trsm_kernel_ln_4x4_penryn.s

Optimized GotoBLAS libraries

#else
	subl	$2, %eax
#endif

	movl	AORIG, AA

	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
	leal	(B,  %eax, 2), BB
#endif

	addps	%xmm5, %xmm4

	pshufd	$0x55, %xmm4, %xmm5
	pshufd	$0x00, %xmm4, %xmm4

#if defined(LN) || defined(LT)
	unpcklps %xmm5, %xmm4

	movsd	-32 * SIZE(BB), %xmm1

	subps	%xmm4,  %xmm1
#else
	movss	-32 * SIZE(AA), %xmm0
	movss	-31 * SIZE(AA), %xmm1

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

#if defined(LN) || defined(LT)
	movss	 -32 * SIZE(AA), %xmm4
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm6, %xmm1
#endif

#ifdef RN
	movaps	-32 * SIZE(BB), %xmm6
	pshufd	 $0x00, %xmm6, %xmm7
	mulss	 %xmm7, %xmm0
	pshufd	 $0x55, %xmm6, %xmm7
	mulss	 %xmm0, %xmm7
	subss	 %xmm7, %xmm1

	pshufd	 $0xff, %xmm6, %xmm7
	mulss	 %xmm7, %xmm1
#endif

#ifdef RT
	movaps	-32 * SIZE(BB), %xmm6
	pshufd	 $0xff, %xmm6, %xmm7
	mulss	 %xmm7, %xmm1
	pshufd	 $0xaa, %xmm6, %xmm7
	mulss	 %xmm1, %xmm7
	subss	 %xmm7, %xmm0

	pshufd	 $0x00, %xmm6, %xmm7
	mulss	 %xmm7, %xmm0
#endif

#if defined(LN) || defined(LT)
	movlps	%xmm1, -32 * SIZE(BB)
#else
	movss	%xmm0, -32 * SIZE(AA)
	movss	%xmm1, -31 * SIZE(AA)
#endif

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

#if defined(LN) || defined(LT)
	pshufd	$1, %xmm1, %xmm3

	movss	%xmm1, 0 * SIZE(CO1)
	movss	%xmm3, 0 * SIZE(CO1, LDC)
#else
	movss	%xmm0, 0 * SIZE(CO1)
	movss	%xmm1, 0 * SIZE(CO1, LDC)
#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, 2), 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

.L60:
	testl	$2, M
	je	.L70

#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	

	movaps	-32 * SIZE(AA), %xmm0
	pxor	%xmm3, %xmm3
	movaps	-32 * SIZE(BB), %xmm1
	pxor	%xmm4, %xmm4
	pxor	%xmm5, %xmm5

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L65
	ALIGN_4

.L62:
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AA)

	pshufd	$0x44, %xmm0, %xmm2
	addps	%xmm3, %xmm4
	pshufd	$0x50, %xmm1, %xmm3
	mulps	%xmm2, %xmm3

	pshufd	$0xee, %xmm0, %xmm2
	movaps	-28 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	pshufd	$0xfa, %xmm1, %xmm3
	movaps	-28 * SIZE(BB), %xmm1
	mulps	%xmm2, %xmm3

	pshufd	$0x44, %xmm0, %xmm2
	addps	%xmm3, %xmm4
	pshufd	$0x50, %xmm1, %xmm3
	mulps	%xmm2, %xmm3

	pshufd	$0xee, %xmm0, %xmm2
	movaps	-24 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	pshufd	$0xfa, %xmm1, %xmm3
	movaps	-24 * SIZE(BB), %xmm1
	mulps	%xmm2, %xmm3

	pshufd	$0x44, %xmm0, %xmm2
	addps	%xmm3, %xmm4
	pshufd	$0x50, %xmm1, %xmm3
	mulps	%xmm2, %xmm3

	pshufd	$0xee, %xmm0, %xmm2
	movaps	-20 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	pshufd	$0xfa, %xmm1, %xmm3
	movaps	-20 * SIZE(BB), %xmm1
	mulps	%xmm2, %xmm3

	pshufd	$0x44, %xmm0, %xmm2
	addps	%xmm3, %xmm4
	pshufd	$0x50, %xmm1, %xmm3
	mulps	%xmm2, %xmm3

	pshufd	$0xee, %xmm0, %xmm2
	movaps	-16 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	pshufd	$0xfa, %xmm1, %xmm3
	movaps	-16 * SIZE(BB), %xmm1
	mulps	%xmm2, %xmm3

	subl   $-16 * SIZE, AA
	subl   $-16 * SIZE, BB

	subl   $1, %eax
	jne    .L62
	ALIGN_4

.L65:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L68
	ALIGN_4

.L66:
	pshufd	$0x44, %xmm0, %xmm2
	movsd	-30 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm4
	pshufd	$0x50, %xmm1, %xmm3
	movsd	-30 * SIZE(BB), %xmm1
	mulps	%xmm2, %xmm3

	addl	$2 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L66
	ALIGN_4

.L68:
#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

	addps	%xmm3, %xmm4
	addps	%xmm5, %xmm4

	movhlps	%xmm4, %xmm5

#if defined(LN) || defined(LT)
	unpcklps %xmm6, %xmm4
	unpcklps %xmm7, %xmm5

	movaps	 %xmm4, %xmm6
	unpcklps %xmm5, %xmm4
	unpckhps %xmm5, %xmm6

	movsd	-32 * SIZE(BB), %xmm1
	movsd	-30 * SIZE(BB), %xmm3

	subps	%xmm4,  %xmm1
	subps	%xmm6,  %xmm3
#else
	movsd	-32 * SIZE(AA), %xmm0
	movsd	-30 * SIZE(AA), %xmm1

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

#ifdef LN
	movaps	-32 * SIZE(AA), %xmm4
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm6, %xmm3
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm3, %xmm6
	subps	 %xmm6, %xmm1

	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm6, %xmm1
#endif

#ifdef LT
	movaps	-32 * SIZE(AA), %xmm4
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm6, %xmm1
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm1, %xmm6
	subps	 %xmm6, %xmm3

	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm6, %xmm3
#endif

#ifdef RN
	movaps	-32 * SIZE(BB), %xmm6
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm7, %xmm0
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm0, %xmm7
	subps	 %xmm7, %xmm1

	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm7, %xmm1
#endif

#ifdef RT
	movaps	-32 * SIZE(BB), %xmm6
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm7, %xmm1
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm1, %xmm7
	subps	 %xmm7, %xmm0

	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm7, %xmm0
#endif

#if defined(LN) || defined(LT)
	movlps	%xmm1, -32 * SIZE(BB)
	movlps	%xmm3, -30 * SIZE(BB)
#else
	movlps	%xmm0, -32 * SIZE(AA)
	movlps	%xmm1, -30 * SIZE(AA)
#endif

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

#if defined(LN) || defined(LT)
	unpcklps %xmm3, %xmm1

	movlps	%xmm1, 0 * SIZE(CO1)
	movhps	%xmm1, 0 * SIZE(CO1, LDC)
#else
	movlps	%xmm0, 0 * SIZE(CO1)
	movlps	%xmm1, 0 * SIZE(CO1, LDC)
#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
	ALIGN_4

.L70:
	movl	M,  %ebx
	sarl	$2, %ebx	# i = (m >> 2)
	jle	.L79
	ALIGN_4

.L51:
#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

	movl	B, BB

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

	movaps	-32 * SIZE(AA), %xmm0
	pxor	%xmm2, %xmm2
	movaps	-32 * SIZE(BB), %xmm1
	pxor	%xmm3, %xmm3

#ifdef LN
	pxor	%xmm4, %xmm4
	prefetcht0     -4 * SIZE(CO1)
	pxor	%xmm5, %xmm5
	prefetcht0     -4 * SIZE(CO1, LDC)
#else
	pxor	%xmm4, %xmm4
	prefetcht0	3 * SIZE(CO1)
	pxor	%xmm5, %xmm5
	prefetcht0	3 * SIZE(CO1, LDC)
#endif
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L55
	ALIGN_4

.L52:
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AA)

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	mulps	%xmm0, %xmm3
	movaps	-28 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm6
	pshufd	$0xaa, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm7
	pshufd	$0xff, %xmm1, %xmm3
	movaps	-28 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	-24 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	mulps	%xmm0, %xmm3
	movaps	-20 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm6
	pshufd	$0xaa, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm7
	pshufd	$0xff, %xmm1, %xmm3
	movaps	-24 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	-16 * SIZE(AA), %xmm0

	PREFETCH (PREFETCHSIZE + 16) * SIZE(AA)

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	mulps	%xmm0, %xmm3
	movaps	-12 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm6
	pshufd	$0xaa, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm7
	pshufd	$0xff, %xmm1, %xmm3
	movaps	-20 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	 -8 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	mulps	%xmm0, %xmm3
	movaps	 -4 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm6
	pshufd	$0xaa, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm7
	pshufd	$0xff, %xmm1, %xmm3
	movaps	-16 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	  0 * SIZE(AA), %xmm0

	subl   $-32 * SIZE, AA
	subl   $-16 * SIZE, BB

	subl   $1, %eax
	jne    .L52
	ALIGN_4

.L55:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L58
	ALIGN_4

.L56:
	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-30 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	-28 * SIZE(AA), %xmm0

	addl	$4 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L56
	ALIGN_4

.L58:
#if defined(LN) || defined(RT)
	movl	KK, %eax
#ifdef LN
	subl	$4, %eax
#else
	subl	$2, %eax
#endif

	movl	AORIG, AA

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

	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

	addps	%xmm2, %xmm4
	addps	%xmm3, %xmm5

#if defined(LN) || defined(LT)
	movaps	 %xmm4, %xmm0
	unpcklps %xmm6, %xmm4
	unpckhps %xmm6, %xmm0

	movaps	 %xmm5, %xmm1
	unpcklps %xmm7, %xmm5
	unpckhps %xmm7, %xmm1

	movaps	 %xmm4, %xmm6
	unpcklps %xmm5, %xmm4
	unpckhps %xmm5, %xmm6

	movaps	 %xmm0, %xmm2
	unpcklps %xmm1, %xmm0
	unpckhps %xmm1, %xmm2

	movsd	-32 * SIZE(BB), %xmm1
	movsd	-30 * SIZE(BB), %xmm3
	movsd	-28 * SIZE(BB), %xmm5
	movsd	-26 * SIZE(BB), %xmm7

	subps	%xmm4,  %xmm1
	subps	%xmm6,  %xmm3
	subps	%xmm0,  %xmm5
	subps	%xmm2,  %xmm7
#else
	movaps	-32 * SIZE(AA), %xmm0
	movaps	-28 * SIZE(AA), %xmm1

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

#ifdef LN
	movaps	-20 * SIZE(AA), %xmm4
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm6, %xmm7
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm7, %xmm6
	subps	 %xmm6, %xmm5
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm7, %xmm6
	subps	 %xmm6, %xmm3
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm7, %xmm6
	subps	 %xmm6, %xmm1

	movaps	-24 * SIZE(AA), %xmm4
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm6, %xmm5
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm5, %xmm6
	subps	 %xmm6, %xmm3
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm5, %xmm6
	subps	 %xmm6, %xmm1

	movaps	-28 * SIZE(AA), %xmm4
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm6, %xmm3
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm3, %xmm6
	subps	 %xmm6, %xmm1

	movaps	-32 * SIZE(AA), %xmm4
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm6, %xmm1
#endif

#ifdef LT
	movaps	-32 * SIZE(AA), %xmm4
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm6, %xmm1

	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm1, %xmm6
	subps	 %xmm6, %xmm3
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm1, %xmm6
	subps	 %xmm6, %xmm5
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm1, %xmm6
	subps	 %xmm6, %xmm7

	movaps	-28 * SIZE(AA), %xmm4
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm6, %xmm3
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm3, %xmm6
	subps	 %xmm6, %xmm5
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm3, %xmm6
	subps	 %xmm6, %xmm7

	movaps	-24 * SIZE(AA), %xmm4
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm6, %xmm5
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm5, %xmm6
	subps	 %xmm6, %xmm7

	movaps	-20 * SIZE(AA), %xmm4
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm6, %xmm7
#endif

#ifdef RN
	movaps	-32 * SIZE(BB), %xmm6
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm7, %xmm0
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm0, %xmm7
	subps	 %xmm7, %xmm1

	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm7, %xmm1
#endif

#ifdef RT
	movaps	-32 * SIZE(BB), %xmm6
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm7, %xmm1
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm1, %xmm7
	subps	 %xmm7, %xmm0

	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm7, %xmm0
#endif

#if defined(LN) || defined(LT)
	movlps	%xmm1, -32 * SIZE(BB)
	movlps	%xmm3, -30 * SIZE(BB)
	movlps	%xmm5, -28 * SIZE(BB)
	movlps	%xmm7, -26 * SIZE(BB)
#else
	movaps	%xmm0, -32 * SIZE(AA)
	movaps	%xmm1, -28 * SIZE(AA)
#endif

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

#if defined(LN) || defined(LT)
	unpcklps %xmm5, %xmm1
	unpcklps %xmm7, %xmm3

	movaps	 %xmm1, %xmm2
	unpcklps %xmm3, %xmm1
	unpckhps %xmm3, %xmm2

	movlps	%xmm1, 0 * SIZE(CO1)
	movhps	%xmm1, 2 * SIZE(CO1)
	movlps	%xmm2, 0 * SIZE(CO1, LDC, 1)
	movhps	%xmm2, 2 * SIZE(CO1, LDC, 1)
#else
	movlps	%xmm0, 0 * SIZE(CO1)
	movhps	%xmm0, 2 * SIZE(CO1)
	movlps	%xmm1, 0 * SIZE(CO1, LDC, 1)
	movhps	%xmm1, 2 * SIZE(CO1, LDC, 1)
#endif

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

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
	leal	(BB, %eax, 2), BB
#endif

#ifdef LN
	subl	$4, KK
#endif

#ifdef LT
	addl	$4, KK
#endif

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

	decl	%ebx			# i --
	jg	.L51
	ALIGN_4

.L79:
#ifdef LN
       movl	K, %eax
       leal	(, %eax, SIZE), %eax
       leal 	(B, %eax, 2), B
#endif

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

#ifdef RN
	addl	$2, KK
#endif

#ifdef RT
	subl	$2, KK
#endif
	ALIGN_4

.L80:
	testl	$1, N