diff --git a/Makefile.system b/Makefile.system
index b8b9ba837..0fd223d60 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -279,7 +279,12 @@ endif
 BINARY_DEFINED = 1
 endif
 
-ifeq ($(CORE), LOONGSON3A)
+ifeq ($(CORE), LOONGSON3A) 
+CCOMMON_OPT += -march=mips64
+FCOMMON_OPT += -march=mips64
+endif
+
+ifeq ($(CORE), LOONGSON3B) 
 CCOMMON_OPT += -march=mips64
 FCOMMON_OPT += -march=mips64
 endif
@@ -534,8 +539,10 @@ ifdef SMP
 CCOMMON_OPT	+= -DSMP_SERVER
 
 ifeq ($(ARCH), mips64)
+ifneq ($(CORE), LOONGSON3B)
 USE_SIMPLE_THREADED_LEVEL3 = 1
 endif
+endif
 
 ifeq ($(USE_OPENMP), 1)
 # USE_SIMPLE_THREADED_LEVEL3 = 1
@@ -600,9 +607,11 @@ endif
 
 ifneq ($(ARCH), x86_64)
 ifneq ($(ARCH), x86)
+ifneq ($(CORE), LOONGSON3B)
 NO_AFFINITY = 1
 endif
 endif
+endif
 
 ifdef NO_AFFINITY
 CCOMMON_OPT	+= -DNO_AFFINITY
diff --git a/common_linux.h b/common_linux.h
index 8b3d44bfa..8d9019a0d 100644
--- a/common_linux.h
+++ b/common_linux.h
@@ -68,9 +68,17 @@ extern long int syscall (long int __sysno, ...);
 static inline int my_mbind(void *addr, unsigned long len, int mode,
 			   unsigned long *nodemask, unsigned long maxnode,
 			   unsigned flags) {
+#if defined (LOONGSON3B) 
+#if defined (__64BIT__)
+	return syscall(SYS_mbind, addr, len, mode, nodemask, maxnode, flags);
+#else
+	return 0; //NULL Implementation on Loongson 3B 32bit.
+#endif
+#else
 //Fixed randomly SEGFAULT when nodemask==NULL with above Linux 2.6.34
 	unsigned long null_nodemask=0;
 	return syscall(SYS_mbind, addr, len, mode, &null_nodemask, maxnode, flags);
+#endif
 }
 
 static inline int my_set_mempolicy(int mode, const unsigned long *addr, unsigned long flag) {
diff --git a/common_mips64.h b/common_mips64.h
index 35d8265bc..85348377e 100644
--- a/common_mips64.h
+++ b/common_mips64.h
@@ -101,10 +101,15 @@ static void INLINE blas_lock(volatile unsigned long *address){
 
 static inline unsigned int rpcc(void){
   unsigned long ret;
-#if defined(LOONGSON3A)
-  unsigned long long tmp;
-  __asm__ __volatile__("dmfc0 %0, $25, 1": "=r"(tmp):: "memory");
-  ret=tmp;
+#if defined(LOONGSON3A) || defined(LOONGSON3B)
+  //  unsigned long long tmp;
+  //__asm__ __volatile__("dmfc0 %0, $25, 1": "=r"(tmp):: "memory");
+  //ret=tmp;
+  __asm__ __volatile__(".set push \n"
+                       ".set mips32r2\n"
+                       "rdhwr %0, $2\n"
+                       ".set pop": "=r"(ret):: "memory");
+
 #else
   __asm__ __volatile__(".set   push    \n"                                     
           ".set   mips32r2\n"                                                  
@@ -114,6 +119,21 @@ static inline unsigned int rpcc(void){
   return ret;
 }
 
+#if defined(LOONGSON3A) || defined(LOONGSON3B)
+#ifndef NO_AFFINITY
+#define WHEREAMI
+static inline int WhereAmI(void){
+  int ret=0;
+  __asm__ __volatile__(".set push \n"
+                       ".set mips32r2\n"
+                       "rdhwr %0, $0\n"
+                       ".set pop": "=r"(ret):: "memory");
+  return ret;
+
+}
+#endif
+#endif
+
 static inline int blas_quickdivide(blasint x, blasint y){
   return x / y;
 }
@@ -234,6 +254,11 @@ REALNAME: ;\
 #define FIXED_PAGESIZE	(16UL << 10)
 #endif
 
+#if defined(LOONGSON3B)
+#define PAGESIZE	(32UL << 10)
+#define FIXED_PAGESIZE	(32UL << 10)
+#endif
+
 #ifndef PAGESIZE
 #define PAGESIZE	(64UL << 10)
 #endif
@@ -245,7 +270,7 @@ REALNAME: ;\
 #define MAP_ANONYMOUS MAP_ANON
 #endif
 
-#if defined(LOONGSON3A)
+#if defined(LOONGSON3A) || defined(LOONGSON3B)
 #define PREFETCHD_(x) ld $0, x
 #define PREFETCHD(x)  PREFETCHD_(x)  
 #else
diff --git a/cpuid_mips.c b/cpuid_mips.c
index f50a4ec3e..217492dd7 100644
--- a/cpuid_mips.c
+++ b/cpuid_mips.c
@@ -72,11 +72,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CPU_UNKNOWN     0
 #define CPU_SICORTEX    1
 #define CPU_LOONGSON3A  2
+#define CPU_LOONGSON3B  3
 
 static char *cpuname[] = {
   "UNKOWN",
   "SICORTEX",
-  "LOONGSON3A"
+  "LOONGSON3A",
+  "LOONGSON3B"
 };
 
 int detect(void){
@@ -101,6 +103,8 @@ int detect(void){
 
   if (strstr(p, "Loongson-3A")){
     return CPU_LOONGSON3A;
+  }else if(strstr(p, "Loongson-3B")){
+    return CPU_LOONGSON3B;
   }else if (strstr(p, "Loongson-3")){
     infile = fopen("/proc/cpuinfo", "r");
     while (fgets(buffer, sizeof(buffer), infile)){
@@ -130,6 +134,8 @@ void get_architecture(void){
 void get_subarchitecture(void){
   if(detect()==CPU_LOONGSON3A) {
     printf("LOONGSON3A");
+  }else if(detect()==CPU_LOONGSON3B){
+    printf("LOONGSON3B");
   }else{
     printf("SICORTEX");
   }
@@ -149,6 +155,15 @@ void get_cpuconfig(void){
     printf("#define DTB_DEFAULT_ENTRIES 64\n");
     printf("#define DTB_SIZE 4096\n");
     printf("#define L2_ASSOCIATIVE 4\n");
+  }else if(detect()==CPU_LOONGSON3B){
+    printf("#define LOONGSON3B\n");
+    printf("#define L1_DATA_SIZE 65536\n");
+    printf("#define L1_DATA_LINESIZE 32\n");
+    printf("#define L2_SIZE 512488\n");
+    printf("#define L2_LINESIZE 32\n");
+    printf("#define DTB_DEFAULT_ENTRIES 64\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 4\n");
   }else{
     printf("#define SICORTEX\n");
     printf("#define L1_DATA_SIZE 32768\n");
@@ -164,6 +179,8 @@ void get_cpuconfig(void){
 void get_libname(void){
   if(detect()==CPU_LOONGSON3A) {
     printf("loongson3a\n");
+  }else if(detect()==CPU_LOONGSON3B) {
+    printf("loongson3b\n");
   }else{
 #ifdef __mips64
   printf("mips64\n");
diff --git a/driver/level3/gemm_thread_mn.c b/driver/level3/gemm_thread_mn.c
index 321e88f0c..b81c6fa40 100644
--- a/driver/level3/gemm_thread_mn.c
+++ b/driver/level3/gemm_thread_mn.c
@@ -77,8 +77,8 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
     range_M[0] = 0;
     i          = arg -> m;
   } else {
-    range_M[0] = range_M[0];
-    i          = range_M[1] - range_M[0];
+    range_M[0] = range_m[0];
+    i          = range_m[1] - range_m[0];
   }
 
   num_cpu_m  = 0;
diff --git a/driver/level3/gemm_thread_variable.c b/driver/level3/gemm_thread_variable.c
index 9d83e950a..9ffe17040 100644
--- a/driver/level3/gemm_thread_variable.c
+++ b/driver/level3/gemm_thread_variable.c
@@ -55,8 +55,8 @@ int CNAME(int mode,
     range_M[0] = 0;
     i          = arg -> m;
   } else {
-    range_M[0] = range_M[0];
-    i          = range_M[1] - range_M[0];
+    range_M[0] = range_m[0];
+    i          = range_m[1] - range_m[0];
   }
 
   num_cpu_m  = 0;
diff --git a/driver/others/memory.c b/driver/others/memory.c
index ac9c87850..feb45eb58 100644
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@@ -389,12 +389,13 @@ static void *alloc_mmap(void *address){
       if (map_address != (void *)-1) {
 		  
 #ifdef OS_LINUX
-#ifdef DEBUG
-		  int ret;
+#if 1
+	//#ifdef DEBUG
+		  int ret=0;
 		  ret=my_mbind(map_address, BUFFER_SIZE * SCALING, MPOL_PREFERRED, NULL, 0, 0);
 		  if(ret==-1){
 			  int errsv=errno;
-			  perror("alloc_mmap:");
+			  perror("OpenBLAS alloc_mmap:");
 			  printf("error code=%d,\tmap_address=%lx\n",errsv,map_address);
 		  }
 
diff --git a/driver/others/parameter.c b/driver/others/parameter.c
index fc7f0447e..21f56e889 100644
--- a/driver/others/parameter.c
+++ b/driver/others/parameter.c
@@ -696,5 +696,20 @@ void blas_set_parameter(void){
   }
 #endif
 #endif
+
+#if defined(LOONGSON3B)
+#ifdef SMP
+  if(blas_num_threads == 1 || blas_num_threads == 2){
+#endif
+    //single thread
+    dgemm_r = 640;
+#ifdef SMP
+  }else{
+    //multi thread
+    dgemm_r = 160;
+  }
+#endif
+#endif 
+
 }
 #endif
diff --git a/getarch.c b/getarch.c
index df052df8a..5b614472a 100644
--- a/getarch.c
+++ b/getarch.c
@@ -117,6 +117,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* #define FORCE_CELL		*/
 /* #define FORCE_SICORTEX	*/
 /* #define FORCE_LOONGSON3A      */
+/* #define FORCE_LOONGSON3B      */
 /* #define FORCE_ITANIUM2	*/
 /* #define FORCE_GENERIC	*/
 /* #define FORCE_SPARC		*/
@@ -548,6 +549,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_LOONGSON3B
+#define FORCE
+#define ARCHITECTURE    "MIPS"
+#define SUBARCHITECTURE "LOONGSON3B"
+#define SUBDIRNAME      "mips64"
+#define ARCHCONFIG   "-DLOONGSON3B " \
+       "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=32 " \
+       "-DL2_SIZE=512488 -DL2_LINESIZE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=4 "
+#define LIBNAME   "loongson3b"
+#define CORENAME  "LOONGSON3B"
+#else
+#endif
+
 #ifdef FORCE_ITANIUM2
 #define FORCE
 #define ARCHITECTURE    "IA64"
diff --git a/kernel/Makefile.L3 b/kernel/Makefile.L3
index 4e331a445..4f419dc80 100644
--- a/kernel/Makefile.L3
+++ b/kernel/Makefile.L3
@@ -498,6 +498,91 @@ $(KDIR)xgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMD
 $(KDIR)xgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
 	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DCC $< -o $@
 
+ifeq ($(TARGET), LOONGSON3B)												 
+$(KDIR)strmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+
+$(KDIR)strmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+
+$(KDIR)strmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+
+$(KDIR)strmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+
+$(KDIR)dtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+
+$(KDIR)dtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+
+$(KDIR)dtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+
+$(KDIR)dtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+
+$(KDIR)qtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+
+$(KDIR)qtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+
+$(KDIR)qtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+
+$(KDIR)qtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+
+$(KDIR)ctrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ctrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ctrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+
+$(KDIR)ctrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+
+$(KDIR)ctrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ctrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ctrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+
+$(KDIR)ctrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+
+$(KDIR)ztrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ztrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ztrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+
+$(KDIR)ztrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+
+$(KDIR)ztrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ztrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+
+$(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+
+$(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
+	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+else
 $(KDIR)strmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
@@ -581,6 +666,7 @@ $(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
 
