📄 trsm_kernel_ln_4x4_core2.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 5 页
字号:
	leaq	(AO, %rax, 2), AO	leaq	(B,  %rax, 2), B	leaq	(BO, %rax, 4), BO#endif#if defined(LN) || defined(LT)	movapd	%xmm8, %xmm0	unpcklpd %xmm9, %xmm8	unpckhpd %xmm9, %xmm0	movapd	 -16 * SIZE(B), %xmm9	movapd	 -14 * SIZE(B), %xmm13	subpd	%xmm8,  %xmm9	subpd	%xmm0,  %xmm13#else	movapd	-16 * SIZE(AO), %xmm0	movapd	-14 * SIZE(AO), %xmm2	subpd	%xmm8, %xmm0	subpd	%xmm9, %xmm2#endif#ifdef LN	movddup	-13 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm13	movddup	-14 * SIZE(AO), %xmm10	mulpd	 %xmm13, %xmm10	subpd	 %xmm10, %xmm9	movddup	-16 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm9#endif#ifdef LT	movddup	-16 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm9	movddup	-15 * SIZE(AO), %xmm10	mulpd	 %xmm9, %xmm10	subpd	 %xmm10, %xmm13	movddup	-13 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm13#endif#ifdef RN	movddup	-16 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm0	movddup	-15 * SIZE(B), %xmm9	mulpd	 %xmm0, %xmm9	subpd	 %xmm9, %xmm2	movddup	-13 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm2#endif#ifdef RT	movddup	-13 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm2	movddup	-14 * SIZE(B), %xmm9	mulpd	 %xmm2, %xmm9	subpd	 %xmm9, %xmm0	movddup	-16 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm0#endif#ifdef LN	subq	$2 * SIZE, CO1	subq	$2 * SIZE, CO2#endif#if defined(LN) || defined(LT)	movsd	%xmm9,  0 * SIZE(CO1)	movsd	%xmm13,  1 * SIZE(CO1)	movhpd	%xmm9,  0 * SIZE(CO2)	movhpd	%xmm13,  1 * SIZE(CO2)#else	movsd	%xmm0,  0 * SIZE(CO1)	movhpd	%xmm0,  1 * SIZE(CO1)	movsd	%xmm2,  0 * SIZE(CO2)	movhpd	%xmm2,  1 * SIZE(CO2)#endif#if defined(LN) || defined(LT)	movapd	%xmm9,  -16 * SIZE(B)	movapd	%xmm13,  -14 * SIZE(B)	movddup	%xmm9,  %xmm8	SHUFPD_3 %xmm9,  %xmm9	movddup	%xmm13,  %xmm12	SHUFPD_3 %xmm13,  %xmm13	movapd	%xmm8,  -16 * SIZE(BO)	movapd	%xmm9,  -14 * SIZE(BO)	movapd	%xmm12,  -12 * SIZE(BO)	movapd	%xmm13,  -10 * SIZE(BO)#else	movapd	%xmm0,  -16 * SIZE(AO)	movapd	%xmm2, -14 * 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:	movq	M,  I	sarq	$2, I	# i = (m >> 2)	jle	.L79	ALIGN_4.L51:#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	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#ifdef LN	prefetcht2    -3 * SIZE(CO1)	pxor	%xmm12, %xmm12	prefetcht2    -3 * SIZE(CO2)	pxor	%xmm13, %xmm13#else	prefetcht2     3 * SIZE(CO1)	pxor	%xmm12, %xmm12	prefetcht2     3 * SIZE(CO2)	pxor	%xmm13, %xmm13#endif#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	sarq	$2, %rax	je	.L55	ALIGN_4.L52:	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)	movapd	-16 * SIZE(AO), %xmm0	movapd	-14 * SIZE(AO), %xmm1	movapd	-16 * SIZE(BO), %xmm2	movapd	 %xmm2, %xmm3	movapd	-14 * SIZE(BO), %xmm4	movapd	 %xmm4, %xmm5	mulpd	%xmm0, %xmm2	mulpd	%xmm1, %xmm3	mulpd	%xmm0, %xmm4	mulpd	%xmm1, %xmm5	addpd	%xmm2, %xmm8	addpd	%xmm3, %xmm12	addpd	%xmm4, %xmm9	addpd	%xmm5, %xmm13	movapd	-12 * SIZE(AO), %xmm0	movapd	-10 * SIZE(AO), %xmm1	movapd	-12 * SIZE(BO), %xmm2	movapd	 %xmm2, %xmm3	movapd	-10 * SIZE(BO), %xmm4	movapd	 %xmm4, %xmm5	mulpd	%xmm0, %xmm2	mulpd	%xmm1, %xmm3	mulpd	%xmm0, %xmm4	mulpd	%xmm1, %xmm5	addpd	%xmm2, %xmm8	addpd	%xmm3, %xmm12	addpd	%xmm4, %xmm9	addpd	%xmm5, %xmm13	movapd	 -8 * SIZE(AO), %xmm0	movapd	 -6 * SIZE(AO), %xmm1	movapd	 -8 * SIZE(BO), %xmm2	movapd	 %xmm2, %xmm3	movapd	 -6 * SIZE(BO), %xmm4	movapd	 %xmm4, %xmm5	mulpd	%xmm0, %xmm2	mulpd	%xmm1, %xmm3	mulpd	%xmm0, %xmm4	mulpd	%xmm1, %xmm5	addpd	%xmm2, %xmm8	addpd	%xmm3, %xmm12	addpd	%xmm4, %xmm9	addpd	%xmm5, %xmm13	movapd	 -4 * SIZE(AO), %xmm0	movapd	 -2 * SIZE(AO), %xmm1	movapd	 -4 * SIZE(BO), %xmm2	movapd	 %xmm2, %xmm3	movapd	 -2 * SIZE(BO), %xmm4	movapd	 %xmm4, %xmm5	mulpd	%xmm0, %xmm2	mulpd	%xmm1, %xmm3	mulpd	%xmm0, %xmm4	mulpd	%xmm1, %xmm5	addpd	%xmm2, %xmm8	addpd	%xmm3, %xmm12	addpd	%xmm4, %xmm9	addpd	%xmm5, %xmm13	subq	$-16 * SIZE, AO	subq	$-16 * SIZE, BO	subq	$1, %rax	jne    .L52	ALIGN_4.L55:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$3, %rax		# if (k & 1)	BRANCH	je .L59	ALIGN_4.L56:	movapd	-16 * SIZE(AO), %xmm0	movapd	-14 * SIZE(AO), %xmm1	movapd	-16 * SIZE(BO), %xmm2	movapd	 %xmm2, %xmm3	movapd	-14 * SIZE(BO), %xmm4	movapd	 %xmm4, %xmm5	mulpd	%xmm0, %xmm2	mulpd	%xmm1, %xmm3	mulpd	%xmm0, %xmm4	mulpd	%xmm1, %xmm5	addpd	%xmm2, %xmm8	addpd	%xmm3, %xmm12	addpd	%xmm4, %xmm9	addpd	%xmm5, %xmm13	addq	$4 * SIZE, AO	addq	$4 * SIZE, BO	subq	$1, %rax	jg	.L56	ALIGN_4.L59:#if defined(LN) || defined(RT)	movq	KK, %rax#ifdef LN	subq	$4, %rax#else	subq	$2, %rax#endif	movq	AORIG, AO	movq	BORIG, B	leaq	16 * SIZE + BUFFER, BO	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 4), AO	leaq	(B,  %rax, 2), B	leaq	(BO, %rax, 4), BO#endif#if defined(LN) || defined(LT)	movapd	%xmm8, %xmm0	unpcklpd %xmm9, %xmm8	unpckhpd %xmm9, %xmm0	movapd	%xmm12, %xmm4	unpcklpd %xmm13, %xmm12	unpckhpd %xmm13, %xmm4	movapd	 -16 * SIZE(B), %xmm9	movapd	 -14 * SIZE(B), %xmm13	movapd	 -12 * SIZE(B), %xmm1	movapd	 -10 * SIZE(B), %xmm5	subpd	%xmm8,  %xmm9	subpd	%xmm0,  %xmm13	subpd	%xmm12,  %xmm1	subpd	%xmm4, %xmm5#else	movapd	 -16 * SIZE(AO), %xmm0	movapd	 -14 * SIZE(AO), %xmm1	movapd	 -12 * SIZE(AO), %xmm2	movapd	 -10 * SIZE(AO), %xmm3	subpd	%xmm8, %xmm0	subpd	%xmm12, %xmm1	subpd	%xmm9, %xmm2	subpd	%xmm13, %xmm3#endif#ifdef LN	movddup	 -1 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm5	movddup	 -2 * SIZE(AO), %xmm10	mulpd	 %xmm5, %xmm10	subpd	 %xmm10, %xmm1	movddup	 -3 * SIZE(AO), %xmm12	mulpd	 %xmm5, %xmm12	subpd	 %xmm12, %xmm13	movddup	 -4 * SIZE(AO), %xmm14	mulpd	 %xmm5, %xmm14	subpd	 %xmm14, %xmm9	movddup	 -6 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm1	movddup	 -7 * SIZE(AO), %xmm10	mulpd	 %xmm1, %xmm10	subpd	 %xmm10, %xmm13	movddup	 -8 * SIZE(AO), %xmm12	mulpd	 %xmm1, %xmm12	subpd	 %xmm12, %xmm9	movddup	-11 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm13	movddup	-12 * SIZE(AO), %xmm10	mulpd	 %xmm13, %xmm10	subpd	 %xmm10, %xmm9	movddup	-16 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm9#endif#ifdef LT	movddup	-16 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm9	movddup	-15 * SIZE(AO), %xmm10	mulpd	 %xmm9, %xmm10	subpd	 %xmm10, %xmm13	movddup	-14 * SIZE(AO), %xmm12	mulpd	 %xmm9, %xmm12	subpd	 %xmm12, %xmm1	movddup	-13 * SIZE(AO), %xmm14	mulpd	 %xmm9, %xmm14	subpd	 %xmm14, %xmm5	