gemm_kernel_8x2_core2.s

Optimized GotoBLAS libraries

	sarl	$3, %eax
	je	.L65
	ALIGN_4

.L62:
	mulps	%xmm1, %xmm0
	mulps	-28 * SIZE(AA), %xmm1
	addps	%xmm0, %xmm4
	movaps	-24 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm6
	movaps	-28 * SIZE(BB), %xmm1
	mulps	%xmm1, %xmm0
	mulps	-20 * SIZE(AA), %xmm1
	addps	%xmm0, %xmm5
	movaps	  0 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm7
	movaps	-24 * SIZE(BB), %xmm1
	mulps	%xmm1, %xmm2
	mulps	-12 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm4
	movaps	 -8 * SIZE(AA), %xmm2
	addps	%xmm1, %xmm6
	movaps	-20 * SIZE(BB), %xmm1
	mulps	%xmm1, %xmm2
	mulps	 -4 * SIZE(AA), %xmm1
	addps	%xmm2, %xmm5
	movaps	 16 * SIZE(AA), %xmm2
	addps	%xmm1, %xmm7
	movaps	  0 * SIZE(BB), %xmm1
	mulps	%xmm3, %xmm0
	mulps	  4 * SIZE(AA), %xmm3
	addps	%xmm0, %xmm4
	movaps	  8 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm6
	movaps	-12 * SIZE(BB), %xmm3
	mulps	%xmm3, %xmm0
	mulps	 12 * SIZE(AA), %xmm3
	addps	%xmm0, %xmm5
	movaps	 32 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	movaps	 -8 * SIZE(BB), %xmm3
	mulps	%xmm3, %xmm2
	mulps	 20 * SIZE(AA), %xmm3
	addps	%xmm2, %xmm4
	movaps	 24 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm6
	movaps	 -4 * SIZE(BB), %xmm3
	mulps	%xmm3, %xmm2
	mulps	 28 * SIZE(AA), %xmm3
	addps	%xmm2, %xmm5
	movaps	 48 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm7
	movaps	 16 * SIZE(BB), %xmm3

	addl   $ 64 * SIZE, AA
	subl   $-32 * SIZE, BB
	decl   %eax
	jne    .L62
	ALIGN_4

.L65:
	movaps	ALPHA,  %xmm3

#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L68
	ALIGN_4

.L66:
	mulps	%xmm1, %xmm0
	mulps	-28 * SIZE(AA), %xmm1
	addps	%xmm0, %xmm4
	movaps	-24 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm6
	movaps	-28 * SIZE(BB), %xmm1

	addl	$8 * SIZE, AA
	addl	$4 * SIZE, BB
	decl	%eax
	jg	.L66
	ALIGN_4

.L68:
	addps	%xmm5, %xmm4
	addps	%xmm7, %xmm6

	mulps	%xmm3, %xmm4
	mulps	%xmm3, %xmm6

#ifndef TRMMKERNEL
	movsd	0 * SIZE(C1), %xmm0
	movhps	2 * SIZE(C1), %xmm0
	movsd	4 * SIZE(C1), %xmm2
	movhps	6 * SIZE(C1), %xmm2

	addps	%xmm0, %xmm4
	addps	%xmm2, %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, 2), BB
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$8, KK
#endif

	addl	$8 * SIZE, C1
	decl	I
	jg	.L61
	ALIGN_4

.L70:
	movl	M,  I
	testl	$4, I
	jle	.L80

.L71:
#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, 2), 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
	movaps	-16 * 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	$4, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L75
	ALIGN_4

.L72:
	mulps	%xmm0,  %xmm1
	movaps	-28 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm4
	movaps	-28 * SIZE(BB), %xmm1
	mulps	%xmm0,  %xmm1
	movaps	-24 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm5
	movaps	-24 * SIZE(BB), %xmm1
	mulps	%xmm0,  %xmm1
	movaps	-20 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm4
	movaps	-20 * SIZE(BB), %xmm1
	mulps	%xmm0,  %xmm1
	movaps	  0 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm5
	movaps	  0 * SIZE(BB), %xmm1
	mulps	%xmm2,  %xmm3
	movaps	-12 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm4
	movaps	-12 * SIZE(BB), %xmm3
	mulps	%xmm2,  %xmm3
	movaps	 -8 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm5
	movaps	 -8 * SIZE(BB), %xmm3
	mulps	%xmm2,  %xmm3
	movaps	 -4 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm4
	movaps	 -4 * SIZE(BB), %xmm3
	mulps	%xmm2,  %xmm3
	movaps	 16 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm5
	movaps	 16 * SIZE(BB), %xmm3

	subl   $-32 * SIZE, AA
	subl   $-32 * SIZE, BB
	decl   %eax
	jne    .L72
	ALIGN_4

.L75:
	movaps	ALPHA,  %xmm3

#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L78
	ALIGN_4

.L76:
	mulps	%xmm0,  %xmm1
	movaps	-28 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm4
	movaps	-28 * SIZE(BB), %xmm1

	addl	$4 * SIZE, AA
	addl	$4 * SIZE, BB
	decl	%eax
	jg	.L76
	ALIGN_4

.L78:
	addps	%xmm5, %xmm4
	mulps	%xmm3, %xmm4

#ifndef TRMMKERNEL
	movsd	0 * SIZE(C1), %xmm0
	movhps	2 * SIZE(C1), %xmm0

	addps	%xmm0, %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, 2), BB
#endif

