902 lines
30 KiB
C
902 lines
30 KiB
C
//========== Copyright (c) Valve Corporation, All rights reserved. ==========//
|
|
//
|
|
// Purpose: This is where all common code for vertex shaders go.
|
|
//
|
|
// $NoKeywords: $
|
|
//
|
|
//===========================================================================//
|
|
|
|
|
|
|
|
#ifndef COMMON_VS_FXC_H_
|
|
#define COMMON_VS_FXC_H_
|
|
|
|
#include "common_fxc.h"
|
|
|
|
// Put global skip commands here. . make sure and check that the appropriate vars are defined
|
|
// so these aren't used on the wrong shaders!
|
|
// --------------------------------------------------------------------------------
|
|
// Ditch all fastpath attemps if we are doing LIGHTING_PREVIEW.
|
|
// SKIP: defined $LIGHTING_PREVIEW && defined $FASTPATH && $LIGHTING_PREVIEW && $FASTPATH
|
|
// --------------------------------------------------------------------------------
|
|
|
|
|
|
#ifndef COMPRESSED_VERTS
|
|
// Default to no vertex compression
|
|
#define COMPRESSED_VERTS 0
|
|
#endif
|
|
|
|
#if ( !defined( SHADER_MODEL_VS_2_0 ) && !defined( SHADER_MODEL_VS_3_0 ) )
|
|
#if COMPRESSED_VERTS == 1
|
|
#error "Vertex compression is only for DX9 and up!"
|
|
#endif
|
|
#endif
|
|
|
|
// We're testing 2 normal compression methods
|
|
// One compressed normals+tangents into a SHORT2 each (8 bytes total)
|
|
// The other compresses them together, into a single UBYTE4 (4 bytes total)
|
|
// FIXME: pick one or the other, compare lighting quality in important cases
|
|
#define COMPRESSED_NORMALS_SEPARATETANGENTS_SHORT2 0
|
|
#define COMPRESSED_NORMALS_COMBINEDTANGENTS_UBYTE4 1
|
|
//#define COMPRESSED_NORMALS_TYPE COMPRESSED_NORMALS_SEPARATETANGENTS_SHORT2
|
|
#define COMPRESSED_NORMALS_TYPE COMPRESSED_NORMALS_COMBINEDTANGENTS_UBYTE4
|
|
|
|
|
|
#define FOGTYPE_RANGE 0
|
|
#define FOGTYPE_HEIGHT 1
|
|
|
|
#define COMPILE_ERROR ( 1/0; )
|
|
|
|
// -------------------------
|
|
// CONSTANTS
|
|
// -------------------------
|
|
|
|
#pragma def ( vs, c0, 0.0f, 1.0f, 2.0f, 0.5f )
|
|
|
|
const float4 cConstants1 : register(c1);
|
|
#define cOOGamma cConstants1.x
|
|
#define cOverbright 2.0f
|
|
#define cOneThird cConstants1.z
|
|
#define cOOOverbright ( 1.0f / 2.0f )
|
|
|
|
|
|
// The g_bLightEnabled registers and g_nLightCountRegister hold the same information regarding
|
|
// enabling lights, but callers internal to this file tend to use the loops, while external
|
|
// callers will end up using the booleans
|
|
const bool g_bLightEnabled[4] : register(b0);
|
|
// through b3
|
|
|
|
const int g_nLightCountRegister : register(i0);
|
|
|
|
|
|
#define g_nLightCount g_nLightCountRegister.x
|
|
|
|
const float4 cEyePos_WaterHeightW : register(c2);
|
|
#define cEyePos cEyePos_WaterHeightW.xyz
|
|
#define cWaterHeight cEyePos_WaterHeightW.w
|
|
|
|
// This is still used by asm stuff.
|
|
const float4 cObsoleteLightIndex : register(c3);
|
|
|
|
const float4x4 cModelViewProj : register(c4);
|
|
const float4x4 cViewProj : register(c8);
|
|
|
|
// Only cFlexScale.x is used
|
|
// It is a binary value used to switch on/off the addition of the flex delta stream
|
|
const float4 cFlexScale : register(c13);
|
|
|
|
const float4 cFogParams : register(c16);
|
|
#define cFogEndOverFogRange cFogParams.x
|
|
// cFogParams.y unused
|
|
#define cRadialFogMaxDensity cFogParams.z //radial fog max density in fractional portion. height fog max density stored in integer portion and is multiplied by 1e10
|
|
#define cOOFogRange cFogParams.w
|
|
|
|
const float4x4 cViewModel : register(c17);
|
|
|
|
const float3 cAmbientCubeX [ 2 ] : register ( c21 ) ;
|
|
const float3 cAmbientCubeY [ 2 ] : register ( c23 ) ;
|
|
const float3 cAmbientCubeZ [ 2 ] : register ( c25 ) ;
|
|
|
|
#ifdef SHADER_MODEL_VS_3_0
|
|
const float4 cFlexWeights [ 512 ] : register ( c1024 ) ;
|
|
#endif
|
|
|
|
struct LightInfo
|
|
{
|
|
float4 color; // {xyz} is color w is light type code (see comment below)
|
|
float4 dir; // {xyz} is dir w is light type code
|
|
float4 pos;
|
|
float4 spotParams;
|
|
float4 atten;
|
|
};
|
|
|
|
// w components of color and dir indicate light type:
|
|
// 1x - directional
|
|
// 01 - spot
|
|
// 00 - point
|
|
|
|
// Four lights x 5 constants each = 20 constants
|
|
LightInfo cLightInfo[4] : register(c27);
|
|
#define LIGHT_0_POSITION_REG c29
|
|
|
|
#ifdef SHADER_MODEL_VS_1_1
|
|
|
|
const float4 cModulationColor : register(c37);
|
|
|
|
#define SHADER_SPECIFIC_CONST_0 c38
|
|
#define SHADER_SPECIFIC_CONST_1 c39
|
|
#define SHADER_SPECIFIC_CONST_2 c40
|
|
#define SHADER_SPECIFIC_CONST_3 c41
|
|
#define SHADER_SPECIFIC_CONST_4 c42
|
|
#define SHADER_SPECIFIC_CONST_5 c43
|
|
#define SHADER_SPECIFIC_CONST_6 c44
|
|
#define SHADER_SPECIFIC_CONST_7 c45
|
|
#define SHADER_SPECIFIC_CONST_8 c46
|
|
#define SHADER_SPECIFIC_CONST_9 c47
|
|
#define SHADER_SPECIFIC_CONST_10 c14
|
|
#define SHADER_SPECIFIC_CONST_11 c15
|
|
|
|
static const int cModel0Index = 48;
|
|
const float4x3 cModel[16] : register(c48);
|
|
// last cmodel is c105 for dx80, c214 for dx90
|
|
|
|
#else // DX9 shaders (vs20 and beyond)
|
|
|
|
const float4 cModulationColor : register( c47 );
|
|
|
|
#define SHADER_SPECIFIC_CONST_0 c48
|
|
#define SHADER_SPECIFIC_CONST_1 c49
|
|
#define SHADER_SPECIFIC_CONST_2 c50
|
|
#define SHADER_SPECIFIC_CONST_3 c51
|
|
#define SHADER_SPECIFIC_CONST_4 c52
|
|
#define SHADER_SPECIFIC_CONST_5 c53
|
|
#define SHADER_SPECIFIC_CONST_6 c54
|
|
#define SHADER_SPECIFIC_CONST_7 c55
|
|
#define SHADER_SPECIFIC_CONST_8 c56
|
|
#define SHADER_SPECIFIC_CONST_9 c57
|
|
#define SHADER_SPECIFIC_CONST_10 c14
|
|
#define SHADER_SPECIFIC_CONST_11 c15
|
|
#define SHADER_SPECIFIC_CONST_12 c12
|
|
|
|
static const int cModel0Index = 58;
|
|
const float4x3 cModel[53] : register( c58 );
|
|
// last cmodel is c105 for dx80, c214 for dx90
|
|
|
|
#define SHADER_SPECIFIC_BOOL_CONST_0 b4
|
|
#define SHADER_SPECIFIC_BOOL_CONST_1 b5
|
|
#define SHADER_SPECIFIC_BOOL_CONST_2 b6
|
|
#define SHADER_SPECIFIC_BOOL_CONST_3 b7
|
|
#define SHADER_SPECIFIC_BOOL_CONST_4 b8
|
|
#define SHADER_SPECIFIC_BOOL_CONST_5 b9
|
|
#define SHADER_SPECIFIC_BOOL_CONST_6 b10
|
|
#define SHADER_SPECIFIC_BOOL_CONST_7 b11
|
|
#endif // vertex shader model constant packing changes
|
|
|
|
|
|
//=======================================================================================
|
|
// Methods to decompress vertex normals
|
|
//=======================================================================================
|
|
|
|
//-----------------------------------------------------------------------------------
|
|
// Decompress a normal from two-component compressed format
|
|
// We expect this data to come from a signed SHORT2 stream in the range of -32768..32767
|
|
//
|
|
// -32678 and 0 are invalid encodings
|
|
// w contains the sign to use in the cross product when generating a binormal
|
|
void _DecompressShort2Tangent( float2 inputTangent, out float4 outputTangent )
|
|
{
|
|
float2 ztSigns = sign( inputTangent ); // sign bits for z and tangent (+1 or -1)
|
|
float2 xyAbs = abs( inputTangent ); // 1..32767
|
|
outputTangent.xy = (xyAbs - 16384.0f) / 16384.0f; // x and y
|
|
outputTangent.z = ztSigns.x * sqrt( saturate( 1.0f - dot( outputTangent.xy, outputTangent.xy ) ) );
|
|
outputTangent.w = ztSigns.y;
|
|
}
|
|
|
|
//-----------------------------------------------------------------------------------
|
|
// Same code as _DecompressShort2Tangent, just one returns a float4, one a float3
|
|
void _DecompressShort2Normal( float2 inputNormal, out float3 outputNormal )
|
|
{
|
|
float4 result;
|
|
_DecompressShort2Tangent( inputNormal, result );
|
|
outputNormal = result.xyz;
|
|
}
|
|
|
|
//-----------------------------------------------------------------------------------
|
|
// Decompress normal+tangent together
|
|
void _DecompressShort2NormalTangent( float2 inputNormal, float2 inputTangent, out float3 outputNormal, out float4 outputTangent )
|
|
{
|
|
// FIXME: if we end up sticking with the SHORT2 format, pack the normal and tangent into a single SHORT4 element
|
|
// (that would make unpacking normal+tangent here together much cheaper than the sum of their parts)
|
|
_DecompressShort2Normal( inputNormal, outputNormal );
|
|
_DecompressShort2Tangent( inputTangent, outputTangent );
|
|
}
|
|
|
|
//=======================================================================================
|
|
// Decompress a normal and tangent from four-component compressed format
|
|
// We expect this data to come from an unsigned UBYTE4 stream in the range of 0..255
|
|
