📄 gemv_t_sse2.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 5 页
字号:
	andq	$2, J	jle	.L40	ALIGN_3.L31:	movq	A, AO1	leaq	(A, LDA, 1), AO2	leaq	(A, LDA, 2), A	movq	BUFFER, BO	movapd	 0 * SIZE(BO), %xmm12	pxor	%xmm0, %xmm0	movapd	 2 * SIZE(BO), %xmm13	pxor	%xmm1, %xmm1	pxor	%xmm2, %xmm2	pxor	%xmm3, %xmm3#if defined(CORE2) || defined(PENRYN)	PREFETCHW	4 * SIZE(CO)#endif	movq	MIN_M, I	sarq	$4,  I	jle	.L34	movapd	 0 * SIZE(AO1), %xmm4	movapd	 0 * SIZE(AO2), %xmm5	movapd	 2 * SIZE(AO1), %xmm8	movapd	 2 * SIZE(AO2), %xmm9	decq	I	jle	.L33	ALIGN_3.L32:	PREFETCH	PREFETCHSIZE * SIZE(AO1)	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	 4 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	movapd	 4 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	movapd	 4 * SIZE(AO2), %xmm5	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm0	movapd	 6 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	movapd	 6 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm1	movapd	 6 * SIZE(AO2), %xmm9	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	 8 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	movapd	 8 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	movapd	 8 * SIZE(AO2), %xmm5	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm0	movapd	10 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	movapd	10 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm1	movapd	10 * SIZE(AO2), %xmm9	PREFETCH	PREFETCHSIZE * SIZE(AO2)	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	12 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	movapd	12 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	movapd	12 * SIZE(AO2), %xmm5	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm0	movapd	14 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	movapd	14 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm1	movapd	14 * SIZE(AO2), %xmm9	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	16 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	movapd	16 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	movapd	16 * SIZE(AO2), %xmm5	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm0	movapd	18 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	movapd	18 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm1	movapd	18 * SIZE(AO2), %xmm9	subq	$-16 * SIZE, AO1	subq	$-16 * SIZE, AO2	subq	$-16 * SIZE, BO	decq	I	jg	.L32	ALIGN_3.L33:	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	 4 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	movapd	 4 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	movapd	 4 * SIZE(AO2), %xmm5	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm2	movapd	 6 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	movapd	 6 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm3	movapd	 6 * SIZE(AO2), %xmm9	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	 8 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	movapd	 8 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	movapd	 8 * SIZE(AO2), %xmm5	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm2	movapd	10 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	