📄 zcopy.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 PREA r8#define INCXM1 r9#define INCYM1 r10#define STACKSIZE 16 PROLOGUE PROFCODE addi SP, SP, -STACKSIZE li r0, 0 stfd f14, 0(SP) stfd f15, 8(SP) slwi INCX, INCX, ZBASE_SHIFT slwi INCY, INCY, ZBASE_SHIFT subi INCXM1, INCX, SIZE subi INCYM1, INCY, SIZE li PREA, 3 * 16 * SIZE cmpwi cr0, N, 0 ble- LL(999) cmpwi cr0, INCX, 2 * SIZE bne- cr0, LL(100) cmpwi cr0, INCY, 2 * SIZE bne- cr0, LL(100) srawi. r0, N, 3 mtspr CTR, r0 beq- cr0, LL(50) .align 4LL(10): LFD f0, 0 * SIZE(X) LFD f1, 1 * SIZE(X) LFD f2, 2 * SIZE(X) LFD f3, 3 * SIZE(X) LFD f4, 4 * SIZE(X) LFD f5, 5 * SIZE(X) LFD f6, 6 * SIZE(X) LFD f7, 7 * SIZE(X) LFD f8, 8 * SIZE(X) LFD f9, 9 * SIZE(X) LFD f10, 10 * SIZE(X) LFD f11, 11 * SIZE(X) LFD f12, 12 * SIZE(X) LFD f13, 13 * SIZE(X) LFD f14, 14 * SIZE(X) LFD f15, 15 * SIZE(X) STFD f0, 0 * SIZE(Y) STFD f1, 1 * SIZE(Y) STFD f2, 2 * SIZE(Y) STFD f3, 3 * SIZE(Y) STFD f4, 4 * SIZE(Y) STFD f5, 5 * SIZE(Y) STFD f6, 6 * SIZE(Y) STFD f7, 7 * SIZE(Y) STFD f8, 8 * SIZE(Y) STFD f9, 9 * SIZE(Y) STFD f10, 10 * SIZE(Y) STFD f11, 11 * SIZE(Y) STFD f12, 12 * SIZE(Y) STFD f13, 13 * SIZE(Y) STFD f14, 14 * SIZE(Y) STFD f15, 15 * SIZE(Y)#ifndef PPC440 dcbt X, PREA dcbtst Y, PREA#endif addi X, X, 16 * SIZE addi Y, Y, 16 * SIZE bdnz LL(10) .align 4LL(50): andi. r0, N, 7 mtspr CTR, r0 beq LL(999) .align 4LL(60): LFD f8, 0 * SIZE(X) LFD f9, 1 * SIZE(X) addi X, X, 2 * SIZE STFD f8, 0 * SIZE(Y) STFD f9, 1 * SIZE(Y) addi Y, Y, 2 * SIZE bdnz LL(60) b LL(999) .align 4LL(100): sub X, X, INCXM1 sub Y, Y, INCYM1 srawi. r0, N, 3 mtspr CTR, r0 beq- LL(150) .align 4LL(110): LFDX f0, X, INCXM1 LFDUX f1, X, INCX LFDX f2, X, INCXM1 LFDUX f3, X, INCX LFDX f4, X, INCXM1 LFDUX f5, X, INCX LFDX f6, X, INCXM1 LFDUX f7, X, INCX LFDX f8, X, INCXM1 LFDUX f9, X, INCX LFDX f10, X, INCXM1 LFDUX f11, X, INCX LFDX f12, X, INCXM1 LFDUX f13, X, INCX LFDX f14, X, INCXM1 LFDUX f15, X, INCX STFDX f0, Y, INCYM1 STFDUX f1, Y, INCY STFDX f2, Y, INCYM1 STFDUX f3, Y, INCY STFDX f4, Y, INCYM1 STFDUX f5, Y, INCY STFDX f6, Y, INCYM1 STFDUX f7, Y, INCY STFDX f8, Y, INCYM1 STFDUX f9, Y, INCY STFDX f10, Y, INCYM1 STFDUX f11, Y, INCY STFDX f12, Y, INCYM1 STFDUX f13, Y, INCY STFDX f14, Y, INCYM1 STFDUX f15, Y, INCY bdnz LL(110) .align 4LL(150): andi. r0, N, 7 mtspr CTR, r0 beq LL(999) .align 4LL(160): LFDX f8, X, INCXM1 LFDUX f9, X, INCX STFDX f8, Y, INCYM1 STFDUX f9, Y, INCY bdnz LL(160) .align 4LL(999): lfd f14, 0(SP) lfd f15, 8(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 + -