// The final vTangent.w contains the sign to use in the cross product when generating a binormal
|
|
void _DecompressUByte4NormalTangent( float4 inputNormal,
|
|
out float3 outputNormal, // {nX, nY, nZ}
|
|
out float4 outputTangent ) // {tX, tY, tZ, sign of binormal}
|
|
{
|
|
float fOne = 1.0f;
|
|
|
|
float4 ztztSignBits = ( inputNormal - 128.0f ) < 0; // sign bits for zs and binormal (1 or 0) set-less-than (slt) asm instruction
|
|
float4 xyxyAbs = abs( inputNormal - 128.0f ) - ztztSignBits; // 0..127
|
|
float4 xyxySignBits = ( xyxyAbs - 64.0f ) < 0; // sign bits for xs and ys (1 or 0)
|
|
float4 normTan = (abs( xyxyAbs - 64.0f ) - xyxySignBits) / 63.0f; // abs({nX, nY, tX, tY})
|
|
outputNormal.xy = normTan.xy; // abs({nX, nY, __, __})
|
|
outputTangent.xy = normTan.zw; // abs({tX, tY, __, __})
|
|
|
|
float4 xyxySigns = 1 - 2*xyxySignBits; // Convert sign bits to signs
|
|
float4 ztztSigns = 1 - 2*ztztSignBits; // ( [1,0] -> [-1,+1] )
|
|
|
|
outputNormal.z = 1.0f - outputNormal.x - outputNormal.y; // Project onto x+y+z=1
|
|
outputNormal.xyz = normalize( outputNormal.xyz ); // Normalize onto unit sphere
|
|
outputNormal.xy *= xyxySigns.xy; // Restore x and y signs
|
|
outputNormal.z *= ztztSigns.x; // Restore z sign
|
|
|
|
outputTangent.z = 1.0f - outputTangent.x - outputTangent.y; // Project onto x+y+z=1
|
|
outputTangent.xyz = normalize( outputTangent.xyz ); // Normalize onto unit sphere
|
|
outputTangent.xy *= xyxySigns.zw; // Restore x and y signs
|
|
outputTangent.z *= ztztSigns.z; // Restore z sign
|
|
outputTangent.w = ztztSigns.w; // Binormal sign
|
|
}
|
|
|
|
|
|
//-----------------------------------------------------------------------------------
|
|
// Decompress just a normal from four-component compressed format (same as above)
|
|
// We expect this data to come from an unsigned UBYTE4 stream in the range of 0..255
|
|
// [ When compiled, this works out to approximately 17 asm instructions ]
|
|
void _DecompressUByte4Normal( float4 inputNormal,
|
|
out float3 outputNormal) // {nX, nY, nZ}
|
|
{
|
|
float fOne = 1.0f;
|
|
|
|
float2 ztSigns = ( inputNormal.xy - 128.0f ) < 0; // sign bits for zs and binormal (1 or 0) set-less-than (slt) asm instruction
|
|
float2 xyAbs = abs( inputNormal.xy - 128.0f ) - ztSigns; // 0..127
|
|
float2 xySigns = ( xyAbs - 64.0f ) < 0; // sign bits for xs and ys (1 or 0)
|
|
outputNormal.xy = ( abs( xyAbs - 64.0f ) - xySigns ) / 63.0f; // abs({nX, nY})
|
|
|
|
outputNormal.z = 1.0f - outputNormal.x - outputNormal.y; // Project onto x+y+z=1
|
|
outputNormal.xyz = normalize( outputNormal.xyz ); // Normalize onto unit sphere
|
|
|
|
outputNormal.xy *= lerp( fOne.xx, -fOne.xx, xySigns ); // Restore x and y signs
|
|
outputNormal.z *= lerp( fOne.x, -fOne.x, ztSigns.x ); // Restore z sign
|
|
}
|
|
|
|
|
|
void DecompressVertex_Normal( float4 inputNormal, out float3 outputNormal )
|
|
{
|
|
if ( COMPRESSED_VERTS == 1 )
|
|
{
|
|
if ( COMPRESSED_NORMALS_TYPE == COMPRESSED_NORMALS_SEPARATETANGENTS_SHORT2 )
|
|
{
|
|
_DecompressShort2Normal( inputNormal.xy, outputNormal );
|
|
}
|
|
else // ( COMPRESSED_NORMALS_TYPE == COMPRESSED_NORMALS_COMBINEDTANGENTS_UBYTE4 )
|
|
{
|
|
_DecompressUByte4Normal( inputNormal, outputNormal );
|
|
}
|
|
}
|
|
else
|
|
{
|
|
outputNormal = inputNormal.xyz;
|
|
}
|
|
}
|
|
|
|
void DecompressVertex_NormalTangent( float4 inputNormal, float4 inputTangent, out float3 outputNormal, out float4 outputTangent )
|
|
{
|
|
if ( COMPRESSED_VERTS == 1 )
|
|
{
|
|
if ( COMPRESSED_NORMALS_TYPE == COMPRESSED_NORMALS_SEPARATETANGENTS_SHORT2 )
|
|
{
|
|
_DecompressShort2NormalTangent( inputNormal.xy, inputTangent.xy, outputNormal, outputTangent );
|
|
}
|
|
else // ( COMPRESSED_NORMALS_TYPE == COMPRESSED_NORMALS_COMBINEDTANGENTS_UBYTE4 )
|
|
{
|
|
_DecompressUByte4NormalTangent( inputNormal, outputNormal, outputTangent );
|
|
}
|
|
}
|
|
else
|
|
{
|
|
outputNormal = inputNormal.xyz;
|
|
outputTangent = inputTangent;
|
|
}
|
|
}
|
|
|
|
|
|
#ifdef SHADER_MODEL_VS_3_0
|
|
|
|
//-----------------------------------------------------------------------------
|
|
// Methods to sample morph data from a vertex texture
|
|
// NOTE: vMorphTargetTextureDim.x = width, cVertexTextureDim.y = height, cVertexTextureDim.z = # of float4 fields per vertex
|
|
// For position + normal morph for example, there will be 2 fields.