movapd	10 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm3	movapd	10 * SIZE(AO2), %xmm9	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	12 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	movapd	12 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	movapd	12 * SIZE(AO2), %xmm5	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm2	movapd	14 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	movapd	14 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm3	movapd	14 * SIZE(AO2), %xmm9	mulpd	%xmm12, %xmm4	subq	$-16 * SIZE, AO1	addpd	%xmm4, %xmm0	subq	$-16 * SIZE, AO2	mulpd	%xmm12, %xmm5	addpd	%xmm5, %xmm1	movapd	16 * SIZE(BO), %xmm12	mulpd	%xmm13, %xmm8	subq	$-16 * SIZE, BO	addpd	%xmm8, %xmm2	mulpd	%xmm13, %xmm9	movapd	 2 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm3.L34:	movq	MIN_M, I	andq	$8,  I	je	.L35	movapd	 0 * SIZE(AO1), %xmm4	movapd	 0 * SIZE(AO2), %xmm5	movapd	 2 * SIZE(AO1), %xmm8	movapd	 2 * SIZE(AO2), %xmm9	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	 4 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	movapd	 4 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	movapd	 4 * SIZE(AO2), %xmm5	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm2	movapd	 6 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	movapd	 6 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm3	movapd	 6 * SIZE(AO2), %xmm9	mulpd	%xmm12, %xmm4	subq	$-8 * SIZE, AO1	addpd	%xmm4, %xmm0	subq	$-8 * SIZE, AO2	mulpd	%xmm12, %xmm5	addpd	%xmm5, %xmm1	movapd	 8 * SIZE(BO), %xmm12	mulpd	%xmm13, %xmm8	subq	$-8 * SIZE, BO	addpd	%xmm8, %xmm2	mulpd	%xmm13, %xmm9	movapd	 2 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm3.L35:	movq	MIN_M, I	andq	$4,  I	je	.L36	movapd	 0 * SIZE(AO1), %xmm4	movapd	 0 * SIZE(AO2), %xmm5	movapd	 2 * SIZE(AO1), %xmm8	movapd	 2 * SIZE(AO2), %xmm9	mulpd	%xmm12, %xmm4	subq	$-4 * SIZE, AO1	addpd	%xmm4, %xmm0	subq	$-4 * SIZE, AO2	mulpd	%xmm12, %xmm5	addpd	%xmm5, %xmm1	movapd	 4 * SIZE(BO), %xmm12	mulpd	%xmm13, %xmm8	subq	$-4 * SIZE, BO	addpd	%xmm8, %xmm2	mulpd	%xmm13, %xmm9	movapd	 2 * SIZE(BO), %xmm13	addpd	%xmm9, %xmm3.L36:	movq	MIN_M, I	andq	$2,  I	je	.L37	movapd	 0 * SIZE(AO1), %xmm4	movapd	 0 * SIZE(AO2), %xmm5	mulpd	%xmm12, %xmm4	subq	$-2 * SIZE, AO1	addpd	%xmm4, %xmm0	subq	$-2 * SIZE, AO2	mulpd	%xmm12, %xmm5	movapd	 2 * SIZE(BO), %xmm12	addpd	%xmm5, %xmm1	subq	$-2 * SIZE, BO.L37:	movq	MIN_M, I	andq	$1,  I	je	.L39	movsd	0 * SIZE(AO1), %xmm8	mulsd	%xmm12, %xmm8	addsd	%xmm8, %xmm0	movsd	0 * SIZE(AO2), %xmm8	mulsd	%xmm12, %xmm8	addsd	%xmm8, %xmm1.L39:	addpd	%xmm2, %xmm0	addpd	%xmm3, %xmm1#ifdef HAVE_SSE3	haddpd	%xmm0, %xmm0	haddpd	%xmm1, %xmm1#else	movapd	%xmm0, %xmm9	movapd	%xmm1, %xmm10	unpckhpd %xmm0, %xmm0	unpckhpd %xmm1, %xmm1	addsd	 %xmm9,  %xmm0	addsd	 %xmm10, %xmm1#endif	mulsd	ALPHA, %xmm0	mulsd	ALPHA, %xmm1	movq	CO, TEMP	addsd	(TEMP), %xmm0	addq	INCY, TEMP	addsd	(TEMP), %xmm1	addq	INCY, TEMP	movsd	%xmm0, (CO)	addq	INCY, CO	movsd	%xmm1, (CO)	addq	INCY, CO	ALIGN_3.L40:	movq	N,  J	andq	$1, J	jle	.L49	ALIGN_3.L41:	movq	A, AO1	addq	LDA, A	movq	BUFFER, BO	movapd	 0 * SIZE(BO), %xmm12	pxor	%xmm0, %xmm0	movapd	 2 * SIZE(BO), %xmm13	