📄 zasum.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 INCXM1 r9#define PREA r8#define FZERO f0#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) stw r0, 144(SP) lfs FZERO,144(SP)#ifdef F_INTERFACE LDINT N, 0(N) LDINT INCX, 0(INCX)#endif slwi INCX, INCX, ZBASE_SHIFT subi INCXM1, INCX, SIZE fmr f1, FZERO fmr f2, FZERO fmr f3, FZERO fmr f4, FZERO fmr f5, FZERO fmr f6, FZERO fmr f7, FZERO li PREA, 4 * 16 * SIZE cmpwi cr0, N, 0 ble- LL(999) cmpwi cr0, INCX, 0 ble- LL(999) cmpwi cr0, INCX, 2 * SIZE bne- cr0, LL(100) srawi. r0, N, 3 mtspr CTR, r0 beq- cr0, LL(50) .align 4 LFD f8, 0 * SIZE(X) LFD f9, 1 * SIZE(X) LFD f10, 2 * SIZE(X) LFD f11, 3 * SIZE(X) LFD f12, 4 * SIZE(X) LFD f13, 5 * SIZE(X) LFD f14, 6 * SIZE(X) LFD f15, 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) fabs f16, f8 fabs f17, f9 fabs f18, f10 fabs f19, f11 fabs f20, f12 fabs f21, f13 fabs f22, f14 fabs f23, f15 bdz LL(20) .align 4LL(10): FADD f0, f0, f16 fabs f16, f24 FADD f1, f1, f17 fabs f17, f25 FADD f2, f2, f18 fabs f18, f26 FADD f3, f3, f19 fabs f19, f27 LFD f8, 16 * SIZE(X) LFD f9, 17 * SIZE(X) LFD f10, 18 * SIZE(X) LFD f11, 19 * SIZE(X) FADD f4, f4, f20 fabs f20, f28 FADD f5, f5, f21 fabs f21, f29 FADD f6, f6, f22 fabs f22, f30 FADD f7, f7, f23 fabs f23, f31 LFD f12, 20 * SIZE(X) LFD f13, 21 * SIZE(X) LFD f14, 22 * SIZE(X) LFD f15, 23 * SIZE(X) FADD f0, f0, f16 fabs f16, f8 FADD f1, f1, f17 fabs f17, f9 FADD f2, f2, f18 fabs f18, f10 FADD f3, f3, f19 fabs f19, f11 LFD f24, 24 * SIZE(X) LFD f25, 25 * SIZE(X) LFD f26, 26 * SIZE(X) LFD f27, 27 * SIZE(X) FADD f4, f4, f20 fabs f20, f12 FADD f5, f5, f21 fabs f21, f13 FADD f6, f6, f22 fabs f22, f14 FADD f7, f7, f23 fabs f23, f15 LFD f28, 28 * SIZE(X) LFD f29, 29 * SIZE(X) LFD f30, 30 * SIZE(X) LFD f31, 31 * SIZE(X) dcbt X, PREA addi X, X, 16 * SIZE bdnz LL(10) .align 4LL(20): FADD f0, f0, f16 fabs f16, f24 FADD f1, f1, f17 fabs f17, f25 FADD f2, f2, f18 fabs f18, f26 FADD f3, f3, f19 fabs f19, f27 FADD f4, f4, f20 fabs f20, f28 FADD f5, f5, f21 fabs f21, f29 FADD f6, f6, f22 fabs f22, f30 FADD f7, f7, f23 fabs f23, f31 FADD f0, f0, f16 FADD f1, f1, f17 FADD f2, f2, f18 FADD f3, f3, f19 FADD f4, f4, f20 FADD f5, f5, f21 FADD f6, f6, f22 FADD f7, f7, f23 addi X, X, 16 * SIZE .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 fabs f8, f8 fabs f9, f9 FADD f0, f0, f8 FADD f1, f1, f9 bdnz LL(60) b LL(999) .align 4LL(100): sub X, X, INCXM1 srawi. r0, N, 3 mtspr CTR, r0 beq- LL(150) 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 LFDX f24, X, INCXM1 LFDUX f25, X, INCX LFDX f26, X, INCXM1 LFDUX f27, X, INCX LFDX f28, X, INCXM1 LFDUX f29, X, INCX LFDX f30, X, INCXM1 LFDUX f31, X, INCX fabs f16, f8 fabs f17, f9 fabs f18, f10 fabs f19, f11 fabs f20, f12 fabs f21, f13 fabs f22, f14 fabs f23, f15 bdz LL(120) .align 4LL(110): FADD f0, f0, f16 fabs f16, f24 FADD f1, f1, f17 fabs f17, f25 FADD f2, f2, f18 fabs f18, f26 FADD f3, f3, f19 fabs f19, f27 LFDX f8, X, INCXM1 LFDUX f9, X, INCX LFDX f10, X, INCXM1 LFDUX f11, X, INCX FADD f4, f4, f20 fabs f20, f28 FADD f5, f5, f21 fabs f21, f29 FADD f6, f6, f22 fabs f22, f30 FADD f7, f7, f23 fabs f23, f31 LFDX f12, X, INCXM1 LFDUX f13, X, INCX LFDX f14, X, INCXM1 LFDUX f15, X, INCX FADD f0, f0, f16 fabs f16, f8 FADD f1, f1, f17 fabs f17, f9 FADD f2, f2, f18 fabs f18, f10 FADD f3, f3, f19 fabs f19, f11 LFDX f24, X, INCXM1 LFDUX f25, X, INCX LFDX f26, X, INCXM1 LFDUX f27, X, INCX FADD f4, f4, f20 fabs f20, f12 FADD f5, f5, f21 fabs f21, f13 FADD f6, f6, f22 fabs f22, f14 FADD f7, f7, f23 fabs f23, f15 LFDX f28, X, INCXM1 LFDUX f29, X, INCX LFDX f30, X, INCXM1 LFDUX f31, X, INCX bdnz LL(110) .align 4LL(120): FADD f0, f0, f16 fabs f16, f24 FADD f1, f1, f17 fabs f17, f25 FADD f2, f2, f18 fabs f18, f26 FADD f3, f3, f19 fabs f19, f27 FADD f4, f4, f20 fabs f20, f28 FADD f5, f5, f21 fabs f21, f29 FADD f6, f6, f22 fabs f22, f30 FADD f7, f7, f23 fabs f23, f31 FADD f0, f0, f16 FADD f1, f1, f17 FADD f2, f2, f18 FADD f3, f3, f19 FADD f4, f4, f20 FADD f5, f5, f21 FADD f6, f6, f22 FADD f7, f7, f23 .align 4LL(150): andi. r0, N, 7 mtspr CTR, r0 beq LL(999) .align 4LL(160): LFDX f8, X, INCXM1 LFDUX f9, X, INCX fabs f8, f8 fabs f9, f9 FADD f0, f0, f8 FADD f1, f1, f9 bdnz LL(160) .align 4LL(999): FADD f0, f0, f1 FADD f2, f2, f3 FADD f4, f4, f5 FADD f6, f6, f7 FADD f0, f0, f2 FADD f4, f4, f6 FADD f1, f0, f4 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 + -