|
|
//-----------------------------------------------------------------------------
|
|
float4 SampleMorphDelta( sampler2D vt, const float3 vMorphTargetTextureDim, const float4 vMorphSubrect, const float flVertexID, const float flField )
|
|
{
|
|
float flColumn = floor( flVertexID / vMorphSubrect.w );
|
|
|
|
float4 t;
|
|
t.x = vMorphSubrect.x + vMorphTargetTextureDim.z * flColumn + flField + 0.5f;
|
|
t.y = vMorphSubrect.y + flVertexID - flColumn * vMorphSubrect.w + 0.5f;
|
|
t.xy /= vMorphTargetTextureDim.xy;
|
|
t.z = t.w = 0.f;
|
|
|
|
return tex2Dlod( vt, t );
|
|
}
|
|
|
|
// Optimized version which reads 2 deltas
|
|
void SampleMorphDelta2( sampler2D vt, const float3 vMorphTargetTextureDim, const float4 vMorphSubrect, const float flVertexID, out float4 delta1, out float4 delta2 )
|
|
{
|
|
float flColumn = floor( flVertexID / vMorphSubrect.w );
|
|
|
|
float4 t;
|
|
t.x = vMorphSubrect.x + vMorphTargetTextureDim.z * flColumn + 0.5f;
|
|
t.y = vMorphSubrect.y + flVertexID - flColumn * vMorphSubrect.w + 0.5f;
|
|
t.xy /= vMorphTargetTextureDim.xy;
|
|
t.z = t.w = 0.f;
|
|
|
|
delta1 = tex2Dlod( vt, t );
|
|
t.x += 1.0f / vMorphTargetTextureDim.x;
|
|
delta2 = tex2Dlod( vt, t );
|
|
}
|
|
|
|
#endif // SHADER_MODEL_VS_3_0
|
|
|
|
//-----------------------------------------------------------------------------
|
|
// Method to apply morphs
|
|
//-----------------------------------------------------------------------------
|
|
bool ApplyMorph( float3 vPosFlex, inout float3 vPosition )
|
|
{
|
|
// Flexes coming in from a separate stream
|
|
float3 vPosDelta = vPosFlex.xyz * cFlexScale.x;
|
|
vPosition.xyz += vPosDelta;
|
|
return true;
|
|
}
|
|
|
|
bool ApplyMorph( float3 vPosFlex, float3 vNormalFlex, inout float3 vPosition, inout float3 vNormal )
|
|
{
|
|
// Flexes coming in from a separate stream
|
|
float3 vPosDelta = vPosFlex.xyz * cFlexScale.x;
|
|
float3 vNormalDelta = vNormalFlex.xyz * cFlexScale.x;
|
|
vPosition.xyz += vPosDelta;
|
|
vNormal += vNormalDelta;
|
|
return true;
|
|
}
|
|
|
|
bool ApplyMorph( float3 vPosFlex, float3 vNormalFlex,
|
|
inout float3 vPosition, inout float3 vNormal, inout float3 vTangent )
|
|
{
|
|
// Flexes coming in from a separate stream
|
|
float3 vPosDelta = vPosFlex.xyz * cFlexScale.x;
|
|
float3 vNormalDelta = vNormalFlex.xyz * cFlexScale.x;
|
|
vPosition.xyz += vPosDelta;
|
|
vNormal += vNormalDelta;
|
|
vTangent.xyz += vNormalDelta;
|
|
return true;
|
|
}
|
|
|
|
bool ApplyMorph( float4 vPosFlex, float3 vNormalFlex,
|
|
inout float3 vPosition, inout float3 vNormal, inout float3 vTangent, out float flWrinkle )
|
|
{
|
|
// Flexes coming in from a separate stream
|
|
float3 vPosDelta = vPosFlex.xyz * cFlexScale.x;
|
|
float3 vNormalDelta = vNormalFlex.xyz * cFlexScale.x;
|
|
flWrinkle = vPosFlex.w * cFlexScale.y;
|
|
vPosition.xyz += vPosDelta;
|
|
vNormal += vNormalDelta;
|
|
vTangent.xyz += vNormalDelta;
|
|
return true;
|
|
}
|
|
|
|
#ifdef SHADER_MODEL_VS_3_0
|
|
|
|
bool ApplyMorph( sampler2D morphSampler, const float3 vMorphTargetTextureDim, const float4 vMorphSubrect,
|
|
const float flVertexID, const float3 vMorphTexCoord,
|
|
inout float3 vPosition )
|
|
{
|
|
#if MORPHING
|
|
|
|
#if !DECAL
|
|
// Flexes coming in from a separate stream
|
|
float4 vPosDelta = SampleMorphDelta( morphSampler, vMorphTargetTextureDim, vMorphSubrect, flVertexID, 0 );
|
|
vPosition += vPosDelta.xyz;
|
|
#else
|
|
float4 t = float4( vMorphTexCoord.x, vMorphTexCoord.y, 0.0f, 0.0f );
|
|
float3 vPosDelta = tex2Dlod( morphSampler, t );
|
|
vPosition += vPosDelta.xyz * vMorphTexCoord.z;
|
|
#endif // DECAL
|
|
|
|
return true;
|
|
|
|
#else // !MORPHING
|
|
return false;
|
|
#endif
|
|
}
|
|
|
|
bool ApplyMorph( sampler2D morphSampler, const float3 vMorphTargetTextureDim, const float4 vMorphSubrect,
|
|
const float flVertexID, const float3 vMorphTexCoord,
|
|
inout float3 vPosition, inout float3 vNormal )
|
|
{
|
|
#if MORPHING
|
|
|
|
#if !DECAL
|
|
float4 vPosDelta, vNormalDelta;
|
|
SampleMorphDelta2( morphSampler, vMorphTargetTextureDim, vMorphSubrect, flVertexID, vPosDelta, vNormalDelta );
|
|
vPosition += vPosDelta.xyz;
|
|
vNormal += vNormalDelta.xyz;
|
|
#else
|
|
float4 t = float4( vMorphTexCoord.x, vMorphTexCoord.y, 0.0f, 0.0f );
|
|
float3 vPosDelta = tex2Dlod( morphSampler, t );
|
|
t.x += 1.0f / vMorphTargetTextureDim.x;
|
|
float3 vNormalDelta = tex2Dlod( morphSampler, t );
|
|
vPosition += vPosDelta.xyz * vMorphTexCoord.z;
|
|
vNormal += vNormalDelta.xyz * vMorphTexCoord.z;
|
|
#endif // DECAL
|
|
|
|
return true;
|
|
|
|
#else // !MORPHING
|
|
return false;
|
|
#endif
|
|
}
|
|
|
|
bool ApplyMorph( sampler2D morphSampler, const float3 vMorphTargetTextureDim, const float4 vMorphSubrect,
|
|
const float flVertexID, const float3 vMorphTexCoord,
|
|
inout float3 vPosition, inout float3 vNormal, inout float3 vTangent )
|
|
{
|
|
#if MORPHING
|
|
|
|
#if !DECAL
|
|
float4 vPosDelta, vNormalDelta;
|
|
SampleMorphDelta2( morphSampler, vMorphTargetTextureDim, vMorphSubrect, flVertexID, vPosDelta, vNormalDelta );
|
|
vPosition += vPosDelta.xyz;
|
|
vNormal += vNormalDelta.xyz;
|
|
vTangent += vNormalDelta.xyz;
|
|
#else
|
|
float4 t = float4( vMorphTexCoord.x, vMorphTexCoord.y, 0.0f, 0.0f );
|
|
float3 vPosDelta = tex2Dlod( morphSampler, t );
|
|
t.x += 1.0f / vMorphTargetTextureDim.x;
|
|
float3 vNormalDelta = tex2Dlod( morphSampler, t );
|
|
vPosition += vPosDelta.xyz * vMorphTexCoord.z;
|
|
vNormal += vNormalDelta.xyz * vMorphTexCoord.z;
|
|
vTangent += vNormalDelta.xyz * vMorphTexCoord.z;
|
|
#endif // DECAL
|
|
|
|
return true;
|
|
|
|
#else // MORPHING
|
|
|
|
return false;
|
|
#endif
|
|
}
|
|
|
|
bool ApplyMorph( sampler2D morphSampler, const float3 vMorphTargetTextureDim, const float4 vMorphSubrect,
|
|
const float flVertexID, const float3 vMorphTexCoord,
|
|
inout float3 vPosition, inout float3 vNormal, inout float3 vTangent, out float flWrinkle )
|
|
{
|
|
#if MORPHING
|
|
|
|
#if !DECAL
|
|
float4 vPosDelta, vNormalDelta;
|
|
SampleMorphDelta2( morphSampler, vMorphTargetTextureDim, vMorphSubrect, flVertexID, vPosDelta, vNormalDelta );
|
|
vPosition += vPosDelta.xyz;
|
|
vNormal += vNormalDelta.xyz;
|
|
vTangent += vNormalDelta.xyz;
|
|
flWrinkle = vPosDelta.w;
|
|
#else
|
|
float4 t = float4( vMorphTexCoord.x, vMorphTexCoord.y, 0.0f, 0.0f );
|
|
float4 vPosDelta = tex2Dlod( morphSampler, t );
|
|
t.x += 1.0f / vMorphTargetTextureDim.x;
|
|
float3 vNormalDelta = tex2Dlod( morphSampler, t );
|
|
|
|
vPosition += vPosDelta.xyz * vMorphTexCoord.z;
|
|
vNormal += vNormalDelta.xyz * vMorphTexCoord.z;
|
|
vTangent += vNormalDelta.xyz * vMorphTexCoord.z;
|
|
flWrinkle = vPosDelta.w * vMorphTexCoord.z;
|
|
#endif // DECAL
|
|
|
|
return true;
|
|
|
|
#else // MORPHING
|
|
|
|
flWrinkle = 0.0f;
|
|
return false;
|
|
|
|
#endif
|
|
}
|
|
|
|
#endif // SHADER_MODEL_VS_3_0
|
|
|
|
float CalcFixedFunctionFog( const float3 worldPos, const bool bWaterFog )
|
|
{
|
|
if( !bWaterFog )
|
|
{
|
|
return CalcRangeFogFactorFixedFunction( worldPos, cEyePos, cRadialFogMaxDensity, cFogEndOverFogRange, cOOFogRange );
|
|
}
|
|
else
|
|
{
|
|
return 0.0f; //all done in the pixel shader as of ps20 (current min-spec)
|
|
}
|
|
}
|
|
|
|
float CalcFixedFunctionFog( const float3 worldPos, const int fogType )
|
|
{
|
|
return CalcFixedFunctionFog( worldPos, fogType != FOGTYPE_RANGE );
|
|
}
|
|
|
|
float CalcNonFixedFunctionFog( const float3 worldPos, const bool bWaterFog )
|
|
{
|
|
if( !bWaterFog )
|
|
{
|
|
return CalcRangeFogFactorNonFixedFunction( worldPos, cEyePos, cRadialFogMaxDensity, cFogEndOverFogRange, cOOFogRange );
|
|
}
|
|
else
|
|
{
|
|
return 0.0f; //all done in the pixel shader as of ps20 (current min-spec)
|
|
}
|
|
}
|
|
|
|
float CalcNonFixedFunctionFog( const float3 worldPos, const int fogType )
|
|
{
|
|
return CalcNonFixedFunctionFog( worldPos, fogType != FOGTYPE_RANGE );
|
|
}
|
|
|
|
float4 DecompressBoneWeights( const float4 weights )
|
|
{
|
|
float4 result = weights;
|
|
|
|
if ( COMPRESSED_VERTS )
|
|
{
|
|
// Decompress from SHORT2 to float. In our case, [-1, +32767] -> [0, +1]
|
|
// NOTE: we add 1 here so we can divide by 32768 - which is exact (divide by 32767 is not).
|
|
// This avoids cracking between meshes with different numbers of bone weights.
|
|
// We use SHORT2 instead of SHORT2N for a similar reason - the GPU's conversion
|
|
// from [-32768,+32767] to [-1,+1] is imprecise in the same way.
|
|
result += 1;
|
|
result /= 32768;
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
void SkinPosition( bool bSkinning, const float4 modelPos,
|
|
const float4 boneWeights, float4 fBoneIndices,
|
|
out float3 worldPos )
|
|
{
|
|
#if !defined( _X360 )
|
|
int3 boneIndices = D3DCOLORtoUBYTE4( fBoneIndices );
|
|
#else
|
|
int3 boneIndices = fBoneIndices;
|
|
#endif
|
|
|
|
// Needed for invariance issues caused by multipass rendering
|
|
#if defined( _X360 )
|
|
[isolate]
|
|
#endif
|
|
{
|
|
if ( !bSkinning )
|
|
{
|
|
worldPos = mul4x3( modelPos, cModel[0] );
|
|
}
|
|
else // skinning - always three bones
|
|
{
|
|
float4x3 mat1 = cModel[boneIndices[0]];
|
|
float4x3 mat2 = cModel[boneIndices[1]];
|
|
float4x3 mat3 = cModel[boneIndices[2]];
|
|
|
|
float3 weights = DecompressBoneWeights( boneWeights ).xyz;
|
|
weights[2] = 1 - (weights[0] + weights[1]);
|
|
|
|
float4x3 blendMatrix = mat1 * weights[0] + mat2 * weights[1] + mat3 * weights[2];
|
|
worldPos = mul4x3( modelPos, blendMatrix );
|
|
}
|
|
}
|
|
}
|
|
|
|
void SkinPositionAndNormal( bool bSkinning, const float4 modelPos, const float3 modelNormal,
|
|
const float4 boneWeights, float4 fBoneIndices,
|
|
out float3 worldPos, out float3 worldNormal )
|
|
{
|
|
// Needed for invariance issues caused by multipass rendering
|
|
#if defined( _X360 )
|
|
[isolate]
|
|
#endif
|
|
{
|
|
|
|
#if !defined( _X360 )
|
|
int3 boneIndices = D3DCOLORtoUBYTE4( fBoneIndices );
|
|
#else
|
|
int3 boneIndices = fBoneIndices;
|
|
#endif
|
|
|
|
if ( !bSkinning )
|
|
{
|
|
worldPos = mul4x3( modelPos, cModel[0] );
|
|
worldNormal = mul3x3( modelNormal, ( const float3x3 )cModel[0] );
|
|
}
|
|
else // skinning - always three bones
|
|
{
|
|
float4x3 mat1 = cModel[boneIndices[0]];
|
|
float4x3 mat2 = cModel[boneIndices[1]];
|
|
float4x3 mat3 = cModel[boneIndices[2]];
|
|
|
|
float3 weights = DecompressBoneWeights( boneWeights ).xyz;
|
|
weights[2] = 1 - (weights[0] + weights[1]);
|
|
|
|
float4x3 blendMatrix = mat1 * weights[0] + mat2 * weights[1] + mat3 * weights[2];
|
|
worldPos = mul4x3( modelPos, blendMatrix );
|
|
worldNormal = mul3x3( modelNormal, ( float3x3 )blendMatrix );
|
|
}
|
|
|
|
} // end [isolate]
|
|
}
|
|
|
|
// Is it worth keeping SkinPosition and SkinPositionAndNormal around since the optimizer
|
|
// gets rid of anything that isn't used?
