doom3-bfg/neo/renderer/OpenGL/gl_backend.cpp
2012-11-26 12:58:24 -06:00

579 lines
20 KiB
C++

/*
===========================================================================
Doom 3 BFG Edition GPL Source Code
Copyright (C) 1993-2012 id Software LLC, a ZeniMax Media company.
This file is part of the Doom 3 BFG Edition GPL Source Code ("Doom 3 BFG Edition Source Code").
Doom 3 BFG Edition 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 3 of the License, or
(at your option) any later version.
Doom 3 BFG Edition 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 Doom 3 BFG Edition Source Code. If not, see <http://www.gnu.org/licenses/>.
In addition, the Doom 3 BFG Edition Source Code is also subject to certain additional terms. You should have received a copy of these additional terms immediately following the terms and conditions of the GNU General Public License which accompanied the Doom 3 BFG Edition Source Code. If not, please request a copy in writing from id Software at the address below.
If you have questions concerning this license or the applicable additional terms, you may contact in writing id Software LLC, c/o ZeniMax Media Inc., Suite 120, Rockville, Maryland 20850 USA.
===========================================================================
*/
#pragma hdrstop
#include "../../idlib/precompiled.h"
#include "../tr_local.h"
#include "../../framework/Common_local.h"
idCVar r_drawFlickerBox( "r_drawFlickerBox", "0", CVAR_RENDERER | CVAR_BOOL, "visual test for dropping frames" );
idCVar stereoRender_warp( "stereoRender_warp", "0", CVAR_RENDERER | CVAR_ARCHIVE | CVAR_BOOL, "use the optical warping renderprog instead of stereoDeGhost" );
idCVar stereoRender_warpStrength( "stereoRender_warpStrength", "1.45", CVAR_RENDERER | CVAR_ARCHIVE | CVAR_FLOAT, "amount of pre-distortion" );
idCVar stereoRender_warpCenterX( "stereoRender_warpCenterX", "0.5", CVAR_RENDERER | CVAR_FLOAT | CVAR_ARCHIVE, "center for left eye, right eye will be 1.0 - this" );
idCVar stereoRender_warpCenterY( "stereoRender_warpCenterY", "0.5", CVAR_RENDERER | CVAR_FLOAT | CVAR_ARCHIVE, "center for both eyes" );
idCVar stereoRender_warpParmZ( "stereoRender_warpParmZ", "0", CVAR_RENDERER | CVAR_FLOAT | CVAR_ARCHIVE, "development parm" );
idCVar stereoRender_warpParmW( "stereoRender_warpParmW", "0", CVAR_RENDERER | CVAR_FLOAT | CVAR_ARCHIVE, "development parm" );
idCVar stereoRender_warpTargetFraction( "stereoRender_warpTargetFraction", "1.0", CVAR_RENDERER | CVAR_FLOAT | CVAR_ARCHIVE, "fraction of half-width the through-lens view covers" );
idCVar r_showSwapBuffers( "r_showSwapBuffers", "0", CVAR_BOOL, "Show timings from GL_BlockingSwapBuffers" );
idCVar r_syncEveryFrame( "r_syncEveryFrame", "1", CVAR_BOOL, "Don't let the GPU buffer execution past swapbuffers" );
static int swapIndex; // 0 or 1 into renderSync
static GLsync renderSync[2];
void GLimp_SwapBuffers();
void RB_SetMVP( const idRenderMatrix & mvp );
/*
============================================================================
RENDER BACK END THREAD FUNCTIONS
============================================================================
*/
/*
=============
RB_DrawFlickerBox
=============
*/
static void RB_DrawFlickerBox() {
if ( !r_drawFlickerBox.GetBool() ) {
return;
}
if ( tr.frameCount & 1 ) {
qglClearColor( 1, 0, 0, 1 );
} else {
qglClearColor( 0, 1, 0, 1 );
}
qglScissor( 0, 0, 256, 256 );
qglClear( GL_COLOR_BUFFER_BIT );
}
/*
=============
RB_SetBuffer
=============
*/
static void RB_SetBuffer( const void *data ) {
// see which draw buffer we want to render the frame to
const setBufferCommand_t * cmd = (const setBufferCommand_t *)data;
RENDERLOG_PRINTF( "---------- RB_SetBuffer ---------- to buffer # %d\n", cmd->buffer );
GL_Scissor( 0, 0, tr.GetWidth(), tr.GetHeight() );
// clear screen for debugging
// automatically enable this with several other debug tools
// that might leave unrendered portions of the screen
if ( r_clear.GetFloat() || idStr::Length( r_clear.GetString() ) != 1 || r_singleArea.GetBool() || r_showOverDraw.GetBool() ) {
float c[3];
if ( sscanf( r_clear.GetString(), "%f %f %f", &c[0], &c[1], &c[2] ) == 3 ) {
GL_Clear( true, false, false, 0, c[0], c[1], c[2], 1.0f );
} else if ( r_clear.GetInteger() == 2 ) {
GL_Clear( true, false, false, 0, 0.0f, 0.0f, 0.0f, 1.0f );
} else if ( r_showOverDraw.GetBool() ) {
GL_Clear( true, false, false, 0, 1.0f, 1.0f, 1.0f, 1.0f );
} else {
GL_Clear( true, false, false, 0, 0.4f, 0.0f, 0.25f, 1.0f );
}
}
}
/*
=============
GL_BlockingSwapBuffers
We want to exit this with the GPU idle, right at vsync
=============
*/
const void GL_BlockingSwapBuffers() {
RENDERLOG_PRINTF( "***************** GL_BlockingSwapBuffers *****************\n\n\n" );
const int beforeFinish = Sys_Milliseconds();
if ( !glConfig.syncAvailable ) {
glFinish();
}
const int beforeSwap = Sys_Milliseconds();
if ( r_showSwapBuffers.GetBool() && beforeSwap - beforeFinish > 1 ) {
common->Printf( "%i msec to glFinish\n", beforeSwap - beforeFinish );
}
GLimp_SwapBuffers();
const int beforeFence = Sys_Milliseconds();
if ( r_showSwapBuffers.GetBool() && beforeFence - beforeSwap > 1 ) {
common->Printf( "%i msec to swapBuffers\n", beforeFence - beforeSwap );
}
if ( glConfig.syncAvailable ) {
swapIndex ^= 1;
if ( qglIsSync( renderSync[swapIndex] ) ) {
qglDeleteSync( renderSync[swapIndex] );
}
// draw something tiny to ensure the sync is after the swap
const int start = Sys_Milliseconds();
qglScissor( 0, 0, 1, 1 );
qglEnable( GL_SCISSOR_TEST );
qglClear( GL_COLOR_BUFFER_BIT );
renderSync[swapIndex] = qglFenceSync( GL_SYNC_GPU_COMMANDS_COMPLETE, 0 );
const int end = Sys_Milliseconds();
if ( r_showSwapBuffers.GetBool() && end - start > 1 ) {
common->Printf( "%i msec to start fence\n", end - start );
}
GLsync syncToWaitOn;
if ( r_syncEveryFrame.GetBool() ) {
syncToWaitOn = renderSync[swapIndex];
} else {
syncToWaitOn = renderSync[!swapIndex];
}
if ( qglIsSync( syncToWaitOn ) ) {
for ( GLenum r = GL_TIMEOUT_EXPIRED; r == GL_TIMEOUT_EXPIRED; ) {
r = qglClientWaitSync( syncToWaitOn, GL_SYNC_FLUSH_COMMANDS_BIT, 1000 * 1000 );
}
}
}
const int afterFence = Sys_Milliseconds();
if ( r_showSwapBuffers.GetBool() && afterFence - beforeFence > 1 ) {
common->Printf( "%i msec to wait on fence\n", afterFence - beforeFence );
}
const int64 exitBlockTime = Sys_Microseconds();
static int64 prevBlockTime;
if ( r_showSwapBuffers.GetBool() && prevBlockTime ) {
const int delta = (int) ( exitBlockTime - prevBlockTime );
common->Printf( "blockToBlock: %i\n", delta );
}
prevBlockTime = exitBlockTime;
}
/*
====================
R_MakeStereoRenderImage
====================
*/
static void R_MakeStereoRenderImage( idImage *image ) {
idImageOpts opts;
opts.width = renderSystem->GetWidth();
opts.height = renderSystem->GetHeight();
opts.numLevels = 1;
opts.format = FMT_RGBA8;
image->AllocImage( opts, TF_LINEAR, TR_CLAMP );
}
/*
====================
RB_StereoRenderExecuteBackEndCommands
Renders the draw list twice, with slight modifications for left eye / right eye
====================
*/
void RB_StereoRenderExecuteBackEndCommands( const emptyCommand_t * const allCmds ) {
uint64 backEndStartTime = Sys_Microseconds();
// If we are in a monoscopic context, this draws to the only buffer, and is
// the same as GL_BACK. In a quad-buffer stereo context, this is necessary
// to prevent GL from forcing the rendering to go to both BACK_LEFT and
// BACK_RIGHT at a performance penalty.
// To allow stereo deghost processing, the views have to be copied to separate
// textures anyway, so there isn't any benefit to rendering to BACK_RIGHT for
// that eye.
qglDrawBuffer( GL_BACK_LEFT );
// create the stereoRenderImage if we haven't already
static idImage * stereoRenderImages[2];
for ( int i = 0; i < 2; i++ ) {
if ( stereoRenderImages[i] == NULL ) {
stereoRenderImages[i] = globalImages->ImageFromFunction( va("_stereoRender%i",i), R_MakeStereoRenderImage );
}
// resize the stereo render image if the main window has changed size
if ( stereoRenderImages[i]->GetUploadWidth() != renderSystem->GetWidth() ||
stereoRenderImages[i]->GetUploadHeight() != renderSystem->GetHeight() ) {
stereoRenderImages[i]->Resize( renderSystem->GetWidth(), renderSystem->GetHeight() );
}
}
// In stereoRender mode, the front end has generated two RC_DRAW_VIEW commands
// with slightly different origins for each eye.
// TODO: only do the copy after the final view has been rendered, not mirror subviews?
// Render the 3D draw views from the screen origin so all the screen relative
// texture mapping works properly, then copy the portion we are going to use
// off to a texture.
bool foundEye[2] = { false, false };
for ( int stereoEye = 1; stereoEye >= -1; stereoEye -= 2 ) {
// set up the target texture we will draw to
const int targetEye = ( stereoEye == 1 ) ? 1 : 0;
// Set the back end into a known default state to fix any stale render state issues
GL_SetDefaultState();
renderProgManager.Unbind();
renderProgManager.ZeroUniforms();
for ( const emptyCommand_t * cmds = allCmds; cmds != NULL; cmds = (const emptyCommand_t *)cmds->next ) {
switch ( cmds->commandId ) {
case RC_NOP:
break;
case RC_DRAW_VIEW_GUI:
case RC_DRAW_VIEW_3D:
{
const drawSurfsCommand_t * const dsc = (const drawSurfsCommand_t *)cmds;
const viewDef_t & eyeViewDef = *dsc->viewDef;
if ( eyeViewDef.renderView.viewEyeBuffer && eyeViewDef.renderView.viewEyeBuffer != stereoEye ) {
// this is the render view for the other eye
continue;
}
foundEye[ targetEye ] = true;
RB_DrawView( dsc, stereoEye );
if ( cmds->commandId == RC_DRAW_VIEW_GUI ) {
}
}
break;
case RC_SET_BUFFER:
RB_SetBuffer( cmds );
break;
case RC_COPY_RENDER:
RB_CopyRender( cmds );
break;
case RC_POST_PROCESS:
{
postProcessCommand_t * cmd = (postProcessCommand_t *)cmds;
if ( cmd->viewDef->renderView.viewEyeBuffer != stereoEye ) {
break;
}
RB_PostProcess( cmds );
}
break;
default:
common->Error( "RB_ExecuteBackEndCommands: bad commandId" );
break;
}
}
// copy to the target
stereoRenderImages[ targetEye ]->CopyFramebuffer( 0, 0, renderSystem->GetWidth(), renderSystem->GetHeight() );
}
// perform the final compositing / warping / deghosting to the actual framebuffer(s)
assert( foundEye[0] && foundEye[1] );
GL_SetDefaultState();
RB_SetMVP( renderMatrix_identity );
// If we are in quad-buffer pixel format but testing another 3D mode,
// make sure we draw to both eyes. This is likely to be sub-optimal
// performance on most cards and drivers, but it is better than getting
// a confusing, half-ghosted view.
if ( renderSystem->GetStereo3DMode() != STEREO3D_QUAD_BUFFER ) {
glDrawBuffer( GL_BACK );
}
GL_State( GLS_DEPTHFUNC_ALWAYS );
GL_Cull( CT_TWO_SIDED );
// We just want to do a quad pass - so make sure we disable any texgen and
// set the texture matrix to the identity so we don't get anomalies from
// any stale uniform data being present from a previous draw call
const float texS[4] = { 1.0f, 0.0f, 0.0f, 0.0f };
const float texT[4] = { 0.0f, 1.0f, 0.0f, 0.0f };
renderProgManager.SetRenderParm( RENDERPARM_TEXTUREMATRIX_S, texS );
renderProgManager.SetRenderParm( RENDERPARM_TEXTUREMATRIX_T, texT );
// disable any texgen
const float texGenEnabled[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
renderProgManager.SetRenderParm( RENDERPARM_TEXGEN_0_ENABLED, texGenEnabled );
renderProgManager.BindShader_Texture();
GL_Color( 1, 1, 1, 1 );
switch( renderSystem->GetStereo3DMode() ) {
case STEREO3D_QUAD_BUFFER:
glDrawBuffer( GL_BACK_RIGHT );
GL_SelectTexture( 0 );
stereoRenderImages[1]->Bind();
GL_SelectTexture( 1 );
stereoRenderImages[0]->Bind();
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
glDrawBuffer( GL_BACK_LEFT );
GL_SelectTexture( 1 );
stereoRenderImages[1]->Bind();
GL_SelectTexture( 0 );
stereoRenderImages[0]->Bind();
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
break;
case STEREO3D_HDMI_720:
// HDMI 720P 3D
GL_SelectTexture( 0 );
stereoRenderImages[1]->Bind();
GL_SelectTexture( 1 );
stereoRenderImages[0]->Bind();
GL_ViewportAndScissor( 0, 0, 1280, 720 );
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
GL_SelectTexture( 0 );
stereoRenderImages[0]->Bind();
GL_SelectTexture( 1 );
stereoRenderImages[1]->Bind();
GL_ViewportAndScissor( 0, 750, 1280, 720 );
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
// force the HDMI 720P 3D guard band to a constant color
glScissor( 0, 720, 1280, 30 );
glClear( GL_COLOR_BUFFER_BIT );
break;
default:
case STEREO3D_SIDE_BY_SIDE:
if ( stereoRender_warp.GetBool() ) {
// this is the Rift warp
// renderSystem->GetWidth() / GetHeight() have returned equal values (640 for initial Rift)
// and we are going to warp them onto a symetric square region of each half of the screen
renderProgManager.BindShader_StereoWarp();
// clear the entire screen to black
// we could be smart and only clear the areas we aren't going to draw on, but
// clears are fast...
glScissor ( 0, 0, glConfig.nativeScreenWidth, glConfig.nativeScreenHeight );
glClearColor( 0, 0, 0, 0 );
glClear( GL_COLOR_BUFFER_BIT );
// the size of the box that will get the warped pixels
// With the 7" displays, this will be less than half the screen width
const int pixelDimensions = ( glConfig.nativeScreenWidth >> 1 ) * stereoRender_warpTargetFraction.GetFloat();
// Always scissor to the half-screen boundary, but the viewports
// might cross that boundary if the lenses can be adjusted closer
// together.
glViewport( ( glConfig.nativeScreenWidth >> 1 ) - pixelDimensions,
( glConfig.nativeScreenHeight >> 1 ) - ( pixelDimensions >> 1 ),
pixelDimensions, pixelDimensions );
glScissor ( 0, 0, glConfig.nativeScreenWidth >> 1, glConfig.nativeScreenHeight );
idVec4 color( stereoRender_warpCenterX.GetFloat(), stereoRender_warpCenterY.GetFloat(), stereoRender_warpParmZ.GetFloat(), stereoRender_warpParmW.GetFloat() );
// don't use GL_Color(), because we don't want to clamp
renderProgManager.SetRenderParm( RENDERPARM_COLOR, color.ToFloatPtr() );
GL_SelectTexture( 0 );
stereoRenderImages[0]->Bind();
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_BORDER );
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_BORDER );
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
idVec4 color2( stereoRender_warpCenterX.GetFloat(), stereoRender_warpCenterY.GetFloat(), stereoRender_warpParmZ.GetFloat(), stereoRender_warpParmW.GetFloat() );
// don't use GL_Color(), because we don't want to clamp
renderProgManager.SetRenderParm( RENDERPARM_COLOR, color2.ToFloatPtr() );
glViewport( ( glConfig.nativeScreenWidth >> 1 ),
( glConfig.nativeScreenHeight >> 1 ) - ( pixelDimensions >> 1 ),
pixelDimensions, pixelDimensions );
glScissor ( glConfig.nativeScreenWidth >> 1, 0, glConfig.nativeScreenWidth >> 1, glConfig.nativeScreenHeight );
GL_SelectTexture( 0 );
stereoRenderImages[1]->Bind();
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_BORDER );
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_BORDER );
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
break;
}
// a non-warped side-by-side-uncompressed (dual input cable) is rendered
// just like STEREO3D_SIDE_BY_SIDE_COMPRESSED, so fall through.
