/* =========================================================================== Doom 3 GPL Source Code Copyright (C) 1999-2011 id Software LLC, a ZeniMax Media company. This file is part of the Doom 3 GPL Source Code ("Doom 3 Source Code"). Doom 3 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 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 Source Code. If not, see . In addition, the Doom 3 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 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. =========================================================================== */ #include "sys/platform.h" #include "renderer/VertexCache.h" #include "renderer/Cinematic.h" #include "renderer/tr_local.h" /* back end scene + lights rendering functions */ /* ================= RB_DrawElementsImmediate Draws with immediate mode commands, which is going to be very slow. This should never happen if the vertex cache is operating properly. ================= */ void RB_DrawElementsImmediate( const srfTriangles_t *tri ) { backEnd.pc.c_drawElements++; backEnd.pc.c_drawIndexes += tri->numIndexes; backEnd.pc.c_drawVertexes += tri->numVerts; if ( tri->ambientSurface != NULL ) { if ( tri->indexes == tri->ambientSurface->indexes ) { backEnd.pc.c_drawRefIndexes += tri->numIndexes; } if ( tri->verts == tri->ambientSurface->verts ) { backEnd.pc.c_drawRefVertexes += tri->numVerts; } } qglBegin( GL_TRIANGLES ); for ( int i = 0 ; i < tri->numIndexes ; i++ ) { qglTexCoord2fv( tri->verts[ tri->indexes[i] ].st.ToFloatPtr() ); qglVertex3fv( tri->verts[ tri->indexes[i] ].xyz.ToFloatPtr() ); } qglEnd(); } /* ================ RB_DrawElementsWithCounters ================ */ void RB_DrawElementsWithCounters( const srfTriangles_t *tri ) { backEnd.pc.c_drawElements++; backEnd.pc.c_drawIndexes += tri->numIndexes; backEnd.pc.c_drawVertexes += tri->numVerts; if ( tri->ambientSurface != NULL ) { if ( tri->indexes == tri->ambientSurface->indexes ) { backEnd.pc.c_drawRefIndexes += tri->numIndexes; } if ( tri->verts == tri->ambientSurface->verts ) { backEnd.pc.c_drawRefVertexes += tri->numVerts; } } if ( tri->indexCache && r_useIndexBuffers.GetBool() ) { qglDrawElements( GL_TRIANGLES, r_singleTriangle.GetBool() ? 3 : tri->numIndexes, GL_INDEX_TYPE, (int *)vertexCache.Position( tri->indexCache ) ); backEnd.pc.c_vboIndexes += tri->numIndexes; } else { if ( r_useIndexBuffers.GetBool() ) { vertexCache.UnbindIndex(); } qglDrawElements( GL_TRIANGLES, r_singleTriangle.GetBool() ? 3 : tri->numIndexes, GL_INDEX_TYPE, tri->indexes ); } } /* ================ RB_DrawShadowElementsWithCounters May not use all the indexes in the surface if caps are skipped ================ */ void RB_DrawShadowElementsWithCounters( const srfTriangles_t *tri, int numIndexes ) { backEnd.pc.c_shadowElements++; backEnd.pc.c_shadowIndexes += numIndexes; backEnd.pc.c_shadowVertexes += tri->numVerts; if ( tri->indexCache && r_useIndexBuffers.GetBool() ) { qglDrawElements( GL_TRIANGLES, r_singleTriangle.GetBool() ? 3 : numIndexes, GL_INDEX_TYPE, (int *)vertexCache.Position( tri->indexCache ) ); backEnd.pc.c_vboIndexes += numIndexes; } else { if ( r_useIndexBuffers.GetBool() ) { vertexCache.UnbindIndex(); } qglDrawElements( GL_TRIANGLES, r_singleTriangle.GetBool() ? 