From 432c309f63736ab3ac7205cbbbf294070d6d4d93 Mon Sep 17 00:00:00 2001
From: traz <wangqian10@iscas.ac.cn>
Date: Tue, 10 May 2011 12:48:43 +0000
Subject: [PATCH] Finish dtrsm_kernel_Lx.S on Loongson3A.

---
 kernel/mips64/trsm_kernel_LN_loongson3a.S | 1938 +++++++++++++++++++++++++++++
 kernel/mips64/trsm_kernel_LT_loongson3a.S | 1783 ++++++++++++++++++++++++++
 2 files changed, 3721 insertions(+)
 create mode 100644 kernel/mips64/trsm_kernel_LN_loongson3a.S
 create mode 100644 kernel/mips64/trsm_kernel_LT_loongson3a.S

diff --git a/kernel/mips64/trsm_kernel_LN_loongson3a.S b/kernel/mips64/trsm_kernel_LN_loongson3a.S
new file mode 100644
index 0000000..aba86fb
--- /dev/null
+++ b/kernel/mips64/trsm_kernel_LN_loongson3a.S
@@ -0,0 +1,1938 @@
+#define REALNAME ASMNAME
+
+#define ASSEMBLER
+#include "common.h"
+
+
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$8
+#define B	$9
+#define C	$10
+#define LDC	$11
+
+#define AO	$12
+#define BO	$13
+
+#define I	$2
+#define J	$3
+#define L	$7
+
+#define CO1	$14
+#define CO2	$15
+#define CO3	$16
+#define CO4	$17
+
+#define OFFSET	$22
+#define KK	$23
+#define TEMP	$24
+#define AORIG	$25
+
+#define a1	$f0
+#define a2	$f1
+#define a3	$f2
+#define a4	$f3
+#define a5	$f4
+#define a6	$f5
+#define a7	$f6
+#define	a8	$f7
+
+#define b1	$f8
+#define b2	$f9
+#define b3	$f10
+#define b4	$f11
+#define b5	$f12
+#define b6	$f13
+#define b7	$f14
+#define b8	$f15
+
+#define t11	$f16
+#define t21	$f17
+#define t31	$f18
+#define t41	$f19
+
+#define t12	$f20
+#define t22	$f21
+#define t32	$f22
+#define t42	$f23
+
+#define t13	$f24
+#define t23	$f25
+#define t33	$f26
+#define t43	$f27
+
+#define t14	$f28
+#define t24	$f29
+#define t34	$f30
+#define t44	$f31
+
+#define ALPHA	$f15
+
+	PROLOGUE
+	
+	daddiu	$sp, $sp, -144
+
+	SDARG	$16,   0($sp)
+	SDARG	$17,   8($sp)
+	SDARG	$18,  16($sp)
+	SDARG	$19,  24($sp)
+	SDARG	$20,  32($sp)
+	SDARG	$21,  40($sp)
+	sdc1	$f24, 48($sp)
+	sdc1	$f25, 56($sp)
+	sdc1	$f26, 64($sp)
+	sdc1	$f27, 72($sp)
+	sdc1	$f28, 80($sp)
+
+	SDARG	$22,  88($sp)
+	SDARG	$23,  96($sp)
+	SDARG	$24, 104($sp)
+	SDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	sdc1	$f20,112($sp)
+	sdc1	$f21,120($sp)
+	sdc1	$f22,128($sp)
+	sdc1	$f23,136($sp)
+#endif
+											#	LN compute from bottom to top
+	LDARG	OFFSET, 144($sp)					
+	dsll	LDC, LDC, BASE_SHIFT			#	ldc
+
+	mult	M, K
+	mflo	TEMP							#	TEMP=MC*KC
+
+	dsll	TEMP, TEMP, BASE_SHIFT			
+	daddu	A, A, TEMP						#	A move to the end of sa
+
+	dsll	TEMP, M, BASE_SHIFT
+	daddu	C, C, TEMP						#	C+=MC
+
+	dsra	J,  N, 2						#	j = nc/4
+	blez	J, .L30
+	nop
+
+.L10:										#	nr=4
+	daddiu	J, J, -1
+	move	CO1, C
+	daddu	CO2, C,   LDC
+	daddu	CO3, CO2, LDC
+	daddu	CO4, CO3, LDC
+
+	MTC	$0,  t11							#	clear result registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	daddu	KK, M, OFFSET					#	kc - kk is the length of the rectangular data part of panel Ai 
+	move	AORIG, A						#	reset A
+
+	daddu	C,  CO4, LDC					#	fixed pointer C, the write back address
+	
+	andi	I,  M, 1						#	mr=2,nr=4	
+	blez	I, .L50
+	nop
+
+	dsll	TEMP,   K, BASE_SHIFT			#	mr=1
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of Ai
+
+	dsll	L,    KK, BASE_SHIFT			#	mr=1	
+	dsll	TEMP, KK, 2 + BASE_SHIFT		#	nr=4
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+	MOV	t13, t11							#	mr=2
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)				
+
+	dsra	L,  TEMP, 2
+	blez	L, .L55
+	nop
+
+
+	.align	3
+.L52:
+	LD	a5,  1 * SIZE(AO)					
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	LD	a3,   2 * SIZE(AO)
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t12, t12, a5, b6
+	MADD	t13, t13, a5, b7
+	MADD	t14, t14, a5, b8
+
+	LD	a7,   3 * SIZE(AO)
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1				#	3rd compute
+	MADD	t12, t12, a3, b2
+	MADD	t13, t13, a3, b3
+	MADD	t14, t14, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 1mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5				#	4th compute
+	MADD	t12, t12, a7, b6
+	MADD	t13, t13, a7, b7
+	MADD	t14, t14, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L52
+	nop
+
+
+	.align 3
+.L55:
+	andi	L, TEMP, 3
+	blez	L, .L58
+	nop
+	
+	.align	3
+.L56:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 1mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L56
+	nop
+
+
+.L58:										#	deal with the triangular part
+	daddiu	TEMP, KK, -1
+	dsll	L,    TEMP, BASE_SHIFT			#	mr=1
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	Ao point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+
+
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+	MUL	t12, b3, t12
+	MUL	t13, b3, t13
+	MUL	t14, b3, t14
+
+	daddiu	CO1, CO1, -1 * SIZE
+	daddiu	CO2, CO2, -1 * SIZE
+	daddiu	CO3, CO3, -1 * SIZE
+	daddiu	CO4, CO4, -1 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+
+
+	daddiu	KK, KK, -1							#	the length of rectangular data part increases by 1
+	MTC	$0,  t11							#	clear result registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+
+	
+.L50:	
+	andi	I,  M, 2						#	mr=2,nr=4	
+	blez	I, .L20
+	nop
+
+	dsll	TEMP,   K, 1 + BASE_SHIFT
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of Ai
+
+	dsll	L,    KK, 1 + BASE_SHIFT		
+	dsll	TEMP, KK, 2 + BASE_SHIFT
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+	MOV	t13, t11							#	mr=2
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)				
+
+	dsra	L,  TEMP, 2
+	blez	L, .L25
+	nop
+
+
+	.align	3
+.L22:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	LD	a3,   4 * SIZE(AO)
+	LD	a4,   5 * SIZE(AO)
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+
+	LD	a7,   6 * SIZE(AO)
+	LD	a8,   7 * SIZE(AO)
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1				#	3rd compute
+	MADD	t21, t21, a4, b1
+	MADD	t12, t12, a3, b2
+	MADD	t22, t22, a4, b2
+	MADD	t13, t13, a3, b3
+	MADD	t23, t23, a4, b3
+	MADD	t14, t14, a3, b4
+	MADD	t24, t24, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5				#	4th compute
+	MADD	t21, t21, a8, b5
+	MADD	t12, t12, a7, b6
+	MADD	t22, t22, a8, b6
+	MADD	t13, t13, a7, b7
+	MADD	t23, t23, a8, b7
+	MADD	t14, t14, a7, b8
+	MADD	t24, t24, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L22
+	nop
+
+
+	.align 3
+.L25:
+	andi	L, TEMP, 3
+	blez	L, .L28
+	nop
+	
+	.align	3
+.L26:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L26
+	nop
+
+
+.L28:										#	deal with the triangular part
+	daddiu	TEMP, KK, -2
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	Ao point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+	SUB	t21, b5, t21
+	SUB	t22, b6, t22
+	SUB	t23, b7, t23
+	SUB	t24, b8, t24
+
+
+	LD	b1,  3 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  2 * SIZE(AO)
+	MUL	t21, b1, t21
+	MUL	t22, b1, t22
+	MUL	t23, b1, t23
+	MUL	t24, b1, t24
+	NMSUB	t11, t11, b2, t21
+	NMSUB	t12, t12, b2, t22
+	NMSUB	t13, t13, b2, t23
+	NMSUB	t14, t14, b2, t24
+	
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+	MUL	t12, b3, t12
+	MUL	t13, b3, t13
+	MUL	t14, b3, t14
+
+	daddiu	CO1, CO1, -2 * SIZE
+	daddiu	CO2, CO2, -2 * SIZE
+	daddiu	CO3, CO3, -2 * SIZE
+	daddiu	CO4, CO4, -2 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+	ST	t21,  4 * SIZE(BO)
+	ST	t22,  5 * SIZE(BO)
+	ST	t23,  6 * SIZE(BO)
+	ST	t24,  7 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+
+
+
+	daddiu	KK, KK, -2							#	the length of rectangular data part increases by 2
+	MTC	$0,  t11							#	clear result registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+
+.L20:
+	dsra	I,  M, 2						#	I=MC/4
+	blez	I, .L29
+	nop
+
+.L11:										#	mr=4
+	dsll	TEMP,   K,  2 + BASE_SHIFT		#	TEMP=KC*MR*data_Byte
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of panel Ai
+	dsll	L,    KK, 2 + BASE_SHIFT		# 	KC-KK is the length of the rectangular data part of Ai	
+	dsll	TEMP, KK, 2 + BASE_SHIFT		#	KK*NR*data_Byte
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B, TEMP
+
+	dsubu	TEMP, K, KK	
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)				
+
+	MOV	t13, t11							#	clear result registers
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	dsra	L,  TEMP, 2						#	L=(KC-offset)/4
+	blez	L, .L15
+	nop
+
+	.align	3
+.L12:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	4th compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			
+
+	daddiu	L, L, -1
+	bgtz	L, .L12
+	nop
+	
+
+	.align 3
+.L15:
+	andi	L, TEMP, 3
+	blez	L, .L18
+	nop
+
+	.align	3
+.L16:
+	MADD	t11, t11, a1, b1				
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L16
+	nop
+
+
+.L18:										#	deal with the triangular data part of panel Ai
+	daddiu	TEMP, KK, -4					#	
+
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	AO point to the triangular data part
+	daddu	BO, B, TEMP
+	
+	LD	b1,  0 * SIZE(BO)					#	triangular_part*X + rectangular_part = B
+	LD	b2,  1 * SIZE(BO)					#	triangular_part*X = B - rectangular_part
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+
+	LD	b5,  4 * SIZE(BO)					#	sb store in row major
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+	
+	SUB	t21, b5, t21
+	SUB	t22, b6, t22
+	SUB	t23, b7, t23
+	SUB	t24, b8, t24
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3, 10 * SIZE(BO)
+	LD	b4, 11 * SIZE(BO)
+	
+	SUB	t31, b1, t31
+	SUB	t32, b2, t32
+	SUB	t33, b3, t33
+	SUB	t34, b4, t34
+	
+	LD	b5, 12 * SIZE(BO)
+	LD	b6, 13 * SIZE(BO)
+	LD	b7, 14 * SIZE(BO)
+	LD	b8, 15 * SIZE(BO)
+
+	SUB	t41, b5, t41
+	SUB	t42, b6, t42
+	SUB	t43, b7, t43
+	SUB	t44, b8, t44
+
+
+	LD	b1,  15 * SIZE(AO)
+	LD	b2,  14 * SIZE(AO)	
+	LD	b4,  13 * SIZE(AO)
+	LD	b7,  12 * SIZE(AO)
+	
+	MUL	t41, b1, t41
+	MUL	t42, b1, t42
+	MUL	t43, b1, t43
+	MUL	t44, b1, t44
+	NMSUB	t31, t31, b2, t41
+	NMSUB	t32, t32, b2, t42
+	NMSUB	t33, t33, b2, t43
+	NMSUB	t34, t34, b2, t44
+	NMSUB	t21, t21, b4, t41
+	NMSUB	t22, t22, b4, t42
+	NMSUB	t23, t23, b4, t43
+	NMSUB	t24, t24, b4, t44
+	NMSUB	t11, t11, b7, t41
+	NMSUB	t12, t12, b7, t42
+	NMSUB	t13, t13, b7, t43
+	NMSUB	t14, t14, b7, t44
+
+
+
+	LD	b3,  10 * SIZE(AO)	
+	LD	b5,   9 * SIZE(AO)
+	LD	b8,   8 * SIZE(AO)
+	MUL	t31, b3, t31
+	MUL	t32, b3, t32
+	MUL	t33, b3, t33
+	MUL	t34, b3, t34
+	NMSUB	t21, t21, b5, t31
+	NMSUB	t22, t22, b5, t32
+	NMSUB	t23, t23, b5, t33
+	NMSUB	t24, t24, b5, t34
+	NMSUB	t11, t11, b8, t31
+	NMSUB	t12, t12, b8, t32
+	NMSUB	t13, t13, b8, t33
+	NMSUB	t14, t14, b8, t34
+
+
+
+	LD	b6,   5 * SIZE(AO)
+	LD	b1,   4 * SIZE(AO)
+	MUL	t21, b6, t21
+	MUL	t22, b6, t22
+	MUL	t23, b6, t23
+	MUL	t24, b6, t24
+	NMSUB	t11, t11, b1, t21
+	NMSUB	t12, t12, b1, t22
+	NMSUB	t13, t13, b1, t23
+	NMSUB	t14, t14, b1, t24
+
+
+
+	LD	b2,   0 * SIZE(AO)
+	MUL	t11, b2, t11
+	MUL	t12, b2, t12
+	MUL	t13, b2, t13
+	MUL	t14, b2, t14
+
+	daddiu	CO1, CO1, -4 * SIZE				#	modify 
+	daddiu	CO2, CO2, -4 * SIZE
+	daddiu	CO3, CO3, -4 * SIZE
+	daddiu	CO4, CO4, -4 * SIZE
+
+
+	ST	t11,  0 * SIZE(BO)					#	update packed B
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+	ST	t21,  4 * SIZE(BO)
+	ST	t22,  5 * SIZE(BO)
+	ST	t23,  6 * SIZE(BO)
+	ST	t24,  7 * SIZE(BO)
+	ST	t31,  8 * SIZE(BO)
+	ST	t32,  9 * SIZE(BO)
+	ST	t33, 10 * SIZE(BO)
+	ST	t34, 11 * SIZE(BO)
+	ST	t41, 12 * SIZE(BO)
+	ST	t42, 13 * SIZE(BO)
+	ST	t43, 14 * SIZE(BO)
+	ST	t44, 15 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)					#	write back 
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t33,  2 * SIZE(CO3)
+	ST	t43,  3 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+	ST	t34,  2 * SIZE(CO4)
+	ST	t44,  3 * SIZE(CO4)
+
+
+	daddiu	KK, KK, -4						#	KC-KK is the length of the rectangular data part, LN compute from bottom to top so KK-=4
+	daddiu	I, I, -1
+
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+	MOV	t12, a1
+	MOV	t22, a1
+	MOV	t32, a1
+	MOV	t42, a1
+	bgtz	I, .L11
+	nop
+
+	.align 3
+
+.L29:
+	dsll	TEMP, K, 2 + BASE_SHIFT
+	daddu	B, B, TEMP							# B point to next Bj
+
+	bgtz	J, .L10
+	nop
+	
+	
+	.align 3
+.L30:
+	andi	J,  N, 2							#	nr=2
+	blez	J, .L70
+	nop
+
+	move	CO1, C
+	daddu	CO2, C,   LDC
+
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	daddu	KK, M, OFFSET
+	move	AORIG, A							#	reset A
+	
+	daddu	C,  CO2, LDC						#	fixed 
+
+	andi	I,  M, 1							#	mr=1
+	blez	I, .L60
+	nop
+
+	dsll	TEMP,   K, BASE_SHIFT
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of everypanel of Ai
+
+	dsll	L,    KK, BASE_SHIFT			#	mr=1
+	dsll	TEMP, KK, 1 + BASE_SHIFT		#	nr=2
+
+	daddu	AO, AORIG, L					#	AO point to rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+	MOV	t12, t11							#	clear result registers
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L65
+	nop
+
+
+	.align	3
+.L62:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t12, t12, a1, b2
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t12, t12, a5, b6
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t12, t12, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 1mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t12, t12, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L62
+	nop
+	
+	.align 3
+
+.L65:
+	andi	L, TEMP, 3
+	blez	L, .L68
+	nop
+	
+	.align	3
+.L66:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L66
+	nop
+
+.L68:
+	daddiu	TEMP, KK, -1					#	mr=1
+
+	dsll	L,    TEMP, BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	Ao point to the triangular data part
+	daddu	BO, B,     TEMP
+	
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+
+	
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+	MUL	t12, b3, t12
+
+	daddiu	CO1, CO1, -1 * SIZE
+	daddiu	CO2, CO2, -1 * SIZE
+
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+
+
+	daddiu	KK, KK, -1
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+
+
+
+.L60:
+	andi	I,  M, 2
+	blez	I, .L40
+	nop
+
+	dsll	TEMP,   K, 1 + BASE_SHIFT
+	dsubu	AORIG, AORIG, TEMP				#	AORIG point to the beginning address of everypanel of Ai
+
+	dsll	L,    KK, 1 + BASE_SHIFT		#	mr=2
+	dsll	TEMP, KK, 1 + BASE_SHIFT		#	nr=2
+
+	daddu	AO, AORIG, L					#	AO point to rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+
+	MOV	t12, t11							#	clear result registers
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L45
+	nop
+
+
+	.align	3
+.L42:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t21, t21, a4, b3
+	MADD	t12, t12, a3, b4
+	MADD	t22, t22, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t21, t21, a8, b7
+	MADD	t12, t12, a7, b8
+	MADD	t22, t22, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L42
+	nop
+	
+	.align 3
+
+.L45:
+	andi	L, TEMP, 3
+	blez	L, .L48
+	nop
+	
+	.align	3
+.L46:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L46
+	nop
+
+.L48:
+	daddiu	TEMP, KK, -2
+
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AORIG, L					#	Ao point to the triangular data part
+	daddu	BO, B,     TEMP
+	
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t21, b3, t21
+	SUB	t22, b4, t22
+
+	LD	b1,  3 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  2 * SIZE(AO)
+	MUL	t21, b1, t21
+	MUL	t22, b1, t22
+	NMSUB	t11, t11, b2, t21
+	NMSUB	t12, t12, b2, t22
+	
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+	MUL	t12, b3, t12
+
+	daddiu	CO1, CO1, -2 * SIZE
+	daddiu	CO2, CO2, -2 * SIZE
+
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t21,  2 * SIZE(BO)
+	ST	t22,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+
+
+	daddiu	KK, KK, -2
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+
+.L40:
+	dsra	I,  M, 2							#	I = mc/4
+	blez	I, .L49
+	nop
+
+.L31:
+	dsll	TEMP,   K,  2 + BASE_SHIFT
+	dsubu	AORIG, AORIG, TEMP					#	AORIG point to the beginning address of panel Ai
+	dsll	L,    KK, 2 + BASE_SHIFT			#	mr=4
+	dsll	TEMP, KK, 1 + BASE_SHIFT			#	nr=2
+
+	daddu	AO, AORIG, L						#	AO point to the rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+	LD	b2,  1 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L35
+	nop
+
+
+	.align	3
+.L32:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+	LD	b3,   4 * SIZE(BO)
+	LD	b4,   5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3				#	3rd compute
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+	MADD	t12, t12, a1, b4
+	MADD	t22, t22, a2, b4
+	MADD	t32, t32, a3, b4
+	MADD	t42, t42, a4, b4
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7				#	4th compute
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+	MADD	t12, t12, a5, b8
+	MADD	t22, t22, a6, b8
+	MADD	t32, t32, a7, b8
+	MADD	t42, t42, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L32
+	nop
+
+	
+	.align 3
+
+.L35:
+	andi	L, TEMP, 3
+	blez	L, .L38
+	nop
+	
+	.align	3
+.L36:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+	
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L36
+	nop
+
+
+.L38:										#
+	daddiu	TEMP, KK, -4
+	dsll	L,    TEMP, 2 + BASE_SHIFT		#	mr=4
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT		#	nr=2
+	daddu	AO, AORIG, L					#	AO point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t21, b3, t21
+	SUB	t22, b4, t22
+	SUB	t31, b5, t31
+	SUB	t32, b6, t32
+	SUB	t41, b7, t41
+	SUB	t42, b8, t42
+
+
+	LD	b1,  15 * SIZE(AO)
+	LD	b2,  14 * SIZE(AO)	
+	LD	b4,  13 * SIZE(AO)
+	LD	b7,  12 * SIZE(AO)
+	
+	MUL	t41, b1, t41
+	MUL	t42, b1, t42
+	NMSUB	t31, t31, b2, t41
+	NMSUB	t32, t32, b2, t42
+	NMSUB	t21, t21, b4, t41
+	NMSUB	t22, t22, b4, t42
+	NMSUB	t11, t11, b7, t41
+	NMSUB	t12, t12, b7, t42
+
+
+
+	LD	b3,  10 * SIZE(AO)	
+	LD	b5,   9 * SIZE(AO)
+	LD	b8,   8 * SIZE(AO)
+	MUL	t31, b3, t31
+	MUL	t32, b3, t32
+	NMSUB	t21, t21, b5, t31
+	NMSUB	t22, t22, b5, t32
+	NMSUB	t11, t11, b8, t31
+	NMSUB	t12, t12, b8, t32
+
+
+
+	LD	b6,   5 * SIZE(AO)
+	LD	b1,   4 * SIZE(AO)
+	MUL	t21, b6, t21
+	MUL	t22, b6, t22
+	NMSUB	t11, t11, b1, t21
+	NMSUB	t12, t12, b1, t22
+
+
+	LD	b2,   0 * SIZE(AO)
+	MUL	t11, b2, t11
+	MUL	t12, b2, t12
+
+	daddiu	CO1, CO1, -4 * SIZE
+	daddiu	CO2, CO2, -4 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t21,  2 * SIZE(BO)
+	ST	t22,  3 * SIZE(BO)
+	ST	t31,  4 * SIZE(BO)
+	ST	t32,  5 * SIZE(BO)
+	ST	t41,  6 * SIZE(BO)
+	ST	t42,  7 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+
+
+	daddiu	KK, KK, -4
+
+	MTC	$0,  t11
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	daddiu	I, I, -1
+	bgtz	I, .L31
+	nop
+
+
+
+	.align 3
+.L49:
+	dsll	TEMP, K, 1 + BASE_SHIFT		# 	nr=2
+	daddu	B, B, TEMP
+
+	.align 3
+
+.L70:
+	andi	J,  N, 1					#	nr=1
+	blez	J, .L999					#	END
+	nop
+
+	move	CO1, C
+
+	daddu	KK, M, OFFSET
+	move	AORIG, A					#	reset A
+
+	andi	I,  M, 1					#	mr=1
+	blez	I, .L90
+	NOP
+
+	MTC	$0,  t11
+
+	dsll	TEMP,   K, BASE_SHIFT			#	mr=1
+	dsubu	AORIG, AORIG, TEMP
+	
+	dsll	L,    KK, BASE_SHIFT
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B,    L 
+
+	dsubu	TEMP, K, KK
+
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L95
+	nop
+	
+	.align	3
+.L92:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 1mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+
+	daddiu	L, L, -1
+	bgtz	L, .L92
+	nop
+	
+	.align 3
+
+.L95:
+	andi	L, TEMP,  3
+	blez	L, .L98
+	nop
+
+	.align	3
+.L96:
+	MADD	t11, t11, a1, b1				#	3rd compute
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 1mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L96
+	nop
+
+
+.L98:
+	daddiu	TEMP, KK, -1					# 	mr=2
+	dsll	TEMP, TEMP,  BASE_SHIFT
+
+	daddu	AO, AORIG, TEMP					#	AO point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+
+	SUB	t11, b1, t11
+
+
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+
+	daddiu	CO1, CO1, -1 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+
+	daddiu	KK, KK, -1
+
+
+.L90:	
+	andi	I,  M, 2
+	blez	I, .L80
+	NOP
+
+	MTC	$0,  t11
+	MOV	t21, t11							#	clear result registers
+
+	dsll	TEMP,   K, 1+BASE_SHIFT			#	mr=2
+	dsubu	AORIG, AORIG, TEMP
+	
+	dsll	L,    KK, 1 + BASE_SHIFT
+	dsll	TEMP, KK, 0 + BASE_SHIFT
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular data part
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+
+	dsra	L,  TEMP, 2
+	blez	L, .L85
+	nop
+	
+	.align	3
+.L82:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t21, t21, a4, b3
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t21, t21, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L82
+	nop
+	
+	.align 3
+
+.L85:
+	andi	L, TEMP,  3
+	blez	L, .L88
+	nop
+
+	.align	3
+.L86:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L86
+	nop
+
+
+.L88:
+	daddiu	TEMP, KK, -2					# 	mr=2
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+
+	daddu	AO, AORIG, L					#	AO point to the triangular data part
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+	LD	b1,  3 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  2 * SIZE(AO)
+	MUL	t21, b1, t21
+	NMSUB	t11, t11, b2, t21
+	
+	LD	b3,  0 * SIZE(AO)
+	MUL	t11, b3, t11
+
+	daddiu	CO1, CO1, -2 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t21,  1 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+
+	daddiu	KK, KK, -2
+	
+	
+	.align	3
+.L80:
+	dsra	I,  M, 2
+	blez	I, .L89
+	nop
+
+.L71:
+	dsll	TEMP,   K,  2 + BASE_SHIFT		#	mr=4
+	dsubu	AORIG, AORIG, TEMP
+
+	dsll	L,    KK, 2 + BASE_SHIFT		#	mr=4
+	dsll	TEMP, KK, 0 + BASE_SHIFT		#	nr=1
+
+	daddu	AO, AORIG, L					#	AO point to the rectangular
+	daddu	BO, B,     TEMP
+
+	dsubu	TEMP, K, KK
+
+
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(BO)					#	get 4b
+
+	dsra	L,  TEMP, 2
+	blez	L, .L75
+	nop										#	reset B
+
+	.align	3
+.L72:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3				#	3rd compute
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7				#	4th compute
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L72
+	nop
+	
+	.align 3
+
+.L75:
+	andi	L, TEMP,  3
+	blez	L, .L78
+	nop
+
+	.align	3
+.L76:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L76
+	nop
+
+.L78:
+	daddiu	TEMP, KK, -4				#	mr=4
+
+	dsll	L,    TEMP, 2 + BASE_SHIFT	#	mr=4
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT	#	nr=1
+	daddu	AO, AORIG, L				#	AO point to the triangular
+	daddu	BO, B,     TEMP
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+	LD	b1,  15 * SIZE(AO)
+	LD	b2,  14 * SIZE(AO)	
+	LD	b4,  13 * SIZE(AO)
+	LD	b7,  12 * SIZE(AO)
+	MUL	t41, b1, t41
+	NMSUB	t31, t31, b2, t41
+	NMSUB	t21, t21, b4, t41
+	NMSUB	t11, t11, b7, t41
+
+
+
+	LD	b3,  10 * SIZE(AO)	
+	LD	b5,   9 * SIZE(AO)
+	LD	b8,   8 * SIZE(AO)
+	MUL	t31, b3, t31
+	NMSUB	t21, t21, b5, t31
+	NMSUB	t11, t11, b8, t31
+
+
+
+	LD	b6,   5 * SIZE(AO)
+	LD	b1,   4 * SIZE(AO)
+	MUL	t21, b6, t21
+	NMSUB	t11, t11, b1, t21
+
+
+
+	LD	b2,   0 * SIZE(AO)
+	MUL	t11, b2, t11
+
+	daddiu	CO1, CO1, -4 * SIZE
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t21,  1 * SIZE(BO)
+	ST	t31,  2 * SIZE(BO)
+	ST	t41,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+
+
+	daddiu	KK, KK, -4
+	daddiu	I, I, -1
+	bgtz	I, .L71
+	nop
+
+
+	.align 3
+.L89:
+	dsll	TEMP, K, BASE_SHIFT			#	nr=1
+	daddu	B, B, TEMP
+
+
+
+	.align 3
+
+.L999:
+	LDARG	$16,   0($sp)
+	LDARG	$17,   8($sp)
+	LDARG	$18,  16($sp)
+	LDARG	$19,  24($sp)
+	LDARG	$20,  32($sp)
+	LDARG	$21,  40($sp)
+	ldc1	$f24, 48($sp)
+	ldc1	$f25, 56($sp)
+	ldc1	$f26, 64($sp)
+	ldc1	$f27, 72($sp)
+	ldc1	$f28, 80($sp)
+
+	LDARG	$22,  88($sp)
+	LDARG	$23,  96($sp)
+	LDARG	$24, 104($sp)
+	LDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	ldc1	$f20,112($sp)
+	ldc1	$f21,120($sp)
+	ldc1	$f22,128($sp)
+	ldc1	$f23,136($sp)
+#endif
+
+	j	$31
+	daddiu	$sp, $sp, 144
+
+	EPILOGUE
diff --git a/kernel/mips64/trsm_kernel_LT_loongson3a.S b/kernel/mips64/trsm_kernel_LT_loongson3a.S
new file mode 100644
index 0000000..4114d94
--- /dev/null
+++ b/kernel/mips64/trsm_kernel_LT_loongson3a.S
@@ -0,0 +1,1783 @@
+#define REALNAME ASMNAME
+
+#define ASSEMBLER
+#include "common.h"
+
+
+#define M	$4
+#define	N	$5
+#define	K	$6
+#define A	$8
+#define B	$9
+#define C	$10
+#define LDC	$11
+
+#define AO	$12
+#define BO	$13
+
+#define I	$2
+#define J	$3
+#define L	$7
+
+#define CO1	$14
+#define CO2	$15
+#define CO3	$16
+#define CO4	$17
+
+#define OFFSET	$22
+#define KK	$23
+#define TEMP	$24
+#define AORIG	$25
+
+#define a1	$f0
+#define a2	$f1
+#define a3	$f2
+#define a4	$f3
+#define a5	$f4
+#define a6	$f5
+#define a7	$f6
+#define	a8	$f7
+
+#define b1	$f8
+#define b2	$f9
+#define b3	$f10
+#define b4	$f11
+#define b5	$f12
+#define b6	$f13
+#define b7	$f14
+#define b8	$f15
+
+#define t11	$f16
+#define t21	$f17
+#define t31	$f18
+#define t41	$f19
+
+#define t12	$f20
+#define t22	$f21
+#define t32	$f22
+#define t42	$f23
+
+#define t13	$f24
+#define t23	$f25
+#define t33	$f26
+#define t43	$f27
+
+#define t14	$f28
+#define t24	$f29
+#define t34	$f30
+#define t44	$f31
+
+#define ALPHA	$f15
+
+	PROLOGUE
+	
+	daddiu	$sp, $sp, -144
+
+	SDARG	$16,   0($sp)
+	SDARG	$17,   8($sp)
+	SDARG	$18,  16($sp)
+	SDARG	$19,  24($sp)
+	SDARG	$20,  32($sp)
+	SDARG	$21,  40($sp)
+	sdc1	$f24, 48($sp)
+	sdc1	$f25, 56($sp)
+	sdc1	$f26, 64($sp)
+	sdc1	$f27, 72($sp)
+	sdc1	$f28, 80($sp)
+
+	SDARG	$22,  88($sp)
+	SDARG	$23,  96($sp)
+	SDARG	$24, 104($sp)
+	SDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	sdc1	$f20,112($sp)
+	sdc1	$f21,120($sp)
+	sdc1	$f22,128($sp)
+	sdc1	$f23,136($sp)
+#endif
+											#	LT compute from left to right, top to bottom
+	LDARG	OFFSET, 144($sp)					
+	dsll	LDC, LDC, BASE_SHIFT			#	ldc
+
+	dsra	J,  N, 2						#	j = nc/4
+	blez	J, .L30
+	nop
+
+.L10:										#	nr=4
+	daddiu	J, J, -1
+	move	CO1, C
+	daddu	CO2, C,   LDC
+	daddu	CO3, CO2, LDC
+	daddu	CO4, CO3, LDC
+
+	MTC	$0,  t11							#	clear result registers
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+
+	dsra	I,  M, 2						#	i = mc/4
+	move	KK, OFFSET						#	kk is the length of the rectangular data part of panel Ai
+	move	AO, A							#	reset A	
+	daddu	C,  CO4, LDC					#	fixed pointer C, the write back address
+	blez	I, .L20
+	nop
+
+
+.L11:										#	mr=4
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+	LD	b2,  1 * SIZE(B)
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)				
+
+	MOV	t13, t11							#	clear result registers
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	dsra	L,  KK, 2						#	L = kk/4
+	blez	L, .L15
+	move	BO,  B							#
+
+
+	.align	3
+.L12:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	4th compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t33, t33, a7, b7
+	MADD	t43, t43, a8, b7
+
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+	MADD	t34, t34, a7, b8
+	MADD	t44, t44, a8, b8			
+
+	daddiu	L, L, -1
+	bgtz	L, .L12
+	nop
+	
+
+	.align 3
+.L15:
+	andi	L, KK,  3						#	the remainder part: KK-KK/4
+	blez	L, .L18
+	nop
+
+	.align	3
+.L16:
+	MADD	t11, t11, a1, b1				
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t33, t33, a3, b3
+	MADD	t43, t43, a4, b3
+
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+	MADD	t34, t34, a3, b4
+	MADD	t44, t44, a4, b4		
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L16
+	nop
+
+
+.L18:										#	deal with the triangular data part of panel Ai
+	LD	b1,  0 * SIZE(BO)					#	triangular_part*X + rectangular_part = B
+	LD	b2,  1 * SIZE(BO)					#	triangular_part*X = B - rectangular_part
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+
+	LD	b5,  4 * SIZE(BO)					#	sb store in row major
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+	
+	SUB	t21, b5, t21
+	SUB	t22, b6, t22
+	SUB	t23, b7, t23
+	SUB	t24, b8, t24
+
+	LD	b1,  8 * SIZE(BO)
+	LD	b2,  9 * SIZE(BO)
+	LD	b3, 10 * SIZE(BO)
+	LD	b4, 11 * SIZE(BO)
+	
+	SUB	t31, b1, t31
+	SUB	t32, b2, t32
+	SUB	t33, b3, t33
+	SUB	t34, b4, t34
+	
+	LD	b5, 12 * SIZE(BO)
+	LD	b6, 13 * SIZE(BO)
+	LD	b7, 14 * SIZE(BO)
+	LD	b8, 15 * SIZE(BO)
+
+	SUB	t41, b5, t41
+	SUB	t42, b6, t42
+	SUB	t43, b7, t43
+	SUB	t44, b8, t44
+
+
+	LD	a1,  0 * SIZE(AO)					#	sa stores in col major		
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	MUL	t11, a1, t11
+	MUL	t12, a1, t12
+	MUL	t13, a1, t13
+	MUL	t14, a1, t14
+	NMSUB	t21, t21, a2, t11
+	NMSUB	t22, t22, a2, t12
+	NMSUB	t23, t23, a2, t13
+	NMSUB	t24, t24, a2, t14
+	NMSUB	t31, t31, a3, t11
+	NMSUB	t32, t32, a3, t12
+	NMSUB	t33, t33, a3, t13
+	NMSUB	t34, t34, a3, t14
+	NMSUB	t41, t41, a4, t11
+	NMSUB	t42, t42, a4, t12
+	NMSUB	t43, t43, a4, t13
+	NMSUB	t44, t44, a4, t14
+
+	
+	LD	a5,  5 * SIZE(AO)
+	LD	a6,  6 * SIZE(AO)
+	LD	a7,  7 * SIZE(AO)
+	MUL	t21, a5, t21
+	MUL	t22, a5, t22
+	MUL	t23, a5, t23
+	MUL	t24, a5, t24
+	NMSUB	t31, t31, a6, t21
+	NMSUB	t32, t32, a6, t22
+	NMSUB	t33, t33, a6, t23
+	NMSUB	t34, t34, a6, t24
+	NMSUB	t41, t41, a7, t21
+	NMSUB	t42, t42, a7, t22
+	NMSUB	t43, t43, a7, t23
+	NMSUB	t44, t44, a7, t24
+
+
+	LD	a8, 10 * SIZE(AO)
+	LD	a1, 11 * SIZE(AO)
+	MUL	t31, a8, t31
+	MUL	t32, a8, t32
+	MUL	t33, a8, t33
+	MUL	t34, a8, t34
+	NMSUB	t41, t41, a1, t31
+	NMSUB	t42, t42, a1, t32
+	NMSUB	t43, t43, a1, t33
+	NMSUB	t44, t44, a1, t34
+
+
+	LD	a2, 15 * SIZE(AO)
+	MUL	t41, a2, t41
+	MUL	t42, a2, t42
+	MUL	t43, a2, t43
+	MUL	t44, a2, t44
+
+	ST	t11,  0 * SIZE(BO)					#	update packed B
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+	ST	t21,  4 * SIZE(BO)
+	ST	t22,  5 * SIZE(BO)
+	ST	t23,  6 * SIZE(BO)
+	ST	t24,  7 * SIZE(BO)
+	ST	t31,  8 * SIZE(BO)
+	ST	t32,  9 * SIZE(BO)
+	ST	t33, 10 * SIZE(BO)
+	ST	t34, 11 * SIZE(BO)
+	ST	t41, 12 * SIZE(BO)
+	ST	t42, 13 * SIZE(BO)
+	ST	t43, 14 * SIZE(BO)
+	ST	t44, 15 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)					#	write back 
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t33,  2 * SIZE(CO3)
+	ST	t43,  3 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+	ST	t34,  2 * SIZE(CO4)
+	ST	t44,  3 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 4 * SIZE				#	fixed pointers
+	daddiu	CO2, CO2, 4 * SIZE
+	daddiu	CO3, CO3, 4 * SIZE
+	daddiu	CO4, CO4, 4 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L						#	mov AO to the end of panel Ai
+	daddu	BO, BO, TEMP					#	mov BO to the end of panel Bj
+
+	daddiu	KK, KK, 4						#	the length of rectangular data part increases by 4
+	daddiu	I, I, -1
+
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+	MOV	t12, a1
+	MOV	t22, a1
+	MOV	t32, a1
+	MOV	t42, a1
+	bgtz	I, .L11
+	nop
+
+
+	.align 3
+.L20:
+	andi	I,  M, 2						#	mr=2,nr=4	
+	blez	I, .L50
+	nop
+
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+	LD	b2,  1 * SIZE(B)
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)				
+
+	dsra	L,  KK, 2
+	blez	L, .L25
+	move	BO,  B
+
+
+	.align	3
+.L22:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	LD	a3,   4 * SIZE(AO)
+	LD	a4,   5 * SIZE(AO)
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t13, t13, a5, b7
+	MADD	t23, t23, a6, b7
+	MADD	t14, t14, a5, b8
+	MADD	t24, t24, a6, b8
+
+	LD	a7,   6 * SIZE(AO)
+	LD	a8,   7 * SIZE(AO)
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1				#	3rd compute
+	MADD	t21, t21, a4, b1
+	MADD	t12, t12, a3, b2
+	MADD	t22, t22, a4, b2
+	MADD	t13, t13, a3, b3
+	MADD	t23, t23, a4, b3
+	MADD	t14, t14, a3, b4
+	MADD	t24, t24, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5				#	4th compute
+	MADD	t21, t21, a8, b5
+	MADD	t12, t12, a7, b6
+	MADD	t22, t22, a8, b6
+	MADD	t13, t13, a7, b7
+	MADD	t23, t23, a8, b7
+	MADD	t14, t14, a7, b8
+	MADD	t24, t24, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L22
+	nop
+
+
+	.align 3
+.L25:
+	andi	L, KK,  3
+	blez	L, .L28
+	nop
+	
+	.align	3
+.L26:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t13, t13, a1, b3
+	MADD	t23, t23, a2, b3
+	MADD	t14, t14, a1, b4
+	MADD	t24, t24, a2, b4
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L26
+	nop
+
+
+.L28:										#	deal with the triangular part
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+	SUB	t21, b5, t21
+	SUB	t22, b6, t22
+	SUB	t23, b7, t23
+	SUB	t24, b8, t24
+
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  1 * SIZE(AO)
+	MUL	t11, b1, t11
+	MUL	t12, b1, t12
+	MUL	t13, b1, t13
+	MUL	t14, b1, t14
+	NMSUB	t21, t21, b2, t11
+	NMSUB	t22, t22, b2, t12
+	NMSUB	t23, t23, b2, t13
+	NMSUB	t24, t24, b2, t14
+	
+	LD	b3,  3 * SIZE(AO)
+	MUL	t21, b3, t21
+	MUL	t22, b3, t22
+	MUL	t23, b3, t23
+	MUL	t24, b3, t24
+
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+	ST	t21,  4 * SIZE(BO)
+	ST	t22,  5 * SIZE(BO)
+	ST	t23,  6 * SIZE(BO)
+	ST	t24,  7 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t23,  1 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+	ST	t24,  1 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 2 * SIZE
+	daddiu	CO2, CO2, 2 * SIZE
+	daddiu	CO3, CO3, 2 * SIZE
+	daddiu	CO4, CO4, 2 * SIZE
+
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L							#	mov AO to the end of Ai
+	daddu	BO, BO, TEMP						#	mov BO to the end of Bj
+
+	daddiu	KK, KK, 2							#	the length of rectangular data part increases by 2
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+	MOV	t12, a1
+	MOV	t22, a1
+	MOV	t32, a1
+	MOV	t42, a1
+
+
+	.align 3
+.L50:
+	andi	I,  M, 1						#	mr=1,nr=4	
+	blez	I, .L29
+	nop
+
+	MOV	t13, t11
+	MOV	t23, t11
+	MOV	t33, t11
+	MOV	t43, t11
+	MOV	t14, t11
+	MOV	t24, t11
+	MOV	t34, t11
+	MOV	t44, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+	LD	b2,  1 * SIZE(B)
+	LD	b3,  2 * SIZE(B)
+	LD	b4,  3 * SIZE(B)				
+
+	dsra	L,  KK, 2
+	blez	L, .L55
+	move	BO,  B
+
+
+	.align	3
+.L52:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	LD	a3,   2 * SIZE(AO)
+	LD	b1,   8 * SIZE(BO)
+	LD	b2,   9 * SIZE(BO)
+	LD	b3,  10 * SIZE(BO)
+	LD	b4,  11 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t12, t12, a5, b6
+	MADD	t13, t13, a5, b7
+	MADD	t14, t14, a5, b8
+
+	LD	a7,   3 * SIZE(AO)
+	LD	b5,  12 * SIZE(BO)
+	LD	b6,  13 * SIZE(BO)
+	LD	b7,  14 * SIZE(BO)
+	LD	b8,  15 * SIZE(BO)
+
+	MADD	t11, t11, a3, b1				#	3rd compute
+	MADD	t12, t12, a3, b2
+	MADD	t13, t13, a3, b3
+	MADD	t14, t14, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += mr*4kr	
+	daddiu	BO, BO, 16 * SIZE				#	BO += 4nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	MADD	t11, t11, a7, b5				#	4th compute
+	MADD	t12, t12, a7, b6
+	MADD	t13, t13, a7, b7
+	MADD	t14, t14, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L52
+	nop
+
+
+	.align 3
+.L55:
+	andi	L, KK,  3
+	blez	L, .L58
+	nop
+	
+	.align	3
+.L56:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t12, t12, a1, b2
+	MADD	t13, t13, a1, b3
+	MADD	t14, t14, a1, b4
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 4nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L56
+	nop
+
+
+.L58:										#	deal with the triangular part
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t13, b3, t13
+	SUB	t14, b4, t14
+
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	MUL	t11, b1, t11
+	MUL	t12, b1, t12
+	MUL	t13, b1, t13
+	MUL	t14, b1, t14
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t13,  2 * SIZE(BO)
+	ST	t14,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t13,  0 * SIZE(CO3)
+	ST	t14,  0 * SIZE(CO4)
+
+	daddiu	CO1, CO1, 1 * SIZE
+	daddiu	CO2, CO2, 1 * SIZE
+	daddiu	CO3, CO3, 1 * SIZE
+	daddiu	CO4, CO4, 1 * SIZE
+
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, BASE_SHIFT				#	mr=1
+	dsll	TEMP, TEMP, 2 + BASE_SHIFT
+	daddu	AO, AO, L							#	mov AO to the end of Ai
+	daddu	BO, BO, TEMP						#	mov BO to the end of Bj
+
+	daddiu	KK, KK, 1							#	the length of rectangular data part increases by 2
+
+	.align 3
+.L29:
+	move	B,  BO								#	fixed panel Bj
+	bgtz	J, .L10
+	nop
+	
+	
+	.align 3
+.L30:
+	andi	J,  N, 2							#	nr=2
+	blez	J, .L70
+	nop
+
+	move	CO1, C
+	daddu	CO2, C,   LDC
+
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	move	KK, OFFSET							
+	move	AO, A								#	reset A
+	daddu	C,  CO2, LDC						#	fixed 
+
+	dsra	I,  M, 2							#	I = mc/4
+	blez	I, .L40
+	nop
+
+.L31:
+	MOV	t12, t11
+	MOV	t22, t11
+	MOV	t32, t11
+	MOV	t42, t11
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+	LD	b2,  1 * SIZE(B)
+
+	dsra	L,  KK, 2						#	L=kk/4
+	blez	L, .L35
+	move	BO,  B							#	reset B
+
+
+	.align	3
+.L32:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+	LD	b3,   4 * SIZE(BO)
+	LD	b4,   5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+	MADD	t32, t32, a7, b6
+	MADD	t42, t42, a8, b6
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3				#	3rd compute
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+	MADD	t12, t12, a1, b4
+	MADD	t22, t22, a2, b4
+	MADD	t32, t32, a3, b4
+	MADD	t42, t42, a4, b4
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7				#	4th compute
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+	MADD	t12, t12, a5, b8
+	MADD	t22, t22, a6, b8
+	MADD	t32, t32, a7, b8
+	MADD	t42, t42, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L32
+	nop
+
+	
+	.align 3
+
+.L35:
+	andi	L, KK,  3
+	blez	L, .L38
+	nop
+	
+	.align	3
+.L36:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+	
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+	MADD	t32, t32, a3, b2
+	MADD	t42, t42, a4, b2
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L36
+	nop
+
+
+.L38:										#
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+	LD	b5,  4 * SIZE(BO)
+	LD	b6,  5 * SIZE(BO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t21, b3, t21
+	SUB	t22, b4, t22
+	SUB	t31, b5, t31
+	SUB	t32, b6, t32
+	SUB	t41, b7, t41
+	SUB	t42, b8, t42
+
+	LD	a1,  0 * SIZE(AO)					#	sa stores in col major		
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	MUL	t11, a1, t11
+	MUL	t12, a1, t12
+	NMSUB	t21, t21, a2, t11
+	NMSUB	t22, t22, a2, t12
+	NMSUB	t31, t31, a3, t11
+	NMSUB	t32, t32, a3, t12
+	NMSUB	t41, t41, a4, t11
+	NMSUB	t42, t42, a4, t12
+
+	
+	LD	a5,  5 * SIZE(AO)
+	LD	a6,  6 * SIZE(AO)
+	LD	a7,  7 * SIZE(AO)
+	MUL	t21, a5, t21
+	MUL	t22, a5, t22
+	NMSUB	t31, t31, a6, t21
+	NMSUB	t32, t32, a6, t22
+	NMSUB	t41, t41, a7, t21
+	NMSUB	t42, t42, a7, t22
+
+
+	LD	a8, 10 * SIZE(AO)
+	LD	a1, 11 * SIZE(AO)
+	MUL	t31, a8, t31
+	MUL	t32, a8, t32
+	NMSUB	t41, t41, a1, t31
+	NMSUB	t42, t42, a1, t32
+
+
+	LD	a2, 15 * SIZE(AO)
+	MUL	t41, a2, t41
+	MUL	t42, a2, t42
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t21,  2 * SIZE(BO)
+	ST	t22,  3 * SIZE(BO)
+	ST	t31,  4 * SIZE(BO)
+	ST	t32,  5 * SIZE(BO)
+	ST	t41,  6 * SIZE(BO)
+	ST	t42,  7 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+	ST	t32,  2 * SIZE(CO2)
+	ST	t42,  3 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 4 * SIZE
+	daddiu	CO2, CO2, 4 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AO, L						#	move AO to the end of Ai
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 4						#	
+
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+
+	daddiu	I, I, -1
+	bgtz	I, .L31
+	nop
+
+
+	.align 3
+.L40:
+	andi	I,  M, 2
+	blez	I, .L60
+	nop
+
+	MOV	t12, t11							#	clear result registers
+	MOV	t22, t21
+	MOV	t32, t31
+	MOV	t42, t41
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(B)
+	LD	b2,  1 * SIZE(B)
+
+	dsra	L,  KK, 2
+	blez	L, .L45
+	move	BO,  B							#	reset B	
+
+
+	.align	3
+.L42:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t12, t12, a5, b6
+	MADD	t22, t22, a6, b6
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t21, t21, a4, b3
+	MADD	t12, t12, a3, b4
+	MADD	t22, t22, a4, b4
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t21, t21, a8, b7
+	MADD	t12, t12, a7, b8
+	MADD	t22, t22, a8, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L42
+	nop
+	
+	.align 3
+
+.L45:
+	andi	L, KK,  3
+	blez	L, .L48
+	nop
+	
+	.align	3
+.L46:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t12, t12, a1, b2
+	MADD	t22, t22, a2, b2
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L46
+	nop
+
+.L48:
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+	SUB	t21, b3, t21
+	SUB	t22, b4, t22
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  1 * SIZE(AO)
+	MUL	t11, b1, t11
+	MUL	t12, b1, t12
+	NMSUB	t21, t21, b2, t11
+	NMSUB	t22, t22, b2, t12
+	
+	LD	b3,  3 * SIZE(AO)
+	MUL	t21, b3, t21
+	MUL	t22, b3, t22
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+	ST	t21,  2 * SIZE(BO)
+	ST	t22,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+	ST	t22,  1 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 2 * SIZE
+	daddiu	CO2, CO2, 2 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 2
+	MTC	$0,  a1
+	MOV	t11, a1
+	MOV	t21, a1
+	MOV	t31, a1
+	MOV	t41, a1
+
+
+	.align 3
+.L60:
+	andi	I,  M, 1						#	mr=1
+	blez	I, .L49
+	nop
+
+	MOV	t12, t11							#	clear result registers
+	MOV	t22, t21
+	MOV	t32, t31
+	MOV	t42, t41
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(B)
+	LD	b2,  1 * SIZE(B)
+
+	dsra	L,  KK, 2
+	blez	L, .L65
+	move	BO,  B							#	reset B	
+
+
+	.align	3
+.L62:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  2 * SIZE(BO)
+	LD	b6,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t12, t12, a1, b2
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  4 * SIZE(BO)
+	LD	b4,  5 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t12, t12, a5, b6
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  6 * SIZE(BO)
+	LD	b8,  7 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t12, t12, a3, b4
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += mr*4kr	
+	daddiu	BO, BO,  8 * SIZE				#	BO += 2nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t12, t12, a7, b8
+
+	daddiu	L, L, -1
+	bgtz	L, .L62
+	nop
+	
+	.align 3
+
+.L65:
+	andi	L, KK,  3
+	blez	L, .L68
+	nop
+	
+	.align	3
+.L66:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t12, t12, a1, b2
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 1mr	
+	daddiu	BO, BO,  2 * SIZE				#	BO += 2nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L66
+	nop
+
+.L68:
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t12, b2, t12
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	MUL	t11, b1, t11
+	MUL	t12, b1, t12
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t12,  1 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t12,  0 * SIZE(CO2)
+
+	daddiu	CO1, CO1, 1 * SIZE
+	daddiu	CO2, CO2, 1 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, BASE_SHIFT		#	mr=1
+	dsll	TEMP, TEMP, 1 + BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 1
+
+	.align 3
+.L49:
+	move	B,  BO
+	
+	.align 3
+
+.L70:
+	andi	J,  N, 1					#	nr=1
+	blez	J, .L999					#	END
+	nop
+
+	move	CO1, C
+
+	move	KK, OFFSET
+	move	AO, A
+
+	dsra	I,  M, 2
+	blez	I, .L80
+	nop
+
+.L71:
+	MTC	$0,  t11								#	clear result regusters
+	MOV	t21, t11
+	MOV	t31, t11
+	MOV	t41, t11
+
+	LD	a1,  0 * SIZE(AO)					#	this part compute the rectangular data part of Ai
+	LD	a2,  1 * SIZE(AO)					#	mr*KK with nr*KK
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)					#	get 4a
+
+	LD	b1,  0 * SIZE(B)					#	get 4b
+
+	dsra	L,  KK, 2
+	blez	L, .L75
+	move	BO,  B							#	reset B
+
+	.align	3
+.L72:
+	LD	a5,  4 * SIZE(AO)					
+	LD	a6,  5 * SIZE(AO)
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	LD	a1,   8 * SIZE(AO)
+	LD	a2,   9 * SIZE(AO)
+	LD	a3,  10 * SIZE(AO)
+	LD	a4,  11 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+	MADD	t31, t31, a7, b5
+	MADD	t41, t41, a8, b5
+
+	LD	a5,  12 * SIZE(AO)
+	LD	a6,  13 * SIZE(AO)
+	LD	a7,  14 * SIZE(AO)
+	LD	a8,  15 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a1, b3				#	3rd compute
+	MADD	t21, t21, a2, b3
+	MADD	t31, t31, a3, b3
+	MADD	t41, t41, a4, b3
+
+	daddiu	AO, AO, 16 * SIZE				#	AO += 4mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a5, b7				#	4th compute
+	MADD	t21, t21, a6, b7
+	MADD	t31, t31, a7, b7
+	MADD	t41, t41, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L72
+	nop
+	
+	.align 3
+
+.L75:
+	andi	L, KK,  3
+	blez	L, .L78
+	nop
+
+	.align	3
+.L76:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+	MADD	t31, t31, a3, b1
+	MADD	t41, t41, a4, b1
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 4mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L76
+	nop
+
+.L78:
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+	LD	b3,  2 * SIZE(BO)
+	LD	b4,  3 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+	SUB	t31, b3, t31
+	SUB	t41, b4, t41
+
+	LD	a1,  0 * SIZE(AO)					#	sa stores in col major		
+	LD	a2,  1 * SIZE(AO)
+	LD	a3,  2 * SIZE(AO)
+	LD	a4,  3 * SIZE(AO)
+	MUL	t11, a1, t11
+	NMSUB	t21, t21, a2, t11
+	NMSUB	t31, t31, a3, t11
+	NMSUB	t41, t41, a4, t11
+
+	LD	a5,  5 * SIZE(AO)
+	LD	a6,  6 * SIZE(AO)
+	LD	a7,  7 * SIZE(AO)
+	MUL	t21, a5, t21
+	NMSUB	t31, t31, a6, t21
+	NMSUB	t41, t41, a7, t21
+
+	LD	a8, 10 * SIZE(AO)
+	LD	a1, 11 * SIZE(AO)
+	MUL	t31, a8, t31
+	NMSUB	t41, t41, a1, t31
+
+	LD	a2, 15 * SIZE(AO)
+	MUL	t41, a2, t41
+
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t21,  1 * SIZE(BO)
+	ST	t31,  2 * SIZE(BO)
+	ST	t41,  3 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+	ST	t31,  2 * SIZE(CO1)
+	ST	t41,  3 * SIZE(CO1)
+
+	daddiu	CO1, CO1, 4 * SIZE
+
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 2 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 4
+	daddiu	I, I, -1
+	bgtz	I, .L71
+	nop
+
+
+	.align 3
+
+.L80:
+	andi	I,  M, 2
+	blez	I, .L90
+	NOP
+
+	MTC	$0,  t11
+	MOV	t21, t11							#	clear result registers
+
+	LD	a1,  0 * SIZE(AO)
+	LD	a2,  1 * SIZE(AO)
+	LD	b1,  0 * SIZE(B)
+
+	dsra	L,  KK, 2
+	blez	L, .L85
+	move	BO,  B
+	
+	.align	3
+.L82:
+	LD	a5,  2 * SIZE(AO)					
+	LD	a6,  3 * SIZE(AO)
+
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+	MADD	t21, t21, a2, b1
+
+	LD	a3,  4 * SIZE(AO)
+	LD	a4,  5 * SIZE(AO)
+
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+	MADD	t21, t21, a6, b5
+
+	LD	a7,  6 * SIZE(AO)
+	LD	a8,  7 * SIZE(AO)
+
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+	MADD	t21, t21, a4, b3
+
+	daddiu	AO, AO,  8 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+	MADD	t21, t21, a8, b7
+
+	daddiu	L, L, -1
+	bgtz	L, .L82
+	nop
+	
+	.align 3
+
+.L85:
+	andi	L, KK,  3
+	blez	L, .L88
+	nop
+
+	.align	3
+.L86:
+	MADD	t11, t11, a1, b1				#	3rd compute
+	MADD	t21, t21, a2, b1
+
+	daddiu	AO, AO,  2 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	a2,  1 * SIZE(AO)
+
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L86
+	nop
+
+
+.L88:
+	LD	b1,  0 * SIZE(BO)
+	LD	b2,  1 * SIZE(BO)
+
+	SUB	t11, b1, t11
+	SUB	t21, b2, t21
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	LD	b2,  1 * SIZE(AO)
+	MUL	t11, b1, t11
+	NMSUB	t21, t21, b2, t11
+	
+	LD	b3,  3 * SIZE(AO)
+	MUL	t21, b3, t21
+
+	ST	t11,  0 * SIZE(BO)
+	ST	t21,  1 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+	ST	t21,  1 * SIZE(CO1)
+
+
+	daddiu	CO1, CO1, 2 * SIZE
+	
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP, 1 + BASE_SHIFT
+	dsll	TEMP, TEMP, 0 + BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 2
+
+	
+	.align 3
+.L90:
+	andi	I,  M, 1					#	mr=1
+	blez	I, .L89
+	NOP
+
+	MTC	$0,  t11
+
+	LD	a1,  0 * SIZE(AO)
+	LD	b1,  0 * SIZE(B)
+
+	dsra	L,  KK, 2
+	blez	L, .L95
+	move	BO,  B
+	
+	.align	3
+.L92:
+	LD	a5,  1 * SIZE(AO)					
+	LD	b5,  1 * SIZE(BO)
+
+	MADD	t11, t11, a1, b1				#	1st compute
+
+	LD	a3,  2 * SIZE(AO)
+	LD	b3,  2 * SIZE(BO)
+
+	MADD	t11, t11, a5, b5				#	2ed compute
+
+	LD	a7,  3 * SIZE(AO)
+	LD	b7,  3 * SIZE(BO)
+
+	MADD	t11, t11, a3, b3				#	3rd compute
+
+	daddiu	AO, AO,  4 * SIZE				#	AO += 2mr*4kr	
+	daddiu	BO, BO,  4 * SIZE				#	BO += 1nr*4kr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+
+	MADD	t11, t11, a7, b7				#	4th compute
+
+	daddiu	L, L, -1
+	bgtz	L, .L92
+	nop
+	
+	.align 3
+.L95:
+	andi	L, KK,  3
+	blez	L, .L98
+	nop
+
+	.align	3
+.L96:
+	MADD	t11, t11, a1, b1				#	3rd compute
+
+	daddiu	AO, AO,  1 * SIZE				#	AO += 2mr	
+	daddiu	BO, BO,  1 * SIZE				#	BO += 1nr
+	
+	LD	a1,  0 * SIZE(AO)					#	next 
+	LD	b1,  0 * SIZE(BO)
+
+	daddiu	L, L, -1
+	bgtz	L, .L96
+	nop
+
+
+.L98:
+	LD	b1,  0 * SIZE(BO)
+
+	SUB	t11, b1, t11
+
+	LD	b1,  0 * SIZE(AO)				#	computes the triangular_part		
+	MUL	t11, b1, t11
+	
+	ST	t11,  0 * SIZE(BO)
+
+	ST	t11,  0 * SIZE(CO1)
+
+
+	daddiu	CO1, CO1, 1 * SIZE
+	
+	dsubu	TEMP, K, KK
+	dsll	L,    TEMP,  BASE_SHIFT
+	dsll	TEMP, TEMP,  BASE_SHIFT
+	daddu	AO, AO, L
+	daddu	BO, BO, TEMP
+
+	daddiu	KK, KK, 1
+
+
+	.align 3
+.L89:
+	move	B,  BO
+
+
+	.align 3
+
+.L999:
+	LDARG	$16,   0($sp)
+	LDARG	$17,   8($sp)
+	LDARG	$18,  16($sp)
+	LDARG	$19,  24($sp)
+	LDARG	$20,  32($sp)
+	LDARG	$21,  40($sp)
+	ldc1	$f24, 48($sp)
+	ldc1	$f25, 56($sp)
+	ldc1	$f26, 64($sp)
+	ldc1	$f27, 72($sp)
+	ldc1	$f28, 80($sp)
+
+	LDARG	$22,  88($sp)
+	LDARG	$23,  96($sp)
+	LDARG	$24, 104($sp)
+	LDARG	$25, 112($sp)
+
+#ifndef __64BIT__
+	ldc1	$f20,112($sp)
+	ldc1	$f21,120($sp)
+	ldc1	$f22,128($sp)
+	ldc1	$f23,136($sp)
+#endif
+
+	j	$31
+	daddiu	$sp, $sp, 144
+
+	EPILOGUE
-- 
2.7.4