Merge branch 'release-0.2.0'

.gitignore

@@ -1,16 +1,23 @@
 *.obj
 *.lib
 *.dll
+*.dylib
 *.def
 *.o
 lapack-3.1.1
 lapack-3.1.1.tgz
+lapack-3.4.1
+lapack-3.4.1.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 +1,17 @@
 OpenBLAS ChangeLog
+====================================================================
+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
@@ -7,6 +20,8 @@ common:
 	* 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.

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
 

Makefile

@@ -256,12 +256,17 @@ LAPACK_URL=http://www.netlib.org/lapack/lapack-3.4.1.tgz
 
 lapack-3.4.1.tgz :
 ifndef NOFORTRAN
-ifeq ($(OSNAME), Darwin)
+#http://stackoverflow.com/questions/7656425/makefile-ifeq-logical-or
+ifeq ($(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 : 
 ifndef NOFORTRAN

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.1.1
+VERSION = 0.2.0
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library 
@@ -94,8 +94,8 @@ VERSION = 0.1.1
 
 # If any gemm arguement m, n or k is less or equal this threshold, gemm will be execute 
 # with single thread. You can use this flag to avoid the overhead of multi-threading 
-# in small matrix sizes. The default value is 4.
-# GEMM_MULTITHREAD_THRESHOLD = 4
+# 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).

Makefile.system

@@ -14,7 +14,15 @@ NETLIB_LAPACK_DIR = $(TOPDIR)/lapack-3.4.1
 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
@@ -45,7 +53,7 @@ GETARCH_FLAGS	+= -DUSE64BITINT
 endif
 
 ifndef GEMM_MULTITHREAD_THRESHOLD
-GEMM_MULTITHREAD_THRESHOLD=4
+GEMM_MULTITHREAD_THRESHOLD=50
 endif
 GETARCH_FLAGS	+= -DGEMM_MULTITHREAD_THRESHOLD=$(GEMM_MULTITHREAD_THRESHOLD) 
 
@@ -108,6 +116,14 @@ export MACOSX_DEPLOYMENT_TARGET=10.2
 MD5SUM = md5 -r
 endif
 
+ifeq ($(OSNAME), FreeBSD)
+MD5SUM = md5 -r
+endif
+
+ifeq ($(OSNAME), NetBSD)
+MD5SUM = md5 -r
+endif
+
 ifeq ($(OSNAME), Linux)
 EXTRALIB	+= -lm
 endif
@@ -231,11 +247,11 @@ endif
 ifdef DYNAMIC_ARCH
 ifeq ($(ARCH), x86)
 DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
-	       CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA ATOM NANO
+	       CORE2 PENRYN DUNNINGTON NEHALEM SANDYBRIDGE ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
 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 SANDYBRIDGE OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
 endif
 
 ifndef DYNAMIC_CORE
@@ -754,6 +770,7 @@ 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

README

@@ -1,84 +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)
-Please read the documents on OpenBLAS wiki pages(http://github.com/xianyi/OpenBLAS/wiki).
-
-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 //Level 3 BLAS subroutines are optimized.
-
-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.
-* 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. I don't think this is a bug in OpenBLAS. 
-
-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 +1,110 @@
+# 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.
+
+#### 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. So far, we didn't support changing the number of threads on Windows. On Windows, these functions are dummy.
+
+    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 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 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/);

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};

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
 

common_interface.h

@@ -45,6 +45,8 @@ extern "C" {
 
 int    BLASFUNC(xerbla)(char *, blasint *info, blasint);
 
+void    BLASFUNC(openblas_set_num_threads)(int *);
+
 FLOATRET  BLASFUNC(sdot)  (blasint *, float  *, blasint *, float  *, blasint *);
 FLOATRET  BLASFUNC(sdsdot)(blasint *, float  *,        float  *, blasint *, float  *, blasint *);
 

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 +103,8 @@
 #define CORE_NEHALEM	17
 #define CORE_ATOM	18
 #define CORE_NANO	19
+#define CORE_SANDYBRIDGE 20
+#define CORE_BOBCAT     21
 
 #define HAVE_SSE      (1 <<  0)
 #define HAVE_SSE2     (1 <<  1)
@@ -122,6 +124,7 @@
 #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 +191,6 @@ typedef struct {
 #define CPUTYPE_NSGEODE			41
 #define CPUTYPE_VIAC3			42
 #define CPUTYPE_NANO			43
+#define CPUTYPE_SANDYBRIDGE             44
+#define CPUTYPE_BOBCAT                  45
 #endif

cpuid_x86.c

@@ -189,6 +189,7 @@ 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;
+    if ((ecx & (1 << 28)) != 0) feature |= HAVE_AVX;
 
     if (have_excpuid() >= 0x01) {
       cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
@@ -983,13 +984,13 @@ int get_cpuname(void){
 			  return CPUTYPE_NEHALEM;
 		  case 10:
                           //Intel Core i5-2000 /i7-2000 (Sandy Bridge)
-                          return CPUTYPE_NEHALEM;
+                          return CPUTYPE_SANDYBRIDGE;
 		  case 12:
 			  //Xeon Processor 5600 (Westmere-EP)
 			  return CPUTYPE_NEHALEM;
 		  case 13:
                           //Intel Core i7-3000 / Xeon E5 (Sandy Bridge)
-                          return CPUTYPE_NEHALEM;
+                          return CPUTYPE_SANDYBRIDGE;
 		  case 15:
 			  //Xeon Processor E7 (Westmere-EX)
 			  return CPUTYPE_NEHALEM;
@@ -1027,6 +1028,8 @@ int get_cpuname(void){
       case  1:
       case 10:
 	return CPUTYPE_BARCELONA;
+      case  5:
+	return CPUTYPE_BOBCAT;
       }
       break;
     }
@@ -1146,6 +1149,8 @@ static char *cpuname[] = {
   "NSGEODE",
   "VIAC3",
   "NANO",
+  "SANDYBRIDGE",
+  "BOBCAT",
 };
 
 static char *lowercpuname[] = {
@@ -1192,6 +1197,8 @@ static char *lowercpuname[] = {
   "tms3x00",
   "nsgeode",
   "nano",
+  "sandybridge",
+  "bobcat",
 };
 
 static char *corename[] = {
@@ -1215,6 +1222,8 @@ static char *corename[] = {
   "NEHALEM",
   "ATOM",
   "NANO",
+  "SANDYBRIDGE",
+  "BOBCAT",
 };
 
 static char *corename_lower[] = {
@@ -1238,6 +1247,8 @@ static char *corename_lower[] = {
   "nehalem",
   "atom",
   "nano",
+  "sandybridge",
+  "bobcat",
 };
 
 
@@ -1321,13 +1332,13 @@ int get_coretype(void){
 	  return CORE_NEHALEM;
 	case 10:
           //Intel Core i5-2000 /i7-2000 (Sandy Bridge)
-          return CORE_NEHALEM;
+          return CORE_SANDYBRIDGE;
 	case 12:
 	  //Xeon Processor 5600 (Westmere-EP)
 	  return CORE_NEHALEM;
 	case 13:
           //Intel Core i7-3000 / Xeon E5 (Sandy Bridge)
-          return CORE_NEHALEM;
+          return CORE_SANDYBRIDGE;
 	case 15:
 	  //Xeon Processor E7 (Westmere-EX)
 	  return CORE_NEHALEM;
@@ -1346,7 +1357,9 @@ 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 return CORE_BARCELONA;
     }
   }
 
@@ -1426,6 +1439,7 @@ 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");
@@ -1491,6 +1505,7 @@ 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

@@ -35,19 +35,19 @@ OS_LINUX
 #endif
 
 #if defined(__FreeBSD__)
-OS_FreeBSD
+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)
@@ -63,7 +63,7 @@ OS_WINNT
 #endif
 
 #if defined(__CYGWIN__)
-OS_CYGWIN
+OS_CYGWIN_NT
 #endif
 
 #if defined(__INTERIX)

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

driver/others/blas_server_win32.c

@@ -63,6 +63,14 @@ static blas_pool_t   pool;
 static HANDLE	    blas_threads   [MAX_CPU_NUMBER];
 static DWORD	    blas_threads_id[MAX_CPU_NUMBER];
 
+void goto_set_num_threads(int num)
+{
+}
+
+void openblas_set_num_threads(int num)
+{
+}
+
 static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
 
       if (!(mode & BLAS_COMPLEX)){

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;
-      }
-      p++;
-    }
-    p = name;
 
-    //    while ((*p == '0') || (*p == ',')) 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;
+	}
+      }
 
-    affinity = strtoul(p, &dummy, 16);
-   
+    }
+    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];
+    }
     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));
-   
-    p = name;
+    read(infile, cpumap, sizeof(cpumap));
 
-    while ((*p == '0') || (*p == ',')) p++;
+    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;
+	}
+      }
 
-    affinity = strtol(p, &p, 16);
+    }
+    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];
+    }
+   
    
     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 +412,89 @@ 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++){
+      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 +584,7 @@ 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 +594,9 @@ 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 +682,7 @@ void gotoblas_affinity_init(void) {
 #ifndef USE_OPENMP
   cpu_set_t cpu_mask;
 #endif
+  int i;
 
   if (initialized) return;
 
@@ -646,6 +734,11 @@ 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 +747,8 @@ 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 +758,8 @@ 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

@@ -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) || defined(OS_DARWIN)
 
 int get_num_procs(void) {
   
@@ -215,7 +215,7 @@ 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 +223,7 @@ 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 +250,7 @@ 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
 

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
@@ -33,7 +33,6 @@ 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) ;
 
@@ -41,5 +40,13 @@ void NAME(int* num_threads){
 	openblas_set_num_threads(*num_threads);
 }
 
-#endif
+#else
+//Single thread
+
+void openblas_set_num_threads(int num_threads) {
+}
+
+void NAME(int* num_threads){
+
+}
 #endif

driver/others/parameter.c

@@ -163,9 +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);
 
@@ -384,6 +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;
 
@@ -435,7 +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;

exports/Makefile

@@ -70,11 +70,11 @@ dll2 : libgoto2_shared.dll
 	$(RANLIB) ../$(LIBNAME)
 ifeq ($(BINARY32), 1)
 	$(DLLWRAP) -o ../$(LIBDLLNAME) --def libopenblas.def \
-	--entry _dllinit@12 -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(FEXTRALIB)
+	--entry _dllinit@12 -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(EXTRALIB)
 	-lib /machine:i386 /def:libopenblas.def
 else
 	$(DLLWRAP) -o ../$(LIBDLLNAME) --def libopenblas.def \
-	--entry $(FU)dllinit    -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(FEXTRALIB)
+	--entry $(FU)dllinit    -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(EXTRALIB)
 	-lib /machine:X64 /def:libopenblas.def 
 endif
 
@@ -121,7 +121,7 @@ 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
 

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;
             }
         }
     }

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 +102,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* #define FORCE_BARCELONA	*/
 /* #define FORCE_SHANGHAI	*/
 /* #define FORCE_ISTANBUL	*/
+/* #define FORCE_BOBCAT	*/
 /* #define FORCE_SSE_GENERIC	*/
 /* #define FORCE_VIAC3		*/
 /* #define FORCE_NANO		*/
@@ -278,6 +279,20 @@ 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
@@ -349,6 +364,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #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

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/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.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/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/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

kernel/x86/trsm_kernel_RT_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
@@ -1036,7 +1036,7 @@
 
 .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
@@ -1066,7 +1066,7 @@
 	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
@@ -2224,7 +2224,7 @@
 .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
@@ -2273,7 +2273,7 @@
 	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

kernel/x86/trsm_kernel_RT_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_RT_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)
@@ -439,7 +439,7 @@
 
 .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
@@ -454,7 +454,7 @@
 	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
@@ -758,7 +758,7 @@
 
 .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
@@ -993,7 +993,7 @@
 
 .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
@@ -1324,7 +1324,7 @@
 
 .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
@@ -1354,7 +1354,7 @@
 	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
@@ -1718,7 +1718,7 @@
 	ALIGN_4
 
 .L62:
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
 #endif
 
@@ -2031,7 +2031,7 @@
 
 .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
@@ -2859,7 +2859,7 @@
 .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
@@ -3303,7 +3303,7 @@
 .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

kernel/x86/zgemm_kernel_1x2_penryn.S

@@ -64,7 +64,7 @@
 #define PREFETCHB     prefetcht0
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCHSIZE  (8 * 1 - 4)
 #define PREFETCHW     prefetcht0
 #define PREFETCHB     prefetcht0

kernel/x86/zgemm_kernel_2x2_penryn.S

@@ -64,7 +64,7 @@
 #define PREFETCHB     prefetcht0
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCHSIZE  (16 * 1 + 8)
 #define PREFETCHW     prefetcht0
 #define PREFETCHB     prefetcht0

kernel/x86/zgemv_n_sse.S

@@ -58,7 +58,7 @@
 #define PREFETCHSIZE	(16 * 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	(16 * 7)

kernel/x86/zgemv_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/zgemv_t_sse.S

