📄 trsm_kernel_ln_4x4_core2.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 5 页
字号:
	subpd	 %xmm12, %xmm5	movddup	 -13 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm0, %xmm10	subpd	 %xmm10, %xmm6	mulpd	 %xmm1, %xmm12	subpd	 %xmm12, %xmm7	movddup	 -11 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm2	mulpd	 %xmm8, %xmm3	movddup	 -10 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm2, %xmm10	subpd	 %xmm10, %xmm4	mulpd	 %xmm3, %xmm12	subpd	 %xmm12, %xmm5	movddup	  -9 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm2, %xmm10	subpd	 %xmm10, %xmm6	mulpd	 %xmm3, %xmm12	subpd	 %xmm12, %xmm7	movddup	  -6 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm4	mulpd	 %xmm8, %xmm5	movddup	  -5 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm4, %xmm10	subpd	 %xmm10, %xmm6	mulpd	 %xmm5, %xmm12	subpd	 %xmm12, %xmm7	movddup	  -1 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm6	mulpd	 %xmm8, %xmm7#endif#ifdef RT	movddup	 -1 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm6	mulpd	 %xmm8, %xmm7	movddup	 -2 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm6, %xmm10	subpd	 %xmm10, %xmm4	mulpd	 %xmm7, %xmm12	subpd	 %xmm12, %xmm5	movddup	 -3 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm6, %xmm10	subpd	 %xmm10, %xmm2	mulpd	 %xmm7, %xmm12	subpd	 %xmm12, %xmm3	movddup	 -4 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm6, %xmm10	subpd	 %xmm10, %xmm0	mulpd	 %xmm7, %xmm12	subpd	 %xmm12, %xmm1	movddup	 -6 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm4	mulpd	 %xmm8, %xmm5	movddup	 -7 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm4, %xmm10	subpd	 %xmm10, %xmm2	mulpd	 %xmm5, %xmm12	subpd	 %xmm12, %xmm3	movddup	 -8 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm4, %xmm10	subpd	 %xmm10, %xmm0	mulpd	 %xmm5, %xmm12	subpd	 %xmm12, %xmm1	movddup	-11 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm2	mulpd	 %xmm8, %xmm3	movddup	-12 * SIZE(B), %xmm10	movapd	%xmm10, %xmm12	mulpd	 %xmm2, %xmm10	subpd	 %xmm10, %xmm0	mulpd	 %xmm3, %xmm12	subpd	 %xmm12, %xmm1	movddup	-16 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm0	mulpd	 %xmm8, %xmm1#endif#ifdef LN	subq	$4 * SIZE, CO1	subq	$4 * SIZE, CO2#endif#if defined(LN) || defined(LT)	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)	movsd	%xmm11,  0 * SIZE(CO1, LDC, 2)	movsd	%xmm15,  1 * SIZE(CO1, LDC, 2)	movsd	%xmm3, 2 * SIZE(CO1, LDC, 2)	movsd	%xmm7, 3 * SIZE(CO1, LDC, 2)	movhpd	%xmm11,  0 * SIZE(CO2, LDC, 2)	movhpd	%xmm15,  1 * SIZE(CO2, LDC, 2)	movhpd	%xmm3, 2 * SIZE(CO2, LDC, 2)	movhpd	%xmm7, 3 * SIZE(CO2, LDC, 2)#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)	movsd	%xmm4,  0 * SIZE(CO1, LDC, 2)	movhpd	%xmm4,  1 * SIZE(CO1, LDC, 2)	movsd	%xmm5,  2 * SIZE(CO1, LDC, 2)	movhpd	%xmm5,  3 * SIZE(CO1, LDC, 2)	movsd	%xmm6,  0 * SIZE(CO2, LDC, 2)	movhpd	%xmm6,  1 * SIZE(CO2, LDC, 2)	movsd	%xmm7,  2 * SIZE(CO2, LDC, 2)	movhpd	%xmm7,  3 * SIZE(CO2, LDC, 2)#endif#if defined(LN) || defined(LT)	movapd	%xmm9, -16 * SIZE(B)	movapd	%xmm11, -14 * SIZE(B)	movapd	%xmm13, -12 * SIZE(B)	movapd	%xmm15, -10 * SIZE(B)	movapd	%xmm1,  -8 * SIZE(B)	movapd	%xmm3, -6 * SIZE(B)	movapd	%xmm5, -4 * SIZE(B)	movapd	%xmm7, -2 * SIZE(B)	movddup	%xmm9,  %xmm8	SHUFPD_3 %xmm9,  %xmm9	movddup	%xmm11,  %xmm10	SHUFPD_3 %xmm11,  %xmm11	movddup	%xmm13,  %xmm12	SHUFPD_3 %xmm13,  %xmm13	movddup	%xmm15,  %xmm14	SHUFPD_3 %xmm15,  %xmm15	movddup	%xmm1,  %xmm0	SHUFPD_3 %xmm1,  %xmm1	movddup	%xmm3, %xmm2	SHUFPD_3 %xmm3, %xmm3	movddup	%xmm5, %xmm4	SHUFPD_3 %xmm5, %xmm5	movddup	%xmm7, %xmm6	SHUFPD_3 %xmm7, %xmm7	movapd	%xmm8, -16 * SIZE(BO)	movapd	%xmm9, -14 * SIZE(BO)	movapd	%xmm10, -12 * SIZE(BO)	movapd	%xmm11, -10 * SIZE(BO)	movapd	%xmm12,  -8 * SIZE(BO)	movapd	%xmm13,  -6 * SIZE(BO)	movapd	%xmm14,  -4 * SIZE(BO)	movapd	%xmm15,  -2 * SIZE(BO)	movapd	%xmm0,   0 * SIZE(BO)	movapd	%xmm1,   2 * SIZE(BO)	movapd	%xmm2,  4 * SIZE(BO)	movapd	%xmm3,  6 * SIZE(BO)	movapd	%xmm4,  8 * SIZE(BO)	movapd	%xmm5, 10 * SIZE(BO)	movapd	%xmm6, 12 * SIZE(BO)	movapd	%xmm7, 14 * SIZE(BO)#else	movapd	%xmm0,  -16 * SIZE(AO)	movapd	%xmm1,  -14 * SIZE(AO)	movapd	%xmm2, -12 * SIZE(AO)	movapd	%xmm3, -10 * SIZE(AO)	movapd	%xmm4,  -8 * SIZE(AO)	movapd	%xmm5,  -6 * SIZE(AO)	movapd	%xmm6,  -4 * SIZE(AO)	movapd	%xmm7,  -2 * 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	$16 * 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	.L11	ALIGN_4		.L39:#ifdef LN       leaq	(, K, SIZE), %rax       leaq	(B, %rax, 4), B#endif#if defined(LT) || defined(RN)	movq	K,  %rax	subq	KK, %rax	leaq	(,%rax, SIZE), %rax	leaq	(B,  %rax, 4), B#endif#ifdef RN	addq	$4, KK#endif#ifdef RT	subq	$4, KK#endif	decq	J			# j --	jg	.L01	ALIGN_4.L40:	testq	$3, N	je	.L999	testq	$2, N	je	.L80	ALIGN_4.L41:/* 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	$1 + 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, 2), 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	.L43	ALIGN_4	.L42:	movddup	 -16 * SIZE(B), %xmm0	movddup	 -15 * SIZE(B), %xmm1	movddup	 -14 * SIZE(B), %xmm2	movddup	 -13 * SIZE(B), %xmm3	movddup	 -12 * SIZE(B), %xmm4	movddup	 -11 * SIZE(B), %xmm5	movddup	 -10 * SIZE(B), %xmm6	movddup	  -9 * SIZE(B), %xmm7	movapd	%xmm0,  0 * SIZE(BO)	movapd	%xmm1,  2 * SIZE(BO)	movapd	%xmm2,  4 * SIZE(BO)	movapd	%xmm3,  6 * SIZE(BO)	movapd	%xmm4,  8 * SIZE(BO)	movapd	%xmm5, 10 * SIZE(BO)	movapd	%xmm6, 12 * SIZE(BO)	movapd	%xmm7, 14 * SIZE(BO)	addq	$8 * SIZE, B	addq	$16 * SIZE, BO	subq	$1, %rax	jne	.L42	ALIGN_4.L43:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$3, %rax	BRANCH	jle	.L50	ALIGN_4.L44:	movddup	 -16 * SIZE(B), %xmm0	movddup	 -15 * SIZE(B), %xmm1	movapd	%xmm0,  0 * SIZE(BO)	movapd	%xmm1,  2 * SIZE(BO)	addq	$2 * SIZE, B	addq	$4 * SIZE, BO	decq	%rax	jne	.L44	ALIGN_4	.L50:#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	# coffset2 = c + ldc#ifndef RT	leaq	(C, LDC, 2), C#endif	testq	$1, M	je	.L60	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		.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
⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -