81 lines
2 KiB
ArmAsm
81 lines
2 KiB
ArmAsm
|
/*
|
||
|
|
* Copyright (C) 1999-2000 Hewlett-Packard Co
|
||
|
|
* Copyright (C) 1999-2000 David Mosberger-Tang <davidm@hpl.hp.com>
|
||
|
|
*
|
||
|
|
* 64-bit integer division.
|
||
|
|
*
|
||
|
|
* This code is based on the application note entitled "Divide, Square Root
|
||
|
|
* and Remainder Algorithms for the IA-64 Architecture". This document
|
||
|
|
* is available as Intel document number 248725-002 or via the web at
|
||
|
|
* http://developer.intel.com/software/opensource/numerics/
|
||
|
|
*
|
||
|
|
* For more details on the theory behind these algorithms, see "IA-64
|
||
|
|
* and Elementary Functions" by Peter Markstein; HP Professional Books
|
||
|
|
* (http://www.hp.com/go/retailbooks/)
|
||
|
|
*/
|
||
|
|
|
||
|
|
#include <asm/asmmacro.h>
|
||
|
|
|
||
|
|
#ifdef MODULO
|
||
|
|
# define OP mod
|
||
|
|
#else
|
||
|
|
# define OP div
|
||
|
|
#endif
|
||
|
|
|
||
|
|
#ifdef UNSIGNED
|
||
|
|
# define SGN u
|
||
|
|
# define INT_TO_FP(a,b) fcvt.xuf.s1 a=b
|
||
|
|
# define FP_TO_INT(a,b) fcvt.fxu.trunc.s1 a=b
|
||
|
|
#else
|
||
|
|
# define SGN
|
||
|
|
# define INT_TO_FP(a,b) fcvt.xf a=b
|
||
|
|
# define FP_TO_INT(a,b) fcvt.fx.trunc.s1 a=b
|
||
|
|
#endif
|
||
|
|
|
||
|
|
#define PASTE1(a,b) a##b
|
||
|
|
#define PASTE(a,b) PASTE1(a,b)
|
||
|
|
#define NAME PASTE(PASTE(__,SGN),PASTE(OP,di3))
|
||
|
|
|
||
|
|
GLOBAL_ENTRY(NAME)
|
||
|
|
.regstk 2,0,0,0
|
||
|
|
// Transfer inputs to FP registers.
|
||
|
|
setf.sig f8 = in0
|
||
|
|
setf.sig f9 = in1
|
||
|
|
;;
|
||
|
|
// Convert the inputs to FP, to avoid FP software-assist faults.
|
||
|
|
INT_TO_FP(f8, f8)
|
||
|
|
INT_TO_FP(f9, f9)
|
||
|
|
;;
|
||
|
|
frcpa.s1 f11, p6 = f8, f9 // y0 = frcpa(b)
|
||
|
|
;;
|
||
|
|
(p6) fmpy.s1 f7 = f8, f11 // q0 = a*y0
|
||
|
|
(p6) fnma.s1 f6 = f9, f11, f1 // e0 = -b*y0 + 1
|
||
|
|
;;
|
||
|
|
(p6) fma.s1 f10 = f7, f6, f7 // q1 = q0*e0 + q0
|
||
|
|
(p6) fmpy.s1 f7 = f6, f6 // e1 = e0*e0
|
||
|
|
;;
|
||
|
|
#ifdef MODULO
|
||
|
|
sub in1 = r0, in1 // in1 = -b
|
||
|
|
#endif
|
||
|
|
(p6) fma.s1 f10 = f10, f7, f10 // q2 = q1*e1 + q1
|
||
|
|
(p6) fma.s1 f6 = f11, f6, f11 // y1 = y0*e0 + y0
|
||
|
|
;;
|
||
|
|
(p6) fma.s1 f6 = f6, f7, f6 // y2 = y1*e1 + y1
|
||
|
|
(p6) fnma.s1 f7 = f9, f10, f8 // r = -b*q2 + a
|
||
|
|
;;
|
||
|
|
#ifdef MODULO
|
||
|
|
setf.sig f8 = in0 // f8 = a
|
||
|
|
setf.sig f9 = in1 // f9 = -b
|
||
|
|
#endif
|
||
|
|
(p6) fma.s1 f11 = f7, f6, f10 // q3 = r*y2 + q2
|
||
|
|
;;
|
||
|
|
FP_TO_INT(f11, f11) // q = trunc(q3)
|
||
|
|
;;
|
||
|
|
#ifdef MODULO
|
||
|
|
xma.l f11 = f11, f9, f8 // r = q*(-b) + a
|
||
|
|
;;
|
||
|
|
#endif
|
||
|
|
getf.sig r8 = f11 // transfer result to result register
|
||
|
|
br.ret.sptk.many rp
|
||
|
|
END(NAME)
|