2
0
mirror of https://github.com/FWGS/xash3d-fwgs synced 2024-11-23 02:15:55 +01:00
xash3d-fwgs/ref/gl/gl_image.c
Alibek Omarov d56e53a347 Replace hardcoded functions names by __func__ macro
* while we're here, fix some possible bugs
* and fix -Wformat=2

s/__FUNCTION__/__func__/g

awawawa
2024-06-19 06:53:42 +03:00

2402 lines
62 KiB
C

/*
gl_image.c - texture uploading and processing
Copyright (C) 2010 Uncle Mike
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
*/
#include <stdarg.h>
#include "gl_local.h"
#include "crclib.h"
#define TEXTURES_HASH_SIZE (MAX_TEXTURES >> 2)
static gl_texture_t gl_textures[MAX_TEXTURES];
static gl_texture_t* gl_texturesHashTable[TEXTURES_HASH_SIZE];
static uint gl_numTextures;
static byte dottexture[8][8] =
{
{0,1,1,0,0,0,0,0},
{1,1,1,1,0,0,0,0},
{1,1,1,1,0,0,0,0},
{0,1,1,0,0,0,0,0},
{0,0,0,0,0,0,0,0},
{0,0,0,0,0,0,0,0},
{0,0,0,0,0,0,0,0},
{0,0,0,0,0,0,0,0},
};
#define IsLightMap( tex ) ( FBitSet(( tex )->flags, TF_ATLAS_PAGE ))
/*
=================
R_GetTexture
acess to array elem
=================
*/
gl_texture_t *R_GetTexture( GLenum texnum )
{
ASSERT( texnum >= 0 && texnum < MAX_TEXTURES );
return &gl_textures[texnum];
}
/*
=================
GL_TargetToString
=================
*/
static const char *GL_TargetToString( GLenum target )
{
switch( target )
{
case GL_TEXTURE_1D:
return "1D";
case GL_TEXTURE_2D:
return "2D";
case GL_TEXTURE_2D_MULTISAMPLE:
return "2D Multisample";
case GL_TEXTURE_3D:
return "3D";
case GL_TEXTURE_CUBE_MAP_ARB:
return "Cube";
case GL_TEXTURE_2D_ARRAY_EXT:
return "Array";
case GL_TEXTURE_RECTANGLE_EXT:
return "Rect";
}
return "??";
}
/*
=================
GL_Bind
=================
*/
void GL_Bind( GLint tmu, GLenum texnum )
{
gl_texture_t *texture;
GLuint glTarget;
// missed or invalid texture?
if( texnum <= 0 || texnum >= MAX_TEXTURES )
{
if( texnum != 0 )
gEngfuncs.Con_DPrintf( S_ERROR "%s: invalid texturenum %d\n", __func__, texnum );
texnum = tr.defaultTexture;
}
if( tmu != GL_KEEP_UNIT )
GL_SelectTexture( tmu );
else tmu = glState.activeTMU;
texture = &gl_textures[texnum];
glTarget = texture->target;
if( glTarget == GL_TEXTURE_2D_ARRAY_EXT )
glTarget = GL_TEXTURE_2D;
if( glState.currentTextureTargets[tmu] != glTarget )
{
GL_EnableTextureUnit( tmu, false );
glState.currentTextureTargets[tmu] = glTarget;
GL_EnableTextureUnit( tmu, true );
}
if( glState.currentTextures[tmu] == texture->texnum )
return;
pglBindTexture( texture->target, texture->texnum );
glState.currentTextures[tmu] = texture->texnum;
glState.currentTexturesIndex[tmu] = texnum;
}
qboolean GL_TextureFilteringEnabled( const gl_texture_t *tex )
{
if( FBitSet( tex->flags, TF_NEAREST ))
return false;
if( FBitSet( tex->flags, TF_DEPTHMAP ))
return true;
if( FBitSet( tex->flags, TF_NOMIPMAP ) || tex->numMips <= 1 )
{
if( FBitSet( tex->flags, TF_ATLAS_PAGE ))
return gl_lightmap_nearest.value == 0.0f;
if( FBitSet( tex->flags, TF_ALLOW_NEAREST ))
return gl_texture_nearest.value == 0.0f;
return true;
}
return gl_texture_nearest.value == 0.0f;
}
/*
=================
GL_ApplyTextureParams
=================
*/
void GL_ApplyTextureParams( gl_texture_t *tex )
{
vec4_t border = { 0.0f, 0.0f, 0.0f, 1.0f };
qboolean nomipmap;
if( !glw_state.initialized )
return;
Assert( tex != NULL );
// multisample textures does not support any sampling state changing
if( FBitSet( tex->flags, TF_MULTISAMPLE ))
return;
// set texture filter
nomipmap = tex->numMips <= 1 || FBitSet( tex->flags, TF_NOMIPMAP|TF_DEPTHMAP );
if( !GL_TextureFilteringEnabled( tex ))
{
pglTexParameteri( tex->target, GL_TEXTURE_MIN_FILTER, nomipmap ? GL_NEAREST : GL_NEAREST_MIPMAP_NEAREST );
pglTexParameteri( tex->target, GL_TEXTURE_MAG_FILTER, GL_NEAREST );
}
else
{
pglTexParameteri( tex->target, GL_TEXTURE_MIN_FILTER, nomipmap ? GL_LINEAR : GL_LINEAR_MIPMAP_LINEAR );
pglTexParameteri( tex->target, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
}
if( FBitSet( tex->flags, TF_DEPTHMAP ))
{
if( !FBitSet( tex->flags, TF_NOCOMPARE ))
{
pglTexParameteri( tex->target, GL_TEXTURE_COMPARE_MODE_ARB, GL_COMPARE_R_TO_TEXTURE_ARB );
pglTexParameteri( tex->target, GL_TEXTURE_COMPARE_FUNC_ARB, GL_LEQUAL );
}
if( FBitSet( tex->flags, TF_LUMINANCE ))
pglTexParameteri( tex->target, GL_DEPTH_TEXTURE_MODE_ARB, GL_LUMINANCE );
else pglTexParameteri( tex->target, GL_DEPTH_TEXTURE_MODE_ARB, GL_INTENSITY );
// allow max anisotropy as 1.0f on depth textures
if( GL_Support( GL_ANISOTROPY_EXT ))
pglTexParameterf( tex->target, GL_TEXTURE_MAX_ANISOTROPY_EXT, 1.0f );
}
else if( !FBitSet( tex->flags, TF_NOMIPMAP ) && tex->numMips > 1 )
{
// set texture anisotropy if available
if( GL_Support( GL_ANISOTROPY_EXT ) && ( tex->numMips > 1 ) && !FBitSet( tex->flags, TF_ALPHACONTRAST ))
pglTexParameterf( tex->target, GL_TEXTURE_MAX_ANISOTROPY_EXT, gl_texture_anisotropy.value );
// set texture LOD bias if available
if( GL_Support( GL_TEXTURE_LOD_BIAS ) && ( tex->numMips > 1 ))
pglTexParameterf( tex->target, GL_TEXTURE_LOD_BIAS_EXT, gl_texture_lodbias.value );
}
// check if border is not supported
if( FBitSet( tex->flags, TF_BORDER ) && !GL_Support( GL_CLAMP_TEXBORDER_EXT ))
{
ClearBits( tex->flags, TF_BORDER );
SetBits( tex->flags, TF_CLAMP );
}
// only seamless cubemaps allows wrap 'clamp_to_border"
if( tex->target == GL_TEXTURE_CUBE_MAP_ARB && !GL_Support( GL_ARB_SEAMLESS_CUBEMAP ) && FBitSet( tex->flags, TF_BORDER ))
ClearBits( tex->flags, TF_BORDER );
// set texture wrap
if( FBitSet( tex->flags, TF_BORDER ))
{
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_BORDER );
if( tex->target != GL_TEXTURE_1D )
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_BORDER );
if( tex->target == GL_TEXTURE_3D || tex->target == GL_TEXTURE_CUBE_MAP_ARB )
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_R, GL_CLAMP_TO_BORDER );
pglTexParameterfv( tex->target, GL_TEXTURE_BORDER_COLOR, border );
}
else if( FBitSet( tex->flags, TF_CLAMP ))
{
if( GL_Support( GL_CLAMPTOEDGE_EXT ))
{
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE );
if( tex->target != GL_TEXTURE_1D )
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE );
if( tex->target == GL_TEXTURE_3D || tex->target == GL_TEXTURE_CUBE_MAP_ARB )
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_R, GL_CLAMP_TO_EDGE );
}
else
{
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_S, GL_CLAMP );
if( tex->target != GL_TEXTURE_1D )
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_T, GL_CLAMP );
if( tex->target == GL_TEXTURE_3D || tex->target == GL_TEXTURE_CUBE_MAP_ARB )
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_R, GL_CLAMP );
}
}
else
{
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_S, GL_REPEAT );
if( tex->target != GL_TEXTURE_1D )
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_T, GL_REPEAT );
if( tex->target == GL_TEXTURE_3D || tex->target == GL_TEXTURE_CUBE_MAP_ARB )
pglTexParameteri( tex->target, GL_TEXTURE_WRAP_R, GL_REPEAT );
}
}
/*
=================
GL_UpdateTextureParams
=================
*/
static void GL_UpdateTextureParams( int iTexture )
{
gl_texture_t *tex = &gl_textures[iTexture];
qboolean nomipmap;
Assert( tex != NULL );
if( !tex->texnum ) return; // free slot
GL_Bind( XASH_TEXTURE0, iTexture );
// set texture anisotropy if available
if( GL_Support( GL_ANISOTROPY_EXT ) && ( tex->numMips > 1 ) && !FBitSet( tex->flags, TF_DEPTHMAP|TF_ALPHACONTRAST ))
pglTexParameterf( tex->target, GL_TEXTURE_MAX_ANISOTROPY_EXT, gl_texture_anisotropy.value );
// set texture LOD bias if available
if( GL_Support( GL_TEXTURE_LOD_BIAS ) && ( tex->numMips > 1 ) && !FBitSet( tex->flags, TF_DEPTHMAP ))
pglTexParameterf( tex->target, GL_TEXTURE_LOD_BIAS_EXT, gl_texture_lodbias.value );
nomipmap = tex->numMips <= 1 || FBitSet( tex->flags, TF_NOMIPMAP|TF_DEPTHMAP );
