📄 xgemm_kernel_1x1.s

📁 Optimized GotoBLAS libraries
💻 S
字号:
/*********************************************************************//*                                                                   *//*             Optimized BLAS libraries                              *//*                     By Kazushige Goto <kgoto@tacc.utexas.edu>     *//*                                                                   *//* Copyright (c) The University of Texas, 2005. All rights reserved. *//* UNIVERSITY EXPRESSLY DISCLAIMS ANY AND ALL WARRANTIES CONCERNING  *//* THIS SOFTWARE AND DOCUMENTATION, INCLUDING ANY WARRANTIES OF      *//* MERCHANTABILITY, FITNESS FOR ANY PARTICULAR PURPOSE,              *//* NON-INFRINGEMENT AND WARRANTIES OF PERFORMANCE, AND ANY WARRANTY  *//* THAT MIGHT OTHERWISE ARISE FROM COURSE OF DEALING OR USAGE OF     *//* TRADE. NO WARRANTY IS EITHER EXPRESS OR IMPLIED WITH RESPECT TO   *//* THE USE OF THE SOFTWARE OR DOCUMENTATION.                         *//* Under no circumstances shall University be liable for incidental, *//* special, indirect, direct or consequential damages or loss of     *//* profits, interruption of business, or related expenses which may  *//* arise from use of Software or Documentation, including but not    *//* limited to those resulting from defects in Software and/or        *//* Documentation, or loss or inaccuracy of data of any kind.         *//*********************************************************************/#define ASSEMBLER#include "common.h"#if defined(OPTERON) || defined(BARCELONA)#define PREFETCH	prefetch#define PREFETCHW	prefetchw#else#define PREFETCH	prefetcht0#define PREFETCHW	prefetcht0#endif#define PREFETCHSIZE (5 + 4 * 10)#define STACK	16#define ARGS	16	#define J	 0 + STACK(%esp)#define KK	 4 + STACK(%esp)#define KKK	 8 + STACK(%esp)#define M	 4 + STACK + ARGS(%esp)#define N	 8 + STACK + ARGS(%esp)#define K	12 + STACK + ARGS(%esp)#define ALPHA_R	16 + STACK + ARGS(%esp)#define ALPHA_I	32 + STACK + ARGS(%esp)#define A	48 + STACK + ARGS(%esp)#define ARG_B	52 + STACK + ARGS(%esp)#define C	56 + STACK + ARGS(%esp)#define ARG_LDC	60 + STACK + ARGS(%esp)#define OFFSET	64 + STACK + ARGS(%esp)#define I	%esi#define B	%ebx#define CO	%edi#define AO	%edx#define BO	%ecx#define LDC	%ebp#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)#define ADD1	  faddp#define ADD2	  fsubrp#define ADD3	  faddp#define ADD4	  faddp#elif defined(RN) || defined(RT) || defined(CN) || defined(CT)#define ADD1	  faddp#define ADD2	  faddp#define ADD3	  fsubrp#define ADD4	  faddp#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)#define ADD1	  faddp#define ADD2	  faddp#define ADD3	  faddp#define ADD4	  fsubrp#else#define ADD1	  faddp#define ADD2	  fsubrp#define ADD3	  fsubrp#define ADD4	  fsubrp#endif#define PREFETCH_OFFSET 48	PROLOGUE	subl	$ARGS, %esp	# Generate Stack Frame	pushl	%ebp	pushl	%edi	pushl	%esi	pushl	%ebx	PROFCODE#if defined(TRMMKERNEL) && !defined(LEFT)	movl	OFFSET, %eax	negl	%eax	movl	%eax, KK#endif		movl	ARG_LDC, LDC	movl	ARG_B,   B	addl	$8 * SIZE, A	addl	$8 * SIZE, B		sall	$ZBASE_SHIFT, LDC	cmpl	$0, M	jle	.L999		movl	N,   %eax	movl	%eax, J	testl	%eax, %eax	jle	.L999	ALIGN_4.L01:#if defined(TRMMKERNEL) && defined(LEFT)	movl	OFFSET, %eax	movl	%eax, KK#endif		movl	A, AO	movl	C, CO	addl	LDC, C	movl	M,  I	ALIGN_4.L11:#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movl	B, BO#else	movl	KK,   %eax	sall	$ZBASE_SHIFT, %eax	leal	(AO, %eax, 1), AO	leal	(B, %eax,  1), BO#endif		fldz	fldz	fldz	fldz#if   defined(HAVE_3DNOW)	prefetchw	2 * SIZE(CO)#elif defined(HAVE_SSE)	prefetchnta	2 * SIZE(CO)#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	$2, %eax 	je	.L15	ALIGN_4.L12:	PREFETCH	(PREFETCHSIZE + 0) * SIZE(AO)	FLD	 -8 * SIZE(AO)	FLD	 -8 * SIZE(BO)	fld	 %st(1)	fmul	 %st(1), %st	ADD1	 %st, %st(3)	FLD	 -7 * SIZE(BO)	fmul	 %st, %st(2)	 	FLD	 -7 * SIZE(AO)	fmul	 %st, %st(2)	fmulp	 %st, %st(1)	ADD2	 %st, %st(6)	ADD3	 %st, %st(3)	ADD4	 %st, %st(3)	FLD	 -6 * SIZE(AO)	FLD	 -6 * SIZE(BO)	fld	 %st(1)	fmul	 %st(1), %st	ADD1	 %st, %st(3)	FLD	 -5 * SIZE(BO)	fmul	 %st, %st(2)	 	FLD	 -5 * SIZE(AO)	fmul	 %st, %st(2)	fmulp	 %st, %st(1)	ADD2	 %st, %st(6)	ADD3	 %st, %st(3)	ADD4	 %st, %st(3)	PREFETCH	(PREFETCHSIZE + 4) * SIZE(AO)	FLD	 -4 * SIZE(AO)	FLD	 -4 * SIZE(BO)	fld	 %st(1)	fmul	 %st(1), %st	ADD1	 %st, %st(3)	FLD	 -3 * SIZE(BO)	fmul	 %st, %st(2)	 	FLD	 -3 * SIZE(AO)	fmul	 %st, %st(2)	fmulp	 %st, %st(1)	ADD2	 %st, %st(6)	ADD3	 %st, %st(3)	ADD4	 %st, %st(3)	FLD	 -2 * SIZE(AO)	FLD	 -2 * SIZE(BO)	fld	 %st(1)	fmul	 %st(1), %st	ADD1	 %st, %st(3)	FLD	 -1 * SIZE(BO)	fmul	 %st, %st(2)	 	FLD	 -1 * SIZE(AO)	fmul	 %st, %st(2)	fmulp	 %st, %st(1)	ADD2	 %st, %st(6)	ADD3	 %st, %st(3)	ADD4	 %st, %st(3)	addl	$8 * SIZE,AO	addl	$8 * SIZE,BO	decl	%eax	jne	.L12	ALIGN_4.L15:#ifndef TRMMKERNEL	movl	K, %eax#else	movl	KKK, %eax#endif	and	$3,  %eax	je	.L18	ALIGN_4.L16:	FLD	 -8 * SIZE(AO)	FLD	 -8 * SIZE(BO)	fld	 %st(1)	fmul	 %st(1), %st	ADD1	 %st, %st(3)	FLD	 -7 * SIZE(BO)	fmul	 %st, %st(2)	 	FLD	 -7 * SIZE(AO)	fmul	 %st, %st(2)	fmulp	 %st, %st(1)	ADD2	 %st, %st(6)	ADD3	 %st, %st(3)	ADD4	 %st, %st(3)	addl	$2 * SIZE,AO	addl	$2 * SIZE,BO	decl	%eax	jne	 .L16	ALIGN_4.L18:	faddp	%st, %st(3)	faddp	%st, %st(1)#ifndef TRMMKERNEL	FLD	ALPHA_R	fld	%st	fmul	%st(2), %st	fxch	%st(1)	fmul	%st(3), %st	FLD	ALPHA_I	fmul	%st, %st(3)	fmulp	%st, %st(4)	fsubp	%st, %st(2)	faddp	%st, %st(2)	FLD	0 * SIZE(CO)	faddp	%st, %st(1)	FST	0 * SIZE(CO)	FLD	1 * SIZE(CO)	faddp	%st, %st(1)	FST	1 * SIZE(CO)#else	FST	1 * SIZE(CO)	FST	0 * SIZE(CO)#endif	#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movl	K, %eax	subl	KKK, %eax	sall	$ZBASE_SHIFT, %eax	leal	(AO, %eax, 1), AO	leal	(BO, %eax, 1), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addl	$1, KK#endif	addl	$2 * SIZE, CO	decl	I	jne	.L11#if defined(TRMMKERNEL) && !defined(LEFT)	addl	$1, KK#endif	movl	BO, B	decl	J	jne	.L01	ALIGN_4.L999:	popl	%ebx	popl	%esi	popl	%edi	popl	%ebp	addl	$ARGS, %esp	ret	EPILOGUE
⌨️ 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 + -