gemm_kernel_4x4_sse3.s

Optimized GotoBLAS libraries

#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L25
	ALIGN_4

.L22:
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AA)
	addps	%xmm2, %xmm4
	movsd     4 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	movddup	  2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsd     8 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movsd    12 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	movddup	  4 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsd    32 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	movsd    20 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm0, %xmm3
	movddup	  6 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	movsd    24 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	movsd    28 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm0, %xmm3
	movddup	 16 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	movsd    48 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	movsd    36 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm1, %xmm2
	movddup	 10 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	movsd    40 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	movsd    44 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm1, %xmm2
	movddup	 12 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	movsd    64 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movsd    52 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm1, %xmm3
	movddup	 14 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsd    56 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movsd    60 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm1, %xmm3
	movddup	 24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsd    80 * SIZE(BB), %xmm3

	addl	$16 * SIZE, AA
	addl	$64 * SIZE, BB
	decl   %eax
	jne    .L22
	ALIGN_4

.L25:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L28
	ALIGN_4

.L26:
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movsd     4 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
 	movddup	  2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsd     8 * SIZE(BB), %xmm2

	addl	$2 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L26
	ALIGN_4

.L28:
	leal	(LDC, LDC, 2), %eax

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

#ifndef TRMMKERNEL
	movsd	0 * SIZE(%esi), %xmm0
	movhps	0 * SIZE(%esi, LDC, 1), %xmm0
	movsd	0 * SIZE(%esi, LDC, 2), %xmm1
	movhps	0 * SIZE(%esi, %eax, 1), %xmm1

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

	movsd	%xmm4, 0 * SIZE(%esi)
	movhps	%xmm4, 0 * SIZE(%esi, LDC, 1)
	movsd	%xmm5, 0 * SIZE(%esi, LDC, 2)
	movhps	%xmm5, 0 * SIZE(%esi, %eax, 1)

#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

	addl	$2 * SIZE, %esi		# coffset += 2
	ALIGN_4

.L30:
	testl	$1, M
	je	.L39

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

	leal	BUFFER, BB	# boffset1 = boffset
#else
	leal	BUFFER, BB	# boffset1 = boffset
	movl	KK, %eax
	leal	(, %eax,   4), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 8), BB
#endif	

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

#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	$1, %eax
#else
	addl	$4, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L35
	ALIGN_4

.L32:
	shufps	$0, %xmm0, %xmm0
	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AA)
	movhps	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movss	 1 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm4
	movsd	 8 * SIZE(BB), %xmm2
	shufps	$0, %xmm0, %xmm0
	movhps	12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movss	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movhps	20 * SIZE(BB), %xmm3
	shufps	$0, %xmm0, %xmm0
	movsd	32 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm3
	movss	 3 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm4
	movsd	24 * SIZE(BB), %xmm3
	shufps	$0, %xmm0, %xmm0
	movhps	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	movss	 8 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	movsd	48 * SIZE(BB), %xmm3
	shufps	$0, %xmm1, %xmm1
	movhps	36 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	movss	 5 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm4
	movsd	40 * SIZE(BB), %xmm2
	shufps	$0, %xmm1, %xmm1
	movhps	44 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	movss	 6 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	movsd	64 * SIZE(BB), %xmm2
	shufps	$0, %xmm1, %xmm1
	movhps	52 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movss	 7 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm4
	movsd	56 * SIZE(BB), %xmm3
	shufps	$0, %xmm1, %xmm1
	movhps	60 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movss	12 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsd	80 * SIZE(BB), %xmm3

	addl	$ 8 * SIZE, AA
	addl	$64 * SIZE, BB
	decl   %eax
	jne    .L32
	ALIGN_4

.L35:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA,  %xmm7
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L38
	ALIGN_4

.L36:
	shufps	$0, %xmm0, %xmm0
	movhps	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movss	 1 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm4
	movsd	 8 * SIZE(BB), %xmm2

	addl	$1 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L36
	ALIGN_4

.L38:
	leal	(LDC, LDC, 2), %eax

	addps	%xmm5, %xmm4
	mulps	%xmm7, %xmm4

	movhlps %xmm4, %xmm5

#ifndef TRMMKERNEL
	movss	0 * SIZE(%esi), %xmm0
	movss	0 * SIZE(%esi, LDC, 1), %xmm1
	movss	0 * SIZE(%esi, LDC, 2), %xmm2
	movss	0 * SIZE(%esi, %eax, 1), %xmm3

	addss	%xmm4, %xmm0
	psrlq	$32, %xmm4
	addss	%xmm4, %xmm1
	addss	%xmm5, %xmm2
	psrlq	$32, %xmm5
	addss	%xmm5, %xmm3

	movss	%xmm0, 0 * SIZE(%esi)
	movss	%xmm1, 0 * SIZE(%esi, LDC, 1)
	movss	%xmm2, 0 * SIZE(%esi, LDC, 2)
	movss	%xmm3 , 0 * SIZE(%esi, %eax, 1)
#else
	movss	%xmm4, 0 * SIZE(%esi)
	psrlq	$32, %xmm4
	movss	%xmm4, 0 * SIZE(%esi, LDC, 1)

	movss	%xmm5, 0 * SIZE(%esi, LDC, 2)
	psrlq	$32, %xmm5
	movss	%xmm5 , 0 * SIZE(%esi, %eax, 1)
#endif


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

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$1, KK
#endif
	ALIGN_4

.L39:
#if defined(TRMMKERNEL) && !defined(LEFT)
	addl	$4, KK
#endif

	leal	(, LDC, 4), %eax
	addl	%eax, C			# c += 4 * ldc
	decl	J			# j --
	jg	.L01
	ALIGN_4

.L40:
	testl	$2, N
	je	.L80

#if defined(TRMMKERNEL) && defined(LEFT)
	movl	OFFSET, %eax
	movl	%eax, KK
#endif	

	movl	K, %eax
	leal	BUFFER, %ecx
	sarl	$3, %eax
	jle	.L45
	ALIGN_4
	
.L42:
	movddup	 0 * SIZE(%edi), %xmm0
	movddup	 2 * SIZE(%edi), %xmm1
	movddup	 4 * SIZE(%edi), %xmm2
	movddup	 6 * SIZE(%edi), %xmm3
	movddup	 8 * SIZE(%edi), %xmm4
	movddup	10 * SIZE(%edi), %xmm5
	movddup	12 * SIZE(%edi), %xmm6
	movddup	14 * SIZE(%edi), %xmm7

	movaps	%xmm0,  0 * SIZE(%ecx)
	movaps	%xmm1,  4 * SIZE(%ecx)
	movaps	%xmm2,  8 * SIZE(%ecx)
	movaps	%xmm3, 12 * SIZE(%ecx)
	movaps	%xmm4, 16 * SIZE(%ecx)
	movaps	%xmm5, 20 * SIZE(%ecx)
	movaps	%xmm6, 24 * SIZE(%ecx)
	movaps	%xmm7, 28 * SIZE(%ecx)

#	prefetcht1	128 * SIZE(%ecx)
	prefetcht0	112 * SIZE(%edi)

	addl	$16 * SIZE, %edi
	addl	$32 * SIZE, %ecx
	decl	%eax
	jne	.L42
	ALIGN_4

.L45:
	movl	K, %eax
	andl	$7, %eax
	BRANCH
	jle	.L50
	ALIGN_4

.L46:
	movddup	 0 * SIZE(%edi), %xmm0
	movaps	%xmm0,  0 * SIZE(%ecx)

	addl	$2 * SIZE, %edi
	addl	$4 * SIZE, %ecx
	decl	%eax
	jne	.L46
	ALIGN_4
	
.L50:
	movl	C, %esi		# coffset = c
	movl	A, %edx		# aoffset = a
	movl	M,  %ebx
	sarl	$2, %ebx	# i = (m >> 2)
	jle	.L60
	ALIGN_4

.L51:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leal	BUFFER, BB	# boffset1 = boffset
#else
	leal	BUFFER, BB	# boffset1 = boffset
	movl	KK, %eax
	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 2), BB
#endif	

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

	prefetcht2	4 * SIZE(%esi)
	prefetcht2	4 * SIZE(%esi, LDC)

#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	.L55
	ALIGN_4

.L52:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AA)
	movshdup  0 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movaps	  4 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsldup  4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movshdup  4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movaps	  8 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsldup  8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movshdup  8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movaps	 12 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsldup 12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movshdup 12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movaps	 32 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsldup 32 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movshdup 16 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movaps	 20 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsldup 20 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movshdup 20 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movaps	 24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsldup 24 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movshdup 24 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movaps	 28 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsldup 28 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movshdup 28 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movaps	 48 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsldup 48 * SIZE(BB), %xmm3

	addl	$32 * SIZE, AA
	addl	$32 * SIZE, BB
	decl   %eax
	jne    .L52
	ALIGN_4

.L55:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L58
	ALIGN_4

.L56:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movshdup  0 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movaps	  4 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsldup  4 * SIZE(BB), %xmm2

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

.L58:
#ifndef TRMMKERNEL
	movsd	0 * SIZE(%esi), %xmm0
	movhps	2 * SIZE(%esi), %xmm0
	movsd	0 * SIZE(%esi, LDC, 1), %xmm1
	movhps	2 * SIZE(%esi, LDC, 1), %xmm1

	mulps	%xmm3, %xmm4
	mulps	%xmm3, %xmm5
	addps	%xmm0, %xmm4
	addps	%xmm1, %xmm5
#else
	mulps	%xmm3, %xmm4
	mulps	%xmm3, %xmm5
#endif

	movsd	%xmm4, 0 * SIZE(%esi)
	movhps	%xmm4, 2 * SIZE(%esi)
	movsd	%xmm5, 0 * SIZE(%esi, LDC, 1)
	movhps	%xmm5, 2 * SIZE(%esi, LDC, 1)

#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, 2), BB
#endif

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

	addl	$4 * SIZE, %esi		# coffset += 2
	decl	%ebx			# i --
	jg	.L51
	ALIGN_4

.L60:
	testl	$2, M
	je	.L70

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

	leal	BUFFER, BB	# boffset1 = boffset
#else
	leal	BUFFER, BB	# boffset1 = boffset
	movl	KK, %eax
	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 2), BB
#endif	

	movddup	  0 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movddup	  8 * SIZE(AA), %xmm1
	pxor	%xmm5, %xmm5
	movsd     0 * SIZE(BB), %xmm2
	movsd    16 * SIZE(BB), %xmm3

	leal	(LDC, LDC, 2), %eax

#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	.L65
	ALIGN_4

.L62:
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AA)
	movddup	  2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm4
	movsd     4 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	movddup	  4 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsd     8 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	movddup	  6 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm4
	movsd    12 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	movddup	 16 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movsd    32 * SIZE(BB), %xmm2
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm1, %xmm3
	movddup	 10 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm4
	movsd    20 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm1, %xmm3
	movddup	 12 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsd    24 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm1, %xmm3
	movddup	 14 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm4
	movsd    28 * SIZE(BB), %xmm3
	shufps	$0x50, %xmm3, %xmm3
	mulps	%xmm1, %xmm3
	movddup	 24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movsd    48 * SIZE(BB), %xmm3

	addl	$16 * SIZE, AA
	addl	$32 * SIZE, BB
	decl   %eax
	jne    .L62
	ALIGN_4

.L65:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L68
	ALIGN_4

.L66:
	shufps	$0x50, %xmm2, %xmm2
	mulps	%xmm0, %xmm2
	movddup	  2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm4