ztrsm_kernel_rt_4x2_sse.s

Optimized GotoBLAS libraries

	subps	%xmm0,  %xmm3
	subps	%xmm1,  %xmm3

	movaps	 4 * SIZE(AO), %xmm8

	pshufd	 $0xee, %xmm8, %xmm9
	pshufd	 $0xbb, %xmm8, %xmm10

	pshufd	 $0xa0, %xmm3, %xmm0
	pshufd	 $0xf5, %xmm3, %xmm3

#ifndef CONJ
	xorps	 %xmm15, %xmm3
#else
	xorps	 %xmm15, %xmm0
#endif

	mulps	%xmm9,  %xmm0
	mulps	%xmm10, %xmm3
	addps	%xmm0,  %xmm3
#endif

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

	pshufd	 $0x44, %xmm8, %xmm9
	pshufd	 $0x11, %xmm8, %xmm10

	pshufd	 $0xa0, %xmm1, %xmm4
	pshufd	 $0xf5, %xmm1, %xmm1

#ifndef CONJ
	xorps	 %xmm15, %xmm1
#else
	xorps	 %xmm15, %xmm4
#endif

	mulps	%xmm9,  %xmm4
	mulps	%xmm10, %xmm1

	addps	%xmm4,  %xmm1
#endif

#ifdef LN
	subq	$4 * SIZE, CO1
#endif

#if defined(LN) || defined(LT)
	movlps	%xmm2,   0 * SIZE(B)
	movlps	%xmm3,   2 * SIZE(B)

	pshufd	$0x00, %xmm2, %xmm0
	pshufd	$0x55, %xmm2, %xmm1

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

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

	movaps	%xmm0,   8 * SIZE(BO)
	movaps	%xmm1,  12 * SIZE(BO)

	movlps	%xmm2,   0 * SIZE(CO1)
	movlps	%xmm3,   2 * SIZE(CO1)
#else
	movaps	%xmm1,   0 * SIZE(AO)

	movlps	%xmm1,   0 * SIZE(CO1)
	movhps	%xmm1,   2 * SIZE(CO1)
#endif

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

#if defined(LT) || defined(RN)
	movq	K,  %rax
	subq	KK, %rax
	salq	$ZBASE_SHIFT, %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 + ZBASE_SHIFT, %rax
	addq	%rax, AORIG
#endif
	ALIGN_4	
	
.L70:
	testq	$1, M
	je	.L79

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

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

	leaq	BUFFER, BO

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

	movlps	 0 * SIZE(AO), %xmm8
	movhps	 2 * SIZE(AO), %xmm8
	movlps	 8 * SIZE(AO), %xmm10
	movhps	10 * SIZE(AO), %xmm10

	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
	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	.L75
	ALIGN_4

.L72:
	mulps	%xmm8, %xmm9
	addps	%xmm9, %xmm0
#if defined(OPTERON) && defined(HAVE_PREFETCH)
	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)
#endif

	movsd	 4 * SIZE(BO), %xmm9
	mulps	%xmm8, %xmm9
	movsd	 2 * SIZE(AO), %xmm8
	addps	%xmm9, %xmm1
	movsd	 8 * SIZE(BO), %xmm9

	mulps	%xmm8, %xmm9
	addps	%xmm9, %xmm2
	movsd	12 * SIZE(BO), %xmm9
	mulps	%xmm8, %xmm9
	movsd	 4 * SIZE(AO), %xmm8
	addps	%xmm9, %xmm3
	movsd	64 * SIZE(BO), %xmm9

	mulps	%xmm8, %xmm11
	addps	%xmm11, %xmm0
	movsd	20 * SIZE(BO), %xmm11
	mulps	%xmm8, %xmm11
	movsd	 6 * SIZE(AO), %xmm8
	addps	%xmm11, %xmm1
	movsd	24 * SIZE(BO), %xmm11

	mulps	%xmm8, %xmm11
	addps	%xmm11, %xmm2
	movsd	28 * SIZE(BO), %xmm11
	mulps	%xmm8, %xmm11
	movsd	16 * SIZE(AO), %xmm8
	addps	%xmm11, %xmm3
	movsd	 80 * SIZE(BO), %xmm11

	mulps	%xmm10, %xmm13
	addps	%xmm13, %xmm0
	movsd	36 * SIZE(BO), %xmm13
	mulps	%xmm10, %xmm13
	movsd	10 * SIZE(AO), %xmm10
	addps	%xmm13, %xmm1
	movsd	40 * SIZE(BO), %xmm13

	mulps	%xmm10, %xmm13
	addps	%xmm13, %xmm2
	movsd	44 * SIZE(BO), %xmm13
	mulps	%xmm10, %xmm13
	movsd	12 * SIZE(AO), %xmm10
	addps	%xmm13, %xmm3
	movsd	 96 * SIZE(BO), %xmm13

	mulps	%xmm10, %xmm15
	addps	%xmm15, %xmm0
	movsd	52 * SIZE(BO), %xmm15
	mulps	%xmm10, %xmm15
	movsd	14 * SIZE(AO), %xmm10
	addps	%xmm15, %xmm1
	movsd	56 * SIZE(BO), %xmm15

	mulps	%xmm10, %xmm15
	addps	%xmm15, %xmm2
	movsd	60 * SIZE(BO), %xmm15
	mulps	%xmm10, %xmm15
	movsd	24 * SIZE(AO), %xmm10
	addps	%xmm15, %xmm3
	movsd	112 * SIZE(BO), %xmm15

	addq   $16 * SIZE, AO
	addq   $64 * 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
	movaps	POSINV,  %xmm15
	andq	$7, %rax		# if (k & 1)
	BRANCH
	je .L78
	ALIGN_4

.L76:
	mulps	%xmm8, %xmm9
	addps	%xmm9, %xmm0
	movsd	 4 * SIZE(BO), %xmm9
	mulps	%xmm8, %xmm9
	movsd	 2 * SIZE(AO), %xmm8
	addps	%xmm9, %xmm1
	movsd	 8 * SIZE(BO), %xmm9

	addq	$2 * SIZE, AO		# aoffset  += 4
	addq	$8 * SIZE, BO		# boffset1 += 8
	decq	%rax
	jg	.L76
	ALIGN_4

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

	shufps	$0xb1, %xmm1, %xmm1

#if defined(LN) || defined(LT)
#ifndef CONJ
	xorps	 %xmm15, %xmm1
#else
	xorps	 %xmm15, %xmm0
#endif
#else
	xorps	 %xmm15, %xmm1
#endif

	addps	%xmm1, %xmm0

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

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

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

#if defined(LN) || defined(LT)
	movlps	 0 * SIZE(B), %xmm2

	subps	%xmm0,  %xmm2
#else
	movlps	 0 * SIZE(AO), %xmm1

	subps	%xmm0,  %xmm1
#endif

#if defined(LN) || defined(LT)
	movaps	 0 * SIZE(AO), %xmm8

	pshufd	 $0x44, %xmm8, %xmm9
	pshufd	 $0x11, %xmm8, %xmm10

	pshufd	 $0xa0, %xmm2, %xmm0
	pshufd	 $0xf5, %xmm2, %xmm2

#ifndef CONJ
	xorps	 %xmm15, %xmm2
#else
	xorps	 %xmm15, %xmm0
#endif

	mulps	%xmm9,  %xmm0
	mulps	%xmm10, %xmm2
	addps	%xmm0,  %xmm2
#endif

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

	pshufd	 $0x44, %xmm8, %xmm9
	pshufd	 $0x11, %xmm8, %xmm10

	pshufd	 $0xa0, %xmm1, %xmm4
	pshufd	 $0xf5, %xmm1, %xmm1

#ifndef CONJ
	xorps	 %xmm15, %xmm1
#else
	xorps	 %xmm15, %xmm4
#endif

	mulps	%xmm9,  %xmm4
	mulps	%xmm10, %xmm1

	addps	%xmm4,  %xmm1
#endif

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

#if defined(LN) || defined(LT)
	movlps	%xmm2,   0 * SIZE(B)

	pshufd	$0x00, %xmm2, %xmm0
	pshufd	$0x55, %xmm2, %xmm1

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

	movlps	%xmm2,   0 * SIZE(CO1)
#else
	movlps	%xmm1,   0 * SIZE(AO)

	movlps	%xmm1,   0 * SIZE(CO1)
#endif

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

#if defined(LT) || defined(RN)
	movq	K,  %rax
	subq	KK, %rax
	salq	$ZBASE_SHIFT, %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	$ZBASE_SHIFT, %rax
	addq	%rax, AORIG
#endif
	ALIGN_4	

.L79:
#ifdef LN
       leaq	(, K, SIZE), %rax
       leaq	(B, %rax, COMPSIZE), B
#endif

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

#ifdef RN
	addq	$1, KK
#endif

#ifdef RT
	subq	$1, KK
#endif
	ALIGN_4

.L40:
	movq	N,  J
	sarq	$1, J		# j = (n >> 2)
	jle	.L999
	ALIGN_4

.L01:
#ifdef LN
	movq	OFFSET, %rax
	addq	M, %rax
	movq	%rax, KK
#endif	

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

#if defined(LN) || defined(RT)
	movq	KK, %rax
	movq	B, BORIG
	salq	$ZBASE_SHIFT, %rax
	leaq	(B,  %rax, 2), 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:
	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

	movaps	%xmm0,  0 * SIZE(BO)
	movaps	%xmm1,  4 * SIZE(BO)
	movaps	%xmm2,  8 * SIZE(BO)
	movaps	%xmm3, 12 * SIZE(BO)

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

	movaps	%xmm4, 16 * SIZE(BO)
	movaps	%xmm5, 20 * SIZE(BO)
	movaps	%xmm6, 24 * SIZE(BO)
	movaps	%xmm7, 28 * SIZE(BO)

	movaps	 8 * SIZE(B), %xmm3
	movaps	12 * SIZE(B), %xmm7

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

	movaps	%xmm0, 32 * SIZE(BO)
	movaps	%xmm1, 36 * SIZE(BO)
	movaps	%xmm2, 40 * SIZE(BO)
	movaps	%xmm3, 44 * SIZE(BO)

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

	movaps	%xmm4, 48 * SIZE(BO)
	movaps	%xmm5, 52 * SIZE(BO)
	movaps	%xmm6, 56 * SIZE(BO)
	movaps	%xmm7, 60 * SIZE(BO)

	addq	$16 * SIZE, B
	addq	$64 * 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:
	movaps	 0 * SIZE(B), %xmm3

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

	movaps	%xmm0,  0 * SIZE(BO)
	movaps	%xmm1,  4 * SIZE(BO)
	movaps	%xmm2,  8 * SIZE(BO)
	movaps	%xmm3, 12 * SIZE(BO)

	addq	$ 4 * SIZE, B
	addq	$16 * 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, 2), %rax
       subq	%rax, C
#endif

	movq	C, CO1			# coffset1 = c
	leaq	(C, LDC, 1), CO2

#ifndef RT
	leaq	(C, LDC, 2), C
#endif

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

.L11:
#ifdef LN
       movq	K, %rax
       salq	$2 + ZBASE_SHIFT, %rax
       subq	%rax, AORIG
#endif

#if defined(LN) || defined(RT)
	movq	KK, %rax
	movq	AORIG, AO
	salq	$ZBASE_SHIFT, %rax
	leaq	(AO, %rax, 4), AO
#endif

	leaq	BUFFER, BO

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

	movaps	 0 * SIZE(BO), %xmm9
	movaps	 4 * SIZE(BO), %xmm11
	movaps	 8 * SIZE(BO), %xmm13
	movaps	16 * SIZE(BO), %xmm15

	movaps	 0 * SIZE(AO), %xmm8
	pxor	%xmm0, %xmm0
	movaps	 4 * SIZE(AO), %xmm10
	pxor	%xmm1, %xmm1
	movaps	 8 * SIZE(AO), %xmm12
	pxor	%xmm2, %xmm2
	movaps	12 * SIZE(AO), %xmm14
	pxor	%xmm3, %xmm3

