trsm_kernel_lt_2x4_sse3.s

Optimized GotoBLAS libraries

	movapd	48 * SIZE(BB), %xmm2
	addpd	 %xmm1, %xmm7
	movddup	10 * SIZE(AA), %xmm1
	mulpd	 %xmm1, %xmm3
	mulpd	42 * SIZE(BB), %xmm1
	addpd	 %xmm3, %xmm4
	movapd	44 * SIZE(BB), %xmm3
	addpd	 %xmm1, %xmm5
	movddup	11 * SIZE(AA), %xmm1
	mulpd	 %xmm1, %xmm3
	mulpd	46 * SIZE(BB), %xmm1
	addpd	 %xmm3, %xmm6
	movapd	56 * SIZE(BB), %xmm3
	addpd	 %xmm1, %xmm7
	movddup	12 * SIZE(AA), %xmm1
	mulpd	 %xmm1, %xmm2
	mulpd	50 * SIZE(BB), %xmm1
	addpd	 %xmm2, %xmm4
	movapd	52 * SIZE(BB), %xmm2
	addpd	 %xmm1, %xmm5
	movddup	13 * SIZE(AA), %xmm1
	mulpd	 %xmm1, %xmm2
	mulpd	54 * SIZE(BB), %xmm1
	addpd	 %xmm2, %xmm6
	movapd	64 * SIZE(BB), %xmm2
	addpd	 %xmm1, %xmm7
	movddup	14 * SIZE(AA), %xmm1
	mulpd	 %xmm1, %xmm3
	mulpd	58 * SIZE(BB), %xmm1
	addpd	 %xmm3, %xmm4
	movapd	60 * SIZE(BB), %xmm3
	addpd	 %xmm1, %xmm5
	movddup	15 * SIZE(AA), %xmm1
	mulpd	 %xmm1, %xmm3
	mulpd	62 * SIZE(BB), %xmm1
	addpd	 %xmm3, %xmm6
	movapd	72 * SIZE(BB), %xmm3
	addpd	 %xmm1, %xmm7
	movddup	24 * SIZE(AA), %xmm1

	addl   $16 * SIZE, AA
	addl   $64 * SIZE, BB
	decl   %eax
	jne    .L22
	ALIGN_4

.L25:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$15, %eax		# if (k & 1)
	BRANCH
	je .L28

.L26:
	mulpd	 %xmm0, %xmm2
	mulpd	 2 * SIZE(BB), %xmm0
	addpd	 %xmm2, %xmm4
	movapd	 4 * SIZE(BB), %xmm2
	addpd	 %xmm0, %xmm5
	movddup	 1 * SIZE(AA), %xmm0

	addl	$1 * SIZE, AA
	addl	$4 * SIZE, BB

	decl	%eax
	jg	.L26
	ALIGN_4

.L28:
	addpd	%xmm6, %xmm4
	addpd	%xmm7, %xmm5

#if defined(LN) || defined(RT)
	movl	KK, %eax
#ifdef LN
	subl	$1, %eax
#else
	subl	$4, %eax
#endif

	movl	AORIG, AA

	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
	leal	(B,  %eax, 4), BB
#endif

#if defined(LN) || defined(LT)
	movapd	 0 * SIZE(BB), %xmm0
	movapd	 2 * SIZE(BB), %xmm1

	subpd	%xmm4,  %xmm0
	subpd	%xmm5,  %xmm1
#else
	movapd	 0 * SIZE(AA), %xmm1
	movapd	 2 * SIZE(AA), %xmm3

	subpd	%xmm4, %xmm1
	subpd	%xmm5, %xmm3

	movapd	       %xmm1, %xmm0
	unpckhpd       %xmm1, %xmm1
	movapd	       %xmm3, %xmm2
	unpckhpd       %xmm3, %xmm3
#endif

#ifdef LN
	movddup	 0 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm0
	mulpd	 %xmm4, %xmm1
#endif

#ifdef LT
	movddup	 0 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm0
	mulpd	 %xmm4, %xmm1
#endif

#ifdef RN
	movsd	 0 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm0
	movsd	 1 * SIZE(BB), %xmm4
	mulsd	 %xmm0, %xmm4
	subsd	 %xmm4, %xmm1
	movsd	 2 * SIZE(BB), %xmm4
	mulsd	 %xmm0, %xmm4
	subsd	 %xmm4, %xmm2
	movsd	 3 * SIZE(BB), %xmm4
	mulsd	 %xmm0, %xmm4
	subsd	 %xmm4, %xmm3

	movsd	 5 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm1
	movsd	 6 * SIZE(BB), %xmm4
	mulsd	 %xmm1, %xmm4
	subsd	 %xmm4, %xmm2
	movsd	 7 * SIZE(BB), %xmm4
	mulsd	 %xmm1, %xmm4
	subsd	 %xmm4, %xmm3

	movsd	10 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm2
	movsd	11 * SIZE(BB), %xmm4
	mulsd	 %xmm2, %xmm4
	subsd	 %xmm4, %xmm3

	movsd	15 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm3
#endif

#ifdef RT
	movsd	15 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm3
	movsd	14 * SIZE(BB), %xmm4
	mulsd	 %xmm3, %xmm4
	subsd	 %xmm4, %xmm2
	movsd	13 * SIZE(BB), %xmm4
	mulsd	 %xmm3, %xmm4
	subsd	 %xmm4, %xmm1
	movsd	12 * SIZE(BB), %xmm4
	mulsd	 %xmm3, %xmm4
	subsd	 %xmm4, %xmm0

	movsd	10 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm2
	movsd	 9 * SIZE(BB), %xmm4
	mulsd	 %xmm2, %xmm4
	subsd	 %xmm4, %xmm1
	movsd	 8 * SIZE(BB), %xmm4
	mulsd	 %xmm2, %xmm4
	subsd	 %xmm4, %xmm0

	movsd	 5 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm1
	movsd	 4 * SIZE(BB), %xmm4
	mulsd	 %xmm1, %xmm4
	subsd	 %xmm4, %xmm0

	movsd	 0 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm0
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm0,   0 * SIZE(BB)
	movapd	%xmm1,   2 * SIZE(BB)
#else
	movsd	%xmm0,   0 * SIZE(AA)
	movsd	%xmm1,   1 * SIZE(AA)
	movsd	%xmm2,   2 * SIZE(AA)
	movsd	%xmm3,   3 * SIZE(AA)
#endif

#ifdef LN
	subl	$1 * SIZE, CO1
#endif

	leal	(LDC, LDC, 2), %eax

#if defined(LN) || defined(LT)
	movsd	%xmm0, 0 * SIZE(CO1)
	movhpd	%xmm0, 0 * SIZE(CO1, LDC, 1)
	movsd	%xmm1, 0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm1, 0 * SIZE(CO1, %eax, 1)
#else
	movsd	%xmm0, 0 * SIZE(CO1)
	movsd	%xmm1, 0 * SIZE(CO1, LDC, 1)
	movsd	%xmm2, 0 * SIZE(CO1, LDC, 2)
	movsd	%xmm3, 0 * SIZE(CO1, %eax, 1)
#endif

#ifndef LN
	addl	$1 * SIZE, CO1
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 4), BB
#endif

#ifdef LN
	subl	$1, KK
#endif

#ifdef LT
	addl	$1, KK
#endif

#ifdef RT
	movl	K, %eax
	sall	$BASE_SHIFT, %eax
	addl	%eax, AORIG
#endif
	ALIGN_4

.L29:
#ifdef LN
       movl	K, %eax
       leal	(, %eax, SIZE), %eax
       leal 	(B, %eax, 4), B
#endif

#if defined(LT) || defined(RN)
	movl	BB, B
#endif

#ifdef RN
	addl	$4, KK
#endif

#ifdef RT
	subl	$4, KK
#endif

	decl	J			# j --
	jg	.L10
	ALIGN_4

.L30:
	testl	$2, N
	je	.L60

#if defined(LT) || defined(RN)
	movl	A, AA
#else
	movl	A, %eax
	movl	%eax, AORIG
#endif

#ifdef RT
	movl	K, %eax
	sall	$1 + BASE_SHIFT, %eax
	subl	%eax, B
#endif

	leal	(, LDC, 2), %eax

#ifdef RT
	subl	%eax, C
#endif
	movl	C, CO1
#ifndef RT
	addl	%eax, C
#endif

#ifdef LN
	movl	OFFSET, %eax
	addl	M, %eax
	movl	%eax, KK
#endif	

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

	movl	M,  %ebx
	sarl	$1, %ebx	# i = (m >> 2)
	jle	.L50
	ALIGN_4

.L41:
#ifdef LN
       movl	K, %eax
       sall	$1 + 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, 2), AA
#endif

	movl	B, BB

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

	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movapd	 8 * SIZE(AA), %xmm1
	pxor	%xmm5, %xmm5
	movddup	 0 * SIZE(BB), %xmm2
	pxor	%xmm6, %xmm6
	movddup	 8 * SIZE(BB), %xmm3
	pxor	%xmm7, %xmm7

#ifdef LN
	prefetchnta	-2 * SIZE(CO1)
	prefetchnta	-2 * SIZE(CO1, LDC, 1)
#else
	prefetchnta	 2 * SIZE(CO1)
	prefetchnta	 2 * SIZE(CO1, LDC, 1)
#endif

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L45
	ALIGN_4

.L42:
	PREFETCH  (PREFETCHSIZE + 0) * SIZE(AA)
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm4
	movddup	 1 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	movapd	 2 * SIZE(AA), %xmm0
	addpd	 %xmm2, %xmm5
	movddup	 2 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm6
	movddup	 3 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	movapd	 4 * SIZE(AA), %xmm0
	addpd	 %xmm2, %xmm7
	movddup	 4 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm4
	movddup	 5 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	movapd	 6 * SIZE(AA), %xmm0
	addpd	 %xmm2, %xmm5
	movddup	 6 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm6
	movddup	 7 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	movapd	16 * SIZE(AA), %xmm0
	addpd	 %xmm2, %xmm7
	movddup	16 * SIZE(BB), %xmm2
	mulpd	 %xmm1, %xmm3
	addpd	 %xmm3, %xmm4
	movddup	 9 * SIZE(BB), %xmm3
	mulpd	 %xmm1, %xmm3
	movapd	10 * SIZE(AA), %xmm1
	addpd	 %xmm3, %xmm5
	movddup	10 * SIZE(BB), %xmm3
	mulpd	 %xmm1, %xmm3
	addpd	 %xmm3, %xmm6
	movddup	11 * SIZE(BB), %xmm3
	mulpd	 %xmm1, %xmm3
	movapd	12 * SIZE(AA), %xmm1
	addpd	 %xmm3, %xmm7
	movddup	12 * SIZE(BB), %xmm3
	mulpd	 %xmm1, %xmm3
	addpd	 %xmm3, %xmm4
	movddup	13 * SIZE(BB), %xmm3
	mulpd	 %xmm1, %xmm3
	movapd	14 * SIZE(AA), %xmm1
	addpd	 %xmm3, %xmm5
	movddup	14 * SIZE(BB), %xmm3
	mulpd	 %xmm1, %xmm3
	addpd	 %xmm3, %xmm6
	movddup	15 * SIZE(BB), %xmm3
	mulpd	 %xmm1, %xmm3
	movapd	24 * SIZE(AA), %xmm1
	addpd	 %xmm3, %xmm7
	movddup	24 * SIZE(BB), %xmm3

	addl   $16 * SIZE, AA
	addl   $16 * SIZE, BB
	decl   %eax
	jne    .L42
	ALIGN_4

.L45:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L48
	ALIGN_3

.L46:
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm4
	movddup	 1 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	movapd	 2 * SIZE(AA), %xmm0
	addpd	 %xmm2, %xmm5
	movddup	 2 * SIZE(BB), %xmm2

	addl	$2 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L46
	ALIGN_4

.L48:
	addpd	%xmm6, %xmm4
	addpd	%xmm7, %xmm5

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

	movl	AORIG, AA

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

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

	movapd	 0 * SIZE(BB), %xmm2
	movapd	 2 * SIZE(BB), %xmm3

	subpd	%xmm4,  %xmm2
	subpd	%xmm0,  %xmm3
#else
	movapd	 0 * SIZE(AA), %xmm0
	movapd	 2 * SIZE(AA), %xmm1

	subpd	%xmm4, %xmm0
	subpd	%xmm5, %xmm1
#endif

#ifdef LN
	movddup	 3 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm3

	movddup	 2 * SIZE(AA), %xmm4
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm2

	movddup	 0 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm2

#endif

#ifdef LT
	movddup	 0 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm2

	movddup	 1 * SIZE(AA), %xmm4
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm3

	movddup	 3 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm3
#endif

#ifdef RN
	movddup	 0 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm0

	movddup	 1 * SIZE(BB), %xmm4
	mulpd	 %xmm0, %xmm4
	subpd	 %xmm4, %xmm1

	movddup	 3 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm1
#endif

#ifdef RT
	movddup	 3 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm1

	movddup	 2 * SIZE(BB), %xmm4
	mulpd	 %xmm1, %xmm4
	subpd	 %xmm4, %xmm0

	movddup	 0 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm0
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm2,   0 * SIZE(BB)
	movapd	%xmm3,   2 * SIZE(BB)
#else
	movapd	%xmm0,   0 * SIZE(AA)
	movapd	%xmm1,   2 * SIZE(AA)
#endif

#ifdef LN
	subl	$2 * SIZE, CO1
#endif

#if defined(LN) || defined(LT)
	movsd	%xmm2, 0 * SIZE(CO1)
	movsd	%xmm3, 1 * SIZE(CO1)
	movhpd	%xmm2, 0 * SIZE(CO1, LDC, 1)
	movhpd	%xmm3, 1 * SIZE(CO1, LDC, 1)
#else
	movsd	%xmm0, 0 * SIZE(CO1)
	movhpd	%xmm0, 1 * SIZE(CO1)
	movsd	%xmm1, 0 * SIZE(CO1, LDC, 1)
	movhpd	%xmm1, 1 * SIZE(CO1, LDC, 1)
#endif

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

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

#ifdef LN
	subl	$2, KK
#endif

#ifdef LT
	addl	$2, KK
#endif

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

	decl	%ebx			# i --
	jg	.L41
	ALIGN_4

.L50:
	movl	M,  %ebx
	testl	$1, %ebx	# i = (m >> 2)
	jle	.L59

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

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

	movl	B, BB

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

	movddup	 0 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movddup	 8 * SIZE(AA), %xmm1
	pxor	%xmm5, %xmm5
	movapd	 0 * SIZE(BB), %xmm2
	pxor	%xmm6, %xmm6
	movapd	 8 * SIZE(BB), %xmm3
	pxor	%xmm7, %xmm7

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$4, %eax
	je	.L55
	ALIGN_4

.L52:
	PREFETCH  (PREFETCHSIZE + 0) * SIZE(AA)
	mulpd	 %xmm0, %xmm2
	movddup	 1 * SIZE(AA), %xmm0
	addpd	 %xmm2, %xmm4
	mulpd	 2 * SIZE(BB), %xmm0
	movapd	16 * SIZE(BB), %xmm2
	addpd	 %xmm0, %xmm5
	movddup	 2 * SIZE(AA), %xmm0
	mulpd	 4 * SIZE(BB), %xmm0
	addpd	 %xmm0, %xmm6
	movddup	 3 * SIZE(AA), %xmm0
	mulpd	 6 * SIZE(BB), %xmm0
	addpd	 %xmm0, %xmm7
	movddup	 4 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm3
	movddup	 5 * SIZE(AA), %xmm0
	addpd	 %xmm3, %xmm4
	mulpd	10 * SIZE(BB), %xmm0
	movapd	24 * SIZE(BB), %xmm3
	addpd	 %xmm0, %xmm5
	movddup	 6 * SIZE(AA), %xmm0
	mulpd	12 * SIZE(BB), %xmm0
	addpd	 %xmm0, %xmm6
	movddup	 7 * SIZE(AA), %xmm0
	mulpd	14 * SIZE(BB), %xmm0
	addpd	 %xmm0, %xmm7
	movddup	16 * SIZE(AA), %xmm0
	mulpd	 %xmm1, %xmm2
	movddup	 9 * SIZE(AA), %xmm1
	addpd	 %xmm2, %xmm4
	mulpd	18 * SIZE(BB), %xmm1
	movapd	32 * SIZE(BB), %xmm2
	addpd	 %xmm1, %xmm5
	movddup	10 * SIZE(AA), %xmm1
	mulpd	20 * SIZE(BB), %xmm1
	addpd	 %xmm1, %xmm6
	movddup	11 * SIZE(AA), %xmm1
	mulpd	22 * SIZE(BB), %xmm1
	addpd	 %xmm1, %xmm7