zgemm_kernel_1x2_sse2.s

Optimized GotoBLAS libraries

	KERNEL1(16  *  6)
	KERNEL2(16  *  6)
	KERNEL3(16  *  6)
	KERNEL4(16  *  6)
	KERNEL5(16  *  6)
	KERNEL6(16  *  6)
	KERNEL7(16  *  6)
	KERNEL8(16  *  6)
	cmpl	$128 *  7, %eax
	jle	.L12
	KERNEL1(16  *  7)
	KERNEL2(16  *  7)
	KERNEL3(16  *  7)
	KERNEL4(16  *  7)
	KERNEL5(16  *  7)
	KERNEL6(16  *  7)
	KERNEL7(16  *  7)
	KERNEL8(16  *  7)

	addl	$128 * 4  * SIZE, BB
	addl	$128 * 1  * SIZE, AA
	subl	$128 * 8, %eax
	jg	.L1X
	jmp	.L15

.L12:
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 4), BB
	ALIGN_4
#else

	sarl	$3, %eax
	je	.L15
	ALIGN_4

.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)

	addl   $64 * SIZE, BB
	addl   $16 * SIZE, AA
	decl   %eax
	jne    .L11
	ALIGN_4
#endif

.L15:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L14
	ALIGN_4

.L13:
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm4
	movapd	 2 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	addpd	 %xmm2, %xmm5
	movapd	 4 * SIZE(BB), %xmm2
	mulpd	 %xmm0, %xmm2
	mulpd	 6 * SIZE(BB), %xmm0
	addpd	 %xmm2, %xmm6
	movapd	 8 * SIZE(BB), %xmm2
	addpd	 %xmm0, %xmm7
	movapd	 2 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L13
	ALIGN_4

.L14:
	movapd	POSINV,  %xmm1
	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(NR) || defined(NC) || defined(TR) || defined(TC)
	xorpd	%xmm1, %xmm5
	xorpd	%xmm1, %xmm7
#else
	xorpd	%xmm1, %xmm4
	xorpd	%xmm1, %xmm6
#endif

#ifndef TRMMKERNEL
	movlpd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0
	movlpd	0 * SIZE(%esi, LDC), %xmm1
	movhpd	1 * SIZE(%esi, LDC), %xmm1
#endif

#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

	pshufd	$0x4e, %xmm4, %xmm5
	pshufd	$0x4e, %xmm6, %xmm7
	
	mulpd	%xmm2, %xmm4
	mulpd	%xmm3, %xmm5
	mulpd	%xmm2, %xmm6
	mulpd	%xmm3, %xmm7

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

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

	movlpd	%xmm4, 0 * SIZE(%esi)
	movhpd	%xmm4, 1 * SIZE(%esi)
	movlpd	%xmm6, 0 * SIZE(%esi, LDC)
	movhpd	%xmm6, 1 * SIZE(%esi, LDC)

#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, 8), BB
#endif

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

	addl	$2 * SIZE, %esi		# coffset += 4
	decl	%ebx			# i --
	jg	.L10
	ALIGN_4	

.L99:
#if defined(TRMMKERNEL) && !defined(LEFT)
	addl	$2, KK
#endif

	leal	(, LDC, 2), %eax
	addl	%eax, C			# c += ldc
	decl	J			# j --
	jg	.L01
	ALIGN_4

.L100:
	movl	N, %eax
	andl	$1, %eax
	jle	.L500
	ALIGN_4

.L101:
#if defined(TRMMKERNEL) && defined(LEFT)
	movl	OFFSET, %eax
	movl	%eax, KK
#endif	

	leal	BUFFER, %ecx
	movapd	POSINV, %xmm7

	movl	K, %eax
	sarl	$2, %eax
	jle	.L103
	ALIGN_4

.L102:
	prefetchnta	 56 * SIZE(B)

	movlpd	 0 * SIZE(B), %xmm0
	movlpd	 1 * SIZE(B), %xmm1
	movlpd	 2 * SIZE(B), %xmm2
	movlpd	 3 * SIZE(B), %xmm3
	movlpd	 4 * SIZE(B), %xmm4
	movlpd	 5 * SIZE(B), %xmm5
	movlpd	 6 * SIZE(B), %xmm6
	movlpd	 7 * SIZE(B), %xmm7

	movlpd	%xmm0,  0 * SIZE(BB)
	movlpd	%xmm0,  1 * SIZE(BB)
	movlpd	%xmm1,  2 * SIZE(BB)
	movlpd	%xmm1,  3 * SIZE(BB)
	movlpd	%xmm2,  4 * SIZE(BB)
	movlpd	%xmm2,  5 * SIZE(BB)
	movlpd	%xmm3,  6 * SIZE(BB)
	movlpd	%xmm3,  7 * SIZE(BB)
	movlpd	%xmm4,  8 * SIZE(BB)
	movlpd	%xmm4,  9 * SIZE(BB)
	movlpd	%xmm5, 10 * SIZE(BB)
	movlpd	%xmm5, 11 * SIZE(BB)
	movlpd	%xmm6, 12 * SIZE(BB)
	movlpd	%xmm6, 13 * SIZE(BB)
	movlpd	%xmm7, 14 * SIZE(BB)
	movlpd	%xmm7, 15 * SIZE(BB)

