trsm_kernel_lt_4x4_sse.s

Optimized GotoBLAS libraries

	pshufd	$0xaa, %xmm5, %xmm4
	pshufd	$0xff, %xmm5, %xmm6
	movaps	%xmm0,  32 * SIZE(BB)
	movaps	%xmm2,  36 * SIZE(BB)
	movaps	%xmm4,  40 * SIZE(BB)
	movaps	%xmm6,  44 * SIZE(BB)

	pshufd	$0x00, %xmm7, %xmm0
	pshufd	$0x55, %xmm7, %xmm2
	pshufd	$0xaa, %xmm7, %xmm4
	pshufd	$0xff, %xmm7, %xmm6
	movaps	%xmm0,  48 * SIZE(BB)
	movaps	%xmm2,  52 * SIZE(BB)
	movaps	%xmm4,  56 * SIZE(BB)
	movaps	%xmm6,  60 * SIZE(BB)
#else
	movaps	%xmm0,   0 * SIZE(AA)
	movaps	%xmm1,   4 * SIZE(AA)
	movaps	%xmm2,   8 * SIZE(AA)
	movaps	%xmm3,  12 * SIZE(AA)
#endif

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

	leal	(LDC, LDC, 2), %eax

#if defined(LN) || defined(LT)
	movaps	 %xmm1, %xmm0
	unpcklps %xmm5, %xmm1
	unpckhps %xmm5, %xmm0

	movaps	 %xmm3, %xmm4
	unpcklps %xmm7, %xmm3
	unpckhps %xmm7, %xmm4

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

	movaps	 %xmm0, %xmm6
	unpcklps %xmm4, %xmm0
	unpckhps %xmm4, %xmm6

	movlps	%xmm1, 0 * SIZE(CO1)
	movhps	%xmm1, 2 * SIZE(CO1)
	movlps	%xmm2, 0 * SIZE(CO1, LDC, 1)
	movhps	%xmm2, 2 * SIZE(CO1, LDC, 1)
	movlps	%xmm0, 0 * SIZE(CO1, LDC, 2)
	movhps	%xmm0, 2 * SIZE(CO1, LDC, 2)
	movlps	%xmm6, 0 * SIZE(CO1, %eax, 1)
	movhps	%xmm6, 2 * SIZE(CO1, %eax, 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)
	movlps	%xmm2, 0 * SIZE(CO1, LDC, 2)
	movhps	%xmm2, 2 * SIZE(CO1, LDC, 2)
	movlps	%xmm3, 0 * SIZE(CO1, %eax, 1)
	movhps	%xmm3, 2 * SIZE(CO1, %eax, 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
#ifdef LT
	addl	$16 * SIZE, B
#endif
#endif

#ifdef LN
	subl	$4, KK
	movl	BORIG, B
#endif

#ifdef LT
	addl	$4, KK
#endif

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

	decl	%ebx			# i --
	jg	.L11
	ALIGN_4

.L20:
	testl	$2, M
	je	.L30

#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
	leal	(BB, %eax, 4), BB
#endif	

	movsd	 0 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movsd	 8 * SIZE(AA), %xmm1
	pxor	%xmm5, %xmm5
	movsd	 0 * SIZE(BB), %xmm2
	pxor	%xmm6, %xmm6
	movsd	16 * SIZE(BB), %xmm3
	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:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
#if defined(OPTERON) || defined(BARCELONA)
	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
#endif
	movsd	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm5
	movsd	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	movsd	12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movsd	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm7
	movsd	32 * SIZE(BB), %xmm2

	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	movsd	20 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm5
	movsd	24 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	movsd	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	movsd	 4 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	movsd	48 * SIZE(BB), %xmm3

	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movsd	36 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm5
	movsd	40 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	movsd	44 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movsd	 6 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm7
	movsd	64 * SIZE(BB), %xmm2

	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	movsd	52 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm5
	movsd	56 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	movsd	60 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	movsd	 16 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	movsd	80 * SIZE(BB), %xmm3

	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	movsd	68 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm5
	movsd	72 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm6
	movsd	76 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	movsd	10 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm7
	movsd	96 * SIZE(BB), %xmm2

	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movsd	84 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm5
	movsd	88 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm6
	movsd	92 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movsd	12 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm7
	movsd	112 * SIZE(BB), %xmm3

	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	movsd	100 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm5
	movsd	104 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm6
	movsd	108 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	movsd	 14 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm7
	movsd	128 * SIZE(BB), %xmm2

	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movsd	116 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm5
	movsd	120 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm6
	movsd	124 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movsd	 24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm7
	movsd	144 * SIZE(BB), %xmm3

	addl	$ 16 * SIZE, AA
	addl	$128 * 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	$7, %eax		# if (k & 1)
	BRANCH
	je .L28
	ALIGN_4

.L26:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movsd	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm5
	movsd	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	movsd	12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movsd	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm7
	movsd	16 * SIZE(BB), %xmm2

	addl	$ 2 * SIZE, AA
	addl	$16 * SIZE, BB
	decl	%eax
	jg	.L26
	ALIGN_4

.L28:
#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

	sall	$1 + BASE_SHIFT, %eax
	leal	(AA, %eax, 1), AA
	leal	(B,  %eax, 2), B
	leal	(BB, %eax, 8), BB
#endif

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

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

	movaps	 0 * SIZE(B), %xmm1
	movaps	 4 * SIZE(B), %xmm3

	subps	%xmm4,  %xmm1
	subps	%xmm6,  %xmm3
#else
	movsd	 0 * SIZE(AA), %xmm0
	movsd	 2 * SIZE(AA), %xmm1
	movsd	 4 * SIZE(AA), %xmm2
	movsd	 6 * SIZE(AA), %xmm3

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

#ifdef LN
	movaps	  0 * 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	 0 * 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	 0 * SIZE(B), %xmm6
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm7, %xmm0
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm0, %xmm7
	subps	 %xmm7, %xmm1
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm0, %xmm7
	subps	 %xmm7, %xmm2
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm0, %xmm7
	subps	 %xmm7, %xmm3

	movaps	 4 * SIZE(B), %xmm6
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm7, %xmm1
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm1, %xmm7
	subps	 %xmm7, %xmm2
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm1, %xmm7
	subps	 %xmm7, %xmm3

	movaps	 8 * SIZE(B), %xmm6
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm7, %xmm2
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm2, %xmm7
	subps	 %xmm7, %xmm3

	movaps	 12 * SIZE(B), %xmm6
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm7, %xmm3
#endif

#ifdef RT
	movaps	 12 * SIZE(B), %xmm6
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm7, %xmm3
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm3, %xmm7
	subps	 %xmm7, %xmm2
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm3, %xmm7
	subps	 %xmm7, %xmm1
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm3, %xmm7
	subps	 %xmm7, %xmm0

	movaps	  8 * SIZE(B), %xmm6
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm7, %xmm2
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm2, %xmm7
	subps	 %xmm7, %xmm1
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm2, %xmm7
	subps	 %xmm7, %xmm0

	movaps	  4 * SIZE(B), %xmm6
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm7, %xmm1
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm1, %xmm7
	subps	 %xmm7, %xmm0

	movaps	  0 * SIZE(B), %xmm6
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm7, %xmm0
#endif

#if defined(LN) || defined(LT)
	movaps	%xmm1,   0 * SIZE(B)
	movaps	%xmm3,   4 * SIZE(B)

	pshufd	$0x00, %xmm1, %xmm0
	pshufd	$0x55, %xmm1, %xmm2
	pshufd	$0xaa, %xmm1, %xmm4
	pshufd	$0xff, %xmm1, %xmm6
	movaps	%xmm0,   0 * SIZE(BB)
	movaps	%xmm2,   4 * SIZE(BB)
	movaps	%xmm4,   8 * SIZE(BB)
	movaps	%xmm6,  12 * SIZE(BB)

	pshufd	$0x00, %xmm3, %xmm0
	pshufd	$0x55, %xmm3, %xmm2
	pshufd	$0xaa, %xmm3, %xmm4
	pshufd	$0xff, %xmm3, %xmm6
	movaps	%xmm0,  16 * SIZE(BB)
	movaps	%xmm2,  20 * SIZE(BB)
	movaps	%xmm4,  24 * SIZE(BB)
	movaps	%xmm6,  28 * SIZE(BB)
#else
	movlps	%xmm0,   0 * SIZE(AA)
	movlps	%xmm1,   2 * SIZE(AA)
	movlps	%xmm2,   4 * SIZE(AA)
	movlps	%xmm3,   6 * SIZE(AA)
#endif

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

	leal	(LDC, LDC, 2), %eax

#if defined(LN) || defined(LT)
	movaps	 %xmm1, %xmm0
	unpcklps %xmm5, %xmm1
	unpckhps %xmm5, %xmm0

	movaps	 %xmm3, %xmm4
	unpcklps %xmm7, %xmm3
	unpckhps %xmm7, %xmm4

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

	movaps	 %xmm0, %xmm6
	unpcklps %xmm4, %xmm0
	unpckhps %xmm4, %xmm6

	movlps	%xmm1, 0 * SIZE(CO1)
	movlps	%xmm2, 0 * SIZE(CO1, LDC, 1)
	movlps	%xmm0, 0 * SIZE(CO1, LDC, 2)
	movlps	%xmm6, 0 * SIZE(CO1, %eax, 1)
#else
	movlps	%xmm0, 0 * SIZE(CO1)
	movlps	%xmm1, 0 * SIZE(CO1, LDC, 1)
	movlps	%xmm2, 0 * SIZE(CO1, LDC, 2)
	movlps	%xmm3, 0 * 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
	ALIGN_4

.L30:
	testl	$1, M
	je	.L39

#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

	leal	BUFFER, BB

#if defined(LN) || defined(RT)
	movl	KK, %eax
	sall	$2 + BASE_SHIFT, %eax
	leal	(BB, %eax, 4), BB
#endif	

	movss	 0 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movss	 4 * SIZE(AA), %xmm1
	pxor	%xmm5, %xmm5
	movss	 0 * SIZE(BB), %xmm2
	pxor	%xmm6, %xmm6
	movss	16 * SIZE(BB), %xmm3
	pxor	%xmm7, %xmm7

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L35
	ALIGN_4

.L32:
	mulss	%xmm0, %xmm2
	addss	%xmm2, %xmm4
#if defined(OPTERON) || defined(BARCELONA)
	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
#endif
	movss	 4 * SIZE(BB), %xmm2
	mulss	%xmm0, %xmm2
	addss	%xmm2, %xmm5
	movss	 8 * SIZE(BB), %xmm2
	mulss	%xmm0, %xmm2
	mulss	12 * SIZE(BB), %xmm0
	addss	%xmm2, %xmm6
	movss	32 * SIZE(BB), %xmm2
	addss	%xmm0, %xmm7
	movss	 1 * SIZE(AA), %xmm0

	mulss	%xmm0, %xmm3
	addss	%xmm3, %xmm4
	movss	20 * SIZE(BB), %xmm3
	mulss	%xmm0, %xmm3
	addss	%xmm3, %xmm5
	movss	24 * SIZE(BB), %xmm3
	mulss	%xmm0, %xmm3
	mulss	28 * SIZE(BB), %xmm0
	addss	%xmm3, %xmm6
	movss	48 * SIZE(BB), %xmm3
	addss	%xmm0, %xmm7
	movss	 2 * SIZE(AA), %xmm0

	mulss	%xmm0, %xmm2
	addss	%xmm2, %xmm4
	movss	36 * SIZE(BB), %xmm2
	mulss	%xmm0, %xmm2
	addss	%xmm2, %xmm5
	movss	40 * SIZE(BB), %xmm2
	mulss	%xmm0, %xmm2
	mulss	44 * SIZE(BB), %xmm0
	addss	%xmm2, %xmm6
	movss	64 * SIZE(BB), %xmm2
	addss	%xmm0, %xmm7
	movss	 3 * SIZE(AA), %xmm0

	mulss	%xmm0, %xmm3
	addss	%xmm3, %xmm4
	movss	52 * SIZE(BB), %xmm3
	mulss	%xmm0, %xmm3
	addss	%xmm3, %xmm5
	movss	56 * SIZE(BB), %xmm3
	mulss	%xmm0, %xmm3
	mulss	60 * SIZE(BB), %xmm0
	addss	%xmm3, %xmm6
	movss	80 * SIZE(BB), %xmm3
	addss	%xmm0, %xmm7
	movss	 8 * SIZE(AA), %xmm0

	mulss	%xmm1, %xmm2
	addss	%xmm2, %xmm4
	movss	68 * SIZE(BB), %xmm2
	mulss	%xmm1, %xmm2
	addss	%xmm2, %xmm5
	movss	72 * SIZE(BB), %xmm2
	mulss	%xmm1, %xmm2
	mulss	76 * SIZE(BB), %xmm1
	addss	%xmm2, %xmm6
	movss	96 * SIZE(BB), %xmm2
	addss	%xmm1, %xmm7
	movss	 5 * SIZE(AA), %xmm1

	mulss	%xmm1, %xmm3
	addss	%xmm3, %xmm4
	movss	84 * SIZE(BB), %xmm3
	mulss	%xmm1, %xmm3
	addss	%xmm3, %xmm5
	movss	88 * SIZE(BB), %xmm3
	mulss	%xmm1, %xmm3
	mulss	92 * SIZE(BB), %xmm1
	addss	%xmm3, %xmm6
	movss	112 * SIZE(BB), %xmm3
	addss	%xmm1, %xmm7
	movss	 6 * SIZE(AA), %xmm1

	mulss	%xmm1, %xmm2
	addss	%xmm2, %xmm4
	movss	100 * SIZE(BB), %xmm2
	mulss	%xmm1, %xmm2
	addss	%xmm2, %xmm5
	movss	104 * SIZE(BB), %xmm2
	mulss	%xmm1, %xmm2
	mulss	108 * SIZE(BB), %xmm1
	addss	%xmm2, %xmm6
	movss	128 * SIZE(BB), %xmm2
	addss	%xmm1, %xmm7
	movss	 7 * SIZE(AA), %xmm1

	mulss	%xmm1, %xmm3
	addss	%xmm3, %xmm4
	movss	116 * SIZE(BB), %xmm3
	mulss	%xmm1, %xmm3
	addss	%xmm3, %xmm5
	movss	120 * SIZE(BB), %xmm3
	mulss	%xmm1, %xmm3
	mulss	124 * SIZE(BB), %xmm1
	addss	%xmm3, %xmm6
	movss	144 * SIZE(BB), %xmm3
	addss	%xmm1, %xmm7
	movss	12 * SIZE(AA), %xmm1

	addl	$  8 * SIZE, AA
	addl	$128 * SIZE, BB
	decl   %eax
	jne    .L32
	ALIGN_4

.L35:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L38
	ALIGN_4

.L36:
	mulss	%xmm0, %xmm2
	addss	%xmm2, %xmm4
	movss	 4 * SIZE(BB), %xmm2
	mulss	%xmm0, %xmm2
	addss	%xmm2, %xmm5
	movss	 8 * SIZE(BB), %xmm2
	mulss	%xmm0, %xmm2
	mulss	12 * SIZE(BB), %xmm0
	addss	%xmm2, %xmm6
	movss	16 * SIZE(BB), %xmm2
	addss	%xmm0, %xmm7
	movss	 1 * SIZE(AA), %xmm0