📄 zgemm_kernel_4x2_core2.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 3 页
字号:
	mulps	%xmm7, %xmm10	mulps	%xmm6, %xmm13	mulps	%xmm7, %xmm12	mulps	%xmm6, %xmm15	mulps	%xmm7, %xmm14	addps	%xmm9, %xmm8	addps	%xmm11, %xmm10	addps	%xmm13, %xmm12	addps	%xmm15, %xmm14#if! defined(TRMMKERNEL) && !defined(BETAZERO)	movsd	0 * SIZE(CO1), %xmm0	movhps	2 * SIZE(CO1), %xmm0	movsd	4 * SIZE(CO1), %xmm2	movhps	6 * SIZE(CO1), %xmm2	movsd	0 * SIZE(CO2), %xmm1	movhps	2 * SIZE(CO2), %xmm1	movsd	4 * SIZE(CO2), %xmm3	movhps	6 * SIZE(CO2), %xmm3	addps	%xmm0,  %xmm8	addps	%xmm1,  %xmm10	addps	%xmm2, %xmm12	addps	%xmm3, %xmm14#endif	movsd	%xmm8, 0 * SIZE(CO1)	movhps	%xmm8, 2 * SIZE(CO1)	movsd	%xmm12, 4 * SIZE(CO1)	movhps	%xmm12, 6 * SIZE(CO1)	movsd	%xmm10, 0 * SIZE(CO2)	movhps	%xmm10, 2 * SIZE(CO2)	movsd	%xmm14, 4 * SIZE(CO2)	movhps	%xmm14, 6 * SIZE(CO2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax, 8), %rax	leaq	(AO, %rax, 4), AO	leaq	(BO, %rax, 8), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$4, KK#endif	addq	$8 * SIZE, CO1		# coffset += 4	addq	$8 * SIZE, CO2		# coffset += 4	decq	I			# i --	jg	.L11	ALIGN_4	.L20:	testq	$2, M	je	.L30#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leaq	32 * SIZE + BUFFER, BO#else	leaq	32 * SIZE + BUFFER, BO	movq	KK, %rax	leaq	(, %rax,   8), %rax	leaq	(AO, %rax, 2), AO	leaq	(BO, %rax, 8), BO#endif		pxor	%xmm8, %xmm8	pxor	%xmm9, %xmm9	pxor	%xmm10, %xmm10	pxor	%xmm11, %xmm11#ifndef TRMMKERNEL	movq	K, %rax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movq	K, %rax	subq	KK, %rax	movq	%rax, KKK	#else	movq	KK, %rax#ifdef LEFT	addq	$2, %rax#else	addq	$2, %rax#endif	movq	%rax, KKK#endif	sarq	$2, %rax	je	.L25	ALIGN_4.L22:	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)	movaps	-32 * SIZE(AO), %xmm0	movaps	-32 * SIZE(BO), %xmm2	movaps	-28 * SIZE(BO), %xmm3	movaps	-24 * SIZE(BO), %xmm4	movaps	-20 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	movaps	-28 * SIZE(AO), %xmm0	movaps	-16 * SIZE(BO), %xmm2	movaps	-12 * SIZE(BO), %xmm3	movaps	 -8 * SIZE(BO), %xmm4	movaps	 -4 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	movaps	-24 * SIZE(AO), %xmm0	movaps	  0 * SIZE(BO), %xmm2	movaps	  4 * SIZE(BO), %xmm3	movaps	  8 * SIZE(BO), %xmm4	movaps	 12 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	movaps	-20 * SIZE(AO), %xmm0	movaps	 16 * SIZE(BO), %xmm2	movaps	 20 * SIZE(BO), %xmm3	movaps	 24 * SIZE(BO), %xmm4	movaps	 28 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	subq	$-16 * SIZE, AO	subq	$-64 * SIZE, BO	subq	$1, %rax	jne    .L22	ALIGN_4.L25:#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	andq	$3, %rax		# if (k & 1)	BRANCH	je .L28	ALIGN_4.L26:	movaps	-32 * SIZE(AO), %xmm0	movaps	-32 * SIZE(BO), %xmm2	movaps	-28 * SIZE(BO), %xmm3	movaps	-24 * SIZE(BO), %xmm4	movaps	-20 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	addq	