ztrsm_kernel_rt_2x2_sse2.s

Optimized GotoBLAS libraries

	addpd	 %xmm0, %xmm1
	addpd	 %xmm4, %xmm5
#endif

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

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

#if defined(LN) || defined(LT)
	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	%xmm1,   0 * SIZE(AO)
	movapd	%xmm5,   2 * SIZE(AO)
#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

	decq	I			# i --
	jg	.L110
	ALIGN_4	

.L130:
	testq	$1, M
	jle	.L199
	ALIGN_4

.L140:
#ifdef LN
       movq	K, %rax
       salq	$0 + 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	$0 + ZBASE_SHIFT, %rax
	leaq	(BO, %rax, 2), BO
#endif	

	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	$2, %rax
	je	.L142

.L141:
	movapd	 0 * SIZE(AO), %xmm8
	movapd	 0 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm0
	mulpd	 2 * SIZE(BO), %xmm8
	addpd	%xmm8, %xmm1

	movapd	 2 * SIZE(AO), %xmm8
	movapd	 4 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm2
	mulpd	 6 * SIZE(BO), %xmm8
	addpd	%xmm8, %xmm3

	movapd	 4 * SIZE(AO), %xmm8
	movapd	 8 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm0
	mulpd	10 * SIZE(BO), %xmm8
	addpd	%xmm8, %xmm1

	movapd	 6 * SIZE(AO), %xmm8
	movapd	12 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm2
	mulpd	14 * SIZE(BO), %xmm8
	addpd	%xmm8, %xmm3

	addq   $8  * SIZE, AO
	addq   $16 * SIZE, BO
	decq   %rax
	jne    .L141

.L142:
	addpd	%xmm2, %xmm0
	addpd	%xmm3, %xmm1

	movapd	POSINV, %xmm15

#if defined(LT) || defined(RN)
	movq	KK, %rax
#else
	movq	K, %rax
	subq	KK, %rax
#endif
	andq	$3, %rax		# if (k & 1)
	BRANCH
	jle .L144

.L143:
	movapd	 0 * SIZE(AO), %xmm8
	movapd	 0 * SIZE(BO), %xmm9
	mulpd	%xmm8, %xmm9
	addpd	%xmm9, %xmm0
	mulpd	 2 * SIZE(BO), %xmm8
	addpd	%xmm8, %xmm1

	addq	$2 * SIZE, AO		# aoffset  += 4
	addq	$4 * SIZE, BO		# boffset1 += 8
	decq	%rax
	jg	.L143
	ALIGN_4

.L144:
#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, 2), BO
#endif

	SHUFPD_1 %xmm1, %xmm1

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)
	xorpd	%xmm15, %xmm1
#else
	xorpd	%xmm15, %xmm0
#endif

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(RR) || defined(RC) || defined(CR) || defined(CC)
	subpd	%xmm1, %xmm0
#else
	addpd	%xmm1, %xmm0
#endif


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

	subpd	%xmm0,  %xmm1
#else
	movapd	 0 * SIZE(AO), %xmm1

	subpd	%xmm0,  %xmm1
#endif

#ifndef CONJ
	SHUFPD_1 %xmm15, %xmm15
#endif

#ifdef LN
	movlpd	 0 * SIZE(AO), %xmm8
	movhpd	 0 * SIZE(AO), %xmm8
	movlpd	 1 * SIZE(AO), %xmm9
	movhpd	 1 * SIZE(AO), %xmm9

	pshufd	$0x4e, %xmm1, %xmm0
	xorpd	 %xmm15, %xmm0

	mulpd	 %xmm8, %xmm1
	mulpd	 %xmm9, %xmm0

	addpd	 %xmm0, %xmm1
#endif

#ifdef LT
	movlpd	 0 * SIZE(AO), %xmm8
	movhpd	 0 * SIZE(AO), %xmm8
	movlpd	 1 * SIZE(AO), %xmm9
	movhpd	 1 * SIZE(AO), %xmm9