if( !GL_TextureFilteringEnabled( tex ))
{
pglTexParameteri( tex->target, GL_TEXTURE_MIN_FILTER, nomipmap ? GL_NEAREST : GL_NEAREST_MIPMAP_NEAREST );
pglTexParameteri( tex->target, GL_TEXTURE_MAG_FILTER, GL_NEAREST );
}
else
{
pglTexParameteri( tex->target, GL_TEXTURE_MIN_FILTER, nomipmap ? GL_LINEAR : GL_LINEAR_MIPMAP_LINEAR );
pglTexParameteri( tex->target, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
}
}
/*
=================
R_SetTextureParameters
=================
*/
void R_SetTextureParameters( void )
{
int i;
if( GL_Support( GL_ANISOTROPY_EXT ))
{
if( gl_texture_anisotropy.value > glConfig.max_texture_anisotropy )
gEngfuncs.Cvar_SetValue( "gl_anisotropy", glConfig.max_texture_anisotropy );
else if( gl_texture_anisotropy.value < 1.0f )
gEngfuncs.Cvar_SetValue( "gl_anisotropy", 1.0f );
}
if( GL_Support( GL_TEXTURE_LOD_BIAS ))
{
if( gl_texture_lodbias.value < -glConfig.max_texture_lod_bias )
gEngfuncs.Cvar_SetValue( "gl_texture_lodbias", -glConfig.max_texture_lod_bias );
else if( gl_texture_lodbias.value > glConfig.max_texture_lod_bias )
gEngfuncs.Cvar_SetValue( "gl_texture_lodbias", glConfig.max_texture_lod_bias );
}
ClearBits( gl_texture_anisotropy.flags, FCVAR_CHANGED );
ClearBits( gl_texture_lodbias.flags, FCVAR_CHANGED );
ClearBits( gl_texture_nearest.flags, FCVAR_CHANGED );
ClearBits( gl_lightmap_nearest.flags, FCVAR_CHANGED );
// change all the existing mipmapped texture objects
for( i = 0; i < gl_numTextures; i++ )
GL_UpdateTextureParams( i );
}
/*
================
GL_CalcTextureSamples
================
*/
static int GL_CalcTextureSamples( int flags )
{
if( FBitSet( flags, IMAGE_HAS_COLOR ))
return FBitSet( flags, IMAGE_HAS_ALPHA ) ? 4 : 3;
return FBitSet( flags, IMAGE_HAS_ALPHA ) ? 2 : 1;
}
/*
==================
GL_CalcImageSize
==================
*/
static size_t GL_CalcImageSize( pixformat_t format, int width, int height, int depth )
{
size_t size = 0;
// check the depth error
depth = Q_max( 1, depth );
switch( format )
{
case PF_LUMINANCE:
size = width * height * depth;
break;
case PF_RGB_24:
case PF_BGR_24:
size = width * height * depth * 3;
break;
case PF_BGRA_32:
case PF_RGBA_32:
size = width * height * depth * 4;
break;
case PF_DXT1:
size = (((width + 3) >> 2) * ((height + 3) >> 2) * 8) * depth;
break;
case PF_DXT3:
case PF_DXT5:
case PF_BC6H_SIGNED:
case PF_BC6H_UNSIGNED:
case PF_BC7_UNORM:
case PF_BC7_SRGB:
case PF_ATI2:
size = (((width + 3) >> 2) * ((height + 3) >> 2) * 16) * depth;
break;
}
return size;
}
/*
==================
GL_CalcTextureSize
==================
*/
static size_t GL_CalcTextureSize( GLenum format, int width, int height, int depth )
{
size_t size = 0;
// check the depth error
depth = Q_max( 1, depth );
switch( format )
{
case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
size = (((width + 3) >> 2) * ((height + 3) >> 2) * 8) * depth;
break;
case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
case GL_COMPRESSED_RED_GREEN_RGTC2_EXT:
case GL_COMPRESSED_LUMINANCE_ALPHA_ARB:
case GL_COMPRESSED_LUMINANCE_ALPHA_3DC_ATI:
case GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM_ARB:
case GL_COMPRESSED_RGBA_BPTC_UNORM_ARB:
case GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT_ARB:
case GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT_ARB:
size = (((width + 3) >> 2) * ((height + 3) >> 2) * 16) * depth;
break;
case GL_RGBA8:
case GL_RGBA:
size = width * height * depth * 4;
break;
case GL_RGB8:
case GL_RGB:
size = width * height * depth * 3;
break;
case GL_RGB5:
size = (width * height * depth * 3) / 2;
break;
case GL_RGBA4:
size = (width * height * depth * 4) / 2;
break;
case GL_INTENSITY:
case GL_LUMINANCE:
case GL_INTENSITY8:
case GL_LUMINANCE8:
size = (width * height * depth);
break;
case GL_LUMINANCE_ALPHA:
case GL_LUMINANCE8_ALPHA8:
size = width * height * depth * 2;
break;
case GL_R8:
size = width * height * depth;
break;
case GL_RG8:
size = width * height * depth * 2;
break;
case GL_R16:
size = width * height * depth * 2;
break;
case GL_RG16:
size = width * height * depth * 4;
break;
case GL_R16F:
case GL_LUMINANCE16F_ARB:
size = width * height * depth * 2; // half-floats
break;
case GL_R32F:
case GL_LUMINANCE32F_ARB:
size = width * height * depth * 4;
break;
case GL_RG16F:
case GL_LUMINANCE_ALPHA16F_ARB:
size = width * height * depth * 4;
break;
case GL_RG32F:
case GL_LUMINANCE_ALPHA32F_ARB:
size = width * height * depth * 8;
break;
case GL_RGB16F_ARB:
size = width * height * depth * 6;
break;
case GL_RGBA16F_ARB:
size = width * height * depth * 8;
break;
case GL_RGB32F_ARB:
size = width * height * depth * 12;
break;
case GL_RGBA32F_ARB:
size = width * height * depth * 16;
break;
case GL_DEPTH_COMPONENT16:
size = width * height * depth * 2;
break;
case GL_DEPTH_COMPONENT24:
size = width * height * depth * 3;
break;
case GL_DEPTH_COMPONENT32F:
size = width * height * depth * 4;
break;
default:
gEngfuncs.Host_Error( "%s: bad texture internal format (%u)\n", __func__, format );
break;
}
return size;
}
static int GL_CalcMipmapCount( gl_texture_t *tex, qboolean haveBuffer )
{
int width, height;
int mipcount;
Assert( tex != NULL );
if( !haveBuffer || tex->target == GL_TEXTURE_3D )
return 1;
// generate mip-levels by user request
if( FBitSet( tex->flags, TF_NOMIPMAP ))
return 1;
// mip-maps can't exceeds 16
for( mipcount = 0; mipcount < 16; mipcount++ )
{
width = Q_max( 1, ( tex->width >> mipcount ));
height = Q_max( 1, ( tex->height >> mipcount ));
if( width == 1 && height == 1 )
break;
}
return mipcount + 1;
}
/*
================
GL_SetTextureDimensions
================
*/
static void GL_SetTextureDimensions( gl_texture_t *tex, int width, int height, int depth )
{
int maxTextureSize = 0;
int maxDepthSize = 1;
Assert( tex != NULL );
switch( tex->target )
{
case GL_TEXTURE_1D:
case GL_TEXTURE_2D:
case GL_TEXTURE_2D_MULTISAMPLE:
maxTextureSize = glConfig.max_2d_texture_size;
break;
case GL_TEXTURE_2D_ARRAY_EXT:
maxDepthSize = glConfig.max_2d_texture_layers;
maxTextureSize = glConfig.max_2d_texture_size;
break;
case GL_TEXTURE_RECTANGLE_EXT:
maxTextureSize = glConfig.max_2d_rectangle_size;
break;
case GL_TEXTURE_CUBE_MAP_ARB:
maxTextureSize = glConfig.max_cubemap_size;
break;
case GL_TEXTURE_3D:
maxDepthSize = glConfig.max_3d_texture_size;
maxTextureSize = glConfig.max_3d_texture_size;
break;
default:
Assert( false );
}
// store original sizes
tex->srcWidth = width;
tex->srcHeight = height;
if( !GL_Support( GL_ARB_TEXTURE_NPOT_EXT ))
{
int step = (int)gl_round_down.value;
int scaled_width, scaled_height;
for( scaled_width = 1; scaled_width < width; scaled_width <<= 1 );
if( step > 0 && width < scaled_width && ( step == 1 || ( scaled_width - width ) > ( scaled_width >> step )))
scaled_width >>= 1;
for( scaled_height = 1; scaled_height < height; scaled_height <<= 1 );
if( step > 0 && height < scaled_height && ( step == 1 || ( scaled_height - height ) > ( scaled_height >> step )))
scaled_height >>= 1;
width = scaled_width;
height = scaled_height;
}
if( width > maxTextureSize || height > maxTextureSize || depth > maxDepthSize )
{
if( tex->target == GL_TEXTURE_1D )
{
while( width > maxTextureSize )
width >>= 1;
}
else if( tex->target == GL_TEXTURE_3D || tex->target == GL_TEXTURE_2D_ARRAY_EXT )
{
while( width > maxTextureSize || height > maxTextureSize || depth > maxDepthSize )
{
width >>= 1;
height >>= 1;
depth >>= 1;
}
}
else // all remaining cases
{
while( width > maxTextureSize || height > maxTextureSize )
{
width >>= 1;
height >>= 1;
}
}
}
// set the texture dimensions
tex->width = Q_max( 1, width );
tex->height = Q_max( 1, height );
tex->depth = Q_max( 1, depth );
}
/*
===============
GL_SetTextureTarget
===============
*/
static void GL_SetTextureTarget( gl_texture_t *tex, rgbdata_t *pic )
{
Assert( pic != NULL );
Assert( tex != NULL );
// correct depth size
pic->depth = Q_max( 1, pic->depth );
tex->numMips = 0; // begin counting
// correct mip count
pic->numMips = Q_max( 1, pic->numMips );
// trying to determine texture type
#ifndef XASH_GLES
if( pic->width > 1 && pic->height <= 1 )
tex->target = GL_TEXTURE_1D;
else
#endif // just skip first condition
if( FBitSet( pic->flags, IMAGE_CUBEMAP ))
tex->target = GL_TEXTURE_CUBE_MAP_ARB;
else if( FBitSet( pic->flags, IMAGE_MULTILAYER ) && pic->depth >= 1 )
tex->target = GL_TEXTURE_2D_ARRAY_EXT;
else if( pic->width > 1 && pic->height > 1 && pic->depth > 1 )
tex->target = GL_TEXTURE_3D;
else if( FBitSet( tex->flags, TF_RECTANGLE ))
tex->target = GL_TEXTURE_RECTANGLE_EXT;
else if( FBitSet(tex->flags, TF_MULTISAMPLE ))
tex->target = GL_TEXTURE_2D_MULTISAMPLE;
else tex->target = GL_TEXTURE_2D; // default case
// check for hardware support
if(( tex->target == GL_TEXTURE_CUBE_MAP_ARB ) && !GL_Support( GL_TEXTURE_CUBEMAP_EXT ))
tex->target = GL_NONE;
if(( tex->target == GL_TEXTURE_RECTANGLE_EXT ) && !GL_Support( GL_TEXTURE_2D_RECT_EXT ))
tex->target = GL_TEXTURE_2D; // fallback
if(( tex->target == GL_TEXTURE_2D_ARRAY_EXT ) && !GL_Support( GL_TEXTURE_ARRAY_EXT ))
tex->target = GL_NONE;
if(( tex->target == GL_TEXTURE_3D ) && !GL_Support( GL_TEXTURE_3D_EXT ))
tex->target = GL_NONE;
// check if depth textures are not supported
if( FBitSet( tex->flags, TF_DEPTHMAP ) && !GL_Support( GL_DEPTH_TEXTURE ))
tex->target = GL_NONE;
// depth cubemaps only allowed when GL_EXT_gpu_shader4 is supported
if( tex->target == GL_TEXTURE_CUBE_MAP_ARB && !GL_Support( GL_EXT_GPU_SHADER4 ) && FBitSet( tex->flags, TF_DEPTHMAP ))
tex->target = GL_NONE;
if(( tex->target == GL_TEXTURE_2D_MULTISAMPLE ) && !GL_Support( GL_TEXTURE_MULTISAMPLE ))
tex->target = GL_NONE;
}
/*
===============
GL_SetTextureFormat
===============
*/
static void GL_SetTextureFormat( gl_texture_t *tex, pixformat_t format, int channelMask )
{
qboolean haveColor = ( channelMask & IMAGE_HAS_COLOR );
qboolean haveAlpha = ( channelMask & IMAGE_HAS_ALPHA );
Assert( tex != NULL );
if( ImageCompressed( format ))
{
switch( format )
{
case PF_DXT1: tex->format = GL_COMPRESSED_RGB_S3TC_DXT1_EXT; break; // never use DXT1 with 1-bit alpha
case PF_DXT3: tex->format = GL_COMPRESSED_RGBA_S3TC_DXT3_EXT; break;
case PF_DXT5: tex->format = GL_COMPRESSED_RGBA_S3TC_DXT5_EXT; break;
case PF_BC6H_SIGNED: tex->format = GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT_ARB; break;
case PF_BC6H_UNSIGNED: tex->format = GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT_ARB; break;
case PF_BC7_SRGB:
case PF_BC7_UNORM: tex->format = GL_COMPRESSED_RGBA_BPTC_UNORM_ARB; break;
case PF_ATI2:
if( glConfig.hardware_type == GLHW_RADEON )
tex->format = GL_COMPRESSED_LUMINANCE_ALPHA_3DC_ATI;
else tex->format = GL_COMPRESSED_RED_GREEN_RGTC2_EXT;
break;
}
return;
}
else if( FBitSet( tex->flags, TF_DEPTHMAP ))
{
if( FBitSet( tex->flags, TF_ARB_16BIT ))
tex->format = GL_DEPTH_COMPONENT16;
else if( FBitSet( tex->flags, TF_ARB_FLOAT ) && GL_Support( GL_ARB_DEPTH_FLOAT_EXT ))
tex->format = GL_DEPTH_COMPONENT32F;
else tex->format = GL_DEPTH_COMPONENT24;
}
else if( FBitSet( tex->flags, TF_ARB_FLOAT|TF_ARB_16BIT ) && GL_Support( GL_ARB_TEXTURE_FLOAT_EXT ))
{
if( haveColor && haveAlpha )
{
if( FBitSet( tex->flags, TF_ARB_16BIT ) || gpGlobals->desktopBitsPixel == 16 )
tex->format = GL_RGBA16F_ARB;
else tex->format = GL_RGBA32F_ARB;
}
else if( haveColor )
{
if( FBitSet( tex->flags, TF_ARB_16BIT ) || gpGlobals->desktopBitsPixel == 16 )
tex->format = GL_RGB16F_ARB;
else tex->format = GL_RGB32F_ARB;
}
else if( haveAlpha )
{
if( FBitSet( tex->flags, TF_ARB_16BIT ) || gpGlobals->desktopBitsPixel == 16 )
tex->format = GL_RG16F;
else tex->format = GL_RG32F;
}
else
{
if( FBitSet( tex->flags, TF_ARB_16BIT ) || gpGlobals->desktopBitsPixel == 16 )
tex->format = GL_LUMINANCE16F_ARB;
else tex->format = GL_LUMINANCE32F_ARB;
}
}
else
{
// NOTE: not all the types will be compressed
int bits = gpGlobals->desktopBitsPixel;
switch( GL_CalcTextureSamples( channelMask ))
{
case 1:
if( FBitSet( tex->flags, TF_ALPHACONTRAST ))
tex->format = GL_INTENSITY8;
else tex->format = GL_LUMINANCE8;
break;
case 2: tex->format = GL_LUMINANCE8_ALPHA8; break;
case 3:
switch( bits )
{
case 16: tex->format = GL_RGB5; break;
case 32: tex->format = GL_RGB8; break;
default: tex->format = GL_RGB; break;
}
break;
case 4:
default:
switch( bits )
{
case 16: tex->format = GL_RGBA4; break;
case 32: tex->format = GL_RGBA8; break;
default: tex->format = GL_RGBA; break;
}
break;
}
}
}
/*
=================
GL_ResampleTexture
Assume input buffer is RGBA
=================
*/
byte *GL_ResampleTexture( const byte *source, int inWidth, int inHeight, int outWidth, int outHeight, qboolean isNormalMap )
{
uint frac, fracStep;
uint *in = (uint *)source;
uint p1[0x1000], p2[0x1000];
byte *pix1, *pix2, *pix3, *pix4;
uint *out, *inRow1, *inRow2;
static byte *scaledImage = NULL; // pointer to a scaled image
vec3_t normal;
int i, x, y;
if( !source ) return NULL;
scaledImage = Mem_Realloc( r_temppool, scaledImage, outWidth * outHeight * 4 );
fracStep = inWidth * 0x10000 / outWidth;
out = (uint *)scaledImage;
frac = fracStep >> 2;
for( i = 0; i < outWidth; i++ )
{
p1[i] = 4 * (frac >> 16);
frac += fracStep;
}
frac = (fracStep >> 2) * 3;
for( i = 0; i < outWidth; i++ )
{
p2[i] = 4 * (frac >> 16);
frac += fracStep;
}
if( isNormalMap )
{
for( y = 0; y < outHeight; y++, out += outWidth )
{
inRow1 = in + inWidth * (int)(((float)y + 0.25f) * inHeight / outHeight);
inRow2 = in + inWidth * (int)(((float)y + 0.75f) * inHeight / outHeight);
for( x = 0; x < outWidth; x++ )
{
pix1 = (byte *)inRow1 + p1[x];
pix2 = (byte *)inRow1 + p2[x];
pix3 = (byte *)inRow2 + p1[x];
pix4 = (byte *)inRow2 + p2[x];
normal[0] = MAKE_SIGNED( pix1[0] ) + MAKE_SIGNED( pix2[0] ) + MAKE_SIGNED( pix3[0] ) + MAKE_SIGNED( pix4[0] );
normal[1] = MAKE_SIGNED( pix1[1] ) + MAKE_SIGNED( pix2[1] ) + MAKE_SIGNED( pix3[1] ) + MAKE_SIGNED( pix4[1] );
normal[2] = MAKE_SIGNED( pix1[2] ) + MAKE_SIGNED( pix2[2] ) + MAKE_SIGNED( pix3[2] ) + MAKE_SIGNED( pix4[2] );
if( !VectorNormalizeLength( normal ))
VectorSet( normal, 0.5f, 0.5f, 1.0f );
((byte *)(out+x))[0] = 128 + (byte)(127.0f * normal[0]);
((byte *)(out+x))[1] = 128 + (byte)(127.0f * normal[1]);
((byte *)(out+x))[2] = 128 + (byte)(127.0f * normal[2]);
((byte *)(out+x))[3] = 255;
}
}
}
else
{
for( y = 0; y < outHeight; y++, out += outWidth )
{
inRow1 = in + inWidth * (int)(((float)y + 0.25f) * inHeight / outHeight);
inRow2 = in + inWidth * (int)(((float)y + 0.75f) * inHeight / outHeight);
for( x = 0; x < outWidth; x++ )
{
pix1 = (byte *)inRow1 + p1[x];
pix2 = (byte *)inRow1 + p2[x];
pix3 = (byte *)inRow2 + p1[x];
pix4 = (byte *)inRow2 + p2[x];
((byte *)(out+x))[0] = (pix1[0] + pix2[0] + pix3[0] + pix4[0]) >> 2;
((byte *)(out+x))[1] = (pix1[1] + pix2[1] + pix3[1] + pix4[1]) >> 2;
((byte *)(out+x))[2] = (pix1[2] + pix2[2] + pix3[2] + pix4[2]) >> 2;
((byte *)(out+x))[3] = (pix1[3] + pix2[3] + pix3[3] + pix4[3]) >> 2;
}
}
}
return scaledImage;
}
/*
=================
GL_BoxFilter3x3
box filter 3x3
=================
*/
static void GL_BoxFilter3x3( byte *out, const byte *in, int w, int h, int x, int y )
{
int r = 0, g = 0, b = 0, a = 0;
int count = 0, acount = 0;
int i, j, u, v;
const byte *pixel;
for( i = 0; i < 3; i++ )
{
u = ( i - 1 ) + x;
for( j = 0; j < 3; j++ )
{
v = ( j - 1 ) + y;
if( u >= 0 && u < w && v >= 0 && v < h )
{
pixel = &in[( u + v * w ) * 4];
if( pixel[3] != 0 )
{
r += pixel[0];
g += pixel[1];
b += pixel[2];
a += pixel[3];
acount++;
}
}
}
}
if( acount == 0 )
acount = 1;
out[0] = r / acount;
out[1] = g / acount;
out[2] = b / acount;
// out[3] = (int)( SimpleSpline( ( a / 12.0f ) / 255.0f ) * 255 );
}
/*
=================
GL_ApplyFilter
Apply box-filter to 1-bit alpha
=================
*/
static byte *GL_ApplyFilter( const byte *source, int width, int height )
{
byte *in = (byte *)source;
byte *out = (byte *)source;
int i;
if( ENGINE_GET_PARM( PARM_QUAKE_COMPATIBLE ) || glConfig.max_multisamples > 1 )
return in;
for( i = 0; source && i < width * height; i++, in += 4 )
{
if( in[0] == 0 && in[1] == 0 && in[2] == 0 && in[3] == 0 )
GL_BoxFilter3x3( in, source, width, height, i % width, i / width );
}
return out;
}
/*
=================
GL_BuildMipMap
Operates in place, quartering the size of the texture
=================
*/
static void GL_BuildMipMap( byte *in, int srcWidth, int srcHeight, int srcDepth, int flags )
{
byte *out = in;
int instride = ALIGN( srcWidth * 4, 1 );
int mipWidth, mipHeight, outpadding;
int row, x, y, z;
vec3_t normal;
if( !in ) return;
mipWidth = Q_max( 1, ( srcWidth >> 1 ));
mipHeight = Q_max( 1, ( srcHeight >> 1 ));
outpadding = ALIGN( mipWidth * 4, 1 ) - mipWidth * 4;
row = srcWidth << 2;
if( FBitSet( flags, TF_ALPHACONTRAST ))
{
memset( in, mipWidth, mipWidth * mipHeight * 4 );
return;
}
// move through all layers
for( z = 0; z < srcDepth; z++ )
{
if( FBitSet( flags, TF_NORMALMAP ))
{
for( y = 0; y < mipHeight; y++, in += instride * 2, out += outpadding )
{
byte *next = ((( y << 1 ) + 1 ) < srcHeight ) ? ( in + instride ) : in;
for( x = 0, row = 0; x < mipWidth; x++, row += 8, out += 4 )
{
if((( x << 1 ) + 1 ) < srcWidth )
{
normal[0] = MAKE_SIGNED( in[row+0] ) + MAKE_SIGNED( in[row+4] )
+ MAKE_SIGNED( next[row+0] ) + MAKE_SIGNED( next[row+4] );
normal[1] = MAKE_SIGNED( in[row+1] ) + MAKE_SIGNED( in[row+5] )
+ MAKE_SIGNED( next[row+1] ) + MAKE_SIGNED( next[row+5] );
normal[2] = MAKE_SIGNED( in[row+2] ) + MAKE_SIGNED( in[row+6] )
+ MAKE_SIGNED( next[row+2] ) + MAKE_SIGNED( next[row+6] );
}
else
{
normal[0] = MAKE_SIGNED( in[row+0] ) + MAKE_SIGNED( next[row+0] );
normal[1] = MAKE_SIGNED( in[row+1] ) + MAKE_SIGNED( next[row+1] );
normal[2] = MAKE_SIGNED( in[row+2] ) + MAKE_SIGNED( next[row+2] );
}
if( !VectorNormalizeLength( normal ))
VectorSet( normal, 0.5f, 0.5f, 1.0f );
out[0] = 128 + (byte)(127.0f * normal[0]);
out[1] = 128 + (byte)(127.0f * normal[1]);
out[2] = 128 + (byte)(127.0f * normal[2]);
out[3] = 255;
}
}
}
else
{
for( y = 0; y < mipHeight; y++, in += instride * 2, out += outpadding )
{
byte *next = ((( y << 1 ) + 1 ) < srcHeight ) ? ( in + instride ) : in;
for( x = 0, row = 0; x < mipWidth; x++, row += 8, out += 4 )
{
if((( x << 1 ) + 1 ) < srcWidth )
{
out[0] = (in[row+0] + in[row+4] + next[row+0] + next[row+4]) >> 2;
out[1] = (in[row+1] + in[row+5] + next[row+1] + next[row+5]) >> 2;
out[2] = (in[row+2] + in[row+6] + next[row+2] + next[row+6]) >> 2;
out[3] = (in[row+3] + in[row+7] + next[row+3] + next[row+7]) >> 2;
}
else
{
out[0] = (in[row+0] + next[row+0]) >> 1;
out[1] = (in[row+1] + next[row+1]) >> 1;
out[2] = (in[row+2] + next[row+2]) >> 1;
out[3] = (in[row+3] + next[row+3]) >> 1;
}
}
}
}
}
}
static void GL_TextureImageRAW( gl_texture_t *tex, GLint side, GLint level, GLint width, GLint height, GLint depth, GLint type, const void *data )
{
GLuint cubeTarget = GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB;
qboolean subImage = FBitSet( tex->flags, TF_IMG_UPLOADED );
GLenum inFormat = gEngfuncs.Image_GetPFDesc(type)->glFormat;
GLint dataType = GL_UNSIGNED_BYTE;
GLsizei samplesCount = 0;
Assert( tex != NULL );
if( FBitSet( tex->flags, TF_DEPTHMAP ))
inFormat = GL_DEPTH_COMPONENT;
if( FBitSet( tex->flags, TF_ARB_16BIT ))
dataType = GL_HALF_FLOAT_ARB;
else if( FBitSet( tex->flags, TF_ARB_FLOAT ))
dataType = GL_FLOAT;
if( tex->target == GL_TEXTURE_1D )
{
if( subImage ) pglTexSubImage1D( tex->target, level, 0, width, inFormat, dataType, data );
else pglTexImage1D( tex->target, level, tex->format, width, 0, inFormat, dataType, data );
}
else if( tex->target == GL_TEXTURE_CUBE_MAP_ARB )
{
if( subImage ) pglTexSubImage2D( cubeTarget + side, level, 0, 0, width, height, inFormat, dataType, data );
else pglTexImage2D( cubeTarget + side, level, tex->format, width, height, 0, inFormat, dataType, data );
}
else if( tex->target == GL_TEXTURE_3D || tex->target == GL_TEXTURE_2D_ARRAY_EXT )
{
if( subImage ) pglTexSubImage3D( tex->target, level, 0, 0, 0, width, height, depth, inFormat, dataType, data );
else pglTexImage3D( tex->target, level, tex->format, width, height, depth, 0, inFormat, dataType, data );
}
else if( tex->target == GL_TEXTURE_2D_MULTISAMPLE )
{
#if !defined( XASH_GL_STATIC ) || (!defined( XASH_GLES ) && !defined( XASH_GL4ES ))
samplesCount = (GLsizei)gEngfuncs.pfnGetCvarFloat("gl_msaa_samples");
switch (samplesCount)
{
case 2:
case 4:
case 8:
case 16:
break;
default:
samplesCount = 1;
}
pglTexImage2DMultisample( tex->target, samplesCount, tex->format, width, height, GL_TRUE );
#else /* !XASH_GL_STATIC !XASH_GLES && !XASH_GL4ES */
gEngfuncs.Con_Printf( S_ERROR "GLES renderer don't support GL_TEXTURE_2D_MULTISAMPLE!\n" );
#endif /* !XASH_GL_STATIC !XASH_GLES && !XASH_GL4ES */
}
else // 2D or RECT
{
if( subImage ) pglTexSubImage2D( tex->target, level, 0, 0, width, height, inFormat, dataType, data );
else pglTexImage2D( tex->target, level, tex->format, width, height, 0, inFormat, dataType, data );
}
}
static void GL_TextureImageCompressed( gl_texture_t *tex, GLint side, GLint level, GLint width, GLint height, GLint depth, size_t size, const void *data )
{
GLuint cubeTarget = GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB;
qboolean subImage = FBitSet( tex->flags, TF_IMG_UPLOADED );
Assert( tex != NULL );
#ifndef XASH_GLES
if( tex->target == GL_TEXTURE_1D )
{
if( subImage ) pglCompressedTexSubImage1DARB( tex->target, level, 0, width, tex->format, size, data );
else pglCompressedTexImage1DARB( tex->target, level, tex->format, width, 0, size, data );
}
else if( tex->target == GL_TEXTURE_CUBE_MAP_ARB )
{
if( subImage ) pglCompressedTexSubImage2DARB( cubeTarget + side, level, 0, 0, width, height, tex->format, size, data );
else pglCompressedTexImage2DARB( cubeTarget + side, level, tex->format, width, height, 0, size, data );
}
else if( tex->target == GL_TEXTURE_3D || tex->target == GL_TEXTURE_2D_ARRAY_EXT )
{
if( subImage ) pglCompressedTexSubImage3DARB( tex->target, level, 0, 0, 0, width, height, depth, tex->format, size, data );
else pglCompressedTexImage3DARB( tex->target, level, tex->format, width, height, depth, 0, size, data );
}
else // 2D or RECT
{
if( subImage ) pglCompressedTexSubImage2DARB( tex->target, level, 0, 0, width, height, tex->format, size, data );
else pglCompressedTexImage2DARB( tex->target, level, tex->format, width, height, 0, size, data );
}
#endif
}
/*
===============
GL_CheckTexImageError
show GL-errors on load images
===============
*/
static void GL_CheckTexImageError( gl_texture_t *tex )
{
int err;
Assert( tex != NULL );
// catch possible errors
if( gl_check_errors.value && ( err = pglGetError()) != GL_NO_ERROR )
gEngfuncs.Con_Printf( S_OPENGL_ERROR "%s while uploading %s [%s]\n", GL_ErrorString( err ), tex->name, GL_TargetToString( tex->target ));
}
/*
===============
GL_UploadTexture
upload texture into video memory
===============
*/
static qboolean GL_UploadTexture( gl_texture_t *tex, rgbdata_t *pic )
{
byte *buf, *data;
size_t texsize, size;
uint width, height;
uint i, j, numSides;
uint offset = 0;
qboolean normalMap;
const byte *bufend;
// dedicated server
if( !glw_state.initialized )
return true;
Assert( pic != NULL );
Assert( tex != NULL );
GL_SetTextureTarget( tex, pic ); // must be first
// make sure what target is correct
if( tex->target == GL_NONE )
{
gEngfuncs.Con_DPrintf( S_ERROR "%s: %s is not supported by your hardware\n", __func__, tex->name );
return false;
}
if( pic->type == PF_BC6H_SIGNED || pic->type == PF_BC6H_UNSIGNED || pic->type == PF_BC7_UNORM || pic->type == PF_BC7_SRGB )
{
if( !GL_Support( GL_ARB_TEXTURE_COMPRESSION_BPTC ))
{
gEngfuncs.Con_DPrintf( S_ERROR "%s: BC6H/BC7 compression formats is not supported by your hardware\n", __func__ );
return false;
}
}
GL_SetTextureDimensions( tex, pic->width, pic->height, pic->depth );
GL_SetTextureFormat( tex, pic->type, pic->flags );
tex->fogParams[0] = pic->fogParams[0];
tex->fogParams[1] = pic->fogParams[1];
tex->fogParams[2] = pic->fogParams[2];
tex->fogParams[3] = pic->fogParams[3];
if(( pic->width * pic->height ) & 3 )
{
// will be resampled, just tell me for debug targets
gEngfuncs.Con_Reportf( "%s: %s s&3 [%d x %d]\n", __func__, tex->name, pic->width, pic->height );
}
buf = pic->buffer;
bufend = pic->buffer + pic->size; // total image size include all the layers, cube sides, mipmaps
offset = GL_CalcImageSize( pic->type, pic->width, pic->height, pic->depth );
texsize = GL_CalcTextureSize( tex->format, tex->width, tex->height, tex->depth );
normalMap = FBitSet( tex->flags, TF_NORMALMAP ) ? true : false;
numSides = FBitSet( pic->flags, IMAGE_CUBEMAP ) ? 6 : 1;
// uploading texture into video memory, change the binding
glState.currentTextures[glState.activeTMU] = tex->texnum;
glState.currentTexturesIndex[glState.activeTMU] = tex - gl_textures;
pglBindTexture( tex->target, tex->texnum );
for( i = 0; i < numSides; i++ )
{
// track the buffer bounds
if( buf != NULL && buf >= bufend )
gEngfuncs.Host_Error( "%s: %s image buffer overflow\n", __func__, tex->name );
if( ImageCompressed( pic->type ))
{
for( j = 0; j < Q_max( 1, pic->numMips ); j++ )
{
width = Q_max( 1, ( tex->width >> j ));
height = Q_max( 1, ( tex->height >> j ));
texsize = GL_CalcTextureSize( tex->format, width, height, tex->depth );
size = GL_CalcImageSize( pic->type, width, height, tex->depth );
GL_TextureImageCompressed( tex, i, j, width, height, tex->depth, size, buf );
tex->size += texsize;
buf += size; // move pointer
tex->numMips++;
GL_CheckTexImageError( tex );
}
}
else if( Q_max( 1, pic->numMips ) > 1 ) // not-compressed DDS
{
for( j = 0; j < Q_max( 1, pic->numMips ); j++ )
{
width = Q_max( 1, ( tex->width >> j ));
height = Q_max( 1, ( tex->height >> j ));
texsize = GL_CalcTextureSize( tex->format, width, height, tex->depth );
size = GL_CalcImageSize( pic->type, width, height, tex->depth );
GL_TextureImageRAW( tex, i, j, width, height, tex->depth, pic->type, buf );
tex->size += texsize;
buf += size; // move pointer
tex->numMips++;
GL_CheckTexImageError( tex );
}
}
else // RGBA32
{
int mipCount = GL_CalcMipmapCount( tex, ( buf != NULL ));
// NOTE: only single uncompressed textures can be resamples, no mips, no layers, no sides
if(( tex->depth == 1 ) && (( pic->width != tex->width ) || ( pic->height != tex->height )))
data = GL_ResampleTexture( buf, pic->width, pic->height, tex->width, tex->height, normalMap );
else data = buf;
if( !ImageCompressed( pic->type ) && !FBitSet( tex->flags, TF_NOMIPMAP ) && FBitSet( pic->flags, IMAGE_ONEBIT_ALPHA ))
data = GL_ApplyFilter( data, tex->width, tex->height );
// mips will be auto-generated if desired
for( j = 0; j < mipCount; j++ )
{
width = Q_max( 1, ( tex->width >> j ));
height = Q_max( 1, ( tex->height >> j ));
texsize = GL_CalcTextureSize( tex->format, width, height, tex->depth );
size = GL_CalcImageSize( pic->type, width, height, tex->depth );
GL_TextureImageRAW( tex, i, j, width, height, tex->depth, pic->type, data );
if( mipCount > 1 )
GL_BuildMipMap( data, width, height, tex->depth, tex->flags );
tex->size += texsize;
tex->numMips++;
GL_CheckTexImageError( tex );
}
// move to next side
if( numSides > 1 && ( buf != NULL ))
buf += GL_CalcImageSize( pic->type, pic->width, pic->height, 1 );
}
}
SetBits( tex->flags, TF_IMG_UPLOADED ); // done
tex->numMips /= numSides;
return true;
}
/*
===============
GL_ProcessImage
do specified actions on pixels
===============
*/
static void GL_ProcessImage( gl_texture_t *tex, rgbdata_t *pic )
{
uint img_flags = 0;
// force upload texture as RGB or RGBA (detail textures requires this)
if( tex->flags & TF_FORCE_COLOR ) pic->flags |= IMAGE_HAS_COLOR;
if( pic->flags & IMAGE_HAS_ALPHA ) tex->flags |= TF_HAS_ALPHA;
tex->encode = pic->encode; // share encode method
if( ImageCompressed( pic->type ))
{
if( !pic->numMips )
tex->flags |= TF_NOMIPMAP; // disable mipmapping by user request
// clear all the unsupported flags
tex->flags &= ~TF_KEEP_SOURCE;
}
else
{
// copy flag about luma pixels
if( pic->flags & IMAGE_HAS_LUMA )
tex->flags |= TF_HAS_LUMA;
if( pic->flags & IMAGE_QUAKEPAL )
tex->flags |= TF_QUAKEPAL;
// create luma texture from quake texture
if( tex->flags & TF_MAKELUMA )
{
img_flags |= IMAGE_MAKE_LUMA;
tex->flags &= ~TF_MAKELUMA;
}
if( !FBitSet( tex->flags, TF_IMG_UPLOADED ) && FBitSet( tex->flags, TF_KEEP_SOURCE ))
tex->original = gEngfuncs.FS_CopyImage( pic ); // because current pic will be expanded to rgba
// we need to expand image into RGBA buffer
if( pic->type == PF_INDEXED_24 || pic->type == PF_INDEXED_32 )
img_flags |= IMAGE_FORCE_RGBA;
// processing image before uploading (force to rgba, make luma etc)
if( pic->buffer ) gEngfuncs.Image_Process( &pic, 0, 0, img_flags, 0 );
if( FBitSet( tex->flags, TF_LUMINANCE ))
ClearBits( pic->flags, IMAGE_HAS_COLOR );
}
}
/*
================
GL_CheckTexName
================
*/
static qboolean GL_CheckTexName( const char *name )
{
int len;
if( !COM_CheckString( name ))
return false;
len = Q_strlen( name );
// because multi-layered textures can exceed name string
if( len >= sizeof( gl_textures->name ))
{
gEngfuncs.Con_Printf( S_ERROR "LoadTexture: too long name %s (%d)\n", name, len );
return false;
}
return true;
}
/*
================
GL_TextureForName
================
*/
static gl_texture_t *GL_TextureForName( const char *name )
{
gl_texture_t *tex;
uint hash;
// find the texture in array
hash = COM_HashKey( name, TEXTURES_HASH_SIZE );
for( tex = gl_texturesHashTable[hash]; tex != NULL; tex = tex->nextHash )
{
if( !Q_stricmp( tex->name, name ))
return tex;
}
return NULL;
}
/*
================
GL_AllocTexture
================
*/
static gl_texture_t *GL_AllocTexture( const char *name, texFlags_t flags )
{
const qboolean skyboxhack = FBitSet( flags, TF_SKYSIDE ) && glConfig.context != CONTEXT_TYPE_GL_CORE;
gl_texture_t *tex = NULL;
GLuint texnum = 1;
if( !skyboxhack )
{
// keep generating new texture names to avoid collision with predefined skybox objects
do
{
pglGenTextures( 1, &texnum );
}
while( texnum >= SKYBOX_BASE_NUM && texnum <= SKYBOX_BASE_NUM + SKYBOX_MAX_SIDES );
}
else texnum = tr.skyboxbasenum;
// try to match texture slot and texture handle because of buggy games
if( texnum >= MAX_TEXTURES || gl_textures[texnum].texnum != 0 )
{
// find a free texture_t slot
uint i;
for( i = 0; i < MAX_TEXTURES; i++ )
{
if( gl_textures[i].texnum )
continue;
tex = &gl_textures[i];
break;
}
}
else tex = &gl_textures[texnum];
if( tex == NULL )
{
gEngfuncs.Host_Error( "%s: MAX_TEXTURES limit exceeds\n", __func__ );
return NULL;
}
// copy initial params
Q_strncpy( tex->name, name, sizeof( tex->name ));
tex->texnum = texnum;
tex->flags = flags;
// increase counter
gl_numTextures = Q_max(( tex - gl_textures ) + 1, gl_numTextures );
if( skyboxhack )
tr.skyboxbasenum++;
// add to hash table
tex->hashValue = COM_HashKey( name, TEXTURES_HASH_SIZE );
tex->nextHash = gl_texturesHashTable[tex->hashValue];
gl_texturesHashTable[tex->hashValue] = tex;
return tex;
}
/*
================
GL_DeleteTexture
================
*/
static void GL_DeleteTexture( gl_texture_t *tex )
{
gl_texture_t **prev;
gl_texture_t *cur;
ASSERT( tex != NULL );
// already freed?
