trsm_kernel_rt_8x4_sse.s

Optimized GotoBLAS libraries

	leaq	(,%rax, SIZE), %rax
	leaq	(AO, %rax, 2), AO
#ifdef LT
	addq	$2 * 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	

.L140:
	testq	$1, M
	je	.L149

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

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

	leaq	BUFFER, BO

#if defined(LN) || defined(RT)
	movq	KK, %rax
	salq	$BASE_SHIFT, %rax
	leaq	(BO, %rax, 4), BO
#endif	

	movss	 0 * SIZE(AO), %xmm8
	movss	 4 * SIZE(AO), %xmm10

	movss	 0 * SIZE(BO), %xmm9
	movss	16 * SIZE(BO), %xmm11

	pxor	%xmm0, %xmm0
	pxor	%xmm1, %xmm1
	pxor	%xmm2, %xmm2
	pxor	%xmm3, %xmm3

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	sarq	$3, %rax
	je	.L145
	ALIGN_4

.L142:
	mulss	%xmm8, %xmm9
#if defined(OPTERON) && defined(HAVE_PREFETCH)
	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)
#endif
	movss	 1 * SIZE(AO), %xmm8
	mulss	 4 * SIZE(BO), %xmm8
	addss	%xmm9, %xmm0
	movss	32 * SIZE(BO), %xmm9
	addss	%xmm8, %xmm1
	movss	 2 * SIZE(AO), %xmm8
	mulss	 8 * SIZE(BO), %xmm8
	addss	%xmm8, %xmm2
	movss	 3 * SIZE(AO), %xmm8
	mulss	12 * SIZE(BO), %xmm8
	addss	%xmm8, %xmm3
	movss	 8 * SIZE(AO), %xmm8
	mulss	%xmm10, %xmm11
	movss	 5 * SIZE(AO), %xmm10
	mulss	20 * SIZE(BO), %xmm10
	addss	%xmm11, %xmm0
	movss	48 * SIZE(BO), %xmm11
	addss	%xmm10, %xmm1
	movss	 6 * SIZE(AO), %xmm10
	mulss	24 * SIZE(BO), %xmm10
	addss	%xmm10, %xmm2
	movss	 7 * SIZE(AO), %xmm10
	mulss	28 * SIZE(BO), %xmm10
	addss	%xmm10, %xmm3
	movss	12 * SIZE(AO), %xmm10

	addq   $ 8 * SIZE, AO
	addq   $32 * SIZE, BO
	decq   %rax
	jne    .L142
	ALIGN_4

.L145:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$7, %rax		# if (k & 1)
	BRANCH
	je .L148
	ALIGN_4

.L146:
	mulss	%xmm8, %xmm9
	movss	 1 * SIZE(AO), %xmm8
	addss	%xmm9, %xmm0
	movss	 4 * SIZE(BO), %xmm9

	addq	$1 * SIZE, AO
	addq	$4 * SIZE, BO
	decq	%rax
	jg	.L146
	ALIGN_4

.L148:
	addss	%xmm1, %xmm0
	addss	%xmm3, %xmm2
	addss	%xmm2, %xmm0

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

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

	salq	$BASE_SHIFT, %rax
	leaq	(AO, %rax, 1), AO
	leaq	(B,  %rax, 1), B
	leaq	(BO, %rax, 4), BO
#endif

#if defined(LN) || defined(LT)
	movss	 0 * SIZE(B), %xmm1
	subss	%xmm0,  %xmm1
#else
	movss	 0 * SIZE(AO), %xmm8
	subps	%xmm0, %xmm8
#endif

#if defined(LN) || defined(LT)
	mulss	  0 * SIZE(AO), %xmm1
#endif

#if defined(RN) || defined(RT)
	mulss	  0 * SIZE(B),  %xmm8
#endif

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

#if defined(LN) || defined(LT)
	movss	%xmm1,   0 * SIZE(B)

	pshufd	$0x00, %xmm1, %xmm2
	movaps	%xmm2,   0 * SIZE(BO)
#else
	movss	%xmm8,   0 * SIZE(AO)
#endif

