gemm_kernel_8x1_sse2.s

Optimized GotoBLAS libraries

	addpd	%xmm3, %xmm4
	movapd	42 * SIZE(AA), %xmm3
	mulpd	%xmm2, %xmm3
	addpd	%xmm3, %xmm5
	movapd	44 * SIZE(AA), %xmm3
	mulpd	%xmm2, %xmm3
	mulpd	46 * SIZE(AA), %xmm2
	addpd	%xmm3, %xmm6
	movapd	56 * SIZE(AA), %xmm3
	movd	(PRE + 48) * SIZE(AA), %mm0
	addpd	%xmm2, %xmm7
	movapd	12 * SIZE(BB), %xmm2
	mulpd	%xmm2, %xmm1
	addpd	%xmm1, %xmm4
	movapd	50 * SIZE(AA), %xmm1
	mulpd	%xmm2, %xmm1
	addpd	%xmm1, %xmm5
	movapd	52 * SIZE(AA), %xmm1
	mulpd	%xmm2, %xmm1
	mulpd	54 * SIZE(AA), %xmm2
	addpd	%xmm1, %xmm6
	movapd	64 * SIZE(AA), %xmm1
	movd	(PRE + 56) * SIZE(AA), %mm0
	addpd	%xmm2, %xmm7
	movapd	14 * SIZE(BB), %xmm2
	mulpd	%xmm2, %xmm3
	addpd	%xmm3, %xmm4
	movapd	58 * SIZE(AA), %xmm3
	mulpd	%xmm2, %xmm3
	addpd	%xmm3, %xmm5
	movapd	60 * SIZE(AA), %xmm3
	mulpd	%xmm2, %xmm3
	mulpd	62 * SIZE(AA), %xmm2
	addpd	%xmm3, %xmm6
	movapd	72 * SIZE(AA), %xmm3
	addpd	%xmm2, %xmm7
	movapd	24 * SIZE(BB), %xmm2

	addl   $64 * SIZE, AA
	addl   $16 * SIZE, BB
	decl   %eax
	jne    .L11
#endif

.L12:
	movapd	ALPHA,  %xmm3
	movl	K, %eax
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L14

.L13:
	movapd	 0 * SIZE(BB), %xmm0
	movapd	 0 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	movapd	 2 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm5
	movapd	 4 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm6
	mulpd	 6 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm7

	addl	$8 * SIZE, AA		# aoffset  += 8
	addl	$2 * SIZE, BB		# boffset1 += 8
	subl	$1, %eax
	jg	.L13
	ALIGN_4

.L14:
	mulpd	%xmm3, %xmm4
	mulpd	%xmm3, %xmm5
	mulpd	%xmm3, %xmm6
	mulpd	%xmm3, %xmm7

	movsd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0
	movsd	2 * SIZE(%esi), %xmm1
	movhpd	3 * SIZE(%esi), %xmm1
	movsd	4 * SIZE(%esi), %xmm2
	movhpd	5 * SIZE(%esi), %xmm2
	movsd	6 * SIZE(%esi), %xmm3
	movhpd	7 * SIZE(%esi), %xmm3

	addpd	%xmm0, %xmm4
	addpd	%xmm1, %xmm5
	addpd	%xmm2, %xmm6
	addpd	%xmm3, %xmm7

	movsd	%xmm4, 0 * SIZE(%esi)
	movhpd	%xmm4, 1 * SIZE(%esi)
	movsd	%xmm5, 2 * SIZE(%esi)
	movhpd	%xmm5, 3 * SIZE(%esi)
	movsd	%xmm6, 4 * SIZE(%esi)
	movhpd	%xmm6, 5 * SIZE(%esi)
	movsd	%xmm7, 6 * SIZE(%esi)
	movhpd	%xmm7, 7 * SIZE(%esi)

	addl	$8 * SIZE, %esi		# coffset += 4
	BRANCH
	decl	%ebx			# i --
	jg	.L10
	ALIGN_2	

.L20:
	movl	M,  %ebx
	testl	$4, %ebx
	jle	.L30

	leal	BUFFER, %ecx
	movl	K, %eax

	movapd	 0 * SIZE + BUFFER, %xmm2
	movapd	 0 * SIZE(%edx), %xmm0
	movapd	 8 * SIZE + BUFFER, %xmm3
	movapd	 8 * SIZE(%edx), %xmm1

	pxor	%xmm4, %xmm4
	pxor	%xmm5, %xmm5
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7

	sarl	$3, %eax
	je	.L22

.L21:	
	movapd	 0 * SIZE(BB), %xmm0
	movapd	 0 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	 2 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	movapd	 2 * SIZE(BB), %xmm0
	movapd	 4 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	 6 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	movapd	 4 * SIZE(BB), %xmm0
	movapd	 8 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	10 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	movapd	 6 * SIZE(BB), %xmm0
	movapd	12 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	14 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	movapd	 8 * SIZE(BB), %xmm0
	movapd	16 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	18 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	movapd	10 * SIZE(BB), %xmm0
	movapd	20 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	22 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	movapd	12 * SIZE(BB), %xmm0
	movapd	24 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	26 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	movapd	14 * SIZE(BB), %xmm0
	movapd	28 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	30 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	addl   $32 * SIZE, AA
	addl   $16 * SIZE, BB
	decl   %eax
	jne    .L21

.L22:
	movapd	ALPHA,  %xmm3
	movl	K, %eax
	andl	$7, %eax
	BRANCH
	je .L24

.L23:
	movapd	 0 * SIZE(BB), %xmm0
	movapd	 0 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4
	mulpd	 2 * SIZE(AA), %xmm0
	addpd	%xmm0, %xmm5

	addl	$4 * SIZE, AA		# aoffset  += 8
	addl	$2 * SIZE, BB		# boffset1 += 8
	subl	$1, %eax
	jg	.L23
	ALIGN_4

.L24:
	mulpd	%xmm3, %xmm4
	mulpd	%xmm3, %xmm5

	movsd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0
	movsd	2 * SIZE(%esi), %xmm1
	movhpd	3 * SIZE(%esi), %xmm1

	addpd	%xmm0, %xmm4
	addpd	%xmm1, %xmm5

	movsd	%xmm4, 0 * SIZE(%esi)
	movhpd	%xmm4, 1 * SIZE(%esi)
	movsd	%xmm5, 2 * SIZE(%esi)
	movhpd	%xmm5, 3 * SIZE(%esi)
	addl	$4 * SIZE, %esi		# coffset += 4
	ALIGN_4

.L30:
	movl	M,  %ebx
	testl	$2, %ebx
	jle	.L50

	leal	BUFFER, %ecx
	movl	K, %eax

	movapd	 0 * SIZE + BUFFER, %xmm2
	movapd	 0 * SIZE(AA), %xmm0
	movapd	 8 * SIZE + BUFFER, %xmm3
	movapd	 8 * SIZE(AA), %xmm1

	pxor	%xmm4, %xmm4
	pxor	%xmm5, %xmm5
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7

	sarl	$3, %eax
	je	.L32

.L31:	
	movapd	 0 * SIZE(BB), %xmm0
	movapd	 0 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	movapd	 2 * SIZE(BB), %xmm0
	movapd	 2 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	movapd	 4 * SIZE(BB), %xmm0
	movapd	 4 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	movapd	 6 * SIZE(BB), %xmm0
	movapd	 6 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	movapd	 8 * SIZE(BB), %xmm0
	movapd	 8 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	movapd	10 * SIZE(BB), %xmm0
	movapd	10 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	movapd	12 * SIZE(BB), %xmm0
	movapd	12 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	movapd	14 * SIZE(BB), %xmm0
	movapd	14 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	addl   $16 * SIZE, AA
	addl   $16 * SIZE, BB
	BRANCH
	decl   %eax
	jne    .L31

.L32:
	movapd	ALPHA,  %xmm3
	movl	K, %eax
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L34

.L33:
	movapd	 0 * SIZE(BB), %xmm0
	movapd	 0 * SIZE(AA), %xmm1
	mulpd	%xmm0, %xmm1
	addpd	%xmm1, %xmm4

	addl	$2 * SIZE, AA		# aoffset  += 8
	addl	$2 * SIZE, BB		# boffset1 += 8
	decl	%eax
	BRANCH
	jg	.L33
	ALIGN_4

.L34:
	mulpd	%xmm3, %xmm4

	movsd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0

	addpd	%xmm0, %xmm4

	movsd	%xmm4, 0 * SIZE(%esi)
	movhpd	%xmm4, 1 * SIZE(%esi)
	addl	$2 * SIZE, %esi
	ALIGN_2

.L50:
	movl	M,  %ebx
	testl	$1, %ebx
	jle	.L99

	leal	BUFFER, %ecx
	movl	K, %eax

	movsd	 0 * SIZE + BUFFER, %xmm2
	movsd	 0 * SIZE(AA), %xmm0
	movsd	 8 * SIZE + BUFFER, %xmm3
	movsd	 4 * SIZE(AA), %xmm1

	pxor	%xmm4, %xmm4
	pxor	%xmm5, %xmm5
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7

	sarl	$3, %eax
	je	.L52

.L51:	
	movsd	 0 * SIZE(AA), %xmm0
	mulsd	 0 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	movsd	 1 * SIZE(AA), %xmm0
	mulsd	 2 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	movsd	 2 * SIZE(AA), %xmm0
	mulsd	 4 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	movsd	 3 * SIZE(AA), %xmm0
	mulsd	 6 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	movsd	 4 * SIZE(AA), %xmm0
	mulsd	 8 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	movsd	 5 * SIZE(AA), %xmm0
	mulsd	10 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	movsd	 6 * SIZE(AA), %xmm0
	mulsd	12 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	movsd	 7 * SIZE(AA), %xmm0
	mulsd	14 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	addl   $ 8 * SIZE, AA
	addl   $16 * SIZE, BB
	BRANCH
	decl   %eax
	jne    .L51

.L52:
	movsd	ALPHA,  %xmm3
	movl	K, %eax
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L54

.L53:
	movsd	 0 * SIZE(AA), %xmm0
	mulsd	 0 * SIZE(BB), %xmm0
	addsd	%xmm0, %xmm4

	addl	$1 * SIZE, AA		# aoffset  += 8
	addl	$2 * SIZE, BB		# boffset1 += 8
	decl	%eax
	BRANCH
	jg	.L53
	ALIGN_4

.L54:
	movsd	0 * SIZE(%esi), %xmm0
	mulsd	%xmm3, %xmm4
	addsd	%xmm0, %xmm4
	movsd	%xmm4, 0 * SIZE(%esi)
	ALIGN_2

.L99:
	addl	LDC, C
	decl	J			# j --
	jg	.L01
	ALIGN_2

.L999:
	movl	OLD_STACK, %esp

	EMMS

	popl	%ebx
	popl	%esi
	popl	%edi
	popl	%ebp
	ret
	ALIGN_2


	EPILOGUE