trsm_kernel_rt_2x4_penryn.s

Optimized GotoBLAS libraries

	mulpd	%xmm0, %xmm2

	addpd	%xmm1, %xmm5
	movaps	  4 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm4
	pshufd	$0x4e, %xmm3, %xmm2
	mulpd	%xmm0, %xmm3
	mulpd	%xmm0, %xmm2
	movaps	 -6 * SIZE(AA), %xmm0

	addpd	%xmm3, %xmm7
	movaps	  6 * SIZE(BB), %xmm3
	addpd	%xmm2, %xmm6
	pshufd	$0x4e, %xmm1, %xmm2
	mulpd	%xmm0, %xmm1
	mulpd	%xmm0, %xmm2

	addpd	%xmm1, %xmm5
	movaps	  8 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm4
	pshufd	$0x4e, %xmm3, %xmm2
	mulpd	%xmm0, %xmm3
	mulpd	%xmm0, %xmm2
	movaps	 -4 * SIZE(AA), %xmm0

	addpd	%xmm3, %xmm7
	movaps	 10 * SIZE(BB), %xmm3
	addpd	%xmm2, %xmm6
	pshufd	$0x4e, %xmm1, %xmm2
	mulpd	%xmm0, %xmm1
	mulpd	%xmm0, %xmm2

	addpd	%xmm1, %xmm5
	movaps	 12 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm4
	pshufd	$0x4e, %xmm3, %xmm2
	mulpd	%xmm0, %xmm3
	mulpd	%xmm0, %xmm2
	movaps	 -2 * SIZE(AA), %xmm0

	addpd	%xmm3, %xmm7
	movaps	 14 * SIZE(BB), %xmm3
	addpd	%xmm2, %xmm6
	pshufd	$0x4e, %xmm1, %xmm2
	mulpd	%xmm0, %xmm1
	mulpd	%xmm0, %xmm2

	addpd	%xmm1, %xmm5
	movaps	 16 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm4
	pshufd	$0x4e, %xmm3, %xmm2
	mulpd	%xmm0, %xmm3
	subl   $-32 * SIZE, BB
	mulpd	%xmm0, %xmm2
	movaps	  0 * SIZE(AA), %xmm0

	subl   $-16 * SIZE, AA

	subl   $1, %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		# if (k & 1)
	BRANCH
	je .L18
	ALIGN_3

.L16:
	addpd	%xmm3, %xmm7
	movaps	-14 * SIZE(BB), %xmm3
	addpd	%xmm2, %xmm6
	pshufd	$0x4e, %xmm1, %xmm2
	mulpd	%xmm0, %xmm1
	mulpd	%xmm0, %xmm2

	addpd	%xmm1, %xmm5
	movaps	-12 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm4
	pshufd	$0x4e, %xmm3, %xmm2
	mulpd	%xmm0, %xmm3
	mulpd	%xmm0, %xmm2

	movaps	-14 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$4 * 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

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

	addpd	%xmm2, %xmm6
	addpd	%xmm3, %xmm7

	movaps	 %xmm4, %xmm0
	movsd    %xmm5, %xmm4
	movsd    %xmm0, %xmm5

	movaps	 %xmm6, %xmm0
	movsd    %xmm7, %xmm6
	movsd    %xmm0, %xmm7

#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(BB), %xmm2
	movapd	-14 * SIZE(BB), %xmm5
	movapd	-12 * SIZE(BB), %xmm3
	movapd	-10 * SIZE(BB), %xmm7

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

#ifdef LN
	movddup	-13 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm3
	mulpd	 %xmm4, %xmm7

	movddup	-14 * SIZE(AA), %xmm4
	movapd	 %xmm4, %xmm6
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm2
	mulpd	 %xmm7, %xmm6
	subpd	 %xmm6, %xmm5

	movddup	 -16 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm2
	mulpd	 %xmm4, %xmm5

#endif

#ifdef LT
	movddup	 -16 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm2
	mulpd	 %xmm4, %xmm5

	movddup	 -15 * SIZE(AA), %xmm4
	movapd	 %xmm4, %xmm6
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm3
	mulpd	 %xmm5, %xmm6
	subpd	 %xmm6, %xmm7

	movddup	 -13 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm3
	mulpd	 %xmm4, %xmm7
#endif

#ifdef RN
	movddup	 -16 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm0
	movddup	 -15 * SIZE(BB), %xmm4
	mulpd	 %xmm0, %xmm4
	subpd	 %xmm4, %xmm1
	movddup	 -14 * SIZE(BB), %xmm4
	mulpd	 %xmm0, %xmm4
	subpd	 %xmm4, %xmm2
	movddup	 -13 * SIZE(BB), %xmm4
	mulpd	 %xmm0, %xmm4
	subpd	 %xmm4, %xmm3

	movddup	 -11 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm1
	movddup	 -10 * SIZE(BB), %xmm4
	mulpd	 %xmm1, %xmm4
	subpd	 %xmm4, %xmm2
	movddup	  -9 * SIZE(BB), %xmm4
	mulpd	 %xmm1, %xmm4
	subpd	 %xmm4, %xmm3

	movddup	 -6 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm2
	movddup	 -5 * SIZE(BB), %xmm4
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm3

	movddup	 -1 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm3
#endif

#ifdef RT
	movddup	-1 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm3
	movddup	-2 * SIZE(BB), %xmm4
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm2
	movddup	-3 * SIZE(BB), %xmm4
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm1
	movddup	-4 * SIZE(BB), %xmm4
	mulpd	 %xmm3, %xmm4
	subpd	 %xmm4, %xmm0

	movddup	-6 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm2
	movddup	-7 * SIZE(BB), %xmm4
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm1
	movddup	-8 * SIZE(BB), %xmm4
	mulpd	 %xmm2, %xmm4
	subpd	 %xmm4, %xmm0

	movddup	-11 * SIZE(BB), %xmm4
	mulpd	 %xmm4, %xmm1
	movddup	-12 * SIZE(BB), %xmm4
	mulpd	 %xmm1, %xmm4
	subpd	 %xmm4, %xmm0

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

#if defined(LN) || defined(LT)
	movapd	%xmm2,   -16 * SIZE(BB)
	movapd	%xmm5,   -14 * SIZE(BB)
	movapd	%xmm3,   -12 * SIZE(BB)
	movapd	%xmm7,   -10 * SIZE(BB)
#else
	movapd	%xmm0,   -16 * SIZE(AA)
	movapd	%xmm1,   -14 * SIZE(AA)
	movapd	%xmm2,   -12 * SIZE(AA)
	movapd	%xmm3,   -10 * SIZE(AA)
#endif

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

	leal	(LDC, LDC, 2), %eax

#if defined(LN) || defined(LT)
	movsd	%xmm2, 0 * SIZE(CO1)
	movsd	%xmm3, 1 * SIZE(CO1)
	movhps	%xmm2, 0 * SIZE(CO1, LDC, 1)
	movhps	%xmm3, 1 * SIZE(CO1, LDC, 1)
	movsd	%xmm5, 0 * SIZE(CO1, LDC, 2)
	movsd	%xmm7, 1 * SIZE(CO1, LDC, 2)
	movhps	%xmm5, 0 * SIZE(CO1, %eax, 1)
	movhps	%xmm7, 1 * SIZE(CO1, %eax, 1)
#else
	movsd	%xmm0, 0 * SIZE(CO1)
	movhps	%xmm0, 1 * SIZE(CO1)
	movsd	%xmm1, 0 * SIZE(CO1, LDC, 1)
	movhps	%xmm1, 1 * SIZE(CO1, LDC, 1)
	movsd	%xmm2, 0 * SIZE(CO1, LDC, 2)
	movhps	%xmm2, 1 * SIZE(CO1, LDC, 2)
	movsd	%xmm3, 0 * SIZE(CO1, %eax, 1)
	movhps	%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
	leal	(BB, %eax, 4), 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	.L11
	ALIGN_4

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

#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	$2 + BASE_SHIFT, %eax
	addl	%eax, BB
#endif	

	movaps	-16 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movaps	-16 * SIZE(BB), %xmm2
	pxor	%xmm5, %xmm5
	movaps	-14 * SIZE(BB), %xmm3
	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	.L25
	ALIGN_4

.L22:
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AA)

	pshufd	$0x44, %xmm0, %xmm1
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm4
	movaps	-12 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm5
	movaps	-10 * SIZE(BB), %xmm3

	pshufd	$0xee, %xmm0, %xmm1
	movaps	-14 * SIZE(AA), %xmm0
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm6
	movaps	 -8 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm7
	movaps	 -6 * SIZE(BB), %xmm3

	pshufd	$0x44, %xmm0, %xmm1
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm4
	movaps	 -4 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm5
	movaps	 -2 * SIZE(BB), %xmm3

	pshufd	$0xee, %xmm0, %xmm1
	movaps	-12 * SIZE(AA), %xmm0
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm6
	movaps	  0 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm7
	movaps	  2 * SIZE(BB), %xmm3

	pshufd	$0x44, %xmm0, %xmm1
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm4
	movaps	  4 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm5
	movaps	  6 * SIZE(BB), %xmm3

	pshufd	$0xee, %xmm0, %xmm1
	movaps	-10 * SIZE(AA), %xmm0
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm6
	movaps	  8 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm7
	movaps	 10 * SIZE(BB), %xmm3

	pshufd	$0x44, %xmm0, %xmm1
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm4
	movaps	 12 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm5
	movaps	 14 * SIZE(BB), %xmm3

	pshufd	$0xee, %xmm0, %xmm1
	movaps	 -8 * SIZE(AA), %xmm0
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm6
	movaps	 16 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm7
	movaps	 18 * SIZE(BB), %xmm3

	subl   $ -8 * SIZE, AA
	subl   $-32 * SIZE, BB

	subl   $1, %eax
	jne    .L22
	ALIGN_4

.L25:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L28

.L26:
	pshufd	$0x44, %xmm0, %xmm1
	movsd	-15 * SIZE(AA), %xmm0
	mulpd	%xmm1, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm2, %xmm4
	movaps	-12 * SIZE(BB), %xmm2
	addpd	%xmm3, %xmm5
	movaps	-10 * SIZE(BB), %xmm3

	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	-16 * SIZE(BB), %xmm0
	movapd	-14 * SIZE(BB), %xmm1

	subpd	%xmm4,  %xmm0
	subpd	%xmm5,  %xmm1
#else
	movapd	-16 * SIZE(AA), %xmm1
	movapd	-14 * 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	-16 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm0
	mulpd	 %xmm4, %xmm1
#endif

#ifdef LT
	movddup	-16 * SIZE(AA), %xmm4
	mulpd	 %xmm4, %xmm0
	mulpd	 %xmm4, %xmm1
#endif

#ifdef RN
	movsd	-16 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm0
	movsd	-15 * SIZE(BB), %xmm4
	mulsd	 %xmm0, %xmm4
	subsd	 %xmm4, %xmm1
	movsd	-14 * SIZE(BB), %xmm4
	mulsd	 %xmm0, %xmm4
	subsd	 %xmm4, %xmm2
	movsd	-13 * SIZE(BB), %xmm4
	mulsd	 %xmm0, %xmm4
	subsd	 %xmm4, %xmm3

	movsd	-11 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm1
	movsd	-10 * SIZE(BB), %xmm4
	mulsd	 %xmm1, %xmm4
	subsd	 %xmm4, %xmm2
	movsd	 -9 * SIZE(BB), %xmm4
	mulsd	 %xmm1, %xmm4
	subsd	 %xmm4, %xmm3

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

	movsd	 -1 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm3
#endif

#ifdef RT
	movsd	 -1 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm3
	movsd	 -2 * SIZE(BB), %xmm4
	mulsd	 %xmm3, %xmm4
	subsd	 %xmm4, %xmm2
	movsd	 -3 * SIZE(BB), %xmm4
	mulsd	 %xmm3, %xmm4
	subsd	 %xmm4, %xmm1
	movsd	 -4 * SIZE(BB), %xmm4
	mulsd	 %xmm3, %xmm4
	subsd	 %xmm4, %xmm0

	movsd	 -6 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm2
	movsd	 -7 * SIZE(BB), %xmm4
	mulsd	 %xmm2, %xmm4
	subsd	 %xmm4, %xmm1
	movsd	 -8 * SIZE(BB), %xmm4
	mulsd	 %xmm2, %xmm4
	subsd	 %xmm4, %xmm0

	movsd	 -11 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm1
	movsd	 -12 * SIZE(BB), %xmm4
	mulsd	 %xmm1, %xmm4
	subsd	 %xmm4, %xmm0

	movsd	 -16 * SIZE(BB), %xmm4
	mulsd	 %xmm4, %xmm0
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm0,  -16 * SIZE(BB)
	movapd	%xmm1,  -14 * SIZE(BB)
#else
	movsd	%xmm0,  -16 * SIZE(AA)
	movsd	%xmm1,  -15 * SIZE(AA)
	movsd	%xmm2,  -14 * SIZE(AA)
	movsd	%xmm3,  -13 * 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)
	movhps	%xmm0, 0 * SIZE(CO1, LDC, 1)
	movsd	%xmm1, 0 * SIZE(CO1, LDC, 2)
	movhps	%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

.L999:
	popl	%ebx
	popl	%esi
	popl	%edi
	popl	%ebp

	addl	$ARGS, %esp
	ret

	EPILOGUE