gemm_kernel_8x2_core2.s

Optimized GotoBLAS libraries

	addps	%xmm2, %xmm7
	movaps	 -8 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm3
	mulps	 20 * SIZE(BB), %xmm2
	addps	%xmm3, %xmm4
	movaps	 24 * SIZE(BB), %xmm3
	addps	%xmm2, %xmm5
	movaps	 -4 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm3
	mulps	 28 * SIZE(BB), %xmm2
	addps	%xmm3, %xmm6
	movaps	 48 * SIZE(BB), %xmm3
	addps	%xmm2, %xmm7
	movaps	 16 * SIZE(AA), %xmm2

	subl   $-32 * SIZE, AA
	addl   $ 64 * SIZE, BB
	decl   %eax
	jne    .L22
	ALIGN_4

.L25:
	movaps	ALPHA,  %xmm3

#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L28
	ALIGN_4

.L26:
	mulps	%xmm0,  %xmm1
	mulps	-28 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm4
	movaps	-24 * SIZE(BB), %xmm1
	addps	%xmm0, %xmm5
	movaps	-28 * SIZE(AA), %xmm0

	addl	$4 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L26
	ALIGN_4

.L28:
	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

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

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

	movsd	0 * SIZE(C1, LDC), %xmm1
	movhps	2 * SIZE(C1, LDC), %xmm1

	addps	%xmm0, %xmm4
	addps	%xmm1, %xmm5
#endif

	movsd	%xmm4, 0 * SIZE(C1)
	movhps	%xmm4, 2 * SIZE(C1)
	movsd	%xmm5, 0 * SIZE(C1, LDC)
	movhps	%xmm5, 2 * SIZE(C1, LDC)

#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	$4, KK
#endif

	addl	$4 * SIZE, C1
	ALIGN_4

.L30:
	movl	M,  I
	testl	$2, I
	jle	.L40

.L31:
#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, 4), BB /* because it's doubled */
#endif

	movsd	-32 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movsd	-32 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movsd	-24 * SIZE(AA), %xmm2
	pxor	%xmm6, %xmm6
	movsd	-16 * SIZE(BB), %xmm3
	pxor	%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	$2, %eax
#else
	addl	$2, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L35
	ALIGN_4

.L32:
	mulps	%xmm0,  %xmm1
	mulps	-28 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm4
 	movsd	-24 * SIZE(BB), %xmm1
	addps	%xmm0, %xmm5
	movsd	-30 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm1
	mulps	-20 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm6
	movsd	  0 * SIZE(BB), %xmm1
	addps	%xmm0, %xmm7
	movsd	-28 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm3
	mulps	-12 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm4
	movsd	 -8 * SIZE(BB), %xmm3
	addps	%xmm0, %xmm5
	movsd	-26 * SIZE(AA), %xmm0
	mulps	%xmm0,  %xmm3
	mulps	 -4 * SIZE(BB), %xmm0
	addps	%xmm3, %xmm6
	movsd	 16 * SIZE(BB), %xmm3
	addps	%xmm0, %xmm7
	movsd	-16 * SIZE(AA), %xmm0
	mulps	%xmm2,  %xmm1
	mulps	  4 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm4
	movsd	  8 * SIZE(BB), %xmm1
	addps	%xmm2, %xmm5
	movsd	-22 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm1
	mulps	 12 * SIZE(BB), %xmm2
	addps	%xmm1, %xmm6
	movsd	 32 * SIZE(BB), %xmm1
	addps	%xmm2, %xmm7
	movsd	-20 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm3
	mulps	 20 * SIZE(BB), %xmm2
	addps	%xmm3, %xmm4
	movsd	 24 * SIZE(BB), %xmm3
	addps	%xmm2, %xmm5
	movsd	-18 * SIZE(AA), %xmm2
	mulps	%xmm2,  %xmm3
	mulps	 28 * SIZE(BB), %xmm2
	addps	%xmm3, %xmm6
	movsd	 48 * SIZE(BB), %xmm3
	addps	%xmm2, %xmm7
	movsd	 -8 * SIZE(AA), %xmm2

	subl   $-16 * SIZE, AA
	addl   $ 64 * SIZE, BB
	decl   %eax
	jne    .L32
	ALIGN_4

