📄 snrm2.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 PREA r8#define FZERO 144(SP)#define FONE 148(SP)#define STACKSIZE 160 PROLOGUE PROFCODE addi SP, SP, -STACKSIZE li r10, 0 lis r11, 0x3f80 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) stw r10, FZERO stw r11, FONE lfs f1, FZERO#ifdef F_INTERFACE LDINT N, 0(N) LDINT INCX, 0(INCX)#endif slwi INCX, INCX, BASE_SHIFT li PREA, 4 * 16 * SIZE cmpwi cr0, N, 0 ble- LL(9999) cmpwi cr0, INCX, 0 ble- LL(9999) fmr f0, f1 fmr f2, f1 fmr f3, f1 fmr f4, f1 fmr f5, f1 fmr f6, f1 fmr f7, f1 fmr f8, f1 fmr f9, f1 fmr f10, f1 fmr f11, f1 fmr f12, f1 fmr f13, f1 fmr f14, f1 fmr f15, f1 cmpwi cr0, INCX, SIZE bne- cr0, LL(1000) srawi. r0, N, 4 mtspr CTR, r0 beq- cr0, LL(150) LFD f16, 0 * SIZE(X) LFD f17, 1 * SIZE(X) LFD f18, 2 * SIZE(X) LFD f19, 3 * SIZE(X) LFD f20, 4 * SIZE(X) LFD f21, 5 * SIZE(X) LFD f22, 6 * SIZE(X) LFD f23, 7 * SIZE(X) LFD f24, 8 * SIZE(X) LFD f25, 9 * SIZE(X) LFD f26, 10 * SIZE(X) LFD f27, 11 * SIZE(X) LFD f28, 12 * SIZE(X) LFD f29, 13 * SIZE(X) LFD f30, 14 * SIZE(X) LFD f31, 15 * SIZE(X) bdz LL(120) .align 4LL(110): fmadd f0, f16, f16, f0 fmadd f1, f17, f17, f1 fmadd f2, f18, f18, f2 fmadd f3, f19, f19, f3 LFD f16, 16 * SIZE(X) LFD f17, 17 * SIZE(X) LFD f18, 18 * SIZE(X) LFD f19, 19 * SIZE(X) fmadd f4, f20, f20, f4 fmadd f5, f21, f21, f5 fmadd f6, f22, f22, f6 fmadd f7, f23, f23, f7 LFD f20, 20 * SIZE(X) LFD f21, 21 * SIZE(X) LFD f22, 22 * SIZE(X) LFD f23, 23 * SIZE(X) fmadd f8, f24, f24, f8 fmadd f9, f25, f25, f9 fmadd f10, f26, f26, f10 fmadd f11, f27, f27, f11 LFD f24, 24 * SIZE(X) LFD f25, 25 * SIZE(X) LFD f26, 26 * SIZE(X) LFD f27, 27 * SIZE(X) fmadd f12, f28, f28, f12 fmadd f13, f29, f29, f13 fmadd f14, f30, f30, f14 fmadd f15, f31, f31, f15 LFD f28, 28 * SIZE(X) LFD f29, 29 * SIZE(X) LFD f30, 30 * SIZE(X) LFD f31, 31 * SIZE(X) addi X, X, 16 * SIZE dcbt PREA, X bdnz LL(110) .align 4LL(120): fmadd f0, f16, f16, f0 fmadd f1, f17, f17, f1 fmadd f2, f18, f18, f2 fmadd f3, f19, f19, f3 fmadd f4, f20, f20, f4 fmadd f5, f21, f21, f5 fmadd f6, f22, f22, f6 fmadd f7, f23, f23, f7 fmadd f8, f24, f24, f8 fmadd f9, f25, f25, f9 fmadd f10, f26, f26, f10 fmadd f11, f27, f27, f11 fmadd f12, f28, f28, f12 fmadd f13, f29, f29, f13 fmadd f14, f30, f30, f14 fmadd f15, f31, f31, f15 addi X, X, 16 * SIZE .align 4LL(150): andi. r0, N, 15 mtspr CTR, r0 beq- cr0, LL(170) .align 4LL(160): LFD