📄 trsm_kernel_rt_4x4_sse3.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 5 页
字号:
#ifdef LN	movddup	 0 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm2#endif#ifdef LT	movddup	 0 * SIZE(AO), %xmm0	mulpd	 %xmm0, %xmm2#endif#ifdef RN	movapd	%xmm2, %xmm0        unpckhpd %xmm0, %xmm0	movsd	 0 * SIZE(BO), %xmm4	mulsd	 %xmm4, %xmm2	movsd	 1 * SIZE(BO), %xmm5	mulsd	 %xmm2, %xmm5	subsd	 %xmm5, %xmm0	movsd	 3 * SIZE(BO), %xmm4	mulsd	 %xmm4, %xmm0	unpcklpd %xmm0, %xmm2#endif#ifdef RT	movapd	%xmm2, %xmm0        unpckhpd %xmm0, %xmm0	movsd	 3 * SIZE(BO), %xmm4	mulsd	 %xmm4, %xmm0	movsd	 2 * SIZE(BO), %xmm5	mulsd	 %xmm0, %xmm5	subsd	 %xmm5, %xmm2	movsd	 0 * SIZE(BO), %xmm4	mulsd	 %xmm4, %xmm2	unpcklpd %xmm0, %xmm2#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)#else	movsd	%xmm2,  0 * SIZE(CO1)	movhpd	%xmm2,  0 * SIZE(CO2)#endif#if defined(LN) || defined(LT)	movapd	%xmm2,   0 * SIZE(BO)#else	movapd	%xmm2,   0 * 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, 2), 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		.L79:#ifdef LN       leaq	(, K, SIZE), %rax       leaq	(B, %rax, 2), B#endif#if defined(LT) || defined(RN)	movq	BO, B#endif#ifdef RN	addq	$2, KK#endif#ifdef RT	subq	$2, KK#endif	ALIGN_4.L40:	movq	N,  J	sarq	$2, J		# j = (n >> 2)	jle	.L999	ALIGN_4	.L10:#if defined(LT) || defined(RN)	movq	A, AO#else	movq	A, AORIG#endif#ifdef RT       movq	K, %rax       salq	$2 + BASE_SHIFT, %rax       subq	%rax, B              leaq	(, LDC, 4), %rax       subq	%rax, C#endif	movq	C, CO1	leaq	(C, LDC, 1), CO2#ifndef RT	leaq	(C, LDC, 4), C#endif#ifdef LN	movq	OFFSET, %rax	addq	M, %rax	movq	%rax, KK#endif	#ifdef LT	movq	OFFSET, %rax	movq	%rax, KK#endif	movq	M,  I	sarq	$2, I	# i = (m >> 2)	jle	.L20	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	leaq	(, %rax, SIZE), %rax	movq	AORIG, AO	leaq	(AO, %rax, 4), AO	leaq	(B,  %rax, 4), 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	movapd	16 * SIZE(AO), %xmm12	movddup 16 * SIZE(BO), %xmm13	movapd	24 * SIZE(AO), %xmm14	movddup	24 * SIZE(BO), %xmm15	prefetchnta     4 * SIZE(CO1)	pxor	%xmm4, %xmm4	prefetchnta     4 * SIZE(CO2)	pxor	%xmm5, %xmm5	prefetchnta     4 * SIZE(CO1, LDC, 2)	pxor	%xmm6, %xmm6	prefetchnta     4 * SIZE(CO2, LDC, 2)	pxor	%xmm7, %xmm7#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif#if 1	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)	KERNEL9 (16  *  0)	KERNEL10(16  *  0)	KERNEL11(16  *  0)	KERNEL12(16  *  0)	KERNEL13(16  *  0)	KERNEL14(16  *  0)	KERNEL15(16  *  0)	KERNEL16(16  *  0)	cmpq	$128 *  1, %rax	jle,pn	.L12	KERNEL1 (16  *  1)	KERNEL2 (16  *  1)	KERNEL3 (16  *  1)	KERNEL4 (16  *  1)	KERNEL5 (16  *  1)	KERNEL6 (16  *  1)	KERNEL7 (16  *  1)	KERNEL8 (16  *  1)	KERNEL9 (16  *  1)	KERNEL10(16  *  1)	KERNEL11(16  *  1)	KERNEL12(16  *  1)	KERNEL13(16  *  1)	KERNEL14(16  *  1)	KERNEL15(16  *  1)	KERNEL16(16  *  1)	cmpq	$128 *  2, %rax	jle,pn	.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)	KERNEL9 (16  *  2)	KERNEL10(16  *  2)	KERNEL11(16  *  2)	KERNEL12(16  *  2)	KERNEL13(16  *  2)	KERNEL14(16  *  2)	KERNEL15(16  *  2)	KERNEL16(16  *  2)	cmpq	$128 *  3, %rax	jle,pn	.L12	KERNEL1 (16  *  3)	KERNEL2 (16  *  3)	KERNEL3 (16  *  3)	KERNEL4 (16  *  3)	KERNEL5 (16  *  3)	KERNEL6 (16  *  3)	KERNEL7 (16  *  3)	KERNEL8 (16  *  3)	KERNEL9 (16  *  3)	KERNEL10(16  *  3)	KERNEL11(16  *  3)	KERNEL12(16  *  3)	KERNEL13(16  *  3)	KERNEL14(16  *  3)	KERNEL15(16  *  3)	KERNEL16(16  *  3)	cmpq	$128 *  4, %rax	jle,pn	.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)	KERNEL9 (16  *  4)	KERNEL10(16  *  4)	KERNEL11(16  *  4)	KERNEL12(16  *  4)	KERNEL13(16  *  4)	KERNEL14(16  *  4)	KERNEL15(16  *  4)	KERNEL16(16  *  4)	cmpq	$128 *  5, %rax	jle,pn	.L12	KERNEL1 (16  *  5)	KERNEL2 (16  *  5)	KERNEL3 (16  *  5)	KERNEL4 (16  *  5)	KERNEL5 (16  *  5)	KERNEL6 (16  *  5)	KERNEL7 (16  *  5)	KERNEL8 (16  *  5)	KERNEL9 (16  *  5)	KERNEL10(16  *  5)	KERNEL11(16  *  5)	KERNEL12(16  *  5)	KERNEL13(16  *  5)	KERNEL14(16  *  5)	KERNEL15(16  *  5)	KERNEL16(16  *  5)	cmpq	$128 *  6, %rax	jle,pn	.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)	KERNEL9 (16  *  6)	KERNEL10(16  *  6)	KERNEL11(16  *  6)	KERNEL12(16  *  6)	KERNEL13(16  *  6)	KERNEL14(16  *  6)	KERNEL15(16  *  6)	KERNEL16(16  *  6)	cmpq	$128 *  7, %rax	jle,pn	.L12	KERNEL1 (16  *  7)	KERNEL2 (16  *  7)	KERNEL3 (16  *  7)	KERNEL4 (16  *  7)	KERNEL5 (16  *  7)	KERNEL6 (16  *  7)	KERNEL7 (16  *  7)	KERNEL8 (16  *  7)	KERNEL9 (16  *  7)	KERNEL10(16  *  7)	KERNEL11(16  *  7)	KERNEL12(16  *  7)	KERNEL13(16  *  7)	KERNEL14(16  *  7)	KERNEL15(16  *  7)	KERNEL16(16  *  7)	addq	$32 * 8  * SIZE, AO	addq	$32 * 8  * SIZE, BO	subq	$128 * 8, %rax	jg	.L1X.L12:	leaq	(AO, %rax, 2), AO	# * 16	leaq	(BO, %rax, 2), BO	# * 64#else	sarq	$3, %rax	je	.L15	ALIGN_4.L12:	mulpd	%xmm8, %xmm9	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AO)	addpd	%xmm9, %xmm0	movddup	 1 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm1	movddup	 2 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm2	movddup	 3 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 2 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm3	movddup	 0 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm4	movddup	 1 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm5	movddup	 2 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm6	movddup	 3 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 4 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm7	movddup	 4 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm0	movddup	 5 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm1	movddup	 6 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm2	movddup	 7 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 6 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm3	movddup	 4 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm4	movddup	 5 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm5	movddup	 6 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	addpd	%xmm9, %xmm6	movddup	 7 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	32 * SIZE(AO), %xmm8	addpd	%xmm9, %xmm7	movddup	32 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm0	movddup	 9 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm1	movddup	10 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm2	movddup	11 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	movapd	10 * SIZE(AO), %xmm10	addpd	%xmm11, %xmm3	movddup	 8 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm4	movddup	 9 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm5	movddup	10 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm6	movddup	11 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	movapd	12 * SIZE(AO), %xmm10	addpd	%xmm11, %xmm7	movddup	12 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm0	movddup	13 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm1	movddup	14 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm2	movddup	15 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	movapd	14 * SIZE(AO), %xmm10	addpd	%xmm11, %xmm3	movddup	12 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm4	movddup	13 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm5	movddup	14 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm6	movddup	15 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	