📄 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 %i0#define X %i1#define INCX %i2#define I %i3#ifdef DOUBLE#define c1 %f0#define c2 %f2#define t1 %f8#define t2 %f10#define t3 %f12#define t4 %f14#define a1 %f16#define a2 %f18#define a3 %f20#define a4 %f22#define a5 %f24#define a6 %f26#define a7 %f28#define a8 %f30#else#define c1 %f0#define c2 %f1#define t1 %f4#define t2 %f5#define t3 %f6#define t4 %f7#define a1 %f8#define a2 %f9#define a3 %f10#define a4 %f11#define a5 %f12#define a6 %f13#define a7 %f14#define a8 %f15#endif PROLOGUE SAVESP#ifdef F_INTERFACE ld [N], N ld [INCX], INCX#endif st %g0, [%fp + STACK_START + 0]#ifdef DOUBLE st %g0, [%fp + STACK_START + 4]#endif sll INCX, ZBASE_SHIFT, INCX LDF [%fp + STACK_START], c1 FMOV c1, c2 FMOV c1, t1 FMOV c1, t2 FMOV c1, t3 FMOV c1, t4 cmp INCX, 0 ble .LL19 cmp INCX, 2 * SIZE bne .LL50 nop sra N, 2, I cmp I, 0 ble,pn %icc, .LL15 nop LDF [X + 0 * SIZE], a1 add I, -1, I LDF [X + 1 * SIZE], a2 cmp I, 0 LDF [X + 2 * SIZE], a3 LDF [X + 3 * SIZE], a4 LDF [X + 4 * SIZE], a5 LDF [X + 5 * SIZE], a6 LDF [X + 6 * SIZE], a7 LDF [X + 7 * SIZE], a8 ble,pt %icc, .LL12 add X, 8 * SIZE, X#define PREFETCHSIZE 32.LL11: FADD c1, t1, c1 prefetch [X + PREFETCHSIZE * SIZE], 0 FABS a1, t1 LDF [X + 0 * SIZE], a1 FADD c2, t2, c2 add I, -1, I FABS a2, t2 LDF [X + 1 * SIZE], a2 FADD c1, t3, c1 cmp I, 0 FABS a3, t3 LDF [X + 2 * SIZE], a3 FADD c2, t4, c2 nop FABS a4, t4 LDF [X + 3 * SIZE], a4 FADD c1, t1, c1 nop FABS a5, t1 LDF [X + 4 * SIZE], a5 FADD c2, t2, c2 nop FABS a6, t2 LDF [X + 5 * SIZE], a6 FADD c1, t3, c1 FABS a7, t3 LDF [X + 6 * SIZE], a7 add X, 8 * SIZE, X FADD c2, t4, c2 FABS a8, t4 bg,pt %icc, .LL11 LDF [X - 1 * SIZE], a8.LL12: FADD c1, t1, c1 FABS a1, t1 FADD c2, t2, c2 FABS a2, t2 FADD c1, t3, c1 FABS a3, t3 FADD c2, t4, c2 FABS a4, t4 FADD c1, t1, c1 FABS a5, t1 FADD c2, t2, c2 FABS a6, t2 FADD c1, t3, c1 FABS a7, t3 FADD c2, t4, c2 FABS a8, t4.LL15: and N, 3, I cmp I, 0 ble,a,pn %icc, .LL19 nop.LL16: LDF [X + 0 * SIZE], a1 LDF [X + 1 * SIZE], a2 add I, -1, I cmp I, 0 FADD c1, t1, c1 FADD c2, t2, c2 FABS a1, t1 FABS a2, t2 bg,pt %icc, .LL16 add X, 2 * SIZE, X.LL19: FADD c1, t1, c1 FADD c2, t2, c2 FADD c1, t3, c1 FADD c2, t4, c2 FADD c1, c2, c1#if !defined(DOUBLE) && defined(F2CCONV) fstod c1, c1#endif return %i7 + 8 clr %g0.LL50: sra N, 2, I cmp I, 0 ble,pn %icc, .LL55 nop LDF [X + 0 * SIZE], a1 LDF [X + 1 * SIZE], a2 add X, INCX, X LDF [X + 0 * SIZE], a3 LDF [X + 1 * SIZE], a4 add X, INCX, X LDF [X + 0 * SIZE], a5 LDF [X + 1 * SIZE], a6 add X, INCX, X add I, -1, I LDF [X + 0 * SIZE], a7 cmp I, 0 LDF [X + 1 * SIZE], a8 ble,pt %icc, .LL52 add X, INCX, X.LL51: FADD c1, t1, c1 add I, -1, I FABS a1, t1 LDF [X + 0 * SIZE], a1 FADD c2, t2, c2 cmp I, 0 FABS a2, t2 LDF [X + 1 * SIZE], a2 add X, INCX, X FADD c1, t3, c1 FABS a3, t3 LDF [X + 0 * SIZE], a3 FADD c2, t4, c2 FABS a4, t4 LDF [X + 1 * SIZE], a4 add X, INCX, X FADD c1, t1, c1 FABS a5, t1 LDF [X + 0 * SIZE], a5 FADD c2, t2, c2 FABS a6, t2 LDF [X + 1 * SIZE], a6 add X, INCX, X FADD c1, t3, c1 FABS a7, t3 LDF [X + 0 * SIZE], a7 FADD c2, t4, c2 FABS a8, t4 LDF [X + 1 * SIZE], a8 bg,pt %icc, .LL51 add X, INCX, X.LL52: FADD c1, t1, c1 FABS a1, t1 FADD c2, t2, c2 FABS a2, t2 FADD c1, t3, c1 FABS a3, t3 FADD c2, t4, c2 FABS a4, t4 FADD c1, t1, c1 FABS a5, t1 FADD c2, t2, c2 FABS a6, t2 FADD c1, t3, c1 FABS a7, t3 FADD c2, t4, c2 FABS a8, t4.LL55: and N, 3, I cmp I, 0 ble,a,pn %icc, .LL59 nop.LL56: LDF [X + 0 * SIZE], a1 LDF [X + 1 * SIZE], a2 FADD c1, t1, c1 FADD c2, t2, c2 add I, -1, I FABS a1, t1 FABS a2, t2 cmp I, 0 bg,pt %icc, .LL56 add X, INCX, X.LL59: FADD c1, t1, c1 FADD c2, t2, c2 FADD c1, t3, c1 FADD c2, t4, c2 FADD c1, c2, c1#if !defined(DOUBLE) && defined(F2CCONV) fstod c1, c1#endif return %i7 + 8 clr %o0 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 + -