case STEREO3D_SIDE_BY_SIDE_COMPRESSED:
GL_SelectTexture( 0 );
stereoRenderImages[0]->Bind();
GL_SelectTexture( 1 );
stereoRenderImages[1]->Bind();
GL_ViewportAndScissor( 0, 0, renderSystem->GetWidth(), renderSystem->GetHeight() );
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
GL_SelectTexture( 0 );
stereoRenderImages[1]->Bind();
GL_SelectTexture( 1 );
stereoRenderImages[0]->Bind();
GL_ViewportAndScissor( renderSystem->GetWidth(), 0, renderSystem->GetWidth(), renderSystem->GetHeight() );
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
break;
case STEREO3D_TOP_AND_BOTTOM_COMPRESSED:
GL_SelectTexture( 1 );
stereoRenderImages[0]->Bind();
GL_SelectTexture( 0 );
stereoRenderImages[1]->Bind();
GL_ViewportAndScissor( 0, 0, renderSystem->GetWidth(), renderSystem->GetHeight() );
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
GL_SelectTexture( 1 );
stereoRenderImages[1]->Bind();
GL_SelectTexture( 0 );
stereoRenderImages[0]->Bind();
GL_ViewportAndScissor( 0, renderSystem->GetHeight(), renderSystem->GetWidth(), renderSystem->GetHeight() );
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
break;
case STEREO3D_INTERLACED:
// every other scanline
GL_SelectTexture( 0 );
stereoRenderImages[0]->Bind();
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST );
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST );
GL_SelectTexture( 1 );
stereoRenderImages[1]->Bind();
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST );
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST );
GL_ViewportAndScissor( 0, 0, renderSystem->GetWidth(), renderSystem->GetHeight()*2 );
renderProgManager.BindShader_StereoInterlace();
RB_DrawElementsWithCounters( &backEnd.unitSquareSurface );
GL_SelectTexture( 0 );
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
GL_SelectTexture( 1 );
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR );
qglTexParameterf( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR );
break;
}
// debug tool
RB_DrawFlickerBox();
// make sure the drawing is actually started
qglFlush();
// we may choose to sync to the swapbuffers before the next frame
// stop rendering on this thread
uint64 backEndFinishTime = Sys_Microseconds();
backEnd.pc.totalMicroSec = backEndFinishTime - backEndStartTime;
}
/*
====================
RB_ExecuteBackEndCommands
This function will be called syncronously if running without
smp extensions, or asyncronously by another thread.
====================
*/
void RB_ExecuteBackEndCommands( const emptyCommand_t *cmds ) {
// r_debugRenderToTexture
int c_draw3d = 0;
int c_draw2d = 0;
int c_setBuffers = 0;
int c_copyRenders = 0;
resolutionScale.SetCurrentGPUFrameTime( commonLocal.GetRendererGPUMicroseconds() );
renderLog.StartFrame();
if ( cmds->commandId == RC_NOP && !cmds->next ) {
return;
}
if ( renderSystem->GetStereo3DMode() != STEREO3D_OFF ) {
RB_StereoRenderExecuteBackEndCommands( cmds );
renderLog.EndFrame();
return;
}
uint64 backEndStartTime = Sys_Microseconds();
// needed for editor rendering
GL_SetDefaultState();
// If we have a stereo pixel format, this will draw to both
// the back left and back right buffers, which will have a
// performance penalty.
qglDrawBuffer( GL_BACK );
for ( ; cmds != NULL; cmds = (const emptyCommand_t *)cmds->next ) {
switch ( cmds->commandId ) {
case RC_NOP:
break;
case RC_DRAW_VIEW_3D:
case RC_DRAW_VIEW_GUI:
RB_DrawView( cmds, 0 );
if ( ((const drawSurfsCommand_t *)cmds)->viewDef->viewEntitys ) {
c_draw3d++;
} else {
c_draw2d++;
}
break;
case RC_SET_BUFFER:
c_setBuffers++;
break;
case RC_COPY_RENDER:
RB_CopyRender( cmds );
c_copyRenders++;
break;
case RC_POST_PROCESS:
RB_PostProcess( cmds );
break;
default:
common->Error( "RB_ExecuteBackEndCommands: bad commandId" );
break;
}
}
RB_DrawFlickerBox();
// Fix for the steam overlay not showing up while in game without Shell/Debug/Console/Menu also rendering
qglColorMask( 1, 1, 1, 1 );
qglFlush();
// stop rendering on this thread
uint64 backEndFinishTime = Sys_Microseconds();
backEnd.pc.totalMicroSec = backEndFinishTime - backEndStartTime;
if ( r_debugRenderToTexture.GetInteger() == 1 ) {
common->Printf( "3d: %i, 2d: %i, SetBuf: %i, CpyRenders: %i, CpyFrameBuf: %i\n", c_draw3d, c_draw2d, c_setBuffers, c_copyRenders, backEnd.pc.c_copyFrameBuffer );
backEnd.pc.c_copyFrameBuffer = 0;
}
renderLog.EndFrame();
}