📄 zgemm_kernel_4x2_sse.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 4 页
字号:
上一页 1 2 34
	movaps	 -8 * SIZE(BO), %xmm3	addps	%xmm2, %xmm9	movaps	 -4 * SIZE(AO), %xmm2	mulps	%xmm2, %xmm3	mulps	 -4 * SIZE(BO), %xmm2	addps	%xmm3, %xmm12	movaps	 48 * SIZE(BO), %xmm3	addps	%xmm2, %xmm13	movaps	 48 * SIZE(AO), %xmm2	PREFETCH	(PREFETCHSIZE + 32) * SIZE(AO)	mulps	%xmm4, %xmm5	mulps	  4 * SIZE(BO), %xmm4	addps	%xmm5, %xmm8	movaps	  0 * SIZE(BO), %xmm5	addps	%xmm4, %xmm9	movaps	  4 * SIZE(AO), %xmm4	mulps	%xmm4, %xmm5	mulps	  4 * SIZE(BO), %xmm4	addps	%xmm5, %xmm12	movaps	  8 * SIZE(BO), %xmm5	addps	%xmm4, %xmm13	movaps	  8 * SIZE(AO), %xmm4	mulps	%xmm4, %xmm5	mulps	 12 * SIZE(BO), %xmm4	addps	%xmm5, %xmm8	movaps	  8 * SIZE(BO), %xmm5	addps	%xmm4, %xmm9	movaps	 12 * SIZE(AO), %xmm4	mulps	%xmm4, %xmm5	mulps	 12 * SIZE(BO), %xmm4	addps	%xmm5, %xmm12	movaps	 64 * SIZE(BO), %xmm5	addps	%xmm4, %xmm13	movaps	 64 * SIZE(AO), %xmm4	PREFETCH	(PREFETCHSIZE + 48) * SIZE(AO)	mulps	%xmm6, %xmm7	mulps	 20 * SIZE(BO), %xmm6	addps	%xmm7, %xmm8	movaps	 16 * SIZE(BO), %xmm7	addps	%xmm6, %xmm9	movaps	 20 * SIZE(AO), %xmm6	mulps	%xmm6, %xmm7	mulps	 20 * SIZE(BO), %xmm6	addps	%xmm7, %xmm12	movaps	 24 * SIZE(BO), %xmm7	addps	%xmm6, %xmm13	movaps	 24 * SIZE(AO), %xmm6	mulps	%xmm6, %xmm7	mulps	 28 * SIZE(BO), %xmm6	addps	%xmm7, %xmm8	movaps	 24 * SIZE(BO), %xmm7	addps	%xmm6, %xmm9	movaps	 28 * SIZE(AO), %xmm6	mulps	%xmm6, %xmm7	mulps	 28 * SIZE(BO), %xmm6	addps	%xmm7, %xmm12	movaps	 80 * SIZE(BO), %xmm7	addps	%xmm6, %xmm13	movaps	 80 * SIZE(AO), %xmm6	subq   $-64 * SIZE, AO	subq   $-64 * SIZE, BO	decq   %rax	jne    .L52	ALIGN_4.L55:#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	movaps	ALPHA_R, %xmm6	movaps	ALPHA_I, %xmm7	andq	$7, %rax		# if (k & 1)	BRANCH	je .L58	ALIGN_4.L56:	mulps	%xmm0, %xmm1	mulps	-28 * SIZE(BO), %xmm0	addps	%xmm1, %xmm8	movaps	-32 * SIZE(BO), %xmm1	addps	%xmm0, %xmm9	movaps	-28 * SIZE(AO), %xmm0	mulps	%xmm0, %xmm1	mulps	-28 * SIZE(BO), %xmm0	addps	%xmm1, %xmm12	movaps	-24 * SIZE(BO), %xmm1	addps	%xmm0, %xmm13	movaps	-24 * SIZE(AO), %xmm0	addq	$ 8 * SIZE, AO	addq	$ 8 * SIZE, BO	decq	%rax	jg	.L56	ALIGN_4.L58:#ifndef TRMMKERNEL	movsd	0 * SIZE(CO1), %xmm0	movhps	2 * SIZE(CO1), %xmm0	movsd	4 * SIZE(CO1), %xmm2	movhps	6 * SIZE(CO1), %xmm2#endif	shufps	$0xb1, %xmm9, %xmm9	shufps	$0xb1, %xmm13, %xmm13#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(RR) || defined(RC) || defined(CR) || defined(CC)	subps	%xmm9, %xmm8	subps	%xmm13, %xmm12#else	addps	%xmm9, %xmm8	addps	%xmm13, %xmm12#endif	movaps	%xmm8, %xmm9	movaps	%xmm12, %xmm13	shufps	$0xb1, %xmm8, %xmm8	shufps	$0xb1, %xmm12, %xmm12	mulps	%xmm6, %xmm9	mulps	%xmm7, %xmm8	mulps	%xmm6, %xmm13	mulps	%xmm7, %xmm12	addps	%xmm9, %xmm8	addps	%xmm13, %xmm12#ifndef TRMMKERNEL	addps	%xmm0,  %xmm8	addps	%xmm2, %xmm12#endif	movsd	%xmm8, 0 * SIZE(CO1)	movhps	%xmm8, 2 * SIZE(CO1)	movsd	%xmm12, 4 * SIZE(CO1)	movhps	%xmm12, 6 * SIZE(CO1)#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, 4), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$4, KK#endif	addq	$8 * SIZE, CO1		# coffset += 4	decq	I			# i --	jg	.L51	ALIGN_4	.L60:	testq	$2, M	je	.L70#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, 4), BO#endif		movaps	-32 * SIZE(AO), %xmm0	pxor	%xmm8, %xmm8	movaps	-16 * SIZE(AO), %xmm2	pxor	%xmm9, %xmm9	movaps	-32 * SIZE(BO), %xmm1	pxor	%xmm10, %xmm10	movaps	-16 * SIZE(BO), %xmm3	pxor	%xmm11, %xmm11	movaps	  0 * SIZE(BO), %xmm5	movaps	 16 * SIZE(BO), %xmm7#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	$1, %rax#endif	movq	%rax, KKK#endif	sarq	$3, %rax	je	.L65	ALIGN_4.L62:	mulps	%xmm0, %xmm1	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)	mulps	-28 * SIZE(BO), %xmm0	addps	%xmm1, %xmm8	movaps	-24 * SIZE(BO), %xmm1	addps	%xmm0, %xmm9	movaps  -28 * SIZE(AO), %xmm0	mulps	%xmm0, %xmm1	mulps	-20 * SIZE(BO), %xmm0	addps	%xmm1, %xmm10	movaps	 32 * SIZE(BO), %xmm1	addps	%xmm0, %xmm11	movaps  -24 * SIZE(AO), %xmm0	mulps	%xmm0, %xmm3	mulps	-12 * SIZE(BO), %xmm0	addps	%xmm3, %xmm8	movaps	 -8 * SIZE(BO), %xmm3	addps	%xmm0, %xmm9	movaps  -20 * SIZE(AO), %xmm0	mulps	%xmm0, %xmm3	mulps	 -4 * SIZE(BO), %xmm0	addps	%xmm3, %xmm10	movaps	 48 * SIZE(BO), %xmm3	addps	%xmm0, %xmm11	movaps    0 * SIZE(AO), %xmm0	PREFETCH	(PREFETCHSIZE + 16) * SIZE(AO)	mulps	%xmm2, %xmm5	mulps	  4 * SIZE(BO), %xmm2	addps	%xmm5, %xmm8	movaps	  8 * SIZE(BO), %xmm5	addps	%xmm2, %xmm9	movaps  -12 * SIZE(AO), %xmm2	mulps	%xmm2, %xmm5	mulps	 12 * SIZE(BO), %xmm2	addps	%xmm5, %xmm10	movaps	 64 * SIZE(BO), %xmm5	addps	%xmm2, %xmm11	movaps   -8 * SIZE(AO), %xmm2	mulps	%xmm2, %xmm7	mulps	 20 * SIZE(BO), %xmm2	addps	%xmm7, %xmm8	movaps	 24 * SIZE(BO), %xmm7	addps	