📄 gemm_kernel_8x2_sse.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 4 页
字号:
上一页 1 2 34
	leal	BUFFER, BB	movl	KK, %eax	leal	(, %eax,   8), %eax	leal	(AA, %eax, 2), AA	leal	(BB, %eax, 2), BB /* because it's doubled */	movaps	 0 * SIZE(BB), %xmm2	XORPS	%xmm4, %xmm4	movaps	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	movaps	 16 * SIZE(BB), %xmm3	XORPS	%xmm6, %xmm6	movaps	 16 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#endif	#ifndef TRMMKERNEL	movl	K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movl	K, %eax	subl	KK, %eax	movl	%eax, KKK	#else	movl	KK, %eax#ifdef LEFT	addl	$4, %eax#else	addl	$1, %eax#endif	movl	%eax, KKK#endif	sarl	$3, %eax	je	.L132	ALIGN_2.L131:	mulps	%xmm0, %xmm2	movaps	 4 * SIZE(AA), %xmm0	addps	%xmm2, %xmm4	mulps	 4 * SIZE(BB), %xmm0	movaps	32 * SIZE(BB), %xmm2	addps	%xmm0, %xmm5	movaps	 8 * SIZE(AA), %xmm0	mulps	 8 * SIZE(BB), %xmm0	addps	%xmm0, %xmm6	movaps	12 * SIZE(AA), %xmm0	mulps	12 * SIZE(BB), %xmm0	addps	%xmm0, %xmm7	movaps	32 * SIZE(AA), %xmm0	mulps	%xmm1, %xmm3	movaps	20 * SIZE(AA), %xmm1	addps	%xmm3, %xmm4	mulps	20 * SIZE(BB), %xmm1	movaps	48 * SIZE(BB), %xmm3	addps	%xmm1, %xmm5	movaps	24 * SIZE(AA), %xmm1	mulps	24 * SIZE(BB), %xmm1	addps	%xmm1, %xmm6	movaps	28 * SIZE(AA), %xmm1	mulps	28 * SIZE(BB), %xmm1	addps	%xmm1, %xmm7	movaps	48 * SIZE(AA), %xmm1	addl   $32 * SIZE, AA	addl   $32 * SIZE, BB	decl   %eax	jne    .L131	ALIGN_2#else#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leal	BUFFER, BB	movaps	 0 * SIZE + BUFFER, %xmm2	XORPS	%xmm4, %xmm4	movaps	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	movaps	 8 * SIZE + BUFFER, %xmm3	XORPS	%xmm6, %xmm6	movaps	 8 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#else	leal	BUFFER, BB	movl	KK, %eax	leal	(, %eax,   8), %eax	leal	(AA, %eax, 2), AA	leal	(BB, %eax, 2), BB /* because it's doubled */	movaps	 0 * SIZE(BB), %xmm2	XORPS	%xmm4, %xmm4	movaps	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	movaps	 8 * SIZE(BB), %xmm3	XORPS	%xmm6, %xmm6	movaps	 8 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#endif	#ifndef TRMMKERNEL	movl	K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movl	K, %eax	subl	KK, %eax	movl	%eax, KKK	#else	movl	KK, %eax#ifdef LEFT	addl	$4, %eax#else	addl	$1, %eax#endif	movl	%eax, KKK#endif	sarl	$3, %eax	je	.L132	ALIGN_2.L131:	mulps	%xmm0, %xmm2	movaps	 4 * SIZE(AA), %xmm0	addps	%xmm2, %xmm4	mulps	 4 * SIZE(BB), %xmm0	movaps	16 * SIZE(BB), %xmm2	addps	%xmm0, %xmm5	movaps	16 * SIZE(AA), %xmm0	mulps	%xmm1, %xmm3	movaps	12 * SIZE(AA), %xmm1	addps	%xmm3, %xmm6	mulps	12 * SIZE(BB), %xmm1	movaps	24 * SIZE(BB), %xmm3	addps	%xmm1, %xmm7	movaps	24 * SIZE(AA), %xmm1	mulps	%xmm0, %xmm2	movaps	20 * SIZE(AA), %xmm0	addps	%xmm2, %xmm4	mulps	20 * SIZE(BB), %xmm0	movaps	32 * SIZE(BB), %xmm2	addps	%xmm0, %xmm5	movaps	32 * SIZE(AA), %xmm0	mulps	%xmm1, %xmm3	movaps	28 * SIZE(AA), %xmm1	addps	%xmm3, %xmm6	mulps	28 * SIZE(BB), %xmm1	movaps	40 * SIZE(BB), %xmm3	addps	%xmm1, %xmm7	movaps	40 * SIZE(AA), %xmm1	addl   $32 * SIZE, AA	addl   $32 * SIZE, BB	decl   %eax	jne    .L131	ALIGN_2#endif.L132:#ifndef TRMMKERNEL	movl	K, %eax#else	movl	KKK, %eax#endif	movaps	ALPHA,  %xmm3	andl	$7, %eax		# if (k & 1)	BRANCH	je .L134.L133:	movaps	 0 * SIZE(BB), %xmm2	movaps	 0 * SIZE(AA), %xmm0	mulps	%xmm0, %xmm2	addps	%xmm2, %xmm4	addl	$4 * SIZE, AA	addl	$4 * SIZE, BB	decl	%eax	jg	.L133	ALIGN_4.L134:	addps	%xmm5, %xmm4	addps	%xmm7, %xmm6	addps	%xmm6, %xmm4	mulps	%xmm3, %xmm4#ifndef TRMMKERNEL	movlps	0 * SIZE(%esi), %xmm0	movhps	2 * SIZE(%esi), %xmm0	addps	%xmm0, %xmm4#endif#ifdef HAVE_SSE2	movsd	%xmm4, 0 * SIZE(%esi)	unpckhpd %xmm4, %xmm4	movsd	%xmm4, 2 * SIZE(%esi)#else	movlps	%xmm4, 0 * SIZE(%esi)	movhps	%xmm4, 2 * SIZE(%esi)#endif#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movl	K, %eax	subl	KKK, %eax	leal	(,%eax,    8), %eax	leal	(AA, %eax, 2), AA	leal	(BB, %eax, 2), BB#endif#if defined(TRMMKERNEL) && defined(LEFT)	addl	$4, KK#endif	addl	$4 * SIZE, %esi	ALIGN_2	.L150:	testl	$2, %ebx	jle	.L170#if (L1_DATA_LINESIZE == 64)#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leal	BUFFER, BB	MOVSD	 0 * SIZE + BUFFER, %xmm2	XORPS	%xmm4, %xmm4	MOVSD	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	MOVSD	 16 * SIZE + BUFFER, %xmm3	XORPS	%xmm6, %xmm6	MOVSD	 8 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#else	leal	BUFFER, BB	movl	KK, %eax	leal	(, %eax,   8), %eax	leal	(AA, %eax, 1), AA	leal	(BB, %eax, 2), BB /* because it's doubled */	MOVSD	 0 * SIZE(BB), %xmm2	XORPS	%xmm4, %xmm4	MOVSD	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	MOVSD	 16 * SIZE(BB), %xmm3	XORPS	%xmm6, %xmm6	MOVSD	 8 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#endif	#ifndef TRMMKERNEL	movl	K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movl	K, %eax	subl	KK, %eax	movl	%eax, KKK	#else	movl	KK, %eax#ifdef LEFT	addl	$2, %eax#else	addl	$1, %eax#endif	movl	%eax, KKK#endif	sarl	$3, %eax	je	.L152	ALIGN_2.L151:	mulps	%xmm0, %xmm2	MOVSD	 2 * SIZE(AA), %xmm0	addps	%xmm2, %xmm4	MOVSD	 4 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	MOVSD	 4 * SIZE(AA), %xmm0	addps	%xmm2, %xmm5	MOVSD	 8 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	MOVSD	 6 * SIZE(AA), %xmm0	addps	%xmm2, %xmm6	MOVSD	12 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	MOVSD	16 * SIZE(AA), %xmm0	addps	%xmm2, %xmm7	MOVSD	32 * SIZE(BB), %xmm2	mulps	%xmm1, %xmm3	MOVSD	10 * SIZE(AA), %xmm1	addps	%xmm3, %xmm4	MOVSD	20 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm3	MOVSD	12 * SIZE(AA), %xmm1	addps	%xmm3, %xmm5	MOVSD	24 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm3	MOVSD	14 * SIZE(AA), %xmm1	addps	%xmm3, %xmm6	MOVSD	28 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm3	MOVSD	24 * SIZE(AA), %xmm1	addps	%xmm3, %xmm7	MOVSD	48 * SIZE(BB), %xmm3	addl   $16 * SIZE, AA	addl   $32 * SIZE, BB	decl   %eax	jne    .L151	ALIGN_2#else#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leal	BUFFER, BB	MOVSD	 0 * SIZE + BUFFER, %xmm2	XORPS	%xmm4, %xmm4	MOVSD	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	MOVSD	 8 * SIZE + BUFFER, %xmm3	XORPS	%xmm6, %xmm6	MOVSD	 8 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#else	leal	BUFFER, BB	movl	KK, %eax	leal	(, %eax,   8), %eax	leal	(AA, %eax, 1), AA	leal	(BB, %eax, 2), BB /* because it's doubled */	MOVSD	 0 * SIZE(BB), %xmm2	XORPS	%xmm4, %xmm4	MOVSD	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	MOVSD	 8 * SIZE(BB), %xmm3	XORPS	%xmm6, %xmm6	MOVSD	 8 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#endif	#ifndef TRMMKERNEL	movl	K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movl	K, %eax	subl	KK, %eax	movl	%eax, KKK	#else	movl	KK, %eax#ifdef LEFT	addl	$2, %eax#else	addl	$1, %eax#endif	movl	%eax, KKK#endif	sarl	$3, %eax	je	.L152	ALIGN_2.L151:	mulps	%xmm0, %xmm2	MOVSD	 2 * SIZE(AA), %xmm0	addps	%xmm2, %xmm4	MOVSD	 4 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm2	MOVSD	 4 * SIZE(AA), %xmm0	addps	%xmm2, %xmm5	MOVSD	16 * SIZE(BB), %xmm2	mulps	%xmm0, %xmm3	MOVSD	 6 * SIZE(AA), %xmm0	addps	%xmm3, %xmm6	MOVSD	12 * SIZE(BB), %xmm3	mulps	%xmm0, %xmm3	MOVSD	16 * SIZE(AA), %xmm0	addps	%xmm3, %xmm7	MOVSD	24 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm2	MOVSD	10 * SIZE(AA), %xmm1	addps	%xmm2, %xmm4	MOVSD	20 * SIZE(BB), %xmm2	mulps	%xmm1, %xmm2	MOVSD	12 * SIZE(AA), %xmm1	addps	%xmm2, %xmm5	MOVSD	32 * SIZE(BB), %xmm2	mulps	%xmm1, %xmm3	MOVSD	14 * SIZE(AA), %xmm1	addps	%xmm3, %xmm6	MOVSD	28 * SIZE(BB), %xmm3	mulps	%xmm1, %xmm3	MOVSD	24 * SIZE(AA), %xmm1	addps	%xmm3, %xmm7	MOVSD	40 * SIZE(BB), %xmm3	addl   $16 * SIZE, AA	addl   $32 * SIZE, BB	decl   %eax	jne    .L151	ALIGN_2#endif.L152:#ifndef TRMMKERNEL	movl	K, %eax#else	movl	KKK, %eax#endif	movaps	ALPHA,  %xmm3	andl	$7, %eax		# if (k & 1)	BRANCH	je .L154.L153:	mulps	%xmm0, %xmm2	MOVSD	 2 * SIZE(AA), %xmm0	addps	%xmm2, %xmm4	MOVSD	 4 * SIZE(BB), %xmm2	addl	$2 * SIZE, AA	addl	$4 * SIZE, BB	decl	%eax	jg	.L153	ALIGN_4.L154:	addps	%xmm5, %xmm4	addps	%xmm7, %xmm6	addps	%xmm6, %xmm4	mulps	%xmm3, %xmm4#ifndef TRMMKERNEL	movlps	0 * SIZE(%esi), %xmm0	addps	%xmm0, %xmm4#endif#ifdef HAVE_SSE2	movsd	%xmm4, 0 * SIZE(%esi)#else	movlps	%xmm4, 0 * SIZE(%esi)#endif#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movl	K, %eax	subl	KKK, %eax	leal	(,%eax,    8), %eax	leal	(AA, %eax, 1), AA	leal	(BB, %eax, 2), BB#endif#if defined(TRMMKERNEL) && defined(LEFT)	addl	$2, KK#endif	addl	$2 * SIZE, %esi	ALIGN_2	.L170:	testl	$1, %ebx	jle	.L999#if (L1_DATA_LINESIZE == 64)#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leal	BUFFER, BB	movss	 0 * SIZE + BUFFER, %xmm2	XORPS	%xmm4, %xmm4	movss	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	movss	 16 * SIZE + BUFFER, %xmm3	XORPS	%xmm6, %xmm6	movss	 4 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#else	leal	BUFFER, BB	movl	KK, %eax	leal	(, %eax,   4), %eax	leal	(AA, %eax, 1), AA	leal	(BB, %eax, 4), BB /* because it's doubled */	movss	 0 * SIZE(BB), %xmm2	XORPS	%xmm4, %xmm4	movss	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	movss	 16 * SIZE(BB), %xmm3	XORPS	%xmm6, %xmm6	movss	 4 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#endif	#ifndef TRMMKERNEL	movl	K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movl	K, %eax	subl	KK, %eax	movl	%eax, KKK	#else	movl	KK, %eax#ifdef LEFT	addl	$1, %eax#else	addl	$1, %eax#endif	movl	%eax, KKK#endif	sarl	$3, %eax	je	.L172	ALIGN_2.L171:	mulss	%xmm0, %xmm2	movss	 1 * SIZE(AA), %xmm0	addss	%xmm2, %xmm4	mulss	 4 * SIZE(BB), %xmm0	movss	32 * SIZE(BB), %xmm2	addss	%xmm0, %xmm5	movss	 2 * SIZE(AA), %xmm0	mulss	 8 * SIZE(BB), %xmm0	addss	%xmm0, %xmm6	movss	 3 * SIZE(AA), %xmm0	mulss	12 * SIZE(BB), %xmm0	addss	%xmm0, %xmm7	movss	 8 * SIZE(AA), %xmm0	mulss	%xmm1, %xmm3	movss	 5 * SIZE(AA), %xmm1	addss	%xmm3, %xmm4	mulss	20 * SIZE(BB), %xmm1	movss	48 * SIZE(BB), %xmm3	addss	%xmm1, %xmm5	movss	 6 * SIZE(AA), %xmm1	mulss	24 * SIZE(BB), %xmm1	addss	%xmm1, %xmm6	movss	 7 * SIZE(AA), %xmm1	mulss	28 * SIZE(BB), %xmm1	addss	%xmm1, %xmm7	movss	12 * SIZE(AA), %xmm1	addl   $ 8 * SIZE, AA	addl   $32 * SIZE, BB	decl   %eax	jne    .L171	ALIGN_2#else#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leal	BUFFER, BB	movss	 0 * SIZE + BUFFER, %xmm2	XORPS	%xmm4, %xmm4	movss	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	movss	 8 * SIZE + BUFFER, %xmm3	XORPS	%xmm6, %xmm6	movss	 4 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#else	leal	BUFFER, BB	movl	KK, %eax	leal	(, %eax,   4), %eax	leal	(AA, %eax, 1), AA	leal	(BB, %eax, 4), BB /* because it's doubled */	movss	 0 * SIZE(BB), %xmm2	XORPS	%xmm4, %xmm4	movss	 0 * SIZE(AA), %xmm0	XORPS	%xmm5, %xmm5	movss	 8 * SIZE(BB), %xmm3	XORPS	%xmm6, %xmm6	movss	 4 * SIZE(AA), %xmm1	XORPS	%xmm7, %xmm7#endif	#ifndef TRMMKERNEL	movl	K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movl	K, %eax	subl	KK, %eax	movl	%eax, KKK	#else	movl	KK, %eax#ifdef LEFT	addl	$1, %eax#else	addl	$1, %eax#endif	movl	%eax, KKK#endif	sarl	$3, %eax	je	.L172	ALIGN_2.L171:	mulss	%xmm0, %xmm2	movss	 1 * SIZE(AA), %xmm0	addss	%xmm2, %xmm4	mulss	 4 * SIZE(BB), %xmm0	movss	16 * SIZE(BB), %xmm2	addss	%xmm0, %xmm5	movss	 2 * SIZE(AA), %xmm0	mulss	%xmm0, %xmm3	movss	 3 * SIZE(AA), %xmm0	addss	%xmm3, %xmm6	mulss	12 * SIZE(BB), %xmm0	movss	24 * SIZE(BB), %xmm3	addss	%xmm0, %xmm7	movss	 8 * SIZE(AA), %xmm0	mulss	%xmm1, %xmm2	movss	 5 * SIZE(AA), %xmm1	addss	%xmm2, %xmm4	mulss	20 * SIZE(BB), %xmm1	movss	32 * SIZE(BB), %xmm2	addss	%xmm1, %xmm5	movss	 6 * SIZE(AA), %xmm1	mulss	%xmm1, %xmm3	movss	 7 * SIZE(AA), %xmm1	addss	%xmm3, %xmm6	mulss	28 * SIZE(BB), %xmm1	movss	40 * SIZE(BB), %xmm3	addss	%xmm1, %xmm7	movss	12 * SIZE(AA), %xmm1	addl   $ 8 * SIZE, AA	addl   $32 * SIZE, BB	decl   %eax	jne    .L171	ALIGN_2#endif.L172:#ifndef TRMMKERNEL	movl	K, %eax#else	movl	KKK, %eax#endif	movss	ALPHA,  %xmm3	andl	$7, %eax		# if (k & 1)	BRANCH	je .L174.L173:	movss	 0 * SIZE(AA), %xmm0	movss	 0 * SIZE(BB), %xmm2	mulss	%xmm0, %xmm2	addss	%xmm2, %xmm4	addl	$1 * SIZE, AA	addl	$4 * SIZE, BB	decl	%eax	jg	.L173	ALIGN_4.L174:	addss	%xmm5, %xmm4	addss	%xmm7, %xmm6	addss	%xmm6, %xmm4	mulss	%xmm3, %xmm4#ifndef TRMMKERNEL	addss	0 * SIZE(%esi), %xmm4#endif	movss	%xmm4, 0 * SIZE(%esi)	ALIGN_2	.L999:	movl	OLD_STACK, %esp	EMMS	popl	%ebx	popl	%esi	popl	%edi	popl	%ebp	ret	ALIGN_2	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 + -