zgemm_kernel_4x1_core2.s

Optimized GotoBLAS libraries

	shufps	$0xb1, %xmm7, %xmm7

	addsubps	%xmm5, %xmm4
	addsubps	%xmm7, %xmm6

	movaps	%xmm4, %xmm5
	movaps	%xmm6, %xmm7

	shufps	$0xb1, %xmm4, %xmm4
	shufps	$0xb1, %xmm6, %xmm6
#else
	shufps	$0xb1, %xmm4, %xmm4
	shufps	$0xb1, %xmm6, %xmm6

	addsubps	%xmm4, %xmm5
	addsubps	%xmm6, %xmm7

	movaps	%xmm5, %xmm4
	movaps	%xmm7, %xmm6

	shufps	$0xb1, %xmm5, %xmm5
	shufps	$0xb1, %xmm7, %xmm7
#endif

	mulps	%xmm0, %xmm5
	mulps	%xmm1, %xmm4

	mulps	%xmm0, %xmm7
	mulps	%xmm1, %xmm6

	addps	%xmm5, %xmm4
	addps	%xmm7, %xmm6

#if! defined(TRMMKERNEL) && !defined(BETAZERO)
	movsd	0 * SIZE(C1), %xmm2
	movhps	2 * SIZE(C1), %xmm2
	movsd	4 * SIZE(C1), %xmm3
	movhps	6 * SIZE(C1), %xmm3

	addps	%xmm2, %xmm4
	addps	%xmm3, %xmm6
#endif

	movsd	%xmm4, 0 * SIZE(C1)
	movhps	%xmm4, 2 * SIZE(C1)
	movsd	%xmm6, 4 * SIZE(C1)
	movhps	%xmm6, 6 * SIZE(C1)

#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, 4), AA
	leal	(BB, %eax, 4), BB
#endif

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

	addl	$8 * SIZE, C1
	decl	%ebx
	jg	.L10
	ALIGN_2
	
.L20:
	movl	M,  %ebx
	testl	$2, %ebx
	jle	.L30

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

	leal	32 * SIZE + BUFFER, BB
#else

	leal	32 * SIZE + BUFFER, BB
	movl	KK, %eax
	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 4), BB /* because it's doubled */
#endif	

	movaps	-32 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movaps	-32 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movaps	-16 * SIZE(AA), %xmm2
	pxor	%xmm6, %xmm6
	movaps	-16 * SIZE(BB), %xmm3
	pxor	%xmm7, %xmm7

#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	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L25
	ALIGN_4

.L22:
	mulps	%xmm0,  %xmm1
	mulps	-28 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm4
	movaps	-24 * SIZE(BB), %xmm1
	ADDSUB	%xmm0, %xmm5
	movaps	-28 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm1
	mulps	-20 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm6
	movaps	  0 * SIZE(BB), %xmm1
	ADDSUB	%xmm0, %xmm7
	movaps	-24 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm3
	mulps	-12 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm4
	movaps	 -8 * SIZE(BB), %xmm3
	ADDSUB	%xmm0, %xmm5
	movaps	-20 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm3
	mulps	 -4 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm6
	movaps	 16 * SIZE(BB), %xmm3
	ADDSUB	%xmm0, %xmm7
	movaps	  0 * SIZE(AA), %xmm0
	mulps	%xmm2,  %xmm1
	mulps	  4 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm4
	movaps	  8 * SIZE(BB), %xmm1
	ADDSUB	%xmm2, %xmm5
	movaps	-12 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm1
	mulps	 12 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm6
	movaps	 32 * SIZE(BB), %xmm1
	ADDSUB	%xmm2, %xmm7
	movaps	 -8 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm3
	mulps	 20 * SIZE(BB), %xmm2
	addps	%xmm3, %xmm4
	movaps	 24 * SIZE(BB), %xmm3
	ADDSUB	%xmm2, %xmm5
	movaps	 -4 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm3
	mulps	 28 * SIZE(BB), %xmm2
	addps	%xmm3, %xmm6
	movaps	 48 * SIZE(BB), %xmm3
	ADDSUB	%xmm2, %xmm7
	movaps	 16 * SIZE(AA), %xmm2

	subl   $-32 * SIZE, AA
	addl   $ 64 * SIZE, BB
	decl	%eax
	jne	.L22
	ALIGN_2

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

.L26:
	mulps	%xmm0,  %xmm1
	mulps	-28 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm4
	movaps	-24 * SIZE(BB), %xmm1
	ADDSUB	%xmm0, %xmm5
	movaps	-28 * SIZE(AA), %xmm0

	addl	$4 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L26

.L28:
	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

	movaps	ALPHA_R, %xmm0
	movaps	ALPHA_I, %xmm1

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)
	shufps	$0xb1, %xmm5, %xmm5
	addsubps	%xmm5, %xmm4
	movaps	%xmm4, %xmm5
	shufps	$0xb1, %xmm4, %xmm4
#else
	shufps	$0xb1, %xmm4, %xmm4
	addsubps	%xmm4, %xmm5
	movaps	%xmm5, %xmm4
	shufps	$0xb1, %xmm5, %xmm5
#endif

	mulps	%xmm0, %xmm5
	mulps	%xmm1, %xmm4

	addps	%xmm5, %xmm4

#if! defined(TRMMKERNEL) && !defined(BETAZERO)
	movsd	0 * SIZE(C1), %xmm2
	movhps	2 * SIZE(C1), %xmm2

	addps	%xmm2, %xmm4
#endif

	movsd	%xmm4, 0 * SIZE(C1)
	movhps	%xmm4, 2 * SIZE(C1)

#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, 4), BB
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$2, KK
#endif
	addl	$4 * SIZE, C1
	ALIGN_2

.L30:
	testl	$1, %ebx
	jle	.L39

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

	leal	32 * SIZE + BUFFER, BB
#else

	leal	32 * SIZE + BUFFER, BB
	movl	KK, %eax
	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 4), BB /* because it's doubled */
#endif	

	movsd	-32 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movsd	-32 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movsd	-24 * SIZE(AA), %xmm2
	pxor	%xmm6, %xmm6
	movsd	-16 * SIZE(BB), %xmm3
	pxor	%xmm7, %xmm7


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

.L32:
	mulps	%xmm0,  %xmm1
	mulps	-28 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm4
 	movsd	-24 * SIZE(BB), %xmm1
	ADDSUB	%xmm0, %xmm5
	movsd	-30 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm1
	mulps	-20 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm6
	movsd	  0 * SIZE(BB), %xmm1
	ADDSUB	%xmm0, %xmm7
	movsd	-28 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm3
	mulps	-12 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm4
	movsd	 -8 * SIZE(BB), %xmm3
	ADDSUB	%xmm0, %xmm5
	movsd	-26 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm3
	mulps	 -4 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm6
	movsd	 16 * SIZE(BB), %xmm3
	ADDSUB	%xmm0, %xmm7
	movsd	-16 * SIZE(AA), %xmm0
	mulps	%xmm2,  %xmm1
	mulps	  4 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm4
	movsd	  8 * SIZE(BB), %xmm1
	ADDSUB	%xmm2, %xmm5
	movsd	-22 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm1
	mulps	 12 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm6
	movsd	 32 * SIZE(BB), %xmm1
	ADDSUB	%xmm2, %xmm7
	movsd	-20 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm3
	mulps	 20 * SIZE(BB), %xmm2
	addps	%xmm3, %xmm4
	movsd	 24 * SIZE(BB), %xmm3
	ADDSUB	%xmm2, %xmm5
	movsd	-18 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm3
	mulps	 28 * SIZE(BB), %xmm2
	addps	%xmm3, %xmm6
	movsd	 48 * SIZE(BB), %xmm3
	ADDSUB	%xmm2, %xmm7
	movsd	 -8 * SIZE(AA), %xmm2

	subl   $-16 * SIZE, AA
	addl   $ 64 * SIZE, BB

	decl	%eax
	jne	.L32
	ALIGN_2

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

.L36:
	mulps	%xmm0,  %xmm1
	mulps	-28 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm4
	movsd	-24 * SIZE(BB), %xmm1
	ADDSUB	%xmm0, %xmm5
	movsd	-30 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L36

.L38:
	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

	movaps	ALPHA_R, %xmm0
	movaps	ALPHA_I, %xmm1

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)
	shufps	$0xb1, %xmm5, %xmm5
	addsubps	%xmm5, %xmm4
	movaps	%xmm4, %xmm5
	shufps	$0xb1, %xmm4, %xmm4
#else
	shufps	$0xb1, %xmm4, %xmm4
	addsubps	%xmm4, %xmm5
	movaps	%xmm5, %xmm4
	shufps	$0xb1, %xmm5, %xmm5
#endif

	mulps	%xmm0, %xmm5
	mulps	%xmm1, %xmm4

	addps	%xmm5, %xmm4

#if! defined(TRMMKERNEL) && !defined(BETAZERO)
	movsd	0 * SIZE(C1), %xmm2
	addps	%xmm2, %xmm4
#endif

	movsd	%xmm4, 0 * SIZE(C1)

#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	$1, KK
#endif
	ALIGN_2

.L39:
#if defined(TRMMKERNEL) && !defined(LEFT)
	addl	$1, KK
#endif

	addl	LDC, C			# c += ldc
	decl	J			# j --
	jg	.L01
	ALIGN_2

.L999:
	movl	OLD_STACK, %esp
	
	EMMS

	popl	%ebx
	popl	%esi
	popl	%edi
	popl	%ebp
	ret

	EPILOGUE