gemm_kernel_4x4_core2.s

Optimized GotoBLAS libraries

	movapd	  4 * SIZE(BO), %xmm4
	mulpd	%xmm0, %xmm4
	addpd	%xmm5, %xmm11
	movapd	  6 * SIZE(BO), %xmm5
	mulpd	%xmm0, %xmm5
	movapd	-10 * SIZE(AO), %xmm0

	addpd	%xmm2, %xmm8
	movapd	  8 * SIZE(BO), %xmm2
	mulpd	%xmm0, %xmm2
	addpd	%xmm3, %xmm9
	movapd	 10 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	addpd	%xmm4, %xmm10
	movapd	 12 * SIZE(BO), %xmm4
	mulpd	%xmm0, %xmm4
	addpd	%xmm5, %xmm11
	movapd	 14 * SIZE(BO), %xmm5
	mulpd	%xmm0, %xmm5
	movapd	 -8 * SIZE(AO), %xmm0

	subq	$ -8 * SIZE, AO
	subq	$-32 * SIZE, BO
	subq	$1, %rax
	jg,pt	.L21
	ALIGN_4

.L25:
	movapd	ALPHA, %xmm7

#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	je,pt	.L28
	ALIGN_4

.L26:
	addpd	%xmm2, %xmm8
	movapd	-16 * SIZE(BO), %xmm2
	mulpd	%xmm0, %xmm2
	addpd	%xmm3, %xmm9
	movapd	-14 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	addpd	%xmm4, %xmm10
	movapd	-12 * SIZE(BO), %xmm4
	mulpd	%xmm0, %xmm4
	addpd	%xmm5, %xmm11
	movapd	-10 * SIZE(BO), %xmm5
	mulpd	%xmm0, %xmm5
	movapd	-14 * SIZE(AO), %xmm0

	addq	$2 * SIZE, AO
	addq	$8 * SIZE, BO
	subq	$1, %rax
	jg,pt	.L26
	ALIGN_4

.L28:
	addpd	%xmm2, %xmm8
	addpd	%xmm3, %xmm9
	addpd	%xmm4, %xmm10
	addpd	%xmm5, %xmm11

#ifndef TRMMKERNEL
	movsd	0 * SIZE(CO1), %xmm0
	movhpd	1 * SIZE(CO1), %xmm0
	movsd	0 * SIZE(CO2), %xmm2
	movhpd	1 * SIZE(CO2), %xmm2

	movsd	0 * SIZE(CO1, LDC, 2), %xmm4
	movhpd	1 * SIZE(CO1, LDC, 2), %xmm4
	movsd	0 * SIZE(CO2, LDC, 2), %xmm6
	movhpd	1 * SIZE(CO2, LDC, 2), %xmm6
#endif

	mulpd	%xmm7, %xmm8
	mulpd	%xmm7, %xmm9
	mulpd	%xmm7, %xmm10
	mulpd	%xmm7, %xmm11

#ifndef TRMMKERNEL
	addpd	%xmm0,  %xmm8
	addpd	%xmm2, %xmm9
	addpd	%xmm4, %xmm10
	addpd	%xmm6, %xmm11
#endif

	movsd	%xmm8, 0 * SIZE(CO1)
	movhpd	%xmm8, 1 * SIZE(CO1)
	movsd	%xmm9, 0 * SIZE(CO2)
	movhpd	%xmm9, 1 * SIZE(CO2)

	movsd	%xmm10, 0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm10, 1 * SIZE(CO1, LDC, 2)
	movsd	%xmm11, 0 * SIZE(CO2, LDC, 2)
	movhpd	%xmm11, 1 * SIZE(CO2, LDC, 2)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movq	K, %rax
	subq	KKK, %rax
	leaq	(,%rax, SIZE), %rax
	leaq	(AO, %rax, 2), AO
	leaq	(BO, %rax, 8), BO
#endif

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

	addq	$2 * SIZE, CO1		# coffset += 4
	addq	$2 * SIZE, CO2		# coffset += 4
	ALIGN_4
	
.L30:
	testq	$1, M
	jle,pt	.L39

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

	leaq	16 * SIZE + BUFFER, BO
#else
	leaq	16 * SIZE + BUFFER, BO
	movq	KK, %rax
	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 1), AO
	leaq	(BO, %rax, 8), BO
#endif	

	pxor	%xmm8, %xmm8
	movsd	-16 * SIZE(AO), %xmm0

	pxor	%xmm9, %xmm9
	pxor	%xmm10, %xmm10
	pxor	%xmm11, %xmm11

	movapd	%xmm8,  %xmm2
	movapd	%xmm9,  %xmm3
	movapd	%xmm10, %xmm4
	movapd	%xmm11, %xmm5

#ifndef TRMMKERNEL
	movq	K, %rax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
	movq	K, %rax
	subq	KK, %rax
	movq	%rax, KKK	
#else
	movq	KK, %rax
#ifdef LEFT
	addq	$1, %rax
#else
	addq	$4, %rax
#endif
	movq	%rax, KKK
#endif
	sarq	$2, %rax
	jle,pn	.L35
	ALIGN_4

.L31:	
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)
	addsd	%xmm2, %xmm8
	movsd	-16 * SIZE(BO), %xmm2
	mulsd	%xmm0, %xmm2
	addsd	%xmm3, %xmm9
	movsd	-14 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	addsd	%xmm4, %xmm10
	movsd	-12 * SIZE(BO), %xmm4
	mulsd	%xmm0, %xmm4
	addsd	%xmm5, %xmm11
	movsd	-10 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	movsd	-15 * SIZE(AO), %xmm0

	addsd	%xmm2, %xmm8
	movsd	 -8 * SIZE(BO), %xmm2
	mulsd	%xmm0, %xmm2
	addsd	%xmm3, %xmm9
	movsd	 -6 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	addsd	%xmm4, %xmm10
	movsd	 -4 * SIZE(BO), %xmm4
	mulsd	%xmm0, %xmm4
	addsd	%xmm5, %xmm11
	movsd	 -2 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	movsd	-14 * SIZE(AO), %xmm0

	addsd	%xmm2, %xmm8
	movsd	  0 * SIZE(BO), %xmm2
	mulsd	%xmm0, %xmm2
	addsd	%xmm3, %xmm9
	movsd	  2 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	addsd	%xmm4, %xmm10
	movsd	  4 * SIZE(BO), %xmm4
	mulsd	%xmm0, %xmm4
	addsd	%xmm5, %xmm11
	movsd	  6 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	movsd	-13 * SIZE(AO), %xmm0

	addsd	%xmm2, %xmm8
	movsd	  8 * SIZE(BO), %xmm2
	mulsd	%xmm0, %xmm2
	addsd	%xmm3, %xmm9
	movsd	 10 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	addsd	%xmm4, %xmm10
	movsd	 12 * SIZE(BO), %xmm4
	mulsd	%xmm0, %xmm4
	addsd	%xmm5, %xmm11
	movsd	 14 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	movsd	-12 * SIZE(AO), %xmm0

	subq	$ -4 * SIZE, AO
	subq	$-32 * SIZE, BO
	subq	$1, %rax
	jg,pt	.L31
	ALIGN_4

.L35:
	movsd	ALPHA, %xmm7

#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	je,pt	.L38
	ALIGN_4

.L36:
	addsd	%xmm2, %xmm8
	movsd	-16 * SIZE(BO), %xmm2
	mulsd	%xmm0, %xmm2
	addsd	%xmm3, %xmm9
	movsd	-14 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	addsd	%xmm4, %xmm10
	movsd	-12 * SIZE(BO), %xmm4
	mulsd	%xmm0, %xmm4
	addsd	%xmm5, %xmm11
	movsd	-10 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	movsd	-15 * SIZE(AO), %xmm0

	addq	$1 * SIZE, AO
	addq	$8 * SIZE, BO
	subq	$1, %rax
	jg,pt	.L36
	ALIGN_4

.L38:
	addsd	%xmm2, %xmm8
	addsd	%xmm3, %xmm9
	addsd	%xmm4, %xmm10
	addsd	%xmm5, %xmm11


#ifndef TRMMKERNEL
	movsd	0 * SIZE(CO1), %xmm0
	movsd	0 * SIZE(CO2), %xmm2
	movsd	0 * SIZE(CO1, LDC, 2), %xmm4
	movsd	0 * SIZE(CO2, LDC, 2), %xmm6
#endif

	mulsd	%xmm7, %xmm8
	mulsd	%xmm7, %xmm9
	mulsd	%xmm7, %xmm10
	mulsd	%xmm7, %xmm11

#ifndef TRMMKERNEL
	addsd	%xmm0,  %xmm8
	addsd	%xmm2, %xmm9
	addsd	%xmm4, %xmm10
	addsd	%xmm6, %xmm11
#endif

	movsd	%xmm8, 0 * SIZE(CO1)
	movsd	%xmm9, 0 * SIZE(CO2)
	movsd	%xmm10, 0 * SIZE(CO1, LDC, 2)
	movsd	%xmm11, 0 * SIZE(CO2, LDC, 2)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movq	K, %rax
	subq	KKK, %rax
	leaq	(,%rax, SIZE), %rax
	leaq	(AO, %rax, 1), AO
	leaq	(BO, %rax, 8), BO
#endif

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

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

	leaq	(C, LDC, 4), C
	subq	$1, J
	jg,pt	.L01
	ALIGN_4

.L40:
	testq	$2, N
	jle,pt	.L80
	ALIGN_4

.L41:
/* Copying to Sub Buffer */
	leaq	BUFFER, BO
	
#if defined(TRMMKERNEL) && defined(LEFT)
	movq	OFFSET, %rax
	movq	%rax, KK
#endif	

	movq	K, %rax
	sarq	$3, %rax
	jle	.L43

	addq	%rax, %rax
	ALIGN_4
	
.L42:
	movddup	 -16 * SIZE(B), %xmm8
	movddup	 -15 * SIZE(B), %xmm9
	movddup	 -14 * SIZE(B), %xmm10
	movddup	 -13 * SIZE(B), %xmm11
	movddup	 -12 * SIZE(B), %xmm12
	movddup	 -11 * SIZE(B), %xmm13
	movddup	 -10 * SIZE(B), %xmm14
	movddup	  -9 * SIZE(B), %xmm15

	movapd	%xmm8,   0 * SIZE(BO)
	movapd	%xmm9,   2 * SIZE(BO)
	movapd	%xmm10,  4 * SIZE(BO)
	movapd	%xmm11,  6 * SIZE(BO)
	movapd	%xmm12,  8 * SIZE(BO)
	movapd	%xmm13, 10 * SIZE(BO)
	movapd	%xmm14, 12 * SIZE(BO)
	movapd	%xmm15, 14 * SIZE(BO)

	addq	$8 * SIZE, B
	addq	$16 * SIZE, BO

	subq	$1, %rax
	jne	.L42
	ALIGN_4

.L43:
	movq	K, %rax
	andq	$7, %rax
	BRANCH
	jle	.L45
	ALIGN_4

.L44:
	movddup	 -16 * SIZE(B), %xmm8
	movddup	 -15 * SIZE(B), %xmm9

	movapd	%xmm8,  0 * SIZE(BO)
	movapd	%xmm9,  2 * SIZE(BO)

	addq	$2 * SIZE, B
	addq	$4 * SIZE, BO
	subq	$1, %rax
	jne	.L44
	ALIGN_4
	
.L45:
	movq	C, CO1
	leaq	(C, LDC, 1), CO2
	movq	A, AO		# aoffset = a

	movq	M,  I
	sarq	$2, I	# i = (m >> 2)
	jle	.L60
	ALIGN_4

.L50:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leaq	16 * SIZE + BUFFER, BO
#else
	leaq	16 * SIZE + BUFFER, BO
	movq	KK, %rax
	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 4), AO
	leaq	(BO, %rax, 4), BO
#endif	

	pxor	%xmm8, %xmm8
	pxor	%xmm9, %xmm9

	prefetcht0     3 * SIZE(CO1)
	pxor	%xmm12, %xmm12
	prefetcht0     3 * SIZE(CO2)
	pxor	%xmm13, %xmm13

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

	movapd	%xmm8, %xmm2
	movapd	%xmm8, %xmm3
	movapd	%xmm8, %xmm4
	movapd	%xmm8, %xmm5


#ifndef TRMMKERNEL
	movq	K, %rax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
	movq	K, %rax
	subq	KK, %rax
	movq	%rax, KKK	
#else
	movq	KK, %rax
#ifdef LEFT
	addq	$4, %rax
#else
	addq	$2, %rax
#endif
	movq	%rax, KKK
#endif
	sarq	$2, %rax
	jle	.L55
	ALIGN_4

.L51:	
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)

	addpd	%xmm2, %xmm8
	movapd	-16 * SIZE(BO), %xmm2
	addpd	%xmm3, %xmm12
	movapd	 %xmm2, %xmm3
	mulpd	%xmm0, %xmm2
	mulpd	%xmm1, %xmm3
	addpd	%xmm4, %xmm9
	movapd	-14 * SIZE(BO), %xmm4
	addpd	%xmm5, %xmm13
	movapd	 %xmm4, %xmm5
	mulpd	%xmm0, %xmm4
	movapd	-12 * SIZE(AO), %xmm0
	mulpd	%xmm1, %xmm5
	movapd	-10 * SIZE(AO), %xmm1

	addpd	%xmm2, %xmm8
	movapd	-12 * SIZE(BO), %xmm2
	addpd	%xmm3, %xmm12
	movapd	 %xmm2, %xmm3
	mulpd	%xmm0, %xmm2
	mulpd	%xmm1, %xmm3
	addpd	%xmm4, %xmm9
	movapd	-10 * SIZE(BO), %xmm4
	addpd	%xmm5, %xmm13
	movapd	 %xmm4, %xmm5
	mulpd	%xmm0, %xmm4
	movapd	 -8 * SIZE(AO), %xmm0
	mulpd	%xmm1, %xmm5
	movapd	 -6 * SIZE(AO), %xmm1

	PREFETCH (PREFETCHSIZE +  8) * SIZE(AO)

	addpd	%xmm2, %xmm8
	movapd	 -8 * SIZE(BO), %xmm2
	addpd	%xmm3, %xmm12
	movapd	 %xmm2, %xmm3
	mulpd	%xmm0, %xmm2
	mulpd	%xmm1, %xmm3
	addpd	%xmm4, %xmm9
	movapd	 -6 * SIZE(BO), %xmm4
	addpd	%xmm5, %xmm13
	movapd	 %xmm4, %xmm5
	mulpd	%xmm0, %xmm4
	movapd	 -4 * SIZE(AO), %xmm0
	mulpd	%xmm1, %xmm5
	movapd	 -2 * SIZE(AO), %xmm1

	addpd	%xmm2, %xmm8
	movapd	 -4 * SIZE(BO), %xmm2
	addpd	%xmm3, %xmm12
	movapd	 %xmm2, %xmm3
	mulpd	%xmm0, %xmm2
	mulpd	%xmm1, %xmm3
	addpd	%xmm4, %xmm9
	movapd	 -2 * SIZE(BO), %xmm4
	addpd	%xmm5, %xmm13
	movapd	 %xmm4, %xmm5
	mulpd	%xmm0, %xmm4
	movapd	  0 * SIZE(AO), %xmm0
	mulpd	%xmm1, %xmm5
	movapd	  2 * SIZE(AO), %xmm1

	subq	$-16 * SIZE, AO
	subq	$-16 * SIZE, BO
	subq	$1, %rax
	jg	.L51
	ALIGN_4

.L55:
	movapd	ALPHA, %xmm7

#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	je .L58
	ALIGN_4

.L56:
	addpd	%xmm2, %xmm8
	movapd	-16 * SIZE(BO), %xmm2
	addpd	%xmm3, %xmm12
	movapd	 %xmm2, %xmm3
	mulpd	%xmm0, %xmm2
	mulpd	%xmm1, %xmm3

	addpd	%xmm4, %xmm9
	movapd	-14 * SIZE(BO), %xmm4
	addpd	%xmm5, %xmm13
	movapd	 %xmm4, %xmm5
	mulpd	%xmm0, %xmm4
	movapd	-12 * SIZE(AO), %xmm0
	mulpd	%xmm1, %xmm5
	movapd	-10 * SIZE(AO), %xmm1

	addq	$4 * SIZE, AO
	addq	$4 * SIZE, BO
	subq	$1, %rax
	jg	.L56
	ALIGN_4

.L58:
	addpd	%xmm2, %xmm8
	addpd	%xmm3, %xmm12
	addpd	%xmm4, %xmm9
	addpd	%xmm5, %xmm13


#ifndef TRMMKERNEL
	movsd	0 * SIZE(CO1), %xmm0
	movhpd	1 * SIZE(CO1), %xmm0
	movsd	2 * SIZE(CO1), %xmm1
	movhpd	3 * SIZE(CO1), %xmm1
	movsd	0 * SIZE(CO2), %xmm2
	movhpd	1 * SIZE(CO2), %xmm2
	movsd	2 * SIZE(CO2), %xmm3
	movhpd	3 * SIZE(CO2), %xmm3
#endif

	mulpd	%xmm7, %xmm8
	mulpd	%xmm7, %xmm9
	mulpd	%xmm7, %xmm12
	mulpd	%xmm7, %xmm13

#ifndef TRMMKERNEL
	addpd	%xmm0,  %xmm8
	addpd	%xmm1,  %xmm12
	addpd	%xmm2, %xmm9
	addpd	%xmm3, %xmm13
#endif

	movsd	%xmm8, 0 * SIZE(CO1)
	movhpd	%xmm8, 1 * SIZE(CO1)
	movsd	%xmm12, 2 * SIZE(CO1)
	movhpd	%xmm12, 3 * SIZE(CO1)
	movsd	%xmm9, 0 * SIZE(CO2)
	movhpd	%xmm9, 1 * SIZE(CO2)
	movsd	%xmm13, 2 * SIZE(CO2)
	movhpd	%xmm13, 3 * SIZE(CO2)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movq	K, %rax
	subq	KKK, %rax
	leaq	(,%rax, SIZE), %rax
	leaq	(AO, %rax, 4), AO
	leaq	(BO, %rax, 4), BO
#endif

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

	addq	$4 * SIZE, CO1
	addq	$4 * SIZE, CO2
	subq	$1, I
	jg	.L50
	ALIGN_4	

.L60:
	testq	$2, M
	jle	.L70

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

	leaq	16 * SIZE + BUFFER, BO
#else
	leaq	16 * SIZE + BUFFER, BO
	movq	KK, %rax
	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 2), AO
	leaq	(BO, %rax, 4), BO
#endif	

	pxor	%xmm8, %xmm8
	movapd	-16 * SIZE(AO), %xmm0
	pxor	%xmm9, %xmm9
	movapd	-14 * SIZE(AO), %xmm1
	pxor	%xmm10, %xmm10
	pxor	%xmm11, %xmm11

	movapd	%xmm8, %xmm2
	movapd	%xmm8, %xmm3
	movapd	%xmm8, %xmm4
	movapd	%xmm8, %xmm5


#ifndef TRMMKERNEL
	movq	K, %rax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
	movq	K, %rax
	subq	KK, %rax
	movq	%rax, KKK	
#else
	movq	KK, %rax
#ifdef LEFT
	addq	$2, %rax
#else
	addq	$2, %rax
#endif
	movq	%rax, KKK
#endif
	sarq	$2, %rax
	jle	.L65
	ALIGN_4

.L61:	
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)

	addpd	%xmm2, %xmm8
	movapd	-16 * SIZE(BO), %xmm2
	mulpd	%xmm0, %xmm2
	addpd	%xmm3, %xmm9
	movapd	-14 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	movapd	-12 * SIZE(AO), %xmm0
	addpd	%xmm4, %xmm10
	movapd	-12 * SIZE(BO), %xmm4
	mulpd	%xmm1, %xmm4
	addpd	%xmm5, %xmm11
	movapd	-10 * SIZE(BO), %xmm5
	mulpd	%xmm1, %xmm5
	movapd	-10 * SIZE(AO), %xmm1

	addpd	%xmm2, %xmm8
	movapd	 -8 * SIZE(BO), %xmm2
	mulpd	%xmm0, %xmm2
	addpd	%xmm3, %xmm9
	movapd	 -6 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	movapd	 -8 * SIZE(AO), %xmm0
	addpd	%xmm4, %xmm10
	movapd	 -4 * SIZE(BO), %xmm4
	mulpd	%xmm1, %xmm4
	addpd	%xmm5, %xmm11
	movapd	 -2 * SIZE(BO), %xmm5
	mulpd	%xmm1, %xmm5
	movapd	 -6 * SIZE(AO), %xmm1

	subq	$ -8 * SIZE, AO
	subq	$-16 * SIZE, BO
	subq	$1, %rax
	jg	.L61
	ALIGN_4

.L65:
	movapd	ALPHA, %xmm7

#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	je .L68
	ALIGN_4

.L66:
	addpd	%xmm2, %xmm8
	movapd	-16 * SIZE(BO), %xmm2
	mulpd	%xmm0, %xmm2
	addpd	%xmm3, %xmm9
	movapd	-14 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	movapd	-14 * SIZE(AO), %xmm0

	addq	$2 * SIZE, AO
	addq	$4 * SIZE, BO
	subq	$1, %rax
	jg	.L66
	ALIGN_4

.L68:
	addpd	%xmm2, %xmm8
	addpd	%xmm3, %xmm9
	addpd	%xmm4, %xmm10
	addpd	%xmm5, %xmm11