trsm_kernel_rt_2x4_sse2.s

Optimized GotoBLAS libraries

	pxor	%xmm7, %xmm7

	movlpd	 0 * SIZE(AA), %xmm0
	movlpd	 4 * SIZE(AA), %xmm1
	movlpd	 0 * SIZE(BB), %xmm2
	movlpd	 8 * SIZE(BB), %xmm3

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L85
	ALIGN_4

.L82:
	mulsd	%xmm0, %xmm2
	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
	movlpd	 1 * SIZE(AA), %xmm0
	mulsd	 2 * SIZE(BB), %xmm0
	addsd	%xmm2, %xmm4
	movlpd	16 * SIZE(BB), %xmm2
	addsd	%xmm0, %xmm5
	movlpd	 2 * SIZE(AA), %xmm0
	mulsd	 4 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm6
	movlpd	 3 * SIZE(AA), %xmm0
	mulsd	 6 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm7
	movlpd	 8 * SIZE(AA), %xmm0
	mulsd	%xmm1, %xmm3
	movlpd	 5 * SIZE(AA), %xmm1
	mulsd	10 * SIZE(BB), %xmm1
	addsd	%xmm3, %xmm4
	movlpd	24 * SIZE(BB), %xmm3
	addsd	%xmm1, %xmm5
	movlpd	 6 * SIZE(AA), %xmm1
	mulsd	12 * SIZE(BB), %xmm1
	addsd	%xmm1, %xmm6
	movlpd	 7 * SIZE(AA), %xmm1
	mulsd	14 * SIZE(BB), %xmm1
	addsd	%xmm1, %xmm7
	movlpd	12 * SIZE(AA), %xmm1

	addl   $ 8 * SIZE, AA
	addl   $16 * SIZE, BB
	decl   %eax
	jne    .L82
	ALIGN_4

.L85:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L88

.L86:
	mulsd	%xmm0, %xmm2
	addsd	%xmm2, %xmm4
	movlpd	 2 * SIZE(BB), %xmm2
	movlpd	 1 * SIZE(AA), %xmm0

	addl	$1 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L86
	ALIGN_4

.L88:
	addsd	%xmm5, %xmm4
	addsd	%xmm7, %xmm6
	addsd	%xmm6, %xmm4

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

	movl	AORIG, AA
	movl	BORIG, B
	leal	BUFFER, BB

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

#if defined(LN) || defined(LT)
	movlpd	 0 * SIZE(B), %xmm2
	subsd	%xmm4, %xmm2
#else
	movlpd	 0 * SIZE(AA), %xmm0
	subsd	%xmm4, %xmm0
#endif

#ifdef LN
	movlpd	 0 * SIZE(AA), %xmm4
	mulsd	 %xmm4, %xmm2
#endif

#ifdef LT
	movlpd	 0 * SIZE(AA), %xmm4
	mulsd	 %xmm4, %xmm2
#endif

#ifdef RN
	movlpd	 0 * SIZE(B), %xmm4
	mulsd	 %xmm4, %xmm0
#endif

#ifdef RT
	movlpd	 0 * SIZE(B), %xmm4
	mulsd	 %xmm4, %xmm0
#endif

#if defined(LN) || defined(LT)
	movlpd	%xmm2,   0 * SIZE(B)

	movlpd	%xmm2,   0 * SIZE(BB)
	movlpd	%xmm2,   1 * SIZE(BB)
#else
	movlpd	%xmm0,   0 * SIZE(AA)
#endif

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

#if defined(LN) || defined(LT)
	movlpd	%xmm2, 0 * SIZE(CO1)
#else
	movlpd	%xmm0, 0 * SIZE(CO1)
#endif

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

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(AA,%eax, SIZE), AA
#ifdef LT
	addl	$1 * SIZE, B
#endif
#endif