	addl	$  8 * SIZE, B
	subl	$-16 * SIZE, %ecx
	decl	%eax
	jne	.L102
	ALIGN_4

.L103:
	movl	K, %eax
	andl	$3, %eax
	BRANCH
	jle	.L105
	ALIGN_4

.L104:
	movlpd	 0 * SIZE(B), %xmm0
	movlpd	 1 * SIZE(B), %xmm1

	movlpd	%xmm0,  0 * SIZE(BB)
	movlpd	%xmm0,  1 * SIZE(BB)
	movlpd	%xmm1,  2 * SIZE(BB)
	movlpd	%xmm1,  3 * SIZE(BB)

	addl	$2 * SIZE, B
	addl	$4 * SIZE, %ecx
	decl	%eax
	jne	.L104
	ALIGN_4

.L105:
	movl	C, %esi		# coffset = c
	movl	A, AA		# aoffset = a
	movl	M,  %ebx
	testl	%ebx, %ebx	
	jle	.L500
	ALIGN_4

.L110:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leal	BUFFER, BB
#else
	leal	BUFFER, BB
	movl	KK, %eax
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 4), BB
#endif	

	pxor	%xmm4, %xmm4
	pxor	%xmm5, %xmm5
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7

	movapd	 0 * SIZE(AA), %xmm0
	movapd	 8 * SIZE(AA), %xmm1
	movapd	 0 * SIZE(BB), %xmm2
	movapd	 8 * SIZE(BB), %xmm3

#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	$1, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L112
	ALIGN_4

.L111:
	PREFETCH  (PREFETCHSIZE + 0) * SIZE(AA)
	mulpd	%xmm0, %xmm2
	mulpd	 2 * SIZE(BB), %xmm0
	addpd	%xmm2, %xmm4
	movapd	 4 * SIZE(BB), %xmm2
	addpd	%xmm0, %xmm5
	movapd	 2 * SIZE(AA), %xmm0
	mulpd	%xmm0, %xmm2
	mulpd	 6 * SIZE(BB), %xmm0
	addpd	%xmm2, %xmm6
	movapd	16 * SIZE(BB), %xmm2
	addpd	%xmm0, %xmm7
	movapd	 4 * SIZE(AA), %xmm0
	mulpd	%xmm0, %xmm3
	mulpd	10 * SIZE(BB), %xmm0
	addpd	%xmm3, %xmm4
	movapd	12 * SIZE(BB), %xmm3
	addpd	%xmm0, %xmm5
	movapd	 6 * SIZE(AA), %xmm0
	mulpd	%xmm0, %xmm3
	mulpd	14 * SIZE(BB), %xmm0
	addpd	%xmm3, %xmm6
	movapd	24 * SIZE(BB), %xmm3
	addpd	%xmm0, %xmm7
	movapd	16 * SIZE(AA), %xmm0
	mulpd	%xmm1, %xmm2
	mulpd	18 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm4
	movapd	20 * SIZE(BB), %xmm2
	addpd	%xmm1, %xmm5
	movapd	10 * SIZE(AA), %xmm1
	mulpd	%xmm1, %xmm2
	mulpd	22 * SIZE(BB), %xmm1
	addpd	%xmm2, %xmm6
	movapd	32 * SIZE(BB), %xmm2
	addpd	%xmm1, %xmm7
	movapd	12 * SIZE(AA), %xmm1
	mulpd	%xmm1, %xmm3
	mulpd	26 * SIZE(BB), %xmm1
	addpd	%xmm3, %xmm4
	movapd	28 * SIZE(BB), %xmm3
	addpd	%xmm1, %xmm5
	movapd	14 * SIZE(AA), %xmm1
	mulpd	%xmm1, %xmm3
	mulpd	30 * SIZE(BB), %xmm1
	addpd	%xmm3, %xmm6
	movapd	40 * SIZE(BB), %xmm3
	addpd	%xmm1, %xmm7
	movapd	24 * SIZE(AA), %xmm1

	addl   $16 * SIZE, AA
	addl   $32 * SIZE, BB
	decl   %eax
	jne    .L111
	ALIGN_4

.L112:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L114
	ALIGN_4

.L113:
	mulpd	%xmm0, %xmm2
	mulpd	 2 * SIZE(BB), %xmm0
	addpd	%xmm2, %xmm4
	movapd	 4 * SIZE(BB), %xmm2
	addpd	%xmm0, %xmm5
	movapd	 2 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$4 * SIZE, BB
	decl	%eax
	jg	.L113
	ALIGN_4

.L114:
	movapd	POSINV,  %xmm1
	movapd	ALPHA_R, %xmm2
	movapd	ALPHA_I, %xmm3

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

	SHUFPD_1 %xmm5, %xmm5

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)
	xorpd	%xmm1, %xmm5
#else
	xorpd	%xmm1, %xmm4
#endif

#ifndef TRMMKERNEL
	movlpd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0
#endif

#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

	pshufd	$0x4e, %xmm4, %xmm5
	
	mulpd	%xmm2, %xmm4
	mulpd	%xmm3, %xmm5

	addpd	%xmm5, %xmm4

#ifndef TRMMKERNEL
	addpd	%xmm0, %xmm4
#endif

	movlpd	%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

	addl	$2 * SIZE, %esi		# coffset += 4
	decl	%ebx			# i --
	jg	.L110
	ALIGN_4	

.L500:
	movl	OLD_STACK, %esp

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

	EPILOGUE