trsm_kernel_ln_2x4_sse2.s

Optimized GotoBLAS libraries

	movlpd	15 * SIZE(B), %xmm4
	mulsd	 %xmm4, %xmm3
#endif

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

	movlpd	10 * SIZE(B), %xmm4
	mulsd	 %xmm4, %xmm2
	movlpd	 9 * SIZE(B), %xmm4
	mulsd	 %xmm2, %xmm4
	subsd	 %xmm4, %xmm1
	movlpd	 8 * SIZE(B), %xmm4
	mulsd	 %xmm2, %xmm4
	subsd	 %xmm4, %xmm0

	movlpd	 5 * SIZE(B), %xmm4
	mulsd	 %xmm4, %xmm1
	movlpd	 4 * SIZE(B), %xmm4
	mulsd	 %xmm1, %xmm4
	subsd	 %xmm4, %xmm0

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

#if defined(LN) || defined(LT)
	movapd	%xmm2,   0 * SIZE(B)
	movapd	%xmm5,   2 * SIZE(B)

	movlpd	%xmm2,   0 * SIZE(BB)
	movlpd	%xmm2,   1 * SIZE(BB)
	movhpd	%xmm2,   2 * SIZE(BB)
	movhpd	%xmm2,   3 * SIZE(BB)
	movlpd	%xmm5,   4 * SIZE(BB)
	movlpd	%xmm5,   5 * SIZE(BB)
	movhpd	%xmm5,   6 * SIZE(BB)
	movhpd	%xmm5,   7 * SIZE(BB)
#else
	movlpd	%xmm0,   0 * SIZE(AA)
	movlpd	%xmm1,   1 * SIZE(AA)
	movlpd	%xmm2,   2 * SIZE(AA)
	movlpd	%xmm3,   3 * SIZE(AA)
#endif

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

	leal	(LDC, LDC, 2), %eax

#if defined(LN) || defined(LT)
	movlpd	%xmm2, 0 * SIZE(CO1)
	movhpd	%xmm2, 0 * SIZE(CO1, LDC, 1)
	movlpd	%xmm5, 0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm5, 0 * SIZE(CO1, %eax, 1)
#else
	movlpd	%xmm0, 0 * SIZE(CO1)
	movlpd	%xmm1, 0 * SIZE(CO1, LDC, 1)
	movlpd	%xmm2, 0 * SIZE(CO1, LDC, 2)
	movlpd	%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	(AA,%eax, SIZE), AA
#ifdef LT
	addl	$4 * 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

.L20:
	movl	M,  %ebx
	sarl	$1, %ebx	# i = (m >> 2)
	jle	.L29
	ALIGN_4

.L11:
#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	$3 + 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

	leal	(LDC, LDC, 2), %eax

#ifdef LN
	prefetchw	-2 * SIZE(CO1)
	prefetchw	-2 * SIZE(CO1, LDC)
	prefetchw	-2 * SIZE(CO1, LDC, 2)
	prefetchw	-2 * SIZE(CO1, %eax)
#else
	prefetchw	 1 * SIZE(CO1)
	prefetchw	 1 * SIZE(CO1, LDC)
	prefetchw	 1 * SIZE(CO1, LDC, 2)
	prefetchw	 1 * SIZE(CO1, %eax)
#endif

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

#if 1
	andl	$-8, %eax
	sall	$4, %eax
	je	.L15
.L1X:	
	KERNEL1(16  *  0)
	KERNEL2(16  *  0)
	KERNEL3(16  *  0)
	KERNEL4(16  *  0)
	KERNEL5(16  *  0)
	KERNEL6(16  *  0)
	KERNEL7(16  *  0)
	KERNEL8(16  *  0)
	cmpl	$128 *  1, %eax
	jle	.L12
	KERNEL1(16  *  1)
	KERNEL2(16  *  1)
	KERNEL3(16  *  1)
	KERNEL4(16  *  1)
	KERNEL5(16  *  1)
	KERNEL6(16  *  1)
	KERNEL7(16  *  1)
	KERNEL8(16  *  1)
	cmpl	$128 *  2, %eax
	jle	.L12
	KERNEL1(16  *  2)
	KERNEL2(16  *  2)
	KERNEL3(16  *  2)
	KERNEL4(16  *  2)
	KERNEL5(16  *  2)
	KERNEL6(16  *  2)
	KERNEL7(16  *  2)
	KERNEL8(16  *  2)
	cmpl	$128 *  3, %eax
	jle	.L12
	KERNEL1(16  *  3)
	KERNEL2(16  *  3)
	KERNEL3(16  *  3)
	KERNEL4(16  *  3)
	KERNEL5(16  *  3)
	KERNEL6(16  *  3)
	KERNEL7(16  *  3)
	KERNEL8(16  *  3)
	cmpl	$128 *  4, %eax
	jle	.L12
	KERNEL1(16  *  4)
	KERNEL2(16  *  4)
	KERNEL3(16  *  4)
	KERNEL4(16  *  4)
	KERNEL5(16  *  4)
	KERNEL6(16  *  4)
	KERNEL7(16  *  4)
	KERNEL8(16  *  4)
	cmpl	$128 *  5, %eax
	jle	.L12
	KERNEL1(16  *  5)
	KERNEL2(16  *  5)
	KERNEL3(16  *  5)
	KERNEL4(16  *  5)
	KERNEL5(16  *  5)
	KERNEL6(16  *  5)
	KERNEL7(16  *  5)
	KERNEL8(16  *  5)
	cmpl	$128 *  6, %eax
	jle	.L12
	KERNEL1(16  *  6)
	KERNEL2(16  *  6)
	KERNEL3(16  *  6)
	KERNEL4(16  *  6)
	KERNEL5(16  *  6)
	KERNEL6(16  *  6)
	KERNEL7(16  *  6)
	KERNEL8(16  *  6)
	cmpl	$128 *  7, %eax
	jle	.L12
	KERNEL1(16  *  7)
	KERNEL2(16  *  7)
	KERNEL3(16  *  7)
	KERNEL4(16  *  7)
	KERNEL5(16  *  7)
	KERNEL6(16  *  7)
	KERNEL7(16  *  7)
	KERNEL8(16  *  7)

	addl	$128 * 4  * SIZE, BB
	addl	$128 * 1  * SIZE, AA
	subl	$128 * 8, %eax
	jg	.L1X
	jmp	.L15

.L12:
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 4), BB
	ALIGN_4
#else

	sarl	$3, %eax
	je	.L15
	ALIGN_4

.L12:
	KERNEL1(16  *  0)
	KERNEL2(16  *  0)
	KERNEL3(16  *  0)
	KERNEL4(16  *  0)
	KERNEL5(16  *  0)
	KERNEL6(16  *  0)
	KERNEL7(16  *  0)
	KERNEL8(16  *  0)

	addl   $64 * SIZE, BB
	addl   $16 * SIZE, AA
	decl   %eax
	jne    .L12
	ALIGN_4
#endif

.L15:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L18
	ALIGN_3

.L16:
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm4
	movapd	 2 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm5
	movapd	 4 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	mulpd	 6 * SIZE(BB), %xmm0
	addpd	 %xmm2, %xmm6
	movapd	 8 * SIZE(BB), %xmm2
	addpd	 %xmm0, %xmm7
	movapd	 2 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L16
	ALIGN_4

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

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

	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
	leal	(B,  %eax, 4), B
	leal	(BB, %eax, 8), 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	 0 * SIZE(B), %xmm2
	movapd	 2 * SIZE(B), %xmm5
	movapd	 4 * SIZE(B), %xmm3
	movapd	 6 * SIZE(B), %xmm7

	subpd	%xmm4,  %xmm2
	subpd	%xmm6,  %xmm5
	subpd	%xmm0,  %xmm3
	subpd	%xmm1,  %xmm7
#else
	movapd	 0 * SIZE(AA), %xmm0
	movapd	 2 * SIZE(AA), %xmm1
	movapd	 4 * SIZE(AA), %xmm2
	movapd	 6 * SIZE(AA), %xmm3

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

#ifdef LN
	movlpd	 3 * SIZE(AA), %xmm4
	movhpd	 3 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm3
	mulpd	 %xmm4, %xmm7

	movlpd	 2 * SIZE(AA), %xmm4
	movhpd	 2 * SIZE(AA), %xmm4
	movapd	 %xmm4, %xmm6
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm2
	mulpd	 %xmm7, %xmm6
	subpd	 %xmm6, %xmm5

	movlpd	 0 * SIZE(AA), %xmm4
	movhpd	 0 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm2
	mulpd	 %xmm4, %xmm5

