trsm_kernel_rt_4x2_sse2.s

Optimized GotoBLAS libraries

	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:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	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

#if defined(LN) || defined(RT)
	movl	KK, %eax
#ifdef LN
	subl	$2, %eax
#else
	subl	$1, %eax
#endif

	movl	AORIG, AA
	movl	BORIG, B
	leal	BUFFER, BB

	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
	leal	(B,  %eax, 1), B
	leal	(BB, %eax, 2), BB
#endif

#if defined(LN) || defined(LT)
	movapd	 0 * SIZE(B), %xmm0
#else
	movapd	 0 * SIZE(AA), %xmm0
#endif

	subpd	%xmm4, %xmm0

#ifdef LN
	movapd	%xmm0, %xmm2
        unpckhpd %xmm2, %xmm2

	movsd	 3 * SIZE(AA), %xmm4
	mulsd	 %xmm4, %xmm2

	movsd	 2 * SIZE(AA), %xmm5
	mulsd	 %xmm2, %xmm5
	subsd	 %xmm5, %xmm0

	movsd	 0 * SIZE(AA), %xmm4
	mulsd	 %xmm4, %xmm0

	unpcklpd %xmm2, %xmm0
#endif

#ifdef LT
	movapd	%xmm0, %xmm2
        unpckhpd %xmm2, %xmm2

	movsd	 0 * SIZE(AA), %xmm4
	mulsd	 %xmm4, %xmm0

	movsd	 1 * SIZE(AA), %xmm5
	mulsd	 %xmm0, %xmm5
	subsd	 %xmm5, %xmm2
	movsd	 3 * SIZE(AA), %xmm4
	mulsd	 %xmm4, %xmm2

	unpcklpd %xmm2, %xmm0
#endif

#if defined(RN) || defined(RT)
	movsd	 0 * SIZE(B), %xmm4
	movhpd	 0 * SIZE(B), %xmm4
	mulpd	 %xmm4, %xmm0
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm0,   0 * SIZE(B)

	movsd	%xmm0,   0 * SIZE(BB)
	movsd	%xmm0,   1 * SIZE(BB)
	movhpd	%xmm0,   2 * SIZE(BB)
	movhpd	%xmm0,   3 * SIZE(BB)
#else
	movapd	%xmm0,   0 * SIZE(AA)
#endif

#ifdef LN
	subl	$2 * SIZE, %esi
#endif

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

#ifndef LN
	addl	$2 * SIZE, %esi
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
#ifdef LT
	addl	$2 * SIZE, B
#endif
#endif

#ifdef LN
	subl	$2, KK
	movl	BORIG, B
#endif

#ifdef LT
	addl	$2, KK
#endif

#ifdef RT
	movl	K, %eax
	movl	BORIG, B
	sall	$1 + BASE_SHIFT, %eax
	addl	%eax, AORIG
#endif
	ALIGN_2

.L150:
	movl	M,  %ebx
	testl	$1, %ebx
	jle	.L159

#ifdef LN
       movl	K, %eax
       sall	$0 + BASE_SHIFT, %eax
       subl	%eax, AORIG
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	AORIG, AA
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
#endif

	leal	BUFFER, BB

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

#if defined(LN) || defined(RT)
	movl	KK, %eax
	sall	$0 + BASE_SHIFT, %eax
	leal	(BB, %eax, 2), BB
#endif	

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#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:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	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

#if defined(LN) || defined(RT)
	movl	KK, %eax
	subl	$1, %eax

	movl	AORIG, AA
	movl	BORIG, B
	leal	BUFFER, BB

	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
	leal	(B,  %eax, 1), B
	leal	(BB, %eax, 2), BB
#endif

#if defined(LN) || defined(LT)
	movsd	 0 * SIZE(B), %xmm0
#else
	movsd	 0 * SIZE(AA), %xmm0
#endif

	subsd	%xmm4, %xmm0

#if defined(LN) || defined(LT)
	mulsd	 0 * SIZE(AA), %xmm0
#endif

#if defined(RN) || defined(RT)
	mulsd	 0 * SIZE(B), %xmm0
#endif

#if defined(LN) || defined(LT)
	movsd	%xmm0,   0 * SIZE(B)

	movsd	%xmm0,   0 * SIZE(BB)
	movsd	%xmm0,   1 * SIZE(BB)
#else
	movsd	%xmm0,   0 * SIZE(AA)
#endif

#ifdef LN
	subl	$1 * SIZE, %esi
#endif

	movsd	%xmm0,  0 * SIZE(%esi)

#ifndef LN
	addl	$1 * SIZE, %esi
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 1), AA
#ifdef LT
	addl	$1 * SIZE, B
#endif
#endif

#ifdef LN
	subl	$1, KK
	movl	BORIG, B
#endif

#ifdef LT
	addl	$1, KK
#endif

#ifdef RT
	movl	K, %eax
	movl	BORIG, B
	sall	$0 + BASE_SHIFT, %eax
	addl	%eax, AORIG
#endif
	ALIGN_2

.L159:
#ifdef LN
       movl	K, %eax
       leal	(, %eax, SIZE), %eax
       leal 	(B, %eax, 1), B
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(B,  %eax, 1), B
#endif

#ifdef RN
	addl	$1, KK
#endif

#ifdef RT
	subl	$1, KK
#endif
	ALIGN_2

.L100:
	movl	N, %eax
	sarl	$1, %eax	# j = (n >> 1)
	movl	%eax, J
	jle	.L999
	ALIGN_2
	
.L01:
/* Copying to Sub Buffer */
#ifdef LN
	movl	OFFSET, %eax
	addl	M, %eax
	movl	%eax, KK
#endif	

	leal	BUFFER, %ecx

#ifdef RT
       movl	K, %eax
       sall	$1 + BASE_SHIFT, %eax
       subl	%eax, B
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	B, BORIG
	leal	(, %eax, SIZE), %eax
	leal	(B,  %eax, 2), B
	leal	(BB, %eax, 4), BB
#endif	

#ifdef LT
	movl	OFFSET, %eax
	movl	%eax, KK
#endif

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$2, %eax
	jle	.L03
	ALIGN_2
	
.L02:
	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
	jne	.L02
	ALIGN_2

.L03:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$3, %eax
	BRANCH
	jle	.L05
	ALIGN_4

.L04:
	movsd	 0 * SIZE(B), %xmm0
	movsd	 1 * SIZE(B), %xmm1

	unpcklpd  %xmm0, %xmm0
	unpcklpd  %xmm1, %xmm1

	movapd	%xmm0,  0 * SIZE(%ecx)
	movapd	%xmm1,  2 * SIZE(%ecx)

	addl	$2 * SIZE, B
	addl	$4 * SIZE, %ecx
	decl	%eax
	jne	.L04
	ALIGN_4
	
.L05:
#if defined(LT) || defined(RN)
	movl	A, AA
#else
	movl	A, %eax
	movl	%eax, AORIG
#endif

	leal	(, LDC, 2), %eax

#ifdef RT
	subl	%eax, C
#endif
	movl	C, %esi		# coffset = c
#ifndef RT
	addl	%eax, C
#endif

	movl	M,  %ebx
	sarl	$2, %ebx	# i = (m >> 2)
	jle	.L30
	ALIGN_4

.L10:
#ifdef LN
       movl	K, %eax
       sall	$2 + BASE_SHIFT, %eax
       subl	%eax, AORIG
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	AORIG, AA
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
#endif

	leal	BUFFER, BB

#if defined(LN) || defined(RT)
	movl	KK, %eax
	sall	$1 + BASE_SHIFT, %eax
	leal	(BB, %eax, 2), BB
#endif	

	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

	prefetcht2	4 * SIZE(%esi)
	prefetcht2	4 * SIZE(%esi, LDC)

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif

#ifdef PENTIUM4
	andl	$-8, %eax
	je,pn	.L12
	sall	$3, %eax
	
.L1X:
	KERNEL1(32  *  0)
	KERNEL2(32  *  0)
	KERNEL3(32  *  0)
	KERNEL4(32  *  0)
	KERNEL5(32  *  0)
	KERNEL6(32  *  0)
	KERNEL7(32  *  0)
	KERNEL8(32  *  0)
	cmpl	$64 *  1, %eax
	jle,pn	.L11
	KERNEL1(32  *  1)
	KERNEL2(32  *  1)
	KERNEL3(32  *  1)
	KERNEL4(32  *  1)
	KERNEL5(32  *  1)
	KERNEL6(32  *  1)
	KERNEL7(32  *  1)
	KERNEL8(32  *  1)
	cmpl	$64 *  2, %eax
	jle,pn	.L11
	KERNEL1(32  *  2)
	KERNEL2(32  *  2)
	KERNEL3(32  *  2)
	KERNEL4(32  *  2)
	KERNEL5(32  *  2)
	KERNEL6(32  *  2)
	KERNEL7(32  *  2)
	KERNEL8(32  *  2)
	cmpl	$64 *  3, %eax
	jle,pn	.L11
	KERNEL1(32  *  3)
	KERNEL2(32  *  3)
	KERNEL3(32  *  3)
	KERNEL4(32  *  3)
	KERNEL5(32  *  3)
	KERNEL6(32  *  3)
	KERNEL7(32  *  3)
	KERNEL8(32  *  3)
	cmpl	$64 *  4, %eax
	jle,pn	.L11
	KERNEL1(32  *  4)
	KERNEL2(32  *  4)
	KERNEL3(32  *  4)
	KERNEL4(32  *  4)
	KERNEL5(32  *  4)
	KERNEL6(32  *  4)
	KERNEL7(32  *  4)
	KERNEL8(32  *  4)
	cmpl	$64 *  5, %eax
	jle,pn	.L11
	KERNEL1(32  *  5)
	KERNEL2(32  *  5)
	KERNEL3(32  *  5)
	KERNEL4(32  *  5)
	KERNEL5(32  *  5)
	KERNEL6(32  *  5)
	KERNEL7(32  *  5)
	KERNEL8(32  *  5)
	cmpl	$64 *  6, %eax
	jle,pn	.L11
	KERNEL1(32  *  6)
	KERNEL2(32  *  6)
	KERNEL3(32  *  6)
	KERNEL4(32  *  6)
	KERNEL5(32  *  6)
	KERNEL6(32  *  6)
	KERNEL7(32  *  6)
	KERNEL8(32  *  6)
	cmpl	$64 *  7, %eax
	jle,pn	.L11
	KERNEL1(32  *  7)
	KERNEL2(32  *  7)
	KERNEL3(32  *  7)
	KERNEL4(32  *  7)
	KERNEL5(32  *  7)
	KERNEL6(32  *  7)
	KERNEL7(32  *  7)
	KERNEL8(32  *  7)

	addl	$64 * 4  * SIZE, AA
	addl	$64 * 4  * SIZE, BB
	subl	$64 * 8, %eax
	jg,pt	.L1X

.L11:
	leal	(AA, %eax, 4), AA
	leal	(BB, %eax, 4), BB

#else
	sarl	$3, %eax
	je	.L12

.L11:	
	KERNEL1(32  *  0)
	KERNEL2(32  *  0)
	KERNEL3(32  *  0)
	KERNEL4(32  *  0)
	KERNEL5(32  *  0)
	KERNEL6(32  *  0)
	KERNEL7(32  *  0)
	KERNEL8(32  *  0)

	addl   $32 * SIZE, %ecx
	addl   $32 * SIZE, %edx
	decl   %eax
	jne    .L11
#endif

.L12:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH

	je .L14

.L13:
	mulpd	%xmm0, %xmm2
	mulpd	 2 * SIZE(BB), %xmm0
	addpd	%xmm2, %xmm4
	movapd	 0 * SIZE(BB), %xmm2
	addpd	%xmm0, %xmm5
	movapd	 2 * SIZE(AA), %xmm0
	mulpd	%xmm0, %xmm2
	mulpd	 2 * SIZE(BB), %xmm0
	addpd	%xmm2, %xmm6
	movapd	 4 * SIZE(BB), %xmm2
	addpd	%xmm0, %xmm7
	movapd	 4 * SIZE(AA), %xmm0

	addl	$4 * SIZE, AA		# aoffset  += 8
	addl	$4 * SIZE, BB		# boffset1 += 8
	subl	$1, %eax
	jg	.L13
	ALIGN_4 

.L14:
#if defined(LN) || defined(RT)
	movl	KK, %eax
#ifdef LN
	subl	$4, %eax
#else
	subl	$2, %eax
#endif

	movl	AORIG, AA
	movl	BORIG, B
	leal	BUFFER, BB

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

#if defined(LN) || defined(LT)
	movapd	 %xmm4, %xmm0
	unpcklpd %xmm5, %xmm4
	unpckhpd %xmm5, %xmm0

	movapd	 %xmm6, %xmm1
	unpcklpd %xmm7, %xmm6
	unpckhpd %xmm7, %xmm1

	movapd	 0 * SIZE(B), %xmm2
	movapd	 2 * SIZE(B), %xmm3
	movapd	 4 * SIZE(B), %xmm5
	movapd	 6 * SIZE(B), %xmm7

	subpd	%xmm4,  %xmm2
	subpd	%xmm0,  %xmm3
	subpd	%xmm6,  %xmm5
	subpd	%xmm1,  %xmm7
#else
	movapd	 0 * SIZE(AA), %xmm0
	movapd	 2 * SIZE(AA), %xmm1
	movapd	 4 * SIZE(AA), %xmm2
	movapd	 6 * SIZE(AA), %xmm3

	subpd	%xmm4, %xmm0
	subpd	%xmm6, %xmm1
	subpd	%xmm5, %xmm2
	subpd	%xmm7, %xmm3
#endif

#ifdef LN
	movsd	15 * SIZE(AA), %xmm0
	movhpd	15 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm7
	movsd	14 * SIZE(AA), %xmm0
	movhpd	14 * SIZE(AA), %xmm0
	mulpd	 %xmm7, %xmm0
	subpd	 %xmm0, %xmm5
	movsd	13 * SIZE(AA), %xmm0
	movhpd	13 * SIZE(AA), %xmm0
	mulpd	 %xmm7, %xmm0
	subpd	 %xmm0, %xmm3
	movsd	12 * SIZE(AA), %xmm0
	movhpd	12 * SIZE(AA), %xmm0
	mulpd	 %xmm7, %xmm0
	subpd	 %xmm0, %xmm2

	movsd	10 * SIZE(AA), %xmm0
	movhpd	10 * SIZE(AA), %xmm0
	mulpd	 %xmm0, %xmm5
	movsd	 9 * SIZE(AA), %xmm0
	movhpd	 9 * SIZE(AA), %xmm0
	mulpd	 %xmm5, %xmm0
	subpd	 %xmm0, %xmm3
	movsd	 8 * SIZE(AA), %xmm0
	movhpd	 8 * SIZE(AA), %xmm0
	mulpd	 %xmm5, %xmm0