📄 ztrsm_kernel_ln_2x2_sse3.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 3 页
字号:
	movapd	%xmm8,  %xmm12	movapd	%xmm8,  %xmm13	SHUFPD_1 %xmm13, %xmm13	mulpd	%xmm2, %xmm12	mulpd	%xmm3, %xmm13	addsubpd %xmm13, %xmm12	subpd	 %xmm12, %xmm9	movapd	%xmm9,  %xmm12	SHUFPD_1 %xmm12, %xmm12	mulpd	 %xmm4, %xmm9	mulpd	 %xmm5, %xmm12	addsubpd %xmm12, %xmm9#endif#ifdef RT	movddup	 6 * SIZE(BO), %xmm0	movddup	 7 * SIZE(BO), %xmm1	movddup	 4 * SIZE(BO), %xmm2	movddup	 5 * SIZE(BO), %xmm3	movddup	 0 * SIZE(BO), %xmm4	movddup	 1 * SIZE(BO), %xmm5#ifdef CONJ	xorpd	%xmm7, %xmm1	xorpd	%xmm7, %xmm3	xorpd	%xmm7, %xmm5#endif	movapd	%xmm9,  %xmm12	SHUFPD_1 %xmm12, %xmm12	mulpd	 %xmm0, %xmm9	mulpd	 %xmm1, %xmm12	addsubpd %xmm12, %xmm9	movapd	%xmm9,  %xmm12	movapd	%xmm9,  %xmm13	SHUFPD_1 %xmm13, %xmm13	mulpd	%xmm2, %xmm12	mulpd	%xmm3, %xmm13	addsubpd %xmm13, %xmm12	subpd	 %xmm12, %xmm8	movapd	%xmm8,  %xmm12	SHUFPD_1 %xmm12, %xmm12	mulpd	 %xmm4, %xmm8	mulpd	 %xmm5, %xmm12	addsubpd %xmm12, %xmm8#endif#ifdef LN	subq	$2 * SIZE, CO1	subq	$2 * SIZE, CO2#endif#if defined(LN) || defined(LT)	movsd	%xmm8,  0 * SIZE(CO1)	movhpd	%xmm8,  1 * SIZE(CO1)	movsd	%xmm9,  0 * SIZE(CO2)	movhpd	%xmm9,  1 * SIZE(CO2)#else	movsd	%xmm8,  0 * SIZE(CO1)	movhpd	%xmm8,  1 * SIZE(CO1)	movsd	%xmm9,  0 * SIZE(CO2)	movhpd	%xmm9,  1 * SIZE(CO2)#endif#if defined(LN) || defined(LT)	movapd	%xmm8,   0 * SIZE(BO)	movapd	%xmm9,   2 * SIZE(BO)#else	movapd	%xmm8,   0 * SIZE(AO)	movapd	%xmm9,   2 * 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	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 + ZBASE_SHIFT, %rax       addq	%rax, AORIG#endif	ALIGN_4.L30:	movq	M,  I	sarq	$1, I		# i = (m >> 2)	jle	.L99	ALIGN_4.L10:#ifdef LN       movq	K, %rax       salq	$1 + ZBASE_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#ifdef LN	prefetchnta    -4 * SIZE(CO1)	prefetchnta    -4 * SIZE(CO2)#else	prefetchnta     4 * SIZE(CO1)	prefetchnta     4 * SIZE(CO2)#endif	pxor	%xmm4, %xmm4	pxor	%xmm5, %xmm5	pxor	%xmm6, %xmm6	pxor	%xmm7, %xmm7#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$-8, %rax	salq	$4, %rax	je	.L12	.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	.L11	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	.L11	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	.L11	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	.L11	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	.L11	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	.L11	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	.L11	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.L11:	leaq	(AO, %rax, 2), AO	# * 16	leaq	(BO, %rax, 2), BO	# * 64	ALIGN_4.L12:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$7, %rax		# if (k & 1)	BRANCH	je .L14	ALIGN_4.L13:	mulpd	%xmm8, %xmm9	movapd	 2 * SIZE(AO), %xmm10	ADD1	%xmm9, %xmm0	movddup	 1 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movddup	 0 * SIZE(BO), %xmm11	ADD2	%xmm9, %xmm1	movddup	 2 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	ADD1	%xmm9, %xmm2	movddup	 3 * SIZE(BO), %xmm9	mulpd	%xmm8, %xmm9	movapd	 4 * SIZE(AO), %xmm8	ADD2	%xmm9, %xmm3	movddup	 4 * SIZE(BO), %xmm9	mulpd	%xmm10, %xmm11	ADD1	%xmm11, %xmm4	movddup	 1 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	ADD2	%xmm11, %xmm5	movddup	 2 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	ADD1	%xmm11, %xmm6	movddup	 3 * SIZE(BO), %xmm11	mulpd	%xmm10, %xmm11	ADD2	%xmm11, %xmm7	addq	$4 * SIZE, AO		# aoffset  += 4	addq	$4 * SIZE, BO		# boffset1 += 8	decq	%rax	jg	.L13	ALIGN_4.L14:	SHUFPD_1 %xmm1, %xmm1	SHUFPD_1 %xmm3, %xmm3	SHUFPD_1 %xmm5, %xmm5	SHUFPD_1 %xmm7, %xmm7#ifndef CONJ	addsubpd	%xmm1, %xmm0	addsubpd	%xmm3, %xmm2	addsubpd	%xmm5, %xmm4	addsubpd	%xmm7, %xmm6#else	addsubpd	%xmm0, %xmm1	addsubpd	%xmm2, %xmm3	addsubpd	%xmm4, %xmm5	addsubpd	%xmm6, %xmm7#endif#if defined(LN) || defined(RT)	movq	KK, %rax	subq	$2, %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	 0 * SIZE(BO), %xmm8	movapd	 2 * SIZE(BO), %xmm9	movapd	 4 * SIZE(BO), %xmm10	movapd	 6 * SIZE(BO), %xmm11#else	movapd	 0 * SIZE(AO), %xmm8	movapd	 2 * SIZE(AO), %xmm9	movapd	 4 * SIZE(AO), %xmm10	movapd	 6 * SIZE(AO), %xmm11#endif#if   (defined(LN) || defined(LT)) && !defined(CONJ)	subpd	%xmm0,  %xmm8	subpd	%xmm2,  %xmm9	subpd	%xmm4,  %xmm10	subpd	%xmm6,  %xmm11#elif (defined(LN) || defined(LT)) &&  defined(CONJ)	subpd	%xmm1,  %xmm8	subpd	%xmm3,  %xmm9	subpd	%xmm5,  %xmm10	subpd	%xmm7,  %xmm11#elif (defined(RN) || defined(RT)) && !defined(CONJ)	subpd	%xmm0,  %xmm8	subpd	%xmm4,  %xmm9	subpd	%xmm2,  %xmm10	subpd	%xmm6,  %xmm11#else	addsubpd %xmm1,  %xmm8	addsubpd %xmm5,  %xmm9	addsubpd %xmm3,  %xmm10	addsubpd %xmm7,  %xmm11#endif#ifdef CONJ	pcmpeqb	%xmm7,  %xmm7	psllq	$63,    %xmm7#endif#if defined(LN) || defined(RT)#ifdef LN	movddup	 6 * SIZE(AO), %xmm0	movddup	 7 * SIZE(AO), %xmm1	movddup	 4 * SIZE(AO), %xmm2	movddup	 5 * SIZE(AO), %xmm3	movddup	 0 * SIZE(AO), %xmm4	movddup	 1 * SIZE(AO), %xmm5#else	movddup	 6 * SIZE(BO), %xmm0	movddup	 7 * SIZE(BO), %xmm1	movddup	 4 * SIZE(BO), %xmm2	movddup	 5 * SIZE(BO), %xmm3	movddup	 0 * SIZE(BO), %xmm4	movddup	 1 * SIZE(BO), %xmm5#endif#ifdef CONJ	xorpd	%xmm7, %xmm1	xorpd	%xmm7, %xmm3	xorpd	%xmm7, %xmm5#endif	movapd	%xmm10,  %xmm12	movapd	%xmm11,  %xmm13	SHUFPD_1 %xmm12, %xmm12	SHUFPD_1 %xmm13, %xmm13	mulpd	 %xmm0, %xmm10	mulpd	 %xmm0, %xmm11	mulpd	 %xmm1, %xmm12	mulpd	 %xmm1, %xmm13	addsubpd %xmm12, %xmm10	addsubpd %xmm13, %xmm11	movapd	%xmm10,  %xmm12	movapd	%xmm10,  %xmm13	movapd	%xmm11,  %xmm14	movapd	%xmm11,  %xmm15	SHUFPD_1 %xmm13, %xmm13	SHUFPD_1 %xmm15, %xmm15	mulpd	%xmm2, %xmm12	mulpd	%xmm2, %xmm14	mulpd	%xmm3, %xmm13	mulpd	%xmm3, %xmm15	addsubpd %xmm13, %xmm12	addsubpd %xmm15, %xmm14	subpd	 %xmm12, %xmm8	subpd	 %xmm14, %xmm9	movapd	%xmm8,  %xmm12	movapd	%xmm9,  %xmm13	SHUFPD_1 %xmm12, %xmm12	SHUFPD_1 %xmm13, %xmm13	mulpd	 %xmm4, %xmm8	mulpd	 %xmm4, %xmm9	mulpd	 %xmm5, %xmm12	mulpd	 %xmm5, %xmm13	addsubpd %xmm12, %xmm8	addsubpd %xmm13, %xmm9#endif#if defined(LT) || defined(RN)#ifdef LT	movddup	 0 * SIZE(AO), %xmm0	movddup	 1 * SIZE(AO), %xmm1	movddup	 2 * SIZE(AO), %xmm2	movddup	 3 * SIZE(AO), %xmm3	movddup	 6 * SIZE(AO), %xmm4	movddup	 7 * SIZE(AO), %xmm5#else	movddup	 0 * SIZE(BO), %xmm0	movddup	 1 * SIZE(BO), %xmm1	movddup	 2 * SIZE(BO), %xmm2	movddup	 3 * SIZE(BO), %xmm3	movddup	 6 * SIZE(BO), %xmm4	movddup	 7 * SIZE(BO), %xmm5#endif#ifdef CONJ	xorpd	%xmm7, %xmm1	xorpd	%xmm7, %xmm3	xorpd	%xmm7, %xmm5#endif	movapd	%xmm8,  %xmm12	movapd	%xmm9,  %xmm13	SHUFPD_1 %xmm12, %xmm12	SHUFPD_1 %xmm13, %xmm13	mulpd	 %xmm0, %xmm8	mulpd	 %xmm0, %xmm9	mulpd	 %xmm1, %xmm12	mulpd	 %xmm1, %xmm13	addsubpd %xmm12, %xmm8	addsubpd %xmm13, %xmm9	movapd	%xmm8,  %xmm12	movapd	%xmm8,  %xmm13	movapd	%xmm9,  %xmm14	movapd	%xmm9,  %xmm15	SHUFPD_1 %xmm13, %xmm13	SHUFPD_1 %xmm15, %xmm15	mulpd	%xmm2, %xmm12	mulpd	%xmm2, %xmm14	mulpd	%xmm3, %xmm13	mulpd	%xmm3, %xmm15	addsubpd %xmm13, %xmm12	addsubpd %xmm15, %xmm14	subpd	 %xmm12, %xmm10	subpd	 %xmm14, %xmm11	movapd	%xmm10,  %xmm12	movapd	%xmm11,  %xmm13	SHUFPD_1 %xmm12, %xmm12	SHUFPD_1 %xmm13, %xmm13	mulpd	 %xmm4, %xmm10	mulpd	 %xmm4, %xmm11	mulpd	 %xmm5, %xmm12	mulpd	 %xmm5, %xmm13	addsubpd %xmm12, %xmm10	addsubpd %xmm13, %xmm11#endif#ifdef LN	subq	$4 * SIZE, CO1	subq	$4 * SIZE, CO2#endif#if defined(LN) || defined(LT)	movsd	%xmm8,  0 * SIZE(CO1)	movhpd	%xmm8,  1 * SIZE(CO1)	movsd	%xmm10, 2 * SIZE(CO1)	movhpd	%xmm10, 3 * SIZE(CO1)	movsd	%xmm9,   0 * SIZE(CO2)	movhpd	%xmm9,   1 * SIZE(CO2)	movsd	%xmm11,  2 * SIZE(CO2)	movhpd	%xmm11,  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	%xmm8,   0 * SIZE(BO)	movapd	%xmm9,   2 * SIZE(BO)	movapd	%xmm10,  4 * SIZE(BO)	movapd	%xmm11,  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, 4), BO#endif#ifdef LN	subq	$2, KK#endif#ifdef LT	addq	$2, KK#endif#ifdef RT       movq	K, %rax       salq	$1 + ZBASE_SHIFT, %rax       addq	%rax, AORIG#endif	decq	I			# i --	jg	.L10	ALIGN_4	.L99:#ifdef LN       leaq	(, K, SIZE), %rax       leaq	(B, %rax, 4), B#endif#if defined(LT) || defined(RN)	movq	BO, B#endif#ifdef RN	addq	$2, KK#endif#ifdef RT	subq	$2, KK#endif	decq	J			# j --	jg	.L01.L100:	testq	$1, N	jle	.L999.L101:#if defined(LT) || defined(RN)	movq	A, AO#else	movq	A, AORIG#endif#ifdef RT       movq	K, %rax       salq	$0 + ZBASE_SHIFT, %rax       subq	%rax, B              subq	LDC, C#endif	movq	C, CO1		# coffset1 = c#ifndef RT	addq	LDC, C#endif#ifdef LN	movq	OFFSET, KK	addq	M, KK#endif	#ifdef LT	movq	OFFSET, KK#endif	testq	$1, M	jle	.L130	ALIGN_4#ifdef LN       movq	K, %rax       salq	$0 + ZBASE_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
⌨️ 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 + -