trsm_kernel_rt_4x4_sse2.s

Optimized GotoBLAS libraries

	movapd	%xmm1,   0 * SIZE(B)
	movapd	%xmm5,   2 * SIZE(B)

	movlpd	%xmm1,   0 * SIZE(BO)
	movlpd	%xmm1,   1 * SIZE(BO)
	movhpd	%xmm1,   2 * SIZE(BO)
	movhpd	%xmm1,   3 * SIZE(BO)
	movlpd	%xmm5,   4 * SIZE(BO)
	movlpd	%xmm5,   5 * SIZE(BO)
	movhpd	%xmm5,   6 * SIZE(BO)
	movhpd	%xmm5,   7 * SIZE(BO)
#else
	movapd	%xmm8,   0 * SIZE(AO)
	movapd	%xmm10,  2 * 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	BUFFER, BO

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

	movsd	 0 * SIZE(AO), %xmm8
	pxor	%xmm0, %xmm0
	movsd	 0 * SIZE(BO), %xmm9
	pxor	%xmm1, %xmm1
	movsd	 4 * SIZE(AO), %xmm10
	pxor	%xmm2, %xmm2
	movsd	 8 * SIZE(BO), %xmm11
	pxor	%xmm3, %xmm3

	movsd	16 * SIZE(BO), %xmm13
	movsd	24 * SIZE(BO), %xmm15

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	sarq	$3, %rax
	je	.L75
	ALIGN_4

.L72:
	mulsd	%xmm8, %xmm9
	PREFETCH	(PREFETCHSIZE     +  0) * SIZE(AO)
	mulsd	 2 * SIZE(BO), %xmm8
	addsd	%xmm9, %xmm0
	movsd	 4 * SIZE(BO), %xmm9
	addsd	%xmm8, %xmm1
	movsd	 1 * SIZE(AO), %xmm8

	mulsd	%xmm8, %xmm9
	mulsd	 6 * SIZE(BO), %xmm8
	addsd	%xmm9, %xmm2
	movsd	32 * SIZE(BO), %xmm9
	addsd	%xmm8, %xmm3
	movsd	 2 * SIZE(AO), %xmm8

	mulsd	%xmm8, %xmm11
	mulsd	10 * SIZE(BO), %xmm8
	addsd	%xmm11, %xmm0
	movsd	12 * SIZE(BO), %xmm11
	addsd	%xmm8, %xmm1
	movsd	 3 * SIZE(AO), %xmm8

	mulsd	%xmm8, %xmm11
	mulsd	14 * SIZE(BO), %xmm8
	addsd	%xmm11, %xmm2
	movsd	40 * SIZE(BO), %xmm11
	addsd	%xmm8, %xmm3
	movsd	 8 * SIZE(AO), %xmm8

	mulsd	%xmm10, %xmm13
	mulsd	18 * SIZE(BO), %xmm10
	addsd	%xmm13, %xmm0
	movsd	20 * SIZE(BO), %xmm13
	addsd	%xmm10, %xmm1
	movsd	 5 * SIZE(AO), %xmm10

	mulsd	%xmm10, %xmm13
	mulsd	22 * SIZE(BO), %xmm10
	addsd	%xmm13, %xmm2
	movsd	48 * SIZE(BO), %xmm13
	addsd	%xmm10, %xmm3
	movsd	 6 * SIZE(AO), %xmm10

	mulsd	%xmm10, %xmm15
	mulsd	26 * SIZE(BO), %xmm10
	addsd	%xmm15, %xmm0
	movsd	28 * SIZE(BO), %xmm15
	addsd	%xmm10, %xmm1
	movsd	 7 * SIZE(AO), %xmm10

	mulsd	%xmm10, %xmm15
	mulsd	30 * SIZE(BO), %xmm10
	addsd	%xmm15, %xmm2
	movsd	56 * SIZE(BO), %xmm15
	addsd	%xmm10, %xmm3
	movsd	12 * SIZE(AO), %xmm10

	addq   $ 8 * SIZE, AO
	addq   $32 * SIZE, BO
	decq   %rax
	jne    .L72
	ALIGN_4

.L75:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$7, %rax		# if (k & 1)
	BRANCH
	je .L78
	ALIGN_4

.L76:
	mulsd	%xmm8, %xmm9
	mulsd	 2 * SIZE(BO), %xmm8
	addsd	%xmm9, %xmm0
	addsd	%xmm8, %xmm1
	movsd	 1 * SIZE(AO), %xmm8
	movsd	 4 * SIZE(BO), %xmm9

	addq	$1 * SIZE, AO		# aoffset  += 4
	addq	$4 * SIZE, BO		# boffset1 += 8
	decq	%rax
	jg	.L76
	ALIGN_4

.L78:
	addsd	%xmm2, %xmm0
	addsd	%xmm3, %xmm1

#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	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	 0 * SIZE(B), %xmm4
	movsd	 1 * SIZE(B), %xmm5
#else
	movsd	 0 * SIZE(AO), %xmm4
	movsd	 1 * SIZE(AO), %xmm5
#endif

	subsd	%xmm0,  %xmm4
	subsd	%xmm1,  %xmm5

#ifdef LN
	movsd	 0 * SIZE(AO), %xmm0

	mulsd	 %xmm0, %xmm4
	mulsd	 %xmm0, %xmm5
#endif

#ifdef LT
	movsd	 0 * SIZE(AO), %xmm0

	mulsd	 %xmm0, %xmm4
	mulsd	 %xmm0, %xmm5
#endif

#ifdef RN
	mulsd	 0 * SIZE(B), %xmm4
	movsd	 1 * SIZE(B), %xmm1
	mulsd	 %xmm4, %xmm1
	subsd	 %xmm1, %xmm5

	mulsd	 3 * SIZE(B), %xmm5
#endif

#ifdef RT
	mulsd	 3 * SIZE(B), %xmm5

	movlpd	 2 * SIZE(B), %xmm1
	mulsd	 %xmm5, %xmm1
	subsd	 %xmm1, %xmm4

	mulsd	 0 * SIZE(B), %xmm4
#endif

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

	movsd	%xmm4,  0 * SIZE(CO1)
	movsd	%xmm5,  0 * SIZE(CO2)

#if defined(LN) || defined(LT)
	movsd	%xmm4,   0 * SIZE(B)
	movsd	%xmm5,   1 * SIZE(B)

	movsd	%xmm4,   0 * SIZE(BO)
	movsd	%xmm4,   1 * SIZE(BO)
	movsd	%xmm5,   2 * SIZE(BO)
	movsd	%xmm5,   3 * SIZE(BO)
#else
	movsd	%xmm4,   0 * SIZE(AO)
	movsd	%xmm5,   1 * 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	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	

#ifdef 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

	addq	%rax, %rax
	ALIGN_4
	
.L02:
	PREFETCHNTA	 40 * SIZE(B)

	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

	addq	$16 * SIZE, BO
	addq	$ 8 * SIZE, B

	movsd	%xmm0, -16 * SIZE(BO)
	movsd	%xmm0, -15 * SIZE(BO)
	movsd	%xmm1, -14 * SIZE(BO)
	movsd	%xmm1, -13 * SIZE(BO)
	movsd	%xmm2, -12 * SIZE(BO)
	movsd	%xmm2, -11 * SIZE(BO)
	movsd	%xmm3, -10 * SIZE(BO)
	movsd	%xmm3,  -9 * SIZE(BO)
	movsd	%xmm4,  -8 * SIZE(BO)
	movsd	%xmm4,  -7 * SIZE(BO)
	movsd	%xmm5,  -6 * SIZE(BO)
	movsd	%xmm5,  -5 * SIZE(BO)
	movsd	%xmm6,  -4 * SIZE(BO)
	movsd	%xmm6,  -3 * SIZE(BO)
	movsd	%xmm7,  -2 * SIZE(BO)
	movsd	%xmm7,  -1 * SIZE(BO)

	decq	%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:
	movsd	 0 * SIZE(B), %xmm0
	movsd	 1 * SIZE(B), %xmm1
	movsd	 2 * SIZE(B), %xmm2
	movsd	 3 * SIZE(B), %xmm3

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

	addq	$4 * SIZE, B
	addq	$8 * SIZE, BO
	decq	%rax
	jne	.L04
	ALIGN_4
	
.L10:
#if defined(LT) || defined(RN)
	movq	A, AO
#else
	movq	A, AORIG
#endif

#ifdef RT
       leaq	(, LDC, 4), %rax
       subq	%rax, C
#endif

	movq	C, CO1			# coffset1 = c
	leaq	(C, LDC, 1), CO2	# coffset2 = c + ldc
#ifndef RT
	leaq	(C, LDC, 4), C
#endif

	movq	M,  I
	sarq	$2, I	# i = (m >> 2)
	jle	.L20
	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
	movq	AORIG, AO
	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 4), AO
#endif

	leaq	BUFFER, BO

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

	movapd	 0 * SIZE(BO), %xmm9
	movapd	 2 * SIZE(BO), %xmm11
	movapd	 4 * SIZE(BO), %xmm13
	movapd	 8 * SIZE(BO), %xmm15

	movapd	 0 * SIZE(AO), %xmm8
	pxor	%xmm0, %xmm0
	movapd	 2 * SIZE(AO), %xmm10
	pxor	%xmm1, %xmm1
	movapd	 4 * SIZE(AO), %xmm12
	pxor	%xmm2, %xmm2
	movapd	 6 * SIZE(AO), %xmm14
	pxor	%xmm3, %xmm3

	PREFETCHW      4 * SIZE(CO1)
	pxor	%xmm4, %xmm4
	PREFETCHW      4 * SIZE(CO2)
	pxor	%xmm5, %xmm5
	PREFETCHW      4 * SIZE(CO1, LDC, 2)
	pxor	%xmm6, %xmm6
	PREFETCHW      4 * SIZE(CO2, LDC, 2)
	pxor	%xmm7, %xmm7

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	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)
	KERNEL1(16 *  1)
	KERNEL2(16 *  1)
	KERNEL3(16 *  1)
	KERNEL4(16 *  1)
	KERNEL5(16 *  1)
	KERNEL6(16 *  1)
	KERNEL7(16 *  1)
	KERNEL8(16 *  1)
	cmpq	$64 *  2, %rax
	jle	.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)
	KERNEL1(16 *  3)
	KERNEL2(16 *  3)
	KERNEL3(16 *  3)
	KERNEL4(16 *  3)
	KERNEL5(16 *  3)
	KERNEL6(16 *  3)
	KERNEL7(16 *  3)
	KERNEL8(16 *  3)
	cmpq	$64 *  4, %rax
	jle	.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)
	KERNEL1(16 *  5)
	KERNEL2(16 *  5)
	KERNEL3(16 *  5)
	KERNEL4(16 *  5)
	KERNEL5(16 *  5)
	KERNEL6(16 *  5)
	KERNEL7(16 *  5)
	KERNEL8(16 *  5)
	cmpq	$64 *  6, %rax
	jle	.L12
	KERNEL1(16 *  6)
	KERNEL2(16 *  6)
	KERNEL3(16 *  6)
	KERNEL4(16 *  6)
	KERNEL5(16 *  6)
	KERNEL6(16 *  6)
	KERNEL7(16 *  6)
	KERNEL8(16 *  6)
	KERNEL1(16 *  7)
	KERNEL2(16 *  7)
	KERNEL3(16 *  7)
	KERNEL4(16 *  7)
	KERNEL5(16 *  7)
	KERNEL6(16 *  7)
	KERNEL7(16 *  7)
	KERNEL8(16 *  7)

	addq	$16 * 8  * SIZE, AO
	addq	$32 * 8  * SIZE, BO
	subq	$64 * 8, %rax
	jg	.L1X

.L12:
	leaq	(AO, %rax, 2), AO	# * 16
	leaq	(BO, %rax, 4), BO	# * 64
	ALIGN_4

.L15:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$7, %rax		# if (k & 1)
	BRANCH
	je .L19
	ALIGN_4

.L16:
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm0
	movapd	 2 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm1
	movapd	 4 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	mulpd	 6 * SIZE(BO), %xmm8
	addpd	%xmm9, %xmm2
	movapd	 0 * SIZE(BO), %xmm9
	addpd	%xmm8, %xmm3
	movapd	 4 * SIZE(AO), %xmm8
	mulpd	%xmm10, %xmm9
	addpd	%xmm9, %xmm4
	movapd	 2 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	addpd	%xmm9, %xmm5
	movapd	 4 * SIZE(BO), %xmm9
	mulpd	%xmm10, %xmm9
	mulpd	 6 * SIZE(BO), %xmm10
	addpd	%xmm9, %xmm6
	movapd	 8 * SIZE(BO), %xmm9
	addpd	%xmm10, %xmm7
	movapd	 6 * SIZE(AO), %xmm10

	addq	$4 * SIZE, AO		# aoffset  += 4
	addq	$8 * SIZE, BO		# boffset1 += 8
	decq	%rax
	jg	.L16
	ALIGN_4

.L19:
#if defined(LN) || defined(RT)
	movq	KK, %rax
#ifdef LN
	subq	$4, %rax
#else
	subq	$4, %rax
#endif

	movq	AORIG, AO
	movq	BORIG, B
	leaq	BUFFER, BO

	leaq	(, %rax, SIZE), %rax
	leaq	(AO, %rax, 4), AO
	leaq	(B,  %rax, 4), B
	leaq	(BO, %rax, 8), 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	%xmm4, %xmm12
	unpcklpd %xmm5, %xmm4
	unpckhpd %xmm5, %xmm12

	movapd	%xmm6, %xmm14
	unpcklpd %xmm7, %xmm6
	unpckhpd %xmm7, %xmm14

	movapd	 0 * SIZE(B), %xmm1
	movapd	 2 * SIZE(B), %xmm3
	movapd	 4 * SIZE(B), %xmm5
	movapd	 6 * SIZE(B), %xmm7
	movapd	 8 * SIZE(B), %xmm9
	movapd	10 * SIZE(B), %xmm11
	movapd	12 * SIZE(B), %xmm13
	movapd	14 * SIZE(B), %xmm15