gemm_kernel_4x4_sse2.s

Optimized GotoBLAS libraries

	KERNEL8(16 *  0)

	KERNEL1(16 *  1)
	KERNEL2(16 *  1)
	KERNEL3(16 *  1)
	KERNEL4(16 *  1)
	KERNEL5(16 *  1)
	KERNEL6(16 *  1)
	KERNEL7(16 *  1)
	KERNEL8(16 *  1)

	addq	$8 * SIZE, %rax
	je,pn	.L15

	KERNEL1(16 *  0)
	KERNEL2(16 *  0)
	KERNEL3(16 *  0)
	KERNEL4(16 *  0)
	KERNEL5(16 *  0)
	KERNEL6(16 *  0)
	KERNEL7(16 *  0)
	KERNEL8(16 *  0)

	KERNEL1(16 *  1)
	KERNEL2(16 *  1)
	KERNEL3(16 *  1)
	KERNEL4(16 *  1)
	KERNEL5(16 *  1)
	KERNEL6(16 *  1)
	KERNEL7(16 *  1)
	KERNEL8(16 *  1)

	addq	$8 * SIZE, %rax
	je,pn	.L15

	KERNEL1(16 *  0)
	KERNEL2(16 *  0)
	KERNEL3(16 *  0)
	KERNEL4(16 *  0)
	KERNEL5(16 *  0)
	KERNEL6(16 *  0)
	KERNEL7(16 *  0)
	KERNEL8(16 *  0)

	KERNEL1(16 *  1)
	KERNEL2(16 *  1)
	KERNEL3(16 *  1)
	KERNEL4(16 *  1)
	KERNEL5(16 *  1)
	KERNEL6(16 *  1)
	KERNEL7(16 *  1)
	KERNEL8(16 *  1)

	addq	$8 * SIZE, %rax
	je,pn	.L15

	KERNEL1(16 *  0)
	KERNEL2(16 *  0)
	KERNEL3(16 *  0)
	KERNEL4(16 *  0)
	KERNEL5(16 *  0)
	KERNEL6(16 *  0)
	KERNEL7(16 *  0)
	KERNEL8(16 *  0)

	KERNEL1(16 *  1)
	KERNEL2(16 *  1)
	KERNEL3(16 *  1)
	KERNEL4(16 *  1)
	KERNEL5(16 *  1)
	KERNEL6(16 *  1)
	KERNEL7(16 *  1)
	KERNEL8(16 *  1)

	addq	$8 * SIZE, %rax
	je,pn	.L15

	KERNEL1(16 *  0)
	KERNEL2(16 *  0)
	KERNEL3(16 *  0)
	KERNEL4(16 *  0)
	KERNEL5(16 *  0)
	KERNEL6(16 *  0)
	KERNEL7(16 *  0)
	KERNEL8(16 *  0)

	KERNEL1(16 *  1)
	KERNEL2(16 *  1)
	KERNEL3(16 *  1)
	KERNEL4(16 *  1)
	KERNEL5(16 *  1)
	KERNEL6(16 *  1)
	KERNEL7(16 *  1)
	KERNEL8(16 *  1)

	addq	$8 * SIZE, %rax
	jl,pt	.L12
	ALIGN_3

.L15:
#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	testq	$4, %rax
	je .L16
	xorq	%rax, %rax
	ALIGN_3

	KERNEL1(16 *  0)
	KERNEL2(16 *  0)
	KERNEL3(16 *  0)
	KERNEL4(16 *  0)
	KERNEL5(16 *  0)
	KERNEL6(16 *  0)
	KERNEL7(16 *  0)
	KERNEL8(16 *  0)

	addq	$32 * SIZE, BO
	addq	$16 * SIZE, AO
	ALIGN_3

#else
	sarq	$2, %rax
	jle,pn	.L16
	ALIGN_3

.L12:
	KERNEL1(16 *  0)
	KERNEL2(16 *  0)
	KERNEL3(16 *  0)
	KERNEL4(16 *  0)
	KERNEL5(16 *  0)
	KERNEL6(16 *  0)
	KERNEL7(16 *  0)
	KERNEL8(16 *  0)

	addq	$ 32 * SIZE, BO
	subq	$-16 * SIZE, AO
	decq	%rax
	jg,pt	.L12
#endif

.L16:
	movapd	ALPHA, %xmm7

#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	je .L19

	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 4), AO
	leaq	(BO, %rax, 8), BO
	negq	%rax
	ALIGN_3

.L17:
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm8
	movapd	-14 * SIZE(BO, %rax, 8), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm9
	movapd	-12 * SIZE(BO, %rax, 8), %xmm1
	mulpd	%xmm0, %xmm1
	mulpd	-10 * SIZE(BO, %rax, 8), %xmm0
	addpd	%xmm1, %xmm10
	movapd	-16 * SIZE(BO, %rax, 8), %xmm1
	addpd	%xmm0, %xmm11
	movapd	-12 * SIZE(AO, %rax, 4), %xmm0
	mulpd	%xmm2, %xmm1
	addpd	%xmm1, %xmm12
	movapd	-14 * SIZE(BO, %rax, 8), %xmm1
	mulpd	%xmm2, %xmm1
	addpd	%xmm1, %xmm13
	movapd	-12 * SIZE(BO, %rax, 8), %xmm1
	mulpd	%xmm2, %xmm1
	mulpd	-10 * SIZE(BO, %rax, 8), %xmm2
	addpd	%xmm1, %xmm14
	movapd	 -8 * SIZE(BO, %rax, 8), %xmm1
	addpd	%xmm2, %xmm15
	movapd	-10 * SIZE(AO, %rax, 4), %xmm2

	addq	$SIZE, %rax
	jl	.L17
	ALIGN_3

.L19:
#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, %xmm10
	mulpd	%xmm7, %xmm11

	mulpd	%xmm7, %xmm12
	mulpd	%xmm7, %xmm13
	mulpd	%xmm7, %xmm14
	mulpd	%xmm7, %xmm15

#ifndef TRMMKERNEL
	movlpd	0 * SIZE(CO1, LDC, 2), %xmm4
	movhpd	1 * SIZE(CO1, LDC, 2), %xmm4
	movlpd	2 * SIZE(CO1, LDC, 2), %xmm5
	movhpd	3 * SIZE(CO1, LDC, 2), %xmm5

	movlpd	0 * SIZE(CO2, LDC, 2), %xmm6
	movhpd	1 * SIZE(CO2, LDC, 2), %xmm6
	movlpd	2 * SIZE(CO2, LDC, 2), %xmm7
	movhpd	3 * SIZE(CO2, LDC, 2), %xmm7

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

	movlpd	%xmm8, 0 * SIZE(CO1)
	movhpd	%xmm8, 1 * SIZE(CO1)
	movlpd	%xmm12, 2 * SIZE(CO1)
	movhpd	%xmm12, 3 * SIZE(CO1)

	movlpd	%xmm9, 0 * SIZE(CO2)
	movhpd	%xmm9, 1 * SIZE(CO2)
	movlpd	%xmm13, 2 * SIZE(CO2)
	movhpd	%xmm13, 3 * SIZE(CO2)

#ifndef TRMMKERNEL
	addpd	%xmm4, %xmm10
	addpd	%xmm5, %xmm14
	addpd	%xmm6, %xmm11
	addpd	%xmm7, %xmm15
#endif

	movlpd	%xmm10, 0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm10, 1 * SIZE(CO1, LDC, 2)
	movlpd	%xmm14, 2 * SIZE(CO1, LDC, 2)
	movhpd	%xmm14, 3 * SIZE(CO1, LDC, 2)

	movlpd	%xmm11, 0 * SIZE(CO2, LDC, 2)
	movhpd	%xmm11, 1 * SIZE(CO2, LDC, 2)
	movlpd	%xmm15, 2 * SIZE(CO2, LDC, 2)
	movhpd	%xmm15, 3 * 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, 4), AO
	leaq	(BO, %rax, 8), BO
#endif

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

	addq	$4 * SIZE, CO1		# coffset += 4
	addq	$4 * SIZE, CO2		# coffset += 4
	decq	I			# i --
	jg,pt	.L11
	ALIGN_3	

.L20:
	testq	$3, M
	je	.L39

	testq	$2, M
	je	.L30
	ALIGN_3

.L21:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leaq	BUFFER, BO
#else
	leaq	BUFFER, BO
	movq	KK, %rax
	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 2), AO
	leaq	(BO, %rax, 8), BO
#endif	

	movapd	-16 * SIZE(AO), %xmm0
	pxor	%xmm8, %xmm8
	movapd	 0 * SIZE(BO), %xmm1
	pxor	%xmm9, %xmm9
	movapd	 -8 * SIZE(AO), %xmm2
	pxor	%xmm10, %xmm10
	movapd	 8 * SIZE(BO), %xmm3
	pxor	%xmm11, %xmm11

	movapd	16 * SIZE(BO), %xmm5
	movapd	24 * SIZE(BO), %xmm7

#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	$4, %rax
#endif
	movq	%rax, KKK
#endif
	sarq	$3, %rax
	je	.L25
	ALIGN_3

.L22:
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm8
	PREFETCH	(PREFETCHSIZE     +  0) * SIZE(AO)
	movapd	 2 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm9
	movapd	 4 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	mulpd	 6 * SIZE(BO), %xmm0
	addpd	%xmm1, %xmm10
	movapd	32 * SIZE(BO), %xmm1
	addpd	%xmm0, %xmm11
	movapd	-14 * SIZE(AO), %xmm0

	mulpd	%xmm0, %xmm3
	addpd	%xmm3, %xmm8
	movapd	10 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	addpd	%xmm3, %xmm9
	movapd	12 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	mulpd	14 * SIZE(BO), %xmm0
	addpd	%xmm3, %xmm10
	movapd	40 * SIZE(BO), %xmm3
	addpd	%xmm0, %xmm11
	movapd	-12 * SIZE(AO), %xmm0

	mulpd	%xmm0, %xmm5
	addpd	%xmm5, %xmm8
	movapd	18 * SIZE(BO), %xmm5
	mulpd	%xmm0, %xmm5
	addpd	%xmm5, %xmm9
	movapd	20 * SIZE(BO), %xmm5
	mulpd	%xmm0, %xmm5
	mulpd	22 * SIZE(BO), %xmm0
	addpd	%xmm5, %xmm10
	movapd	48 * SIZE(BO), %xmm5
	addpd	%xmm0, %xmm11
	movapd	-10 * SIZE(AO), %xmm0

	mulpd	%xmm0, %xmm7
	addpd	%xmm7, %xmm8
	movapd	26 * SIZE(BO), %xmm7
	mulpd	%xmm0, %xmm7
	addpd	%xmm7, %xmm9
	movapd	28 * SIZE(BO), %xmm7
	mulpd	%xmm0, %xmm7
	mulpd	30 * SIZE(BO), %xmm0
	addpd	%xmm7, %xmm10
	movapd	56 * SIZE(BO), %xmm7
	addpd	%xmm0, %xmm11
	movapd	  0 * SIZE(AO), %xmm0

	PREFETCH	(PREFETCHSIZE     +  8) * SIZE(AO)
	mulpd	%xmm2, %xmm1
	addpd	%xmm1, %xmm8
	movapd	34 * SIZE(BO), %xmm1
	mulpd	%xmm2, %xmm1
	addpd	%xmm1, %xmm9
	movapd	36 * SIZE(BO), %xmm1
	mulpd	%xmm2, %xmm1
	mulpd	38 * SIZE(BO), %xmm2
	addpd	%xmm1, %xmm10
	movapd	64 * SIZE(BO), %xmm1
	addpd	%xmm2, %xmm11
	movapd	-6 * SIZE(AO), %xmm2

	mulpd	%xmm2, %xmm3
	addpd	%xmm3, %xmm8
	movapd	42 * SIZE(BO), %xmm3
	mulpd	%xmm2, %xmm3
	addpd	%xmm3, %xmm9
	movapd	44 * SIZE(BO), %xmm3
	mulpd	%xmm2, %xmm3
	mulpd	46 * SIZE(BO), %xmm2
	addpd	%xmm3, %xmm10
	movapd	72 * SIZE(BO), %xmm3
	addpd	%xmm2, %xmm11
	movapd	-4 * SIZE(AO), %xmm2

	mulpd	%xmm2, %xmm5
	addpd	%xmm5, %xmm8
	movapd	50 * SIZE(BO), %xmm5
	mulpd	%xmm2, %xmm5
	addpd	%xmm5, %xmm9
	movapd	52 * SIZE(BO), %xmm5
	mulpd	%xmm2, %xmm5
	mulpd	54 * SIZE(BO), %xmm2
	addpd	%xmm5, %xmm10
	movapd	80 * SIZE(BO), %xmm5
	addpd	%xmm2, %xmm11
	movapd	-2 * SIZE(AO), %xmm2

	mulpd	%xmm2, %xmm7
	addpd	%xmm7, %xmm8
	movapd	58 * SIZE(BO), %xmm7
	mulpd	%xmm2, %xmm7
	addpd	%xmm7, %xmm9
	movapd	60 * SIZE(BO), %xmm7
	mulpd	%xmm2, %xmm7
	mulpd	62 * SIZE(BO), %xmm2
	addpd	%xmm7, %xmm10
	movapd	88 * SIZE(BO), %xmm7
	addpd	%xmm2, %xmm11
	movapd	 8 * SIZE(AO), %xmm2

	addq   $16 * SIZE, AO
	addq   $64 * SIZE, BO
	decq   %rax
	jne    .L22
	ALIGN_3

.L25:
#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	movapd	ALPHA, %xmm7
	andq	$7, %rax		# if (k & 1)
	BRANCH
	je .L29
	ALIGN_3

.L26:
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm8
	movapd	  2 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm9
	movapd	  4 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	mulpd	  6 * SIZE(BO), %xmm0
	addpd	%xmm1, %xmm10
	movapd	  8 * SIZE(BO), %xmm1
	addpd	%xmm0, %xmm11
	movapd	-14 * SIZE(AO), %xmm0

	addq	$2 * SIZE, AO		# aoffset  += 4
	addq	$8 * SIZE, BO		# boffset1 += 8
	decq	%rax
	jg	.L26
	ALIGN_3

.L29:
#ifndef TRMMKERNEL
	movlpd	0 * SIZE(CO1), %xmm0
	movhpd	1 * SIZE(CO1), %xmm0
	movlpd	0 * SIZE(CO2), %xmm2
	movhpd	1 * SIZE(CO2), %xmm2

	movlpd	0 * SIZE(CO1, LDC, 2), %xmm4
	movhpd	1 * SIZE(CO1, LDC, 2), %xmm4
	movlpd	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

	movlpd	%xmm8, 0 * SIZE(CO1)
	movhpd	%xmm8, 1 * SIZE(CO1)
	movlpd	%xmm9, 0 * SIZE(CO2)
	movhpd	%xmm9, 1 * SIZE(CO2)
	movlpd	%xmm10, 0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm10, 1 * SIZE(CO1, LDC, 2)
	movlpd	%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_3	

.L30:
	testq	$1, M
	je	.L39
	ALIGN_3

.L31:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

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

	movsd	-16 * SIZE(AO), %xmm0
	pxor	%xmm8, %xmm8
	movsd	 0 * SIZE(BO), %xmm1
	pxor	%xmm9, %xmm9
	movsd	 -8 * SIZE(AO), %xmm2
	pxor	%xmm10, %xmm10
	movsd	 8 * SIZE(BO), %xmm3
	pxor	%xmm11, %xmm11

	movsd	16 * SIZE(BO), %xmm5
	movsd	24 * SIZE(BO), %xmm7

#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	$3, %rax
	je	.L35
	ALIGN_3

.L32:
	mulsd	%xmm0, %xmm1
	addsd	%xmm1, %xmm8
	PREFETCH	(PREFETCHSIZE     +  0) * SIZE(AO)
	movsd	 2 * SIZE(BO), %xmm1
	mulsd	%xmm0, %xmm1
	addsd	%xmm1, %xmm9
	movsd	 4 * SIZE(BO), %xmm1
	mulsd	%xmm0, %xmm1
	mulsd	 6 * SIZE(BO), %xmm0
	addsd	%xmm1, %xmm10
	movsd	32 * SIZE(BO), %xmm1
	addsd	%xmm0, %xmm11
	movsd	-15 * SIZE(AO), %xmm0

	mulsd	%xmm0, %xmm3
	addsd	%xmm3, %xmm8
	movsd	10 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	addsd	%xmm3, %xmm9
	movsd	12 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	mulsd	14 * SIZE(BO), %xmm0
	addsd	%xmm3, %xmm10
	movsd	40 * SIZE(BO), %xmm3
	addsd	%xmm0, %xmm11
	movsd	-14 * SIZE(AO), %xmm0

	mulsd	%xmm0, %xmm5
	addsd	%xmm5, %xmm8
	movsd	18 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	addsd	%xmm5, %xmm9
	movsd	20 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	mulsd	22 * SIZE(BO), %xmm0
	addsd	%xmm5, %xmm10
	movsd	48 * SIZE(BO), %xmm5
	addsd	%xmm0, %xmm11
	movsd	-13 * SIZE(AO), %xmm0

	mulsd	%xmm0, %xmm7
	addsd	%xmm7, %xmm8
	movsd	26 * SIZE(BO), %xmm7
	mulsd	%xmm0, %xmm7
	addsd	%xmm7, %xmm9
	movsd	28 * SIZE(BO), %xmm7
	mulsd	%xmm0, %xmm7
	mulsd	30 * SIZE(BO), %xmm0
	addsd	%xmm7, %xmm10
	movsd	56 * SIZE(BO), %xmm7
	addsd	%xmm0, %xmm11
	movsd	-12 * SIZE(AO), %xmm0

	mulsd	%xmm0, %xmm1
	addsd	%xmm1, %xmm8
	movsd	34 * SIZE(BO), %xmm1
	mulsd	%xmm0, %xmm1
	addsd	%xmm1, %xmm9
	movsd	36 * SIZE(BO), %xmm1
	mulsd	%xmm0, %xmm1
	mulsd	38 * SIZE(BO), %xmm0
	addsd	%xmm1, %xmm10
	movsd	64 * SIZE(BO), %xmm1
	addsd	%xmm0, %xmm11
	movsd	-11 * SIZE(AO), %xmm0

	mulsd	%xmm0, %xmm3
	addsd	%xmm3, %xmm8
	movsd	42 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	addsd	%xmm3, %xmm9
	movsd	44 * SIZE(BO), %xmm3
	mulsd	%xmm0, %xmm3
	mulsd	46 * SIZE(BO), %xmm0
	addsd	%xmm3, %xmm10
	movsd	72 * SIZE(BO), %xmm3
	addsd	%xmm0, %xmm11
	movsd	-10 * SIZE(AO), %xmm0

	mulsd	%xmm0, %xmm5
	addsd	%xmm5, %xmm8
	movsd	50 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	addsd	%xmm5, %xmm9
	movsd	52 * SIZE(BO), %xmm5
	mulsd	%xmm0, %xmm5
	mulsd	54 * SIZE(BO), %xmm0
	addsd	%xmm5, %xmm10
	movsd	80 * SIZE(BO), %xmm5
	addsd	%xmm0, %xmm11
	movsd	-9 * SIZE(AO), %xmm0

	mulsd	%xmm0, %xmm7
	addsd	%xmm7, %xmm8
	movsd	58 * SIZE(BO), %xmm7
	mulsd	%xmm0, %xmm7
	addsd	%xmm7, %xmm9