 $(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+endif
 
 $(KDIR)xtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
diff --git a/kernel/generic/gemmkernel_2x2.c b/kernel/generic/gemmkernel_2x2.c
new file mode 100644
index 000000000..3645ef154
--- /dev/null
+++ b/kernel/generic/gemmkernel_2x2.c
@@ -0,0 +1,157 @@
+#include "common.h"
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
+#ifdef TRMMKERNEL
+		,BLASLONG offset
+#endif
+		) 
+{
+   BLASLONG i,j,k;
+   FLOAT *C0,*C1,*ptrba,*ptrbb;
+   FLOAT res0,res1,res2,res3,load0,load1,load2,load3,load4,load5,load6,load7;
+   for (j=0; j<bn/2; j+=1) 
+     {
+        C0 = C;
+        C1 = C0+ldc;
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+             for (k=0; k<bk/4; k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  load4 = ptrba[2*1+0];
+                  load5 = ptrbb[2*1+0];
+                  res0 = res0+load4*load5;
+                  load6 = ptrba[2*1+1];
+                  res1 = res1+load6*load5;
+                  load7 = ptrbb[2*1+1];
+                  res2 = res2+load4*load7;
+                  res3 = res3+load6*load7;
+                  load0 = ptrba[2*2+0];
+                  load1 = ptrbb[2*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*2+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*2+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  load4 = ptrba[2*3+0];
+                  load5 = ptrbb[2*3+0];
+                  res0 = res0+load4*load5;
+                  load6 = ptrba[2*3+1];
+                  res1 = res1+load6*load5;
+                  load7 = ptrbb[2*3+1];
+                  res2 = res2+load4*load7;
+                  res3 = res3+load6*load7;
+                  ptrba = ptrba+8;
+                  ptrbb = ptrbb+8;
+               }
+             for (k=0; k<(bk&3); k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+2;
+               }
+             res0 = res0*alpha;
+             C0[0] = C0[0]+res0;
+             res1 = res1*alpha;
+             C0[1] = C0[1]+res1;
+             res2 = res2*alpha;
+             C1[0] = C1[0]+res2;
+             res3 = res3*alpha;
+             C1[1] = C1[1]+res3;
+             C0 = C0+2;
+             C1 = C1+2;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+                  load0 = ptrba[0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrbb[2*0+1];
+                  res1 = res1+load0*load2;
+                  ptrba = ptrba+1;
+                  ptrbb = ptrbb+2;
+               }
+             res0 = res0*alpha;
+             C0[0] = C0[0]+res0;
+             res1 = res1*alpha;
+             C1[0] = C1[0]+res1;
+             C0 = C0+1;
+             C1 = C1+1;
+          }
+        k = (bk<<1);
+        bb = bb+k;
+        i = (ldc<<1);
+        C = C+i;
+     }
+   for (j=0; j<(bn&1); j+=1) 
+     {
+        C0 = C;
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+1;
+               }
+             res0 = res0*alpha;
+             C0[0] = C0[0]+res0;
+             res1 = res1*alpha;
+             C0[1] = C0[1]+res1;
+             C0 = C0+2;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+                  load0 = ptrba[0+0];
+                  load1 = ptrbb[0+0];
+                  res0 = res0+load0*load1;
+                  ptrba = ptrba+1;
+                  ptrbb = ptrbb+1;
+               }
+             res0 = res0*alpha;
+             C0[0] = C0[0]+res0;
+             C0 = C0+1;
+          }
+        k = (bk<<0);
+        bb = bb+k;
+        C = C+ldc;
+     }
+   return 0;
+}
diff --git a/kernel/generic/trmmkernel_2x2.c b/kernel/generic/trmmkernel_2x2.c
new file mode 100644
index 000000000..5b16806dc
--- /dev/null
+++ b/kernel/generic/trmmkernel_2x2.c
@@ -0,0 +1,280 @@
+#include "common.h"
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
+#ifdef TRMMKERNEL
+		,BLASLONG offset
+#endif
+		) 
+{
+   BLASLONG i,j,k;
+   FLOAT *C0,*C1,*ptrba,*ptrbb;
+   FLOAT res0,res1,res2,res3,load0,load1,load2,load3,load4,load5,load6,load7;
+   BLASLONG off, temp;
+#if defined(TRMMKERNEL) && !defined(LEFT)
+   off = -offset; 
+#endif
+   for (j=0; j<bn/2; j+=1) 
+     {
+        C0 = C;
+        C1 = C0+ldc;
+#if defined(TRMMKERNEL) && defined(LEFT)
+		off = offset;
+#endif
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+#if (defined(LEFT) &&  defined(TRANSA)) || \
+			  (!defined(LEFT) && !defined(TRANSA))
+             ptrbb = bb;
+#else
+			  ptrba += off*2;
+			  ptrbb = bb + off*2;
+#endif
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || \
+			 (!defined(LEFT) && defined(TRANSA))
+			 temp = bk-off;
+#elif defined(LEFT) 
+			 temp = off+2;
+#else
+			 temp = off+2;
+#endif
+             for (k=0; k<temp/4; k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  load4 = ptrba[2*1+0];
+                  load5 = ptrbb[2*1+0];
+                  res0 = res0+load4*load5;
+                  load6 = ptrba[2*1+1];
+                  res1 = res1+load6*load5;
+                  load7 = ptrbb[2*1+1];
+                  res2 = res2+load4*load7;
+                  res3 = res3+load6*load7;
+                  load0 = ptrba[2*2+0];
+                  load1 = ptrbb[2*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*2+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*2+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  load4 = ptrba[2*3+0];
+                  load5 = ptrbb[2*3+0];
+                  res0 = res0+load4*load5;
+                  load6 = ptrba[2*3+1];
+                  res1 = res1+load6*load5;
+                  load7 = ptrbb[2*3+1];
+                  res2 = res2+load4*load7;
+                  res3 = res3+load6*load7;
+                  ptrba = ptrba+8;
+                  ptrbb = ptrbb+8;
+               }
+             for (k=0; k<(temp&3); k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res2 = res2+load0*load3;
+                  res3 = res3+load2*load3;
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+2;
+               }
+             res0 = res0*alpha;
+             C0[0] = res0;
+             res1 = res1*alpha;
+             C0[1] = res1;
+             res2 = res2*alpha;
+             C1[0] = res2;
+             res3 = res3*alpha;
+             C1[1] = res3;
+#if ( defined(LEFT) && defined(TRANSA)) || \
+			 (!defined(LEFT) && !defined(TRANSA)) 
+			 temp = bk - off;
+#ifdef LEFT
+			 temp -= 2;
+#else 
+			 temp -= 2;
+#endif
+			 ptrba += temp*2;
+			 ptrbb += temp*2;
+#endif
+#ifdef LEFT
+			 off += 2;
+#endif
+             C0 = C0+2;
+             C1 = C1+2;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+#if (defined(LEFT) &&  defined(TRANSA)) ||(!defined(LEFT) && !defined(TRANSA))
+             ptrbb = bb;
+#else 
+			 ptrba += off;
+			 ptrbb = bb+off*2;
+#endif
+             res0 = 0;
+             res1 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			 temp = bk-off;
+#elif defined(LEFT)
+			 temp = off+1;
+#else 
+			 temp = off+2;
+#endif
+             for (k=0; k<temp; k+=1) 
+               {
+                  load0 = ptrba[0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrbb[2*0+1];
+                  res1 = res1+load0*load2;
+                  ptrba = ptrba+1;
+                  ptrbb = ptrbb+2;
+               }
+             res0 = res0*alpha;
+             C0[0] = res0;
+             res1 = res1*alpha;
+             C1[0] = res1;
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA)) 
+			 temp = bk-off;
+#ifdef LEFT
+			 temp -= 1;
+#else 
+			 temp -= 2;
+#endif
+			 ptrba += temp;
+			 ptrbb += temp*2;
+#endif
+#ifdef LEFT	
+			 off += 1;
+#endif
+             C0 = C0+1;
+             C1 = C1+1;
+          }
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 2;
+#endif
+        k = (bk<<1);
+        bb = bb+k;
+        i = (ldc<<1);
+        C = C+i;
+     }
+   for (j=0; j<(bn&1); j+=1) 
+     {
+        C0 = C;
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+		off = offset;
+#endif
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+#if (defined(LEFT) &&  defined(TRANSA)) || \
+			  (!defined(LEFT) && !defined(TRANSA))
+			  ptrbb = bb;
+#else 
+			  ptrba += off*2;
+			  ptrbb = bb + off;
+#endif
+             res0 = 0;
+             res1 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || \
+			 (!defined(LEFT) && defined(TRANSA))
+			 temp = bk-off;
+#elif defined(LEFT)
+			 temp = off+2;
+#else
+			 temp = off+1;
+#endif
+			 for (k=0; k<temp; k+=1) 
+               {
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+1;
+               }
+             res0 = res0*alpha;
+             C0[0] = res0;
+             res1 = res1*alpha;
+			 C0[1] = res1;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			 (!defined(LEFT) && !defined(TRANSA))
+			 temp = bk - off;
+#ifdef LEFT
+			 temp -= 2;
+#else 
+			 temp -= 1;
+#endif
+			 ptrba += temp*2;
+			 ptrbb += temp;
+#endif
+#ifdef LEFT
+			 off += 2;
+#endif
+
+             C0 = C0+2;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+             ptrbb = bb;
+#else 
+			 ptrba += off;
+			 ptrbb = bb+off;
+#endif
+             res0 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			 temp = bk-off;
+#elif defined(LEFT)
+			 temp = off + 1;
+#else 
+			 temp = off + 1;
+#endif
+			 for (k=0; k<temp; k+=1) 
+               {
+                  load0 = ptrba[0+0];
+                  load1 = ptrbb[0+0];
+                  res0 = res0+load0*load1;
+                  ptrba = ptrba+1;
+                  ptrbb = ptrbb+1;
+               }
+             res0 = res0*alpha;
+             C0[0] = res0;
+#if ( defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			 temp = bk-off;
+#ifdef LEFT
+			 temp -= 1;
+#else 
+			 temp -= 1;
+#endif
+			 ptrba += temp;
+			 ptrbb += temp;
+#endif
+#ifdef LEFT
+			 off += 1;
+#endif
+             C0 = C0+1;
+          }
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 1;
+#endif
+        k = (bk<<0);
+        bb = bb+k;
+        C = C+ldc;
+     }
+   return 0;
+}
diff --git a/kernel/generic/zgemmkernel_2x2.c b/kernel/generic/zgemmkernel_2x2.c
new file mode 100644
index 000000000..cb2a26e58
--- /dev/null
+++ b/kernel/generic/zgemmkernel_2x2.c
@@ -0,0 +1,838 @@
+#include "common.h"
+/********************************
+  ADD1 a*c
+  ADD2 b*c
+  ADD3 a*d
+  ADD4 b*d
+*********************************/
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alphar,FLOAT alphai,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
+#ifdef	TRMMKERNEL
+		, BLASLONG offset
+#endif
+		)
+{
+   BLASLONG i,j,k;
+   FLOAT *C0,*C1,*ptrba,*ptrbb;
+   FLOAT res0,res1,res2,res3,res4,res5,res6,res7,load0,load1,load2,load3,load4,load5,load6,load7,load8,load9,load10,load11,load12,load13,load14,load15;
+   for (j=0; j<bn/2; j+=1) 
+     {
+        C0 = C;
+        C1 = C0+2*ldc;
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+             res4 = 0;
+             res5 = 0;
+             res6 = 0;
+             res7 = 0;
+             for (k=0; k<bk/4; k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7+load4*load7;
+                  load8 = ptrba[4*1+0];
+                  load9 = ptrbb[4*1+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*1+1];
+                  res1 = res1+load10*load9;
+                  load11 = ptrbb[4*1+1];
+                  res0 = res0-load10*load11;
+                  res1 = res1+load8*load11;
+                  load12 = ptrba[4*1+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*1+3];
+                  res3 = res3+load13*load9;
+                  res2 = res2-load13*load11;
+                  res3 = res3+load12*load11;
+                  load14 = ptrbb[4*1+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5+load10*load14;
+                  load15 = ptrbb[4*1+3];
+                  res4 = res4-load10*load15;
+                  res5 = res5+load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7+load13*load14;
+                  res6 = res6-load13*load15;
+                  res7 = res7+load12*load15;
+                  load0 = ptrba[4*2+0];
+                  load1 = ptrbb[4*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*2+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*2+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*2+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*2+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*2+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*2+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7+load4*load7;
+                  load8 = ptrba[4*3+0];
+                  load9 = ptrbb[4*3+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*3+1];
+                  res1 = res1+load10*load9;
+                  load11 = ptrbb[4*3+1];
+                  res0 = res0-load10*load11;
+                  res1 = res1+load8*load11;
+                  load12 = ptrba[4*3+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*3+3];
+                  res3 = res3+load13*load9;
+                  res2 = res2-load13*load11;
+                  res3 = res3+load12*load11;
+                  load14 = ptrbb[4*3+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5+load10*load14;
+                  load15 = ptrbb[4*3+3];
+                  res4 = res4-load10*load15;
+                  res5 = res5+load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7+load13*load14;
+                  res6 = res6-load13*load15;
+                  res7 = res7+load12*load15;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7-load4*load7;
+                  load8 = ptrba[4*1+0];
+                  load9 = ptrbb[4*1+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*1+1];
+                  res1 = res1+load10*load9;
+                  load11 = ptrbb[4*1+1];
+                  res0 = res0+load10*load11;
+                  res1 = res1-load8*load11;
+                  load12 = ptrba[4*1+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*1+3];
+                  res3 = res3+load13*load9;
+                  res2 = res2+load13*load11;
+                  res3 = res3-load12*load11;
+                  load14 = ptrbb[4*1+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5+load10*load14;
+                  load15 = ptrbb[4*1+3];
+                  res4 = res4+load10*load15;
+                  res5 = res5-load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7+load13*load14;
+                  res6 = res6+load13*load15;
+                  res7 = res7-load12*load15;
+                  load0 = ptrba[4*2+0];
+                  load1 = ptrbb[4*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*2+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*2+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*2+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*2+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*2+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*2+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7-load4*load7;
+                  load8 = ptrba[4*3+0];
+                  load9 = ptrbb[4*3+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*3+1];
+                  res1 = res1+load10*load9;
+                  load11 = ptrbb[4*3+1];
+                  res0 = res0+load10*load11;
+                  res1 = res1-load8*load11;
+                  load12 = ptrba[4*3+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*3+3];
+                  res3 = res3+load13*load9;
+                  res2 = res2+load13*load11;
+                  res3 = res3-load12*load11;
+                  load14 = ptrbb[4*3+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5+load10*load14;
+                  load15 = ptrbb[4*3+3];
+                  res4 = res4+load10*load15;
+                  res5 = res5-load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7+load13*load14;
+                  res6 = res6+load13*load15;
+                  res7 = res7-load12*load15;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7+load4*load7;
+                  load8 = ptrba[4*1+0];
+                  load9 = ptrbb[4*1+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*1+1];
+                  res1 = res1-load10*load9;
+                  load11 = ptrbb[4*1+1];
+                  res0 = res0+load10*load11;
+                  res1 = res1+load8*load11;
+                  load12 = ptrba[4*1+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*1+3];
+                  res3 = res3-load13*load9;
+                  res2 = res2+load13*load11;
+                  res3 = res3+load12*load11;
+                  load14 = ptrbb[4*1+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5-load10*load14;
+                  load15 = ptrbb[4*1+3];
+                  res4 = res4+load10*load15;
+                  res5 = res5+load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7-load13*load14;
+                  res6 = res6+load13*load15;
+                  res7 = res7+load12*load15;
+                  load0 = ptrba[4*2+0];
+                  load1 = ptrbb[4*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*2+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*2+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*2+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*2+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*2+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*2+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7+load4*load7;
+                  load8 = ptrba[4*3+0];
+                  load9 = ptrbb[4*3+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*3+1];
+                  res1 = res1-load10*load9;
+                  load11 = ptrbb[4*3+1];
+                  res0 = res0+load10*load11;
+                  res1 = res1+load8*load11;
+                  load12 = ptrba[4*3+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*3+3];
+                  res3 = res3-load13*load9;
+                  res2 = res2+load13*load11;
+                  res3 = res3+load12*load11;
+                  load14 = ptrbb[4*3+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5-load10*load14;
+                  load15 = ptrbb[4*3+3];
+                  res4 = res4+load10*load15;
+                  res5 = res5+load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7-load13*load14;
+                  res6 = res6+load13*load15;
+                  res7 = res7+load12*load15;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7-load4*load7;
+                  load8 = ptrba[4*1+0];
+                  load9 = ptrbb[4*1+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*1+1];
+                  res1 = res1-load10*load9;
+                  load11 = ptrbb[4*1+1];
+                  res0 = res0-load10*load11;
+                  res1 = res1-load8*load11;
+                  load12 = ptrba[4*1+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*1+3];
+                  res3 = res3-load13*load9;
+                  res2 = res2-load13*load11;
+                  res3 = res3-load12*load11;
+                  load14 = ptrbb[4*1+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5-load10*load14;
+                  load15 = ptrbb[4*1+3];
+                  res4 = res4-load10*load15;
+                  res5 = res5-load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7-load13*load14;
+                  res6 = res6-load13*load15;
+                  res7 = res7-load12*load15;
+                  load0 = ptrba[4*2+0];
+                  load1 = ptrbb[4*2+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*2+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*2+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*2+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*2+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*2+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*2+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7-load4*load7;
+                  load8 = ptrba[4*3+0];
+                  load9 = ptrbb[4*3+0];
+                  res0 = res0+load8*load9;
+                  load10 = ptrba[4*3+1];
+                  res1 = res1-load10*load9;
+                  load11 = ptrbb[4*3+1];
+                  res0 = res0-load10*load11;
+                  res1 = res1-load8*load11;
+                  load12 = ptrba[4*3+2];
+                  res2 = res2+load12*load9;
+                  load13 = ptrba[4*3+3];
+                  res3 = res3-load13*load9;
+                  res2 = res2-load13*load11;
+                  res3 = res3-load12*load11;
+                  load14 = ptrbb[4*3+2];
+                  res4 = res4+load8*load14;
+                  res5 = res5-load10*load14;
+                  load15 = ptrbb[4*3+3];
+                  res4 = res4-load10*load15;
+                  res5 = res5-load8*load15;
+                  res6 = res6+load12*load14;
+                  res7 = res7-load13*load14;
+                  res6 = res6-load13*load15;
+                  res7 = res7-load12*load15;
+#endif
+                  ptrba = ptrba+16;
+                  ptrbb = ptrbb+16;
+               }
+             for (k=0; k<(bk&3); k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7+load4*load7;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5+load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7+load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7-load4*load7;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3+load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4+load2*load7;
+                  res5 = res5+load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6+load5*load7;
+                  res7 = res7+load4*load7;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3-load4*load3;
+                  load6 = ptrbb[4*0+2];
+                  res4 = res4+load0*load6;
+                  res5 = res5-load2*load6;
+                  load7 = ptrbb[4*0+3];
+                  res4 = res4-load2*load7;
+                  res5 = res5-load0*load7;
+                  res6 = res6+load4*load6;
+                  res7 = res7-load5*load6;
+                  res6 = res6-load5*load7;
+                  res7 = res7-load4*load7;
+#endif
+                  ptrba = ptrba+4;
+                  ptrbb = ptrbb+4;
+               }
+             load0 = res0*alphar;
+             C0[0] = C0[0]+load0;
+             load1 = res1*alphar;
+             C0[1] = C0[1]+load1;
+             load0 = res1*alphai;
+             C0[0] = C0[0]-load0;
+             load1 = res0*alphai;
+             C0[1] = C0[1]+load1;
+             load2 = res2*alphar;
+             C0[2] = C0[2]+load2;
+             load3 = res3*alphar;
+             C0[3] = C0[3]+load3;
+             load2 = res3*alphai;
+             C0[2] = C0[2]-load2;
+             load3 = res2*alphai;
+             C0[3] = C0[3]+load3;
+             load4 = res4*alphar;
+             C1[0] = C1[0]+load4;
+             load5 = res5*alphar;
+             C1[1] = C1[1]+load5;
+             load4 = res5*alphai;
+             C1[0] = C1[0]-load4;
+             load5 = res4*alphai;
+             C1[1] = C1[1]+load5;
+             load6 = res6*alphar;
+             C1[2] = C1[2]+load6;
+             load7 = res7*alphar;
+             C1[3] = C1[3]+load7;
+             load6 = res7*alphai;
+             C1[2] = C1[2]-load6;
+             load7 = res6*alphai;
+             C1[3] = C1[3]+load7;
+             C0 = C0+4;
+             C1 = C1+4;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrbb[4*0+2];
+                  res2 = res2+load0*load4;
+                  res3 = res3+load2*load4;
+                  load5 = ptrbb[4*0+3];
+                  res2 = res2-load2*load5;
+                  res3 = res3+load0*load5;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrbb[4*0+2];
+                  res2 = res2+load0*load4;
+                  res3 = res3+load2*load4;
+                  load5 = ptrbb[4*0+3];
+                  res2 = res2+load2*load5;
+                  res3 = res3-load0*load5;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrbb[4*0+2];
+                  res2 = res2+load0*load4;
+                  res3 = res3-load2*load4;
+                  load5 = ptrbb[4*0+3];
+                  res2 = res2+load2*load5;
+                  res3 = res3+load0*load5;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[4*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[4*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrbb[4*0+2];
+                  res2 = res2+load0*load4;
+                  res3 = res3-load2*load4;
+                  load5 = ptrbb[4*0+3];
+                  res2 = res2-load2*load5;
+                  res3 = res3-load0*load5;
+#endif
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+4;
+               }
+             load0 = res0*alphar;
+             C0[0] = C0[0]+load0;
+             load1 = res1*alphar;
+             C0[1] = C0[1]+load1;
+             load0 = res1*alphai;
+             C0[0] = C0[0]-load0;
+             load1 = res0*alphai;
+             C0[1] = C0[1]+load1;
+             load2 = res2*alphar;
+             C1[0] = C1[0]+load2;
+             load3 = res3*alphar;
+             C1[1] = C1[1]+load3;
+             load2 = res3*alphai;
+             C1[0] = C1[0]-load2;
+             load3 = res2*alphai;
+             C1[1] = C1[1]+load3;
+             C0 = C0+2;
+             C1 = C1+2;
+          }
+        k = (bk<<2);
+        bb = bb+k;
+        i = (ldc<<2);
+        C = C+i;
+     }
+   for (j=0; j<(bn&1); j+=1) 
+     {
+        C0 = C;
+        ptrba = ba;
+        for (i=0; i<bm/2; i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             res2 = 0;
+             res3 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3+load4*load3;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3+load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3-load4*load3;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2+load5*load3;
+                  res3 = res3+load4*load3;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[4*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[4*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+                  load4 = ptrba[4*0+2];
+                  res2 = res2+load4*load1;
+                  load5 = ptrba[4*0+3];
+                  res3 = res3-load5*load1;
+                  res2 = res2-load5*load3;
+                  res3 = res3-load4*load3;
+#endif
+                  ptrba = ptrba+4;
+                  ptrbb = ptrbb+2;
+               }
+             load0 = res0*alphar;
+             C0[0] = C0[0]+load0;
+             load1 = res1*alphar;
+             C0[1] = C0[1]+load1;
+             load0 = res1*alphai;
+             C0[0] = C0[0]-load0;
+             load1 = res0*alphai;
+             C0[1] = C0[1]+load1;
+             load2 = res2*alphar;
+             C0[2] = C0[2]+load2;
+             load3 = res3*alphar;
+             C0[3] = C0[3]+load3;
+             load2 = res3*alphai;
+             C0[2] = C0[2]-load2;
+             load3 = res2*alphai;
+             C0[3] = C0[3]+load3;
+             C0 = C0+4;
+          }
+        for (i=0; i<(bm&1); i+=1) 
+          {
+             ptrbb = bb;
+             res0 = 0;
+             res1 = 0;
+             for (k=0; k<bk; k+=1) 
+               {
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1+load0*load3;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1+load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1-load0*load3;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0+load2*load3;
+                  res1 = res1+load0*load3;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+                  load0 = ptrba[2*0+0];
+                  load1 = ptrbb[2*0+0];
+                  res0 = res0+load0*load1;
+                  load2 = ptrba[2*0+1];
+                  res1 = res1-load2*load1;
+                  load3 = ptrbb[2*0+1];
+                  res0 = res0-load2*load3;
+                  res1 = res1-load0*load3;
+#endif
+                  ptrba = ptrba+2;
+                  ptrbb = ptrbb+2;
+               }
+             load0 = res0*alphar;
+             C0[0] = C0[0]+load0;
+             load1 = res1*alphar;
+             C0[1] = C0[1]+load1;
+             load0 = res1*alphai;
+             C0[0] = C0[0]-load0;
+             load1 = res0*alphai;
+             C0[1] = C0[1]+load1;
+             C0 = C0+2;
+          }
+        k = (bk<<1);
+        bb = bb+k;
+        i = (ldc<<1);
+        C = C+i;
+     }
+   return 0;
+}
diff --git a/kernel/generic/ztrmmkernel_2x2.c b/kernel/generic/ztrmmkernel_2x2.c
new file mode 100644
index 000000000..b7c6539c2
--- /dev/null
+++ b/kernel/generic/ztrmmkernel_2x2.c
@@ -0,0 +1,923 @@
+#include "common.h"
+/********************************
+  ADD1 a*c
+  ADD2 b*c
+  ADD3 a*d
+  ADD4 b*d
+ *********************************/
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alphar,FLOAT alphai,FLOAT* ba,FLOAT* bb,
+		FLOAT* C,BLASLONG ldc, BLASLONG offset)
+{
+	BLASLONG i,j,k;
+	FLOAT *C0,*C1,*ptrba,*ptrbb;
+	FLOAT res0,res1,res2,res3,res4,res5,res6,res7,load0,load1,load2,load3,load4,load5,load6,load7,load8,load9,load10,load11,load12,load13,load14,load15;
+	BLASLONG off, temp;
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	off = -offset;
+#endif
+	for (j=0; j<bn/2; j+=1) 
+	{
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+		off = offset;
+#endif
+		C0 = C;
+		C1 = C0+2*ldc;
+		ptrba = ba;
+		for (i=0; i<bm/2; i+=1) 
+		{
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			ptrbb = bb;
+#else
+			ptrba += off*2*2;
+			ptrbb = bb+off*2*2;
+#endif
+			res0 = 0;
+			res1 = 0;
+			res2 = 0;
+			res3 = 0;
+			res4 = 0;
+			res5 = 0;
+			res6 = 0;
+			res7 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			temp = bk - off;
+#elif defined(LEFT)
+			temp = off + 2;
+#else 
+			temp = off + 2;
+#endif
+			for (k=0; k<temp/4; k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4-load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7+load4*load7;
+				load8 = ptrba[4*1+0];
+				load9 = ptrbb[4*1+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*1+1];
+				res1 = res1+load10*load9;
+				load11 = ptrbb[4*1+1];
+				res0 = res0-load10*load11;
+				res1 = res1+load8*load11;
+				load12 = ptrba[4*1+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*1+3];
+				res3 = res3+load13*load9;
+				res2 = res2-load13*load11;
+				res3 = res3+load12*load11;
+				load14 = ptrbb[4*1+2];
+				res4 = res4+load8*load14;
+				res5 = res5+load10*load14;
+				load15 = ptrbb[4*1+3];
+				res4 = res4-load10*load15;
+				res5 = res5+load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7+load13*load14;
+				res6 = res6-load13*load15;
+				res7 = res7+load12*load15;
+				load0 = ptrba[4*2+0];
+				load1 = ptrbb[4*2+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*2+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*2+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*2+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*2+3];
+				res3 = res3+load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*2+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*2+3];
+				res4 = res4-load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7+load4*load7;
+				load8 = ptrba[4*3+0];
+				load9 = ptrbb[4*3+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*3+1];
+				res1 = res1+load10*load9;
+				load11 = ptrbb[4*3+1];
+				res0 = res0-load10*load11;
+				res1 = res1+load8*load11;
+				load12 = ptrba[4*3+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*3+3];
+				res3 = res3+load13*load9;
+				res2 = res2-load13*load11;
+				res3 = res3+load12*load11;
+				load14 = ptrbb[4*3+2];
+				res4 = res4+load8*load14;
+				res5 = res5+load10*load14;
+				load15 = ptrbb[4*3+3];
+				res4 = res4-load10*load15;
+				res5 = res5+load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7+load13*load14;
+				res6 = res6-load13*load15;
+				res7 = res7+load12*load15;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4+load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7-load4*load7;
+				load8 = ptrba[4*1+0];
+				load9 = ptrbb[4*1+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*1+1];
+				res1 = res1+load10*load9;
+				load11 = ptrbb[4*1+1];
+				res0 = res0+load10*load11;
+				res1 = res1-load8*load11;
+				load12 = ptrba[4*1+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*1+3];
+				res3 = res3+load13*load9;
+				res2 = res2+load13*load11;
+				res3 = res3-load12*load11;
+				load14 = ptrbb[4*1+2];
+				res4 = res4+load8*load14;
+				res5 = res5+load10*load14;
+				load15 = ptrbb[4*1+3];
+				res4 = res4+load10*load15;
+				res5 = res5-load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7+load13*load14;
+				res6 = res6+load13*load15;
+				res7 = res7-load12*load15;
+				load0 = ptrba[4*2+0];
+				load1 = ptrbb[4*2+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*2+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*2+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*2+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*2+3];
+				res3 = res3+load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*2+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*2+3];
+				res4 = res4+load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7-load4*load7;
+				load8 = ptrba[4*3+0];
+				load9 = ptrbb[4*3+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*3+1];
+				res1 = res1+load10*load9;
+				load11 = ptrbb[4*3+1];
+				res0 = res0+load10*load11;
+				res1 = res1-load8*load11;
+				load12 = ptrba[4*3+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*3+3];
+				res3 = res3+load13*load9;
+				res2 = res2+load13*load11;
