Merge branch 'release-v0.1alpha2'v0.1alpha2

41 files changed, 12958 insertions, 50 deletions

diff --git a/.gitignore b/.gitignore
index 44af57166..6cfc5b3c1 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,8 +1,13 @@
+*.obj
+*.lib
+*.dll
+*.def
 *.o
 lapack-3.1.1
 lapack-3.1.1.tgz
 *.so
 *.a
+.svn
 *~
 config.h
 Makefile.conf
diff --git a/Changelog.txt b/Changelog.txt
index 4f83fdf97..b54949ec5 100644
--- a/Changelog.txt
+++ b/Changelog.txt
@@ -1,6 +1,7 @@
 OpenBLAS ChangeLog
 ====================================================================
-Version 0.1 alpha2(in development)
+Version 0.1 alpha2
+23-Jun-2011
 
 common:
 	* Fixed blasint undefined bug in <cblas.h> file. Other software 
@@ -15,11 +16,25 @@ common:
 	* Provided an error message when the arch is not supported.(Refs 
 	  issue #19 on github)
 	* Fixed issue #23. Fixed a bug of f_check script about generating link flags.
+	* Added openblas_set_num_threads for Fortran.
+	* Fixed #25 a wrong result of rotmg.
+	* Fixed a bug about detecting underscore prefix in c_check.
+	* Print the wall time (cycles) with enabling FUNCTION_PROFILE
+	* Fixed #35 a build bug with NO_LAPACK=1 & DYNAMIC_ARCH=1
+	* Added install target. You can use "make install". (Refs #20)
+
 
 x86/x86_64:
-	* 
+	* Fixed #28 a wrong result of dsdot on x86_64.
+	* Fixed #32 a SEGFAULT bug of zdotc with gcc-4.6.
+	* Fixed #33 ztrmm bug on Nehalem.
+	* Walk round #27 the low performance axpy issue with small imput size & multithreads.
+
 MIPS64:
-	* 
+	* Fixed #28 a wrong result of dsdot on Loongson3A/MIPS64. 
+	* Optimized single/double precision BLAS Level3 on Loongson3A/MIPS64. (Refs #2)
+	* Optimized single/double precision axpy function on Loongson3A/MIPS64. (Refs #3)
+
 ====================================================================
 Version 0.1 alpha1
 20-Mar-2011
diff --git a/Makefile b/Makefile
index 77dd3c2e7..798c56192 100644
--- a/Makefile
+++ b/Makefile
@@ -15,6 +15,10 @@ ifdef SANITY_CHECK
 BLASDIRS += reference
 endif
 
+ifndef PREFIX
+PREFIX = /opt/OpenBLAS
+endif
+
 SUBDIRS	= $(BLASDIRS)
 ifneq ($(NO_LAPACK), 1)
 SUBDIRS	+= lapack
@@ -22,8 +26,8 @@ endif
 
 SUBDIRS_ALL = $(SUBDIRS) test ctest utest exports benchmark ../laswp ../bench
 
-.PHONY : all libs netlib test ctest shared
-.NOTPARALLEL : all libs prof lapack-test
+.PHONY : all libs netlib test ctest shared install 
+.NOTPARALLEL : all libs prof lapack-test install
 
 all :: libs netlib tests shared
 	@echo
@@ -70,7 +74,7 @@ ifeq ($(OSNAME), Darwin)
 endif
 ifeq ($(OSNAME), WINNT)
 	$(MAKE) -C exports dll
-#	-ln -fs $(LIBDLLNAME) libopenblas.dll
+	-ln -fs $(LIBDLLNAME) libopenblas.dll
 endif
 ifeq ($(OSNAME), CYGWIN_NT)
 	$(MAKE) -C exports dll
@@ -105,12 +109,17 @@ endif
 	  $(MAKE) -C $$d $(@F) || exit 1 ; \
 	fi; \
 	done
+#Save the config files for installation
+	cp Makefile.conf Makefile.conf_last
+	cp config.h config_last.h
 ifdef DYNAMIC_ARCH
 	  $(MAKE) -C kernel commonlibs || exit 1
 	for d in $(DYNAMIC_CORE) ; \
 	do  $(MAKE) GOTOBLAS_MAKEFILE= -C kernel TARGET_CORE=$$d kernel || exit 1 ;\
 	done
+	echo DYNAMIC_ARCH=1 >> Makefile.conf_last
 endif
+	touch lib.grd
 
 prof : prof_blas prof_lapack
 
@@ -230,19 +239,23 @@ lapack-test :
 
 dummy :
 
+install :
+	$(MAKE) -f Makefile.install install
+
 clean ::
 	@for d in $(SUBDIRS_ALL) ; \
 	do if test -d $$d; then \
 	  $(MAKE) -C $$d $(@F) || exit 1 ; \
 	fi; \
 	done
-ifdef DYNAMIC_ARCH
+#ifdef DYNAMIC_ARCH
 	@$(MAKE) -C kernel clean
-endif
+#endif
 	@rm -f *.$(LIBSUFFIX) *.so *~ *.exe getarch getarch_2nd *.dll *.lib *.$(SUFFIX) *.dwf libopenblas.$(LIBSUFFIX) libopenblas_p.$(LIBSUFFIX) *.lnk myconfig.h
 	@rm -f Makefile.conf config.h Makefile_kernel.conf config_kernel.h st* *.dylib
 	@if test -d lapack-3.1.1; then \
 	echo deleting lapack-3.1.1; \
 	rm -rf lapack-3.1.1 ;\
 	fi
+	@rm -f *.grd Makefile.conf_last config_last.h
 	@echo Done.
\ No newline at end of file
diff --git a/Makefile.install b/Makefile.install
new file mode 100644
index 000000000..80dafc9c6
--- /dev/null
+++ b/Makefile.install
@@ -0,0 +1,65 @@
+TOPDIR	= .
+export GOTOBLAS_MAKEFILE = 1
+-include $(TOPDIR)/Makefile.conf_last
+include ./Makefile.system
+
+.PHONY : install
+.NOTPARALLEL : install
+
+lib.grd :
+	$(error OpenBLAS: Please run "make" firstly)
+
+install : 	lib.grd
+	@-mkdir -p $(PREFIX)
+	@echo Generating openblas_config.h in $(PREFIX)
+#for inc 
+	@echo \#ifndef OPENBLAS_CONFIG_H > $(PREFIX)/openblas_config.h
+	@echo \#define OPENBLAS_CONFIG_H >> $(PREFIX)/openblas_config.h
+	@cat config_last.h >> $(PREFIX)/openblas_config.h
+	@echo \#define VERSION \" OpenBLAS $(VERSION) \" >> $(PREFIX)/openblas_config.h
+	@cat openblas_config_template.h >> $(PREFIX)/openblas_config.h
+	@echo \#endif >> $(PREFIX)/openblas_config.h
+
+	@echo Generating f77blas.h in $(PREFIX)
+	@echo \#ifndef OPENBLAS_F77BLAS_H > $(PREFIX)/f77blas.h
+	@echo \#define OPENBLAS_F77BLAS_H >> $(PREFIX)/f77blas.h
+	@echo \#include \"openblas_config.h\" >> $(PREFIX)/f77blas.h
+	@cat common_interface.h >> $(PREFIX)/f77blas.h
+	@echo \#endif >> $(PREFIX)/f77blas.h
+
+	@echo Generating cblas.h in $(PREFIX)
+	@sed 's/common/openblas_config/g' cblas.h > $(PREFIX)/cblas.h
+
+#for install static library 
+	@echo Copy the static library to $(PREFIX)
+	@cp $(LIBNAME) $(PREFIX)
+	@-ln -fs $(PREFIX)/$(LIBNAME) $(PREFIX)/libopenblas.$(LIBSUFFIX)
+#for install shared library 
+	@echo Copy the shared library to $(PREFIX)
+ifeq ($(OSNAME), Linux)
+	-cp $(LIBSONAME) $(PREFIX)
+	-ln -fs $(PREFIX)/$(LIBSONAME) $(PREFIX)/libopenblas.so
+endif
+ifeq ($(OSNAME), FreeBSD)
+	-cp $(LIBSONAME) $(PREFIX)
+	-ln -fs $(PREFIX)/$(LIBSONAME) $(PREFIX)/libopenblas.so
+endif
+ifeq ($(OSNAME), NetBSD)
+	-cp $(LIBSONAME) $(PREFIX)
+	-ln -fs $(PREFIX)/$(LIBSONAME) $(PREFIX)/libopenblas.so
+endif
+ifeq ($(OSNAME), Darwin)
+	-cp $(LIBDYNNAME) $(PREFIX)
+	-ln -fs $(PREFIX)/$(LIBDYNNAME) $(PREFIX)/libopenblas.dylib
+endif
+ifeq ($(OSNAME), WINNT)
+	-cp $(LIBDLLNAME) $(PREFIX)
+	-ln -fs $(PREFIX)/$(LIBDLLNAME) $(PREFIX)/libopenblas.dll
+endif
+ifeq ($(OSNAME), CYGWIN_NT)
+	-cp $(LIBDLLNAME) $(PREFIX)
+	-ln -fs $(PREFIX)/$(LIBDLLNAME) $(PREFIX)/libopenblas.dll
+endif
+
+	@echo Install OK!
+
diff --git a/Makefile.rule b/Makefile.rule
index 61f9eb91d..88d552495 100644
--- a/Makefile.rule
+++ b/Makefile.rule
@@ -91,6 +91,9 @@ VERSION = 0.1alpha2
 # SANITY_CHECK to compare the result with reference BLAS.
 # UTEST_CHECK = 1
 
+# The installation directory.
+# PREFIX = /opt/OpenBLAS
+
 # Common Optimization Flag; -O2 is enough.
 # DEBUG = 1
 
diff --git a/Makefile.system b/Makefile.system
index ca752623c..6cf65c7bd 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -515,6 +515,10 @@ ifeq ($(DYNAMIC_ARCH), 1)
 CCOMMON_OPT	+= -DDYNAMIC_ARCH
 endif
 
+ifeq ($(NO_LAPACK), 1)
+CCOMMON_OPT	+= -DNO_LAPACK
+endif
+
 ifdef SMP
 CCOMMON_OPT	+= -DSMP_SERVER
 
diff --git a/README b/README
index c18b6c502..9a7b16326 100644
--- a/README
+++ b/README
@@ -22,6 +22,11 @@ make BINARY=64 CC=mips64el-unknown-linux-gnu-gcc FC=mips64el-unknown-linux-gnu-g
 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
 
@@ -67,6 +72,7 @@ Please see Changelog.txt to obtain the differences between GotoBLAS2 1.13 BSD ve
 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.
+* This library is not compatible with EKOPath Compiler Suite 4.0.10 (http://www.pathscale.com/ekopath-compiler-suite). However, Path64 (https://github.com/path64/compiler) could compile the codes successfully.
 
 10. Specification of Git Branches
 We used the git branching model in this article (http://nvie.com/posts/a-successful-git-branching-model/). 
@@ -74,4 +80,4 @@ 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
\ No newline at end of file
+  * The gh-pages branch. This is for web pages
diff --git a/c_check b/c_check
index d8025f9f3..263efeb3d 100644
--- a/c_check
+++ b/c_check
@@ -149,7 +149,7 @@ $binformat    = bin64  if ($data =~ /BINARY_64/);
 
 $data = `$compiler_name -S ctest1.c && grep globl ctest1.s | head -n 1 && rm -f ctest1.s`;
 
-$data =~ /globl\ ([_\.]*)(.*)/;
+$data =~ /globl\s([_\.]*)(.*)/;
 
 $need_fu      = $1;
 
diff --git a/common_mips64.h b/common_mips64.h
index 7c7a70ba5..acea79011 100644
--- a/common_mips64.h
+++ b/common_mips64.h
@@ -220,6 +220,11 @@ REALNAME: ;\
 
 #define BUFFER_SIZE     ( 8 << 20)
 
+#if defined(LOONGSON3A)
+#define PAGESIZE	(16UL << 10)
+#define FIXED_PAGESIZE	(16UL << 10)
+#endif
+
 #ifndef PAGESIZE
 #define PAGESIZE	(64UL << 10)
 #endif
diff --git a/common_reference.h b/common_reference.h
index 04b11f80f..4cc4be4fd 100644
--- a/common_reference.h
+++ b/common_reference.h
@@ -60,4 +60,8 @@ float   _Complex BLASFUNC_REF(cdotc)  (blasint *, float  *, blasint *, float  *,
 double  _Complex BLASFUNC_REF(zdotu)  (blasint *, double  *, blasint *, double  *, blasint *);
 double  _Complex BLASFUNC_REF(zdotc)  (blasint *, double  *, blasint *, double  *, blasint *);
 
+void BLASFUNC_REF(drotmg)(double *, double *, double *, double *, double *);
+
+double BLASFUNC_REF(dsdot)(blasint *, float *, blasint *, float *, blasint*);
+												   
 #endif
diff --git a/driver/others/Makefile b/driver/others/Makefile
index bc5de3848..75b552b65 100644
--- a/driver/others/Makefile
+++ b/driver/others/Makefile
@@ -6,7 +6,7 @@ COMMONOBJS	 = memory.$(SUFFIX) xerbla.$(SUFFIX) c_abs.$(SUFFIX) z_abs.$(SUFFIX)
 COMMONOBJS	+= slamch.$(SUFFIX) slamc3.$(SUFFIX) dlamch.$(SUFFIX)  dlamc3.$(SUFFIX)
 
 ifdef SMP
-COMMONOBJS	+= blas_server.$(SUFFIX) divtable.$(SUFFIX) blasL1thread.$(SUFFIX) 
+COMMONOBJS	+= blas_server.$(SUFFIX) divtable.$(SUFFIX) blasL1thread.$(SUFFIX) openblas_set_num_threads.$(SUFFIX)
 ifndef NO_AFFINITY
 COMMONOBJS	+= init.$(SUFFIX) 
 endif
@@ -100,6 +100,9 @@ memory.$(SUFFIX) : $(MEMORY) ../../common.h ../../param.h
 blas_server.$(SUFFIX) : $(BLAS_SERVER) ../../common.h ../../common_thread.h ../../param.h
 	$(CC) $(CFLAGS) -c $< -o $(@F)
 
+openblas_set_num_threads.$(SUFFIX) : openblas_set_num_threads.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
 blasL1thread.$(SUFFIX) : blas_l1_thread.c ../../common.h ../../common_thread.h
 	$(CC) $(CFLAGS) -c $< -o $(@F)
 
diff --git a/driver/others/blas_server_omp.c b/driver/others/blas_server_omp.c
index 17d886e52..4fd4cd440 100644
--- a/driver/others/blas_server_omp.c
+++ b/driver/others/blas_server_omp.c
@@ -38,7 +38,7 @@
 
 #include <stdio.h>
 #include <stdlib.h>
-#include <sys/mman.h>
+//#include <sys/mman.h>
 #include "common.h"
 
 #ifndef USE_OPENMP
diff --git a/driver/others/openblas_set_num_threads.c b/driver/others/openblas_set_num_threads.c
new file mode 100644
index 000000000..7ca3b7114
--- /dev/null
+++ b/driver/others/openblas_set_num_threads.c
@@ -0,0 +1,45 @@
+/*****************************************************************************
+Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+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 "common.h"
+
+#ifdef SMP_SERVER
+#ifdef OS_LINUX
+
+extern  void openblas_set_num_threads(int num_threads) ;
+
+void NAME(int* num_threads){
+	openblas_set_num_threads(*num_threads);
+}
+
+#endif
+#endif
diff --git a/driver/others/profile.c b/driver/others/profile.c
index f65550c9f..f464c0b6a 100644
--- a/driver/others/profile.c
+++ b/driver/others/profile.c
@@ -74,20 +74,21 @@ void gotoblas_profile_quit(void) {
   if (cycles > 0) {
 
     fprintf(stderr, "\n\t====== BLAS Profiling Result =======\n\n");
-    fprintf(stderr, "  Function      No. of Calls   Time Consumption   Efficiency  Bytes/cycle\n");
+    fprintf(stderr, "  Function      No. of Calls   Time Consumption   Efficiency  Bytes/cycle  Wall Time(Cycles)\n");
     
     for (i = 0; i < MAX_PROF_TABLE; i ++) {
       if (function_profile_table[i].calls) {
 #ifndef OS_WINDOWS
-	fprintf(stderr, "%-12s  : %10Ld        %8.2f%%      %10.3f%%  %8.2f\n", 
+	fprintf(stderr, "%-12s  : %10Ld        %8.2f%%      %10.3f%%  %8.2f   %Ld\n", 
 #else
-	fprintf(stderr, "%-12s  : %10lld        %8.2f%%      %10.3f%%  %8.2f\n", 
+	fprintf(stderr, "%-12s  : %10lld        %8.2f%%      %10.3f%%  %8.2f   %lld\n", 
 #endif
 		func_table[i],
 		function_profile_table[i].calls,
 		(double)function_profile_table[i].cycles  / (double)cycles * 100.,
 		(double)function_profile_table[i].fops    / (double)function_profile_table[i].tcycles * 100.,
-		(double)function_profile_table[i].area    / (double)function_profile_table[i].cycles
+			(double)function_profile_table[i].area    / (double)function_profile_table[i].cycles,
+			function_profile_table[i].cycles
 		);
       }
     }
diff --git a/exports/Makefile b/exports/Makefile
index 24cdc41c8..f4c9314f9 100644
--- a/exports/Makefile
+++ b/exports/Makefile
@@ -53,18 +53,19 @@ dyn : $(LIBDYNNAME)
 zip : dll
 	zip $(LIBZIPNAME) $(LIBDLLNAME) $(LIBNAME)
 
-dll  : libgoto2.dll
+dll  : ../$(LIBDLLNAME)
+#libgoto2.dll
 
 dll2 : libgoto2_shared.dll
 
-libgoto2.dll : ../$(LIBNAME) libgoto2.def dllinit.$(SUFFIX)
+../$(LIBDLLNAME) : ../$(LIBNAME) libgoto2.def dllinit.$(SUFFIX)
 	$(RANLIB) ../$(LIBNAME)
 ifeq ($(BINARY32), 1)
-	$(DLLWRAP) -o $(@F) --def libgoto2.def \
+	$(DLLWRAP) -o ../$(LIBDLLNAME) --def libgoto2.def \
 	--entry _dllinit@12 -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(FEXTRALIB)
 	-lib /machine:i386 /def:libgoto2.def
 else
-	$(DLLWRAP) -o $(@F) --def libgoto2.def \
+	$(DLLWRAP) -o ../$(LIBDLLNAME) --def libgoto2.def \
 	--entry _dllinit    -s dllinit.$(SUFFIX) --dllname $(@F) ../$(LIBNAME) $(FEXTRALIB)
 	-lib /machine:X64 /def:libgoto2.def 
 endif
@@ -84,7 +85,7 @@ libgoto_hpl.def : gensymbol
 	perl ./gensymbol win2khpl $(ARCH) dummy $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK)  > $(@F)
 
 $(LIBDYNNAME) : ../$(LIBNAME) osx.def
-	$(PREFIX)gcc $(CFLAGS) -all_load -dynamiclib -o $(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
+	$(PREFIX)gcc $(CFLAGS) -all_load -dynamiclib -o ../$(LIBDYNNAME) $< -Wl,-exported_symbols_list,osx.def  $(FEXTRALIB)
 
 symbol.$(SUFFIX) : symbol.S
 	$(CC) $(CFLAGS) -c -o $(@F) $^
diff --git a/interface/axpy.c b/interface/axpy.c
index dd75b758c..82b0ee234 100644
--- a/interface/axpy.c
+++ b/interface/axpy.c
@@ -85,7 +85,11 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx, FLOAT *y, blasint inc
   //In that case, the threads would be dependent.
   if (incx == 0 || incy == 0)
 	  nthreads = 1;
-
+ 
+  //Temporarily walk around the low performance issue with small imput size & multithreads.
+  if (n <= 10000)
+	  nthreads = 1;
+  
   if (nthreads == 1) {
 #endif
 
diff --git a/interface/create b/interface/create
index b7be8ab6e..b7be8ab6e 100644..100755
--- a/interface/create
+++ b/interface/create
diff --git a/interface/dsdot.c b/interface/dsdot.c
index 66f7917d5..94237e0c4 100644
--- a/interface/dsdot.c
+++ b/interface/dsdot.c
@@ -49,6 +49,7 @@ double NAME(blasint *N, float *x, blasint *INCX, float *y, blasint *INCY){
   BLASLONG n    = *N;
   BLASLONG incx = *INCX;
   BLASLONG incy = *INCY;
+  double ret = 0.0;
 
   PRINT_DEBUG_NAME;
 
@@ -61,19 +62,21 @@ double NAME(blasint *N, float *x, blasint *INCX, float *y, blasint *INCY){
   if (incx < 0) x -= (n - 1) * incx;
   if (incy < 0) y -= (n - 1) * incy;
 
-  return DSDOT_K(n, x, incx, y, incy);
+  ret=DSDOT_K(n, x, incx, y, incy);
 
   FUNCTION_PROFILE_END(1, n, n);
 
   IDEBUG_END;
 
-  return 0;
+  return ret;
   
 }
 
 #else
 
 double CNAME(blasint n, float *x, blasint incx, float *y, blasint incy){
+
+  double ret = 0.0;
   
   PRINT_DEBUG_CNAME;
 
@@ -86,13 +89,13 @@ double CNAME(blasint n, float *x, blasint incx, float *y, blasint incy){
   if (incx < 0) x -= (n - 1) * incx;
   if (incy < 0) y -= (n - 1) * incy;
 
-  return DSDOT_K(n, x, incx, y, incy);
+  ret=DSDOT_K(n, x, incx, y, incy);
 
   FUNCTION_PROFILE_END(1, n, n);
 
   IDEBUG_END;
 
-  return 0;
+  return ret;
   
 }
 
diff --git a/interface/rotmg.c b/interface/rotmg.c
index c37c09914..3db891714 100644
--- a/interface/rotmg.c
+++ b/interface/rotmg.c
@@ -7,6 +7,12 @@
 #define  GAMSQ   16777216.e0
 #define  RGAMSQ  5.9604645e-8
 
+#ifdef DOUBLE
+#define ABS(x) fabs(x)
+#else
+#define ABS(x) fabsf(x)
+#endif
+
 #ifndef CBLAS
 
 void NAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT *DY1, FLOAT *dparam){
@@ -47,7 +53,7 @@ void CNAME(FLOAT *dd1, FLOAT *dd2, FLOAT *dx1, FLOAT dy1, FLOAT *dparam){
     dq2 =  dp2 * dy1;
     dq1 =  dp1 * *dx1;
 
-    if (! (abs(dq1) > abs(dq2))) goto L40;
+    if (! (ABS(dq1) > ABS(dq2))) goto L40;
 
     dh21 = -(dy1) / *dx1;
     dh12 = dp2 / dp1;
@@ -140,7 +146,7 @@ L150:
     goto L130;
 
 L160:
-    if (! (abs(*dd2) <= RGAMSQ)) {
+    if (! (ABS(*dd2) <= RGAMSQ)) {
 	goto L190;
     }
     if (*dd2 == ZERO) {
@@ -157,7 +163,7 @@ L180:
     goto L160;
 
 L190:
-    if (! (abs(*dd2) >= GAMSQ)) {
+    if (! (ABS(*dd2) >= GAMSQ)) {
 	goto L220;
     }
     igo = 3;
diff --git a/kernel/Makefile b/kernel/Makefile
index 6084cbc3f..aed145b60 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -53,6 +53,11 @@ SBLASOBJS	+= setparam$(TSUFFIX).$(SUFFIX)
 CCOMMON_OPT	+= -DTS=$(TSUFFIX)
 endif
 
+KERNEL_INTERFACE = ../common_level1.h ../common_level2.h ../common_level3.h 
+ifneq ($(NO_LAPACK), 1)
+KERNEL_INTERFACE += ../common_lapack.h 
+endif
+
 ifeq ($(ARCH), x86)
 COMMONOBJS	+= cpuid.$(SUFFIX)
 endif
@@ -88,9 +93,10 @@ setparam$(TSUFFIX).$(SUFFIX): setparam$(TSUFFIX).c kernel$(TSUFFIX).h
 setparam$(TSUFFIX).c : setparam-ref.c
 	sed 's/TS/$(TSUFFIX)/g' $< > $(@F)
 
-kernel$(TSUFFIX).h : ../common_level1.h ../common_level2.h ../common_level3.h ../common_lapack.h 
+kernel$(TSUFFIX).h : $(KERNEL_INTERFACE) 
 	sed 's/\ *(/$(TSUFFIX)(/g' $^ > $(@F)
 
+
 cpuid.$(SUFFIX): $(KERNELDIR)/cpuid.S
 	$(CC) -c $(CFLAGS) $< -o $(@F)
 
@@ -112,10 +118,10 @@ lsame.$(PSUFFIX): $(KERNELDIR)/$(LSAME_KERNEL)
 cpuid.$(PSUFFIX): $(KERNELDIR)/cpuid.S
 	$(CC) -c $(PFLAGS) $< -o $(@F)
 
-ifdef DYNAMIC_ARCH
+#ifdef DYNAMIC_ARCH
 clean ::
 	@rm -f setparam_*.c kernel_*.h setparam.h kernel.h
 
-endif
+#endif
 
 include $(TOPDIR)/Makefile.tail
diff --git a/kernel/Makefile.L1 b/kernel/Makefile.L1
index 317f14363..b08664a8e 100644
--- a/kernel/Makefile.L1
+++ b/kernel/Makefile.L1
@@ -668,7 +668,7 @@ $(KDIR)qdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)qdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNEL
 	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)dsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)dsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL) 
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
 
 $(KDIR)sdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)sdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL) 
 	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
diff --git a/kernel/mips64/KERNEL b/kernel/mips64/KERNEL
index dd0d2cfea..ebb447b11 100644
--- a/kernel/mips64/KERNEL
+++ b/kernel/mips64/KERNEL
@@ -91,15 +91,37 @@ ifndef ZGEMM_BETA
 ZGEMM_BETA = ../generic/zgemm_beta.c
 endif
 
+ifndef	STRSMKERNEL_LN
 STRSMKERNEL_LN	=  trsm_kernel_LN.S
+endif
+
+ifndef	STRSMKERNEL_LT
 STRSMKERNEL_LT	=  trsm_kernel_LT.S
+endif
+
+ifndef	STRSMKERNEL_RN
 STRSMKERNEL_RN	=  trsm_kernel_LT.S
+endif
+
+ifndef	STRSMKERNEL_RT
 STRSMKERNEL_RT	=  trsm_kernel_RT.S
+endif
 
+ifndef DTRSMKERNEL_LN
 DTRSMKERNEL_LN	=  trsm_kernel_LN.S
+endif
+
+ifndef DTRSMKERNEL_LT
 DTRSMKERNEL_LT	=  trsm_kernel_LT.S
+endif
+
+ifndef DTRSMKERNEL_RN
 DTRSMKERNEL_RN	=  trsm_kernel_LT.S
+endif
+
+ifndef DTRSMKERNEL_RT
 DTRSMKERNEL_RT	=  trsm_kernel_RT.S
+endif
 
 CTRSMKERNEL_LN	=  ztrsm_kernel_LT.S
 CTRSMKERNEL_LT	=  ztrsm_kernel_LT.S
diff --git a/kernel/mips64/KERNEL.LOONGSON3A b/kernel/mips64/KERNEL.LOONGSON3A
index b295070d9..e72ac142e 100644
--- a/kernel/mips64/KERNEL.LOONGSON3A
+++ b/kernel/mips64/KERNEL.LOONGSON3A
@@ -1,2 +1,24 @@
 SAXPYKERNEL=axpy_loongson3a.S
 DAXPYKERNEL=daxpy_loongson3a_simd.S
+
+SGEMMKERNEL    =  sgemm_kernel_loongson3a.S
+SGEMMONCOPY    = ../generic/gemm_ncopy_4.c
+SGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
+SGEMMONCOPYOBJ =  sgemm_oncopy.o
+SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+
+DGEMMKERNEL    =  gemm_kernel_loongson3a.S
+DGEMMONCOPY    = ../generic/gemm_ncopy_4.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
+DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy.o
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
diff --git a/kernel/mips64/dot.S b/kernel/mips64/dot.S
index b1f599172..6220b6ac9 100644
--- a/kernel/mips64/dot.S
+++ b/kernel/mips64/dot.S
@@ -300,7 +300,11 @@
 	.align 3
 
 .L999:
-	j	$31
 	ADD	s1, s1, s2
-
+#ifdef DSDOT
+	cvt.d.s s1, s1
+#endif
+	j	$31
+	NOP
+	
 	EPILOGUE
diff --git a/kernel/mips64/gemm_kernel_loongson3a.S b/kernel/mips64/gemm_kernel_loongson3a.S
new file mode 100644
index 000000000..3e95a3ed4
--- /dev/null
+++ b/kernel/mips64/gemm_kernel_loongson3a.S
@@ -0,0 +1,2390 @@
+#define REALNAME ASMNAME
+#define ASSEMBLER
+#include "common.h"
+#define FETCH	ld
+#define gsLQC1(base,fq,ft,offset) .word(0x32<<26|base<<21|ft<<16|0x1<<15|offset<<6|0x1<<5|fq)
+#define gsSQC1(base,fq,ft,offset) .word(0x3A<<26|base<<21|ft<<16|0x1<<15|offset<<6|0x1<<5|fq)
+
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$8
+#define B	$9
+#define C	$10
+#define LDC	$11
+
+#define AO	$12
+#define BO	$13
+
+#define CO1	$14
+#define CO2	$15
+#define CO3	$16
+#define CO4	$17
+
+#define KCO	$18
+#define MCO	$19
+#define NCO	$20
+
+#define SPANB	$21
+#define PREB	$23
+#define PREA	$24
+#define SPANA	$25
+
+#define ALPHA	$f15
+
+#if defined(TRMMKERNEL)
+#define	OFFSET	$2
+#define	KK	$3
+#define	TEMP	$7
+#endif
+
+#define R8	8
+#define	R9	9
+#define R14	14
+#define R15	15
+#define R16	16
+#define R17 17
+
+#define	t11	$f30
+#define	t21	$f31
+#define	t31	$f28
+#define	t41	$f29
+
+#define	t12	$f26
+#define	t22	$f27
+#define	t32	$f24
+#define	t42	$f25
+
+#define	t13	$f22
+#define	t23	$f23
+#define	t33	$f20
+#define	t43	$f21
+
+#define	t14	$f18
+#define	t24	$f19
+#define	t34	$f16
+#define	t44	$f17
+
+#define	c11	$f0
+#define	c21	$f1
+#define	c31	$f2
+#define	c41	$f3
+
+#define	c12	$f4
+#define	c22	$f5
+#define	c32	$f6
+#define	c42	$f7
+
+#define	c13	$f8
+#define	c23	$f9
+#define	c33	$f10
+#define c43	$f11
+
+#define	c14	$f12
+#define	c24	$f13
+#define	c34	$f14
+#define	c44	$f0
+
+#define	a0	$f0
+#define	a1	$f1
+#define	a2	$f2
+#define	a3	$f3
+#define	a4	$f4
+#define	a5	$f5
+#define	a6	$f6
+#define	a7	$f7
+#define	b0	$f8
+#define	b1	$f9
+#define	b2	$f10
+#define b3	$f11
+#define	b4	$f12
+#define	b5	$f13
+#define	b6	$f14
+#define	b7	$f15
+
+#define F31 31
+#define F30 30
+#define F29 29
+#define F28 28
+#define F27 27
+#define F26 26
+#define F25 25
+#define F24 24 
+#define F23 23
+#define F22 22
+#define F21 21
+#define F20 20
+#define F19 19
+#define F18 18
+#define F17 17
+#define F16 16 
+#define F15 15
+#define F14 14
+#define F13 13
+#define F12 12
+#define F11 11
+#define F10 10
+#define F9 9
+#define F8 8
+#define F7 7
+#define F6 6
+#define F5 5
+#define F4 4 
+#define F3 3 
+#define F2 2 
+#define F1 1 
+#define F0 0
+
+	PROLOGUE
+	
+	daddiu	$sp, $sp, -160
+	sd	$16,   0($sp)
+	sd	$17,   8($sp)
+	sd	$18,  16($sp)
+	sd	$19,  24($sp)
+	sd	$20,  32($sp)
+	sd	$21,  40($sp)
+	sd	$22,  48($sp)
+	ST	$f24, 56($sp)
+	ST	$f25, 64($sp)
+	ST	$f26, 72($sp)
+	ST	$f27, 80($sp)
+	ST	$f28, 88($sp)
+	sd	$23,  96($sp)
+	sd	$24, 104($sp)
+	sd	$25, 112($sp)
+	ST	$f20,120($sp)
+	ST	$f21,128($sp)
+	ST	$f22,136($sp)
+	ST	$f23,144($sp)
+
+
+	.align	5					
+.L0_N4:									#  Loop N
+	ST	ALPHA,152($sp)					#  Backup	ALPHA
+	move	MCO,M						#  Backup	M
+
+	move	NCO,N						#  Backup	N
+	move	KCO,K						#  Backup	K
+
+	move	AO,A						#  Backup	A_addr
+	dsra	N,NCO,2						#  N=NCO/2
+	
+	dsll	LDC,LDC,BASE_SHIFT			#  LDC*8Byte
+	dsll	SPANB,KCO,2+BASE_SHIFT		#  SPANB=KC*4nr*8Byte=KC*2^5
+	
+#if defined(TRMMKERNEL)
+	LDARG	OFFSET,160($sp)				#	OFFSET is relate to the data part		
+#endif
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	neg		KK,OFFSET				
+#endif
+	
+	move	BO,B						#  Backup	B_addr
+	beq		N,$0,.L0_N2					#  N=0,NCO<4
+	dsll	SPANA,KCO,1+BASE_SHIFT		#  SPANA = KCO*2mr*8Byte
+
+.L0_N4_Lb:								#	mr=4,nr=4
+	move	CO1,C							
+	dsra	M,MCO,2						#  M=MCO/2
+	
+	move	A,AO						#  Reset A
+	daddu	CO2,C,LDC
+
+	daddu	PREB,BO,SPANB				#  PreB point next panelB
+	daddu	CO3,CO2,LDC
+
+	daddu	PREA,AO,SPANA
+	daddu	CO4,CO3,LDC
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+	move	KK,OFFSET					
+#endif
+	beqz	M,.L14_M2
+	daddu	C,CO4,LDC					#	move C to next panel Cj
+
+.L10:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO						#	(SIDE=L and UPLO=L) or (SIZE=R and UPLO=U)
+#else
+	dsll	K,KK,2 + BASE_SHIFT			#  KK is the length that needs to span to the data part
+	dsll	TEMP,KK,2 + BASE_SHIFT
+
+	daddu	A,A,K						#  move A B to data part
+	daddu	B,BO,TEMP
+#endif
+
+	MTC		$0,t11
+	MOV	t21,t11
+	gsLQC1(R8,F1,F0,0)					#	a0,a1
+
+	MOV	t31,t11
+	MOV	t41,t11
+	gsLQC1(R9,F9,F8,0)					#	b0,b1
+
+	MOV	t12,t11
+	MOV	t22,t11
+	gsLQC1(R8,F3,F2,1)					#	a2,a3
+	
+	MOV	t32,t11
+	MOV	t42,t11
+	gsLQC1(R9,F11,F10,1)				#	b2,b3
+
+	MOV	t13,t11
+	MOV	t23,t11
+	
+	MOV	t33,t11
+	MOV	t43,t11
+
+	MOV	t14,t11
+	MOV	t24,t11
+	
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP,KCO,KK					#  temp is the length of the data part
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 4					#	S=L,U=L	
+#else
+	daddiu	TEMP, KK, 4					#	S=R,U=U,for this two situation KK is the length of the data part
+#endif
+	dsra	K,TEMP,2					#  K=KCO/2
+	MOV	t34,t11
+	beqz	K,.L15
+	MOV	t44,t11
+
+#else							
+	move	B,BO						#	Reset B
+	MTC		$0,t11						# 	GEMM part	NR=4,MR=4
+	gsLQC1(R8,F1,F0,0)					#	a0,a1
+
+	MOV	t21,t11
+	MOV	t31,t11
+	gsLQC1(R9,F9,F8,0)					#	b0,b1
+
+	MOV	t41,t11
+	MOV	t12,t11
+	gsLQC1(R8,F3,F2,1)					#	a2,a3
+	
+	MOV	t22,t11
+	MOV	t32,t11
+	gsLQC1(R9,F11,F10,1)				#	b2,b3
+
+	MOV	t42,t11
+	dsra	K,KCO,2						#  K=KCO/2
+	
+	MOV	t13,t11
+	MOV	t23,t11
+	
+	MOV	t33,t11
+	MOV	t43,t11
+
+	MOV	t14,t11
+	MOV	t24,t11
+	
+	MOV	t34,t11
+	beqz	K,.L15
+	MOV	t44,t11							#	clear 16 results registers
+#endif
+	
+	.align	5
+.L11:									#  kr=4
+	gsLQC1(R8,F5,F4,2)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R9,F13,F12,2)		
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	gsLQC1(R8,F7,F6,3)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	
+	gsLQC1(R9,F15,F14,3)
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+
+	FETCH		$0,(PREB)
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+	
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+	
+	FETCH		$0,(PREA)
+	MADD	t33,t33,a2,b2
+	MADD	t43,t43,a3,b2
+
+	MADD	t34,t34,a2,b3
+	MADD	t44,t44,a3,b3
+								
+.L12:
+	gsLQC1(R8,F1,F0,4)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F9,F8,4)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+
+	gsLQC1(R8,F3,F2,5)
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+
+	gsLQC1(R9,F11,F10,5)
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+
+	FETCH		$0,4*SIZE(PREB)
+	MADD	t13,t13,a4,b6
+	MADD	t23,t23,a5,b6
+
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t33,t33,a6,b6
+	MADD	t43,t43,a7,b6
+	
+	MADD	t34,t34,a6,b7
+	MADD	t44,t44,a7,b7
+
+.L13:
+	gsLQC1(R8,F5,F4,6)	
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R9,F13,F12,6)
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	gsLQC1(R8,F7,F6,7)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+
+	gsLQC1(R9,F15,F14,7)
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+	daddu	A,A,16*SIZE					#  4mr*4kr
+
+	FETCH		$0,8*SIZE(PREB)
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+	daddu	B,B,16*SIZE					#	4nr*4kr
+
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+
+	FETCH		$0,8*SIZE(PREA)
+	MADD	t33,t33,a2,b2
+	MADD	t43,t43,a3,b2
+
+	MADD	t34,t34,a2,b3
+	MADD	t44,t44,a3,b3
+	
+.L14:
+	gsLQC1(R8,F1,F0,0)						
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+
+	gsLQC1(R8,F3,F2,1)
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+	daddiu	K,K,-1
+
+	gsLQC1(R9,F11,F10,1)
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+
+	FETCH		$0,12*SIZE(PREB)
+	MADD	t13,t13,a4,b6
+	MADD	t23,t23,a5,b6
+
+	FETCH		$0,12*SIZE(PREA)
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+
+	MADD	t33,t33,a6,b6
+	MADD	t43,t43,a7,b6
+	daddu	PREB,PREB,16*SIZE
+	
+	MADD	t34,t34,a6,b7
+	MADD	t44,t44,a7,b7
+	bnez 	K,.L11
+	daddu	PREA,PREA,16*SIZE
+
+.L15:									#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP, 2
+#endif
+	beqz	K,.L18
+	nop
+
+.L16:			
+	gsLQC1(R8,F5,F4,2)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R9,F13,F12,2)		
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	gsLQC1(R8,F7,F6,3)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+
+	gsLQC1(R9,F15,F14,3)
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+	daddu	A,A,8*SIZE					#	4mr*2kr
+
+	FETCH		$0,0(PREB)
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+	daddu	B,B,8*SIZE					#	4nr*2kr
+
+	FETCH		$0,0(PREA)
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+
+	MADD	t33,t33,a2,b2
+	MADD	t43,t43,a3,b2
+
+	MADD	t34,t34,a2,b3
+	MADD	t44,t44,a3,b3
+								
+.L17:
+	gsLQC1(R8,F1,F0,0)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+
+	gsLQC1(R8,F3,F2,1)
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+
+	gsLQC1(R9,F11,F10,1)
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+
+	FETCH		$0,4*SIZE(PREB)
+	MADD	t13,t13,a4,b6
+	MADD	t23,t23,a5,b6
+
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+	daddu	PREB,PREB,8*SIZE
+
+	MADD	t33,t33,a6,b6
+	MADD	t43,t43,a7,b6
+	daddu	PREA,PREA,8*SIZE
+	
+	MADD	t34,t34,a6,b7
+	MADD	t44,t44,a7,b7
+	
+.L18:									#	kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L19				  
+	LD	ALPHA,152($sp)					#  Get ALPHA
+	
+	FETCH		$0,0(PREB)
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,4*SIZE					#  	4mr*kr
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+	daddu	B,B,4*SIZE					#	4nr*kr
+
+	FETCH		$0,0(PREA)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	daddu	PREB,PREB,4*SIZE
+
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+	daddu	PREA,PREA,4*SIZE
+
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+
+	MADD	t33,t33,a2,b2
+	MADD	t43,t43,a3,b2
+
+	MADD	t34,t34,a2,b3
+	MADD	t44,t44,a3,b3
+
+.L19:									#  Write Back to C
+#ifndef TRMMKERNEL				
+	LD	c11,0(CO1)						#  GEMM write part 
+	LD	c21,1*SIZE(CO1)					#  get 16 C
+	LD	c31,2*SIZE(CO1)
+	LD	c41,3*SIZE(CO1)
+
+	LD	c12,0(CO2)
+	MADD	t11,c11,t11,ALPHA
+	LD	c22,1*SIZE(CO2)
+	MADD	t21,c21,t21,ALPHA
+	LD	c32,2*SIZE(CO2)
+	MADD	t31,c31,t31,ALPHA
+	LD	c42,3*SIZE(CO2)
+	MADD	t41,c41,t41,ALPHA
+
+	LD	c13,0(CO3)
+	MADD	t12,c12,t12,ALPHA
+	LD	c23,1*SIZE(CO3)
+	MADD	t22,c22,t22,ALPHA
+	LD	c33,2*SIZE(CO3)
+	MADD	t32,c32,t32,ALPHA
+	LD	c43,3*SIZE(CO3)
+	MADD	t42,c42,t42,ALPHA
+
+	LD	c14,0(CO4)
+	MADD	t13,c13,t13,ALPHA
+	LD	c24,1*SIZE(CO4)
+	MADD	t23,c23,t23,ALPHA
+	LD	c34,2*SIZE(CO4)
+	MADD	t33,c33,t33,ALPHA
+	LD	c44,3*SIZE(CO4)
+	MADD	t43,c43,t43,ALPHA
+
+	ST	t11,0(CO1)
+	MADD	t14,c14,t14,ALPHA
+	ST	t21,1*SIZE(CO1)
+	MADD	t24,c24,t24,ALPHA
+	ST	t31,2*SIZE(CO1)
+	MADD	t34,c34,t34,ALPHA
+	ST	t41,3*SIZE(CO1)
+	MADD	t44,c44,t44,ALPHA
+	daddiu	M,M,-1						#  M--
+
+	ST	t12,0(CO2)
+	ST	t22,1*SIZE(CO2)
+	ST	t32,2*SIZE(CO2)
+	ST	t42,3*SIZE(CO2)
+
+	ST	t13,0(CO3)
+	ST	t23,1*SIZE(CO3)
+	ST	t33,2*SIZE(CO3)
+	ST	t43,3*SIZE(CO3)
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,4*SIZE(CO2)
+	FETCH	$0,4*SIZE(CO3)
+	FETCH	$0,4*SIZE(CO4)
+
+	FETCH	$0,8*SIZE(CO1)
+	FETCH	$0,8*SIZE(CO2)
+	FETCH	$0,8*SIZE(CO3)
+	FETCH	$0,8*SIZE(CO4)
+
+	ST	t14,0(CO4)
+	daddu	CO1,CO1,4*SIZE				#  COi += 4
+	ST	t24,1*SIZE(CO4)
+	daddu	CO2,CO2,4*SIZE
+	ST	t34,2*SIZE(CO4)
+	daddu	CO3,CO3,4*SIZE
+	ST	t44,3*SIZE(CO4)
+	daddu	PREB,BO,SPANB
+	
+	bnez	M,.L10				
+	daddu	CO4,CO4,4*SIZE
+
+#else							
+	MUL	t11, ALPHA, t11					#	TRMM write back part
+	MUL	t21, ALPHA, t21
+	MUL	t31, ALPHA, t31
+	MUL	t41, ALPHA, t41
+
+	ST	t11, 0 * SIZE(CO1)
+	MUL	t12, ALPHA, t12
+	ST	t21, 1 * SIZE(CO1)
+	MUL	t22, ALPHA, t22
+	ST	t31, 2 * SIZE(CO1)
+	MUL	t32, ALPHA, t32
+	ST	t41, 3 * SIZE(CO1)
+	MUL	t42, ALPHA, t42
+
+	ST	t12, 0 * SIZE(CO2)
+	MUL	t13, ALPHA, t13
+	ST	t22, 1 * SIZE(CO2)
+	MUL	t23, ALPHA, t23
+	ST	t32, 2 * SIZE(CO2)
+	MUL	t33, ALPHA, t33
+	ST	t42, 3 * SIZE(CO2)
+	MUL	t43, ALPHA, t43
+
+	ST	t13, 0 * SIZE(CO3)
+	MUL	t14, ALPHA, t14
+	ST	t23, 1 * SIZE(CO3)
+	MUL	t24, ALPHA, t24
+	ST	t33, 2 * SIZE(CO3)
+	MUL	t34, ALPHA, t34
+	ST	t43, 3 * SIZE(CO3)
+	MUL	t44, ALPHA, t44
+
+	ST	t14, 0 * SIZE(CO4)
+	daddiu	M,M,-1						#  M--
+	ST	t24, 1 * SIZE(CO4)
+	ST	t34, 2 * SIZE(CO4)
+	ST	t44, 3 * SIZE(CO4)
+	daddiu	CO1,CO1, 4 * SIZE
+	daddiu	CO2,CO2, 4 * SIZE
+	daddiu	CO3,CO3, 4 * SIZE
+	daddiu	CO4,CO4, 4 * SIZE	
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,4*SIZE(CO2)
+	FETCH	$0,4*SIZE(CO3)
+	FETCH	$0,4*SIZE(CO4)
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+	FETCH	$0,0(CO3)
+	FETCH	$0,0(CO4)
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP,KCO,KK								
+#ifdef LEFT
+	daddiu	TEMP,TEMP, -4
+#else
+	daddiu	TEMP,TEMP, -4
+#endif
+	dsll	K,TEMP,2 + BASE_SHIFT
+	dsll	TEMP,TEMP,2 + BASE_SHIFT
+	daddu	A,A,K						# 	mov A to the end of panel Ai
+	daddu	B,B,TEMP					# 	mov B to the end of panel Bj
+#endif
+
+#ifdef LEFT										
+	daddiu	KK, KK,4
+#endif
+	bnez	M,.L10					
+	nop
+#endif
+
+
+	.align 3
+.L14_M2:
+	andi	M, MCO, 2					#	nr=4,mr=2
+	beqz	M,.L14_M1			
+	nop
+
+.L20:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO						#	Reset B
+#else
+	dsll	K,KK,1 + BASE_SHIFT			#	mr=2	
+	dsll	TEMP,KK,2 + BASE_SHIFT		#	nr=4
+	daddu	A,A,K
+	daddu	B,BO,TEMP
+#endif
+
+	MTC		$0,t11
+	MOV	t21,t11
+	gsLQC1(R8,F1,F0,0)					#	a0,a1
+
+	MOV	t12,t11
+	MOV	t22,t11
+	gsLQC1(R9,F9,F8,0)					#	b0,b1
+	
+	MOV	t13,t11
+	MOV	t23,t11
+	gsLQC1(R9,F11,F10,1)				#	b2,b3
+	
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP,KCO,KK
+#elif defined(LEFT)
+	daddiu	TEMP,KK,2					#	left part,controlled by mr, mr=2
+#else
+	daddiu	TEMP,KK,4					#  	right part,controlled by nr,nr=4
+#endif
+	dsra	K,TEMP,2
+	MOV	t14,t11
+	beqz	K,.L25
+	MOV	t24,t11							#	clear 2*4=8 results registers
+
+#else
+	move	B,BO						#	Reset B 
+	MTC		$0,t11
+	gsLQC1(R8,F1,F0,0)		
+	
+	MOV	t21,t11
+	MOV	t12,t11
+	gsLQC1(R9,F9,F8,0)			
+
+	MOV	t22,t11
+	dsra	K,KCO,2				
+	gsLQC1(R9,F11,F10,1)		
+	
+	MOV	t13,t11
+	MOV	t23,t11
+	
+	MOV	t14,t11
+	beqz	K,.L25
+	MOV	t24,t11
+#endif
+
+.L21:									#  nr=4,mr=2,kr=4
+	gsLQC1(R8,F5,F4,1)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R9,F13,F12,2)		
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	gsLQC1(R9,F15,F14,3)
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+	
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+	
+	gsLQC1(R8,F3,F2,2)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F9,F8,4)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+
+	gsLQC1(R9,F11,F10,5)
+	MADD	t13,t13,a4,b6
+	MADD	t23,t23,a5,b6
+
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+	daddiu	K,K,-1
+
+	gsLQC1(R8,F7,F6,3)	
+	MADD	t11,t11,a2,b0
+	MADD	t21,t21,a3,b0
+
+	gsLQC1(R9,F13,F12,6)
+	MADD	t12,t12,a2,b1
+	MADD	t22,t22,a3,b1
+
+	gsLQC1(R9,F15,F14,7)
+	MADD	t13,t13,a2,b2
+	MADD	t23,t23,a3,b2
+	daddu	A,A,8*SIZE					#  2mr*4kr
+
+	MADD	t14,t14,a2,b3
+	MADD	t24,t24,a3,b3
+	daddu	B,B,16*SIZE					#	4nr*4kr
+
+	gsLQC1(R8,F1,F0,0)
+	MADD	t11,t11,a6,b4
+	MADD	t21,t21,a7,b4
+
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a6,b5
+	MADD	t22,t22,a7,b5
+
+	gsLQC1(R9,F11,F10,1)
+	MADD	t13,t13,a6,b6
+	MADD	t23,t23,a7,b6
+
+	MADD	t14,t14,a6,b7
+	bnez 	K,.L21
+	MADD	t24,t24,a7,b7
+
+.L25:										
+#ifndef TRMMKERNEL
+	andi	K,KCO,2						#	kr=2
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L28
+	nop
+
+.L26:			
+	gsLQC1(R8,F5,F4,1)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R9,F13,F12,2)		
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	gsLQC1(R9,F15,F14,3)
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+	daddu	A,A,4*SIZE					#  	2mr*2kr
+	
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+	daddu	B,B,8*SIZE					#	4nr*2kr
+
+.L27:
+	gsLQC1(R8,F1,F0,0)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+
+	gsLQC1(R9,F11,F10,1)
+	MADD	t13,t13,a4,b6
+	MADD	t23,t23,a5,b6
+
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+	
+.L28:									#	kr=1	
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L29				  
+	LD	ALPHA,152($sp)					#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,2*SIZE					#  2mr*kr
+	daddu	B,B,4*SIZE					#  4nr*kr
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+
+.L29:									#  Write Back to C
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)						#	GEMM write back part
+	LD	c21,1*SIZE(CO1)			
+
+	LD	c12,0(CO2)
+	LD	c22,1*SIZE(CO2)
+	
+	LD	c13,0(CO3)
+	MADD	t11,c11,t11,ALPHA
+	LD	c23,1*SIZE(CO3)
+	MADD	t21,c21,t21,ALPHA
+
+	LD	c14,0(CO4)
+	MADD	t12,c12,t12,ALPHA
+	LD	c24,1*SIZE(CO4)
+	MADD	t22,c22,t22,ALPHA
+
+	ST	t11,0(CO1)
+	MADD	t13,c13,t13,ALPHA
+	ST	t21,1*SIZE(CO1)
+	MADD	t23,c23,t23,ALPHA
+
+	ST	t12,0(CO2)
+	MADD	t14,c14,t14,ALPHA
+	ST	t22,1*SIZE(CO2)
+	MADD	t24,c24,t24,ALPHA
+
+	ST	t13,0(CO3)
+	daddu	CO1,CO1,2*SIZE				#  COi += 2
+	ST	t23,1*SIZE(CO3)
+	daddu	CO2,CO2,2*SIZE
+
+	ST	t14,0(CO4)
+	daddu	CO3,CO3,2*SIZE
+	ST	t24,1*SIZE(CO4)
+	daddu	CO4,CO4,2*SIZE
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+	FETCH	$0,0(CO3)
+	FETCH	$0,0(CO4)
+
+#else
+	MUL	t11, ALPHA, t11					#	TRMM write back part
+	MUL	t21, ALPHA, t21
+	
+	ST	t11, 0 * SIZE(CO1)
+	MUL	t12, ALPHA, t12
+	ST	t21, 1 * SIZE(CO1)
+	MUL	t22, ALPHA, t22
+	
+	ST	t12, 0 * SIZE(CO2)
+	MUL	t13, ALPHA, t13
+	ST	t22, 1 * SIZE(CO2)
+	MUL	t23, ALPHA, t23
+	
+	ST	t13, 0 * SIZE(CO3)
+	MUL	t14, ALPHA, t14
+	ST	t23, 1 * SIZE(CO3)
+	MUL	t24, ALPHA, t24
+	
+	ST	t14, 0 * SIZE(CO4)
+	ST	t24, 1 * SIZE(CO4)
+	
+	daddiu	CO1,CO1, 2 * SIZE
+	daddiu	CO2,CO2, 2 * SIZE
+	daddiu	CO3,CO3, 2 * SIZE
+	daddiu	CO4,CO4, 2 * SIZE
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+	FETCH	$0,0(CO3)
+	FETCH	$0,0(CO4)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP,KCO,KK
+#ifdef LEFT
+	daddiu	TEMP,TEMP,-2
+#else
+	daddiu	TEMP,TEMP,-4
+#endif
+	dsll	K,TEMP,1 + BASE_SHIFT
+	dsll	TEMP,TEMP,2 + BASE_SHIFT
+
+	daddu	A,A,K						#	move A to next panel Ai
+	daddu	B,B,TEMP					#	move B to next panel Bj
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+
+
+	.align 3
+.L14_M1:
+	andi	M,MCO,1						#	mr=1	
+	beqz	M,.L0_N4_Loop				#  	M = 0, finishing one panel Bj
+	nop
+
+.L30:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO						#	Reset B
+#else
+	dsll	K,KK, 0 + BASE_SHIFT
+	dsll	TEMP,KK,2 + BASE_SHIFT
+
+	daddu	A,A,K
+	daddu	B,BO,TEMP
+#endif
+	MTC		$0,t11
+	MOV	t12,t11
+	LD	a0,	0 * SIZE(A)					#	a0
+
+	MOV	t13,t11
+	gsLQC1(R9,F9,F8,0)					#	b0,b1
+
+	MOV	t14,t11							#	clear result registers
+	gsLQC1(R9,F11,F10,1)				#	b2,b3
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1
+#else
+	daddiu	TEMP, KK, 4
+#endif
+	dsra	K,TEMP, 2
+	nop
+	beqz	K,.L35
+	nop
+								
+#else							
+	move	B,BO						#	Reset B, GEMM part
+	dsra	K,KCO,2						#  	K=KCO/2
+	LD	a0, 0 * SIZE(A)					#	a0
+
+	MTC		$0,t11
+	MOV	t12,t11
+	gsLQC1(R9,F9,F8,0)					#	b0,b1
+
+	MOV	t13,t11
+	MOV	t14,t11
+	gsLQC1(R9,F11,F10,1)				#	b2,b3
+	
+	beqz	K,.L35
+	nop
+#endif
+
+.L31:									#	nr=4,mr=1,kr=4	
+	LD	a1,	1*SIZE(A)					#	load a1
+	MADD	t11,t11,a0,b0
+	
+	gsLQC1(R9,F13,F12,2)				#	b4,b5
+	MADD	t12,t12,a0,b1
+	
+	gsLQC1(R9,F15,F14,3)				#	b6,b7
+	MADD	t13,t13,a0,b2
+	MADD	t14,t14,a0,b3
+
+	LD	a2,	2*SIZE(A)					#	a2
+	MADD	t11,t11,a1,b4
+	
+	gsLQC1(R9,F9,F8,4)
+	MADD	t12,t12,a1,b5
+	
+	gsLQC1(R9,F11,F10,5)
+	MADD	t13,t13,a1,b6
+	MADD	t14,t14,a1,b7
+	daddiu	K,K,-1
+
+	LD	a3,	3*SIZE(A)					#	a3
+	MADD	t11,t11,a2,b0
+	
+	gsLQC1(R9,F13,F12,6)
+	MADD	t12,t12,a2,b1
+	daddu	A,A,4*SIZE					#	1mr*4kr
+	
+	gsLQC1(R9,F15,F14,7)
+	MADD	t13,t13,a2,b2
+	MADD	t14,t14,a2,b3
+	daddu	B,B,16*SIZE					#	4nr*4kr
+
+	LD	a0,	0*SIZE(A)					#	a0
+	MADD	t11,t11,a3,b4
+	
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a3,b5
+	
+	gsLQC1(R9,F11,F10,1)
+	MADD	t13,t13,a3,b6
+	bnez 	K,.L31
+	MADD	t14,t14,a3,b7
+
+.L35:									#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2			
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L38
+	nop
+
+.L36:			
+	LD	a1,1*SIZE(A)					#	load a1
+	MADD	t11,t11,a0,b0
+	
+	gsLQC1(R9,F13,F12,2)				
+	MADD	t12,t12,a0,b1
+	daddu	A,A,2*SIZE					#  	mr*2kr
+	
+	gsLQC1(R9,F15,F14,3)
+	MADD	t13,t13,a0,b2
+	MADD	t14,t14,a0,b3
+	daddu	B,B,8*SIZE					#	4nr*2kr
+
+
+.L37:
+	LD	a0,0(A)
+	MADD	t11,t11,a1,b4
+	
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a1,b5
+	
+	gsLQC1(R9,F11,F10,1)
+	MADD	t13,t13,a1,b6
+	MADD	t14,t14,a1,b7
+	
+.L38:									#  	kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L39				
+	LD	ALPHA,152($sp)					#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t12,t12,a0,b1
+	daddu	A,A,1*SIZE				
+	daddu	B,B,4*SIZE
+	
+	MADD	t13,t13,a0,b2
+	MADD	t14,t14,a0,b3
+
+.L39:									#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)			
+	LD	c12,0(CO2)
+	LD	c13,0(CO3)
+	LD	c14,0(CO4)
+	
+	MADD	t11,c11,t11,ALPHA
+	MADD	t12,c12,t12,ALPHA
+	MADD	t13,c13,t13,ALPHA
+	MADD	t14,c14,t14,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t12,0(CO2)
+	ST	t13,0(CO3)
+	ST	t14,0(CO4)
+#else
+	MUL	t11, ALPHA, t11
+	MUL	t12, ALPHA, t12
+	MUL	t13, ALPHA, t13
+	MUL	t14, ALPHA, t14
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -1
+#else
+	daddiu	TEMP, TEMP, -4
+#endif
+
+	dsll	K,TEMP, 0 + BASE_SHIFT
+	dsll	TEMP,TEMP, 2 + BASE_SHIFT
+
+	daddu	A,A,K
+	daddu	B,B,TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 1
+#endif
+#endif
+
+
+	.align	3
+.L0_N4_Loop:								#	mc finished
+	daddiu	N,N,-1							#  N--
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	daddiu	KK, KK,4 
+#endif
+	bnez	N,.L0_N4_Lb			
+	move	BO,B							#  Set BO point to next panel Bj
+
+	.align	5					
+.L0_N2:
+	andi	N,NCO,2							#  	nr = 2
+	beqz	N,.L0_N1		
+	nop
+
+.L0_N2_Lb:
+	move	CO1,C					
+	daddu	CO2,C,LDC
+
+	dsra	M,MCO,2				
+	move	A,AO							#  Reset A
+
+	daddu	PREA,AO,SPANA
+	daddu	C,CO2,LDC
+
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	move	KK, OFFSET
+#endif
+	beqz	M,.L12_M2
+	nop
+
+.L40:						
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO							#	Reset B
+#else
+	dsll	K,KK, 2 + BASE_SHIFT
+	dsll	TEMP, KK,1 + BASE_SHIFT	
+
+	daddu	A,A,K
+	daddu	B,BO,TEMP
+#endif
+	MTC		$0,t11
+	MOV	t21,t11
+	gsLQC1(R8,F1,F0,0)						#	a0,a1
+
+	MOV	t31,t11
+	MOV	t41,t11
+	gsLQC1(R9,F9,F8,0)						#	b0,b1
+
+	MOV	t12,t11
+	MOV	t22,t11
+	gsLQC1(R8,F3,F2,1)						#	a2,a3
+	
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP,KCO,KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 4
+#else
+	daddiu	TEMP, KK, 2
+#endif
+	dsra	K,TEMP,2				
+	MOV	t32,t11
+	beqz	K,.L45
+	MOV	t42,t11
+
+#else
+	move	B,BO							#	Reset B
+	MTC		$0,t11							#  	gemm part
+	gsLQC1(R8,F1,F0,0)						#	a0,a1
+
+	MOV	t21,t11
+	MOV	t31,t11
+	gsLQC1(R9,F9,F8,0)						#	b0,b1
+
+	MOV	t41,t11
+	dsra	K,KCO,2							#	K=KCO/2
+	gsLQC1(R8,F3,F2,1)						#	a2,a3
+	
+	MOV	t12,t11
+	MOV	t22,t11
+	
+	MOV	t32,t11
+	beqz	K,.L45
+	MOV	t42,t11
+#endif
+
+.L41:										#  	nr=2,mr=kr=4
+	gsLQC1(R8,F5,F4,2)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R9,F13,F12,1)	
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	gsLQC1(R8,F7,F6,3)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	
+	FETCH		$0,(PREA)
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+
+.L42:
+	gsLQC1(R8,F1,F0,4)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F11,F10,2)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+
+	gsLQC1(R8,F3,F2,5)
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+
+.L43:
+	gsLQC1(R8,F5,F4,6)	
+	MADD	t11,t11,a0,b2
+	MADD	t21,t21,a1,b2
+
+	gsLQC1(R9,F15,F14,3)
+	MADD	t12,t12,a0,b3
+	MADD	t22,t22,a1,b3
+
+	gsLQC1(R8,F7,F6,7)
+	MADD	t31,t31,a2,b2
+	MADD	t41,t41,a3,b2
+	daddu	B,B,8*SIZE						#	2nr*4kr	
+
+	FETCH		$0,8*SIZE(PREA)
+	MADD	t32,t32,a2,b3
+	MADD	t42,t42,a3,b3
+	daddu	A,A,16*SIZE						#	4mr*4kr
+
+.L44:
+	gsLQC1(R8,F1,F0,0)
+	MADD	t11,t11,a4,b6
+	MADD	t21,t21,a5,b6
+	daddiu	K,K,-1
+
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a4,b7
+	MADD	t22,t22,a5,b7
+	daddu	PREA,PREA,16*SIZE
+
+	gsLQC1(R8,F3,F2,1)
+	MADD	t31,t31,a6,b6
+	MADD	t41,t41,a7,b6
+
+	FETCH		$0,-4*SIZE(PREA)
+	MADD	t32,t32,a6,b7
+	bnez 	K,.L41
+	MADD	t42,t42,a7,b7
+
+
+.L45:										#  	kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L48
+	nop
+
+.L46:			
+	gsLQC1(R8,F5,F4,2)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R9,F13,F12,1)		
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	gsLQC1(R8,F7,F6,3)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	daddu	B,B,4*SIZE						#  B+=2(nr)*2(kr)*8Byte=32
+
+	FETCH		$0,0(PREA)
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+	daddu	A,A,8*SIZE						#  A+=4(mr)*2(kr)*8Byte=8*SIZE
+
+.L47:
+	gsLQC1(R8,F1,F0,0)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+
+	gsLQC1(R8,F3,F2,1)
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+	daddu	PREA,PREA,8*SIZE
+
+	
+.L48:										#	 kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L49				 
+	LD	ALPHA,152($sp)						#  Get ALPHA
+	
+	FETCH		$0,0(PREA)
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,4*SIZE						#  A+=4(mr)*1(kr)*8Byte=32
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+	daddu	B,B,2*SIZE
+	daddu	PREA,PREA,4*SIZE
+
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+
+.L49:										#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)							#  gemm write back part Fetch 16 C
+	LD	c21,1*SIZE(CO1)			
+	LD	c31,2*SIZE(CO1)
+	LD	c41,3*SIZE(CO1)
+
+	LD	c12,0(CO2)
+	MADD	t11,c11,t11,ALPHA
+	LD	c22,1*SIZE(CO2)
+	MADD	t21,c21,t21,ALPHA
+	LD	c32,2*SIZE(CO2)
+	MADD	t31,c31,t31,ALPHA
+	LD	c42,3*SIZE(CO2)
+	MADD	t41,c41,t41,ALPHA
+
+	ST	t11,0(CO1)
+	MADD	t12,c12,t12,ALPHA
+	ST	t21,1*SIZE(CO1)
+	MADD	t22,c22,t22,ALPHA
+	ST	t31,2*SIZE(CO1)
+	MADD	t32,c32,t32,ALPHA
+	ST	t41,3*SIZE(CO1)
+	MADD	t42,c42,t42,ALPHA
+	daddiu	M,M,-1				
+
+	ST	t12,0(CO2)
+	ST	t22,1*SIZE(CO2)
+	ST	t32,2*SIZE(CO2)
+	ST	t42,3*SIZE(CO2)
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,4*SIZE(CO2)
+	FETCH	$0,8*SIZE(CO1)
+	FETCH	$0,8*SIZE(CO2)
+
+	daddu	CO1,CO1,4*SIZE			
+	bnez	M,.L40				
+	daddu	CO2,CO2,4*SIZE
+
+#else
+	MUL	t11, ALPHA, t11
+	MUL	t21, ALPHA, t21
+	MUL	t31, ALPHA, t31
+	MUL	t41, ALPHA, t41
+	
+	MUL	t12, ALPHA, t12
+	ST	t11, 0 * SIZE(CO1)
+	MUL	t22, ALPHA, t22
+	ST	t21, 1 * SIZE(CO1)
+	MUL	t32, ALPHA, t32
+	ST	t31, 2 * SIZE(CO1)
+	MUL	t42, ALPHA, t42
+	ST	t41, 3 * SIZE(CO1)
+	
+	ST	t12, 0 * SIZE(CO2)
+	daddiu	M,M,-1
+	ST	t22, 1 * SIZE(CO2)
+	ST	t32, 2 * SIZE(CO2)
+	ST	t42, 3 * SIZE(CO2)
+	
+	daddiu	CO1,CO1, 4*SIZE
+	daddiu	CO2,CO2, 4*SIZE
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+	FETCH	$0,4(CO1)
+	FETCH	$0,4(CO2)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) ||(!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -4
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+	dsll	K,TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+
+	daddu	A,A,K
+	daddu	B,B,TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 4
+#endif
+	bnez	M,.L40
+	nop
+#endif
+
+
+	.align 3
+.L12_M2:
+	andi	M,MCO,2						#  	mr = 2
+	beqz	M,.L12_M1			
+	nop
+
+.L50:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO
+#else
+	dsll	K,    KK, 1 + BASE_SHIFT	#mr=2
+	dsll	TEMP, KK, 1 + BASE_SHIFT	#nr=2
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+	MTC		$0,t11
+	gsLQC1(R8,F1,F0,0)					#a0,a1
+
+	MOV	t21,t11
+	gsLQC1(R9,F9,F8,0)					#b0,b1
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 2
+#else
+	daddiu	TEMP, KK, 2
+#endif
+	dsra	K,TEMP,2			
+	MOV	t12,t11
+	beqz	K,.L55
+	MOV	t22,t11
+
+#else
+	move	B,BO
+	dsra	K,KCO,2						#  K=KCO/2
+	gsLQC1(R8,F1,F0,0)					#a0,a1
+
+	MTC		$0,t11
+	MOV	t21,t11
+	gsLQC1(R9,F9,F8,0)					#b0,b1
+	
+	MOV	t12,t11
+	beqz	K,.L55
+	MOV	t22,t11
+#endif
+
+.L51:									#  nr=2 mr=2,kr=4
+	gsLQC1(R8,F5,F4,1)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R9,F13,F12,1)	
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	gsLQC1(R8,F3,F2,2)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F11,F10,2)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+	daddiu	K,K,-1
+
+	gsLQC1(R8,F7,F6,3)	
+	MADD	t11,t11,a2,b2
+	MADD	t21,t21,a3,b2
+	daddu	A,A,8*SIZE					#  A+=2(mr)*4(kr)*8Byte=8*SIZE
+
+	gsLQC1(R9,F15,F14,3)
+	MADD	t12,t12,a2,b3
+	MADD	t22,t22,a3,b3
+	daddu	B,B,8*SIZE					#  B+=2(nr)*4(kr)*8Byte=16*SIZE
+
+	gsLQC1(R8,F1,F0,0)
+	MADD	t11,t11,a6,b6
+	MADD	t21,t21,a7,b6
+
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a6,b7
+	bnez 	K,.L51
+	MADD	t22,t22,a7,b7
+
+.L55:									#  	kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L58
+	nop
+
+.L56:			
+	gsLQC1(R8,F5,F4,1)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,4*SIZE					#  A+=2(mr)*2(kr)*8Byte=32
+
+	gsLQC1(R9,F13,F12,1)		
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+	daddu	B,B,4*SIZE					#	2nr*2kr
+
+.L57:
+	gsLQC1(R8,F1,F0,0)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+
+	
+.L58:									#  kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP, 1
+#endif
+	beqz	K,.L59				
+	LD	ALPHA,152($sp)					#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,2*SIZE					#  	A+=2(mr)*1(kr)*8Byte=16
+	daddu	B,B,2*SIZE					#	2nr*kr
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+
+.L59:									#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)						#  write gemm part back Fetch 16 C
+	LD	c21,1*SIZE(CO1)			
+	LD	c12,0(CO2)
+	LD	c22,1*SIZE(CO2)
+	
+	MADD	t11,c11,t11,ALPHA
+	MADD	t21,c21,t21,ALPHA
+	MADD	t12,c12,t12,ALPHA
+	MADD	t22,c22,t22,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t21,1*SIZE(CO1)
+	ST	t12,0(CO2)
+	ST	t22,1*SIZE(CO2)
+
+	daddu	CO1,CO1,2*SIZE			
+	daddu	CO2,CO2,2*SIZE
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+#else
+	daddiu	M, M, -1
+	daddiu	CO1,CO1, 2 * SIZE
+	daddiu	CO2,CO2, 2 * SIZE
+	MUL	t11, ALPHA, t11
+	MUL	t21, ALPHA, t21
+	MUL	t12, ALPHA, t12
+	MUL	t22, ALPHA, t22
+
+	ST	t11, -2 * SIZE(CO1)
+	ST	t21, -1 * SIZE(CO1)
+	ST	t12, -2 * SIZE(CO2)
+	ST	t22, -1 * SIZE(CO2)
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -2
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+
+	dsll	K,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, B, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+
+
+	.align 3
+.L12_M1:
+	andi	M,MCO,1					#  	mr = 1
+	beqz	M,.L0_N2_Loop		
+	nop
+
+.L60:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO					#	Reset B
+#else
+	dsll	K,    KK, 0 + BASE_SHIFT
+	dsll	TEMP, KK, 1 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+	MTC		$0,t11
+	LD	a0, 0*SIZE(A)				#	a0
+	
+	MOV	t21,t11
+	gsLQC1(R9,F9,F8,0)				#	b0,b1
+	
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1
+#else
+	daddiu	TEMP, KK, 2
+#endif
+	dsra	K,TEMP,2				
+	MOV	t12,t11
+	beqz	K,.L65
+	MOV	t22,t11
+
+#else
+	dsra	K,KCO,2				
+	move	B,BO					#  Reset B
+	LD	a0,0*SIZE(A)
+	
+	MTC		$0,t11
+	MOV	t21,t11
+	gsLQC1(R9,F9,F8,0)		
+
+	MOV	t12,t11
+	beqz	K,.L65
+	MOV	t22,t11
+#endif
+
+.L61:								#	nr=2,mr=1,kr=4	
+	LD	a4,	1*SIZE(A)				#	a2
+	MADD	t11,t11,a0,b0
+	
+	gsLQC1(R9,F13,F12,1)		
+	MADD	t12,t12,a0,b1
+
+	LD	a2,	2*SIZE(A)				#	a3
+	MADD	t11,t11,a4,b4
+	
+	gsLQC1(R9,F11,F10,2)
+	MADD	t12,t12,a4,b5
+
+	LD	a6,	3*SIZE(A)				#	a4
+	MADD	t11,t11,a2,b2
+	daddiu	K,K,-1
+	
+	gsLQC1(R9,F15,F14,3)
+	MADD	t12,t12,a2,b3
+	daddu	A,A,4*SIZE				#  A+=1(mr)*4(kr)*8Byte=32
+
+	LD	a0,	0*SIZE(A)
+	MADD	t11,t11,a6,b6
+	daddu	B,B,8*SIZE				#  B+=2(nr)*4(kr)*8Byte=8*SIZE
+	
+	gsLQC1(R9,F9,F8,0)				#	a0
+	bnez 	K,.L61
+	MADD	t12,t12,a6,b7
+
+.L65:								#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L68
+	nop
+
+.L66:			
+	LD	a4,	1*SIZE(A)				#	a1
+	MADD	t11,t11,a0,b0
+	daddu	A,A,2*SIZE				#  A+=1(mr)*2(kr)*8Byte=16
+	
+	gsLQC1(R9,F13,F12,1)	
+	MADD	t12,t12,a0,b1
+	daddu	B,B,4*SIZE
+
+.L67:
+	LD	a0,0(A)						#	a0
+	MADD	t11,t11,a4,b4
+	
+	gsLQC1(R9,F9,F8,0)
+	MADD	t12,t12,a4,b5
+
+	
+.L68:								#   kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L69				  
+	LD	ALPHA,152($sp)				#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t12,t12,a0,b1
+	daddu	A,A,1*SIZE				#  A+=1(mr)*1(kr)*8Byte=16
+	daddu	B,B,2*SIZE
+
+
+.L69:								#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)					#  Fetch 16 C
+	LD	c12,0(CO2)
+	
+	MADD	t11,c11,t11,ALPHA
+	MADD	t12,c12,t12,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t12,0(CO2)
+
+	daddu	CO1,CO1,1*SIZE		
+	daddu	CO2,CO2,1*SIZE
+
+#else
+	MUL	t11, ALPHA, t11
+	MUL	t12, ALPHA, t12
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+
+	daddu	CO1,CO1,1*SIZE			
+	daddu	CO2,CO2,1*SIZE
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -1
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+
+	dsll	K,    TEMP, 0 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, B, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 1
+#endif
+#endif
+
+.L0_N2_Loop:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	daddiu	KK, KK, 2
+#endif
+	move	BO, B
+
+
+	.align	5					
+.L0_N1:
+	andi	N,NCO,1					#  nr = 1
+	beqz	N,.L999					
+	nop
+
+	move	CO1,C				
+	dsra	M,MCO,2				
+	
+	move	A,AO					#  Reset A
+	daddu	PREA,AO,SPANA
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	move	KK, OFFSET
+#endif
+
+	beqz	M,.L11_M2
+	daddu	C,CO1,LDC
+
+.L70:						
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B, BO					#	Reset B
+#else
+	dsll	K,    KK, 2 + BASE_SHIFT
+	dsll	TEMP, KK, 0 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+
+	MTC		$0,t11
+	LD	b0,	0*SIZE(B)
+	
+	MOV	t21,t11
+	gsLQC1(R8,F1,F0,0)				#a0,a1
+
+	MOV	t31,t11
+	gsLQC1(R8,F3,F2,1)				#a2,a3
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 4
+#else
+	daddiu	TEMP, KK, 1
+#endif
+	dsra	K,TEMP,2		
+	MOV	t41,t11
+	beqz	K,.L75
+	nop
+#else
+	move	B, BO					#	Reset B
+	dsra	K,KCO,2			
+	LD	b0,	0*SIZE(B)
+	
+	MTC		$0,t11
+	MOV	t21,t11
+	gsLQC1(R8,F1,F0,0)				#a0,a1
+	
+	MOV	t31,t11
+	MOV	t41,t11
+	gsLQC1(R8,F3,F2,1)				#a2,a3
+	
+	beqz	K,.L75
+	nop
+#endif
+
+.L71:								#  nr=1,mr=kr=4
+	LD	b4,	1*SIZE(B)				#	b1
+	MADD	t11,t11,a0,b0
+	
+	gsLQC1(R8,F5,F4,2)			
+	MADD	t21,t21,a1,b0
+
+	gsLQC1(R8,F7,F6,3)
+	FETCH		$0,(PREA)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+
+.L72:
+	LD	b2,	2*SIZE(B)				#	b2
+	MADD	t11,t11,a4,b4
+	gsLQC1(R8,F1,F0,4)
+	MADD	t21,t21,a5,b4
+
+	gsLQC1(R8,F3,F2,5)
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+
+.L73:
+	LD	b6,	3*SIZE(B)
+	MADD	t11,t11,a0,b2
+	daddu	B,B,4*SIZE				#  B+=1(nr)*4(kr)*8Byte=32
+	
+	gsLQC1(R8,F5,F4,6)	
+	MADD	t21,t21,a1,b2
+	FETCH		$0,8*SIZE(PREA)
+
+	gsLQC1(R8,F7,F6,7)
+	MADD	t31,t31,a2,b2
+	MADD	t41,t41,a3,b2
+	daddu	A,A,16*SIZE				#  A+=4(mr)*4(kr)*8Byte=16*SIZE
+
+.L74:
+	LD	b0,	0*SIZE(B)
+	MADD	t11,t11,a4,b6
+	daddu	PREA,PREA,16*SIZE
+	
+	gsLQC1(R8,F1,F0,0)
+	MADD	t21,t21,a5,b6
+	daddiu	K,K,-1
+	FETCH		$0,-32(PREA)
+
+	gsLQC1(R8,F3,F2,1)
+	MADD	t31,t31,a6,b6
+	bnez 	K,.L71
+	MADD	t41,t41,a7,b6
+
+
+.L75:								#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L78
+	nop
+
+.L76:			
+	LD	b4,	1*SIZE(B)
+	MADD	t11,t11,a0,b0
+	daddu	B,B,2*SIZE				#  B+=1(nr)*2(kr)*8Byte=32
+	
+	gsLQC1(R8,F5,F4,2)			
+	MADD	t21,t21,a1,b0
+	FETCH		$0,0(PREA)
+
+	gsLQC1(R8,F7,F6,3)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	daddu	A,A,8*SIZE				#  A+=4(mr)*2(kr)*8Byte=8*SIZE
+
+.L77:
+	LD	b0,0(B)
+	MADD	t11,t11,a4,b4
+	
+	gsLQC1(R8,F1,F0,0)
+	MADD	t21,t21,a5,b4
+	FETCH		$0,4*SIZE(PREA)
+
+	gsLQC1(R8,F3,F2,1)
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+	daddu	PREA,PREA,8*SIZE
+
+	
+.L78:								#   kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L79				 
+	LD	ALPHA,152($sp)				#  Get ALPHA
+	
+	FETCH		$0,0(PREA)
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,4*SIZE				#  A+=4(mr)*1(kr)*8Byte=32
+
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	daddu	B,B,1*SIZE
+	daddu	PREA,PREA,4*SIZE
+
+
+.L79:								#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)					#  Fetch 16 C
+	LD	c21,1*SIZE(CO1)			
+	LD	c31,2*SIZE(CO1)
+	LD	c41,3*SIZE(CO1)
+
+	MADD	t11,c11,t11,ALPHA
+	MADD	t21,c21,t21,ALPHA
+	MADD	t31,c31,t31,ALPHA
+	MADD	t41,c41,t41,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t21,1*SIZE(CO1)
+	ST	t31,2*SIZE(CO1)
+	ST	t41,3*SIZE(CO1)
+	daddiu	M,M,-1					#  M--
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,8*SIZE(CO1)
+
+	bnez	M,.L70					#  M!=0
+	daddu	CO1,CO1,4*SIZE			#  COx += 4*8Byte
+#else
+	daddiu	M,M,-1					#  M--
+	MUL	t11, ALPHA, t11
+	MUL	t21, ALPHA, t21
+	MUL	t31, ALPHA, t31
+	MUL	t41, ALPHA, t41
+
+	ST	t11,0(CO1)
+	ST	t21,1*SIZE(CO1)
+	ST	t31,2*SIZE(CO1)
+	ST	t41,3*SIZE(CO1)
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,8*SIZE(CO1)
+
+	daddu	CO1,CO1,4*SIZE			
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -4
+#else
+	daddiu	TEMP, TEMP, -1
+#endif
+
+	dsll	K,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+
+	daddu	A, A,K
+	daddu	B, B, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 4
+#endif
+	bnez	M,.L70				
+	nop
+#endif
+
+
+	.align 3
+.L11_M2:
+	andi	M,MCO,2					#  mr = 2
+	beqz	M,.L11_M1			
+	nop
+
+.L80:						
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B, BO
+#else
+	dsll	K,    KK, 1 + BASE_SHIFT
+	dsll	TEMP, KK, 0 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+
+	LD	b0,	0*SIZE(B)
+	MTC		$0,t11
+	
+	gsLQC1(R8,F1,F0,0)				#a0,a1
+	MOV		t21,t11
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 2
+#else
+	daddiu	TEMP, KK, 1
+#endif
+	dsra	K,TEMP,2				#  K=KCO/2
+	beqz	K,.L85
+	nop
+#else
+	move	B, BO
+	dsra	K,KCO,2				
+	LD	b0,	0*SIZE(B)
+
+	MTC		$0,t11
+	MOV		t21,t11
+	gsLQC1(R8,F1,F0,0)				#a0,a1
+	
+	beqz	K,.L85
+	nop
+#endif
+
+.L81:								#  nr=1,mr=2,kr=4
+	LD	b4,	1*SIZE(B)
+	gsLQC1(R8,F5,F4,1)			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+
+	LD	b2,	2*SIZE(B)
+	gsLQC1(R8,F3,F2,2)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+	
+	LD	b6,	3*SIZE(B)
+	gsLQC1(R8,F7,F6,3)
+	MADD	t11,t11,a2,b2
+	MADD	t21,t21,a3,b2
+
+	daddu	A,A,8*SIZE				#  A+=2(mr)*4(kr)*8Byte=8*SIZE
+	daddu	B,B,4*SIZE				#  B+=1(nr)*4(kr)*8Byte=32
+
+	LD	b0,	0*SIZE(B)
+	gsLQC1(R8,F1,F0,0)
+	MADD	t11,t11,a6,b6
+	MADD	t21,t21,a7,b6
+	
+	daddiu	K,K,-1
+	bnez 	K,.L81
+	nop
+
+.L85:								#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L88
+	nop
+
+.L86:			
+	gsLQC1(R8,F5,F4,1)		
+	LD	b4,	1*SIZE(B)
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	
+	daddu	A,A,4*SIZE				#  A+=2(mr)*2(kr)*8Byte=32
+	daddu	B,B,2*SIZE				#  B+=1(nr)*2(kr)*8Byte=16
+	
+	gsLQC1(R8,F1,F0,0)
+	LD	b0,0(B)
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+
+	
+.L88:								#  kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L89				
+	LD	ALPHA,152($sp)				#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,2*SIZE				#  A+=2(mr)*1(kr)*8Byte=16
+	daddu	B,B,1*SIZE
+
+
+.L89:								#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)					#  Fetch 16 C
+	LD	c21,1*SIZE(CO1)			
+
+	MADD	t11,c11,t11,ALPHA
+	MADD	t21,c21,t21,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t21,1*SIZE(CO1)
+
+	FETCH	$0,2*SIZE(CO1)
+	
+	daddu	CO1,CO1,2*SIZE			#  COx += 2*8Byte
+
+#else
+	daddu	CO1,CO1,2*SIZE			#  COx += 2*8Byte
+	MUL	t11, ALPHA, t11
+	MUL	t21, ALPHA, t21
+
+	FETCH	$0,0(CO1)
+	ST	t11, -2 * SIZE(CO1)
+	ST	t21, -1 * SIZE(CO1)
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -2
+#else
+	daddiu	TEMP, TEMP, -1
+#endif
+
+	dsll	K,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, B, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+
+
+	.align 3
+.L11_M1:
+	andi		M,MCO,1				#   mr = 1
+	beqz	M,.L999			
+	nop
+
+.L90:			
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,  BO
+#else
+	dsll	K,    KK, 0 + BASE_SHIFT
+	dsll	TEMP, KK, 0 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+	LD	a0,	0*SIZE(A)
+	LD	b0,	0*SIZE(B)
+	MTC		$0,t11
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1
+#else
+	daddiu	TEMP, KK, 1
+#endif
+	dsra	K,  TEMP, 2
+	beqz	K,.L95
+	nop
+
+#else
+	move	B,  BO
+	LD	a0,	0*SIZE(A)
+	LD	b0,	0*SIZE(B)
+	dsra	K,KCO,2				
+	beqz	K,.L95
+	MTC		$0,t11
+#endif
+
+.L91:								#  nr=mr=1,kr=4
+	LD	a4,	1*SIZE(A)
+	LD	b4,	1*SIZE(B)
+	MADD	t11,t11,a0,b0
+	
+	LD	a2,	2*SIZE(A)
+	LD	b2,	2*SIZE(B)
+	MADD	t11,t11,a4,b4
+
+	LD	a6,	3*SIZE(A)
+	LD	b6,	3*SIZE(B)
+	MADD	t11,t11,a2,b2
+	
+	daddu	A,A,4*SIZE				#  A+=1(mr)*4(kr)*8Byte=32
+	daddu	B,B,4*SIZE				#  B+=1(nr)*4(kr)*8Byte=32
+
+	LD	a0,	0*SIZE(A)
+	LD	b0,	0*SIZE(B)
+	MADD	t11,t11,a6,b6
+	
+	daddiu	K,K,-1
+	bnez 	K,.L91
+	nop
+
+.L95:								#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2			
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L98
+	nop
+
+.L96:			
+	LD	a4,	1*SIZE(A)
+	LD	b4,	1*SIZE(B)
+	MADD	t11,t11,a0,b0
+	daddu	B,B,2*SIZE				#  B+=1(nr)*2(kr)*8Byte=16
+	daddu	A,A,2*SIZE				#  A+=1(mr)*2(kr)*8Byte=32
+
+	LD	b0,0(B)
+	LD	a0,0(A)
+	MADD	t11,t11,a4,b4
+	
+.L98:								#  kr=1
+#ifndef TRMMKERNEL
+	andi		K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L99				
+	LD	ALPHA,152($sp)				#  Get ALPHA
+
+	MADD	t11,t11,a0,b0
+
+
+.L99:								#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)					#  Fetch 16 C
+	MADD	t11,c11,t11,ALPHA
+	ST	t11,0(CO1)
+
+#else
+	MUL	t11, ALPHA, t11
+
+	ST	t11,  0 * SIZE(CO1)
+#endif
+
+
+.L999:							#  End
+	ld	$16,   0($sp)
+	ld	$17,   8($sp)
+	ld	$18,  16($sp)
+	ld	$19,  24($sp)
+	ld	$20,  32($sp)
+	ld	$21,  40($sp)
+	ld	$22,  48($sp)
+	LD	$f24, 56($sp)
+	LD	$f25, 64($sp)
+	LD	$f26, 72($sp)
+	LD	$f27, 80($sp)
+	LD	$f28, 88($sp)
+	ld	$23,  96($sp)
+	ld	$24, 104($sp)
+	ld	$25, 112($sp)
+	LD	$f20,120($sp)
+	LD	$f21,128($sp)
+	LD	$f22,136($sp)
+	LD	$f23,144($sp)
+
+	j	$31
+	daddiu	$sp, $sp, 160
+
+	EPILOGUE
diff --git a/kernel/mips64/sgemm_kernel_loongson3a.S b/kernel/mips64/sgemm_kernel_loongson3a.S
new file mode 100644
index 000000000..4a8c9b0e4
--- /dev/null
+++ b/kernel/mips64/sgemm_kernel_loongson3a.S
@@ -0,0 +1,2579 @@
+#define REALNAME ASMNAME
+#define ASSEMBLER
+#include "common.h"
+
+#define FETCH	ld
+#define gsLQC1(base,fq,ft,offset) .word(0x32<<26|base<<21|ft<<16|0x1<<15|offset<<6|0x1<<5|fq)
+#define gsSQC1(base,fq,ft,offset) .word(0x3A<<26|base<<21|ft<<16|0x1<<15|offset<<6|0x1<<5|fq)
+
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$8
+#define B	$9
+#define C	$10
+#define LDC	$11
+
+#define AO	$12
+#define BO	$13
+
+#define CO1	$14
+#define CO2	$15
+#define CO3	$16
+#define CO4	$17
+
+#define KCO	$18
+#define MCO	$19
+#define NCO	$20
+
+#define SPANB	$21
+#define PREB	$23
+#define PREA	$24
+#define SPANA	$25
+
+#define ALPHA	$f15
+
+#if defined(TRMMKERNEL)
+#define	OFFSET	$2
+#define	KK	$3
+#define	TEMP	$7
+#endif
+
+#define R8	8
+#define	R9	9
+#define R14	14
+#define R15	15
+#define R16	16
+#define R17 17
+
+#define	t11	$f30
+#define	t21	$f31
+#define	t31	$f28
+#define	t41	$f29
+
+#define	t12	$f26
+#define	t22	$f27
+#define	t32	$f24
+#define	t42	$f25
+
+#define	t13	$f22
+#define	t23	$f23
+#define	t33	$f20
+#define	t43	$f21
+
+#define	t14	$f18
+#define	t24	$f19
+#define	t34	$f16
+#define	t44	$f17
+
+#define	c11	$f0
+#define	c21	$f1
+#define	c31	$f2
+#define	c41	$f3
+
+#define	c12	$f4
+#define	c22	$f5
+#define	c32	$f6
+#define	c42	$f7
+
+#define	c13	$f8
+#define	c23	$f9
+#define	c33	$f10
+#define c43	$f11
+
+#define	c14	$f12
+#define	c24	$f13
+#define	c34	$f14
+#define	c44	$f0
+
+#define	a0	$f0
+#define	a1	$f1
+#define	a2	$f2
+#define	a3	$f3
+#define	a4	$f4
+#define	a5	$f5
+#define	a6	$f6
+#define	a7	$f7
+#define	b0	$f8
+#define	b1	$f9
+#define	b2	$f10
+#define b3	$f11
+#define	b4	$f12
+#define	b5	$f13
+#define	b6	$f14
+#define	b7	$f15
+
+#define F31 31
+#define F30 30
+#define F29 29
+#define F28 28
+#define F27 27
+#define F26 26
+#define F25 25
+#define F24 24 
+#define F23 23
+#define F22 22
+#define F21 21
+#define F20 20
+#define F19 19
+#define F18 18
+#define F17 17
+#define F16 16 
+#define F15 15
+#define F14 14
+#define F13 13
+#define F12 12
+#define F11 11
+#define F10 10
+#define F9 9
+#define F8 8
+#define F7 7
+#define F6 6
+#define F5 5
+#define F4 4 
+#define F3 3 
+#define F2 2 
+#define F1 1 
+#define F0 0
+
+	PROLOGUE
+	
+	daddiu	$sp, $sp, -160
+	sd	$16,   0($sp)
+	sd	$17,   8($sp)
+	sd	$18,  16($sp)
+	sd	$19,  24($sp)
+	sd	$20,  32($sp)
+	sd	$21,  40($sp)
+	sd	$22,  48($sp)
+	ST	$f24, 56($sp)
+	ST	$f25, 64($sp)
+	ST	$f26, 72($sp)
+	ST	$f27, 80($sp)
+	ST	$f28, 88($sp)
+	sd	$23,  96($sp)
+	sd	$24, 104($sp)
+	sd	$25, 112($sp)
+	ST	$f20,120($sp)
+	ST	$f21,128($sp)
+	ST	$f22,136($sp)
+	ST	$f23,144($sp)
+
+
+	.align	5					
+.L0_N4:									#  Loop N
+	ST	ALPHA,152($sp)					#  Backup	ALPHA
+	move	MCO,M						#  Backup	M
+
+	move	NCO,N						#  Backup	N
+	move	KCO,K						#  Backup	K
+
+	move	AO,A						#  Backup	A_addr
+	dsra	N,NCO,2						#  N=NCO/2
+	
+	dsll	LDC,LDC,BASE_SHIFT			#  LDC*8Byte
+	dsll	SPANB,KCO,2+BASE_SHIFT		#  SPANB=KC*4nr*8Byte=KC*2^5
+	
+#if defined(TRMMKERNEL)
+	LDARG	OFFSET,160($sp)				#	OFFSET is relate to the data part		
+#endif
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	neg		KK,OFFSET				
+#endif
+	
+	move	BO,B						#  Backup	B_addr
+	beq		N,$0,.L0_N2					#  N=0,NCO<4
+	dsll	SPANA,KCO,1+BASE_SHIFT		#  SPANA = KCO*2mr*8Byte
+
+.L0_N4_Lb:								#	mr=4,nr=4
+	move	CO1,C							
+	dsra	M,MCO,2						#  M=MCO/2
+	
+	move	A,AO						#  Reset A
+	daddu	CO2,C,LDC
+
+	daddu	PREB,BO,SPANB				#  PreB point next panelB
+	daddu	CO3,CO2,LDC
+
+	daddu	PREA,AO,SPANA
+	daddu	CO4,CO3,LDC
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+	move	KK,OFFSET					
+#endif
+	beqz	M,.L14_M2
+	daddu	C,CO4,LDC					#	move C to next panel Cj
+
+.L10:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO						#	(SIDE=L and UPLO=L) or (SIZE=R and UPLO=U)
+#else
+	dsll	K,KK,2 + BASE_SHIFT			#  KK is the length that needs to span to the data part
+	dsll	TEMP,KK,2 + BASE_SHIFT
+
+	daddu	A,A,K						#  move A B to data part
+	daddu	B,BO,TEMP
+#endif
+
+	MTC		$0,t11						# 	GEMM part	NR=4,MR=4
+	LD	a0,0(A)
+
+	MOV	t21,t11
+	MOV	t31,t11
+	LD	a1,1*SIZE(A)
+
+	MOV	t41,t11
+	MOV	t12,t11
+	LD	b0,0(B)
+	
+	MOV	t22,t11
+	MOV	t32,t11
+	LD	b1,1*SIZE(B)
+
+	MOV	t42,t11
+	LD	a2,2*SIZE(A)
+	
+	MOV	t13,t11
+	MOV	t23,t11
+	LD	b2,2*SIZE(B)
+	
+	MOV	t33,t11
+	MOV	t43,t11
+	LD	a3,3*SIZE(A)
+
+	MOV	t14,t11
+	MOV	t24,t11
+	LD	b3,3*SIZE(B)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP,KCO,KK					#  temp is the length of the data part
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 4					#	S=L,U=L	
+#else
+	daddiu	TEMP, KK, 4					#	S=R,U=U,for this two situation KK is the length of the data part
+#endif
+	dsra	K,TEMP,2					#  K=KCO/2
+	MOV	t34,t11
+	beqz	K,.L15
+	MOV	t44,t11
+
+#else							
+	move	B,BO						#	Reset B
+	MTC		$0,t11						# 	GEMM part	NR=4,MR=4
+	LD	a0,0(A)
+
+	MOV	t21,t11
+	MOV	t31,t11
+	LD	a1,1*SIZE(A)
+
+	MOV	t41,t11
+	MOV	t12,t11
+	LD	b0,0(B)
+	
+	MOV	t22,t11
+	MOV	t32,t11
+	LD	b1,1*SIZE(B)
+
+	MOV	t42,t11
+	dsra	K,KCO,2						#  K=KCO/2
+	LD	a2,2*SIZE(A)
+	
+	MOV	t13,t11
+	MOV	t23,t11
+	LD	b2,2*SIZE(B)
+	
+	MOV	t33,t11
+	MOV	t43,t11
+	LD	a3,3*SIZE(A)
+
+	MOV	t14,t11
+	MOV	t24,t11
+	LD	b3,3*SIZE(B)
+
+	MOV	t34,t11
+	beqz	K,.L15
+	MOV	t44,t11							#	clear 16 results registers
+#endif
+	
+	.align	5
+.L11:									#  kr=4
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	LD	a4,4*SIZE(A)
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+	LD	a5,5*SIZE(A)
+	
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	LD	b4,4*SIZE(B)
+	
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+	LD	b5,5*SIZE(B)
+	FETCH		$0,(PREB)
+	
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+	LD	a6,6*SIZE(A)
+	
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+	LD	b6,6*SIZE(B)
+	FETCH		$0,(PREA)
+	
+	MADD	t33,t33,a2,b2
+	MADD	t43,t43,a3,b2
+	LD	a7,7*SIZE(A)
+	
+	MADD	t34,t34,a2,b3
+	MADD	t44,t44,a3,b3
+	LD	b7,7*SIZE(B)
+
+.L12:
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+	LD	a0,8*SIZE(A)
+
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+	LD	a1,9*SIZE(A)
+
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+	LD	b0,8*SIZE(B)
+
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+	LD	b1,9*SIZE(B)
+
+	FETCH		$0,4*SIZE(PREB)
+	MADD	t13,t13,a4,b6
+	MADD	t23,t23,a5,b6
+	LD	a2,10*SIZE(A)
+
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+	LD	b2,10*SIZE(B)
+
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t33,t33,a6,b6
+	MADD	t43,t43,a7,b6
+	LD	a3,11*SIZE(A)
+
+	MADD	t34,t34,a6,b7
+	MADD	t44,t44,a7,b7
+	LD	b3,11*SIZE(B)
+
+.L13:
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	LD	a4,12*SIZE(A)
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+	LD	a5,13*SIZE(A)
+
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	LD	b4,12*SIZE(B)
+
+	FETCH		$0,8*SIZE(PREA)
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+	LD	b5,13*SIZE(B)
+
+	FETCH		$0,8*SIZE(PREB)
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+	LD	a6,14*SIZE(A)
+
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+	daddu	A,A,16*SIZE					#  4mr*4kr
+	LD	b6,14*SIZE(B)
+
+	MADD	t33,t33,a2,b2
+	MADD	t43,t43,a3,b2
+	daddu	B,B,16*SIZE					#	4nr*4kr
+	LD	a7,-1*SIZE(A)
+
+	MADD	t34,t34,a2,b3
+	MADD	t44,t44,a3,b3
+	LD	b7,-1*SIZE(B)
+
+.L14:
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+	LD	a0,0(A)
+
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+	LD	a1,1*SIZE(A)
+
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+	daddiu	K,K,-1
+	LD	b0,0(B)
+
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+	daddu	PREA,PREA,16*SIZE
+	LD	b1,1*SIZE(B)
+
+	FETCH		$0,12*SIZE(PREB)
+	MADD	t13,t13,a4,b6
+	MADD	t23,t23,a5,b6
+	LD	a2,2*SIZE(A)
+
+	FETCH		$0,-4*SIZE(PREA)
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+	LD	b2,2*SIZE(B)
+
+	MADD	t33,t33,a6,b6
+	MADD	t43,t43,a7,b6
+	daddu	PREB,PREB,16*SIZE
+	LD	a3,3*SIZE(A)
+
+	MADD	t34,t34,a6,b7
+	MADD	t44,t44,a7,b7
+	bnez 	K,.L11
+	LD	b3,3*SIZE(B)
+
+
+.L15:									#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP, 2
+#endif
+	beqz	K,.L18
+	nop
+
+.L16:			
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	LD	a4,4*SIZE(A)
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+	LD	a5,5*SIZE(A)
+
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	LD	b4,4*SIZE(B)
+
+	FETCH		$0,0(PREA)
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+	LD	b5,5*SIZE(B)
+
+	FETCH		$0,0(PREB)
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+	LD	a6,6*SIZE(A)
+
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+	daddu	A,A,8*SIZE					#	4mr*2kr
+	LD	b6,6*SIZE(B)
+
+	MADD	t33,t33,a2,b2
+	MADD	t43,t43,a3,b2
+	daddu	B,B,8*SIZE					#	4nr*2kr
+	LD	a7,-1*SIZE(A)
+
+	MADD	t34,t34,a2,b3
+	MADD	t44,t44,a3,b3
+	LD	b7,-1*SIZE(B)
+
+.L17:
+	MADD	t11,t11,a4,b4
+	MADD	t21,t21,a5,b4
+	LD	a0,0*SIZE(A)
+
+	MADD	t12,t12,a4,b5
+	MADD	t22,t22,a5,b5
+	LD	a1,1*SIZE(A)
+
+	MADD	t31,t31,a6,b4
+	MADD	t41,t41,a7,b4
+	LD	b0,0*SIZE(B)
+
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+	LD	b1,1*SIZE(B)
+
+	FETCH		$0,4*SIZE(PREB)
+	MADD	t13,t13,a4,b6
+	MADD	t23,t23,a5,b6
+	LD	a2,2*SIZE(A)
+
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+	LD	b2,2*SIZE(B)
+
+	MADD	t33,t33,a6,b6
+	MADD	t43,t43,a7,b6
+	daddu	PREA,PREA,8*SIZE
+	LD	a3,3*SIZE(A)
+
+	MADD	t34,t34,a6,b7
+	MADD	t44,t44,a7,b7
+	daddu	PREB,PREB,8*SIZE
+	LD	b3,3*SIZE(B)
+
+	
+.L18:									#	kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L19				  
+	LD	ALPHA,152($sp)					#  Get ALPHA
+	
+	FETCH		$0,0(PREB)
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,4*SIZE					#  	4mr*kr
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+	daddu	B,B,4*SIZE					#	4nr*kr
+
+	FETCH		$0,0(PREA)
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	daddu	PREB,PREB,4*SIZE
+
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+	daddu	PREA,PREA,4*SIZE
+
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+
+	MADD	t33,t33,a2,b2
+	MADD	t43,t43,a3,b2
+
+	MADD	t34,t34,a2,b3
+	MADD	t44,t44,a3,b3
+
+.L19:									#  Write Back to C
+#ifndef TRMMKERNEL				
+	LD	c11,0(CO1)						#  GEMM write part 
+	LD	c21,1*SIZE(CO1)					#  get 16 C
+	LD	c31,2*SIZE(CO1)
+	LD	c41,3*SIZE(CO1)
+
+	LD	c12,0(CO2)
+	MADD	t11,c11,t11,ALPHA
+	LD	c22,1*SIZE(CO2)
+	MADD	t21,c21,t21,ALPHA
+	LD	c32,2*SIZE(CO2)
+	MADD	t31,c31,t31,ALPHA
+	LD	c42,3*SIZE(CO2)
+	MADD	t41,c41,t41,ALPHA
+
+	LD	c13,0(CO3)
+	MADD	t12,c12,t12,ALPHA
+	LD	c23,1*SIZE(CO3)
+	MADD	t22,c22,t22,ALPHA
+	LD	c33,2*SIZE(CO3)
+	MADD	t32,c32,t32,ALPHA
+	LD	c43,3*SIZE(CO3)
+	MADD	t42,c42,t42,ALPHA
+
+	LD	c14,0(CO4)
+	MADD	t13,c13,t13,ALPHA
+	LD	c24,1*SIZE(CO4)
+	MADD	t23,c23,t23,ALPHA
+	LD	c34,2*SIZE(CO4)
+	MADD	t33,c33,t33,ALPHA
+	LD	c44,3*SIZE(CO4)
+	MADD	t43,c43,t43,ALPHA
+
+	ST	t11,0(CO1)
+	MADD	t14,c14,t14,ALPHA
+	ST	t21,1*SIZE(CO1)
+	MADD	t24,c24,t24,ALPHA
+	ST	t31,2*SIZE(CO1)
+	MADD	t34,c34,t34,ALPHA
+	ST	t41,3*SIZE(CO1)
+	MADD	t44,c44,t44,ALPHA
+	daddiu	M,M,-1						#  M--
+
+	ST	t12,0(CO2)
+	ST	t22,1*SIZE(CO2)
+	ST	t32,2*SIZE(CO2)
+	ST	t42,3*SIZE(CO2)
+
+	ST	t13,0(CO3)
+	ST	t23,1*SIZE(CO3)
+	ST	t33,2*SIZE(CO3)
+	ST	t43,3*SIZE(CO3)
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,4*SIZE(CO2)
+	FETCH	$0,4*SIZE(CO3)
+	FETCH	$0,4*SIZE(CO4)
+
+	FETCH	$0,8*SIZE(CO1)
+	FETCH	$0,8*SIZE(CO2)
+	FETCH	$0,8*SIZE(CO3)
+	FETCH	$0,8*SIZE(CO4)
+
+	ST	t14,0(CO4)
+	daddu	CO1,CO1,4*SIZE				#  COi += 4
+	ST	t24,1*SIZE(CO4)
+	daddu	CO2,CO2,4*SIZE
+	ST	t34,2*SIZE(CO4)
+	daddu	CO3,CO3,4*SIZE
+	ST	t44,3*SIZE(CO4)
+	daddu	PREB,BO,SPANB
+	
+	bnez	M,.L10				
+	daddu	CO4,CO4,4*SIZE
+
+#else							
+	MUL	t11, ALPHA, t11					#	TRMM write back part
+	MUL	t21, ALPHA, t21
+	MUL	t31, ALPHA, t31
+	MUL	t41, ALPHA, t41
+
+	ST	t11, 0 * SIZE(CO1)
+	MUL	t12, ALPHA, t12
+	ST	t21, 1 * SIZE(CO1)
+	MUL	t22, ALPHA, t22
+	ST	t31, 2 * SIZE(CO1)
+	MUL	t32, ALPHA, t32
+	ST	t41, 3 * SIZE(CO1)
+	MUL	t42, ALPHA, t42
+
+	ST	t12, 0 * SIZE(CO2)
+	MUL	t13, ALPHA, t13
+	ST	t22, 1 * SIZE(CO2)
+	MUL	t23, ALPHA, t23
+	ST	t32, 2 * SIZE(CO2)
+	MUL	t33, ALPHA, t33
+	ST	t42, 3 * SIZE(CO2)
+	MUL	t43, ALPHA, t43
+
+	ST	t13, 0 * SIZE(CO3)
+	MUL	t14, ALPHA, t14
+	ST	t23, 1 * SIZE(CO3)
+	MUL	t24, ALPHA, t24
+	ST	t33, 2 * SIZE(CO3)
+	MUL	t34, ALPHA, t34
+	ST	t43, 3 * SIZE(CO3)
+	MUL	t44, ALPHA, t44
+
+	ST	t14, 0 * SIZE(CO4)
+	daddiu	M,M,-1						#  M--
+	ST	t24, 1 * SIZE(CO4)
+	ST	t34, 2 * SIZE(CO4)
+	ST	t44, 3 * SIZE(CO4)
+	daddiu	CO1,CO1, 4 * SIZE
+	daddiu	CO2,CO2, 4 * SIZE
+	daddiu	CO3,CO3, 4 * SIZE
+	daddiu	CO4,CO4, 4 * SIZE	
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,4*SIZE(CO2)
+	FETCH	$0,4*SIZE(CO3)
+	FETCH	$0,4*SIZE(CO4)
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+	FETCH	$0,0(CO3)
+	FETCH	$0,0(CO4)
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP,KCO,KK								
+#ifdef LEFT
+	daddiu	TEMP,TEMP, -4
+#else
+	daddiu	TEMP,TEMP, -4
+#endif
+	dsll	K,TEMP,2 + BASE_SHIFT
+	dsll	TEMP,TEMP,2 + BASE_SHIFT
+	daddu	A,A,K						# 	mov A to the end of panel Ai
+	daddu	B,B,TEMP					# 	mov B to the end of panel Bj
+#endif
+
+#ifdef LEFT										
+	daddiu	KK, KK,4
+#endif
+	bnez	M,.L10					
+	nop
+#endif
+
+
+	.align 3
+.L14_M2:
+	andi	M, MCO, 2					#	nr=4,mr=2
+	beqz	M,.L14_M1			
+	nop
+
+.L20:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO						#	Reset B
+#else
+	dsll	K,KK,1 + BASE_SHIFT			#	mr=2	
+	dsll	TEMP,KK,2 + BASE_SHIFT		#	nr=4
+	daddu	A,A,K
+	daddu	B,BO,TEMP
+#endif
+
+	LD	a0,0*SIZE(A)
+	MTC		$0,t11
+	LD	a1,1*SIZE(A)
+	
+	MOV	t21,t11
+	LD	b0,0*SIZE(B)
+	MOV	t12,t11
+	LD	b1,1*SIZE(B)
+
+	MOV	t22,t11
+	LD	b2,2*SIZE(B)
+
+	MOV	t13,t11
+	MOV	t23,t11
+	LD	b3,3*SIZE(B)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP,KCO,KK
+#elif defined(LEFT)
+	daddiu	TEMP,KK,2					#	left part,controlled by mr, mr=2
+#else
+	daddiu	TEMP,KK,4					#  	right part,controlled by nr,nr=4
+#endif
+	dsra	K,TEMP,2
+	MOV	t14,t11
+	beqz	K,.L25
+	MOV	t24,t11							#	clear 2*4=8 results registers
+
+#else
+	move	B,BO						#	Reset B 
+	LD	a0,0*SIZE(A)
+	MTC		$0,t11
+	LD	a1,1*SIZE(A)
+	
+	MOV	t21,t11
+	LD	b0,0*SIZE(B)
+	MOV	t12,t11
+	LD	b1,1*SIZE(B)
+
+	MOV	t22,t11
+	dsra	K,KCO,2				
+	LD	b2,2*SIZE(B)
+
+	MOV	t13,t11
+	MOV	t23,t11
+	LD	b3,3*SIZE(B)
+
+	MOV	t14,t11
+	beqz	K,.L25
+	MOV	t24,t11
+
+#endif
+
+.L21:									#  nr=4,mr=2,kr=4
+	MADD	t11,t11,a0,b0
+	LD	a4,2*SIZE(A)
+	MADD	t21,t21,a1,b0
+	LD	a5,3*SIZE(A)
+
+	MADD	t12,t12,a0,b1
+	LD	b4,4*SIZE(B)
+	MADD	t22,t22,a1,b1
+	LD	b5,5*SIZE(B)
+
+	MADD	t13,t13,a0,b2
+	LD	b6,6*SIZE(B)
+	MADD	t23,t23,a1,b2
+	LD	b7,7*SIZE(B)
+
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+	
+	MADD	t11,t11,a4,b4
+	LD	a2,4*SIZE(A)
+	MADD	t21,t21,a5,b4
+	LD	a3,5*SIZE(A)
+
+	MADD	t12,t12,a4,b5
+	LD	b0,8*SIZE(B)
+	MADD	t22,t22,a5,b5
+	LD	b1,9*SIZE(B)
+
+	MADD	t13,t13,a4,b6
+	LD	b2,10*SIZE(B)
+	MADD	t23,t23,a5,b6
+	LD	b3,11*SIZE(B)
+
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+	daddiu	K,K,-1
+
+	MADD	t11,t11,a2,b0
+	LD	a6,6*SIZE(A)
+	MADD	t21,t21,a3,b0
+	LD	a7,7*SIZE(A)
+
+	MADD	t12,t12,a2,b1
+	LD	b4,12*SIZE(B)
+	MADD	t22,t22,a3,b1
+	LD	b5,13*SIZE(B)
+
+	MADD	t13,t13,a2,b2
+	LD	b6,14*SIZE(B)
+	MADD	t23,t23,a3,b2
+	LD	b7,15*SIZE(B)
+
+	MADD	t14,t14,a2,b3
+	MADD	t24,t24,a3,b3
+	daddu	A,A,8*SIZE					#  2mr*4kr
+	daddu	B,B,16*SIZE					#	4nr*4kr
+
+	MADD	t11,t11,a6,b4
+	LD	a0,0*SIZE(A)
+	MADD	t21,t21,a7,b4
+	LD	a1,1*SIZE(A)
+
+	MADD	t12,t12,a6,b5
+	LD	b0,0*SIZE(B)
+	MADD	t22,t22,a7,b5
+	LD	b1,1*SIZE(B)
+
+	MADD	t13,t13,a6,b6
+	LD	b2,2*SIZE(B)
+	MADD	t23,t23,a7,b6
+	LD	b3,3*SIZE(B)
+
+	MADD	t14,t14,a6,b7
+	bnez 	K,.L21
+	MADD	t24,t24,a7,b7
+
+
+.L25:										
+#ifndef TRMMKERNEL
+	andi	K,KCO,2						#	kr=2
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L28
+	nop
+
+.L26:			
+	MADD	t11,t11,a0,b0
+	LD	a4,2*SIZE(A)
+	MADD	t21,t21,a1,b0
+	LD	a5,3*SIZE(A)
+
+	MADD	t12,t12,a0,b1
+	LD	b4,4*SIZE(B)
+	MADD	t22,t22,a1,b1
+	LD	b5,5*SIZE(B)
+
+	MADD	t13,t13,a0,b2
+	LD	b6,6*SIZE(B)
+	MADD	t23,t23,a1,b2
+	LD	b7,7*SIZE(B)
+	
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+	daddu	A,A,4*SIZE					#  	2mr*2kr
+	daddu	B,B,8*SIZE					#	4nr*2kr
+
+.L27:
+	MADD	t11,t11,a4,b4
+	LD	a0,0*SIZE(A)
+	MADD	t21,t21,a5,b4
+	LD	a1,1*SIZE(A)
+
+	MADD	t12,t12,a4,b5
+	LD	b0,0*SIZE(B)
+	MADD	t22,t22,a5,b5
+	LD	b1,1*SIZE(B)
+
+	MADD	t13,t13,a4,b6
+	LD	b2,2*SIZE(B)
+	MADD	t23,t23,a5,b6
+	LD	b3,3*SIZE(B)
+
+	MADD	t14,t14,a4,b7
+	MADD	t24,t24,a5,b7
+
+	
+.L28:									#	kr=1	
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L29				  
+	LD	ALPHA,152($sp)					#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,2*SIZE					#  2mr*kr
+	daddu	B,B,4*SIZE					#  4nr*kr
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+	MADD	t13,t13,a0,b2
+	MADD	t23,t23,a1,b2
+
+	MADD	t14,t14,a0,b3
+	MADD	t24,t24,a1,b3
+
+.L29:									#  Write Back to C
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)						#	GEMM write back part
+	LD	c21,1*SIZE(CO1)			
+
+	LD	c12,0(CO2)
+	LD	c22,1*SIZE(CO2)
+	
+	LD	c13,0(CO3)
+	MADD	t11,c11,t11,ALPHA
+	LD	c23,1*SIZE(CO3)
+	MADD	t21,c21,t21,ALPHA
+
+	LD	c14,0(CO4)
+	MADD	t12,c12,t12,ALPHA
+	LD	c24,1*SIZE(CO4)
+	MADD	t22,c22,t22,ALPHA
+
+	ST	t11,0(CO1)
+	MADD	t13,c13,t13,ALPHA
+	ST	t21,1*SIZE(CO1)
+	MADD	t23,c23,t23,ALPHA
+
+	ST	t12,0(CO2)
+	MADD	t14,c14,t14,ALPHA