	PREFETCHW      7 * SIZE(CO1)
	pxor	%xmm4, %xmm4
	PREFETCHW      7 * SIZE(CO2)
	pxor	%xmm5, %xmm5
	pxor	%xmm6, %xmm6
	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(32 *  0)
	KERNEL2(32 *  0)
	KERNEL3(32 *  0)
	KERNEL4(32 *  0)
	KERNEL5(32 *  0)
	KERNEL6(32 *  0)
	KERNEL7(32 *  0)
	KERNEL8(32 *  0)
	KERNEL1(32 *  1)
	KERNEL2(32 *  1)
	KERNEL3(32 *  1)
	KERNEL4(32 *  1)
	KERNEL5(32 *  1)
	KERNEL6(32 *  1)
	KERNEL7(32 *  1)
	KERNEL8(32 *  1)

	addq	$32 * 2  * SIZE, AO
	addq	$64 * 2  * SIZE, BO
	subq	$64 * 2, %rax
	jg	.L1X

.L12:
	leaq	(AO, %rax, 2), AO
	leaq	(BO, %rax, 4), BO
	ALIGN_4

.L15:
#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	movaps	POSINV,  %xmm15
	andq	$7, %rax		# if (k & 1)
	BRANCH
	je .L18
	ALIGN_4

.L16:
	mulps	%xmm8, %xmm9
	addps	%xmm9, %xmm0
	movaps	 4 * SIZE(BO), %xmm9
	mulps	%xmm8, %xmm9
	addps	%xmm9, %xmm1
	movaps	 8 * SIZE(BO), %xmm9
	mulps	%xmm8, %xmm9
	mulps	12 * SIZE(BO), %xmm8
	addps	%xmm9, %xmm2
	movaps	 0 * SIZE(BO), %xmm9
	addps	%xmm8, %xmm3
	movaps	 8 * SIZE(AO), %xmm8
	mulps	%xmm10, %xmm9
	addps	%xmm9, %xmm4
	movaps	 4 * SIZE(BO), %xmm9
	mulps	%xmm10, %xmm9
	addps	%xmm9, %xmm5
	movaps	 8 * SIZE(BO), %xmm9
	mulps	%xmm10, %xmm9
	mulps	12 * SIZE(BO), %xmm10
	addps	%xmm9, %xmm6
	movaps	16 * SIZE(BO), %xmm9
	addps	%xmm10, %xmm7
	movaps	12 * SIZE(AO), %xmm10

	addq	$ 8 * SIZE, AO		# aoffset  += 4
	addq	$16 * SIZE, BO		# boffset1 += 8
	decq	%rax
	jg	.L16
	ALIGN_4

.L18:
	shufps	$0xb1, %xmm1, %xmm1
	shufps	$0xb1, %xmm3, %xmm3
	shufps	$0xb1, %xmm5, %xmm5
	shufps	$0xb1, %xmm7, %xmm7

#if defined(LN) || defined(LT)
#ifndef CONJ
	xorps	 %xmm15, %xmm1
	xorps	 %xmm15, %xmm3
	xorps	 %xmm15, %xmm5
	xorps	 %xmm15, %xmm7
#else
	xorps	 %xmm15, %xmm0
	xorps	 %xmm15, %xmm2
	xorps	 %xmm15, %xmm4
	xorps	 %xmm15, %xmm6
#endif
#else
	xorps	 %xmm15, %xmm1
	xorps	 %xmm15, %xmm3
	xorps	 %xmm15, %xmm5
	xorps	 %xmm15, %xmm7
#endif

	addps	%xmm1, %xmm0
	addps	%xmm3, %xmm2
	addps	%xmm5, %xmm4
	addps	%xmm7, %xmm6

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

	movq	AORIG, AO
	movq	BORIG, B