movapd	40 * SIZE(AO), %xmm10	addpd	%xmm11, %xmm7	movddup	40 * SIZE(BO), %xmm11	mulpd	%xmm12, %xmm13	PREFETCH  (PREFETCHSIZE + 16) * SIZE(AO)	addpd	%xmm13, %xmm0	movddup	17 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm1	movddup	18 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm2	movddup	19 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	movapd	18 * SIZE(AO), %xmm12	addpd	%xmm13, %xmm3	movddup	16 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm4	movddup	17 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm5	movddup	18 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm6	movddup	19 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	movapd	20 * SIZE(AO), %xmm12	addpd	%xmm13, %xmm7	movddup	20 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm0	movddup	21 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm1	movddup	22 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm2	movddup	23 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	movapd	22 * SIZE(AO), %xmm12	addpd	%xmm13, %xmm3	movddup	20 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm4	movddup 21 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm5	movddup	22 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	addpd	%xmm13, %xmm6	movddup	23 * SIZE(BO), %xmm13	mulpd	%xmm12, %xmm13	movapd	48 * SIZE(AO), %xmm12	addpd	%xmm13, %xmm7	movddup	48 * SIZE(BO), %xmm13	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm0	movddup	25 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm1	movddup	26 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm2	movddup	27 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	movapd	26 * SIZE(AO), %xmm14	addpd	%xmm15, %xmm3	movddup	24 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm4	movddup	25 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm5	movddup	26 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm6	movddup	27 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	movapd	28 * SIZE(AO), %xmm14	addpd	%xmm15, %xmm7	movddup	28 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm0	movddup	29 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm1	movddup	30 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm2	movddup	31 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	movapd	30 * SIZE(AO), %xmm14	addpd	%xmm15, %xmm3	movddup	28 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm4	movddup	29 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm5	movddup	30 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	addpd	%xmm15, %xmm6	movddup	31 * SIZE(BO), %xmm15	mulpd	%xmm14, %xmm15	movapd	56 * SIZE(AO), %xmm14	addpd	%xmm15, %xmm7	movddup	56 * SIZE(BO), %xmm15	addq   $32 * SIZE, BO	addq   $32 * SIZE, AO	decq   %rax	BRANCH	jne    .L12#endif	ALIGN_4.L15:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$7, %rax		# if (k & 1)	BRANCH	je .L19	ALIGN_4.L16:	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	%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	%xmm10, %xmm11	addpd	%xmm11, %xmm4	movddup	 1 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm5	movddup	 2 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm6	movddup	 3 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	addpd	%xmm11, %xmm7	addq	$4 * SIZE, AO		# aoffset  += 4	addq	$4 * SIZE, BO		# boffset1 += 8	decq	%rax	jg	.L16	ALIGN_4.L19:#if defined(LN) || defined(RT)	movq	KK, %rax	subq	$4, %rax	leaq	(, %rax, SIZE), %rax	movq	AORIG, AO	leaq	(AO, %rax, 4), AO	leaq	(B,  %rax, 4), 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	%xmm4, %xmm12	unpcklpd %xmm5, %xmm4
⌨️ 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 + -