+				res3 = res3-load12*load11;
+				load14 = ptrbb[4*3+2];
+				res4 = res4+load8*load14;
+				res5 = res5+load10*load14;
+				load15 = ptrbb[4*3+3];
+				res4 = res4+load10*load15;
+				res5 = res5-load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7+load13*load14;
+				res6 = res6+load13*load15;
+				res7 = res7-load12*load15;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4+load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7+load4*load7;
+				load8 = ptrba[4*1+0];
+				load9 = ptrbb[4*1+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*1+1];
+				res1 = res1-load10*load9;
+				load11 = ptrbb[4*1+1];
+				res0 = res0+load10*load11;
+				res1 = res1+load8*load11;
+				load12 = ptrba[4*1+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*1+3];
+				res3 = res3-load13*load9;
+				res2 = res2+load13*load11;
+				res3 = res3+load12*load11;
+				load14 = ptrbb[4*1+2];
+				res4 = res4+load8*load14;
+				res5 = res5-load10*load14;
+				load15 = ptrbb[4*1+3];
+				res4 = res4+load10*load15;
+				res5 = res5+load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7-load13*load14;
+				res6 = res6+load13*load15;
+				res7 = res7+load12*load15;
+				load0 = ptrba[4*2+0];
+				load1 = ptrbb[4*2+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*2+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*2+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*2+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*2+3];
+				res3 = res3-load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*2+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*2+3];
+				res4 = res4+load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7+load4*load7;
+				load8 = ptrba[4*3+0];
+				load9 = ptrbb[4*3+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*3+1];
+				res1 = res1-load10*load9;
+				load11 = ptrbb[4*3+1];
+				res0 = res0+load10*load11;
+				res1 = res1+load8*load11;
+				load12 = ptrba[4*3+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*3+3];
+				res3 = res3-load13*load9;
+				res2 = res2+load13*load11;
+				res3 = res3+load12*load11;
+				load14 = ptrbb[4*3+2];
+				res4 = res4+load8*load14;
+				res5 = res5-load10*load14;
+				load15 = ptrbb[4*3+3];
+				res4 = res4+load10*load15;
+				res5 = res5+load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7-load13*load14;
+				res6 = res6+load13*load15;
+				res7 = res7+load12*load15;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4-load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7-load4*load7;
+				load8 = ptrba[4*1+0];
+				load9 = ptrbb[4*1+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*1+1];
+				res1 = res1-load10*load9;
+				load11 = ptrbb[4*1+1];
+				res0 = res0-load10*load11;
+				res1 = res1-load8*load11;
+				load12 = ptrba[4*1+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*1+3];
+				res3 = res3-load13*load9;
+				res2 = res2-load13*load11;
+				res3 = res3-load12*load11;
+				load14 = ptrbb[4*1+2];
+				res4 = res4+load8*load14;
+				res5 = res5-load10*load14;
+				load15 = ptrbb[4*1+3];
+				res4 = res4-load10*load15;
+				res5 = res5-load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7-load13*load14;
+				res6 = res6-load13*load15;
+				res7 = res7-load12*load15;
+				load0 = ptrba[4*2+0];
+				load1 = ptrbb[4*2+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*2+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*2+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*2+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*2+3];
+				res3 = res3-load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*2+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*2+3];
+				res4 = res4-load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7-load4*load7;
+				load8 = ptrba[4*3+0];
+				load9 = ptrbb[4*3+0];
+				res0 = res0+load8*load9;
+				load10 = ptrba[4*3+1];
+				res1 = res1-load10*load9;
+				load11 = ptrbb[4*3+1];
+				res0 = res0-load10*load11;
+				res1 = res1-load8*load11;
+				load12 = ptrba[4*3+2];
+				res2 = res2+load12*load9;
+				load13 = ptrba[4*3+3];
+				res3 = res3-load13*load9;
+				res2 = res2-load13*load11;
+				res3 = res3-load12*load11;
+				load14 = ptrbb[4*3+2];
+				res4 = res4+load8*load14;
+				res5 = res5-load10*load14;
+				load15 = ptrbb[4*3+3];
+				res4 = res4-load10*load15;
+				res5 = res5-load8*load15;
+				res6 = res6+load12*load14;
+				res7 = res7-load13*load14;
+				res6 = res6-load13*load15;
+				res7 = res7-load12*load15;
+#endif
+				ptrba = ptrba+16;
+				ptrbb = ptrbb+16;
+			}
+			for (k=0; k<(temp&3); k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4-load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7+load4*load7;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5+load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4+load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7+load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7-load4*load7;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3+load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4+load2*load7;
+				res5 = res5+load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6+load5*load7;
+				res7 = res7+load4*load7;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3-load4*load3;
+				load6 = ptrbb[4*0+2];
+				res4 = res4+load0*load6;
+				res5 = res5-load2*load6;
+				load7 = ptrbb[4*0+3];
+				res4 = res4-load2*load7;
+				res5 = res5-load0*load7;
+				res6 = res6+load4*load6;
+				res7 = res7-load5*load6;
+				res6 = res6-load5*load7;
+				res7 = res7-load4*load7;
+#endif
+				ptrba = ptrba+4;
+				ptrbb = ptrbb+4;
+			}
+			load0 = res0*alphar-res1*alphai;
+			load1 = res1*alphar+res0*alphai;
+			C0[0] = load0;
+			C0[1] = load1;
+
+			load2 = res2*alphar-res3*alphai;
+			load3 = res3*alphar+res2*alphai;
+			C0[2] = load2;
+			C0[3] = load3;
+
+			load4 = res4*alphar-res5*alphai;
+			load5 = res5*alphar+res4*alphai;
+			C1[0] = load4;
+			C1[1] = load5;
+
+			load6 = res6*alphar-res7*alphai;
+			load7 = res7*alphar+res6*alphai;
+			C1[2] = load6;
+			C1[3] = load7;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			(!defined(LEFT) && !defined(TRANSA))
+			temp = bk - off;
+#ifdef LEFT
+			temp -= 2;
+#else
+			temp -= 2;
+#endif
+			ptrba += temp*2*2;
+			ptrbb += temp*2*2;
+#endif
+#ifdef LEFT
+			off += 2;
+#endif
+
+			C0 = C0+4;
+			C1 = C1+4;
+		}
+		for (i=0; i<(bm&1); i+=1) 
+		{
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			ptrbb = bb;
+#else 
+			ptrba += off*2;
+			ptrbb = bb + off*2*2;
+#endif
+			res0 = 0;
+			res1 = 0;
+			res2 = 0;
+			res3 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			temp = bk - off;
+#elif defined(LEFT)
+			temp = off+1;
+#else
+			temp = off+2;
+#endif
+			for (k=0; k<temp; k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrbb[4*0+2];
+				res2 = res2+load0*load4;
+				res3 = res3+load2*load4;
+				load5 = ptrbb[4*0+3];
+				res2 = res2-load2*load5;
+				res3 = res3+load0*load5;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrbb[4*0+2];
+				res2 = res2+load0*load4;
+				res3 = res3+load2*load4;
+				load5 = ptrbb[4*0+3];
+				res2 = res2+load2*load5;
+				res3 = res3-load0*load5;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrbb[4*0+2];
+				res2 = res2+load0*load4;
+				res3 = res3-load2*load4;
+				load5 = ptrbb[4*0+3];
+				res2 = res2+load2*load5;
+				res3 = res3+load0*load5;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[4*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[4*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrbb[4*0+2];
+				res2 = res2+load0*load4;
+				res3 = res3-load2*load4;
+				load5 = ptrbb[4*0+3];
+				res2 = res2-load2*load5;
+				res3 = res3-load0*load5;
+#endif
+				ptrba = ptrba+2;
+				ptrbb = ptrbb+4;
+			}
+			load0 = res0*alphar-res1*alphai;
+			load1 = res1*alphar+res0*alphai;
+			C0[0] = load0;
+			C0[1] = load1;
+
+			load2 = res2*alphar-res3*alphai;
+			load3 = res3*alphar+res2*alphai;
+			C1[0] = load2;
+			C1[1] = load3;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			(!defined(LEFT) && !defined(TRANSA))
+			temp = bk - off;
+#ifdef LEFT
+			temp -= 1;
+#else 
+			temp -= 2;
+#endif
+			ptrba += temp*2;
+			ptrbb += temp*2*2;
+#endif
+#ifdef LEFT
+			off += 1;
+#endif
+			C0 = C0+2;
+			C1 = C1+2;
+		}
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 2;
+#endif
+		k = (bk<<2);
+		bb = bb+k;
+		i = (ldc<<2);
+		C = C+i;
+	}
+	for (j=0; j<(bn&1); j+=1) 
+	{
+		C0 = C;
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+		off = offset;
+#endif
+		ptrba = ba;
+		for (i=0; i<bm/2; i+=1) 
+		{
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			ptrbb = bb;
+#else 
+			ptrba += off*2*2;
+			ptrbb = bb+off*2;
+#endif
+			res0 = 0;
+			res1 = 0;
+			res2 = 0;
+			res3 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			temp = bk - off;
+#elif defined(LEFT)
+			temp = off + 2;
+#else
+			temp = off + 1;
+#endif
+			for (k=0; k<temp; k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3+load4*load3;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3+load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3-load4*load3;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2+load5*load3;
+				res3 = res3+load4*load3;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[4*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[4*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+				load4 = ptrba[4*0+2];
+				res2 = res2+load4*load1;
+				load5 = ptrba[4*0+3];
+				res3 = res3-load5*load1;
+				res2 = res2-load5*load3;
+				res3 = res3-load4*load3;
+#endif
+				ptrba = ptrba+4;
+				ptrbb = ptrbb+2;
+			}
+			load0 = res0*alphar-res1*alphai;
+			load1 = res1*alphar+res0*alphai;
+			C0[0] = load0;
+			C0[1] = load1;
+
+			load2 = res2*alphar-res3*alphai;
+			load3 = res3*alphar+res2*alphai;
+			C0[2] = load2;
+			C0[3] = load3;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			(!defined(LEFT) && !defined(TRANSA))
+			temp = bk-off;
+#ifdef LEFT
+			temp -= 2;
+#else
+			temp -= 1;
+#endif
+			ptrba += temp*2*2;
+			ptrbb += temp*2;
+#endif
+#ifdef LEFT
+			off += 2;
+#endif
+			C0 = C0+4;
+		}
+		for (i=0; i<(bm&1); i+=1) 
+		{
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+			ptrbb = bb;
+#else 
+			ptrba += off*2;
+			ptrbb = bb + off*2;
+#endif
+			res0 = 0;
+			res1 = 0;
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+			temp = bk-off;
+#elif defined(LEFT)
+			temp = off + 1;
+#else 
+			temp = off + 1;
+#endif
+			for (k=0; k<temp; k+=1) 
+			{
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1+load0*load3;
+#endif
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1+load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1-load0*load3;
+#endif
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0+load2*load3;
+				res1 = res1+load0*load3;
+#endif
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+				load0 = ptrba[2*0+0];
+				load1 = ptrbb[2*0+0];
+				res0 = res0+load0*load1;
+				load2 = ptrba[2*0+1];
+				res1 = res1-load2*load1;
+				load3 = ptrbb[2*0+1];
+				res0 = res0-load2*load3;
+				res1 = res1-load0*load3;
+#endif
+				ptrba = ptrba+2;
+				ptrbb = ptrbb+2;
+			}
+			load0 = res0*alphar-res1*alphai;
+			load1 = res1*alphar+res0*alphai;
+			C0[0] = load0;
+			C0[1] = load1;
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+			(!defined(LEFT) && !defined(TRANSA))
+			temp = bk - off;
+#ifdef LEFT
+			temp -= 1;
+#else 
+			temp -= 1;
+#endif
+			ptrba += temp*2;
+			ptrbb += temp*2;
+#endif
+#ifdef LEFT
+			off += 1;
+#endif
+			C0 = C0+2;
+		}
+		k = (bk<<1);
+		bb = bb+k;
+		i = (ldc<<1);
+		C = C+i;
+	}
+	return 0;
+}
diff --git a/kernel/mips64/KERNEL.LOONGSON3B b/kernel/mips64/KERNEL.LOONGSON3B
new file mode 100644
index 000000000..df4380d11
--- /dev/null
+++ b/kernel/mips64/KERNEL.LOONGSON3B
@@ -0,0 +1,64 @@
+SAXPYKERNEL=axpy_loongson3a.S
+DAXPYKERNEL=daxpy_loongson3a_simd.S
+
+SGEMVNKERNEL = gemv_n_loongson3a.c
+SGEMVTKERNEL = gemv_t_loongson3a.c
+DGEMVNKERNEL = gemv_n_loongson3a.c
+DGEMVTKERNEL = gemv_t_loongson3a.c
+CGEMVNKERNEL = zgemv_n_loongson3a.c
+CGEMVTKERNEL = zgemv_t_loongson3a.c
+ZGEMVNKERNEL = zgemv_n_loongson3a.c
+ZGEMVTKERNEL = zgemv_t_loongson3a.c
+
+STRMMKERNEL	= ../generic/trmmkernel_2x2.c
+DTRMMKERNEL	= ../generic/trmmkernel_2x2.c
+CTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
+ZTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
+
+SGEMMKERNEL    =  ../generic/gemmkernel_2x2.c		
+SGEMMONCOPY    =  ../generic/gemm_ncopy_2.c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_2.c
+SGEMMONCOPYOBJ =  sgemm_oncopy.o
+SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+
+DGEMMKERNEL    =  ../generic/gemmkernel_2x2.c		
+DGEMMONCOPY    = ../generic/gemm_ncopy_2.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy.o
+
+CGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+
+ZGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+ZGEMMONCOPYOBJ =  zgemm_oncopy.o
+ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+
+
+
diff --git a/kernel/mips64/cgemm_kernel_loongson3b_2x2.S b/kernel/mips64/cgemm_kernel_loongson3b_2x2.S
new file mode 100644
index 000000000..5ded7aed0
--- /dev/null
+++ b/kernel/mips64/cgemm_kernel_loongson3b_2x2.S
@@ -0,0 +1,1468 @@
+#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 STACKSIZE	160
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$9
+#define B	$10
+#define C	$11
+#define LDC	$8
+
+#define AO	$12
+#define BO	$13
+
+#define	R12	12
+#define	R13	13
+
+#define I	$2
+#define J	$3
+#define L	$7
+
+#define CO1	$14
+#define CO2	$15
+#define PREA	$16
+#define PREB	$17
+
+#if defined(TRMMKERNEL)
+#define OFFSET	$18
+#define KK	$19
+#define TEMP	$20
+#endif
+
+#define a1	$f0
+#define a2	$f1
+#define a3	$f2
+#define a4	$f3
+
+#define b1	$f4
+#define b2	$f5
+#define b3	$f6
+#define b4	$f7
+
+#define a5	$f8
+#define a6	$f9
+#define a7	$f10
+#define a8	$f11
+
+#define b5	$f12
+#define b6	$f13
+#define b7	$f15
+#define b8	$f16
+
+#define c11	$f14
+#define c12	$f17
+#define c13	$f18
+#define c14	$f19
+#define c21	$f20
+#define c22	$f21
+#define c23	$f22
+#define c24	$f23
+#define c31	$f24
+#define c32	$f25
+#define c33	$f26
+#define c34	$f27
+#define c41	$f28
+#define c42	$f29
+#define c43	$f30
+#define c44	$f31
+
+#define F0	0
+#define F1	1
+#define F2	2
+#define F3	3
+#define F4	4
+#define F5	5
+#define	F6	6
+#define	F7	7
+#define	F8	8
+#define	F9	9
+#define	F10	10
+#define	F11	11
+#define	F12	12
+#define	F13	13
+#define	F14	14
+#define	F15	15
+#define	F16	16
+#define	F17	17
+#define	F18	18
+#define	F19	19
+#define	F20	20
+#define	F21	21
+#define	F22	22
+#define	F23	23
+#define	F24	24
+#define	F25	25
+#define	F26	26
+#define	F27	27
+#define	F28	28
+#define	F29	29
+#define	F30	30
+#define	F31	31
+
+#define ALPHA_R	$f15
+#define ALPHA_I	$f16
+
+#################################
+##	MADD1	a*c
+##	MADD2	b*c
+##	MADD3	a*d
+##	MADD4	d*b
+##################################
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+#define MADD1	  MADD
+#define MADD2	  MADD
+#define MADD3	  MADD
+#define MADD4	  NMSUB
+#endif
+
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+#define MADD1	  MADD
+#define MADD2	  MADD
+#define MADD3	  NMSUB
+#define MADD4	  MADD
+#endif
+
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+#define MADD1	  MADD
+#define MADD2	  NMSUB
+#define MADD3	  MADD
+#define MADD4	  MADD
+#endif
+
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+#define MADD1	  MADD
+#define MADD2	  NMSUB
+#define MADD3	  NMSUB
+#define MADD4	  NMSUB
+#endif
+
+	PROLOGUE
+	
+	LDARG	LDC,   0($sp)
+	daddiu	$sp, $sp, -STACKSIZE
+
+	SDARG	$16,   0($sp)
+	SDARG	$17,   8($sp)
+	sdc1	$f24, 16($sp)
+	sdc1	$f25, 24($sp)
+	sdc1	$f26, 32($sp)
+	sdc1	$f27, 40($sp)
+	sdc1	$f28, 48($sp)
+	sdc1	$f29, 56($sp)
+
+#if defined(TRMMKERNEL)
+	SDARG	$18,  64($sp)
+	SDARG	$19,  72($sp)
+	SDARG	$20,  80($sp)
+
+	LDARG	OFFSET, STACKSIZE + 8($sp)
+#endif
+
+#ifndef __64BIT__
+	sdc1	$f20, 88($sp)
+	sdc1	$f21, 96($sp)
+	sdc1	$f22,104($sp)
+	sdc1	$f23,112($sp)
+#endif
+
+	dsra	J,  N, 1							#	J=N/2
+	ST	ALPHA_R, 128($sp)						#	store alpha_r & alpha_i
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	neg	KK, OFFSET
+#endif
+
+	dsll	LDC, LDC, ZBASE_SHIFT				#	LDC*SIZE*COMPSIZE
+	blez	J, .L20
+	ST	ALPHA_I, 136($sp)
+
+
+	.align	5
+.L10:
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	move	KK, OFFSET
+#endif
+
+	daddiu	J, J, -1							
+	dsra	I,  M, 1							#	I=M/2
+
+	dsll	PREB, K, 1 + ZBASE_SHIFT				#	PREA=K*2*2^4
+	dsll	PREA, K, 1 + ZBASE_SHIFT				#	PREA=K*2*2^4
+
+	move	CO1, C								#	Fix pointer Cx
+	daddu	CO2, C,   LDC
+
+	move	AO, A								#	Reset AO
+	blez	I, .L30
+	daddu	PREA, PREA, A						#	PREA=A+panel size
+
+.L11:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	BO,  B
+#else
+	dsll	TEMP, KK, 1 + ZBASE_SHIFT
+
+	daddu	AO, AO, TEMP
+	daddu	BO, B,  TEMP
+#endif
+
+	MTC	$0,  c11								#	Clear results regs
+	LD	a1,	0 * SIZE(AO)
+	MOV	c12, c11
+	LD	a2,	1 * SIZE(AO)
+
+	MOV	c13, c11
+	LD	b1, 0 * SIZE(BO)
+	MOV	c14, c11
+	LD	b2, 1 * SIZE(BO)
+
+	MOV	c21, c11
+	LD	a3,	2 * SIZE(AO)
+	MOV	c22, c11
+	LD	a4, 3 * SIZE(AO)
+	
+	MOV	c23, c11
+	LD	b3, 2 * SIZE(BO)
+	MOV	c24, c11
+	LD	b4,	3 * SIZE(BO)
+
+	FETCH	$0, 0 * SIZE(CO2)
+	MOV	c31, c11
+	MOV	c32, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	MOV	c33, c11
+	MOV	c34, c11
+	
+	FETCH	$0, 4 * SIZE(CO2)
+	MOV	c41, c11
+	MOV	c42, c11
+
+	FETCH	$0, 4 * SIZE(CO1)
+	MOV	c43, c11
+	MOV	c44, c11
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, K, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 2
+#else
+	daddiu	TEMP, KK, 2
+#endif
+	dsra	L,  TEMP, 2
+	daddu	PREB, PREB, B						#	PREA=A+panel size
+	blez	L, .L15
+	NOP
+
+#else
+
+	dsra	L,  K, 2							#	Unroll K 4 times	
+	move	BO,  B
+
+	MTC	$0,  c11								#	Clear results regs
+	LD	a1,	0 * SIZE(AO)
+	MOV	c12, c11
+	LD	a2,	1 * SIZE(AO)
+
+	MOV	c13, c11
+	LD	b1,	0 * SIZE(BO)
+	MOV	c14, c11
+	LD	b2,	1 * SIZE(BO)
+
+	MOV	c21, c11
+	LD	a3,	2 * SIZE(AO)
+	MOV	c22, c11
+	LD	a4, 3 * SIZE(AO)
+	
+	MOV	c23, c11
+	LD	b3, 2 * SIZE(BO)
+	MOV	c24, c11
+	LD	b4,	3 * SIZE(BO)
+
+	MOV	c31, c11
+	MOV	c32, c11
+	FETCH	$0, 0 * SIZE(CO2)
+
+	MOV	c33, c11
+	MOV	c34, c11
+	FETCH	$0, 0 * SIZE(CO1)
+	
+	MOV	c41, c11
+	MOV	c42, c11
+	FETCH	$0, 4 * SIZE(CO2)
+
+	MOV	c43, c11
+	NOP
+	FETCH	$0, 4 * SIZE(CO1)
+
+	daddu	PREB, PREB, B						#	PREA=A+panel size
+	blez	L, .L15
+	MOV	c44, c11
+#endif
+
+	.align	5
+
+.L12:
+	LD	a5,	4 * SIZE(AO)
+	LD	a6,	5 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	LD	b5,	4 * SIZE(BO)
+	LD	b6,	5 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	LD	a7,	6 * SIZE(AO)
+	LD	a8,	7 * SIZE(AO)
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+
+	LD	b7,	6 * SIZE(BO)
+	LD	b8, 7 * SIZE(BO)
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+	FETCH	$0, 4 * SIZE(PREA)
+	FETCH	$0, 4 * SIZE(PREB)
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+
+	MADD1	c41, c41, a3, b3 					#	A2xB2
+	MADD3	c43, c43, a3, b4
+	MADD2	c42, c42, a4, b3
+	MADD4	c44, c44, a4, b4
+
+	LD	a1,	8 * SIZE(AO)
+	LD	a2, 9 * SIZE(AO)
+	MADD1	c11, c11, a5, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b6					#	axd
+
+	LD	b1,	8 * SIZE(BO)
+	LD	b2, 9 * SIZE(BO)
+	MADD2	c12, c12, a6, b5					#	bxc
+	MADD4	c14, c14, a6, b6					#	bxd
+
+	LD	a3,	10 * SIZE(AO)
+	LD	a4,	11 * SIZE(AO)	
+	MADD1	c21, c21, a7, b5 					#	A2xB1
+	MADD3	c23, c23, a7, b6
+
+	LD	b3,	10 * SIZE(BO)
+	LD	b4,	11 * SIZE(BO)
+	MADD2	c22, c22, a8, b5
+	MADD4	c24, c24, a8, b6
+
+	FETCH	$0, 8 * SIZE(PREA)
+	FETCH	$0, 8 * SIZE(PREB)
+	MADD1	c31, c31, a5, b7 					#	A1xB2
+	MADD3	c33, c33, a5, b8
+
+	MADD2	c32, c32, a6, b7
+	MADD4	c34, c34, a6, b8
+
+	MADD1	c41, c41, a7, b7 					#	A2xB2
+	MADD3	c43, c43, a7, b8
+	MADD2	c42, c42, a8, b7
+	MADD4	c44, c44, a8, b8
+
+	LD	a5,	12 * SIZE(AO)
+	LD	a6,	13 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	LD	b5,	12 * SIZE(BO)
+	LD	b6,	13 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	LD	a7,	14 * SIZE(AO)
+	LD	a8,	15 * SIZE(AO)
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+
+	LD	b7, 14 * SIZE(BO)
+	LD	b8,	15 * SIZE(BO)
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+	FETCH	$0, 12 * SIZE(PREA)
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+	daddiu	L, L, -1
+
+	FETCH	$0, 12 * SIZE(PREB)
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+	daddiu	AO, AO, 16 * SIZE
+
+	daddiu	BO, BO, 16 * SIZE  					#	2nr*4kr*cmpx
+	MADD1	c41, c41, a3, b3 					#	A2xB2
+	MADD3	c43, c43, a3, b4
+	daddu	PREA, PREA, 16 * SIZE
+
+	MADD2	c42, c42, a4, b3
+	MADD4	c44, c44, a4, b4
+	daddu	PREB, PREB, 16 * SIZE
+
+	LD	a1,	0 * SIZE(AO)
+	LD	a2,	1 * SIZE(AO)
+	MADD1	c11, c11, a5, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b6					#	axd
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2,	1 * SIZE(BO)
+	MADD2	c12, c12, a6, b5					#	bxc
+	MADD4	c14, c14, a6, b6					#	bxd
+
+	LD	a3,	2 * SIZE(AO)
+	LD	a4,	3 * SIZE(AO)
+	MADD1	c21, c21, a7, b5 					#	A2xB1
+	MADD3	c23, c23, a7, b6
+
+	LD	b3, 2 * SIZE(BO)
+	LD	b4,	3 * SIZE(BO)
+	MADD2	c22, c22, a8, b5
+	MADD4	c24, c24, a8, b6
+
+	FETCH	$0, 0 * SIZE(PREA)
+	FETCH	$0, 0 * SIZE(PREB)
+	MADD1	c31, c31, a5, b7 					#	A1xB2
+	MADD3	c33, c33, a5, b8
+
+	MADD2	c32, c32, a6, b7
+	MADD4	c34, c34, a6, b8
+
+	MADD1	c41, c41, a7, b7 					#	A2xB2
+	MADD3	c43, c43, a7, b8
+
+	MADD2	c42, c42, a8, b7
+	bgtz	L, .L12
+	MADD4	c44, c44, a8, b8
+
+	.align	5
+
+.L15:
+#ifndef	TRMMKERNEL
+	andi	L,  K, 3
+	LD	ALPHA_R, 128($sp)
+#else
+	andi	L,  TEMP, 3
+	LD	ALPHA_R, 128($sp)
+#endif
+	blez	L, .L18
+	LD	ALPHA_I, 136($sp)
+
+	.align	5
+
+.L16:
+	daddiu	BO, BO, 4 * SIZE					#	2nr*1kr*cmpx
+	daddiu	AO, AO, 4 * SIZE					#	2mr*1kr*cmpx
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	daddiu	PREA, PREA, 4 * SIZE			
+	daddiu	PREB, PREB, 4 * SIZE			
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+	
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+	FETCH	$0, 0 * SIZE(PREA)
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+	daddiu	L, L, -1
+
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+
+	FETCH	$0, 0 * SIZE(PREB)
+	MADD1	c41, c41, a3, b3 					#	A2xB2
+	MADD3	c43, c43, a3, b4
+
+	MADD2	c42, c42, a4, b3
+	MADD4	c44, c44, a4, b4
+
+	LD	a1,	0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	LD	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	bgtz	L, .L16
+	NOP
+
+.L18:
+#ifndef TRMMKERNEL
+ 	ADD	c11, c14, c11
+	LD	a1,  0 * SIZE(CO1)
+	ADD	c12, c13, c12
+	LD	a2,  1 * SIZE(CO1)
+	ADD	c21, c24, c21
+	LD	b1,  2 * SIZE(CO1)
+	ADD	c22, c23, c22
+	LD	b2,  3 * SIZE(CO1)
+
+	ADD	c31, c34, c31
+	LD	a3,  0 * SIZE(CO2)
+	ADD	c32, c33, c32
+	LD	a4,  1 * SIZE(CO2)
+	ADD	c41, c44, c41
+	LD	b3,  2 * SIZE(CO2)
+	ADD	c42, c43, c42
+	LD	b4,  3 * SIZE(CO2)
+
+	daddiu	I, I, -1
+	MADD	a1, a1, ALPHA_R, c11
+	MADD	a2, a2, ALPHA_R, c12
+	MADD	b1, b1, ALPHA_R, c21
+	MADD	b2, b2, ALPHA_R, c22
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+	NMSUB	b1, b1, ALPHA_I, c22
+	MADD	b2, b2, ALPHA_I, c21
+
+	MADD	a3, a3, ALPHA_R, c31
+	MADD	a4, a4, ALPHA_R, c32
+	ST	a1, 0 * SIZE(CO1)
+	MADD	b3, b3, ALPHA_R, c41
+	MADD	b4, b4, ALPHA_R, c42
+	ST	a2, 1 * SIZE(CO1)
+
+	NMSUB	a3, a3, ALPHA_I, c32
+	MADD	a4, a4, ALPHA_I, c31
+	ST	b1, 2 * SIZE(CO1)
+
+	NMSUB	b3, b3, ALPHA_I, c42
+	MADD	b4, b4, ALPHA_I, c41
+	ST	b2, 3 * SIZE(CO1)
+
+	ST	a3, 0 * SIZE(CO2)
+	ST	a4, 1 * SIZE(CO2)
+	ST	b3, 2 * SIZE(CO2)
+	ST	b4, 3 * SIZE(CO2)
+
+#else
+ 	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+	ADD	c21, c24, c21
+	ADD	c22, c23, c22
+
+	ADD	c31, c34, c31
+	ADD	c32, c33, c32
+	ADD	c41, c44, c41
+	ADD	c42, c43, c42
+
+	daddiu	I, I, -1
+	MUL	a1, ALPHA_R, c11
+	MUL	a2, ALPHA_R, c12
+	MUL	b1, ALPHA_R, c21
+	MUL	b2, ALPHA_R, c22
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+	NMSUB	b1, b1, ALPHA_I, c22
+	MADD	b2, b2, ALPHA_I, c21
+
+	MUL	a3, ALPHA_R, c31
+	MUL	a4, ALPHA_R, c32
+	MUL	b3, ALPHA_R, c41
+	MUL	b4, ALPHA_R, c42
+
+	NMSUB	a3, a3, ALPHA_I, c32
+	MADD	a4, a4, ALPHA_I, c31
+	NMSUB	b3, b3, ALPHA_I, c42
+	MADD	b4, b4, ALPHA_I, c41
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+	ST	b1, 2 * SIZE(CO1)
+	ST	b2, 3 * SIZE(CO1)
+
+	ST	a3, 0 * SIZE(CO2)
+	ST	a4, 1 * SIZE(CO2)
+	ST	b3, 2 * SIZE(CO2)
+	ST	b4, 3 * SIZE(CO2)
+
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+    (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, K, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -2
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+
+	dsll	TEMP, TEMP, 1 + ZBASE_SHIFT
+
+	daddu	AO, AO, TEMP
+	daddu	BO, BO, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+
+	dsll	PREB, K, 1 + ZBASE_SHIFT				#	PREA=K*2*2^4
+	daddiu	CO1,CO1, 4 * SIZE 
+	bgtz	I, .L11
+	daddiu	CO2,CO2, 4 * SIZE 
+
+	.align	5
+.L30:
+	andi	I, M, 1
+	daddu	C,   C, LDC						#	Change C to next panel
+
+	daddu	PREB, PREB, B						#	PREA=A+panel size
+	blez	I, .L19
+	daddu	C,   C, LDC						#	Change C to next panel
+
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	BO,  B
+#else
+	dsll	L,    KK, ZBASE_SHIFT			#	MR=1
+	dsll	TEMP, KK, 1 + ZBASE_SHIFT		#	NR=2
+
+	daddu	AO, AO, L
+	daddu	BO, B,  TEMP
+#endif
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2,	1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+	
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	MOV	c31, c11
+	MOV	c32, c11
+
+	FETCH	$0, 0 * SIZE(PREB)
+	MOV	c33, c11
+	MOV	c34, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	FETCH	$0, 0 * SIZE(CO2)
+	FETCH	$0, 4 * SIZE(CO1)
+	FETCH	$0, 4 * SIZE(CO2)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, K, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1							#	MR=1
+#else	
+	daddiu	TEMP, KK, 2							#	NR=2
+#endif
+	dsra	L,  TEMP, 2
+	blez	L, .L35
+	NOP
+
+#else
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2,	1 * SIZE(AO)
+	dsra	L,  K, 2							#	Unroll K 4 times	
+	move	BO,  B
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+	
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	FETCH	$0, 0 * SIZE(PREB)
+	MOV	c31, c11
+	MOV	c32, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	FETCH	$0, 0 * SIZE(CO2)
+	FETCH	$0, 4 * SIZE(CO1)
+	FETCH	$0, 4 * SIZE(CO2)
+
+	MOV	c33, c11
+	blez	L, .L35
+	MOV	c34, c11
+#endif
+
+	.align	5
+
+.L32:
+	LD	a3,	2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	LD	b5, 4 * SIZE(BO)
+	LD	b6, 5 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+	
+	LD	b7, 6 * SIZE(BO)
+	LD	b8, 7 * SIZE(BO)
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+	
+	FETCH	$0, 4 * SIZE(PREB)
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+	NOP
+
+	LD	a5,	4 * SIZE(AO)
+	LD	a6, 5 * SIZE(AO)
+	MADD1	c11, c11, a3, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a3, b6					#	axd
+
+	LD	b1, 8 * SIZE(BO)
+	LD	b2, 9 * SIZE(BO)
+	MADD2	c12, c12, a4, b5					#	bxc
+	MADD4	c14, c14, a4, b6					#	bxd
+	
+	LD	b3, 10 * SIZE(BO)
+	LD	b4, 11 * SIZE(BO)
+	MADD1	c31, c31, a3, b7 					#	A1xB2
+	MADD3	c33, c33, a3, b8
+
+	FETCH	$0, 8 * SIZE(PREB)
+	MADD2	c32, c32, a4, b7
+	MADD4	c34, c34, a4, b8
+	daddiu	L, L, -1
+
+	LD	a7, 6 * SIZE(AO)
+	LD	a8, 7 * SIZE(AO)
+	MADD1	c11, c11, a5, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b2					#	axd
+
+	LD	b5, 12 * SIZE(BO)
+	LD	b6, 13 * SIZE(BO)
+	MADD2	c12, c12, a6, b1					#	bxc
+	MADD4	c14, c14, a6, b2					#	bxd
+
+	LD	b7, 14 * SIZE(BO)
+	LD	b8, 15 * SIZE(BO)
+	MADD1	c31, c31, a5, b3 					#	A1xB2
+	MADD3	c33, c33, a5, b4
+
+	daddiu	AO, AO, 8 * SIZE 					#	2mr*4kr*cmpx
+	daddiu	BO, BO, 16 * SIZE  					#	2nr*4kr*cmpx
+
+	FETCH	$0, 12 * SIZE(PREB)
+	MADD2	c32, c32, a6, b3
+	MADD4	c34, c34, a6, b4
+	NOP
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MADD1	c11, c11, a7, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a7, b6					#	axd
+
+	LD	b1,	0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MADD2	c12, c12, a8, b5					#	bxc
+	MADD4	c14, c14, a8, b6					#	bxd
+
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	MADD1	c31, c31, a7, b7 					#	A1xB2
+	NOP
+
+	MADD3	c33, c33, a7, b8
+	daddiu	PREB, PREB, 16 * SIZE
+
+	FETCH	$0, 0 * SIZE(PREB)
+	MADD2	c32, c32, a8, b7
+	bgtz	L, .L32
+	MADD4	c34, c34, a8, b8
+
+
+.L35:
+#ifndef TRMMKERNEL
+	andi	L,  K, 3
+	LD	ALPHA_R, 128($sp)
+#else
+	andi	L,  TEMP, 3
+	LD	ALPHA_R, 128($sp)
+#endif
+	blez	L, .L38
+	LD	ALPHA_I, 136($sp)
+	.align	5
+
+.L36:
+	daddiu	L, L, -1
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	daddiu	BO, BO, 4 * SIZE 					#	2nr*1kr*cmpx
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	daddiu	AO, AO, 2 * SIZE					#	2mr*1kr*cmpx
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+	
+	daddiu	PREB, PREB, 4 * SIZE 
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+
+	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)
+	bgtz	L, .L36
+	NOP
+
+.L38:
+#ifndef	TRMMKERNEL
+ 	ADD	c11, c14, c11
+	LD	a1,  0 * SIZE(CO1)
+	ADD	c12, c13, c12
+	LD	a2,  1 * SIZE(CO1)
+
+	ADD	c31, c34, c31
+	LD	a3,  0 * SIZE(CO2)
+	ADD	c32, c33, c32
+	LD	a4,  1 * SIZE(CO2)
+
+	MADD	a1, a1, ALPHA_R, c11
+	MADD	a2, a2, ALPHA_R, c12
+
+	MADD	a3, a3, ALPHA_R, c31
+	MADD	a4, a4, ALPHA_R, c32
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+
+	NMSUB	a3, a3, ALPHA_I, c32
+	MADD	a4, a4, ALPHA_I, c31
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+
+	ST	a3, 0 * SIZE(CO2)
+	ST	a4, 1 * SIZE(CO2)
+
+	daddiu	CO1,CO1, 2 * SIZE 
+	daddiu	CO2,CO2, 2 * SIZE 
+
+#else
+	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+
+	ADD	c31, c34, c31
+	ADD	c32, c33, c32
+
+	MUL	a1, ALPHA_R, c11
+	MUL	a2, ALPHA_R, c12
+
+	MUL	a3, ALPHA_R, c31
+	MUL	a4, ALPHA_R, c32
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+
+	NMSUB	a3, a3, ALPHA_I, c32
+	MADD	a4, a4, ALPHA_I, c31
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+
+	ST	a3, 0 * SIZE(CO2)
+	ST	a4, 1 * SIZE(CO2)
+
+	daddiu	CO1,CO1, 2 * SIZE 
+	daddiu	CO2,CO2, 2 * SIZE 
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+    (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, K, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -1
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+	dsll	L,    TEMP, ZBASE_SHIFT
+	dsll	TEMP, TEMP, 1 + ZBASE_SHIFT
+
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 1
+#endif
+#endif
+
+	.align 5
+
+.L19:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	daddiu	KK, KK, 2
+#endif
+
+	bgtz	J, .L10
+	move	B, BO
+
+	.align 5
+	
+.L20:
+	andi	J,  N, 1
+	blez	J, .L999
+	dsll	PREA, K, 1+ZBASE_SHIFT				#	PREA=K*2*2^4
+
+	dsra	I,  M, 1							#	I=M/2
+	move	CO1, C
+
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	move	KK, OFFSET
+#endif
+
+	move	AO, A								#	Reset AO
+	blez	I, .L29
+	daddu	PREA, PREA, A
+
+.L21:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	BO,  B
+#else
+	dsll	L,    KK, 1 + ZBASE_SHIFT
+	dsll	TEMP, KK, ZBASE_SHIFT
+
+	daddu	AO, AO, L
+	daddu	BO, B,  TEMP
+#endif
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	LD	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	MOV	c21, c11
+	MOV	c22, c11
+
+	FETCH	$0, 0 * SIZE(PREA)
+	MOV	c23, c11
+	MOV	c24, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	FETCH	$0, 4 * SIZE(CO1)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, K, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 2						#	define Mr=2
+#else
+	daddiu	TEMP, KK, 1						#	define	NR=1
+#endif
+	dsra	L,  TEMP, 2
+	blez	L, .L25
+	NOP
+
+#else
+	dsra	L,  K, 2							#	Unroll K 4 times	
+	move	BO,  B
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	LD	a3, 2 * SIZE(AO)
+	LD 	a4, 3 * SIZE(AO)
+	MOV	c21, c11
+	MOV	c22, c11
+
+	FETCH	$0, 0 * SIZE(PREA)
+	MOV	c23, c11
+	MOV	c24, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	FETCH	$0, 4 * SIZE(CO1)
+
+	blez	L, .L25
+	NOP
+#endif
+
+	.align	5
+
+.L22:
+	LD	 a5, 4 * SIZE(AO)
+	LD	 a6, 5 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	LD	 b3, 2 * SIZE(BO)
+	LD	 b4, 3 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	LD 	 a7, 6 * SIZE(AO)
+	LD 	 a8, 7 * SIZE(AO)
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+	
+	FETCH	$0, 4 * SIZE(PREA)
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+	LD 	a1, 8 * SIZE(AO)
+	LD	a2, 9 * SIZE(AO)
+	MADD1	c11, c11, a5, b3 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b4					#	axd
+
+	LD	b5, 4 * SIZE(BO)
+	LD	b6, 5 * SIZE(BO)
+	MADD2	c12, c12, a6, b3					#	bxc
+	MADD4	c14, c14, a6, b4					#	bxd
+
+	LD	a3, 10 * SIZE(AO)
+	LD	a4, 11 * SIZE(AO)
+	MADD1	c21, c21, a7, b3 					#	A2xB1
+	MADD3	c23, c23, a7, b4
+
+	FETCH	$0, 8 * SIZE(PREA)
+	MADD2	c22, c22, a8, b3
+	MADD4	c24, c24, a8, b4
+	daddiu	L, L, -1
+
+	LD	a5, 12 * SIZE(AO)
+	LD	a6, 13 * SIZE(AO)
+	MADD1	c11, c11, a1, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b6					#	axd
+
+	LD	b7, 6 * SIZE(BO)
+	LD	b8, 7 * SIZE(BO)
+	MADD2	c12, c12, a2, b5					#	bxc
+	MADD4	c14, c14, a2, b6					#	bxd
+
+	LD	a7, 14 * SIZE(AO)
+	LD	a8, 15 * SIZE(AO)
+	MADD1	c21, c21, a3, b5 					#	A2xB1
+	MADD3	c23, c23, a3, b6
+
+	daddiu	BO, BO,  8 * SIZE  					#	1nr*4kr*cmpx
+	daddiu	AO, AO, 16 * SIZE 					#	2mr*4kr*cmpx
+
+	FETCH	$0, 12 * SIZE(PREA)
+	MADD2	c22, c22, a4, b5
+	MADD4	c24, c24, a4, b6
+	daddiu	PREA, PREA, 16 * SIZE 		
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MADD1	c11, c11, a5, b7 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b8					#	axd
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MADD2	c12, c12, a6, b7					#	bxc
+	MADD4	c14, c14, a6, b8					#	bxd
+
+	LD	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	MADD1	c21, c21, a7, b7 					#	A2xB1
+	MADD3	c23, c23, a7, b8
+
+	FETCH	$0, 0 * SIZE(PREA)
+	MADD2	c22, c22, a8, b7
+	bgtz	L, .L22
+	MADD4	c24, c24, a8, b8
+
+
+.L25:
+#ifndef TRMMKERNEL
+	andi	L,  K, 3
+	LD	ALPHA_R, 128($sp)
+#else
+	andi	L, TEMP, 3
