📄 rot_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" #define N r3#define X r4#define INCX r5 #define Y r6#define INCY r7#define PRE r8#define XX r9#define YY r10#define C f1#define S f2#define STACKSIZE 32 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) 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 XX, X mr YY, Y srawi. r0, N, 3 mtspr CTR, r0 beq- LL(150) LFDUX f0, X, INCX LFDUX f3, Y, INCY LFDUX f4, X, INCX FMUL f10, C, f0 LFDUX f5, Y, INCY FMUL f11, C, f3 LFDUX f6, X, INCX FMUL f12, C, f4 LFDUX f7, Y, INCY FMUL f13, C, f5 LFDUX f8, X, INCX FMADD f10, S, f3, f10 LFDUX f9, Y, INCY FNMSUB f11, S, f0, f11 LFDUX f0, X, INCX FMADD f12, S, f5, f12 LFDUX f3, Y, INCY FNMSUB f13, S, f4, f13 LFDUX f4, X, INCX bdz LL(111) .align 4LL(110): FMUL f14, C, f6 LFDUX f5, Y, INCY FMUL f15, C, f7 STFDUX f10, XX, INCX FMUL f16, C, f8 STFDUX f11, YY, INCY FMUL f17, C, f9 STFDUX f12, XX, INCX#ifdef PPCG4 dcbtst X, PRE#endif FMADD f14, S, f7, f14 STFDUX f13, YY, INCY FNMSUB f15, S, f6, f15 LFDUX f6, X, INCX FMADD f16, S, f9, f16 LFDUX f7, Y, INCY FNMSUB f17, S, f8, f17 LFDUX f8, X, INCX FMUL f10, C, f0 LFDUX f9, Y, INCY FMUL f11, C, f3 STFDUX f14, XX, INCX FMUL f12, C, f4 STFDUX f15, YY, INCY FMUL f13, C, f5 STFDUX f16, XX, INCX#ifdef PPCG4 dcbtst Y, PRE#endif FMADD f10, S, f3, f10 STFDUX f17, YY, INCY FNMSUB f11, S, f0, f11 LFDUX f0, X, INCX FMADD f12, S, f5, f12 LFDUX f3, Y, INCY FNMSUB f13, S, f4, f13 LFDUX f4, X, INCX FMUL f14, C, f6 LFDUX f5, Y, INCY FMUL f15, C, f7 STFDUX f10, XX, INCX FMUL f16, C, f8 STFDUX f11, YY, INCY FMUL f17, C, f9 STFDUX f12, XX, INCX#if defined(PPCG4) && defined(DOUBLE) dcbt X, PRE#endif FMADD f14, S, f7, f14 STFDUX f13, YY, INCY FNMSUB f15, S, f6, f15 LFDUX f6, X, INCX FMADD f16, S, f9, f16 LFDUX f7, Y, INCY FNMSUB f17, S, f8, f17 LFDUX f8, X, INCX FMUL f10, C, f0 LFDUX f9, Y, INCY FMUL f11, C, f3 STFDUX f14, XX, INCX FMUL f12, C, f4 STFDUX f15, YY, INCY FMUL f13, C, f5 STFDUX f16, XX, INCX#if defined(PPCG4) && defined(DOUBLE) dcbtst Y, PRE#endif FMADD f10, S, f3, f10 STFDUX f17, YY, INCY FNMSUB f11, S, f0, f11 LFDUX f0, X, INCX FMADD f12, S, f5, f12 LFDUX f3, Y, INCY FNMSUB f13, S, f4, f13 LFDUX f4, X, INCX bdnz LL(110) .align 4LL(111): FMUL f14, C, f6 LFDUX f5, Y, INCY FMUL f15, C, f7 STFDUX f10, XX, INCX FMUL f16, C, f8 STFDUX f11, YY, INCY FMUL f17, C, f9 STFDUX f12, XX, INCX FMADD f14, S, f7, f14 STFDUX f13, YY, INCY FNMSUB f15, S, f6, f15 LFDUX f6, X, INCX FMADD f16, S, f9, f16 LFDUX f7, Y, INCY FNMSUB f17, S, f8, f17 LFDUX f8, X, INCX FMUL f10, C, f0 LFDUX f9, Y, INCY FMUL f11, C, f3 STFDUX f14, XX, INCX FMUL f12, C, f4 STFDUX f15, YY, INCY FMUL f13, C, f5 STFDUX f16, XX, INCX FMUL f14, C, f6 STFDUX f17, YY, INCY FMUL f15, C, f7 FMUL f16, C, f8 FMUL f17, C, f9 FMADD f10, S, f3, f10 FNMSUB f11, S, f0, f11 FMADD f12, S, f5, f12 FNMSUB f13, S, f4, f13 FMADD f14, S, f7, f14 STFDUX f10, XX, INCX FNMSUB f15, S, f6, f15 STFDUX f11, YY, INCY FMADD f16, S, f9, f16 STFDUX f12, XX, INCX FNMSUB f17, S, f8, f17 STFDUX f13, YY, INCY STFDUX f14, XX, INCX STFDUX f15, YY, INCY STFDUX f16, XX, INCX STFDUX f17, YY, INCY .align 4LL(150): andi. r0, N, 7 mtspr CTR, r0 beq LL(999) .align 4LL(160): LFDUX f0, X, INCX LFDUX f3, Y, INCY FMUL f10, C, f0 FMUL f11, C, f3 FMADD f10, S, f3, f10 FNMSUB f11, S, f0, f11 STFDUX f10, XX, INCX STFDUX f11, YY, INCY bdnz LL(160) .align 4LL(999): lfd f14, 0(SP) lfd f15, 8(SP) lfd f16, 16(SP) lfd f17, 24(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 + -