gemm_kernel_4x2_sse2.s

Optimized GotoBLAS libraries

	movsd	%xmm5, 0 * SIZE(%esi, LDC)
	movhpd	%xmm5, 1 * SIZE(%esi, LDC)

	addl	$2 * SIZE, %esi		# coffset += 4

#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, 4), BB
#endif

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

	ALIGN_2

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


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

	leal	BUFFER, %ecx

	movapd	 0 * SIZE + BUFFER, %xmm2
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE + BUFFER, %xmm3
	pxor	%xmm6, %xmm6
	movsd	 4 * SIZE(AA), %xmm1
	pxor	%xmm7, %xmm7

#else

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

	movapd	 0 * SIZE(BB), %xmm2
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE(BB), %xmm3
	pxor	%xmm6, %xmm6
	movsd	 4 * SIZE(AA), %xmm1
	pxor	%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	$1, %eax
#else
	addl	$2, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L52

.L51:	
	mulsd	%xmm0, %xmm2
	mulsd	 2 * SIZE(BB), %xmm0
	addsd	%xmm2, %xmm4
	movsd	 4 * SIZE(BB), %xmm2
	addsd	%xmm0, %xmm5
	movsd	 1 * SIZE(AA), %xmm0
	mulsd	%xmm0, %xmm2
	mulsd	 6 * SIZE(BB), %xmm0
	addsd	%xmm2, %xmm4
	movsd	16 * SIZE(BB), %xmm2
	addsd	%xmm0, %xmm5
	movsd	 2 * SIZE(AA), %xmm0
	mulsd	%xmm0, %xmm3
	mulsd	10 * SIZE(BB), %xmm0
	addsd	%xmm3, %xmm4
	movsd	12 * SIZE(BB), %xmm3
	addsd	%xmm0, %xmm5
	movsd	 3 * SIZE(AA), %xmm0
	mulsd	%xmm0, %xmm3
	mulsd	14 * SIZE(BB), %xmm0
	addsd	%xmm3, %xmm4
	movsd	24 * SIZE(BB), %xmm3
	addsd	%xmm0, %xmm5
	movsd	 8 * SIZE(AA), %xmm0
	mulsd	%xmm1, %xmm2
	mulsd	18 * SIZE(BB), %xmm1
	addsd	%xmm2, %xmm4
	movsd	20 * SIZE(BB), %xmm2
	addsd	%xmm1, %xmm5
	movsd	 5 * SIZE(AA), %xmm1
	mulsd	%xmm1, %xmm2
	mulsd	22 * SIZE(BB), %xmm1
	addsd	%xmm2, %xmm4
	movsd	32 * SIZE(BB), %xmm2
	addsd	%xmm1, %xmm5
	movsd	 6 * SIZE(AA), %xmm1
	mulsd	%xmm1, %xmm3
	mulsd	26 * SIZE(BB), %xmm1
	addsd	%xmm3, %xmm4
	movsd	28 * SIZE(BB), %xmm3
	addsd	%xmm1, %xmm5
	movsd	 7 * SIZE(AA), %xmm1
	mulsd	%xmm1, %xmm3
	mulsd	30 * SIZE(BB), %xmm1
	addsd	%xmm3, %xmm4
	movsd	40 * SIZE(BB), %xmm3
	addsd	%xmm1, %xmm5
	movsd	12 * SIZE(AA), %xmm1

	addl   $ 8 * SIZE, AA
	addl   $32 * SIZE, BB
	BRANCH
	decl   %eax
	jne    .L51

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

.L53:
	mulsd	%xmm0, %xmm2
	mulsd	 2 * SIZE(BB), %xmm0
	addsd	%xmm2, %xmm4
	movsd	 4 * SIZE(BB), %xmm2
	addsd	%xmm0, %xmm5
	movsd	 1 * SIZE(AA), %xmm0

	addl	$1 * SIZE, AA		# aoffset  += 8
	addl	$4 * SIZE, BB		# boffset1 += 8
	decl	%eax
	BRANCH
	jg	.L53
	ALIGN_4

.L54:
	addsd	%xmm6, %xmm4
	addsd	%xmm7, %xmm5

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

#ifndef TRMMKERNEL
	addsd	0 * SIZE(%esi), %xmm4
	addsd	0 * SIZE(%esi, LDC), %xmm5
#endif

	movsd	%xmm4, 0 * SIZE(%esi)
	movsd	%xmm5, 0 * SIZE(%esi, LDC)
	addl	$1 * SIZE, %esi

#if defined(TRMMKERNEL) && defined(LEFT)
	addl	$1, KK
#endif
	ALIGN_2

.L99:
#if defined(TRMMKERNEL) && !defined(LEFT)
	addl	$2, KK
#endif

	leal	(, LDC, 2), %eax
	addl	%eax, C			# c += 2 * ldc
	BRANCH
	decl	J			# j --
	jg	.L01
	ALIGN_2

.L100:
	movl	N, %eax
	testl	$1, %eax
	jle	.L999
	ALIGN_2
	
.L101:
#if defined(TRMMKERNEL) && defined(LEFT)
	movl	OFFSET, %eax
	movl	%eax, KK
#endif	

/* Copying to Sub Buffer */
	leal	BUFFER, %ecx

	movl	K, %eax
	sarl	$3, %eax
	jle	.L103
	ALIGN_4
	
.L102:
	movsd	 0 * SIZE(B), %xmm0
	movsd	 1 * SIZE(B), %xmm1
	movsd	 2 * SIZE(B), %xmm2
	movsd	 3 * SIZE(B), %xmm3
	movsd	 4 * SIZE(B), %xmm4
	movsd	 5 * SIZE(B), %xmm5
	movsd	 6 * SIZE(B), %xmm6
	movsd	 7 * SIZE(B), %xmm7

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

	movapd	%xmm0,  0 * SIZE(%ecx)
	movapd	%xmm1,  2 * SIZE(%ecx)
	movapd	%xmm2,  4 * SIZE(%ecx)
	movapd	%xmm3,  6 * SIZE(%ecx)
	movapd	%xmm4,  8 * SIZE(%ecx)
	movapd	%xmm5, 10 * SIZE(%ecx)
	movapd	%xmm6, 12 * SIZE(%ecx)
	movapd	%xmm7, 14 * SIZE(%ecx)

	prefetcht0	104 * SIZE(B)

	addl	$ 8 * SIZE, B
	addl	$16 * SIZE, %ecx
	decl	%eax
	BRANCH
	jne	.L102
	ALIGN_2

.L103:
	movl	K, %eax
	andl	$7, %eax
	BRANCH
	jle	.L105
	ALIGN_2

.L104:
	movsd	 0 * SIZE(B), %xmm0

	unpcklpd  %xmm0, %xmm0

	movapd	%xmm0,  0 * SIZE(%ecx)

	addl	$1 * SIZE, B
	addl	$2 * SIZE, %ecx
	decl	%eax
	jne	.L104
	ALIGN_4
	
.L105:
	movl	C, %esi		# coffset = c
	movl	A, %edx		# aoffset = a
	movl	M,  %ebx
	sarl	$2, %ebx	# i = (m >> 2)
	jle	.L130
	ALIGN_4

.L110:
#if !defined(TRMMKERNEL) || \
	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))

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

#else

	leal	BUFFER, BB
	movl	KK, %eax
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
	leal	(BB, %eax, 2), BB

	movapd	 0 * SIZE(BB), %xmm2
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE(BB), %xmm3
	pxor	%xmm6, %xmm6
	movapd	 8 * SIZE(AA), %xmm1
	pxor	%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	$4, %eax
#else
	addl	$1, %eax
#endif
	movl	%eax, KKK
#endif
	sarl	$3, %eax
	je	.L112

.L111:	
	mulpd	%xmm2, %xmm0
	mulpd	 2 * SIZE(AA), %xmm2
	addpd	%xmm0, %xmm4
	movapd	 4 * SIZE(AA), %xmm0
	addpd	%xmm2, %xmm6
	movapd	 2 * SIZE(BB), %xmm2
	mulpd	%xmm2, %xmm0
	mulpd	 6 * SIZE(AA), %xmm2
	addpd	%xmm0, %xmm5
	movapd	16 * SIZE(AA), %xmm0
	addpd	%xmm2, %xmm7
	movapd	 4 * SIZE(BB), %xmm2
	mulpd	%xmm2, %xmm1
	mulpd	10 * SIZE(AA), %xmm2
	addpd	%xmm1, %xmm4
	movapd	12 * SIZE(AA), %xmm1
	addpd	%xmm2, %xmm6
	movapd	 6 * SIZE(BB), %xmm2
	mulpd	%xmm2, %xmm1
	mulpd	14 * SIZE(AA), %xmm2
	addpd	%xmm1, %xmm5
	movapd	24 * SIZE(AA), %xmm1
	addpd	%xmm2, %xmm7
	movapd	16 * SIZE(BB), %xmm2
	mulpd	%xmm3, %xmm0
	mulpd	18 * SIZE(AA), %xmm3
	addpd	%xmm0, %xmm4
	movapd	20 * SIZE(AA), %xmm0
	addpd	%xmm3, %xmm6
	movapd	10 * SIZE(BB), %xmm3
	mulpd	%xmm3, %xmm0
	mulpd	22 * SIZE(AA), %xmm3
	addpd	%xmm0, %xmm5
	movapd	32 * SIZE(AA), %xmm0
	addpd	%xmm3, %xmm7
	movapd	12 * SIZE(BB), %xmm3
	mulpd	%xmm3, %xmm1
	mulpd	26 * SIZE(AA), %xmm3
	addpd	%xmm1, %xmm4
	movapd	28 * SIZE(AA), %xmm1
	addpd	%xmm3, %xmm6
	movapd	14 * SIZE(BB), %xmm3
	mulpd	%xmm3, %xmm1
	mulpd	30 * SIZE(AA), %xmm3
	addpd	%xmm1, %xmm5
	movapd	40 * SIZE(AA), %xmm1
	addpd	%xmm3, %xmm7
	movapd	24 * SIZE(BB), %xmm3

	addl   $32 * SIZE, AA
	addl   $16 * SIZE, BB
	decl   %eax
	jne    .L111

.L112:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movapd	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L114

.L113:
	mulpd	%xmm2, %xmm0
	mulpd	 2 * SIZE(AA), %xmm2
	addpd	%xmm0, %xmm4
	movapd	 4 * SIZE(AA), %xmm0
	addpd	%xmm2, %xmm6
	movapd	 2 * SIZE(BB), %xmm2

	addl	$4 * SIZE, AA		# aoffset  += 8
	addl	$2 * SIZE, BB		# boffset1 += 8
	subl	$1, %eax
	jg	.L113
	ALIGN_4

.L114:
	addpd	%xmm5, %xmm4
	addpd	%xmm7, %xmm6

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

#ifndef TRMMKERNEL
	movsd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0
	addpd	%xmm0, %xmm4

	movsd	2 * SIZE(%esi), %xmm1
	movhpd	3 * SIZE(%esi), %xmm1
	addpd	%xmm1, %xmm6
#endif

	movsd	%xmm4, 0 * SIZE(%esi)
	unpckhpd %xmm4, %xmm4
	movsd	%xmm4, 1 * SIZE(%esi)

	movsd	%xmm6, 2 * SIZE(%esi)
	unpckhpd %xmm6, %xmm6
	movsd	%xmm6, 3 * SIZE(%esi)

#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, %esi		# coffset += 4
	BRANCH
	decl	%ebx			# i --
	jg	.L110
	ALIGN_2	

.L130:
	movl	M,  %ebx
	testl	$2, %ebx
	jle	.L150

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

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

#else

	leal	BUFFER, BB
	movl	KK, %eax
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 2), BB

	movapd	 0 * SIZE(BB), %xmm2
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE(BB), %xmm3
	pxor	%xmm6, %xmm6
	movapd	 8 * SIZE(AA), %xmm1
	pxor	%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	.L132

.L131:	
	mulpd	%xmm0, %xmm2
	movapd	 2 * SIZE(AA), %xmm0
	addpd	%xmm2, %xmm4
	mulpd	 2 * SIZE(BB), %xmm0
	movapd	16 * SIZE(BB), %xmm2
	addpd	%xmm0, %xmm5
	movapd	 4 * SIZE(AA), %xmm0
	mulpd	 4 * SIZE(BB), %xmm0
	addpd	%xmm0, %xmm6
	movapd	 6 * SIZE(AA), %xmm0
	mulpd	 6 * SIZE(BB), %xmm0
	addpd	%xmm0, %xmm7
	movapd	16 * SIZE(AA), %xmm0
	mulpd	%xmm1, %xmm3
	movapd	10 * SIZE(AA), %xmm1
	addpd	%xmm3, %xmm4
	mulpd	10 * SIZE(BB), %xmm1
	movapd	24 * SIZE(BB), %xmm3
	addpd	%xmm1, %xmm5
	movapd	12 * SIZE(AA), %xmm1
	mulpd	12 * SIZE(BB), %xmm1
	addpd	%xmm1, %xmm6
	movapd	14 * SIZE(AA), %xmm1
	mulpd	14 * SIZE(BB), %xmm1
	addpd	%xmm1, %xmm7
	movapd	24 * SIZE(AA), %xmm1

	addl   $16 * SIZE, AA
	addl   $16 * SIZE, BB
	BRANCH
	decl   %eax
	jne    .L131

.L132:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movapd	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L134

