gemm_kernel_4x2_core2.s

Optimized GotoBLAS libraries

#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	.L35
	ALIGN_4

.L32:
	mulsd	%xmm0,  %xmm1
	mulsd	-14 * SIZE(BB), %xmm0
	addsd	%xmm1, %xmm4
 	movsd	-12 * SIZE(BB), %xmm1
	addsd	%xmm0, %xmm5
	movsd	-15 * SIZE(AA), %xmm0
	mulsd	%xmm0,  %xmm1
	mulsd	-10 * SIZE(BB), %xmm0
	addsd	%xmm1, %xmm6
	movsd	  0 * SIZE(BB), %xmm1
	addsd	%xmm0, %xmm7
	movsd	-14 * SIZE(AA), %xmm0
	mulsd	%xmm0,  %xmm3
	mulsd	 -6 * SIZE(BB), %xmm0
	addsd	%xmm3, %xmm4
	movsd	 -4 * SIZE(BB), %xmm3
	addsd	%xmm0, %xmm5
	movsd	-13 * SIZE(AA), %xmm0
	mulsd	%xmm0,  %xmm3
	mulsd	 -2 * SIZE(BB), %xmm0
	addsd	%xmm3, %xmm6
	movsd	  8 * SIZE(BB), %xmm3
	addsd	%xmm0, %xmm7
	movsd	 -8 * SIZE(AA), %xmm0
	mulsd	%xmm2,  %xmm1
	mulsd	  2 * SIZE(BB), %xmm2
	addsd	%xmm1, %xmm4
	movsd	  4 * SIZE(BB), %xmm1
	addsd	%xmm2, %xmm5
	movsd	-11 * SIZE(AA), %xmm2
	mulsd	%xmm2,  %xmm1
	mulsd	  6 * SIZE(BB), %xmm2
	addsd	%xmm1, %xmm6
	movsd	 16 * SIZE(BB), %xmm1
	addsd	%xmm2, %xmm7
	movsd	-10 * SIZE(AA), %xmm2
	mulsd	%xmm2,  %xmm3
	mulsd	 10 * SIZE(BB), %xmm2
	addsd	%xmm3, %xmm4
	movsd	 12 * SIZE(BB), %xmm3
	addsd	%xmm2, %xmm5
	movsd	 -9 * SIZE(AA), %xmm2
	mulsd	%xmm2,  %xmm3
	mulsd	 14 * SIZE(BB), %xmm2
	addsd	%xmm3, %xmm6
	movsd	 24 * SIZE(BB), %xmm3
	addsd	%xmm2, %xmm7
	movsd	 -4 * SIZE(AA), %xmm2

	subl   $-8 * SIZE, AA
	addl   $32 * 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:
	mulsd	%xmm0,  %xmm1
	mulsd	-14 * SIZE(BB), %xmm0
	addsd	%xmm1, %xmm4
	movsd	-12 * SIZE(BB), %xmm1
	addsd	%xmm0, %xmm5
	movsd	-15 * SIZE(AA), %xmm0

	addl	$1 * SIZE, AA
	addl	$4 * SIZE, BB
	decl	%eax
	jg	.L36
	ALIGN_4

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

	mulsd	%xmm3, %xmm4
	mulsd	%xmm3, %xmm5

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

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

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

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

.L39:
#if defined(TRMMKERNEL) && !defined(LEFT)
	addl	$2, KK
#endif

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

.L40:
	movl	N, %eax
	testl	$1, %eax
	jle	.L999
	ALIGN_4

.L41:
	leal	16 * SIZE + BUFFER, BB

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

	movl	K, %eax
	sarl	$3, %eax
	jle	.L45
	ALIGN_4
	
.L42:
	movddup	 -16 * SIZE(B), %xmm0
	movddup	 -15 * SIZE(B), %xmm1
	movddup	 -14 * SIZE(B), %xmm2
	movddup	 -13 * SIZE(B), %xmm3
	movddup	 -12 * SIZE(B), %xmm4
	movddup	 -11 * SIZE(B), %xmm5
	movddup	 -10 * SIZE(B), %xmm6
	movddup	  -9 * SIZE(B), %xmm7

	movapd	%xmm0,  -16 * SIZE(BB)
	movapd	%xmm1,  -14 * SIZE(BB)
	movapd	%xmm2,  -12 * SIZE(BB)
	movapd	%xmm3,  -10 * SIZE(BB)
	movapd	%xmm4,   -8 * SIZE(BB)
	movapd	%xmm5,   -6 * SIZE(BB)
	movapd	%xmm6,   -4 * SIZE(BB)
	movapd	%xmm7,   -2 * SIZE(BB)

	addl	$ 8 * SIZE, B
	addl	$16 * SIZE, BB
	decl	%eax
	jne	.L42
	ALIGN_4

.L45:
	movl	K, %eax
	andl	$7, %eax
	BRANCH
	jle	.L50
	ALIGN_4

.L46:
	movddup	 -16 * SIZE(B), %xmm0

	movapd	%xmm0,  -16 * SIZE(BB)
	addl	$1 * SIZE, B
	addl	$2 * SIZE, BB
	decl	%eax
	jne	.L46
	ALIGN_4
	
.L50:
	movl	C, C1
	movl	A, AA
	movl	M,  I
	sarl	$2, I
	jle	.L60
	ALIGN_4

.L51:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	leal	16 * SIZE + BUFFER, BB
#else
	leal	16 * SIZE + BUFFER, BB
	movl	KK, %eax
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
	leal	(BB, %eax, 2), BB
#endif

	movapd	-16 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movapd	-16 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movapd	 -8 * SIZE(AA), %xmm2
	pxor	%xmm6, %xmm6
	movapd	 -8 * 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	$4, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L55
	ALIGN_4

.L52:
	mulpd	%xmm1, %xmm0
	mulpd	-14 * SIZE(AA), %xmm1
	addpd	%xmm0, %xmm4
	movapd	-12 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm6
	movapd	-14 * SIZE(BB), %xmm1
	mulpd	%xmm1, %xmm0
	mulpd	-10 * SIZE(AA), %xmm1
	addpd	%xmm0, %xmm5
	movapd	  0 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm7
	movapd	-12 * SIZE(BB), %xmm1
	mulpd	%xmm1, %xmm2
	mulpd	 -6 * SIZE(AA), %xmm1
	addpd	%xmm2, %xmm4
	movapd	 -4 * SIZE(AA), %xmm2
	addpd	%xmm1, %xmm6
	movapd	-10 * SIZE(BB), %xmm1
	mulpd	%xmm1, %xmm2
	mulpd	 -2 * SIZE(AA), %xmm1
	addpd	%xmm2, %xmm5
	movapd	  8 * SIZE(AA), %xmm2
	addpd	%xmm1, %xmm7
	movapd	  0 * SIZE(BB), %xmm1
	mulpd	%xmm3, %xmm0
	mulpd	  2 * SIZE(AA), %xmm3
	addpd	%xmm0, %xmm4
	movapd	  4 * SIZE(AA), %xmm0
	addpd	%xmm3, %xmm6
	movapd	 -6 * SIZE(BB), %xmm3
	mulpd	%xmm3, %xmm0
	mulpd	  6 * SIZE(AA), %xmm3
	addpd	%xmm0, %xmm5
	movapd	 16 * SIZE(AA), %xmm0
	addpd	%xmm3, %xmm7
	movapd	 -4 * SIZE(BB), %xmm3
	mulpd	%xmm3, %xmm2
	mulpd	 10 * SIZE(AA), %xmm3
	addpd	%xmm2, %xmm4
	movapd	 12 * SIZE(AA), %xmm2
	addpd	%xmm3, %xmm6
	movapd	 -2 * SIZE(BB), %xmm3
	mulpd	%xmm3, %xmm2
	mulpd	 14 * SIZE(AA), %xmm3
	addpd	%xmm2, %xmm5
	movapd	 24 * SIZE(AA), %xmm2
	addpd	%xmm3, %xmm7
	movapd	  8 * SIZE(BB), %xmm3

	addl   $ 32 * SIZE, AA
	subl   $-16 * SIZE, BB
	decl   %eax
	jne    .L52
	ALIGN_4

.L55:
	movapd	ALPHA,  %xmm3

#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	andl	$7, %eax
	BRANCH
	je .L58
	ALIGN_4

.L56:
	mulpd	%xmm1, %xmm0
	mulpd	-14 * SIZE(AA), %xmm1
	addpd	%xmm0, %xmm4
	movapd	-12 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm6
	movapd	-14 * SIZE(BB), %xmm1

	addl	$4 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L56
	ALIGN_4

.L58:
	addpd	%xmm5, %xmm4
	addpd	%xmm7, %xmm6

	mulpd	%xmm3, %xmm4
	mulpd	%xmm3, %xmm6

#ifndef TRMMKERNEL
	movsd	0 * SIZE(C1), %xmm0
	movhpd	1 * SIZE(C1), %xmm0
	movsd	2 * SIZE(C1), %xmm2
	movhpd	3 * SIZE(C1), %xmm2

	addpd	%xmm0, %xmm4
	addpd	%xmm2, %xmm6
#endif

	movsd	%xmm4, 0 * SIZE(C1)
	movhpd	%xmm4, 1 * SIZE(C1)
	movsd	%xmm6, 2 * SIZE(C1)
	movhpd	%xmm6, 3 * SIZE(C1)

#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, 4), AA
	leal	(BB, %eax, 2), BB
#endif

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

	addl	$4 * SIZE, C1
	decl	I
	jg	.L51
	ALIGN_4

.L60:
	movl	M,  I
	testl	$2, I
	jle	.L70

.L61:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	leal	16 * SIZE + BUFFER, BB
#else
	leal	16 * SIZE + BUFFER, BB
	movl	KK, %eax
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 2), BB
#endif

	movapd	-16 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movapd	-16 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movapd	 -8 * SIZE(AA), %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	$2, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L65
	ALIGN_4

.L62:
	mulpd	%xmm0,  %xmm1
	movapd	-14 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm4
	movapd	-14 * SIZE(BB), %xmm1
	mulpd	%xmm0,  %xmm1
	movapd	-12 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm5
	movapd	-12 * SIZE(BB), %xmm1
	mulpd	%xmm0,  %xmm1
	movapd	-10 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm4
	movapd	-10 * SIZE(BB), %xmm1
	mulpd	%xmm0,  %xmm1
	movapd	  0 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm5
	movapd	  0 * SIZE(BB), %xmm1
	mulpd	%xmm2,  %xmm3
	movapd	 -6 * SIZE(AA), %xmm2
	addpd	%xmm3, %xmm4
	movapd	 -6 * SIZE(BB), %xmm3
	mulpd	%xmm2,  %xmm3
	movapd	 -4 * SIZE(AA), %xmm2
	addpd	%xmm3, %xmm5
	movapd	 -4 * SIZE(BB), %xmm3
	mulpd	%xmm2,  %xmm3
	movapd	 -2 * SIZE(AA), %xmm2
	addpd	%xmm3, %xmm4
	movapd	 -2 * SIZE(BB), %xmm3
	mulpd	%xmm2,  %xmm3
	movapd	  8 * SIZE(AA), %xmm2
	addpd	%xmm3, %xmm5
	movapd	  8 * SIZE(BB), %xmm3

	subl   $-16 * SIZE, AA
	subl   $-16 * SIZE, BB
	decl   %eax
	jne    .L62
	ALIGN_4

.L65:
	movapd	ALPHA,  %xmm3

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

.L66:
	mulpd	%xmm0,  %xmm1
	movapd	-14 * SIZE(AA), %xmm0
	addpd	%xmm1, %xmm4
	movapd	-14 * SIZE(BB), %xmm1

	addl	$2 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L66
	ALIGN_4

.L68:
	addpd	%xmm5, %xmm4
	mulpd	%xmm3, %xmm4

#ifndef TRMMKERNEL
	movsd	0 * SIZE(C1), %xmm0
	movhpd	1 * SIZE(C1), %xmm0

	addpd	%xmm0, %xmm4
#endif

	movsd	%xmm4, 0 * SIZE(C1)
	movhpd	%xmm4, 1 * SIZE(C1)

#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, 2), BB
#endif

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$2, KK
#endif

	addl	$2 * SIZE, C1
	ALIGN_4

.L70:
	movl	M,  I
	testl	$1, I
	jle	.L79

.L71:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	leal	16 * SIZE + BUFFER, BB
#else
	leal	16 * SIZE + BUFFER, BB
	movl	KK, %eax
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 2), BB
#endif

	movsd	-16 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movsd	-16 * SIZE(BB), %xmm1
	pxor	%xmm5, %xmm5
	movsd	 -8 * SIZE(BB), %xmm3
	movsd	-12 * 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
	addl	$1, %eax
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L75
	ALIGN_4

.L72:
	mulsd	%xmm0,  %xmm1
	movsd	-15 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm4
	movsd	-14 * SIZE(BB), %xmm1
	mulsd	%xmm0,  %xmm1
	movsd	-14 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm5
	movsd	-12 * SIZE(BB), %xmm1
	mulsd	%xmm0,  %xmm1
	movsd	-13 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm4
	movsd	-10 * SIZE(BB), %xmm1
	mulsd	%xmm0,  %xmm1
	movsd	 -8 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm5
	movsd	 -0 * SIZE(BB), %xmm1
	mulsd	%xmm2,  %xmm3
	movsd	-11 * SIZE(AA), %xmm2
	addsd	%xmm3, %xmm4
	movsd	 -6 * SIZE(BB), %xmm3
	mulsd	%xmm2,  %xmm3
	movsd	-10 * SIZE(AA), %xmm2
	addsd	%xmm3, %xmm5
	movsd	 -4 * SIZE(BB), %xmm3
	mulsd	%xmm2,  %xmm3
	movsd	 -9 * SIZE(AA), %xmm2
	addsd	%xmm3, %xmm4
	movsd	 -2 * SIZE(BB), %xmm3
	mulsd	%xmm2,  %xmm3
	movsd	 -4 * SIZE(AA), %xmm2
	addsd	%xmm3, %xmm5
	movsd	  8 * SIZE(BB), %xmm3

	subl   $ -8 * SIZE, AA
	subl   $-16 * SIZE, BB
	decl   %eax
	jne    .L72
	ALIGN_4

.L75:
	movsd	ALPHA,  %xmm3

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

.L76:
	mulsd	%xmm0,  %xmm1
	movsd	-15 * SIZE(AA), %xmm0
	addsd	%xmm1, %xmm4
	movsd	-14 * SIZE(BB), %xmm1

	addl	$1 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L76
	ALIGN_4

.L78:
	addsd	%xmm5, %xmm4
	mulsd	%xmm3, %xmm4

#ifndef TRMMKERNEL
	movsd	0 * SIZE(C1), %xmm0
	addsd	%xmm0, %xmm4
#endif
	movsd	%xmm4, 0 * SIZE(C1)
	ALIGN_4

.L79:
	addl	LDC, C
	ALIGN_4


.L999:
	movl	OLD_STACK, %esp
	
	EMMS

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

	EPILOGUE