@@ -58,7 +58,7 @@
 #define PREFETCHSIZE	(16 * 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	(16 * 7)

kernel/x86/zgemv_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/zscal_sse.S

@@ -55,7 +55,7 @@
 #define XX	%edi
 #define FLAG	%ebp
 
-#if defined(NEHALEM) || defined(PENRYN) || defined(DUNNINGTON)
+#if defined(NEHALEM) || defined(PENRYN) || defined(DUNNINGTON) || defined(SANDYBRIDGE)
 #define  USE_PSHUFD
 #else
 #define USE_PSHUFD_HALF
@@ -697,7 +697,7 @@
 	cmpl	$2 * SIZE, INCX
 	jne	.L120
 
-#if defined(ALIGNED_ACCESS) && !defined(NEHALEM)
+#if defined(ALIGNED_ACCESS) && !defined(NEHALEM) &&  !defined(SANDYBRIDGE)
 
 	PSHUFD2($0, %xmm0, %xmm6)
 	PSHUFD2($0, %xmm1, %xmm1)

kernel/x86/zscal_sse2.S

@@ -57,7 +57,7 @@
 
 #include "l1param.h"
 
-#if defined(NEHALEM) || defined(PENRYN) || defined(DUNNINGTON)
+#if defined(NEHALEM) || defined(PENRYN) || defined(DUNNINGTON) || defined(SANDYBRIDGE)
 #define  USE_PSHUFD
 #else
 #define USE_PSHUFD_HALF
@@ -860,7 +860,7 @@
 	cmpl	$2 * SIZE, INCX
 	jne	.L220
 
-#if defined(ALIGNED_ACCESS) && !defined(NEHALEM)
+#if defined(ALIGNED_ACCESS) && !defined(NEHALEM) && !defined(SANDYBRIDGE)
 
 #ifdef HAVE_SSE3
 	movddup	%xmm0, %xmm6

kernel/x86/ztrsm_kernel_LN_2x2_penryn.S

@@ -61,7 +61,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_LN_2x2_sse.S

@@ -75,7 +75,7 @@
 #define STACK_ALIGN	4096
 #define STACK_OFFSET	1024
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 #define PREFETCHSIZE (16 * 10 + 8)
 #define WPREFETCHSIZE 112
 #define PREFETCH      prefetch
@@ -533,7 +533,7 @@
 	addps	%xmm0, %xmm7
 	movsd	16 * SIZE(AA), %xmm0
 	mulps	%xmm1, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht1     (PREFETCHSIZE + 16) * SIZE(AA)
 #endif
 	addps	%xmm2, %xmm4

kernel/x86/ztrsm_kernel_LT_1x2_penryn.S

@@ -63,7 +63,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_LT_2x2_penryn.S

@@ -61,7 +61,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_LT_2x2_sse.S

@@ -75,7 +75,7 @@
 #define STACK_ALIGN	4096
 #define STACK_OFFSET	1024
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 #define PREFETCHSIZE (16 * 10 + 8)
 #define WPREFETCHSIZE 112
 #define PREFETCH      prefetch
@@ -994,7 +994,7 @@
 	addps	%xmm0, %xmm7
 	movsd	16 * SIZE(AA), %xmm0
 	mulps	%xmm1, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht1     (PREFETCHSIZE + 16) * SIZE(AA)
 #endif
 	addps	%xmm2, %xmm4

kernel/x86/ztrsm_kernel_RT_1x2_penryn.S

@@ -63,7 +63,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_RT_2x2_penryn.S

@@ -61,7 +61,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_RT_2x2_sse.S

@@ -75,7 +75,7 @@
 #define STACK_ALIGN	4096
 #define STACK_OFFSET	1024
 
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 #define PREFETCHSIZE (16 * 10 + 8)
 #define WPREFETCHSIZE 112
 #define PREFETCH      prefetch
@@ -1820,7 +1820,7 @@
 	addps	%xmm0, %xmm7
 	movsd	16 * SIZE(AA), %xmm0
 	mulps	%xmm1, %xmm2
-#if defined(OPTERON) || defined(BARCELONA)
+#if defined(OPTERON) || defined(BARCELONA) || defined(BOBCAT)
 	prefetcht1     (PREFETCHSIZE + 16) * SIZE(AA)
 #endif
 	addps	%xmm2, %xmm4

kernel/x86_64/KERNEL.BOBCAT

@@ -0,0 +1,62 @@
+ZGEMVNKERNEL = zgemv_n_dup.S
+ZGEMVTKERNEL = zgemv_t_dup.S
+
+SGEMMKERNEL    =  gemm_kernel_8x4_barcelona.S
+SGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
+SGEMMITCOPY    =  ../generic/gemm_tcopy_8.c
+SGEMMONCOPY    =  gemm_ncopy_4_opteron.S
+SGEMMOTCOPY    =  gemm_tcopy_4_opteron.S
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX) 
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX) 
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+DGEMMKERNEL    =  gemm_kernel_4x4_barcelona.S
+DGEMMINCOPY    =
+DGEMMITCOPY    =
+DGEMMONCOPY    =  gemm_ncopy_4_opteron.S
+DGEMMOTCOPY    =  gemm_tcopy_4_opteron.S
+DGEMMINCOPYOBJ =
+DGEMMITCOPYOBJ =
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+CGEMMKERNEL    =  zgemm_kernel_4x2_barcelona.S
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_4.c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_4.c
+CGEMMONCOPY    =  zgemm_ncopy_2.S
+CGEMMOTCOPY    =  zgemm_tcopy_2.S
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMKERNEL    =  zgemm_kernel_2x2_barcelona.S
+ZGEMMINCOPY    =
+ZGEMMITCOPY    =
+ZGEMMONCOPY    =  zgemm_ncopy_2.S
+ZGEMMOTCOPY    =  zgemm_tcopy_2.S
+ZGEMMINCOPYOBJ =
+ZGEMMITCOPYOBJ =
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+STRSMKERNEL_LN	=  trsm_kernel_LN_8x4_sse.S
+STRSMKERNEL_LT	=  trsm_kernel_LT_8x4_sse.S
+STRSMKERNEL_RN	=  trsm_kernel_LT_8x4_sse.S
+STRSMKERNEL_RT	=  trsm_kernel_RT_8x4_sse.S
+
+DTRSMKERNEL_LN	=  trsm_kernel_LN_4x4_barcelona.S
+DTRSMKERNEL_LT	=  trsm_kernel_LT_4x4_barcelona.S
+DTRSMKERNEL_RN	=  trsm_kernel_LT_4x4_barcelona.S
+DTRSMKERNEL_RT	=  trsm_kernel_RT_4x4_barcelona.S
+
+CTRSMKERNEL_LN	=  ztrsm_kernel_LN_4x2_sse.S
+CTRSMKERNEL_LT	=  ztrsm_kernel_LT_4x2_sse.S
+CTRSMKERNEL_RN	=  ztrsm_kernel_LT_4x2_sse.S
+CTRSMKERNEL_RT	=  ztrsm_kernel_RT_4x2_sse.S
+
+ZTRSMKERNEL_LN	=  ztrsm_kernel_LN_2x2_sse2.S
+ZTRSMKERNEL_LT	=  ztrsm_kernel_LT_2x2_sse2.S
+ZTRSMKERNEL_RN	=  ztrsm_kernel_LT_2x2_sse2.S
+ZTRSMKERNEL_RT	=  ztrsm_kernel_RT_2x2_sse2.S
+
+CGEMM3MKERNEL    =  zgemm3m_kernel_8x4_barcelona.S
+ZGEMM3MKERNEL    =  zgemm3m_kernel_4x4_barcelona.S

kernel/x86_64/KERNEL.SANDYBRIDGE

@@ -0,0 +1,84 @@
+SGEMMKERNEL    =  sgemm_kernel_8x8_sandy.S
+SGEMMINCOPY    =  
+SGEMMITCOPY    =  
+SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_8.c
+SGEMMINCOPYOBJ =   
+SGEMMITCOPYOBJ =   
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+DGEMMKERNEL    =  dgemm_kernel_4x8_sandy.S
+DGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_8.c
+#DGEMMONCOPY    =  gemm_ncopy_4.S
+DGEMMONCOPY    =  ../generic/gemm_ncopy_4.c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+#DGEMMOTCOPY    =  gemm_tcopy_4.S
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+#CGEMMKERNEL    =  zgemm_kernel_2x4_nehalem.S
+CGEMMKERNEL    =  cgemm_kernel_4x8_sandy.S
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_8_sandy.c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_8_sandy.c
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_4_sandy.c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_4_sandy.c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+#ZGEMMKERNEL    =  zgemm_kernel_1x4_nehalem.S
+ZGEMMKERNEL    =  zgemm_kernel_4x4_sandy.S
+ZGEMMINCOPY    =  
+ZGEMMITCOPY    = 
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_4.c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_4.c
+ZGEMMINCOPYOBJ =  
+ZGEMMITCOPYOBJ =  
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+#STRSMKERNEL_LN	=  trsm_kernel_LN_4x8_nehalem.S
+#STRSMKERNEL_LT	=  trsm_kernel_LT_4x8_nehalem.S
+#STRSMKERNEL_RN	=  trsm_kernel_LT_4x8_nehalem.S
+#STRSMKERNEL_RT	=  trsm_kernel_RT_4x8_nehalem.S
+
+#DTRSMKERNEL_LN	=  trsm_kernel_LN_2x8_nehalem.S
+#DTRSMKERNEL_LT	=  trsm_kernel_LT_2x8_nehalem.S
+#DTRSMKERNEL_RN	=  trsm_kernel_LT_2x8_nehalem.S
+#DTRSMKERNEL_RT	=  trsm_kernel_RT_2x8_nehalem.S
+
+#CTRSMKERNEL_LN	=  ztrsm_kernel_LN_2x4_nehalem.S
+#CTRSMKERNEL_LT	=  ztrsm_kernel_LT_2x4_nehalem.S
+#CTRSMKERNEL_RN	=  ztrsm_kernel_LT_2x4_nehalem.S
+#CTRSMKERNEL_RT	=  ztrsm_kernel_RT_2x4_nehalem.S
+
+#ZTRSMKERNEL_LN	=  ztrsm_kernel_LT_1x4_nehalem.S
+#ZTRSMKERNEL_LT	=  ztrsm_kernel_LT_1x4_nehalem.S
+#ZTRSMKERNEL_RN	=  ztrsm_kernel_LT_1x4_nehalem.S
+#ZTRSMKERNEL_RT	=  ztrsm_kernel_RT_1x4_nehalem.S
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../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
+
+
+
+CGEMM3MKERNEL    =  zgemm3m_kernel_4x8_nehalem.S
+ZGEMM3MKERNEL    =  zgemm3m_kernel_2x8_nehalem.S

kernel/x86_64/dgemm_ncopy_2.S

@@ -45,6 +45,12 @@
 #define PREFETCHW	prefetcht0
 #endif
 
+#ifdef SANDYBRIDGE
+#define PREFETCHSIZE	16
+#define PREFETCH	prefetcht0
+#define PREFETCHW	prefetcht0
+#endif
+
 #ifndef MOVAPS
 #define MOVAPS	movaps
 #endif

kernel/x86_64/dgemm_ncopy_4.S

@@ -45,7 +45,7 @@
 #define PREFETCHW     prefetcht0
 #endif
 
-#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM)
+#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCHSIZE	16
 #define PREFETCH      prefetcht0
 #define PREFETCHW     prefetcht0

kernel/x86_64/dgemm_ncopy_8.S

@@ -45,6 +45,12 @@
 #define PREFETCHW	prefetcht0
 #endif
 
+#ifdef SANDYBRIDGE
+#define PREFETCHSIZE	12
+#define PREFETCH	prefetcht0
+#define PREFETCHW	prefetcht0
+#endif
+
 #ifndef MOVAPS
 #define MOVAPS	movaps
 #endif

kernel/x86_64/dgemm_tcopy_2.S

@@ -52,6 +52,13 @@
 #define MOVUPS_A	movups
 #endif
 
+#ifdef SANDYBRIDGE
+#define PREFETCHSIZE	12
+#define PREFETCH	prefetcht0
+#define PREFETCHW	prefetcht0
+#define MOVUPS_A	movups
+#endif
+
 #if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON)
 #define PREFETCHSIZE	16
 #define PREFETCH      prefetcht0

kernel/x86_64/dgemm_tcopy_4.S

@@ -51,6 +51,12 @@
 #define MOVUPS_A	movups
 #endif
 
+#ifdef SANDYBRIDGE
+#define PREFETCHSIZE	12
+#define PREFETCH      prefetcht0
+#define MOVUPS_A	movups
+#endif
+
 #if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON)
 #define PREFETCHSIZE	16
 #define PREFETCH      prefetcht0