if( !tex->texnum ) return;
// debug
if( !tex->name[0] )
{
gEngfuncs.Con_Printf( S_ERROR "%s: trying to free unnamed texture with texnum %i\n", __func__, tex->texnum );
return;
}
// remove from hash table
prev = &gl_texturesHashTable[tex->hashValue];
while( 1 )
{
cur = *prev;
if( !cur ) break;
if( cur == tex )
{
*prev = cur->nextHash;
break;
}
prev = &cur->nextHash;
}
// invalidate texture units state cache
for( int i = 0; i < MAX_TEXTURE_UNITS; i++ )
{
if( glState.currentTextures[i] == tex->texnum )
{
if( glState.currentTextureTargets[i] != GL_NONE )
{
GL_SelectTexture( i );
pglDisable( glState.currentTextureTargets[i] );
}
glState.currentTextureTargets[i] = GL_NONE;
glState.currentTextures[i] = -1;
glState.currentTexturesIndex[i] = 0;
}
}
// release source
if( tex->original )
gEngfuncs.FS_FreeImage( tex->original );
if( glw_state.initialized )
pglDeleteTextures( 1, &tex->texnum );
memset( tex, 0, sizeof( *tex ));
}
/*
================
GL_UpdateTexSize
recalc image room
================
*/
void GL_UpdateTexSize( int texnum, int width, int height, int depth )
{
int i, j, texsize;
int numSides;
gl_texture_t *tex;
if( texnum <= 0 || texnum >= MAX_TEXTURES )
return;
tex = &gl_textures[texnum];
numSides = FBitSet( tex->flags, TF_CUBEMAP ) ? 6 : 1;
GL_SetTextureDimensions( tex, width, height, depth );
tex->size = 0; // recompute now
for( i = 0; i < numSides; i++ )
{
for( j = 0; j < Q_max( 1, tex->numMips ); j++ )
{
width = Q_max( 1, ( tex->width >> j ));
height = Q_max( 1, ( tex->height >> j ));
texsize = GL_CalcTextureSize( tex->format, width, height, tex->depth );
tex->size += texsize;
}
}
}
/*
================
GL_LoadTexture
================
*/
int GL_LoadTexture( const char *name, const byte *buf, size_t size, int flags )
{
gl_texture_t *tex;
rgbdata_t *pic;
uint picFlags = 0;
if( !GL_CheckTexName( name ))
return 0;
// see if already loaded
if(( tex = GL_TextureForName( name )))
return (tex - gl_textures);
if( FBitSet( flags, TF_NOFLIP_TGA ))
SetBits( picFlags, IL_DONTFLIP_TGA );
if( FBitSet( flags, TF_KEEP_SOURCE ) && !FBitSet( flags, TF_EXPAND_SOURCE ))
SetBits( picFlags, IL_KEEP_8BIT );
// set some image flags
gEngfuncs.Image_SetForceFlags( picFlags );
pic = gEngfuncs.FS_LoadImage( name, buf, size );
if( !pic ) return 0; // couldn't loading image
// allocate the new one
tex = GL_AllocTexture( name, flags );
GL_ProcessImage( tex, pic );
if( !GL_UploadTexture( tex, pic ))
{
memset( tex, 0, sizeof( gl_texture_t ));
gEngfuncs.FS_FreeImage( pic ); // release source texture
return 0;
}
GL_ApplyTextureParams( tex ); // update texture filter, wrap etc
gEngfuncs.FS_FreeImage( pic ); // release source texture
// NOTE: always return texnum as index in array or engine will stop work !!!
return tex - gl_textures;
}
/*
================
GL_LoadTextureArray
================
*/
int GL_LoadTextureArray( const char **names, int flags )
{
rgbdata_t *pic, *src;
char basename[256];
uint numLayers = 0;
uint picFlags = 0;
char name[256];
gl_texture_t *tex;
size_t len = 0;
int ret = 0;
uint i, j;
if( !names || !names[0] || !glw_state.initialized )
return 0;
// count layers (g-cont. this is pontentially unsafe loop)
for( i = 0; i < glConfig.max_2d_texture_layers && ( *names[i] != '\0' ); i++ )
numLayers++;
name[0] = '\0';
if( numLayers <= 0 ) return 0;
// create complexname from layer names
for( i = 0; i < numLayers - 1; i++ )
{
COM_FileBase( names[i], basename, sizeof( basename ));
ret = Q_snprintf( &name[len], sizeof( name ) - len, "%s|", basename );
if( ret == -1 )
return 0;
len += ret;
}
COM_FileBase( names[i], basename, sizeof( basename ));
ret = Q_snprintf( &name[len], sizeof( name ) - len, "%s[%i]", basename, numLayers );
if( ret == -1 )
return 0;
if( !GL_CheckTexName( name ))
return 0;
// see if already loaded
if(( tex = GL_TextureForName( name )))
return (tex - gl_textures);
// load all the images and pack it into single image
for( i = 0, pic = NULL; i < numLayers; i++ )
{
size_t srcsize, dstsize, mipsize;
src = gEngfuncs.FS_LoadImage( names[i], NULL, 0 );
if( !src ) break; // coldn't find layer
if( pic )
{
// mixed mode: DXT + RGB
if( pic->type != src->type )
{
gEngfuncs.Con_Printf( S_ERROR "%s: mismatch image format for %s and %s\n", __func__, names[0], names[i] );
break;
}
// different mipcount
if( pic->numMips != src->numMips )
{
gEngfuncs.Con_Printf( S_ERROR "%s: mismatch mip count for %s and %s\n", __func__, names[0], names[i] );
break;
}
if( pic->encode != src->encode )
{
gEngfuncs.Con_Printf( S_ERROR "%s: mismatch custom encoding for %s and %s\n", __func__, names[0], names[i] );
break;
}
// but allow to rescale raw images
if( ImageRAW( pic->type ) && ImageRAW( src->type ) && ( pic->width != src->width || pic->height != src->height ))
gEngfuncs.Image_Process( &src, pic->width, pic->height, IMAGE_RESAMPLE, 0.0f );
if( pic->size != src->size )
{
gEngfuncs.Con_Printf( S_ERROR "%s: mismatch image size for %s and %s\n", __func__, names[0], names[i] );
break;
}
}
else
{
// create new image
pic = Mem_Malloc( r_temppool, sizeof( rgbdata_t ));
memcpy( pic, src, sizeof( rgbdata_t ));
// expand pic buffer for all layers
pic->buffer = Mem_Malloc( r_temppool, pic->size * numLayers );
pic->depth = 0;
}
mipsize = srcsize = dstsize = 0;
for( j = 0; j < Q_max( 1, pic->numMips ); j++ )
{
int width = Q_max( 1, ( pic->width >> j ));
int height = Q_max( 1, ( pic->height >> j ));
mipsize = GL_CalcImageSize( pic->type, width, height, 1 );
memcpy( pic->buffer + dstsize + mipsize * i, src->buffer + srcsize, mipsize );
dstsize += mipsize * numLayers;
srcsize += mipsize;
}
gEngfuncs.FS_FreeImage( src );
// increase layers
pic->depth++;
}
// there were errors
if( !pic || ( pic->depth != numLayers ))
{
gEngfuncs.Con_Printf( S_ERROR "%s: not all layers were loaded. Texture array is not created\n", __func__ );
if( pic ) gEngfuncs.FS_FreeImage( pic );
return 0;
}
// it's multilayer image!
SetBits( pic->flags, IMAGE_MULTILAYER );
pic->size *= numLayers;
// allocate the new one
tex = GL_AllocTexture( name, flags );
GL_ProcessImage( tex, pic );
if( !GL_UploadTexture( tex, pic ))
{
memset( tex, 0, sizeof( gl_texture_t ));
gEngfuncs.FS_FreeImage( pic ); // release source texture
return 0;
}
GL_ApplyTextureParams( tex ); // update texture filter, wrap etc
gEngfuncs.FS_FreeImage( pic ); // release source texture
// NOTE: always return texnum as index in array or engine will stop work !!!
return tex - gl_textures;
}
/*
================
GL_LoadTextureFromBuffer
================
*/
int GL_LoadTextureFromBuffer( const char *name, rgbdata_t *pic, texFlags_t flags, qboolean update )
{
gl_texture_t *tex;
if( !GL_CheckTexName( name ))
return 0;
// see if already loaded
if(( tex = GL_TextureForName( name )) && !update )
return (tex - gl_textures);
// couldn't loading image
if( !pic ) return 0;
if( update )
{
if( tex == NULL )
gEngfuncs.Host_Error( "%s: couldn't find texture %s for update\n", __func__, name );
SetBits( tex->flags, flags );
}
else
{
// allocate the new one
tex = GL_AllocTexture( name, flags );
}
GL_ProcessImage( tex, pic );
if( !GL_UploadTexture( tex, pic ))
{
memset( tex, 0, sizeof( gl_texture_t ));
return 0;
}
GL_ApplyTextureParams( tex ); // update texture filter, wrap etc
return (tex - gl_textures);
}
/*
================
GL_CreateTexture
creates texture from buffer
================
*/
int GL_CreateTexture( const char *name, int width, int height, const void *buffer, texFlags_t flags )
{
qboolean update = FBitSet( flags, TF_UPDATE ) ? true : false;
int datasize = 1;
rgbdata_t r_empty;
if( FBitSet( flags, TF_ARB_16BIT ))
datasize = 2;
else if( FBitSet( flags, TF_ARB_FLOAT ))
datasize = 4;
ClearBits( flags, TF_UPDATE );
memset( &r_empty, 0, sizeof( r_empty ));
r_empty.width = width;
r_empty.height = height;
r_empty.type = PF_RGBA_32;
r_empty.size = r_empty.width * r_empty.height * datasize * 4;
r_empty.buffer = (byte *)buffer;
// clear invalid combinations
ClearBits( flags, TF_TEXTURE_3D );
// if image not luminance and not alphacontrast it will have color
if( !FBitSet( flags, TF_LUMINANCE ) && !FBitSet( flags, TF_ALPHACONTRAST ))
SetBits( r_empty.flags, IMAGE_HAS_COLOR );
if( FBitSet( flags, TF_HAS_ALPHA ))
SetBits( r_empty.flags, IMAGE_HAS_ALPHA );
if( FBitSet( flags, TF_CUBEMAP ))
{
if( !GL_Support( GL_TEXTURE_CUBEMAP_EXT ))
return 0;
SetBits( r_empty.flags, IMAGE_CUBEMAP );
r_empty.size *= 6;
}
return GL_LoadTextureFromBuffer( name, &r_empty, flags, update );
}
/*
================
GL_CreateTextureArray
creates texture array from buffer
================
*/
int GL_CreateTextureArray( const char *name, int width, int height, int depth, const void *buffer, texFlags_t flags )
{
rgbdata_t r_empty;
memset( &r_empty, 0, sizeof( r_empty ));
r_empty.width = Q_max( width, 1 );
r_empty.height = Q_max( height, 1 );
r_empty.depth = Q_max( depth, 1 );
r_empty.type = PF_RGBA_32;
r_empty.size = r_empty.width * r_empty.height * r_empty.depth * 4;
r_empty.buffer = (byte *)buffer;
// clear invalid combinations
ClearBits( flags, TF_CUBEMAP|TF_SKYSIDE|TF_HAS_LUMA|TF_MAKELUMA|TF_ALPHACONTRAST );
// if image not luminance it will have color
if( !FBitSet( flags, TF_LUMINANCE ))
SetBits( r_empty.flags, IMAGE_HAS_COLOR );
if( FBitSet( flags, TF_HAS_ALPHA ))
SetBits( r_empty.flags, IMAGE_HAS_ALPHA );
if( FBitSet( flags, TF_TEXTURE_3D ))
{
if( !GL_Support( GL_TEXTURE_3D_EXT ))
return 0;
}
else
{
if( !GL_Support( GL_TEXTURE_ARRAY_EXT ))
return 0;
SetBits( r_empty.flags, IMAGE_MULTILAYER );
}
return GL_LoadTextureInternal( name, &r_empty, flags );
}
/*
================
GL_FindTexture
================
*/
int GL_FindTexture( const char *name )
{
gl_texture_t *tex;
if( !GL_CheckTexName( name ))
return 0;
// see if already loaded
if(( tex = GL_TextureForName( name )))
return (tex - gl_textures);
return 0;
}
/*
================
GL_FreeTexture
================
*/
void GL_FreeTexture( GLenum texnum )
{
// number 0 it's already freed
if( texnum <= 0 ) return;
GL_DeleteTexture( &gl_textures[texnum] );
}
/*
================
GL_ProcessTexture
================
*/
void GL_ProcessTexture( int texnum, float gamma, int topColor, int bottomColor )
{
gl_texture_t *image;
rgbdata_t *pic;
int flags = 0;
if( texnum <= 0 || texnum >= MAX_TEXTURES )
return; // missed image
image = &gl_textures[texnum];
// select mode
if( gamma != -1.0f )
{
flags = IMAGE_LIGHTGAMMA;
}
else if( topColor != -1 && bottomColor != -1 )
{
flags = IMAGE_REMAP;
}
else
{
gEngfuncs.Con_Printf( S_ERROR "%s: bad operation for %s\n", __func__, image->name );
return;
}
if( !image->original )
{
gEngfuncs.Con_Printf( S_ERROR "%s: no input data for %s\n", __func__, image->name );
return;
}
if( ImageCompressed( image->original->type ))
{
gEngfuncs.Con_Printf( S_ERROR "%s: can't process compressed texture %s\n", __func__, image->name );
return;
}
// all the operations makes over the image copy not an original
pic = gEngfuncs.FS_CopyImage( image->original );
// we need to expand image into RGBA buffer
if( pic->type == PF_INDEXED_24 || pic->type == PF_INDEXED_32 )
flags |= IMAGE_FORCE_RGBA;
gEngfuncs.Image_Process( &pic, topColor, bottomColor, flags, 0.0f );
GL_UploadTexture( image, pic );
GL_ApplyTextureParams( image ); // update texture filter, wrap etc
gEngfuncs.FS_FreeImage( pic );
}
/*
================
GL_TexMemory
return size of all uploaded textures
================
*/
int GL_TexMemory( void )
{
int i, total = 0;
for( i = 0; i < gl_numTextures; i++ )
total += gl_textures[i].size;
return total;
}
/*
==============================================================================
INTERNAL TEXTURES
==============================================================================
*/
/*
==================
GL_FakeImage
==================
*/
static rgbdata_t *GL_FakeImage( int width, int height, int depth, int flags )
{
static byte data2D[1024]; // 16x16x4
static rgbdata_t r_image;
// also use this for bad textures, but without alpha
r_image.width = Q_max( 1, width );
r_image.height = Q_max( 1, height );
r_image.depth = Q_max( 1, depth );
r_image.flags = flags;
r_image.type = PF_RGBA_32;
r_image.size = r_image.width * r_image.height * r_image.depth * 4;
r_image.buffer = (r_image.size > sizeof( data2D )) ? NULL : data2D;
r_image.palette = NULL;
r_image.numMips = 1;
r_image.encode = 0;
if( FBitSet( r_image.flags, IMAGE_CUBEMAP ))
r_image.size *= 6;
memset( data2D, 0xFF, sizeof( data2D ));
return &r_image;
}
/*
==================
R_InitDlightTexture
==================
*/
void R_InitDlightTexture( void )
{
rgbdata_t r_image;
if( tr.dlightTexture != 0 )
return; // already initialized
memset( &r_image, 0, sizeof( r_image ));
r_image.width = BLOCK_SIZE;
r_image.height = BLOCK_SIZE;
r_image.flags = IMAGE_HAS_COLOR;
r_image.type = PF_RGBA_32;
r_image.size = r_image.width * r_image.height * 4;
tr.dlightTexture = GL_LoadTextureInternal( "*dlight", &r_image, TF_NOMIPMAP|TF_CLAMP|TF_ATLAS_PAGE );
}
/*
==================
GL_CreateInternalTextures
==================
*/
static void GL_CreateInternalTextures( void )
{
int dx2, dy, d;
int x, y;
rgbdata_t *pic;
// emo-texture from quake1
pic = GL_FakeImage( 16, 16, 1, IMAGE_HAS_COLOR );
for( y = 0; y < 16; y++ )
{
for( x = 0; x < 16; x++ )
{
if(( y < 8 ) ^ ( x < 8 ))
((uint *)pic->buffer)[y*16+x] = 0xFFFF00FF;
else ((uint *)pic->buffer)[y*16+x] = 0xFF000000;
}
}
tr.defaultTexture = GL_LoadTextureInternal( REF_DEFAULT_TEXTURE, pic, TF_COLORMAP );
// particle texture from quake1
pic = GL_FakeImage( 8, 8, 1, IMAGE_HAS_COLOR|IMAGE_HAS_ALPHA );
for( x = 0; x < 8; x++ )
{
for( y = 0; y < 8; y++ )
{
if( dottexture[x][y] )
pic->buffer[( y * 8 + x ) * 4 + 3] = 255;
else pic->buffer[( y * 8 + x ) * 4 + 3] = 0;
}
}
tr.particleTexture = GL_LoadTextureInternal( REF_PARTICLE_TEXTURE, pic, TF_CLAMP );
