gemm_kernel_8x2_sse.s

Optimized GotoBLAS libraries

	decl   %eax
	jne    .L11
	ALIGN_2
#endif

.L12:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L14

.L13:
	movaps	 4 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movaps	 0 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm1
	movaps	 4 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm5
	movaps	 4 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	movaps	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm1
	movaps	 8 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm7

	addl	$8 * SIZE, AA
	addl	$8 * SIZE, BB
	subl	$1, %eax
	jg	.L13
	ALIGN_4

.L14:
	mulps	%xmm3, %xmm4
	mulps	%xmm3, %xmm5
	mulps	%xmm3, %xmm6
	mulps	%xmm3, %xmm7

#ifndef TRMMKERNEL
	shufps	$0xe4, %xmm4, %xmm4
	shufps	$0xe4, %xmm5, %xmm5

	MOVSD	0 * SIZE(%esi), %xmm0
	movhps	2 * SIZE(%esi), %xmm0
	MOVSD	4 * SIZE(%esi), %xmm1
	movhps	6 * SIZE(%esi), %xmm1

	shufps	$0xe4, %xmm6, %xmm6
	shufps	$0xe4, %xmm7, %xmm7

	MOVSD	0 * SIZE(%esi, LDC), %xmm2
	movhps	2 * SIZE(%esi, LDC), %xmm2
	MOVSD	4 * SIZE(%esi, LDC), %xmm3
	movhps	6 * SIZE(%esi, LDC), %xmm3

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

	MOVSD	%xmm4, 0 * SIZE(%esi)
	movhps	%xmm4, 2 * SIZE(%esi)
	MOVSD	%xmm6, 4 * SIZE(%esi)
	movhps	%xmm6, 6 * SIZE(%esi)

	MOVSD	%xmm5, 0 * SIZE(%esi, LDC)
	movhps	%xmm5, 2 * SIZE(%esi, LDC)
	MOVSD	%xmm7, 4 * SIZE(%esi, LDC)
	movhps	%xmm7, 6 * SIZE(%esi, LDC)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movl	K, %eax
	subl	KKK, %eax
	leal	(,%eax,    8), %eax
	leal	(AA, %eax, 4), AA
	leal	(BB, %eax, 4), BB
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$8, KK
#endif

	addl	$8 * SIZE, %esi
	BRANCH
	decl	%ebx			# i --
	jg	.L10
	ALIGN_2	

.L30:
	movl	M,  %ebx
	andl	$7, %ebx
	jle	.L99

	testl	$4, %ebx
	jle	.L50

#if (L1_DATA_LINESIZE == 64)
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leal	BUFFER, BB
	movaps	 0 * SIZE + BUFFER, %xmm2
	XORPS	%xmm4, %xmm4
	movaps	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	movaps	 16 * SIZE + BUFFER, %xmm3
	XORPS	%xmm6, %xmm6
	movaps	 16 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7
#else
	leal	BUFFER, BB
	movl	KK, %eax
	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 4), BB /* because it's doubled */

	movaps	 0 * SIZE(BB), %xmm2
	XORPS	%xmm4, %xmm4
	movaps	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	movaps	 16 * SIZE(BB), %xmm3
	XORPS	%xmm6, %xmm6
	movaps	 16 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7
#endif	

#ifndef TRMMKERNEL
	movl	K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
	movl	K, %eax
	subl	KK, %eax
	movl	%eax, KKK	
#else
	movl	KK, %eax
#ifdef LEFT
	addl	$4, %eax
#else
	addl	$2, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L32
	ALIGN_2

.L31:
	mulps	%xmm0, %xmm2
	mulps	 4 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm4
	movaps	 8 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm5
	movaps	 4 * SIZE(AA), %xmm0
	mulps	%xmm0, %xmm2
	mulps	12 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm6
	movaps	32 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm7
	movaps	 8 * SIZE(AA), %xmm0
	mulps	%xmm0, %xmm3
	mulps	20 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm4
	movaps	24 * SIZE(BB), %xmm3
	addps	%xmm0, %xmm5
	movaps	12 * SIZE(AA), %xmm0
	mulps	%xmm0, %xmm3
	mulps	28 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm6
	movaps	48 * SIZE(BB), %xmm3
	addps	%xmm0, %xmm7
	movaps	32 * SIZE(AA), %xmm0
	mulps	%xmm1, %xmm2
	mulps	36 * SIZE(BB), %xmm1
	addps	%xmm2, %xmm4
	movaps	40 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm5
	movaps	20 * SIZE(AA), %xmm1
	mulps	%xmm1, %xmm2
	mulps	44 * SIZE(BB), %xmm1
	addps	%xmm2, %xmm6
	movaps	64 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm7
	movaps	24 * SIZE(AA), %xmm1
	mulps	%xmm1, %xmm3
	mulps	52 * SIZE(BB), %xmm1
	addps	%xmm3, %xmm4
	movaps	56 * SIZE(BB), %xmm3
	addps	%xmm1, %xmm5
	movaps	28 * SIZE(AA), %xmm1
	mulps	%xmm1, %xmm3
	mulps	60 * SIZE(BB), %xmm1
	addps	%xmm3, %xmm6
	movaps	80 * SIZE(BB), %xmm3
	addps	%xmm1, %xmm7
	movaps	48 * SIZE(AA), %xmm1

	addl   $32 * SIZE, AA
	addl   $64 * SIZE, BB
	decl   %eax
	jne    .L31
	ALIGN_2

#else
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leal	BUFFER, BB
	movaps	 0 * SIZE + BUFFER, %xmm2
	XORPS	%xmm4, %xmm4
	movaps	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	movaps	 8 * SIZE + BUFFER, %xmm3
	XORPS	%xmm6, %xmm6
	movaps	 8 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7
#else
	leal	BUFFER, BB
	movl	KK, %eax
	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 4), BB /* because it's doubled */

	movaps	 0 * SIZE(BB), %xmm2
	XORPS	%xmm4, %xmm4
	movaps	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	movaps	 8 * SIZE(BB), %xmm3
	XORPS	%xmm6, %xmm6
	movaps	 8 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7
#endif	

#ifndef TRMMKERNEL
	movl	K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
	movl	K, %eax
	subl	KK, %eax
	movl	%eax, KKK	
#else
	movl	KK, %eax
#ifdef LEFT
	addl	$4, %eax
#else
	addl	$2, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L32
	ALIGN_2

.L31:
	mulps	%xmm0, %xmm2
	mulps	 4 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm4
	movaps	16 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm5
	movaps	 4 * SIZE(AA), %xmm0
	mulps	%xmm0, %xmm3
	mulps	12 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm6
	movaps	24 * SIZE(BB), %xmm3
	addps	%xmm0, %xmm7
	movaps	16 * SIZE(AA), %xmm0
	mulps	%xmm1, %xmm2
	mulps	20 * SIZE(BB), %xmm1
	addps	%xmm2, %xmm4
	movaps	32 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm5
	movaps	12 * SIZE(AA), %xmm1
	mulps	%xmm1, %xmm3
	mulps	28 * SIZE(BB), %xmm1
	addps	%xmm3, %xmm6
	movaps	40 * SIZE(BB), %xmm3
	addps	%xmm1, %xmm7
	movaps	24 * SIZE(AA), %xmm1
	mulps	%xmm0, %xmm2
	mulps	36 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm4
	movaps	48 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm5
	movaps	20 * SIZE(AA), %xmm0
	mulps	%xmm0, %xmm3
	mulps	44 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm6
	movaps	56 * SIZE(BB), %xmm3
	addps	%xmm0, %xmm7
	movaps	32 * SIZE(AA), %xmm0
	mulps	%xmm1, %xmm2
	mulps	52 * SIZE(BB), %xmm1
	addps	%xmm2, %xmm4
	movaps	64 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm5
	movaps	28 * SIZE(AA), %xmm1
	mulps	%xmm1, %xmm3
	mulps	60 * SIZE(BB), %xmm1
	addps	%xmm3, %xmm6
	movaps	72 * SIZE(BB), %xmm3
	addps	%xmm1, %xmm7
	movaps	40 * SIZE(AA), %xmm1

	addl   $32 * SIZE, AA
	addl   $64 * SIZE, BB
	decl   %eax
	jne    .L31
	ALIGN_2
#endif

.L32:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L34

.L33:
	mulps	%xmm0, %xmm2
	mulps	 4 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm4
	movaps	 8 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm5
	movaps	 4 * SIZE(AA), %xmm0

	addl	$4 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L33
	ALIGN_4

.L34:
	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

	mulps	%xmm3, %xmm4
	mulps	%xmm3, %xmm5

#ifndef TRMMKERNEL
	movlps	0 * SIZE(%esi), %xmm0
	movhps	2 * SIZE(%esi), %xmm0
	addps	%xmm0, %xmm4

	movlps	0 * SIZE(%esi, LDC), %xmm0
	movhps	2 * SIZE(%esi, LDC), %xmm0
	addps	%xmm0, %xmm5
#endif

#ifdef HAVE_SSE2
	movsd	%xmm4, 0 * SIZE(%esi)
	unpckhpd %xmm4, %xmm4
	movsd	%xmm4, 2 * SIZE(%esi)

	movsd	%xmm5, 0 * SIZE(%esi, LDC)
	unpckhpd %xmm5, %xmm5
	movsd	%xmm5, 2 * SIZE(%esi, LDC)
#else
	movlps	%xmm4, 0 * SIZE(%esi)
	movhps	%xmm4, 2 * SIZE(%esi)

	movlps	%xmm5, 0 * SIZE(%esi, LDC)
	movhps	%xmm5, 2 * SIZE(%esi, LDC)
#endif

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movl	K, %eax
	subl	KKK, %eax
	leal	(,%eax,    8), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 4), BB
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$4, KK
#endif

	addl	$4 * SIZE, %esi
	ALIGN_2	

.L50:
	testl	$2, %ebx
	jle	.L70


#if (L1_DATA_LINESIZE == 64)
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leal	BUFFER, BB
	MOVSD	 0 * SIZE + BUFFER, %xmm2
	XORPS	%xmm4, %xmm4
	MOVSD	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	MOVSD	 16 * SIZE + BUFFER, %xmm3
	XORPS	%xmm6, %xmm6
	MOVSD	 8 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7
#else
	leal	BUFFER, BB
	movl	KK, %eax
	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 4), BB /* because it's doubled */

	MOVSD	 0 * SIZE(BB), %xmm2
	XORPS	%xmm4, %xmm4
	MOVSD	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	MOVSD	 16 * SIZE(BB), %xmm3
	XORPS	%xmm6, %xmm6
	MOVSD	 8 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7
#endif	

#ifndef TRMMKERNEL
	movl	K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
	movl	K, %eax
	subl	KK, %eax
	movl	%eax, KKK	
#else
	movl	KK, %eax
#ifdef LEFT
	addl	$2, %eax
#else
	addl	$2, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L52
	ALIGN_2

.L51:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	MOVSD	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	MOVSD	12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 4 * 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
	MOVSD	 6 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	MOVSD	24 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	MOVSD	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	MOVSD	16 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	MOVSD	48 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	36 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	MOVSD	10 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	MOVSD	40 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm6
	MOVSD	44 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	MOVSD	12 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm7
	MOVSD	64 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	MOVSD	52 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	MOVSD	14 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	MOVSD	56 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm6
	MOVSD	60 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	MOVSD	24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm7
	MOVSD	80 * SIZE(BB), %xmm3

	addl   $16 * SIZE, AA
	addl   $64 * SIZE, BB
	decl   %eax
	jne    .L51
	ALIGN_2

#else
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leal	BUFFER, BB
	MOVSD	 0 * SIZE + BUFFER, %xmm2
	XORPS	%xmm4, %xmm4
	MOVSD	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	MOVSD	 8 * SIZE + BUFFER, %xmm3
	XORPS	%xmm6, %xmm6
	MOVSD	 8 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7
#else
	leal	BUFFER, BB
	movl	KK, %eax
	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 4), BB /* because it's doubled */

	MOVSD	 0 * SIZE(BB), %xmm2
	XORPS	%xmm4, %xmm4
	MOVSD	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	MOVSD	 8 * SIZE(BB), %xmm3
	XORPS	%xmm6, %xmm6
	MOVSD	 8 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7
#endif	

#ifndef TRMMKERNEL
	movl	K, %eax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
	movl	K, %eax
	subl	KK, %eax
	movl	%eax, KKK	
#else
	movl	KK, %eax
#ifdef LEFT
	addl	$2, %eax
#else
	addl	$2, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L52
	ALIGN_2

.L51:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	MOVSD	16 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	MOVSD	12 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	MOVSD	 4 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	MOVSD	24 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	20 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 6 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	MOVSD	32 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	MOVSD	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	MOVSD	16 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	MOVSD	40 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	36 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	MOVSD	10 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	MOVSD	48 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm6
	MOVSD	44 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	MOVSD	12 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm7
	MOVSD	56 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	52 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	MOVSD	14 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	MOVSD	64 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm6
	MOVSD	60 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	MOVSD	24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm7
	MOVSD	72 * SIZE(BB), %xmm3

	addl   $16 * SIZE, AA
	addl   $64 * SIZE, BB
	decl   %eax
	jne    .L51
	ALIGN_2
#endif

.L52:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L54

.L53:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	MOVSD	 8 * SIZE(BB), %xmm2

	addl	$2 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L53
	ALIGN_4

.L54:
	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

	mulps	%xmm3, %xmm4
	mulps	%xmm3, %xmm5

#ifndef TRMMKERNEL
	movlps	0 * SIZE(%esi), %xmm0
	addps	%xmm0, %xmm4
	movlps	0 * SIZE(%esi, LDC), %xmm0
	addps	%xmm0, %xmm5
#endif

#ifdef HAVE_SSE2
	movsd	%xmm4, 0 * SIZE(%esi)
	movsd	%xmm5, 0 * SIZE(%esi, LDC)
#else
	movlps	%xmm4, 0 * SIZE(%esi)
	movlps	%xmm5, 0 * SIZE(%esi, LDC)
#endif

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movl	K, %eax
	subl	KKK, %eax
	leal	(,%eax,    8), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 4), BB
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$2, KK
#endif