.L133:
	movapd	 0 * SIZE(AA), %xmm0
	mulpd	 0 * SIZE(BB), %xmm0
	addpd	%xmm0, %xmm4

	addl	$2 * SIZE, AA		# aoffset  += 8
	addl	$2 * SIZE, BB		# boffset1 += 8
	decl	%eax
	BRANCH
	jg	.L133
	ALIGN_4

.L134:
	addpd	%xmm5, %xmm4
	addpd	%xmm7, %xmm6
	addpd	%xmm6, %xmm4

	mulpd	%xmm3, %xmm4

#ifndef TRMMKERNEL
	movsd	0 * SIZE(%esi), %xmm0
	movhpd	1 * SIZE(%esi), %xmm0
	addpd	%xmm0, %xmm4
#endif

	movsd	%xmm4, 0 * SIZE(%esi)
	unpckhpd %xmm4, %xmm4
	movsd	%xmm4, 1 * SIZE(%esi)

	addl	$2 * SIZE, %esi		# coffset += 4

#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
	ALIGN_2

.L150:
	movl	M,  %ebx
	testl	$1, %ebx
	jle	.L999

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

	leal	BUFFER, BB
	movapd	 0 * SIZE + BUFFER, %xmm2
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE + BUFFER, %xmm3
	pxor	%xmm6, %xmm6
	movapd	 4 * SIZE(AA), %xmm1
	pxor	%xmm7, %xmm7

#else

	leal	BUFFER, BB
	movl	KK, %eax
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 2), BB

	movapd	 0 * SIZE(BB), %xmm2
	pxor	%xmm4, %xmm4
	movapd	 0 * SIZE(AA), %xmm0
	pxor	%xmm5, %xmm5
	movapd	 8 * SIZE(BB), %xmm3
	pxor	%xmm6, %xmm6
	movapd	 4 * SIZE(AA), %xmm1
	pxor	%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	.L152

.L151:	
	mulsd	%xmm0, %xmm2
	movsd	 1 * SIZE(AA), %xmm0
	addsd	%xmm2, %xmm4
	mulsd	 2 * SIZE(BB), %xmm0
	movsd	16 * SIZE(BB), %xmm2
	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	 8 * SIZE(AA), %xmm0
	mulsd	%xmm1, %xmm3
	movsd	 5 * SIZE(AA), %xmm1
	addsd	%xmm3, %xmm4
	mulsd	10 * SIZE(BB), %xmm1
	movsd	24 * SIZE(BB), %xmm3
	addsd	%xmm1, %xmm4
	movsd	 6 * SIZE(AA), %xmm1
	mulsd	12 * SIZE(BB), %xmm1
	addsd	%xmm1, %xmm4
	movsd	 7 * SIZE(AA), %xmm1
	mulsd	14 * SIZE(BB), %xmm1
	addsd	%xmm1, %xmm4
	movsd	12 * SIZE(AA), %xmm1

	addl   $ 8 * SIZE, AA
	addl   $16 * SIZE, BB
	BRANCH
	decl   %eax
	jne    .L151

.L152:
#ifndef TRMMKERNEL
	movl	K, %eax
#else
	movl	KKK, %eax
#endif
	movsd	ALPHA,  %xmm3
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L154

.L153:
	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	.L153
	ALIGN_4

.L154:
	addsd	%xmm6, %xmm4
	addsd	%xmm7, %xmm5

	mulsd	%xmm3, %xmm4

#ifndef TRMMKERNEL
	addsd	0 * SIZE(%esi), %xmm4
#endif
	movsd	%xmm4, 0 * SIZE(%esi)
	ALIGN_2

.L999:
	movl	OLD_STACK, %esp

	EMMS

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


	EPILOGUE