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

	addl	$4 * SIZE, C1
	ALIGN_4

.L80:
	movl	M,  I
	testl	$2, I
	jle	.L90

.L81:
#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, 2), BB /* because it's doubled */
#endif

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

#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	.L85
	ALIGN_4

.L82:
	mulps	%xmm0,  %xmm1
	movsd	-30 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm4
	movsd	-28 * SIZE(BB), %xmm1
	mulps	%xmm0,  %xmm1
	movsd	-28 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm5
	movsd	-24 * SIZE(BB), %xmm1
	mulps	%xmm0,  %xmm1
	movsd	-26 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm4
	movsd	-20 * SIZE(BB), %xmm1
	mulps	%xmm0,  %xmm1
	movsd	-16 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm5
	movsd	 -0 * SIZE(BB), %xmm1
	mulps	%xmm2,  %xmm3
	movsd	-22 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm4
	movsd	-12 * SIZE(BB), %xmm3
	mulps	%xmm2,  %xmm3
	movsd	-20 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm5
	movsd	 -8 * SIZE(BB), %xmm3
	mulps	%xmm2,  %xmm3
	movsd	-18 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm4
	movsd	 -4 * SIZE(BB), %xmm3
	mulps	%xmm2,  %xmm3
	movsd	 -8 * SIZE(AA), %xmm2
	addps	%xmm3, %xmm5
	movsd	 16 * SIZE(BB), %xmm3

	subl   $-16 * SIZE, AA
	subl   $-32 * SIZE, BB
	decl   %eax
	jne    .L82
	ALIGN_4

.L85:
	movsd	ALPHA,  %xmm3

#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L88
	ALIGN_4

.L86:
	mulps	%xmm0,  %xmm1
	movsd	-30 * SIZE(AA), %xmm0
	addps	%xmm1, %xmm4
	movsd	-28 * SIZE(BB), %xmm1

	addl	$2 * SIZE, AA
	addl	$4 * SIZE, BB
	decl	%eax
	jg	.L86
	ALIGN_4

.L88:
	addps	%xmm5, %xmm4
	mulps	%xmm3, %xmm4

#ifndef TRMMKERNEL
	movsd	0 * SIZE(C1), %xmm0
	addps	%xmm0, %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, 2), BB
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$2, KK
#endif
	addl	$2 * SIZE, C1
	ALIGN_4

.L90:
	movl	M,  I
	testl	$1, I
	jle	.L99

.L91:
#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,   4), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 4), BB /* because it's doubled */
#endif

	movss	-32 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movss	-32 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movss	-16 * SIZE(BB), %xmm3
	movss	-28 * SIZE(AA), %xmm2

#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	.L95
	ALIGN_4

.L92:
	mulss	%xmm0,  %xmm1
	movss	-31 * SIZE(AA), %xmm0
	addss	%xmm1, %xmm4
	movss	-28 * SIZE(BB), %xmm1
	mulss	%xmm0,  %xmm1
	movss	-30 * SIZE(AA), %xmm0
	addss	%xmm1, %xmm5
	movss	-24 * SIZE(BB), %xmm1
	mulss	%xmm0,  %xmm1
	movss	-29 * SIZE(AA), %xmm0
	addss	%xmm1, %xmm4
	movss	-20 * SIZE(BB), %xmm1
	mulss	%xmm0,  %xmm1
	movss	-24 * SIZE(AA), %xmm0
	addss	%xmm1, %xmm5
	movss	 -0 * SIZE(BB), %xmm1
	mulss	%xmm2,  %xmm3
	movss	-27 * SIZE(AA), %xmm2
	addss	%xmm3, %xmm4
	movss	-12 * SIZE(BB), %xmm3
	mulss	%xmm2,  %xmm3
	movss	-26 * SIZE(AA), %xmm2
	addss	%xmm3, %xmm5
	movss	 -8 * SIZE(BB), %xmm3
	mulss	%xmm2,  %xmm3
	movss	-25 * SIZE(AA), %xmm2
	addss	%xmm3, %xmm4
	movss	 -4 * SIZE(BB), %xmm3
	mulss	%xmm2,  %xmm3
	movss	-20 * SIZE(AA), %xmm2
	addss	%xmm3, %xmm5
	movss	 16 * SIZE(BB), %xmm3

	subl   $ -8 * SIZE, AA
	subl   $-32 * SIZE, BB
	decl   %eax
	jne    .L92
	ALIGN_4

.L95:
	movss	ALPHA,  %xmm3

#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L98
	ALIGN_4

.L96:
	mulss	%xmm0,  %xmm1
	movss	-31 * SIZE(AA), %xmm0
	addss	%xmm1, %xmm4
	movss	-28 * SIZE(BB), %xmm1

	addl	$1 * SIZE, AA
	addl	$4 * SIZE, BB
	decl	%eax
	jg	.L96
	ALIGN_4

.L98:
	addss	%xmm5, %xmm4
	mulss	%xmm3, %xmm4

#ifndef TRMMKERNEL
	movss	0 * SIZE(C1), %xmm0
	addss	%xmm0, %xmm4
#endif
	movss	%xmm4, 0 * SIZE(C1)
	ALIGN_4

.L99:
	addl	LDC, C
	ALIGN_4


.L999:
	movl	OLD_STACK, %esp
	
	EMMS

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

	EPILOGUE