📄 gemm_kernel_4x4_barcelona.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 3 页
字号:
	addpd	%xmm0, %xmm8	addpd	%xmm1, %xmm12#endif	movlpd	%xmm8,  (CO1)	movhpd	%xmm8,  1 * SIZE(CO1)	movlpd	%xmm12, 2 * SIZE(CO1)	movhpd	%xmm12, 3 * SIZE(CO1)#ifndef TRMMKERNEL	movupd	(CO2), %xmm2	movupd	2 * SIZE(CO2), %xmm3#endif	mulpd	%xmm7, %xmm9	mulpd	%xmm7, %xmm13#ifndef TRMMKERNEL	addpd	%xmm2, %xmm9	addpd	%xmm3, %xmm13#endif	movlpd	%xmm9,  (CO2)	movhpd	%xmm9,  1 * SIZE(CO2)	movlpd	%xmm13, 2 * SIZE(CO2)	movhpd	%xmm13, 3 * SIZE(CO2)#ifndef TRMMKERNEL	movupd	(CO1, LDC, 2), %xmm0	movupd	2 * SIZE(CO1, LDC, 2), %xmm1#endif	mulpd	%xmm7, %xmm10	mulpd	%xmm7, %xmm14#ifndef TRMMKERNEL	addpd	%xmm0, %xmm10	addpd	%xmm1, %xmm14#endif	movlpd	%xmm10, (CO1, LDC, 2)	movhpd	%xmm10, 1 * SIZE(CO1, LDC, 2)	movlpd	%xmm14, 2 * SIZE(CO1, LDC, 2)	movhpd	%xmm14, 3 * SIZE(CO1, LDC, 2)#ifndef TRMMKERNEL	movupd	(CO2, LDC, 2), %xmm2	movupd	2 * SIZE(CO2, LDC, 2), %xmm3#endif	mulpd	%xmm7, %xmm11	mulpd	%xmm7, %xmm15#ifndef TRMMKERNEL	addpd	%xmm2, %xmm11	addpd	%xmm3, %xmm15#endif	movlpd	%xmm11, (CO2, LDC, 2)	movhpd	%xmm11, 1 * SIZE(CO2, LDC, 2)	movlpd	%xmm15, 2 * SIZE(CO2, LDC, 2)	movhpd	%xmm15, 3 * SIZE(CO2, LDC, 2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax, SIZE), %rax	leaq	(AO, %rax, 4), AO	leaq	(BO, %rax, 4), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$4, KK#endif	addq	$4 * SIZE, CO1		# coffset += 4	addq	$4 * SIZE, CO2		# coffset += 4	decq	I			# i --	jg,pt	.L11	ALIGN_4	.L20:	testq	$3, M	je	.L39	testq	$2, M	je	.L30	ALIGN_4.L21:#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	B, BO#else	movq	KK, %rax	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 2), AO	leaq	(B,  %rax, 4), BO#endif		movapd	-16 * SIZE(AO), %xmm0	pxor	%xmm8, %xmm8	movapd	-12 * SIZE(AO), %xmm2	pxor	%xmm9, %xmm9	movddup	-16 * SIZE(BO), %xmm1	pxor	%xmm10, %xmm10	movddup	-15 * SIZE(BO), %xmm5	pxor	%xmm11, %xmm11	movddup	 -8 * SIZE(BO), %xmm3#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	$4, %rax#endif	movq	%rax, KKK#endif	andq	$-4, %rax	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 2), AO	leaq	(BO, %rax, 4), BO	negq	%rax	je,pn	.L26	ALIGN_4.L22:	mulpd	%xmm0, %xmm1	addpd	%xmm1, %xmm8	movddup	-14 * SIZE(BO, %rax, 4), %xmm1	mulpd	%xmm0, %xmm5	addpd	%xmm5, %xmm9	movddup	-13 * SIZE(BO, %rax, 4), %xmm5	mulpd	%xmm0, %xmm1	addpd	%xmm1, %xmm10	movddup	-12 * SIZE(BO, %rax, 4), %xmm1	mulpd	%xmm0, %xmm5	movapd	-14 * SIZE(AO, %rax, 2), %xmm0	addpd	%xmm5, %xmm11	movddup	-11 * SIZE(BO, %rax, 4), %xmm5	mulpd	%xmm0, %xmm1	addpd	%xmm1, %xmm8	movddup	-10 * SIZE(BO, %rax, 4), %xmm1	mulpd	%xmm0, %xmm5	addpd	%xmm5, %xmm9	movddup	 -9 * SIZE(BO, %rax, 4), %xmm5	mulpd	%xmm0, %xmm1	addpd	%xmm1, %xmm10	movddup	  (BO, %rax, 4), %xmm1	mulpd	%xmm0, %xmm5	movapd	 -8 * SIZE(AO, %rax, 2), %xmm0	addpd	%xmm5, %xmm11	movddup	 -7 * SIZE(BO, %rax, 4), %xmm5	mulpd	%xmm2, %xmm3	addpd	%xmm3, %xmm8	movddup	 -6 * SIZE(BO, %rax, 4), %xmm3	mulpd	%xmm2, %xmm5	addpd	%xmm5, %xmm9	movddup	 -5 * SIZE(BO, %rax, 4), %xmm5	mulpd	%xmm2, %xmm3	addpd	%xmm3, %xmm10	movddup	 -4 * SIZE(BO, %rax, 4), %xmm3	mulpd	%xmm2, %xmm5	movapd	-10 * SIZE(AO, %rax, 2), %xmm2	addpd	%xmm5, %xmm11	movddup	 -3 * SIZE(BO, %rax, 4), %xmm5	mulpd	%xmm2, %xmm3	addpd	%xmm3, %xmm8	movddup	 -2 * SIZE(BO, %rax, 4), %xmm3	mulpd	%xmm2, %xmm5	addpd	%xmm5, %xmm9	movddup	 -1 * SIZE(BO, %rax, 4), %xmm5	mulpd	%xmm2, %xmm3	addpd	%xmm3, %xmm10	movddup	  8 * SIZE(BO, %rax, 4), %xmm3	mulpd	%xmm2, %xmm5	movapd	 -4 * SIZE(AO, %rax, 2), %xmm2	addpd	%xmm5, %xmm11	movddup	  1 * SIZE(BO, %rax, 4), %xmm5	addq	$4 * SIZE, %rax	jl,pt	.L22	ALIGN_4.L26:	movddup	ALPHA, %xmm7#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	andq	$3, %rax		# if (k & 1)	je .L29	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 2), AO	leaq	(BO, %rax, 4), BO	negq	%rax	ALIGN_4.L27:	mulpd	%xmm0, %xmm1	addpd	%xmm1, %xmm8	movddup	-14 * SIZE(BO, %rax, 4), %xmm1	mulpd	%xmm0, %xmm5	addpd	%xmm5, %xmm9	movddup	-13 * SIZE(BO, %rax, 4), %xmm5	mulpd	%xmm0, %xmm1	addpd	%xmm1, %xmm10	movddup	-12 * SIZE(BO, %rax, 4), %xmm1	mulpd	%xmm0, %xmm5	movapd	-14 * SIZE(AO, %rax, 2), %xmm0	addpd	%xmm5, %xmm11	movddup	-11 * SIZE(BO, %rax, 4), %xmm5	addq	$SIZE, %rax	jl	.L27	ALIGN_4.L29:#ifndef TRMMKERNEL	movupd	(CO1), %xmm0	movupd	(CO2), %xmm2	movupd	(CO1, LDC, 2), %xmm4	movupd	(CO2, LDC, 2), %xmm6#endif	mulpd	%xmm7, %xmm8	mulpd	%xmm7, %xmm9	mulpd	%xmm7, %xmm10	mulpd	%xmm7, %xmm11#ifndef TRMMKERNEL	addpd	%xmm0, %xmm8	addpd	%xmm2, %xmm9	addpd	%xmm4, %xmm10	addpd	%xmm6, %xmm11#endif	movlpd	%xmm8,  (CO1)	movhpd	%xmm8,  1 * SIZE(CO1)	movlpd	%xmm9,  (CO2)	movhpd	%xmm9,  1 * SIZE(CO2)	movlpd	%xmm10, (CO1, LDC, 2)	movhpd	%xmm10, 1 * SIZE(CO1, LDC, 2)	movlpd	%xmm11, (CO2, LDC, 2)	movhpd	%xmm11, 1 * SIZE(CO2, LDC, 2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax, SIZE), %rax	leaq	(AO, %rax, 2), AO	leaq	(BO, %rax, 4), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$2, KK#endif	addq	$2 * SIZE, CO1	addq	$2 * SIZE, CO2	ALIGN_4	.L30:	testq	$1, M	je	.L39#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	B, BO#else	movq	KK, %rax	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 1), AO	leaq	(B,  %rax, 4), BO#endif		movddup	-16 * SIZE(AO), %xmm0	pxor	%xmm8, %xmm8	movddup	-14 * SIZE(AO), %xmm2	pxor	%xmm9, %xmm9	movddup	-15 * SIZE(AO), %xmm4	pxor	%xmm10, %xmm10	movapd	-16 * SIZE(BO), %xmm1	pxor	%xmm11, %xmm11	movapd	 -8 * SIZE(BO), %xmm3#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	$4, %rax#endif	movq	%rax, KKK#endif	andq	$-4, %rax	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 1), AO	leaq	(BO, %rax, 4), BO	negq	%rax	je,pn	.L36	ALIGN_4.L32:	mulpd	%xmm0, %xmm1	mulpd	-14 * SIZE(BO, %rax, 4), %xmm0	addpd	%xmm1, %xmm8	movapd	-12 * SIZE(BO, %rax, 4), %xmm1	addpd	%xmm0, %xmm9	movddup	-12 * SIZE(AO, %rax, 1), %xmm0	mulpd	%xmm4, %xmm1	mulpd	-10 * SIZE(BO, %rax, 4), %xmm4	addpd	%xmm1, %xmm10	movapd	  (BO, %rax, 4), %xmm1	addpd	%xmm4, %xmm11	movddup	-11 * SIZE(AO, %rax, 1), %xmm4	mulpd	%xmm2, %xmm3	mulpd	 -6 * SIZE(BO, %rax, 4), %xmm2	addpd	%xmm3, %xmm8	movapd	 -4 * SIZE(BO, %rax, 4), %xmm3	addpd	%xmm2, %xmm9	movddup	-13 * SIZE(AO, %rax, 1), %xmm2	mulpd	%xmm2, %xmm3	mulpd	 -2 * SIZE(BO, %rax, 4), %xmm2	addpd	%xmm3, %xmm10	movapd	  8 * SIZE(BO, %rax, 4), %xmm3	addpd	%xmm2, %xmm11	movddup	-10 * SIZE(AO, %rax, 1), %xmm2	addq	$4 * SIZE, %rax	jl,pt	.L32	ALIGN_4.L36:	movddup	ALPHA, %xmm7#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	andq	$3, %rax		# if (k & 1)	je .L38	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 1), AO	leaq	(BO, %rax, 4), BO	negq	%rax	ALIGN_4.L37:	mulpd	%xmm0, %xmm1	mulpd	-14 * SIZE(BO, %rax, 4), %xmm0	addpd	%xmm1, %xmm8	movapd	-12 * SIZE(BO, %rax, 4), %xmm1	addpd	%xmm0, %xmm9	movddup	-15 * SIZE(AO, %rax, 1), %xmm0	addq	$SIZE, %rax	jl	.L37	ALIGN_4.L38:	addpd	%xmm10, %xmm8	addpd	%xmm11, %xmm9#ifndef TRMMKERNEL	movlpd	(CO1), %xmm0	movhpd	(CO2), %xmm0	movlpd	(CO1, LDC, 2), %xmm1	movhpd	(CO2, LDC, 2), %xmm1#endif	mulpd	%xmm7, %xmm8	mulpd	%xmm7, %xmm9#ifndef TRMMKERNEL	addpd	%xmm0, %xmm8	addpd	%xmm1, %xmm9#endif	movlpd	%xmm8,  (CO1)	movhpd	%xmm8,  (CO2)	movlpd	%xmm9,  (CO1, LDC, 2)	movhpd	%xmm9,  (CO2, LDC, 2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax, SIZE), %rax	leaq	(AO, %rax, 1), AO	leaq	(BO, %rax, 4), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$1, KK#endif	ALIGN_4		.L39:#if defined(TRMMKERNEL) && !defined(LEFT)	addq	$4, KK#endif	movq	BO, B	leaq	(C, LDC, 4), C		# c += 4 * ldc	decq	J			# j --	jg	.L01	ALIGN_4.L40:	testq	$3, N	je	.L999	testq	$2, N	je	.L80	ALIGN_4.L41:#if defined(TRMMKERNEL) && defined(LEFT)	movq	OFFSET, %rax	movq	%rax, KK#endif		movq	C, CO1			# coffset1 = c	leaq	(C, LDC, 1), CO2	# coffset2 = c + ldc	movq	A, AO		# aoffset = a	movq	K, %rax	salq	$BASE_SHIFT + 1, %rax	leaq	(B, %rax), BB	movq	M,  I	sarq	$2, I	# i = (m >> 2)	jle	.L60	ALIGN_4.L51:#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	B, BO#else	movq	KK, %rax	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 4), AO	leaq	(B,  %rax, 2), BO#endif		movddup	-16 * SIZE(BO), %xmm1	movddup	-15 * SIZE(BO), %xmm5	pxor	%xmm8, %xmm8	movddup	-12 * SIZE(BO), %xmm3	pxor	%xmm9, %xmm9	movapd	-16 * SIZE(AO), %xmm0	pxor	%xmm12, %xmm12	movapd	 -8 * SIZE(AO), %xmm4	pxor	%xmm13, %xmm13	prefetchw      3 * SIZE(CO1)	movapd	%xmm0, %xmm2	prefetchw      3 * SIZE(CO2)	prefetch	 -16 * SIZE(BB)	subq		 $-8 * SIZE, BB#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	$4, %rax#else	addq	$2, %rax#endif	movq	%rax, KKK#endif	andq	$-4, %rax	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 4), AO	leaq	(BO, %rax, 2), BO	negq	%rax	je,pn	.L56	ALIGN_4.L52:	mulpd	%xmm1, %xmm0	mulpd	-14 * SIZE(AO, %rax, 4), %xmm1	addpd	%xmm0, %xmm8	movapd	-12 * SIZE(AO, %rax, 4), %xmm0	addpd	%xmm1, %xmm12	movddup	-14 * SIZE(BO, %rax, 2), %xmm1	mulpd	%xmm5, %xmm2	mulpd	-14 * SIZE(AO, %rax, 4), %xmm5	addpd	%xmm2, %xmm9	addpd	%xmm5, %xmm13	movddup	-13 * SIZE(BO, %rax, 2), %xmm5	movapd	%xmm0, %xmm2	mulpd	%xmm1, %xmm0	mulpd	-10 * SIZE(AO, %rax, 4), %xmm1	addpd	%xmm0, %xmm8	movapd	  (AO, %rax, 4), %xmm0	addpd	%xmm1, %xmm12	movddup	 -8 * SIZE(BO, %rax, 2), %xmm1	mulpd	%xmm5, %xmm2	mulpd	-10 * SIZE(AO, %rax, 4), %xmm5	addpd	%xmm2, %xmm9	addpd	%xmm5, %xmm13	movddup	-11 * SIZE(BO, %rax, 2), %xmm5	movapd	%xmm4, %xmm2	mulpd	%xmm3, %xmm4	mulpd	 -6 * SIZE(AO, %rax, 4), %xmm3	addpd	%xmm4, %xmm8	movapd	 -4 * SIZE(AO, %rax, 4), %xmm4	addpd	%xmm3, %xmm12	movddup	-10 * SIZE(BO, %rax, 2), %xmm3	mulpd	%xmm5, %xmm2	mulpd	 -6 * SIZE(AO, %rax, 4), %xmm5	addpd	%xmm2, %xmm9	addpd	%xmm5, %xmm13	movddup	 -9 * SIZE(BO, %rax, 2), %xmm5	movapd	%xmm4, %xmm2	mulpd	%xmm3, %xmm4	mulpd	 -2 * SIZE(AO, %rax, 4), %xmm3	addpd	%xmm4, %xmm8	movapd	  8 * SIZE(AO, %rax, 4), %xmm4	addpd	%xmm3, %xmm12	movddup	 -4 * SIZE(BO, %rax, 2), %xmm3	mulpd	%xmm5, %xmm2	mulpd	 -2 * SIZE(AO, %rax, 4), %xmm5	addpd	%xmm2, %xmm9	addpd	%xmm5, %xmm13	movddup	 -7 * SIZE(BO, %rax, 2), %xmm5	movapd	%xmm0, %xmm2	addq	$4 * SIZE, %rax	jl,pt	.L52	ALIGN_4.L56:	movddup	ALPHA, %xmm7#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	andq	$3, %rax		# if (k & 1)	je .L59	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 4), AO	leaq	(BO, %rax, 2), BO	negq	%rax	ALIGN_4.L57:	mulpd	%xmm1, %xmm0	mulpd	-14 * SIZE(AO, %rax, 4), %xmm1	addpd	%xmm0, %xmm8	movapd	-12 * SIZE(AO, %rax, 4), %xmm0	addpd	%xmm1, %xmm12	movddup	-14 * SIZE(BO, %rax, 2), %xmm1	mulpd	%xmm5, %xmm2	mulpd	-14 * SIZE(AO, %rax, 4), %xmm5	addpd	%xmm2, %xmm9	addpd	%xmm5, %xmm13	movddup	-13 * SIZE(BO, %rax, 2), %xmm5	movapd	%xmm0, %xmm2	addq	$SIZE, %rax	jl	.L57	ALIGN_4.L59:#ifndef TRMMKERNEL	movupd	(CO1), %xmm0	movupd	2 * SIZE(CO1), %xmm1	movupd	(CO2), %xmm2	movupd	2 * SIZE(CO2), %xmm3#endif	mulpd	%xmm7, %xmm8	mulpd	%xmm7, %xmm9	mulpd	%xmm7, %xmm12	mulpd	%xmm7, %xmm13#ifndef TRMMKERNEL	addpd	%xmm0, %xmm8	addpd	%xmm1, %xmm12	addpd	%xmm2, %xmm9	addpd	%xmm3, %xmm13#endif	movlpd	%xmm8,  (CO1)	movhpd	%xmm8,  1 * SIZE(CO1)	movlpd	%xmm12, 2 * SIZE(CO1)	movhpd	%xmm12, 3 * SIZE(CO1)	movlpd	%xmm9,  (CO2)	movhpd	%xmm9,  1 * SIZE(CO2)	movlpd	%xmm13, 2 * SIZE(CO2)	movhpd	%xmm13, 3 * SIZE(CO2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax, SIZE), %rax	leaq	(AO, %rax, 4), AO	leaq	(BO, %rax, 2), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$4, KK#endif	addq	$4 * SIZE, CO1		# coffset += 4	addq	$4 * SIZE, CO2		# coffset += 4	decq	I			# i --	jg	.L51	ALIGN_4	.L60:	testq	$2, M	je	.L70	ALIGN_4.L61:#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	B, BO#else	movq	KK, %rax	leaq	(, %rax, SIZE), %rax	leaq	(AO, %rax, 2), AO	leaq	(B,  %rax, 2), BO#endif		movapd	-16 * SIZE(AO), %xmm0	pxor	%xmm8, %xmm8	movapd	-12 * SIZE(AO), %xmm2	pxor	%xmm9, %xmm9	movddup	-16 * SIZE(BO), %xmm1	pxor	%xmm10, %xmm10	movddup	-15 * SIZE(BO), %xmm3	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
⌨️ 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 + -