qzdoom/src/m_fixed.h
2016-03-01 09:47:10 -06:00

152 lines
3.4 KiB
C

// "Build Engine & Tools" Copyright (c) 1993-1997 Ken Silverman
// Ken Silverman's official web site: "http://www.advsys.net/ken"
// See the included license file "BUILDLIC.TXT" for license info.
//
// This file is based on pragmas.h from Ken Silverman's original Build
// source code release and contains routines that were originally
// inline assembly but are not now.
#ifndef __M_FIXED__
#define __M_FIXED__
#include <stdlib.h>
#include "doomtype.h"
#if defined(__GNUC__) && defined(__i386__) && !defined(__clang__)
#include "gccinlines.h"
#elif defined(_MSC_VER) && defined(_M_IX86)
#include "mscinlines.h"
#else
#include "basicinlines.h"
#endif
#include "xs_Float.h"
#define MAKESAFEDIVSCALE(x) \
inline SDWORD SafeDivScale##x (SDWORD a, SDWORD b) \
{ \
if ((DWORD)abs(a) >> (31-x) >= (DWORD)abs (b)) \
return (a^b)<0 ? FIXED_MIN : FIXED_MAX; \
return DivScale##x (a, b); \
}
MAKESAFEDIVSCALE(1)
MAKESAFEDIVSCALE(2)
MAKESAFEDIVSCALE(3)
MAKESAFEDIVSCALE(4)
MAKESAFEDIVSCALE(5)
MAKESAFEDIVSCALE(6)
MAKESAFEDIVSCALE(7)
MAKESAFEDIVSCALE(8)
MAKESAFEDIVSCALE(9)
MAKESAFEDIVSCALE(10)
MAKESAFEDIVSCALE(11)
MAKESAFEDIVSCALE(12)
MAKESAFEDIVSCALE(13)
MAKESAFEDIVSCALE(14)
MAKESAFEDIVSCALE(15)
MAKESAFEDIVSCALE(16)
MAKESAFEDIVSCALE(17)
MAKESAFEDIVSCALE(18)
MAKESAFEDIVSCALE(19)
MAKESAFEDIVSCALE(20)
MAKESAFEDIVSCALE(21)
MAKESAFEDIVSCALE(22)
MAKESAFEDIVSCALE(23)
MAKESAFEDIVSCALE(24)
MAKESAFEDIVSCALE(25)
MAKESAFEDIVSCALE(26)
MAKESAFEDIVSCALE(27)
MAKESAFEDIVSCALE(28)
MAKESAFEDIVSCALE(29)
MAKESAFEDIVSCALE(30)
#undef MAKESAFEDIVSCALE
inline SDWORD SafeDivScale31 (SDWORD a, SDWORD b)
{
if ((DWORD)abs(a) >= (DWORD)abs (b))
return (a^b)<0 ? FIXED_MIN : FIXED_MAX;
return DivScale31 (a, b);
}
inline SDWORD SafeDivScale32 (SDWORD a, SDWORD b)
{
if ((DWORD)abs(a) >= (DWORD)abs (b) >> 1)
return (a^b)<0 ? FIXED_MIN : FIXED_MAX;
return DivScale32 (a, b);
}
#define FixedMul MulScale16
#define FixedDiv SafeDivScale16
inline void qinterpolatedown16 (SDWORD *out, DWORD count, SDWORD val, SDWORD delta)
{
if (count & 1)
{
out[0] = val >> 16;
val += delta;
}
count >>= 1;
while (count-- != 0)
{
int temp = val + delta;
out[0] = val >> 16;
val = temp + delta;
out[1] = temp >> 16;
out += 2;
}
}
inline void qinterpolatedown16short (short *out, DWORD count, SDWORD val, SDWORD delta)
{
if (count)
{
if ((size_t)out & 2)
{ // align to dword boundary
*out++ = (short)(val >> 16);
count--;
val += delta;
}
DWORD *o2 = (DWORD *)out;
DWORD c2 = count>>1;
while (c2-- != 0)
{
SDWORD temp = val + delta;
*o2++ = (temp & 0xffff0000) | ((DWORD)val >> 16);
val = temp + delta;
}
if (count & 1)
{
*(short *)o2 = (short)(val >> 16);
}
}
}
//returns num/den, dmval = num%den
inline SDWORD DivMod (SDWORD num, SDWORD den, SDWORD *dmval)
{
*dmval = num % den;
return num / den;
}
//returns num%den, dmval = num/den
inline SDWORD ModDiv (SDWORD num, SDWORD den, SDWORD *dmval)
{
*dmval = num / den;
return num % den;
}
#define FLOAT2FIXED(f) ((fixed_t)xs_Fix<16>::ToFix(f))
#define FIXED2FLOAT(f) ((f) / float(65536))
#define FIXED2DBL(f) ((f) / double(65536))
#define ANGLE2DBL(f) ((f) * (90./ANGLE_90))
#define ANGLE2FLOAT(f) (float((f) * (90./ANGLE_90)))
#define FLOAT2ANGLE(f) ((angle_t)xs_CRoundToInt((f) * (ANGLE_90/90.)))
#define ANGLE2RAD(f) ((f) * (M_PI/ANGLE_180))
#define ANGLE2RADF(f) ((f) * float(M_PI/ANGLE_180))
#define RAD2ANGLE(f) ((angle_t)xs_CRoundToInt((f) * (ANGLE_180/M_PI)))
#endif