kernel/x86_64/dgemm_tcopy_8.S

@@ -46,6 +46,13 @@
 #define MOVUPS_A	movups
 #endif
 
+#ifdef SANDYBRIDGE
+#define PREFETCHSIZE	16
+#define PREFETCH	prefetcht0
+#define PREFETCHW	prefetcht0
+#define MOVUPS_A	movups
+#endif
+
 #ifdef MOVUPS_A
 #define MOVUPS_A1(OFF, ADDR, REGS)		MOVUPS_A	OFF(ADDR), REGS
 #define MOVUPS_A2(OFF, ADDR, BASE, SCALE, REGS)	MOVUPS_A	OFF(ADDR, BASE, SCALE), REGS

kernel/x86_64/gemm_ncopy_2.S

@@ -46,6 +46,13 @@
 #define PREFETCHW     prefetcht0
 #endif
 
+#if defined(SANDYBRIDGE)
+#define RPREFETCHSIZE	12
+#define WPREFETCHSIZE (RPREFETCHSIZE * 2)
+#define PREFETCH      prefetcht0
+#define PREFETCHW     prefetcht0
+#endif
+
 #ifndef WINDOWS_ABI
 
 #define M	ARG1	/* rdi */

kernel/x86_64/gemm_ncopy_4.S

@@ -46,7 +46,7 @@
 #define PREFETCHW     prefetcht0
 #endif
 
-#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM)
+#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) || defined(SANDYBRIDGE)
 #define RPREFETCHSIZE	12
 #define WPREFETCHSIZE (RPREFETCHSIZE * 4)
 #define PREFETCH      prefetcht0

kernel/x86_64/gemm_tcopy_2.S

@@ -46,6 +46,13 @@
 #define PREFETCHW     prefetcht0
 #endif
 
+#if defined(SANDYBRIDGE)
+#define RPREFETCHSIZE	12
+#define WPREFETCHSIZE (RPREFETCHSIZE * 2)
+#define PREFETCH      prefetcht0
+#define PREFETCHW     prefetcht0
+#endif
+
 #ifndef WINDOWS_ABI
 
 #define M	ARG1	/* rdi */

kernel/x86_64/gemm_tcopy_4.S

@@ -46,7 +46,7 @@
 #define PREFETCHW     prefetcht0
 #endif
 
-#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM)
+#if defined(CORE2) || defined(PENRYN) || defined(DUNNINGTON) || defined(NEHALEM) || defined(SANDYBRIDGE)
 #define RPREFETCHSIZE	12
 #define WPREFETCHSIZE (RPREFETCHSIZE * 4)
 #define PREFETCH      prefetcht0

kernel/x86_64/symv_L_sse.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 12)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 12)
@@ -76,7 +76,7 @@
 #define movsd		movlps
 #endif
 
-#if defined(BARCELONA) || defined(SHANGHAI)
+#if defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH	prefetch
 #define PREFETCHW	prefetchw
 #define PREFETCHSIZE	(16 * 16)

kernel/x86_64/symv_L_sse2.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 12)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 12)
@@ -76,7 +76,7 @@
 #define movsd		movlpd
 #endif
 
-#if defined(BARCELONA) || defined(SHANGHAI)
+#if defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH	prefetch
 #define PREFETCHW	prefetchw
 #define PREFETCHSIZE	(16 * 16)

kernel/x86_64/symv_U_sse.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 12)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 12)
@@ -76,7 +76,7 @@
 #define movsd		movlps
 #endif
 
-#if defined(BARCELONA) || defined(SHANGHAI)
+#if defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH	prefetch
 #define PREFETCHW	prefetchw
 #define PREFETCHSIZE	(16 * 16)

kernel/x86_64/symv_U_sse2.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 12)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)
@@ -76,7 +76,7 @@
 #define movsd		movlpd
 #endif
 
-#if defined(BARCELONA) || defined(SHANGHAI)
+#if defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH	prefetch
 #define PREFETCHW	prefetchw
 #define PREFETCHSIZE	(16 * 16)

kernel/x86_64/zscal_sse.S

@@ -685,7 +685,7 @@
 	cmpq	$2 * SIZE, INCX
 	jne	.L120
 
-#if defined(ALIGNED_ACCESS) && !defined(NEHALEM)
+#if defined(ALIGNED_ACCESS) && !defined(NEHALEM) && !defined(SANDYBRIDGE)
 
 	pshufd	$0, %xmm0, %xmm14
 	pshufd	$0, %xmm1, %xmm1

kernel/x86_64/zscal_sse2.S

@@ -55,7 +55,7 @@
 
 #include "l1param.h"
 