+	LD	ALPHA_R, 128($sp)
+#endif
+	blez	L, .L28
+	LD	ALPHA_I, 136($sp)
+	.align	3
+
+.L26:
+	daddiu	L, L, -1
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	daddiu	BO, BO, 2 * SIZE 					#	2nr*1kr*cmpx
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	daddiu	AO, AO, 4 * SIZE 					#	2mr*1kr*cmpx
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+
+	daddiu	PREA, PREA, 4 * SIZE 				#	2mr*1kr*cmpx
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2	Unroll K=4
+#	gsLQC1(R12, F3, F2, 1)						#	R:a3	I:a4
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	LD	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+
+	bgtz	L, .L26
+	FETCH	$0, 0 * SIZE(PREA)
+
+.L28:
+#ifndef	TRMMKERNEL
+ 	ADD	c11, c14, c11
+	LD	a1,  0 * SIZE(CO1)
+	ADD	c12, c13, c12
+	LD	a2,  1 * SIZE(CO1)
+	ADD	c21, c24, c21
+	LD	b1,  2 * SIZE(CO1)
+	ADD	c22, c23, c22
+	LD	b2,  3 * SIZE(CO1)
+
+	daddiu	I, I, -1
+	MADD	a1, a1, ALPHA_R, c11
+	MADD	a2, a2, ALPHA_R, c12
+	MADD	b1, b1, ALPHA_R, c21
+	MADD	b2, b2, ALPHA_R, c22
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+	NMSUB	b1, b1, ALPHA_I, c22
+	MADD	b2, b2, ALPHA_I, c21
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+	ST	b1, 2 * SIZE(CO1)
+	ST	b2, 3 * SIZE(CO1)
+
+#else
+ 	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+	ADD	c21, c24, c21
+	ADD	c22, c23, c22
+
+	daddiu	I, I, -1
+	MUL	a1, ALPHA_R, c11
+	MUL	a2, ALPHA_R, c12
+	MUL	b1, ALPHA_R, c21
+	MUL	b2, ALPHA_R, c22
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+	NMSUB	b1, b1, ALPHA_I, c22
+	MADD	b2, b2, ALPHA_I, c21
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+	ST	b1, 2 * SIZE(CO1)
+	ST	b2, 3 * SIZE(CO1)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+    (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, K, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -2
+#else
+	daddiu	TEMP, TEMP, -1
+#endif
+
+	dsll	L,    TEMP, 1 + ZBASE_SHIFT
+	dsll	TEMP, TEMP, ZBASE_SHIFT
+
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+	daddiu	CO1,CO1, 4 * SIZE 
+	bgtz	I, .L21
+	NOP
+
+.L29:
+	andi	I, M, 1
+	blez	I, .L999
+	NOP
+
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	BO,  B
+#else
+	dsll	TEMP, KK,  ZBASE_SHIFT
+
+	daddu	AO, AO, TEMP
+	daddu	BO, B,  TEMP
+#endif
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	b1, 0 * SIZE(BO)
+	LD 	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	FETCH	$0, 0 * SIZE(PREA)
+	FETCH	$0, 4 * SIZE(PREA)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, K, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1
+#else
+	daddiu	TEMP, KK, 1
+#endif
+	dsra	L,  TEMP, 2
+	blez	L, .L45
+	NOP
+
+#else
+	dsra	L,  K, 2							#	Unroll K 4 times	
+	move	BO,  B
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	b1, 0 * SIZE(BO)
+	LD 	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	FETCH	$0, 0 * SIZE(PREA)
+	FETCH	$0, 4 * SIZE(PREA)
+	blez	L, .L45
+	NOP
+#endif
+
+	.align	3
+
+.L42:
+#	gsLQC1(R12, F3, F2, 1)						#	R:a3	I:a4
+	LD 	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+#	gsLQC1(R13, F7, F6, 1)						#	R:b2	I:b3
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+#	gsLQC1(R12, F9, F8, 2)						#	Unroll K=1	
+	LD	a5, 4 * SIZE(AO)
+	LD	a6, 5 * SIZE(AO)
+	MADD1	c11, c11, a3, b3 					#	axc		A1xB1	
+	MADD3	c13, c13, a3, b4					#	axd
+
+#	gsLQC1(R13, F13, F12, 2)				
+	LD	b5, 4 * SIZE(BO)
+	LD 	b6, 5 * SIZE(BO)
+	MADD2	c12, c12, a4, b3					#	bxc
+	MADD4	c14, c14, a4, b4					#	bxd
+
+#	gsLQC1(R12, F11, F10, 3)					
+	LD 	a7, 6 * SIZE(AO)
+	LD	a8, 7 * SIZE(AO)
+	MADD1	c11, c11, a5, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b6					#	axd
+
+	daddiu	L, L, -1
+
+#	gsLQC1(R13, F16, F15, 3)			
+	LD	b7, 6 * SIZE(BO)
+	LD	b8, 7 * SIZE(BO)
+	MADD2	c12, c12, a6, b5					#	bxc
+	MADD4	c14, c14, a6, b6					#	bxd
+
+	daddiu	AO, AO, 8 * SIZE 					#	2mr*4kr*cmpx
+	daddiu	BO, BO, 8 * SIZE  					#	2nr*4kr*cmpx
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2	Unroll K=4
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MADD1	c11, c11, a7, b7 					#	axc		A1xB1	
+	MADD3	c13, c13, a7, b8					#	axd
+
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MADD2	c12, c12, a8, b7					#	bxc
+	MADD4	c14, c14, a8, b8					#	bxd
+
+	bgtz	L, .L42
+	NOP
+
+
+	.align 5
+
+.L45:
+#ifndef	TRMMKERNEL
+	andi	L,  K, 3
+	LD	ALPHA_R, 128($sp)
+#else
+	andi	L,  TEMP, 3
+	LD	ALPHA_R, 128($sp)
+#endif
+	blez	L, .L48
+	LD	ALPHA_I, 136($sp)
+
+.L46:
+	daddiu	L, L, -1
+	daddiu	BO, BO, 1 * SIZE * COMPSIZE			#	2nr*1kr*cmpx
+	daddiu	AO, AO, 1 * SIZE * COMPSIZE			#	2mr*1kr*cmpx
+
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2	Unroll K=4
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	bgtz	L, .L46
+	NOP
+
+.L48:
+#ifndef	TRMMKERNEL
+ 	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+
+	LD	a1,  0 * SIZE(CO1)
+	LD	a2,  1 * SIZE(CO1)
+
+	MADD	a1, a1, ALPHA_R, c11
+	MADD	a2, a2, ALPHA_R, c12
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+
+#else
+	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+
+	MUL	a1, ALPHA_R, c11
+	MUL	a2, ALPHA_R, c12
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+    (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, K, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -1
+#else
+	daddiu	TEMP, TEMP, -1
+#endif
+
+	dsll	TEMP, TEMP, ZBASE_SHIFT
+
+	daddu	AO, AO, TEMP
+	daddu	BO, BO, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 1
+#endif
+
+	daddiu	CO1,CO1, 2 * SIZE 
+#endif
+
+
+
+	.align 5
+
+.L999:
+	LDARG	$16,   0($sp)
+	LDARG	$17,   8($sp)
+	ldc1	$f24, 16($sp)
+	ldc1	$f25, 24($sp)
+	ldc1	$f26, 32($sp)
+	ldc1	$f27, 40($sp)
+	ldc1	$f28, 48($sp)
+	ldc1	$f29, 56($sp)
+
+#if defined(TRMMKERNEL)
+	LDARG	$18,  64($sp)
+	LDARG	$19,  72($sp)
+	LDARG	$20,  80($sp)
+#endif
+
+#ifndef __64BIT__
+	ldc1	$f20, 88($sp)
+	ldc1	$f21, 96($sp)
+	ldc1	$f22,104($sp)
+	ldc1	$f23,112($sp)
+#endif
+
+	j	$31
+	daddiu	$sp, $sp, STACKSIZE
+
+	EPILOGUE
diff --git a/kernel/mips64/dgemm_kernel_loongson3b_4x4.S b/kernel/mips64/dgemm_kernel_loongson3b_4x4.S
new file mode 100644
index 000000000..4a8c9b0e4
--- /dev/null
+++ b/kernel/mips64/dgemm_kernel_loongson3b_4x4.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/sgemm_kernel_loongson3b_4x4.S b/kernel/mips64/sgemm_kernel_loongson3b_4x4.S
new file mode 100644
index 000000000..4a8c9b0e4
--- /dev/null
+++ b/kernel/mips64/sgemm_kernel_loongson3b_4x4.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/zgemm_kernel_loongson3b_2x2.S b/kernel/mips64/zgemm_kernel_loongson3b_2x2.S
new file mode 100644
index 000000000..5ded7aed0
--- /dev/null
+++ b/kernel/mips64/zgemm_kernel_loongson3b_2x2.S
@@ -0,0 +1,1468 @@
+#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 STACKSIZE	160
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$9
+#define B	$10
+#define C	$11
+#define LDC	$8
+
+#define AO	$12
+#define BO	$13
+
+#define	R12	12
+#define	R13	13
+
+#define I	$2
+#define J	$3
+#define L	$7
+
+#define CO1	$14
+#define CO2	$15
+#define PREA	$16
+#define PREB	$17
+
+#if defined(TRMMKERNEL)
+#define OFFSET	$18
+#define KK	$19
+#define TEMP	$20
+#endif
+
+#define a1	$f0
+#define a2	$f1
+#define a3	$f2
+#define a4	$f3
+
+#define b1	$f4
+#define b2	$f5
+#define b3	$f6
+#define b4	$f7
+
+#define a5	$f8
+#define a6	$f9
+#define a7	$f10
+#define a8	$f11
+
+#define b5	$f12
+#define b6	$f13
+#define b7	$f15
+#define b8	$f16
+
+#define c11	$f14
+#define c12	$f17
+#define c13	$f18
+#define c14	$f19
+#define c21	$f20
+#define c22	$f21
+#define c23	$f22
+#define c24	$f23
+#define c31	$f24
+#define c32	$f25
+#define c33	$f26
+#define c34	$f27
+#define c41	$f28
+#define c42	$f29
+#define c43	$f30
+#define c44	$f31
+
+#define F0	0
+#define F1	1
+#define F2	2
+#define F3	3
+#define F4	4
+#define F5	5
+#define	F6	6
+#define	F7	7
+#define	F8	8
+#define	F9	9
+#define	F10	10
+#define	F11	11
+#define	F12	12
+#define	F13	13
+#define	F14	14
+#define	F15	15
+#define	F16	16
+#define	F17	17
+#define	F18	18
+#define	F19	19
+#define	F20	20
+#define	F21	21
+#define	F22	22
+#define	F23	23
+#define	F24	24
+#define	F25	25
+#define	F26	26
+#define	F27	27
+#define	F28	28
+#define	F29	29
+#define	F30	30
+#define	F31	31
+
+#define ALPHA_R	$f15
+#define ALPHA_I	$f16
+
+#################################
+##	MADD1	a*c
+##	MADD2	b*c
+##	MADD3	a*d
+##	MADD4	d*b
+##################################
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+#define MADD1	  MADD
+#define MADD2	  MADD
+#define MADD3	  MADD
+#define MADD4	  NMSUB
+#endif
+
+#if   defined(NR) || defined(NC) || defined(TR) || defined(TC)
+#define MADD1	  MADD
+#define MADD2	  MADD
+#define MADD3	  NMSUB
+#define MADD4	  MADD
+#endif
+
+#if   defined(RN) || defined(RT) || defined(CN) || defined(CT)
+#define MADD1	  MADD
+#define MADD2	  NMSUB
+#define MADD3	  MADD
+#define MADD4	  MADD
+#endif
+
+#if   defined(RR) || defined(RC) || defined(CR) || defined(CC)
+#define MADD1	  MADD
+#define MADD2	  NMSUB
+#define MADD3	  NMSUB
+#define MADD4	  NMSUB
+#endif
+
+	PROLOGUE
+	
+	LDARG	LDC,   0($sp)
+	daddiu	$sp, $sp, -STACKSIZE
+
+	SDARG	$16,   0($sp)
+	SDARG	$17,   8($sp)
+	sdc1	$f24, 16($sp)
+	sdc1	$f25, 24($sp)
+	sdc1	$f26, 32($sp)
+	sdc1	$f27, 40($sp)
+	sdc1	$f28, 48($sp)
+	sdc1	$f29, 56($sp)
+
+#if defined(TRMMKERNEL)
+	SDARG	$18,  64($sp)
+	SDARG	$19,  72($sp)
+	SDARG	$20,  80($sp)
+
+	LDARG	OFFSET, STACKSIZE + 8($sp)
+#endif
+
+#ifndef __64BIT__
+	sdc1	$f20, 88($sp)
+	sdc1	$f21, 96($sp)
+	sdc1	$f22,104($sp)
+	sdc1	$f23,112($sp)
+#endif
+
+	dsra	J,  N, 1							#	J=N/2
+	ST	ALPHA_R, 128($sp)						#	store alpha_r & alpha_i
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	neg	KK, OFFSET
+#endif
+
+	dsll	LDC, LDC, ZBASE_SHIFT				#	LDC*SIZE*COMPSIZE
+	blez	J, .L20
+	ST	ALPHA_I, 136($sp)
+
+
+	.align	5
+.L10:
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	move	KK, OFFSET
+#endif
+
+	daddiu	J, J, -1							
+	dsra	I,  M, 1							#	I=M/2
+
+	dsll	PREB, K, 1 + ZBASE_SHIFT				#	PREA=K*2*2^4
+	dsll	PREA, K, 1 + ZBASE_SHIFT				#	PREA=K*2*2^4
+
+	move	CO1, C								#	Fix pointer Cx
+	daddu	CO2, C,   LDC
+
+	move	AO, A								#	Reset AO
+	blez	I, .L30
+	daddu	PREA, PREA, A						#	PREA=A+panel size
+
+.L11:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	BO,  B
+#else
+	dsll	TEMP, KK, 1 + ZBASE_SHIFT
+
+	daddu	AO, AO, TEMP
+	daddu	BO, B,  TEMP
+#endif
+
+	MTC	$0,  c11								#	Clear results regs
+	LD	a1,	0 * SIZE(AO)
+	MOV	c12, c11
+	LD	a2,	1 * SIZE(AO)
+
+	MOV	c13, c11
+	LD	b1, 0 * SIZE(BO)
+	MOV	c14, c11
+	LD	b2, 1 * SIZE(BO)
+
+	MOV	c21, c11
+	LD	a3,	2 * SIZE(AO)
+	MOV	c22, c11
+	LD	a4, 3 * SIZE(AO)
+	
+	MOV	c23, c11
+	LD	b3, 2 * SIZE(BO)
+	MOV	c24, c11
+	LD	b4,	3 * SIZE(BO)
+
+	FETCH	$0, 0 * SIZE(CO2)
+	MOV	c31, c11
+	MOV	c32, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	MOV	c33, c11
+	MOV	c34, c11
+	
+	FETCH	$0, 4 * SIZE(CO2)
+	MOV	c41, c11
+	MOV	c42, c11
+
+	FETCH	$0, 4 * SIZE(CO1)
+	MOV	c43, c11
+	MOV	c44, c11
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, K, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 2
+#else
+	daddiu	TEMP, KK, 2
+#endif
+	dsra	L,  TEMP, 2
+	daddu	PREB, PREB, B						#	PREA=A+panel size
+	blez	L, .L15
+	NOP
+
+#else
+
+	dsra	L,  K, 2							#	Unroll K 4 times	
+	move	BO,  B
+
+	MTC	$0,  c11								#	Clear results regs
+	LD	a1,	0 * SIZE(AO)
+	MOV	c12, c11
+	LD	a2,	1 * SIZE(AO)
+
+	MOV	c13, c11
+	LD	b1,	0 * SIZE(BO)
+	MOV	c14, c11
+	LD	b2,	1 * SIZE(BO)
+
+	MOV	c21, c11
+	LD	a3,	2 * SIZE(AO)
+	MOV	c22, c11
+	LD	a4, 3 * SIZE(AO)
+	
+	MOV	c23, c11
+	LD	b3, 2 * SIZE(BO)
+	MOV	c24, c11
+	LD	b4,	3 * SIZE(BO)
+
+	MOV	c31, c11
+	MOV	c32, c11
+	FETCH	$0, 0 * SIZE(CO2)
+
+	MOV	c33, c11
+	MOV	c34, c11
+	FETCH	$0, 0 * SIZE(CO1)
+	
+	MOV	c41, c11
+	MOV	c42, c11
+	FETCH	$0, 4 * SIZE(CO2)
+
+	MOV	c43, c11
+	NOP
+	FETCH	$0, 4 * SIZE(CO1)
+
+	daddu	PREB, PREB, B						#	PREA=A+panel size
+	blez	L, .L15
+	MOV	c44, c11
+#endif
+
+	.align	5
+
+.L12:
+	LD	a5,	4 * SIZE(AO)
+	LD	a6,	5 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	LD	b5,	4 * SIZE(BO)
+	LD	b6,	5 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	LD	a7,	6 * SIZE(AO)
+	LD	a8,	7 * SIZE(AO)
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+
+	LD	b7,	6 * SIZE(BO)
+	LD	b8, 7 * SIZE(BO)
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+	FETCH	$0, 4 * SIZE(PREA)
+	FETCH	$0, 4 * SIZE(PREB)
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+
+	MADD1	c41, c41, a3, b3 					#	A2xB2
+	MADD3	c43, c43, a3, b4
+	MADD2	c42, c42, a4, b3
+	MADD4	c44, c44, a4, b4
+
+	LD	a1,	8 * SIZE(AO)
+	LD	a2, 9 * SIZE(AO)
+	MADD1	c11, c11, a5, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b6					#	axd
+
+	LD	b1,	8 * SIZE(BO)
+	LD	b2, 9 * SIZE(BO)
+	MADD2	c12, c12, a6, b5					#	bxc
+	MADD4	c14, c14, a6, b6					#	bxd
+
+	LD	a3,	10 * SIZE(AO)
+	LD	a4,	11 * SIZE(AO)	
+	MADD1	c21, c21, a7, b5 					#	A2xB1
+	MADD3	c23, c23, a7, b6
+
+	LD	b3,	10 * SIZE(BO)
+	LD	b4,	11 * SIZE(BO)
+	MADD2	c22, c22, a8, b5
+	MADD4	c24, c24, a8, b6
+
+	FETCH	$0, 8 * SIZE(PREA)
+	FETCH	$0, 8 * SIZE(PREB)
+	MADD1	c31, c31, a5, b7 					#	A1xB2
+	MADD3	c33, c33, a5, b8
+
+	MADD2	c32, c32, a6, b7
+	MADD4	c34, c34, a6, b8
+
+	MADD1	c41, c41, a7, b7 					#	A2xB2
+	MADD3	c43, c43, a7, b8
+	MADD2	c42, c42, a8, b7
+	MADD4	c44, c44, a8, b8
+
+	LD	a5,	12 * SIZE(AO)
+	LD	a6,	13 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	LD	b5,	12 * SIZE(BO)
+	LD	b6,	13 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	LD	a7,	14 * SIZE(AO)
+	LD	a8,	15 * SIZE(AO)
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+
+	LD	b7, 14 * SIZE(BO)
+	LD	b8,	15 * SIZE(BO)
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+	FETCH	$0, 12 * SIZE(PREA)
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+	daddiu	L, L, -1
+
+	FETCH	$0, 12 * SIZE(PREB)
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+	daddiu	AO, AO, 16 * SIZE
+
+	daddiu	BO, BO, 16 * SIZE  					#	2nr*4kr*cmpx
+	MADD1	c41, c41, a3, b3 					#	A2xB2
+	MADD3	c43, c43, a3, b4
+	daddu	PREA, PREA, 16 * SIZE
+
+	MADD2	c42, c42, a4, b3
+	MADD4	c44, c44, a4, b4
+	daddu	PREB, PREB, 16 * SIZE
+
+	LD	a1,	0 * SIZE(AO)
+	LD	a2,	1 * SIZE(AO)
+	MADD1	c11, c11, a5, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b6					#	axd
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2,	1 * SIZE(BO)
+	MADD2	c12, c12, a6, b5					#	bxc
+	MADD4	c14, c14, a6, b6					#	bxd
+
+	LD	a3,	2 * SIZE(AO)
+	LD	a4,	3 * SIZE(AO)
+	MADD1	c21, c21, a7, b5 					#	A2xB1
+	MADD3	c23, c23, a7, b6
+
+	LD	b3, 2 * SIZE(BO)
+	LD	b4,	3 * SIZE(BO)
+	MADD2	c22, c22, a8, b5
+	MADD4	c24, c24, a8, b6
+
+	FETCH	$0, 0 * SIZE(PREA)
+	FETCH	$0, 0 * SIZE(PREB)
+	MADD1	c31, c31, a5, b7 					#	A1xB2
+	MADD3	c33, c33, a5, b8
+
+	MADD2	c32, c32, a6, b7
+	MADD4	c34, c34, a6, b8
+
+	MADD1	c41, c41, a7, b7 					#	A2xB2
+	MADD3	c43, c43, a7, b8
+
+	MADD2	c42, c42, a8, b7
+	bgtz	L, .L12
+	MADD4	c44, c44, a8, b8
+
+	.align	5
+
+.L15:
+#ifndef	TRMMKERNEL
+	andi	L,  K, 3
+	LD	ALPHA_R, 128($sp)
+#else
+	andi	L,  TEMP, 3
+	LD	ALPHA_R, 128($sp)
+#endif
+	blez	L, .L18
+	LD	ALPHA_I, 136($sp)
+
+	.align	5
+
+.L16:
+	daddiu	BO, BO, 4 * SIZE					#	2nr*1kr*cmpx
+	daddiu	AO, AO, 4 * SIZE					#	2mr*1kr*cmpx
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	daddiu	PREA, PREA, 4 * SIZE			
+	daddiu	PREB, PREB, 4 * SIZE			
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+	
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+	FETCH	$0, 0 * SIZE(PREA)
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+	daddiu	L, L, -1
+
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+
+	FETCH	$0, 0 * SIZE(PREB)
+	MADD1	c41, c41, a3, b3 					#	A2xB2
+	MADD3	c43, c43, a3, b4
+
+	MADD2	c42, c42, a4, b3
+	MADD4	c44, c44, a4, b4
+
+	LD	a1,	0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	LD	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	bgtz	L, .L16
+	NOP
+
+.L18:
+#ifndef TRMMKERNEL
+ 	ADD	c11, c14, c11
+	LD	a1,  0 * SIZE(CO1)
+	ADD	c12, c13, c12
+	LD	a2,  1 * SIZE(CO1)
+	ADD	c21, c24, c21
+	LD	b1,  2 * SIZE(CO1)
+	ADD	c22, c23, c22
+	LD	b2,  3 * SIZE(CO1)
+
+	ADD	c31, c34, c31
+	LD	a3,  0 * SIZE(CO2)
+	ADD	c32, c33, c32
+	LD	a4,  1 * SIZE(CO2)
+	ADD	c41, c44, c41
+	LD	b3,  2 * SIZE(CO2)
+	ADD	c42, c43, c42
+	LD	b4,  3 * SIZE(CO2)
+
+	daddiu	I, I, -1
+	MADD	a1, a1, ALPHA_R, c11
+	MADD	a2, a2, ALPHA_R, c12
+	MADD	b1, b1, ALPHA_R, c21
+	MADD	b2, b2, ALPHA_R, c22
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+	NMSUB	b1, b1, ALPHA_I, c22
+	MADD	b2, b2, ALPHA_I, c21
+
+	MADD	a3, a3, ALPHA_R, c31
+	MADD	a4, a4, ALPHA_R, c32
+	ST	a1, 0 * SIZE(CO1)
+	MADD	b3, b3, ALPHA_R, c41
+	MADD	b4, b4, ALPHA_R, c42
+	ST	a2, 1 * SIZE(CO1)
+
+	NMSUB	a3, a3, ALPHA_I, c32
+	MADD	a4, a4, ALPHA_I, c31
+	ST	b1, 2 * SIZE(CO1)
+
+	NMSUB	b3, b3, ALPHA_I, c42
+	MADD	b4, b4, ALPHA_I, c41
+	ST	b2, 3 * SIZE(CO1)
+
+	ST	a3, 0 * SIZE(CO2)
+	ST	a4, 1 * SIZE(CO2)
+	ST	b3, 2 * SIZE(CO2)
+	ST	b4, 3 * SIZE(CO2)
+
+#else
+ 	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+	ADD	c21, c24, c21
+	ADD	c22, c23, c22
+
+	ADD	c31, c34, c31
+	ADD	c32, c33, c32
+	ADD	c41, c44, c41
+	ADD	c42, c43, c42
+
+	daddiu	I, I, -1
+	MUL	a1, ALPHA_R, c11
+	MUL	a2, ALPHA_R, c12
+	MUL	b1, ALPHA_R, c21
+	MUL	b2, ALPHA_R, c22
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+	NMSUB	b1, b1, ALPHA_I, c22
+	MADD	b2, b2, ALPHA_I, c21
+
+	MUL	a3, ALPHA_R, c31
+	MUL	a4, ALPHA_R, c32
+	MUL	b3, ALPHA_R, c41
+	MUL	b4, ALPHA_R, c42
+
+	NMSUB	a3, a3, ALPHA_I, c32
+	MADD	a4, a4, ALPHA_I, c31
+	NMSUB	b3, b3, ALPHA_I, c42
+	MADD	b4, b4, ALPHA_I, c41
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+	ST	b1, 2 * SIZE(CO1)
+	ST	b2, 3 * SIZE(CO1)
+
+	ST	a3, 0 * SIZE(CO2)
+	ST	a4, 1 * SIZE(CO2)
+	ST	b3, 2 * SIZE(CO2)
+	ST	b4, 3 * SIZE(CO2)
+
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+    (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, K, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -2
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+
+	dsll	TEMP, TEMP, 1 + ZBASE_SHIFT
+
+	daddu	AO, AO, TEMP
+	daddu	BO, BO, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+
+	dsll	PREB, K, 1 + ZBASE_SHIFT				#	PREA=K*2*2^4
+	daddiu	CO1,CO1, 4 * SIZE 
+	bgtz	I, .L11
+	daddiu	CO2,CO2, 4 * SIZE 
+
+	.align	5
+.L30:
+	andi	I, M, 1
+	daddu	C,   C, LDC						#	Change C to next panel
+
+	daddu	PREB, PREB, B						#	PREA=A+panel size
+	blez	I, .L19
+	daddu	C,   C, LDC						#	Change C to next panel
+
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	BO,  B
+#else
+	dsll	L,    KK, ZBASE_SHIFT			#	MR=1
+	dsll	TEMP, KK, 1 + ZBASE_SHIFT		#	NR=2
+
+	daddu	AO, AO, L
+	daddu	BO, B,  TEMP
+#endif
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2,	1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+	
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	MOV	c31, c11
+	MOV	c32, c11
+
+	FETCH	$0, 0 * SIZE(PREB)
+	MOV	c33, c11
+	MOV	c34, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	FETCH	$0, 0 * SIZE(CO2)
+	FETCH	$0, 4 * SIZE(CO1)
+	FETCH	$0, 4 * SIZE(CO2)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, K, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1							#	MR=1
+#else	
+	daddiu	TEMP, KK, 2							#	NR=2
+#endif
+	dsra	L,  TEMP, 2
+	blez	L, .L35
+	NOP
+
+#else
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2,	1 * SIZE(AO)
+	dsra	L,  K, 2							#	Unroll K 4 times	
+	move	BO,  B
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+	
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	FETCH	$0, 0 * SIZE(PREB)
+	MOV	c31, c11
+	MOV	c32, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	FETCH	$0, 0 * SIZE(CO2)
+	FETCH	$0, 4 * SIZE(CO1)
+	FETCH	$0, 4 * SIZE(CO2)
+
+	MOV	c33, c11
+	blez	L, .L35
+	MOV	c34, c11
+#endif
+
+	.align	5
+
+.L32:
+	LD	a3,	2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	LD	b5, 4 * SIZE(BO)
+	LD	b6, 5 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+	
+	LD	b7, 6 * SIZE(BO)
+	LD	b8, 7 * SIZE(BO)
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+	
+	FETCH	$0, 4 * SIZE(PREB)
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+	NOP
+
+	LD	a5,	4 * SIZE(AO)
+	LD	a6, 5 * SIZE(AO)
+	MADD1	c11, c11, a3, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a3, b6					#	axd
+
+	LD	b1, 8 * SIZE(BO)
+	LD	b2, 9 * SIZE(BO)
+	MADD2	c12, c12, a4, b5					#	bxc
+	MADD4	c14, c14, a4, b6					#	bxd
+	
+	LD	b3, 10 * SIZE(BO)
+	LD	b4, 11 * SIZE(BO)
+	MADD1	c31, c31, a3, b7 					#	A1xB2
+	MADD3	c33, c33, a3, b8
+
+	FETCH	$0, 8 * SIZE(PREB)
+	MADD2	c32, c32, a4, b7
+	MADD4	c34, c34, a4, b8
+	daddiu	L, L, -1
+
+	LD	a7, 6 * SIZE(AO)
+	LD	a8, 7 * SIZE(AO)
+	MADD1	c11, c11, a5, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b2					#	axd
+
+	LD	b5, 12 * SIZE(BO)
+	LD	b6, 13 * SIZE(BO)
+	MADD2	c12, c12, a6, b1					#	bxc
+	MADD4	c14, c14, a6, b2					#	bxd
+
+	LD	b7, 14 * SIZE(BO)
+	LD	b8, 15 * SIZE(BO)
+	MADD1	c31, c31, a5, b3 					#	A1xB2
+	MADD3	c33, c33, a5, b4
+
+	daddiu	AO, AO, 8 * SIZE 					#	2mr*4kr*cmpx
+	daddiu	BO, BO, 16 * SIZE  					#	2nr*4kr*cmpx
+
+	FETCH	$0, 12 * SIZE(PREB)
+	MADD2	c32, c32, a6, b3
+	MADD4	c34, c34, a6, b4
+	NOP
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MADD1	c11, c11, a7, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a7, b6					#	axd
+
+	LD	b1,	0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MADD2	c12, c12, a8, b5					#	bxc
+	MADD4	c14, c14, a8, b6					#	bxd
+
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	MADD1	c31, c31, a7, b7 					#	A1xB2
+	NOP
+
+	MADD3	c33, c33, a7, b8
+	daddiu	PREB, PREB, 16 * SIZE
+
+	FETCH	$0, 0 * SIZE(PREB)
+	MADD2	c32, c32, a8, b7
+	bgtz	L, .L32
+	MADD4	c34, c34, a8, b8
+
+
+.L35:
+#ifndef TRMMKERNEL
+	andi	L,  K, 3
+	LD	ALPHA_R, 128($sp)
+#else
+	andi	L,  TEMP, 3
+	LD	ALPHA_R, 128($sp)
+#endif
+	blez	L, .L38
+	LD	ALPHA_I, 136($sp)
+	.align	5
+
+.L36:
+	daddiu	L, L, -1
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	daddiu	BO, BO, 4 * SIZE 					#	2nr*1kr*cmpx
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	daddiu	AO, AO, 2 * SIZE					#	2mr*1kr*cmpx
+	MADD1	c31, c31, a1, b3 					#	A1xB2
+	MADD3	c33, c33, a1, b4
+	
+	daddiu	PREB, PREB, 4 * SIZE 
+	MADD2	c32, c32, a2, b3
+	MADD4	c34, c34, a2, b4
+
+	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)
+	bgtz	L, .L36
+	NOP
+
+.L38:
+#ifndef	TRMMKERNEL
+ 	ADD	c11, c14, c11
+	LD	a1,  0 * SIZE(CO1)
+	ADD	c12, c13, c12
+	LD	a2,  1 * SIZE(CO1)
+
+	ADD	c31, c34, c31
+	LD	a3,  0 * SIZE(CO2)
+	ADD	c32, c33, c32
+	LD	a4,  1 * SIZE(CO2)
+
+	MADD	a1, a1, ALPHA_R, c11
+	MADD	a2, a2, ALPHA_R, c12
+
+	MADD	a3, a3, ALPHA_R, c31
+	MADD	a4, a4, ALPHA_R, c32
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+
+	NMSUB	a3, a3, ALPHA_I, c32
+	MADD	a4, a4, ALPHA_I, c31
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+
+	ST	a3, 0 * SIZE(CO2)
+	ST	a4, 1 * SIZE(CO2)
+
+	daddiu	CO1,CO1, 2 * SIZE 
+	daddiu	CO2,CO2, 2 * SIZE 
+
+#else
+	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+
+	ADD	c31, c34, c31
+	ADD	c32, c33, c32
+
+	MUL	a1, ALPHA_R, c11
+	MUL	a2, ALPHA_R, c12
+
+	MUL	a3, ALPHA_R, c31
+	MUL	a4, ALPHA_R, c32
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+
+	NMSUB	a3, a3, ALPHA_I, c32
+	MADD	a4, a4, ALPHA_I, c31
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+
+	ST	a3, 0 * SIZE(CO2)
+	ST	a4, 1 * SIZE(CO2)
+
+	daddiu	CO1,CO1, 2 * SIZE 
+	daddiu	CO2,CO2, 2 * SIZE 
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+    (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, K, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -1
+#else
+	daddiu	TEMP, TEMP, -2
+#endif
+	dsll	L,    TEMP, ZBASE_SHIFT
+	dsll	TEMP, TEMP, 1 + ZBASE_SHIFT
+
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 1
+#endif
+#endif
+
+	.align 5
+
+.L19:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+	daddiu	KK, KK, 2
+#endif
+
+	bgtz	J, .L10
+	move	B, BO
+
+	.align 5
+	
+.L20:
+	andi	J,  N, 1
+	blez	J, .L999
+	dsll	PREA, K, 1+ZBASE_SHIFT				#	PREA=K*2*2^4
+
+	dsra	I,  M, 1							#	I=M/2
+	move	CO1, C
+
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	move	KK, OFFSET
+#endif
+
+	move	AO, A								#	Reset AO
+	blez	I, .L29
+	daddu	PREA, PREA, A
+
+.L21:
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	BO,  B
+#else
+	dsll	L,    KK, 1 + ZBASE_SHIFT
+	dsll	TEMP, KK, ZBASE_SHIFT
+
+	daddu	AO, AO, L
+	daddu	BO, B,  TEMP
+#endif
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	LD	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	MOV	c21, c11
+	MOV	c22, c11
+
+	FETCH	$0, 0 * SIZE(PREA)
+	MOV	c23, c11
+	MOV	c24, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	FETCH	$0, 4 * SIZE(CO1)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, K, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 2						#	define Mr=2
+#else
+	daddiu	TEMP, KK, 1						#	define	NR=1
+#endif
+	dsra	L,  TEMP, 2
+	blez	L, .L25
+	NOP
+
+#else
+	dsra	L,  K, 2							#	Unroll K 4 times	
+	move	BO,  B
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	LD	a3, 2 * SIZE(AO)
+	LD 	a4, 3 * SIZE(AO)
+	MOV	c21, c11
+	MOV	c22, c11
+
+	FETCH	$0, 0 * SIZE(PREA)
+	MOV	c23, c11
+	MOV	c24, c11
+
+	FETCH	$0, 0 * SIZE(CO1)
+	FETCH	$0, 4 * SIZE(CO1)
+
+	blez	L, .L25
+	NOP
+#endif
+
+	.align	5
+
+.L22:
+	LD	 a5, 4 * SIZE(AO)
+	LD	 a6, 5 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	LD	 b3, 2 * SIZE(BO)
+	LD	 b4, 3 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	LD 	 a7, 6 * SIZE(AO)
+	LD 	 a8, 7 * SIZE(AO)
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+	
+	FETCH	$0, 4 * SIZE(PREA)
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+	LD 	a1, 8 * SIZE(AO)
+	LD	a2, 9 * SIZE(AO)
+	MADD1	c11, c11, a5, b3 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b4					#	axd
+
+	LD	b5, 4 * SIZE(BO)
+	LD	b6, 5 * SIZE(BO)
+	MADD2	c12, c12, a6, b3					#	bxc
+	MADD4	c14, c14, a6, b4					#	bxd
+
+	LD	a3, 10 * SIZE(AO)
+	LD	a4, 11 * SIZE(AO)
+	MADD1	c21, c21, a7, b3 					#	A2xB1
+	MADD3	c23, c23, a7, b4
+
+	FETCH	$0, 8 * SIZE(PREA)
+	MADD2	c22, c22, a8, b3
+	MADD4	c24, c24, a8, b4
+	daddiu	L, L, -1
+
+	LD	a5, 12 * SIZE(AO)
+	LD	a6, 13 * SIZE(AO)
+	MADD1	c11, c11, a1, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b6					#	axd
+
+	LD	b7, 6 * SIZE(BO)
+	LD	b8, 7 * SIZE(BO)
+	MADD2	c12, c12, a2, b5					#	bxc
+	MADD4	c14, c14, a2, b6					#	bxd
+
+	LD	a7, 14 * SIZE(AO)
+	LD	a8, 15 * SIZE(AO)
+	MADD1	c21, c21, a3, b5 					#	A2xB1
+	MADD3	c23, c23, a3, b6
+
+	daddiu	BO, BO,  8 * SIZE  					#	1nr*4kr*cmpx
+	daddiu	AO, AO, 16 * SIZE 					#	2mr*4kr*cmpx
+
+	FETCH	$0, 12 * SIZE(PREA)
+	MADD2	c22, c22, a4, b5
+	MADD4	c24, c24, a4, b6
+	daddiu	PREA, PREA, 16 * SIZE 		
+
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MADD1	c11, c11, a5, b7 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b8					#	axd
+
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MADD2	c12, c12, a6, b7					#	bxc
+	MADD4	c14, c14, a6, b8					#	bxd
+
+	LD	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	MADD1	c21, c21, a7, b7 					#	A2xB1
+	MADD3	c23, c23, a7, b8
+
+	FETCH	$0, 0 * SIZE(PREA)
+	MADD2	c22, c22, a8, b7
+	bgtz	L, .L22
+	MADD4	c24, c24, a8, b8
+
+
+.L25:
+#ifndef TRMMKERNEL
+	andi	L,  K, 3
+	LD	ALPHA_R, 128($sp)
+#else
+	andi	L, TEMP, 3
+	LD	ALPHA_R, 128($sp)
+#endif
+	blez	L, .L28
+	LD	ALPHA_I, 136($sp)
+	.align	3
+
+.L26:
+	daddiu	L, L, -1
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+	daddiu	BO, BO, 2 * SIZE 					#	2nr*1kr*cmpx
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+	daddiu	AO, AO, 4 * SIZE 					#	2mr*1kr*cmpx
+	MADD1	c21, c21, a3, b1 					#	A2xB1
+	MADD3	c23, c23, a3, b2
+
+	daddiu	PREA, PREA, 4 * SIZE 				#	2mr*1kr*cmpx
+	MADD2	c22, c22, a4, b1
+	MADD4	c24, c24, a4, b2
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2	Unroll K=4
+#	gsLQC1(R12, F3, F2, 1)						#	R:a3	I:a4
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	LD	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+
+	bgtz	L, .L26
+	FETCH	$0, 0 * SIZE(PREA)
+
+.L28:
+#ifndef	TRMMKERNEL
+ 	ADD	c11, c14, c11
+	LD	a1,  0 * SIZE(CO1)
+	ADD	c12, c13, c12
+	LD	a2,  1 * SIZE(CO1)
+	ADD	c21, c24, c21
+	LD	b1,  2 * SIZE(CO1)
+	ADD	c22, c23, c22
+	LD	b2,  3 * SIZE(CO1)
+
+	daddiu	I, I, -1
+	MADD	a1, a1, ALPHA_R, c11
+	MADD	a2, a2, ALPHA_R, c12
+	MADD	b1, b1, ALPHA_R, c21
+	MADD	b2, b2, ALPHA_R, c22
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+	NMSUB	b1, b1, ALPHA_I, c22
+	MADD	b2, b2, ALPHA_I, c21
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+	ST	b1, 2 * SIZE(CO1)
+	ST	b2, 3 * SIZE(CO1)
+
+#else
+ 	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+	ADD	c21, c24, c21
+	ADD	c22, c23, c22
+
+	daddiu	I, I, -1
+	MUL	a1, ALPHA_R, c11
+	MUL	a2, ALPHA_R, c12
+	MUL	b1, ALPHA_R, c21
+	MUL	b2, ALPHA_R, c22
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+	NMSUB	b1, b1, ALPHA_I, c22
+	MADD	b2, b2, ALPHA_I, c21
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+	ST	b1, 2 * SIZE(CO1)
+	ST	b2, 3 * SIZE(CO1)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+    (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, K, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -2
+#else
+	daddiu	TEMP, TEMP, -1
+#endif
+
+	dsll	L,    TEMP, 1 + ZBASE_SHIFT
+	dsll	TEMP, TEMP, ZBASE_SHIFT
+
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 2
+#endif
+#endif
+	daddiu	CO1,CO1, 4 * SIZE 
+	bgtz	I, .L21
+	NOP
+
+.L29:
+	andi	I, M, 1
+	blez	I, .L999
+	NOP
+
+#if defined(TRMMKERNEL)
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	move	BO,  B
+#else
+	dsll	TEMP, KK,  ZBASE_SHIFT
+
+	daddu	AO, AO, TEMP
+	daddu	BO, B,  TEMP
+#endif
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	b1, 0 * SIZE(BO)
+	LD 	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	FETCH	$0, 0 * SIZE(PREA)
+	FETCH	$0, 4 * SIZE(PREA)
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	dsubu	TEMP, K, KK
+#elif defined(LEFT)
+	daddiu	TEMP, KK, 1
+#else
+	daddiu	TEMP, KK, 1
+#endif
+	dsra	L,  TEMP, 2
+	blez	L, .L45
+	NOP
+
+#else
+	dsra	L,  K, 2							#	Unroll K 4 times	
+	move	BO,  B
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MTC	$0,  c11								#	Clear results regs
+	MOV	c12, c11
+
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	b1, 0 * SIZE(BO)
+	LD 	b2, 1 * SIZE(BO)
+	MOV	c13, c11
+	MOV	c14, c11
+
+	FETCH	$0, 0 * SIZE(PREA)
+	FETCH	$0, 4 * SIZE(PREA)
+	blez	L, .L45
+	NOP
+#endif
+
+	.align	3
+
+.L42:
+#	gsLQC1(R12, F3, F2, 1)						#	R:a3	I:a4
+	LD 	a3, 2 * SIZE(AO)
+	LD	a4, 3 * SIZE(AO)
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+
+#	gsLQC1(R13, F7, F6, 1)						#	R:b2	I:b3
+	LD	b3, 2 * SIZE(BO)
+	LD	b4, 3 * SIZE(BO)
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+#	gsLQC1(R12, F9, F8, 2)						#	Unroll K=1	
+	LD	a5, 4 * SIZE(AO)
+	LD	a6, 5 * SIZE(AO)
+	MADD1	c11, c11, a3, b3 					#	axc		A1xB1	
+	MADD3	c13, c13, a3, b4					#	axd
+
+#	gsLQC1(R13, F13, F12, 2)				
+	LD	b5, 4 * SIZE(BO)
+	LD 	b6, 5 * SIZE(BO)
+	MADD2	c12, c12, a4, b3					#	bxc
+	MADD4	c14, c14, a4, b4					#	bxd
+
+#	gsLQC1(R12, F11, F10, 3)					
+	LD 	a7, 6 * SIZE(AO)
+	LD	a8, 7 * SIZE(AO)
+	MADD1	c11, c11, a5, b5 					#	axc		A1xB1	
+	MADD3	c13, c13, a5, b6					#	axd
+
+	daddiu	L, L, -1
+
+#	gsLQC1(R13, F16, F15, 3)			
+	LD	b7, 6 * SIZE(BO)
+	LD	b8, 7 * SIZE(BO)
+	MADD2	c12, c12, a6, b5					#	bxc
+	MADD4	c14, c14, a6, b6					#	bxd
+
+	daddiu	AO, AO, 8 * SIZE 					#	2mr*4kr*cmpx
+	daddiu	BO, BO, 8 * SIZE  					#	2nr*4kr*cmpx
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2	Unroll K=4
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	MADD1	c11, c11, a7, b7 					#	axc		A1xB1	
+	MADD3	c13, c13, a7, b8					#	axd
+
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	MADD2	c12, c12, a8, b7					#	bxc
+	MADD4	c14, c14, a8, b8					#	bxd
+
+	bgtz	L, .L42
+	NOP
+
+
+	.align 5
+
+.L45:
+#ifndef	TRMMKERNEL
+	andi	L,  K, 3
+	LD	ALPHA_R, 128($sp)
+#else
+	andi	L,  TEMP, 3
+	LD	ALPHA_R, 128($sp)
+#endif
+	blez	L, .L48
+	LD	ALPHA_I, 136($sp)
+
+.L46:
+	daddiu	L, L, -1
+	daddiu	BO, BO, 1 * SIZE * COMPSIZE			#	2nr*1kr*cmpx
+	daddiu	AO, AO, 1 * SIZE * COMPSIZE			#	2mr*1kr*cmpx
+
+	MADD1	c11, c11, a1, b1 					#	axc		A1xB1	
+	MADD3	c13, c13, a1, b2					#	axd
+	MADD2	c12, c12, a2, b1					#	bxc
+	MADD4	c14, c14, a2, b2					#	bxd
+
+#	gsLQC1(R12, F1, F0, 0)						#	R:a1	I:a2	Unroll K=4
+#	gsLQC1(R13, F5, F4, 0)						#	R:b1	I:b2
+	LD	a1, 0 * SIZE(AO)
+	LD	a2, 1 * SIZE(AO)
+	LD	b1, 0 * SIZE(BO)
+	LD	b2, 1 * SIZE(BO)
+	bgtz	L, .L46
+	NOP
+
+.L48:
+#ifndef	TRMMKERNEL
+ 	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+
+	LD	a1,  0 * SIZE(CO1)
+	LD	a2,  1 * SIZE(CO1)
+
+	MADD	a1, a1, ALPHA_R, c11
+	MADD	a2, a2, ALPHA_R, c12
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+
+#else
+	ADD	c11, c14, c11
+	ADD	c12, c13, c12
+
+	MUL	a1, ALPHA_R, c11
+	MUL	a2, ALPHA_R, c12
+
+	NMSUB	a1, a1, ALPHA_I, c12
+	MADD	a2, a2, ALPHA_I, c11
+
+	ST	a1, 0 * SIZE(CO1)
+	ST	a2, 1 * SIZE(CO1)
+
+#if ( defined(LEFT) &&  defined(TRANSA)) || \
+    (!defined(LEFT) && !defined(TRANSA))
+	dsubu	TEMP, K, KK
+#ifdef LEFT
+	daddiu	TEMP, TEMP, -1
+#else
+	daddiu	TEMP, TEMP, -1
+#endif
+
+	dsll	TEMP, TEMP, ZBASE_SHIFT
+
+	daddu	AO, AO, TEMP
+	daddu	BO, BO, TEMP
+#endif
+
+#ifdef LEFT
+	daddiu	KK, KK, 1
+#endif
+
+	daddiu	CO1,CO1, 2 * SIZE 
+#endif
+
+
+
+	.align 5
+
+.L999:
+	LDARG	$16,   0($sp)
+	LDARG	$17,   8($sp)
+	ldc1	$f24, 16($sp)
+	ldc1	$f25, 24($sp)
+	ldc1	$f26, 32($sp)
+	ldc1	$f27, 40($sp)
+	ldc1	$f28, 48($sp)
+	ldc1	$f29, 56($sp)
+
+#if defined(TRMMKERNEL)
+	LDARG	$18,  64($sp)
+	LDARG	$19,  72($sp)
+	LDARG	$20,  80($sp)
+#endif
+
+#ifndef __64BIT__
+	ldc1	$f20, 88($sp)
+	ldc1	$f21, 96($sp)
+	ldc1	$f22,104($sp)
+	ldc1	$f23,112($sp)
+#endif
+
+	j	$31
+	daddiu	$sp, $sp, STACKSIZE
+
+	EPILOGUE
diff --git a/param.h b/param.h
index 4ffe05cf8..72d721d4e 100644
--- a/param.h
+++ b/param.h
@@ -1502,10 +1502,51 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_DEFAULT_Q 128
 #define ZGEMM_DEFAULT_Q 80
 
-#define SGEMM_DEFAULT_R 1024
+#define SGEMM_DEFAULT_R 640 
 #define DGEMM_DEFAULT_R dgemm_r 
-#define CGEMM_DEFAULT_R 1024
-#define ZGEMM_DEFAULT_R 1024
+#define CGEMM_DEFAULT_R 640
+#define ZGEMM_DEFAULT_R 640
+
+#define GEMM_OFFSET_A1	0x10000
+#define	GEMM_OFFSET_B1	0x100000
+
+#define SYMV_P	16
+#endif
+
+#ifdef LOONGSON3B
+#define SNUMOPT		2
+#define DNUMOPT		2
+
+#define GEMM_DEFAULT_OFFSET_A 0
+#define GEMM_DEFAULT_OFFSET_B 0
+#define GEMM_DEFAULT_ALIGN 0x03fffUL
+
+#define SGEMM_DEFAULT_UNROLL_M  2
+#define SGEMM_DEFAULT_UNROLL_N  2
+
+#define DGEMM_DEFAULT_UNROLL_M  2
+#define DGEMM_DEFAULT_UNROLL_N  2
+
+#define CGEMM_DEFAULT_UNROLL_M  2
+#define CGEMM_DEFAULT_UNROLL_N  2
+
+#define ZGEMM_DEFAULT_UNROLL_M  2
+#define ZGEMM_DEFAULT_UNROLL_N  2
+
+#define SGEMM_DEFAULT_P	64
+#define DGEMM_DEFAULT_P	24
+#define CGEMM_DEFAULT_P 24
+#define ZGEMM_DEFAULT_P 20
+
+#define SGEMM_DEFAULT_Q 192
+#define DGEMM_DEFAULT_Q 128
+#define CGEMM_DEFAULT_Q 128
+#define ZGEMM_DEFAULT_Q 64
+
+#define SGEMM_DEFAULT_R 512
+#define DGEMM_DEFAULT_R 512 
+#define CGEMM_DEFAULT_R 512
+#define ZGEMM_DEFAULT_R 512
 
 #define GEMM_OFFSET_A1	0x10000
 #define	GEMM_OFFSET_B1	0x100000