📄 swap.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 PREA r4#define XX r10#define YY r11#else#define N r3#define X r7#define INCX r8 #define Y r9#define INCY r10#define PREA r4#define XX r5#define YY r6#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 PREA r5#define XX r6#define YY r11#else#define N r3#define X r7#define INCX r8 #define Y r9#define INCY r10#define PREA r4#define XX r5#define YY r6#endif#endif#define STACKSIZE 160 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) stfd f25, 88(SP) stfd f26, 96(SP) stfd f27, 104(SP) stfd f28, 112(SP) stfd f29, 120(SP) stfd f30, 128(SP) stfd f31, 136(SP)#if (defined(_AIX) || defined(__APPLE__)) && !defined(__64BIT__) && defined(DOUBLE) lwz INCY, 56 + STACKSIZE(SP)#endif slwi INCX, INCX, BASE_SHIFT slwi INCY, INCY, BASE_SHIFT li PREA, 3 * 16 * SIZE cmpwi cr0, N, 0 ble- LL(999) cmpwi cr0, INCX, SIZE bne- cr0, LL(100) cmpwi cr0, INCY, SIZE bne- cr0, LL(100) srawi. r0, N, 4 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 f16, 0 * SIZE(Y) LFD f17, 1 * SIZE(Y) LFD f18, 2 * SIZE(Y) LFD f19, 3 * SIZE(Y) LFD f4, 4 * SIZE(X) LFD f5, 5 * SIZE(X) LFD f6, 6 * SIZE(X) LFD f7, 7 * SIZE(X) LFD f20, 4 * SIZE(Y) LFD f21, 5 * SIZE(Y) LFD f22, 6 * SIZE(Y) LFD f23, 7 * SIZE(Y) LFD f8, 8 * SIZE(X) LFD f9, 9 * SIZE(X) LFD f10, 10 * SIZE(X) LFD f11, 11 * SIZE(X) LFD f24, 8 * SIZE(Y) LFD f25, 9 * SIZE(Y) LFD f26, 10 * SIZE(Y) LFD f27, 11 * SIZE(Y) LFD f12, 12 * SIZE(X) LFD f13, 13 * SIZE(X) LFD f14, 14 * SIZE(X) LFD f15, 15 * SIZE(X) LFD f28, 12 * SIZE(Y) LFD f29, 13 * SIZE(Y) LFD f30, 14 * SIZE(Y) LFD f31, 15 * SIZE(Y) STFD f16, 0 * SIZE(X) STFD f17, 1 * SIZE(X) STFD f18, 2 * SIZE(X) STFD f19, 3 * SIZE(X) STFD f0, 0 * SIZE(Y) STFD f1, 1 * SIZE(Y) STFD f2, 2 * SIZE(Y) STFD f3, 3 * SIZE(Y) STFD f20, 4 * SIZE(X) STFD f21, 5 * SIZE(X) STFD f22, 6 * SIZE(X) STFD f23, 7 * SIZE(X) STFD f4, 4 * SIZE(Y) STFD f5, 5 * SIZE(Y) STFD f6, 6 * SIZE(Y) STFD f7, 7 * SIZE(Y) STFD f24, 8 * SIZE(X) STFD f25, 9 * SIZE(X) STFD f26, 10 * SIZE(X) STFD f27, 11 * SIZE(X) STFD f8, 8 * SIZE(Y) STFD f9, 9 * SIZE(Y) STFD f10, 10 * SIZE(Y) STFD f11, 11 * SIZE(Y) STFD f28, 12 * SIZE(X) STFD f29, 13 * SIZE(X) STFD f30, 14 * SIZE(X) STFD f31, 15 * SIZE(X) STFD f12, 12 * SIZE(Y) STFD f13, 13 * SIZE(Y) STFD f14, 14 * SIZE(Y) STFD f15, 15 * SIZE(Y) dcbtst X, PREA dcbtst Y, PREA addi X, X, 16 * SIZE addi Y, Y, 16 * SIZE bdnz LL(10) .align 4LL(50): andi. r0, N, 15 mtspr CTR, r0 beq LL(999) .align 4LL(60): LFD f8, 0 * SIZE(X) LFD f9, 0 * SIZE(Y) STFD f9, 0 * SIZE(X) STFD f8, 0 * SIZE(Y) addi X, X, 1 * SIZE addi Y, Y, 1 * SIZE bdnz LL(60) b LL(999) .align 4LL(100): sub X, X, INCX sub Y, Y, INCY mr XX, X mr YY, Y srawi. r0, N, 4 mtspr CTR, r0 beq- LL(150) .align 4LL(110): LFDUX f0, X, INCX LFDUX f1, X, INCX LFDUX f2, X, INCX LFDUX f3, X, INCX LFDUX f16, Y, INCY LFDUX f17, Y, INCY LFDUX f18, Y, INCY LFDUX f19, Y, INCY LFDUX f4, X, INCX LFDUX f5, X, INCX LFDUX f6, X, INCX LFDUX f7, X, INCX LFDUX f20, Y, INCY LFDUX f21, Y, INCY LFDUX f22, Y, INCY LFDUX f23, Y, INCY LFDUX f8, X, INCX LFDUX f9, X, INCX LFDUX f10, X, INCX LFDUX f11, X, INCX LFDUX f24, Y, INCY LFDUX f25, Y, INCY LFDUX f26, Y, INCY LFDUX f27, Y, INCY LFDUX f12, X, INCX LFDUX f13, X, INCX LFDUX f14, X, INCX LFDUX f15, X, INCX LFDUX f28, Y, INCY LFDUX f29, Y, INCY LFDUX f30, Y, INCY LFDUX f31, Y, INCY STFDUX f16, XX, INCX STFDUX f17, XX, INCX STFDUX f18, XX, INCX STFDUX f19, XX, INCX STFDUX f0, YY, INCY STFDUX f1, YY, INCY STFDUX f2, YY, INCY STFDUX f3, YY, INCY STFDUX f20, XX, INCX STFDUX f21, XX, INCX STFDUX f22, XX, INCX STFDUX f23, XX, INCX STFDUX f4, YY, INCY STFDUX f5, YY, INCY STFDUX f6, YY, INCY STFDUX f7, YY, INCY STFDUX f24, XX, INCX STFDUX f25, XX, INCX STFDUX f26, XX, INCX STFDUX f27, XX, INCX STFDUX f8, YY, INCY STFDUX f9, YY, INCY STFDUX f10, YY, INCY STFDUX f11, YY, INCY STFDUX f28, XX, INCX STFDUX f29, XX, INCX STFDUX f30, XX, INCX STFDUX f31, XX, INCX STFDUX f12, YY, INCY STFDUX f13, YY, INCY STFDUX f14, YY, INCY STFDUX f15, YY, INCY bdnz LL(110) .align 4LL(150): andi. r0, N, 15 mtspr CTR, r0 beq LL(999) .align 4LL(160): LFDUX f8, X, INCX LFDUX f9, Y, INCY STFDUX f9, XX, INCX STFDUX f8, 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) lfd f25, 88(SP) lfd f26, 96(SP) lfd f27, 104(SP) lfd f28, 112(SP) lfd f29, 120(SP) lfd f30, 128(SP) lfd f31, 136(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 + -