trsm_kernel_rt_4x4_core2.s

Optimized GotoBLAS libraries

	movsd	%xmm9,  0 * SIZE(CO1)
	movsd	%xmm13,  1 * SIZE(CO1)
	movsd	%xmm1,  2 * SIZE(CO1)
	movsd	%xmm5, 3 * SIZE(CO1)

	movhpd	%xmm9,  0 * SIZE(CO2)
	movhpd	%xmm13,  1 * SIZE(CO2)
	movhpd	%xmm1,  2 * SIZE(CO2)
	movhpd	%xmm5, 3 * SIZE(CO2)
#else
	movsd	%xmm0,  0 * SIZE(CO1)
	movhpd	%xmm0,  1 * SIZE(CO1)
	movsd	%xmm1,  2 * SIZE(CO1)
	movhpd	%xmm1,  3 * SIZE(CO1)

	movsd	%xmm2,  0 * SIZE(CO2)
	movhpd	%xmm2,  1 * SIZE(CO2)
	movsd	%xmm3,  2 * SIZE(CO2)
	movhpd	%xmm3,  3 * SIZE(CO2)
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm9,  -16 * SIZE(B)
	movapd	%xmm13,  -14 * SIZE(B)
	movapd	%xmm1,  -12 * SIZE(B)
	movapd	%xmm5, -10 * SIZE(B)

	movddup	%xmm9,  %xmm8
	SHUFPD_3 %xmm9,  %xmm9
	movddup	%xmm13,  %xmm12
	SHUFPD_3 %xmm13,  %xmm13
	movddup	%xmm1,  %xmm0
	SHUFPD_3 %xmm1,  %xmm1
	movddup	%xmm5, %xmm4
	SHUFPD_3 %xmm5, %xmm5

	movapd	%xmm8,  -16 * SIZE(BO)
	movapd	%xmm9,  -14 * SIZE(BO)
	movapd	%xmm12,  -12 * SIZE(BO)
	movapd	%xmm13,  -10 * SIZE(BO)
	movapd	%xmm0,   -8 * SIZE(BO)
	movapd	%xmm1,   -6 * SIZE(BO)
	movapd	%xmm4,  -4 * SIZE(BO)
	movapd	%xmm5,  -2 * SIZE(BO)
#else
	movapd	%xmm0,  -16 * SIZE(AO)
	movapd	%xmm1,  -14 * SIZE(AO)
	movapd	%xmm2, -12 * SIZE(AO)
	movapd	%xmm3, -10 * SIZE(AO)
#endif

#ifndef LN
	addq	$4 * SIZE, CO1
	addq	$4 * SIZE, CO2
#endif

#if defined(LT) || defined(RN)
	movq	K,  %rax
	subq	KK, %rax
	leaq	(,%rax, SIZE), %rax
	leaq	(AO, %rax, 4), AO
#ifdef LT
	addq	$8 * SIZE, B
#endif
#endif

#ifdef LN
	subq	$4, KK
	movq	BORIG, B
#endif

#ifdef LT
	addq	$4, KK
#endif

#ifdef RT
       movq	K, %rax
       movq	BORIG, B
       salq	$2 + BASE_SHIFT, %rax
       addq	%rax, AORIG
#endif

	decq	I			# i --
	jg	.L51
	ALIGN_4	

.L60:
	testq	$2, M
	je	.L70
	ALIGN_4

.L61:
#ifdef LN
       movq	K, %rax
       salq	$1 + BASE_SHIFT, %rax
       subq	%rax, AORIG
#endif

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

	leaq	16 * SIZE + BUFFER, BO

#if defined(LN) || defined(RT)
	movq	KK, %rax
	salq	$1 + BASE_SHIFT, %rax
	leaq	(BO, %rax, 2), BO
#endif	

	pxor	%xmm8, %xmm8
	pxor	%xmm9, %xmm9
	pxor	%xmm10, %xmm10
	pxor	%xmm11, %xmm11

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	sarq	$2, %rax
	je	.L65
	ALIGN_4

.L62:
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)

	movapd	-16 * SIZE(AO), %xmm0
	movapd	-14 * SIZE(AO), %xmm1
	movapd	-16 * SIZE(BO), %xmm2
	movapd	-14 * SIZE(BO), %xmm3
	movapd	-12 * SIZE(BO), %xmm4
	movapd	-10 * SIZE(BO), %xmm5

	mulpd	%xmm0, %xmm2
	mulpd	%xmm0, %xmm3
	mulpd	%xmm1, %xmm4
	mulpd	%xmm1, %xmm5

	addpd	%xmm2, %xmm8
	addpd	%xmm3, %xmm9
	addpd	%xmm4, %xmm10
	addpd	%xmm5, %xmm11

	movapd	-12 * SIZE(AO), %xmm0
	movapd	-10 * SIZE(AO), %xmm1
	movapd	 -8 * SIZE(BO), %xmm2
	movapd	 -6 * SIZE(BO), %xmm3
	movapd	 -4 * SIZE(BO), %xmm4
	movapd	 -2 * SIZE(BO), %xmm5

	mulpd	%xmm0, %xmm2
	mulpd	%xmm0, %xmm3
	mulpd	%xmm1, %xmm4
	mulpd	%xmm1, %xmm5

	addpd	%xmm2, %xmm8
	addpd	%xmm3, %xmm9
	addpd	%xmm4, %xmm10
	addpd	%xmm5, %xmm11

	subq	$ -8 * SIZE, AO
	subq	$-16 * SIZE, BO
	subq	$1, %rax
	jne    .L62
	ALIGN_4

.L65:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	BRANCH
	je .L69
	ALIGN_4

.L66:
	movapd	-16 * SIZE(AO), %xmm0
	movapd	-16 * SIZE(BO), %xmm2
	movapd	-14 * SIZE(BO), %xmm3

	mulpd	%xmm0, %xmm2
	mulpd	%xmm0, %xmm3

	addpd	%xmm2, %xmm8
	addpd	%xmm3, %xmm9

	addq	$2 * SIZE, AO
	addq	$4 * SIZE, BO
	subq	$1, %rax
	jg	.L66
	ALIGN_4

.L69:
	addpd	%xmm10, %xmm8
	addpd	%xmm11, %xmm9

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

	movq	AORIG, AO
	movq	BORIG, B
	leaq	16 * SIZE + BUFFER, BO

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

#if defined(LN) || defined(LT)
	movapd	%xmm8, %xmm0
	unpcklpd %xmm9, %xmm8
	unpckhpd %xmm9, %xmm0

	movapd	 -16 * SIZE(B), %xmm9
	movapd	 -14 * SIZE(B), %xmm13

	subpd	%xmm8,  %xmm9
	subpd	%xmm0,  %xmm13
#else
	movapd	-16 * SIZE(AO), %xmm0
	movapd	-14 * SIZE(AO), %xmm2

	subpd	%xmm8, %xmm0
	subpd	%xmm9, %xmm2
#endif


#ifdef LN
	movddup	-13 * SIZE(AO), %xmm8
	mulpd	 %xmm8, %xmm13

	movddup	-14 * SIZE(AO), %xmm10
	mulpd	 %xmm13, %xmm10
	subpd	 %xmm10, %xmm9

	movddup	-16 * SIZE(AO), %xmm8
	mulpd	 %xmm8, %xmm9
#endif

#ifdef LT
	movddup	-16 * SIZE(AO), %xmm8
	mulpd	 %xmm8, %xmm9

	movddup	-15 * SIZE(AO), %xmm10
	mulpd	 %xmm9, %xmm10
	subpd	 %xmm10, %xmm13

	movddup	-13 * SIZE(AO), %xmm8
	mulpd	 %xmm8, %xmm13
#endif

#ifdef RN
	movddup	-16 * SIZE(B), %xmm8
	mulpd	 %xmm8, %xmm0

	movddup	-15 * SIZE(B), %xmm9
	mulpd	 %xmm0, %xmm9
	subpd	 %xmm9, %xmm2

	movddup	-13 * SIZE(B), %xmm8
	mulpd	 %xmm8, %xmm2
#endif

#ifdef RT
	movddup	-13 * SIZE(B), %xmm8
	mulpd	 %xmm8, %xmm2

	movddup	-14 * SIZE(B), %xmm9
	mulpd	 %xmm2, %xmm9
	subpd	 %xmm9, %xmm0

	movddup	-16 * SIZE(B), %xmm8
	mulpd	 %xmm8, %xmm0
#endif

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

#if defined(LN) || defined(LT)
	movsd	%xmm9,  0 * SIZE(CO1)
	movsd	%xmm13,  1 * SIZE(CO1)

	movhpd	%xmm9,  0 * SIZE(CO2)
	movhpd	%xmm13,  1 * SIZE(CO2)
#else
	movsd	%xmm0,  0 * SIZE(CO1)
	movhpd	%xmm0,  1 * SIZE(CO1)

	movsd	%xmm2,  0 * SIZE(CO2)
	movhpd	%xmm2,  1 * SIZE(CO2)
#endif

#if defined(LN) || defined(LT)
	movapd	%xmm9,  -16 * SIZE(B)
	movapd	%xmm13,  -14 * SIZE(B)

	movddup	%xmm9,  %xmm8
	SHUFPD_3 %xmm9,  %xmm9
	movddup	%xmm13,  %xmm12
	SHUFPD_3 %xmm13,  %xmm13

	movapd	%xmm8,  -16 * SIZE(BO)
	movapd	%xmm9,  -14 * SIZE(BO)
	movapd	%xmm12,  -12 * SIZE(BO)
	movapd	%xmm13,  -10 * SIZE(BO)
#else
	movapd	%xmm0,  -16 * SIZE(AO)
	movapd	%xmm2, -14 * 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
#ifdef LT
	addq	$4 * SIZE, B
#endif
#endif

#ifdef LN
	subq	$2, KK
	movq	BORIG, B
#endif

#ifdef LT
	addq	$2, KK
#endif

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

.L70:
	testq	$1, M
	je	.L79
	ALIGN_4

.L71:
#ifdef LN
       movq	K, %rax
       salq	$0 + BASE_SHIFT, %rax
       subq	%rax, AORIG
#endif

#if defined(LN) || defined(RT)
	movq	KK, %rax
	movq	AORIG, AO
	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 1), AO
#endif

	leaq	16 * SIZE + BUFFER, BO

#if defined(LN) || defined(RT)
	movq	KK, %rax
	salq	$1 + BASE_SHIFT, %rax
	leaq	(BO, %rax, 2), BO
#endif	

	pxor	%xmm8, %xmm8
	pxor	%xmm9, %xmm9
	pxor	%xmm10, %xmm10
	pxor	%xmm11, %xmm11

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	sarq	$2, %rax
	je	.L75
	ALIGN_4

.L72:
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)

	movsd	-16 * SIZE(AO), %xmm0
	movsd	-15 * SIZE(AO), %xmm1
	movsd	-16 * SIZE(BO), %xmm2
	movsd	-14 * SIZE(BO), %xmm3
	movsd	-12 * SIZE(BO), %xmm4
	movsd	-10 * SIZE(BO), %xmm5

	mulsd	%xmm0, %xmm2
	mulsd	%xmm0, %xmm3
	mulsd	%xmm1, %xmm4
	mulsd	%xmm1, %xmm5

	addsd	%xmm2, %xmm8
	addsd	%xmm3, %xmm9
	addsd	%xmm4, %xmm10
	addsd	%xmm5, %xmm11

	movsd	-14 * SIZE(AO), %xmm0
	movsd	-13 * SIZE(AO), %xmm1
	movsd	 -8 * SIZE(BO), %xmm2
	movsd	 -6 * SIZE(BO), %xmm3
	movsd	 -4 * SIZE(BO), %xmm4
	movsd	 -2 * SIZE(BO), %xmm5

	mulsd	%xmm0, %xmm2
	mulsd	%xmm0, %xmm3
	mulsd	%xmm1, %xmm4
	mulsd	%xmm1, %xmm5

	addsd	%xmm2, %xmm8
	addsd	%xmm3, %xmm9
	addsd	%xmm4, %xmm10
	addsd	%xmm5, %xmm11

	subq	$ -4 * SIZE, AO
	subq	$-16 * SIZE, BO
	subq	$1, %rax
	jne    .L72
	ALIGN_4

.L75:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	BRANCH
	je .L78
	ALIGN_4

.L76:
	movsd	-16 * SIZE(AO), %xmm0
	movsd	-16 * SIZE(BO), %xmm2
	movsd	-14 * SIZE(BO), %xmm3

	mulsd	%xmm0, %xmm2
	mulsd	%xmm0, %xmm3

	addsd	%xmm2, %xmm8
	addsd	%xmm3, %xmm9

	addq	$1 * SIZE, AO
	addq	$4 * SIZE, BO
	subq	$1, %rax
	jg	.L76
	ALIGN_4

.L78:
	addsd	%xmm10, %xmm8
	addsd	%xmm11, %xmm9

#if defined(LN) || defined(RT)
	movq	KK, %rax
#ifdef LN
	subq	$1, %rax
#else
	subq	$2, %rax
#endif

	movq	AORIG, AO
	movq	BORIG, B
	leaq	16 * SIZE + BUFFER, BO

	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 1), AO
	leaq	(B,  %rax, 2), B
	leaq	(BO, %rax, 4), BO
#endif

#if defined(LN) || defined(LT)
	movsd	-16 * SIZE(B), %xmm12
	movsd	-15 * SIZE(B), %xmm13
#else
	movsd	-16 * SIZE(AO), %xmm12
	movsd	-15 * SIZE(AO), %xmm13
#endif

	subsd	%xmm8,  %xmm12
	subsd	%xmm9,  %xmm13

#ifdef LN
	movsd	-16 * SIZE(AO), %xmm8

	mulsd	 %xmm8, %xmm12
	mulsd	 %xmm8, %xmm13
#endif

#ifdef LT
	movsd	-16 * SIZE(AO), %xmm8

	mulsd	 %xmm8, %xmm12
	mulsd	 %xmm8, %xmm13
#endif

#ifdef RN
	mulsd	-16 * SIZE(B), %xmm12
	movsd	-15 * SIZE(B), %xmm9
	mulsd	 %xmm12, %xmm9
	subsd	 %xmm9, %xmm13

	mulsd	-13 * SIZE(B), %xmm13
#endif

#ifdef RT
	mulsd	-13 * SIZE(B), %xmm13

	movlpd	-14 * SIZE(B), %xmm9
	mulsd	 %xmm13, %xmm9
	subsd	 %xmm9, %xmm12

	mulsd	-16 * SIZE(B), %xmm12
#endif

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

	movsd	%xmm12,  0 * SIZE(CO1)
	movsd	%xmm13,  0 * SIZE(CO2)

#if defined(LN) || defined(LT)
	movsd	%xmm12,  -16 * SIZE(B)
	movsd	%xmm13,  -15 * SIZE(B)

	movsd	%xmm12,  -16 * SIZE(BO)
	movsd	%xmm12,  -15 * SIZE(BO)
	movsd	%xmm13,  -14 * SIZE(BO)
	movsd	%xmm13,  -13 * SIZE(BO)
#else
	movsd	%xmm12,  -16 * SIZE(AO)
	movsd	%xmm13,  -15 * 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
#ifdef LT
	addq	$2 * SIZE, B
#endif
#endif

#ifdef LN
	subq	$1, KK
	movq	BORIG, B
#endif

#ifdef LT
	addq	$1, KK
#endif

#ifdef RT
	movq	K, %rax
	movq	BORIG, B
	salq	$0 + BASE_SHIFT, %rax
	addq	%rax, AORIG
#endif
	ALIGN_4	
	
.L79:
#ifdef LN
       leaq	(, K, SIZE), %rax
       leaq	(B, %rax, 2), B
#endif

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

#ifdef RN
	addq	$2, KK
#endif

#ifdef RT
	subq	$2, KK
#endif
	ALIGN_4

.L80:
	movq	N,  J
	sarq	$2, J		# j = (n >> 2)
	jle	.L999

.L01:
/* Copying to Sub Buffer */

#ifdef LN
	movq	OFFSET, %rax
	addq	M, %rax
	movq	%rax, KK
#endif	

	leaq	16 * SIZE + BUFFER, BO

#ifdef RT
       movq	K, %rax
       salq	$2 + BASE_SHIFT, %rax
       subq	%rax, B
#endif

#if defined(LN) || defined(RT)
	movq	KK, %rax
	movq	B, BORIG
	leaq	(, %rax, SIZE), %rax
	leaq	(B,  %rax, 4), B
	leaq	(BO, %rax, 8), BO
#endif	

#if defined(LT)
	movq	OFFSET, %rax
	movq	%rax, KK
#endif

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	sarq	$2, %rax
	jle	.L03
	ALIGN_4
	
.L02:
	prefetcht0	(PREFETCH_R + 0) * SIZE(B)
	movapd	 -16 * SIZE(B), %xmm0
	movapd	 -14 * SIZE(B), %xmm1
	movapd	 -12 * SIZE(B), %xmm2
	movapd	 -10 * SIZE(B), %xmm3
	movapd	  -8 * SIZE(B), %xmm4
	movapd	  -6 * SIZE(B), %xmm5
	movapd	  -4 * SIZE(B), %xmm6
	movapd	  -2 * SIZE(B), %xmm7

	prefetcht0	(PREFETCH_R + 8) * SIZE(B)
	movddup	  %xmm0, %xmm8
	unpckhpd  %xmm0, %xmm0
	movddup	  %xmm1, %xmm9
	unpckhpd  %xmm1, %xmm1
	movddup	  %xmm2, %xmm10
	unpckhpd  %xmm2, %xmm2
	movddup	  %xmm3, %xmm11
	unpckhpd  %xmm3, %xmm3
	movddup	  %xmm4, %xmm12
	unpckhpd  %xmm4, %xmm4
	movddup	  %xmm5, %xmm13
	unpckhpd  %xmm5, %xmm5
	movddup	  %xmm6, %xmm14
	unpckhpd  %xmm6, %xmm6
	movddup	  %xmm7, %xmm15
	unpckhpd  %xmm7, %xmm7
	
	prefetcht0	(PREFETCH_W +  0) * SIZE(BO)
	movapd	%xmm8,  -16 * SIZE(BO)
	movapd	%xmm0,  -14 * SIZE(BO)
	movapd	%xmm9,  -12 * SIZE(BO)
	movapd	%xmm1,  -10 * SIZE(BO)

	prefetcht0	(PREFETCH_W +  8) * SIZE(BO)
	movapd	%xmm10,  -8 * SIZE(BO)
	movapd	%xmm2,   -6 * SIZE(BO)
	movapd	%xmm11,  -4 * SIZE(BO)
	movapd	%xmm3,   -2 * SIZE(BO)

	prefetcht0	(PREFETCH_W + 16) * SIZE(BO)
	movapd	%xmm12,   0 * SIZE(BO)
	movapd	%xmm4,    2 * SIZE(BO)
	movapd	%xmm13,   4 * SIZE(BO)
	movapd	%xmm5,    6 * SIZE(BO)

	prefetcht0	(PREFETCH_W + 24) * SIZE(BO)
	movapd	%xmm14,   8 * SIZE(BO)
	movapd	%xmm6,   10 * SIZE(BO)
	movapd	%xmm15,  12 * SIZE(BO)
	movapd	%xmm7,   14 * SIZE(BO)

	subq	$-16  * SIZE, B
	subq	$-32 * SIZE, BO
	subq	$1, %rax
	jne	.L02
	ALIGN_4

.L03:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$3, %rax
	BRANCH
	jle	.L10
	ALIGN_4

.L04:
	movapd	 -16 * SIZE(B), %xmm0
	movapd	 -14 * SIZE(B), %xmm1

	movddup	  %xmm0, %xmm8
	unpckhpd  %xmm0, %xmm0
	movddup	  %xmm1, %xmm9
	unpckhpd  %xmm1, %xmm1

	movapd	%xmm8,  -16 * SIZE(BO)
	movapd	%xmm0,  -14 * SIZE(BO)
	movapd	%xmm9,  -12 * SIZE(BO)
	movapd	%xmm1,  -10 * SIZE(BO)