pxor	%xmm1, %xmm1	pxor	%xmm2, %xmm2	pxor	%xmm3, %xmm3#if defined(CORE2) || defined(PENRYN)	PREFETCHW	4 * SIZE(CO)#endif	movq	MIN_M, I	sarq	$4,  I	jle	.L44	movapd	 0 * SIZE(AO1), %xmm4	movapd	 2 * SIZE(AO1), %xmm8	decq	I	jle	.L43	ALIGN_3.L42:	PREFETCH	PREFETCHSIZE * SIZE(AO1)	mulpd	%xmm12, %xmm4	movapd	 4 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	movapd	 4 * SIZE(AO1), %xmm4	mulpd	%xmm13, %xmm8	movapd	 6 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm0	movapd	 6 * SIZE(AO1), %xmm8	mulpd	%xmm12, %xmm4	movapd	 8 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	movapd	 8 * SIZE(AO1), %xmm4	mulpd	%xmm13, %xmm8	movapd	10 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm0	movapd	10 * SIZE(AO1), %xmm8	mulpd	%xmm12, %xmm4	movapd	12 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	movapd	12 * SIZE(AO1), %xmm4	mulpd	%xmm13, %xmm8	movapd	14 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm0	movapd	14 * SIZE(AO1), %xmm8	mulpd	%xmm12, %xmm4	movapd	16 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	movapd	16 * SIZE(AO1), %xmm4	mulpd	%xmm13, %xmm8	movapd	18 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm0	movapd	18 * SIZE(AO1), %xmm8	subq	$-16 * SIZE, AO1	subq	$-16 * SIZE, BO	decq	I	jg	.L42	ALIGN_3.L43:	mulpd	%xmm12, %xmm4	movapd	 4 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	movapd	 4 * SIZE(AO1), %xmm4	mulpd	%xmm13, %xmm8	movapd	 6 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm1	movapd	 6 * SIZE(AO1), %xmm8	mulpd	%xmm12, %xmm4	movapd	 8 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm2	movapd	 8 * SIZE(AO1), %xmm4	mulpd	%xmm13, %xmm8	movapd	10 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm3	movapd	10 * SIZE(AO1), %xmm8	mulpd	%xmm12, %xmm4	movapd	12 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	movapd	12 * SIZE(AO1), %xmm4	mulpd	%xmm13, %xmm8	movapd	14 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm1	movapd	14 * SIZE(AO1), %xmm8	mulpd	%xmm12, %xmm4	movapd	16 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm2	subq	$-16 * SIZE, AO1	mulpd	%xmm13, %xmm8	movapd	18 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm3	subq	$-16 * SIZE, BO.L44:	movq	MIN_M, I	andq	$8,  I	je	.L45	movapd	 0 * SIZE(AO1), %xmm4	movapd	 2 * SIZE(AO1), %xmm8	mulpd	%xmm12, %xmm4	movapd	 4 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	movapd	 4 * SIZE(AO1), %xmm4	mulpd	%xmm13, %xmm8	movapd	 6 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm1	movapd	 6 * SIZE(AO1), %xmm8	mulpd	%xmm12, %xmm4	movapd	 8 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm2	subq	$-8 * SIZE, AO1	mulpd	%xmm13, %xmm8	movapd	10 * SIZE(BO), %xmm13	addpd	%xmm8, %xmm3	subq	$-8 * SIZE, BO.L45:	movq	MIN_M, I	andq	$4,  I	je	.L46	movapd	 0 * SIZE(AO1), %xmm4	movapd	 2 * SIZE(AO1), %xmm8	subq	$-4 * SIZE, AO1	mulpd	%xmm12, %xmm4	movapd	 4 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	mulpd	%xmm13, %xmm8	subq	$-4 * SIZE, BO	addpd	%xmm8, %xmm1.L46:	movq	MIN_M, I	andq	$2,  I	je	.L47	movapd	 0 * SIZE(AO1), %xmm4	subq	$-2 * SIZE, AO1	mulpd	%xmm12, %xmm4	movapd	 2 * SIZE(BO), %xmm12	addpd	%xmm4, %xmm0	subq	$-2 * SIZE, BO.L47:	movq	MIN_M, I	