📄 zasum_sse.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"	#define STACK	12#define ARGS     8#define V	 0 + STACK(%esp)#define STACK_M		 4 + STACK + ARGS(%esp)#define STACK_X		 8 + STACK + ARGS(%esp)#define STACK_INCX	12 + STACK + ARGS(%esp)#define I	%eax#define M	%ecx#define X	%esi#define INCX	%ebx#define FLAG	%edi	#ifndef HAVE_SSE2#define movsd	movlps#define clear(a)	xorps a, a#else#define clear(a)#endif	PROLOGUE	subl	$ARGS, %esp	pushl	%esi	pushl	%edi	pushl	%ebx	PROFCODE	movl	STACK_M, M	movl	STACK_X, X	movl	STACK_INCX, INCX#ifdef F_INTERFACE	movl	(M), M	movl	(INCX), INCX#endif#ifdef HAVE_SSE2	pxor	%xmm0, %xmm0	pxor	%xmm7, %xmm7#else	xorps	%xmm0, %xmm0	xorps	%xmm7, %xmm7#endif	testl	M, M	jle	.L999	testl	INCX, INCX	jle	.L999#ifdef HAVE_SSE2	pxor	%xmm1, %xmm1	cmpeqps	%xmm7, %xmm7	psrld	$1, %xmm7#else	xorps	%xmm1, %xmm1	movl	$0x7fffffff, V	movss	V, %xmm7	shufps	$0, %xmm7, %xmm7#endif	xor	FLAG, FLAG	sall	$ZBASE_SHIFT, INCX	cmpl	$2 * SIZE, INCX	jne	.L100		# incx != 1	cmpl	$8, M	jle	.L50/* Analigned Check */	testl	$3, X	jne	.L50		# Unaligned Mode/* Aligned check */	testl	$4, X	je	.L05	movss	0 * SIZE(X), %xmm0	addl	$SIZE, X	andps	%xmm7, %xmm0	decl	M	mov	$1, FLAG	ALIGN_3.L05:	testl	$8, X	je	.L10	clear	(%xmm1)	movsd	0 * SIZE(X), %xmm1	andps	%xmm7, %xmm1	decl	M	addl	$2 * SIZE, X	ALIGN_3/* aligned mode */.L10:	movl	M,  I	sarl	$5, I	jle	.L14		movaps	 0 * SIZE(X), %xmm2	movaps	16 * SIZE(X), %xmm3	movaps	32 * SIZE(X), %xmm4	movaps	48 * SIZE(X), %xmm5	decl	I	jle	.L12	ALIGN_4#ifdef PENTIUM3#define PREFETCHSIZE 216#endif#ifdef CORE2#define PREFETCHSIZE 216#endif#ifdef PENRYN#define PREFETCHSIZE 216#endif#ifdef PENTIUM4#define PREFETCHSIZE 216#endif#ifdef PENTIUMM#define PREFETCHSIZE 216#endif#if defined(OPTERON) || defined(BARCELONA)#define PREFETCHSIZE 344#endif.L11:#if defined(OPTERON) || defined(BARCELONA)       prefetcht0	(PREFETCHSIZE +  0) * SIZE(X)#endif	andps	%xmm7, %xmm2	addps	%xmm2, %xmm0	movaps	 4 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm1	movaps	 8 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm0	movaps	12 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm1#if defined(OPTERON) || defined(BARCELONA)       prefetcht0	(PREFETCHSIZE + 16) * SIZE(X)#endif	movaps	 64 * SIZE(X), %xmm2	andps	%xmm7, %xmm3	addps	%xmm3, %xmm0	movaps	20 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm1	movaps	24 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm0	movaps	28 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm1#ifdef PENTIUM4       prefetcht0	(PREFETCHSIZE +  0) * SIZE(X)#endif#if defined(OPTERON) || defined(BARCELONA)       prefetcht0	(PREFETCHSIZE + 32) * SIZE(X)#endif	movaps	 80 * SIZE(X), %xmm3	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movaps	36 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	movaps	40 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movaps	44 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1#if defined(OPTERON) || defined(BARCELONA)       prefetcht0	(PREFETCHSIZE + 48) * SIZE(X)#endif	movaps	 96 * SIZE(X), %xmm4	andps	%xmm7, %xmm5	addps	%xmm5, %xmm0	movaps	52 * SIZE(X), %xmm5	andps	%xmm7, %xmm5	addps	%xmm5, %xmm1	movaps	56 * SIZE(X), %xmm5	andps	%xmm7, %xmm5	addps	%xmm5, %xmm0	movaps	60 * SIZE(X), %xmm5	andps	%xmm7, %xmm5	addps	%xmm5, %xmm1	movaps	112 * SIZE(X), %xmm5#ifdef PENTIUM4	prefetcht0	(PREFETCHSIZE + 32) * SIZE(X)#endif	addl	$64 * SIZE, X	decl	I	jg	.L11	ALIGN_4.L12:	andps	%xmm7, %xmm2	addps	%xmm2, %xmm0	movaps	 4 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm1	movaps	 8 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm0	movaps	12 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm1	andps	%xmm7, %xmm3	addps	%xmm3, %xmm0	movaps	20 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm1	movaps	24 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm0	movaps	28 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm1	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movaps	36 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	movaps	40 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movaps	44 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	andps	%xmm7, %xmm5	addps	%xmm5, %xmm0	movaps	52 * SIZE(X), %xmm5	andps	%xmm7, %xmm5	addps	%xmm5, %xmm1	movaps	56 * SIZE(X), %xmm5	andps	%xmm7, %xmm5	addps	%xmm5, %xmm0	movaps	60 * SIZE(X), %xmm5	andps	%xmm7, %xmm5	addps	%xmm5, %xmm1	addl	$64 * SIZE, X	ALIGN_2.L14:	testl	$31,  M	je	.L19	testl	$16, M	je	.L15	movaps	 0 * SIZE(X), %xmm2	movaps	16 * SIZE(X), %xmm3	andps	%xmm7, %xmm2	addps	%xmm2, %xmm0	movaps	 4 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm1	movaps	 8 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm0	movaps	12 * SIZE(X), %xmm2	andps	%xmm7, %xmm2	addps	%xmm2, %xmm1	andps	%xmm7, %xmm3	addps	%xmm3, %xmm0	movaps	20 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm1	movaps	24 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm0	movaps	28 * SIZE(X), %xmm3	andps	%xmm7, %xmm3	addps	%xmm3, %xmm1	addl	$32 * SIZE, X	ALIGN_3.L15:	testl	$8, M	je	.L16	movaps	 0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movaps	 4 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	movaps	 8 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movaps	12 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	addl	$16 * SIZE, X	ALIGN_3.L16:	testl	$4, M	je	.L17	movaps	0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movaps	4 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	addl	$8 * SIZE, X	ALIGN_3.L17:	testl	$2, M	je	.L18	movaps	0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	addl	$4 * SIZE, X	ALIGN_3.L18:	testl	$1, M	je	.L19	clear	(%xmm4)	movsd	0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	addl	$2 * SIZE, X	ALIGN_4.L19:	testl	$1, FLAG	je	.L998	movss	0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addss	%xmm4, %xmm0	jmp	.L998	ALIGN_4/* unaligned mode */.L50:	movl	M,  I	sarl	$3, I	jle	.L55	ALIGN_4	.L51:	prefetchnta	PREFETCHSIZE * SIZE(X)	movsd	 0 * SIZE(X), %xmm4	movhps	 2 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movsd	 4 * SIZE(X), %xmm4	movhps	 6 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	movsd	 8 * SIZE(X), %xmm4	movhps	10 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movsd	12 * SIZE(X), %xmm4	movhps	14 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	addl	$16 * SIZE, X	decl	I	jg	.L51	ALIGN_4.L55:	clear	(%xmm4)	testl	$7,  M	je	.L998	testl	$4, M	je	.L56	movsd	0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movsd	2 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	movsd	4 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movsd	6 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	addl	$8 * SIZE, X	ALIGN_3.L56:	testl	$2, M	je	.L57	movsd	0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movsd	2 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	addl	$4 * SIZE, X	ALIGN_3.L57:	testl	$1, M	je	.L998	movsd	0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	jmp	.L998	ALIGN_4/* incx != 1 */.L100:	clear	(%xmm4)	movl	M,  I	sarl	$2, I	jle	.L105	ALIGN_4	.L101:	movsd	0 * SIZE(X), %xmm4	addl	INCX, X	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movsd	0 * SIZE(X), %xmm4	addl	INCX, X	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	movsd	0 * SIZE(X), %xmm4	addl	INCX, X	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	movsd	0 * SIZE(X), %xmm4	addl	INCX, X	andps	%xmm7, %xmm4	addps	%xmm4, %xmm1	decl	I	jg	.L101	ALIGN_4.L105:	andl	$3,  M	jle	.L998	ALIGN_4.L106:	movsd	0 * SIZE(X), %xmm4	andps	%xmm7, %xmm4	addps	%xmm4, %xmm0	addl	INCX, X	decl	M	jg	.L106	ALIGN_4.L998:	addps	%xmm1, %xmm0	movhlps	%xmm0, %xmm1	addps	%xmm1, %xmm0		movaps	%xmm0, %xmm1	shufps  $1, %xmm0, %xmm0	addss	 %xmm1, %xmm0		ALIGN_4.L999:	movss	%xmm0, V	flds	V	popl	%ebx	popl	%edi	popl	%esi	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 + -