📄 axpy_ppc440.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" #ifdef linux#ifndef __64BIT__#define N r3#define X r6#define INCX r7 #define Y r8#define INCY r9#define YY r5#define PRE r4#else#define N r3#define X r7#define INCX r8 #define Y r9#define INCY r10#define YY r5#define PRE r4#endif#endif#if defined(_AIX) || defined(__APPLE__)#if !defined(__64BIT__) && defined(DOUBLE)#define N r3#define X r8#define INCX r9 #define Y r10#define INCY r4#define YY r6#define PRE r5#else#define N r3#define X r7#define INCX r8 #define Y r9#define INCY r10#define YY r5#define PRE r4#endif#endif#define ALPHA f24#define STACKSIZE 96 PROLOGUE PROFCODE addi SP, SP, -STACKSIZE li r0, 0 stfd f14, 0(SP) stfd f15, 8(SP) stfd f16, 16(SP) stfd f17, 24(SP) stfd f18, 32(SP) stfd f19, 40(SP) stfd f20, 48(SP) stfd f21, 56(SP) stfd f22, 64(SP) stfd f23, 72(SP) stfd f24, 80(SP)#if (defined(_AIX) || defined(__APPLE__)) && !defined(__64BIT__) && defined(DOUBLE) lwz INCY, 56 + STACKSIZE(SP)#endif fmr ALPHA, f1 slwi INCX, INCX, BASE_SHIFT slwi INCY, INCY, BASE_SHIFT li PRE, 2 * 16 * SIZE cmpwi cr0, N, 0 ble- LL(999) sub X, X, INCX sub Y, Y, INCY mr YY, Y srawi. r0, N, 4 mtspr CTR, r0 beq- LL(150) .align 4 LFDUX f0, X, INCX LFDUX f1, X, INCX LFDUX f2, X, INCX LFDUX f3, X, INCX LFDUX f8, Y, INCY LFDUX f9, Y, INCY LFDUX f10, Y, INCY LFDUX f11, Y, INCY LFDUX f4, X, INCX LFDUX f5, X, INCX LFDUX f6, X, INCX LFDUX f7, X, INCX LFDUX f12, Y, INCY LFDUX f13, Y, INCY LFDUX f14, Y, INCY LFDUX f15, Y, INCY bdz LL(120) .align 4LL(110): FMADD f16, ALPHA, f0, f8 LFDUX f0, X, INCX LFDUX f8, Y, INCY#ifdef PPCG4 dcbt X, PRE#endif FMADD f17, ALPHA, f1, f9 LFDUX f1, X, INCX LFDUX f9, Y, INCY FMADD f18, ALPHA, f2, f10 LFDUX f2, X, INCX LFDUX f10, Y, INCY#ifdef PPCG4 dcbtst Y, PRE#endif FMADD f19, ALPHA, f3, f11 LFDUX f3, X, INCX LFDUX f11, Y, INCY FMADD f20, ALPHA, f4, f12 LFDUX f4, X, INCX LFDUX f12, Y, INCY#if defined(PPCG4) && defined(DOUBLE) dcbt X, PRE#endif FMADD f21, ALPHA, f5, f13 LFDUX f5, X, INCX LFDUX f13, Y, INCY FMADD f22, ALPHA, f6, f14 LFDUX f6, X, INCX LFDUX f14, Y, INCY#if defined(PPCG4) && defined(DOUBLE) dcbtst Y, PRE#endif FMADD f23, ALPHA, f7, f15 LFDUX f7, X, INCX LFDUX f15, Y, INCY STFDUX f16, YY, INCY STFDUX f17, YY, INCY STFDUX f18, YY, INCY STFDUX f19, YY, INCY FMADD f16, ALPHA, f0, f8 LFDUX f0, X, INCX LFDUX f8, Y, INCY#ifdef PPCG4 dcbt X, PRE#endif FMADD f17, ALPHA, f1, f9 LFDUX f1, X, INCX LFDUX f9, Y, INCY FMADD f18, ALPHA, f2, f10 LFDUX f2, X, INCX LFDUX f10, Y, INCY#ifdef PPCG4 dcbtst Y, PRE#endif FMADD f19, ALPHA, f3, f11 LFDUX f3, X, INCX LFDUX f11, Y, INCY STFDUX f20, YY, INCY STFDUX f21, YY, INCY STFDUX f22, YY, INCY STFDUX f23, YY, INCY FMADD f20, ALPHA, f4, f12 LFDUX f4, X, INCX LFDUX f12, Y, INCY#if defined(PPCG4) && defined(DOUBLE) dcbt X, PRE#endif FMADD f21, ALPHA, f5, f13 LFDUX f5, X, INCX LFDUX f13, Y, INCY FMADD f22, ALPHA, f6, f14 LFDUX f6, X, INCX LFDUX f14, Y, INCY#if defined(PPCG4) && defined(DOUBLE) dcbtst Y, PRE#endif FMADD f23, ALPHA, f7, f15 LFDUX f7, X, INCX LFDUX f15, Y, INCY STFDUX f16, YY, INCY STFDUX f17, YY, INCY STFDUX f18, YY, INCY STFDUX f19, YY, INCY STFDUX f20, YY, INCY STFDUX f21, YY, INCY STFDUX f22, YY, INCY STFDUX f23, YY, INCY bdnz LL(110) .align 4LL(120): FMADD f16, ALPHA, f0, f8 LFDUX f0, X, INCX LFDUX f8, Y, INCY FMADD f17, ALPHA, f1, f9 LFDUX f1, X, INCX LFDUX f9, Y, INCY FMADD f18, ALPHA, f2, f10 LFDUX f2, X, INCX LFDUX f10, Y, INCY FMADD f19, ALPHA, f3, f11 LFDUX f3, X, INCX LFDUX f11, Y, INCY FMADD f20, ALPHA, f4, f12 LFDUX f4, X, INCX LFDUX f12, Y, INCY FMADD f21, ALPHA, f5, f13 LFDUX f5, X, INCX LFDUX f13, Y, INCY FMADD f22, ALPHA, f6, f14 LFDUX f6, X, INCX LFDUX f14, Y, INCY FMADD f23, ALPHA, f7, f15 LFDUX f7, X, INCX LFDUX f15, Y, INCY STFDUX f16, YY, INCY STFDUX f17, YY, INCY STFDUX f18, YY, INCY STFDUX f19, YY, INCY FMADD f16, ALPHA, f0, f8 STFDUX f20, YY, INCY FMADD f17, ALPHA, f1, f9 STFDUX f21, YY, INCY FMADD f18, ALPHA, f2, f10 STFDUX f22, YY, INCY FMADD f19, ALPHA, f3, f11 STFDUX f23, YY, INCY FMADD f20, ALPHA, f4, f12 STFDUX f16, YY, INCY FMADD f21, ALPHA, f5, f13 STFDUX f17, YY, INCY FMADD f22, ALPHA, f6, f14 STFDUX f18, YY, INCY FMADD f23, ALPHA, f7, f15 STFDUX f19, YY, INCY STFDUX f20, YY, INCY STFDUX f21, YY, INCY STFDUX f22, YY, INCY STFDUX f23, YY, INCY .align 4LL(150): andi. r0, N, 15 mtspr CTR, r0 beq LL(999) .align 4LL(160): LFDUX f0, X, INCX LFDUX f8, Y, INCY FMADD f16, ALPHA, f0, f8 STFDUX f16, YY, INCY bdnz LL(160) .align 4LL(999): lfd f14, 0(SP) lfd f15, 8(SP) lfd f16, 16(SP) lfd f17, 24(SP) lfd f18, 32(SP) lfd f19, 40(SP) lfd f20, 48(SP) lfd f21, 56(SP) lfd f22, 64(SP) lfd f23, 72(SP) lfd f24, 80(SP) addi SP, SP, STACKSIZE blr 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 + -