%xmm2, %xmm9	movaps   -4 * SIZE(AO), %xmm2	mulps	%xmm2, %xmm7	mulps	 28 * SIZE(BO), %xmm2	addps	%xmm7, %xmm10	movaps	 80 * SIZE(BO), %xmm7	addps	%xmm2, %xmm11	movaps   16 * SIZE(AO), %xmm2	subq   $-32 * SIZE, AO	subq   $-64 * SIZE, BO	decq   %rax	jne    .L62	ALIGN_4.L65:#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	movaps	ALPHA_R, %xmm6	movaps	ALPHA_I, %xmm7	andq	$7, %rax		# if (k & 1)	BRANCH	je .L68	ALIGN_4.L66:	mulps	%xmm0, %xmm1	mulps	-28 * SIZE(BO), %xmm0	addps	%xmm1, %xmm8	movaps	-24 * SIZE(BO), %xmm1	addps	%xmm0, %xmm9	movaps  -28 * SIZE(AO), %xmm0	addq	$4 * SIZE, AO		# aoffset  += 4	addq	$8 * SIZE, BO		# boffset1 += 8	decq	%rax	jg	.L66	ALIGN_4.L68:#ifndef TRMMKERNEL	movsd	0 * SIZE(CO1), %xmm0	movhps	2 * SIZE(CO1), %xmm0#endif	addps	%xmm10, %xmm8	addps	%xmm11, %xmm9	shufps	$0xb1, %xmm9, %xmm9#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(RR) || defined(RC) || defined(CR) || defined(CC)	subps	%xmm9, %xmm8#else	addps	%xmm9, %xmm8#endif	movaps	%xmm8, %xmm9	shufps	$0xb1, %xmm8, %xmm8	mulps	%xmm6, %xmm9	mulps	%xmm7, %xmm8	addps	%xmm9, %xmm8#ifndef TRMMKERNEL	addps	%xmm0,  %xmm8#endif	movsd	%xmm8, 0 * SIZE(CO1)	movhps	%xmm8, 2 * SIZE(CO1)#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, 4), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$2, KK#endif	addq	$4 * SIZE, CO1		# coffset += 4	ALIGN_4		.L70:	testq	$1, M	je	.L999#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, 4), BO#endif		movaps	-32 * SIZE(AO), %xmm0	pxor	%xmm8, %xmm8	movaps	-24 * SIZE(AO), %xmm2	pxor	%xmm9, %xmm9	movaps	-32 * SIZE(BO), %xmm1	pxor	%xmm10, %xmm10	movaps	-16 * SIZE(BO), %xmm3	pxor	%xmm11, %xmm11	movaps	  0 * SIZE(BO), %xmm5	movaps	 16 * SIZE(BO), %xmm7#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	$1, %rax#endif	movq	%rax, KKK#endif	sarq	$3, %rax	je	.L75	ALIGN_4.L72:	mulps	%xmm0, %xmm1	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)	addps	%xmm1, %xmm8	movsd	-28 * SIZE(BO), %xmm1	mulps	%xmm0, %xmm1	movsd	-30 * SIZE(AO), %xmm0	addps	%xmm1, %xmm9	movsd	-24 * SIZE(BO), %xmm1	mulps	%xmm0, %xmm1	addps	%xmm1, %xmm10	movsd	-20 * SIZE(BO), %xmm1	mulps	%xmm0, %xmm1	movsd	-28 * SIZE(AO), %xmm0	addps	%xmm1, %xmm11	movsd	 32 * SIZE(BO), %xmm1	mulps	%xmm0, %xmm3	addps	%xmm3, %xmm8	movsd	-12 * SIZE(BO), %xmm3	mulps	%xmm0, %xmm3	movsd	-26 * SIZE(AO), %xmm0	addps	%xmm3, %xmm9	movsd	 -8 * SIZE(BO), %xmm3	mulps	%xmm0, %xmm3	addps	%xmm3, %xmm10	movsd	 -4 * SIZE(BO), %xmm3	mulps	%xmm0, %xmm3	movsd	-16 * SIZE(AO), %xmm0	addps	%xmm3, %xmm11	movsd	 48 * SIZE(BO), %xmm3	mulps	%xmm2, %xmm5	addps	%xmm5, %xmm8	movsd	  4 * SIZE(BO), %xmm5	mulps	%xmm2, %xmm5	movsd	-22 * SIZE(AO), %xmm2	addps	%xmm5, %xmm9	movsd	  8 * SIZE(BO), %xmm5	mulps	%xmm2, %xmm5	addps	%xmm5, %xmm10	movsd	 12 * SIZE(BO), %xmm5	mulps	%xmm2, %xmm5	movsd	-20 * SIZE(AO), %xmm2	addps	%xmm5, %xmm11	movsd	 64 * SIZE(BO), %xmm5	mulps	%xmm2, %xmm7	addps	%xmm7, %xmm8	movsd	 20 * SIZE(BO), %xmm7	mulps	%xmm2, %xmm7	movsd	-18 * SIZE(AO), %xmm2	addps	%xmm7, %xmm9	movsd	 24 * SIZE(BO), %xmm7	mulps	%xmm2, %xmm7	addps	%xmm7, %xmm10	movsd	 28 * SIZE(BO), %xmm7	mulps	%xmm2, %xmm7	movsd	 -8 * SIZE(AO), %xmm2	addps	%xmm7, %xmm11	movsd	 80 * SIZE(BO), %xmm7	subq   $-16 * SIZE, AO	subq   $-64 * SIZE, BO	decq   %rax	jne    .L72	ALIGN_4.L75:#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	movaps	ALPHA_R, %xmm6	movaps	ALPHA_I, %xmm7	andq	$7, %rax		# if (k & 1)	BRANCH	je .L78	ALIGN_4.L76:	mulps	%xmm0, %xmm1	addps	%xmm1, %xmm8	movsd	-28 * SIZE(BO), %xmm1	mulps	%xmm0, %xmm1	movsd	-30 * SIZE(AO), %xmm0	addps	%xmm1, %xmm9	movsd	-24 * SIZE(BO), %xmm1	addq	$2 * SIZE, AO	addq	$8 * SIZE, BO	decq	%rax	jg	.L76	ALIGN_4.L78:#ifndef TRMMKERNEL	movsd	0 * SIZE(CO1), %xmm0#endif	addps	%xmm10, %xmm8	addps	%xmm11, %xmm9	shufps	$0xb1, %xmm9, %xmm9#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(RR) || defined(RC) || defined(CR) || defined(CC)	subps	%xmm9, %xmm8#else	addps	%xmm9, %xmm8#endif	movaps	%xmm8, %xmm9	shufps	$0xb1, %xmm8, %xmm8	mulps	%xmm6, %xmm9	mulps	%xmm7, %xmm8	addps	%xmm9, %xmm8#ifndef TRMMKERNEL	addps	%xmm0,  %xmm8#endif	movsd	%xmm8, 0 * SIZE(CO1)	ALIGN_4	.L999:	movq	%rbx, %rsp	EMMS	movq	  0(%rsp), %rbx	movq	  8(%rsp), %rbp	movq	 16(%rsp), %r12	movq	 24(%rsp), %r13	movq	 32(%rsp), %r14	movq	 40(%rsp), %r15#ifdef WINDOWS_ABI	movq	 48(%rsp), %rdi	movq	 56(%rsp), %rsi	movups	 64(%rsp), %xmm6	movups	 80(%rsp), %xmm7	movups	 96(%rsp), %xmm8	movups	112(%rsp), %xmm9	movups	128(%rsp), %xmm10	movups	144(%rsp), %xmm11	movups	160(%rsp), %xmm12	movups	176(%rsp), %xmm13	movups	192(%rsp), %xmm14	movups	208(%rsp), %xmm15#endif	addq	$STACKSIZE, %rsp	ret	EPILOGUE
上一页 1 2 34
⌨️ 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 + -