movddup	-11 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm13	movddup	-10 * SIZE(AO), %xmm10	mulpd	 %xmm13, %xmm10	subpd	 %xmm10, %xmm1	movddup	 -9 * SIZE(AO), %xmm12	mulpd	 %xmm13, %xmm12	subpd	 %xmm12, %xmm5	movddup	 -6 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm1	movddup	 -5 * SIZE(AO), %xmm10	mulpd	 %xmm1, %xmm10	subpd	 %xmm10, %xmm5	movddup	 -1 * SIZE(AO), %xmm8	mulpd	 %xmm8, %xmm5#endif#ifdef RN	movddup	-16 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm0	mulpd	 %xmm8, %xmm1	movddup	-15 * SIZE(B), %xmm9	movapd	%xmm9, %xmm10	mulpd	 %xmm0, %xmm9	subpd	 %xmm9, %xmm2	mulpd	 %xmm1, %xmm10	subpd	 %xmm10, %xmm3	movddup	-13 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm2	mulpd	 %xmm8, %xmm3#endif#ifdef RT	movddup	-13 * SIZE(B), %xmm8	mulpd	 %xmm8, %xmm2	mulpd	 %xmm8, %xmm3	movddup	-14 * SIZE(B), %xmm9	movapd	%xmm9, %xmm10	mulpd	 %xmm2, %xmm9	subpd	 %xmm9, %xmm0	mulpd	 %xmm3, %xmm10	subpd	 %xmm10, %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)#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)#endif#if defined(LN) || defined(LT)	movapd	%xmm9,  -16 * SIZE(B)	movapd	%xmm13,  -14 * SIZE(B)	movapd	%xmm1,  -12 * SIZE(B)	movapd	%xmm5, -10 * SIZE(B)	movddup	%xmm9,  %xmm8	SHUFPD_3 %xmm9,  %xmm9	movddup	%xmm13,  %xmm12	SHUFPD_3 %xmm13,  %xmm13	movddup	%xmm1,  %xmm0	SHUFPD_3 %xmm1,  %xmm1	movddup	%xmm5, %xmm4	SHUFPD_3 %xmm5, %xmm5	movapd	%xmm8,  -16 * SIZE(BO)	movapd	%xmm9,  -14 * SIZE(BO)	movapd	%xmm12,  -12 * SIZE(BO)	movapd	%xmm13,  -10 * SIZE(BO)	movapd	%xmm0,   -8 * SIZE(BO)	movapd	%xmm1,   -6 * SIZE(BO)	movapd	%xmm4,  -4 * SIZE(BO)	movapd	%xmm5,  -2 * SIZE(BO)#else	movapd	%xmm0,  -16 * SIZE(AO)	movapd	%xmm1,  -14 * SIZE(AO)	movapd	%xmm2, -12 * SIZE(AO)	movapd	%xmm3, -10 * 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	$8 * 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	.L51	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:	testq	$1, N	je	.L999	ALIGN_4.L81:/* 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	$0 + 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, 1), B	leaq	(BO, %rax, 2), 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	$3, %rax	jle	.L83	ALIGN_4	.L82:	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	subq	$-16 * SIZE, BO	subq	$1, %rax	jne	.L82	ALIGN_4.L83:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$7, %rax	BRANCH	jle	.L90	ALIGN_4.L84:	movddup	 -16 * SIZE(B), %xmm0	movapd	%xmm0,  0 * SIZE(BO)	addq	$1 * SIZE, B	addq	$2 * SIZE, BO	subq	$1, %rax	jne	.L84	ALIGN_4	.L90:#if defined(LT) || defined(RN)	movq	A, AO#else	movq	A, AORIG#endif#ifdef RT	subq	LDC, C#endif	movq	C, CO1			# coffset1 = c#ifndef RT	addq	LDC, C#endif	testq	$1, M	je	.L100	ALIGN_4.L111:#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	$0 + 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	.L115	ALIGN_4.L112:	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	mulsd	%xmm0, %xmm2	mulsd	%xmm1, %xmm3	addsd	%xmm2, %xmm8	addsd	%xmm3, %xmm9	movsd	-14 * SIZE(AO), %xmm0	movsd	-13 * SIZE(AO), %xmm1	movsd	-12 * SIZE(BO), %xmm2	movsd	-10 * SIZE(BO), %xmm3	mulsd	%xmm0, %xmm2
⌨️ 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 + -