📄 zaxpy_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 r4#define PRE r5#else#define N r3#define X r8#define INCX r9 #define Y r5#define INCY r4#define YY r6#define PRE r7#endif#endif#if defined(_AIX) || defined(__APPLE__)#if !defined(__64BIT__) && defined(DOUBLE)#define N r3#define X r10#define INCX r4#define Y r5#define INCY r6#define YY r7#define PRE r8#else#define N r3#define X r8#define INCX r9#define Y r10#define INCY r4#define YY r5#define PRE r6#endif#endif#define ALPHA_R f24#define ALPHA_I f25#ifndef CONJ#define ADD1 FNMSUB#define ADD2 FMADD#else#define ADD1 FMADD#define ADD2 FNMSUB#endif#define STACKSIZE 96 PROLOGUE PROFCODE subi SP, SP, STACKSIZE 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) stfd f25, 88(SP)#if defined(linux) && defined(__64BIT__) ld INCY, 112 + STACKSIZE(SP)#endif#if defined(_AIX) || defined(__APPLE__)#ifdef __64BIT__ ld INCY, 112 + STACKSIZE(SP)#else#ifdef DOUBLE lwz INCX, 56 + STACKSIZE(SP) lwz Y, 60 + STACKSIZE(SP) lwz INCY, 64 + STACKSIZE(SP)#else lwz INCY, 56 + STACKSIZE(SP)#endif#endif#endif fmr ALPHA_R, f1 slwi INCX, INCX, ZBASE_SHIFT fmr ALPHA_I, f2 slwi INCY, INCY, ZBASE_SHIFT subi INCX, INCX, SIZE subi INCY, INCY, SIZE 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, 3 mtspr CTR, r0 ble- LL(150) .align 4 LFDUX f0, X, INCX LFDU f1, 1 * SIZE(X) LFDUX f2, X, INCX LFDU f3, 1 * SIZE(X) LFDUX f8, Y, INCY LFDU f9, 1 * SIZE(Y) LFDUX f10, Y, INCY LFDU f11, 1 * SIZE(Y) LFDUX f4, X, INCX LFDU f5, 1 * SIZE(X) LFDUX f6, X, INCX LFDU f7, 1 * SIZE(X) LFDUX f12, Y, INCY LFDU f13, 1 * SIZE(Y) LFDUX f14, Y, INCY LFDU f15, 1 * SIZE(Y) bdz LL(120) .align 4LL(110): FMADD f16, ALPHA_R, f0, f8 LFDUX f8, Y, INCY FMADD f17, ALPHA_I, f0, f9 LFDU f9, 1 * SIZE(Y) FMADD f18, ALPHA_R, f2, f10 LFDUX f10, Y, INCY FMADD f19, ALPHA_I, f2, f11 LFDU f11, 1 * SIZE(Y)#ifdef PPCG4 dcbt X, PRE#endif ADD1 f16, ALPHA_I, f1, f16 LFDUX f0, X, INCX ADD2 f17, ALPHA_R, f1, f17 LFDU f1, 1 * SIZE(X) ADD1 f18, ALPHA_I, f3, f18 LFDUX f2, X, INCX ADD2 f19, ALPHA_R, f3, f19 LFDU f3, 1 * SIZE(X)#ifdef PPCG4 dcbtst Y, PRE#endif FMADD f20, ALPHA_R, f4, f12 LFDUX f12, Y, INCY FMADD f21, ALPHA_I, f4, f13 LFDU f13, 1 * SIZE(Y) FMADD f22, ALPHA_R, f6, f14 LFDUX f14, Y, INCY FMADD f23, ALPHA_I, f6, f15 LFDU f15, 1 * SIZE(Y)#if defined(PPCG4) && defined(DOUBLE) dcbt X, PRE#endif ADD1 f20, ALPHA_I, f5, f20 LFDUX f4, X, INCX ADD2 f21, ALPHA_R, f5, f21 LFDU f5, 1 * SIZE(X) ADD1 f22, ALPHA_I, f7, f22 LFDUX f6, X, INCX ADD2 f23, ALPHA_R, f7, f23 LFDU f7, 1 * SIZE(X)#if defined(PPCG4) && defined(DOUBLE) dcbtst Y, PRE#endif STFDUX f16, YY, INCY STFDU f17, 1 * SIZE(YY) STFDUX f18, YY, INCY STFDU f19, 1 * SIZE(YY) FMADD f16, ALPHA_R, f0, f8 LFDUX f8, Y, INCY FMADD f17, ALPHA_I, f0, f9 LFDU f9, 1 * SIZE(Y) FMADD f18, ALPHA_R, f2, f10 LFDUX f10, Y, INCY FMADD f19, ALPHA_I, f2, f11 LFDU f11, 1 * SIZE(Y)#ifdef PPCG4 dcbt X, PRE#endif ADD1 f16, ALPHA_I, f1, f16 LFDUX f0, X, INCX ADD2 f17, ALPHA_R, f1, f17 LFDU f1, 1 * SIZE(X) ADD1 f18, ALPHA_I, f3, f18 LFDUX f2, X, INCX ADD2 f19, ALPHA_R, f3, f19 LFDU f3, 1 * SIZE(X)#ifdef PPCG4 dcbtst Y, PRE#endif STFDUX f20, YY, INCY STFDU f21, 1 * SIZE(YY) STFDUX f22, YY, INCY STFDU f23, 1 * SIZE(YY) FMADD f20, ALPHA_R, f4, f12 LFDUX f12, Y, INCY FMADD f21, ALPHA_I, f4, f13 LFDU f13, 1 * SIZE(Y) FMADD f22, ALPHA_R, f6, f14 LFDUX f14, Y, INCY FMADD f23, ALPHA_I, f6, f15 LFDU f15, 1 * SIZE(Y)#if defined(PPCG4) && defined(DOUBLE) dcbt X, PRE#endif ADD1 f20, ALPHA_I, f5, f20 LFDUX f4, X, INCX ADD2 f21, ALPHA_R, f5, f21 LFDU f5, 1 * SIZE(X) ADD1 f22, ALPHA_I, f7, f22 LFDUX f6, X, INCX ADD2 f23, ALPHA_R, f7, f23 LFDU f7, 1 * SIZE(X)#if defined(PPCG4) && defined(DOUBLE) dcbtst Y, PRE#endif STFDUX f16, YY, INCY STFDU f17, 1 * SIZE(YY) STFDUX f18, YY, INCY STFDU f19, 1 * SIZE(YY) STFDUX f20, YY, INCY STFDU f21, 1 * SIZE(YY) STFDUX f22, YY, INCY STFDU f23, 1 * SIZE(YY) bdnz LL(110) .align 4LL(120): FMADD f16, ALPHA_R, f0, f8 LFDUX f8, Y, INCY FMADD f17, ALPHA_I, f0, f9 LFDU f9, 1 * SIZE(Y) FMADD f18, ALPHA_R, f2, f10 LFDUX f10, Y, INCY FMADD f19, ALPHA_I, f2, f11 LFDU f11, 1 * SIZE(Y) ADD1 f16, ALPHA_I, f1, f16 LFDUX f0, X, INCX ADD2 f17, ALPHA_R, f1, f17 LFDU f1, 1 * SIZE(X) ADD1 f18, ALPHA_I, f3, f18 LFDUX f2, X, INCX ADD2 f19, ALPHA_R, f3, f19 LFDU f3, 1 * SIZE(X) FMADD f20, ALPHA_R, f4, f12 LFDUX f12, Y, INCY FMADD f21, ALPHA_I, f4, f13 LFDU f13, 1 * SIZE(Y) FMADD f22, ALPHA_R, f6, f14 LFDUX f14, Y, INCY FMADD f23, ALPHA_I, f6, f15 LFDU f15, 1 * SIZE(Y) ADD1 f20, ALPHA_I, f5, f20 LFDUX f4, X, INCX ADD2 f21, ALPHA_R, f5, f21 LFDU f5, 1 * SIZE(X) ADD1 f22, ALPHA_I, f7, f22 LFDUX f6, X, INCX ADD2 f23, ALPHA_R, f7, f23 LFDU f7, 1 * SIZE(X) STFDUX f16, YY, INCY FMADD f16, ALPHA_R, f0, f8 STFDU f17, 1 * SIZE(YY) FMADD f17, ALPHA_I, f0, f9 STFDUX f18, YY, INCY FMADD f18, ALPHA_R, f2, f10 STFDU f19, 1 * SIZE(YY) FMADD f19, ALPHA_I, f2, f11 ADD1 f16, ALPHA_I, f1, f16 ADD2 f17, ALPHA_R, f1, f17 ADD1 f18, ALPHA_I, f3, f18 ADD2 f19, ALPHA_R, f3, f19 STFDUX f20, YY, INCY FMADD f20, ALPHA_R, f4, f12 STFDU f21, 1 * SIZE(YY) FMADD f21, ALPHA_I, f4, f13 STFDUX f22, YY, INCY FMADD f22, ALPHA_R, f6, f14 STFDU f23, 1 * SIZE(YY) FMADD f23, ALPHA_I, f6, f15 ADD1 f20, ALPHA_I, f5, f20 STFDUX f16, YY, INCY ADD2 f21, ALPHA_R, f5, f21 STFDU f17, 1 * SIZE(YY) ADD1 f22, ALPHA_I, f7, f22 STFDUX f18, YY, INCY ADD2 f23, ALPHA_R, f7, f23 STFDU f19, 1 * SIZE(YY) STFDUX f20, YY, INCY STFDU f21, 1 * SIZE(YY) STFDUX f22, YY, INCY STFDU f23, 1 * SIZE(YY) .align 4LL(150): andi. r0, N, 7 mtspr CTR, r0 ble LL(999) .align 4LL(160): LFDUX f0, X, INCX LFDU f1, 1 * SIZE(X) LFDUX f8, Y, INCY LFDU f9, 1 * SIZE(Y) FMADD f16, ALPHA_R, f0, f8 FMADD f17, ALPHA_I, f0, f9 ADD1 f16, ALPHA_I, f1, f16 ADD2 f17, ALPHA_R, f1, f17 STFDUX f16, YY, INCY STFDU f17, 1 * SIZE(YY) 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) lfd f25, 88(SP) addi SP, SP, STACKSIZE li r0, 0 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 + -