.L35:
	movsd	ALPHA,  %xmm3

#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L38
	ALIGN_4

.L36:
	mulps	%xmm0,  %xmm1
	mulps	-28 * SIZE(BB), %xmm0
	addps	%xmm1, %xmm4
	movsd	-24 * SIZE(BB), %xmm1
	addps	%xmm0, %xmm5
	movsd	-30 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L36
	ALIGN_4

.L38:
	addps	%xmm6, %xmm4
	addps	%xmm7, %xmm5

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

#ifndef TRMMKERNEL
	movsd	0 * SIZE(C1), %xmm0
	movsd	0 * SIZE(C1, LDC), %xmm1

	addps	%xmm0, %xmm4
	addps	%xmm1, %xmm5
#endif

	movsd	%xmm4, 0 * SIZE(C1)
	movsd	%xmm5, 0 * SIZE(C1, LDC)

#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	$2, KK
#endif

	addl	$2 * SIZE, C1
	ALIGN_4

.L40:
	movl	M,  I
	testl	$1, I
	jle	.L49

.L41:
#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, 8), BB /* because it's doubled */
#endif

	movss	-32 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movss	-32 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movss	-28 * SIZE(AA), %xmm2
	pxor	%xmm6, %xmm6
	movss	-16 * SIZE(BB), %xmm3
	pxor	%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	$2, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L45
	ALIGN_4

.L42:
	mulss	%xmm0,  %xmm1
	mulss	-28 * SIZE(BB), %xmm0
	addss	%xmm1, %xmm4
 	movss	-24 * SIZE(BB), %xmm1
	addss	%xmm0, %xmm5
	movss	-31 * SIZE(AA), %xmm0
	mulss	%xmm0,  %xmm1
	mulss	-20 * SIZE(BB), %xmm0
	addss	%xmm1, %xmm6
	movss	  0 * SIZE(BB), %xmm1
	addss	%xmm0, %xmm7
	movss	-30 * SIZE(AA), %xmm0
	mulss	%xmm0,  %xmm3
	mulss	-12 * SIZE(BB), %xmm0
	addss	%xmm3, %xmm4
	movss	 -8 * SIZE(BB), %xmm3
	addss	%xmm0, %xmm5
	movss	-29 * SIZE(AA), %xmm0
	mulss	%xmm0,  %xmm3
	mulss	 -4 * SIZE(BB), %xmm0
	addss	%xmm3, %xmm6
	movss	 16 * SIZE(BB), %xmm3
	addss	%xmm0, %xmm7
	movss	-24 * SIZE(AA), %xmm0
	mulss	%xmm2,  %xmm1
	mulss	  4 * SIZE(BB), %xmm2
	addss	%xmm1, %xmm4
	movss	  8 * SIZE(BB), %xmm1
	addss	%xmm2, %xmm5
	movss	-27 * SIZE(AA), %xmm2
	mulss	%xmm2,  %xmm1
	mulss	 12 * SIZE(BB), %xmm2
	addss	%xmm1, %xmm6
	movss	 32 * SIZE(BB), %xmm1
	addss	%xmm2, %xmm7
	movss	-26 * SIZE(AA), %xmm2
	mulss	%xmm2,  %xmm3
	mulss	 20 * SIZE(BB), %xmm2
	addss	%xmm3, %xmm4
	movss	 24 * SIZE(BB), %xmm3
	addss	%xmm2, %xmm5
	movss	-25 * SIZE(AA), %xmm2
	mulss	%xmm2,  %xmm3
	mulss	 28 * SIZE(BB), %xmm2
	addss	%xmm3, %xmm6
	movss	 48 * SIZE(BB), %xmm3
	addss	%xmm2, %xmm7
	movss	-20 * SIZE(AA), %xmm2

