfamily == 2)) {
 	if (ecx & (1 <<  0)) return &gotoblas_OPTERON_SSE3; 
 	else return &gotoblas_OPTERON;
-      }  else {
+      }  else if (exfamily == 5) {
+	return &gotoblas_BOBCAT;
+      } else {
 	return &gotoblas_BARCELONA;
       }
     }
@@ -178,6 +236,8 @@ static char *corename[] = {
     "Opteron(SSE3)",
     "Barcelona",
     "Nano",
+    "Sandybridge",
+    "Bobcat",
 };
 
 char *gotoblas_corename(void) {
@@ -197,7 +257,9 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_OPTERON)      return corename[13];
   if (gotoblas == &gotoblas_BARCELONA)    return corename[14];
   if (gotoblas == &gotoblas_NANO)         return corename[15];
-  
+  if (gotoblas == &gotoblas_SANDYBRIDGE)  return corename[16];
+  if (gotoblas == &gotoblas_BOBCAT)       return corename[17];
+
   return corename[0];
 }
 
@@ -216,7 +278,7 @@ void gotoblas_dynamic_init(void) {
   if (gotoblas && gotoblas -> init) {
     gotoblas -> init();
   } else {
-    fprintf(stderr, "GotoBLAS : Architecture Initialization failed. No initialization function found.\n");
+    fprintf(stderr, "OpenBLAS : Architecture Initialization failed. No initialization function found.\n");
     exit(1);
   }
   

driver/others/init.c

@@ -1,5 +1,5 @@
 /*****************************************************************************
-Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -85,6 +85,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define MAX_NODES	16
 #define MAX_CPUS	256
+#define NCPUBITS        (8*sizeof(unsigned long))
+#define MAX_BITMASK_LEN (MAX_CPUS/NCPUBITS)
+#define CPUELT(cpu)	((cpu) / NCPUBITS)
+#define CPUMASK(cpu)	((unsigned long) 1UL << ((cpu) % NCPUBITS))
+
 
 #define SH_MAGIC	0x510510
 
@@ -103,10 +108,10 @@ typedef struct {
   int num_nodes;
   int num_procs;
   int final_num_procs;
-  unsigned long avail;
-  
+  unsigned long avail [MAX_BITMASK_LEN];
+  int avail_count;
   unsigned long cpu_info   [MAX_CPUS];
-  unsigned long node_info  [MAX_NODES];
+  unsigned long node_info  [MAX_NODES][MAX_BITMASK_LEN];
   int cpu_use[MAX_CPUS];
 
 } shm_t;
@@ -126,7 +131,8 @@ static shm_t *common = (void *)-1;
 static int shmid, pshmid;
 static void *paddr;
 
-static unsigned long lprocmask, lnodemask;
+static unsigned long lprocmask[MAX_BITMASK_LEN], lnodemask;
+static int lprocmask_count = 0;
 static int numprocs = 1;
 static int numnodes = 1;
 
@@ -177,70 +183,114 @@ static inline int rcount(unsigned long number) {
   than sizeof(unsigned long). On 64 bits, the limit 
   is 64. On 32 bits, it is 32.
 ***/
-static inline unsigned long get_cpumap(int node) {
+static inline void get_cpumap(int node, unsigned long * node_info) {
 
   int infile;
-  unsigned long affinity;
+  unsigned long affinity[32];
   char name[160];
   char cpumap[160];
-  char *p, *dummy;
+  char *dummy;
   int i=0;
+  int count=0;
+  int k=0;
 
   sprintf(name, CPUMAP_NAME, node);
   
   infile = open(name, O_RDONLY);
+  for(i=0; i<32; i++){
+    affinity[i] = 0;
+  }
 
-  affinity = 0;
-    
   if (infile != -1) {
     
     read(infile, cpumap, sizeof(cpumap));
-    p = cpumap;
-    while (*p != '\n' && i<160){
-      if(*p != ',') {
-	name[i++]=*p;
+
+    for(i=0; i<160; i++){
+      if(cpumap[i] == '\n')
+	break;
+      if(cpumap[i] != ','){
+	name[k++]=cpumap[i];
+	
+	//Enough data for Hex
+	if(k >= NCPUBITS/4){
+	  affinity[count++] = strtoul(name, &dummy, 16);
+	  k=0;
+	}
       }
-      p++;
+
+    }
+    if(k!=0){
+      name[k]='\0';
+      affinity[count++] = strtoul(name, &dummy, 16);
+      k=0;
+    }
+    // 0-63bit -> node_info[0], 64-128bit -> node_info[1] ....
+    // revert the sequence
+    for(i=0; i<count && i<MAX_BITMASK_LEN; i++){
+      node_info[i]=affinity[count-i-1];
     }
-    p = name;
-
-    //    while ((*p == '0') || (*p == ',')) p++;
-
-    affinity = strtoul(p, &dummy, 16);
-   
     close(infile);
   }
 
-  return affinity;
+  return ;
 }
 
-static inline unsigned long get_share(int cpu, int level) {
+static inline void get_share(int cpu, int level, unsigned long * share) {
 
   int infile;
-  unsigned long affinity;
+  unsigned long affinity[32];
+  char cpumap[160];
   char name[160];
-  char *p;
-  
+  char *dummy;
+  int count=0;
+  int i=0,k=0;
+  int bitmask_idx = 0;
+
   sprintf(name, SHARE_NAME, cpu, level);
   
   infile = open(name, O_RDONLY);
 
-  affinity = (1UL << cpu);
-    
+  //  Init share
+  for(i=0; i<MAX_BITMASK_LEN; i++){
+    share[i]=0;
+  }
+  bitmask_idx = CPUELT(cpu);
+  share[bitmask_idx] = CPUMASK(cpu);
+
   if (infile != -1) {
     
-    read(infile, name, sizeof(name));
+    read(infile, cpumap, sizeof(cpumap));
+
+    for(i=0; i<160; i++){
+      if(cpumap[i] == '\n')
+	break;
+      if(cpumap[i] != ','){
+	name[k++]=cpumap[i];
+	
+	//Enough data 
+	if(k >= NCPUBITS/4){
+	  affinity[count++] = strtoul(name, &dummy, 16);
+	  k=0;
+	}
+      }
+
+    }
+    if(k!=0){
+      name[k]='\0';
+      affinity[count++] = strtoul(name, &dummy, 16);
+      k=0;
+    }
+    // 0-63bit -> node_info[0], 64-128bit -> node_info[1] ....
+    // revert the sequence
+    for(i=0; i<count && i<MAX_BITMASK_LEN; i++){
+      share[i]=affinity[count-i-1];
+    }
    
-    p = name;
-
-    while ((*p == '0') || (*p == ',')) p++;
-
-    affinity = strtol(p, &p, 16);
    
     close(infile);
   }
 
-  return affinity;
+  return ;
 }
 
 static int numa_check(void) {
@@ -248,6 +298,7 @@ static int numa_check(void) {
   DIR *dp;
   struct dirent *dir;
   int node;
+  int j;
 
   common -> num_nodes = 0;
 
@@ -258,7 +309,9 @@ static int numa_check(void) {
     return 0;
   }
 
-  for (node = 0; node < MAX_NODES; node ++) common -> node_info[node] = 0;
+  for (node = 0; node < MAX_NODES; node ++) {
+    for (j = 0; j<MAX_BITMASK_LEN; j++) common -> node_info[node][j] = 0;
+  }
 
   while ((dir = readdir(dp)) != NULL) {
     if (*(unsigned int *) dir -> d_name == 0x065646f6eU) {
@@ -266,12 +319,12 @@ static int numa_check(void) {
       node = atoi(&dir -> d_name[4]);
 
       if (node > MAX_NODES) {
-	fprintf(stderr, "\nGotoBLAS Warining : MAX_NODES (NUMA) is too small. Terminated.\n");
+	fprintf(stderr, "\nOpenBLAS Warning : MAX_NODES (NUMA) is too small. Terminated.\n");
 	exit(1);
       }
 
       common -> num_nodes ++;
-      common -> node_info[node] = get_cpumap(node);
+      get_cpumap(node, common->node_info[node]);
 
     }
   }
@@ -284,7 +337,7 @@ static int numa_check(void) {
   fprintf(stderr, "Numa found : number of Nodes = %2d\n", common -> num_nodes);
 
   for (node = 0; node < common -> num_nodes; node ++)
-    fprintf(stderr, "MASK (%2d) : %08lx\n", node, common -> node_info[node]);
+    fprintf(stderr, "MASK (%2d) : %08lx\n", node, common -> node_info[node][0]);
 #endif
 
   return common -> num_nodes;
@@ -296,11 +349,13 @@ static void numa_mapping(void) {
   int i, j, h;
   unsigned long work, bit;
   int count = 0;
+  int bitmask_idx = 0;
 
   for (node = 0; node < common -> num_nodes; node ++) {
     core = 0;
     for (cpu = 0; cpu < common -> num_procs; cpu ++) {
-      if (common -> node_info[node] & common -> avail & (1UL << cpu)) {
+      bitmask_idx = CPUELT(cpu);
+      if (common -> node_info[node][bitmask_idx] & common -> avail[bitmask_idx] & CPUMASK(cpu)) {
 	common -> cpu_info[count] = WRITE_CORE(core) | WRITE_NODE(node) | WRITE_CPU(cpu);
 	count ++;
 	core ++;
@@ -357,58 +412,92 @@ static void numa_mapping(void) {
 
 static void disable_hyperthread(void) {
 
-  unsigned long share;
+  unsigned long share[MAX_BITMASK_LEN];
   int cpu;
+  int bitmask_idx = 0;
+  int i=0, count=0;
+  bitmask_idx = CPUELT(common -> num_procs);
 
-  if(common->num_procs > 64){
-    fprintf(stderr, "\nOpenBLAS Warining : The number of CPU/Cores(%d) is beyond the limit(64). Terminated.\n", common->num_procs);
-    exit(1);
-  }else if(common->num_procs == 64){
-    common -> avail = 0xFFFFFFFFFFFFFFFFUL;
-  }else
-    common -> avail = (1UL << common -> num_procs) - 1;
+  for(i=0; i< bitmask_idx; i++){
+    common -> avail[count++] = 0xFFFFFFFFFFFFFFFFUL;
+  }
+  if(CPUMASK(common -> num_procs) != 1){
+    common -> avail[count++] = CPUMASK(common -> num_procs) - 1;
+  }
+  common -> avail_count = count;
+
+  /* if(common->num_procs > 64){ */
+  /*   fprintf(stderr, "\nOpenBLAS Warning : The number of CPU/Cores(%d) is beyond the limit(64). Terminated.\n", common->num_procs); */
+  /*   exit(1); */
+  /* }else if(common->num_procs == 64){ */
+  /*   common -> avail = 0xFFFFFFFFFFFFFFFFUL; */
+  /* }else */
+  /*   common -> avail = (1UL << common -> num_procs) - 1; */
 
 #ifdef DEBUG
-  fprintf(stderr, "\nAvail CPUs    : %04lx.\n", common -> avail);
+  fprintf(stderr, "\nAvail CPUs    : ");
+  for(i=0; i<count; i++)
+    fprintf(stderr, "%04lx ", common -> avail[i]);
+  fprintf(stderr, ".\n");
 #endif
 
   for (cpu = 0; cpu < common -> num_procs; cpu ++) {
-    
-    share = (get_share(cpu, 1) & common -> avail);
-    
-    if (popcount(share) > 1) {
+
+    get_share(cpu, 1, share);
+
+    //When the shared cpu are in different element of share & avail array, this may be a bug.
+    for (i = 0; i < count ; i++){
+
+      share[i] &= common->avail[i];
+
+      if (popcount(share[i]) > 1) {
       
 #ifdef DEBUG
-      fprintf(stderr, "Detected Hyper Threading on CPU %4x; disabled CPU %04lx.\n",
-	      cpu, share & ~(1UL << cpu));
+	fprintf(stderr, "Detected Hyper Threading on CPU %4x; disabled CPU %04lx.\n",
+		cpu, share[i] & ~(CPUMASK(cpu)));
 #endif
       
-      common -> avail &= ~((share & ~(1UL << cpu)));
+	common -> avail[i] &= ~((share[i] & ~ CPUMASK(cpu)));
+      }
     }
   }
 }
 
 static void disable_affinity(void) {
-
+  int i=0;
+  int bitmask_idx=0;
+  int count=0;
 #ifdef DEBUG
-    fprintf(stderr, "Final all available CPUs  : %04lx.\n\n", common -> avail);
+    fprintf(stderr, "Final all available CPUs  : %04lx.\n\n", common -> avail[0]);
     fprintf(stderr, "CPU mask                  : %04lx.\n\n", *(unsigned long *)&cpu_orig_mask[0]);
 #endif
 
-  if(common->final_num_procs > 64){
-    fprintf(stderr, "\nOpenBLAS Warining : The number of CPU/Cores(%d) is beyond the limit(64). Terminated.\n", common->final_num_procs);
-    exit(1);
-  }else if(common->final_num_procs == 64){
-    lprocmask = 0xFFFFFFFFFFFFFFFFUL;
-  }else
-    lprocmask = (1UL << common -> final_num_procs) - 1;
+  /* if(common->final_num_procs > 64){ */
+  /*   fprintf(stderr, "\nOpenBLAS Warining : The number of CPU/Cores(%d) is beyond the limit(64). Terminated.\n", common->final_num_procs); */
+  /*   exit(1); */
+  /* }else if(common->final_num_procs == 64){ */
+  /*   lprocmask = 0xFFFFFFFFFFFFFFFFUL; */
+  /* }else */
+  /*   lprocmask = (1UL << common -> final_num_procs) - 1; */
+
+  bitmask_idx = CPUELT(common -> final_num_procs);
+
+  for(i=0; i< bitmask_idx; i++){
+    lprocmask[count++] = 0xFFFFFFFFFFFFFFFFUL;
+  }
+  if(CPUMASK(common -> final_num_procs) != 1){
+    lprocmask[count++] = CPUMASK(common -> final_num_procs) - 1;
+  }
+  lprocmask_count = count;
 
 #ifndef USE_OPENMP
-  lprocmask &= *(unsigned long *)&cpu_orig_mask[0];
+  for(i=0; i< count; i++){
+    lprocmask[i] &= ((unsigned long *)&cpu_orig_mask[0])[i];
+  }
 #endif
 
 #ifdef DEBUG
-    fprintf(stderr, "I choose these CPUs  : %04lx.\n\n", lprocmask);
+    fprintf(stderr, "I choose these CPUs  : %04lx.\n\n", lprocmask[0]);
 #endif
 
 }
@@ -498,7 +587,7 @@ static void create_pshmem(void) {
 static void local_cpu_map(void) {
 
   int cpu, id, mapping;
-
+  int bitmask_idx = 0;
   cpu = 0;
   mapping = 0;
 
@@ -508,8 +597,9 @@ static void local_cpu_map(void) {
     if (id > 0) {
       if (is_dead(id)) common -> cpu_use[cpu] = 0;
     }
-
-    if ((common -> cpu_use[cpu] == 0) && (lprocmask & (1UL << cpu))) {
+    
+    bitmask_idx = CPUELT(cpu);
+    if ((common -> cpu_use[cpu] == 0) && (lprocmask[bitmask_idx] & CPUMASK(cpu))) {
 
       common -> cpu_use[cpu] = pshmid;
       cpu_mapping[mapping] = READ_CPU(common -> cpu_info[cpu]);
@@ -595,6 +685,7 @@ void gotoblas_affinity_init(void) {
 #ifndef USE_OPENMP
   cpu_set_t cpu_mask;
 #endif
+  int i;
 
   if (initialized) return;
 
@@ -646,6 +737,11 @@ void gotoblas_affinity_init(void) {
 
     common -> num_procs = get_nprocs();
 
+    if(common -> num_procs > MAX_CPUS) {
+      fprintf(stderr, "\nOpenBLAS Warining : The number of CPU/Cores(%d) is beyond the limit(%d). Terminated.\n", common->num_procs, MAX_CPUS);
+      exit(1);
+    }
+
     for (cpu = 0; cpu < common -> num_procs; cpu++) common -> cpu_info[cpu] = cpu;
     
     numa_check();
@@ -654,7 +750,8 @@ void gotoblas_affinity_init(void) {
 
     if (common -> num_nodes > 1) numa_mapping();
 
-    common -> final_num_procs = popcount(common -> avail);
+    common -> final_num_procs = 0;
+    for(i = 0; i < common -> avail_count; i++) common -> final_num_procs += popcount(common -> avail[i]);
 
     for (cpu = 0; cpu < common -> final_num_procs; cpu ++) common -> cpu_use[cpu] =  0;
 
@@ -664,7 +761,8 @@ void gotoblas_affinity_init(void) {
 
   disable_affinity();
 
-  num_avail = popcount(lprocmask);
+  num_avail = 0;
+  for(i=0; i<lprocmask_count; i++) num_avail += popcount(lprocmask[i]);
 
   if ((numprocs <= 0) || (numprocs > num_avail)) numprocs = num_avail;
 

driver/others/memory.c

@@ -1,5 +1,5 @@
 /*****************************************************************************
-Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -103,7 +103,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <sys/syscall.h>
 #endif
 
-#if defined(OS_FreeBSD) || defined(OS_Darwin)
+#if defined(OS_FREEBSD) || defined(OS_DARWIN)
 #include <sys/sysctl.h>
 #endif
 
@@ -185,7 +185,7 @@ int get_num_procs(void) {
 
 #endif
 
-#if defined(OS_FreeBSD) || defined(OS_Darwin)
+#if defined(OS_FREEBSD) 
 
 int get_num_procs(void) {
   
@@ -206,7 +206,27 @@ int get_num_procs(void) {
 
 #endif
 
+#if defined(OS_DARWIN)
+int get_num_procs(void) {
+  static int nums = 0;
+  size_t len;
+  if (nums == 0){
+    len = sizeof(int);
+    sysctlbyname("hw.physicalcpu", &nums, &len, NULL, 0);
+  }
+  return nums;
+}
+#endif
+
+/*
+OpenBLAS uses the numbers of CPU cores in multithreading. 
+It can be set by openblas_set_num_threads(int num_threads);
+*/
 int blas_cpu_number  = 0;
+/*
+The numbers of threads in the thread pool. 
+This value is equal or large than blas_cpu_number. This means some threads are sleep.
+*/
 int blas_num_threads = 0;
 
 int  goto_get_num_procs  (void) {
@@ -215,7 +235,7 @@ int  goto_get_num_procs  (void) {
 
 int blas_get_cpu_number(void){
   char *p;
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FreeBSD) || defined(OS_Darwin)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_DARWIN)
   int max_num;
 #endif
   int blas_goto_num   = 0;
@@ -223,7 +243,7 @@ int blas_get_cpu_number(void){
 
   if (blas_num_threads) return blas_num_threads;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FreeBSD) || defined(OS_Darwin)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_DARWIN)
   max_num = get_num_procs();
 #endif
 
@@ -250,7 +270,7 @@ int blas_get_cpu_number(void){
   else if (blas_omp_num > 0) blas_num_threads = blas_omp_num;
   else blas_num_threads = MAX_CPU_NUMBER;
 
-#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FreeBSD) || defined(OS_Darwin)
+#if defined(OS_LINUX) || defined(OS_WINDOWS) || defined(OS_FREEBSD) || defined(OS_DARWIN)
   if (blas_num_threads > max_num) blas_num_threads = max_num;
 #endif
 
@@ -390,11 +410,11 @@ static void *alloc_mmap(void *address){
 		  
 #ifdef OS_LINUX
 #ifdef DEBUG
-		  int ret;
+		  int ret=0;
 		  ret=my_mbind(map_address, BUFFER_SIZE * SCALING, MPOL_PREFERRED, NULL, 0, 0);
 		  if(ret==-1){
 			  int errsv=errno;
-			  perror("alloc_mmap:");
+			  perror("OpenBLAS alloc_mmap:");
 			  printf("error code=%d,\tmap_address=%lx\n",errsv,map_address);
 		  }
 
@@ -884,7 +904,7 @@ void *blas_memory_alloc(int procpos){
       if (!blas_num_threads) blas_cpu_number = blas_get_cpu_number();
 #endif
       
-#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64)
+#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64)
 #ifndef DYNAMIC_ARCH
       blas_set_parameter();
 #endif
@@ -1128,7 +1148,7 @@ static BLASULONG   init_lock = 0UL;
 static void _touch_memory(blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, 
 			  void *sa, void *sb, BLASLONG pos) {
 
-#ifndef ARCH_POWER
+#if !defined(ARCH_POWER) && !defined(ARCH_SPARC)
 
   long size;
   BLASULONG buffer;
@@ -1289,6 +1309,7 @@ void DESTRUCTOR gotoblas_quit(void) {
    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_set_num_threads.c

@@ -1,5 +1,5 @@
 /*****************************************************************************
-Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -33,13 +33,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #ifdef SMP_SERVER
-#ifdef OS_LINUX
 
 extern  void openblas_set_num_threads(int num_threads) ;
 
-void NAME(int* num_threads){
+void openblas_set_num_threads_(int* num_threads){
 	openblas_set_num_threads(*num_threads);
 }
 
-#endif
+#else
+//Single thread
+
+void openblas_set_num_threads(int num_threads) {
+}
+
+void openblas_set_num_threads_(int* num_threads){
+
+}
 #endif

driver/others/parameter.c

@@ -45,8 +45,22 @@ int get_L2_size(void);
 #define DEFAULT_GEMM_P 128
 #define DEFAULT_GEMM_Q 128
 #define DEFAULT_GEMM_R 128
+#define DEFAULT_GEMM_OFFSET_A 0
+#define DEFAULT_GEMM_OFFSET_B 0
 
 /* Global Parameter */
+#if GEMM_OFFSET_A == gemm_offset_a
+BLASLONG gemm_offset_a = DEFAULT_GEMM_OFFSET_A;
+#else
+BLASLONG gemm_offset_a = GEMM_OFFSET_A;
+#endif
+
+#if GEMM_OFFSET_B == gemm_offset_b
+BLASLONG gemm_offset_b = DEFAULT_GEMM_OFFSET_B;
+#else
+BLASLONG gemm_offset_b = GEMM_OFFSET_B;
+#endif
+
 #if SGEMM_P == sgemm_p
 BLASLONG sgemm_p = DEFAULT_GEMM_P;
 #else
@@ -149,9 +163,9 @@ int get_L2_size(void){
 
   int eax, ebx, ecx, edx;
 
-#if defined(ATHLON) || defined(OPTERON) || defined(BARCELONA) || \
+#if defined(ATHLON) || defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT) || \
     defined(CORE_PRESCOTT) || defined(CORE_CORE2) || defined(PENRYN) || defined(DUNNINGTON) || \
-    defined(CORE_NEHALEM) || defined(ATOM) || defined(GENERIC)
+  defined(CORE_NEHALEM) || defined(CORE_SANDYBRIDGE) || defined(ATOM) || defined(GENERIC)
 
   cpuid(0x80000006, &eax, &ebx, &ecx, &edx);
 
@@ -370,6 +384,17 @@ void blas_set_parameter(void){
 #endif
 #endif
 
+#if defined(SANDYBRIDGE)
+  sgemm_p = 1024;
+  dgemm_p =  512;
+  cgemm_p =  512;
+  zgemm_p =  256;
+#ifdef EXPRECISION
+  qgemm_p =  256;
+  xgemm_p =  128;
+#endif
+#endif
+
 #if defined(CORE_PRESCOTT)  || defined(GENERIC)
   size >>= 6;
 
@@ -421,7 +446,7 @@ void blas_set_parameter(void){
 #endif
 #endif
 
-#if defined(CORE_BARCELONA)
+#if defined(CORE_BARCELONA) || defined(CORE_BOBCAT)
   size >>= 8;
 
   sgemm_p = 232 * size;
@@ -666,3 +691,36 @@ void blas_set_parameter(void){
 #endif
 
 #endif
+
+#if defined(ARCH_MIPS64) 
+void blas_set_parameter(void){
+#if defined(LOONGSON3A)
+#ifdef SMP
+  if(blas_num_threads == 1){
+#endif
+    //single thread
+    dgemm_r = 1024;
+#ifdef SMP
+  }else{
+    //multi thread
+    dgemm_r = 200;
+  }
+#endif
+#endif
+
+#if defined(LOONGSON3B)
+#ifdef SMP
+  if(blas_num_threads == 1 || blas_num_threads == 2){
+#endif
+    //single thread
+    dgemm_r = 640;
+#ifdef SMP
+  }else{
+    //multi thread
+    dgemm_r = 160;
+  }
+#endif
+#endif 
+
+}
+#endif

exports/Makefile

@@ -10,6 +10,14 @@ ifndef NO_CBLAS
 NO_CBLAS	= 0
 endif
 
+ifndef NO_LAPACK
+NO_LAPACK	= 0
+endif
+
+ifndef NO_LAPACKE
+NO_LAPACKE	= 0
+endif
+
 ifeq ($(OSNAME), WINNT)
 ifeq ($(F_COMPILER), GFORTRAN)
 EXTRALIB += -lgfortran
@@ -58,16 +66,21 @@ dll  : ../$(LIBDLLNAME)
 
 dll2 : libgoto2_shared.dll
 
-../$(LIBDLLNAME) : ../$(LIBNAME) libgoto2.def dllinit.$(SUFFIX)
+# On Windows, we only generate a DLL without a version suffix.  This is because
+# applications which link against the dynamic library reference a fixed DLL name
+# in their import table.  By instead using a stable name it is possible to
+# upgrade between library versions, without needing to re-link an application.
+# For more details see: https://github.com/xianyi/OpenBLAS/issues/127.
+../$(LIBDLLNAME) : ../$(LIBNAME) libopenblas.def dllinit.$(SUFFIX)
 	$(RANLIB) ../$(LIBNAME)
 ifeq ($(BINARY32), 1)
-	$(DLLWRAP) -o ../$(LIBDLLNAME) --def libgoto2.def \
-	--entry _dllinit@12 -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(FEXTRALIB)
-	-lib /machine:i386 /def:libgoto2.def
+	$(DLLWRAP) -o ../$(LIBDLLNAME) --def libopenblas.def \
+	--entry _dllinit@12 -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(EXTRALIB)
+	-lib /machine:i386 /def:libopenblas.def
 else
-	$(DLLWRAP) -o ../$(LIBDLLNAME) --def libgoto2.def \
-	--entry $(FU)dllinit    -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(FEXTRALIB)
-	-lib /machine:X64 /def:libgoto2.def 
+	$(DLLWRAP) -o ../$(LIBDLLNAME) --def libopenblas.def \
+	--entry $(FU)dllinit    -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(EXTRALIB)
+	-lib /machine:X64 /def:libopenblas.def 
 endif
 
 libgoto2_shared.dll : ../$(LIBNAME) libgoto2_shared.def
@@ -75,14 +88,14 @@ libgoto2_shared.dll : ../$(LIBNAME) libgoto2_shared.def
 	-Wl,--whole-archive ../$(LIBNAME) -Wl,--no-whole-archive \
 	-Wl,--out-implib,libgoto2_shared.lib $(FEXTRALIB)
 
-libgoto2.def : gensymbol
-	perl ./gensymbol win2k    $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) > $(@F)
+libopenblas.def : gensymbol
+	perl ./gensymbol win2k    $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) > $(@F)
 
 libgoto2_shared.def : gensymbol
-	perl ./gensymbol win2k    $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK)  > $(@F)
+	perl ./gensymbol win2k    $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) > $(@F)
 
 libgoto_hpl.def : gensymbol
-	perl ./gensymbol win2khpl $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK)  > $(@F)
+	perl ./gensymbol win2khpl $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) > $(@F)
 
 $(LIBDYNNAME) : ../$(LIBNAME) osx.def
 	$(CC) $(CFLAGS) -all_load -headerpad_max_install_names -install_name $(CURDIR)/../$(LIBDYNNAME) -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
@@ -100,20 +113,21 @@ so : ../$(LIBSONAME)
 ../$(LIBSONAME) : ../$(LIBNAME) linux.def linktest.c
 	$(CC) $(CFLAGS) -shared -o ../$(LIBSONAME) \
 	-Wl,--whole-archive ../$(LIBNAME) -Wl,--no-whole-archive \
-	-Wl,--retain-symbols-file=linux.def $(EXTRALIB)
+	-Wl,--retain-symbols-file=linux.def -Wl,-soname,$(LIBPREFIX).so.$(MAJOR_VERSION) $(EXTRALIB)
 	$(CC) $(CFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
 	rm -f linktest
 
 endif
 
-ifeq ($(OSNAME), FreeBSD)
+#http://stackoverflow.com/questions/7656425/makefile-ifeq-logical-or
+ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD NetBSD))
 
 so : ../$(LIBSONAME)
 
 ../$(LIBSONAME) : ../$(LIBNAME) linux.def linktest.c
 	$(CC) $(CFLAGS)  -shared -o ../$(LIBSONAME) \
 	-Wl,--whole-archive ../$(LIBNAME) -Wl,--no-whole-archive \
-	-Wl,--retain-symbols-file=linux.def $(EXTRALIB)
+	-Wl,--retain-symbols-file=linux.def $(FEXTRALIB) $(EXTRALIB)
 	$(CC) $(CFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
 	rm -f linktest
 
@@ -163,23 +177,23 @@ static : ../$(LIBNAME)
 	rm -f goto.$(SUFFIX)
 
 linux.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol linux $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) > $(@F)
+	perl ./gensymbol linux $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) > $(@F)
 
 osx.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol osx $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) > $(@F)
+	perl ./gensymbol osx $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) > $(@F)
 
 aix.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol aix $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) > $(@F)
+	perl ./gensymbol aix $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) > $(@F)
 
 symbol.S : gensymbol
-	perl ./gensymbol win2kasm noarch dummy $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) > symbol.S
+	perl ./gensymbol win2kasm noarch dummy $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) > symbol.S
 
 test : linktest.c
 	$(CC) $(CFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) -lm && echo OK.
 	rm -f linktest
 
 linktest.c : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol linktest  $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) > linktest.c
+	perl ./gensymbol linktest  $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) > linktest.c
 
 clean ::
 	@rm -f *.def *.dylib __.SYMDEF*

f_check

@@ -32,11 +32,12 @@ if ($compiler eq "") {
 	      "pgf95", "pgf90", "pgf77",
               "ifort");
     
+OUTER:
     foreach $lists (@lists) {
         foreach $path (@path) {
-            if (-f $path . "/" . $lists) {
+            if (-x $path . "/" . $lists) {
                 $compiler = $lists;
-                break;
+                last OUTER;
             }
         }
     }
@@ -210,6 +211,10 @@ if (!$?) {
 	if ($?) {
 	    $link = `$compiler $openmp -q32 -v ftest2.f 2>&1 && rm -f a.out a.exe`;
 	}
+       #For gfortran MIPS
+	if ($?) {
+	    $link = `$compiler $openmp -mabi=n32 -v ftest2.f 2>&1 && rm -f a.out a.exe`;
+	}
 	$binary = "" if ($?);
     }
     
@@ -218,6 +223,10 @@ if (!$?) {
 	if ($?) {
 	    $link = `$compiler $openmp -q64 -v ftest2.f 2>&1 && rm -f a.out a.exe`;
 	}
+       #For gfortran MIPS
+	if ($?) {
+	    $link = `$compiler $openmp -mabi=64 -v ftest2.f 2>&1 && rm -f a.out a.exe`;
+	}
 	$binary = "" if ($?);
     }
     
@@ -237,6 +246,8 @@ if ($link ne "") {
     $link =~ s/\-rpath\s+/\-rpath\@/g;
 
     @flags = split(/[\s\,\n]/, $link);
+    # remove leading and trailing quotes from each flag.
+    @flags = map {s/^['"]|['"]$//g; $_} @flags;
 
     foreach $flags (@flags) {
 	if (
@@ -284,6 +295,10 @@ if ($link ne "") {
 
 }
 
+if ($vendor eq "INTEL"){
+    $linker_a .= "-lgfortran"
+}
+
 open(MAKEFILE, ">> $makefile") || die "Can't append $makefile";
 open(CONFFILE, ">> $config"  ) || die "Can't append $config";
 

getarch.c

@@ -1,5 +1,5 @@
 /*****************************************************************************
-Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -102,6 +102,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* #define FORCE_BARCELONA	*/
 /* #define FORCE_SHANGHAI	*/
 /* #define FORCE_ISTANBUL	*/
+/* #define FORCE_BULLDOZER	*/
+/* #define FORCE_BOBCAT	*/
 /* #define FORCE_SSE_GENERIC	*/
 /* #define FORCE_VIAC3		*/
 /* #define FORCE_NANO		*/
@@ -117,6 +119,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* #define FORCE_CELL		*/
 /* #define FORCE_SICORTEX	*/
 /* #define FORCE_LOONGSON3A      */
+/* #define FORCE_LOONGSON3B      */
 /* #define FORCE_ITANIUM2	*/
 /* #define FORCE_GENERIC	*/
 /* #define FORCE_SPARC		*/
@@ -277,6 +280,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "NEHALEM"
 #endif
 
+#ifdef FORCE_SANDYBRIDGE
+#define FORCE
+#define FORCE_INTEL
+#define ARCHITECTURE    "X86"
+#define SUBARCHITECTURE "SANDYBRIDGE"
+#define ARCHCONFIG   "-DSANDYBRIDGE " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 -DHAVE_AVX"
+#define LIBNAME   "sandybridge"
+#define CORENAME  "SANDYBRIDGE"
+#endif
+
 #ifdef FORCE_ATOM
 #define FORCE
 #define FORCE_INTEL
@@ -333,7 +350,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "OPTERON"
 #endif
 
-#if defined(FORCE_BARCELONA) || defined(FORCE_SHANGHAI) || defined(FORCE_ISTANBUL)
+#if defined(FORCE_BARCELONA) || defined(FORCE_SHANGHAI) || defined(FORCE_ISTANBUL) || defined (FORCE_BULLDOZER)
 #define FORCE
 #define FORCE_INTEL
 #define ARCHITECTURE    "X86"
@@ -341,13 +358,28 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ARCHCONFIG   "-DBARCELONA " \
 		     "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=64 " \
 		     "-DL2_SIZE=524288 -DL2_LINESIZE=64  -DL3_SIZE=2097152 " \
-		     "-DDTB_DEFAULT_ENTRIES=48 -DDTB_SIZE=4096 -DHAVE_3DNOW " \
-		     "-DHAVE_3DNOWEX -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 " \
+		     "-DDTB_DEFAULT_ENTRIES=48 -DDTB_SIZE=4096 " \
+		     "-DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 " \
 		     "-DHAVE_SSE4A -DHAVE_MISALIGNSSE -DHAVE_128BITFPU -DHAVE_FASTMOVU"
 #define LIBNAME   "barcelona"
 #define CORENAME  "BARCELONA"
 #endif
 
+#if defined(FORCE_BOBCAT) 
+#define FORCE
+#define FORCE_INTEL
+#define ARCHITECTURE    "X86"
+#define SUBARCHITECTURE "BOBCAT"
+#define ARCHCONFIG   "-DBOBCAT " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=524288 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=40 -DDTB_SIZE=4096 " \
+		     "-DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 " \
+		     "-DHAVE_SSE4A -DHAVE_MISALIGNSSE -DHAVE_CFLUSH -DHAVE_CMOV"
+#define LIBNAME   "bobcat"
+#define CORENAME  "BOBCAT"
+#endif
+
 #ifdef FORCE_SSE_GENERIC
 #define FORCE
 #define FORCE_INTEL
@@ -548,6 +580,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_LOONGSON3B
+#define FORCE
+#define ARCHITECTURE    "MIPS"
+#define SUBARCHITECTURE "LOONGSON3B"
+#define SUBDIRNAME      "mips64"
+#define ARCHCONFIG   "-DLOONGSON3B " \
+       "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=32 " \
+       "-DL2_SIZE=512488 -DL2_LINESIZE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=4 "
+#define LIBNAME   "loongson3b"
+#define CORENAME  "LOONGSON3B"
+#else
+#endif
+
 #ifdef FORCE_ITANIUM2
 #define FORCE
 #define ARCHITECTURE    "IA64"

getarch_2nd.c

@@ -34,6 +34,7 @@ int main(int argc, char **argv) {
 #ifdef USE64BITINT
 	printf("#define USE64BITINT\n");
 #endif
+	printf("#define GEMM_MULTITHREAD_THRESHOLD\t%ld\n", GEMM_MULTITHREAD_THRESHOLD);
   }
 
   return 0;

interface/Makefile

@@ -318,7 +318,7 @@ CZBLAS3OBJS   = \
 
 ifndef NO_CBLAS
 
-CFLAGS += -I.
+override CFLAGS += -I.
 
 SBLAS1OBJS   += $(CSBLAS1OBJS)
 SBLAS2OBJS   += $(CSBLAS2OBJS)
@@ -400,7 +400,7 @@ all :: libs
 
 ifdef FUNCTION_PROFILE
 $(BLASOBJS) $(BLASOBJS_P) : functable.h
-$(BLASOBJS) $(BLASOBJS_P) : CFLAGS += -DPROFILE_FUNC_NAME=interface_$(*F)
+$(BLASOBJS) $(BLASOBJS_P) : override CFLAGS += -DPROFILE_FUNC_NAME=interface_$(*F)
 
 functable.h : Makefile
 	./create $(FUNCALLFILES) > functable.h
@@ -420,7 +420,7 @@ level3 : $(SBLAS3OBJS) $(DBLAS3OBJS) $(QBLAS3OBJS) $(CBLAS3OBJS) $(ZBLAS3OBJS) $
 	$(AR) $(ARFLAGS) -ru $(TOPDIR)/$(LIBNAME) $^
 
 $(CSBLASOBJS) $(CSBLASOBJS_P) $(CDBLASOBJS) $(CDBLASOBJS_P) $(CQBLASOBJS) $(CQBLASOBJS_P) \
-$(CCBLASOBJS) $(CCBLASOBJS_P) $(CZBLASOBJS) $(CZBLASOBJS_P) $(CXBLASOBJS) $(CXBLASOBJS_P) : CFLAGS += -DCBLAS
+$(CCBLASOBJS) $(CCBLASOBJS_P) $(CZBLASOBJS) $(CZBLASOBJS_P) $(CXBLASOBJS) $(CXBLASOBJS_P) : override CFLAGS += -DCBLAS
 
 srot.$(SUFFIX) srot.$(PSUFFIX) : rot.c
 	$(CC) $(CFLAGS) -c $< -o $(@F)
@@ -770,20 +770,36 @@ xgeru.$(SUFFIX) xgeru.$(PSUFFIX) : zger.c
 xgerc.$(SUFFIX) xgerc.$(PSUFFIX) : zger.c
 	$(CC) -c $(CFLAGS) -DCONJ $< -o $(@F)
 
+ifndef USE_NETLIB_GEMV
 sgemv.$(SUFFIX) sgemv.$(PSUFFIX): gemv.c
 	$(CC) -c $(CFLAGS) -o $(@F) $<
 
 dgemv.$(SUFFIX) dgemv.$(PSUFFIX): gemv.c
 	$(CC) -c $(CFLAGS) -o $(@F) $<
+else
+sgemv.$(SUFFIX) sgemv.$(PSUFFIX): netlib/sgemv.f
+	$(FC) -c $(FFLAGS) -o $(@F) $<
+
+dgemv.$(SUFFIX) dgemv.$(PSUFFIX): netlib/dgemv.f
+	$(FC) -c $(FFLAGS) -o $(@F) $<
+endif
 
 qgemv.$(SUFFIX) qgemv.$(PSUFFIX): gemv.c
 	$(CC) -c $(CFLAGS) -o $(@F) $<
-
+	
+ifndef USE_NETLIB_GEMV
 cgemv.$(SUFFIX) cgemv.$(PSUFFIX): zgemv.c
 	$(CC) -c $(CFLAGS) -o $(@F) $<
 
 zgemv.$(SUFFIX) zgemv.$(PSUFFIX): zgemv.c
 	$(CC) -c $(CFLAGS) -o $(@F) $<
+else
+cgemv.$(SUFFIX) cgemv.$(PSUFFIX): netlib/cgemv.f
+	$(FC) -c $(FFLAGS) -o $(@F) $<
+
+zgemv.$(SUFFIX) zgemv.$(PSUFFIX): netlib/zgemv.f
+	$(FC) -c $(FFLAGS) -o $(@F) $<
+endif
 
 xgemv.$(SUFFIX) xgemv.$(PSUFFIX): zgemv.c
 	$(CC) -c $(CFLAGS) -o $(@F) $<

interface/gemm.c

@@ -397,8 +397,13 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
   mode |= (transb << BLAS_TRANSB_SHIFT);
 
   args.common = NULL;
-  args.nthreads = num_cpu_avail(3);
 
+  if(args.m <= GEMM_MULTITHREAD_THRESHOLD || args.n <= GEMM_MULTITHREAD_THRESHOLD 
+     || args.k <=GEMM_MULTITHREAD_THRESHOLD){
+    args.nthreads = 1;
+  }else{
+    args.nthreads = num_cpu_avail(3);
+  }
  if (args.nthreads == 1) {
 #endif
     

interface/netlib/cgemv.f

@@ -0,0 +1,285 @@
+      SUBROUTINE CGEMV(TRANS,M,N,ALPHA,A,LDA,X,INCX,BETA,Y,INCY)
+*     .. Scalar Arguments ..
+      COMPLEX ALPHA,BETA
+      INTEGER INCX,INCY,LDA,M,N
+      CHARACTER TRANS
+*     ..
+*     .. Array Arguments ..
+      COMPLEX A(LDA,*),X(*),Y(*)
+*     ..
+*
+*  Purpose
+*  =======
+*
+*  CGEMV performs one of the matrix-vector operations
+*
+*     y := alpha*A*x + beta*y,   or   y := alpha*A**T*x + beta*y,   or
+*
+*     y := alpha*A**H*x + beta*y,
+*
+*  where alpha and beta are scalars, x and y are vectors and A is an
+*  m by n matrix.
+*
+*  Arguments
+*  ==========
+*
+*  TRANS  - CHARACTER*1.
+*           On entry, TRANS specifies the operation to be performed as
+*           follows:
+*
+*              TRANS = 'N' or 'n'   y := alpha*A*x + beta*y.
+*
+*              TRANS = 'T' or 't'   y := alpha*A**T*x + beta*y.
+*
+*              TRANS = 'C' or 'c'   y := alpha*A**H*x + beta*y.
+*
+*           Unchanged on exit.
+*
+*  M      - INTEGER.
+*           On entry, M specifies the number of rows of the matrix A.
+*           M must be at least zero.
+*           Unchanged on exit.
+*
+*  N      - INTEGER.
+*           On entry, N specifies the number of columns of the matrix A.
+*           N must be at least zero.
+*           Unchanged on exit.
+*
+*  ALPHA  - COMPLEX         .
+*           On entry, ALPHA specifies the scalar alpha.
+*           Unchanged on exit.
+*
+*  A      - COMPLEX          array of DIMENSION ( LDA, n ).
+*           Before entry, the leading m by n part of the array A must
+*           contain the matrix of coefficients.
+*           Unchanged on exit.
+*
+*  LDA    - INTEGER.
+*           On entry, LDA specifies the first dimension of A as declared
+*           in the calling (sub) program. LDA must be at least
+*           max( 1, m ).
+*           Unchanged on exit.
+*
+*  X      - COMPLEX          array of DIMENSION at least
+*           ( 1 + ( n - 1 )*abs( INCX ) ) when TRANS = 'N' or 'n'
+*           and at least
+*           ( 1 + ( m - 1 )*abs( INCX ) ) otherwise.
+*           Before entry, the incremented array X must contain the
+*           vector x.
+*           Unchanged on exit.
+*
+*  INCX   - INTEGER.
+*           On entry, INCX specifies the increment for the elements of
+*           X. INCX must not be zero.
+*           Unchanged on exit.
+*
+*  BETA   - COMPLEX         .
+*           On entry, BETA specifies the scalar beta. When BETA is
+*           supplied as zero then Y need not be set on input.
+*           Unchanged on exit.
+*
+*  Y      - COMPLEX          array of DIMENSION at least
+*           ( 1 + ( m - 1 )*abs( INCY ) ) when TRANS = 'N' or 'n'
+*           and at least
+*           ( 1 + ( n - 1 )*abs( INCY ) ) otherwise.
+*           Before entry with BETA non-zero, the incremented array Y
+*           must contain the vector y. On exit, Y is overwritten by the
+*           updated vector y.
+*
+*  INCY   - INTEGER.
+*           On entry, INCY specifies the increment for the elements of
+*           Y. INCY must not be zero.
+*           Unchanged on exit.
+*
+*  Further Details
+*  ===============
+*
+*  Level 2 Blas routine.
+*  The vector and matrix arguments are not referenced when N = 0, or M = 0
+*
+*  -- Written on 22-October-1986.
+*     Jack Dongarra, Argonne National Lab.
+*     Jeremy Du Croz, Nag Central Office.
+*     Sven Hammarling, Nag Central Office.
+*     Richard Hanson, Sandia National Labs.
+*
+*  =====================================================================
+*
+*     .. Parameters ..
+      COMPLEX ONE
+      PARAMETER (ONE= (1.0E+0,0.0E+0))
+      COMPLEX ZERO
+      PARAMETER (ZERO= (0.0E+0,0.0E+0))
+*     ..
+*     .. Local Scalars ..
+      COMPLEX TEMP
+      INTEGER I,INFO,IX,IY,J,JX,JY,KX,KY,LENX,LENY
+      LOGICAL NOCONJ
+*     ..
+*     .. External Functions ..
+      LOGICAL LSAME
+      EXTERNAL LSAME
+*     ..
+*     .. External Subroutines ..
+      EXTERNAL XERBLA
+*     ..
+*     .. Intrinsic Functions ..
+      INTRINSIC CONJG,MAX
+*     ..
+*
+*     Test the input parameters.
+*
+      INFO = 0
+      IF (.NOT.LSAME(TRANS,'N') .AND. .NOT.LSAME(TRANS,'T') .AND.
+     +    .NOT.LSAME(TRANS,'C')) THEN
+          INFO = 1
+      ELSE IF (M.LT.0) THEN
+          INFO = 2
+      ELSE IF (N.LT.0) THEN
+          INFO = 3
+      ELSE IF (LDA.LT.MAX(1,M)) THEN
+          INFO = 6
+      ELSE IF (INCX.EQ.0) THEN
+          INFO = 8
+      ELSE IF (INCY.EQ.0) THEN
+          INFO = 11
+      END IF
+      IF (INFO.NE.0) THEN
+          CALL XERBLA('CGEMV ',INFO)
+          RETURN
+      END IF
+*
+*     Quick return if possible.
+*
+      IF ((M.EQ.0) .OR. (N.EQ.0) .OR.
+     +    ((ALPHA.EQ.ZERO).AND. (BETA.EQ.ONE))) RETURN
+*
+      NOCONJ = LSAME(TRANS,'T')
+*
+*     Set  LENX  and  LENY, the lengths of the vectors x and y, and set
+*     up the start points in  X  and  Y.
+*
+      IF (LSAME(TRANS,'N')) THEN
+          LENX = N
+          LENY = M
+      ELSE
+          LENX = M
+          LENY = N
+      END IF
+      IF (INCX.GT.0) THEN
+          KX = 1
+      ELSE
+          KX = 1 - (LENX-1)*INCX
+      END IF
+      IF (INCY.GT.0) THEN
+          KY = 1
+      ELSE
+          KY = 1 - (LENY-1)*INCY
+      END IF
+*
+*     Start the operations. In this version the elements of A are
+*     accessed sequentially with one pass through A.
+*
+*     First form  y := beta*y.
+*
+      IF (BETA.NE.ONE) THEN
+          IF (INCY.EQ.1) THEN
+              IF (BETA.EQ.ZERO) THEN
+                  DO 10 I = 1,LENY
+                      Y(I) = ZERO
+   10             CONTINUE
+              ELSE
+                  DO 20 I = 1,LENY
+                      Y(I) = BETA*Y(I)
+   20             CONTINUE
+              END IF
+          ELSE
+              IY = KY
+              IF (BETA.EQ.ZERO) THEN
+                  DO 30 I = 1,LENY
+                      Y(IY) = ZERO
+                      IY = IY + INCY
+   30             CONTINUE
+              ELSE
+                  DO 40 I = 1,LENY
+                      Y(IY) = BETA*Y(IY)
+                      IY = IY + INCY
+   40             CONTINUE
+              END IF
+          END IF
+      END IF
+      IF (ALPHA.EQ.ZERO) RETURN
+      IF (LSAME(TRANS,'N')) THEN
+*
+*        Form  y := alpha*A*x + y.
+*
+          JX = KX
+          IF (INCY.EQ.1) THEN
+              DO 60 J = 1,N
+                  IF (X(JX).NE.ZERO) THEN
+                      TEMP = ALPHA*X(JX)
+                      DO 50 I = 1,M
+                          Y(I) = Y(I) + TEMP*A(I,J)
+   50                 CONTINUE
+                  END IF
+                  JX = JX + INCX
+   60         CONTINUE
+          ELSE
+              DO 80 J = 1,N
+                  IF (X(JX).NE.ZERO) THEN
+                      TEMP = ALPHA*X(JX)
+                      IY = KY
+                      DO 70 I = 1,M
+                          Y(IY) = Y(IY) + TEMP*A(I,J)
+                          IY = IY + INCY
+   70                 CONTINUE
+                  END IF
+                  JX = JX + INCX
+   80         CONTINUE
+          END IF
+      ELSE
+*
+*        Form  y := alpha*A**T*x + y  or  y := alpha*A**H*x + y.
+*
+          JY = KY
+          IF (INCX.EQ.1) THEN
+              DO 110 J = 1,N
+                  TEMP = ZERO
+                  IF (NOCONJ) THEN
+                      DO 90 I = 1,M
+                          TEMP = TEMP + A(I,J)*X(I)
+   90                 CONTINUE
+                  ELSE
+                      DO 100 I = 1,M
+                          TEMP = TEMP + CONJG(A(I,J))*X(I)
+  100                 CONTINUE
+                  END IF
+                  Y(JY) = Y(JY) + ALPHA*TEMP
+                  JY = JY + INCY
+  110         CONTINUE
+          ELSE
+              DO 140 J = 1,N
+                  TEMP = ZERO
+                  IX = KX
+                  IF (NOCONJ) THEN
+                      DO 120 I = 1,M
+                          TEMP = TEMP + A(I,J)*X(IX)
+                          IX = IX + INCX
+  120                 CONTINUE
+                  ELSE
+                      DO 130 I = 1,M
+                          TEMP = TEMP + CONJG(A(I,J))*X(IX)
+                          IX = IX + INCX
+  130                 CONTINUE
+                  END IF
+                  Y(JY) = Y(JY) + ALPHA*TEMP
+                  JY = JY + INCY
+  140         CONTINUE
+          END IF
+      END IF
+*
+      RETURN
+*
+*     End of CGEMV .
+*
+      END

interface/netlib/dgemv.f

@@ -0,0 +1,265 @@
+      SUBROUTINE DGEMV(TRANS,M,N,ALPHA,A,LDA,X,INCX,BETA,Y,INCY)
+*     .. Scalar Arguments ..
+      DOUBLE PRECISION ALPHA,BETA
+      INTEGER INCX,INCY,LDA,M,N
+      CHARACTER TRANS
+*     ..
+*     .. Array Arguments ..
+      DOUBLE PRECISION A(LDA,*),X(*),Y(*)
+*     ..
+*
+*  Purpose
+*  =======
+*
+*  DGEMV  performs one of the matrix-vector operations
+*
+*     y := alpha*A*x + beta*y,   or   y := alpha*A**T*x + beta*y,
+*
+*  where alpha and beta are scalars, x and y are vectors and A is an
+*  m by n matrix.
+*
+*  Arguments
+*  ==========
+*
+*  TRANS  - CHARACTER*1.
+*           On entry, TRANS specifies the operation to be performed as
+*           follows:
+*
+*              TRANS = 'N' or 'n'   y := alpha*A*x + beta*y.
+*
+*              TRANS = 'T' or 't'   y := alpha*A**T*x + beta*y.
+*
+*              TRANS = 'C' or 'c'   y := alpha*A**T*x + beta*y.
+*
+*           Unchanged on exit.
+*
+*  M      - INTEGER.
+*           On entry, M specifies the number of rows of the matrix A.
+*           M must be at least zero.
+*           Unchanged on exit.
+*
+*  N      - INTEGER.
+*           On entry, N specifies the number of columns of the matrix A.
+*           N must be at least zero.
+*           Unchanged on exit.
+*
+*  ALPHA  - DOUBLE PRECISION.
+*           On entry, ALPHA specifies the scalar alpha.
+*           Unchanged on exit.
+*
+*  A      - DOUBLE PRECISION array of DIMENSION ( LDA, n ).
+*           Before entry, the leading m by n part of the array A must
+*           contain the matrix of coefficients.
+*           Unchanged on exit.
+*
+*  LDA    - INTEGER.
+*           On entry, LDA specifies the first dimension of A as declared
+*           in the calling (sub) program. LDA must be at least
+*           max( 1, m ).
+*           Unchanged on exit.
+*
+*  X      - DOUBLE PRECISION array of DIMENSION at least
+*           ( 1 + ( n - 1 )*abs( INCX ) ) when TRANS = 'N' or 'n'
+*           and at least
+*           ( 1 + ( m - 1 )*abs( INCX ) ) otherwise.
+*           Before entry, the incremented array X must contain the
+*           vector x.
+*           Unchanged on exit.
+*
+*  INCX   - INTEGER.
+*           On entry, INCX specifies the increment for the elements of
+*           X. INCX must not be zero.
+*           Unchanged on exit.
+*
+*  BETA   - DOUBLE PRECISION.
+*           On entry, BETA specifies the scalar beta. When BETA is
+*           supplied as zero then Y need not be set on input.
+*           Unchanged on exit.
+*
+*  Y      - DOUBLE PRECISION array of DIMENSION at least
+*           ( 1 + ( m - 1 )*abs( INCY ) ) when TRANS = 'N' or 'n'
+*           and at least
+*           ( 1 + ( n - 1 )*abs( INCY ) ) otherwise.
+*           Before entry with BETA non-zero, the incremented array Y
+*           must contain the vector y. On exit, Y is overwritten by the
+*           updated vector y.
+*
+*  INCY   - INTEGER.
+*           On entry, INCY specifies the increment for the elements of
+*           Y. INCY must not be zero.
+*           Unchanged on exit.
+*
+*  Further Details
+*  ===============
+*
+*  Level 2 Blas routine.
+*  The vector and matrix arguments are not referenced when N = 0, or M = 0
+*
+*  -- Written on 22-October-1986.
+*     Jack Dongarra, Argonne National Lab.
+*     Jeremy Du Croz, Nag Central Office.
+*     Sven Hammarling, Nag Central Office.
+*     Richard Hanson, Sandia National Labs.
+*
+*  =====================================================================
+*
+*     .. Parameters ..
+      DOUBLE PRECISION ONE,ZERO
+      PARAMETER (ONE=1.0D+0,ZERO=0.0D+0)
+*     ..
+*     .. Local Scalars ..
+      DOUBLE PRECISION TEMP
+      INTEGER I,INFO,IX,IY,J,JX,JY,KX,KY,LENX,LENY
+*     ..
+*     .. External Functions ..
+      LOGICAL LSAME
+      EXTERNAL LSAME
+*     ..
+*     .. External Subroutines ..
+      EXTERNAL XERBLA
+*     ..
+*     .. Intrinsic Functions ..
+      INTRINSIC MAX
+*     ..
+*
+*     Test the input parameters.
+*
+      INFO = 0
+      IF (.NOT.LSAME(TRANS,'N') .AND. .NOT.LSAME(TRANS,'T') .AND.
+     +    .NOT.LSAME(TRANS,'C')) THEN
+          INFO = 1
+      ELSE IF (M.LT.0) THEN
+          INFO = 2
+      ELSE IF (N.LT.0) THEN
+          INFO = 3
+      ELSE IF (LDA.LT.MAX(1,M)) THEN
+          INFO = 6
+      ELSE IF (INCX.EQ.0) THEN
+          INFO = 8
+      ELSE IF (INCY.EQ.0) THEN
+          INFO = 11
+      END IF
+      IF (INFO.NE.0) THEN
+          CALL XERBLA('DGEMV ',INFO)
+          RETURN
+      END IF
+*
+*     Quick return if possible.
+*
+      IF ((M.EQ.0) .OR. (N.EQ.0) .OR.
+     +    ((ALPHA.EQ.ZERO).AND. (BETA.EQ.ONE))) RETURN
+*
+*     Set  LENX  and  LENY, the lengths of the vectors x and y, and set
+*     up the start points in  X  and  Y.
+*
+      IF (LSAME(TRANS,'N')) THEN
+          LENX = N
+          LENY = M
+      ELSE
+          LENX = M
+          LENY = N
+      END IF
+      IF (INCX.GT.0) THEN
+          KX = 1
+      ELSE
+          KX = 1 - (LENX-1)*INCX
+      END IF
+      IF (INCY.GT.0) THEN
+          KY = 1
+      ELSE
+          KY = 1 - (LENY-1)*INCY
+      END IF
+*
+*     Start the operations. In this version the elements of A are
+*     accessed sequentially with one pass through A.
+*
+*     First form  y := beta*y.
+*
+      IF (BETA.NE.ONE) THEN
+          IF (INCY.EQ.1) THEN
+              IF (BETA.EQ.ZERO) THEN
+                  DO 10 I = 1,LENY
+                      Y(I) = ZERO
+   10             CONTINUE
+              ELSE
+                  DO 20 I = 1,LENY
+                      Y(I) = BETA*Y(I)
+   20             CONTINUE
+              END IF
+          ELSE
+              IY = KY
+              IF (BETA.EQ.ZERO) THEN
+                  DO 30 I = 1,LENY
+                      Y(IY) = ZERO
+                      IY = IY + INCY
+   30             CONTINUE
+              ELSE
+                  DO 40 I = 1,LENY
+                      Y(IY) = BETA*Y(IY)
+                      IY = IY + INCY
+   40             CONTINUE
+              END IF
+          END IF
+      END IF
+      IF (ALPHA.EQ.ZERO) RETURN
+      IF (LSAME(TRANS,'N')) THEN
+*
+*        Form  y := alpha*A*x + y.
+*
+          JX = KX
+          IF (INCY.EQ.1) THEN
+              DO 60 J = 1,N
+                  IF (X(JX).NE.ZERO) THEN
+                      TEMP = ALPHA*X(JX)
+                      DO 50 I = 1,M
+                          Y(I) = Y(I) + TEMP*A(I,J)
+   50                 CONTINUE
+                  END IF
+                  JX = JX + INCX
+   60         CONTINUE
+          ELSE
+              DO 80 J = 1,N
+                  IF (X(JX).NE.ZERO) THEN
+                      TEMP = ALPHA*X(JX)
+                      IY = KY
+                      DO 70 I = 1,M
+                          Y(IY) = Y(IY) + TEMP*A(I,J)
+                          IY = IY + INCY
+   70                 CONTINUE
+                  END IF
+                  JX = JX + INCX
+   80         CONTINUE
+          END IF
+      ELSE
+*
+*        Form  y := alpha*A**T*x + y.
+*
+          JY = KY
+          IF (INCX.EQ.1) THEN
+              DO 100 J = 1,N
+                  TEMP = ZERO
+                  DO 90 I = 1,M
+                      TEMP = TEMP + A(I,J)*X(I)
+   90             CONTINUE
+                  Y(JY) = Y(JY) + ALPHA*TEMP
+                  JY = JY + INCY
+  100         CONTINUE
+          ELSE
+              DO 120 J = 1,N
+                  TEMP = ZERO
+                  IX = KX
+                  DO 110 I = 1,M
+                      TEMP = TEMP + A(I,J)*X(IX)
+                      IX = IX + INCX
+  110             CONTINUE
+                  Y(JY) = Y(JY) + ALPHA*TEMP
+                  JY = JY + INCY
+  120         CONTINUE
+          END IF
+      END IF
+*
+      RETURN
+*
+*     End of DGEMV .
+*
+      END

interface/netlib/sgemv.f

@@ -0,0 +1,265 @@
+      SUBROUTINE SGEMV(TRANS,M,N,ALPHA,A,LDA,X,INCX,BETA,Y,INCY)
+*     .. Scalar Arguments ..
+      REAL ALPHA,BETA
+      INTEGER INCX,INCY,LDA,M,N
+      CHARACTER TRANS
+*     ..
+*     .. Array Arguments ..
+      REAL A(LDA,*),X(*),Y(*)
+*     ..
+*
+*  Purpose
+*  =======
+*
+*  SGEMV  performs one of the matrix-vector operations
+*
+*     y := alpha*A*x + beta*y,   or   y := alpha*A**T*x + beta*y,
+*
+*  where alpha and beta are scalars, x and y are vectors and A is an
+*  m by n matrix.
+*
+*  Arguments
+*  ==========
+*
+*  TRANS  - CHARACTER*1.
+*           On entry, TRANS specifies the operation to be performed as
+*           follows:
+*
+*              TRANS = 'N' or 'n'   y := alpha*A*x + beta*y.
+*
+*              TRANS = 'T' or 't'   y := alpha*A**T*x + beta*y.
+*
+*              TRANS = 'C' or 'c'   y := alpha*A**T*x + beta*y.
+*
+*           Unchanged on exit.
+*
+*  M      - INTEGER.
+*           On entry, M specifies the number of rows of the matrix A.
+*           M must be at least zero.
+*           Unchanged on exit.
+*
+*  N      - INTEGER.
+*           On entry, N specifies the number of columns of the matrix A.
+*           N must be at least zero.
+*           Unchanged on exit.
+*
+*  ALPHA  - REAL            .
+*           On entry, ALPHA specifies the scalar alpha.
+*           Unchanged on exit.
+*
+*  A      - REAL             array of DIMENSION ( LDA, n ).
+*           Before entry, the leading m by n part of the array A must
+*           contain the matrix of coefficients.
+*           Unchanged on exit.
+*
+*  LDA    - INTEGER.
+*           On entry, LDA specifies the first dimension of A as declared
+*           in the calling (sub) program. LDA must be at least
+*           max( 1, m ).
+*           Unchanged on exit.
+*
+*  X      - REAL             array of DIMENSION at least
+*           ( 1 + ( n - 1 )*abs( INCX ) ) when TRANS = 'N' or 'n'
+*           and at least
+*           ( 1 + ( m - 1 )*abs( INCX ) ) otherwise.
+*           Before entry, the incremented array X must contain the
+*           vector x.
+*           Unchanged on exit.
+*
+*  INCX   - INTEGER.
+*           On entry, INCX specifies the increment for the elements of
+*           X. INCX must not be zero.
+*           Unchanged on exit.
+*
+*  BETA   - REAL            .
+*           On entry, BETA specifies the scalar beta. When BETA is
+*           supplied as zero then Y need not be set on input.
+*           Unchanged on exit.
+*
+*  Y      - REAL             array of DIMENSION at least
+*           ( 1 + ( m - 1 )*abs( INCY ) ) when TRANS = 'N' or 'n'
+*           and at least
+*           ( 1 + ( n - 1 )*abs( INCY ) ) otherwise.
+*           Before entry with BETA non-zero, the incremented array Y
+*           must contain the vector y. On exit, Y is overwritten by the
+*           updated vector y.
+*
+*  INCY   - INTEGER.
+*           On entry, INCY specifies the increment for the elements of
+*           Y. INCY must not be zero.
+*           Unchanged on exit.
+*
+*  Further Details
+*  ===============
+*
+*  Level 2 Blas routine.
+*  The vector and matrix arguments are not referenced when N = 0, or M = 0
+*
+*  -- Written on 22-October-1986.
+*     Jack Dongarra, Argonne National Lab.
+*     Jeremy Du Croz, Nag Central Office.
+*     Sven Hammarling, Nag Central Office.
+*     Richard Hanson, Sandia National Labs.
+*
+*  =====================================================================
+*
+*     .. Parameters ..
+      REAL ONE,ZERO
+      PARAMETER (ONE=1.0E+0,ZERO=0.0E+0)
+*     ..
+*     .. Local Scalars ..
+      REAL TEMP
+      INTEGER I,INFO,IX,IY,J,JX,JY,KX,KY,LENX,LENY
+*     ..
+*     .. External Functions ..
+      LOGICAL LSAME
+      EXTERNAL LSAME
+*     ..
+*     .. External Subroutines ..
+      EXTERNAL XERBLA
+*     ..
+*     .. Intrinsic Functions ..
+      INTRINSIC MAX
+*     ..
+*
+*     Test the input parameters.
+*
+      INFO = 0
+      IF (.NOT.LSAME(TRANS,'N') .AND. .NOT.LSAME(TRANS,'T') .AND.
+     +    .NOT.LSAME(TRANS,'C')) THEN
+          INFO = 1
+      ELSE IF (M.LT.0) THEN
+          INFO = 2
+      ELSE IF (N.LT.0) THEN
+          INFO = 3
+      ELSE IF (LDA.LT.MAX(1,M)) THEN
+          INFO = 6
+      ELSE IF (INCX.EQ.0) THEN
+          INFO = 8
+      ELSE IF (INCY.EQ.0) THEN
+          INFO = 11
+      END IF
+      IF (INFO.NE.0) THEN
+          CALL XERBLA('SGEMV ',INFO)
+          RETURN
+      END IF
+*
+*     Quick return if possible.
+*
+      IF ((M.EQ.0) .OR. (N.EQ.0) .OR.
+     +    ((ALPHA.EQ.ZERO).AND. (BETA.EQ.ONE))) RETURN
+*
+*     Set  LENX  and  LENY, the lengths of the vectors x and y, and set
+*     up the start points in  X  and  Y.
+*
+      IF (LSAME(TRANS,'N')) THEN
+          LENX = N
+          LENY = M
+      ELSE
+          LENX = M
+          LENY = N
+      END IF
+      IF (INCX.GT.0) THEN
+          KX = 1
+      ELSE
+          KX = 1 - (LENX-1)*INCX
+      END IF
+      IF (INCY.GT.0) THEN
+          KY = 1
+      ELSE
+          KY = 1 - (LENY-1)*INCY
+      END IF
+*
+*     Start the operations. In this version the elements of A are
+*     accessed sequentially with one pass through A.
+*
+*     First form  y := beta*y.
+*
+      IF (BETA.NE.ONE) THEN
+          IF (INCY.EQ.1) THEN
+              IF (BETA.EQ.ZERO) THEN
+                  DO 10 I = 1,LENY
+                      Y(I) = ZERO
+   10             CONTINUE
+              ELSE
+                  DO 20 I = 1,LENY
+                      Y(I) = BETA*Y(I)
+   20             CONTINUE
+              END IF
+          ELSE
+              IY = KY
+              IF (BETA.EQ.ZERO) THEN
+                  DO 30 I = 1,LENY
+                      Y(IY) = ZERO
+                      IY = IY + INCY
+   30             CONTINUE
+              ELSE
+                  DO 40 I = 1,LENY
+                      Y(IY) = BETA*Y(IY)
+                      IY = IY + INCY
+   40             CONTINUE
+              END IF
+          END IF
+      END IF
+      IF (ALPHA.EQ.ZERO) RETURN
+      IF (LSAME(TRANS,'N')) THEN
+*
+*        Form  y := alpha*A*x + y.
+*
+          JX = KX
+          IF (INCY.EQ.1) THEN
+              DO 60 J = 1,N
+                  IF (X(JX).NE.ZERO) THEN
+                      TEMP = ALPHA*X(JX)
+                      DO 50 I = 1,M
+                          Y(I) = Y(I) + TEMP*A(I,J)
+   50                 CONTINUE
+                  END IF
+                  JX = JX + INCX
+   60         CONTINUE
+          ELSE
+              DO 80 J = 1,N
+                  IF (X(JX).NE.ZERO) THEN
+                      TEMP = ALPHA*X(JX)
+                      IY = KY
+                      DO 70 I = 1,M
+                          Y(IY) = Y(IY) + TEMP*A(I,J)
+                          IY = IY + INCY
+   70                 CONTINUE
+                  END IF
+                  JX = JX + INCX
+   80         CONTINUE
+          END IF
+      ELSE
+*
+*        Form  y := alpha*A**T*x + y.
+*
+          JY = KY
+          IF (INCX.EQ.1) THEN
+              DO 100 J = 1,N
+                  TEMP = ZERO
+                  DO 90 I = 1,M
+                      TEMP = TEMP + A(I,J)*X(I)
+   90             CONTINUE
+                  Y(JY) = Y(JY) + ALPHA*TEMP
+                  JY = JY + INCY
+  100         CONTINUE
+          ELSE
+              DO 120 J = 1,N
+                  TEMP = ZERO
+                  IX = KX
+                  DO 110 I = 1,M
+                      TEMP = TEMP + A(I,J)*X(IX)
+                      IX = IX + INCX
+  110             CONTINUE
+                  Y(JY) = Y(JY) + ALPHA*TEMP
+                  JY = JY + INCY
+  120         CONTINUE
+          END IF
+      END IF
+*
+      RETURN
+*
+*     End of SGEMV .
+*
+      END

interface/netlib/zgemv.f

@@ -0,0 +1,285 @@
+      SUBROUTINE ZGEMV(TRANS,M,N,ALPHA,A,LDA,X,INCX,BETA,Y,INCY)
+*     .. Scalar Arguments ..
+      DOUBLE COMPLEX ALPHA,BETA
+      INTEGER INCX,INCY,LDA,M,N
+      CHARACTER TRANS
+*     ..
+*     .. Array Arguments ..
+      DOUBLE COMPLEX A(LDA,*),X(*),Y(*)
+*     ..
+*
+*  Purpose
+*  =======
+*
+*  ZGEMV  performs one of the matrix-vector operations
+*
+*     y := alpha*A*x + beta*y,   or   y := alpha*A**T*x + beta*y,   or
+*
+*     y := alpha*A**H*x + beta*y,
+*
+*  where alpha and beta are scalars, x and y are vectors and A is an
+*  m by n matrix.
+*
+*  Arguments
+*  ==========
+*
+*  TRANS  - CHARACTER*1.
+*           On entry, TRANS specifies the operation to be performed as
+*           follows:
+*
+*              TRANS = 'N' or 'n'   y := alpha*A*x + beta*y.
+*
+*              TRANS = 'T' or 't'   y := alpha*A**T*x + beta*y.
+*
+*              TRANS = 'C' or 'c'   y := alpha*A**H*x + beta*y.
+*
+*           Unchanged on exit.
+*
+*  M      - INTEGER.
+*           On entry, M specifies the number of rows of the matrix A.
+*           M must be at least zero.
+*           Unchanged on exit.
+*
+*  N      - INTEGER.
+*           On entry, N specifies the number of columns of the matrix A.
+*           N must be at least zero.
+*           Unchanged on exit.
+*
+*  ALPHA  - COMPLEX*16      .
+*           On entry, ALPHA specifies the scalar alpha.
+*           Unchanged on exit.
+*
+*  A      - COMPLEX*16       array of DIMENSION ( LDA, n ).
+*           Before entry, the leading m by n part of the array A must
+*           contain the matrix of coefficients.
+*           Unchanged on exit.
+*
+*  LDA    - INTEGER.
+*           On entry, LDA specifies the first dimension of A as declared
+*           in the calling (sub) program. LDA must be at least
+*           max( 1, m ).
+*           Unchanged on exit.
+*
+*  X      - COMPLEX*16       array of DIMENSION at least
+*           ( 1 + ( n - 1 )*abs( INCX ) ) when TRANS = 'N' or 'n'
+*           and at least
+*           ( 1 + ( m - 1 )*abs( INCX ) ) otherwise.
+*           Before entry, the incremented array X must contain the
+*           vector x.
+*           Unchanged on exit.
+*
+*  INCX   - INTEGER.
+*           On entry, INCX specifies the increment for the elements of
+*           X. INCX must not be zero.
+*           Unchanged on exit.
+*
+*  BETA   - COMPLEX*16      .
+*           On entry, BETA specifies the scalar beta. When BETA is
+*           supplied as zero then Y need not be set on input.
+*           Unchanged on exit.
+*
+*  Y      - COMPLEX*16       array of DIMENSION at least
+*           ( 1 + ( m - 1 )*abs( INCY ) ) when TRANS = 'N' or 'n'
+*           and at least
+*           ( 1 + ( n - 1 )*abs( INCY ) ) otherwise.
+*           Before entry with BETA non-zero, the incremented array Y
+*           must contain the vector y. On exit, Y is overwritten by the
+*           updated vector y.
+*
+*  INCY   - INTEGER.
+*           On entry, INCY specifies the increment for the elements of
+*           Y. INCY must not be zero.
+*           Unchanged on exit.
+*
+*  Further Details
+*  ===============
+*
+*  Level 2 Blas routine.
+*  The vector and matrix arguments are not referenced when N = 0, or M = 0
+*
+*  -- Written on 22-October-1986.
+*     Jack Dongarra, Argonne National Lab.
+*     Jeremy Du Croz, Nag Central Office.
+*     Sven Hammarling, Nag Central Office.
+*     Richard Hanson, Sandia National Labs.
+*
+*  =====================================================================
+*
+*     .. Parameters ..
+      DOUBLE COMPLEX ONE
+      PARAMETER (ONE= (1.0D+0,0.0D+0))
+      DOUBLE COMPLEX ZERO
+      PARAMETER (ZERO= (0.0D+0,0.0D+0))
+*     ..
+*     .. Local Scalars ..
+      DOUBLE COMPLEX TEMP
+      INTEGER I,INFO,IX,IY,J,JX,JY,KX,KY,LENX,LENY
+      LOGICAL NOCONJ
+*     ..
+*     .. External Functions ..
+      LOGICAL LSAME
+      EXTERNAL LSAME
+*     ..
+*     .. External Subroutines ..
+      EXTERNAL XERBLA
+*     ..
+*     .. Intrinsic Functions ..
+      INTRINSIC DCONJG,MAX
+*     ..
+*
+*     Test the input parameters.
+*
+      INFO = 0
+      IF (.NOT.LSAME(TRANS,'N') .AND. .NOT.LSAME(TRANS,'T') .AND.
+     +    .NOT.LSAME(TRANS,'C')) THEN
+          INFO = 1
+      ELSE IF (M.LT.0) THEN
+          INFO = 2
+      ELSE IF (N.LT.0) THEN
+          INFO = 3
+      ELSE IF (LDA.LT.MAX(1,M)) THEN
+          INFO = 6
+      ELSE IF (INCX.EQ.0) THEN
+          INFO = 8
+      ELSE IF (INCY.EQ.0) THEN
+          INFO = 11
+      END IF
+      IF (INFO.NE.0) THEN
+          CALL XERBLA('ZGEMV ',INFO)
+          RETURN
+      END IF
+*
+*     Quick return if possible.
+*
+      IF ((M.EQ.0) .OR. (N.EQ.0) .OR.
+     +    ((ALPHA.EQ.ZERO).AND. (BETA.EQ.ONE))) RETURN
+*
+      NOCONJ = LSAME(TRANS,'T')
+*
+*     Set  LENX  and  LENY, the lengths of the vectors x and y, and set
+*     up the start points in  X  and  Y.
+*
+      IF (LSAME(TRANS,'N')) THEN
+          LENX = N
+          LENY = M
+      ELSE
+          LENX = M
+          LENY = N
+      END IF
+      IF (INCX.GT.0) THEN
+          KX = 1
+      ELSE
+          KX = 1 - (LENX-1)*INCX
+      END IF
+      IF (INCY.GT.0) THEN
+          KY = 1
+      ELSE
+          KY = 1 - (LENY-1)*INCY
+      END IF
+*
+*     Start the operations. In this version the elements of A are
+*     accessed sequentially with one pass through A.
+*
+*     First form  y := beta*y.
+*
+      IF (BETA.NE.ONE) THEN
+          IF (INCY.EQ.1) THEN
+              IF (BETA.EQ.ZERO) THEN
+                  DO 10 I = 1,LENY
+                      Y(I) = ZERO
+   10             CONTINUE
+              ELSE
+                  DO 20 I = 1,LENY
+                      Y(I) = BETA*Y(I)
+   20             CONTINUE
+              END IF
+          ELSE
+              IY = KY
+              IF (BETA.EQ.ZERO) THEN
+                  DO 30 I = 1,LENY
+                      Y(IY) = ZERO
+                      IY = IY + INCY
+   30             CONTINUE
+              ELSE
+                  DO 40 I = 1,LENY
+                      Y(IY) = BETA*Y(IY)
+                      IY = IY + INCY
+   40             CONTINUE
+              END IF
+          END IF
+      END IF
+      IF (ALPHA.EQ.ZERO) RETURN
+      IF (LSAME(TRANS,'N')) THEN
+*
+*        Form  y := alpha*A*x + y.
+*
+          JX = KX
+          IF (INCY.EQ.1) THEN
+              DO 60 J = 1,N
+                  IF (X(JX).NE.ZERO) THEN
+                      TEMP = ALPHA*X(JX)
+                      DO 50 I = 1,M
+                          Y(I) = Y(I) + TEMP*A(I,J)
+   50                 CONTINUE
+                  END IF
+                  JX = JX + INCX
+   60         CONTINUE
+          ELSE
+              DO 80 J = 1,N
+                  IF (X(JX).NE.ZERO) THEN
+                      TEMP = ALPHA*X(JX)
+                      IY = KY
+                      DO 70 I = 1,M
+                          Y(IY) = Y(IY) + TEMP*A(I,J)
+                          IY = IY + INCY
+   70                 CONTINUE
+                  END IF
+                  JX = JX + INCX
+   80         CONTINUE
+          END IF
+      ELSE
+*
+*        Form  y := alpha*A**T*x + y  or  y := alpha*A**H*x + y.
+*
+          JY = KY
+          IF (INCX.EQ.1) THEN
+              DO 110 J = 1,N
+                  TEMP = ZERO
+                  IF (NOCONJ) THEN
+                      DO 90 I = 1,M
+                          TEMP = TEMP + A(I,J)*X(I)
+   90                 CONTINUE
+                  ELSE
+                      DO 100 I = 1,M
+                          TEMP = TEMP + DCONJG(A(I,J))*X(I)
+  100                 CONTINUE
+                  END IF
+                  Y(JY) = Y(JY) + ALPHA*TEMP
+                  JY = JY + INCY
+  110         CONTINUE
+          ELSE
+              DO 140 J = 1,N
+                  TEMP = ZERO
+                  IX = KX
+                  IF (NOCONJ) THEN
+                      DO 120 I = 1,M
+                          TEMP = TEMP + A(I,J)*X(IX)
+                          IX = IX + INCX
+  120                 CONTINUE
+                  ELSE
+                      DO 130 I = 1,M
+                          TEMP = TEMP + DCONJG(A(I,J))*X(IX)
+                          IX = IX + INCX
+  130                 CONTINUE
+                  END IF
+                  Y(JY) = Y(JY) + ALPHA*TEMP
+                  JY = JY + INCY
+  140         CONTINUE
+          END IF
+      END IF
+*
+      RETURN
+*
+*     End of ZGEMV .
+*
+      END

interface/symm.c

@@ -136,6 +136,7 @@ void NAME(char *SIDE, char *UPLO,
   FLOAT *sa, *sb;
 
 #ifdef SMP
+#ifndef COMPLEX
 #ifdef XDOUBLE
   int mode  =  BLAS_XDOUBLE | BLAS_REAL;
 #elif defined(DOUBLE)
@@ -143,6 +144,15 @@ void NAME(char *SIDE, char *UPLO,
 #else
   int mode  =  BLAS_SINGLE  | BLAS_REAL;
 #endif  
+#else
+#ifdef XDOUBLE
+  int mode  =  BLAS_XDOUBLE | BLAS_COMPLEX;
+#elif defined(DOUBLE)
+  int mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
+#else
+  int mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
+#endif
+#endif
 #endif
 
 #if defined(SMP) && !defined(NO_AFFINITY)
@@ -237,6 +247,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_SIDE Side, enum CBLAS_UPLO Uplo,
   FLOAT *sa, *sb;
 
 #ifdef SMP
+#ifndef COMPLEX
 #ifdef XDOUBLE
   int mode  =  BLAS_XDOUBLE | BLAS_REAL;
 #elif defined(DOUBLE)
@@ -244,6 +255,15 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_SIDE Side, enum CBLAS_UPLO Uplo,
 #else
   int mode  =  BLAS_SINGLE  | BLAS_REAL;
 #endif  
+#else
+#ifdef XDOUBLE
+  int mode  =  BLAS_XDOUBLE | BLAS_COMPLEX;
+#elif defined(DOUBLE)
+  int mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
+#else
+  int mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
+#endif  
+#endif
 #endif
 
 #if defined(SMP) && !defined(NO_AFFINITY)

kernel/Makefile

@@ -6,7 +6,7 @@ TOPDIR	= ..
 include $(TOPDIR)/Makefile.system
 
 ifdef TARGET_CORE
-CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
+override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
 BUILD_KERNEL = 1
 KDIR =
 TSUFFIX = _$(TARGET_CORE)
@@ -48,7 +48,7 @@ HPLOBJS = \
 
 COMMONOBJS	+= lsame.$(SUFFIX) scabs1.$(SUFFIX) dcabs1.$(SUFFIX)
 
-ifdef DYNAMIC_ARCH
+ifeq ($(DYNAMIC_ARCH), 1)
 SBLASOBJS	+= setparam$(TSUFFIX).$(SUFFIX)
 CCOMMON_OPT	+= -DTS=$(TSUFFIX)
 endif

kernel/Makefile.L3

@@ -498,6 +498,91 @@ $(KDIR)xgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMD
 $(KDIR)xgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
 	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DCC $< -o $@
 
+ifeq ($(TARGET), LOONGSON3B)												 
+$(KDIR)strmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+
+$(KDIR)strmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+
+$(KDIR)strmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+
+$(KDIR)strmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+
+$(KDIR)dtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+
+$(KDIR)dtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+
+$(KDIR)dtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+
+$(KDIR)dtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+
+$(KDIR)qtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+
+$(KDIR)qtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+
+$(KDIR)qtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+
+$(KDIR)qtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+
+$(KDIR)ctrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ctrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ctrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+
+$(KDIR)ctrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+
+$(KDIR)ctrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ctrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ctrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+
+$(KDIR)ctrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+
+$(KDIR)ztrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ztrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ztrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+
+$(KDIR)ztrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+
+$(KDIR)ztrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ztrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+
+$(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+else
 $(KDIR)strmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
@@ -581,6 +666,7 @@ $(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
 
 $(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+endif
 
 $(KDIR)xtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@

kernel/generic/gemmkernel_2x2.c

@@ -0,0 +1,157 @@
+#include "common.h"
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
+#ifdef TRMMKERNEL
+		,BLASLONG offset
+#endif
+		) 
+{
+   BLASLONG i,j,k;
+   FLOAT *C0,*C1,*ptrba,*ptrbb;
+   FLOAT res0,res1,res2,res3,load0,load1,load2,load3,load4,load5,load6,load7;
+   for (j=0; j<bn/2; j+=1) 
+     {
+        C0 = C;
+        C1 = C0+ldc;
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+             for (k=0; k<bk/4; k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  load4 = ptrba[2*1+0];
+                  load5 = ptrbb[2*1+0];
+                  res0 = res0+load4*load5;
+                  load6 = ptrba[2*1+1];
+                  res1 = res1+load6*load5;
+                  load7 = ptrbb[2*1+1];
+                  res2 = res2+load4*load7;
+                  res3 = res3+load6*load7;
+                  load0 = ptrba[2*2+0];
+                  load1 = ptrbb[2*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*2+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*2+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  load4 = ptrba[2*3+0];
+                  load5 = ptrbb[2*3+0];
+                  res0 = res0+load4*load5;
+                  load6 = ptrba[2*3+1];
+                  res1 = res1+load6*load5;
+                  load7 = ptrbb[2*3+1];
+                  res2 = res2+load4*load7;
+                  res3 = res3+load6*load7;
+                  ptrba = ptrba+8;
+                  ptrbb = ptrbb+8;
+               }
+             for (k=0; k<(bk&3); k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+2;
+               }
+             res0 = res0*alpha;
+             C0[0] = C0[0]+res0;
+             res1 = res1*alpha;
+             C0[1] = C0[1]+res1;
+             res2 = res2*alpha;
+             C1[0] = C1[0]+res2;
+             res3 = res3*alpha;
+             C1[1] = C1[1]+res3;
+             C0 = C0+2;
+             C1 = C1+2;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+                  load0 = ptrba[0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrbb[2*0+1];
+                  res1 = res1+load0*load2;
+                  ptrba = ptrba+1;
+                  ptrbb = ptrbb+2;
+               }
+             res0 = res0*alpha;
+             C0[0] = C0[0]+res0;
+             res1 = res1*alpha;
+             C1[0] = C1[0]+res1;
+             C0 = C0+1;
+             C1 = C1+1;
+          }
+        k = (bk<<1);
+        bb = bb+k;
+        i = (ldc<<1);
+        C = C+i;
+     }
+   for (j=0; j<(bn&1); j+=1) 
+     {
+        C0 = C;
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+1;
+               }
+             res0 = res0*alpha;
+             C0[0] = C0[0]+res0;
+             res1 = res1*alpha;
+             C0[1] = C0[1]+res1;
+             C0 = C0+2;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+                  load0 = ptrba[0+0];
+                  load1 = ptrbb[0+0];
+                  res0 = res0+load0*load1;
+                  ptrba = ptrba+1;
+                  ptrbb = ptrbb+1;
+               }
+             res0 = res0*alpha;
+             C0[0] = C0[0]+res0;
+             C0 = C0+1;
+          }
+        k = (bk<<0);
+        bb = bb+k;
+        C = C+ldc;
+     }
+   return 0;
+}

kernel/generic/trmmkernel_2x2.c

@@ -0,0 +1,280 @@
+#include "common.h"
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
+#ifdef TRMMKERNEL
+		,BLASLONG offset
+#endif
+		) 
+{
+   BLASLONG i,j,k;
+   FLOAT *C0,*C1,*ptrba,*ptrbb;
+   FLOAT res0,res1,res2,res3,load0,load1,load2,load3,load4,load5,load6,load7;
+   BLASLONG off, temp;
+#if defined(TRMMKERNEL) && !defined(LEFT)
+   off = -offset; 
+#endif
+   for (j=0; j<bn/2; j+=1) 
+     {
+        C0 = C;
+        C1 = C0+ldc;
+#if defined(TRMMKERNEL) && defined(LEFT)
+		off = offset;
+#endif
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+#if (defined(LEFT) &&  defined(TRANSA)) || \
+			  (!defined(LEFT) && !defined(TRANSA))
+             ptrbb = bb;
+#else
+			  ptrba += off*2;
+			  ptrbb = bb + off*2;
+#endif
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || \
+			 (!defined(LEFT) && defined(TRANSA))
+			 temp = bk-off;
+#elif defined(LEFT) 
+			 temp = off+2;
+#else
+			 temp = off+2;
+#endif
+             for (k=0; k<temp/4; k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  load4 = ptrba[2*1+0];
+                  load5 = ptrbb[2*1+0];
+                  res0 = res0+load4*load5;
+                  load6 = ptrba[2*1+1];
+                  res1 = res1+load6*load5;
+                  load7 = ptrbb[2*1+1];
+                  res2 = res2+load4*load7;
+                  res3 = res3+load6*load7;
+                  load0 = ptrba[2*2+0];
+                  load1 = ptrbb[2*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*2+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*2+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  load4 = ptrba[2*3+0];
+                  load5 = ptrbb[2*3+0];
+                  res0 = res0+load4*load5;
+                  load6 = ptrba[2*3+1];
+                  res1 = res1+load6*load5;
+                  load7 = ptrbb[2*3+1];
+                  res2 = res2+load4*load7;
+                  res3 = res3+load6*load7;
+                  ptrba = ptrba+8;
+                  ptrbb = ptrbb+8;
+               }
+             for (k=0; k<(temp&3); k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+2;
+               }
+             res0 = res0*alpha;
+             C0[0] = res0;
+             res1 = res1*alpha;
+             C0[1] = res1;
+             res2 = res2*alpha;
+             C1[0] = res2;
+             res3 = res3*alpha;
+             C1[1] = res3;
+#if ( defined(LEFT) && defined(TRANSA)) || \
+			 (!defined(LEFT) && !defined(TRANSA)) 
+			 temp = bk - off;
+#ifdef LEFT
+			 temp -= 2;
+#else 
+			 temp -= 2;
+#endif
+			 ptrba += temp*2;
+			 ptrbb += temp*2;
+#endif
+#ifdef LEFT
+			 off += 2;
+#endif
+             C0 = C0+2;
+             C1 = C1+2;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+#if (defined(LEFT) &&  defined(TRANSA)) ||(!defined(LEFT) && !defined(TRANSA))
+             ptrbb = bb;
+#else 
+			 ptrba += off;
+			 ptrbb = bb+off*2;
+#endif
+             res0 = 0;
+             res1 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			 temp = bk-off;
+#elif defined(LEFT)
+			 temp = off+1;
+#else 
+			 temp = off+2;
+#endif
+             for (k=0; k<temp; k+=1) 
+               {
+                  load0 = ptrba[0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrbb[2*0+1];
+                  res1 = res1+load0*load2;
+                  ptrba = ptrba+1;
+                  ptrbb = ptrbb+2;
+               }
+             res0 = res0*alpha;
+             C0[0] = res0;
+             res1 = res1*alpha;
+             C1[0] = res1;
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) 
+			 temp = bk-off;
+#ifdef LEFT
+			 temp -= 1;
+#else 
+			 temp -= 2;
+#endif
+			 ptrba += temp;
+			 ptrbb += temp*2;
+#endif
+#ifdef LEFT	
+			 off += 1;
+#endif
+             C0 = C0+1;
+             C1 = C1+1;
+          }
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 2;
+#endif
+        k = (bk<<1);
+        bb = bb+k;
+        i = (ldc<<1);
+        C = C+i;
+     }
+   for (j=0; j<(bn&1); j+=1) 
+     {
+        C0 = C;
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+		off = offset;
+#endif
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+#if (defined(LEFT) &&  defined(TRANSA)) || \
+			  (!defined(LEFT) && !defined(TRANSA))
+			  ptrbb = bb;
+#else 
+			  ptrba += off*2;
+			  ptrbb = bb + off;
+#endif
+             res0 = 0;
+             res1 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || \
+			 (!defined(LEFT) && defined(TRANSA))
+			 temp = bk-off;
+#elif defined(LEFT)
+			 temp = off+2;
+#else
+			 temp = off+1;
+#endif
+			 for (k=0; k<temp; k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+1;
+               }
+             res0 = res0*alpha;
+             C0[0] = res0;
+             res1 = res1*alpha;
+			 C0[1] = res1;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			 (!defined(LEFT) && !defined(TRANSA))
+			 temp = bk - off;
+#ifdef LEFT
+			 temp -= 2;
+#else 
+			 temp -= 1;
+#endif
+			 ptrba += temp*2;
+			 ptrbb += temp;
+#endif
+#ifdef LEFT
+			 off += 2;
+#endif
+
+             C0 = C0+2;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+             ptrbb = bb;
+#else 
+			 ptrba += off;
+			 ptrbb = bb+off;
+#endif
+             res0 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			 temp = bk-off;
+#elif defined(LEFT)
+			 temp = off + 1;
+#else 
+			 temp = off + 1;
+#endif
+			 for (k=0; k<temp; k+=1) 
+               {
+                  load0 = ptrba[0+0];
+                  load1 = ptrbb[0+0];
+                  res0 = res0+load0*load1;
+                  ptrba = ptrba+1;
+                  ptrbb = ptrbb+1;
+               }
+             res0 = res0*alpha;
+             C0[0] = res0;
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			 temp = bk-off;
+#ifdef LEFT
+			 temp -= 1;
+#else 
+			 temp -= 1;
+#endif
+			 ptrba += temp;
+			 ptrbb += temp;
+#endif
+#ifdef LEFT
+			 off += 1;
+#endif
+             C0 = C0+1;
+          }
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 1;
+#endif
+        k = (bk<<0);
+        bb = bb+k;
+        C = C+ldc;
+     }
+   return 0;
+}

kernel/generic/zgemm_ncopy_4_sandy.c

@@ -0,0 +1,235 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 <stdio.h>
+#include "common.h"
+
+int CNAME(BLASLONG row,BLASLONG col,FLOAT* src,BLASLONG srcdim,FLOAT* dest) 
+{
+   BLASLONG i,j;
+   BLASLONG idx=0;
+   BLASLONG ii;
+   FLOAT *src0,*src1,*src2,*src3,*dest0;
+   for (j=0; j<col/4; j+=1) 
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src = src3+2*srcdim;
+        dest0 = dest;
+        ii = (row<<3);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src0[2];
+             dest0[9] = src0[3];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src2[2];
+             dest0[13] = src2[3];
+             dest0[14] = src3[2];
+             dest0[15] = src3[3];
+             dest0[16] = src0[4];
+             dest0[17] = src0[5];
+             dest0[18] = src1[4];
+             dest0[19] = src1[5];
+             dest0[20] = src2[4];
+             dest0[21] = src2[5];
+             dest0[22] = src3[4];
+             dest0[23] = src3[5];
+             dest0[24] = src0[6];
+             dest0[25] = src0[7];
+             dest0[26] = src1[6];
+             dest0[27] = src1[7];
+             dest0[28] = src2[6];
+             dest0[29] = src2[7];
+             dest0[30] = src3[6];
+             dest0[31] = src3[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             ii = (4<<3);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src0[2];
+             dest0[9] = src0[3];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src2[2];
+             dest0[13] = src2[3];
+             dest0[14] = src3[2];
+             dest0[15] = src3[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             ii = (2<<3);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             ii = (1<<3);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&2)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src = src1+2*srcdim;
+        dest0 = dest;
+        ii = (row<<2);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src0[2];
+             dest0[5] = src0[3];
+             dest0[6] = src1[2];
+             dest0[7] = src1[3];
+             dest0[8] = src0[4];
+             dest0[9] = src0[5];
+             dest0[10] = src1[4];
+             dest0[11] = src1[5];
+             dest0[12] = src0[6];
+             dest0[13] = src0[7];
+             dest0[14] = src1[6];
+             dest0[15] = src1[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             ii = (4<<2);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src0[2];
+             dest0[5] = src0[3];
+             dest0[6] = src1[2];
+             dest0[7] = src1[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             ii = (2<<2);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             ii = (1<<2);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&1)  
+     {
+        src0 = src;
+        src = src0+2*srcdim;
+        dest0 = dest;
+        ii = (row<<1);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             src0 = src0+8;
+             ii = (4<<1);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             src0 = src0+4;
+             ii = (2<<1);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             src0 = src0+2;
+             ii = (1<<1);
+             dest0 = dest0+ii;
+          }
+     }
+   return 0;
+}

kernel/generic/zgemm_ncopy_8_sandy.c

@@ -0,0 +1,401 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 <stdio.h>
+#include "common.h"
+
+int CNAME(BLASLONG row,BLASLONG col,FLOAT* src,BLASLONG srcdim,FLOAT* dest) 
+{
+   BLASLONG i,j;
+   BLASLONG idx=0;
+   BLASLONG ii;
+   FLOAT *src0,*src1,*src2,*src3,*src4,*src5,*src6,*src7,*dest0;
+   for (j=0; j<col/8; j+=1) 
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src4 = src3+2*srcdim;
+        src5 = src4+2*srcdim;
+        src6 = src5+2*srcdim;
+        src7 = src6+2*srcdim;
+        src = src7+2*srcdim;
+        dest0 = dest;
+        ii = (row<<4);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src4[0];
+             dest0[9] = src4[1];
+             dest0[10] = src5[0];
+             dest0[11] = src5[1];
+             dest0[12] = src6[0];
+             dest0[13] = src6[1];
+             dest0[14] = src7[0];
+             dest0[15] = src7[1];
+             dest0[16] = src0[2];
+             dest0[17] = src0[3];
+             dest0[18] = src1[2];
+             dest0[19] = src1[3];
+             dest0[20] = src2[2];
+             dest0[21] = src2[3];
+             dest0[22] = src3[2];
+             dest0[23] = src3[3];
+             dest0[24] = src4[2];
+             dest0[25] = src4[3];
+             dest0[26] = src5[2];
+             dest0[27] = src5[3];
+             dest0[28] = src6[2];
+             dest0[29] = src6[3];
+             dest0[30] = src7[2];
+             dest0[31] = src7[3];
+             dest0[32] = src0[4];
+             dest0[33] = src0[5];
+             dest0[34] = src1[4];
+             dest0[35] = src1[5];
+             dest0[36] = src2[4];
+             dest0[37] = src2[5];
+             dest0[38] = src3[4];
+             dest0[39] = src3[5];
+             dest0[40] = src4[4];
+             dest0[41] = src4[5];
+             dest0[42] = src5[4];
+             dest0[43] = src5[5];
+             dest0[44] = src6[4];
+             dest0[45] = src6[5];
+             dest0[46] = src7[4];
+             dest0[47] = src7[5];
+             dest0[48] = src0[6];
+             dest0[49] = src0[7];
+             dest0[50] = src1[6];
+             dest0[51] = src1[7];
+             dest0[52] = src2[6];
+             dest0[53] = src2[7];
+             dest0[54] = src3[6];
+             dest0[55] = src3[7];
+             dest0[56] = src4[6];
+             dest0[57] = src4[7];
+             dest0[58] = src5[6];
+             dest0[59] = src5[7];
+             dest0[60] = src6[6];
+             dest0[61] = src6[7];
+             dest0[62] = src7[6];
+             dest0[63] = src7[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             src4 = src4+8;
+             src5 = src5+8;
+             src6 = src6+8;
+             src7 = src7+8;
+             ii = (4<<4);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src4[0];
+             dest0[9] = src4[1];
+             dest0[10] = src5[0];
+             dest0[11] = src5[1];
+             dest0[12] = src6[0];
+             dest0[13] = src6[1];
+             dest0[14] = src7[0];
+             dest0[15] = src7[1];
+             dest0[16] = src0[2];
+             dest0[17] = src0[3];
+             dest0[18] = src1[2];
+             dest0[19] = src1[3];
+             dest0[20] = src2[2];
+             dest0[21] = src2[3];
+             dest0[22] = src3[2];
+             dest0[23] = src3[3];
+             dest0[24] = src4[2];
+             dest0[25] = src4[3];
+             dest0[26] = src5[2];
+             dest0[27] = src5[3];
+             dest0[28] = src6[2];
+             dest0[29] = src6[3];
+             dest0[30] = src7[2];
+             dest0[31] = src7[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             src4 = src4+4;
+             src5 = src5+4;
+             src6 = src6+4;
+             src7 = src7+4;
+             ii = (2<<4);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src4[0];
+             dest0[9] = src4[1];
+             dest0[10] = src5[0];
+             dest0[11] = src5[1];
+             dest0[12] = src6[0];
+             dest0[13] = src6[1];
+             dest0[14] = src7[0];
+             dest0[15] = src7[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             src4 = src4+2;
+             src5 = src5+2;
+             src6 = src6+2;
+             src7 = src7+2;
+             ii = (1<<4);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&4)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src = src3+2*srcdim;
+        dest0 = dest;
+        ii = (row<<3);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src0[2];
+             dest0[9] = src0[3];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src2[2];
+             dest0[13] = src2[3];
+             dest0[14] = src3[2];
+             dest0[15] = src3[3];
+             dest0[16] = src0[4];
+             dest0[17] = src0[5];
+             dest0[18] = src1[4];
+             dest0[19] = src1[5];
+             dest0[20] = src2[4];
+             dest0[21] = src2[5];
+             dest0[22] = src3[4];
+             dest0[23] = src3[5];
+             dest0[24] = src0[6];
+             dest0[25] = src0[7];
+             dest0[26] = src1[6];
+             dest0[27] = src1[7];
+             dest0[28] = src2[6];
+             dest0[29] = src2[7];
+             dest0[30] = src3[6];
+             dest0[31] = src3[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             ii = (4<<3);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src0[2];
+             dest0[9] = src0[3];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src2[2];
+             dest0[13] = src2[3];
+             dest0[14] = src3[2];
+             dest0[15] = src3[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             ii = (2<<3);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             ii = (1<<3);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&2)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src = src1+2*srcdim;
+        dest0 = dest;
+        ii = (row<<2);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src0[2];
+             dest0[5] = src0[3];
+             dest0[6] = src1[2];
+             dest0[7] = src1[3];
+             dest0[8] = src0[4];
+             dest0[9] = src0[5];
+             dest0[10] = src1[4];
+             dest0[11] = src1[5];
+             dest0[12] = src0[6];
+             dest0[13] = src0[7];
+             dest0[14] = src1[6];
+             dest0[15] = src1[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             ii = (4<<2);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src0[2];
+             dest0[5] = src0[3];
+             dest0[6] = src1[2];
+             dest0[7] = src1[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             ii = (2<<2);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             ii = (1<<2);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&1)  
+     {
+        src0 = src;
+        src = src0+2*srcdim;
+        dest0 = dest;
+        ii = (row<<1);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             src0 = src0+8;
+             ii = (4<<1);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             src0 = src0+4;
+             ii = (2<<1);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             src0 = src0+2;
+             ii = (1<<1);
+             dest0 = dest0+ii;
+          }
+     }
+   return 0;
+}

kernel/generic/zgemm_tcopy_4_sandy.c

@@ -0,0 +1,237 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 <stdio.h>
+#include "common.h"
+
+int CNAME(BLASLONG row,BLASLONG col,FLOAT* src,BLASLONG srcdim,FLOAT* dest) 
+{
+   BLASLONG i,j;
+   BLASLONG idx=0;
+   BLASLONG ii;
+   FLOAT *src0,*src1,*src2,*src3,*dest0;
+   FLOAT *dest1,*dest2;
+   ii = col&-4;
+   ii = ii*(2*row);
+   dest2 = dest+ii;
+   ii = col&-2;
+   ii = ii*(2*row);
+   dest1 = dest+ii;
+   for (j=0; j<row/4; j+=1) 
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src = src3+2*srcdim;
+        dest0 = dest;
+        ii = (4<<3);
+        dest = dest+ii;
+        for (i=0; i<col/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src1[0];
+             dest0[9] = src1[1];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src1[4];
+             dest0[13] = src1[5];
+             dest0[14] = src1[6];
+             dest0[15] = src1[7];
+             dest0[16] = src2[0];
+             dest0[17] = src2[1];
+             dest0[18] = src2[2];
+             dest0[19] = src2[3];
+             dest0[20] = src2[4];
+             dest0[21] = src2[5];
+             dest0[22] = src2[6];
+             dest0[23] = src2[7];
+             dest0[24] = src3[0];
+             dest0[25] = src3[1];
+             dest0[26] = src3[2];
+             dest0[27] = src3[3];
+             dest0[28] = src3[4];
+             dest0[29] = src3[5];
+             dest0[30] = src3[6];
+             dest0[31] = src3[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             ii = (row<<3);
+             dest0 = dest0+ii;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             dest2[4] = src1[0];
+             dest2[5] = src1[1];
+             dest2[6] = src1[2];
+             dest2[7] = src1[3];
+             dest2[8] = src2[0];
+             dest2[9] = src2[1];
+             dest2[10] = src2[2];
+             dest2[11] = src2[3];
+             dest2[12] = src3[0];
+             dest2[13] = src3[1];
+             dest2[14] = src3[2];
+             dest2[15] = src3[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             dest2 = dest2+16;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             dest1[2] = src1[0];
+             dest1[3] = src1[1];
+             dest1[4] = src2[0];
+             dest1[5] = src2[1];
+             dest1[6] = src3[0];
+             dest1[7] = src3[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             dest1 = dest1+8;
+          }
+     }
+   if (row&2)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src = src1+2*srcdim;
+        dest0 = dest;
+        ii = (2<<3);
+        dest = dest+ii;
+        for (i=0; i<col/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src1[0];
+             dest0[9] = src1[1];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src1[4];
+             dest0[13] = src1[5];
+             dest0[14] = src1[6];
+             dest0[15] = src1[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             ii = (row<<3);
+             dest0 = dest0+ii;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             dest2[4] = src1[0];
+             dest2[5] = src1[1];
+             dest2[6] = src1[2];
+             dest2[7] = src1[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             dest2 = dest2+8;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             dest1[2] = src1[0];
+             dest1[3] = src1[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             dest1 = dest1+4;
+          }
+     }
+   if (row&1)  
+     {
+        src0 = src;
+        src = src0+2*srcdim;
+        dest0 = dest;
+        ii = (1<<3);
+        dest = dest+ii;
+        for (i=0; i<col/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             src0 = src0+8;
+             ii = (row<<3);
+             dest0 = dest0+ii;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             src0 = src0+4;
+             dest2 = dest2+4;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             src0 = src0+2;
+             dest1 = dest1+2;
+          }
+     }
+
+   return 0;
+}

kernel/generic/zgemm_tcopy_8_sandy.c

@@ -0,0 +1,370 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 <stdio.h>
+#include "common.h"
+
+int CNAME(BLASLONG row,BLASLONG col,FLOAT* src,BLASLONG srcdim,FLOAT* dest) 
+{
+   BLASLONG i,j;
+   BLASLONG idx=0;
+   BLASLONG ii;
+   FLOAT *src0,*src1,*src2,*src3,*dest0;
+   FLOAT *dest1,*dest2,*dest4;
+   ii = col&-8;
+   ii = ii*(2*row);
+   dest4 = dest+ii;
+   ii = col&-4;
+   ii = ii*(2*row);
+   dest2 = dest+ii;
+   ii = col&-2;
+   ii = ii*(2*row);
+   dest1 = dest+ii;
+   for (j=0; j<row/4; j+=1) 
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src = src3+2*srcdim;
+        dest0 = dest;
+        ii = (4<<4);
+        dest = dest+ii;
+        for (i=0; i<col/8; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src0[8];
+             dest0[9] = src0[9];
+             dest0[10] = src0[10];
+             dest0[11] = src0[11];
+             dest0[12] = src0[12];
+             dest0[13] = src0[13];
+             dest0[14] = src0[14];
+             dest0[15] = src0[15];
+             dest0[16] = src1[0];
+             dest0[17] = src1[1];
+             dest0[18] = src1[2];
+             dest0[19] = src1[3];
+             dest0[20] = src1[4];
+             dest0[21] = src1[5];
+             dest0[22] = src1[6];
+             dest0[23] = src1[7];
+             dest0[24] = src1[8];
+             dest0[25] = src1[9];
+             dest0[26] = src1[10];
+             dest0[27] = src1[11];
+             dest0[28] = src1[12];
+             dest0[29] = src1[13];
+             dest0[30] = src1[14];
+             dest0[31] = src1[15];
+             dest0[32] = src2[0];
+             dest0[33] = src2[1];
+             dest0[34] = src2[2];
+             dest0[35] = src2[3];
+             dest0[36] = src2[4];
+             dest0[37] = src2[5];
+             dest0[38] = src2[6];
+             dest0[39] = src2[7];
+             dest0[40] = src2[8];
+             dest0[41] = src2[9];
+             dest0[42] = src2[10];
+             dest0[43] = src2[11];
+             dest0[44] = src2[12];
+             dest0[45] = src2[13];
+             dest0[46] = src2[14];
+             dest0[47] = src2[15];
+             dest0[48] = src3[0];
+             dest0[49] = src3[1];
+             dest0[50] = src3[2];
+             dest0[51] = src3[3];
+             dest0[52] = src3[4];
+             dest0[53] = src3[5];
+             dest0[54] = src3[6];
+             dest0[55] = src3[7];
+             dest0[56] = src3[8];
+             dest0[57] = src3[9];
+             dest0[58] = src3[10];
+             dest0[59] = src3[11];
+             dest0[60] = src3[12];
+             dest0[61] = src3[13];
+             dest0[62] = src3[14];
+             dest0[63] = src3[15];
+             src0 = src0+16;
+             src1 = src1+16;
+             src2 = src2+16;
+             src3 = src3+16;
+             ii = (row<<4);
+             dest0 = dest0+ii;
+          }
+        if (col&4)  
+          {
+             dest4[0] = src0[0];
+             dest4[1] = src0[1];
+             dest4[2] = src0[2];
+             dest4[3] = src0[3];
+             dest4[4] = src0[4];
+             dest4[5] = src0[5];
+             dest4[6] = src0[6];
+             dest4[7] = src0[7];
+             dest4[8] = src1[0];
+             dest4[9] = src1[1];
+             dest4[10] = src1[2];
+             dest4[11] = src1[3];
+             dest4[12] = src1[4];
+             dest4[13] = src1[5];
+             dest4[14] = src1[6];
+             dest4[15] = src1[7];
+             dest4[16] = src2[0];
+             dest4[17] = src2[1];
+             dest4[18] = src2[2];
+             dest4[19] = src2[3];
+             dest4[20] = src2[4];
+             dest4[21] = src2[5];
+             dest4[22] = src2[6];
+             dest4[23] = src2[7];
+             dest4[24] = src3[0];
+             dest4[25] = src3[1];
+             dest4[26] = src3[2];
+             dest4[27] = src3[3];
+             dest4[28] = src3[4];
+             dest4[29] = src3[5];
+             dest4[30] = src3[6];
+             dest4[31] = src3[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             dest4 = dest4+32;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             dest2[4] = src1[0];
+             dest2[5] = src1[1];
+             dest2[6] = src1[2];
+             dest2[7] = src1[3];
+             dest2[8] = src2[0];
+             dest2[9] = src2[1];
+             dest2[10] = src2[2];
+             dest2[11] = src2[3];
+             dest2[12] = src3[0];
+             dest2[13] = src3[1];
+             dest2[14] = src3[2];
+             dest2[15] = src3[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             dest2 = dest2+16;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             dest1[2] = src1[0];
+             dest1[3] = src1[1];
+             dest1[4] = src2[0];
+             dest1[5] = src2[1];
+             dest1[6] = src3[0];
+             dest1[7] = src3[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             dest1 = dest1+8;
+          }
+     }
+   if (row&2)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src = src1+2*srcdim;
+        dest0 = dest;
+        ii = (2<<4);
+        dest = dest+ii;
+        for (i=0; i<col/8; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src0[8];
+             dest0[9] = src0[9];
+             dest0[10] = src0[10];
+             dest0[11] = src0[11];
+             dest0[12] = src0[12];
+             dest0[13] = src0[13];
+             dest0[14] = src0[14];
+             dest0[15] = src0[15];
+             dest0[16] = src1[0];
+             dest0[17] = src1[1];
+             dest0[18] = src1[2];
+             dest0[19] = src1[3];
+             dest0[20] = src1[4];
+             dest0[21] = src1[5];
+             dest0[22] = src1[6];
+             dest0[23] = src1[7];
+             dest0[24] = src1[8];
+             dest0[25] = src1[9];
+             dest0[26] = src1[10];
+             dest0[27] = src1[11];
+             dest0[28] = src1[12];
+             dest0[29] = src1[13];
+             dest0[30] = src1[14];
+             dest0[31] = src1[15];
+             src0 = src0+16;
+             src1 = src1+16;
+             ii = (row<<4);
+             dest0 = dest0+ii;
+          }
+        if (col&4)  
+          {
+             dest4[0] = src0[0];
+             dest4[1] = src0[1];
+             dest4[2] = src0[2];
+             dest4[3] = src0[3];
+             dest4[4] = src0[4];
+             dest4[5] = src0[5];
+             dest4[6] = src0[6];
+             dest4[7] = src0[7];
+             dest4[8] = src1[0];
+             dest4[9] = src1[1];
+             dest4[10] = src1[2];
+             dest4[11] = src1[3];
+             dest4[12] = src1[4];
+             dest4[13] = src1[5];
+             dest4[14] = src1[6];
+             dest4[15] = src1[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             dest4 = dest4+16;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             dest2[4] = src1[0];
+             dest2[5] = src1[1];
+             dest2[6] = src1[2];
+             dest2[7] = src1[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             dest2 = dest2+8;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             dest1[2] = src1[0];
+             dest1[3] = src1[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             dest1 = dest1+4;
+          }
+     }
+   if (row&1)  
+     {
+        src0 = src;
+        src = src0+2*srcdim;
+        dest0 = dest;
+        ii = (1<<4);
+        dest = dest+ii;
+        for (i=0; i<col/8; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src0[8];
+             dest0[9] = src0[9];
+             dest0[10] = src0[10];
+             dest0[11] = src0[11];
+             dest0[12] = src0[12];
+             dest0[13] = src0[13];
+             dest0[14] = src0[14];
+             dest0[15] = src0[15];
+             src0 = src0+16;
+             ii = (row<<4);
+             dest0 = dest0+ii;
+          }
+        if (col&4)  
+          {
+             dest4[0] = src0[0];
+             dest4[1] = src0[1];
+             dest4[2] = src0[2];
+             dest4[3] = src0[3];
+             dest4[4] = src0[4];
+             dest4[5] = src0[5];
+             dest4[6] = src0[6];
+             dest4[7] = src0[7];
+             src0 = src0+8;
+             dest4 = dest4+8;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             src0 = src0+4;
+             dest2 = dest2+4;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             src0 = src0+2;
+             dest1 = dest1+2;
+          }
+     }
+   return 0;
+}

kernel/generic/zgemmkernel_2x2.c

@@ -0,0 +1,838 @@
+#include "common.h"
+/********************************
+  ADD1 a*c
+  ADD2 b*c
+  ADD3 a*d
+  ADD4 b*d
+*********************************/
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alphar,FLOAT alphai,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
+#ifdef	TRMMKERNEL
+		, BLASLONG offset
+#endif
+		)
+{
+   BLASLONG i,j,k;
+   FLOAT *C0,*C1,*ptrba,*ptrbb;
+   FLOAT res0,res1,res2,res3,res4,res5,res6,res7,load0,load1,load2,load3,load4,load5,load6,load7,load8,load9,load10,load11,load12,load13,load14,load15;
+   for (j=0; j<bn/2; j+=1) 
+     {
+        C0 = C;
+        C1 = C0+2*ldc;
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+             res4 = 0;
+             res5 = 0;
+             res6 = 0;
+             res7 = 0;
+             for (k=0; k<bk/4; k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7+load4*load7;
+                  load8 = ptrba[4*1+0];
+                  load9 = ptrbb[4*1+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*1+1];
+                  res1 = res1+load10*load9;
+                  load11 = ptrbb[4*1+1];
+                  res0 = res0-load10*load11;
+                  res1 = res1+load8*load11;
+                  load12 = ptrba[4*1+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*1+3];
+                  res3 = res3+load13*load9;
+                  res2 = res2-load13*load11;
+                  res3 = res3+load12*load11;
+                  load14 = ptrbb[4*1+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5+load10*load14;
+                  load15 = ptrbb[4*1+3];
+                  res4 = res4-load10*load15;
+                  res5 = res5+load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7+load13*load14;
+                  res6 = res6-load13*load15;
+                  res7 = res7+load12*load15;
+                  load0 = ptrba[4*2+0];
+                  load1 = ptrbb[4*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*2+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*2+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*2+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*2+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*2+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*2+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7+load4*load7;
+                  load8 = ptrba[4*3+0];
+                  load9 = ptrbb[4*3+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*3+1];
+                  res1 = res1+load10*load9;
+                  load11 = ptrbb[4*3+1];
+                  res0 = res0-load10*load11;
+                  res1 = res1+load8*load11;
+                  load12 = ptrba[4*3+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*3+3];
+                  res3 = res3+load13*load9;
+                  res2 = res2-load13*load11;
+                  res3 = res3+load12*load11;
+                  load14 = ptrbb[4*3+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5+load10*load14;
+                  load15 = ptrbb[4*3+3];
+                  res4 = res4-load10*load15;
+                  res5 = res5+load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7+load13*load14;
+                  res6 = res6-load13*load15;
+                  res7 = res7+load12*load15;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7-load4*load7;
+                  load8 = ptrba[4*1+0];
+                  load9 = ptrbb[4*1+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*1+1];
+                  res1 = res1+load10*load9;
+                  load11 = ptrbb[4*1+1];
+                  res0 = res0+load10*load11;
+                  res1 = res1-load8*load11;
+                  load12 = ptrba[4*1+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*1+3];
+                  res3 = res3+load13*load9;
+                  res2 = res2+load13*load11;
+                  res3 = res3-load12*load11;
+                  load14 = ptrbb[4*1+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5+load10*load14;
+                  load15 = ptrbb[4*1+3];
+                  res4 = res4+load10*load15;
+                  res5 = res5-load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7+load13*load14;
+                  res6 = res6+load13*load15;
+                  res7 = res7-load12*load15;
+                  load0 = ptrba[4*2+0];
+                  load1 = ptrbb[4*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*2+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*2+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*2+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*2+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*2+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*2+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7-load4*load7;
+                  load8 = ptrba[4*3+0];
+                  load9 = ptrbb[4*3+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*3+1];
+                  res1 = res1+load10*load9;
+                  load11 = ptrbb[4*3+1];
+                  res0 = res0+load10*load11;
+                  res1 = res1-load8*load11;
+                  load12 = ptrba[4*3+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*3+3];
+                  res3 = res3+load13*load9;
+                  res2 = res2+load13*load11;
+                  res3 = res3-load12*load11;
+                  load14 = ptrbb[4*3+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5+load10*load14;
+                  load15 = ptrbb[4*3+3];
+                  res4 = res4+load10*load15;
+                  res5 = res5-load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7+load13*load14;
+                  res6 = res6+load13*load15;
+                  res7 = res7-load12*load15;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7+load4*load7;
+                  load8 = ptrba[4*1+0];
+                  load9 = ptrbb[4*1+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*1+1];
+                  res1 = res1-load10*load9;
+                  load11 = ptrbb[4*1+1];
+                  res0 = res0+load10*load11;
+                  res1 = res1+load8*load11;
+                  load12 = ptrba[4*1+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*1+3];
+                  res3 = res3-load13*load9;
+                  res2 = res2+load13*load11;
+                  res3 = res3+load12*load11;
+                  load14 = ptrbb[4*1+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5-load10*load14;
+                  load15 = ptrbb[4*1+3];
+                  res4 = res4+load10*load15;
+                  res5 = res5+load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7-load13*load14;
+                  res6 = res6+load13*load15;
+                  res7 = res7+load12*load15;
+                  load0 = ptrba[4*2+0];
+                  load1 = ptrbb[4*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*2+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*2+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*2+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*2+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*2+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*2+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7+load4*load7;
+                  load8 = ptrba[4*3+0];
+                  load9 = ptrbb[4*3+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*3+1];
+                  res1 = res1-load10*load9;
+                  load11 = ptrbb[4*3+1];
+                  res0 = res0+load10*load11;
+                  res1 = res1+load8*load11;
+                  load12 = ptrba[4*3+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*3+3];
+                  res3 = res3-load13*load9;
+                  res2 = res2+load13*load11;
+                  res3 = res3+load12*load11;
+                  load14 = ptrbb[4*3+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5-load10*load14;
+                  load15 = ptrbb[4*3+3];
+                  res4 = res4+load10*load15;
+                  res5 = res5+load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7-load13*load14;
+                  res6 = res6+load13*load15;
+                  res7 = res7+load12*load15;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7-load4*load7;
+                  load8 = ptrba[4*1+0];
+                  load9 = ptrbb[4*1+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*1+1];
+                  res1 = res1-load10*load9;
+                  load11 = ptrbb[4*1+1];
+                  res0 = res0-load10*load11;
+                  res1 = res1-load8*load11;
+                  load12 = ptrba[4*1+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*1+3];
+                  res3 = res3-load13*load9;
+                  res2 = res2-load13*load11;
+                  res3 = res3-load12*load11;
+                  load14 = ptrbb[4*1+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5-load10*load14;
+                  load15 = ptrbb[4*1+3];
+                  res4 = res4-load10*load15;
+                  res5 = res5-load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7-load13*load14;
+                  res6 = res6-load13*load15;
+                  res7 = res7-load12*load15;
+                  load0 = ptrba[4*2+0];
+                  load1 = ptrbb[4*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*2+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*2+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*2+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*2+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*2+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*2+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7-load4*load7;
+                  load8 = ptrba[4*3+0];
+                  load9 = ptrbb[4*3+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*3+1];
+                  res1 = res1-load10*load9;
+                  load11 = ptrbb[4*3+1];
+                  res0 = res0-load10*load11;
+                  res1 = res1-load8*load11;
+                  load12 = ptrba[4*3+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*3+3];
+                  res3 = res3-load13*load9;
+                  res2 = res2-load13*load11;
+                  res3 = res3-load12*load11;
+                  load14 = ptrbb[4*3+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5-load10*load14;
+                  load15 = ptrbb[4*3+3];
+                  res4 = res4-load10*load15;
+                  res5 = res5-load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7-load13*load14;
+                  res6 = res6-load13*load15;
+                  res7 = res7-load12*load15;
+#endif
+                  ptrba = ptrba+16;
+                  ptrbb = ptrbb+16;
+               }
+             for (k=0; k<(bk&3); k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7+load4*load7;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7-load4*load7;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7+load4*load7;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7-load4*load7;
+#endif
+                  ptrba = ptrba+4;
+                  ptrbb = ptrbb+4;
+               }
+             load0 = res0*alphar;
+             C0[0] = C0[0]+load0;
+             load1 = res1*alphar;
+             C0[1] = C0[1]+load1;
+             load0 = res1*alphai;
+             C0[0] = C0[0]-load0;
+             load1 = res0*alphai;
+             C0[1] = C0[1]+load1;
+             load2 = res2*alphar;
+             C0[2] = C0[2]+load2;
+             load3 = res3*alphar;
+             C0[3] = C0[3]+load3;
+             load2 = res3*alphai;
+             C0[2] = C0[2]-load2;
+             load3 = res2*alphai;
+             C0[3] = C0[3]+load3;
+             load4 = res4*alphar;
+             C1[0] = C1[0]+load4;
+             load5 = res5*alphar;
+             C1[1] = C1[1]+load5;
+             load4 = res5*alphai;
+             C1[0] = C1[0]-load4;
+             load5 = res4*alphai;
+             C1[1] = C1[1]+load5;
+             load6 = res6*alphar;
+             C1[2] = C1[2]+load6;
+             load7 = res7*alphar;
+             C1[3] = C1[3]+load7;
+             load6 = res7*alphai;
+             C1[2] = C1[2]-load6;
+             load7 = res6*alphai;
+             C1[3] = C1[3]+load7;
+             C0 = C0+4;
+             C1 = C1+4;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrbb[4*0+2];
+                  res2 = res2+load0*load4;
+                  res3 = res3+load2*load4;
+                  load5 = ptrbb[4*0+3];
+                  res2 = res2-load2*load5;
+                  res3 = res3+load0*load5;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrbb[4*0+2];
+                  res2 = res2+load0*load4;
+                  res3 = res3+load2*load4;
+                  load5 = ptrbb[4*0+3];
+                  res2 = res2+load2*load5;
+                  res3 = res3-load0*load5;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrbb[4*0+2];
+                  res2 = res2+load0*load4;
+                  res3 = res3-load2*load4;
+                  load5 = ptrbb[4*0+3];
+                  res2 = res2+load2*load5;
+                  res3 = res3+load0*load5;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrbb[4*0+2];
+                  res2 = res2+load0*load4;
+                  res3 = res3-load2*load4;
+                  load5 = ptrbb[4*0+3];
+                  res2 = res2-load2*load5;
+                  res3 = res3-load0*load5;
+#endif
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+4;
+               }
+             load0 = res0*alphar;
+             C0[0] = C0[0]+load0;
+             load1 = res1*alphar;
+             C0[1] = C0[1]+load1;
+             load0 = res1*alphai;
+             C0[0] = C0[0]-load0;
+             load1 = res0*alphai;
+             C0[1] = C0[1]+load1;
+             load2 = res2*alphar;
+             C1[0] = C1[0]+load2;
+             load3 = res3*alphar;
+             C1[1] = C1[1]+load3;
+             load2 = res3*alphai;
+             C1[0] = C1[0]-load2;
+             load3 = res2*alphai;
+             C1[1] = C1[1]+load3;
+             C0 = C0+2;
+             C1 = C1+2;
+          }
+        k = (bk<<2);
+        bb = bb+k;
+        i = (ldc<<2);
+        C = C+i;
+     }
+   for (j=0; j<(bn&1); j+=1) 
+     {
+        C0 = C;
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3+load4*load3;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3-load4*load3;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3+load4*load3;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3-load4*load3;
+#endif
+                  ptrba = ptrba+4;
+                  ptrbb = ptrbb+2;
+               }
+             load0 = res0*alphar;
+             C0[0] = C0[0]+load0;
+             load1 = res1*alphar;
+             C0[1] = C0[1]+load1;
+             load0 = res1*alphai;
+             C0[0] = C0[0]-load0;
+             load1 = res0*alphai;
+             C0[1] = C0[1]+load1;
+             load2 = res2*alphar;
+             C0[2] = C0[2]+load2;
+             load3 = res3*alphar;
+             C0[3] = C0[3]+load3;
+             load2 = res3*alphai;
+             C0[2] = C0[2]-load2;
+             load3 = res2*alphai;
+             C0[3] = C0[3]+load3;
+             C0 = C0+4;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+#endif
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+2;
+               }
+             load0 = res0*alphar;
+             C0[0] = C0[0]+load0;
+             load1 = res1*alphar;
+             C0[1] = C0[1]+load1;
+             load0 = res1*alphai;
+             C0[0] = C0[0]-load0;
+             load1 = res0*alphai;
+             C0[1] = C0[1]+load1;
+             C0 = C0+2;
+          }
+        k = (bk<<1);
+        bb = bb+k;
+        i = (ldc<<1);
+        C = C+i;
+     }
+   return 0;
+}

kernel/generic/ztrmmkernel_2x2.c

@@ -0,0 +1,923 @@
+#include "common.h"
+/********************************
+  ADD1 a*c
+  ADD2 b*c
+  ADD3 a*d
+  ADD4 b*d
+ *********************************/
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alphar,FLOAT alphai,FLOAT* ba,FLOAT* bb,
+		FLOAT* C,BLASLONG ldc, BLASLONG offset)
+{
+	BLASLONG i,j,k;
+	FLOAT *C0,*C1,*ptrba,*ptrbb;
+	FLOAT res0,res1,res2,res3,res4,res5,res6,res7,load0,load1,load2,load3,load4,load5,load6,load7,load8,load9,load10,load11,load12,load13,load14,load15;
+	BLASLONG off, temp;
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	off = -offset;
+#endif
+	for (j=0; j<bn/2; j+=1) 
+	{
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+		off = offset;
+#endif
+		C0 = C;
+		C1 = C0+2*ldc;
+		ptrba = ba;
+		for (i=0; i<bm/2; i+=1) 
+		{
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			ptrbb = bb;
+#else
+			ptrba += off*2*2;
+			ptrbb = bb+off*2*2;
+#endif
+			res0 = 0;
+			res1 = 0;
+			res2 = 0;
+			res3 = 0;
+			res4 = 0;
+			res5 = 0;
+			res6 = 0;
+			res7 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			temp = bk - off;
+#elif defined(LEFT)
+			temp = off + 2;
+#else 
+			temp = off + 2;
+#endif
+			for (k=0; k<temp/4; k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4-load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7+load4*load7;
+				load8 = ptrba[4*1+0];
+				load9 = ptrbb[4*1+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*1+1];
+				res1 = res1+load10*load9;
+				load11 = ptrbb[4*1+1];
+				res0 = res0-load10*load11;
+				res1 = res1+load8*load11;
+				load12 = ptrba[4*1+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*1+3];
+				res3 = res3+load13*load9;
+				res2 = res2-load13*load11;
+				res3 = res3+load12*load11;
+				load14 = ptrbb[4*1+2];
+				res4 = res4+load8*load14;
+				res5 = res5+load10*load14;
+				load15 = ptrbb[4*1+3];
+				res4 = res4-load10*load15;
+				res5 = res5+load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7+load13*load14;
+				res6 = res6-load13*load15;
+				res7 = res7+load12*load15;
+				load0 = ptrba[4*2+0];
+				load1 = ptrbb[4*2+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*2+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*2+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*2+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*2+3];
+				res3 = res3+load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*2+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*2+3];
+				res4 = res4-load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7+load4*load7;
+				load8 = ptrba[4*3+0];
+				load9 = ptrbb[4*3+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*3+1];
+				res1 = res1+load10*load9;
+				load11 = ptrbb[4*3+1];
+				res0 = res0-load10*load11;
+				res1 = res1+load8*load11;
+				load12 = ptrba[4*3+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*3+3];
+				res3 = res3+load13*load9;
+				res2 = res2-load13*load11;
+				res3 = res3+load12*load11;
+				load14 = ptrbb[4*3+2];
+				res4 = res4+load8*load14;
+				res5 = res5+load10*load14;
+				load15 = ptrbb[4*3+3];
+				res4 = res4-load10*load15;
+				res5 = res5+load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7+load13*load14;
+				res6 = res6-load13*load15;
+				res7 = res7+load12*load15;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4+load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7-load4*load7;
+				load8 = ptrba[4*1+0];
+				load9 = ptrbb[4*1+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*1+1];
+				res1 = res1+load10*load9;
+				load11 = ptrbb[4*1+1];
+				res0 = res0+load10*load11;
+				res1 = res1-load8*load11;
+				load12 = ptrba[4*1+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*1+3];
+				res3 = res3+load13*load9;
+				res2 = res2+load13*load11;
+				res3 = res3-load12*load11;
+				load14 = ptrbb[4*1+2];
+				res4 = res4+load8*load14;
+				res5 = res5+load10*load14;
+				load15 = ptrbb[4*1+3];
+				res4 = res4+load10*load15;
+				res5 = res5-load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7+load13*load14;
+				res6 = res6+load13*load15;
+				res7 = res7-load12*load15;
+				load0 = ptrba[4*2+0];
+				load1 = ptrbb[4*2+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*2+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*2+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*2+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*2+3];
+				res3 = res3+load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*2+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*2+3];
+				res4 = res4+load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7-load4*load7;
+				load8 = ptrba[4*3+0];
+				load9 = ptrbb[4*3+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*3+1];
+				res1 = res1+load10*load9;
+				load11 = ptrbb[4*3+1];
+				res0 = res0+load10*load11;
+				res1 = res1-load8*load11;
+				load12 = ptrba[4*3+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*3+3];
+				res3 = res3+load13*load9;
+				res2 = res2+load13*load11;
+				res3 = res3-load12*load11;
+				load14 = ptrbb[4*3+2];
+				res4 = res4+load8*load14;
+				res5 = res5+load10*load14;
+				load15 = ptrbb[4*3+3];
+				res4 = res4+load10*load15;
+				res5 = res5-load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7+load13*load14;
+				res6 = res6+load13*load15;
+				res7 = res7-load12*load15;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4+load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7+load4*load7;
+				load8 = ptrba[4*1+0];
+				load9 = ptrbb[4*1+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*1+1];
+				res1 = res1-load10*load9;
+				load11 = ptrbb[4*1+1];
+				res0 = res0+load10*load11;
+				res1 = res1+load8*load11;
+				load12 = ptrba[4*1+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*1+3];
+				res3 = res3-load13*load9;
+				res2 = res2+load13*load11;
+				res3 = res3+load12*load11;
+				load14 = ptrbb[4*1+2];
+				res4 = res4+load8*load14;
+				res5 = res5-load10*load14;
+				load15 = ptrbb[4*1+3];
+				res4 = res4+load10*load15;
+				res5 = res5+load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7-load13*load14;
+				res6 = res6+load13*load15;
+				res7 = res7+load12*load15;
+				load0 = ptrba[4*2+0];
+				load1 = ptrbb[4*2+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*2+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*2+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*2+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*2+3];
+				res3 = res3-load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*2+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*2+3];
+				res4 = res4+load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7+load4*load7;
+				load8 = ptrba[4*3+0];
+				load9 = ptrbb[4*3+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*3+1];
+				res1 = res1-load10*load9;
+				load11 = ptrbb[4*3+1];
+				res0 = res0+load10*load11;
+				res1 = res1+load8*load11;
+				load12 = ptrba[4*3+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*3+3];
+				res3 = res3-load13*load9;
+				res2 = res2+load13*load11;
+				res3 = res3+load12*load11;
+				load14 = ptrbb[4*3+2];
+				res4 = res4+load8*load14;
+				res5 = res5-load10*load14;
+				load15 = ptrbb[4*3+3];
+				res4 = res4+load10*load15;
+				res5 = res5+load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7-load13*load14;
+				res6 = res6+load13*load15;
+				res7 = res7+load12*load15;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4-load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7-load4*load7;
+				load8 = ptrba[4*1+0];
+				load9 = ptrbb[4*1+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*1+1];
+				res1 = res1-load10*load9;
+				load11 = ptrbb[4*1+1];
+				res0 = res0-load10*load11;
+				res1 = res1-load8*load11;
+				load12 = ptrba[4*1+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*1+3];
+				res3 = res3-load13*load9;
+				res2 = res2-load13*load11;
+				res3 = res3-load12*load11;
+				load14 = ptrbb[4*1+2];
+				res4 = res4+load8*load14;
+				res5 = res5-load10*load14;
+				load15 = ptrbb[4*1+3];
+				res4 = res4-load10*load15;
+				res5 = res5-load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7-load13*load14;
+				res6 = res6-load13*load15;
+				res7 = res7-load12*load15;
+				load0 = ptrba[4*2+0];
+				load1 = ptrbb[4*2+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*2+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*2+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*2+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*2+3];
+				res3 = res3-load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*2+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*2+3];
+				res4 = res4-load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7-load4*load7;
+				load8 = ptrba[4*3+0];
+				load9 = ptrbb[4*3+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*3+1];
+				res1 = res1-load10*load9;
+				load11 = ptrbb[4*3+1];
+				res0 = res0-load10*load11;
+				res1 = res1-load8*load11;
+				load12 = ptrba[4*3+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*3+3];
+				res3 = res3-load13*load9;
+				res2 = res2-load13*load11;
+				res3 = res3-load12*load11;
+				load14 = ptrbb[4*3+2];
+				res4 = res4+load8*load14;
+				res5 = res5-load10*load14;
+				load15 = ptrbb[4*3+3];
+				res4 = res4-load10*load15;
+				res5 = res5-load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7-load13*load14;
+				res6 = res6-load13*load15;
+				res7 = res7-load12*load15;
+#endif
+				ptrba = ptrba+16;
+				ptrbb = ptrbb+16;
+			}
+			for (k=0; k<(temp&3); k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4-load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7+load4*load7;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4+load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7-load4*load7;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4+load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7+load4*load7;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4-load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7-load4*load7;
+#endif
+				ptrba = ptrba+4;
+				ptrbb = ptrbb+4;
+			}
+			load0 = res0*alphar-res1*alphai;
+			load1 = res1*alphar+res0*alphai;
+			C0[0] = load0;
+			C0[1] = load1;
+
+			load2 = res2*alphar-res3*alphai;
+			load3 = res3*alphar+res2*alphai;
+			C0[2] = load2;
+			C0[3] = load3;
+
+			load4 = res4*alphar-res5*alphai;
+			load5 = res5*alphar+res4*alphai;
+			C1[0] = load4;
+			C1[1] = load5;
+
+			load6 = res6*alphar-res7*alphai;
+			load7 = res7*alphar+res6*alphai;
+			C1[2] = load6;
+			C1[3] = load7;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			(!defined(LEFT) && !defined(TRANSA))
+			temp = bk - off;
+#ifdef LEFT
+			temp -= 2;
+#else
+			temp -= 2;
+#endif
+			ptrba += temp*2*2;
+			ptrbb += temp*2*2;
+#endif
+#ifdef LEFT
+			off += 2;
+#endif
+
+			C0 = C0+4;
+			C1 = C1+4;
+		}
+		for (i=0; i<(bm&1); i+=1) 
+		{
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			ptrbb = bb;
+#else 
+			ptrba += off*2;
+			ptrbb = bb + off*2*2;
+#endif
+			res0 = 0;
+			res1 = 0;
+			res2 = 0;
+			res3 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			temp = bk - off;
+#elif defined(LEFT)
+			temp = off+1;
+#else
+			temp = off+2;
+#endif
+			for (k=0; k<temp; k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrbb[4*0+2];
+				res2 = res2+load0*load4;
+				res3 = res3+load2*load4;
+				load5 = ptrbb[4*0+3];
+				res2 = res2-load2*load5;
+				res3 = res3+load0*load5;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrbb[4*0+2];
+				res2 = res2+load0*load4;
+				res3 = res3+load2*load4;
+				load5 = ptrbb[4*0+3];
+				res2 = res2+load2*load5;
+				res3 = res3-load0*load5;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrbb[4*0+2];
+				res2 = res2+load0*load4;
+				res3 = res3-load2*load4;
+				load5 = ptrbb[4*0+3];
+				res2 = res2+load2*load5;
+				res3 = res3+load0*load5;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrbb[4*0+2];
+				res2 = res2+load0*load4;
+				res3 = res3-load2*load4;
+				load5 = ptrbb[4*0+3];
+				res2 = res2-load2*load5;
+				res3 = res3-load0*load5;
+#endif
+				ptrba = ptrba+2;
+				ptrbb = ptrbb+4;
+			}
+			load0 = res0*alphar-res1*alphai;
+			load1 = res1*alphar+res0*alphai;
+			C0[0] = load0;
+			C0[1] = load1;
+
+			load2 = res2*alphar-res3*alphai;
+			load3 = res3*alphar+res2*alphai;
+			C1[0] = load2;
+			C1[1] = load3;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			(!defined(LEFT) && !defined(TRANSA))
+			temp = bk - off;
+#ifdef LEFT
+			temp -= 1;
+#else 
+			temp -= 2;
+#endif
+			ptrba += temp*2;
+			ptrbb += temp*2*2;
+#endif
+#ifdef LEFT
+			off += 1;
+#endif
+			C0 = C0+2;
+			C1 = C1+2;
+		}
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 2;
+#endif
+		k = (bk<<2);
+		bb = bb+k;
+		i = (ldc<<2);
+		C = C+i;
+	}
+	for (j=0; j<(bn&1); j+=1) 
+	{
+		C0 = C;
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+		off = offset;
+#endif
+		ptrba = ba;
+		for (i=0; i<bm/2; i+=1) 
+		{
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			ptrbb = bb;
+#else 
+			ptrba += off*2*2;
+			ptrbb = bb+off*2;
+#endif
+			res0 = 0;
+			res1 = 0;
+			res2 = 0;
+			res3 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			temp = bk - off;
+#elif defined(LEFT)
+			temp = off + 2;
+#else
+			temp = off + 1;
+#endif
+			for (k=0; k<temp; k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3+load4*load3;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3-load4*load3;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3+load4*load3;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3-load4*load3;
+#endif
+				ptrba = ptrba+4;
+				ptrbb = ptrbb+2;
+			}
+			load0 = res0*alphar-res1*alphai;
+			load1 = res1*alphar+res0*alphai;
+			C0[0] = load0;
+			C0[1] = load1;
+
+			load2 = res2*alphar-res3*alphai;
+			load3 = res3*alphar+res2*alphai;
+			C0[2] = load2;
+			C0[3] = load3;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			(!defined(LEFT) && !defined(TRANSA))
+			temp = bk-off;
+#ifdef LEFT
+			temp -= 2;
+#else
+			temp -= 1;
+#endif
+			ptrba += temp*2*2;
+			ptrbb += temp*2;
+#endif
+#ifdef LEFT
+			off += 2;
+#endif
+			C0 = C0+4;
+		}
+		for (i=0; i<(bm&1); i+=1) 
+		{
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			ptrbb = bb;
+#else 
+			ptrba += off*2;
+			ptrbb = bb + off*2;
+#endif
+			res0 = 0;
+			res1 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			temp = bk-off;
+#elif defined(LEFT)
+			temp = off + 1;
+#else 
+			temp = off + 1;
+#endif
+			for (k=0; k<temp; k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+#endif
+				ptrba = ptrba+2;
+				ptrbb = ptrbb+2;
+			}
+			load0 = res0*alphar-res1*alphai;
+			load1 = res1*alphar+res0*alphai;
+			C0[0] = load0;
+			C0[1] = load1;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			(!defined(LEFT) && !defined(TRANSA))
+			temp = bk - off;
+#ifdef LEFT
+			temp -= 1;
+#else 
+			temp -= 1;
+#endif
+			ptrba += temp*2;
+			ptrbb += temp*2;
+#endif
+#ifdef LEFT
+			off += 1;
+#endif
+			C0 = C0+2;
+		}
+		k = (bk<<1);
+		bb = bb+k;
+		i = (ldc<<1);
+		C = C+i;
+	}
+	return 0;
+}

kernel/mips64/KERNEL

@@ -123,15 +123,37 @@ ifndef DTRSMKERNEL_RT
 DTRSMKERNEL_RT	=  trsm_kernel_RT.S
 endif
 
+ifndef	CTRSMKERNEL_LN
 CTRSMKERNEL_LN	=  ztrsm_kernel_LT.S
-CTRSMKERNEL_LT	=  ztrsm_kernel_LT.S
-CTRSMKERNEL_RN	=  ztrsm_kernel_LT.S
-CTRSMKERNEL_RT	=  ztrsm_kernel_RT.S
+endif
 
+ifndef	CTRSMKERNEL_LT
+CTRSMKERNEL_LT	=  ztrsm_kernel_LT.S
+endif
+
+ifndef	CTRSMKERNEL_RN
+CTRSMKERNEL_RN	=  ztrsm_kernel_LT.S
+endif
+
+ifndef	CTRSMKERNEL_RT
+CTRSMKERNEL_RT	=  ztrsm_kernel_RT.S
+endif
+
+ifndef	ZTRSMKERNEL_LN
 ZTRSMKERNEL_LN	=  ztrsm_kernel_LT.S
+endif
+
+ifndef	ZTRSMKERNEL_LT
 ZTRSMKERNEL_LT	=  ztrsm_kernel_LT.S
+endif
+
+ifndef	ZTRSMKERNEL_RN
 ZTRSMKERNEL_RN	=  ztrsm_kernel_LT.S
+endif
+
+ifndef	ZTRSMKERNEL_RT
 ZTRSMKERNEL_RT	=  ztrsm_kernel_RT.S
+endif
 
 CGEMM3MKERNEL    =  zgemm3m_kernel.S
 ZGEMM3MKERNEL    =  zgemm3m_kernel.S

kernel/mips64/KERNEL.LOONGSON3A

@@ -1,18 +1,48 @@
 SAXPYKERNEL=axpy_loongson3a.S
 DAXPYKERNEL=daxpy_loongson3a_simd.S
 
-SGEMMKERNEL    =  sgemm_kernel_loongson3a.S
-SGEMMONCOPY    = ../generic/gemm_ncopy_4.c
-SGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
+SGEMVNKERNEL = gemv_n_loongson3a.c
+SGEMVTKERNEL = gemv_t_loongson3a.c
+DGEMVNKERNEL = gemv_n_loongson3a.c
+DGEMVTKERNEL = gemv_t_loongson3a.c
+CGEMVNKERNEL = zgemv_n_loongson3a.c
+CGEMVTKERNEL = zgemv_t_loongson3a.c
+ZGEMVNKERNEL = zgemv_n_loongson3a.c
+ZGEMVTKERNEL = zgemv_t_loongson3a.c
+
+
+SGEMMKERNEL    =  sgemm_kernel_8x4_ps.S		
+SGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
+SGEMMITCOPY    =  ../generic/gemm_tcopy_8.c
+SGEMMONCOPY    =  ../generic/gemm_ncopy_4.c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+SGEMMINCOPYOBJ =  sgemm_incopy.o
+SGEMMITCOPYOBJ =  sgemm_itcopy.o
 SGEMMONCOPYOBJ =  sgemm_oncopy.o
 SGEMMOTCOPYOBJ =  sgemm_otcopy.o
 
-DGEMMKERNEL    =  gemm_kernel_loongson3a.S
+DGEMMKERNEL    =  dgemm_kernel_loongson3a_4x4.S
 DGEMMONCOPY    = ../generic/gemm_ncopy_4.c
 DGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
 DGEMMONCOPYOBJ = dgemm_oncopy.o
 DGEMMOTCOPYOBJ = dgemm_otcopy.o
 
+CGEMMKERNEL    =  cgemm_kernel_loongson3a_4x2_ps.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.o
+CGEMMITCOPYOBJ =  cgemm_itcopy.o
+CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+
+ZGEMMKERNEL    =  zgemm_kernel_loongson3a_2x2.S
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+ZGEMMONCOPYOBJ =  zgemm_oncopy.o
+ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+
 STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
 STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
 STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
@@ -22,3 +52,17 @@ DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
 DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
 DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
 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/mips64/KERNEL.LOONGSON3B

@@ -0,0 +1,64 @@
+SAXPYKERNEL=axpy_loongson3a.S
+DAXPYKERNEL=daxpy_loongson3a_simd.S
+
+SGEMVNKERNEL = gemv_n_loongson3a.c
+SGEMVTKERNEL = gemv_t_loongson3a.c
+DGEMVNKERNEL = gemv_n_loongson3a.c
+DGEMVTKERNEL = gemv_t_loongson3a.c
+CGEMVNKERNEL = zgemv_n_loongson3a.c
+CGEMVTKERNEL = zgemv_t_loongson3a.c
+ZGEMVNKERNEL = zgemv_n_loongson3a.c
+ZGEMVTKERNEL = zgemv_t_loongson3a.c
+
+STRMMKERNEL	= ../generic/trmmkernel_2x2.c
+DTRMMKERNEL	= ../generic/trmmkernel_2x2.c
+CTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
+ZTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
+
+SGEMMKERNEL    =  ../generic/gemmkernel_2x2.c		
+SGEMMONCOPY    =  ../generic/gemm_ncopy_2.c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_2.c
+SGEMMONCOPYOBJ =  sgemm_oncopy.o
+SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+
+DGEMMKERNEL    =  ../generic/gemmkernel_2x2.c		
+DGEMMONCOPY    = ../generic/gemm_ncopy_2.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy.o
+
+CGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+
+ZGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+ZGEMMONCOPYOBJ =  zgemm_oncopy.o
+ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+
+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	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+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/mips64/gemv_n_loongson3a.c

@@ -0,0 +1,101 @@
+#include "common.h"
+
+//These are auto-tuning codes on Loongson-3A platform. 
+
+//#define prefetch(x) __builtin_prefetch(x)
+//#define prefetch(x) do {_mm_prefetch((char *)(x), _MM_HINT_T0);} while(0)
+#define prefetch(x) __asm__ __volatile__("ld $0, %0"::"m"(x))
+#define likely(x) __builtin_expect(!!(x), 1)
+#define unlikely(x) __builtin_expect(!!(x), 0)
+
+#define spec_loop_alpha1 do {Y[i] += A[LDA * j + i] * X[k]; i++;} while(0)
+#define spec_loop do {Y[i] += ALPHA * A[LDA * j + i] * X[k]; i++;} while(0)
+#define norm_loop_alpha1 do {Y[h] += A[LDA * j + i] * X[k]; i++; h += INCY;} while(0)
+#define norm_loop do {Y[h] += ALPHA * A[LDA * j + i] * X[k]; i++; h += INCY;} while(0)
+
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG UNUSED, FLOAT ALPHA, FLOAT *A, BLASLONG LDA, FLOAT *X, BLASLONG INCX, FLOAT *Y, BLASLONG INCY, FLOAT *BUFFER) 
+{
+
+	BLASLONG kx=0, ky=0;
+	if(!ALPHA)
+		return 0;
+
+	//if(INCX < 0)
+	//	kx = (1-N) * INCX;
+	//	INCX = -INCX;
+	//if(INCY < 0)
+	//	ky = (1-M) * INCY;
+	//	INCY = -INCY;
+
+	BLASLONG fahead = 30;
+	BLASLONG spec_unroll = 4;
+	BLASLONG tMQ = M - M % spec_unroll;
+	BLASLONG j = 0, k = 0;
+
+	if(ALPHA == 1) {
+		if(INCY == 1) {
+			for(k=kx; likely(j < N); j++, k += INCX) {
+				BLASLONG i = 0;
+				for(; likely(i < tMQ);) {
+					prefetch(A[LDA * j + i + fahead]);
+					prefetch(Y[i + fahead]);
+					/*loop_mark*/ spec_loop_alpha1;
+					/*loop_mark*/ spec_loop_alpha1;
+					/*loop_mark*/ spec_loop_alpha1;
+					/*loop_mark*/ spec_loop_alpha1;
+				}
+				for(; likely(i < M);) {
+					spec_loop_alpha1;
+				}
+			}
+		} else {
+			for(k=kx; likely(j < N); j++, k += INCX) {
+				BLASLONG i = 0, h = ky;
+				for(; likely(i < tMQ);) {
+					prefetch(A[LDA * j + i + fahead]);
+					prefetch(Y[h + fahead]);
+					/*loop_mark*/ norm_loop_alpha1;
+					/*loop_mark*/ norm_loop_alpha1;
+					/*loop_mark*/ norm_loop_alpha1;
+					/*loop_mark*/ norm_loop_alpha1;
+				}
+				for(; likely(i < M);) {
+					norm_loop_alpha1;
+				}
+			}
+		}
+	} else {
+		if(INCY == 1) {
+			for(k=kx; likely(j < N); j++, k += INCX) {
+				BLASLONG i = 0;
+				for(; likely(i < tMQ);) {
+					prefetch(A[LDA * j + i + fahead]);
+					prefetch(Y[i + fahead]);
+					/*loop_mark*/ spec_loop;
+					/*loop_mark*/ spec_loop;
+					/*loop_mark*/ spec_loop;
+					/*loop_mark*/ spec_loop;
+				}
+				for(; likely(i < M);) {
+					spec_loop;
+				}
+			}
+		} else {
+			for(k=kx; likely(j < N); j++, k += INCX) {
+				BLASLONG i = 0, h = ky;
+				for(; likely(i < tMQ);) {
+					prefetch(A[LDA * j + i + fahead]);
+					prefetch(Y[h + fahead]);
+					/*loop_mark*/ norm_loop;
+					/*loop_mark*/ norm_loop;
+					/*loop_mark*/ norm_loop;
+					/*loop_mark*/ norm_loop;
+				}
+				for(; likely(i < M);) {
+					norm_loop;
+				}
+			}
+		}
+	}
+	return 0;
+}

kernel/mips64/gemv_t_loongson3a.c

@@ -0,0 +1,93 @@
+#include "common.h"
+
+//These are auto-tuning codes on Loongson-3A platform. 
+
+//#define prefetch(x) __builtin_prefetch(x)
+//#define prefetch(x) do {_mm_prefetch((char *)(x), _MM_HINT_T0);} while(0)
+#define prefetch(x) __asm__ __volatile__("ld $0, %0"::"m"(x))
+#define likely(x) __builtin_expect(!!(x), 1)
+#define unlikely(x) __builtin_expect(!!(x), 0)
+
+#define spec_loop_alpha1 do {Y[k] += A[LDA * j + i] * X[i]; i++;} while(0)
+#define spec_loop do {Y[k] += ALPHA * A[LDA * j + i] * X[i]; i++;} while(0)
+#define norm_loop_alpha1 do {Y[k] += A[LDA * j + i] * X[h]; i++; h += INCX;} while(0)
+#define norm_loop do {Y[k] += ALPHA * A[LDA * j + i] * X[h]; i++; h += INCX;} while(0)
+
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG UNUSED, FLOAT ALPHA, FLOAT *A, BLASLONG LDA, FLOAT *X, BLASLONG INCX, FLOAT *Y, BLASLONG INCY, FLOAT *BUFFER) {
+
+	if(!ALPHA)
+		return 0;
+
+//	if(INCX < 0)
+//		INCX = -INCX;
+//	if(INCY < 0)
+//		INCY = -INCY;
+
+	BLASLONG fahead = 30;
+	BLASLONG spec_unroll = 3;
+	BLASLONG tMQ = M - M % spec_unroll;
+	BLASLONG j = 0, k = 0;
+
+	if(ALPHA == 1) {
+		if(INCX == 1) {
+			for(; likely(j < N); j++, k += INCY) {
+				BLASLONG i = 0;
+				for(; likely(i < tMQ);) {
+					prefetch(A[LDA * j + i + fahead]);
+					prefetch(X[i + fahead]);
+					/*loop_mark*/ spec_loop_alpha1;
+					/*loop_mark*/ spec_loop_alpha1;
+					/*loop_mark*/ spec_loop_alpha1;
+				}
+				for(; likely(i < M);) {
+					spec_loop_alpha1;
+				}
+			}
+		} else {
+			for(; likely(j < N); j++, k += INCY) {
+				BLASLONG i = 0, h = 0;
+				for(; likely(i < tMQ);) {
+					prefetch(A[LDA * j + i + fahead]);
+					prefetch(X[h + fahead]);
+					/*loop_mark*/ norm_loop_alpha1;
+					/*loop_mark*/ norm_loop_alpha1;
+					/*loop_mark*/ norm_loop_alpha1;
+				}
+				for(; likely(i < M);) {
+					norm_loop_alpha1;
+				}
+			}
+		}
+	} else {
+		if(INCX == 1) {
+			for(; likely(j < N); j++, k += INCY) {
+				BLASLONG i = 0;
+				for(; likely(i < tMQ);) {
+					prefetch(A[LDA * j + i + fahead]);
+					prefetch(X[i + fahead]);
+					/*loop_mark*/ spec_loop;
+					/*loop_mark*/ spec_loop;
+					/*loop_mark*/ spec_loop;
+				}
+				for(; likely(i < M);) {
+					spec_loop;
+				}
+			}
+		} else {
+			for(; likely(j < N); j++, k += INCY) {
+				BLASLONG i = 0, h = 0;
+				for(; likely(i < tMQ);) {
+					prefetch(A[LDA * j + i + fahead]);
+					prefetch(X[h + fahead]);
+					/*loop_mark*/ norm_loop;
+					/*loop_mark*/ norm_loop;
+					/*loop_mark*/ norm_loop;
+				}
+				for(; likely(i < M);) {
+					norm_loop;
+				}
+			}
+		}
+	}
+	return 0;
+}

kernel/mips64/zgemv_n_loongson3a.c

@@ -0,0 +1,139 @@
+#include "common.h" 
+
+//typedef int BLASLONG;
+//typedef double FLOAT;
+
+#define prefetch(x) __asm__ __volatile__("ld $0, %0"::"m"(x))
+#define likely(x) __builtin_expect(!!(x), 1)
+#define unlikely(x) __builtin_expect(!!(x), 0)
+
+#if !defined(CONJ) && !defined(XCONJ)
+#define spec_loop_alpha1	spec_loop_alpha1_0
+#define spec_loop			spec_loop_0
+#define norm_loop_alpha1	norm_loop_alpha1_0
+#define norm_loop			norm_loop_0
+#endif
+
+#if  defined(CONJ) && !defined(XCONJ)
+#define spec_loop_alpha1 	spec_loop_alpha1_1
+#define spec_loop			spec_loop_1
+#define norm_loop_alpha1	norm_loop_alpha1_1
+#define norm_loop			norm_loop_1
+#endif
+
+#if  !defined(CONJ) && defined(XCONJ)
+#define spec_loop_alpha1 	spec_loop_alpha1_2
+#define spec_loop			spec_loop_2
+#define norm_loop_alpha1	norm_loop_alpha1_2
+#define norm_loop			norm_loop_2
+#endif
+
+#if  defined(CONJ) && defined(XCONJ)
+#define spec_loop_alpha1 	spec_loop_alpha1_3
+#define spec_loop			spec_loop_3
+#define norm_loop_alpha1	norm_loop_alpha1_3
+#define norm_loop			norm_loop_3
+#endif
+
+#define spec_loop_alpha1_0 do {Y[ii] += A[jj + ii] * X[k]; Y[ii + 1] += A[jj + ii + 1] * X[k]; Y[ii + 1] += A[jj + ii] * X[k + 1]; Y[ii] -= A[jj + ii + 1] * X[k + 1]; ii += 2;} while(0)
+
+#define spec_loop_alpha1_1 do {Y[ii] += A[jj + ii] * X[k]; Y[ii + 1] -= A[jj + ii + 1] * X[k]; Y[ii + 1] += A[jj + ii] * X[k + 1]; Y[ii] += A[jj + ii + 1] * X[k + 1]; ii += 2;} while(0)
+
+#define spec_loop_alpha1_2 do {Y[ii] += A[jj + ii] * X[k]; Y[ii + 1] += A[jj + ii + 1] * X[k]; Y[ii + 1] -= A[jj + ii] * X[k + 1]; Y[ii] += A[jj + ii + 1] * X[k + 1]; ii += 2;} while(0)
+
+#define spec_loop_alpha1_3 do {Y[ii] += A[jj + ii] * X[k]; Y[ii + 1] -= A[jj + ii + 1] * X[k]; Y[ii + 1] -= A[jj + ii] * X[k + 1]; Y[ii] -= A[jj + ii + 1] * X[k + 1]; ii += 2;} while(0)
+
+#define spec_loop_0 do {rTmp = A[jj + ii] * X[k] - A[jj + ii + 1] * X[k + 1]; iTmp = A[jj + ii] * X[k + 1] + A[jj + ii + 1] * X[k]; Y[ii] += rTmp * rALPHA - iTmp * iALPHA; Y[ii + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2;} while(0)
+
+#define spec_loop_1 do {rTmp = A[jj + ii] * X[k] + A[jj + ii + 1] * X[k + 1]; iTmp = A[jj + ii] * X[k + 1] - A[jj + ii + 1] * X[k]; Y[ii] += rTmp * rALPHA - iTmp * iALPHA; Y[ii + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2;} while(0)
+
+#define spec_loop_2 do {rTmp = A[jj + ii] * X[k] + A[jj + ii + 1] * X[k + 1]; iTmp = -A[jj + ii] * X[k + 1] + A[jj + ii + 1] * X[k]; Y[ii] += rTmp * rALPHA - iTmp * iALPHA; Y[ii + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2;} while(0)
+
+#define spec_loop_3 do {rTmp = A[jj + ii] * X[k] - A[jj + ii + 1] * X[k + 1]; iTmp = -A[jj + ii] * X[k + 1] - A[jj + ii + 1] * X[k]; Y[ii] += rTmp * rALPHA - iTmp * iALPHA; Y[ii + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2;} while(0)
+
+#define norm_loop_alpha1_0 do {Y[iii] += A[jj + ii] * X[k] - A[jj + ii + 1] * X[k + 1]; Y[iii + 1] += A[jj + ii] * X[k + 1] + A[jj + ii + 1] * X[k]; ii += 2; iii += INCY * 2;} while(0)
+
+#define norm_loop_alpha1_1 do {Y[iii] += A[jj + ii] * X[k] + A[jj + ii + 1] * X[k + 1]; Y[iii + 1] += A[jj + ii] * X[k + 1] - A[jj + ii + 1] * X[k]; ii += 2; iii += INCY * 2;} while(0)
+
+#define norm_loop_alpha1_2 do {Y[iii] += A[jj + ii] * X[k] + A[jj + ii + 1] * X[k + 1]; Y[iii + 1] += -A[jj + ii] * X[k + 1] + A[jj + ii + 1] * X[k]; ii += 2; iii += INCY * 2;} while(0)
+
+#define norm_loop_alpha1_3 do {Y[iii] += A[jj + ii] * X[k] - A[jj + ii + 1] * X[k + 1]; Y[iii + 1] += -A[jj + ii] * X[k + 1] - A[jj + ii + 1] * X[k]; ii += 2; iii += INCY * 2;} while(0)
+
+#define norm_loop_0 do {rTmp = A[jj + ii] * X[k] - A[jj + ii + 1] * X[k + 1]; iTmp = A[jj + ii] * X[k + 1] + A[jj + ii + 1] * X[k]; Y[iii] += rTmp * rALPHA - iTmp * iALPHA; Y[iii + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2; iii += INCY * 2;} while(0)
+
+#define norm_loop_1 do {rTmp = A[jj + ii] * X[k] + A[jj + ii + 1] * X[k + 1]; iTmp = A[jj + ii] * X[k + 1] - A[jj + ii + 1] * X[k]; Y[iii] += rTmp * rALPHA - iTmp * iALPHA; Y[iii + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2; iii += INCY * 2;} while(0)
+
+#define norm_loop_2 do {rTmp = A[jj + ii] * X[k] + A[jj + ii + 1] * X[k + 1]; iTmp = -A[jj + ii] * X[k + 1] + A[jj + ii + 1] * X[k]; Y[iii] += rTmp * rALPHA - iTmp * iALPHA; Y[iii + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2; iii += INCY * 2;} while(0)
+
+#define norm_loop_3 do {rTmp = A[jj + ii] * X[k] - A[jj + ii + 1] * X[k + 1]; iTmp = -A[jj + ii] * X[k + 1] - A[jj + ii + 1] * X[k]; Y[iii] += rTmp * rALPHA - iTmp * iALPHA; Y[iii + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2; iii += INCY * 2;} while(0)
+
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG UNUSED, FLOAT rALPHA, FLOAT iALPHA, FLOAT *A, BLASLONG LDA, FLOAT *X, BLASLONG INCX, FLOAT *Y, BLASLONG INCY, FLOAT *BUFFER) {
+
+	if(!rALPHA && iALPHA)
+		return 0;
+
+	BLASLONG fahead = 60;
+	BLASLONG spec_unroll = 2;
+	BLASLONG tMQ = M - M % spec_unroll;
+	BLASLONG j = 0, k = 0, jj = 0;
+
+	if(rALPHA == 1 && iALPHA == 0) {
+		if(INCY == 1) {
+			for(; likely(j < N); j++, k += INCX * 2, jj += LDA * 2) {
+				BLASLONG i = 0, ii = 0;
+				for(; likely(i < tMQ); i += spec_unroll) {
+					prefetch(A[jj + ii + fahead]);
+					prefetch(Y[ii + fahead]);
+					/*loop_mark*/ spec_loop_alpha1;
+					/*loop_mark*/ spec_loop_alpha1;
+				}
+				for(; likely(i < M); i++) {
+					spec_loop_alpha1;
+				}
+			}
+		} else {
+			for(; likely(j < N); j++, k += INCX * 2, jj += LDA * 2) {
+				BLASLONG i = 0, ii = 0, iii = 0;
+				for(; likely(i < tMQ); i += spec_unroll) {
+					prefetch(A[jj + ii + fahead]);
+					prefetch(Y[iii + fahead]);
+					/*loop_mark*/ norm_loop_alpha1;
+					/*loop_mark*/ norm_loop_alpha1;
+				}
+				for(; likely(i < M); i++) {
+					norm_loop_alpha1;
+				}
+			}
+		}
+	} else {
+		FLOAT rTmp, iTmp;
+		if(INCY == 1) {
+			for(; likely(j < N); j++, k += INCX * 2, jj += LDA * 2) {
+				BLASLONG i = 0, ii = 0;
+				for(; likely(i < tMQ); i += spec_unroll) {
+					prefetch(A[jj + ii + fahead]);
+					prefetch(Y[ii + fahead]);
+					/*loop_mark*/ spec_loop;
+					/*loop_mark*/ spec_loop;
+				}
+				for(; likely(i < M); i++) {
+					spec_loop;
+				}
+			}
+		} else {
+			for(; likely(j < N); j++, k += INCX * 2, jj += LDA * 2) {
+				BLASLONG i = 0, ii = 0, iii = 0;
+				for(; likely(i < tMQ); i += spec_unroll) {
+					prefetch(A[jj + ii + fahead]);
+					prefetch(Y[iii + fahead]);
+					/*loop_mark*/ norm_loop;
+					/*loop_mark*/ norm_loop;
+				}
+				for(; likely(i < M); i++) {
+					norm_loop;
+				}
+			}
+		}
+	}
+	return 0;
+}

kernel/mips64/zgemv_t_loongson3a.c

@@ -0,0 +1,125 @@
+#include "common.h" 
+
+#define prefetch(x) __asm__ __volatile__("ld $0, %0"::"m"(x))
+#define likely(x) __builtin_expect(!!(x), 1)
+#define unlikely(x) __builtin_expect(!!(x), 0)
+
+#if !defined(CONJ) && !defined(XCONJ)
+#define spec_loop_alpha1	spec_loop_alpha1_0
+#define spec_loop			spec_loop_0
+#define norm_loop_alpha1	norm_loop_alpha1_0
+#define norm_loop			norm_loop_0
+#endif
+
+#if  defined(CONJ) && !defined(XCONJ)
+#define spec_loop_alpha1 	spec_loop_alpha1_1
+#define spec_loop			spec_loop_1
+#define norm_loop_alpha1	norm_loop_alpha1_1
+#define norm_loop			norm_loop_1
+#endif
+
+#if  !defined(CONJ) && defined(XCONJ)
+#define spec_loop_alpha1 	spec_loop_alpha1_2
+#define spec_loop			spec_loop_2
+#define norm_loop_alpha1	norm_loop_alpha1_2
+#define norm_loop			norm_loop_2
+#endif
+
+#if  defined(CONJ) && defined(XCONJ)
+#define spec_loop_alpha1 	spec_loop_alpha1_3
+#define spec_loop			spec_loop_3
+#define norm_loop_alpha1	norm_loop_alpha1_3
+#define norm_loop			norm_loop_3
+#endif
+
+
+#define spec_loop_alpha1_0 do {Y[k] += A[jj + ii] * X[ii]; Y[k + 1] += A[jj + ii + 1] * X[ii]; Y[k + 1] += A[jj + ii] * X[ii + 1]; Y[k] -= A[jj + ii + 1] * X[ii + 1]; ii += 2;} while(0)
+#define spec_loop_alpha1_1 do {Y[k] += A[jj + ii] * X[ii]; Y[k + 1] -= A[jj + ii + 1] * X[ii]; Y[k + 1] += A[jj + ii] * X[ii + 1]; Y[k] += A[jj + ii + 1] * X[ii + 1]; ii += 2;} while(0)
+#define spec_loop_alpha1_2 do {Y[k] += A[jj + ii] * X[ii]; Y[k + 1] += A[jj + ii + 1] * X[ii]; Y[k + 1] -= A[jj + ii] * X[ii + 1]; Y[k] += A[jj + ii + 1] * X[ii + 1]; ii += 2;} while(0)
+#define spec_loop_alpha1_3 do {Y[k] += A[jj + ii] * X[ii]; Y[k + 1] -= A[jj + ii + 1] * X[ii]; Y[k + 1] -= A[jj + ii] * X[ii + 1]; Y[k] -= A[jj + ii + 1] * X[ii + 1]; ii += 2;} while(0)
+
+#define spec_loop_0 do {rTmp = A[jj + ii] * X[ii] - A[jj + ii + 1] * X[ii + 1]; iTmp = A[jj + ii] * X[ii + 1] + A[jj + ii + 1] * X[ii]; Y[k] += rTmp * rALPHA - iTmp * iALPHA; Y[k + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2;} while(0)
+#define spec_loop_1 do {rTmp = A[jj + ii] * X[ii] + A[jj + ii + 1] * X[ii + 1]; iTmp = A[jj + ii] * X[ii + 1] - A[jj + ii + 1] * X[ii]; Y[k] += rTmp * rALPHA - iTmp * iALPHA; Y[k + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2;} while(0)
+#define spec_loop_2 do {rTmp = A[jj + ii] * X[ii] + A[jj + ii + 1] * X[ii + 1]; iTmp = -A[jj + ii] * X[ii + 1] + A[jj + ii + 1] * X[ii]; Y[k] += rTmp * rALPHA - iTmp * iALPHA; Y[k + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2;} while(0)
+#define spec_loop_3 do {rTmp = A[jj + ii] * X[ii] - A[jj + ii + 1] * X[ii + 1]; iTmp = -A[jj + ii] * X[ii + 1] - A[jj + ii + 1] * X[ii]; Y[k] += rTmp * rALPHA - iTmp * iALPHA; Y[k + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2;} while(0)
+
+#define norm_loop_alpha1_0 do {Y[k] += A[jj + ii] * X[iii] - A[jj + ii + 1] * X[iii + 1]; Y[k + 1] += A[jj + ii] * X[iii + 1] + A[jj + ii + 1] * X[iii]; ii += 2; iii += INCX * 2;} while(0)
+#define norm_loop_alpha1_1 do {Y[k] += A[jj + ii] * X[iii] + A[jj + ii + 1] * X[iii + 1]; Y[k + 1] += A[jj + ii] * X[iii + 1] - A[jj + ii + 1] * X[iii]; ii += 2; iii += INCX * 2;} while(0)
+#define norm_loop_alpha1_2 do {Y[k] += A[jj + ii] * X[iii] + A[jj + ii + 1] * X[iii + 1]; Y[k + 1] += -A[jj + ii] * X[iii + 1] + A[jj + ii + 1] * X[iii]; ii += 2; iii += INCX * 2;} while(0)
+#define norm_loop_alpha1_3 do {Y[k] += A[jj + ii] * X[iii] - A[jj + ii + 1] * X[iii + 1]; Y[k + 1] += -A[jj + ii] * X[iii + 1] - A[jj + ii + 1] * X[iii]; ii += 2; iii += INCX * 2;} while(0)
+
+#define norm_loop_0 do {rTmp = A[jj + ii] * X[iii] - A[jj + ii + 1] * X[iii + 1]; iTmp = A[jj + ii] * X[iii + 1] + A[jj + ii + 1] * X[iii]; Y[k] += rTmp * rALPHA - iTmp * iALPHA; Y[k + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2; iii += INCX * 2;} while(0)
+#define norm_loop_1 do {rTmp = A[jj + ii] * X[iii] + A[jj + ii + 1] * X[iii + 1]; iTmp = A[jj + ii] * X[iii + 1] - A[jj + ii + 1] * X[iii]; Y[k] += rTmp * rALPHA - iTmp * iALPHA; Y[k + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2; iii += INCX * 2;} while(0)
+#define norm_loop_2 do {rTmp = A[jj + ii] * X[iii] + A[jj + ii + 1] * X[iii + 1]; iTmp = -A[jj + ii] * X[iii + 1] + A[jj + ii + 1] * X[iii]; Y[k] += rTmp * rALPHA - iTmp * iALPHA; Y[k + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2; iii += INCX * 2;} while(0)
+#define norm_loop_3 do {rTmp = A[jj + ii] * X[iii] - A[jj + ii + 1] * X[iii + 1]; iTmp = -A[jj + ii] * X[iii + 1] - A[jj + ii + 1] * X[iii]; Y[k] += rTmp * rALPHA - iTmp * iALPHA; Y[k + 1] += rTmp * iALPHA + iTmp * rALPHA; ii += 2; iii += INCX * 2;} while(0)
+
+int CNAME(BLASLONG M, BLASLONG N, BLASLONG UNUSED, FLOAT rALPHA, FLOAT iALPHA, FLOAT *A, BLASLONG LDA, FLOAT *X, BLASLONG INCX, FLOAT *Y, BLASLONG INCY, FLOAT *BUFFER) {
+
+	if(!rALPHA && iALPHA)
+		return 0;
+
+	BLASLONG fahead = 30;
+	BLASLONG spec_unroll = 2;
+	BLASLONG tMQ = M - M % spec_unroll;
+	BLASLONG j = 0, k = 0, jj = 0;
+
+	if(rALPHA == 1 && iALPHA == 0) {
+		if(INCX == 1) {
+			for(; likely(j < N); j++, k += INCY * 2, jj += LDA * 2) {
+				BLASLONG i = 0, ii = 0;
+				for(; likely(i < tMQ); i += spec_unroll) {
+					prefetch(A[jj + ii + fahead]);
+					prefetch(X[ii + fahead]);
+					/*loop_mark*/ spec_loop_alpha1;
+					/*loop_mark*/ spec_loop_alpha1;
+				}
+				for(; likely(i < M); i++) {
+					spec_loop_alpha1;
+				}
+			}
+		} else {
+			for(; likely(j < N); j++, k += INCY * 2, jj += LDA * 2) {
+				BLASLONG i = 0, ii = 0, iii = 0;
+				for(; likely(i < tMQ); i += spec_unroll) {
+					prefetch(A[jj + ii + fahead]);
+					prefetch(X[iii + fahead]);
+					/*loop_mark*/ norm_loop_alpha1;
+					/*loop_mark*/ norm_loop_alpha1;
+				}
+				for(; likely(i < M); i++) {
+					norm_loop_alpha1;
+				}
+			}
+		}
+	} else {
+		FLOAT rTmp, iTmp;
+		if(INCX == 1) {
+			for(; likely(j < N); j++, k += INCY * 2, jj += LDA * 2) {
+				BLASLONG i = 0, ii = 0;
+				for(; likely(i < tMQ); i += spec_unroll) {
+					prefetch(A[jj + ii + fahead]);
+					prefetch(X[ii + fahead]);
+					/*loop_mark*/ spec_loop;
+					/*loop_mark*/ spec_loop;
+				}
+				for(; likely(i < M); i++) {
+					spec_loop;
+				}
+			}
+		} else {
+			for(; likely(j < N); j++, k += INCY * 2, jj += LDA * 2) {
+				BLASLONG i = 0, ii = 0, iii = 0;
+				for(; likely(i < tMQ); i += spec_unroll) {
+					prefetch(A[jj + ii + fahead]);
+					prefetch(X[iii + fahead]);
+					/*loop_mark*/ norm_loop;
+					/*loop_mark*/ norm_loop;
+				}
+				for(; likely(i < M); i++) {
+					norm_loop;
+				}
+			}
+		}
+	}
+	return 0;
+}

kernel/setparam-ref.c

@@ -746,6 +746,22 @@ static void init_parameter(void) {
 #endif
 #endif
 
+#ifdef SANDYBRIDGE
+
+#ifdef DEBUG
+  fprintf(stderr, "Sandybridge\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 OPTERON
 
 #ifdef DEBUG
@@ -778,6 +794,22 @@ static void init_parameter(void) {
 #endif
 #endif
 
+#ifdef BOBCAT
+
+#ifdef DEBUG
+  fprintf(stderr, "Bobcate\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/KERNEL

@@ -239,6 +239,22 @@ ifndef ZSWAPKERNEL
 ZSWAPKERNEL = zswap_sse2.S
 endif
 
+ifndef DGEMVNKERNEL
+DGEMVNKERNEL = gemv_n_sse2.S
+endif
+
+ifndef DGEMVTKERNEL
+DGEMVTKERNEL = gemv_t_sse2.S
+endif
+
+ifndef ZGEMVNKERNEL
+ZGEMVNKERNEL = zgemv_n_sse2.S
+endif
+
+ifndef ZGEMVTKERNEL
+ZGEMVTKERNEL = zgemv_t_sse2.S
+endif
+
 endif
 
 

kernel/x86/KERNEL.BOBCAT

@@ -0,0 +1,59 @@
+SGEMMKERNEL    =  gemm_kernel_4x4_barcelona.S
+SGEMMINCOPY    =  
+SGEMMITCOPY    =  
+SGEMMONCOPY    =  ../generic/gemm_ncopy_4.c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+SGEMMINCOPYOBJ =  
+SGEMMITCOPYOBJ =  
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+DGEMMKERNEL    =  gemm_kernel_2x4_barcelona.S
+DGEMMINCOPY    =  ../generic/gemm_ncopy_2.c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_2.c
+DGEMMONCOPY    =  ../generic/gemm_ncopy_4.c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+CGEMMKERNEL    =  zgemm_kernel_2x2_barcelona.S
+CGEMMINCOPY    =  
+CGEMMITCOPY    =  
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
+CGEMMINCOPYOBJ =
+CGEMMITCOPYOBJ =  
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMKERNEL    =  zgemm_kernel_1x2_barcelona.S
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_1.c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_1.c
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+STRSMKERNEL_LN	=  trsm_kernel_LN_4x4_sse.S
+STRSMKERNEL_LT	=  trsm_kernel_LT_4x4_sse.S
+STRSMKERNEL_RN	=  trsm_kernel_LT_4x4_sse.S
+STRSMKERNEL_RT	=  trsm_kernel_RT_4x4_sse.S
+
+DTRSMKERNEL_LN	=  trsm_kernel_LN_2x4_sse2.S
+DTRSMKERNEL_LT	=  trsm_kernel_LT_2x4_sse2.S
+DTRSMKERNEL_RN	=  trsm_kernel_LT_2x4_sse2.S
+DTRSMKERNEL_RT	=  trsm_kernel_RT_2x4_sse2.S
+
+CTRSMKERNEL_LN	=  ztrsm_kernel_LN_2x2_sse.S
+CTRSMKERNEL_LT	=  ztrsm_kernel_LT_2x2_sse.S
+CTRSMKERNEL_RN	=  ztrsm_kernel_LT_2x2_sse.S
+CTRSMKERNEL_RT	=  ztrsm_kernel_RT_2x2_sse.S
+
+ZTRSMKERNEL_LN	=  ztrsm_kernel_LT_1x2_sse2.S
+ZTRSMKERNEL_LT	=  ztrsm_kernel_LT_1x2_sse2.S
+ZTRSMKERNEL_RN	=  ztrsm_kernel_LT_1x2_sse2.S
+ZTRSMKERNEL_RT	=  ztrsm_kernel_RT_1x2_sse2.S
+
+CGEMM3MKERNEL    =  zgemm3m_kernel_4x4_barcelona.S
+ZGEMM3MKERNEL    =  zgemm3m_kernel_2x4_barcelona.S

kernel/x86/KERNEL.SANDYBRIDGE

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

kernel/x86/amax_sse.S

@@ -495,7 +495,6 @@
 	ALIGN_4
 	
 .L999:
-	RESTOREREGISTERS
 
 	subl	$8, %esp
 	movss	%xmm0, (%esp)

kernel/x86/gemm_kernel_2x4_penryn.S

@@ -76,6 +76,12 @@
 #define PREFETCHB     prefetcht0
 #endif
 
+#ifdef SANDYBRIDGE
+#define PREFETCHSIZE  (8 * 1 - 4)
+#define PREFETCHW     prefetcht0
+#define PREFETCHB     prefetcht0
+#endif
+
 #ifndef PREFETCH
 #define PREFETCH      prefetcht0
 #endif

kernel/x86/gemm_kernel_4x4_penryn.S

@@ -69,6 +69,12 @@
 #define PREFETCHB     prefetcht0
 #endif
 
+#ifdef SANDYBRIDGE
+#define PREFETCHSIZE  (16 * 1 - 8)
+#define PREFETCHW     prefetcht0
+#define PREFETCHB     prefetcht0
+#endif
+
 #ifndef PREFETCH
 #define PREFETCH      prefetcht0
 #endif
@@ -262,7 +268,7 @@
 	movaps	-16 * SIZE(AA), %xmm0
 
 	addps	%xmm2, %xmm7
-#ifndef NEHALEM
+#if !(defined(NEHALEM) || defined(SANDYBRIDGE))
 	PREFETCH (PREFETCHSIZE + 16) * SIZE(AA)
 #endif
 	pshufd	$0x93, %xmm1, %xmm2

kernel/x86/gemv_n_sse.S

@@ -58,7 +58,7 @@
 #define PREFETCHSIZE	(16 * 4)
 #endif
 
-#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM)
+#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 7)

kernel/x86/gemv_n_sse2.S

@@ -45,7 +45,7 @@
 #define PREFETCHSIZE	(8 * 2)
 #endif
 
-#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM)
+#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(8 * 7)

kernel/x86/gemv_t_sse.S

@@ -58,7 +58,7 @@
 #define PREFETCHSIZE	(16 * 4)
 #endif
 
-#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM)
+#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 7)

kernel/x86/gemv_t_sse2.S

@@ -45,7 +45,7 @@
 #define PREFETCHSIZE	(8 * 2)
 #endif
 
-#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM)
+#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(8 * 7)

kernel/x86/scal_sse2.S

@@ -76,7 +76,8 @@
 	xorps	%xmm1, %xmm1
 	comisd	%xmm0, %xmm1
 	jne	.L100		# Alpha != ZERO
-
+        jp      .L100           # For Alpha = NaN
+	
 /* Alpha == ZERO */
 	cmpl	$SIZE, INCX
 	jne	.L50

kernel/x86/trsm_kernel_LN_2x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_LN_2x4_sse2.S

@@ -69,7 +69,7 @@
 #define STACK_ALIGN	4096
 #define STACK_OFFSET	1024
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 #define PREFETCH     prefetch
 #define PREFETCHSIZE (8 * 10 + 4)
 #endif
@@ -439,7 +439,7 @@
 .L22:
 	mulsd	%xmm0, %xmm2
 	addsd	%xmm2, %xmm4
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	PREFETCH (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movlpd	 2 * SIZE(BB), %xmm2
@@ -488,7 +488,7 @@
 	movlpd	40 * SIZE(BB), %xmm3
 	addsd	%xmm0, %xmm7
 	movlpd	 8 * SIZE(AA), %xmm0
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	PREFETCH (PREFETCHSIZE  + 8) * SIZE(AA)
 #endif
 	mulsd	%xmm1, %xmm2
@@ -1697,7 +1697,7 @@
 
 .L42:
 	mulpd	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	mulpd	 2 * SIZE(BB), %xmm0
@@ -1727,7 +1727,7 @@
 	addpd	%xmm0, %xmm7
 	movapd	16 * SIZE(AA), %xmm0
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 8) * SIZE(AA)
 #endif
 	mulpd	%xmm1, %xmm2

kernel/x86/trsm_kernel_LN_4x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_LN_4x4_sse.S

@@ -64,7 +64,7 @@
 #define BORIG	60(%esp)
 #define BUFFER 128(%esp)
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 #define PREFETCH     prefetch
 #define PREFETCHW    prefetchw
 #define PREFETCHSIZE (16 * 10 + 8)
@@ -437,7 +437,7 @@
 .L32:
 	mulss	%xmm0, %xmm2
 	addss	%xmm2, %xmm4
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movss	 4 * SIZE(BB), %xmm2
@@ -833,7 +833,7 @@
 .L22:
 	mulps	%xmm0, %xmm2
 	addps	%xmm2, %xmm4
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movaps	 4 * SIZE(BB), %xmm2
@@ -1848,7 +1848,7 @@
 
 .L72:
 	mulss	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	mulss	 4 * SIZE(BB), %xmm0
@@ -2109,7 +2109,7 @@
 	ALIGN_4
 
 .L62:
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 
@@ -2429,7 +2429,7 @@
 
 .L52:
 	mulps	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	mulps	 4 * SIZE(BB), %xmm0
@@ -2459,7 +2459,7 @@
 	addps	%xmm0, %xmm5
 	movaps	32 * SIZE(AA), %xmm0
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 16) * SIZE(AA)
 #endif
 	mulps	%xmm1, %xmm2
@@ -2952,7 +2952,7 @@
 
 .L112:
 	mulss	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movss	 1 * SIZE(AA), %xmm0
@@ -3148,7 +3148,7 @@
 
 .L102:
 	mulps	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movsd	 2 * SIZE(AA), %xmm0
@@ -3389,7 +3389,7 @@
 
 .L92:
 	mulps	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movaps	 4 * SIZE(AA), %xmm0
@@ -3404,7 +3404,7 @@
 	mulps	12 * SIZE(BB), %xmm0
 	addps	%xmm0, %xmm7
 	movaps	32 * SIZE(AA), %xmm0
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 16) * SIZE(AA)
 #endif
 	mulps	%xmm1, %xmm3

kernel/x86/trsm_kernel_LT_2x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_LT_2x4_sse2.S

@@ -69,7 +69,7 @@
 #define STACK_ALIGN	4096
 #define STACK_OFFSET	1024
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 #define PREFETCH     prefetch
 #define PREFETCHSIZE (8 * 10 + 4)
 #endif
@@ -910,7 +910,7 @@
 .L22:
 	mulsd	%xmm0, %xmm2
 	addsd	%xmm2, %xmm4
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	PREFETCH (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movlpd	 2 * SIZE(BB), %xmm2
@@ -959,7 +959,7 @@
 	movlpd	40 * SIZE(BB), %xmm3
 	addsd	%xmm0, %xmm7
 	movlpd	 8 * SIZE(AA), %xmm0
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	PREFETCH (PREFETCHSIZE  + 8) * SIZE(AA)
 #endif
 	mulsd	%xmm1, %xmm2
@@ -1439,7 +1439,7 @@
 
 .L42:
 	mulpd	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	mulpd	 2 * SIZE(BB), %xmm0
@@ -1469,7 +1469,7 @@
 	addpd	%xmm0, %xmm7
 	movapd	16 * SIZE(AA), %xmm0
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 8) * SIZE(AA)
 #endif
 	mulpd	%xmm1, %xmm2

kernel/x86/trsm_kernel_LT_4x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_LT_4x4_sse.S

@@ -64,7 +64,7 @@
 #define BORIG	60(%esp)
 #define BUFFER 128(%esp)
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 #define PREFETCH     prefetch
 #define PREFETCHW    prefetchw
 #define PREFETCHSIZE (16 * 10 + 8)
@@ -872,7 +872,7 @@
 .L22:
 	mulps	%xmm0, %xmm2
 	addps	%xmm2, %xmm4
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movaps	 4 * SIZE(BB), %xmm2
@@ -1316,7 +1316,7 @@
 .L32:
 	mulss	%xmm0, %xmm2
 	addss	%xmm2, %xmm4
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movss	 4 * SIZE(BB), %xmm2
@@ -1855,7 +1855,7 @@
 
 .L52:
 	mulps	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	mulps	 4 * SIZE(BB), %xmm0
@@ -1885,7 +1885,7 @@
 	addps	%xmm0, %xmm5
 	movaps	32 * SIZE(AA), %xmm0
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 16) * SIZE(AA)
 #endif
 	mulps	%xmm1, %xmm2
@@ -2249,7 +2249,7 @@
 	ALIGN_4
 
 .L62:
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 
@@ -2562,7 +2562,7 @@
 
 .L72:
 	mulss	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	mulss	 4 * SIZE(BB), %xmm0
@@ -2957,7 +2957,7 @@
 
 .L92:
 	mulps	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movaps	 4 * SIZE(AA), %xmm0
@@ -2972,7 +2972,7 @@
 	mulps	12 * SIZE(BB), %xmm0
 	addps	%xmm0, %xmm7
 	movaps	32 * SIZE(AA), %xmm0
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 16) * SIZE(AA)
 #endif
 	mulps	%xmm1, %xmm3
@@ -3280,7 +3280,7 @@
 
 .L102:
 	mulps	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movsd	 2 * SIZE(AA), %xmm0
@@ -3515,7 +3515,7 @@
 
 .L112:
 	mulss	%xmm0, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 	movss	 1 * SIZE(AA), %xmm0

kernel/x86/trsm_kernel_RT_2x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif