zgemm_kernel_2x2_sse2.s

Optimized GotoBLAS libraries

	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
	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	POSINV,  %xmm5
	movapd	ALPHA_R, %xmm6
	movapd	ALPHA_I, %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
	movlpd	0 * SIZE(CO1), %xmm0
	movhpd	1 * SIZE(CO1), %xmm0
	movlpd	2 * SIZE(CO1), %xmm2
	movhpd	3 * SIZE(CO1), %xmm2

	movlpd	0 * SIZE(CO2), %xmm1
	movhpd	1 * SIZE(CO2), %xmm1
	movlpd	2 * SIZE(CO2), %xmm3
	movhpd	3 * SIZE(CO2), %xmm3
#endif

	SHUFPD_1 %xmm9, %xmm9
	SHUFPD_1 %xmm11, %xmm11
	SHUFPD_1 %xmm13, %xmm13
	SHUFPD_1 %xmm15, %xmm15

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)
	xorpd	%xmm5, %xmm9
	xorpd	%xmm5, %xmm11
	xorpd	%xmm5, %xmm13
	xorpd	%xmm5, %xmm15
#else
	xorpd	%xmm5, %xmm8
	xorpd	%xmm5, %xmm10
	xorpd	%xmm5, %xmm12
	xorpd	%xmm5, %xmm14
#endif

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(RR) || defined(RC) || defined(CR) || defined(CC)
	subpd	%xmm9, %xmm8
	subpd	%xmm11, %xmm10
	subpd	%xmm13, %xmm12
	subpd	%xmm15, %xmm14
#else
	addpd	%xmm9, %xmm8
	addpd	%xmm11, %xmm10
	addpd	%xmm13, %xmm12
	addpd	%xmm15, %xmm14
#endif

	pshufd	$0x4e, %xmm8, %xmm9
	pshufd	$0x4e, %xmm10, %xmm11
	pshufd	$0x4e, %xmm12, %xmm13
	pshufd	$0x4e, %xmm14, %xmm15
	
	mulpd	%xmm6, %xmm8
	mulpd	%xmm7, %xmm9
	mulpd	%xmm6, %xmm10
	mulpd	%xmm7, %xmm11

	mulpd	%xmm6, %xmm12
	mulpd	%xmm7, %xmm13
	mulpd	%xmm6, %xmm14
	mulpd	%xmm7, %xmm15

	addpd	%xmm9, %xmm8
	addpd	%xmm11, %xmm10
	addpd	%xmm13, %xmm12
	addpd	%xmm15, %xmm14

#ifndef TRMMKERNEL
	addpd	%xmm0,  %xmm8
	addpd	%xmm2, %xmm12
	addpd	%xmm1,  %xmm10
	addpd	%xmm3, %xmm14
#endif

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

	movlpd	%xmm10, 0 * SIZE(CO2)
	movhpd	%xmm10, 1 * SIZE(CO2)
	movlpd	%xmm14, 2 * SIZE(CO2)
	movhpd	%xmm14, 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, 8), BO
#endif

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

	addq	$4 * SIZE, CO1		# coffset += 4
	addq	$4 * SIZE, CO2		# coffset += 4
	decq	I			# i --
	jg	.L10
	ALIGN_4	

.L30:
	testq	$1, M
	jle	.L99

#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, 8), BO
#endif	

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

#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	$2, %rax
#endif
	movq	%rax, KKK
#endif
	sarq	$3, %rax
	je	.L44
	ALIGN_4	

.L41:
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm8
	movapd	-14 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	PREFETCH	(PREFETCHSIZE     +  0) * SIZE(AO)
	addpd	%xmm1, %xmm9
	movapd	-12 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	mulpd	-10 * SIZE(BO), %xmm0
	addpd	%xmm1, %xmm10
	movapd	  0 * SIZE(BO), %xmm1
	addpd	%xmm0, %xmm11
	movapd	-14 * SIZE(AO), %xmm0
	mulpd	%xmm0, %xmm3
	addpd	%xmm3, %xmm8
	movapd	 -6 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	addpd	%xmm3, %xmm9
	movapd	 -4 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	mulpd	 -2 * SIZE(BO), %xmm0
	addpd	%xmm3, %xmm10
	movapd	  8 * SIZE(BO), %xmm3
	addpd	%xmm0, %xmm11
	movapd	-12 * SIZE(AO), %xmm0
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm8
	movapd	  2 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	PREFETCH	(PREFETCHSIZE     +  8) * SIZE(AO)
	addpd	%xmm1, %xmm9
	movapd	  4 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	mulpd	  6 * SIZE(BO), %xmm0
	addpd	%xmm1, %xmm10
	movapd	 16 * SIZE(BO), %xmm1
	addpd	%xmm0, %xmm11
	movapd	-10 * 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	 24 * SIZE(BO), %xmm3
	addpd	%xmm0, %xmm11
	movapd	  0 * SIZE(AO), %xmm0
	mulpd	%xmm2, %xmm1
	addpd	%xmm1, %xmm8
	movapd	 18 * SIZE(BO), %xmm1
	mulpd	%xmm2, %xmm1
	addpd	%xmm1, %xmm9
	movapd	 20 * SIZE(BO), %xmm1
	mulpd	%xmm2, %xmm1
	mulpd	 22 * SIZE(BO), %xmm2
	addpd	%xmm1, %xmm10
	movapd	 32 * SIZE(BO), %xmm1
	addpd	%xmm2, %xmm11
	movapd	 -6 * SIZE(AO), %xmm2
	mulpd	%xmm2, %xmm3
	addpd	%xmm3, %xmm8
	movapd	 26 * SIZE(BO), %xmm3
	mulpd	%xmm2, %xmm3
	addpd	%xmm3, %xmm9
	movapd	 28 * SIZE(BO), %xmm3
	mulpd	%xmm2, %xmm3
	mulpd	 30 * SIZE(BO), %xmm2
	addpd	%xmm3, %xmm10
	movapd	 40 * SIZE(BO), %xmm3
	addpd	%xmm2, %xmm11
	movapd	 -4 * SIZE(AO), %xmm2
	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	 48 * SIZE(BO), %xmm1
	addpd	%xmm2, %xmm11
	movapd	 -2 * 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	 56 * SIZE(BO), %xmm3
	addpd	%xmm2, %xmm11
	movapd	  8 * SIZE(AO), %xmm2

	subq   $-16 * SIZE, AO
	addq   $64 * SIZE, BO
	decq   %rax
	jne    .L41
	ALIGN_4	

.L44:
#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$4, %rax
	BRANCH
	jle .L45

	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm8
	movapd	-14 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm9
	movapd	-12 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	mulpd	-10 * SIZE(BO), %xmm0
	addpd	%xmm1, %xmm10
	movapd	  0 * SIZE(BO), %xmm1
	addpd	%xmm0, %xmm11
	movapd	-14 * SIZE(AO), %xmm0
	mulpd	%xmm0, %xmm3
	addpd	%xmm3, %xmm8
	movapd	 -6 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	addpd	%xmm3, %xmm9
	movapd	 -4 * SIZE(BO), %xmm3
	mulpd	%xmm0, %xmm3
	mulpd	 -2 * SIZE(BO), %xmm0
	addpd	%xmm3, %xmm10
	movapd	  8 * SIZE(BO), %xmm3
	addpd	%xmm0, %xmm11
	movapd	-12 * SIZE(AO), %xmm0
	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	 16 * SIZE(BO), %xmm1
	addpd	%xmm0, %xmm11
	movapd	-10 * 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	 24 * SIZE(BO), %xmm3
	addpd	%xmm0, %xmm11
	movapd	 -8 * SIZE(AO), %xmm0

	addq   $ 8 * SIZE, AO
	addq   $32 * SIZE, BO
	ALIGN_4	

.L45:
#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	movapd	POSINV,  %xmm5
	movapd	ALPHA_R, %xmm6
	movapd	ALPHA_I, %xmm7
	andq	$3, %rax		# if (k & 1)
	BRANCH
	jle .L47
	ALIGN_4	

.L46:
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm8
	movapd	-14 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm9
	movapd	-12 * SIZE(BO), %xmm1
	mulpd	%xmm0, %xmm1
	mulpd	-10 * SIZE(BO), %xmm0
	addpd	%xmm1, %xmm10
	movapd	 -8 * SIZE(BO), %xmm1
	addpd	%xmm0, %xmm11
	movapd	-14 * SIZE(AO), %xmm0

	addq	$2 * SIZE, AO
	addq	$8 * SIZE, BO

	decq	%rax
	jg	.L46
	ALIGN_4

.L47:
#ifndef TRMMKERNEL
	movlpd	0 * SIZE(CO1), %xmm0
	movhpd	1 * SIZE(CO1), %xmm0