andq	$1,  I	je	.L48	movsd	0 * SIZE(AO1), %xmm8	mulsd	%xmm12, %xmm8	addsd	%xmm8, %xmm0.L48:	addpd	%xmm2, %xmm0	addpd	%xmm3, %xmm1	addpd	%xmm1, %xmm0#ifdef HAVE_SSE3	haddpd	%xmm0, %xmm0#else	movapd	%xmm0, %xmm9	unpckhpd %xmm0, %xmm0	addsd	 %xmm9,  %xmm0#endif	mulsd	ALPHA, %xmm0	addsd	(CO), %xmm0	movsd	%xmm0, (CO)	ALIGN_3.L49:	addq	NLDA, A	addq	$P, IS	cmpq	M,  IS	jl	.L10	jmp	.L999	ALIGN_3.L50:	movq	$P,   TEMP	movq	M,    MIN_M	subq	IS,   MIN_M	cmpq	TEMP, MIN_M	cmovg	TEMP, MIN_M	movq	BUFFER, BO	movsd	(X), %xmm0	movsd	%xmm0, 0 * SIZE(BO)	addq	INCX, X	addq	$2 * SIZE, BO	leaq	-1(MIN_M), I	sarq	$3,    I	jle	.L55	ALIGN_3.L52:	movsd	(X), %xmm0	addq	INCX, X	movhpd	(X), %xmm0	addq	INCX, X	movsd	(X), %xmm2	addq	INCX, X	movhpd	(X), %xmm2	addq	INCX, X	movsd	(X), %xmm4	addq	INCX, X	movhpd	(X), %xmm4	addq	INCX, X	movsd	(X), %xmm6	addq	INCX, X	movhpd	(X), %xmm6	addq	INCX, X	movapd	%xmm0, 0 * SIZE(BO)	movapd	%xmm2, 2 * SIZE(BO)	movapd	%xmm4, 4 * SIZE(BO)	movapd	%xmm6, 6 * SIZE(BO)	addq	$8 * SIZE, BO	decq	I	jg	.L52	ALIGN_3.L55:	leaq	-1(MIN_M), I	andq	$7,    I	jle	.L60	ALIGN_2.L56:	movsd	(X), %xmm0	addq	INCX, X	movsd	%xmm0, 0 * SIZE(BO)	addq	$SIZE, BO	decq	I	jg	.L56	ALIGN_3.L60:	movq	Y, CO	movq	N,  J	sarq	$2, J	jle	.L70	ALIGN_3.L61:	movq	A, AO1	leaq	(A, LDA, 1), AO2	leaq	(A, LDA, 4), A	movq	BUFFER, BO	movapd	(BO), %xmm12	movsd	(AO1), %xmm0	mulsd	%xmm12, %xmm0	movsd	(AO2), %xmm1	mulsd	%xmm12, %xmm1	movsd	(AO1, LDA, 2), %xmm2	mulsd	%xmm12, %xmm2	movsd	(AO2, LDA, 2), %xmm3	mulsd	%xmm12, %xmm3	movapd	 2 * SIZE(BO), %xmm12	movapd	 4 * SIZE(BO), %xmm13	addq	$SIZE, AO1	addq	$SIZE, AO2	addq	$2 * SIZE, BO#if defined(CORE2) || defined(PENRYN)	PREFETCHW	4 * SIZE(CO)#endif	leaq	-1(MIN_M), I	sarq	$4,  I	jle	.L64	movapd	 0 * SIZE(AO1), %xmm4	movapd	 0 * SIZE(AO2), %xmm5	movapd	 0 * SIZE(AO1, LDA, 2), %xmm6	movapd	 0 * SIZE(AO2, LDA, 2), %xmm7	movapd	 2 * SIZE(AO1), %xmm8	movapd	 2 * SIZE(AO2), %xmm9	movapd	 2 * SIZE(AO1, LDA, 2), %xmm10	movapd	 2 * SIZE(AO2, LDA, 2), %xmm11	decq	I	jle	.L63	ALIGN_3.L62:	PREFETCH	PREFETCHSIZE * SIZE(AO1)	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	 4 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	addpd	%xmm5, %xmm1	movapd	 4 * SIZE(AO2), %xmm5	mulpd	%xmm12, %xmm6	addpd	%xmm6, %xmm2	movapd	 4 * SIZE(AO1, LDA, 2), %xmm6	mulpd	%xmm12, %xmm7	movapd	 4 * SIZE(BO), %xmm12	addpd	%xmm7, %xmm3	movapd	 4 * SIZE(AO2, LDA, 2), %xmm7#if defined(CORE2) || defined(PENRYN)	PREFETCH	(PREFETCHSIZE + 8) * SIZE(AO1)#endif	mulpd	%xmm13, %xmm8	addpd	%xmm8, %xmm0	movapd	 6 * SIZE(AO1), %xmm8	mulpd	%xmm13, %xmm9	addpd	%xmm9, %xmm1	movapd	 6 * SIZE(AO2), %xmm9	mulpd	%xmm13, %xmm10	addpd	%xmm10, %xmm2	movapd	 6 * SIZE(AO1, LDA, 2), %xmm10	mulpd	%xmm13, %xmm11	movapd	 6 * SIZE(BO), %xmm13	addpd	%xmm11, %xmm3	movapd	 6 * SIZE(AO2, LDA, 2), %xmm11	PREFETCH	PREFETCHSIZE * SIZE(AO2)	mulpd	%xmm12, %xmm4	addpd	%xmm4, %xmm0	movapd	 8 * SIZE(AO1), %xmm4	mulpd	%xmm12, %xmm5	addpd	%xmm5, %xmm1	movapd	 8 * SIZE(AO2), %xmm5	mulpd	%xmm12, %xmm6	addpd	%xmm6, %xmm2	movapd	 8 * SIZE(AO1, LDA, 2), %xmm6	mulpd	%xmm12, %xmm7	movapd	 8 * SIZE(BO), %xmm12
⌨️ 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 + -