+	ST	t22,1*SIZE(CO2)
+	MADD	t24,c24,t24,ALPHA
+
+	ST	t13,0(CO3)
+	daddu	CO1,CO1,2*SIZE				#  COi += 2
+	ST	t23,1*SIZE(CO3)
+	daddu	CO2,CO2,2*SIZE
+
+	ST	t14,0(CO4)
+	daddu	CO3,CO3,2*SIZE
+	ST	t24,1*SIZE(CO4)
+	daddu	CO4,CO4,2*SIZE
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+	FETCH	$0,0(CO3)
+	FETCH	$0,0(CO4)
+
+#else
+	MUL	t11, ALPHA, t11					#	TRMM write back part
+	MUL	t21, ALPHA, t21
+	
+	ST	t11, 0 * SIZE(CO1)
+	MUL	t12, ALPHA, t12
+	ST	t21, 1 * SIZE(CO1)
+	MUL	t22, ALPHA, t22
+	
+	ST	t12, 0 * SIZE(CO2)
+	MUL	t13, ALPHA, t13
+	ST	t22, 1 * SIZE(CO2)
+	MUL	t23, ALPHA, t23
+	
+	ST	t13, 0 * SIZE(CO3)
+	MUL	t14, ALPHA, t14
+	ST	t23, 1 * SIZE(CO3)
+	MUL	t24, ALPHA, t24
+	
+	ST	t14, 0 * SIZE(CO4)
+	ST	t24, 1 * SIZE(CO4)
+	
+	daddiu	CO1,CO1, 2 * SIZE
+	daddiu	CO2,CO2, 2 * SIZE
+	daddiu	CO3,CO3, 2 * SIZE
+	daddiu	CO4,CO4, 2 * SIZE
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+	FETCH	$0,0(CO3)
+	FETCH	$0,0(CO4)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP,KCO,KK
+#ifdef LEFT
+	daddiu	TEMP,TEMP,-2
+#else
+	daddiu	TEMP,TEMP,-4
+#endif
+	dsll	K,TEMP,1 + BASE_SHIFT
+	dsll	TEMP,TEMP,2 + BASE_SHIFT
+
+	daddu	A,A,K						#	move A to next panel Ai
+	daddu	B,B,TEMP					#	move B to next panel Bj
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+
+
+	.align 3
+.L14_M1:
+	andi	M,MCO,1						#	mr=1	
+	beqz	M,.L0_N4_Loop				#  	M = 0, finishing one panel Bj
+	nop
+
+.L30:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO						#	Reset B
+#else
+	dsll	K,KK, BASE_SHIFT
+	dsll	TEMP,KK,2 + BASE_SHIFT
+
+	daddu	A,A,K
+	daddu	B,BO,TEMP
+#endif
+
+	LD	a0, 0 * SIZE(A)					#	a0
+
+	MTC		$0,t11
+	LD	b0,0*SIZE(B)
+	
+	MOV	t12,t11
+	LD	b1,1*SIZE(B)
+
+	MOV	t13,t11
+	LD	b2,2*SIZE(B)
+	
+	MOV	t14,t11
+	LD	b3,3*SIZE(B)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1
+#else
+	daddiu	TEMP, KK, 4
+#endif
+	dsra	K,TEMP, 2
+	nop
+	beqz	K,.L35
+	nop
+								
+#else							
+	move	B,BO						#	Reset B, GEMM part
+	dsra	K,KCO,2						#  	K=KCO/2
+	LD	a0, 0 * SIZE(A)					#	a0
+
+	MTC		$0,t11
+	LD	b0,0*SIZE(B)
+	
+	MOV	t12,t11
+	LD	b1,1*SIZE(B)
+
+	MOV	t13,t11
+	LD	b2,2*SIZE(B)
+	
+	MOV	t14,t11
+	beqz	K,.L35
+	LD	b3,3*SIZE(B)
+
+#endif
+
+.L31:									#	nr=4,mr=1,kr=4	
+	LD	a1,	1*SIZE(A)					#	load a1
+	MADD	t11,t11,a0,b0
+	
+	LD	b4,4*SIZE(B)
+	LD	b5,5*SIZE(B)
+	MADD	t12,t12,a0,b1
+	
+	LD	b6,6*SIZE(B)
+	LD	b7,7*SIZE(B)
+	MADD	t13,t13,a0,b2
+	MADD	t14,t14,a0,b3
+
+	LD	a2,	2*SIZE(A)					#	a2
+	MADD	t11,t11,a1,b4
+	
+	LD	b0,8*SIZE(B)
+	LD	b1,9*SIZE(B)
+	MADD	t12,t12,a1,b5
+	
+	LD	b2,10*SIZE(B)
+	LD	b3,11*SIZE(B)
+	MADD	t13,t13,a1,b6
+	MADD	t14,t14,a1,b7
+
+	LD	a3,	3*SIZE(A)					#	a3
+	MADD	t11,t11,a2,b0
+	daddiu	K,K,-1
+	
+	LD	b4,12*SIZE(B)
+	LD	b5,13*SIZE(B)
+	MADD	t12,t12,a2,b1
+	daddu	A,A,4*SIZE					#	1mr*4kr
+	
+	LD	b6,14*SIZE(B)
+	LD	b7,15*SIZE(B)
+	MADD	t13,t13,a2,b2
+	MADD	t14,t14,a2,b3
+
+	LD	a0,	0*SIZE(A)					#	a0
+	daddu	B,B,16*SIZE					#	4nr*4kr
+	MADD	t11,t11,a3,b4
+	
+	LD	b0,0*SIZE(B)
+	MADD	t12,t12,a3,b5
+	LD	b1,1*SIZE(B)
+	MADD	t13,t13,a3,b6
+
+	LD	b2,2*SIZE(B)
+	MADD	t14,t14,a3,b7
+	bnez 	K,.L31
+	LD	b3,3*SIZE(B)
+
+
+.L35:									#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2			
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L38
+	nop
+
+.L36:			
+	LD	a1,1*SIZE(A)					#	load a1
+	MADD	t11,t11,a0,b0
+
+	LD	b4,4*SIZE(B)
+	LD	b5,5*SIZE(B)
+	MADD	t12,t12,a0,b1
+	daddu	A,A,2*SIZE					#  	mr*2kr
+	
+	LD	b6,6*SIZE(B)
+	MADD	t13,t13,a0,b2
+	
+	LD	b7,7*SIZE(B)
+	MADD	t14,t14,a0,b3
+	daddu	B,B,8*SIZE					#	4nr*2kr
+
+
+.L37:
+	LD	a0,0(A)
+	MADD	t11,t11,a1,b4
+	
+	LD	b0,0*SIZE(B)
+	LD	b1,1*SIZE(B)
+	MADD	t12,t12,a1,b5
+	
+	LD	b2,2*SIZE(B)
+	LD	b3,3*SIZE(B)
+	MADD	t13,t13,a1,b6
+	MADD	t14,t14,a1,b7
+	
+	
+.L38:									#  	kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L39				
+	LD	ALPHA,152($sp)					#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t12,t12,a0,b1
+	daddu	A,A,1*SIZE				
+	daddu	B,B,4*SIZE
+	
+	MADD	t13,t13,a0,b2
+	MADD	t14,t14,a0,b3
+
+.L39:									#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)			
+	LD	c12,0(CO2)
+	LD	c13,0(CO3)
+	LD	c14,0(CO4)
+	
+	MADD	t11,c11,t11,ALPHA
+	MADD	t12,c12,t12,ALPHA
+	MADD	t13,c13,t13,ALPHA
+	MADD	t14,c14,t14,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t12,0(CO2)
+	ST	t13,0(CO3)
+	ST	t14,0(CO4)
+#else
+	MUL	t11, ALPHA, t11
+	MUL	t12, ALPHA, t12
+	MUL	t13, ALPHA, t13
+	MUL	t14, ALPHA, t14
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -1
+#else
+	daddiu	TEMP, TEMP, -4
+#endif
+
+	dsll	K,TEMP, BASE_SHIFT
+	dsll	TEMP,TEMP, 2 + BASE_SHIFT
+
+	daddu	A,A,K
+	daddu	B,B,TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 1
+#endif
+#endif
+
+
+	.align	3
+.L0_N4_Loop:								#	mc finished
+	daddiu	N,N,-1							#  N--
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	daddiu	KK, KK,4 
+#endif
+	bnez	N,.L0_N4_Lb			
+	move	BO,B							#  Set BO point to next panel Bj
+
+	.align	5					
+.L0_N2:
+	andi	N,NCO,2							#  	nr = 2
+	beqz	N,.L0_N1		
+	nop
+
+.L0_N2_Lb:
+	move	CO1,C					
+	daddu	CO2,C,LDC
+
+	dsra	M,MCO,2				
+	move	A,AO							#  Reset A
+
+	daddu	PREA,AO,SPANA
+	daddu	C,CO2,LDC
+
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	move	KK, OFFSET
+#endif
+	beqz	M,.L12_M2
+	nop
+
+.L40:						
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO							#	Reset B
+#else
+	dsll	K,KK, 2 + BASE_SHIFT
+	dsll	TEMP, KK,1 + BASE_SHIFT	
+
+	daddu	A,A,K
+	daddu	B,BO,TEMP
+#endif
+	LD	a0,0*SIZE(A)
+	MTC		$0,t11							#  	gemm part
+	LD	a1,1*SIZE(A)
+
+	MOV	t21,t11
+	LD	b0,0*SIZE(B)
+	MOV	t31,t11
+	LD	b1,1*SIZE(B)
+
+	MOV	t41,t11
+	LD	a2,2*SIZE(A)
+	LD	a3,3*SIZE(A)
+	
+	MOV	t12,t11
+	MOV	t22,t11
+	
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP,KCO,KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 4
+#else
+	daddiu	TEMP, KK, 2
+#endif
+	dsra	K,TEMP,2				
+	MOV	t32,t11
+	beqz	K,.L45
+	MOV	t42,t11
+
+#else
+	move	B,BO							#	Reset B
+	LD	a0,0*SIZE(A)
+	MTC		$0,t11							#  	gemm part
+	LD	a1,1*SIZE(A)
+
+	MOV	t21,t11
+	LD	b0,0*SIZE(B)
+	MOV	t31,t11
+	LD	b1,1*SIZE(B)
+
+	MOV	t41,t11
+	LD	a2,2*SIZE(A)
+	dsra	K,KCO,2							#	K=KCO/2
+	LD	a3,3*SIZE(A)
+	
+	MOV	t12,t11
+	MOV	t22,t11
+	
+	MOV	t32,t11
+	beqz	K,.L45
+	MOV	t42,t11
+
+#endif
+
+.L41:										#  	nr=2,mr=kr=4
+	MADD	t11,t11,a0,b0
+	LD	a4,4*SIZE(A)
+	MADD	t21,t21,a1,b0
+	LD	a5,5*SIZE(A)
+
+	MADD	t12,t12,a0,b1
+	LD	b4,2*SIZE(B)
+	MADD	t22,t22,a1,b1
+	LD	b5,3*SIZE(B)
+
+	MADD	t31,t31,a2,b0
+	LD	a6,6*SIZE(A)
+	MADD	t41,t41,a3,b0
+	LD	a7,7*SIZE(A)
+
+	FETCH		$0,(PREA)
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+
+.L42:
+	MADD	t11,t11,a4,b4
+	LD	a0,8*SIZE(A)
+	MADD	t21,t21,a5,b4
+	LD	a1,9*SIZE(A)
+
+	MADD	t12,t12,a4,b5
+	LD	b2,4*SIZE(B)
+	MADD	t22,t22,a5,b5
+	LD	b3,5*SIZE(B)
+
+	MADD	t31,t31,a6,b4
+	LD	a2,10*SIZE(A)
+	MADD	t41,t41,a7,b4
+	LD	a3,11*SIZE(A)
+
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+
+.L43:
+	MADD	t11,t11,a0,b2
+	LD	a4,12*SIZE(A)
+	MADD	t21,t21,a1,b2
+	LD	a5,13*SIZE(A)
+
+	MADD	t12,t12,a0,b3
+	LD	b6,6*SIZE(B)
+	MADD	t22,t22,a1,b3
+	LD	b7,7*SIZE(B)
+
+	MADD	t31,t31,a2,b2
+	LD	a6,14*SIZE(A)
+	MADD	t41,t41,a3,b2
+	LD	a7,15*SIZE(A)
+
+	FETCH		$0,8*SIZE(PREA)
+	MADD	t32,t32,a2,b3
+	MADD	t42,t42,a3,b3
+	
+	daddu	A,A,16*SIZE						#	4mr*4kr
+	daddu	B,B,8*SIZE						#	2nr*4kr	
+
+.L44:
+	MADD	t11,t11,a4,b6
+	LD	a0,0*SIZE(A)
+	MADD	t21,t21,a5,b6
+	LD	a1,1*SIZE(A)
+
+
+	MADD	t12,t12,a4,b7
+	LD	b0,0*SIZE(B)
+	MADD	t22,t22,a5,b7
+	LD	b1,1*SIZE(B)
+
+	daddiu	K,K,-1
+	daddu	PREA,PREA,16*SIZE
+
+	MADD	t31,t31,a6,b6
+	LD	a2,2*SIZE(A)
+	MADD	t41,t41,a7,b6
+	LD	a3,3*SIZE(A)
+
+	FETCH		$0,-4*SIZE(PREA)
+	MADD	t32,t32,a6,b7
+	bnez 	K,.L41
+	MADD	t42,t42,a7,b7
+
+
+.L45:										#  	kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L48
+	nop
+
+.L46:			
+	MADD	t11,t11,a0,b0
+	LD	a4,4*SIZE(A)
+	MADD	t21,t21,a1,b0
+	LD	a5,5*SIZE(A)
+
+	MADD	t12,t12,a0,b1
+	LD	b4,2*SIZE(B)
+	MADD	t22,t22,a1,b1
+	LD	b5,3*SIZE(B)
+
+	MADD	t31,t31,a2,b0
+	LD	a6,6*SIZE(A)
+	MADD	t41,t41,a3,b0
+	LD	a7,7*SIZE(A)
+
+	FETCH		$0,0(PREA)
+	MADD	t32,t32,a2,b1
+	daddu	B,B,4*SIZE						#  B+=2(nr)*2(kr)*8Byte=32
+	
+	MADD	t42,t42,a3,b1
+	daddu	A,A,8*SIZE						#  A+=4(mr)*2(kr)*8Byte=8*SIZE
+
+.L47:
+	MADD	t11,t11,a4,b4
+	LD	a0,0*SIZE(A)
+	MADD	t21,t21,a5,b4
+	LD	a1,1*SIZE(A)
+
+	MADD	t12,t12,a4,b5
+	LD	b0,0*SIZE(B)
+	MADD	t22,t22,a5,b5
+	LD	b1,1*SIZE(B)
+
+	MADD	t31,t31,a6,b4
+	LD	a2,2*SIZE(A)
+	MADD	t41,t41,a7,b4
+	LD	a3,3*SIZE(A)
+
+	FETCH		$0,4*SIZE(PREA)
+	MADD	t32,t32,a6,b5
+	MADD	t42,t42,a7,b5
+	daddu	PREA,PREA,8*SIZE
+
+
+	
+.L48:										#	 kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L49				 
+	LD	ALPHA,152($sp)						#  Get ALPHA
+	
+	FETCH		$0,0(PREA)
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,4*SIZE						#  A+=4(mr)*1(kr)*8Byte=32
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+	daddu	B,B,2*SIZE
+	daddu	PREA,PREA,4*SIZE
+
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+
+	MADD	t32,t32,a2,b1
+	MADD	t42,t42,a3,b1
+
+.L49:										#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)							#  gemm write back part Fetch 16 C
+	LD	c21,1*SIZE(CO1)			
+	LD	c31,2*SIZE(CO1)
+	LD	c41,3*SIZE(CO1)
+
+	LD	c12,0(CO2)
+	MADD	t11,c11,t11,ALPHA
+	LD	c22,1*SIZE(CO2)
+	MADD	t21,c21,t21,ALPHA
+	LD	c32,2*SIZE(CO2)
+	MADD	t31,c31,t31,ALPHA
+	LD	c42,3*SIZE(CO2)
+	MADD	t41,c41,t41,ALPHA
+
+	ST	t11,0(CO1)
+	MADD	t12,c12,t12,ALPHA
+	ST	t21,1*SIZE(CO1)
+	MADD	t22,c22,t22,ALPHA
+	ST	t31,2*SIZE(CO1)
+	MADD	t32,c32,t32,ALPHA
+	ST	t41,3*SIZE(CO1)
+	MADD	t42,c42,t42,ALPHA
+	daddiu	M,M,-1				
+
+	ST	t12,0(CO2)
+	ST	t22,1*SIZE(CO2)
+	ST	t32,2*SIZE(CO2)
+	ST	t42,3*SIZE(CO2)
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,4*SIZE(CO2)
+	FETCH	$0,8*SIZE(CO1)
+	FETCH	$0,8*SIZE(CO2)
+
+	daddu	CO1,CO1,4*SIZE			
+	bnez	M,.L40				
+	daddu	CO2,CO2,4*SIZE
+
+#else
+	MUL	t11, ALPHA, t11
+	MUL	t21, ALPHA, t21
+	MUL	t31, ALPHA, t31
+	MUL	t41, ALPHA, t41
+	
+	MUL	t12, ALPHA, t12
+	ST	t11, 0 * SIZE(CO1)
+	MUL	t22, ALPHA, t22
+	ST	t21, 1 * SIZE(CO1)
+	MUL	t32, ALPHA, t32
+	ST	t31, 2 * SIZE(CO1)
+	MUL	t42, ALPHA, t42
+	ST	t41, 3 * SIZE(CO1)
+	
+	ST	t12, 0 * SIZE(CO2)
+	daddiu	M,M,-1
+	ST	t22, 1 * SIZE(CO2)
+	ST	t32, 2 * SIZE(CO2)
+	ST	t42, 3 * SIZE(CO2)
+	
+	daddiu	CO1,CO1, 4*SIZE
+	daddiu	CO2,CO2, 4*SIZE
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+	FETCH	$0,4(CO1)
+	FETCH	$0,4(CO2)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) ||(!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -4
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+	dsll	K,TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+
+	daddu	A,A,K
+	daddu	B,B,TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 4
+#endif
+	bnez	M,.L40
+	nop
+#endif
+
+
+	.align 3
+.L12_M2:
+	andi	M,MCO,2						#  	mr = 2
+	beqz	M,.L12_M1			
+	nop
+
+.L50:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO
+#else
+	dsll	K,    KK, 1 + BASE_SHIFT	#mr=2
+	dsll	TEMP, KK, 1 + BASE_SHIFT	#nr=2
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+	LD	a0,0*SIZE(A)
+	LD	a1,1*SIZE(A)
+
+	MTC		$0,t11
+	LD	b0,0*SIZE(B)
+	MOV	t21,t11
+	LD	b1,1*SIZE(B)
+	
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 2
+#else
+	daddiu	TEMP, KK, 2
+#endif
+	dsra	K,TEMP,2			
+	MOV	t12,t11
+	beqz	K,.L55
+	MOV	t22,t11
+
+#else
+	move	B,BO
+	LD	a0,0*SIZE(A)
+	dsra	K,KCO,2						#  K=KCO/2
+	LD	a1,1*SIZE(A)
+
+	MTC		$0,t11
+	LD	b0,0*SIZE(B)
+	MOV	t21,t11
+	LD	b1,1*SIZE(B)
+	
+	MOV	t12,t11
+	beqz	K,.L55
+	MOV	t22,t11
+
+#endif
+
+.L51:									#  nr=2 mr=2,kr=4
+	MADD	t11,t11,a0,b0
+	LD	a4,2*SIZE(A)
+	MADD	t21,t21,a1,b0
+	LD	b4,2*SIZE(B)
+
+	MADD	t12,t12,a0,b1
+	LD	a5,3*SIZE(A)
+	MADD	t22,t22,a1,b1
+	LD	b5,3*SIZE(B)
+
+	MADD	t11,t11,a4,b4
+	LD	a2,4*SIZE(A)
+	MADD	t21,t21,a5,b4
+	LD	b2,4*SIZE(B)
+
+	MADD	t12,t12,a4,b5
+	LD	a3,5*SIZE(A)
+	MADD	t22,t22,a5,b5
+	daddiu	K,K,-1
+	LD	b3,5*SIZE(B)
+
+	MADD	t11,t11,a2,b2
+	LD	a6,6*SIZE(A)
+	MADD	t21,t21,a3,b2
+	daddu	A,A,8*SIZE					#  A+=2(mr)*4(kr)*8Byte=8*SIZE
+	LD	b6,6*SIZE(B)
+
+	MADD	t12,t12,a2,b3
+	daddu	B,B,8*SIZE					#  B+=2(nr)*4(kr)*8Byte=16*SIZE
+	LD	a7,-1*SIZE(A)
+	MADD	t22,t22,a3,b3
+	LD	b7,-1*SIZE(B)
+
+	MADD	t11,t11,a6,b6
+	LD	a0,0*SIZE(A)
+	MADD	t21,t21,a7,b6
+	LD	b0,0*SIZE(B)
+
+	MADD	t12,t12,a6,b7
+	LD	a1,1*SIZE(A)
+
+	MADD	t22,t22,a7,b7
+	bnez 	K,.L51
+	LD	b1,1*SIZE(B)
+
+
+.L55:									#  	kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L58
+	nop
+
+.L56:			
+	MADD	t11,t11,a0,b0
+	LD	a4,2*SIZE(A)
+	MADD	t21,t21,a1,b0
+	daddu	A,A,4*SIZE					#  A+=2(mr)*2(kr)*8Byte=32
+	LD	b4,2*SIZE(B)
+
+	MADD	t12,t12,a0,b1
+	daddu	B,B,4*SIZE					#	2nr*2kr
+	LD	a5,-1*SIZE(A)
+	MADD	t22,t22,a1,b1
+	LD	b5,-1*SIZE(B)
+
+.L57:
+	MADD	t11,t11,a4,b4
+	LD	a0,0*SIZE(A)
+	MADD	t21,t21,a5,b4
+	LD	b0,0*SIZE(B)
+
+	MADD	t12,t12,a4,b5
+	LD	a1,1*SIZE(A)
+	MADD	t22,t22,a5,b5
+	LD	b1,1*SIZE(B)
+
+.L58:									#  kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP, 1
+#endif
+	beqz	K,.L59				
+	LD	ALPHA,152($sp)					#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,2*SIZE					#  	A+=2(mr)*1(kr)*8Byte=16
+	daddu	B,B,2*SIZE					#	2nr*kr
+
+	MADD	t12,t12,a0,b1
+	MADD	t22,t22,a1,b1
+
+
+.L59:									#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)						#  write gemm part back Fetch 16 C
+	LD	c21,1*SIZE(CO1)			
+	LD	c12,0(CO2)
+	LD	c22,1*SIZE(CO2)
+	
+	MADD	t11,c11,t11,ALPHA
+	MADD	t21,c21,t21,ALPHA
+	MADD	t12,c12,t12,ALPHA
+	MADD	t22,c22,t22,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t21,1*SIZE(CO1)
+	ST	t12,0(CO2)
+	ST	t22,1*SIZE(CO2)
+
+	daddu	CO1,CO1,2*SIZE			
+	daddu	CO2,CO2,2*SIZE
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+#else
+	daddiu	M, M, -1
+	daddiu	CO1,CO1, 2 * SIZE
+	daddiu	CO2,CO2, 2 * SIZE
+	MUL	t11, ALPHA, t11
+	MUL	t21, ALPHA, t21
+	MUL	t12, ALPHA, t12
+	MUL	t22, ALPHA, t22
+
+	ST	t11, -2 * SIZE(CO1)
+	ST	t21, -1 * SIZE(CO1)
+	ST	t12, -2 * SIZE(CO2)
+	ST	t22, -1 * SIZE(CO2)
+
+	FETCH	$0,0(CO1)
+	FETCH	$0,0(CO2)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -2
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+
+	dsll	K,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, B, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+
+
+	.align 3
+.L12_M1:
+	andi	M,MCO,1					#  	mr = 1
+	beqz	M,.L0_N2_Loop		
+	nop
+
+.L60:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,BO					#	Reset B
+#else
+	dsll	K,    KK, 0 + BASE_SHIFT
+	dsll	TEMP, KK, 1 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+	LD	a0,0*SIZE(A)
+	
+	MTC		$0,t11
+	MOV	t21,t11
+	LD	b0,0*SIZE(B)
+
+	MOV	t12,t11
+	LD	b1,1*SIZE(B)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1
+#else
+	daddiu	TEMP, KK, 2
+#endif
+	dsra	K,TEMP,2				
+	MOV	t22,t11
+	beqz	K,.L65
+	nop
+
+#else
+	dsra	K,KCO,2				
+	move	B,BO					#  Reset B
+	LD	a0,0*SIZE(A)
+	
+	MTC		$0,t11
+	MOV	t21,t11
+	LD	b0,0*SIZE(B)
+
+	MOV	t12,t11
+	LD	b1,1*SIZE(B)
+	beqz	K,.L65
+	MOV	t22,t11
+
+#endif
+
+.L61:								#	nr=2,mr=1,kr=4	
+	LD	a4,	1*SIZE(A)				#	a2
+	LD	b4, 2*SIZE(B)
+	MADD	t11,t11,a0,b0
+	
+	LD	b5,3*SIZE(B)
+	MADD	t12,t12,a0,b1
+
+	LD	a2,	2*SIZE(A)				#	a3
+	LD	b2,4*SIZE(B)
+	MADD	t11,t11,a4,b4
+	
+	LD	b3,5*SIZE(B)
+	MADD	t12,t12,a4,b5
+
+	LD	a6,	3*SIZE(A)				#	a4
+	daddiu	K,K,-1
+	LD	b6,6*SIZE(B)
+	MADD	t11,t11,a2,b2
+	
+	LD	b7,7*SIZE(B)
+	MADD	t12,t12,a2,b3
+	daddu	A,A,4*SIZE				#  A+=1(mr)*4(kr)*8Byte=32
+
+	LD	a0,	0*SIZE(A)
+	daddu	B,B,8*SIZE				#  B+=2(nr)*4(kr)*8Byte=8*SIZE
+	
+	LD	b0,0*SIZE(B)	
+	MADD	t11,t11,a6,b6
+	
+	LD	b1,1*SIZE(B)
+	bnez 	K,.L61
+	MADD	t12,t12,a6,b7
+
+
+
+.L65:								#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L68
+	nop
+
+.L66:			
+	LD	a4,	1*SIZE(A)				#	a1
+	MADD	t11,t11,a0,b0
+	LD	b4,2*SIZE(B)
+	daddu	A,A,2*SIZE				#  A+=1(mr)*2(kr)*8Byte=16
+	
+	LD	b5,3*SIZE(B)
+	MADD	t12,t12,a0,b1
+	daddu	B,B,4*SIZE
+
+.L67:
+	LD	a0,0(A)						#	a0
+	LD	b0,0*SIZE(B)
+	MADD	t11,t11,a4,b4
+	
+	LD	b1,1*SIZE(B)
+	MADD	t12,t12,a4,b5
+
+
+.L68:								#   kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L69				  
+	LD	ALPHA,152($sp)				#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t12,t12,a0,b1
+	daddu	A,A,1*SIZE				#  A+=1(mr)*1(kr)*8Byte=16
+	daddu	B,B,2*SIZE
+
+
+.L69:								#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)					#  Fetch 16 C
+	LD	c12,0(CO2)
+	
+	MADD	t11,c11,t11,ALPHA
+	MADD	t12,c12,t12,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t12,0(CO2)
+
+	daddu	CO1,CO1,1*SIZE		
+	daddu	CO2,CO2,1*SIZE
+
+#else
+	MUL	t11, ALPHA, t11
+	MUL	t12, ALPHA, t12
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+
+	daddu	CO1,CO1,1*SIZE			
+	daddu	CO2,CO2,1*SIZE
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -1
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+
+	dsll	K,    TEMP, 0 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, B, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 1
+#endif
+#endif
+
+.L0_N2_Loop:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	daddiu	KK, KK, 2
+#endif
+	move	BO, B
+
+
+	.align	5					
+.L0_N1:
+	andi	N,NCO,1					#  nr = 1
+	beqz	N,.L999					
+	nop
+
+	move	CO1,C				
+	dsra	M,MCO,2				
+	
+	move	A,AO					#  Reset A
+	daddu	PREA,AO,SPANA
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	move	KK, OFFSET
+#endif
+
+	beqz	M,.L11_M2
+	daddu	C,CO1,LDC
+
+.L70:						
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B, BO					#	Reset B
+#else
+	dsll	K,    KK, 2 + BASE_SHIFT
+	dsll	TEMP, KK, 0 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+	LD	b0,	0*SIZE(B)
+	
+	MTC		$0,t11
+	LD	a0,0*SIZE(A)
+	MOV	t21,t11
+	LD	a1,1*SIZE(A)
+	
+	MOV	t31,t11
+	LD	a2,2*SIZE(A)
+	MOV	t41,t11
+	LD	a3,3*SIZE(A)
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 4
+#else
+	daddiu	TEMP, KK, 1
+#endif
+	dsra	K,TEMP,2		
+	beqz	K,.L75
+	nop
+#else
+	move	B, BO					#	Reset B
+	dsra	K,KCO,2			
+	LD	b0,	0*SIZE(B)
+	
+	MTC		$0,t11
+	LD	a0,0*SIZE(A)
+	MOV	t21,t11
+	LD	a1,1*SIZE(A)
+	
+	MOV	t31,t11
+	LD	a2,2*SIZE(A)
+	MOV	t41,t11
+	beqz	K,.L75
+	LD	a3,3*SIZE(A)
+
+#endif
+
+.L71:								#  nr=1,mr=kr=4
+	LD	b4,	1*SIZE(B)				#	b1
+	MADD	t11,t11,a0,b0
+	
+	LD	a4,	4*SIZE(A)
+	MADD	t21,t21,a1,b0
+
+	LD	a5,	5*SIZE(A)
+	FETCH		$0,(PREA)
+
+	LD	a6,6*SIZE(A)
+	MADD	t31,t31,a2,b0
+
+	LD	a7,7*SIZE(A)
+	MADD	t41,t41,a3,b0
+
+.L72:
+	LD	b2,	2*SIZE(B)				#	b2
+	MADD	t11,t11,a4,b4
+	
+	LD	a0,8*SIZE(A)
+	MADD	t21,t21,a5,b4
+
+	LD	a1,9*SIZE(A)
+	FETCH		$0,4*SIZE(PREA)
+
+	LD	a2,10*SIZE(A)
+	MADD	t31,t31,a6,b4
+	
+	LD	a3,11*SIZE(A)
+	MADD	t41,t41,a7,b4
+
+.L73:
+	LD	b6,	3*SIZE(B)
+	MADD	t11,t11,a0,b2
+	
+	LD	a4,12*SIZE(A)
+	daddu	B,B,4*SIZE				#  B+=1(nr)*4(kr)*8Byte=32
+	
+	LD	a5,13*SIZE(A)
+	MADD	t21,t21,a1,b2
+
+	LD	a6,14*SIZE(A)
+	FETCH		$0,8*SIZE(PREA)
+	MADD	t31,t31,a2,b2
+
+	LD	a7,15*SIZE(A)
+	MADD	t41,t41,a3,b2
+	daddu	A,A,16*SIZE				#  A+=4(mr)*4(kr)*8Byte=16*SIZE
+
+.L74:
+	LD	b0,	0*SIZE(B)
+	MADD	t11,t11,a4,b6
+	
+	LD	a0,0*SIZE(A)
+	daddu	PREA,PREA,16*SIZE
+
+	LD	a1,1*SIZE(A)
+	MADD	t21,t21,a5,b6
+
+	LD	a2,2*SIZE(A)
+	daddiu	K,K,-1
+	MADD	t31,t31,a6,b6
+
+	LD	a3,3*SIZE(A)
+	MADD	t41,t41,a7,b6
+	bnez 	K,.L71
+	FETCH		$0,-32(PREA)
+
+
+.L75:								#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L78
+	nop
+
+.L76:			
+	LD	b4,	1*SIZE(B)
+	MADD	t11,t11,a0,b0
+	
+	LD	a4,4*SIZE(A)
+	daddu	B,B,2*SIZE				#  B+=1(nr)*2(kr)*8Byte=32
+	
+	LD	a5,5*SIZE(A)
+	MADD	t21,t21,a1,b0
+	FETCH		$0,0(PREA)
+
+	LD	a6,6*SIZE(A)
+	MADD	t31,t31,a2,b0
+
+	LD	a7,7*SIZE(A)
+	MADD	t41,t41,a3,b0
+	daddu	A,A,8*SIZE				#  A+=4(mr)*2(kr)*8Byte=8*SIZE
+
+.L77:
+	LD	b0,0(B)
+	MADD	t11,t11,a4,b4
+
+	LD	a0,0*SIZE(A)
+	MADD	t21,t21,a5,b4
+	FETCH		$0,4*SIZE(PREA)
+
+	LD	a1,1*SIZE(A)
+	MADD	t31,t31,a6,b4
+
+	LD	a2,2*SIZE(A)
+	MADD	t41,t41,a7,b4
+
+	LD	a3,3*SIZE(A)
+	daddu	PREA,PREA,8*SIZE
+
+
+	
+.L78:								#   kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L79				 
+	LD	ALPHA,152($sp)				#  Get ALPHA
+	
+	FETCH		$0,0(PREA)
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,4*SIZE				#  A+=4(mr)*1(kr)*8Byte=32
+
+	MADD	t31,t31,a2,b0
+	MADD	t41,t41,a3,b0
+	daddu	B,B,1*SIZE
+	daddu	PREA,PREA,4*SIZE
+
+
+.L79:								#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)					#  Fetch 16 C
+	LD	c21,1*SIZE(CO1)			
+	LD	c31,2*SIZE(CO1)
+	LD	c41,3*SIZE(CO1)
+
+	MADD	t11,c11,t11,ALPHA
+	MADD	t21,c21,t21,ALPHA
+	MADD	t31,c31,t31,ALPHA
+	MADD	t41,c41,t41,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t21,1*SIZE(CO1)
+	ST	t31,2*SIZE(CO1)
+	ST	t41,3*SIZE(CO1)
+	daddiu	M,M,-1					#  M--
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,8*SIZE(CO1)
+
+	bnez	M,.L70					#  M!=0
+	daddu	CO1,CO1,4*SIZE			#  COx += 4*8Byte
+#else
+	daddiu	M,M,-1					#  M--
+	MUL	t11, ALPHA, t11
+	MUL	t21, ALPHA, t21
+	MUL	t31, ALPHA, t31
+	MUL	t41, ALPHA, t41
+
+	ST	t11,0(CO1)
+	ST	t21,1*SIZE(CO1)
+	ST	t31,2*SIZE(CO1)
+	ST	t41,3*SIZE(CO1)
+
+	FETCH	$0,4*SIZE(CO1)
+	FETCH	$0,8*SIZE(CO1)
+
+	daddu	CO1,CO1,4*SIZE			
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -4
+#else
+	daddiu	TEMP, TEMP, -1
+#endif
+
+	dsll	K,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+
+	daddu	A, A,K
+	daddu	B, B, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 4
+#endif
+	bnez	M,.L70				
+	nop
+#endif
+
+
+	.align 3
+.L11_M2:
+	andi	M,MCO,2					#  mr = 2
+	beqz	M,.L11_M1			
+	nop
+
+.L80:						
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B, BO
+#else
+	dsll	K,    KK, 1 + BASE_SHIFT
+	dsll	TEMP, KK, 0 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+	LD	b0,	0*SIZE(B)
+
+	MTC		$0,t11
+	MOV		t21,t11
+	LD	a0,0*SIZE(A)
+	LD	a1,1*SIZE(A)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 2
+#else
+	daddiu	TEMP, KK, 1
+#endif
+	dsra	K,TEMP,2				#  K=KCO/2
+	beqz	K,.L85
+	nop
+#else
+	move	B, BO
+	dsra	K,KCO,2				
+	LD	b0,	0*SIZE(B)
+
+	MTC		$0,t11
+	MOV		t21,t11
+	LD	a0,0*SIZE(A)
+	
+	beqz	K,.L85
+	LD	a1,1*SIZE(A)
+
+#endif
+
+.L81:								#  nr=1,mr=2,kr=4
+	LD	b4,	1*SIZE(B)
+	LD	a4,2*SIZE(A)
+	MADD	t11,t11,a0,b0
+	LD	a5,3*SIZE(A)
+	MADD	t21,t21,a1,b0
+
+	LD	b2,	2*SIZE(B)
+	LD	a2,4*SIZE(A)
+	MADD	t11,t11,a4,b4
+	LD	a3,5*SIZE(A)
+	MADD	t21,t21,a5,b4
+	
+	LD	b6,	3*SIZE(B)
+	LD	a6,6*SIZE(A)
+	MADD	t11,t11,a2,b2
+	LD	a7,7*SIZE(A)
+	MADD	t21,t21,a3,b2
+
+	daddu	A,A,8*SIZE				#  A+=2(mr)*4(kr)*8Byte=8*SIZE
+	daddu	B,B,4*SIZE				#  B+=1(nr)*4(kr)*8Byte=32
+
+	LD	b0,	0*SIZE(B)
+	daddiu	K,K,-1
+
+	LD	a0,0*SIZE(A)
+	MADD	t11,t11,a6,b6
+
+	LD	a1,1*SIZE(A)
+	bnez 	K,.L81
+	MADD	t21,t21,a7,b6
+
+.L85:								#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2				
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L88
+	nop
+
+.L86:			
+	LD	b4,	1*SIZE(B)
+	LD	a4,2*SIZE(A)
+	MADD	t11,t11,a0,b0
+	LD	a5,3*SIZE(A)
+	MADD	t21,t21,a1,b0
+	
+	daddu	A,A,4*SIZE				#  A+=2(mr)*2(kr)*8Byte=32
+	daddu	B,B,2*SIZE				#  B+=1(nr)*2(kr)*8Byte=16
+	
+	LD	b0,0(B)
+	LD	a0,0*SIZE(A)
+	MADD	t11,t11,a4,b4
+	LD	a1,1*SIZE(A)
+	MADD	t21,t21,a5,b4
+
+
+	
+.L88:								#  kr=1
+#ifndef TRMMKERNEL
+	andi	K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L89				
+	LD	ALPHA,152($sp)				#  Get ALPHA
+	
+	MADD	t11,t11,a0,b0
+	MADD	t21,t21,a1,b0
+	daddu	A,A,2*SIZE				#  A+=2(mr)*1(kr)*8Byte=16
+	daddu	B,B,1*SIZE
+
+
+.L89:								#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)					#  Fetch 16 C
+	LD	c21,1*SIZE(CO1)			
+
+	MADD	t11,c11,t11,ALPHA
+	MADD	t21,c21,t21,ALPHA
+
+	ST	t11,0(CO1)
+	ST	t21,1*SIZE(CO1)
+
+	FETCH	$0,2*SIZE(CO1)
+	
+	daddu	CO1,CO1,2*SIZE			#  COx += 2*8Byte
+
+#else
+	daddu	CO1,CO1,2*SIZE			#  COx += 2*8Byte
+	MUL	t11, ALPHA, t11
+	MUL	t21, ALPHA, t21
+
+	FETCH	$0,0(CO1)
+	ST	t11, -2 * SIZE(CO1)
+	ST	t21, -1 * SIZE(CO1)
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -2
+#else
+	daddiu	TEMP, TEMP, -1
+#endif
+
+	dsll	K,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, B, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+
+
+	.align 3
+.L11_M1:
+	andi		M,MCO,1				#   mr = 1
+	beqz	M,.L999			
+	nop
+
+.L90:			
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	B,  BO
+#else
+	dsll	K,    KK, 0 + BASE_SHIFT
+	dsll	TEMP, KK, 0 + BASE_SHIFT
+
+	daddu	A, A, K
+	daddu	B, BO,  TEMP
+#endif
+	LD	a0,	0*SIZE(A)
+	LD	b0,	0*SIZE(B)
+	MTC		$0,t11
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, KCO, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1
+#else
+	daddiu	TEMP, KK, 1
+#endif
+	dsra	K,  TEMP, 2
+	beqz	K,.L95
+	nop
+
+#else
+	move	B,  BO
+	LD	a0,	0*SIZE(A)
+	LD	b0,	0*SIZE(B)
+	dsra	K,KCO,2				
+	beqz	K,.L95
+	MTC		$0,t11
+#endif
+
+.L91:								#  nr=mr=1,kr=4
+	LD	a4,	1*SIZE(A)
+	LD	b4,	1*SIZE(B)
+	MADD	t11,t11,a0,b0
+	
+	LD	a2,	2*SIZE(A)
+	LD	b2,	2*SIZE(B)
+	MADD	t11,t11,a4,b4
+
+	LD	a6,	3*SIZE(A)
+	LD	b6,	3*SIZE(B)
+	MADD	t11,t11,a2,b2
+	
+	daddu	A,A,4*SIZE				#  A+=1(mr)*4(kr)*8Byte=32
+	daddu	B,B,4*SIZE				#  B+=1(nr)*4(kr)*8Byte=32
+
+	LD	a0,	0*SIZE(A)
+	LD	b0,	0*SIZE(B)
+	MADD	t11,t11,a6,b6
+	
+	daddiu	K,K,-1
+	bnez 	K,.L91
+	nop
+
+.L95:								#  kr=2
+#ifndef TRMMKERNEL
+	andi	K,KCO,2			
+#else
+	andi	K,TEMP,2
+#endif
+	beqz	K,.L98
+	nop
+
+.L96:			
+	LD	a4,	1*SIZE(A)
+	LD	b4,	1*SIZE(B)
+	MADD	t11,t11,a0,b0
+	daddu	B,B,2*SIZE				#  B+=1(nr)*2(kr)*8Byte=16
+	daddu	A,A,2*SIZE				#  A+=1(mr)*2(kr)*8Byte=32
+
+	LD	b0,0(B)
+	LD	a0,0(A)
+	MADD	t11,t11,a4,b4
+	
+.L98:								#  kr=1
+#ifndef TRMMKERNEL
+	andi		K,KCO,1
+#else
+	andi	K,TEMP,1
+#endif
+	beqz	K,.L99				
+	LD	ALPHA,152($sp)				#  Get ALPHA
+
+	MADD	t11,t11,a0,b0
+
+
+.L99:								#  Write Back
+#ifndef TRMMKERNEL
+	LD	c11,0(CO1)					#  Fetch 16 C
+	MADD	t11,c11,t11,ALPHA
+	ST	t11,0(CO1)
+
+#else
+	MUL	t11, ALPHA, t11
+
+	ST	t11,  0 * SIZE(CO1)
+#endif
+
+
+.L999:							#  End
+	ld	$16,   0($sp)
+	ld	$17,   8($sp)
+	ld	$18,  16($sp)
+	ld	$19,  24($sp)
+	ld	$20,  32($sp)
+	ld	$21,  40($sp)
+	ld	$22,  48($sp)
+	LD	$f24, 56($sp)
+	LD	$f25, 64($sp)
+	LD	$f26, 72($sp)
+	LD	$f27, 80($sp)
+	LD	$f28, 88($sp)
+	ld	$23,  96($sp)
+	ld	$24, 104($sp)
+	ld	$25, 112($sp)
+	LD	$f20,120($sp)
+	LD	$f21,128($sp)
+	LD	$f22,136($sp)
+	LD	$f23,144($sp)
+
+	j	$31
+	daddiu	$sp, $sp, 160
+
+	EPILOGUE
diff --git a/kernel/mips64/trsm_kernel_LN_loongson3a.S b/kernel/mips64/trsm_kernel_LN_loongson3a.S
new file mode 100644
index 000000000..aba86fbce
--- /dev/null
+++ b/kernel/mips64/trsm_kernel_LN_loongson3a.S
@@ -0,0 +1,1938 @@
+#define REALNAME ASMNAME
+
+#define ASSEMBLER
+#include "common.h"
+
+
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$8
+#define B	$9
+#define C	$10
+#define LDC	$11
+
+#define AO	$12
+#define BO	$13
+
+#define I	$2
+#define J	$3
+#define L	$7
+
+#define CO1	$14
+#define CO2	$15
+#define CO3	$16
+#define CO4	$17
+
+#define OFFSET	$22
+#define KK	$23
+#define TEMP	$24
+#define AORIG	$25
+
+#define a1	$f0
+#define a2	$f1
+#define a3	$f2
+#define a4	$f3
+#define a5	$f4
+#define a6	$f5
+#define a7	$f6
+#define	a8	$f7
+
+#define b1	$f8
+#define b2	$f9
+#define b3	$f10
+#define b4	$f11
+#define b5	$f12
+#define b6	$f13
+#define b7	$f14
+#define b8	$f15
+
+#define t11	$f16
+#define t21	$f17
+#define t31	$f18
+#define t41	$f19
+
+#define t12	$f20
+#define t22	$f21
+#define t32	$f22
+#define t42	$f23
+
+#define t13	$f24
+#define t23	$f25
+#define t33	$f26
+#define t43	$f27
+
+#define t14	$f28
+#define t24	$f29
+#define t34	$f30
+#define t44	$f31
+
+#define ALPHA	$f15
+
+	PROLOGUE
+	
+	daddiu	$sp, $sp, -144
+
+	SDARG	$16,   0($sp)
+	SDARG	$17,   8($sp)
+	SDARG	$18,  16($sp)
+	SDARG	$19,  24($sp)
+	SDARG	$20,  32($sp)
+	SDARG	$21,  40($sp)
+	sdc1	$f24, 48($sp)
+	sdc1	$f25, 56($sp)
+	sdc1	$f26, 64($sp)
+	sdc1	$f27, 72($sp)
+	sdc1	$f28, 80($sp)
+
+	SDARG	$22,  88($sp)
+	SDARG	$23,  96($sp)
+	SDARG	$24, 104($sp)
+	SDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	sdc1	$f20,112($sp)
+	sdc1	$f21,120($sp)
+	sdc1	$f22,128($sp)
+	sdc1	$f23,136($sp)
+#endif
+											#	LN compute from bottom to top
+	LDARG	OFFSET, 144($sp)					
+	dsll	LDC, LDC, BASE_SHIFT			#	ldc
+
+	mult	M, K
+	mflo	TEMP							#	TEMP=MC*KC
+
+	dsll	TEMP, TEMP, BASE_SHIFT			
+	daddu	A, A, TEMP						#	A move to the end of sa
+
+	dsll	TEMP, M, BASE_SHIFT
+	daddu	C, C, TEMP						#	C+=MC
+
+	dsra	J,  N, 2						#	j = nc/4
+	blez	J, .L30
+	nop
+
+.L10:										#	nr=4
+	daddiu	J, J, -1
+	move	CO1, C
+	daddu	CO2, C,   LDC
+	daddu	CO3, CO2, LDC
+	daddu	CO4, CO3, LDC
+
+	MTC	$0,  t11							#	clear result registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	daddu	KK, M, OFFSET					#	kc - kk is the length of the rectangular data part of panel Ai 
+	move	AORIG, A						#	reset A
+
+	daddu	C,  CO4, LDC					#	fixed pointer C, the write back address
+	
+	andi	I,  M, 1						#	mr=2,nr=4	
+	blez	I, .L50
+	nop
+
+	dsll	TEMP,   K, BASE_SHIFT			#	mr=1
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of Ai
+
+	dsll	L,    KK, BASE_SHIFT			#	mr=1	
+	dsll	TEMP, KK, 2 + BASE_SHIFT		#	nr=4
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+	MOV	t13, t11							#	mr=2
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)				
+
+	dsra	L,  TEMP, 2
+	blez	L, .L55
+	nop
+
+
+	.align	3
+.L52:
+	LD	a5,  1 * SIZE(AO)					
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	LD	a3,   2 * SIZE(AO)
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t12, t12, a5, b6
+	MADD	t13, t13, a5, b7
+	MADD	t14, t14, a5, b8
+
+	LD	a7,   3 * SIZE(AO)
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1				#	3rd compute
+	MADD	t12, t12, a3, b2
+	MADD	t13, t13, a3, b3
+	MADD	t14, t14, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 1mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5				#	4th compute