#endif

#ifdef LT
	movlpd	 0 * SIZE(AA), %xmm4
	movhpd	 0 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm2
	mulpd	 %xmm4, %xmm5

	movlpd	 1 * SIZE(AA), %xmm4
	movhpd	 1 * SIZE(AA), %xmm4
	movapd	 %xmm4, %xmm6
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm3
	mulpd	 %xmm5, %xmm6
	subpd	 %xmm6, %xmm7

	movlpd	 3 * SIZE(AA), %xmm4
	movhpd	 3 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm3
	mulpd	 %xmm4, %xmm7
#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	 2 * SIZE(B), %xmm4
	movhpd	 2 * SIZE(B), %xmm4
	mulpd	 %xmm0, %xmm4
	subpd	 %xmm4, %xmm2
	movlpd	 3 * SIZE(B), %xmm4
	movhpd	 3 * SIZE(B), %xmm4
	mulpd	 %xmm0, %xmm4
	subpd	 %xmm4, %xmm3

	movlpd	 5 * SIZE(B), %xmm4
	movhpd	 5 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm1
	movlpd	 6 * SIZE(B), %xmm4
	movhpd	 6 * SIZE(B), %xmm4
	mulpd	 %xmm1, %xmm4
	subpd	 %xmm4, %xmm2
	movlpd	 7 * SIZE(B), %xmm4
	movhpd	 7 * SIZE(B), %xmm4
	mulpd	 %xmm1, %xmm4
	subpd	 %xmm4, %xmm3

	movlpd	10 * SIZE(B), %xmm4
	movhpd	10 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm2
	movlpd	11 * SIZE(B), %xmm4
	movhpd	11 * SIZE(B), %xmm4
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm3

	movlpd	15 * SIZE(B), %xmm4
	movhpd	15 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm3
#endif

#ifdef RT
	movlpd	15 * SIZE(B), %xmm4
	movhpd	15 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm3
	movlpd	14 * SIZE(B), %xmm4
	movhpd	14 * SIZE(B), %xmm4
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm2
	movlpd	13 * SIZE(B), %xmm4
	movhpd	13 * SIZE(B), %xmm4
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm1
	movlpd	12 * SIZE(B), %xmm4
	movhpd	12 * SIZE(B), %xmm4
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm0

	movlpd	10 * SIZE(B), %xmm4
	movhpd	10 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm2
	movlpd	 9 * SIZE(B), %xmm4
	movhpd	 9 * SIZE(B), %xmm4
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm1
	movlpd	 8 * SIZE(B), %xmm4
	movhpd	 8 * SIZE(B), %xmm4
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm0

	movlpd	 5 * SIZE(B), %xmm4
	movhpd	 5 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm1
	movlpd	 4 * SIZE(B), %xmm4
	movhpd	 4 * 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	%xmm5,   2 * SIZE(B)
	movapd	%xmm3,   4 * SIZE(B)
	movapd	%xmm7,   6 * SIZE(B)

	movlpd	%xmm2,   0 * SIZE(BB)
	movlpd	%xmm2,   1 * SIZE(BB)
	movhpd	%xmm2,   2 * SIZE(BB)
	movhpd	%xmm2,   3 * SIZE(BB)
	movlpd	%xmm5,   4 * SIZE(BB)
	movlpd	%xmm5,   5 * SIZE(BB)
	movhpd	%xmm5,   6 * SIZE(BB)
	movhpd	%xmm5,   7 * SIZE(BB)
	movlpd	%xmm3,   8 * SIZE(BB)
	movlpd	%xmm3,   9 * SIZE(BB)
	movhpd	%xmm3,  10 * SIZE(BB)
	movhpd	%xmm3,  11 * SIZE(BB)
	movlpd	%xmm7,  12 * SIZE(BB)
	movlpd	%xmm7,  13 * SIZE(BB)
	movhpd	%xmm7,  14 * SIZE(BB)
	movhpd	%xmm7,  15 * SIZE(BB)
#else
	movapd	%xmm0,   0 * SIZE(AA)
	movapd	%xmm1,   2 * SIZE(AA)
	movapd	%xmm2,   4 * SIZE(AA)
	movapd	%xmm3,   6 * SIZE(AA)
#endif

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

	leal	(LDC, LDC, 2), %eax

#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)
	movlpd	%xmm5, 0 * SIZE(CO1, LDC, 2)
	movlpd	%xmm7, 1 * SIZE(CO1, LDC, 2)
	movhpd	%xmm5, 0 * SIZE(CO1, %eax, 1)
	movhpd	%xmm7, 1 * SIZE(CO1, %eax, 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)
	movlpd	%xmm2, 0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm2, 1 * SIZE(CO1, LDC, 2)
	movlpd	%xmm3, 0 * SIZE(CO1, %eax, 1)
	movhpd	%xmm3, 1 * SIZE(CO1, %eax, 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	$8 * 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	.L11
	ALIGN_4

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

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

#ifdef RN
	addl	$4, KK
#endif

#ifdef RT
	subl	$4, KK
#endif

	decl	J			# j --
	jg	.L01
	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