trsm_kernel_ln_4x4_sse3.s

Optimized GotoBLAS libraries

	subsd	 %xmm6, %xmm0
	movsd	12 * SIZE(BO), %xmm7
	mulsd	 %xmm1, %xmm7
	subsd	 %xmm7, %xmm2

	movsd	10 * SIZE(BO), %xmm4
	mulsd	 %xmm4, %xmm3

	movsd	 9 * SIZE(BO), %xmm5
	mulsd	 %xmm3, %xmm5
	subsd	 %xmm5, %xmm0
	movsd	 8 * SIZE(BO), %xmm6
	mulsd	 %xmm3, %xmm6
	subsd	 %xmm6, %xmm2

	movsd	 5 * SIZE(BO), %xmm4
	mulsd	 %xmm4, %xmm0

	movsd	 4 * SIZE(BO), %xmm5
	mulsd	 %xmm0, %xmm5
	subsd	 %xmm5, %xmm2

	movsd	 0 * SIZE(BO), %xmm4
	mulsd	 %xmm4, %xmm2

	unpcklpd %xmm0, %xmm2
	unpcklpd %xmm1, %xmm3

#endif

#ifdef LN
	subq	$1 * SIZE, CO1
	subq	$1 * SIZE, CO2
#endif

#if defined(LN) || defined(LT)
	movsd	%xmm2,  0 * SIZE(CO1)
	movhpd	%xmm2,  0 * SIZE(CO2)
	movsd	%xmm3,  0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm3,  0 * SIZE(CO2, LDC, 2)
#else
	movsd	%xmm2,  0 * SIZE(CO1)
	movhpd	%xmm2,  0 * SIZE(CO2)
	movsd	%xmm3,  0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm3,  0 * SIZE(CO2, LDC, 2)
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm2,   0 * SIZE(BO)
	movapd	%xmm3,   2 * SIZE(BO)
#else
	movapd	%xmm2,   0 * SIZE(AO)
	movapd	%xmm3,   2 * SIZE(AO)
#endif

#ifndef LN
	addq	$1 * SIZE, CO1
	addq	$1 * SIZE, CO2
#endif

#if defined(LT) || defined(RN)
	movq	K,  %rax
	subq	KK, %rax
	leaq	(,%rax, SIZE), %rax
	leaq	(AO, %rax, 1), AO
	leaq	(BO, %rax, 4), BO
#endif

#ifdef LN
	subq	$1, KK
#endif

#ifdef LT
	addq	$1, KK
#endif

#ifdef RT
       movq	K, %rax
       salq	$0 + BASE_SHIFT, %rax
       addq	%rax, AORIG
#endif
	ALIGN_4	
	
.L30:
	testq	$2, M
	BRANCH
	je	.L20
	ALIGN_4

.L21:

#ifdef LN
       movq	K, %rax
       salq	$1 + BASE_SHIFT, %rax
       subq	%rax, AORIG
#endif

#if defined(LN) || defined(RT)
	movq	KK, %rax
	leaq	(, %rax, SIZE), %rax
	movq	AORIG, AO
	leaq	(AO, %rax, 2), AO
	leaq	(B,  %rax, 4), BO
#else
	movq	B, BO
#endif	

	movapd	 0 * SIZE(AO), %xmm8
	pxor	%xmm0, %xmm0
	movddup	 0 * SIZE(BO), %xmm9
	pxor	%xmm1, %xmm1
	movapd	 8 * SIZE(AO), %xmm10
	pxor	%xmm2, %xmm2
	movddup	 8 * SIZE(BO), %xmm11
	pxor	%xmm3, %xmm3

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	sarq	$3, %rax
	je	.L25
	ALIGN_4

.L22:
	mulpd	%xmm8, %xmm9
	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AO)
	addpd	%xmm9, %xmm0
	movddup	 1 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm1
	movddup	 2 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm2
	movddup	 3 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	movapd	 2 * SIZE(AO), %xmm8
	addpd	%xmm9, %xmm3
	movddup	 4 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm0
	movddup	 5 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm1
	movddup	 6 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm2
	movddup	 7 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	movapd	 4 * SIZE(AO), %xmm8
	addpd	%xmm9, %xmm3
	movddup	16 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm11
	addpd	%xmm11, %xmm0
	movddup	 9 * SIZE(BO), %xmm11
	mulpd	%xmm8, %xmm11
	addpd	%xmm11, %xmm1
	movddup	10 * SIZE(BO), %xmm11
	mulpd	%xmm8, %xmm11
	addpd	%xmm11, %xmm2
	movddup	11 * SIZE(BO), %xmm11
	mulpd	%xmm8, %xmm11
	movapd	 6 * SIZE(AO), %xmm8
	addpd	%xmm11, %xmm3
	movddup	12 * SIZE(BO), %xmm11
	mulpd	%xmm8, %xmm11
	addpd	%xmm11, %xmm0
	movddup	13 * SIZE(BO), %xmm11
	mulpd	%xmm8, %xmm11
	addpd	%xmm11, %xmm1
	movddup	14 * SIZE(BO), %xmm11
	mulpd	%xmm8, %xmm11
	addpd	%xmm11, %xmm2
	movddup	15 * SIZE(BO), %xmm11
	mulpd	%xmm8, %xmm11
	movapd	16 * SIZE(AO), %xmm8
	addpd	%xmm11, %xmm3
	movddup	24 * SIZE(BO), %xmm11
	mulpd	%xmm10, %xmm9
	addpd	%xmm9, %xmm0
	movddup	17 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	addpd	%xmm9, %xmm1
	movddup	18 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	addpd	%xmm9, %xmm2
	movddup	19 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	movapd	10 * SIZE(AO), %xmm10
	addpd	%xmm9, %xmm3
	movddup	20 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	addpd	%xmm9, %xmm0
	movddup	21 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	addpd	%xmm9, %xmm1
	movddup	22 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	addpd	%xmm9, %xmm2
	movddup	23 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	movapd	12 * SIZE(AO), %xmm10
	addpd	%xmm9, %xmm3
	movddup	32 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm11
	addpd	%xmm11, %xmm0
	movddup	25 * SIZE(BO), %xmm11
	mulpd	%xmm10, %xmm11
	addpd	%xmm11, %xmm1
	movddup	26 * SIZE(BO), %xmm11
	mulpd	%xmm10, %xmm11
	addpd	%xmm11, %xmm2
	movddup	27 * SIZE(BO), %xmm11
	mulpd	%xmm10, %xmm11
	movapd	14 * SIZE(AO), %xmm10
	addpd	%xmm11, %xmm3
	movddup	28 * SIZE(BO), %xmm11
	mulpd	%xmm10, %xmm11
	addpd	%xmm11, %xmm0
	movddup	29 * SIZE(BO), %xmm11
	mulpd	%xmm10, %xmm11
	addpd	%xmm11, %xmm1
	movddup	30 * SIZE(BO), %xmm11
	mulpd	%xmm10, %xmm11
	addpd	%xmm11, %xmm2
	movddup	31 * SIZE(BO), %xmm11
	mulpd	%xmm10, %xmm11
	movapd	24 * SIZE(AO), %xmm10
	addpd	%xmm11, %xmm3
	movddup	40 * SIZE(BO), %xmm11

	addq   $16 * SIZE, AO
	addq   $32 * SIZE, BO
	decq   %rax
	jne    .L22
	ALIGN_4

.L25:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$7, %rax		# if (k & 1)
	BRANCH
	je .L29
	ALIGN_4

.L26:
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm0
	movddup	 1 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm1
	movddup	 2 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm2
	movddup	 3 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	movapd	 2 * SIZE(AO), %xmm8
	addpd	%xmm9, %xmm3
	movddup	 4 * SIZE(BO), %xmm9

	addq	$2 * SIZE, AO		# aoffset  += 4
	addq	$4 * SIZE, BO		# boffset1 += 8
	decq	%rax
	jg	.L26
	ALIGN_4

.L29:
#if defined(LN) || defined(RT)
	movq	KK, %rax
#ifdef LN
	subq	$2, %rax
#else
	subq	$4, %rax
#endif

	leaq	(, %rax, SIZE), %rax
	movq	AORIG, AO
	leaq	(AO, %rax, 2), AO
	leaq	(B,  %rax, 4), BO
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm0, %xmm8
	unpcklpd %xmm1, %xmm0
	unpckhpd %xmm1, %xmm8

	movapd	%xmm2, %xmm10
	unpcklpd %xmm3, %xmm2
	unpckhpd %xmm3, %xmm10

	movapd	 0 * SIZE(BO), %xmm1
	movapd	 2 * SIZE(BO), %xmm3
	movapd	 4 * SIZE(BO), %xmm5
	movapd	 6 * SIZE(BO), %xmm7

	subpd	%xmm0,  %xmm1
	subpd	%xmm2,  %xmm3
	subpd	%xmm8,  %xmm5
	subpd	%xmm10, %xmm7
#else

	movapd	 0 * SIZE(AO), %xmm8
	movapd	 2 * SIZE(AO), %xmm10
	movapd	 4 * SIZE(AO), %xmm12
	movapd	 6 * SIZE(AO), %xmm14

	subpd	%xmm0, %xmm8
	subpd	%xmm1, %xmm10
	subpd	%xmm2, %xmm12
	subpd	%xmm3, %xmm14
#endif

#ifdef LN
	movddup	 3 * SIZE(AO), %xmm0
	mulpd	 %xmm0, %xmm5
	mulpd	 %xmm0, %xmm7

	movddup	 2 * SIZE(AO), %xmm2
	mulpd	 %xmm5, %xmm2
	subpd	 %xmm2, %xmm1
	movddup	 2 * SIZE(AO), %xmm2
	mulpd	 %xmm7, %xmm2
	subpd	 %xmm2, %xmm3

	movddup	 0 * SIZE(AO), %xmm0
	mulpd	 %xmm0, %xmm1
	mulpd	 %xmm0, %xmm3
#endif

#ifdef LT
	movddup	 0 * SIZE(AO), %xmm0
	mulpd	 %xmm0, %xmm1
	mulpd	 %xmm0, %xmm3

	movddup	 1 * SIZE(AO), %xmm2
	mulpd	 %xmm1, %xmm2
	subpd	 %xmm2, %xmm5
	movddup	 1 * SIZE(AO), %xmm2
	mulpd	 %xmm3, %xmm2
	subpd	 %xmm2, %xmm7

	movddup	 3 * SIZE(AO), %xmm0
	mulpd	 %xmm0, %xmm5
	mulpd	 %xmm0, %xmm7
#endif

#ifdef RN
	movddup	 0 * SIZE(BO), %xmm0
	mulpd	 %xmm0, %xmm8

	movddup	 1 * SIZE(BO), %xmm1
	mulpd	 %xmm8, %xmm1
	subpd	 %xmm1, %xmm10
	movddup	 2 * SIZE(BO), %xmm2
	mulpd	 %xmm8, %xmm2
	subpd	 %xmm2, %xmm12
	movddup	 3 * SIZE(BO), %xmm3
	mulpd	 %xmm8, %xmm3
	subpd	 %xmm3, %xmm14

	movddup	 5 * SIZE(BO), %xmm0
	mulpd	 %xmm0, %xmm10
	movddup	 6 * SIZE(BO), %xmm1
	mulpd	 %xmm10, %xmm1
	subpd	 %xmm1, %xmm12
	movddup	 7 * SIZE(BO), %xmm2
	mulpd	 %xmm10, %xmm2
	subpd	 %xmm2, %xmm14

	movddup	10 * SIZE(BO), %xmm0
	mulpd	 %xmm0, %xmm12

	movddup	11 * SIZE(BO), %xmm1
	mulpd	 %xmm12, %xmm1
	subpd	 %xmm1, %xmm14

	movddup	15 * SIZE(BO), %xmm0
	mulpd	 %xmm0, %xmm14
#endif

#ifdef RT
	movddup	15 * SIZE(BO), %xmm0
	mulpd	 %xmm0, %xmm14

	movddup	14 * SIZE(BO), %xmm1
	mulpd	 %xmm14, %xmm1
	subpd	 %xmm1, %xmm12
	movddup	13 * SIZE(BO), %xmm2
	mulpd	 %xmm14, %xmm2
	subpd	 %xmm2, %xmm10
	movddup	12 * SIZE(BO), %xmm3
	mulpd	 %xmm14, %xmm3
	subpd	 %xmm3, %xmm8

	movddup	10 * SIZE(BO), %xmm0
	mulpd	 %xmm0, %xmm12
	movddup	 9 * SIZE(BO), %xmm1
	mulpd	 %xmm12, %xmm1
	subpd	 %xmm1, %xmm10
	movddup	 8 * SIZE(BO), %xmm2
	mulpd	 %xmm12, %xmm2
	subpd	 %xmm2, %xmm8

	movddup	 5 * SIZE(BO), %xmm0
	mulpd	 %xmm0, %xmm10
	movddup	 4 * SIZE(BO), %xmm1
	mulpd	 %xmm10, %xmm1
	subpd	 %xmm1, %xmm8

	movddup	 0 * SIZE(BO), %xmm0
	mulpd	 %xmm0, %xmm8
#endif

#ifdef LN
	subq	$2 * SIZE, CO1
	subq	$2 * SIZE, CO2
#endif

#if defined(LN) || defined(LT)
	movsd	%xmm1,  0 * SIZE(CO1)
	movsd	%xmm5,  1 * SIZE(CO1)
	movhpd	%xmm1,  0 * SIZE(CO2)
	movhpd	%xmm5,  1 * SIZE(CO2)

	movsd	%xmm3,  0 * SIZE(CO1, LDC, 2)
	movsd	%xmm7,  1 * SIZE(CO1, LDC, 2)
	movhpd	%xmm3,  0 * SIZE(CO2, LDC, 2)
	movhpd	%xmm7,  1 * SIZE(CO2, LDC, 2)
#else
	movsd	%xmm8,  0 * SIZE(CO1)
	movhpd	%xmm8,  1 * SIZE(CO1)
	movsd	%xmm10,  0 * SIZE(CO2)
	movhpd	%xmm10,  1 * SIZE(CO2)

	movsd	%xmm12,  0 * SIZE(CO1, LDC, 2)
	movhpd	%xmm12,  1 * SIZE(CO1, LDC, 2)
	movsd	%xmm14,  0 * SIZE(CO2, LDC, 2)
	movhpd	%xmm14,  1 * SIZE(CO2, LDC, 2)
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm1,   0 * SIZE(BO)
	movapd	%xmm3,   2 * SIZE(BO)
	movapd	%xmm5,   4 * SIZE(BO)
	movapd	%xmm7,   6 * SIZE(BO)
#else
	movapd	%xmm8,   0 * SIZE(AO)
	movapd	%xmm10,  2 * SIZE(AO)
	movapd	%xmm12,  4 * SIZE(AO)
	movapd	%xmm14,  6 * SIZE(AO)
#endif

#ifndef LN
	addq	$2 * SIZE, CO1
	addq	$2 * SIZE, CO2
#endif

#if defined(LT) || defined(RN)
	movq	K,  %rax
	subq	KK, %rax
	leaq	(,%rax, SIZE), %rax
	leaq	(AO, %rax, 2), AO
	leaq	(BO, %rax, 4), BO
#endif

#ifdef LN
	subq	$2, KK
#endif

#ifdef LT
	addq	$2, KK
#endif

#ifdef RT
       movq	K, %rax
       salq	$1 + BASE_SHIFT, %rax
       addq	%rax, AORIG
#endif
	ALIGN_4	

.L20:
	movq	M,  I
	sarq	$2, I	# i = (m >> 2)
	jle	.L39
	ALIGN_4

.L11:

#ifdef LN
       movq	K, %rax
       salq	$2 + BASE_SHIFT, %rax
       subq	%rax, AORIG
#endif

#if defined(LN) || defined(RT)
	movq	KK, %rax
	leaq	(, %rax, SIZE), %rax
	movq	AORIG, AO
	leaq	(AO, %rax, 4), AO
	leaq	(B,  %rax, 4), BO
#else
	movq	B, BO
#endif	

	prefetcht0	  0 * SIZE(BB)
	subq	   $-8 * SIZE, BB

	movapd	 0 * SIZE(AO), %xmm8
	pxor	%xmm0, %xmm0
	movddup	 0 * SIZE(BO), %xmm9
	pxor	%xmm1, %xmm1
	movapd	 8 * SIZE(AO), %xmm10
	pxor	%xmm2, %xmm2
	movddup	 8 * SIZE(BO), %xmm11
	pxor	%xmm3, %xmm3

	movapd	16 * SIZE(AO), %xmm12
	movddup 16 * SIZE(BO), %xmm13
	movapd	24 * SIZE(AO), %xmm14
	movddup	24 * SIZE(BO), %xmm15

#ifdef LN
	prefetchnta    -4 * SIZE(CO1)
	pxor	%xmm4, %xmm4
	prefetchnta    -4 * SIZE(CO2)
	pxor	%xmm5, %xmm5
	prefetchnta    -4 * SIZE(CO1, LDC, 2)
	pxor	%xmm6, %xmm6
	prefetchnta    -4 * SIZE(CO2, LDC, 2)
	pxor	%xmm7, %xmm7
#else
	prefetchnta     4 * SIZE(CO1)
	pxor	%xmm4, %xmm4
	prefetchnta     4 * SIZE(CO2)
	pxor	%xmm5, %xmm5
	prefetchnta     4 * SIZE(CO1, LDC, 2)
	pxor	%xmm6, %xmm6
	prefetchnta     4 * SIZE(CO2, LDC, 2)
	pxor	%xmm7, %xmm7
#endif

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif

#if 1
	andq	$-8, %rax
	salq	$4, %rax
	je	.L15
.L1X:	
	KERNEL1 (16  *  0)
	KERNEL2 (16  *  0)
	KERNEL3 (16  *  0)
	KERNEL4 (16  *  0)
	KERNEL5 (16  *  0)
	KERNEL6 (16  *  0)
	KERNEL7 (16  *  0)
	KERNEL8 (16  *  0)
	KERNEL9 (16  *  0)
	KERNEL10(16  *  0)
	KERNEL11(16  *  0)
	KERNEL12(16  *  0)
	KERNEL13(16  *  0)
	KERNEL14(16  *  0)
	KERNEL15(16  *  0)
	KERNEL16(16  *  0)
	cmpq	$128 *  1, %rax
	jle,pn	.L12
	KERNEL1 (16  *  1)
	KERNEL2 (16  *  1)
	KERNEL3 (16  *  1)
	KERNEL4 (16  *  1)
	KERNEL5 (16  *  1)
	KERNEL6 (16  *  1)
	KERNEL7 (16  *  1)
	KERNEL8 (16  *  1)
	KERNEL9 (16  *  1)
	KERNEL10(16  *  1)
	KERNEL11(16  *  1)
	KERNEL12(16  *  1)
	KERNEL13(16  *  1)
	KERNEL14(16  *  1)
	KERNEL15(16  *  1)
	KERNEL16(16  *  1)
	cmpq	$128 *  2, %rax
	jle,pn	.L12
	KERNEL1 (16  *  2)
	KERNEL2 (16  *  2)
	KERNEL3 (16  *  2)
	KERNEL4 (16  *  2)
	KERNEL5 (16  *  2)
	KERNEL6 (16  *  2)
	KERNEL7 (16  *  2)
	KERNEL8 (16  *  2)
	KERNEL9 (16  *  2)
	KERNEL10(16  *  2)
	KERNEL11(16  *  2)
	KERNEL12(16  *  2)
	KERNEL13(16  *  2)
	KERNEL14(16  *  2)
	KERNEL15(16  *  2)
	KERNEL16(16  *  2)
	cmpq	$128 *  3, %rax
	jle,pn	.L12
	KERNEL1 (16  *  3)
	KERNEL2 (16  *  3)
	KERNEL3 (16  *  3)
	KERNEL4 (16  *  3)
	KERNEL5 (16  *  3)
	KERNEL6 (16  *  3)
	KERNEL7 (16  *  3)
	KERNEL8 (16  *  3)
	KERNEL9 (16  *  3)
	KERNEL10(16  *  3)
	KERNEL11(16  *  3)
	KERNEL12(16  *  3)
	KERNEL13(16  *  3)
	KERNEL14(16  *  3)
	KERNEL15(16  *  3)
	KERNEL16(16  *  3)
	cmpq	$128 *  4, %rax
	jle,pn	.L12
	KERNEL1 (16  *  4)
	KERNEL2 (16  *  4)
	KERNEL3 (16  *  4)
	KERNEL4 (16  *  4)
	KERNEL5 (16  *  4)
	KERNEL6 (16  *  4)
	KERNEL7 (16  *  4)
	KERNEL8 (16  *  4)
	KERNEL9 (16  *  4)
	KERNEL10(16  *  4)
	KERNEL11(16  *  4)
	KERNEL12(16  *  4)
	KERNEL13(16  *  4)
	KERNEL14(16  *  4)
	KERNEL15(16  *  4)
	KERNEL16(16  *  4)
	cmpq	$128 *  5, %rax
	jle,pn	.L12
	KERNEL1 (16  *  5)
	KERNEL2 (16  *  5)
	KERNEL3 (16  *  5)
	KERNEL4 (16  *  5)
	KERNEL5 (16  *  5)
	KERNEL6 (16  *  5)
	KERNEL7 (16  *  5)
	KERNEL8 (16  *  5)
	KERNEL9 (16  *  5)
	KERNEL10(16  *  5)
	KERNEL11(16  *  5)