+	MADD	t12, t12, a7, b6
+	MADD	t13, t13, a7, b7
+	MADD	t14, t14, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L52
+	nop
+
+
+	.align 3
+.L55:
+	andi	L, TEMP, 3
+	blez	L, .L58
+	nop
+	
+	.align	3
+.L56:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 1mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L56
+	nop
+
+
+.L58:										#	deal with the triangular part
+	daddiu	TEMP, KK, -1
+	dsll	L,    TEMP, BASE_SHIFT			#	mr=1
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	Ao point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+
+
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+	MUL	t12, b3, t12
+	MUL	t13, b3, t13
+	MUL	t14, b3, t14
+
+	daddiu	CO1, CO1, -1 * SIZE
+	daddiu	CO2, CO2, -1 * SIZE
+	daddiu	CO3, CO3, -1 * SIZE
+	daddiu	CO4, CO4, -1 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+
+
+	daddiu	KK, KK, -1							#	the length of rectangular data part increases by 1
+	MTC	$0,  t11							#	clear result registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+
+	
+.L50:	
+	andi	I,  M, 2						#	mr=2,nr=4	
+	blez	I, .L20
+	nop
+
+	dsll	TEMP,   K, 1 + BASE_SHIFT
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of Ai
+
+	dsll	L,    KK, 1 + BASE_SHIFT		
+	dsll	TEMP, KK, 2 + BASE_SHIFT
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+	MOV	t13, t11							#	mr=2
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)				
+
+	dsra	L,  TEMP, 2
+	blez	L, .L25
+	nop
+
+
+	.align	3
+.L22:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	LD	a3,   4 * SIZE(AO)
+	LD	a4,   5 * SIZE(AO)
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+
+	LD	a7,   6 * SIZE(AO)
+	LD	a8,   7 * SIZE(AO)
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1				#	3rd compute
+	MADD	t21, t21, a4, b1
+	MADD	t12, t12, a3, b2
+	MADD	t22, t22, a4, b2
+	MADD	t13, t13, a3, b3
+	MADD	t23, t23, a4, b3
+	MADD	t14, t14, a3, b4
+	MADD	t24, t24, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5				#	4th compute
+	MADD	t21, t21, a8, b5
+	MADD	t12, t12, a7, b6
+	MADD	t22, t22, a8, b6
+	MADD	t13, t13, a7, b7
+	MADD	t23, t23, a8, b7
+	MADD	t14, t14, a7, b8
+	MADD	t24, t24, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L22
+	nop
+
+
+	.align 3
+.L25:
+	andi	L, TEMP, 3
+	blez	L, .L28
+	nop
+	
+	.align	3
+.L26:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L26
+	nop
+
+
+.L28:										#	deal with the triangular part
+	daddiu	TEMP, KK, -2
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	Ao point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+	SUB	t21, b5, t21
+	SUB	t22, b6, t22
+	SUB	t23, b7, t23
+	SUB	t24, b8, t24
+
+
+	LD	b1,  3 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  2 * SIZE(AO)
+	MUL	t21, b1, t21
+	MUL	t22, b1, t22
+	MUL	t23, b1, t23
+	MUL	t24, b1, t24
+	NMSUB	t11, t11, b2, t21
+	NMSUB	t12, t12, b2, t22
+	NMSUB	t13, t13, b2, t23
+	NMSUB	t14, t14, b2, t24
+	
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+	MUL	t12, b3, t12
+	MUL	t13, b3, t13
+	MUL	t14, b3, t14
+
+	daddiu	CO1, CO1, -2 * SIZE
+	daddiu	CO2, CO2, -2 * SIZE
+	daddiu	CO3, CO3, -2 * SIZE
+	daddiu	CO4, CO4, -2 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+	ST	t21,  4 * SIZE(BO)
+	ST	t22,  5 * SIZE(BO)
+	ST	t23,  6 * SIZE(BO)
+	ST	t24,  7 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+
+
+
+	daddiu	KK, KK, -2							#	the length of rectangular data part increases by 2
+	MTC	$0,  t11							#	clear result registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+
+.L20:
+	dsra	I,  M, 2						#	I=MC/4
+	blez	I, .L29
+	nop
+
+.L11:										#	mr=4
+	dsll	TEMP,   K,  2 + BASE_SHIFT		#	TEMP=KC*MR*data_Byte
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of panel Ai
+	dsll	L,    KK, 2 + BASE_SHIFT		# 	KC-KK is the length of the rectangular data part of Ai	
+	dsll	TEMP, KK, 2 + BASE_SHIFT		#	KK*NR*data_Byte
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B, TEMP
+
+	dsubu	TEMP, K, KK	
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)				
+
+	MOV	t13, t11							#	clear result registers
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L15
+	nop
+
+	.align	3
+.L12:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	4th compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			
+
+	daddiu	L, L, -1
+	bgtz	L, .L12
+	nop
+	
+
+	.align 3
+.L15:
+	andi	L, TEMP, 3
+	blez	L, .L18
+	nop
+
+	.align	3
+.L16:
+	MADD	t11, t11, a1, b1				
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L16
+	nop
+
+
+.L18:										#	deal with the triangular data part of panel Ai
+	daddiu	TEMP, KK, -4					#	
+
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	AO point to the triangular data part
+	daddu	BO, B, TEMP
+	
+	LD	b1,  0 * SIZE(BO)					#	triangular_part*X + rectangular_part = B
+	LD	b2,  1 * SIZE(BO)					#	triangular_part*X = B - rectangular_part
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+
+	LD	b5,  4 * SIZE(BO)					#	sb store in row major
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+	
+	SUB	t21, b5, t21
+	SUB	t22, b6, t22
+	SUB	t23, b7, t23
+	SUB	t24, b8, t24
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3, 10 * SIZE(BO)
+	LD	b4, 11 * SIZE(BO)
+	
+	SUB	t31, b1, t31
+	SUB	t32, b2, t32
+	SUB	t33, b3, t33
+	SUB	t34, b4, t34
+	
+	LD	b5, 12 * SIZE(BO)
+	LD	b6, 13 * SIZE(BO)
+	LD	b7, 14 * SIZE(BO)
+	LD	b8, 15 * SIZE(BO)
+
+	SUB	t41, b5, t41
+	SUB	t42, b6, t42
+	SUB	t43, b7, t43
+	SUB	t44, b8, t44
+
+
+	LD	b1,  15 * SIZE(AO)
+	LD	b2,  14 * SIZE(AO)	
+	LD	b4,  13 * SIZE(AO)
+	LD	b7,  12 * SIZE(AO)
+	
+	MUL	t41, b1, t41
+	MUL	t42, b1, t42
+	MUL	t43, b1, t43
+	MUL	t44, b1, t44
+	NMSUB	t31, t31, b2, t41
+	NMSUB	t32, t32, b2, t42
+	NMSUB	t33, t33, b2, t43
+	NMSUB	t34, t34, b2, t44
+	NMSUB	t21, t21, b4, t41
+	NMSUB	t22, t22, b4, t42
+	NMSUB	t23, t23, b4, t43
+	NMSUB	t24, t24, b4, t44
+	NMSUB	t11, t11, b7, t41
+	NMSUB	t12, t12, b7, t42
+	NMSUB	t13, t13, b7, t43
+	NMSUB	t14, t14, b7, t44
+
+
+
+	LD	b3,  10 * SIZE(AO)	
+	LD	b5,   9 * SIZE(AO)
+	LD	b8,   8 * SIZE(AO)
+	MUL	t31, b3, t31
+	MUL	t32, b3, t32
+	MUL	t33, b3, t33
+	MUL	t34, b3, t34
+	NMSUB	t21, t21, b5, t31
+	NMSUB	t22, t22, b5, t32
+	NMSUB	t23, t23, b5, t33
+	NMSUB	t24, t24, b5, t34
+	NMSUB	t11, t11, b8, t31
+	NMSUB	t12, t12, b8, t32
+	NMSUB	t13, t13, b8, t33
+	NMSUB	t14, t14, b8, t34
+
+
+
+	LD	b6,   5 * SIZE(AO)
+	LD	b1,   4 * SIZE(AO)
+	MUL	t21, b6, t21
+	MUL	t22, b6, t22
+	MUL	t23, b6, t23
+	MUL	t24, b6, t24
+	NMSUB	t11, t11, b1, t21
+	NMSUB	t12, t12, b1, t22
+	NMSUB	t13, t13, b1, t23
+	NMSUB	t14, t14, b1, t24
+
+
+
+	LD	b2,   0 * SIZE(AO)
+	MUL	t11, b2, t11
+	MUL	t12, b2, t12
+	MUL	t13, b2, t13
+	MUL	t14, b2, t14
+
+	daddiu	CO1, CO1, -4 * SIZE				#	modify 
+	daddiu	CO2, CO2, -4 * SIZE
+	daddiu	CO3, CO3, -4 * SIZE
+	daddiu	CO4, CO4, -4 * SIZE
+
+
+	ST	t11,  0 * SIZE(BO)					#	update packed B
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+	ST	t21,  4 * SIZE(BO)
+	ST	t22,  5 * SIZE(BO)
+	ST	t23,  6 * SIZE(BO)
+	ST	t24,  7 * SIZE(BO)
+	ST	t31,  8 * SIZE(BO)
+	ST	t32,  9 * SIZE(BO)
+	ST	t33, 10 * SIZE(BO)
+	ST	t34, 11 * SIZE(BO)
+	ST	t41, 12 * SIZE(BO)
+	ST	t42, 13 * SIZE(BO)
+	ST	t43, 14 * SIZE(BO)
+	ST	t44, 15 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)					#	write back 
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t33,  2 * SIZE(CO3)
+	ST	t43,  3 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+	ST	t34,  2 * SIZE(CO4)
+	ST	t44,  3 * SIZE(CO4)
+
+
+	daddiu	KK, KK, -4						#	KC-KK is the length of the rectangular data part, LN compute from bottom to top so KK-=4
+	daddiu	I, I, -1
+
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+	MOV	t12, a1
+	MOV	t22, a1
+	MOV	t32, a1
+	MOV	t42, a1
+	bgtz	I, .L11
+	nop
+
+	.align 3
+
+.L29:
+	dsll	TEMP, K, 2 + BASE_SHIFT
+	daddu	B, B, TEMP							# B point to next Bj
+
+	bgtz	J, .L10
+	nop
+	
+	
+	.align 3
+.L30:
+	andi	J,  N, 2							#	nr=2
+	blez	J, .L70
+	nop
+
+	move	CO1, C
+	daddu	CO2, C,   LDC
+
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	daddu	KK, M, OFFSET
+	move	AORIG, A							#	reset A
+	
+	daddu	C,  CO2, LDC						#	fixed 
+
+	andi	I,  M, 1							#	mr=1
+	blez	I, .L60
+	nop
+
+	dsll	TEMP,   K, BASE_SHIFT
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of everypanel of Ai
+
+	dsll	L,    KK, BASE_SHIFT			#	mr=1
+	dsll	TEMP, KK, 1 + BASE_SHIFT		#	nr=2
+
+	daddu	AO, AORIG, L					#	AO point to rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+	MOV	t12, t11							#	clear result registers
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L65
+	nop
+
+
+	.align	3
+.L62:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t12, t12, a1, b2
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t12, t12, a5, b6
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t12, t12, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 1mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t12, t12, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L62
+	nop
+	
+	.align 3
+
+.L65:
+	andi	L, TEMP, 3
+	blez	L, .L68
+	nop
+	
+	.align	3
+.L66:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L66
+	nop
+
+.L68:
+	daddiu	TEMP, KK, -1					#	mr=1
+
+	dsll	L,    TEMP, BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	Ao point to the triangular data part
+	daddu	BO, B,     TEMP
+	
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+
+	
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+	MUL	t12, b3, t12
+
+	daddiu	CO1, CO1, -1 * SIZE
+	daddiu	CO2, CO2, -1 * SIZE
+
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+
+
+	daddiu	KK, KK, -1
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+
+
+
+.L60:
+	andi	I,  M, 2
+	blez	I, .L40
+	nop
+
+	dsll	TEMP,   K, 1 + BASE_SHIFT
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of everypanel of Ai
+
+	dsll	L,    KK, 1 + BASE_SHIFT		#	mr=2
+	dsll	TEMP, KK, 1 + BASE_SHIFT		#	nr=2
+
+	daddu	AO, AORIG, L					#	AO point to rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+
+	MOV	t12, t11							#	clear result registers
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L45
+	nop
+
+
+	.align	3
+.L42:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t21, t21, a4, b3
+	MADD	t12, t12, a3, b4
+	MADD	t22, t22, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t21, t21, a8, b7
+	MADD	t12, t12, a7, b8
+	MADD	t22, t22, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L42
+	nop
+	
+	.align 3
+
+.L45:
+	andi	L, TEMP, 3
+	blez	L, .L48
+	nop
+	
+	.align	3
+.L46:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L46
+	nop
+
+.L48:
+	daddiu	TEMP, KK, -2
+
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	Ao point to the triangular data part
+	daddu	BO, B,     TEMP
+	
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t21, b3, t21
+	SUB	t22, b4, t22
+
+	LD	b1,  3 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  2 * SIZE(AO)
+	MUL	t21, b1, t21
+	MUL	t22, b1, t22
+	NMSUB	t11, t11, b2, t21
+	NMSUB	t12, t12, b2, t22
+	
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+	MUL	t12, b3, t12
+
+	daddiu	CO1, CO1, -2 * SIZE
+	daddiu	CO2, CO2, -2 * SIZE
+
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t21,  2 * SIZE(BO)
+	ST	t22,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+
+
+	daddiu	KK, KK, -2
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+
+.L40:
+	dsra	I,  M, 2							#	I = mc/4
+	blez	I, .L49
+	nop
+
+.L31:
+	dsll	TEMP,   K,  2 + BASE_SHIFT
+	dsubu	AORIG, AORIG, TEMP					#	AORIG point to the beginning address of panel Ai
+	dsll	L,    KK, 2 + BASE_SHIFT			#	mr=4
+	dsll	TEMP, KK, 1 + BASE_SHIFT			#	nr=2
+
+	daddu	AO, AORIG, L						#	AO point to the rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L35
+	nop
+
+
+	.align	3
+.L32:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+	LD	b3,   4 * SIZE(BO)
+	LD	b4,   5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3				#	3rd compute
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+	MADD	t12, t12, a1, b4
+	MADD	t22, t22, a2, b4
+	MADD	t32, t32, a3, b4
+	MADD	t42, t42, a4, b4
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7				#	4th compute
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+	MADD	t12, t12, a5, b8
+	MADD	t22, t22, a6, b8
+	MADD	t32, t32, a7, b8
+	MADD	t42, t42, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L32
+	nop
+
+	
+	.align 3
+
+.L35:
+	andi	L, TEMP, 3
+	blez	L, .L38
+	nop
+	
+	.align	3
+.L36:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+	
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L36
+	nop
+
+
+.L38:										#
+	daddiu	TEMP, KK, -4
+	dsll	L,    TEMP, 2 + BASE_SHIFT		#	mr=4
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT		#	nr=2
+	daddu	AO, AORIG, L					#	AO point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t21, b3, t21
+	SUB	t22, b4, t22
+	SUB	t31, b5, t31
+	SUB	t32, b6, t32
+	SUB	t41, b7, t41
+	SUB	t42, b8, t42
+
+
+	LD	b1,  15 * SIZE(AO)
+	LD	b2,  14 * SIZE(AO)	
+	LD	b4,  13 * SIZE(AO)
+	LD	b7,  12 * SIZE(AO)
+	
+	MUL	t41, b1, t41
+	MUL	t42, b1, t42
+	NMSUB	t31, t31, b2, t41
+	NMSUB	t32, t32, b2, t42
+	NMSUB	t21, t21, b4, t41
+	NMSUB	t22, t22, b4, t42
+	NMSUB	t11, t11, b7, t41
+	NMSUB	t12, t12, b7, t42
+
+
+
+	LD	b3,  10 * SIZE(AO)	
+	LD	b5,   9 * SIZE(AO)
+	LD	b8,   8 * SIZE(AO)
+	MUL	t31, b3, t31
+	MUL	t32, b3, t32
+	NMSUB	t21, t21, b5, t31
+	NMSUB	t22, t22, b5, t32
+	NMSUB	t11, t11, b8, t31
+	NMSUB	t12, t12, b8, t32
+
+
+
+	LD	b6,   5 * SIZE(AO)
+	LD	b1,   4 * SIZE(AO)
+	MUL	t21, b6, t21
+	MUL	t22, b6, t22
+	NMSUB	t11, t11, b1, t21
+	NMSUB	t12, t12, b1, t22
+
+
+	LD	b2,   0 * SIZE(AO)
+	MUL	t11, b2, t11
+	MUL	t12, b2, t12
+
+	daddiu	CO1, CO1, -4 * SIZE
+	daddiu	CO2, CO2, -4 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t21,  2 * SIZE(BO)
+	ST	t22,  3 * SIZE(BO)
+	ST	t31,  4 * SIZE(BO)
+	ST	t32,  5 * SIZE(BO)
+	ST	t41,  6 * SIZE(BO)
+	ST	t42,  7 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+
+
+	daddiu	KK, KK, -4
+
+	MTC	$0,  t11
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	daddiu	I, I, -1
+	bgtz	I, .L31
+	nop
+
+
+
+	.align 3
+.L49:
+	dsll	TEMP, K, 1 + BASE_SHIFT		# 	nr=2
+	daddu	B, B, TEMP
+
+	.align 3
+
+.L70:
+	andi	J,  N, 1					#	nr=1
+	blez	J, .L999					#	END
+	nop
+
+	move	CO1, C
+
+	daddu	KK, M, OFFSET
+	move	AORIG, A					#	reset A
+
+	andi	I,  M, 1					#	mr=1
+	blez	I, .L90
+	NOP
+
+	MTC	$0,  t11
+
+	dsll	TEMP,   K, BASE_SHIFT			#	mr=1
+	dsubu	AORIG, AORIG, TEMP
+	
+	dsll	L,    KK, BASE_SHIFT
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B,    L 
+
+	dsubu	TEMP, K, KK
+
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L95
+	nop
+	
+	.align	3
+.L92:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 1mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+
+	daddiu	L, L, -1
+	bgtz	L, .L92
+	nop
+	
+	.align 3
+
+.L95:
+	andi	L, TEMP,  3
+	blez	L, .L98
+	nop
+
+	.align	3
+.L96:
+	MADD	t11, t11, a1, b1				#	3rd compute
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 1mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L96
+	nop
+
+
+.L98:
+	daddiu	TEMP, KK, -1					# 	mr=2
+	dsll	TEMP, TEMP,  BASE_SHIFT
+
+	daddu	AO, AORIG, TEMP					#	AO point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+
+	SUB	t11, b1, t11
+
+
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+
+	daddiu	CO1, CO1, -1 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+
+	daddiu	KK, KK, -1
+
+
+.L90:	
+	andi	I,  M, 2
+	blez	I, .L80
+	NOP
+
+	MTC	$0,  t11
+	MOV	t21, t11							#	clear result registers
+
+	dsll	TEMP,   K, 1+BASE_SHIFT			#	mr=2
+	dsubu	AORIG, AORIG, TEMP
+	
+	dsll	L,    KK, 1 + BASE_SHIFT
+	dsll	TEMP, KK, 0 + BASE_SHIFT
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L85
+	nop
+	
+	.align	3
+.L82:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t21, t21, a4, b3
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t21, t21, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L82
+	nop
+	
+	.align 3
+
+.L85:
+	andi	L, TEMP,  3
+	blez	L, .L88
+	nop
+
+	.align	3
+.L86:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L86
+	nop
+
+
+.L88:
+	daddiu	TEMP, KK, -2					# 	mr=2
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+
+	daddu	AO, AORIG, L					#	AO point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+	LD	b1,  3 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  2 * SIZE(AO)
+	MUL	t21, b1, t21
+	NMSUB	t11, t11, b2, t21
+	
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+
+	daddiu	CO1, CO1, -2 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t21,  1 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+
+	daddiu	KK, KK, -2
+	
+	
+	.align	3
+.L80:
+	dsra	I,  M, 2
+	blez	I, .L89
+	nop
+
+.L71:
+	dsll	TEMP,   K,  2 + BASE_SHIFT		#	mr=4
+	dsubu	AORIG, AORIG, TEMP
+
+	dsll	L,    KK, 2 + BASE_SHIFT		#	mr=4
+	dsll	TEMP, KK, 0 + BASE_SHIFT		#	nr=1
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+
+	dsra	L,  TEMP, 2
+	blez	L, .L75
+	nop										#	reset B
+
+	.align	3
+.L72:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3				#	3rd compute
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7				#	4th compute
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L72
+	nop
+	
+	.align 3
+
+.L75:
+	andi	L, TEMP,  3
+	blez	L, .L78
+	nop
+
+	.align	3
+.L76:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L76
+	nop
+
+.L78:
+	daddiu	TEMP, KK, -4				#	mr=4
+
+	dsll	L,    TEMP, 2 + BASE_SHIFT	#	mr=4
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT	#	nr=1
+	daddu	AO, AORIG, L				#	AO point to the triangular
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+	LD	b1,  15 * SIZE(AO)
+	LD	b2,  14 * SIZE(AO)	
+	LD	b4,  13 * SIZE(AO)
+	LD	b7,  12 * SIZE(AO)
+	MUL	t41, b1, t41
+	NMSUB	t31, t31, b2, t41
+	NMSUB	t21, t21, b4, t41
+	NMSUB	t11, t11, b7, t41
+
+
+
+	LD	b3,  10 * SIZE(AO)	
+	LD	b5,   9 * SIZE(AO)
+	LD	b8,   8 * SIZE(AO)
+	MUL	t31, b3, t31
+	NMSUB	t21, t21, b5, t31
+	NMSUB	t11, t11, b8, t31
+
+
+
+	LD	b6,   5 * SIZE(AO)
+	LD	b1,   4 * SIZE(AO)
+	MUL	t21, b6, t21
+	NMSUB	t11, t11, b1, t21
+
+
+
+	LD	b2,   0 * SIZE(AO)
+	MUL	t11, b2, t11
+
+	daddiu	CO1, CO1, -4 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t21,  1 * SIZE(BO)
+	ST	t31,  2 * SIZE(BO)
+	ST	t41,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+
+
+	daddiu	KK, KK, -4
+	daddiu	I, I, -1
+	bgtz	I, .L71
+	nop
+
+
+	.align 3
+.L89:
+	dsll	TEMP, K, BASE_SHIFT			#	nr=1
+	daddu	B, B, TEMP
+
+
+
+	.align 3
+
+.L999:
+	LDARG	$16,   0($sp)
+	LDARG	$17,   8($sp)
+	LDARG	$18,  16($sp)
+	LDARG	$19,  24($sp)
+	LDARG	$20,  32($sp)
+	LDARG	$21,  40($sp)
+	ldc1	$f24, 48($sp)
+	ldc1	$f25, 56($sp)
+	ldc1	$f26, 64($sp)
+	ldc1	$f27, 72($sp)
+	ldc1	$f28, 80($sp)
+
+	LDARG	$22,  88($sp)
+	LDARG	$23,  96($sp)
+	LDARG	$24, 104($sp)
+	LDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	ldc1	$f20,112($sp)
+	ldc1	$f21,120($sp)
+	ldc1	$f22,128($sp)
+	ldc1	$f23,136($sp)
+#endif
+
+	j	$31
+	daddiu	$sp, $sp, 144
+
+	EPILOGUE
diff --git a/kernel/mips64/trsm_kernel_LT_loongson3a.S b/kernel/mips64/trsm_kernel_LT_loongson3a.S
new file mode 100644
index 000000000..4114d94ef
--- /dev/null
+++ b/kernel/mips64/trsm_kernel_LT_loongson3a.S
@@ -0,0 +1,1783 @@
+#define REALNAME ASMNAME
+
+#define ASSEMBLER
+#include "common.h"
+
+
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$8
+#define B	$9
+#define C	$10
+#define LDC	$11
+
+#define AO	$12
+#define BO	$13
+
+#define I	$2
+#define J	$3
+#define L	$7
+
+#define CO1	$14
+#define CO2	$15
+#define CO3	$16
+#define CO4	$17
+
+#define OFFSET	$22
+#define KK	$23
+#define TEMP	$24
+#define AORIG	$25
+
+#define a1	$f0
+#define a2	$f1
+#define a3	$f2
+#define a4	$f3
+#define a5	$f4
+#define a6	$f5
+#define a7	$f6
+#define	a8	$f7
+
+#define b1	$f8
+#define b2	$f9
+#define b3	$f10
+#define b4	$f11
+#define b5	$f12
+#define b6	$f13
+#define b7	$f14
+#define b8	$f15
+
+#define t11	$f16
+#define t21	$f17
+#define t31	$f18
+#define t41	$f19
+
+#define t12	$f20
+#define t22	$f21
+#define t32	$f22
+#define t42	$f23
+
+#define t13	$f24
+#define t23	$f25
+#define t33	$f26
+#define t43	$f27
+
+#define t14	$f28
+#define t24	$f29
+#define t34	$f30
+#define t44	$f31
+
+#define ALPHA	$f15
+
+	PROLOGUE
+	
+	daddiu	$sp, $sp, -144
+
+	SDARG	$16,   0($sp)
+	SDARG	$17,   8($sp)
+	SDARG	$18,  16($sp)
+	SDARG	$19,  24($sp)
+	SDARG	$20,  32($sp)
+	SDARG	$21,  40($sp)
+	sdc1	$f24, 48($sp)
+	sdc1	$f25, 56($sp)
+	sdc1	$f26, 64($sp)
+	sdc1	$f27, 72($sp)
+	sdc1	$f28, 80($sp)
+
+	SDARG	$22,  88($sp)
+	SDARG	$23,  96($sp)
+	SDARG	$24, 104($sp)
+	SDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	sdc1	$f20,112($sp)
+	sdc1	$f21,120($sp)
+	sdc1	$f22,128($sp)
+	sdc1	$f23,136($sp)
+#endif
+											#	LT compute from left to right, top to bottom
+	LDARG	OFFSET, 144($sp)					
+	dsll	LDC, LDC, BASE_SHIFT			#	ldc
+
+	dsra	J,  N, 2						#	j = nc/4
+	blez	J, .L30
+	nop
+
+.L10:										#	nr=4
+	daddiu	J, J, -1
+	move	CO1, C
+	daddu	CO2, C,   LDC
+	daddu	CO3, CO2, LDC
+	daddu	CO4, CO3, LDC
+
+	MTC	$0,  t11							#	clear result registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	dsra	I,  M, 2						#	i = mc/4
+	move	KK, OFFSET						#	kk is the length of the rectangular data part of panel Ai
+	move	AO, A							#	reset A	
+	daddu	C,  CO4, LDC					#	fixed pointer C, the write back address
+	blez	I, .L20
+	nop
+
+
+.L11:										#	mr=4
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+	LD	b2,  1 * SIZE(B)
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)				
+
+	MOV	t13, t11							#	clear result registers
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	dsra	L,  KK, 2						#	L = kk/4
+	blez	L, .L15
+	move	BO,  B							#
+
+
+	.align	3
+.L12:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	4th compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			
+
+	daddiu	L, L, -1
+	bgtz	L, .L12
+	nop
+	
+
+	.align 3
+.L15:
+	andi	L, KK,  3						#	the remainder part: KK-KK/4
+	blez	L, .L18
+	nop
+
+	.align	3
+.L16:
+	MADD	t11, t11, a1, b1				
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L16
+	nop
+
+
+.L18:										#	deal with the triangular data part of panel Ai
+	LD	b1,  0 * SIZE(BO)					#	triangular_part*X + rectangular_part = B
+	LD	b2,  1 * SIZE(BO)					#	triangular_part*X = B - rectangular_part
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+
+	LD	b5,  4 * SIZE(BO)					#	sb store in row major
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+	
+	SUB	t21, b5, t21
+	SUB	t22, b6, t22
+	SUB	t23, b7, t23
+	SUB	t24, b8, t24
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3, 10 * SIZE(BO)
+	LD	b4, 11 * SIZE(BO)
+	
+	SUB	t31, b1, t31
+	SUB	t32, b2, t32
+	SUB	t33, b3, t33
+	SUB	t34, b4, t34
+	
+	LD	b5, 12 * SIZE(BO)
+	LD	b6, 13 * SIZE(BO)
+	LD	b7, 14 * SIZE(BO)
+	LD	b8, 15 * SIZE(BO)
+
+	SUB	t41, b5, t41
+	SUB	t42, b6, t42
+	SUB	t43, b7, t43
+	SUB	t44, b8, t44
+
+
+	LD	a1,  0 * SIZE(AO)					#	sa stores in col major		
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	MUL	t11, a1, t11
+	MUL	t12, a1, t12
+	MUL	t13, a1, t13
+	MUL	t14, a1, t14
+	NMSUB	t21, t21, a2, t11
+	NMSUB	t22, t22, a2, t12
+	NMSUB	t23, t23, a2, t13
+	NMSUB	t24, t24, a2, t14
+	NMSUB	t31, t31, a3, t11
+	NMSUB	t32, t32, a3, t12
+	NMSUB	t33, t33, a3, t13
+	NMSUB	t34, t34, a3, t14
+	NMSUB	t41, t41, a4, t11
+	NMSUB	t42, t42, a4, t12
+	NMSUB	t43, t43, a4, t13
+	NMSUB	t44, t44, a4, t14
+
+	
+	LD	a5,  5 * SIZE(AO)
+	LD	a6,  6 * SIZE(AO)
+	LD	a7,  7 * SIZE(AO)
+	MUL	t21, a5, t21
+	MUL	t22, a5, t22
+	MUL	t23, a5, t23
+	MUL	t24, a5, t24
+	NMSUB	t31, t31, a6, t21
+	NMSUB	t32, t32, a6, t22
+	NMSUB	t33, t33, a6, t23
+	NMSUB	t34, t34, a6, t24
+	NMSUB	t41, t41, a7, t21
+	NMSUB	t42, t42, a7, t22
+	NMSUB	t43, t43, a7, t23
+	NMSUB	t44, t44, a7, t24
+
+
+	LD	a8, 10 * SIZE(AO)
+	LD	a1, 11 * SIZE(AO)
+	MUL	t31, a8, t31
+	MUL	t32, a8, t32
+	MUL	t33, a8, t33
+	MUL	t34, a8, t34
+	NMSUB	t41, t41, a1, t31
+	NMSUB	t42, t42, a1, t32
+	NMSUB	t43, t43, a1, t33
+	NMSUB	t44, t44, a1, t34
+
+
+	LD	a2, 15 * SIZE(AO)
+	MUL	t41, a2, t41
+	MUL	t42, a2, t42
+	MUL	t43, a2, t43
+	MUL	t44, a2, t44
+
+	ST	t11,  0 * SIZE(BO)					#	update packed B
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+	ST	t21,  4 * SIZE(BO)
+	ST	t22,  5 * SIZE(BO)
+	ST	t23,  6 * SIZE(BO)
+	ST	t24,  7 * SIZE(BO)
+	ST	t31,  8 * SIZE(BO)
+	ST	t32,  9 * SIZE(BO)
+	ST	t33, 10 * SIZE(BO)
+	ST	t34, 11 * SIZE(BO)
+	ST	t41, 12 * SIZE(BO)
+	ST	t42, 13 * SIZE(BO)
+	ST	t43, 14 * SIZE(BO)
+	ST	t44, 15 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)					#	write back 
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t33,  2 * SIZE(CO3)
+	ST	t43,  3 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+	ST	t34,  2 * SIZE(CO4)
+	ST	t44,  3 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 4 * SIZE				#	fixed pointers
+	daddiu	CO2, CO2, 4 * SIZE
+	daddiu	CO3, CO3, 4 * SIZE
+	daddiu	CO4, CO4, 4 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L						#	mov AO to the end of panel Ai
+	daddu	BO, BO, TEMP					#	mov BO to the end of panel Bj
+
+	daddiu	KK, KK, 4						#	the length of rectangular data part increases by 4
+	daddiu	I, I, -1
+
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+	MOV	t12, a1
+	MOV	t22, a1
+	MOV	t32, a1
+	MOV	t42, a1
+	bgtz	I, .L11
+	nop
+
+
+	.align 3
+.L20:
+	andi	I,  M, 2						#	mr=2,nr=4	
+	blez	I, .L50
+	nop
+
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+	LD	b2,  1 * SIZE(B)
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)				
+
+	dsra	L,  KK, 2
+	blez	L, .L25
+	move	BO,  B
+
+
+	.align	3
+.L22:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	LD	a3,   4 * SIZE(AO)
+	LD	a4,   5 * SIZE(AO)
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+
+	LD	a7,   6 * SIZE(AO)
+	LD	a8,   7 * SIZE(AO)
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1				#	3rd compute
+	MADD	t21, t21, a4, b1
+	MADD	t12, t12, a3, b2
+	MADD	t22, t22, a4, b2
+	MADD	t13, t13, a3, b3
+	MADD	t23, t23, a4, b3
+	MADD	t14, t14, a3, b4
+	MADD	t24, t24, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5				#	4th compute
+	MADD	t21, t21, a8, b5
+	MADD	t12, t12, a7, b6
+	MADD	t22, t22, a8, b6
+	MADD	t13, t13, a7, b7
+	MADD	t23, t23, a8, b7
+	MADD	t14, t14, a7, b8
+	MADD	t24, t24, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L22
+	nop
+
+
+	.align 3
+.L25:
+	andi	L, KK,  3
+	blez	L, .L28
+	nop
+	
+	.align	3
+.L26:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L26
+	nop
+
+
+.L28:										#	deal with the triangular part
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+	SUB	t21, b5, t21
+	SUB	t22, b6, t22
+	SUB	t23, b7, t23
+	SUB	t24, b8, t24
+
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  1 * SIZE(AO)
+	MUL	t11, b1, t11
+	MUL	t12, b1, t12
+	MUL	t13, b1, t13
+	MUL	t14, b1, t14
+	NMSUB	t21, t21, b2, t11
+	NMSUB	t22, t22, b2, t12
+	NMSUB	t23, t23, b2, t13
+	NMSUB	t24, t24, b2, t14
+	
+	LD	b3,  3 * SIZE(AO)
+	MUL	t21, b3, t21
+	MUL	t22, b3, t22
+	MUL	t23, b3, t23
+	MUL	t24, b3, t24
+
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+	ST	t21,  4 * SIZE(BO)
+	ST	t22,  5 * SIZE(BO)
+	ST	t23,  6 * SIZE(BO)
+	ST	t24,  7 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 2 * SIZE
+	daddiu	CO2, CO2, 2 * SIZE
+	daddiu	CO3, CO3, 2 * SIZE
+	daddiu	CO4, CO4, 2 * SIZE
+
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L							#	mov AO to the end of Ai
+	daddu	BO, BO, TEMP						#	mov BO to the end of Bj
+
+	daddiu	KK, KK, 2							#	the length of rectangular data part increases by 2
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+	MOV	t12, a1
+	MOV	t22, a1
+	MOV	t32, a1
+	MOV	t42, a1
+
+
+	.align 3
+.L50:
+	andi	I,  M, 1						#	mr=1,nr=4	
+	blez	I, .L29
+	nop
+
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+	LD	b2,  1 * SIZE(B)
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)				
+
+	dsra	L,  KK, 2
+	blez	L, .L55
+	move	BO,  B
+
+
+	.align	3
+.L52:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	LD	a3,   2 * SIZE(AO)
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t12, t12, a5, b6
+	MADD	t13, t13, a5, b7
+	MADD	t14, t14, a5, b8
+
+	LD	a7,   3 * SIZE(AO)
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1				#	3rd compute
+	MADD	t12, t12, a3, b2
+	MADD	t13, t13, a3, b3
+	MADD	t14, t14, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5				#	4th compute
+	MADD	t12, t12, a7, b6
+	MADD	t13, t13, a7, b7
+	MADD	t14, t14, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L52
+	nop
+
+
+	.align 3
+.L55:
+	andi	L, KK,  3
+	blez	L, .L58
+	nop
+	
+	.align	3
+.L56:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L56
+	nop
+
+
+.L58:										#	deal with the triangular part
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	MUL	t11, b1, t11
+	MUL	t12, b1, t12
+	MUL	t13, b1, t13
+	MUL	t14, b1, t14
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 1 * SIZE
+	daddiu	CO2, CO2, 1 * SIZE
+	daddiu	CO3, CO3, 1 * SIZE
+	daddiu	CO4, CO4, 1 * SIZE
+
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, BASE_SHIFT				#	mr=1
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L							#	mov AO to the end of Ai
+	daddu	BO, BO, TEMP						#	mov BO to the end of Bj
+
+	daddiu	KK, KK, 1							#	the length of rectangular data part increases by 2
+
+	.align 3
+.L29:
+	move	B,  BO								#	fixed panel Bj
+	bgtz	J, .L10
+	nop
+	
+	
+	.align 3
+.L30:
+	andi	J,  N, 2							#	nr=2
+	blez	J, .L70
+	nop
+
+	move	CO1, C
+	daddu	CO2, C,   LDC
+