#ifdef LN
	subl	$1, KK
	movl	BORIG, B
#endif

#ifdef LT
	addl	$1, KK
#endif

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

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

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

#ifdef RN
	addl	$1, KK
#endif

#ifdef RT
	subl	$1, KK
#endif
	ALIGN_4


.L30:
	testl	$2, N
	je	.L60

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

	leal	BUFFER, BB

#ifdef RT
       movl	K, %eax
       sall	$1 + 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, 2), B
	leal	(BB, %eax, 4), 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	$2, %eax
	jle	.L35
	ALIGN_4
	
.L32:
#define COPYPREFETCH 40

	prefetchnta	(COPYPREFETCH) * SIZE(B)

	movq	 0 * SIZE(B), %mm0
	movq	 1 * SIZE(B), %mm1
	movq	 2 * SIZE(B), %mm2
	movq	 3 * SIZE(B), %mm3
	movq	 4 * SIZE(B), %mm4
	movq	 5 * SIZE(B), %mm5
	movq	 6 * SIZE(B), %mm6
	movq	 7 * SIZE(B), %mm7

	movq	%mm0,  0 * SIZE(BB)
	movq	%mm0,  1 * SIZE(BB)
	movq	%mm1,  2 * SIZE(BB)
	movq	%mm1,  3 * SIZE(BB)
	movq	%mm2,  4 * SIZE(BB)
	movq	%mm2,  5 * SIZE(BB)
	movq	%mm3,  6 * SIZE(BB)
	movq	%mm3,  7 * SIZE(BB)

	movq	%mm4,  8 * SIZE(BB)
	movq	%mm4,  9 * SIZE(BB)
	movq	%mm5, 10 * SIZE(BB)
	movq	%mm5, 11 * SIZE(BB)
	movq	%mm6, 12 * SIZE(BB)
	movq	%mm6, 13 * SIZE(BB)
	movq	%mm7, 14 * SIZE(BB)
	movq	%mm7, 15 * SIZE(BB)

	addl	$ 8 * SIZE, B
	addl	$16 * SIZE, BB
	decl	%eax
	jne	.L32
	ALIGN_2

.L35:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$3, %eax
	BRANCH
	jle	.L40
	ALIGN_2

.L36:
	movq	 0 * SIZE(B), %mm0
	movq	 1 * SIZE(B), %mm1

	movq	%mm0,  0 * SIZE(BB)
	movq	%mm0,  1 * SIZE(BB)
	movq	%mm1,  2 * SIZE(BB)
	movq	%mm1,  3 * SIZE(BB)

	addl	$2 * SIZE, B
	addl	$4 * SIZE, BB
	decl	%eax
	jne	.L36
	ALIGN_4
	
.L40:
#if defined(LT) || defined(RN)
	movl	A, AA
#else
	movl	A, %eax
	movl	%eax, AORIG
#endif

	leal	(, LDC, 2), %eax

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

	leal	BUFFER, BB

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

	pxor	%xmm4, %xmm4
	pxor	%xmm5, %xmm5
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7

	movapd	 0 * SIZE(AA), %xmm0
	movapd	 8 * SIZE(AA), %xmm1
	movapd	 0 * SIZE(BB), %xmm2
	movapd	 8 * SIZE(BB), %xmm3

#ifdef LN
	prefetchw	-2 * SIZE(CO1)
	prefetchw	-2 * SIZE(CO1, LDC)
#else
	prefetchw	 1 * SIZE(CO1)
	prefetchw	 1 * SIZE(CO1, LDC)
#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:
	mulpd	%xmm0, %xmm2
#if defined(OPTERON) || defined(BARCELONA)
	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
#endif
	mulpd	 2 * SIZE(BB), %xmm0
	addpd	%xmm2, %xmm4
	movapd	 4 * SIZE(BB), %xmm2
	addpd	%xmm0, %xmm5
	movapd	 2 * SIZE(AA), %xmm0

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

	mulpd	%xmm0, %xmm3
	mulpd	10 * SIZE(BB), %xmm0
	addpd	%xmm3, %xmm4
	movapd	12 * SIZE(BB), %xmm3
	addpd	%xmm0, %xmm5
	movapd	 6 * SIZE(AA), %xmm0

	mulpd	%xmm0, %xmm3
	mulpd	14 * SIZE(BB), %xmm0
	addpd	%xmm3, %xmm6
	movapd	24 * SIZE(BB), %xmm3
	addpd	%xmm0, %xmm7
	movapd	16 * SIZE(AA), %xmm0

#if defined(OPTERON) || defined(BARCELONA)
	prefetcht0 (PREFETCHSIZE  + 8) * SIZE(AA)
#endif
	mulpd	%xmm1, %xmm2
	mulpd	18 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm4
	movapd	20 * SIZE(BB), %xmm2
	addpd	%xmm1, %xmm5
	movapd	10 * SIZE(AA), %xmm1

	mulpd	%xmm1, %xmm2
	mulpd	22 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm6
	movapd	32 * SIZE(BB), %xmm2
	addpd	%xmm1, %xmm7
	movapd	12 * SIZE(AA), %xmm1

	mulpd	%xmm1, %xmm3
	mulpd	26 * SIZE(BB), %xmm1
	addpd	%xmm3, %xmm4
	movapd	28 * SIZE(BB), %xmm3
	addpd	%xmm1, %xmm5
	movapd	14 * SIZE(AA), %xmm1

	mulpd	%xmm1, %xmm3
	mulpd	30 * SIZE(BB), %xmm1
	addpd	%xmm3, %xmm6
	movapd	40 * SIZE(BB), %xmm3
	addpd	%xmm1, %xmm7
	movapd	24 * SIZE(AA), %xmm1

	addl   $16 * SIZE, AA
	addl   $32 * 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
	mulpd	 2 * SIZE(BB), %xmm0
	addpd	%xmm2, %xmm4
	movapd	 4 * SIZE(BB), %xmm2
	addpd	%xmm0, %xmm5
	movapd	 2 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$4 * 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
	movl	BORIG, B
	leal	BUFFER, BB

	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 2), 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	 0 * SIZE(B), %xmm2
	movapd	 2 * SIZE(B), %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
	movlpd	 3 * SIZE(AA), %xmm4
	movhpd	 3 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm3

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

	movlpd	 0 * SIZE(AA), %xmm4
	movhpd	 0 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm2

#endif

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

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

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

#ifdef RN
	movlpd	 0 * SIZE(B), %xmm4
	movhpd	 0 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm0
	movlpd	 1 * SIZE(B), %xmm4
	movhpd	 1 * SIZE(B), %xmm4
	mulpd	 %xmm0, %xmm4
	subpd	 %xmm4, %xmm1

	movlpd	 3 * SIZE(B), %xmm4
	movhpd	 3 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm1
#endif

#ifdef RT
	movlpd	 3 * SIZE(B), %xmm4
	movhpd	 3 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm1
	movlpd	 2 * SIZE(B), %xmm4
	movhpd	 2 * SIZE(B), %xmm4
	mulpd	 %xmm1, %xmm4
	subpd	 %xmm4, %xmm0

	movlpd	 0 * SIZE(B), %xmm4
	movhpd	 0 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm0
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm2,   0 * SIZE(B)
	movapd	%xmm3,   2 * SIZE(B)

	movlpd	%xmm2,   0 * SIZE(BB)
	movlpd	%xmm2,   1 * SIZE(BB)
	movhpd	%xmm2,   2 * SIZE(BB)
	movhpd	%xmm2,   3 * SIZE(BB)
	movlpd	%xmm3,   4 * SIZE(BB)
	movlpd	%xmm3,   5 * SIZE(BB)
	movhpd	%xmm3,   6 * SIZE(BB)
	movhpd	%xmm3,   7 * 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)
	movlpd	%xmm2, 0 * SIZE(CO1)
	movlpd	%xmm3, 1 * SIZE(CO1)
	movhpd	%xmm2, 0 * SIZE(CO1, LDC, 1)
	movhpd	%xmm3, 1 * SIZE(CO1, LDC, 1)
#else
	movlpd	%xmm0, 0 * SIZE(CO1)
	movhpd	%xmm0, 1 * SIZE(CO1)
	movlpd	%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
#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

	decl	%ebx			# i --
	jg	.L41
	ALIGN_4

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

#ifdef LN