#if defined(LN) || defined(LT)
	movss	%xmm1,  0 * SIZE(CO1)
#else
	movss	%xmm8,  0 * SIZE(CO1)
#endif

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

#if defined(LT) || defined(RN)
	movq	K,  %rax
	subq	KK, %rax
	leaq	(,%rax, SIZE), %rax
	leaq	(AO, %rax, 1), AO
#ifdef LT
	addq	$1 * 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	$BASE_SHIFT, %rax
	addq	%rax, AORIG
#endif
	ALIGN_4	

.L149:
#ifdef LN
       leaq	(, K, SIZE), %rax
       leaq	(B, %rax, 1), B
#endif

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

#ifdef RN
	addq	$1, KK
#endif

#ifdef RT
	subq	$1, KK
#endif
	ALIGN_4

.L50:
	testq	$2, N
	je	.L100

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

	leaq	BUFFER, BO
	
#ifdef RT
       movq	K, %rax
       salq	$1 + BASE_SHIFT, %rax
       subq	%rax, B
#endif

#if defined(LN) || defined(RT)
	movq	KK, %rax
	movq	B, BORIG
	salq	$1 + BASE_SHIFT, %rax
	leaq	(B,  %rax, 1), B
	leaq	(BO, %rax, 4), 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	.L53
	ALIGN_4
	
.L52:
	movaps	 0 * SIZE(B), %xmm3
	movaps	 4 * SIZE(B), %xmm7

	pshufd	 $0x00, %xmm3, %xmm0
	pshufd	 $0x55, %xmm3, %xmm1
	pshufd	 $0xaa, %xmm3, %xmm2
	pshufd	 $0xff, %xmm3, %xmm3

	pshufd	 $0x00, %xmm7, %xmm4
	pshufd	 $0x55, %xmm7, %xmm5
	pshufd	 $0xaa, %xmm7, %xmm6
	pshufd	 $0xff, %xmm7, %xmm7

	movaps	%xmm0,  0 * SIZE(BO)
	movaps	%xmm1,  4 * SIZE(BO)
	movaps	%xmm2,  8 * SIZE(BO)
	movaps	%xmm3, 12 * SIZE(BO)
	movaps	%xmm4, 16 * SIZE(BO)
	movaps	%xmm5, 20 * SIZE(BO)
	movaps	%xmm6, 24 * SIZE(BO)
	movaps	%xmm7, 28 * SIZE(BO)

	addq	$ 8 * SIZE, B
	addq	$32 * SIZE, BO

	decq	%rax
	jne	.L52
	ALIGN_4

.L53:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$3, %rax
	BRANCH
	jle	.L60
	ALIGN_4

.L54:
	movsd	 0 * SIZE(B), %xmm3

	pshufd	 $0x00, %xmm3, %xmm0
	pshufd	 $0x55, %xmm3, %xmm1

	movaps	%xmm0,  0 * SIZE(BO)
	movaps	%xmm1,  4 * SIZE(BO)

	addq	$2 * SIZE, B
	addq	$8 * SIZE, BO
	decq	%rax
	jne	.L54
	ALIGN_4
	
.L60:
#if defined(LT) || defined(RN)
	movq	A, AO
#else
	movq	A, AORIG
#endif

#ifdef RT
       leaq	(, LDC, 2), %rax
       subq	%rax, C
#endif

	movq	C, CO1			# coffset1 = c
#ifndef RT
	leaq	(C, LDC, 2), C
#endif

	movq	M,  I
	sarq	$3, I	# i = (m >> 3)
	jle	.L70
	ALIGN_4

.L61:
#ifdef LN
       movq	K, %rax
       salq	$3 + 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, 8), AO
#endif

	leaq	BUFFER, BO

#if defined(LN) || defined(RT)
	movq	KK, %rax
	salq	$1 + BASE_SHIFT, %rax
	leaq	(BO, %rax, 4), BO
#endif	

	movaps	 0 * SIZE(AO), %xmm8
	movaps	16 * SIZE(AO), %xmm10
	movaps	32 * SIZE(AO), %xmm12
	movaps	48 * SIZE(AO), %xmm14

	movaps	 0 * SIZE(BO), %xmm9
	movaps	16 * SIZE(BO), %xmm11
	movaps	32 * SIZE(BO), %xmm13
	movaps	48 * SIZE(BO), %xmm15

	pxor	%xmm0, %xmm0
	pxor	%xmm1, %xmm1

	PREFETCHW      4 * SIZE(CO1)
	pxor	%xmm4, %xmm4
	PREFETCHW      4 * SIZE(CO2)
	pxor	%xmm5, %xmm5

