zgemm_kernel_4x2_core2.s

Optimized GotoBLAS libraries

	
.L50:
	movq	C, CO1			# coffset1 = c
	movq	A, AO		# aoffset = a

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

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

	leaq	32 * SIZE + BUFFER, BO
#else
	leaq	32 * SIZE + BUFFER, BO
	movq	KK, %rax
	leaq	(, %rax, 8), %rax
	leaq	(AO, %rax, 4), AO
	leaq	(BO, %rax, 4), BO
#endif	

	pxor	%xmm8, %xmm8
	prefetcht0     3 * SIZE(CO1)
	pxor	%xmm9, %xmm9
	pxor	%xmm12, %xmm12
	pxor	%xmm13, %xmm13

#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	$1, %rax
#endif
	movq	%rax, KKK
#endif
	sarq	$2, %rax
	je	.L55
	ALIGN_4

.L52:
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)

	movaps	-32 * SIZE(AO), %xmm0
	movaps	-28 * SIZE(AO), %xmm1

	movaps	-32 * SIZE(BO), %xmm2
	movaps	 %xmm2, %xmm3
	movaps	-28 * SIZE(BO), %xmm4
	movaps	 %xmm4, %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm1, %xmm3
	mulps	%xmm0, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	addps	%xmm3, %xmm12
	ADDSUB	%xmm4, %xmm9
	ADDSUB	%xmm5, %xmm13

	movaps	-24 * SIZE(AO), %xmm0
	movaps	-20 * SIZE(AO), %xmm1

	movaps	-24 * SIZE(BO), %xmm2
	movaps	 %xmm2, %xmm3
	movaps	-20 * SIZE(BO), %xmm4
	movaps	 %xmm4, %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm1, %xmm3
	mulps	%xmm0, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	addps	%xmm3, %xmm12
	ADDSUB	%xmm4, %xmm9
	ADDSUB	%xmm5, %xmm13

	movaps	-16 * SIZE(AO), %xmm0
	movaps	-12 * SIZE(AO), %xmm1

	movaps	-16 * SIZE(BO), %xmm2
	movaps	 %xmm2, %xmm3
	movaps	-12 * SIZE(BO), %xmm4
	movaps	 %xmm4, %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm1, %xmm3
	mulps	%xmm0, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	addps	%xmm3, %xmm12
	ADDSUB	%xmm4, %xmm9
	ADDSUB	%xmm5, %xmm13

	movaps	 -8 * SIZE(AO), %xmm0
	movaps	 -4 * SIZE(AO), %xmm1

	movaps	 -8 * SIZE(BO), %xmm2
	movaps	 %xmm2, %xmm3
	movaps	 -4 * SIZE(BO), %xmm4
	movaps	 %xmm4, %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm1, %xmm3
	mulps	%xmm0, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	addps	%xmm3, %xmm12
	ADDSUB	%xmm4, %xmm9
	ADDSUB	%xmm5, %xmm13

	subq	$-32 * SIZE, AO
	subq	$-32 * SIZE, BO
	subq	$1, %rax
	jne    .L52
	ALIGN_4

.L55:
#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	BRANCH
	je .L58
	ALIGN_4

.L56:
	movaps	-32 * SIZE(AO), %xmm0
	movaps	-28 * SIZE(AO), %xmm1

	movaps	-32 * SIZE(BO), %xmm2
	movaps	 %xmm2, %xmm3
	movaps	-28 * SIZE(BO), %xmm4
	movaps	 %xmm4, %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm1, %xmm3
	mulps	%xmm0, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	addps	%xmm3, %xmm12
	ADDSUB	%xmm4, %xmm9
	ADDSUB	%xmm5, %xmm13

	addq	$8 * SIZE, AO
	addq	$8 * SIZE, BO
	subq	$1, %rax
	jg	.L56
	ALIGN_4

.L58:
	movaps	ALPHA_R, %xmm6
	movaps	ALPHA_I, %xmm7

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)

	shufps	$0xb1, %xmm9, %xmm9
	shufps	$0xb1, %xmm13, %xmm13

	addsubps	%xmm9, %xmm8
	addsubps	%xmm13, %xmm12

	movaps	%xmm8, %xmm9
	movaps	%xmm12, %xmm13

	shufps	$0xb1, %xmm8, %xmm8
	shufps	$0xb1, %xmm12, %xmm12
#else
	shufps	$0xb1, %xmm8, %xmm8
	shufps	$0xb1, %xmm12, %xmm12

	addsubps	%xmm8, %xmm9
	addsubps	%xmm12, %xmm13

	movaps	%xmm9, %xmm8
	movaps	%xmm13, %xmm12

	shufps	$0xb1, %xmm9, %xmm9
	shufps	$0xb1, %xmm13, %xmm13
#endif

	mulps	%xmm6, %xmm9
	mulps	%xmm7, %xmm8
	mulps	%xmm6, %xmm13
	mulps	%xmm7, %xmm12

	addps	%xmm9, %xmm8
	addps	%xmm13, %xmm12

#if! defined(TRMMKERNEL) && !defined(BETAZERO)
	movsd	0 * SIZE(CO1), %xmm0
	movhps	2 * SIZE(CO1), %xmm0
	movsd	4 * SIZE(CO1), %xmm1
	movhps	6 * SIZE(CO1), %xmm1

	addps	%xmm0, %xmm8
	addps	%xmm1, %xmm12
#endif

	movsd	%xmm8, 0 * SIZE(CO1)
	movhps	%xmm8, 2 * SIZE(CO1)
	movsd	%xmm12, 4 * SIZE(CO1)
	movhps	%xmm12, 6 * SIZE(CO1)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movq	K, %rax
	subq	KKK, %rax
	leaq	(,%rax, 8), %rax
	leaq	(AO, %rax, 4), AO
	leaq	(BO, %rax, 4), BO
#endif

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

	addq	$8 * SIZE, CO1		# coffset += 4
	decq	I			# i --
	jg	.L51
	ALIGN_4	

.L60:
	testq	$2, M
	je	.L70

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

	leaq	32 * SIZE + BUFFER, BO
#else
	leaq	32 * SIZE + BUFFER, BO
	movq	KK, %rax
	leaq	(, %rax, 8), %rax
	leaq	(AO, %rax, 2), AO
	leaq	(BO, %rax, 4), BO
#endif	

	pxor	%xmm8, %xmm8
	pxor	%xmm9, %xmm9
	pxor	%xmm10, %xmm10
	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	$2, %rax
#else
	addq	$1, %rax
#endif
	movq	%rax, KKK
#endif
	sarq	$2, %rax
	je	.L65
	ALIGN_4

.L62:
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)

	movaps	-32 * SIZE(AO), %xmm0
	movaps	-28 * SIZE(AO), %xmm1
	movaps	-32 * SIZE(BO), %xmm2
	movaps	-28 * SIZE(BO), %xmm3
	movaps	-24 * SIZE(BO), %xmm4
	movaps	-20 * SIZE(BO), %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm0, %xmm3
	mulps	%xmm1, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	ADDSUB	%xmm3, %xmm9
	addps	%xmm4, %xmm10
	ADDSUB	%xmm5, %xmm11

	movaps	-24 * SIZE(AO), %xmm0
	movaps	-20 * SIZE(AO), %xmm1
	movaps	-16 * SIZE(BO), %xmm2
	movaps	-12 * SIZE(BO), %xmm3
	movaps	 -8 * SIZE(BO), %xmm4
	movaps	 -4 * SIZE(BO), %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm0, %xmm3
	mulps	%xmm1, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	ADDSUB	%xmm3, %xmm9
	addps	%xmm4, %xmm10
	ADDSUB	%xmm5, %xmm11

	subq	$-16 * SIZE, AO
	subq	$-32 * SIZE, BO
	subq	$1, %rax
	jne    .L62
	ALIGN_4

.L65:
#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	BRANCH
	je .L68
	ALIGN_4

.L66:
	movaps	-32 * SIZE(AO), %xmm0
	movaps	-32 * SIZE(BO), %xmm2
	movaps	-28 * SIZE(BO), %xmm3

	mulps	%xmm0, %xmm2
	mulps	%xmm0, %xmm3

	addps	%xmm2, %xmm8
	ADDSUB	%xmm3, %xmm9

	addq	$4 * SIZE, AO
	addq	$8 * SIZE, BO
	subq	$1, %rax
	jg	.L66
	ALIGN_4

.L68:
	movaps	ALPHA_R, %xmm6
	movaps	ALPHA_I, %xmm7

	addps	%xmm10, %xmm8
	addps	%xmm11, %xmm9

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)
	shufps	$0xb1, %xmm9, %xmm9
	addsubps	%xmm9, %xmm8
	movaps	%xmm8, %xmm9
	shufps	$0xb1, %xmm8, %xmm8
#else
	shufps	$0xb1, %xmm8, %xmm8
	addsubps	%xmm8, %xmm9
	movaps	%xmm9, %xmm8
	shufps	$0xb1, %xmm9, %xmm9
#endif

	mulps	%xmm6, %xmm9
	mulps	%xmm7, %xmm8
	addps	%xmm9, %xmm8

#if! defined(TRMMKERNEL) && !defined(BETAZERO)
	movsd	0 * SIZE(CO1), %xmm0
	movhps	2 * SIZE(CO1), %xmm0

	addps	%xmm0, %xmm8
#endif

	movsd	%xmm8, 0 * SIZE(CO1)
	movhps	%xmm8, 2 * SIZE(CO1)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movq	K, %rax
	subq	KKK, %rax
	leaq	(,%rax,  8), %rax
	leaq	(AO, %rax, 2), AO
	leaq	(BO, %rax, 4), BO
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addq	$2, KK
#endif
	addq	$4 * SIZE, CO1		# coffset += 4
	ALIGN_4	
	
.L70:
	testq	$1, M
	je	.L999

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

	leaq	32 * SIZE + BUFFER, BO
#else
	leaq	32 * SIZE + BUFFER, BO
	movq	KK, %rax
	leaq	(, %rax, 8), %rax
	leaq	(AO, %rax, 1), AO
	leaq	(BO, %rax, 4), BO
#endif	

	pxor	%xmm8, %xmm8
	pxor	%xmm9, %xmm9
	pxor	%xmm10, %xmm10
	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	$1, %rax
#endif
	movq	%rax, KKK
#endif
	sarq	$2, %rax
	je	.L75
	ALIGN_4

.L72:
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)

	movsd	-32 * SIZE(AO), %xmm0
	movsd	-30 * SIZE(AO), %xmm1
	movsd	-32 * SIZE(BO), %xmm2
	movsd	-28 * SIZE(BO), %xmm3
	movsd	-24 * SIZE(BO), %xmm4
	movsd	-20 * SIZE(BO), %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm0, %xmm3
	mulps	%xmm1, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	ADDSUB	%xmm3, %xmm9
	addps	%xmm4, %xmm10
	ADDSUB	%xmm5, %xmm11

	movsd	-28 * SIZE(AO), %xmm0
	movsd	-26 * SIZE(AO), %xmm1
	movsd	-16 * SIZE(BO), %xmm2
	movsd	-12 * SIZE(BO), %xmm3
	movsd	 -8 * SIZE(BO), %xmm4
	movsd	 -4 * SIZE(BO), %xmm5

	mulps	%xmm0, %xmm2
	mulps	%xmm0, %xmm3
	mulps	%xmm1, %xmm4
	mulps	%xmm1, %xmm5

	addps	%xmm2, %xmm8
	ADDSUB	%xmm3, %xmm9
	addps	%xmm4, %xmm10
	ADDSUB	%xmm5, %xmm11

	subq	$ -8 * SIZE, AO
	subq	$-32 * SIZE, BO
	subq	$1, %rax
	jne    .L72
	ALIGN_4

.L75:
#ifndef TRMMKERNEL
	movq	K, %rax
#else
	movq	KKK, %rax
#endif
	andq	$3, %rax
	BRANCH
	je .L78
	ALIGN_4

.L76:
	movsd	-32 * SIZE(AO), %xmm0
	movsd	-32 * SIZE(BO), %xmm2
	movsd	-28 * SIZE(BO), %xmm3

	mulps	%xmm0, %xmm2
	mulps	%xmm0, %xmm3

	addps	%xmm2, %xmm8
	ADDSUB	%xmm3, %xmm9

	addq	$2 * SIZE, AO
	addq	$8 * SIZE, BO
	subq	$1, %rax
	jg	.L76
	ALIGN_4

.L78:
	movaps	ALPHA_R, %xmm6
	movaps	ALPHA_I, %xmm7

	addps	%xmm10, %xmm8
	addps	%xmm11, %xmm9

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)
	shufps	$0xb1, %xmm9, %xmm9
	addsubps	%xmm9, %xmm8
	movaps	%xmm8, %xmm9
	shufps	$0xb1, %xmm8, %xmm8
#else
	shufps	$0xb1, %xmm8, %xmm8
	addsubps	%xmm8, %xmm9
	movaps	%xmm9, %xmm8
	shufps	$0xb1, %xmm9, %xmm9
#endif

	mulps	%xmm6, %xmm9
	mulps	%xmm7, %xmm8
	addps	%xmm9, %xmm8

#if! defined(TRMMKERNEL) && !defined(BETAZERO)
	movsd	0 * SIZE(CO1), %xmm0

	addps	%xmm0, %xmm8
#endif

	movsd	%xmm8, 0 * SIZE(CO1)
	ALIGN_4	

.L999:
	movq	%r15, %rsp

	movq	  0(%rsp), %rbx
	movq	  8(%rsp), %rbp
	movq	 16(%rsp), %r12
	movq	 24(%rsp), %r13
	movq	 32(%rsp), %r14
	movq	 40(%rsp), %r15

#ifdef WINDOWS_ABI
	movq	 48(%rsp), %rdi
	movq	 56(%rsp), %rsi
	movups	 64(%rsp), %xmm6
	movups	 80(%rsp), %xmm7
	movups	 96(%rsp), %xmm8
	movups	112(%rsp), %xmm9
	movups	128(%rsp), %xmm10
	movups	144(%rsp), %xmm11
	movups	160(%rsp), %xmm12
	movups	176(%rsp), %xmm13
	movups	192(%rsp), %xmm14
	movups	208(%rsp), %xmm15
#endif

	addq	$STACKSIZE, %rsp
	ret

	EPILOGUE