	subl   $-8 * SIZE, AA
	addl   $64 * SIZE, BB
	decl   %eax
	jne    .L42
	ALIGN_4

.L45:
	movss	ALPHA,  %xmm3

#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L48
	ALIGN_4

.L46:
	mulss	%xmm0,  %xmm1
	mulss	-28 * SIZE(BB), %xmm0
	addss	%xmm1, %xmm4
	movss	-24 * SIZE(BB), %xmm1
	addss	%xmm0, %xmm5
	movss	-31 * SIZE(AA), %xmm0

	addl	$1 * SIZE, AA
	addl	$8 * SIZE, BB
	decl	%eax
	jg	.L46
	ALIGN_4

.L48:
	addss	%xmm6, %xmm4
	addss	%xmm7, %xmm5

	mulss	%xmm3, %xmm4
	mulss	%xmm3, %xmm5

#ifndef TRMMKERNEL
	movss	0 * SIZE(C1), %xmm0
	movss	0 * SIZE(C1, LDC), %xmm1

	addss	%xmm0, %xmm4
	addss	%xmm1, %xmm5
#endif

	movss	%xmm4, 0 * SIZE(C1)
	movss	%xmm5, 0 * SIZE(C1, LDC)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	movl	K, %eax
	subl	KKK, %eax
	leal	(,%eax,    4), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 8), BB
#endif

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

.L49:
#if defined(TRMMKERNEL) && !defined(LEFT)
	addl	$2, KK
#endif

	leal	(, LDC, 2), %eax
	addl	%eax, C
	decl	J
	jg	.L01
	ALIGN_4

.L50:
	movl	N, %eax
	testl	$1, %eax
	jle	.L999
	ALIGN_4

.L51:
	leal	32 * SIZE + BUFFER, BB

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

	movl	K, %eax
	sarl	$3, %eax
	jle	.L55
	ALIGN_4
	
.L52:
	movss	-32 * SIZE(B), %xmm0
	movss	-31 * SIZE(B), %xmm1
	movss	-30 * SIZE(B), %xmm2
	movss	-29 * SIZE(B), %xmm3
	movss	-28 * SIZE(B), %xmm4
	movss	-27 * SIZE(B), %xmm5
	movss	-26 * SIZE(B), %xmm6
	movss	-25 * SIZE(B), %xmm7

	shufps	$0, %xmm0, %xmm0
	shufps	$0, %xmm1, %xmm1
	shufps	$0, %xmm2, %xmm2
	shufps	$0, %xmm3, %xmm3
	shufps	$0, %xmm4, %xmm4
	shufps	$0, %xmm5, %xmm5
	shufps	$0, %xmm6, %xmm6
	shufps	$0, %xmm7, %xmm7

	movaps	%xmm0,  -32 * SIZE(BB)
	movaps	%xmm1,  -28 * SIZE(BB)
	movaps	%xmm2,  -24 * SIZE(BB)
	movaps	%xmm3,  -20 * SIZE(BB)
	movaps	%xmm4,  -16 * SIZE(BB)
	movaps	%xmm5,  -12 * SIZE(BB)
	movaps	%xmm6,   -8 * SIZE(BB)
	movaps	%xmm7,   -4 * SIZE(BB)

	addl	$  8 * SIZE, B
	subl	$-32 * SIZE, BB
	decl	%eax
	jne	.L52
	ALIGN_4

.L55:
	movl	K, %eax
	andl	$7, %eax
	BRANCH
	jle	.L60
	ALIGN_4

.L56:
	movss	-32 * SIZE(B), %xmm0
	shufps	$0, %xmm0, %xmm0
	movaps	%xmm0,  -32 * SIZE(BB)

	addl	$1 * SIZE, B
	addl	$4 * SIZE, BB
	decl	%eax
	jne	.L56
	ALIGN_4
	
.L60:
	movl	C, C1
	movl	A, AA
	movl	M,  I
	sarl	$3, I
	jle	.L70
	ALIGN_4

.L61:
#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, 4), 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
	pxor	%xmm6, %xmm6
	movaps	-16 * SIZE(BB), %xmm3
	pxor	%xmm7, %xmm7

	prefetcht0	3 * SIZE(C1)

#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	$8, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif