zgemm_kernel_2x1_sse2.s

Optimized GotoBLAS libraries

	movapd	 0 * SIZE(BB), %xmm2
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE(BB), %xmm3
	pxor	%xmm6, %xmm6
	movapd	 8 * SIZE(AA), %xmm1
	pxor	%xmm7, %xmm7

#endif	

	prefetchnta	 3 * SIZE(%esi)

#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
	andl	$-8, %eax
	je,pn	.L12
	sall	$3, %eax
	
.L1X:
	KERNEL1(32  *  0)
	KERNEL2(32  *  0)
	KERNEL3(32  *  0)
	KERNEL4(32  *  0)
	KERNEL5(32  *  0)
	KERNEL6(32  *  0)
	KERNEL7(32  *  0)
	KERNEL8(32  *  0)
	cmpl	$64 *  1, %eax
	jle,pn	.L11
	KERNEL1(32  *  1)
	KERNEL2(32  *  1)
	KERNEL3(32  *  1)
	KERNEL4(32  *  1)
	KERNEL5(32  *  1)
	KERNEL6(32  *  1)
	KERNEL7(32  *  1)
	KERNEL8(32  *  1)
	cmpl	$64 *  2, %eax
	jle,pn	.L11
	KERNEL1(32  *  2)
	KERNEL2(32  *  2)
	KERNEL3(32  *  2)
	KERNEL4(32  *  2)
	KERNEL5(32  *  2)
	KERNEL6(32  *  2)
	KERNEL7(32  *  2)
	KERNEL8(32  *  2)
	cmpl	$64 *  3, %eax
	jle,pn	.L11
	KERNEL1(32  *  3)
	KERNEL2(32  *  3)
	KERNEL3(32  *  3)
	KERNEL4(32  *  3)
	KERNEL5(32  *  3)
	KERNEL6(32  *  3)
	KERNEL7(32  *  3)
	KERNEL8(32  *  3)
	cmpl	$64 *  4, %eax
	jle,pn	.L11
	KERNEL1(32  *  4)
	KERNEL2(32  *  4)
	KERNEL3(32  *  4)
	KERNEL4(32  *  4)
	KERNEL5(32  *  4)
	KERNEL6(32  *  4)
	KERNEL7(32  *  4)
	KERNEL8(32  *  4)
	cmpl	$64 *  5, %eax
	jle,pn	.L11
	KERNEL1(32  *  5)
	KERNEL2(32  *  5)
	KERNEL3(32  *  5)
	KERNEL4(32  *  5)
	KERNEL5(32  *  5)
	KERNEL6(32  *  5)
	KERNEL7(32  *  5)
	KERNEL8(32  *  5)
	cmpl	$64 *  6, %eax
	jle,pn	.L11
	KERNEL1(32  *  6)
	KERNEL2(32  *  6)
	KERNEL3(32  *  6)
	KERNEL4(32  *  6)
	KERNEL5(32  *  6)
	KERNEL6(32  *  6)
	KERNEL7(32  *  6)
	KERNEL8(32  *  6)
	cmpl	$64 *  7, %eax
	jle,pn	.L11
	KERNEL1(32  *  7)
	KERNEL2(32  *  7)
	KERNEL3(32  *  7)
	KERNEL4(32  *  7)
	KERNEL5(32  *  7)
	KERNEL6(32  *  7)
	KERNEL7(32  *  7)
	KERNEL8(32  *  7)

	addl	$64 * 4  * SIZE, AA
	addl	$64 * 4  * SIZE, BB
	subl	$64 * 8, %eax
	jg,pt	.L1X

.L11:
	leal	(BB, %eax, 4), BB
	leal	(AA, %eax, 4), AA
	
.L12:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L14

.L13:
	movapd	 2 * SIZE(BB), %xmm1
	mulpd	%xmm0, %xmm2
	addpd	%xmm2, %xmm4
	movapd	 0 * SIZE(BB), %xmm2
	mulpd	%xmm0, %xmm1
	movapd	 2 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm5
	movapd	 2 * SIZE(BB), %xmm1
	mulpd	%xmm0, %xmm2
	addpd	%xmm2, %xmm6
	movapd	 4 * SIZE(BB), %xmm2
	mulpd	%xmm0, %xmm1
	movapd	 4 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm7

	addl	$4 * SIZE, AA		# aoffset  += 8
	addl	$4 * SIZE, BB		# boffset1 += 8
	subl	$1, %eax
	jg	.L13

.L14:
	movapd	ALPHA_R, %xmm2
	movapd	ALPHA_I, %xmm3

	SHUFPD_1 %xmm5, %xmm5
	SHUFPD_1 %xmm7, %xmm7

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(RR) || defined(RC) || defined(CR) || defined(CC)
	subpd	%xmm5, %xmm4
	subpd	%xmm7, %xmm6
#else
	addpd	%xmm5, %xmm4
	addpd	%xmm7, %xmm6
#endif

	movapd	%xmm4, %xmm5
	movapd	%xmm6, %xmm7

	SHUFPD_1 %xmm4, %xmm4
	SHUFPD_1 %xmm6, %xmm6

	mulpd	%xmm2, %xmm5
	mulpd	%xmm3, %xmm4

	mulpd	%xmm2, %xmm7
	mulpd	%xmm3, %xmm6

	addpd	%xmm5, %xmm4
	addpd	%xmm7, %xmm6

#ifndef TRMMKERNEL
	movsd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0
	movsd	2 * SIZE(%esi), %xmm1
	movhpd	3 * SIZE(%esi), %xmm1

	addpd	%xmm0, %xmm4
	addpd	%xmm1, %xmm6
#endif

	movsd	%xmm4, 0 * SIZE(%esi)
	movhpd	%xmm4, 1 * SIZE(%esi)
	movsd	%xmm6, 2 * SIZE(%esi)
	movhpd	%xmm6, 3 * SIZE(%esi)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movl	K, %eax
	subl	KKK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
	leal	(BB, %eax, 4), BB
#endif

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

	addl	$4 * SIZE, %esi		# coffset += 4
	decl	%ebx			# i --
	jg	.L10

.L50:
	movl	M,  %ebx
	testl	$1, %ebx
	je	.L99

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

	leal	BUFFER, %ecx

	movapd	 0 * SIZE + BUFFER, %xmm1
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE + BUFFER, %xmm2
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7
#else

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

	movapd	 0 * SIZE(BB), %xmm1
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE(BB), %xmm2
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7
#endif	


#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	$2, %eax		# l = (k >> 2)
	jle	.L52

.L51:
	mulpd	%xmm0, %xmm1
	movapd	 2 * SIZE(BB), %xmm3
	addpd	%xmm1, %xmm4
	movapd	16 * SIZE(BB), %xmm1

	mulpd	%xmm0, %xmm3
	movapd	 2 * SIZE(AA), %xmm0
	addpd	%xmm3, %xmm5
	movapd	 4 * SIZE(BB), %xmm3

	mulpd	%xmm0, %xmm3
	mulpd	 6 * SIZE(BB), %xmm0
	addpd	%xmm3, %xmm4

	addpd	%xmm0, %xmm5
	movapd	 4 * SIZE(AA), %xmm0
	mulpd	%xmm0, %xmm2
	mulpd	10 * SIZE(BB), %xmm0

	addpd	%xmm2, %xmm4
	addpd	%xmm0, %xmm5
	movapd	6 * SIZE(AA), %xmm0

	movapd	12 * SIZE(BB), %xmm2
	mulpd	%xmm0, %xmm2
	addpd	%xmm2, %xmm4
	movapd	24 * SIZE(BB), %xmm2

	mulpd	14 * SIZE(BB), %xmm0
	addpd	%xmm0, %xmm5
	movapd	 8 * SIZE(AA), %xmm0

	addl	$ 8 * SIZE, AA	# aoffset  += 2
	addl	$16 * SIZE, BB	# boffset1 += 4

	decl	%eax			# l--
	jg	.L51
	ALIGN_2

.L52:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$3, %eax		# l = (k & 3)
	jle	.L54
	ALIGN_2

.L53:
	movapd	0 * SIZE(BB), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	movapd	2 * SIZE(BB), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm5
	movapd	2 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA		# aoffset  += 2
	addl	$4 * SIZE, BB		# boffset1 += 4
	decl	%eax			# l--
	jg	.L53

.L54:
	movapd	ALPHA_R, %xmm2
	movapd	ALPHA_I, %xmm3

	SHUFPD_1 %xmm5, %xmm5

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(RR) || defined(RC) || defined(CR) || defined(CC)
	subpd	%xmm5, %xmm4
#else
	addpd	%xmm5, %xmm4
#endif

	movapd	%xmm4, %xmm5

	SHUFPD_1 %xmm4, %xmm4

	mulpd	%xmm2, %xmm5
	mulpd	%xmm3, %xmm4

	addpd	%xmm5, %xmm4

#ifndef TRMMKERNEL
	SHUFPD_2 %xmm4, %xmm4

	movsd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0

	addpd	%xmm0, %xmm4
#endif

	movsd	%xmm4, 0 * SIZE(%esi)
	movhpd	%xmm4, 1 * SIZE(%esi)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movl	K, %eax
	subl	KKK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 4), BB
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$1, KK
#endif
	ALIGN_2

.L99:
#if defined(TRMMKERNEL) && !defined(LEFT)
	addl	$1, KK
#endif

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

.L999:
	movl	OLD_STACK, %esp

	EMMS

	popl	%ebx
	popl	%esi
	popl	%edi
	popl	%ebp
	ret
	ALIGN_2

	EPILOGUE