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	sarq	$3, %rax
	je	.L65
	ALIGN_4

.L62:
	mulps	%xmm8, %xmm9
#if defined(OPTERON) && defined(HAVE_PREFETCH)
	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)
#endif
	mulps	 4 * SIZE(BO), %xmm8
	addps	%xmm9, %xmm0
	movaps	 0 * SIZE(BO), %xmm9
	addps	%xmm8, %xmm1
	movaps	 4 * SIZE(AO), %xmm8
	mulps	%xmm8, %xmm9
	mulps	 4 * SIZE(BO), %xmm8
	addps	%xmm9, %xmm4
	movaps	 8 * SIZE(BO), %xmm9
	addps	%xmm8, %xmm5
	movaps	 8 * SIZE(AO), %xmm8

	mulps	%xmm8, %xmm9
	mulps	12 * SIZE(BO), %xmm8
	addps	%xmm9, %xmm0
	movaps	 8 * SIZE(BO), %xmm9
	addps	%xmm8, %xmm1
	movaps	12 * SIZE(AO), %xmm8
	mulps	%xmm8, %xmm9
	mulps	12 * SIZE(BO), %xmm8
	addps	%xmm9, %xmm4
	movaps	64 * SIZE(BO), %xmm9
	addps	%xmm8, %xmm5
	movaps	64 * SIZE(AO), %xmm8

#if defined(OPTERON) && defined(HAVE_PREFETCH)
	PREFETCH	(PREFETCHSIZE + 16) * SIZE(AO)
#endif
	mulps	%xmm10, %xmm11
	mulps	20 * SIZE(BO), %xmm10
	addps	%xmm11, %xmm0
	movaps	16 * SIZE(BO), %xmm11
	addps	%xmm10, %xmm1
	movaps	20 * SIZE(AO), %xmm10
	mulps	%xmm10, %xmm11
	mulps	20 * SIZE(BO), %xmm10
	addps	%xmm11, %xmm4
	movaps	24 * SIZE(BO), %xmm11
	addps	%xmm10, %xmm5
	movaps	24 * SIZE(AO), %xmm10

	mulps	%xmm10, %xmm11
	mulps	28 * SIZE(BO), %xmm10
	addps	%xmm11, %xmm0
	movaps	24 * SIZE(BO), %xmm11
	addps	%xmm10, %xmm1
	movaps	28 * SIZE(AO), %xmm10
	mulps	%xmm10, %xmm11
	mulps	28 * SIZE(BO), %xmm10
	addps	%xmm11, %xmm4
	movaps	80 * SIZE(BO), %xmm11
	addps	%xmm10, %xmm5
	movaps	80 * SIZE(AO), %xmm10

#if defined(OPTERON) && defined(HAVE_PREFETCH)
	PREFETCH	(PREFETCHSIZE + 32) * SIZE(AO)
#endif
	mulps	%xmm12, %xmm13
	mulps	36 * SIZE(BO), %xmm12
	addps	%xmm13, %xmm0
	movaps	32 * SIZE(BO), %xmm13
	addps	%xmm12, %xmm1
	movaps	36 * SIZE(AO), %xmm12
	mulps	%xmm12, %xmm13
	mulps	36 * SIZE(BO), %xmm12
	addps	%xmm13, %xmm4
	movaps	40 * SIZE(BO), %xmm13
	addps	%xmm12, %xmm5
	movaps	40 * SIZE(AO), %xmm12

	mulps	%xmm12, %xmm13
	mulps	44 * SIZE(BO), %xmm12
	addps	%xmm13, %xmm0
	movaps	40 * SIZE(BO), %xmm13
	addps	%xmm12, %xmm1
	movaps	44 * SIZE(AO), %xmm12
	mulps	%xmm12, %xmm13
	mulps	44 * SIZE(BO), %xmm12
	addps	%xmm13, %xmm4
	addps	%xmm12, %xmm5
	movaps	96 * SIZE(BO), %xmm13
	movaps	96 * SIZE(AO), %xmm12

