trsm_kernel_ln_4x2_core2.s

Optimized GotoBLAS libraries

	movhpd	%xmm3,  1 * SIZE(CO1, LDC)
#else
	movsd	%xmm0,  0 * SIZE(CO1)
	movhpd	%xmm0,  1 * SIZE(CO1)
	movsd	%xmm1,  0 * SIZE(CO1, LDC)
	movhpd	%xmm1,  1 * SIZE(CO1, LDC)
#endif

#ifndef LN
	addl	$2 * SIZE, CO1
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm2,  -16 * SIZE(B)
	movapd	%xmm3,  -14 * SIZE(B)

	movddup	%xmm2, %xmm0
	movddup	%xmm3, %xmm1

	unpckhpd %xmm2, %xmm2
	unpckhpd %xmm3, %xmm3

	movapd	%xmm0,  -16 * SIZE(BB)
	movapd	%xmm2,  -14 * SIZE(BB)
	movapd	%xmm1,  -12 * SIZE(BB)
	movapd	%xmm3,  -10 * SIZE(BB)
#else
	movapd	%xmm0,  -16 * SIZE(AA)
	movapd	%xmm1,  -14 * SIZE(AA)
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
#ifdef LT
	addl	$4 * SIZE, B
#endif
#endif

#ifdef LN
	subl	$2, KK
	movl	BORIG, B
#endif

#ifdef LT
	addl	$2, KK
#endif

#ifdef RT
	movl	K, %eax
	movl	BORIG, B
	sall	$1 + BASE_SHIFT, %eax
	addl	%eax, AORIG
#endif
	ALIGN_2

.L50:
	movl	M,  %ebx
	sarl	$2, %ebx	# i = (m >> 2)
	jle	.L99
	ALIGN_4

.L10:
#ifdef LN
       movl	K, %eax
       sall	$2 + BASE_SHIFT, %eax
       subl	%eax, AORIG
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	AORIG, AA
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
#endif

	leal	16 * SIZE + BUFFER, BB

#if defined(LN) || defined(RT)
	movl	KK, %eax
	sall	$1 + BASE_SHIFT, %eax
	leal	(BB, %eax, 2), BB
#endif	

	movapd	-16 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movapd	-16 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movapd	 -8 * SIZE(AA), %xmm3
	pxor	%xmm6, %xmm6
#ifdef LN
	prefetcht2	-3 * SIZE(CO1)
	pxor	%xmm7, %xmm7
	prefetcht2	-3 * SIZE(CO1, LDC)
#else
	prefetcht2	 3 * SIZE(CO1)
	pxor	%xmm7, %xmm7
	prefetcht2	 3 * SIZE(CO1, LDC)
#endif

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L15
	ALIGN_4

.L12:
	movapd	%xmm1,  %xmm2
	mulpd	%xmm0,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	-14 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm0
	addpd	%xmm0,  %xmm5
	movapd	-14 * SIZE(AA), %xmm0
	mulpd	%xmm0,  %xmm2
	mulpd	%xmm0,  %xmm1
	movapd	-12 * SIZE(AA), %xmm0
	addpd	%xmm2,  %xmm6
	addpd	%xmm1,  %xmm7

	movapd	-12 * SIZE(BB), %xmm1
	movapd	%xmm1,  %xmm2
	mulpd	%xmm0,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	-10 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm0
	addpd	%xmm0,  %xmm5
	movapd	-10 * SIZE(AA), %xmm0
	mulpd	%xmm0,  %xmm2
	mulpd	%xmm0,  %xmm1
	movapd	  0 * SIZE(AA), %xmm0
	addpd	%xmm2,  %xmm6
	addpd	%xmm1,  %xmm7

	movapd	 -8 * SIZE(BB), %xmm1
	movapd	%xmm1,  %xmm2
	mulpd	%xmm3,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	 -6 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm3
	addpd	%xmm3,  %xmm5
	movapd	 -6 * SIZE(AA), %xmm3
	mulpd	%xmm3,  %xmm2
	mulpd	%xmm3,  %xmm1
	movapd	 -4 * SIZE(AA), %xmm3
	addpd	%xmm2,  %xmm6
	addpd	%xmm1,  %xmm7

	movapd	 -4 * SIZE(BB), %xmm1
	movapd	%xmm1,  %xmm2
	mulpd	%xmm3,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	 -2 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm3
	addpd	%xmm3,  %xmm5
	movapd	 -2 * SIZE(AA), %xmm3
	mulpd	%xmm3,  %xmm2
	mulpd	%xmm3,  %xmm1
	movapd	  8 * SIZE(AA), %xmm3
	addpd	%xmm2,  %xmm6
	addpd	%xmm1,  %xmm7
	movapd	  0 * SIZE(BB), %xmm1

	movapd	%xmm1,  %xmm2
	mulpd	%xmm0,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	  2 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm0
	addpd	%xmm0,  %xmm5
	movapd	  2 * SIZE(AA), %xmm0
	mulpd	%xmm0,  %xmm2
	mulpd	%xmm0,  %xmm1
	movapd	  4 * SIZE(AA), %xmm0
	addpd	%xmm2,  %xmm6
	addpd	%xmm1,  %xmm7

	movapd	  4 * SIZE(BB), %xmm1
	movapd	%xmm1,  %xmm2
	mulpd	%xmm0,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	  6 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm0
	addpd	%xmm0,  %xmm5
	movapd	  6 * SIZE(AA), %xmm0
	mulpd	%xmm0,  %xmm2
	mulpd	%xmm0,  %xmm1
	movapd	 16 * SIZE(AA), %xmm0
	addpd	%xmm2,  %xmm6
	addpd	%xmm1,  %xmm7

	movapd	  8 * SIZE(BB), %xmm1
	movapd	%xmm1,  %xmm2
	mulpd	%xmm3,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	 10 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm3
	addpd	%xmm3,  %xmm5
	movapd	 10 * SIZE(AA), %xmm3
	mulpd	%xmm3,  %xmm2
	mulpd	%xmm3,  %xmm1
	addpd	%xmm2,  %xmm6
	movapd	 12 * SIZE(AA), %xmm3
	addpd	%xmm1,  %xmm7

	movapd	 12 * SIZE(BB), %xmm1
	movapd	%xmm1,  %xmm2
	mulpd	%xmm3,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	 14 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm3
	addpd	%xmm3,  %xmm5
	movapd	 14 * SIZE(AA), %xmm3
	mulpd	%xmm3,  %xmm2
	mulpd	%xmm3,  %xmm1
	subl   $-32 * SIZE, BB
	movapd	 24 * SIZE(AA), %xmm3
	subl   $-32 * SIZE, AA
	addpd	%xmm2,  %xmm6
	addpd	%xmm1,  %xmm7
	movapd	-16 * SIZE(BB), %xmm1

	decl   %eax
	jne    .L12
	ALIGN_4

.L15:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L18
	ALIGN_4

.L16:
	movapd	%xmm1,  %xmm2
	mulpd	%xmm0,  %xmm1
	addpd	%xmm1,  %xmm4
	movapd	-14 * SIZE(BB), %xmm1
	mulpd	%xmm1,  %xmm0
	addpd	%xmm0,  %xmm5
	movapd	-14 * SIZE(AA), %xmm0
	mulpd	%xmm0,  %xmm2
	mulpd	%xmm0,  %xmm1
	movapd	-12 * SIZE(AA), %xmm0
	addpd	%xmm2,  %xmm6
	addpd	%xmm1,  %xmm7
	movapd	-12 * SIZE(BB), %xmm1

	addl	$4 * SIZE, AA
	addl	$4 * SIZE, BB
	decl	%eax
	jg	.L16
	ALIGN_4

.L18:
#if defined(LN) || defined(RT)
	movl	KK, %eax
#ifdef LN
	subl	$4, %eax
#else
	subl	$2, %eax
#endif

	movl	AORIG, AA
	movl	BORIG, B
	leal	16 * SIZE + BUFFER, BB

	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
	leal	(B,  %eax, 2), B
	leal	(BB, %eax, 4), BB
#endif

#if defined(LN) || defined(LT)
	movapd	 %xmm4, %xmm0
	unpcklpd %xmm5, %xmm4
	unpckhpd %xmm5, %xmm0

	movapd	 %xmm6, %xmm1
	unpcklpd %xmm7, %xmm6
	unpckhpd %xmm7, %xmm1

	movapd	-16 * SIZE(B), %xmm2
	movapd	-14 * SIZE(B), %xmm3
	movapd	-12 * SIZE(B), %xmm5
	movapd	-10 * SIZE(B), %xmm7

	subpd	%xmm4,  %xmm2
	subpd	%xmm0,  %xmm3
	subpd	%xmm6,  %xmm5
	subpd	%xmm1,  %xmm7
#else
	movapd	-16 * SIZE(AA), %xmm0
	movapd	-14 * SIZE(AA), %xmm1
	movapd	-12 * SIZE(AA), %xmm2
	movapd	-10 * SIZE(AA), %xmm3

	subpd	%xmm4, %xmm0
	subpd	%xmm6, %xmm1
	subpd	%xmm5, %xmm2
	subpd	%xmm7, %xmm3
#endif

#ifdef LN
	movddup	 -1 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm7
	movddup	 -2 * SIZE(AA), %xmm0
	mulpd	 %xmm7, %xmm0
	subpd	 %xmm0, %xmm5
	movddup	 -3 * SIZE(AA), %xmm0
	mulpd	 %xmm7, %xmm0
	subpd	 %xmm0, %xmm3
	movddup	 -4 * SIZE(AA), %xmm0
	mulpd	 %xmm7, %xmm0
	subpd	 %xmm0, %xmm2

	movddup	 -6 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm5
	movddup	 -7 * SIZE(AA), %xmm0
	mulpd	 %xmm5, %xmm0
	subpd	 %xmm0, %xmm3
	movddup	 -8 * SIZE(AA), %xmm0
	mulpd	 %xmm5, %xmm0
	subpd	 %xmm0, %xmm2

	movddup	-11 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm3
	movddup	-12 * SIZE(AA), %xmm0
	mulpd	 %xmm3, %xmm0
	subpd	 %xmm0, %xmm2

	movddup	-16 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm2
#endif

#ifdef LT
	movddup	-16 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm2
	movddup	-15 * SIZE(AA), %xmm0
	mulpd	 %xmm2, %xmm0
	subpd	 %xmm0, %xmm3
	movddup	-14 * SIZE(AA), %xmm0
	mulpd	 %xmm2, %xmm0
	subpd	 %xmm0, %xmm5
	movddup	-13 * SIZE(AA), %xmm0
	mulpd	 %xmm2, %xmm0
	subpd	 %xmm0, %xmm7

	movddup	-11 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm3
	movddup	-10 * SIZE(AA), %xmm0
	mulpd	 %xmm3, %xmm0
	subpd	 %xmm0, %xmm5
	movddup	 -9 * SIZE(AA), %xmm0
	mulpd	 %xmm3, %xmm0
	subpd	 %xmm0, %xmm7

	movddup	 -6 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm5
	movddup	 -5 * SIZE(AA), %xmm0
	mulpd	 %xmm5, %xmm0
	subpd	 %xmm0, %xmm7

	movddup	 -1 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm7
#endif

#ifdef RN
	movddup	-16 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm0
	mulpd	 %xmm4, %xmm1

	movddup	-15 * SIZE(B), %xmm4
	movapd	 %xmm4, %xmm5
	mulpd	 %xmm0, %xmm4
	subpd	 %xmm4, %xmm2
	mulpd	 %xmm1, %xmm5
	subpd	 %xmm5, %xmm3

	movddup	-13 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm2
	mulpd	 %xmm4, %xmm3
#endif

#ifdef RT
	movddup	-13 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm2
	mulpd	 %xmm4, %xmm3

	movddup	-14 * SIZE(B), %xmm4
	movapd	 %xmm4, %xmm5
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm0
	mulpd	 %xmm3, %xmm5
	subpd	 %xmm5, %xmm1

	movddup	-16 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm0
	mulpd	 %xmm4, %xmm1
#endif

#ifdef LN
	subl	$4 * SIZE, CO1
#endif

#if defined(LN) || defined(LT)
	movsd	%xmm2,  0 * SIZE(CO1)
	movsd	%xmm3,  1 * SIZE(CO1)
	movsd	%xmm5,  2 * SIZE(CO1)
	movsd	%xmm7,  3 * SIZE(CO1)

	movhpd	%xmm2,  0 * SIZE(CO1, LDC)
	movhpd	%xmm3,  1 * SIZE(CO1, LDC)
	movhpd	%xmm5,  2 * SIZE(CO1, LDC)
	movhpd	%xmm7,  3 * SIZE(CO1, LDC)
#else
	movsd	%xmm0,  0 * SIZE(CO1)
	movhpd	%xmm0,  1 * SIZE(CO1)
	movsd	%xmm1,  2 * SIZE(CO1)
	movhpd	%xmm1,  3 * SIZE(CO1)

	movsd	%xmm2,  0 * SIZE(CO1, LDC)
	movhpd	%xmm2,  1 * SIZE(CO1, LDC)
	movsd	%xmm3,  2 * SIZE(CO1, LDC)
	movhpd	%xmm3,  3 * SIZE(CO1, LDC)
#endif

#ifndef LN
	addl	$4 * SIZE, CO1
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm2,  -16 * SIZE(B)
	movapd	%xmm3,  -14 * SIZE(B)
	movapd	%xmm5,  -12 * SIZE(B)
	movapd	%xmm7,  -10 * SIZE(B)

	movddup	%xmm2, %xmm0
	movddup	%xmm3, %xmm1
	movddup	%xmm5, %xmm4
	movddup	%xmm7, %xmm6

	unpckhpd %xmm2, %xmm2
	unpckhpd %xmm3, %xmm3
	unpckhpd %xmm5, %xmm5
	unpckhpd %xmm7, %xmm7

	movapd	%xmm0,  -16 * SIZE(BB)
	movapd	%xmm2,  -14 * SIZE(BB)
	movapd	%xmm1,  -12 * SIZE(BB)
	movapd	%xmm3,  -10 * SIZE(BB)
	movapd	%xmm4,   -8 * SIZE(BB)
	movapd	%xmm5,   -6 * SIZE(BB)
	movapd	%xmm6,   -4 * SIZE(BB)
	movapd	%xmm7,   -2 * SIZE(BB)
#else
	movapd	%xmm0,  -16 * SIZE(AA)
	movapd	%xmm1,  -14 * SIZE(AA)
	movapd	%xmm2,  -12 * SIZE(AA)
	movapd	%xmm3,  -10 * SIZE(AA)
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
#ifdef LT
	addl	$8 * SIZE, B
#endif
#endif

#ifdef LN
	subl	$4, KK
	movl	BORIG, B
#endif

#ifdef LT
	addl	$4, KK
#endif

#ifdef RT
	movl	K, %eax
	movl	BORIG, B
	sall	$2 + BASE_SHIFT, %eax
	addl	%eax, AORIG
#endif

	decl	%ebx			# i --
	jg	.L10
	ALIGN_2	

.L99:
#ifdef LN
       movl	K, %eax
       leal	(, %eax, SIZE), %eax
       leal 	(B, %eax, 2), B
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(B,  %eax, 2), B
#endif

#ifdef RN
	addl	$2, KK
#endif

#ifdef RT
	subl	$2, KK
#endif

	decl	J			# j --
	jg	.L01
	ALIGN_2

.L100:
	movl	N, %eax
	testl	$1, %eax
	jle	.L999
	ALIGN_2
	
.L101:
#ifdef LN
	movl	OFFSET, %eax
	addl	M, %eax
	movl	%eax, KK
#endif	

	leal	16 * SIZE + BUFFER, BB

#ifdef RT
       movl	K, %eax
       sall	$0 + BASE_SHIFT, %eax
       subl	%eax, B
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	B, BORIG
	leal	(, %eax, SIZE), %eax
	leal	(B,  %eax, 1), B
	leal	(BB, %eax, 2), BB
#endif	

#ifdef LT
	movl	OFFSET, %eax
	movl	%eax, KK
#endif

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	jle	.L103
	ALIGN_4
	
.L102:
	movddup	 -16 * SIZE(B), %xmm0
	movddup	 -15 * SIZE(B), %xmm1
	movddup	 -14 * SIZE(B), %xmm2
	movddup	 -13 * SIZE(B), %xmm3
	movddup	 -12 * SIZE(B), %xmm4
	movddup	 -11 * SIZE(B), %xmm5
	movddup	 -10 * SIZE(B), %xmm6
	movddup	  -9 * SIZE(B), %xmm7

	movapd	%xmm0,  -16 * SIZE(BB)
	movapd	%xmm1,  -14 * SIZE(BB)
	movapd	%xmm2,  -12 * SIZE(BB)
	movapd	%xmm3,  -10 * SIZE(BB)
	movapd	%xmm4,   -8 * SIZE(BB)
	movapd	%xmm5,   -6 * SIZE(BB)
	movapd	%xmm6,   -4 * SIZE(BB)
	movapd	%xmm7,   -2 * SIZE(BB)

	addl	$ 8 * SIZE, B
	addl	$16 * SIZE, %ecx
	decl	%eax
	BRANCH
	jne	.L102
	ALIGN_2

.L103:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax
	BRANCH
	jle	.L105
	ALIGN_2

.L104:
	movddup	 -16 * SIZE(B), %xmm0

	movapd	%xmm0,  -16 * SIZE(BB)
	addl	$1 * SIZE, B
	addl	$2 * SIZE, BB
	decl	%eax
	jne	.L104
	ALIGN_4
	
.L105:
#if defined(LT) || defined(RN)
	movl	A, AA
#else
	movl	A, %eax
	movl	%eax, AORIG
#endif

#ifdef RT
	subl	LDC, C
#endif
	movl	C, CO1		# coffset = c
#ifndef RT
	addl	LDC, C
#endif

	movl	M,  %ebx
	testl	$1, %ebx
	jle	.L130

#ifdef LN
       movl	K, %eax
       sall	$0 + BASE_SHIFT, %eax
       subl	%eax, AORIG
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	AORIG, AA
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
#endif

	leal	16 * SIZE + BUFFER, BB

#if defined(LN) || defined(RT)
	movl	KK, %eax
	sall	$0 + BASE_SHIFT, %eax
	leal	(BB, %eax, 2), BB
#endif	

	movsd	-16 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movsd	-16 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movsd	 -8 * SIZE(BB), %xmm3
	movsd	-12 * SIZE(AA), %xmm2

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L152

.L151:	
	mulsd	%xmm0,  %xmm1
	movsd	-15 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm4
	movsd	-14 * SIZE(BB), %xmm1
	mulsd	%xmm0,  %xmm1
	movsd	-14 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm5
	movsd	-12 * SIZE(BB), %xmm1
	mulsd	%xmm0,  %xmm1
	movsd	-13 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm4
	movsd	-10 * SIZE(BB), %xmm1
	mulsd	%xmm0,  %xmm1
	movsd	 -8 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm5
	movsd	 -0 * SIZE(BB), %xmm1
	mulsd	%xmm2,  %xmm3
	movsd	-11 * SIZE(AA), %xmm2
	addsd	%xmm3, %xmm4