$ 4 * SIZE, AO	addq	$16 * SIZE, BO	subq	$1, %rax	jg	.L26	ALIGN_4.L28:	movaps	ALPHA_R, %xmm6	movaps	ALPHA_I, %xmm7#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(NR) || defined(NC) || defined(TR) || defined(TC)	shufps	$0xb1, %xmm9, %xmm9	shufps	$0xb1, %xmm11, %xmm11	addsubps	%xmm9, %xmm8	addsubps	%xmm11, %xmm10	movaps	%xmm8, %xmm9	movaps	%xmm10, %xmm11	shufps	$0xb1, %xmm8, %xmm8	shufps	$0xb1, %xmm10, %xmm10#else	shufps	$0xb1, %xmm8, %xmm8	shufps	$0xb1, %xmm10, %xmm10	addsubps	%xmm8, %xmm9	addsubps	%xmm10, %xmm11	movaps	%xmm9, %xmm8	movaps	%xmm11, %xmm10	shufps	$0xb1, %xmm9, %xmm9	shufps	$0xb1, %xmm11, %xmm11#endif	mulps	%xmm6, %xmm9	mulps	%xmm7, %xmm8	mulps	%xmm6, %xmm11	mulps	%xmm7, %xmm10	addps	%xmm9, %xmm8	addps	%xmm11, %xmm10#if! defined(TRMMKERNEL) && !defined(BETAZERO)	movsd	0 * SIZE(CO1), %xmm0	movhps	2 * SIZE(CO1), %xmm0	movsd	0 * SIZE(CO2), %xmm2	movhps	2 * SIZE(CO2), %xmm2	addps	%xmm0,  %xmm8	addps	%xmm2, %xmm10#endif	movsd	%xmm8, 0 * SIZE(CO1)	movhps	%xmm8, 2 * SIZE(CO1)	movsd	%xmm10, 0 * SIZE(CO2)	movhps	%xmm10, 2 * SIZE(CO2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax,    8), %rax	leaq	(AO, %rax, 2), AO	leaq	(BO, %rax, 8), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$2, KK#endif	addq	$4 * SIZE, CO1		# coffset += 4	addq	$4 * SIZE, CO2		# coffset += 4	ALIGN_4		.L30:	testq	$1, M	je	.L39#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leaq	32 * SIZE + BUFFER, BO#else	leaq	32 * SIZE + BUFFER, BO	movq	KK, %rax	leaq	(, %rax,   8), %rax	leaq	(AO, %rax, 1), AO	leaq	(BO, %rax, 8), BO#endif		pxor	%xmm8, %xmm8	pxor	%xmm9, %xmm9	pxor	%xmm10, %xmm10	pxor	%xmm11, %xmm11#ifndef TRMMKERNEL	movq	K, %rax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movq	K, %rax	subq	KK, %rax	movq	%rax, KKK	#else	movq	KK, %rax#ifdef LEFT	addq	$1, %rax#else	addq	$2, %rax#endif	movq	%rax, KKK#endif	sarq	$2, %rax	je	.L35	ALIGN_4.L32:	PREFETCH (PREFETCHSIZE +  0) * SIZE(AO)	movsd	-32 * SIZE(AO), %xmm0	movsd	-32 * SIZE(BO), %xmm2	movsd	-28 * SIZE(BO), %xmm3	movsd	-24 * SIZE(BO), %xmm4	movsd	-20 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	movsd	-30 * SIZE(AO), %xmm0	movsd	-16 * SIZE(BO), %xmm2	movsd	-12 * SIZE(BO), %xmm3	movsd	 -8 * SIZE(BO), %xmm4	movsd	 -4 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	movsd	-28 * SIZE(AO), %xmm0	movsd	  0 * SIZE(BO), %xmm2	movsd	  4 * SIZE(BO), %xmm3	movsd	  8 * SIZE(BO), %xmm4	movsd	 12 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	movsd	-26 * SIZE(AO), %xmm0	movsd	 16 * SIZE(BO), %xmm2	movsd	 20 * SIZE(BO), %xmm3	movsd	 24 * SIZE(BO), %xmm4	movsd	 28 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	subq	$ -8 * SIZE, AO	subq	$-64 * SIZE, BO	subq	$1, %rax	jne    .L32	ALIGN_4.L35:#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	andq	$3, %rax		# if (k & 1)	BRANCH	je .L38	ALIGN_4.L36:	movsd	-32 * SIZE(AO), %xmm0	movsd	-32 * SIZE(BO), %xmm2	movsd	-28 * SIZE(BO), %xmm3	movsd	-24 * SIZE(BO), %xmm4	movsd	-20 * SIZE(BO), %xmm5	mulps	%xmm0, %xmm2	mulps	%xmm0, %xmm3	mulps	%xmm0, %xmm4	mulps	%xmm0, %xmm5	addps	%xmm2, %xmm8	ADDSUB	%xmm3, %xmm9	addps	%xmm4, %xmm10	ADDSUB	%xmm5, %xmm11	addq	$ 2 * SIZE, AO	addq	$16 * SIZE, BO	subq	$1, %rax	jg	.L36	ALIGN_4.L38:	movaps	ALPHA_R, %xmm6	movaps	ALPHA_I, %xmm7#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(NR) || defined(NC) || defined(TR) || defined(TC)	shufps	$0xb1, %xmm9, %xmm9	shufps	$0xb1, %xmm11, %xmm11	addsubps	%xmm9, %xmm8	addsubps	%xmm11, %xmm10	movaps	%xmm8, %xmm9	movaps	%xmm10, %xmm11	shufps	$0xb1, %xmm8, %xmm8	shufps	$0xb1, %xmm10, %xmm10#else	shufps	$0xb1, %xmm8, %xmm8	shufps	$0xb1, %xmm10, %xmm10	addsubps	%xmm8, %xmm9	addsubps	%xmm10, %xmm11	movaps	%xmm9, %xmm8	movaps	%xmm11, %xmm10	shufps	$0xb1, %xmm9, %xmm9	shufps	$0xb1, %xmm11, %xmm11#endif	mulps	%xmm6, %xmm9	mulps	%xmm7, %xmm8	mulps	%xmm6, %xmm11	mulps	%xmm7, %xmm10	addps	%xmm9, %xmm8	addps	%xmm11, %xmm10#if! defined(TRMMKERNEL) && !defined(BETAZERO)	movsd	0 * SIZE(CO1), %xmm0	movsd	0 * SIZE(CO2), %xmm2	addps	%xmm0,  %xmm8	addps	%xmm2, %xmm10#endif	movsd	%xmm8, 0 * SIZE(CO1)	movsd	%xmm10, 0 * SIZE(CO2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax,    8), %rax	leaq	(AO, %rax, 1), AO	leaq	(BO, %rax, 8), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$1, KK#endif	ALIGN_4	.L39:#if defined(TRMMKERNEL) && !defined(LEFT)	addl	$2, KK#endif	leaq	(C, LDC, 2), C		# c += 2 * ldc	decq	J			# j --	jg	.L01	ALIGN_4.L40:	testq	$1, N	je	.L999	ALIGN_4.L41:#if defined(TRMMKERNEL) && defined(LEFT)	movq	OFFSET, %rax	movq	%rax, KK#endif	/* Copying to Sub Buffer */	leaq	BUFFER, BO		movq	K, %rax	sarq	$2, %rax	jle	.L43	ALIGN_4	.L42:	movss	 -32 * SIZE(B), %xmm8	movss	 -31 * SIZE(B), %xmm9	movss	 -30 * SIZE(B), %xmm10	movss	 -29 * SIZE(B), %xmm11	movss	 -28 * SIZE(B), %xmm12	movss	 -27 * SIZE(B), %xmm13	movss	 -26 * SIZE(B), %xmm14	movss	 -25 * SIZE(B), %xmm15	shufps	$0, %xmm8,  %xmm8	shufps	$0, %xmm9,  %xmm9	shufps	$0, %xmm10,  %xmm10	shufps	$0, %xmm11,  %xmm11	shufps	$0, %xmm12,  %xmm12	shufps	$0, %xmm13,  %xmm13	shufps	$0, %xmm14,  %xmm14	shufps	$0, %xmm15,  %xmm15	movaps	%xmm8,  0 * SIZE(BO)	movaps	%xmm9,  4 * SIZE(BO)	movaps	%xmm10,  8 * SIZE(BO)	movaps	%xmm11, 12 * SIZE(BO)	movaps	%xmm12, 16 * SIZE(BO)	movaps	%xmm13, 20 * SIZE(BO)	movaps	%xmm14, 24 * SIZE(BO)	movaps	%xmm15, 28 * SIZE(BO)	addq	$ 8 * SIZE, B	addq	$32 * SIZE, BO	subq	$1, %rax	jne	.L42	ALIGN_4.L43:	movq	K, %rax	andq	$3, %rax	BRANCH	jle	.L50	ALIGN_4.L44:	movss	 -32 * SIZE(B), %xmm8	movss	 -31 * SIZE(B), %xmm9	shufps	$0, %xmm8,  %xmm8	shufps	$0, %xmm9,  %xmm9	movaps	%xmm8,  0 * SIZE(BO)	movaps	%xmm9,  4 * SIZE(BO)	addq	$2 * SIZE, B	addq	$8 * SIZE, BO	subq	$1, %rax	jne	.L44	ALIGN_4
⌨️ 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 + -