q3rally/engine/code/renderergl2/tr_backend.c
Zack Middleton 22a0949a26 ioquake3 resync to commit 4610a240 from bc8737d7
Upgrade to build and run from VS2019
Upgrades to Xcode project and Apple Silicon support
Update SDL2 to 2.0.14
Updated SDL2 include files to fix Mac build in GitHub Actions
Added another mention of arm64 to command line help
Restored original opus sse files, excluded from Xcode
Added arm64 to the post-build symlinking step
Merge branch 'main' into xcode
Merge branch 'main' into vs2019
Added shell script to compile Universal 2 binary (x86_64+arm64)
Reverting alert style to deprecated methods
Upgrades to Xcode project and Apple Silicon support
Update SDL2 to 2.0.14
Added another mention of arm64 to command line help
Restored original opus sse files, excluded from Xcode
Added arm64 to the post-build symlinking step
Added shell script to compile Universal 2 binary (x86_64+arm64)
Reverting alert style to deprecated methods
Merge branch 'xcode' of https://github.com/tomkidd/ioq3 into xcode
Removed signature from SDL dylib, enabled dark mode on macOS.
spaces > tabs
Ad-hoc signed libSDL2-2.0.0.dylib
Fix compiling against SDL 2.0.17
UB2 now signs and notarizes, upgraded to SDL 2.0.16
Architectures in libSDL2 restored for ppc and i386
Merge remote-tracking branch 'upstream/main' into vs2019
Update SDL2 to 2.0.16
Added rudimentary support for automatically finding Microsoft Store version of Quake 3
GHA deprecated Ubuntu 16.04 - update to 18.04
qsort cannot be called with NULL
Merge remote-tracking branch 'upstream/main' into vs2019
Addressed string concatenation issue and added dummy method for Mac/Linux
Added missing variable.
Merge remote-tracking branch 'upstream/main' into xcode
Updated SDL 2.0.16 headers and Mac version of libraries to fix GitHub actions
Addressed PR suggestions
Modified MS Store path handling to better follow the pattern of Steam/GOG
Merge pull request #481 from tomkidd/xcode
Merge pull request #482 from tomkidd/vs2019
OpenGL2: Fix r_grayscale 1 making everything solid black
Print full GL_EXTENSIONS list for OpenGL contexts before 3.0
Fix being unable to enter Team Arena CD key
OpenGL2: GL_DEPTH_TEXTURE_MODE was removed from OpenGL 3.0/Core
Improve setting Microsoft Store path
Update building for macOS in README
Make macOS arm64 default to target macOS 11 in Makefile
Fix error when cross-compiling for macOS arm64 using Makefile
Fix passing arguments to VM dylib on Apple M1
Fix compiling on older macOS
Fix memory corruption in S_TransferPaintBuffer
Fix memset
Fix hex digit
Fix uninitialized variable
some old URL and doc updates
Update README.md
Update FUNDING.yml
code/curl: update ifdef condition for MCST-LCC compiler in mcst-lcc compiler => 1.25 added a new macro definition to determine compiler
Revert "code/curl: update ifdef condition for MCST-LCC compiler"
Revert "E2K: fixed build by MCST lcc compiler when using USE_CURL=1 option"
More predictable mesh normals generation
vm_x86.c: Add `defined(_M_IX86) || defined(_M_X64)` (fix for VS2019)
Add keys for SDL 2.0.14's new gamepad buttons
Fix in_availableJoysticks cvar not updating
Fix (disabled) Wavelet sound decompression
Update to SDL 2.24.0 and add separate macOS UB2 dylib
Update macOS UB1 to SDL 2.0.22
Fix running make-macosx{,-ub2}.sh on Linux
Update MSVC .lib files to SDL 2.24.0
2023-03-02 23:13:18 -06:00

1795 lines
47 KiB
C

/*
===========================================================================
Copyright (C) 1999-2005 Id Software, Inc.
This file is part of Quake III Arena source code.
Quake III Arena source code 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 2 of the License,
or (at your option) any later version.
Quake III Arena source code 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.
You should have received a copy of the GNU General Public License
along with Quake III Arena source code; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
===========================================================================
*/
#include "tr_local.h"
#include "tr_fbo.h"
#include "tr_dsa.h"
backEndData_t *backEndData;
backEndState_t backEnd;
static float s_flipMatrix[16] = {
// convert from our coordinate system (looking down X)
// to OpenGL's coordinate system (looking down -Z)
0, 0, -1, 0,
-1, 0, 0, 0,
0, 1, 0, 0,
0, 0, 0, 1
};
/*
** GL_BindToTMU
*/
void GL_BindToTMU( image_t *image, int tmu )
{
GLuint texture = (tmu == TB_COLORMAP) ? tr.defaultImage->texnum : 0;
GLenum target = GL_TEXTURE_2D;
if (image)
{
if (image->flags & IMGFLAG_CUBEMAP)
target = GL_TEXTURE_CUBE_MAP;
image->frameUsed = tr.frameCount;
texture = image->texnum;
}
else
{
ri.Printf(PRINT_WARNING, "GL_BindToTMU: NULL image\n");
}
GL_BindMultiTexture(GL_TEXTURE0 + tmu, target, texture);
}
/*
** GL_Cull
*/
void GL_Cull( int cullType ) {
if ( glState.faceCulling == cullType ) {
return;
}
if ( cullType == CT_TWO_SIDED )
{
qglDisable( GL_CULL_FACE );
}
else
{
qboolean cullFront = (cullType == CT_FRONT_SIDED);
if ( glState.faceCulling == CT_TWO_SIDED )
qglEnable( GL_CULL_FACE );
if ( glState.faceCullFront != cullFront )
qglCullFace( cullFront ? GL_FRONT : GL_BACK );
glState.faceCullFront = cullFront;
}
glState.faceCulling = cullType;
}
/*
** GL_State
**
** This routine is responsible for setting the most commonly changed state
** in Q3.
*/
void GL_State( unsigned long stateBits )
{
unsigned long diff = stateBits ^ glState.glStateBits;
if ( !diff )
{
return;
}
//
// check depthFunc bits
//
if ( diff & GLS_DEPTHFUNC_BITS )
{
if ( stateBits & GLS_DEPTHFUNC_EQUAL )
{
qglDepthFunc( GL_EQUAL );
}
else if ( stateBits & GLS_DEPTHFUNC_GREATER)
{
qglDepthFunc( GL_GREATER );
}
else
{
qglDepthFunc( GL_LEQUAL );
}
}
//
// check blend bits
//
if ( diff & ( GLS_SRCBLEND_BITS | GLS_DSTBLEND_BITS ) )
{
uint32_t oldState = glState.glStateBits & ( GLS_SRCBLEND_BITS | GLS_DSTBLEND_BITS );
uint32_t newState = stateBits & ( GLS_SRCBLEND_BITS | GLS_DSTBLEND_BITS );
uint32_t storedState = glState.storedGlState & ( GLS_SRCBLEND_BITS | GLS_DSTBLEND_BITS );
if (oldState == 0)
{
qglEnable( GL_BLEND );
}
else if (newState == 0)
{
qglDisable( GL_BLEND );
}
if (newState != 0 && storedState != newState)
{
GLenum srcFactor = GL_ONE, dstFactor = GL_ONE;
glState.storedGlState &= ~( GLS_SRCBLEND_BITS | GLS_DSTBLEND_BITS );
glState.storedGlState |= newState;
switch ( stateBits & GLS_SRCBLEND_BITS )
{
case GLS_SRCBLEND_ZERO:
srcFactor = GL_ZERO;
break;
case GLS_SRCBLEND_ONE:
srcFactor = GL_ONE;
break;
case GLS_SRCBLEND_DST_COLOR:
srcFactor = GL_DST_COLOR;
break;
case GLS_SRCBLEND_ONE_MINUS_DST_COLOR:
srcFactor = GL_ONE_MINUS_DST_COLOR;
break;
case GLS_SRCBLEND_SRC_ALPHA:
srcFactor = GL_SRC_ALPHA;
break;
case GLS_SRCBLEND_ONE_MINUS_SRC_ALPHA:
srcFactor = GL_ONE_MINUS_SRC_ALPHA;
break;
case GLS_SRCBLEND_DST_ALPHA:
srcFactor = GL_DST_ALPHA;
break;
case GLS_SRCBLEND_ONE_MINUS_DST_ALPHA:
srcFactor = GL_ONE_MINUS_DST_ALPHA;
break;
case GLS_SRCBLEND_ALPHA_SATURATE:
srcFactor = GL_SRC_ALPHA_SATURATE;
break;
default:
ri.Error( ERR_DROP, "GL_State: invalid src blend state bits" );
break;
}
switch ( stateBits & GLS_DSTBLEND_BITS )
{
case GLS_DSTBLEND_ZERO:
dstFactor = GL_ZERO;
break;
case GLS_DSTBLEND_ONE:
dstFactor = GL_ONE;
break;
case GLS_DSTBLEND_SRC_COLOR:
dstFactor = GL_SRC_COLOR;
break;
case GLS_DSTBLEND_ONE_MINUS_SRC_COLOR:
dstFactor = GL_ONE_MINUS_SRC_COLOR;
break;
case GLS_DSTBLEND_SRC_ALPHA:
dstFactor = GL_SRC_ALPHA;
break;
case GLS_DSTBLEND_ONE_MINUS_SRC_ALPHA:
dstFactor = GL_ONE_MINUS_SRC_ALPHA;
break;
case GLS_DSTBLEND_DST_ALPHA:
dstFactor = GL_DST_ALPHA;
break;
case GLS_DSTBLEND_ONE_MINUS_DST_ALPHA:
dstFactor = GL_ONE_MINUS_DST_ALPHA;
break;
default:
ri.Error( ERR_DROP, "GL_State: invalid dst blend state bits" );
break;
}
qglBlendFunc( srcFactor, dstFactor );
}
}
//
// check depthmask
//
if ( diff & GLS_DEPTHMASK_TRUE )
{
if ( stateBits & GLS_DEPTHMASK_TRUE )
{
qglDepthMask( GL_TRUE );
}
else
{
qglDepthMask( GL_FALSE );
}
}
//
// fill/line mode
//
if ( diff & GLS_POLYMODE_LINE )
{
if ( stateBits & GLS_POLYMODE_LINE )
{
qglPolygonMode( GL_FRONT_AND_BACK, GL_LINE );
}
else
{
qglPolygonMode( GL_FRONT_AND_BACK, GL_FILL );
}
}
//
// depthtest
//
if ( diff & GLS_DEPTHTEST_DISABLE )
{
if ( stateBits & GLS_DEPTHTEST_DISABLE )
{
qglDisable( GL_DEPTH_TEST );
}
else
{
qglEnable( GL_DEPTH_TEST );
}
}
glState.glStateBits = stateBits;
}
void GL_SetProjectionMatrix(mat4_t matrix)
{
Mat4Copy(matrix, glState.projection);
Mat4Multiply(glState.projection, glState.modelview, glState.modelviewProjection);
}
void GL_SetModelviewMatrix(mat4_t matrix)
{
Mat4Copy(matrix, glState.modelview);
Mat4Multiply(glState.projection, glState.modelview, glState.modelviewProjection);
}
/*
================
RB_Hyperspace
A player has predicted a teleport, but hasn't arrived yet
================
*/
static void RB_Hyperspace( void ) {
float c;
if ( !backEnd.isHyperspace ) {
// do initialization shit
}
c = ( backEnd.refdef.time & 255 ) / 255.0f;
qglClearColor( c, c, c, 1 );
qglClear( GL_COLOR_BUFFER_BIT );
qglClearColor(0.0f, 0.0f, 0.0f, 1.0f);
backEnd.isHyperspace = qtrue;
}
static void SetViewportAndScissor( void ) {
GL_SetProjectionMatrix( backEnd.viewParms.projectionMatrix );
// set the window clipping
qglViewport( backEnd.viewParms.viewportX, backEnd.viewParms.viewportY,
backEnd.viewParms.viewportWidth, backEnd.viewParms.viewportHeight );
qglScissor( backEnd.viewParms.viewportX, backEnd.viewParms.viewportY,
backEnd.viewParms.viewportWidth, backEnd.viewParms.viewportHeight );
}
/*
=================
RB_BeginDrawingView
Any mirrored or portaled views have already been drawn, so prepare
to actually render the visible surfaces for this view
=================
*/
void RB_BeginDrawingView (void) {
int clearBits = 0;
// sync with gl if needed
if ( r_finish->integer == 1 && !glState.finishCalled ) {
qglFinish ();
glState.finishCalled = qtrue;
}
if ( r_finish->integer == 0 ) {
glState.finishCalled = qtrue;
}
// we will need to change the projection matrix before drawing
// 2D images again
backEnd.projection2D = qfalse;
if (glRefConfig.framebufferObject)
{
FBO_t *fbo = backEnd.viewParms.targetFbo;
// FIXME: HUGE HACK: render to the screen fbo if we've already postprocessed the frame and aren't drawing more world
// drawing more world check is in case of double renders, such as skyportals
if (fbo == NULL && !(backEnd.framePostProcessed && (backEnd.refdef.rdflags & RDF_NOWORLDMODEL)))
fbo = tr.renderFbo;
if (tr.renderCubeFbo && fbo == tr.renderCubeFbo)
{
cubemap_t *cubemap = &tr.cubemaps[backEnd.viewParms.targetFboCubemapIndex];
FBO_AttachImage(fbo, cubemap->image, GL_COLOR_ATTACHMENT0_EXT, backEnd.viewParms.targetFboLayer);
}
FBO_Bind(fbo);
}
//
// set the modelview matrix for the viewer
//
SetViewportAndScissor();
// ensures that depth writes are enabled for the depth clear
GL_State( GLS_DEFAULT );
// clear relevant buffers
clearBits = GL_DEPTH_BUFFER_BIT;
if ( r_measureOverdraw->integer || r_shadows->integer == 2 )
{
clearBits |= GL_STENCIL_BUFFER_BIT;
}
if ( r_fastsky->integer && !( backEnd.refdef.rdflags & RDF_NOWORLDMODEL ) )
{
clearBits |= GL_COLOR_BUFFER_BIT; // FIXME: only if sky shaders have been used
}
// clear to black for cube maps
if (tr.renderCubeFbo && backEnd.viewParms.targetFbo == tr.renderCubeFbo)
{
clearBits |= GL_COLOR_BUFFER_BIT;
}
qglClear( clearBits );
if ( ( backEnd.refdef.rdflags & RDF_HYPERSPACE ) )
{
RB_Hyperspace();
return;
}
else
{
backEnd.isHyperspace = qfalse;
}
// we will only draw a sun if there was sky rendered in this view
backEnd.skyRenderedThisView = qfalse;
// clip to the plane of the portal
if ( backEnd.viewParms.isPortal ) {
#if 0
float plane[4];
GLdouble plane2[4];
plane[0] = backEnd.viewParms.portalPlane.normal[0];
plane[1] = backEnd.viewParms.portalPlane.normal[1];
plane[2] = backEnd.viewParms.portalPlane.normal[2];
plane[3] = backEnd.viewParms.portalPlane.dist;
plane2[0] = DotProduct (backEnd.viewParms.or.axis[0], plane);
plane2[1] = DotProduct (backEnd.viewParms.or.axis[1], plane);
plane2[2] = DotProduct (backEnd.viewParms.or.axis[2], plane);
plane2[3] = DotProduct (plane, backEnd.viewParms.or.origin) - plane[3];
#endif
GL_SetModelviewMatrix( s_flipMatrix );
}
}
/*
==================
RB_RenderDrawSurfList
==================
*/
void RB_RenderDrawSurfList( drawSurf_t *drawSurfs, int numDrawSurfs ) {
shader_t *shader = NULL, *oldShader;
int fogNum, oldFogNum;
int entityNum, oldEntityNum;
int dlighted, oldDlighted;
int pshadowed, oldPshadowed;
int cubemapIndex, oldCubemapIndex;
qboolean depthRange, oldDepthRange, isCrosshair, wasCrosshair;
int i;
drawSurf_t *drawSurf;
int oldSort;
double originalTime;
FBO_t* fbo = NULL;
// save original time for entity shader offsets
originalTime = backEnd.refdef.floatTime;
fbo = glState.currentFBO;
// draw everything
oldEntityNum = -1;
backEnd.currentEntity = &tr.worldEntity;
oldShader = NULL;
oldFogNum = -1;
oldDepthRange = qfalse;
wasCrosshair = qfalse;
oldDlighted = qfalse;
oldPshadowed = qfalse;
oldCubemapIndex = -1;
oldSort = -1;
backEnd.pc.c_surfaces += numDrawSurfs;
for (i = 0, drawSurf = drawSurfs ; i < numDrawSurfs ; i++, drawSurf++) {
if ( drawSurf->sort == oldSort && drawSurf->cubemapIndex == oldCubemapIndex) {
if (backEnd.depthFill && shader && shader->sort != SS_OPAQUE)
continue;
// fast path, same as previous sort
rb_surfaceTable[ *drawSurf->surface ]( drawSurf->surface );
continue;
}
oldSort = drawSurf->sort;
R_DecomposeSort( drawSurf->sort, &entityNum, &shader, &fogNum, &dlighted, &pshadowed );
cubemapIndex = drawSurf->cubemapIndex;
//
// change the tess parameters if needed
// a "entityMergable" shader is a shader that can have surfaces from separate
// entities merged into a single batch, like smoke and blood puff sprites
if ( shader != NULL && ( shader != oldShader || fogNum != oldFogNum || dlighted != oldDlighted || pshadowed != oldPshadowed || cubemapIndex != oldCubemapIndex
|| ( entityNum != oldEntityNum && !shader->entityMergable ) ) ) {
if (oldShader != NULL) {
RB_EndSurface();
}
RB_BeginSurface( shader, fogNum, cubemapIndex );
backEnd.pc.c_surfBatches++;
oldShader = shader;
oldFogNum = fogNum;
oldDlighted = dlighted;
oldPshadowed = pshadowed;
oldCubemapIndex = cubemapIndex;
}
if (backEnd.depthFill && shader && shader->sort != SS_OPAQUE)
continue;
//
// change the modelview matrix if needed
//
if ( entityNum != oldEntityNum ) {
depthRange = isCrosshair = qfalse;
if ( entityNum != REFENTITYNUM_WORLD ) {
backEnd.currentEntity = &backEnd.refdef.entities[entityNum];
// FIXME: e.shaderTime must be passed as int to avoid fp-precision loss issues
backEnd.refdef.floatTime = originalTime - (double)backEnd.currentEntity->e.shaderTime;
// we have to reset the shaderTime as well otherwise image animations start
// from the wrong frame
tess.shaderTime = backEnd.refdef.floatTime - tess.shader->timeOffset;
// set up the transformation matrix
R_RotateForEntity( backEnd.currentEntity, &backEnd.viewParms, &backEnd.or );
// set up the dynamic lighting if needed
if ( backEnd.currentEntity->needDlights ) {
R_TransformDlights( backEnd.refdef.num_dlights, backEnd.refdef.dlights, &backEnd.or );
}
if(backEnd.currentEntity->e.renderfx & RF_DEPTHHACK)
{
// hack the depth range to prevent view model from poking into walls
depthRange = qtrue;
if(backEnd.currentEntity->e.renderfx & RF_CROSSHAIR)
isCrosshair = qtrue;
}
} else {
backEnd.currentEntity = &tr.worldEntity;
backEnd.refdef.floatTime = originalTime;
backEnd.or = backEnd.viewParms.world;
// we have to reset the shaderTime as well otherwise image animations on
// the world (like water) continue with the wrong frame
tess.shaderTime = backEnd.refdef.floatTime - tess.shader->timeOffset;
R_TransformDlights( backEnd.refdef.num_dlights, backEnd.refdef.dlights, &backEnd.or );
}
GL_SetModelviewMatrix( backEnd.or.modelMatrix );
//
// change depthrange. Also change projection matrix so first person weapon does not look like coming
// out of the screen.
//
if (oldDepthRange != depthRange || wasCrosshair != isCrosshair)
{
if (depthRange)
{
if(backEnd.viewParms.stereoFrame != STEREO_CENTER)
{
if(isCrosshair)
{
if(oldDepthRange)
{
// was not a crosshair but now is, change back proj matrix
GL_SetProjectionMatrix( backEnd.viewParms.projectionMatrix );
}
}
else
{
viewParms_t temp = backEnd.viewParms;
R_SetupProjection(&temp, r_znear->value, 0, qfalse);
GL_SetProjectionMatrix( temp.projectionMatrix );
}
}
if(!oldDepthRange)
qglDepthRange (0, 0.3);
}
else
{
if(!wasCrosshair && backEnd.viewParms.stereoFrame != STEREO_CENTER)
{
GL_SetProjectionMatrix( backEnd.viewParms.projectionMatrix );
}
qglDepthRange (0, 1);
}
oldDepthRange = depthRange;
wasCrosshair = isCrosshair;
}
oldEntityNum = entityNum;
}
// add the triangles for this surface
rb_surfaceTable[ *drawSurf->surface ]( drawSurf->surface );
}
backEnd.refdef.floatTime = originalTime;
// draw the contents of the last shader batch
if (oldShader != NULL) {
RB_EndSurface();
}
if (glRefConfig.framebufferObject)
FBO_Bind(fbo);
// go back to the world modelview matrix
GL_SetModelviewMatrix( backEnd.viewParms.world.modelMatrix );
qglDepthRange (0, 1);
}
/*
============================================================================
RENDER BACK END FUNCTIONS
============================================================================
*/
/*
================
RB_SetGL2D
================
*/
void RB_SetGL2D (void) {
mat4_t matrix;
int width, height;
if (backEnd.projection2D && backEnd.last2DFBO == glState.currentFBO)
return;
backEnd.projection2D = qtrue;
backEnd.last2DFBO = glState.currentFBO;
if (glState.currentFBO)
{
width = glState.currentFBO->width;
height = glState.currentFBO->height;
}
else
{
width = glConfig.vidWidth;
height = glConfig.vidHeight;
}
// set 2D virtual screen size
qglViewport( 0, 0, width, height );
qglScissor( 0, 0, width, height );
Mat4Ortho(0, width, height, 0, 0, 1, matrix);
GL_SetProjectionMatrix(matrix);
Mat4Identity(matrix);
GL_SetModelviewMatrix(matrix);
GL_State( GLS_DEPTHTEST_DISABLE |
GLS_SRCBLEND_SRC_ALPHA |
GLS_DSTBLEND_ONE_MINUS_SRC_ALPHA );
GL_Cull( CT_TWO_SIDED );
// set time for 2D shaders
backEnd.refdef.time = ri.Milliseconds();
backEnd.refdef.floatTime = backEnd.refdef.time * 0.001;
}
/*
=============
RE_StretchRaw
FIXME: not exactly backend
Stretches a raw 32 bit power of 2 bitmap image over the given screen rectangle.
Used for cinematics.