|
|
void SkinPositionNormalAndTangentSpace(
|
|
bool bSkinning,
|
|
const float4 modelPos, const float3 modelNormal,
|
|
const float4 modelTangentS,
|
|
const float4 boneWeights, float4 fBoneIndices,
|
|
out float3 worldPos, out float3 worldNormal,
|
|
out float3 worldTangentS, out float3 worldTangentT )
|
|
{
|
|
#if !defined( _X360 )
|
|
int3 boneIndices = D3DCOLORtoUBYTE4( fBoneIndices );
|
|
#else
|
|
int3 boneIndices = fBoneIndices;
|
|
#endif
|
|
|
|
// Needed for invariance issues caused by multipass rendering
|
|
#if defined( _X360 )
|
|
[isolate]
|
|
#endif
|
|
{
|
|
if ( !bSkinning )
|
|
{
|
|
worldPos = mul4x3( modelPos, cModel[0] );
|
|
worldNormal = mul3x3( modelNormal, ( const float3x3 )cModel[0] );
|
|
worldTangentS = mul3x3( ( float3 )modelTangentS, ( const float3x3 )cModel[0] );
|
|
}
|
|
else // skinning - always three bones
|
|
{
|
|
float4x3 mat1 = cModel[boneIndices[0]];
|
|
float4x3 mat2 = cModel[boneIndices[1]];
|
|
float4x3 mat3 = cModel[boneIndices[2]];
|
|
|
|
float3 weights = DecompressBoneWeights( boneWeights ).xyz;
|
|
weights[2] = 1 - (weights[0] + weights[1]);
|
|
|
|
float4x3 blendMatrix = mat1 * weights[0] + mat2 * weights[1] + mat3 * weights[2];
|
|
worldPos = mul4x3( modelPos, blendMatrix );
|
|
worldNormal = mul3x3( modelNormal, ( const float3x3 )blendMatrix );
|
|
worldTangentS = mul3x3( ( float3 )modelTangentS, ( const float3x3 )blendMatrix );
|
|
}
|
|
worldTangentT = cross( worldNormal, worldTangentS ) * modelTangentS.w;
|
|
}
|
|
}
|
|
|
|
|
|
//-----------------------------------------------------------------------------
|
|
// Lighting helper functions
|
|
//-----------------------------------------------------------------------------
|
|
|
|
float3 AmbientLight( const float3 worldNormal )
|
|
{
|
|
float3 nSquared = worldNormal * worldNormal;
|
|
int3 isNegative = ( worldNormal < 0.0 );
|
|
float3 linearColor;
|
|
linearColor = nSquared.x * cAmbientCubeX[isNegative.x] +
|
|
nSquared.y * cAmbientCubeY[isNegative.y] +
|
|
nSquared.z * cAmbientCubeZ[isNegative.z];
|
|
return linearColor;
|
|
}
|
|
|
|
// The following "internal" routines are called "privately" by other routines in this file which
|
|
// handle the particular flavor of vs20 control flow appropriate to the original caller
|
|
float VertexAttenInternal( const float3 worldPos, int lightNum )
|
|
{
|
|
float result = 0.0f;
|
|
|
|
// Get light direction
|
|
float3 lightDir = cLightInfo[lightNum].pos - worldPos;
|
|
|
|
// Get light distance squared.
|
|
float lightDistSquared = dot( lightDir, lightDir );
|
|
|
|
// Get 1/lightDistance
|
|
float ooLightDist = rsqrt( lightDistSquared );
|
|
|
|
// Normalize light direction
|
|
lightDir *= ooLightDist;
|
|
|
|
float3 vDist;
|
|
# if defined( _X360 )
|
|
{
|
|
//X360 dynamic compile hits an internal compiler error using dst(), this is the breakdown of how dst() works from the 360 docs.
|
|
vDist.x = 1;
|
|
vDist.y = lightDistSquared * ooLightDist;
|
|
vDist.z = lightDistSquared;
|
|
//flDist.w = ooLightDist;
|
|
}
|
|
# else
|
|
{
|
|
vDist = dst( lightDistSquared, ooLightDist );
|
|
}
|
|
# endif
|
|
|
|
float flDistanceAtten = 1.0f / dot( cLightInfo[lightNum].atten.xyz, vDist );
|
|
|
|
// Spot attenuation
|
|
float flCosTheta = dot( cLightInfo[lightNum].dir.xyz, -lightDir );
|
|
float flSpotAtten = (flCosTheta - cLightInfo[lightNum].spotParams.z) * cLightInfo[lightNum].spotParams.w;
|
|
flSpotAtten = max( 0.0001f, flSpotAtten );
|
|
flSpotAtten = pow( flSpotAtten, cLightInfo[lightNum].spotParams.x );
|
|
flSpotAtten = saturate( flSpotAtten );
|
|
|
|
// Select between point and spot
|
|
float flAtten = lerp( flDistanceAtten, flDistanceAtten * flSpotAtten, cLightInfo[lightNum].dir.w );
|
|
|
|
// Select between above and directional (no attenuation)
|
|
result = lerp( flAtten, 1.0f, cLightInfo[lightNum].color.w );
|
|
|
|
return result;
|
|
}
|
|
|
|
float CosineTermInternal( const float3 worldPos, const float3 worldNormal, int lightNum, bool bHalfLambert )
|
|
{
|
|
// Calculate light direction assuming this is a point or spot
|
|
float3 lightDir = normalize( cLightInfo[lightNum].pos - worldPos );
|
|
|
|
// Select the above direction or the one in the structure, based upon light type
|
|
lightDir = lerp( lightDir, -cLightInfo[lightNum].dir, cLightInfo[lightNum].color.w );
