zgemm_kernel_2x2_sse.s

Optimized GotoBLAS libraries

#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$1, KK
#endif
	ALIGN_4

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

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

.L100:
	movl	N, %eax
	andl	$1, %eax
	jle	.L999
	ALIGN_4

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

/* Copying to Sub Buffer */
	leal	BUFFER, %ecx
	movaps	POSINV, %xmm7

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

.L102:
	movss	0 * SIZE(B), %xmm0
	movss	1 * SIZE(B), %xmm1
	movss	2 * SIZE(B), %xmm2
	movss	3 * SIZE(B), %xmm3

	shufps	$0, %xmm0, %xmm0
	shufps	$0, %xmm1, %xmm1
	shufps	$0, %xmm2, %xmm2
	shufps	$0, %xmm3, %xmm3

#if defined(NN) || defined(NT) || defined(NR) || defined(NC)  || \
    defined(TN) || defined(TT) || defined(TR) || defined(TC) 
	xorps	 %xmm7, %xmm1
	xorps	 %xmm7, %xmm3
#else
	xorps	 %xmm7, %xmm0
	xorps	 %xmm7, %xmm2
#endif

	movaps	%xmm0,  0 * SIZE(BB)
	movaps	%xmm1,  4 * SIZE(BB)
	movaps	%xmm2,  8 * SIZE(BB)
	movaps	%xmm3, 12 * SIZE(BB)

	movss	4 * SIZE(B), %xmm0
	movss	5 * SIZE(B), %xmm1
	movss	6 * SIZE(B), %xmm2
	movss	7 * SIZE(B), %xmm3

	shufps	$0, %xmm0, %xmm0
	shufps	$0, %xmm1, %xmm1
	shufps	$0, %xmm2, %xmm2
	shufps	$0, %xmm3, %xmm3

#if defined(NN) || defined(NT) || defined(NR) || defined(NC)  || \
    defined(TN) || defined(TT) || defined(TR) || defined(TC) 
	xorps	 %xmm7, %xmm1
	xorps	 %xmm7, %xmm3
#else
	xorps	 %xmm7, %xmm0
	xorps	 %xmm7, %xmm2
#endif

	movaps	%xmm0, 16 * SIZE(BB)
	movaps	%xmm1, 20 * SIZE(BB)
	movaps	%xmm2, 24 * SIZE(BB)
	movaps	%xmm3, 28 * SIZE(BB)

	prefetcht0	 104 * SIZE(B)

	addl	$ 8 * SIZE, B
	addl	$32 * SIZE, BB
	decl	%eax
	jne	.L102
	ALIGN_4

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

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

	shufps	$0, %xmm0, %xmm0
	shufps	$0, %xmm1, %xmm1

#if defined(NN) || defined(NT) || defined(NR) || defined(NC)  || \
    defined(TN) || defined(TT) || defined(TR) || defined(TC) 
	xorps	 %xmm7, %xmm1
#else
	xorps	 %xmm7, %xmm0
#endif

	movaps	%xmm0,  0 * SIZE(BB)
	movaps	%xmm1,  4 * SIZE(BB)

	addl	$ 2 * SIZE, %edi
	addl	$ 8 * SIZE, %ecx
	decl	%eax
	jne	.L104
	ALIGN_4

.L105:
	movl	C,  %esi
	movl	A,  AA
	movl	M,  %ebx
	sarl	$1, %ebx
	jle	.L130
	ALIGN_4

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

	leal	BUFFER, BB	# boffset1 = boffset
#else
	leal	BUFFER, BB	# boffset1 = boffset
	movl	KK, %eax
	leal	(, %eax,    8), %eax
	leal	(AA, %eax,  2), AA
	leal	(BB,  %eax, 4), BB
#endif	

	XORPS	%xmm4, %xmm4
	XORPS	%xmm5, %xmm5
	XORPS	%xmm6, %xmm6
	XORPS	%xmm7, %xmm7

	movaps	 0 * SIZE(AA), %xmm0
	movaps	16 * SIZE(AA), %xmm1
	movaps	 0 * SIZE(BB), %xmm2
	movaps	16 * SIZE(BB), %xmm3

#if defined(OPTERON) || defined(BARCELONA)
	prefetchw   4 * SIZE(%esi)
#endif

#ifdef PENTIUM4
	prefetchnta 4 * SIZE(%esi)
#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
#ifdef LEFT
	addl	$2, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L112
	ALIGN_4

.L111:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movaps	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movaps	 4 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movaps	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	movaps	12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movaps	 8 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm7
	movaps	32 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	movaps	20 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	movaps	12  * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	movaps	24 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	movaps	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	movaps	32 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	movaps	48 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	movaps	36 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	movaps	20 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	movaps	40 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm6
	movaps	44 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	movaps	24 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm7
	movaps	64 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movaps	52 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movaps	28 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movaps	56 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm6
	movaps	60 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movaps	48 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm7
	movaps	80 * SIZE(BB), %xmm3

	addl	$ 32 * SIZE, AA
	addl	$ 64 * SIZE, BB
	decl	%eax
	jne	.L111
	ALIGN_4
	
.L112:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA_R, %xmm1
	movaps	ALPHA_I, %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L114
	ALIGN_4

.L113:
	mulps	%xmm0, %xmm2
	mulps	 4 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm4
	movaps	 8 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm5
	movaps	 4 * SIZE(AA), %xmm0

	addl	$ 4 * SIZE, AA
	addl	$ 8 * SIZE, BB
	decl	%eax
	jg	.L113
	ALIGN_4

.L114:
	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

	shufps	$0xb1, %xmm5, %xmm5

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

	movaps	%xmm4, %xmm5

	shufps	$0xb1, %xmm4, %xmm4

	mulps	%xmm1, %xmm5
	mulps	%xmm3, %xmm4

	addps	%xmm5, %xmm4

#ifndef TRMMKERNEL
	shufps	$0xe4, %xmm4, %xmm4

	movsd	0 * SIZE(%esi), %xmm0
	movhps	2 * SIZE(%esi), %xmm0

	addps	%xmm0, %xmm4
#endif

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

#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, %esi		# coffset += 4
	decl	%ebx			# i --
	jg	.L110
	ALIGN_4

.L130:
	movl	M,  %ebx
	andl	$1, %ebx
	jle	.L999
	ALIGN_4

.L140:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

	leal	BUFFER, BB	# boffset1 = boffset
#else
	leal	BUFFER, BB	# boffset1 = boffset
	movl	KK, %eax
	leal	(, %eax,    8), %eax
	leal	(AA, %eax,  1), AA
	leal	(BB,  %eax, 4), BB
#endif	

	movlps	 0 * SIZE(AA), %xmm0
	XORPS	%xmm4, %xmm4
	movlps	 8 * SIZE(AA), %xmm1
	XORPS	%xmm5, %xmm5
	movaps	 0 * SIZE(BB), %xmm2
	XORPS	%xmm6, %xmm6
	movaps	16 * SIZE(BB), %xmm3
	XORPS	%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	$1, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L142
	ALIGN_4

.L141:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movaps	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movlps	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	movaps	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	movaps	12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movlps	 4 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm7
	movaps	32 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	movaps	20 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	movlps	 6 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	movaps	24 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	movaps	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	movlps	16 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	movaps	48 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	movaps	36 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	movlps	10 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	movaps	40 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm6
	movaps	44 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	movlps	12 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm7
	movaps	64 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	movaps	52 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movlps	14 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	movaps	56 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm6
	movaps	60 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	movlps	24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm7
	movaps	80 * SIZE(BB), %xmm3

	addl	$ 16 * SIZE, AA
	addl	$ 64 * SIZE, BB
	decl	%eax
	jne	.L141
	ALIGN_4
	
.L142:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movaps	ALPHA_R, %xmm1
	movaps	ALPHA_I, %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L144
	ALIGN_4

.L143:
	mulps	%xmm0, %xmm2
	mulps	 4 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm4
	movaps	 8 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm5
	movlps	 2 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L143
	ALIGN_4

.L144:
	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

	shufps	$0xb1, %xmm5, %xmm5

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

	movaps	%xmm4, %xmm5

	shufps	$0xb1, %xmm4, %xmm4

	mulps	%xmm1, %xmm5
	mulps	%xmm3, %xmm4

	addps	%xmm5, %xmm4

#ifndef TRMMKERNEL
	shufps	$0xe4, %xmm4, %xmm4

	movsd	0 * SIZE(%esi), %xmm0

	addps	%xmm0, %xmm4
#endif

	movsd	%xmm4, 0 * SIZE(%esi)
	ALIGN_4

.L999:
	EMMS

	movl	OLD_STACK, %esp
	popl	%ebx
	popl	%esi
	popl	%edi
	popl	%ebp
	ret

	EPILOGUE