Source code of Windows XP (NT5)
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

299 lines
6.4 KiB

//----------------------------------------------------------------------------
// File trans.c
// Author Timothy David Corrie Jr. ([email protected])
// Copyright (C) 1995-96 Microsoft
// Date 01-16-95
//
//
// Description
//
// Contains sin, cos and tan for rationals
//
//
//----------------------------------------------------------------------------
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#if defined( DOS )
#include <dosstub.h>
#else
#include <windows.h>
#endif
#include <ratpak.h>
void scalerat( IN OUT PRAT *pa, IN ANGLE_TYPE angletype )
{
switch ( angletype )
{
case ANGLE_RAD:
scale2pi( pa );
break;
case ANGLE_DEG:
scale( pa, rat_360 );
break;
case ANGLE_GRAD:
scale( pa, rat_400 );
break;
}
}
//-----------------------------------------------------------------------------
//
// FUNCTION: sinrat, _sinrat
//
// ARGUMENTS: x PRAT representation of number to take the sine of
//
// RETURN: sin of x in PRAT form.
//
// EXPLANATION: This uses Taylor series
//
// n
// ___ 2j+1 j
// \ ] X -1
// \ ---------
// / (2j+1)!
// /__]
// j=0
// or,
// n
// ___ 2
// \ ] -X
// \ thisterm ; where thisterm = thisterm * ---------
// / j j+1 j (2j)*(2j+1)
// /__]
// j=0
//
// thisterm = X ; and stop when thisterm < precision used.
// 0 n
//
//-----------------------------------------------------------------------------
void _sinrat( PRAT *px )
{
CREATETAYLOR();
DUPRAT(pret,*px);
DUPRAT(thisterm,*px);
DUPNUM(n2,num_one);
xx->pp->sign *= -1;
do {
NEXTTERM(xx,INC(n2) DIVNUM(n2) INC(n2) DIVNUM(n2));
} while ( !SMALL_ENOUGH_RAT( thisterm ) );
DESTROYTAYLOR();
// Since *px might be epsilon above 1 or below -1, due to TRIMIT we need
// this trick here.
inbetween(px,rat_one);
// Since *px might be epsilon near zero we must set it to zero.
if ( rat_le(*px,rat_smallest) && rat_ge(*px,rat_negsmallest) )
{
DUPRAT(*px,rat_zero);
}
}
void sinrat( PRAT *px )
{
scale2pi(px);
_sinrat(px);
}
void sinanglerat( IN OUT PRAT *pa, IN ANGLE_TYPE angletype )
{
scalerat( pa, angletype );
switch ( angletype )
{
case ANGLE_DEG:
if ( rat_gt( *pa, rat_180 ) )
{
subrat(pa,rat_360);
}
divrat( pa, rat_180 );
mulrat( pa, pi );
break;
case ANGLE_GRAD:
if ( rat_gt( *pa, rat_200 ) )
{
subrat(pa,rat_400);
}
divrat( pa, rat_200 );
mulrat( pa, pi );
break;
}
_sinrat( pa );
}
//-----------------------------------------------------------------------------
//
// FUNCTION: cosrat, _cosrat
//
// ARGUMENTS: x PRAT representation of number to take the cosine of
//
// RETURN: cosin of x in PRAT form.
//
// EXPLANATION: This uses Taylor series
//
// n
// ___ 2j j
// \ ] X -1
// \ ---------
// / (2j)!
// /__]
// j=0
// or,
// n
// ___ 2
// \ ] -X
// \ thisterm ; where thisterm = thisterm * ---------
// / j j+1 j (2j)*(2j+1)
// /__]
// j=0
//
// thisterm = 1 ; and stop when thisterm < precision used.
// 0 n
//
//-----------------------------------------------------------------------------
void _cosrat( PRAT *px )
{
CREATETAYLOR();
pret->pp=longtonum( 1L, nRadix );
pret->pq=longtonum( 1L, nRadix );
DUPRAT(thisterm,pret)
n2=longtonum(0L, nRadix);
xx->pp->sign *= -1;
do {
NEXTTERM(xx,INC(n2) DIVNUM(n2) INC(n2) DIVNUM(n2));
} while ( !SMALL_ENOUGH_RAT( thisterm ) );
DESTROYTAYLOR();
// Since *px might be epsilon above 1 or below -1, due to TRIMIT we need
// this trick here.
inbetween(px,rat_one);
// Since *px might be epsilon near zero we must set it to zero.
if ( rat_le(*px,rat_smallest) && rat_ge(*px,rat_negsmallest) )
{
DUPRAT(*px,rat_zero);
}
}
void cosrat( PRAT *px )
{
scale2pi(px);
_cosrat(px);
}
void cosanglerat( IN OUT PRAT *pa, IN ANGLE_TYPE angletype )
{
scalerat( pa, angletype );
switch ( angletype )
{
case ANGLE_DEG:
if ( rat_gt( *pa, rat_180 ) )
{
PRAT ptmp=NULL;
DUPRAT(ptmp,rat_360);
subrat(&ptmp,*pa);
destroyrat(*pa);
*pa=ptmp;
}
divrat( pa, rat_180 );
mulrat( pa, pi );
break;
case ANGLE_GRAD:
if ( rat_gt( *pa, rat_200 ) )
{
PRAT ptmp=NULL;
DUPRAT(ptmp,rat_400);
subrat(&ptmp,*pa);
destroyrat(*pa);
*pa=ptmp;
}
divrat( pa, rat_200 );
mulrat( pa, pi );
break;
}
_cosrat( pa );
}
//-----------------------------------------------------------------------------
//
// FUNCTION: tanrat, _tanrat
//
// ARGUMENTS: x PRAT representation of number to take the tangent of
//
// RETURN: tan of x in PRAT form.
//
// EXPLANATION: This uses sinrat and cosrat
//
//-----------------------------------------------------------------------------
void _tanrat( PRAT *px )
{
PRAT ptmp=NULL;
DUPRAT(ptmp,*px);
_sinrat(px);
_cosrat(&ptmp);
if ( zerrat( ptmp ) )
{
destroyrat(ptmp);
throw( CALC_E_DOMAIN );
}
divrat(px,ptmp);
destroyrat(ptmp);
}
void tanrat( PRAT *px )
{
scale2pi(px);
_tanrat(px);
}
void tananglerat( IN OUT PRAT *pa, IN ANGLE_TYPE angletype )
{
scalerat( pa, angletype );
switch ( angletype )
{
case ANGLE_DEG:
if ( rat_gt( *pa, rat_180 ) )
{
subrat(pa,rat_180);
}
divrat( pa, rat_180 );
mulrat( pa, pi );
break;
case ANGLE_GRAD:
if ( rat_gt( *pa, rat_200 ) )
{
subrat(pa,rat_200);
}
divrat( pa, rat_200 );
mulrat( pa, pi );
break;
}
_tanrat( pa );
}