|
|
|
|
// compute N dot L
|
|
float NDotL = dot( worldNormal, lightDir );
|
|
|
|
if ( !bHalfLambert )
|
|
{
|
|
NDotL = max( 0.0f, NDotL );
|
|
}
|
|
else // Half-Lambert
|
|
{
|
|
NDotL = NDotL * 0.5 + 0.5;
|
|
NDotL = NDotL * NDotL;
|
|
}
|
|
return NDotL;
|
|
}
|
|
|
|
// This routine uses booleans to do early-outs and is meant to be called by routines OUTSIDE of this file
|
|
float GetVertexAttenForLight( const float3 worldPos, int lightNum )
|
|
{
|
|
float result = 0.0f;
|
|
if ( g_bLightEnabled[lightNum] )
|
|
{
|
|
result = VertexAttenInternal( worldPos, lightNum );
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
// This routine uses booleans to do early-outs and is meant to be called by routines OUTSIDE of this file
|
|
float CosineTerm( const float3 worldPos, const float3 worldNormal, int lightNum, bool bHalfLambert )
|
|
{
|
|
float flResult = 0.0f;
|
|
if ( g_bLightEnabled[lightNum] )
|
|
{
|
|
flResult = CosineTermInternal( worldPos, worldNormal, lightNum, bHalfLambert );
|
|
}
|
|
|
|
return flResult;
|
|
}
|
|
|
|
float3 DoLightInternal( const float3 worldPos, const float3 worldNormal, int lightNum, bool bHalfLambert )
|
|
{
|
|
return cLightInfo[lightNum].color *
|
|
CosineTermInternal( worldPos, worldNormal, lightNum, bHalfLambert ) *
|
|
VertexAttenInternal( worldPos, lightNum );
|
|
}
|
|
|
|
// This routine
|
|
float3 DoLighting( const float3 worldPos, const float3 worldNormal,
|
|
const float3 staticLightingColor, const bool bStaticLight,
|
|
const bool bDynamicLight, bool bHalfLambert )
|
|
{
|
|
float3 linearColor = float3( 0.0f, 0.0f, 0.0f );
|
|
|
|
if( bStaticLight ) // Static light
|
|
{
|
|
linearColor += GammaToLinear( staticLightingColor.rgb * cOverbright );
|
|
}
|
|
|
|
if( bDynamicLight ) // Dynamic light
|
|
{
|
|
for (int i = 0; i < g_nLightCount; i++)
|
|
{
|
|
linearColor += DoLightInternal( worldPos, worldNormal, i, bHalfLambert );
|
|
}
|
|
}
|
|
|
|
if( bDynamicLight )
|
|
{
|
|
linearColor += AmbientLight( worldNormal ); //ambient light is already remapped
|
|
}
|
|
|
|
return linearColor;
|
|
}
|
|
|
|
float3 DoLightingUnrolled( const float3 worldPos, const float3 worldNormal,
|
|
const float3 staticLightingColor, const bool bStaticLight,
|
|
const bool bDynamicLight, bool bHalfLambert, const int nNumLights )
|
|
{
|
|
float3 linearColor = float3( 0.0f, 0.0f, 0.0f );
|
|
|
|
if( bStaticLight ) // Static light
|
|
{
|
|
linearColor += GammaToLinear( staticLightingColor * cOverbright );
|
|
}
|
|
|
|
if( bDynamicLight ) // Ambient light
|
|
{
|
|
if ( nNumLights >= 1 )
|
|
linearColor += DoLightInternal( worldPos, worldNormal, 0, bHalfLambert );
|
|
if ( nNumLights >= 2 )
|
|
linearColor += DoLightInternal( worldPos, worldNormal, 1, bHalfLambert );
|
|
if ( nNumLights >= 3 )
|
|
linearColor += DoLightInternal( worldPos, worldNormal, 2, bHalfLambert );
|
|
if ( nNumLights >= 4 )
|
|
linearColor += DoLightInternal( worldPos, worldNormal, 3, bHalfLambert );
|
|
}
|
|
|
|
if( bDynamicLight )
|
|
{
|
|
linearColor += AmbientLight( worldNormal ); //ambient light is already remapped
|
|
}
|
|
|
|
return linearColor;
|
|
}
|
|
|
|
int4 FloatToInt( in float4 floats )
|
|
{
|
|
return D3DCOLORtoUBYTE4( floats.zyxw / 255.001953125 );
|
|
}
|
|
|
|
float2 ComputeSphereMapTexCoords( in float3 reflectionVector )
|
|
{
|
|
// transform reflection vector into view space
|
|
reflectionVector = mul( reflectionVector, ( float3x3 )cViewModel );
|
|
|
|
// generate <rx ry rz+1>
|
|
float3 tmp = float3( reflectionVector.x, reflectionVector.y, reflectionVector.z + 1.0f );
|
|
|
|
// find 1 / len
|
|
float ooLen = dot( tmp, tmp );
|
|
ooLen = 1.0f / sqrt( ooLen );
|
|
|
|
// tmp = tmp/|tmp| + 1
|
|
tmp.xy = ooLen * tmp.xy + 1.0f;
|
|
|
|
return tmp.xy * 0.5f;
|
|
}
|
|
|
|
|
|
#define DEFORMATION_CLAMP_TO_BOX_IN_WORLDSPACE 1
|
|
// minxyz.minsoftness / maxxyz.maxsoftness
|
|
float3 ApplyDeformation( float3 worldpos, int deftype, float4 defparms0, float4 defparms1,
|
|
float4 defparms2, float4 defparms3 )
|
|
{
|
|
float3 ret = worldpos;
|
|
if ( deftype == DEFORMATION_CLAMP_TO_BOX_IN_WORLDSPACE )
|
|
{
|
|
ret=max( ret, defparms2.xyz );
|
|
ret=min( ret, defparms3.xyz );
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
|
|
#endif //#ifndef COMMON_VS_FXC_H_
|