f16, 0 * SIZE(X) addi X, X, 1 * SIZE fmadd f0, f16, f16, f0 bdnz LL(160) .align 4LL(170): fadd f0, f0, f1 fadd f2, f2, f3 fadd f4, f4, f5 fadd f6, f6, f7 fadd f8, f8, f9 fadd f10, f10, f11 fadd f12, f12, f13 fadd f14, f14, f15 fadd f0, f0, f2 fadd f4, f4, f6 fadd f8, f8, f10 fadd f12, f12, f14 fadd f0, f0, f4 fadd f8, f8, f12 fadd f0, f0, f8 fsqrts f1, f0 b LL(9999) .align 4LL(1000): sub X, X, INCX srawi. r0, N, 4 mtspr CTR, r0 beq- cr0, LL(1150) LFDUX f16, X, INCX LFDUX f17, X, INCX LFDUX f18, X, INCX LFDUX f19, X, INCX LFDUX f20, X, INCX LFDUX f21, X, INCX LFDUX f22, X, INCX LFDUX f23, X, INCX LFDUX f24, X, INCX LFDUX f25, X, INCX LFDUX f26, X, INCX LFDUX f27, X, INCX LFDUX f28, X, INCX LFDUX f29, X, INCX LFDUX f30, X, INCX LFDUX f31, X, INCX bdz LL(1120) .align 4LL(1110): fmadd f0, f16, f16, f0 fmadd f1, f17, f17, f1 fmadd f2, f18, f18, f2 fmadd f3, f19, f19, f3 LFDUX f16, X, INCX LFDUX f17, X, INCX LFDUX f18, X, INCX LFDUX f19, X, INCX fmadd f4, f20, f20, f4 fmadd f5, f21, f21, f5 fmadd f6, f22, f22, f6 fmadd f7, f23, f23, f7 LFDUX f20, X, INCX LFDUX f21, X, INCX LFDUX f22, X, INCX LFDUX f23, X, INCX fmadd f8, f24, f24, f8 fmadd f9, f25, f25, f9 fmadd f10, f26, f26, f10 fmadd f11, f27, f27, f11 LFDUX f24, X, INCX LFDUX f25, X, INCX LFDUX f26, X, INCX LFDUX f27, X, INCX fmadd f12, f28, f28, f12 fmadd f13, f29, f29, f13 fmadd f14, f30, f30, f14 fmadd f15, f31, f31, f15 LFDUX f28, X, INCX LFDUX f29, X, INCX LFDUX f30, X, INCX LFDUX f31, X, INCX bdnz LL(1110) .align 4LL(1120): fmadd f0, f16, f16, f0 fmadd f1, f17, f17, f1 fmadd f2, f18, f18, f2 fmadd f3, f19, f19, f3 fmadd f4, f20, f20, f4 fmadd f5, f21, f21, f5 fmadd f6, f22, f22, f6 fmadd f7, f23, f23, f7 fmadd f8, f24, f24, f8 fmadd f9, f25, f25, f9 fmadd f10, f26, f26, f10 fmadd f11, f27, f27, f11 fmadd f12, f28, f28, f12 fmadd f13, f29, f29, f13 fmadd f14, f30, f30, f14 fmadd f15, f31, f31, f15 .align 4LL(1150): andi. r0, N, 15 mtspr CTR, r0 beq- cr0, LL(1170) .align 4LL(1160): LFDUX f16, X, INCX fmadd f0, f16, f16, f0 bdnz LL(1160) .align 4LL(1170): fadd f0, f0, f1 fadd f2, f2, f3 fadd f4, f4, f5 fadd f6, f6, f7 fadd f8, f8, f9 fadd f10, f10, f11 fadd f12, f12, f13 fadd f14, f14, f15 fadd f0, f0, f2 fadd f4, f4, f6 fadd f8, f8, f10 fadd f12, f12, f14 fadd f0, f0, f4 fadd f8, f8, f12 fadd f0, f0, f8 fsqrts f1, f0 .align 4LL(9999): 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 + -