=============
*/
void RE_StretchRaw (int x, int y, int w, int h, int cols, int rows, const byte *data, int client, qboolean dirty) {
int i, j;
int start, end;
vec4_t quadVerts[4];
vec2_t texCoords[4];
if ( !tr.registered ) {
return;
}
R_IssuePendingRenderCommands();
if ( tess.numIndexes ) {
RB_EndSurface();
}
// we definitely want to sync every frame for the cinematics
qglFinish();
start = 0;
if ( r_speeds->integer ) {
start = ri.Milliseconds();
}
// make sure rows and cols are powers of 2
for ( i = 0 ; ( 1 << i ) < cols ; i++ ) {
}
for ( j = 0 ; ( 1 << j ) < rows ; j++ ) {
}
if ( ( 1 << i ) != cols || ( 1 << j ) != rows) {
ri.Error (ERR_DROP, "Draw_StretchRaw: size not a power of 2: %i by %i", cols, rows);
}
RE_UploadCinematic (w, h, cols, rows, data, client, dirty);
GL_BindToTMU(tr.scratchImage[client], TB_COLORMAP);
if ( r_speeds->integer ) {
end = ri.Milliseconds();
ri.Printf( PRINT_ALL, "qglTexSubImage2D %i, %i: %i msec\n", cols, rows, end - start );
}
// FIXME: HUGE hack
if (glRefConfig.framebufferObject)
{
FBO_Bind(backEnd.framePostProcessed ? NULL : tr.renderFbo);
}
RB_SetGL2D();
VectorSet4(quadVerts[0], x, y, 0.0f, 1.0f);
VectorSet4(quadVerts[1], x + w, y, 0.0f, 1.0f);
VectorSet4(quadVerts[2], x + w, y + h, 0.0f, 1.0f);
VectorSet4(quadVerts[3], x, y + h, 0.0f, 1.0f);
VectorSet2(texCoords[0], 0.5f / cols, 0.5f / rows);
VectorSet2(texCoords[1], (cols - 0.5f) / cols, 0.5f / rows);
VectorSet2(texCoords[2], (cols - 0.5f) / cols, (rows - 0.5f) / rows);
VectorSet2(texCoords[3], 0.5f / cols, (rows - 0.5f) / rows);
GLSL_BindProgram(&tr.textureColorShader);
GLSL_SetUniformMat4(&tr.textureColorShader, UNIFORM_MODELVIEWPROJECTIONMATRIX, glState.modelviewProjection);
GLSL_SetUniformVec4(&tr.textureColorShader, UNIFORM_COLOR, colorWhite);
RB_InstantQuad2(quadVerts, texCoords);
}
void RE_UploadCinematic (int w, int h, int cols, int rows, const byte *data, int client, qboolean dirty) {
GLuint texture;
if (!tr.scratchImage[client])
{
ri.Printf(PRINT_WARNING, "RE_UploadCinematic: scratch images not initialized\n");
return;
}
texture = tr.scratchImage[client]->texnum;
// if the scratchImage isn't in the format we want, specify it as a new texture
if ( cols != tr.scratchImage[client]->width || rows != tr.scratchImage[client]->height ) {
tr.scratchImage[client]->width = tr.scratchImage[client]->uploadWidth = cols;
tr.scratchImage[client]->height = tr.scratchImage[client]->uploadHeight = rows;
qglTextureImage2DEXT(texture, GL_TEXTURE_2D, 0, GL_RGB8, cols, rows, 0, GL_RGBA, GL_UNSIGNED_BYTE, data);
qglTextureParameterfEXT(texture, GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
qglTextureParameterfEXT(texture, GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
qglTextureParameterfEXT(texture, GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
qglTextureParameterfEXT(texture, GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
} else {
if (dirty) {
// otherwise, just subimage upload it so that drivers can tell we are going to be changing
// it and don't try and do a texture compression
qglTextureSubImage2DEXT(texture, GL_TEXTURE_2D, 0, 0, 0, cols, rows, GL_RGBA, GL_UNSIGNED_BYTE, data);
}
}
}
/*
=============
RB_SetColor
=============
*/
const void *RB_SetColor( const void *data ) {
const setColorCommand_t *cmd;
cmd = (const setColorCommand_t *)data;
backEnd.color2D[0] = cmd->color[0] * 255;
backEnd.color2D[1] = cmd->color[1] * 255;
backEnd.color2D[2] = cmd->color[2] * 255;
backEnd.color2D[3] = cmd->color[3] * 255;
return (const void *)(cmd + 1);
}
/*
=============
RB_StretchPic
=============
*/
const void *RB_StretchPic ( const void *data ) {
const stretchPicCommand_t *cmd;
shader_t *shader;
int numVerts, numIndexes;
cmd = (const stretchPicCommand_t *)data;
// FIXME: HUGE hack
if (glRefConfig.framebufferObject)
FBO_Bind(backEnd.framePostProcessed ? NULL : tr.renderFbo);
RB_SetGL2D();
shader = cmd->shader;
if ( shader != tess.shader ) {
if ( tess.numIndexes ) {
RB_EndSurface();
}
backEnd.currentEntity = &backEnd.entity2D;
RB_BeginSurface( shader, 0, 0 );
}
RB_CHECKOVERFLOW( 4, 6 );
numVerts = tess.numVertexes;
numIndexes = tess.numIndexes;
tess.numVertexes += 4;
tess.numIndexes += 6;
tess.indexes[ numIndexes ] = numVerts + 3;
tess.indexes[ numIndexes + 1 ] = numVerts + 0;
tess.indexes[ numIndexes + 2 ] = numVerts + 2;
tess.indexes[ numIndexes + 3 ] = numVerts + 2;
tess.indexes[ numIndexes + 4 ] = numVerts + 0;
tess.indexes[ numIndexes + 5 ] = numVerts + 1;
{
uint16_t color[4];
VectorScale4(backEnd.color2D, 257, color);
VectorCopy4(color, tess.color[ numVerts ]);
VectorCopy4(color, tess.color[ numVerts + 1]);
VectorCopy4(color, tess.color[ numVerts + 2]);
VectorCopy4(color, tess.color[ numVerts + 3 ]);
}
tess.xyz[ numVerts ][0] = cmd->x;
tess.xyz[ numVerts ][1] = cmd->y;
tess.xyz[ numVerts ][2] = 0;
tess.texCoords[ numVerts ][0] = cmd->s1;
tess.texCoords[ numVerts ][1] = cmd->t1;
tess.xyz[ numVerts + 1 ][0] = cmd->x + cmd->w;
tess.xyz[ numVerts + 1 ][1] = cmd->y;
tess.xyz[ numVerts + 1 ][2] = 0;
tess.texCoords[ numVerts + 1 ][0] = cmd->s2;
tess.texCoords[ numVerts + 1 ][1] = cmd->t1;
tess.xyz[ numVerts + 2 ][0] = cmd->x + cmd->w;
tess.xyz[ numVerts + 2 ][1] = cmd->y + cmd->h;
tess.xyz[ numVerts + 2 ][2] = 0;
tess.texCoords[ numVerts + 2 ][0] = cmd->s2;
tess.texCoords[ numVerts + 2 ][1] = cmd->t2;
tess.xyz[ numVerts + 3 ][0] = cmd->x;
tess.xyz[ numVerts + 3 ][1] = cmd->y + cmd->h;
tess.xyz[ numVerts + 3 ][2] = 0;
tess.texCoords[ numVerts + 3 ][0] = cmd->s1;
tess.texCoords[ numVerts + 3 ][1] = cmd->t2;
return (const void *)(cmd + 1);
}
/*
=============
RB_DrawSurfs
=============
*/
const void *RB_DrawSurfs( const void *data ) {
const drawSurfsCommand_t *cmd;
qboolean isShadowView;
// finish any 2D drawing if needed
if ( tess.numIndexes ) {
RB_EndSurface();
}
cmd = (const drawSurfsCommand_t *)data;
backEnd.refdef = cmd->refdef;
backEnd.viewParms = cmd->viewParms;
isShadowView = !!(backEnd.viewParms.flags & VPF_DEPTHSHADOW);
// clear the z buffer, set the modelview, etc
RB_BeginDrawingView ();
if (glRefConfig.framebufferObject && (backEnd.viewParms.flags & VPF_DEPTHCLAMP) && glRefConfig.depthClamp)
{
qglEnable(GL_DEPTH_CLAMP);
}
if (glRefConfig.framebufferObject && !(backEnd.refdef.rdflags & RDF_NOWORLDMODEL) && (r_depthPrepass->integer || isShadowView))
{
FBO_t *oldFbo = glState.currentFBO;
vec4_t viewInfo;
VectorSet4(viewInfo, backEnd.viewParms.zFar / r_znear->value, backEnd.viewParms.zFar, 0.0, 0.0);
backEnd.depthFill = qtrue;
qglColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
RB_RenderDrawSurfList( cmd->drawSurfs, cmd->numDrawSurfs );
qglColorMask(!backEnd.colorMask[0], !backEnd.colorMask[1], !backEnd.colorMask[2], !backEnd.colorMask[3]);
backEnd.depthFill = qfalse;
if (!isShadowView)
{
if (tr.msaaResolveFbo)
{
// If we're using multisampling, resolve the depth first
FBO_FastBlit(tr.renderFbo, NULL, tr.msaaResolveFbo, NULL, GL_DEPTH_BUFFER_BIT, GL_NEAREST);
}
else if (tr.renderFbo == NULL && tr.renderDepthImage)
{
// If we're rendering directly to the screen, copy the depth to a texture
// This is incredibly slow on Intel Graphics, so just skip it on there
if (!glRefConfig.intelGraphics)
qglCopyTextureSubImage2DEXT(tr.renderDepthImage->texnum, GL_TEXTURE_2D, 0, 0, 0, 0, 0, glConfig.vidWidth, glConfig.vidHeight);
}
if (tr.hdrDepthFbo)
{
// need the depth in a texture we can do GL_LINEAR sampling on, so copy it to an HDR image
vec4_t srcTexCoords;
VectorSet4(srcTexCoords, 0.0f, 0.0f, 1.0f, 1.0f);
FBO_BlitFromTexture(tr.renderDepthImage, srcTexCoords, NULL, tr.hdrDepthFbo, NULL, NULL, NULL, 0);
}
if (r_sunlightMode->integer && backEnd.viewParms.flags & VPF_USESUNLIGHT)
{
vec4_t quadVerts[4];
vec2_t texCoords[4];
vec4_t box;
FBO_Bind(tr.screenShadowFbo);
box[0] = backEnd.viewParms.viewportX * tr.screenShadowFbo->width / (float)glConfig.vidWidth;
box[1] = backEnd.viewParms.viewportY * tr.screenShadowFbo->height / (float)glConfig.vidHeight;
box[2] = backEnd.viewParms.viewportWidth * tr.screenShadowFbo->width / (float)glConfig.vidWidth;
box[3] = backEnd.viewParms.viewportHeight * tr.screenShadowFbo->height / (float)glConfig.vidHeight;
qglViewport(box[0], box[1], box[2], box[3]);
qglScissor(box[0], box[1], box[2], box[3]);
box[0] = backEnd.viewParms.viewportX / (float)glConfig.vidWidth;
box[1] = backEnd.viewParms.viewportY / (float)glConfig.vidHeight;
box[2] = box[0] + backEnd.viewParms.viewportWidth / (float)glConfig.vidWidth;
box[3] = box[1] + backEnd.viewParms.viewportHeight / (float)glConfig.vidHeight;
texCoords[0][0] = box[0]; texCoords[0][1] = box[3];
texCoords[1][0] = box[2]; texCoords[1][1] = box[3];
texCoords[2][0] = box[2]; texCoords[2][1] = box[1];
texCoords[3][0] = box[0]; texCoords[3][1] = box[1];
box[0] = -1.0f;
box[1] = -1.0f;
box[2] = 1.0f;
box[3] = 1.0f;
VectorSet4(quadVerts[0], box[0], box[3], 0, 1);
VectorSet4(quadVerts[1], box[2], box[3], 0, 1);
VectorSet4(quadVerts[2], box[2], box[1], 0, 1);
VectorSet4(quadVerts[3], box[0], box[1], 0, 1);
GL_State(GLS_DEPTHTEST_DISABLE);
GLSL_BindProgram(&tr.shadowmaskShader);
GL_BindToTMU(tr.renderDepthImage, TB_COLORMAP);
if (r_shadowCascadeZFar->integer != 0)
{
GL_BindToTMU(tr.sunShadowDepthImage[0], TB_SHADOWMAP);
GL_BindToTMU(tr.sunShadowDepthImage[1], TB_SHADOWMAP2);
GL_BindToTMU(tr.sunShadowDepthImage[2], TB_SHADOWMAP3);
GL_BindToTMU(tr.sunShadowDepthImage[3], TB_SHADOWMAP4);
GLSL_SetUniformMat4(&tr.shadowmaskShader, UNIFORM_SHADOWMVP, backEnd.refdef.sunShadowMvp[0]);
GLSL_SetUniformMat4(&tr.shadowmaskShader, UNIFORM_SHADOWMVP2, backEnd.refdef.sunShadowMvp[1]);
GLSL_SetUniformMat4(&tr.shadowmaskShader, UNIFORM_SHADOWMVP3, backEnd.refdef.sunShadowMvp[2]);
GLSL_SetUniformMat4(&tr.shadowmaskShader, UNIFORM_SHADOWMVP4, backEnd.refdef.sunShadowMvp[3]);
}
else
{
GL_BindToTMU(tr.sunShadowDepthImage[3], TB_SHADOWMAP);
GLSL_SetUniformMat4(&tr.shadowmaskShader, UNIFORM_SHADOWMVP, backEnd.refdef.sunShadowMvp[3]);
}
GLSL_SetUniformVec3(&tr.shadowmaskShader, UNIFORM_VIEWORIGIN, backEnd.refdef.vieworg);
{
vec3_t viewVector;
float zmax = backEnd.viewParms.zFar;
float ymax = zmax * tan(backEnd.viewParms.fovY * M_PI / 360.0f);
float xmax = zmax * tan(backEnd.viewParms.fovX * M_PI / 360.0f);
VectorScale(backEnd.refdef.viewaxis[0], zmax, viewVector);
GLSL_SetUniformVec3(&tr.shadowmaskShader, UNIFORM_VIEWFORWARD, viewVector);
VectorScale(backEnd.refdef.viewaxis[1], xmax, viewVector);
GLSL_SetUniformVec3(&tr.shadowmaskShader, UNIFORM_VIEWLEFT, viewVector);
VectorScale(backEnd.refdef.viewaxis[2], ymax, viewVector);
GLSL_SetUniformVec3(&tr.shadowmaskShader, UNIFORM_VIEWUP, viewVector);
GLSL_SetUniformVec4(&tr.shadowmaskShader, UNIFORM_VIEWINFO, viewInfo);
}
RB_InstantQuad2(quadVerts, texCoords); //, color, shaderProgram, invTexRes);
if (r_shadowBlur->integer)
{
viewInfo[2] = 1.0f / (float)(tr.screenScratchFbo->width);
viewInfo[3] = 1.0f / (float)(tr.screenScratchFbo->height);
FBO_Bind(tr.screenScratchFbo);
GLSL_BindProgram(&tr.depthBlurShader[0]);
GL_BindToTMU(tr.screenShadowImage, TB_COLORMAP);
GL_BindToTMU(tr.hdrDepthImage, TB_LIGHTMAP);
GLSL_SetUniformVec4(&tr.depthBlurShader[0], UNIFORM_VIEWINFO, viewInfo);
RB_InstantQuad2(quadVerts, texCoords);
FBO_Bind(tr.screenShadowFbo);
GLSL_BindProgram(&tr.depthBlurShader[1]);
GL_BindToTMU(tr.screenScratchImage, TB_COLORMAP);
GL_BindToTMU(tr.hdrDepthImage, TB_LIGHTMAP);
GLSL_SetUniformVec4(&tr.depthBlurShader[1], UNIFORM_VIEWINFO, viewInfo);
RB_InstantQuad2(quadVerts, texCoords);
}
}
if (r_ssao->integer)
{
vec4_t quadVerts[4];
vec2_t texCoords[4];
viewInfo[2] = 1.0f / ((float)(tr.quarterImage[0]->width) * tan(backEnd.viewParms.fovX * M_PI / 360.0f) * 2.0f);
viewInfo[3] = 1.0f / ((float)(tr.quarterImage[0]->height) * tan(backEnd.viewParms.fovY * M_PI / 360.0f) * 2.0f);
viewInfo[3] *= (float)backEnd.viewParms.viewportHeight / (float)backEnd.viewParms.viewportWidth;
FBO_Bind(tr.quarterFbo[0]);
qglViewport(0, 0, tr.quarterFbo[0]->width, tr.quarterFbo[0]->height);
qglScissor(0, 0, tr.quarterFbo[0]->width, tr.quarterFbo[0]->height);
VectorSet4(quadVerts[0], -1, 1, 0, 1);
VectorSet4(quadVerts[1], 1, 1, 0, 1);
VectorSet4(quadVerts[2], 1, -1, 0, 1);
VectorSet4(quadVerts[3], -1, -1, 0, 1);
texCoords[0][0] = 0; texCoords[0][1] = 1;
texCoords[1][0] = 1; texCoords[1][1] = 1;
texCoords[2][0] = 1; texCoords[2][1] = 0;
texCoords[3][0] = 0; texCoords[3][1] = 0;
GL_State( GLS_DEPTHTEST_DISABLE );
GLSL_BindProgram(&tr.ssaoShader);
GL_BindToTMU(tr.hdrDepthImage, TB_COLORMAP);
GLSL_SetUniformVec4(&tr.ssaoShader, UNIFORM_VIEWINFO, viewInfo);
RB_InstantQuad2(quadVerts, texCoords); //, color, shaderProgram, invTexRes);
viewInfo[2] = 1.0f / (float)(tr.quarterImage[0]->width);
viewInfo[3] = 1.0f / (float)(tr.quarterImage[0]->height);
FBO_Bind(tr.quarterFbo[1]);
qglViewport(0, 0, tr.quarterFbo[1]->width, tr.quarterFbo[1]->height);
qglScissor(0, 0, tr.quarterFbo[1]->width, tr.quarterFbo[1]->height);
GLSL_BindProgram(&tr.depthBlurShader[0]);
GL_BindToTMU(tr.quarterImage[0], TB_COLORMAP);
GL_BindToTMU(tr.hdrDepthImage, TB_LIGHTMAP);
GLSL_SetUniformVec4(&tr.depthBlurShader[0], UNIFORM_VIEWINFO, viewInfo);
RB_InstantQuad2(quadVerts, texCoords); //, color, shaderProgram, invTexRes);
FBO_Bind(tr.screenSsaoFbo);
qglViewport(0, 0, tr.screenSsaoFbo->width, tr.screenSsaoFbo->height);
qglScissor(0, 0, tr.screenSsaoFbo->width, tr.screenSsaoFbo->height);
GLSL_BindProgram(&tr.depthBlurShader[1]);
GL_BindToTMU(tr.quarterImage[1], TB_COLORMAP);
GL_BindToTMU(tr.hdrDepthImage, TB_LIGHTMAP);
GLSL_SetUniformVec4(&tr.depthBlurShader[1], UNIFORM_VIEWINFO, viewInfo);
RB_InstantQuad2(quadVerts, texCoords); //, color, shaderProgram, invTexRes);
}
}
// reset viewport and scissor
FBO_Bind(oldFbo);
SetViewportAndScissor();
}
if (glRefConfig.framebufferObject && (backEnd.viewParms.flags & VPF_DEPTHCLAMP) && glRefConfig.depthClamp)
{
qglDisable(GL_DEPTH_CLAMP);
}
if (!isShadowView)
{
RB_RenderDrawSurfList( cmd->drawSurfs, cmd->numDrawSurfs );
if (r_drawSun->integer)
{
RB_DrawSun(0.1, tr.sunShader);
}
if (glRefConfig.framebufferObject && r_drawSunRays->integer)
{
FBO_t *oldFbo = glState.currentFBO;
FBO_Bind(tr.sunRaysFbo);
qglClearColor( 0.0f, 0.0f, 0.0f, 1.0f );
qglClear( GL_COLOR_BUFFER_BIT );
if (glRefConfig.occlusionQuery)
{
tr.sunFlareQueryActive[tr.sunFlareQueryIndex] = qtrue;
qglBeginQuery(GL_SAMPLES_PASSED, tr.sunFlareQuery[tr.sunFlareQueryIndex]);
}
RB_DrawSun(0.3, tr.sunFlareShader);
if (glRefConfig.occlusionQuery)
{
qglEndQuery(GL_SAMPLES_PASSED);
}
FBO_Bind(oldFbo);
}
// darken down any stencil shadows
RB_ShadowFinish();
// add light flares on lights that aren't obscured
RB_RenderFlares();
}
if (glRefConfig.framebufferObject && tr.renderCubeFbo && backEnd.viewParms.targetFbo == tr.renderCubeFbo)
{
cubemap_t *cubemap = &tr.cubemaps[backEnd.viewParms.targetFboCubemapIndex];
FBO_Bind(NULL);
if (cubemap && cubemap->image)
qglGenerateTextureMipmapEXT(cubemap->image->texnum, GL_TEXTURE_CUBE_MAP);
}
return (const void *)(cmd + 1);
}
/*
=============
RB_DrawBuffer
=============
*/
const void *RB_DrawBuffer( const void *data ) {
const drawBufferCommand_t *cmd;
cmd = (const drawBufferCommand_t *)data;
// finish any 2D drawing if needed
if(tess.numIndexes)
RB_EndSurface();
if (glRefConfig.framebufferObject)
FBO_Bind(NULL);
qglDrawBuffer( cmd->buffer );
// clear screen for debugging
if ( r_clear->integer ) {
qglClearColor( 1, 0, 0.5, 1 );
qglClear( GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT );
}
return (const void *)(cmd + 1);
}
/*
===============
RB_ShowImages
Draw all the images to the screen, on top of whatever
was there. This is used to test for texture thrashing.
Also called by RE_EndRegistration
===============
*/
void RB_ShowImages( void ) {
int i;
image_t *image;
float x, y, w, h;
int start, end;
RB_SetGL2D();
qglClear( GL_COLOR_BUFFER_BIT );
qglFinish();
start = ri.Milliseconds();
for ( i=0 ; i<tr.numImages ; i++ ) {
image = tr.images[i];
w = glConfig.vidWidth / 20;
h = glConfig.vidHeight / 15;
x = i % 20 * w;
y = i / 20 * h;
// show in proportional size in mode 2
if ( r_showImages->integer == 2 ) {
w *= image->uploadWidth / 512.0f;
h *= image->uploadHeight / 512.0f;
}
{
vec4_t quadVerts[4];
GL_BindToTMU(image, TB_COLORMAP);
VectorSet4(quadVerts[0], x, y, 0, 1);
VectorSet4(quadVerts[1], x + w, y, 0, 1);
VectorSet4(quadVerts[2], x + w, y + h, 0, 1);
VectorSet4(quadVerts[3], x, y + h, 0, 1);
RB_InstantQuad(quadVerts);
}
}
qglFinish();
end = ri.Milliseconds();
ri.Printf( PRINT_ALL, "%i msec to draw all images\n", end - start );
}
/*
=============
RB_ColorMask
=============
*/
const void *RB_ColorMask(const void *data)
{
const colorMaskCommand_t *cmd = data;
// finish any 2D drawing if needed
if(tess.numIndexes)
RB_EndSurface();
if (glRefConfig.framebufferObject)
{
// reverse color mask, so 0 0 0 0 is the default
backEnd.colorMask[0] = !cmd->rgba[0];
backEnd.colorMask[1] = !cmd->rgba[1];
backEnd.colorMask[2] = !cmd->rgba[2];
backEnd.colorMask[3] = !cmd->rgba[3];
}
qglColorMask(cmd->rgba[0], cmd->rgba[1], cmd->rgba[2], cmd->rgba[3]);
return (const void *)(cmd + 1);
}
/*
=============
RB_ClearDepth
=============
*/
const void *RB_ClearDepth(const void *data)
{
const clearDepthCommand_t *cmd = data;
// finish any 2D drawing if needed
if(tess.numIndexes)
RB_EndSurface();
// texture swapping test
if (r_showImages->integer)
RB_ShowImages();
if (glRefConfig.framebufferObject)
{
if (!tr.renderFbo || backEnd.framePostProcessed)
{
FBO_Bind(NULL);
}
else
{
FBO_Bind(tr.renderFbo);
}
}
qglClear(GL_DEPTH_BUFFER_BIT);
// if we're doing MSAA, clear the depth texture for the resolve buffer
if (tr.msaaResolveFbo)
{
FBO_Bind(tr.msaaResolveFbo);
qglClear(GL_DEPTH_BUFFER_BIT);
}
return (const void *)(cmd + 1);
}
/*
=============
RB_SwapBuffers
=============
*/
const void *RB_SwapBuffers( const void *data ) {
const swapBuffersCommand_t *cmd;
// finish any 2D drawing if needed
if ( tess.numIndexes ) {
RB_EndSurface();
}
// texture swapping test
if ( r_showImages->integer ) {
RB_ShowImages();
}
cmd = (const swapBuffersCommand_t *)data;
// we measure overdraw by reading back the stencil buffer and
// counting up the number of increments that have happened
if ( r_measureOverdraw->integer ) {
int i;
long sum = 0;
unsigned char *stencilReadback;
stencilReadback = ri.Hunk_AllocateTempMemory( glConfig.vidWidth * glConfig.vidHeight );
qglReadPixels( 0, 0, glConfig.vidWidth, glConfig.vidHeight, GL_STENCIL_INDEX, GL_UNSIGNED_BYTE, stencilReadback );
for ( i = 0; i < glConfig.vidWidth * glConfig.vidHeight; i++ ) {
sum += stencilReadback[i];
}
backEnd.pc.c_overDraw += sum;
ri.Hunk_FreeTempMemory( stencilReadback );
}
if (glRefConfig.framebufferObject)
{
if (!backEnd.framePostProcessed)
{
if (tr.msaaResolveFbo && r_hdr->integer)
{
// Resolving an RGB16F MSAA FBO to the screen messes with the brightness, so resolve to an RGB16F FBO first
FBO_FastBlit(tr.renderFbo, NULL, tr.msaaResolveFbo, NULL, GL_COLOR_BUFFER_BIT, GL_NEAREST);
FBO_FastBlit(tr.msaaResolveFbo, NULL, NULL, NULL, GL_COLOR_BUFFER_BIT, GL_NEAREST);
}
else if (tr.renderFbo)
{
FBO_FastBlit(tr.renderFbo, NULL, NULL, NULL, GL_COLOR_BUFFER_BIT, GL_NEAREST);
}
}
}
if ( !glState.finishCalled ) {
qglFinish();
}
GLimp_LogComment( "***************** RB_SwapBuffers *****************\n\n\n" );
GLimp_EndFrame();
backEnd.framePostProcessed = qfalse;
backEnd.projection2D = qfalse;
return (const void *)(cmd + 1);
}
/*
=============
RB_CapShadowMap
=============
*/
const void *RB_CapShadowMap(const void *data)
{
const capShadowmapCommand_t *cmd = data;
// finish any 2D drawing if needed
if(tess.numIndexes)
RB_EndSurface();
if (cmd->map != -1)
{
if (cmd->cubeSide != -1)
{
if (tr.shadowCubemaps[cmd->map])
{
qglCopyTextureSubImage2DEXT(tr.shadowCubemaps[cmd->map]->texnum, GL_TEXTURE_CUBE_MAP_POSITIVE_X + cmd->cubeSide, 0, 0, 0, backEnd.refdef.x, glConfig.vidHeight - ( backEnd.refdef.y + PSHADOW_MAP_SIZE ), PSHADOW_MAP_SIZE, PSHADOW_MAP_SIZE);
}
}
else
{
if (tr.pshadowMaps[cmd->map])
{
qglCopyTextureSubImage2DEXT(tr.pshadowMaps[cmd->map]->texnum, GL_TEXTURE_2D, 0, 0, 0, backEnd.refdef.x, glConfig.vidHeight - (backEnd.refdef.y + PSHADOW_MAP_SIZE), PSHADOW_MAP_SIZE, PSHADOW_MAP_SIZE);
}
}
}
return (const void *)(cmd + 1);
}
/*
=============
RB_PostProcess
=============
*/
const void *RB_PostProcess(const void *data)
{
const postProcessCommand_t *cmd = data;
FBO_t *srcFbo;
ivec4_t srcBox, dstBox;
qboolean autoExposure;
// finish any 2D drawing if needed
if(tess.numIndexes)
RB_EndSurface();
if (!glRefConfig.framebufferObject || !r_postProcess->integer)
{
// do nothing
return (const void *)(cmd + 1);
}
if (cmd)
{
backEnd.refdef = cmd->refdef;
backEnd.viewParms = cmd->viewParms;
}
srcFbo = tr.renderFbo;
if (tr.msaaResolveFbo)
{
// Resolve the MSAA before anything else
// Can't resolve just part of the MSAA FBO, so multiple views will suffer a performance hit here
FBO_FastBlit(tr.renderFbo, NULL, tr.msaaResolveFbo, NULL, GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT, GL_NEAREST);
srcFbo = tr.msaaResolveFbo;
}
dstBox[0] = backEnd.viewParms.viewportX;
dstBox[1] = backEnd.viewParms.viewportY;
dstBox[2] = backEnd.viewParms.viewportWidth;
dstBox[3] = backEnd.viewParms.viewportHeight;
if (r_ssao->integer)
{
srcBox[0] = backEnd.viewParms.viewportX * tr.screenSsaoImage->width / (float)glConfig.vidWidth;
srcBox[1] = backEnd.viewParms.viewportY * tr.screenSsaoImage->height / (float)glConfig.vidHeight;
srcBox[2] = backEnd.viewParms.viewportWidth * tr.screenSsaoImage->width / (float)glConfig.vidWidth;
srcBox[3] = backEnd.viewParms.viewportHeight * tr.screenSsaoImage->height / (float)glConfig.vidHeight;
FBO_Blit(tr.screenSsaoFbo, srcBox, NULL, srcFbo, dstBox, NULL, NULL, GLS_SRCBLEND_DST_COLOR | GLS_DSTBLEND_ZERO);
}
srcBox[0] = backEnd.viewParms.viewportX;
srcBox[1] = backEnd.viewParms.viewportY;
srcBox[2] = backEnd.viewParms.viewportWidth;
srcBox[3] = backEnd.viewParms.viewportHeight;
if (srcFbo)
{
if (r_hdr->integer && (r_toneMap->integer || r_forceToneMap->integer))
{
autoExposure = r_autoExposure->integer || r_forceAutoExposure->integer;
RB_ToneMap(srcFbo, srcBox, NULL, dstBox, autoExposure);
}
else if (r_cameraExposure->value == 0.0f)
{
FBO_FastBlit(srcFbo, srcBox, NULL, dstBox, GL_COLOR_BUFFER_BIT, GL_NEAREST);
}
else
{
vec4_t color;
color[0] =
color[1] =
color[2] = pow(2, r_cameraExposure->value); //exp2(r_cameraExposure->value);
color[3] = 1.0f;
FBO_Blit(srcFbo, srcBox, NULL, NULL, dstBox, NULL, color, 0);
}
}
if (r_drawSunRays->integer)
RB_SunRays(NULL, srcBox, NULL, dstBox);
if (1)
RB_BokehBlur(NULL, srcBox, NULL, dstBox, backEnd.refdef.blurFactor);
else
RB_GaussianBlur(backEnd.refdef.blurFactor);
#if 0
if (0)
{
vec4_t quadVerts[4];
vec2_t texCoords[4];
ivec4_t iQtrBox;
vec4_t box;
vec4_t viewInfo;
static float scale = 5.0f;
scale -= 0.005f;
if (scale < 0.01f)
scale = 5.0f;
FBO_FastBlit(NULL, NULL, tr.quarterFbo[0], NULL, GL_COLOR_BUFFER_BIT, GL_LINEAR);
iQtrBox[0] = backEnd.viewParms.viewportX * tr.quarterImage[0]->width / (float)glConfig.vidWidth;
iQtrBox[1] = backEnd.viewParms.viewportY * tr.quarterImage[0]->height / (float)glConfig.vidHeight;
iQtrBox[2] = backEnd.viewParms.viewportWidth * tr.quarterImage[0]->width / (float)glConfig.vidWidth;
iQtrBox[3] = backEnd.viewParms.viewportHeight * tr.quarterImage[0]->height / (float)glConfig.vidHeight;
qglViewport(iQtrBox[0], iQtrBox[1], iQtrBox[2], iQtrBox[3]);
qglScissor(iQtrBox[0], iQtrBox[1], iQtrBox[2], iQtrBox[3]);
VectorSet4(box, 0.0f, 0.0f, 1.0f, 1.0f);
texCoords[0][0] = box[0]; texCoords[0][1] = box[3];
texCoords[1][0] = box[2]; texCoords[1][1] = box[3];
texCoords[2][0] = box[2]; texCoords[2][1] = box[1];
texCoords[3][0] = box[0]; texCoords[3][1] = box[1];
VectorSet4(box, -1.0f, -1.0f, 1.0f, 1.0f);
VectorSet4(quadVerts[0], box[0], box[3], 0, 1);
VectorSet4(quadVerts[1], box[2], box[3], 0, 1);
VectorSet4(quadVerts[2], box[2], box[1], 0, 1);
VectorSet4(quadVerts[3], box[0], box[1], 0, 1);
GL_State(GLS_DEPTHTEST_DISABLE);
VectorSet4(viewInfo, backEnd.viewParms.zFar / r_znear->value, backEnd.viewParms.zFar, 0.0, 0.0);
viewInfo[2] = scale / (float)(tr.quarterImage[0]->width);
viewInfo[3] = scale / (float)(tr.quarterImage[0]->height);
FBO_Bind(tr.quarterFbo[1]);
GLSL_BindProgram(&tr.depthBlurShader[2]);
GL_BindToTMU(tr.quarterImage[0], TB_COLORMAP);
GLSL_SetUniformVec4(&tr.depthBlurShader[2], UNIFORM_VIEWINFO, viewInfo);
RB_InstantQuad2(quadVerts, texCoords);
FBO_Bind(tr.quarterFbo[0]);
GLSL_BindProgram(&tr.depthBlurShader[3]);
GL_BindToTMU(tr.quarterImage[1], TB_COLORMAP);
GLSL_SetUniformVec4(&tr.depthBlurShader[3], UNIFORM_VIEWINFO, viewInfo);
RB_InstantQuad2(quadVerts, texCoords);
SetViewportAndScissor();
FBO_FastBlit(tr.quarterFbo[1], NULL, NULL, NULL, GL_COLOR_BUFFER_BIT, GL_LINEAR);
FBO_Bind(NULL);
}
#endif
if (0 && r_sunlightMode->integer)
{
ivec4_t dstBox;
VectorSet4(dstBox, 0, glConfig.vidHeight - 128, 128, 128);
FBO_BlitFromTexture(tr.sunShadowDepthImage[0], NULL, NULL, NULL, dstBox, NULL, NULL, 0);
VectorSet4(dstBox, 128, glConfig.vidHeight - 128, 128, 128);
FBO_BlitFromTexture(tr.sunShadowDepthImage[1], NULL, NULL, NULL, dstBox, NULL, NULL, 0);
VectorSet4(dstBox, 256, glConfig.vidHeight - 128, 128, 128);
FBO_BlitFromTexture(tr.sunShadowDepthImage[2], NULL, NULL, NULL, dstBox, NULL, NULL, 0);
VectorSet4(dstBox, 384, glConfig.vidHeight - 128, 128, 128);
FBO_BlitFromTexture(tr.sunShadowDepthImage[3], NULL, NULL, NULL, dstBox, NULL, NULL, 0);
}
if (0 && r_shadows->integer == 4)
{
ivec4_t dstBox;
VectorSet4(dstBox, 512 + 0, glConfig.vidHeight - 128, 128, 128);
FBO_BlitFromTexture(tr.pshadowMaps[0], NULL, NULL, NULL, dstBox, NULL, NULL, 0);
VectorSet4(dstBox, 512 + 128, glConfig.vidHeight - 128, 128, 128);
FBO_BlitFromTexture(tr.pshadowMaps[1], NULL, NULL, NULL, dstBox, NULL, NULL, 0);
VectorSet4(dstBox, 512 + 256, glConfig.vidHeight - 128, 128, 128);
FBO_BlitFromTexture(tr.pshadowMaps[2], NULL, NULL, NULL, dstBox, NULL, NULL, 0);
VectorSet4(dstBox, 512 + 384, glConfig.vidHeight - 128, 128, 128);
FBO_BlitFromTexture(tr.pshadowMaps[3], NULL, NULL, NULL, dstBox, NULL, NULL, 0);
}
if (0)
{
ivec4_t dstBox;
VectorSet4(dstBox, 256, glConfig.vidHeight - 256, 256, 256);
FBO_BlitFromTexture(tr.renderDepthImage, NULL, NULL, NULL, dstBox, NULL, NULL, 0);
VectorSet4(dstBox, 512, glConfig.vidHeight - 256, 256, 256);
FBO_BlitFromTexture(tr.screenShadowImage, NULL, NULL, NULL, dstBox, NULL, NULL, 0);
}
if (0)
{
ivec4_t dstBox;
VectorSet4(dstBox, 256, glConfig.vidHeight - 256, 256, 256);
FBO_BlitFromTexture(tr.sunRaysImage, NULL, NULL, NULL, dstBox, NULL, NULL, 0);
}
#if 0
if (r_cubeMapping->integer && tr.numCubemaps)
{
ivec4_t dstBox;
int cubemapIndex = R_CubemapForPoint( backEnd.viewParms.or.origin );
if (cubemapIndex)
{
VectorSet4(dstBox, 0, glConfig.vidHeight - 256, 256, 256);
//FBO_BlitFromTexture(tr.renderCubeImage, NULL, NULL, NULL, dstBox, &tr.testcubeShader, NULL, 0);
FBO_BlitFromTexture(tr.cubemaps[cubemapIndex - 1].image, NULL, NULL, NULL, dstBox, &tr.testcubeShader, NULL, 0);
}
}
#endif
backEnd.framePostProcessed = qtrue;
return (const void *)(cmd + 1);
}
// FIXME: put this function declaration elsewhere
void R_SaveDDS(const char *filename, byte *pic, int width, int height, int depth);
/*
=============
RB_ExportCubemaps
=============
*/
const void *RB_ExportCubemaps(const void *data)
{
const exportCubemapsCommand_t *cmd = data;
// finish any 2D drawing if needed
if (tess.numIndexes)
RB_EndSurface();
if (!glRefConfig.framebufferObject || !tr.world || tr.numCubemaps == 0)
{
// do nothing
ri.Printf(PRINT_ALL, "Nothing to export!\n");
return (const void *)(cmd + 1);
}
if (cmd)
{
FBO_t *oldFbo = glState.currentFBO;
int sideSize = r_cubemapSize->integer * r_cubemapSize->integer * 4;
byte *cubemapPixels = ri.Malloc(sideSize * 6);
int i, j;
FBO_Bind(tr.renderCubeFbo);
for (i = 0; i < tr.numCubemaps; i++)
{
char filename[MAX_QPATH];
cubemap_t *cubemap = &tr.cubemaps[i];
byte *p = cubemapPixels;
for (j = 0; j < 6; j++)
{
FBO_AttachImage(tr.renderCubeFbo, cubemap->image, GL_COLOR_ATTACHMENT0_EXT, j);
qglReadPixels(0, 0, r_cubemapSize->integer, r_cubemapSize->integer, GL_RGBA, GL_UNSIGNED_BYTE, p);
p += sideSize;
}
if (cubemap->name[0])
{
COM_StripExtension(cubemap->name, filename, MAX_QPATH);
Q_strcat(filename, MAX_QPATH, ".dds");
}
else
{
Com_sprintf(filename, MAX_QPATH, "cubemaps/%s/%03d.dds", tr.world->baseName, i);
}
R_SaveDDS(filename, cubemapPixels, r_cubemapSize->integer, r_cubemapSize->integer, 6);
ri.Printf(PRINT_ALL, "Saved cubemap %d as %s\n", i, filename);
}
FBO_Bind(oldFbo);
ri.Free(cubemapPixels);
}
return (const void *)(cmd + 1);
}
/*
====================
RB_ExecuteRenderCommands
====================
*/
void RB_ExecuteRenderCommands( const void *data ) {
int t1, t2;
t1 = ri.Milliseconds ();
while ( 1 ) {
data = PADP(data, sizeof(void *));
switch ( *(const int *)data ) {
case RC_SET_COLOR:
data = RB_SetColor( data );
break;
case RC_STRETCH_PIC:
data = RB_StretchPic( data );
break;
case RC_DRAW_SURFS:
data = RB_DrawSurfs( data );
break;
case RC_DRAW_BUFFER:
data = RB_DrawBuffer( data );
break;
case RC_SWAP_BUFFERS:
data = RB_SwapBuffers( data );
break;
case RC_SCREENSHOT:
data = RB_TakeScreenshotCmd( data );
break;
case RC_VIDEOFRAME:
data = RB_TakeVideoFrameCmd( data );
break;
case RC_COLORMASK:
data = RB_ColorMask(data);
break;
case RC_CLEARDEPTH:
data = RB_ClearDepth(data);
break;
case RC_CAPSHADOWMAP:
data = RB_CapShadowMap(data);
break;
case RC_POSTPROCESS:
data = RB_PostProcess(data);
break;
case RC_EXPORT_CUBEMAPS:
data = RB_ExportCubemaps(data);
break;
case RC_END_OF_LIST:
default:
// finish any 2D drawing if needed
if(tess.numIndexes)
RB_EndSurface();
// stop rendering
t2 = ri.Milliseconds ();
backEnd.pc.msec = t2 - t1;
return;
}
}
}