📄 trsm_kernel_ln_4x4_sse2.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 5 页
字号:
	movlpd	 3 * SIZE(B), %xmm3	mulsd	 %xmm4, %xmm3	subsd	 %xmm3, %xmm7	mulsd	 5 * SIZE(B), %xmm5	movlpd	 6 * SIZE(B), %xmm1	mulsd	 %xmm5, %xmm1	subsd	 %xmm1, %xmm6	movlpd	 7 * SIZE(B), %xmm2	mulsd	 %xmm5, %xmm2	subsd	 %xmm2, %xmm7	mulsd	10 * SIZE(B), %xmm6	movlpd	11 * SIZE(B), %xmm1	mulsd	 %xmm6, %xmm1	subsd	 %xmm1, %xmm7	mulsd	15 * SIZE(B), %xmm7#endif#ifdef RT	mulsd	 15 * SIZE(B), %xmm7	movlpd	14 * SIZE(B), %xmm1	mulsd	 %xmm7, %xmm1	subsd	 %xmm1, %xmm6	movlpd	13 * SIZE(B), %xmm2	mulsd	 %xmm7, %xmm2	subsd	 %xmm2, %xmm5	movlpd	12 * SIZE(B), %xmm3	mulsd	 %xmm7, %xmm3	subsd	 %xmm3, %xmm4	mulsd	 10 * SIZE(B), %xmm6	movlpd	 9 * SIZE(B), %xmm1	mulsd	 %xmm6, %xmm1	subsd	 %xmm1, %xmm5	movlpd	 8 * SIZE(B), %xmm2	mulsd	 %xmm6, %xmm2	subsd	 %xmm2, %xmm4	mulsd	 5 * SIZE(B), %xmm5	movlpd	 4 * SIZE(B), %xmm1	mulsd	 %xmm5, %xmm1	subsd	 %xmm1, %xmm4	mulsd	 0 * SIZE(B), %xmm4#endif#ifdef LN	subq	$1 * SIZE, CO1	subq	$1 * SIZE, CO2#endif	movsd	%xmm4,  0 * SIZE(CO1)	movsd	%xmm5,  0 * SIZE(CO2)	movsd	%xmm6,  0 * SIZE(CO1, LDC, 2)	movsd	%xmm7,  0 * SIZE(CO2, LDC, 2)#if defined(LN) || defined(LT)	movsd	%xmm4,   0 * SIZE(B)	movsd	%xmm5,   1 * SIZE(B)	movsd	%xmm6,   2 * SIZE(B)	movsd	%xmm7,   3 * SIZE(B)	movsd	%xmm4,   0 * SIZE(BO)	movsd	%xmm4,   1 * SIZE(BO)	movsd	%xmm5,   2 * SIZE(BO)	movsd	%xmm5,   3 * SIZE(BO)	movsd	%xmm6,   4 * SIZE(BO)	movsd	%xmm6,   5 * SIZE(BO)	movsd	%xmm7,   6 * SIZE(BO)	movsd	%xmm7,   7 * SIZE(BO)#else	movsd	%xmm4,   0 * SIZE(AO)	movsd	%xmm5,   1 * SIZE(AO)	movsd	%xmm6,   2 * SIZE(AO)	movsd	%xmm7,   3 * 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	$4 * 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		.L20:	testq	$2, M	je	.L30	ALIGN_4.L21:#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	BUFFER, BO#if defined(LN) || defined(RT)	movq	KK, %rax	salq	$2 + BASE_SHIFT, %rax	leaq	(BO, %rax, 2), BO#endif		movapd	 0 * SIZE(AO), %xmm8	pxor	%xmm0, %xmm0	movapd	 0 * SIZE(BO), %xmm9	pxor	%xmm1, %xmm1	movapd	 8 * SIZE(AO), %xmm10	pxor	%xmm2, %xmm2	movapd	 8 * SIZE(BO), %xmm11	pxor	%xmm3, %xmm3	movapd	16 * SIZE(BO), %xmm13	movapd	24 * SIZE(BO), %xmm15#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	sarq	$3, %rax	je	.L25	ALIGN_4.L22:	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm0	PREFETCH	(PREFETCHSIZE     +  0) * SIZE(AO)	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	32 * SIZE(BO), %xmm9	addpd	%xmm8, %xmm3	movapd	 2 * SIZE(AO), %xmm8	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm0	movapd	10 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	addpd	%xmm11, %xmm1	movapd	12 * SIZE(BO), %xmm11	mulpd	%xmm8, %xmm11	mulpd	14 * SIZE(BO), %xmm8	addpd	%xmm11, %xmm2	movapd	40 * SIZE(BO), %xmm11	addpd	%xmm8, %xmm3	movapd	 4 * SIZE(AO), %xmm8	mulpd	%xmm8, %xmm13	addpd	%xmm13, %xmm0	movapd	18 * SIZE(BO), %xmm13	mulpd	%xmm8, %xmm13	addpd	%xmm13, %xmm1	movapd	20 * SIZE(BO), %xmm13	mulpd	%xmm8, %xmm13	mulpd	22 * SIZE(BO), %xmm8	addpd	%xmm13, %xmm2	movapd	48 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm3	movapd	 6 * SIZE(AO), %xmm8	mulpd	%xmm8, %xmm15	addpd	%xmm15, %xmm0	movapd	26 * SIZE(BO), %xmm15	mulpd	%xmm8, %xmm15	addpd	%xmm15, %xmm1	movapd	28 * SIZE(BO), %xmm15	mulpd	%xmm8, %xmm15	mulpd	30 * SIZE(BO), %xmm8	addpd	%xmm15, %xmm2	movapd	56 * SIZE(BO), %xmm15	addpd	%xmm8, %xmm3	movapd	16 * SIZE(AO), %xmm8	PREFETCH	(PREFETCHSIZE     +  8) * SIZE(AO)	mulpd	%xmm10, %xmm9	addpd	%xmm9, %xmm0	movapd	34 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	addpd	%xmm9, %xmm1	movapd	36 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm9	mulpd	38 * SIZE(BO), %xmm10	addpd	%xmm9, %xmm2	movapd	64 * SIZE(BO), %xmm9	addpd	%xmm10, %xmm3	movapd	10 * SIZE(AO), %xmm10	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm0	movapd	42 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm1	movapd	44 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	mulpd	46 * SIZE(BO), %xmm10	addpd	%xmm11, %xmm2	movapd	72 * SIZE(BO), %xmm11	addpd	%xmm10, %xmm3	movapd	12 * SIZE(AO), %xmm10	mulpd	%xmm10, %xmm13	addpd	%xmm13, %xmm0	movapd	50 * SIZE(BO), %xmm13	mulpd	%xmm10, %xmm13	addpd	%xmm13, %xmm1	movapd	52 * SIZE(BO), %xmm13	mulpd	%xmm10, %xmm13	mulpd	54 * SIZE(BO), %xmm10	addpd	%xmm13, %xmm2	movapd	80 * SIZE(BO), %xmm13	addpd	%xmm10, %xmm3	movapd	14 * SIZE(AO), %xmm10	mulpd	%xmm10, %xmm15	addpd	%xmm15, %xmm0	movapd	58 * SIZE(BO), %xmm15	mulpd	%xmm10, %xmm15	addpd	%xmm15, %xmm1	movapd	60 * SIZE(BO), %xmm15	mulpd	%xmm10, %xmm15	mulpd	62 * SIZE(BO), %xmm10	addpd	%xmm15, %xmm2	movapd	88 * SIZE(BO), %xmm15	addpd	%xmm10, %xmm3	movapd	24 * SIZE(AO), %xmm10	addq   $16 * SIZE, AO	addq   $64 * SIZE, BO	decq   %rax	jne    .L22	ALIGN_4.L25:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$7, %rax		# if (k & 1)	BRANCH	je .L29	ALIGN_4.L26:	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	 8 * SIZE(BO), %xmm9	addpd	%xmm8, %xmm3	movapd	 2 * SIZE(AO), %xmm8	addq	$2 * SIZE, AO		# aoffset  += 4	addq	$8 * SIZE, BO		# boffset1 += 8	decq	%rax	jg	.L26	ALIGN_4.L29:#if defined(LN) || defined(RT)	movq	KK, %rax#ifdef LN	subq	$2, %rax#else	subq	$4, %rax#endif	movq	AORIG, AO	movq	BORIG, B	leaq	BUFFER, BO	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 2), AO	leaq	(B,  %rax, 4), B	leaq	(BO, %rax, 8), BO#endif#if defined(LN) || defined(LT)	movapd	%xmm0, %xmm8	unpcklpd %xmm1, %xmm0	unpckhpd %xmm1, %xmm8	movapd	%xmm2, %xmm10	unpcklpd %xmm3, %xmm2	unpckhpd %xmm3, %xmm10	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	%xmm8,  %xmm5	subpd	%xmm10, %xmm7#else	movapd	 0 * SIZE(AO), %xmm8	movapd	 2 * SIZE(AO), %xmm10	movapd	 4 * SIZE(AO), %xmm12	movapd	 6 * SIZE(AO), %xmm14	subpd	%xmm0, %xmm8	subpd	%xmm1, %xmm10	subpd	%xmm2, %xmm12	subpd	%xmm3, %xmm14#endif#ifdef LN	movlpd	 3 * SIZE(AO), %xmm0	movhpd	 3 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm5	mulpd	 %xmm0, %xmm7	movlpd	 2 * SIZE(AO), %xmm2	movhpd	 2 * SIZE(AO), %xmm2	mulpd	 %xmm5, %xmm2	subpd	 %xmm2, %xmm1	movlpd	 2 * SIZE(AO), %xmm2	movhpd	 2 * SIZE(AO), %xmm2	mulpd	 %xmm7, %xmm2	subpd	 %xmm2, %xmm3	movlpd	 0 * SIZE(AO), %xmm0	movhpd	 0 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm1	mulpd	 %xmm0, %xmm3#endif#ifdef LT	movlpd	 0 * SIZE(AO), %xmm0	movhpd	 0 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm1	mulpd	 %xmm0, %xmm3	movlpd	 1 * SIZE(AO), %xmm2	movhpd	 1 * SIZE(AO), %xmm2	mulpd	 %xmm1, %xmm2	subpd	 %xmm2, %xmm5	movlpd	 1 * SIZE(AO), %xmm2	movhpd	 1 * SIZE(AO), %xmm2	mulpd	 %xmm3, %xmm2	subpd	 %xmm2, %xmm7	movlpd	 3 * SIZE(AO), %xmm0	movhpd	 3 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm5	mulpd	 %xmm0, %xmm7#endif#ifdef RN	movlpd	 0 * SIZE(B), %xmm0	movhpd	 0 * SIZE(B), %xmm0	mulpd	 %xmm0, %xmm8	movlpd	 1 * SIZE(B), %xmm1	movhpd	 1 * SIZE(B), %xmm1	mulpd	 %xmm8, %xmm1	subpd	 %xmm1, %xmm10	movlpd	 2 * SIZE(B), %xmm2	movhpd	 2 * SIZE(B), %xmm2	mulpd	 %xmm8, %xmm2	subpd	 %xmm2, %xmm12	movlpd	 3 * SIZE(B), %xmm3	movhpd	 3 * SIZE(B), %xmm3	mulpd	 %xmm8, %xmm3	subpd	 %xmm3, %xmm14	movlpd	 5 * SIZE(B), %xmm0	movhpd	 5 * SIZE(B), %xmm0	mulpd	 %xmm0, %xmm10	movlpd	 6 * SIZE(B), %xmm1	movhpd	 6 * SIZE(B), %xmm1	mulpd	 %xmm10, %xmm1	subpd	 %xmm1, %xmm12	movlpd	 7 * SIZE(B), %xmm2	movhpd	 7 * SIZE(B), %xmm2	mulpd	 %xmm10, %xmm2	subpd	 %xmm2, %xmm14	movlpd	10 * SIZE(B), %xmm0	movhpd	10 * SIZE(B), %xmm0	mulpd	 %xmm0, %xmm12	movlpd	11 * SIZE(B), %xmm1	movhpd	11 * SIZE(B), %xmm1	mulpd	 %xmm12, %xmm1	subpd	 %xmm1, %xmm14	movlpd	15 * SIZE(B), %xmm0	movhpd	15 * SIZE(B), %xmm0	mulpd	 %xmm0, %xmm14#endif#ifdef RT	movlpd	15 * SIZE(B), %xmm0	movhpd	15 * SIZE(B), %xmm0	mulpd	 %xmm0, %xmm14	movlpd	14 * SIZE(B), %xmm1	movhpd	14 * SIZE(B), %xmm1	mulpd	 %xmm14, %xmm1	subpd	 %xmm1, %xmm12	movlpd	13 * SIZE(B), %xmm2	movhpd	13 * SIZE(B), %xmm2	mulpd	 %xmm14, %xmm2	subpd	 %xmm2, %xmm10	movlpd	12 * SIZE(B), %xmm3	movhpd	12 * SIZE(B), %xmm3	mulpd	 %xmm14, %xmm3	subpd	 %xmm3, %xmm8	movlpd	10 * SIZE(B), %xmm0	movhpd	10 * SIZE(B), %xmm0	mulpd	 %xmm0, %xmm12	movlpd	 9 * SIZE(B), %xmm1	movhpd	 9 * SIZE(B), %xmm1	mulpd	 %xmm12, %xmm1	subpd	 %xmm1, %xmm10	movlpd	 8 * SIZE(B), %xmm2	movhpd	 8 * SIZE(B), %xmm2	mulpd	 %xmm12, %xmm2	subpd	 %xmm2, %xmm8	movlpd	 5 * SIZE(B), %xmm0	movhpd	 5 * SIZE(B), %xmm0	mulpd	 %xmm0, %xmm10	movlpd	 4 * SIZE(B), %xmm1	movhpd	 4 * SIZE(B), %xmm1	mulpd	 %xmm10, %xmm1	subpd	 %xmm1, %xmm8	movlpd	 0 * SIZE(B), %xmm0	movhpd	 0 * SIZE(B), %xmm0	mulpd	 %xmm0, %xmm8#endif#ifdef LN	subq	$2 * SIZE, CO1	subq	$2 * SIZE, CO2#endif#if defined(LN) || defined(LT)	movsd	%xmm1,  0 * SIZE(CO1)	movsd	%xmm5,  1 * SIZE(CO1)	movhpd	%xmm1,  0 * SIZE(CO2)	movhpd	%xmm5,  1 * SIZE(CO2)	movsd	%xmm3,  0 * SIZE(CO1, LDC, 2)	movsd	%xmm7,  1 * SIZE(CO1, LDC, 2)	movhpd	%xmm3,  0 * SIZE(CO2, LDC, 2)	movhpd	%xmm7,  1 * SIZE(CO2, LDC, 2)#else	movsd	%xmm8,  0 * SIZE(CO1)	movhpd	%xmm8,  1 * SIZE(CO1)	movsd	%xmm10,  0 * SIZE(CO2)	movhpd	%xmm10,  1 * SIZE(CO2)	movsd	%xmm12,  0 * SIZE(CO1, LDC, 2)	movhpd	%xmm12,  1 * SIZE(CO1, LDC, 2)	movsd	%xmm14,  0 * SIZE(CO2, LDC, 2)	movhpd	%xmm14,  1 * SIZE(CO2, LDC, 2)#endif#if defined(LN) || defined(LT)	movapd	%xmm1,   0 * SIZE(B)	movapd	%xmm3,   2 * SIZE(B)	movapd	%xmm5,   4 * SIZE(B)	movapd	%xmm7,   6 * SIZE(B)	movlpd	%xmm1,   0 * SIZE(BO)	movlpd	%xmm1,   1 * SIZE(BO)	movhpd	%xmm1,   2 * SIZE(BO)	movhpd	%xmm1,   3 * SIZE(BO)	movlpd	%xmm3,   4 * SIZE(BO)	movlpd	%xmm3,   5 * SIZE(BO)	movhpd	%xmm3,   6 * SIZE(BO)	movhpd	%xmm3,   7 * SIZE(BO)	movlpd	%xmm5,   8 * SIZE(BO)	movlpd	%xmm5,   9 * SIZE(BO)	movhpd	%xmm5,  10 * SIZE(BO)	movhpd	%xmm5,  11 * SIZE(BO)	movlpd	%xmm7,  12 * SIZE(BO)	movlpd	%xmm7,  13 * SIZE(BO)	movhpd	%xmm7,  14 * SIZE(BO)	movhpd	%xmm7,  15 * SIZE(BO)#else	movapd	%xmm8,   0 * SIZE(AO)	movapd	%xmm10,  2 * SIZE(AO)	movapd	%xmm12,  4 * SIZE(AO)	movapd	%xmm14,  6 * 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	$8 * 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	.L30:	movq	M,  I	sarq	$2, I	# i = (m >> 2)	jle	.L39	ALIGN_4.L11:#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	BUFFER, BO#if defined(LN) || defined(RT)	movq	KK, %rax	salq	$2 + BASE_SHIFT, %rax	leaq	(BO, %rax, 2), BO#endif		movapd	 0 * SIZE(BO), %xmm9	movapd	 2 * SIZE(BO), %xmm11	movapd	 4 * SIZE(BO), %xmm13	movapd	 8 * SIZE(BO), %xmm15	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#ifdef LN	PREFETCHW     -4 * SIZE(CO1)	pxor	%xmm4, %xmm4	PREFETCHW     -4 * SIZE(CO2)	pxor	%xmm5, %xmm5	PREFETCHW     -4 * SIZE(CO1, LDC, 2)	pxor	%xmm6, %xmm6	PREFETCHW     -4 * SIZE(CO2, LDC, 2)	pxor	%xmm7, %xmm7#else	PREFETCHW      4 * SIZE(CO1)	pxor	%xmm4, %xmm4	PREFETCHW      4 * SIZE(CO2)	pxor	%xmm5, %xmm5	PREFETCHW      4 * SIZE(CO1, LDC, 2)	pxor	%xmm6, %xmm6	PREFETCHW      4 * SIZE(CO2, LDC, 2)	pxor	%xmm7, %xmm7#endif#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)
⌨️ 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 + -