zgemm_kernel_4x1_sse.s

Optimized GotoBLAS libraries

	leal	(, %eax,   8), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 4), BB /* because it's doubled */

	movaps	 0 * SIZE(BB), %xmm2
	XORPS	%xmm4, %xmm4
	movaps	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	movaps	 8 * SIZE(BB), %xmm3
	XORPS	%xmm6, %xmm6
	movaps	 8 * SIZE(AA), %xmm1
	XORPS	%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
#ifdef LEFT
	addl	$2, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L52
	ALIGN_4

.L51:
	mulps	%xmm0, %xmm2
	mulps	 4 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm4
	movaps	16 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm5
	movaps	 4 * SIZE(AA), %xmm0
	mulps	%xmm0, %xmm3
	mulps	12 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm6
	movaps	24 * SIZE(BB), %xmm3
	addps	%xmm0, %xmm7
	movaps	16 * SIZE(AA), %xmm0
	mulps	%xmm1, %xmm2
	mulps	20 * SIZE(BB), %xmm1
	addps	%xmm2, %xmm4
	movaps	32 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm5
	movaps	12 * SIZE(AA), %xmm1
	mulps	%xmm1, %xmm3
	mulps	28 * SIZE(BB), %xmm1
	addps	%xmm3, %xmm6
	movaps	40 * SIZE(BB), %xmm3
	addps	%xmm1, %xmm7
	movaps	24 * SIZE(AA), %xmm1
	mulps	%xmm0, %xmm2
	mulps	36 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm4
	movaps	48 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm5
	movaps	20 * SIZE(AA), %xmm0
	mulps	%xmm0, %xmm3
	mulps	44 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm6
	movaps	56 * SIZE(BB), %xmm3
	addps	%xmm0, %xmm7
	movaps	 32 * SIZE(AA), %xmm0
	mulps	%xmm1, %xmm2
	mulps	52 * SIZE(BB), %xmm1
	addps	%xmm2, %xmm4
	movaps	 64 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm5
	movaps	28 * SIZE(AA), %xmm1
	mulps	%xmm1, %xmm3
	mulps	60 * SIZE(BB), %xmm1
	addps	%xmm3, %xmm6
	movaps	 72 * SIZE(BB), %xmm3
	addps	%xmm1, %xmm7
	movaps	 40 * SIZE(AA), %xmm1

	addl	$32 * SIZE, AA
	addl	$64 * SIZE, BB
	decl	%eax
	jne	.L51
#endif
	
.L52:
#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 .L54

.L53:
	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		# aoffset  += 8
	addl	$8 * SIZE, BB		# boffset1 += 8
	decl	%eax
	jg	.L53

.L54:
	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
	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
	ALIGN_2

.L70:
	testl	$1, %ebx
	jle	.L99


#if (L1_DATA_LINESIZE == 64)

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

	leal	BUFFER, BB
	MOVSD	 0 * SIZE + BUFFER, %xmm2
	XORPS	%xmm4, %xmm4
	MOVSD	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	MOVSD	 16 * SIZE + BUFFER, %xmm3
	XORPS	%xmm6, %xmm6
	MOVSD	 8 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7

#else

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

	MOVSD	 0 * SIZE(BB), %xmm2
	XORPS	%xmm4, %xmm4
	MOVSD	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	MOVSD	 16 * SIZE(BB), %xmm3
	XORPS	%xmm6, %xmm6
	MOVSD	 8 * SIZE(AA), %xmm1
	XORPS	%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	$3, %eax
	je	.L72
	ALIGN_4

.L71:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	MOVSD	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	MOVSD	12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 4 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm7
	MOVSD	32 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	MOVSD	20 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	MOVSD	 6 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	MOVSD	24 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	MOVSD	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	MOVSD	16 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	MOVSD	48 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	36 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	MOVSD	10 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	MOVSD	40 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm6
	MOVSD	44 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	MOVSD	12 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm7
	MOVSD	64 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	MOVSD	52 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	MOVSD	14 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	MOVSD	56 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm6
	MOVSD	60 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	MOVSD	24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm7
	MOVSD	80 * SIZE(BB), %xmm3

	addl	$16 * SIZE, AA
	addl	$64 * SIZE, BB
	decl	%eax
	jne	.L71
	ALIGN_2

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

	leal	BUFFER, BB
	MOVSD	 0 * SIZE + BUFFER, %xmm2
	XORPS	%xmm4, %xmm4
	MOVSD	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	MOVSD	 8 * SIZE + BUFFER, %xmm3
	XORPS	%xmm6, %xmm6
	MOVSD	 8 * SIZE(AA), %xmm1
	XORPS	%xmm7, %xmm7

#else

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

	MOVSD	 0 * SIZE(BB), %xmm2
	XORPS	%xmm4, %xmm4
	MOVSD	 0 * SIZE(AA), %xmm0
	XORPS	%xmm5, %xmm5
	MOVSD	 8 * SIZE(BB), %xmm3
	XORPS	%xmm6, %xmm6
	MOVSD	 8 * SIZE(AA), %xmm1
	XORPS	%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	$3, %eax
	je	.L72
	ALIGN_4

.L71:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	MOVSD	16 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	MOVSD	12 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	MOVSD	 4 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	MOVSD	24 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	20 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 6 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	MOVSD	32 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	MOVSD	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	MOVSD	16 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm5
	MOVSD	40 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	36 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	MOVSD	10 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	MOVSD	48 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	MOVSD	44 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
	MOVSD	12 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	MOVSD	56 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	52 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm2
	MOVSD	14 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	MOVSD	64 * SIZE(BB), %xmm2
	mulps	%xmm1, %xmm3
	addps	%xmm3, %xmm4
	MOVSD	60 * SIZE(BB), %xmm3
	mulps	%xmm1, %xmm3
 	MOVSD	24 * SIZE(AA), %xmm1
	addps	%xmm3, %xmm5
	MOVSD	72 * SIZE(BB), %xmm3

	addl	$16 * SIZE, AA
	addl	$64 * SIZE, BB
	decl	%eax
	jne	.L71
	ALIGN_2
#endif
	
.L72:
#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 .L74

.L73:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	MOVSD	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	MOVSD	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm5
	MOVSD	 8 * SIZE(BB), %xmm2

	addl	$2 * SIZE, AA		# aoffset  += 8
	addl	$8 * SIZE, BB		# boffset1 += 8
	decl	%eax
	jg	.L73

.L74:
	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
	MOVSD	0 * SIZE(%esi), %xmm0

	addps	%xmm0, %xmm4
#endif

	MOVSD	%xmm4, 0 * 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, 1), 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
	ALIGN_2

.L999:
	movl	OLD_STACK, %esp
	
	EMMS

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

	EPILOGUE