📄 trsm_kernel_lt_4x4_sse3.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 5 页
字号:
	movsd	15 * SIZE(BO), %xmm4	mulsd	 %xmm4, %xmm1	movsd	14 * SIZE(BO), %xmm5	mulsd	 %xmm1, %xmm5	subsd	 %xmm5, %xmm3	movsd	13 * SIZE(BO), %xmm6	mulsd	 %xmm1, %xmm6	subsd	 %xmm6, %xmm0	movsd	12 * SIZE(BO), %xmm7	mulsd	 %xmm1, %xmm7	subsd	 %xmm7, %xmm2	movsd	10 * SIZE(BO), %xmm4	mulsd	 %xmm4, %xmm3	movsd	 9 * SIZE(BO), %xmm5	mulsd	 %xmm3, %xmm5	subsd	 %xmm5, %xmm0	movsd	 8 * SIZE(BO), %xmm6	mulsd	 %xmm3, %xmm6	subsd	 %xmm6, %xmm2	movsd	 5 * SIZE(BO), %xmm4	mulsd	 %xmm4, %xmm0	movsd	 4 * SIZE(BO), %xmm5	mulsd	 %xmm0, %xmm5	subsd	 %xmm5, %xmm2	movsd	 0 * SIZE(BO), %xmm4	mulsd	 %xmm4, %xmm2	unpcklpd %xmm0, %xmm2	unpcklpd %xmm1, %xmm3#endif#ifdef LN	subq	$1 * SIZE, CO1	subq	$1 * SIZE, CO2#endif#if defined(LN) || defined(LT)	movsd	%xmm2,  0 * SIZE(CO1)	movhpd	%xmm2,  0 * SIZE(CO2)	movsd	%xmm3,  0 * SIZE(CO1, LDC, 2)	movhpd	%xmm3,  0 * SIZE(CO2, LDC, 2)#else	movsd	%xmm2,  0 * SIZE(CO1)	movhpd	%xmm2,  0 * SIZE(CO2)	movsd	%xmm3,  0 * SIZE(CO1, LDC, 2)	movhpd	%xmm3,  0 * SIZE(CO2, LDC, 2)#endif#if defined(LN) || defined(LT)	movapd	%xmm2,   0 * SIZE(BO)	movapd	%xmm3,   2 * SIZE(BO)#else	movapd	%xmm2,   0 * SIZE(AO)	movapd	%xmm3,   2 * 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	leaq	(BO, %rax, 4), BO#endif#ifdef LN	subq	$1, KK#endif#ifdef LT	addq	$1, KK#endif#ifdef RT       movq	K, %rax       salq	$0 + BASE_SHIFT, %rax       addq	%rax, AORIG#endif	ALIGN_4		.L39:#ifdef LN       leaq	(, K, SIZE), %rax       leaq	(B, %rax, 4), B#endif#if defined(LT) || defined(RN)	movq	BO, B#endif#ifdef RN	addq	$4, KK#endif#ifdef RT	subq	$4, KK#endif	decq	J			# j --	jg	.L10	ALIGN_4.L40:	testq	$2, N	je	.L80	ALIGN_4#if defined(LT) || defined(RN)	movq	A, AO#else	movq	A, AORIG#endif#ifdef RT       movq	K, %rax       salq	$1 + BASE_SHIFT, %rax       subq	%rax, B              leaq	(, LDC, 2), %rax       subq	%rax, C#endif	movq	C, CO1	leaq	(C, LDC, 1), CO2#ifndef RT	leaq	(C, LDC, 2), C#endif#ifdef LN	movq	OFFSET, %rax	addq	M, %rax	movq	%rax, KK#endif		movq	K, %rax	salq	$BASE_SHIFT + 1, %rax	leaq	(B, %rax), BB#ifdef LT	movq	OFFSET, %rax	movq	%rax, KK#endif	movq	M,  I	sarq	$2, I	# i = (m >> 2)	jle	.L60	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	leaq	(, %rax, SIZE), %rax	movq	AORIG, AO	leaq	(AO, %rax, 4), AO	leaq	(B,  %rax, 2), BO#else	movq	B, BO#endif		prefetcht0	  0 * SIZE(BB)	subq	   $-4 * SIZE, BB	movapd	 0 * SIZE(AO), %xmm8	pxor	%xmm0, %xmm0	movddup	 0 * SIZE(BO), %xmm9	pxor	%xmm1, %xmm1	movapd	 8 * SIZE(AO), %xmm10	pxor	%xmm4, %xmm4	movddup	 8 * SIZE(BO), %xmm11	pxor	%xmm5, %xmm5#ifdef HAVE_3DNOW	prefetchw      4 * SIZE(CO1)	prefetchw      4 * SIZE(CO2)#else	prefetchnta     4 * SIZE(CO1)	prefetchnta     4 * SIZE(CO2)#endif#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	sarq	$3, %rax	je	.L55	ALIGN_4.L52:	mulpd	%xmm8, %xmm9	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AO)	addpd	%xmm9, %xmm0	movddup	 1 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 2 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm1	movddup	 0 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm4	movddup	 1 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 4 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm5	movddup	 2 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm0	movddup	 3 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 6 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm1	movddup	 2 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm4	movddup	 3 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	16 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm5	movddup	 4 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	addpd	%xmm9, %xmm0	movddup	 5 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	movapd	10 * SIZE(AO), %xmm10	addpd	%xmm9, %xmm1	movddup	 4 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	addpd	%xmm9, %xmm4	movddup	 5 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	