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	move	KK, OFFSET							
+	move	AO, A								#	reset A
+	daddu	C,  CO2, LDC						#	fixed 
+
+	dsra	I,  M, 2							#	I = mc/4
+	blez	I, .L40
+	nop
+
+.L31:
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+	LD	b2,  1 * SIZE(B)
+
+	dsra	L,  KK, 2						#	L=kk/4
+	blez	L, .L35
+	move	BO,  B							#	reset B
+
+
+	.align	3
+.L32:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+	LD	b3,   4 * SIZE(BO)
+	LD	b4,   5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3				#	3rd compute
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+	MADD	t12, t12, a1, b4
+	MADD	t22, t22, a2, b4
+	MADD	t32, t32, a3, b4
+	MADD	t42, t42, a4, b4
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7				#	4th compute
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+	MADD	t12, t12, a5, b8
+	MADD	t22, t22, a6, b8
+	MADD	t32, t32, a7, b8
+	MADD	t42, t42, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L32
+	nop
+
+	
+	.align 3
+
+.L35:
+	andi	L, KK,  3
+	blez	L, .L38
+	nop
+	
+	.align	3
+.L36:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+	
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L36
+	nop
+
+
+.L38:										#
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t21, b3, t21
+	SUB	t22, b4, t22
+	SUB	t31, b5, t31
+	SUB	t32, b6, t32
+	SUB	t41, b7, t41
+	SUB	t42, b8, t42
+
+	LD	a1,  0 * SIZE(AO)					#	sa stores in col major		
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	MUL	t11, a1, t11
+	MUL	t12, a1, t12
+	NMSUB	t21, t21, a2, t11
+	NMSUB	t22, t22, a2, t12
+	NMSUB	t31, t31, a3, t11
+	NMSUB	t32, t32, a3, t12
+	NMSUB	t41, t41, a4, t11
+	NMSUB	t42, t42, a4, t12
+
+	
+	LD	a5,  5 * SIZE(AO)
+	LD	a6,  6 * SIZE(AO)
+	LD	a7,  7 * SIZE(AO)
+	MUL	t21, a5, t21
+	MUL	t22, a5, t22
+	NMSUB	t31, t31, a6, t21
+	NMSUB	t32, t32, a6, t22
+	NMSUB	t41, t41, a7, t21
+	NMSUB	t42, t42, a7, t22
+
+
+	LD	a8, 10 * SIZE(AO)
+	LD	a1, 11 * SIZE(AO)
+	MUL	t31, a8, t31
+	MUL	t32, a8, t32
+	NMSUB	t41, t41, a1, t31
+	NMSUB	t42, t42, a1, t32
+
+
+	LD	a2, 15 * SIZE(AO)
+	MUL	t41, a2, t41
+	MUL	t42, a2, t42
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t21,  2 * SIZE(BO)
+	ST	t22,  3 * SIZE(BO)
+	ST	t31,  4 * SIZE(BO)
+	ST	t32,  5 * SIZE(BO)
+	ST	t41,  6 * SIZE(BO)
+	ST	t42,  7 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 4 * SIZE
+	daddiu	CO2, CO2, 4 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AO, L						#	move AO to the end of Ai
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 4						#	
+
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+
+	daddiu	I, I, -1
+	bgtz	I, .L31
+	nop
+
+
+	.align 3
+.L40:
+	andi	I,  M, 2
+	blez	I, .L60
+	nop
+
+	MOV	t12, t11							#	clear result registers
+	MOV	t22, t21
+	MOV	t32, t31
+	MOV	t42, t41
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(B)
+	LD	b2,  1 * SIZE(B)
+
+	dsra	L,  KK, 2
+	blez	L, .L45
+	move	BO,  B							#	reset B	
+
+
+	.align	3
+.L42:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t21, t21, a4, b3
+	MADD	t12, t12, a3, b4
+	MADD	t22, t22, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t21, t21, a8, b7
+	MADD	t12, t12, a7, b8
+	MADD	t22, t22, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L42
+	nop
+	
+	.align 3
+
+.L45:
+	andi	L, KK,  3
+	blez	L, .L48
+	nop
+	
+	.align	3
+.L46:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L46
+	nop
+
+.L48:
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t21, b3, t21
+	SUB	t22, b4, t22
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  1 * SIZE(AO)
+	MUL	t11, b1, t11
+	MUL	t12, b1, t12
+	NMSUB	t21, t21, b2, t11
+	NMSUB	t22, t22, b2, t12
+	
+	LD	b3,  3 * SIZE(AO)
+	MUL	t21, b3, t21
+	MUL	t22, b3, t22
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t21,  2 * SIZE(BO)
+	ST	t22,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 2 * SIZE
+	daddiu	CO2, CO2, 2 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 2
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+
+
+	.align 3
+.L60:
+	andi	I,  M, 1						#	mr=1
+	blez	I, .L49
+	nop
+
+	MOV	t12, t11							#	clear result registers
+	MOV	t22, t21
+	MOV	t32, t31
+	MOV	t42, t41
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(B)
+	LD	b2,  1 * SIZE(B)
+
+	dsra	L,  KK, 2
+	blez	L, .L65
+	move	BO,  B							#	reset B	
+
+
+	.align	3
+.L62:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t12, t12, a1, b2
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t12, t12, a5, b6
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t12, t12, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t12, t12, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L62
+	nop
+	
+	.align 3
+
+.L65:
+	andi	L, KK,  3
+	blez	L, .L68
+	nop
+	
+	.align	3
+.L66:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t12, t12, a1, b2
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 1mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L66
+	nop
+
+.L68:
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	MUL	t11, b1, t11
+	MUL	t12, b1, t12
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 1 * SIZE
+	daddiu	CO2, CO2, 1 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, BASE_SHIFT		#	mr=1
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 1
+
+	.align 3
+.L49:
+	move	B,  BO
+	
+	.align 3
+
+.L70:
+	andi	J,  N, 1					#	nr=1
+	blez	J, .L999					#	END
+	nop
+
+	move	CO1, C
+
+	move	KK, OFFSET
+	move	AO, A
+
+	dsra	I,  M, 2
+	blez	I, .L80
+	nop
+
+.L71:
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+
+	dsra	L,  KK, 2
+	blez	L, .L75
+	move	BO,  B							#	reset B
+
+	.align	3
+.L72:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3				#	3rd compute
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7				#	4th compute
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L72
+	nop
+	
+	.align 3
+
+.L75:
+	andi	L, KK,  3
+	blez	L, .L78
+	nop
+
+	.align	3
+.L76:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L76
+	nop
+
+.L78:
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+	LD	a1,  0 * SIZE(AO)					#	sa stores in col major		
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	MUL	t11, a1, t11
+	NMSUB	t21, t21, a2, t11
+	NMSUB	t31, t31, a3, t11
+	NMSUB	t41, t41, a4, t11
+
+	LD	a5,  5 * SIZE(AO)
+	LD	a6,  6 * SIZE(AO)
+	LD	a7,  7 * SIZE(AO)
+	MUL	t21, a5, t21
+	NMSUB	t31, t31, a6, t21
+	NMSUB	t41, t41, a7, t21
+
+	LD	a8, 10 * SIZE(AO)
+	LD	a1, 11 * SIZE(AO)
+	MUL	t31, a8, t31
+	NMSUB	t41, t41, a1, t31
+
+	LD	a2, 15 * SIZE(AO)
+	MUL	t41, a2, t41
+
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t21,  1 * SIZE(BO)
+	ST	t31,  2 * SIZE(BO)
+	ST	t41,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+
+	daddiu	CO1, CO1, 4 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 4
+	daddiu	I, I, -1
+	bgtz	I, .L71
+	nop
+
+
+	.align 3
+
+.L80:
+	andi	I,  M, 2
+	blez	I, .L90
+	NOP
+
+	MTC	$0,  t11
+	MOV	t21, t11							#	clear result registers
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(B)
+
+	dsra	L,  KK, 2
+	blez	L, .L85
+	move	BO,  B
+	
+	.align	3
+.L82:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t21, t21, a4, b3
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t21, t21, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L82
+	nop
+	
+	.align 3
+
+.L85:
+	andi	L, KK,  3
+	blez	L, .L88
+	nop
+
+	.align	3
+.L86:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L86
+	nop
+
+
+.L88:
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  1 * SIZE(AO)
+	MUL	t11, b1, t11
+	NMSUB	t21, t21, b2, t11
+	
+	LD	b3,  3 * SIZE(AO)
+	MUL	t21, b3, t21
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t21,  1 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+
+
+	daddiu	CO1, CO1, 2 * SIZE
+	
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 2
+
+	
+	.align 3
+.L90:
+	andi	I,  M, 1					#	mr=1
+	blez	I, .L89
+	NOP
+
+	MTC	$0,  t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(B)
+
+	dsra	L,  KK, 2
+	blez	L, .L95
+	move	BO,  B
+	
+	.align	3
+.L92:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+
+	daddiu	L, L, -1
+	bgtz	L, .L92
+	nop
+	
+	.align 3
+.L95:
+	andi	L, KK,  3
+	blez	L, .L98
+	nop
+
+	.align	3
+.L96:
+	MADD	t11, t11, a1, b1				#	3rd compute
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L96
+	nop
+
+
+.L98:
+	LD	b1,  0 * SIZE(BO)
+
+	SUB	t11, b1, t11
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	MUL	t11, b1, t11
+	
+	ST	t11,  0 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+
+
+	daddiu	CO1, CO1, 1 * SIZE
+	
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP,  BASE_SHIFT
+	dsll	TEMP, TEMP,  BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 1
+
+
+	.align 3
+.L89:
+	move	B,  BO
+
+
+	.align 3
+
+.L999:
+	LDARG	$16,   0($sp)
+	LDARG	$17,   8($sp)
+	LDARG	$18,  16($sp)
+	LDARG	$19,  24($sp)
+	LDARG	$20,  32($sp)
+	LDARG	$21,  40($sp)
+	ldc1	$f24, 48($sp)
+	ldc1	$f25, 56($sp)
+	ldc1	$f26, 64($sp)
+	ldc1	$f27, 72($sp)
+	ldc1	$f28, 80($sp)
+
+	LDARG	$22,  88($sp)
+	LDARG	$23,  96($sp)
+	LDARG	$24, 104($sp)
+	LDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	ldc1	$f20,112($sp)
+	ldc1	$f21,120($sp)
+	ldc1	$f22,128($sp)
+	ldc1	$f23,136($sp)
+#endif
+
+	j	$31
+	daddiu	$sp, $sp, 144
+
+	EPILOGUE
diff --git a/kernel/mips64/trsm_kernel_RN_loongson3a.S b/kernel/mips64/trsm_kernel_RN_loongson3a.S
new file mode 100644
index 000000000..790d7c981
--- /dev/null
+++ b/kernel/mips64/trsm_kernel_RN_loongson3a.S
@@ -0,0 +1,1852 @@
+#define REALNAME ASMNAME
+
+#define ASSEMBLER
+#include "common.h"
+
+
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$8
+#define B	$9
+#define C	$10
+#define LDC	$11
+
+#define AO	$12
+#define BO	$13
+
+#define I	$2
+#define J	$3
+#define L	$7
+
+#define CO1	$14
+#define CO2	$15
+#define CO3	$16
+#define CO4	$17
+
+#define OFFSET	$22
+#define KK	$23
+#define TEMP	$24
+#define AORIG	$25
+
+#define a1	$f0
+#define a2	$f1
+#define a3	$f26
+#define a4	$f27
+
+#define a5	$f28
+#define a6	$f29
+#define	a7	$f30
+#define	a8	$f31
+
+#define b1	$f2
+#define b2	$f3
+#define b3	$f4
+#define b4	$f5
+
+#define b5	$f6
+#define b6	$f7
+#define b7	$f8
+#define b8	$f9
+
+#define t11	$f10
+#define t21	$f11
+#define t31	$f12
+#define	t41	$f13
+
+#define t12	$f14
+#define	t22	$f15
+#define t32	$f16
+#define	t42	$f17
+
+#define	t13	$f18
+#define	t23	$f19
+#define	t33	$f20
+#define	t43	$f21
+
+#define	t14	$f22
+#define	t24	$f23
+#define	t34	$f24
+#define t44	$f25
+
+	PROLOGUE
+	
+	daddiu	$sp, $sp, -144
+
+	SDARG	$16,   0($sp)
+	SDARG	$17,   8($sp)
+	SDARG	$18,  16($sp)
+	SDARG	$19,  24($sp)
+	SDARG	$20,  32($sp)
+	SDARG	$21,  40($sp)
+	sdc1	$f24, 48($sp)
+	sdc1	$f25, 56($sp)
+	sdc1	$f26, 64($sp)
+	sdc1	$f27, 72($sp)
+	sdc1	$f28, 80($sp)
+
+	SDARG	$22,  88($sp)
+	SDARG	$23,  96($sp)
+	SDARG	$24, 104($sp)
+	SDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	sdc1	$f20,112($sp)
+	sdc1	$f21,120($sp)
+	sdc1	$f22,128($sp)
+	sdc1	$f23,136($sp)
+#endif
+
+											#	RN compute from top to bottom left to right
+	.align	3
+	LDARG	OFFSET, 144($sp)				#	get the last parameter
+	dsll	LDC, LDC, BASE_SHIFT			#	LDC * data_Byte
+
+	neg	KK, OFFSET							#	for RN OFFSET always 0
+
+	dsra	J,  N, 2						#	J = NC/4 
+	blez	J, .L30
+	NOP
+
+.L10:
+	daddiu	J, J, -1
+	
+	move	CO1, C
+	daddu	CO2, C,   LDC
+	daddu	CO3, CO2, LDC
+	daddu	CO4, CO3, LDC
+	
+	move	AO, A							#	A is the retangular matrix and B is the trigular matrix	
+	daddu	C,  CO4, LDC					#	Fixed pointer C
+
+	dsra	I,  M, 2						#	I=MC/4
+	blez	I, .L20
+	NOP
+
+	.align 3
+.L11:
+	MTC	$0,  t11							#	clear results registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+	
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+	LD	a2,  1 * SIZE(AO)					#	get 4 a	
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+	LD	b2,  1 * SIZE(B)					#	get 4 b
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L15
+	move	BO,  B							#	reset B
+
+.L12:
+	LD	a5,  4 * SIZE(AO)
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4			#	fisrt 	
+
+	LD	a1,  8 * SIZE(AO)
+	LD	a2,  9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			#	second
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4			#	third
+
+	daddiu	AO, AO, 16 * SIZE			#	AO += 4mr*4kr	
+	daddiu	BO, BO, 16 * SIZE			#	BP += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			#	fouth
+
+	daddiu	L, L, -1
+	bgtz	L, .L12
+	NOP
+
+
+.L15:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L18
+	NOP
+
+	.align	3
+.L16:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO, 4 * SIZE			#	AO += 4mr	
+	daddiu	BO, BO, 4 * SIZE			#	BP += 4nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L16
+	NOP
+
+
+	.align 3
+.L18:									#	.L18 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+	LD	b2,  1 * SIZE(AO)				#	Fixed results
+	LD	b3,  2 * SIZE(AO)				
+	LD	b4,  3 * SIZE(AO)				#	sa stored as col major
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+ 	LD	b5,  4 * SIZE(AO)
+	LD	b6,  5 * SIZE(AO)
+	LD	b7,  6 * SIZE(AO)
+	LD	b8,  7 * SIZE(AO)
+
+	SUB	t12, b5, t12
+	SUB	t22, b6, t22
+	SUB	t32, b7, t32
+	SUB	t42, b8, t42
+
+ 	LD	b1,  8 * SIZE(AO)
+	LD	b2,  9 * SIZE(AO)
+	LD	b3, 10 * SIZE(AO)
+	LD	b4, 11 * SIZE(AO)
+
+	SUB	t13, b1, t13
+	SUB	t23, b2, t23
+	SUB	t33, b3, t33
+	SUB	t43, b4, t43
+
+ 	LD	b5, 12 * SIZE(AO)
+	LD	b6, 13 * SIZE(AO)
+	LD	b7, 14 * SIZE(AO)
+	LD	b8, 15 * SIZE(AO)
+
+	SUB	t14, b5, t14
+	SUB	t24, b6, t24
+	SUB	t34, b7, t34
+	SUB	t44, b8, t44
+
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	MUL	t11, b1, t11
+	MUL	t21, b1, t21
+	MUL	t31, b1, t31
+	MUL	t41, b1, t41
+	NMSUB	t12, t12, b2, t11
+	NMSUB	t22, t22, b2, t21
+	NMSUB	t32, t32, b2, t31
+	NMSUB	t42, t42, b2, t41
+	NMSUB	t13, t13, b3, t11
+	NMSUB	t23, t23, b3, t21
+	NMSUB	t33, t33, b3, t31
+	NMSUB	t43, t43, b3, t41
+	NMSUB	t14, t14, b4, t11
+	NMSUB	t24, t24, b4, t21
+	NMSUB	t34, t34, b4, t31
+	NMSUB	t44, t44, b4, t41
+
+	
+	LD	b5,  5 * SIZE(BO)
+	LD	b6,  6 * SIZE(BO)
+	LD	b7,  7 * SIZE(BO)
+	MUL	t12, b5, t12
+	MUL	t22, b5, t22
+	MUL	t32, b5, t32
+	MUL	t42, b5, t42
+	NMSUB	t13, t13, b6, t12
+	NMSUB	t23, t23, b6, t22
+	NMSUB	t33, t33, b6, t32
+	NMSUB	t43, t43, b6, t42
+	NMSUB	t14, t14, b7, t12
+	NMSUB	t24, t24, b7, t22
+	NMSUB	t34, t34, b7, t32
+	NMSUB	t44, t44, b7, t42
+
+
+
+	LD	b8,  10 * SIZE(BO)
+	LD	b1,  11 * SIZE(BO)
+	MUL	t13, b8, t13
+	MUL	t23, b8, t23
+	MUL	t33, b8, t33
+	MUL	t43, b8, t43
+	NMSUB	t14, t14, b1, t13
+	NMSUB	t24, t24, b1, t23
+	NMSUB	t34, t34, b1, t33
+	NMSUB	t44, t44, b1, t43
+
+
+
+	LD	b2,  15 * SIZE(BO)
+	MUL	t14, b2, t14
+	MUL	t24, b2, t24
+	MUL	t34, b2, t34
+	MUL	t44, b2, t44
+
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+	ST	t21,  1 * SIZE(AO)
+	ST	t31,  2 * SIZE(AO)
+	ST	t41,  3 * SIZE(AO)
+
+	ST	t12,  4 * SIZE(AO)
+	ST	t22,  5 * SIZE(AO)
+	ST	t32,  6 * SIZE(AO)
+	ST	t42,  7 * SIZE(AO)
+
+	ST	t13,  8 * SIZE(AO)
+	ST	t23,  9 * SIZE(AO)
+	ST	t33, 10 * SIZE(AO)
+	ST	t43, 11 * SIZE(AO)
+
+	ST	t14, 12 * SIZE(AO)
+	ST	t24, 13 * SIZE(AO)
+	ST	t34, 14 * SIZE(AO)
+	ST	t44, 15 * SIZE(AO)
+
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+	
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t33,  2 * SIZE(CO3)
+	ST	t43,  3 * SIZE(CO3)
+	
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+	ST	t34,  2 * SIZE(CO4)
+	ST	t44,  3 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 4 * SIZE			#	fixed address
+	daddiu	CO2, CO2, 4 * SIZE
+	daddiu	CO3, CO3, 4 * SIZE
+	daddiu	CO4, CO4, 4 * SIZE
+
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+	daddiu	I, I, -1
+	bgtz	I, .L11
+	NOP
+
+	.align 3
+.L20:									
+	andi	I,  M, 2					#	mr=2
+	blez	I, .L50
+	nop
+
+	MTC	$0,  t11							#	clear results registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+	
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+	LD	a2,  1 * SIZE(AO)					#	get 4 a	
+
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+	LD	b2,  1 * SIZE(B)					#	get 4 b
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L25
+	move	BO,  B							#	reset B
+
+.L22:
+	LD	a5,  2 * SIZE(AO)
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1
+	MADD	t21, t21, a4, b1
+
+	MADD	t12, t12, a3, b2
+	MADD	t22, t22, a4, b2
+
+	MADD	t13, t13, a3, b3
+	MADD	t23, t23, a4, b3
+
+	MADD	t14, t14, a3, b4
+	MADD	t24, t24, a4, b4
+
+	daddiu	AO, AO, 8 * SIZE			#	AO += 2mr*4kr	
+	daddiu	BO, BO, 16 * SIZE			#	BP += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5
+	MADD	t21, t21, a8, b5
+
+	MADD	t12, t12, a7, b6
+	MADD	t22, t22, a8, b6
+
+	MADD	t13, t13, a7, b7
+	MADD	t23, t23, a8, b7
+
+	MADD	t14, t14, a7, b8
+	MADD	t24, t24, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L22
+	NOP
+
+
+.L25:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L28
+	NOP
+
+	.align	3
+.L26:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	daddiu	AO, AO, 2 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 4 * SIZE			#	BP += 4nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L26
+	NOP
+
+
+	.align 3
+.L28:									#	.L18 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+	LD	b2,  1 * SIZE(AO)				#	Fixed results
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+ 	LD	b5,  2 * SIZE(AO)
+	LD	b6,  3 * SIZE(AO)
+
+	SUB	t12, b5, t12
+	SUB	t22, b6, t22
+
+	LD	b3,  4 * SIZE(AO)
+	LD	b4,  5 * SIZE(AO)
+
+	SUB	t13, b3, t13
+	SUB	t23, b4, t23
+
+	LD	b7,  6 * SIZE(AO)
+	LD	b8,  7 * SIZE(AO)
+
+	SUB	t14, b7, t14
+	SUB	t24, b8, t24
+
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	MUL	t11, b1, t11
+	MUL	t21, b1, t21
+	NMSUB	t12, t12, b2, t11
+	NMSUB	t22, t22, b2, t21
+	NMSUB	t13, t13, b3, t11
+	NMSUB	t23, t23, b3, t21
+	NMSUB	t14, t14, b4, t11
+	NMSUB	t24, t24, b4, t21
+
+	
+	LD	b5,  5 * SIZE(BO)
+	LD	b6,  6 * SIZE(BO)
+	LD	b7,  7 * SIZE(BO)
+	MUL	t12, b5, t12
+	MUL	t22, b5, t22
+	NMSUB	t13, t13, b6, t12
+	NMSUB	t23, t23, b6, t22
+	NMSUB	t14, t14, b7, t12
+	NMSUB	t24, t24, b7, t22
+
+
+
+	LD	b8,  10 * SIZE(BO)
+	LD	b1,  11 * SIZE(BO)
+	MUL	t13, b8, t13
+	MUL	t23, b8, t23
+	NMSUB	t14, t14, b1, t13
+	NMSUB	t24, t24, b1, t23
+
+
+
+	LD	b2,  15 * SIZE(BO)
+	MUL	t14, b2, t14
+	MUL	t24, b2, t24
+
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+	ST	t21,  1 * SIZE(AO)
+
+	ST	t12,  2 * SIZE(AO)
+	ST	t22,  3 * SIZE(AO)
+
+	ST	t13,  4 * SIZE(AO)
+	ST	t23,  5 * SIZE(AO)
+
+	ST	t14,  6 * SIZE(AO)
+	ST	t24,  7 * SIZE(AO)
+
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	ST	t21,  1 * SIZE(CO1)
+	
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 2 * SIZE			#	fixed address
+	daddiu	CO2, CO2, 2 * SIZE			#	mr=2
+	daddiu	CO3, CO3, 2 * SIZE
+	daddiu	CO4, CO4, 2 * SIZE
+
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, 1 + BASE_SHIFT	#	mr=2
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+	.align 3
+.L50:									
+	andi	I,  M, 1					#	mr=1
+	blez	I, .L29
+	nop
+
+	MTC	$0,  t11							#	clear results registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+	
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+	LD	b2,  1 * SIZE(B)					#	get 4 b
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L55
+	move	BO,  B							#	reset B
+
+.L52:
+	LD	a5,  1 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	LD	a3,  2 * SIZE(AO)
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t12, t12, a5, b6
+	MADD	t13, t13, a5, b7
+	MADD	t14, t14, a5, b8
+
+	LD	a7,  3 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1
+	MADD	t12, t12, a3, b2
+	MADD	t13, t13, a3, b3
+	MADD	t14, t14, a3, b4
+
+	daddiu	AO, AO, 4 * SIZE			#	AO += 1mr*4kr	
+	daddiu	BO, BO, 16 * SIZE			#	BP += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5
+	MADD	t12, t12, a7, b6
+	MADD	t13, t13, a7, b7
+	MADD	t14, t14, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L52
+	NOP
+
+
+.L55:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L58
+	NOP
+
+	.align	3
+.L56:
+	MADD	t11, t11, a1, b1
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	daddiu	AO, AO, 1 * SIZE			#	AO += 1mr	
+	daddiu	BO, BO, 4 * SIZE			#	BP += 4nr
+	
+	LD	a1,  0 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L56
+	NOP
+
+
+	.align 3
+.L58:									#	.L18 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+ 	LD	b5,  1 * SIZE(AO)
+	LD	b3,  2 * SIZE(AO)
+	LD	b7,  3 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b5, t12
+	SUB	t13, b3, t13
+	SUB	t14, b7, t14
+
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	MUL	t11, b1, t11
+	NMSUB	t12, t12, b2, t11
+	NMSUB	t13, t13, b3, t11
+	NMSUB	t14, t14, b4, t11
+
+	
+	LD	b5,  5 * SIZE(BO)
+	LD	b6,  6 * SIZE(BO)
+	LD	b7,  7 * SIZE(BO)
+	MUL	t12, b5, t12
+	NMSUB	t13, t13, b6, t12
+	NMSUB	t14, t14, b7, t12
+
+
+	LD	b8,  10 * SIZE(BO)
+	LD	b1,  11 * SIZE(BO)
+	MUL	t13, b8, t13
+	NMSUB	t14, t14, b1, t13
+
+
+	LD	b2,  15 * SIZE(BO)
+	MUL	t14, b2, t14
+
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+	ST	t12,  1 * SIZE(AO)
+	ST	t13,  2 * SIZE(AO)
+	ST	t14,  3 * SIZE(AO)
+
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	ST	t12,  0 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 1 * SIZE			#	fixed address
+	daddiu	CO2, CO2, 1 * SIZE			#	mr=2
+	daddiu	CO3, CO3, 1 * SIZE
+	daddiu	CO4, CO4, 1 * SIZE
+
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, BASE_SHIFT	#	mr=2
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+
+	.align 3
+.L29:
+	move	B,  BO						#	change to next panel of Bj
+	daddiu	KK, KK,  4					#	rectangular data length increase by 4	
+	bgtz	J, .L10
+	NOP
+
+
+	.align 3
+	
+.L30:
+	andi	J,  N, 2
+	blez	J, .L70
+	nop
+
+	move	CO1, C
+	daddu	CO2, C,   LDC
+	
+	move	AO, A							#	A is the retangular matrix and B is the trigular matrix	
+	daddu	C,  CO2, LDC					#	Fixed pointer C
+
+	dsra	I,  M, 2						#	I=MC/4
+	blez	I, .L40
+	NOP
+
+	.align 3
+.L31:
+	MTC	$0,  t11							#	clear results registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+	LD	a2,  1 * SIZE(AO)					#	get 4 a	
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+	LD	b2,  1 * SIZE(B)					#	get 4 b
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L35
+	move	BO,  B							#	reset B
+
+.L32:
+	LD	a5,  4 * SIZE(AO)
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	LD	a1,  8 * SIZE(AO)
+	LD	a2,  9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+
+	MADD	t12, t12, a1, b4
+	MADD	t22, t22, a2, b4
+	MADD	t32, t32, a3, b4
+	MADD	t42, t42, a4, b4
+
+	daddiu	AO, AO, 16 * SIZE			#	AO += 4mr*4kr	
+	daddiu	BO, BO,  8 * SIZE			#	BP += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+
+	MADD	t12, t12, a5, b8
+	MADD	t22, t22, a6, b8
+	MADD	t32, t32, a7, b8
+	MADD	t42, t42, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L32
+	NOP
+
+
+.L35:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L38
+	NOP
+
+	.align	3
+.L36:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	daddiu	AO, AO, 4 * SIZE			#	AO += 4mr	
+	daddiu	BO, BO, 2 * SIZE			#	BP += 2nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L36
+	NOP
+
+
+	.align 3
+.L38:									#	.L38 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+	LD	b2,  1 * SIZE(AO)				#	Fixed results
+	LD	b3,  2 * SIZE(AO)				
+	LD	b4,  3 * SIZE(AO)				#	sa stored as col major
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+ 	LD	b5,  4 * SIZE(AO)
+	LD	b6,  5 * SIZE(AO)
+	LD	b7,  6 * SIZE(AO)
+	LD	b8,  7 * SIZE(AO)
+
+	SUB	t12, b5, t12
+	SUB	t22, b6, t22
+	SUB	t32, b7, t32
+	SUB	t42, b8, t42
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	LD	b2,  1 * SIZE(BO)
+	MUL	t11, b1, t11
+	MUL	t21, b1, t21
+	MUL	t31, b1, t31
+	MUL	t41, b1, t41
+	NMSUB	t12, t12, b2, t11
+	NMSUB	t22, t22, b2, t21
+	NMSUB	t32, t32, b2, t31
+	NMSUB	t42, t42, b2, t41
+	
+	LD	b5,  3 * SIZE(BO)
+	MUL	t12, b5, t12
+	MUL	t22, b5, t22
+	MUL	t32, b5, t32
+	MUL	t42, b5, t42
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+	ST	t21,  1 * SIZE(AO)
+	ST	t31,  2 * SIZE(AO)
+	ST	t41,  3 * SIZE(AO)
+
+	ST	t12,  4 * SIZE(AO)
+	ST	t22,  5 * SIZE(AO)
+	ST	t32,  6 * SIZE(AO)
+	ST	t42,  7 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 4 * SIZE			#	fixed address
+	daddiu	CO2, CO2, 4 * SIZE
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT	#	nr=2
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+	daddiu	I, I, -1
+	bgtz	I, .L31
+	NOP
+
+	.align 3
+.L40:
+	andi	I, M,2
+	blez	I,.L60
+	nop
+
+	MTC	$0,  t11							#	clear results registers
+	MOV	t21, t11
+
+	MOV	t12, t11
+	MOV	t22, t11
+
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+	LD	a2,  1 * SIZE(AO)					#	get 4 a	
+
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+	LD	b2,  1 * SIZE(B)					#	get 4 b
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L45
+	move	BO,  B							#	reset B
+
+.L42:
+	LD	a5,  2 * SIZE(AO)
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3
+	MADD	t21, t21, a4, b3
+	MADD	t12, t12, a3, b4
+	MADD	t22, t22, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE			#	AO += 2mr*4kr	
+	daddiu	BO, BO,  8 * SIZE			#	BP += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7
+	MADD	t21, t21, a8, b7
+	MADD	t12, t12, a7, b8
+	MADD	t22, t22, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L42
+	NOP
+
+
+.L45:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L48
+	NOP
+
+	.align	3
+.L46:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	daddiu	AO, AO, 2 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 2 * SIZE			#	BP += 2nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L46
+	NOP
+
+
+	.align 3
+.L48:									#	.L48 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+	LD	b2,  1 * SIZE(AO)				#	Fixed results
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+ 	LD	b5,  2 * SIZE(AO)
+	LD	b6,  3 * SIZE(AO)
+
+	SUB	t12, b5, t12
+	SUB	t22, b6, t22
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	LD	b2,  1 * SIZE(BO)
+	MUL	t11, b1, t11
+	MUL	t21, b1, t21
+	NMSUB	t12, t12, b2, t11
+	NMSUB	t22, t22, b2, t21
+	
+	LD	b5,  3 * SIZE(BO)
+	MUL	t12, b5, t12
+	MUL	t22, b5, t22
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+	ST	t21,  1 * SIZE(AO)
+	ST	t12,  2 * SIZE(AO)
+	ST	t22,  3 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 2 * SIZE			#	fixed address
+	daddiu	CO2, CO2, 2 * SIZE
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT	#	nr=2
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+
+	.align 3
+.L60:
+	andi	I,M,1						#	nr=2 mr=1
+	blez	I,.L39
+	nop
+
+	MTC	$0,  t11							#	clear results registers
+	MOV	t12, t11
+
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+	LD	b2,  1 * SIZE(B)					#	get 4 b
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L65
+	move	BO,  B							#	reset B
+
+.L62:
+	LD	a5,  1 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t12, t12, a1, b2
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t12, t12, a5, b6
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3
+	MADD	t12, t12, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE			#	AO += 1mr*4kr	
+	daddiu	BO, BO,  8 * SIZE			#	BP += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7
+	MADD	t12, t12, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L62
+	NOP
+
+
+.L65:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L68
+	NOP
+
+	.align	3
+.L66:
+	MADD	t11, t11, a1, b1
+	MADD	t12, t12, a1, b2
+
+	daddiu	AO, AO, 1 * SIZE			#	AO += mr	
+	daddiu	BO, BO, 2 * SIZE			#	BP += 2nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L66
+	NOP
+
+
+	.align 3
+.L68:									#	.L48 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+	LD	b5,  1 * SIZE(AO)				#	Fixed results
+
+	SUB	t11, b1, t11
+	SUB	t12, b5, t12
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	LD	b2,  1 * SIZE(BO)
+	MUL	t11, b1, t11
+	NMSUB	t12, t12, b2, t11
+	
+	LD	b5,  3 * SIZE(BO)
+	MUL	t12, b5, t12
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+	ST	t12,  1 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	ST	t12,  0 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 1 * SIZE			#	fixed address
+	daddiu	CO2, CO2, 1 * SIZE
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, BASE_SHIFT		#	mr=1
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT	#	nr=2
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+
+	.align 3
+.L39:
+	move	B,  BO						#	change to next panel of Bj
+	daddiu	KK, KK,  2					#	rectangular data length increase by 4	
+
+
+
+	.align 3
+
+.L70:
+	andi	J,  N, 1					#	nr=1
+	blez	J, .L999
+	NOP
+
+	move	CO1, C
+	move	AO, A
+
+	daddu	C,  CO1, LDC
+
+	dsra	I,  M, 2					#	I=MC/4
+	blez	I, .L80
+	NOP
+
+	.align 3
+.L71:
+	MTC	$0,  t11							#	clear results registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+	LD	a2,  1 * SIZE(AO)					#	get 4 a	
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L75
+	move	BO,  B							#	reset B
+
+.L72:
+	LD	a5,  4 * SIZE(AO)
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	LD	a1,  8 * SIZE(AO)
+	LD	a2,  9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+
+	daddiu	AO, AO, 16 * SIZE			#	AO += 4mr*4kr	
+	daddiu	BO, BO,  4 * SIZE			#	BP += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L72
+	NOP
+
+
+.L75:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L78
+	NOP
+
+	.align	3
+.L76:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	daddiu	AO, AO, 4 * SIZE			#	AO += 4mr	
+	daddiu	BO, BO, 1 * SIZE			#	BP += 1nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L76
+	NOP
+
+
+	.align 3
+.L78:									#	.L78 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+	LD	b2,  1 * SIZE(AO)				#	Fixed results
+	LD	b3,  2 * SIZE(AO)				
+	LD	b4,  3 * SIZE(AO)				#	sa stored as col major
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	MUL	t11, b1, t11
+	MUL	t21, b1, t21
+	MUL	t31, b1, t31
+	MUL	t41, b1, t41
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+	ST	t21,  1 * SIZE(AO)
+	ST	t31,  2 * SIZE(AO)
+	ST	t41,  3 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	
+
+	daddiu	CO1, CO1, 4 * SIZE			#	fixed address
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, BASE_SHIFT		#	nr=1
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+	daddiu	I, I, -1
+	bgtz	I, .L71
+	NOP
+
+
+	.align 3
+.L80:
+	andi	I, M, 2						#	mr=2
+	blez	I, .L90
+	nop
+
+	MTC	$0,  t11							#	clear results registers
+	MOV	t21, t11
+
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+	LD	a2,  1 * SIZE(AO)					#	get 4 a	
+
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L85
+	move	BO,  B							#	reset B
+
+.L82:
+	LD	a5,  2 * SIZE(AO)
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3
+	MADD	t21, t21, a4, b3
+
+	daddiu	AO, AO,  8 * SIZE			#	AO += 2mr*4kr	
+	daddiu	BO, BO,  4 * SIZE			#	BP += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7
+	MADD	t21, t21, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L82
+	NOP
+
+
+.L85:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L88
+	NOP
+
+	.align	3
+.L86:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	daddiu	AO, AO, 2 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 1 * SIZE			#	BP += 1nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L86
+	NOP
+
+
+	.align 3
+.L88:									#	.L88 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+	LD	b2,  1 * SIZE(AO)				#	Fixed results
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	MUL	t11, b1, t11
+	MUL	t21, b1, t21
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+	ST	t21,  1 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	ST	t21,  1 * SIZE(CO1)
+	
+
+	daddiu	CO1, CO1, 2 * SIZE			#	fixed address
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, BASE_SHIFT		#	nr=1
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+
+	.align 3
+.L90:
+	andi	I, M, 1						#	mr=1
+	blez	I, .L79
+	nop
+
+	MTC	$0,  t11							#	clear results registers
+
+	LD	a1,  0 * SIZE(AO)					#	AO point to the beginning address of sa
+	LD	b1,  0 * SIZE(B)					#	B point to the beginning address of every panel Bj
+
+	dsra	L,  KK, 2						#	L=KK/4, KK is the length of the retangular data part of Bj
+	blez	L, .L95
+	move	BO,  B							#	reset B
+
+.L92:
+	LD	a5,  1 * SIZE(AO)
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3
+
+	daddiu	AO, AO,  4 * SIZE			#	AO += 1mr*4kr	
+	daddiu	BO, BO,  4 * SIZE			#	BP += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L92
+	NOP
+
+
+.L95:
+	andi	L, KK, 3					#	deal with kc remainder part
+	blez	L, .L98
+	NOP
+
+	.align	3
+.L96:
+	MADD	t11, t11, a1, b1
+
+	daddiu	AO, AO, 1 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 1 * SIZE			#	BP += 1nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L96
+	NOP
+
+
+	.align 3
+.L98:									#	.L98 always deal with the trigular data part
+ 	LD	b1,  0 * SIZE(AO)				#	for RN & RT A is the result matrix
+
+	SUB	t11, b1, t11
+
+
+	LD	b1,  0 * SIZE(BO)				#	BO point to the beginning of the trigular data part of Bj
+	MUL	t11, b1, t11
+
+
+	ST	t11,  0 * SIZE(AO)				#	update packed blockA for follow-up compute
+
+	ST	t11,  0 * SIZE(CO1)				#	write back results
+	
+
+	daddiu	CO1, CO1, 1 * SIZE			#	fixed address
+
+	dsubu	TEMP, K, KK					#	temp = kc - retangular data length of every panel
+	dsll	L,    TEMP, BASE_SHIFT
+	dsll	TEMP, TEMP, BASE_SHIFT		#	nr=1
+	daddu	AO, AO, L					#	move AO to the end of this panel. also the beginning of next panel
+	daddu	BO, BO, TEMP				#	move BO to the end of this panel
+
+
+	.align 3
+.L79:
+	move	B,  BO
+	daddiu	KK, KK, 1
+
+
+	.align 3
+
+
+.L999:
+	LDARG	$16,   0($sp)
+	LDARG	$17,   8($sp)
+	LDARG	$18,  16($sp)
+	LDARG	$19,  24($sp)
+	LDARG	$20,  32($sp)
+	LDARG	$21,  40($sp)
+	ldc1	$f24, 48($sp)
+	ldc1	$f25, 56($sp)
+	ldc1	$f26, 64($sp)
+	ldc1	$f27, 72($sp)
+	ldc1	$f28, 80($sp)
+
+	LDARG	$22,  88($sp)
+	LDARG	$23,  96($sp)
+	LDARG	$24, 104($sp)
+	LDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	ldc1	$f20,112($sp)
+	ldc1	$f21,120($sp)
+	ldc1	$f22,128($sp)
+	ldc1	$f23,136($sp)
+#endif
+
+	j	$31
+	daddiu	$sp, $sp, 144
+
+	EPILOGUE
diff --git a/kernel/mips64/trsm_kernel_RT_loongson3a.S b/kernel/mips64/trsm_kernel_RT_loongson3a.S
new file mode 100644
index 000000000..cf20cf9e0
--- /dev/null
+++ b/kernel/mips64/trsm_kernel_RT_loongson3a.S
@@ -0,0 +1,1958 @@
+#define REALNAME ASMNAME
+
+#define ASSEMBLER
+#include "common.h"
+
+
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$8
+#define B	$9
+#define C	$10
+#define LDC	$11
+
+#define AO	$12
+#define BO	$13
+
+#define I	$2
+#define J	$3
+#define L	$7
+
+#define CO1	$14
+#define CO2	$15
+#define CO3	$16
+#define CO4	$17
+
+#define OFFSET	$22
+#define KK	$23
+#define TEMP	$24
+#define AORIG	$25
+
+#define a1	$f0
+#define a2	$f1
+#define a3	$f26
+#define a4	$f27
+
+#define a5	$f28
+#define a6	$f29
+#define	a7	$f30
+#define	a8	$f31
+
+#define b1	$f2
+#define b2	$f3
+#define b3	$f4
+#define b4	$f5
+
+#define b5	$f6
+#define b6	$f7
+#define b7	$f8
+#define b8	$f9
+
+#define t11	$f10
+#define t21	$f11
+#define t31	$f12
+#define	t41	$f13
+
+#define t12	$f14
+#define	t22	$f15
+#define t32	$f16
+#define	t42	$f17
+
+#define	t13	$f18
+#define	t23	$f19
+#define	t33	$f20
+#define	t43	$f21
+
+#define	t14	$f22
+#define	t24	$f23
+#define	t34	$f24
+#define t44	$f25
+
+	PROLOGUE
+	
+	daddiu	$sp, $sp, -144
+
+	SDARG	$16,   0($sp)
+	SDARG	$17,   8($sp)
+	SDARG	$18,  16($sp)
+	SDARG	$19,  24($sp)
+	SDARG	$20,  32($sp)
+	SDARG	$21,  40($sp)
+	sdc1	$f24, 48($sp)
+	sdc1	$f25, 56($sp)
+	sdc1	$f26, 64($sp)
+	sdc1	$f27, 72($sp)
+	sdc1	$f28, 80($sp)
+
+	SDARG	$22,  88($sp)
+	SDARG	$23,  96($sp)
+	SDARG	$24, 104($sp)
+	SDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	sdc1	$f20,112($sp)
+	sdc1	$f21,120($sp)
+	sdc1	$f22,128($sp)
+	sdc1	$f23,136($sp)
+#endif
+
+
+	.align	3								#	RT compute from right to left
+	LDARG	OFFSET, 144($sp)				#	get the last parameter
+	dsll	LDC, LDC, BASE_SHIFT			#	LDC * data_Byte
+
+	mult	N, K
+	mflo	TEMP
+
+	dsll	TEMP, TEMP, BASE_SHIFT			#	B Representative triangle matrix!!!
+	daddu	B, B, TEMP						#	B point to the end of sb
+											#	Be carefull B has no effeck of mc!!
+	mult	N, LDC
+	mflo	TEMP
+	daddu	C, C, TEMP						#	C point to the last colum of blockB 
+
+	dsubu	KK, K, OFFSET					#	KC-KK is the length of rectangular data part of Bj
+
+	andi	J,  N, 1
+	blez	J, .L30
+	nop
+
+	dsll	TEMP, K, BASE_SHIFT
+	dsubu	B, B, TEMP						#	move B to the beginning address of Bj
+
+	dsubu	C, C, LDC
+
+	move	CO1, C
+
+	move	AORIG, A
+	
+	dsra	I,  M, 2
+	blez	I, .L80
+	NOP
+
+.L31:										#	mr=4,nr=1
+	dsll	L,    KK, 2 + BASE_SHIFT		#	mr=4	
+	dsll	TEMP, KK, BASE_SHIFT			#	nr=1
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 4 results registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L35
+	NOP
+	
+	.align	3
+
+.L32:
+	LD	a5,  4 * SIZE(AO)
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	LD	a1,  8 * SIZE(AO)
+	LD	a2,  9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+
+	daddiu	AO, AO, 16 * SIZE			#	AO += 4mr*4kr	
+	daddiu	BO, BO,  4 * SIZE			#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L32
+	NOP
+
+
+	.align 3
+
+.L35:
+	andi	L, TEMP, 3
+	blez	L, .L38
+	NOP
+	.align	3
+
+.L36:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	daddiu	AO, AO, 4 * SIZE			#	AO += 4mr	
+	daddiu	BO, BO, 1 * SIZE			#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L36
+	NOP
+
+
+	.align
+.L38:
+	daddiu	TEMP, KK,  -1				#	deal with the triangular data part
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, BASE_SHIFT		#	nr=1
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+	LD	b2,  1 * SIZE(AO)
+	LD	b3,  2 * SIZE(AO)
+	LD	b4,  3 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+
+	LD	b2,   0 * SIZE(BO)
+	MUL	t11, b2, t11
+	MUL	t21, b2, t21
+	MUL	t31, b2, t31
+	MUL	t41, b2, t41
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+	ST	t21,  1 * SIZE(AO)
+	ST	t31,  2 * SIZE(AO)
+	ST	t41,  3 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	
+
+	daddiu	CO1, CO1, 4 * SIZE			#	fixed pointer
+
+	dsll	TEMP, K, 2 + BASE_SHIFT		
+	daddu	AORIG, AORIG, TEMP			# move to next panel Ai
+
+	daddiu	I, I, -1
+	bgtz	I, .L31
+	NOP
+
+
+	.align 3
+.L80:
+	andi	I, M, 2
+	blez	I, .L90
+	nop
+
+	dsll	L,    KK, 1 + BASE_SHIFT		#	mr=2	
+	dsll	TEMP, KK, BASE_SHIFT			#	nr=1
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 4 results registers
+	MOV	t21, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L85
+	NOP
+	
+	.align	3
+
+.L82:
+	LD	a5,  2 * SIZE(AO)
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3
+	MADD	t21, t21, a4, b3
+
+	daddiu	AO, AO, 16 * SIZE			#	AO += 4mr*4kr	
+	daddiu	BO, BO,  4 * SIZE			#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7
+	MADD	t21, t21, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L82
+	NOP
+
+
+	.align 3
+
+.L85:
+	andi	L, TEMP, 3
+	blez	L, .L88
+	NOP
+	.align	3
+
+.L86:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	daddiu	AO, AO, 2 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 1 * SIZE			#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L86
+	NOP
+
+
+	.align
+.L88:
+	daddiu	TEMP, KK,  -1				#	deal with the triangular data part
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, BASE_SHIFT		#	nr=1
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+	LD	b2,  1 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+
+	LD	b2,   0 * SIZE(BO)
+	MUL	t11, b2, t11
+	MUL	t21, b2, t21
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+	ST	t21,  1 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	ST	t21,  1 * SIZE(CO1)
+	
+
+	daddiu	CO1, CO1, 2 * SIZE			#	fixed pointer
+
+	dsll	TEMP, K, 1 + BASE_SHIFT		
+	daddu	AORIG, AORIG, TEMP			# move to next panel Ai
+
+
+	.align 3
+.L90:
+	andi	I, M, 1
+	blez	I, .L39
+	nop
+
+	dsll	L,    KK, BASE_SHIFT			#	mr=1	
+	dsll	TEMP, KK, BASE_SHIFT			#	nr=1
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 4 results registers
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L95
+	NOP
+	
+	.align	3
+
+.L92:
+	LD	a5,  1 * SIZE(AO)
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3
+
+	daddiu	AO, AO,  4 * SIZE			#	AO += 1mr*4kr	
+	daddiu	BO, BO,  4 * SIZE			#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L92
+	NOP
+
+
+	.align 3
+
+.L95:
+	andi	L, TEMP, 3
+	blez	L, .L98
+	NOP
+	.align	3
+
+.L96:
+	MADD	t11, t11, a1, b1
+
+	daddiu	AO, AO, 1 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 1 * SIZE			#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L96
+	NOP
+
+
+	.align
+.L98:
+	daddiu	TEMP, KK,  -1				#	deal with the triangular data part
+	dsll	L,    TEMP, BASE_SHIFT
+	dsll	TEMP, TEMP, BASE_SHIFT		#	nr=1
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+
+	SUB	t11, b1, t11
+
+
+	LD	b2,   0 * SIZE(BO)
+	MUL	t11, b2, t11
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	
+
+	daddiu	CO1, CO1, 1 * SIZE			#	fixed pointer
+
+	dsll	TEMP, K, BASE_SHIFT		
+	daddu	AORIG, AORIG, TEMP			# move to next panel Ai
+
+
+.L39:
+	daddiu	KK, KK, -1						#	rectangular data length increased by 1
+
+
+	.align 3
+.L30:										#	nr=2
+	andi	J,  N, 2
+	blez	J, .L50
+	nop
+
+	dsll	TEMP, K, 1 + BASE_SHIFT			#	Kc*2nr	move B to the beginning address of Bj
+	dsubu	B, B, TEMP
+
+	dsll	TEMP, LDC, 1					#	C 
+	dsubu	C, C, TEMP
+
+	move	CO1, C
+	daddu	CO2, C,   LDC
+
+	move	AORIG, A
+
+	dsra	I,  M, 2
+	blez	I, .L60
+	NOP
+
+.L51:										#	mr=4,nr=2
+	dsll	L,    KK, 2 + BASE_SHIFT		#	mr=4	
+	dsll	TEMP, KK, 1 + BASE_SHIFT		#	nr=2
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 8 results registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L55
+	NOP
+	
+	.align	3
+
+.L52:
+	LD	a5,  4 * SIZE(AO)
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	LD	a1,  8 * SIZE(AO)
+	LD	a2,  9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+
+	MADD	t12, t12, a1, b4
+	MADD	t22, t22, a2, b4
+	MADD	t32, t32, a3, b4
+	MADD	t42, t42, a4, b4
+
+	daddiu	AO, AO, 16 * SIZE			#	AO += 4mr*4kr	
+	daddiu	BO, BO,  8 * SIZE			#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+
+	MADD	t12, t12, a5, b8
+	MADD	t22, t22, a6, b8
+	MADD	t32, t32, a7, b8
+	MADD	t42, t42, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L52
+	NOP
+
+
+	.align 3
+
+.L55:
+	andi	L, TEMP, 3
+	blez	L, .L58
+	NOP
+	.align	3
+
+.L56:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	daddiu	AO, AO, 4 * SIZE			#	AO += 4mr	
+	daddiu	BO, BO, 2 * SIZE			#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L56
+	NOP
+
+
+	.align
+.L58:
+	daddiu	TEMP, KK,  -2				#	deal with the triangular data part
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT	#	nr=2
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+	LD	b2,  1 * SIZE(AO)
+	LD	b3,  2 * SIZE(AO)
+	LD	b4,  3 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+ 	LD	b5,  4 * SIZE(AO)
+	LD	b6,  5 * SIZE(AO)
+	LD	b7,  6 * SIZE(AO)
+	LD	b8,  7 * SIZE(AO)
+
+	SUB	t12, b5, t12
+	SUB	t22, b6, t22
+	SUB	t32, b7, t32
+	SUB	t42, b8, t42
+
+
+	LD	b8,   3 * SIZE(BO)
+	LD	b1,   2 * SIZE(BO)
+	MUL	t12, b8, t12
+	MUL	t22, b8, t22
+	MUL	t32, b8, t32
+	MUL	t42, b8, t42
+	NMSUB	t11, t11, b1, t12
+	NMSUB	t21, t21, b1, t22
+	NMSUB	t31, t31, b1, t32
+	NMSUB	t41, t41, b1, t42
+
+
+	LD	b2,   0 * SIZE(BO)
+	MUL	t11, b2, t11
+	MUL	t21, b2, t21
+	MUL	t31, b2, t31
+	MUL	t41, b2, t41
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+	ST	t21,  1 * SIZE(AO)
+	ST	t31,  2 * SIZE(AO)
+	ST	t41,  3 * SIZE(AO)
+
+	ST	t12,  4 * SIZE(AO)
+	ST	t22,  5 * SIZE(AO)
+	ST	t32,  6 * SIZE(AO)
+	ST	t42,  7 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+	
+	daddiu	CO1, CO1, 4 * SIZE			#	fixed pointer
+	daddiu	CO2, CO2, 4 * SIZE
+
+	dsll	TEMP, K, 2 + BASE_SHIFT		
+	daddu	AORIG, AORIG, TEMP			# move to next panel Ai
+
+	daddiu	I, I, -1
+	bgtz	I, .L51
+	NOP
+
+
+
+	.align 3
+.L60:
+	andi	I,  M, 2						#	mr=2
+	blez	I, .L70
+	nop
+
+	dsll	L,    KK, 1 + BASE_SHIFT		#	mr=2	
+	dsll	TEMP, KK, 1 + BASE_SHIFT		#	nr=2
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 8 results registers
+	MOV	t21, t11
+	MOV	t12, t11
+	MOV	t22, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L65
+	NOP
+	
+	.align	3
+
+.L62:
+	LD	a5,  2 * SIZE(AO)
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3
+	MADD	t21, t21, a4, b3
+
+	MADD	t12, t12, a3, b4
+	MADD	t22, t22, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE			#	AO += 2mr*4kr	
+	daddiu	BO, BO,  8 * SIZE			#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7
+	MADD	t21, t21, a8, b7
+
+	MADD	t12, t12, a7, b8
+	MADD	t22, t22, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L62
+	NOP
+
+
+	.align 3
+
+.L65:
+	andi	L, TEMP, 3
+	blez	L, .L68
+	NOP
+	.align	3
+
+.L66:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	daddiu	AO, AO, 2 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 2 * SIZE			#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L66
+	NOP
+
+
+	.align
+.L68:
+	daddiu	TEMP, KK,  -2				#	deal with the triangular data part
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT	#	nr=2
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+	LD	b2,  1 * SIZE(AO)
+	LD	b3,  2 * SIZE(AO)
+	LD	b4,  3 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t12, b3, t12
+	SUB	t22, b4, t22
+
+
+	LD	b8,   3 * SIZE(BO)
+	LD	b7,   2 * SIZE(BO)
+	MUL	t12, b8, t12
+	MUL	t22, b8, t22
+	NMSUB	t11, t11, b7, t12
+	NMSUB	t21, t21, b7, t22
+
+
+	LD	b6,   0 * SIZE(BO)
+	MUL	t11, b6, t11
+	MUL	t21, b6, t21
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+	ST	t21,  1 * SIZE(AO)
+	ST	t12,  2 * SIZE(AO)
+	ST	t22,  3 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	ST	t21,  1 * SIZE(CO1)
+	
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	
+	daddiu	CO1, CO1, 2 * SIZE			#	fixed pointer
+	daddiu	CO2, CO2, 2 * SIZE
+
+	dsll	TEMP, K, 1 + BASE_SHIFT		#	mr=2
+	daddu	AORIG, AORIG, TEMP			# 	move to next panel Ai
+
+
+
+	.align 3
+.L70:
+	andi	I, M, 1							#	mr=1
+	blez	I, .L59
+	nop
+
+	dsll	L,    KK, BASE_SHIFT			#	mr=1	
+	dsll	TEMP, KK, 1 + BASE_SHIFT		#	nr=2
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 8 results registers
+	MOV	t12, t11
+
+	LD	a1,  0 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L75
+	NOP
+	
+	.align	3
+
+.L72:
+	LD	a5,  1 * SIZE(AO)
+
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t12, t12, a1, b2
+
+	LD	a3,  2 * SIZE(AO)
+
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t12, t12, a5, b6
+
+	LD	a7,  3 * SIZE(AO)
+
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3
+	MADD	t12, t12, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE			#	AO += 1mr*4kr	
+	daddiu	BO, BO,  8 * SIZE			#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7
+	MADD	t12, t12, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L72
+	NOP
+
+
+	.align 3
+
+.L75:
+	andi	L, TEMP, 3
+	blez	L, .L78
+	NOP
+	.align	3
+
+.L76:
+	MADD	t11, t11, a1, b1
+	MADD	t12, t12, a1, b2
+
+	daddiu	AO, AO, 1 * SIZE			#	AO += 1mr	
+	daddiu	BO, BO, 2 * SIZE			#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L76
+	NOP
+
+
+	.align
+.L78:
+	daddiu	TEMP, KK,  -2				#	deal with the triangular data part
+	dsll	L,    TEMP, BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT	#	nr=2
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+	LD	b2,  1 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+
+
+	LD	b8,   3 * SIZE(BO)
+	LD	b7,   2 * SIZE(BO)
+	MUL	t12, b8, t12
+	NMSUB	t11, t11, b7, t12
+
+
+	LD	b6,   0 * SIZE(BO)
+	MUL	t11, b6, t11
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+	ST	t12,  1 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	ST	t12,  0 * SIZE(CO2)
+	
+	daddiu	CO1, CO1, 1 * SIZE			#	fixed pointer
+	daddiu	CO2, CO2, 1 * SIZE
+
+	dsll	TEMP, K, BASE_SHIFT		#	mr=2
+	daddu	AORIG, AORIG, TEMP			# 	move to next panel Ai
+
+
+.L59:
+	daddiu	KK, KK, -2						#	rectangular data length increased by 2
+
+
+
+	.align 3
+.L50:
+	dsra	J,  N, 2						#	J = NC/4 
+	blez	J, .L999
+	NOP
+
+.L10:
+	dsll	TEMP, K, 2 + BASE_SHIFT
+	dsubu	B, B, TEMP						#	move B to the beginning address of Bj	
+
+	dsll	TEMP, LDC, 2
+	dsubu	C, C, TEMP						#	move C to the beginning address of Cj
+
+	daddiu	J, J, -1
+
+	move	CO1, C
+	daddu	CO2, C,   LDC
+	daddu	CO3, CO2, LDC
+	daddu	CO4, CO3, LDC
+	
+	move	AORIG, A						#	reset A
+
+	dsra	I,  M, 2						#	I=MC/4
+	blez	I, .L20
+	NOP
+
+	.align 3
+.L11:
+	dsll	L,    KK, 2 + BASE_SHIFT		#	mr=4	
+	dsll	TEMP, KK, 2 + BASE_SHIFT		#	nr=4
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 16 results registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L15
+	NOP
+	
+	.align	3
+
+.L12:
+	LD	a5,  4 * SIZE(AO)
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4			#	fisrt 	
+
+	LD	a1,  8 * SIZE(AO)
+	LD	a2,  9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			#	second
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4			#	third
+
+	daddiu	AO, AO, 16 * SIZE			#	AO += 4mr*4kr	
+	daddiu	BO, BO, 16 * SIZE			#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			#	fouth
+
+	daddiu	L, L, -1
+	bgtz	L, .L12
+	NOP
+
+
+	.align 3
+
+.L15:
+	andi	L, TEMP, 3
+	blez	L, .L18
+	NOP
+	.align	3
+
+.L16:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4			#	third
+
+	daddiu	AO, AO, 4 * SIZE			#	AO += 4mr	
+	daddiu	BO, BO, 4 * SIZE			#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L16
+	NOP
+
+
+	.align
+.L18:
+	daddiu	TEMP, KK, -4				#	deal with the triangular data part
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+	LD	b2,  1 * SIZE(AO)
+	LD	b3,  2 * SIZE(AO)
+	LD	b4,  3 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+ 	LD	b5,  4 * SIZE(AO)
+	LD	b6,  5 * SIZE(AO)
+	LD	b7,  6 * SIZE(AO)
+	LD	b8,  7 * SIZE(AO)
+
+	SUB	t12, b5, t12
+	SUB	t22, b6, t22
+	SUB	t32, b7, t32
+	SUB	t42, b8, t42
+
+ 	LD	b1,  8 * SIZE(AO)
+	LD	b2,  9 * SIZE(AO)
+	LD	b3, 10 * SIZE(AO)
+	LD	b4, 11 * SIZE(AO)
+
+	SUB	t13, b1, t13
+	SUB	t23, b2, t23
+	SUB	t33, b3, t33
+	SUB	t43, b4, t43
+
+ 	LD	b5, 12 * SIZE(AO)
+	LD	b6, 13 * SIZE(AO)
+	LD	b7, 14 * SIZE(AO)
+	LD	b8, 15 * SIZE(AO)
+
+	SUB	t14, b5, t14
+	SUB	t24, b6, t24
+	SUB	t34, b7, t34
+	SUB	t44, b8, t44
+
+
+	LD	b1, 15 * SIZE(BO)
+	LD	b2, 14 * SIZE(BO)
+	LD	b3, 13 * SIZE(BO)
+	LD	b4, 12 * SIZE(BO)
+	MUL	t14, b1, t14
+	MUL	t24, b1, t24
+	MUL	t34, b1, t34
+	MUL	t44, b1, t44
+	NMSUB	t13, t13, b2, t14
+	NMSUB	t23, t23, b2, t24
+	NMSUB	t33, t33, b2, t34
+	NMSUB	t43, t43, b2, t44
+	NMSUB	t12, t12, b3, t14
+	NMSUB	t22, t22, b3, t24
+	NMSUB	t32, t32, b3, t34
+	NMSUB	t42, t42, b3, t44
+	NMSUB	t11, t11, b4, t14
+	NMSUB	t21, t21, b4, t24
+	NMSUB	t31, t31, b4, t34
+	NMSUB	t41, t41, b4, t44
+
+	
+	LD	b5,  10 * SIZE(BO)
+	LD	b6,   9 * SIZE(BO)
+	LD	b7,   8 * SIZE(BO)
+	MUL	t13, b5, t13
+	MUL	t23, b5, t23
+	MUL	t33, b5, t33
+	MUL	t43, b5, t43
+	NMSUB	t12, t12, b6, t13
+	NMSUB	t22, t22, b6, t23
+	NMSUB	t32, t32, b6, t33
+	NMSUB	t42, t42, b6, t43
+	NMSUB	t11, t11, b7, t13
+	NMSUB	t21, t21, b7, t23
+	NMSUB	t31, t31, b7, t33
+	NMSUB	t41, t41, b7, t43
+
+
+	LD	b8,   5 * SIZE(BO)
+	LD	b1,   4 * SIZE(BO)
+	MUL	t12, b8, t12
+	MUL	t22, b8, t22
+	MUL	t32, b8, t32
+	MUL	t42, b8, t42
+	NMSUB	t11, t11, b1, t12
+	NMSUB	t21, t21, b1, t22
+	NMSUB	t31, t31, b1, t32
+	NMSUB	t41, t41, b1, t42
+
+
+	LD	b2,   0 * SIZE(BO)
+	MUL	t11, b2, t11
+	MUL	t21, b2, t21
+	MUL	t31, b2, t31
+	MUL	t41, b2, t41
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+	ST	t21,  1 * SIZE(AO)
+	ST	t31,  2 * SIZE(AO)
+	ST	t41,  3 * SIZE(AO)
+
+	ST	t12,  4 * SIZE(AO)
+	ST	t22,  5 * SIZE(AO)
+	ST	t32,  6 * SIZE(AO)
+	ST	t42,  7 * SIZE(AO)
+
+	ST	t13,  8 * SIZE(AO)
+	ST	t23,  9 * SIZE(AO)
+	ST	t33, 10 * SIZE(AO)
+	ST	t43, 11 * SIZE(AO)
+
+	ST	t14, 12 * SIZE(AO)
+	ST	t24, 13 * SIZE(AO)
+	ST	t34, 14 * SIZE(AO)
+	ST	t44, 15 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+	
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t33,  2 * SIZE(CO3)
+	ST	t43,  3 * SIZE(CO3)
+	
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+	ST	t34,  2 * SIZE(CO4)
+	ST	t44,  3 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 4 * SIZE			#	fixed pointer
+	daddiu	CO2, CO2, 4 * SIZE
+	daddiu	CO3, CO3, 4 * SIZE
+	daddiu	CO4, CO4, 4 * SIZE
+
+	dsll	TEMP, K, 2 + BASE_SHIFT		
+	daddu	AORIG, AORIG, TEMP			# move to next panel Ai
+
+	daddiu	I, I, -1
+	bgtz	I, .L11
+	NOP
+
+	.align 3
+.L20:
+	andi	I,  M, 2						#	mr=2
+	blez	I, .L40
+	NOP
+
+	dsll	L,    KK, 1 + BASE_SHIFT		#	mr=2	
+	dsll	TEMP, KK, 2 + BASE_SHIFT		#	nr=4
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 8 results registers
+	MOV	t21, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t14, t11
+	MOV	t24, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L25
+	NOP
+	
+	.align	3
+
+.L22:
+	LD	a5,  2 * SIZE(AO)
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t21, t21, a6, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1
+	MADD	t21, t21, a4, b1
+
+	MADD	t12, t12, a3, b2
+	MADD	t22, t22, a4, b2
+
+	MADD	t13, t13, a3, b3
+	MADD	t23, t23, a4, b3
+
+	MADD	t14, t14, a3, b4
+	MADD	t24, t24, a4, b4
+
+	daddiu	AO, AO, 8 * SIZE			#	AO += 2mr*4kr	
+	daddiu	BO, BO, 16 * SIZE			#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5
+	MADD	t21, t21, a8, b5
+
+	MADD	t12, t12, a7, b6
+	MADD	t22, t22, a8, b6
+
+	MADD	t13, t13, a7, b7
+	MADD	t23, t23, a8, b7
+
+	MADD	t14, t14, a7, b8
+	MADD	t24, t24, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L22
+	NOP
+
+
+	.align 3
+
+.L25:
+	andi	L, TEMP, 3
+	blez	L, .L28
+	NOP
+	.align	3
+
+.L26:
+	MADD	t11, t11, a1, b1
+	MADD	t21, t21, a2, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	daddiu	AO, AO, 2 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 4 * SIZE			#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L26
+	NOP
+
+
+	.align
+.L28:
+	daddiu	TEMP, KK, -4				#	deal with the triangular data part
+	dsll	L,    TEMP, 1 + BASE_SHIFT	#	mr=2
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+	LD	b2,  1 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+ 	LD	b5,  2 * SIZE(AO)
+	LD	b6,  3 * SIZE(AO)
+
+	SUB	t12, b5, t12
+	SUB	t22, b6, t22
+
+	LD	b3,  4 * SIZE(AO)
+	LD	b4,  5 * SIZE(AO)
+
+	SUB	t13, b3, t13
+	SUB	t23, b4, t23
+
+	LD	b7,  6 * SIZE(AO)
+	LD	b8,  7 * SIZE(AO)
+
+	SUB	t14, b7, t14
+	SUB	t24, b8, t24
+
+
+	LD	b1, 15 * SIZE(BO)
+	LD	b2, 14 * SIZE(BO)
+	LD	b3, 13 * SIZE(BO)
+	LD	b4, 12 * SIZE(BO)
+	MUL	t14, b1, t14
+	MUL	t24, b1, t24
+	NMSUB	t13, t13, b2, t14
+	NMSUB	t23, t23, b2, t24
+	NMSUB	t12, t12, b3, t14
+	NMSUB	t22, t22, b3, t24
+	NMSUB	t11, t11, b4, t14
+	NMSUB	t21, t21, b4, t24
+
+	
+	LD	b5,  10 * SIZE(BO)
+	LD	b6,   9 * SIZE(BO)
+	LD	b7,   8 * SIZE(BO)
+	MUL	t13, b5, t13
+	MUL	t23, b5, t23
+	NMSUB	t12, t12, b6, t13
+	NMSUB	t22, t22, b6, t23
+	NMSUB	t11, t11, b7, t13
+	NMSUB	t21, t21, b7, t23
+
+
+	LD	b8,   5 * SIZE(BO)
+	LD	b1,   4 * SIZE(BO)
+	MUL	t12, b8, t12
+	MUL	t22, b8, t22
+	NMSUB	t11, t11, b1, t12
+	NMSUB	t21, t21, b1, t22
+
+
+	LD	b2,   0 * SIZE(BO)
+	MUL	t11, b2, t11
+	MUL	t21, b2, t21
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+	ST	t21,  1 * SIZE(AO)
+
+	ST	t12,  2 * SIZE(AO)
+	ST	t22,  3 * SIZE(AO)
+
+	ST	t13,  4 * SIZE(AO)
+	ST	t23,  5 * SIZE(AO)
+
+	ST	t14,  6 * SIZE(AO)
+	ST	t24,  7 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	ST	t21,  1 * SIZE(CO1)
+	
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 2 * SIZE			#	fixed pointer
+	daddiu	CO2, CO2, 2 * SIZE
+	daddiu	CO3, CO3, 2 * SIZE
+	daddiu	CO4, CO4, 2 * SIZE
+
+	dsll	TEMP, K, 1 + BASE_SHIFT		#	mr=2
+	daddu	AORIG, AORIG, TEMP			# 	move to next panel Ai
+
+
+	.align 3
+.L40:
+	andi	I,  M, 1
+	blez	I, .L29
+	NOP
+
+	dsll	L,    KK, BASE_SHIFT			#	mr=1	
+	dsll	TEMP, KK, 2 + BASE_SHIFT		#	nr=4
+	daddu	AO, AORIG, L					
+	daddu	BO, B, TEMP						#	BO point to the retangular data part,also reset BO
+	dsubu	TEMP, K, KK						#	temp = the length of rectangular data part 
+
+	MTC	$0,  t11							#	clear 4 results registers
+	MOV	t12, t11
+	MOV	t13, t11
+	MOV	t14, t11
+
+	LD	a1,  0 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L45
+	NOP
+	
+	.align	3
+
+.L42:
+	LD	a5,  1 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	LD	a3,  2 * SIZE(AO)
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5
+	MADD	t12, t12, a5, b6
+	MADD	t13, t13, a5, b7
+	MADD	t14, t14, a5, b8
+
+	LD	a7,  3 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1
+	MADD	t12, t12, a3, b2
+	MADD	t13, t13, a3, b3
+	MADD	t14, t14, a3, b4
+
+	daddiu	AO, AO, 4 * SIZE			#	AO += 1mr*4kr	
+	daddiu	BO, BO, 16 * SIZE			#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5
+	MADD	t12, t12, a7, b6
+	MADD	t13, t13, a7, b7
+	MADD	t14, t14, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L42
+	NOP
+
+
+	.align 3
+
+.L45:
+	andi	L, TEMP, 3
+	blez	L, .L48
+	NOP
+	.align	3
+
+.L46:
+	MADD	t11, t11, a1, b1
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	daddiu	AO, AO, 1 * SIZE			#	AO += 2mr	
+	daddiu	BO, BO, 4 * SIZE			#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L46
+	NOP
+
+
+	.align
+.L48:
+	daddiu	TEMP, KK, -4				#	deal with the triangular data part
+	dsll	L,    TEMP, BASE_SHIFT		#	mr=1
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L
+	daddu	BO, B, TEMP					#	BO point to the trigular data part
+
+ 	LD	b1,  0 * SIZE(AO)				#	fixed results
+ 	LD	b5,  1 * SIZE(AO)
+	LD	b3,  2 * SIZE(AO)
+	LD	b7,  3 * SIZE(AO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b5, t12
+	SUB	t13, b3, t13
+	SUB	t14, b7, t14
+
+
+	LD	b1, 15 * SIZE(BO)
+	LD	b2, 14 * SIZE(BO)
+	LD	b3, 13 * SIZE(BO)
+	LD	b4, 12 * SIZE(BO)
+	MUL	t14, b1, t14
+	NMSUB	t13, t13, b2, t14
+	NMSUB	t12, t12, b3, t14
+	NMSUB	t11, t11, b4, t14
+
+	
+	LD	b5,  10 * SIZE(BO)
+	LD	b6,   9 * SIZE(BO)
+	LD	b7,   8 * SIZE(BO)
+	MUL	t13, b5, t13
+	NMSUB	t12, t12, b6, t13
+	NMSUB	t11, t11, b7, t13
+
+
+	LD	b8,   5 * SIZE(BO)
+	LD	b1,   4 * SIZE(BO)
+	MUL	t12, b8, t12
+	NMSUB	t11, t11, b1, t12
+
+
+	LD	b2,   0 * SIZE(BO)
+	MUL	t11, b2, t11
+
+
+	ST	t11,  0 * SIZE(AO)				# updata packed A
+	ST	t12,  1 * SIZE(AO)
+	ST	t13,  2 * SIZE(AO)
+	ST	t14,  3 * SIZE(AO)
+
+	ST	t11,  0 * SIZE(CO1)				#	write back
+	ST	t12,  0 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 1 * SIZE			#	fixed pointer
+	daddiu	CO2, CO2, 1 * SIZE
+	daddiu	CO3, CO3, 1 * SIZE
+	daddiu	CO4, CO4, 1 * SIZE
+
+	dsll	TEMP, K, BASE_SHIFT			#	mr=2
+	daddu	AORIG, AORIG, TEMP			# 	move to next panel Ai
+
+
+.L29:
+	daddiu	KK, KK, -4					#	rectangular data part increased by 4
+	bgtz	J, .L10
+	NOP
+
+
+
+	.align 3
+
+
+.L999:
+	LDARG	$16,   0($sp)
+	LDARG	$17,   8($sp)
+	LDARG	$18,  16($sp)
+	LDARG	$19,  24($sp)
+	LDARG	$20,  32($sp)
+	LDARG	$21,  40($sp)
+	ldc1	$f24, 48($sp)
+	ldc1	$f25, 56($sp)
+	ldc1	$f26, 64($sp)
+	ldc1	$f27, 72($sp)
+	ldc1	$f28, 80($sp)
+
+	LDARG	$22,  88($sp)
+	LDARG	$23,  96($sp)
+	LDARG	$24, 104($sp)
+	LDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	ldc1	$f20,112($sp)
+	ldc1	$f21,120($sp)
+	ldc1	$f22,128($sp)
+	ldc1	$f23,136($sp)
+#endif
+
+	j	$31
+	daddiu	$sp, $sp, 144
+
+	EPILOGUE
diff --git a/kernel/setparam-ref.c b/kernel/setparam-ref.c
index 0ab57f3b3..d3734bbd9 100644
--- a/kernel/setparam-ref.c
+++ b/kernel/setparam-ref.c
@@ -101,7 +101,11 @@ gotoblas_t TABLE_NAME = {
 #endif
   ssymm_outcopyTS, ssymm_oltcopyTS,
 
+#ifndef NO_LAPACK
   sneg_tcopyTS, slaswp_ncopyTS,
+#else
+  NULL,NULL,
+#endif
 
   0, 0, 0,
   DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N, MAX(DGEMM_DEFAULT_UNROLL_M, DGEMM_DEFAULT_UNROLL_N),
@@ -147,7 +151,11 @@ gotoblas_t TABLE_NAME = {
 #endif
   dsymm_outcopyTS, dsymm_oltcopyTS,
 
+#ifndef NO_LAPACK
   dneg_tcopyTS, dlaswp_ncopyTS,
+#else
+  NULL, NULL,
+#endif
 
 #ifdef EXPRECISION
 
@@ -195,7 +203,11 @@ gotoblas_t TABLE_NAME = {
 #endif
   qsymm_outcopyTS, qsymm_oltcopyTS,
 
+#ifndef NO_LAPACK
   qneg_tcopyTS, qlaswp_ncopyTS,
+#else
+  NULL, NULL,
+#endif
 
 #endif
 
@@ -286,7 +298,11 @@ gotoblas_t TABLE_NAME = {
   chemm3m_oucopyrTS,  chemm3m_olcopyrTS,
   chemm3m_oucopyiTS,  chemm3m_olcopyiTS,
 
+#ifndef NO_LAPACK
   cneg_tcopyTS, claswp_ncopyTS,
+#else
+  NULL, NULL,
+#endif
 
   0, 0, 0,
   ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N, MAX(ZGEMM_DEFAULT_UNROLL_M, ZGEMM_DEFAULT_UNROLL_N),
@@ -375,7 +391,11 @@ gotoblas_t TABLE_NAME = {
   zhemm3m_oucopyrTS,  zhemm3m_olcopyrTS,
   zhemm3m_oucopyiTS,  zhemm3m_olcopyiTS,
 
+#ifndef NO_LAPACK
   zneg_tcopyTS, zlaswp_ncopyTS,
+#else
+  NULL, NULL,
+#endif
 
 #ifdef EXPRECISION
 
@@ -466,7 +486,11 @@ gotoblas_t TABLE_NAME = {
   xhemm3m_oucopyrTS,  xhemm3m_olcopyrTS,
   xhemm3m_oucopyiTS,  xhemm3m_olcopyiTS,
 
+#ifndef NO_LAPACK
   xneg_tcopyTS, xlaswp_ncopyTS,
+#else
+  NULL, NULL,
+#endif
 
 #endif
 
diff --git a/kernel/x86/zdot_sse2.S b/kernel/x86/zdot_sse2.S
index 5aeefde31..2a174fb5d 100644
--- a/kernel/x86/zdot_sse2.S
+++ b/kernel/x86/zdot_sse2.S
@@ -1541,5 +1541,8 @@
 	popl	%ebx
 	popl	%esi
 	popl	%edi
+/*remove the hidden return value address from the stack.*/
+	popl	%ecx
+	xchgl	%ecx, 0(%esp)
 	ret
 	EPILOGUE
diff --git a/kernel/x86_64/dot_sse.S b/kernel/x86_64/dot_sse.S
index cc866a9c5..61c481064 100644
--- a/kernel/x86_64/dot_sse.S
+++ b/kernel/x86_64/dot_sse.S
@@ -1286,6 +1286,10 @@
 	haddps	%xmm0, %xmm0
 #endif
 
+#ifdef DSDOT
+	cvtss2sd %xmm0, %xmm0
+#endif
+
 	RESTOREREGISTERS
 
 	ret
diff --git a/kernel/x86_64/zgemm_kernel_1x4_nehalem.S b/kernel/x86_64/zgemm_kernel_1x4_nehalem.S
index e72a19c96..4ddfc488b 100644
--- a/kernel/x86_64/zgemm_kernel_1x4_nehalem.S
+++ b/kernel/x86_64/zgemm_kernel_1x4_nehalem.S
@@ -544,7 +544,7 @@
 	jg	.L11
 
 #if defined(TRMMKERNEL) && !defined(LEFT)
-	addq	$1, KK
+	addq	$4, KK
 #endif
 
 	leaq	(C, LDC, 4), C
@@ -594,7 +594,7 @@
 	jg	.L11
 
 #if defined(TRMMKERNEL) && !defined(LEFT)
-	addq	$1, KK
+	addq	$4, KK
 #endif
 
 	leaq	(C, LDC, 4), C
diff --git a/openblas_config_template.h b/openblas_config_template.h
new file mode 100644
index 000000000..9fb80aa4f
--- /dev/null
+++ b/openblas_config_template.h
@@ -0,0 +1,21 @@
+/*This is only for "make install" target.*/
+
+#ifdef NEEDBUNDERSCORE
+#define BLASFUNC(FUNC) FUNC##_
+#else
+#define BLASFUNC(FUNC) FUNC
+#endif
+
+#if defined(OS_WINDOWS) && defined(__64BIT__)
+typedef long long BLASLONG;
+typedef unsigned long long BLASULONG;
+#else
+typedef long BLASLONG;
+typedef unsigned long BLASULONG;
+#endif
+
+#ifdef USE64BITINT
+typedef BLASLONG blasint;
+#else
+typedef int blasint;
+#endif
diff --git a/param.h b/param.h
index 8fcd19358..603caab46 100644
--- a/param.h
+++ b/param.h
@@ -1480,27 +1480,29 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define GEMM_DEFAULT_OFFSET_B 0
 #define GEMM_DEFAULT_ALIGN 0x03fffUL
 
-#define SGEMM_DEFAULT_UNROLL_M  2
-#define SGEMM_DEFAULT_UNROLL_N  8
-#define DGEMM_DEFAULT_UNROLL_M  2
-#define DGEMM_DEFAULT_UNROLL_N  8
+#define SGEMM_DEFAULT_UNROLL_M  4
+#define SGEMM_DEFAULT_UNROLL_N  4
+
+#define DGEMM_DEFAULT_UNROLL_M  4
+#define DGEMM_DEFAULT_UNROLL_N  4
+
 #define CGEMM_DEFAULT_UNROLL_M  1
 #define CGEMM_DEFAULT_UNROLL_N  4
 #define ZGEMM_DEFAULT_UNROLL_M  1
 #define ZGEMM_DEFAULT_UNROLL_N  4
 
-#define SGEMM_DEFAULT_P 108
-#define DGEMM_DEFAULT_P 112
+#define SGEMM_DEFAULT_P 32
+#define DGEMM_DEFAULT_P	32 
 #define CGEMM_DEFAULT_P 108
 #define ZGEMM_DEFAULT_P 112
 
-#define SGEMM_DEFAULT_Q 288
-#define DGEMM_DEFAULT_Q 144
+#define SGEMM_DEFAULT_Q 116
+#define DGEMM_DEFAULT_Q 116
 #define CGEMM_DEFAULT_Q 144
 #define ZGEMM_DEFAULT_Q  72
 
-#define SGEMM_DEFAULT_R 2000
-#define DGEMM_DEFAULT_R 2000
+#define SGEMM_DEFAULT_R 1000
+#define DGEMM_DEFAULT_R 1000
 #define CGEMM_DEFAULT_R 2000
 #define ZGEMM_DEFAULT_R 2000
 
diff --git a/utest/Makefile b/utest/Makefile
index 9d512b877..e7c5f3412 100644
--- a/utest/Makefile
+++ b/utest/Makefile
@@ -5,12 +5,12 @@ include $(TOPDIR)/Makefile.system
 TARGET=openblas_utest
 CUNIT_LIB=/usr/local/lib/libcunit.a
 
-OBJS=main.o test_rot.o test_swap.o test_axpy.o test_dotu.o
+OBJS=main.o test_rot.o test_swap.o test_axpy.o test_dotu.o test_rotmg.o test_dsdot.o
 
 all : run_test
 
 $(TARGET): $(OBJS)
-	$(CC) -o $@ $^ ../$(LIBNAME) $(CUNIT_LIB) $(EXTRALIB) 
+	$(FC) -o $@ $^ ../$(LIBNAME) $(CUNIT_LIB) $(EXTRALIB)
 
 run_test: $(TARGET)
 	./$(TARGET)
diff --git a/utest/common_utest.h b/utest/common_utest.h
index 3e9ecb422..1332ef6ab 100644
--- a/utest/common_utest.h
+++ b/utest/common_utest.h
@@ -57,4 +57,8 @@ void test_caxpy_inc_0(void);
 void test_zdotu_n_1(void);
 void test_zdotu_offset_1(void);
 
+void test_drotmg(void);
+
+void test_dsdot_n_1(void);
+
 #endif
diff --git a/utest/main.c b/utest/main.c
index f6ecf3cc0..135709507 100644
--- a/utest/main.c
+++ b/utest/main.c
@@ -54,7 +54,10 @@ CU_TestInfo test_level1[]={
 
 	{"Testing zdotu with n == 1",test_zdotu_n_1},
 	{"Testing zdotu with input x & y offset == 1",test_zdotu_offset_1},
-	
+
+	{"Testing drotmg",test_drotmg},
+
+	{"Testing dsdot with n == 1",test_dsdot_n_1},
 	CU_TEST_INFO_NULL,
 };
 
diff --git a/utest/test_dsdot.c b/utest/test_dsdot.c
new file mode 100644
index 000000000..8df7380be
--- /dev/null
+++ b/utest/test_dsdot.c
@@ -0,0 +1,50 @@
+/*****************************************************************************
+Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+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 "common_utest.h"
+
+void test_dsdot_n_1()
+{
+	float x= 0.172555164;
+	float y= -0.0138700781;
+	int incx=1;
+	int incy=1;
+	int n=1;
+	
+	double res1=0.0f, res2=0.0f;
+
+	res1=BLASFUNC(dsdot)(&n, &x, &incx, &y, &incy);
+	res2=BLASFUNC_REF(dsdot)(&n, &x, &incx, &y, &incy);
+
+	CU_ASSERT_DOUBLE_EQUAL(res1, res2, CHECK_EPS);
+	
+}
diff --git a/utest/test_rotmg.c b/utest/test_rotmg.c
new file mode 100644
index 000000000..e51e6b299
--- /dev/null
+++ b/utest/test_rotmg.c
@@ -0,0 +1,60 @@
+/*****************************************************************************
+Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
+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 "common_utest.h"
+
+void test_drotmg()
+{
+	double te_d1, tr_d1;
+	double te_d2, tr_d2;
+	double te_x1, tr_x1;
+	double te_y1, tr_y1;
+	double te_param[5],tr_param[5];
+	int i=0;
+	te_d1= tr_d1=0.21149573940783739;
+	te_d2= tr_d2=0.046892057172954082;
+	te_x1= tr_x1=-0.42272687517106533;
+	te_y1= tr_y1=0.42211309121921659;
+	//OpenBLAS
+	BLASFUNC(drotmg)(&te_d1, &te_d2, &te_x1, &te_y1, te_param);
+	//reference
+	BLASFUNC_REF(drotmg)(&tr_d1, &tr_d2, &tr_x1, &tr_y1, tr_param);
+	
+	CU_ASSERT_DOUBLE_EQUAL(te_d1, tr_d1, CHECK_EPS);
+	CU_ASSERT_DOUBLE_EQUAL(te_d2, tr_d2, CHECK_EPS);
+	CU_ASSERT_DOUBLE_EQUAL(te_x1, tr_x1, CHECK_EPS);
+	CU_ASSERT_DOUBLE_EQUAL(te_y1, tr_y1, CHECK_EPS);
+
+	for(i=0; i<5; i++){
+		CU_ASSERT_DOUBLE_EQUAL(te_param[i], tr_param[i], CHECK_EPS);
+	}
+}