#if defined(OPTERON) && defined(HAVE_PREFETCH)
	PREFETCH	(PREFETCHSIZE + 48) * SIZE(AO)
#endif
	mulps	%xmm14, %xmm15
	mulps	52 * SIZE(BO), %xmm14
	addps	%xmm15, %xmm0
	movaps	48 * SIZE(BO), %xmm15
	addps	%xmm14, %xmm1
	movaps	52 * SIZE(AO), %xmm14
	mulps	%xmm14, %xmm15
	mulps	52 * SIZE(BO), %xmm14
	addps	%xmm15, %xmm4
	movaps	56 * SIZE(BO), %xmm15
	addps	%xmm14, %xmm5
	movaps	56 * SIZE(AO), %xmm14

	mulps	%xmm14, %xmm15
	mulps	60 * SIZE(BO), %xmm14
	addps	%xmm15, %xmm0
	movaps	56 * SIZE(BO), %xmm15
	addps	%xmm14, %xmm1
	movaps	60 * SIZE(AO), %xmm14
	mulps	%xmm14, %xmm15
	mulps	60 * SIZE(BO), %xmm14
	addps	%xmm15, %xmm4
	movaps	112 * SIZE(BO), %xmm15
	addps	%xmm14, %xmm5
	movaps	112 * SIZE(AO), %xmm14

	addq   $64 * SIZE, AO
	addq   $64 * SIZE, BO
	decq   %rax
	jne    .L62
	ALIGN_4

.L65:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$7, %rax		# if (k & 1)
	BRANCH
	je .L68
	ALIGN_4

.L66:
	mulps	%xmm8, %xmm9
	mulps	 4 * SIZE(BO), %xmm8
	addps	%xmm9, %xmm0
	movaps	 0 * SIZE(BO), %xmm9
	addps	%xmm8, %xmm1
	movaps	 4 * SIZE(AO), %xmm8
	mulps	%xmm8, %xmm9
	mulps	 4 * SIZE(BO), %xmm8
	addps	%xmm9, %xmm4
	movaps	 8 * SIZE(BO), %xmm9
	addps	%xmm8, %xmm5
	movaps	 8 * SIZE(AO), %xmm8

	addq	$8 * SIZE, AO		# aoffset  += 4
	addq	$8 * SIZE, BO		# boffset1 += 8
	decq	%rax
	jg	.L66
	ALIGN_4

.L68:
#if defined(LN) || defined(RT)
	movq	KK, %rax
#ifdef LN
	subq	$8, %rax
#else
	subq	$2, %rax
#endif

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

	salq	$1 + BASE_SHIFT, %rax
	leaq	(AO, %rax, 4), AO
	leaq	(B,  %rax, 1), B
	leaq	(BO, %rax, 4), BO
#endif

#if defined(LN) || defined(LT)
	movaps	 %xmm0,  %xmm8
	unpcklps %xmm2,  %xmm0
	unpckhps %xmm2,  %xmm8

	movaps	 %xmm1,  %xmm14
	unpcklps %xmm3,  %xmm1
	unpckhps %xmm3,  %xmm14

	movaps	 %xmm0,  %xmm2
	unpcklps %xmm1,  %xmm0
	unpckhps %xmm1,  %xmm2

	movaps	 %xmm8,  %xmm3
	unpcklps %xmm14, %xmm8
	unpckhps %xmm14, %xmm3

	movaps	 %xmm4,  %xmm9
	unpcklps %xmm6,  %xmm4
	unpckhps %xmm6,  %xmm9

	movaps	 %xmm5,  %xmm14
	unpcklps %xmm7,  %xmm5
	unpckhps %xmm7,  %xmm14

	movaps	 %xmm4,  %xmm6
	unpcklps %xmm5,  %xmm4
	unpckhps %xmm5,  %xmm6

	movaps	 %xmm9,  %xmm7
	unpcklps %xmm14, %xmm9
	unpckhps %xmm14, %xmm7

	movsd	 0 * SIZE(B), %xmm1
	movsd	 2 * SIZE(B), %xmm5
	movsd	 4 * SIZE(B), %xmm10
	movsd	 6 * SIZE(B), %xmm11
	movsd	 8 * SIZE(B), %xmm12
	movsd	10 * SIZE(B), %xmm13
	movsd	12 * SIZE(B), %xmm14
	movsd	14 * SIZE(B), %xmm15

	subps	%xmm0,  %xmm1
	subps	%xmm2,  %xmm5
	subps	%xmm8,  %xmm10
	subps	%xmm3,  %xmm11
	subps	%xmm4,  %xmm12
	subps	%xmm6,  %xmm13
	subps	%xmm9,  %xmm14
	subps	%xmm7,  %xmm15
#else
	movaps	 0 * SIZE(AO), %xmm8
	movaps	 4 * SIZE(AO), %xmm9
	movaps	 8 * SIZE(AO), %xmm10
	movaps	12 * SIZE(AO), %xmm11

	subps	%xmm0, %xmm8
	subps	%xmm4, %xmm9
	subps	%xmm1, %xmm10
	subps	%xmm5, %xmm11
#endif

#ifdef LN
	movaps	 60 * SIZE(AO), %xmm6
	pshufd	 $0xff, %xmm6, %xmm8
	mulps	 %xmm8, %xmm15
	pshufd	 $0xaa, %xmm6, %xmm8
	mulps	 %xmm15, %xmm8
	subps	 %xmm8, %xmm14
	pshufd	 $0x55, %xmm6, %xmm8
	mulps	 %xmm15, %xmm8
	subps	 %xmm8, %xmm13
	pshufd	 $0x00, %xmm6, %xmm8
	mulps	 %xmm15, %xmm8
	subps	 %xmm8, %xmm12

	movaps	 56 * SIZE(AO), %xmm6
	pshufd	 $0xff, %xmm6, %xmm8
	mulps	 %xmm15, %xmm8
	subps	 %xmm8, %xmm11
	pshufd	 $0xaa, %xmm6, %xmm8
	mulps	 %xmm15, %xmm8
	subps	 %xmm8, %xmm10
	pshufd	 $0x55, %xmm6, %xmm8
	mulps	 %xmm15, %xmm8
	subps	 %xmm8, %xmm5
	pshufd	 $0x00, %xmm6, %xmm8
	mulps	 %xmm15, %xmm8
	subps	 %xmm8, %xmm1

	movaps	 52 * SIZE(AO), %xmm6
	pshufd	 $0xaa, %xmm6, %xmm8
	mulps	 %xmm8, %xmm14
	pshufd	 $0x55, %xmm6, %xmm8
	mulps	 %xmm14, %xmm8
	subps	 %xmm8, %xmm13
	pshufd	 $0x00, %xmm6, %xmm8
	mulps	 %xmm14, %xmm8
	subps	 %xmm8, %xmm12

	movaps	 48 * SIZE(AO), %xmm6
	pshufd	 $0xff, %xmm6, %xmm8
	mulps	 %xmm14, %xmm8
	subps	 %xmm8, %xmm11
	pshufd	 $0xaa, %xmm6, %xmm8
	mulps	 %xmm14, %xmm8
	subps	 %xmm8, %xmm10
	pshufd	 $0x55, %xmm6, %xmm8
	mulps	 %xmm14, %xmm8
	subps	 %xmm8, %xmm5
	pshufd	 $0x00, %xmm6, %xmm8
	mulps	 %xmm14, %xmm8
	subps	 %xmm8, %xmm1

	movaps	 44 * SIZE(AO), %xmm6
	pshufd	 $0x55, %xmm6, %xmm8
	mulps	 %xmm8, %xmm13
	pshufd	 $0x00, %xmm6, %xmm8
	mulps	 %xmm13, %xmm8
	subps	 %xmm8, %xmm12