movapd	12 * SIZE(AO), %xmm10	addpd	%xmm9, %xmm5	movddup	 6 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	addpd	%xmm9, %xmm0	movddup	 7 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	movapd	14 * SIZE(AO), %xmm10	addpd	%xmm9, %xmm1	movddup	 6 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	addpd	%xmm9, %xmm4	movddup	 7 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	movapd	40 * SIZE(AO), %xmm10	addpd	%xmm9, %xmm5	movddup	16 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm11	PREFETCH  (PREFETCHSIZE + 16) * SIZE(AO)	addpd	%xmm11, %xmm0	movddup	 9 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	movapd	18 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm1	movddup	 8 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm4	movddup	 9 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	movapd	20 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm5	movddup	10 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm0	movddup	11 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	movapd	22 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm1	movddup	10 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm4	movddup	11 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	movapd	24 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm5	movddup	12 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm0	movddup	13 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	movapd	26 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm1	movddup	12 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm4	movddup	13 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	movapd	28 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm5	movddup	14 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm0	movddup	15 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	movapd	30 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm1	movddup	14 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm4	movddup	15 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	movapd	32 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm5	movddup	24 * SIZE(BO), %xmm11	addq   $32 * SIZE, AO	addq   $16 * SIZE, BO	decq   %rax	jne    .L52	ALIGN_4.L55:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$7, %rax		# if (k & 1)	BRANCH	je .L59	ALIGN_4.L56:	mulpd	%xmm8, %xmm9	movapd	 2 * SIZE(AO), %xmm10	addpd	%xmm9, %xmm0	movddup	 1 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movddup	 0 * SIZE(BO), %xmm11	addpd	%xmm9, %xmm1	movddup	 2 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm11	movapd	 4 * SIZE(AO), %xmm8	addpd	%xmm11, %xmm4	movddup	 1 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm5	addq	$4 * SIZE, AO		# aoffset  += 4	addq	$2 * SIZE, BO		# boffset1 += 8	decq	%rax	jg	.L56	ALIGN_4.L59:#if defined(LN) || defined(RT)	movq	KK, %rax#ifdef LN	subq	$4, %rax#else	subq	$2, %rax#endif	leaq	(, %rax, SIZE), %rax	movq	AORIG, AO	leaq	(AO, %rax, 4), AO	leaq	(B,  %rax, 2), BO#endif#if defined(LN) || defined(LT)	movapd	%xmm0, %xmm8	unpcklpd %xmm1, %xmm0	unpckhpd %xmm1, %xmm8	movapd	%xmm4, %xmm12	unpcklpd %xmm5, %xmm4	unpckhpd %xmm5, %xmm12	movapd	 0 * SIZE(BO), %xmm1	movapd	 2 * SIZE(BO), %xmm5	movapd	 4 * SIZE(BO), %xmm9	movapd	 6 * SIZE(BO), %xmm13	subpd	%xmm0,  %xmm1	subpd	%xmm8,  %xmm5	subpd	%xmm4,  %xmm9	subpd	%xmm12, %xmm13#else	movapd	 0 * SIZE(AO), %xmm8	movapd	 2 * SIZE(AO), %xmm9	movapd	 4 * SIZE(AO), %xmm10	movapd	 6 * SIZE(AO), %xmm11	subpd	%xmm0, %xmm8	subpd	%xmm4, %xmm9	subpd	%xmm1, %xmm10	subpd	%xmm5, %xmm11#endif#ifdef LN	movddup	15 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm13	movddup	14 * SIZE(AO), %xmm2	mulpd	 %xmm13, %xmm2	subpd	 %xmm2, %xmm9	movddup	13 * SIZE(AO), %xmm4	mulpd	 %xmm13, %xmm4	subpd	 %xmm4, %xmm5	movddup	12 * SIZE(AO), %xmm6	mulpd	 %xmm13, %xmm6	subpd	 %xmm6, %xmm1	movddup	10 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm9	movddup	 9 * SIZE(AO), %xmm2	mulpd	 %xmm9, %xmm2	subpd	 %xmm2, %xmm5	movddup	 8 * SIZE(AO), %xmm4	mulpd	 %xmm9, %xmm4	subpd	 %xmm4, %xmm1	movddup	 5 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm5	movddup	 4 * SIZE(AO), %xmm2	mulpd	 %xmm5, %xmm2	subpd	 %xmm2, %xmm1	movddup	 0 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm1#endif#ifdef LT	movddup	 0 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm1	movddup	 1 * SIZE(AO), %xmm2	mulpd	 %xmm1, %xmm2	subpd	 %xmm2, %xmm5	movddup	 2 * SIZE(AO), %xmm4	mulpd	 %xmm1, %xmm4	subpd	 %xmm4, %xmm9	movddup	 3 * SIZE(AO), %xmm6	mulpd	 %xmm1, %xmm6	subpd	 %xmm6, %xmm13	movddup	 5 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm5	movddup	 6 * SIZE(AO), %xmm2	mulpd	 %xmm5, %xmm2	subpd	 %xmm2, %xmm9	movddup	 7 * SIZE(AO), %xmm4	mulpd	 %xmm5, %xmm4	subpd	 %xmm4, %xmm13	movddup	10 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm9	movddup	11 * SIZE(AO), %xmm2	mulpd	 %xmm9, %xmm2	subpd	 %xmm2, %xmm13	movddup	15 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm13#endif#ifdef RN	movddup	 0 * SIZE(BO), %xmm0	mulpd	 %xmm0, %xmm8	mulpd	 %xmm0, %xmm9	movddup	 1 * SIZE(BO), %xmm1	mulpd	 %xmm8, %xmm1	subpd	 %xmm1, %xmm10	movddup	 1 * SIZE(BO), %xmm1	mulpd	 %xmm9, %xmm1	subpd	 %xmm1, %xmm11	movddup	 3 * SIZE(BO), %xmm0	mulpd	 %xmm0, %xmm10	mulpd	 %xmm0, %xmm11#endif#ifdef RT	movddup	 3 * SIZE(BO), %xmm0	mulpd	 %xmm0, %xmm10	mulpd	 %xmm0, %xmm11	movddup	 2 * SIZE(BO), %xmm1	mulpd	 %xmm10, %xmm1	subpd	 %xmm1, %xmm8	movddup	 2 * SIZE(BO), %xmm1	mulpd	 %xmm11, %xmm1	subpd	 %xmm1, %xmm9	movddup	 0 * SIZE(BO), %xmm0	mulpd	 %xmm0, %xmm8	mulpd	 %xmm0, %xmm9#endif#ifdef LN	subq	$4 * SIZE, CO1	subq	$4 * SIZE, CO2#endif#if defined(LN) || defined(LT)	movsd	%xmm1,  0 * SIZE(CO1)	movsd	%xmm5,  1 * SIZE(CO1)	movsd	%xmm9,  2 * SIZE(CO1)	movsd	%xmm13, 3 * SIZE(CO1)	movhpd	%xmm1,  0 * SIZE(CO2)	movhpd	%xmm5,  1 * SIZE(CO2)	movhpd	%xmm9,  2 * SIZE(CO2)	movhpd	%xmm13, 3 * SIZE(CO2)#else	movsd	%xmm8,  0 * SIZE(CO1)	movhpd	%xmm8,  1 * SIZE(CO1)	movsd	%xmm9,  2 * SIZE(CO1)	movhpd	%xmm9,  3 * SIZE(CO1)	movsd	%xmm10,  0 * SIZE(CO2)	movhpd	%xmm10,  1 * SIZE(CO2)	movsd	%xmm11,  2 * SIZE(CO2)	movhpd	%xmm11,  3 * SIZE(CO2)#endif#if defined(LN) || defined(LT)	movapd	%xmm1,   0 * SIZE(BO)	movapd	%xmm5,   2 * SIZE(BO)	movapd	%xmm9,   4 * SIZE(BO)	movapd	%xmm13,  6 * SIZE(BO)#else	movapd	%xmm8,   0 * SIZE(AO)	movapd	%xmm9,   2 * SIZE(AO)	movapd	%xmm10,  4 * SIZE(AO)	movapd	%xmm11,  6 * 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	leaq	(BO, %rax, 2), BO#endif#ifdef LN	subq	$4, KK#endif#ifdef LT	addq	$4, KK#endif#ifdef RT       movq	K, %rax       salq	$2 + BASE_SHIFT, %rax       addq	%rax, AORIG#endif	decq	I			# i --	jg	.L51	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	leaq	(, %rax, SIZE), %rax	movq	AORIG, AO	leaq	(AO, %rax, 2), AO	leaq	(B,  %rax, 2), BO#else	movq	B, BO#endif		movapd	 0 * SIZE(AO), %xmm8	pxor	%xmm0, %xmm0	movddup	 0 * SIZE(BO), %xmm9	pxor	%xmm1, %xmm1	movapd	 8 * SIZE(AO), %xmm10	pxor	%xmm2, %xmm2	movddup	 8 * SIZE(BO), %xmm11	pxor	%xmm3, %xmm3#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	sarq	$3, %rax	je	.L65	ALIGN_4.L62:	mulpd	%xmm8, %xmm9	PREFETCH  (PREFETCHSIZE + 0) * SIZE(AO)	addpd	%xmm9, %xmm0	movddup	 1 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 2 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm1	movddup	 2 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm2	movddup	 3 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 4 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm3	movddup	 4 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm0	movddup	 5 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 6 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm1	movddup	 6 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm2	movddup	 7 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	16 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm3	movddup	16 * SIZE(BO), %xmm9
⌨️ 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 + -