JurassicParkTrespasser/jp2_pc/Source/Lib/Renderer/Primitives/AMDK6/WalkEx.hpp

331 lines
8.0 KiB
C++
Raw Permalink Normal View History

2018-01-01 22:07:24 +00:00
/***********************************************************************************************
*
* Copyright <EFBFBD> DreamWorks Interactive. 1996
*
* Contents:
* Overloaded functions for "Walk.hpp" optimized for the Pentium.
* Optimized for the AMD K6-3D Processor
*
* Bugs:
*
* To do:
* Changed the name of the module to 'TexWalkEx.hpp.' Make functions member functions of
* CWalk1D and CWalk2D.
*
***********************************************************************************************
*
* $Log:: /JP2_PC/Source/Lib/Renderer/Primitives/AMDK6/WalkEx.hpp $
*
* 4 97.10.27 1:25p Mmouni
* Initial K6-3D version, copied from the Pentium versions.
*
* 8 9/01/97 8:04p Rwyatt
*
* 7 8/15/97 2:33a Rwyatt
* Fixed local variable problem
*
* 6 8/15/97 12:58a Rwyatt
* Removed local floating point constants and used the global ones.
*
* 5 1/15/97 1:02p Pkeet
* Moved the assembly versions of the 2d initializing functions to WalkEx.hpp. Removed fog
* parameters.
*
* 4 8/12/96 7:15p Pkeet
* Added comments.
*
* 3 7/10/96 1:48p Pkeet
* Added additional comments.
*
* 2 7/09/96 6:38p Pkeet
* Added a '+=' operator for 'CWalk1D.'
*
* 1 7/09/96 6:00p Pkeet
* Initial implementation.
*
**********************************************************************************************/
#ifndef HEADER_LIB_RENDERER_PRIMITIVES_P5_WALKEX_HPP
#define HEADER_LIB_RENDERER_PRIMITIVES_P5_WALKEX_HPP
// Make sure this file is being included from 'FixedP.hpp'.
#ifndef HEADER_LIB_RENDERER_PRIMITIVES_WALK_HPP
#error The header file Walk.hpp has not been included!
#endif
//
// Global functions.
//
//*********************************************************************************************
//
inline void operator+=(CWalk1D& w1d_a, const CWalk1D& w1d_b)
//
// Adds two CWalk1D parameters together and stores the result in the first parameter.
//
// Notes:
// The assembly version is equivalent to the following 'C' code:
//
// w1d_a.bfx.i4Int += w1d_b.bfx.i4Int;
// if (iAddCarry(w1d_a.bfx.u4Frac, w1d_a.bfx.u4Frac, w1d_b.bfx.u4Frac))
// w1d_a.bfx.i4Int += w1d_a.iOffsetPerLine;
//
//**************************************
{
//
// Register usage:
//
// Register Variable
//
// ebx w1d_a.bfx.u4Frac
// ecx w1d_b.bfx.i4Int
// edi w1d_a
// esi w1d_b
//
__asm
{
// Load variables into the registers.
mov edi, w1d_a
mov esi, w1d_b
mov ebx, [edi]CWalk1D.bfxValue.u4Frac
mov ecx, [edi]CWalk1D.bfxValue.i4Int
// Add the integer portions together.
add ecx, [esi]CWalk1D.bfxValue.i4Int
//
// Add the fractional portions together, and add an extra step value if a carry
// results.
//
add ebx, [esi]CWalk1D.bfxValue.u4Frac
jnc short SKIP_STEP_ADD
add ecx, [edi]CWalk1D.iOffsetPerLine
SKIP_STEP_ADD:
// Store the modified variables.
mov [edi]CWalk1D.bfxValue.u4Frac, ebx
mov [edi]CWalk1D.bfxValue.i4Int, ecx
}
}
//*********************************************************************************************
//
inline void operator+=
(
CWalk2D& w2d_a, // First 2D walk parameter, cumulative results stored here.
const CWalk2D& w2d_b // Second 2D walk parameter.
)
//
// Adds two CWalk2D parameters together and stores the result in the first parameter.
//
// Notes:
// This code demonstrates an implementation Michael Abrash's trick of walking through
// a 2D texture map with arbitrary widths and heights. The assembly version is equivalent
// to the following 'C' code:
//
// int i_carry = iAddCarry(w2d_a.uVFrac, w2d_a.uVFrac, w2d_b.uVFrac);
// w2d_a.iUVInt[1] += w2d_b.iUVInt[i_carry + 1] - iAddCarry(w2d_a.uUFrac, w2d_a.uUFrac,
// w2d_b.uUFrac);
//
// The algorithm can be described as follows:
//
// 1. Add the fractional components of V together, and store the carry in a register.
//
// 2. Add the fractional components of U together, and add the carry of the U
// additions to the integer UV portion.
//
// 3. If there was no carry from the V fraction addition, add the combined integer
// components of U and V. If there was a carry from the V fraction addition, add
// the combined integer components of U and V plus a value representing the V line
// offset.
//
//**************************************
{
//
// Register usage:
//
// Register Variable
//
// eax Temporary
// ebx w2d_a.uUFrac
// ecx w2d_a.uVFrac
// edx w2d_a.iUVInt[1]
// edi w2d_a
// esi w2d_b
//
__asm
{
// Load variables into the registers.
mov edi, w2d_a
mov esi, w2d_b
mov ebx, [edi]CWalk2D.uUFrac
mov ecx, [edi]CWalk2D.uVFrac
mov edx, [edi + 4]CWalk2D.iUVInt
// Add the V fractional values together and store the carry.
add ecx, [esi]CWalk2D.uVFrac
sbb eax, eax
// Add the U fractional values, setting the carry flag.
add ebx, [esi]CWalk2D.uUFrac
// Add the integer values from the integer table, and add a possible carry.
adc edx, [esi + eax*4 + 4]CWalk2D.iUVInt
// Store the modified variables.
mov [edi]CWalk2D.uUFrac, ebx
mov [edi]CWalk2D.uVFrac, ecx
mov [edi + 4]CWalk2D.iUVInt, edx
}
}
#if VER_ASM
//*****************************************************************************************
//
inline void Initialize
(
CWalk2D& rw2d,
float f_u, // The 'U' component.
float f_v, // The 'V' component.
int i_width // The width of the texture map.
)
//
// Sets the fractional and integer parts of the CWalk2D object.
//
//**************************************
{
const float fFixed16Scale = 65536.0;
int32 i4_temp;
__asm
{
mov esi, [rw2d]
fld [fFixed16Scale]
fmul [f_u]
fistp [i4_temp]
fld [fFixed16Scale]
fmul [f_v]
mov ecx, [i4_temp]
sar ecx, 16
mov edx, [i4_temp]
fistp [i4_temp]
shl edx, 16
mov eax, [i4_temp]
sar eax, 16
mov [esi]CWalk2D.uUFrac, edx
imul [i_width]
mov ebx, [i4_temp]
shl ebx, 16
add eax, ecx
mov [esi]CWalk2D.uVFrac, ebx
mov [esi + 4]CWalk2D.iUVInt, eax
add eax, i_width
mov [esi]CWalk2D.iUVInt, eax
}
}
//*****************************************************************************************
//
inline void InitializePos
(
CWalk2D& rw2d,
float f_u, // The 'U' component.
float f_v, // The 'V' component.
int i_width // The width of the texture map.
)
//
// Sets the fractional and integer parts of the CWalk2D object.
//
//**************************************
{
#define iBITS_ACCURACY 11
const float fFastFixed16Conversion = float(1 << (23 - iBITS_ACCURACY));
int32 i4_temp_a;
int32 i4_temp_b;
__asm
{
mov esi, [rw2d]
fld [fFastFixed16Conversion]
fld st(0)
fadd [f_u]
fstp float ptr[i4_temp_a]
mov ecx, [i4_temp_a]
fadd [f_v]
sar ecx, iBITS_ACCURACY
fstp [i4_temp_b]
and ecx, 0x007FFFFF >> iBITS_ACCURACY
mov edx, [i4_temp_a]
shl edx, 32 - iBITS_ACCURACY
mov eax, [i4_temp_b]
sar eax, iBITS_ACCURACY
and eax, 0x007FFFFF >> iBITS_ACCURACY
mov [esi]CWalk2D.uUFrac, edx
imul [i_width]
mov ebx, [i4_temp_b]
shl ebx, 32 - iBITS_ACCURACY
add eax, ecx
mov [esi]CWalk2D.uVFrac, ebx
mov [esi + 4]CWalk2D.iUVInt, eax
//add eax, i_width
//mov [esi]CWalk2D.iUVInt, eax
}
}
#endif // VER_ASM
//
// Additional notes.
//
/**********************************************************************************************
*
* This function is included only to illustrate how the 'C' versions of the overloaded '+='
* operators for CWalk1D and CWalk2D classes work. Reference is made in the notes of these
* respective functions to this 'iAddCarry' function.
*
* inline int iAddCarry(uint& u_result, uint i_a, uint i_b)
* {
* // Add the two values together.
* u_result = i_a + i_b;
*
* // Return -1 if a carry is present.
* if ((u_result < i_a) || (u_result < i_b))
* return -1;
*
* // Return 0 if there is no carry.
* return 0;
* }
*
**********************************************************************************************/
#endif