	pshufd	$0x4e, %xmm1, %xmm0

	xorpd	 %xmm15, %xmm0

	mulpd	 %xmm8, %xmm1
	mulpd	 %xmm9, %xmm0

	addpd	 %xmm0, %xmm1
#endif

#ifdef RN
	movlpd	 0 * SIZE(B), %xmm8
	movhpd	 0 * SIZE(B), %xmm8
	movlpd	 1 * SIZE(B), %xmm9
	movhpd	 1 * SIZE(B), %xmm9

	pshufd	$0x4e, %xmm1, %xmm0

	xorpd	 %xmm15, %xmm0

	mulpd	 %xmm8, %xmm1
	mulpd	 %xmm9, %xmm0

	addpd	 %xmm0, %xmm1
#endif

#ifdef RT
	movlpd	 0 * SIZE(B), %xmm8
	movhpd	 0 * SIZE(B), %xmm8
	movlpd	 1 * SIZE(B), %xmm9
	movhpd	 1 * SIZE(B), %xmm9

	pshufd	$0x4e, %xmm1, %xmm0

	xorpd	 %xmm15, %xmm0

	mulpd	 %xmm8, %xmm1
	mulpd	 %xmm9, %xmm0

	addpd	 %xmm0, %xmm1
#endif

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

	movsd	%xmm1,   0 * SIZE(CO1)
	movhpd	%xmm1,   1 * SIZE(CO1)

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

	movlpd	%xmm1,   0 * SIZE(BO)
	movlpd	%xmm1,   1 * SIZE(BO)
	movhpd	%xmm1,   2 * SIZE(BO)
	movhpd	%xmm1,   3 * SIZE(BO)
#else
	movapd	%xmm1,   0 * SIZE(AO)
#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	$0 + ZBASE_SHIFT, %rax
	addq	%rax, AORIG
#endif
	ALIGN_4
	
.L199:
#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, 1 * COMPSIZE), B
#endif

#ifdef RN
	addq	$1, KK
#endif

#ifdef RT
	subq	$1, KK
#endif
	ALIGN_4

.L100:	
	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	

/* Copying to Sub Buffer */
	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, 4), 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

	addq	%rax, %rax
	ALIGN_4
	
.L02:
	PREFETCHNTA	 56 * SIZE(B)

	movlpd	 0 * SIZE(B), %xmm0
	movlpd	 1 * SIZE(B), %xmm1
	movlpd	 2 * SIZE(B), %xmm2
	movlpd	 3 * SIZE(B), %xmm3
	movlpd	 4 * SIZE(B), %xmm4
	movlpd	 5 * SIZE(B), %xmm5
	movlpd	 6 * SIZE(B), %xmm6
	movlpd	 7 * SIZE(B), %xmm7

	movlpd	%xmm0,  0 * SIZE(BO)
	movlpd	%xmm0,  1 * SIZE(BO)
	movlpd	%xmm1,  2 * SIZE(BO)
	movlpd	%xmm1,  3 * SIZE(BO)
	movlpd	%xmm2,  4 * SIZE(BO)
	movlpd	%xmm2,  5 * SIZE(BO)
	movlpd	%xmm3,  6 * SIZE(BO)
	movlpd	%xmm3,  7 * SIZE(BO)
	movlpd	%xmm4,  8 * SIZE(BO)
	movlpd	%xmm4,  9 * SIZE(BO)
	movlpd	%xmm5, 10 * SIZE(BO)
	movlpd	%xmm5, 11 * SIZE(BO)
	movlpd	%xmm6, 12 * SIZE(BO)
	movlpd	%xmm6, 13 * SIZE(BO)
	movlpd	%xmm7, 14 * SIZE(BO)
	movlpd	%xmm7, 15 * SIZE(BO)

	subq	$-16 * SIZE, BO
	addq	$  8 * SIZE, B
	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	.L05
	ALIGN_4

.L04:
	movlpd	 0 * SIZE(B), %xmm0
	movlpd	 1 * SIZE(B), %xmm1
	movlpd	 2 * SIZE(B), %xmm2
	movlpd	 3 * SIZE(B), %xmm3

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

	addq	$ 4 * SIZE, B
	addq	$ 8 * SIZE, BO

	decq	%rax
	jne	.L04
	ALIGN_4
	
.L05:
#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
	leaq	(C, LDC, 1), CO2

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

	movq	M,  I
	sarq	$1, I		# i = (m >> 2)
	jle	.L30
	ALIGN_4

.L10:
#ifdef LN
       movq	K, %rax
       salq	$1 + 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, 2), AO
#endif

	leaq	BUFFER, BO

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

	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

	movapd	 0 * SIZE(BO), %xmm9
	pxor	%xmm4, %xmm4
	movapd	 2 * SIZE(BO), %xmm11
	pxor	%xmm5, %xmm5
	movapd	 4 * SIZE(BO), %xmm13
	movapd	 8 * SIZE(BO), %xmm15

	PREFETCHW      4 * SIZE(CO1)
	pxor	%xmm6, %xmm6
	PREFETCHW      4 * SIZE(CO2)
	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
	movapd	POSINV,  %xmm15
	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	$2, %rax
#else
	subq	$2, %rax
#endif

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

	salq	$ZBASE_SHIFT, %rax
	leaq	(AO, %rax, 2), AO
	leaq	(B,  %rax, 2), B
	leaq	(BO, %rax, 4), BO
#endif

	SHUFPD_1 %xmm1, %xmm1
	SHUFPD_1 %xmm3, %xmm3
	SHUFPD_1 %xmm5, %xmm5
	SHUFPD_1 %xmm7, %xmm7

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)
	xorpd	%xmm15, %xmm1
	xorpd	%xmm15, %xmm3
	xorpd	%xmm15, %xmm5
	xorpd	%xmm15, %xmm7
#else
	xorpd	%xmm15, %xmm0
	xorpd	%xmm15, %xmm2
	xorpd	%xmm15, %xmm4
	xorpd	%xmm15, %xmm6
#endif

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(RR) || defined(RC) || defined(CR) || defined(CC)
	subpd	%xmm1, %xmm0
	subpd	%xmm3, %xmm2
	subpd	%xmm5, %xmm4
	subpd	%xmm7, %xmm6
#else
	addpd	%xmm1, %xmm0
	addpd	%xmm3, %xmm2
	addpd	%xmm5, %xmm4
	addpd	%xmm7, %xmm6
#endif

#if defined(LN) || defined(LT)
	movapd	 0 * SIZE(B), %xmm1
	movapd	 2 * SIZE(B), %xmm3
	movapd	 4 * SIZE(B), %xmm5
	movapd	 6 * SIZE(B), %xmm7

	subpd	%xmm0,  %xmm1
	subpd	%xmm2,  %xmm3
	subpd	%xmm4,  %xmm5
	subpd	%xmm6,  %xmm7
#else
	movapd	 0 * SIZE(AO), %xmm1
	movapd	 2 * SIZE(AO), %xmm5
	movapd	 4 * SIZE(AO), %xmm3
	movapd	 6 * SIZE(AO), %xmm7

	subpd	%xmm0,  %xmm1
	subpd	%xmm2,  %xmm3
	subpd	%xmm4,  %xmm5
	subpd	%xmm6,  %xmm7
#endif

#ifndef CONJ
	SHUFPD_1 %xmm15, %xmm15
#endif

#ifdef LN
	movlpd	 6 * SIZE(AO), %xmm8
	movhpd	 6 * SIZE(AO), %xmm8
	movlpd	 7 * SIZE(AO), %xmm9
	movhpd	 7 * SIZE(AO), %xmm9
	movlpd	 4 * SIZE(AO), %xmm10
	movhpd	 4 * SIZE(AO), %xmm10