-#if defined(NEHALEM) || defined(PENRYN) || defined(DUNNINGTON) || defined(BARCELONA) || defined(NANO)
+#if defined(NEHALEM) || defined(PENRYN) || defined(DUNNINGTON) || defined(BARCELONA) || defined(NANO)  || defined(SANDYBRIDGE)
 #define  USE_PSHUFD
 #else
 #define  USE_PSHUFD_HALF
@@ -803,7 +803,7 @@
 	cmpq	$2 * SIZE, INCX
 	jne	.L220
 
-#if defined(ALIGNED_ACCESS) && !defined(NEHALEM)
+#if defined(ALIGNED_ACCESS) && !defined(NEHALEM) && !defined(SANDYBRIDGE)
 	
 	movddup	%xmm0,  %xmm14
 	pxor	%xmm15, %xmm15

kernel/x86_64/zsymv_L_sse.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 24)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)
@@ -160,7 +160,7 @@
 #define a3     %xmm14
 #define	xt1    %xmm15
 
-#if (defined(HAVE_SSE3) && !defined(CORE_OPTERON)) || defined(BARCELONA) || defined(SHANGHAI)
+#if (defined(HAVE_SSE3) && !defined(CORE_OPTERON)) || defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define MOVDDUP(a, b, c)	movddup	a(b), c
 #define MOVDDUP2(a, b, c)	movddup	a##b, c
 #else

kernel/x86_64/zsymv_L_sse2.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 24)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)
@@ -76,7 +76,7 @@
 #define movsd		movlpd
 #endif
 
-#if defined(BARCELONA)  || defined(SHANGHAI)
+#if defined(BARCELONA)  || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH	prefetch
 #define PREFETCHW	prefetchw
 #define PREFETCHSIZE	(16 * 16)

kernel/x86_64/zsymv_U_sse.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 24)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)
@@ -76,7 +76,7 @@
 #define movsd		movlpd
 #endif
 
-#if defined(BARCELONA)  || defined(SHANGHAI)
+#if defined(BARCELONA)  || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH	prefetch
 #define PREFETCHW	prefetchw
 #define PREFETCHSIZE	(16 * 16)

kernel/x86_64/zsymv_U_sse2.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 24)
 #endif
 
-#ifdef NEHALEM
+#if defined(NEHALEM) || defined(SANDYBRIDGE)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)
@@ -76,7 +76,7 @@
 #define movsd		movlpd
 #endif
 
-#if defined(BARCELONA)  || defined(SHANGHAI)
+#if defined(BARCELONA)  || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH	prefetch
 #define PREFETCHW	prefetchw
 #define PREFETCHSIZE	(16 * 16)

kernel/x86_64/ztrsm_kernel_LN_2x2_sse2.S

@@ -86,7 +86,7 @@
 #define BORIG	 72(%rsp)
 #define BUFFER	128(%rsp)
 
-#if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI)
+#if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH     prefetch
 #define PREFETCHW    prefetchw
 #define PREFETCHNTA  prefetchnta

kernel/x86_64/ztrsm_kernel_LN_4x2_sse.S

@@ -95,7 +95,7 @@
 #define PREFETCHSIZE (8 * 6 + 4)
 #endif
 
-#if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI)
+#if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH     prefetch
 #define PREFETCHW    prefetchw
 #define PREFETCHNTA  prefetchnta

kernel/x86_64/ztrsm_kernel_LT_2x2_sse2.S

@@ -86,7 +86,7 @@
 #define BORIG	 72(%rsp)
 #define BUFFER	128(%rsp)
 
-#if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI)
+#if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH     prefetch
 #define PREFETCHW    prefetchw
 #define PREFETCHNTA  prefetchnta

kernel/x86_64/ztrsm_kernel_LT_4x2_sse.S

@@ -95,7 +95,7 @@
 #define PREFETCHSIZE (8 * 6 + 4)
 #endif
 
-#if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI)
+#if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI) || defined(BOBCAT)
 #define PREFETCH     prefetch
 #define PREFETCHW    prefetchw
 #define PREFETCHNTA  prefetchnta