// white texture
pic = GL_FakeImage( 4, 4, 1, IMAGE_HAS_COLOR );
for( x = 0; x < 16; x++ )
((uint *)pic->buffer)[x] = 0xFFFFFFFF;
tr.whiteTexture = GL_LoadTextureInternal( REF_WHITE_TEXTURE, pic, TF_COLORMAP );
// gray texture
pic = GL_FakeImage( 4, 4, 1, IMAGE_HAS_COLOR );
for( x = 0; x < 16; x++ )
((uint *)pic->buffer)[x] = 0xFF7F7F7F;
tr.grayTexture = GL_LoadTextureInternal( REF_GRAY_TEXTURE, pic, TF_COLORMAP );
// black texture
pic = GL_FakeImage( 4, 4, 1, IMAGE_HAS_COLOR );
for( x = 0; x < 16; x++ )
((uint *)pic->buffer)[x] = 0xFF000000;
tr.blackTexture = GL_LoadTextureInternal( REF_BLACK_TEXTURE, pic, TF_COLORMAP );
// cinematic dummy
pic = GL_FakeImage( 640, 100, 1, IMAGE_HAS_COLOR );
tr.cinTexture = GL_LoadTextureInternal( "*cintexture", pic, TF_NOMIPMAP|TF_CLAMP );
}
/*
===============
R_TextureList_f
===============
*/
void R_TextureList_f( void )
{
gl_texture_t *image;
int i, texCount, bytes = 0;
gEngfuncs.Con_Printf( "\n" );
gEngfuncs.Con_Printf( " -id- -w- -h- -size- -fmt- -type- -data- -encode- -wrap- -depth- -name--------\n" );
for( i = texCount = 0, image = gl_textures; i < gl_numTextures; i++, image++ )
{
if( !image->texnum ) continue;
bytes += image->size;
texCount++;
gEngfuncs.Con_Printf( "%4i: ", i );
gEngfuncs.Con_Printf( "%4i %4i ", image->width, image->height );
gEngfuncs.Con_Printf( "%12s ", Q_memprint( image->size ));
switch( image->format )
{
case GL_COMPRESSED_RGBA_ARB:
gEngfuncs.Con_Printf( "CRGBA " );
break;
case GL_COMPRESSED_RGB_ARB:
gEngfuncs.Con_Printf( "CRGB " );
break;
case GL_COMPRESSED_LUMINANCE_ALPHA_ARB:
gEngfuncs.Con_Printf( "CLA " );
break;
case GL_COMPRESSED_LUMINANCE_ARB:
gEngfuncs.Con_Printf( "CL " );
break;
case GL_COMPRESSED_ALPHA_ARB:
gEngfuncs.Con_Printf( "CA " );
break;
case GL_COMPRESSED_INTENSITY_ARB:
gEngfuncs.Con_Printf( "CI " );
break;
case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
gEngfuncs.Con_Printf( "DXT1c " );
break;
case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
gEngfuncs.Con_Printf( "DXT1a " );
break;
case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
gEngfuncs.Con_Printf( "DXT3 " );
break;
case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
gEngfuncs.Con_Printf( "DXT5 " );
break;
case GL_COMPRESSED_RED_GREEN_RGTC2_EXT:
case GL_COMPRESSED_LUMINANCE_ALPHA_3DC_ATI:
gEngfuncs.Con_Printf( "ATI2 " );
break;
case GL_RGBA:
gEngfuncs.Con_Printf( "RGBA " );
break;
case GL_RGBA8:
gEngfuncs.Con_Printf( "RGBA8 " );
break;
case GL_RGBA4:
gEngfuncs.Con_Printf( "RGBA4 " );
break;
case GL_RGB:
gEngfuncs.Con_Printf( "RGB " );
break;
case GL_RGB8:
gEngfuncs.Con_Printf( "RGB8 " );
break;
case GL_RGB5:
gEngfuncs.Con_Printf( "RGB5 " );
break;
case GL_LUMINANCE4_ALPHA4:
gEngfuncs.Con_Printf( "L4A4 " );
break;
case GL_LUMINANCE_ALPHA:
case GL_LUMINANCE8_ALPHA8:
gEngfuncs.Con_Printf( "L8A8 " );
break;
case GL_LUMINANCE4:
gEngfuncs.Con_Printf( "L4 " );
break;
case GL_LUMINANCE:
case GL_LUMINANCE8:
gEngfuncs.Con_Printf( "L8 " );
break;
case GL_ALPHA8:
gEngfuncs.Con_Printf( "A8 " );
break;
case GL_INTENSITY8:
gEngfuncs.Con_Printf( "I8 " );
break;
case GL_DEPTH_COMPONENT:
case GL_DEPTH_COMPONENT24:
gEngfuncs.Con_Printf( "DPTH24" );
break;
case GL_DEPTH_COMPONENT32F:
gEngfuncs.Con_Printf( "DPTH32" );
break;
case GL_LUMINANCE16F_ARB:
gEngfuncs.Con_Printf( "L16F " );
break;
case GL_LUMINANCE32F_ARB:
gEngfuncs.Con_Printf( "L32F " );
break;
case GL_LUMINANCE_ALPHA16F_ARB:
gEngfuncs.Con_Printf( "LA16F " );
break;
case GL_LUMINANCE_ALPHA32F_ARB:
gEngfuncs.Con_Printf( "LA32F " );
break;
case GL_RG16F:
gEngfuncs.Con_Printf( "RG16F " );
break;
case GL_RG32F:
gEngfuncs.Con_Printf( "RG32F " );
break;
case GL_RGB16F_ARB:
gEngfuncs.Con_Printf( "RGB16F" );
break;
case GL_RGB32F_ARB:
gEngfuncs.Con_Printf( "RGB32F" );
break;
case GL_RGBA16F_ARB:
gEngfuncs.Con_Printf( "RGBA16F" );
break;
case GL_RGBA32F_ARB:
gEngfuncs.Con_Printf( "RGBA32F" );
break;
default:
gEngfuncs.Con_Printf( " ^1ERROR^7 " );
break;
}
switch( image->target )
{
case GL_TEXTURE_1D:
gEngfuncs.Con_Printf( " 1D " );
break;
case GL_TEXTURE_2D:
gEngfuncs.Con_Printf( " 2D " );
break;
case GL_TEXTURE_3D:
gEngfuncs.Con_Printf( " 3D " );
break;
case GL_TEXTURE_CUBE_MAP_ARB:
gEngfuncs.Con_Printf( "CUBE " );
break;
case GL_TEXTURE_RECTANGLE_EXT:
gEngfuncs.Con_Printf( "RECT " );
break;
case GL_TEXTURE_2D_ARRAY_EXT:
gEngfuncs.Con_Printf( "ARRAY " );
break;
case GL_TEXTURE_2D_MULTISAMPLE:
gEngfuncs.Con_Printf( "MSAA ");
break;
default:
gEngfuncs.Con_Printf( "???? " );
break;
}
if( image->flags & TF_NORMALMAP )
gEngfuncs.Con_Printf( "normal " );
else gEngfuncs.Con_Printf( "diffuse " );
switch( image->encode )
{
case DXT_ENCODE_COLOR_YCoCg:
gEngfuncs.Con_Printf( "YCoCg " );
break;
case DXT_ENCODE_NORMAL_AG_ORTHO:
gEngfuncs.Con_Printf( "ortho " );
break;
case DXT_ENCODE_NORMAL_AG_STEREO:
gEngfuncs.Con_Printf( "stereo " );
break;
case DXT_ENCODE_NORMAL_AG_PARABOLOID:
gEngfuncs.Con_Printf( "parabolic " );
break;
case DXT_ENCODE_NORMAL_AG_QUARTIC:
gEngfuncs.Con_Printf( "quartic " );
break;
case DXT_ENCODE_NORMAL_AG_AZIMUTHAL:
gEngfuncs.Con_Printf( "azimuthal " );
break;
default:
gEngfuncs.Con_Printf( "default " );
break;
}
if( image->flags & TF_CLAMP )
gEngfuncs.Con_Printf( "clamp " );
else if( image->flags & TF_BORDER )
gEngfuncs.Con_Printf( "border " );
else gEngfuncs.Con_Printf( "repeat " );
gEngfuncs.Con_Printf( " %d ", image->depth );
gEngfuncs.Con_Printf( " %s\n", image->name );
}
gEngfuncs.Con_Printf( "---------------------------------------------------------\n" );
gEngfuncs.Con_Printf( "%i total textures\n", texCount );
gEngfuncs.Con_Printf( "%s total memory used\n", Q_memprint( bytes ));
gEngfuncs.Con_Printf( "\n" );
}
/*
===============
R_InitImages
===============
*/
void R_InitImages( void )
{
memset( gl_textures, 0, sizeof( gl_textures ));
memset( gl_texturesHashTable, 0, sizeof( gl_texturesHashTable ));
gl_numTextures = 0;
// create unused 0-entry
Q_strncpy( gl_textures->name, "*unused*", sizeof( gl_textures->name ));
gl_textures->hashValue = COM_HashKey( gl_textures->name, TEXTURES_HASH_SIZE );
gl_textures->nextHash = gl_texturesHashTable[gl_textures->hashValue];
gl_texturesHashTable[gl_textures->hashValue] = gl_textures;
gl_numTextures = 1;
// validate cvars
R_SetTextureParameters();
GL_CreateInternalTextures();
R_InitRipples();
gEngfuncs.Cmd_AddCommand( "texturelist", R_TextureList_f, "display loaded textures list" );
}
/*
===============
R_ShutdownImages
===============
*/
void R_ShutdownImages( void )
{
gl_texture_t *tex;
int i;
gEngfuncs.Cmd_RemoveCommand( "texturelist" );
GL_CleanupAllTextureUnits();
for( i = 0, tex = gl_textures; i < gl_numTextures; i++, tex++ )
GL_DeleteTexture( tex );
memset( tr.lightmapTextures, 0, sizeof( tr.lightmapTextures ));
memset( gl_texturesHashTable, 0, sizeof( gl_texturesHashTable ));
memset( gl_textures, 0, sizeof( gl_textures ));
gl_numTextures = 0;
}
void R_TextureReplacementReport( const char *modelname, int gl_texturenum, const char *foundpath )
{
if( host_allow_materials->value != 2.0f )
return;
if( gl_texturenum > 0 )
gEngfuncs.Con_Printf( "Looking for %s tex replacement..." S_GREEN "OK (%s)\n", modelname, foundpath );
else if( gl_texturenum < 0 )
gEngfuncs.Con_Printf( "Looking for %s tex replacement..." S_YELLOW "MISS (%s)\n", modelname, foundpath );
else
gEngfuncs.Con_Printf( "Looking for %s tex replacement..." S_RED "FAIL (%s)\n", modelname, foundpath );
}
qboolean R_SearchForTextureReplacement( char *out, size_t size, const char *modelname, const char *fmt, ... )
{
va_list ap;
int ret;
va_start( ap, fmt );
ret = Q_vsnprintf( out, size, fmt, ap );
va_end( ap );
if( ret < 0 )
{
R_TextureReplacementReport( modelname, -1, "overflow" );
return false;
}
if( gEngfuncs.fsapi->FileExists( out, false ))
return true;
R_TextureReplacementReport( modelname, -1, out );
return false;
}