3 : numIndexes, GL_INDEX_TYPE, tri->indexes ); } } /* =============== RB_RenderTriangleSurface Sets texcoord and vertex pointers =============== */ void RB_RenderTriangleSurface( const srfTriangles_t *tri ) { if ( !tri->ambientCache ) { RB_DrawElementsImmediate( tri ); return; } idDrawVert *ac = (idDrawVert *)vertexCache.Position( tri->ambientCache ); qglVertexPointer( 3, GL_FLOAT, sizeof( idDrawVert ), ac->xyz.ToFloatPtr() ); qglTexCoordPointer( 2, GL_FLOAT, sizeof( idDrawVert ), ac->st.ToFloatPtr() ); RB_DrawElementsWithCounters( tri ); } /* =============== RB_T_RenderTriangleSurface =============== */ void RB_T_RenderTriangleSurface( const drawSurf_t *surf ) { RB_RenderTriangleSurface( surf->geo ); } /* =============== RB_EnterWeaponDepthHack =============== */ void RB_EnterWeaponDepthHack() { qglDepthRange( 0, 0.5 ); float matrix[16]; memcpy( matrix, backEnd.viewDef->projectionMatrix, sizeof( matrix ) ); matrix[14] *= 0.25; qglMatrixMode(GL_PROJECTION); qglLoadMatrixf( matrix ); qglMatrixMode(GL_MODELVIEW); } /* =============== RB_EnterModelDepthHack =============== */ void RB_EnterModelDepthHack( float depth ) { qglDepthRange( 0.0f, 1.0f ); float matrix[16]; memcpy( matrix, backEnd.viewDef->projectionMatrix, sizeof( matrix ) ); matrix[14] -= depth; qglMatrixMode(GL_PROJECTION); qglLoadMatrixf( matrix ); qglMatrixMode(GL_MODELVIEW); } /* =============== RB_LeaveDepthHack =============== */ void RB_LeaveDepthHack() { qglDepthRange( 0, 1 ); qglMatrixMode(GL_PROJECTION); qglLoadMatrixf( backEnd.viewDef->projectionMatrix ); qglMatrixMode(GL_MODELVIEW); } /* ==================== RB_RenderDrawSurfListWithFunction The triangle functions can check backEnd.currentSpace != surf->space to see if they need to perform any new matrix setup. The modelview matrix will already have been loaded, and backEnd.currentSpace will be updated after the triangle function completes. ==================== */ void RB_RenderDrawSurfListWithFunction( drawSurf_t **drawSurfs, int numDrawSurfs, void (*triFunc_)( const drawSurf_t *) ) { int i; const drawSurf_t *drawSurf; backEnd.currentSpace = NULL; for (i = 0 ; i < numDrawSurfs ; i++ ) { drawSurf = drawSurfs[i]; // change the matrix if needed if ( drawSurf->space != backEnd.currentSpace ) { qglLoadMatrixf( drawSurf->space->modelViewMatrix ); } if ( drawSurf->space->weaponDepthHack ) { RB_EnterWeaponDepthHack(); } if ( drawSurf->space->modelDepthHack != 0.0f ) { RB_EnterModelDepthHack( drawSurf->space->modelDepthHack ); } // change the scissor if needed if ( r_useScissor.GetBool() && !backEnd.currentScissor.Equals( drawSurf->scissorRect ) ) { backEnd.currentScissor = drawSurf->scissorRect; qglScissor( backEnd.viewDef->viewport.x1 + backEnd.currentScissor.x1, backEnd.viewDef->viewport.y1 + backEnd.currentScissor.y1, backEnd.currentScissor.x2 + 1 - backEnd.currentScissor.x1, backEnd.currentScissor.y2 + 1 - backEnd.currentScissor.y1 ); } // render it triFunc_( drawSurf ); if ( drawSurf->space->weaponDepthHack || drawSurf->space->modelDepthHack != 0.0f ) { RB_LeaveDepthHack(); } backEnd.currentSpace = drawSurf->space; } } /* ====================== RB_RenderDrawSurfChainWithFunction ====================== */ void RB_RenderDrawSurfChainWithFunction( const drawSurf_t *drawSurfs, void (*triFunc_)( const drawSurf_t *) ) { const drawSurf_t *drawSurf; backEnd.currentSpace = NULL; for ( drawSurf = drawSurfs ; drawSurf ; drawSurf = drawSurf->nextOnLight ) { // change the matrix if needed if ( drawSurf->space != backEnd.currentSpace ) { qglLoadMatrixf( drawSurf->space->modelViewMatrix ); } if ( drawSurf->space->weaponDepthHack ) { RB_EnterWeaponDepthHack(); } if ( drawSurf->space->modelDepthHack ) { RB_EnterModelDepthHack( drawSurf->space->modelDepthHack ); } // change the scissor if needed if ( r_useScissor.GetBool() && !backEnd.currentScissor.Equals( drawSurf->scissorRect ) ) { backEnd.currentScissor = drawSurf->scissorRect; qglScissor( backEnd.viewDef->viewport.x1 + backEnd.currentScissor.x1, backEnd.viewDef->viewport.y1 + backEnd.currentScissor.y1, backEnd.currentScissor.x2 + 1 - backEnd.currentScissor.x1, backEnd.currentScissor.y2 + 1 - backEnd.currentScissor.y1 ); } // render it triFunc_( drawSurf ); if ( drawSurf->space->weaponDepthHack || drawSurf->space->modelDepthHack != 0.0f ) { RB_LeaveDepthHack(); } backEnd.currentSpace = drawSurf->space; } } /* ====================== RB_GetShaderTextureMatrix ====================== */ void RB_GetShaderTextureMatrix( const float *shaderRegisters, const textureStage_t *texture, float matrix[16] ) { matrix[0] = shaderRegisters[ texture->matrix[0][0] ]; matrix[4] = shaderRegisters[ texture->matrix[0][1] ]; matrix[8] = 0; matrix[12] = shaderRegisters[ texture->matrix[0][2] ]; // we attempt to keep scrolls from generating incredibly large texture values, but // center rotations and center scales can still generate offsets that need to be > 1 if ( matrix[12] < -40 || matrix[12] > 40 ) { matrix[12] -= (int)matrix[12]; } matrix[1] = shaderRegisters[ texture->matrix[1][0] ]; matrix[5] = shaderRegisters[ texture->matrix[1][1] ]; matrix[9] = 0; matrix[13] = shaderRegisters[ texture->matrix[1][2] ]; if ( matrix[13] < -40 || matrix[13] > 40 ) { matrix[13] -= (int)matrix[13]; } matrix[2] = 0; matrix[6] = 0; matrix[10] = 1; matrix[14] = 0; matrix[3] = 0; matrix[7] = 0; matrix[11] = 0; matrix[15] = 1; } /* ====================== RB_LoadShaderTextureMatrix ====================== */ void RB_LoadShaderTextureMatrix( const float *shaderRegisters, const textureStage_t *texture ) { float matrix[16]; RB_GetShaderTextureMatrix( shaderRegisters, texture, matrix ); qglMatrixMode( GL_TEXTURE ); qglLoadMatrixf( matrix ); qglMatrixMode( GL_MODELVIEW ); } /* ====================== RB_BindVariableStageImage Handles generating a cinematic frame if needed ====================== */ void RB_BindVariableStageImage( const textureStage_t *texture, const float *shaderRegisters ) { if ( texture->cinematic ) { cinData_t cin; if ( r_skipDynamicTextures.GetBool() ) { globalImages->defaultImage->Bind(); return; } // offset time by shaderParm[7] (FIXME: make the time offset a parameter of the shader?) // We make no attempt to optimize for multiple identical cinematics being in view, or // for cinematics going at a lower framerate than the renderer. cin = texture->cinematic->ImageForTime( (int)(1000 * ( backEnd.viewDef->floatTime + backEnd.viewDef->renderView.shaderParms[11] ) ) ); if ( cin.image ) { globalImages->cinematicImage->UploadScratch( cin.image, cin.imageWidth, cin.imageHeight ); } else { globalImages->blackImage->Bind(); } } else { //FIXME: see why image is invalid if (texture->image) { texture->image->Bind(); } } } /* ====================== RB_BindStageTexture ====================== */ void RB_BindStageTexture( const float *shaderRegisters, const textureStage_t *texture, const drawSurf_t *surf ) { // image RB_BindVariableStageImage( texture, shaderRegisters ); // texgens if ( texture->texgen == TG_DIFFUSE_CUBE ) { qglTexCoordPointer( 3, GL_FLOAT, sizeof( idDrawVert ), ((idDrawVert *)vertexCache.Position( surf->geo->ambientCache ))->normal.ToFloatPtr() ); } if ( texture->texgen == TG_SKYBOX_CUBE || texture->texgen == TG_WOBBLESKY_CUBE ) { qglTexCoordPointer( 3, GL_FLOAT, 0, vertexCache.Position( surf->dynamicTexCoords ) ); } if ( texture->texgen == TG_REFLECT_CUBE ) { qglEnable( GL_TEXTURE_GEN_S ); qglEnable( GL_TEXTURE_GEN_T ); qglEnable( GL_TEXTURE_GEN_R ); qglTexGenf( GL_S, GL_TEXTURE_GEN_MODE, GL_REFLECTION_MAP_EXT ); qglTexGenf( GL_T, GL_TEXTURE_GEN_MODE, GL_REFLECTION_MAP_EXT ); qglTexGenf( GL_R, GL_TEXTURE_GEN_MODE, GL_REFLECTION_MAP_EXT ); qglEnableClientState( GL_NORMAL_ARRAY ); qglNormalPointer( GL_FLOAT, sizeof( idDrawVert ), ((idDrawVert *)vertexCache.Position( surf->geo->ambientCache ))->normal.ToFloatPtr() ); qglMatrixMode( GL_TEXTURE ); float mat[16]; R_TransposeGLMatrix( backEnd.viewDef->worldSpace.modelViewMatrix, mat ); qglLoadMatrixf( mat ); qglMatrixMode( GL_MODELVIEW ); } // matrix if ( texture->hasMatrix ) { RB_LoadShaderTextureMatrix( shaderRegisters, texture ); } } /* ====================== RB_FinishStageTexture ====================== */ void RB_FinishStageTexture( const textureStage_t *texture, const drawSurf_t *surf ) { if ( texture->texgen == TG_DIFFUSE_CUBE || texture->texgen == TG_SKYBOX_CUBE || texture->texgen == TG_WOBBLESKY_CUBE ) { qglTexCoordPointer( 2, GL_FLOAT, sizeof( idDrawVert ), (void *)&(((idDrawVert *)vertexCache.Position( surf->geo->ambientCache ))->st) ); } if ( texture->texgen == TG_REFLECT_CUBE ) { qglDisable( GL_TEXTURE_GEN_S ); qglDisable( GL_TEXTURE_GEN_T ); qglDisable( GL_TEXTURE_GEN_R ); qglTexGenf( GL_S, GL_TEXTURE_GEN_MODE, GL_OBJECT_LINEAR ); qglTexGenf( GL_T, GL_TEXTURE_GEN_MODE, GL_OBJECT_LINEAR ); qglTexGenf( GL_R, GL_TEXTURE_GEN_MODE, GL_OBJECT_LINEAR ); qglDisableClientState( GL_NORMAL_ARRAY ); qglMatrixMode( GL_TEXTURE ); qglLoadIdentity(); qglMatrixMode( GL_MODELVIEW ); } if ( texture->hasMatrix ) { qglMatrixMode( GL_TEXTURE ); qglLoadIdentity(); qglMatrixMode( GL_MODELVIEW ); } } //============================================================================================= /* ================= RB_DetermineLightScale Sets: backEnd.lightScale backEnd.overBright Find out how much we are going to need to overscale the lighting, so we can down modulate the pre-lighting passes. We only look at light calculations, but an argument could be made that we should also look at surface evaluations, which would let surfaces overbright past 1.0 ================= */ void RB_DetermineLightScale( void ) { viewLight_t *vLight; const idMaterial *shader; float max; int i, j, numStages; const shaderStage_t *stage; // the light scale will be based on the largest color component of any surface // that will be drawn. // should we consider separating rgb scales? // if there are no lights, this will remain at 1.0, so GUI-only // rendering will not lose any bits of precision max = 1.0; for ( vLight = backEnd.viewDef->viewLights ; vLight ; vLight = vLight->next ) { // lights with no surfaces or shaderparms may still be present // for debug display if ( !vLight->localInteractions && !vLight->globalInteractions && !vLight->translucentInteractions ) { continue; } shader = vLight->lightShader; numStages = shader->GetNumStages(); for ( i = 0 ; i < numStages ; i++ ) { stage = shader->GetStage( i ); for ( j = 0 ; j < 3 ; j++ ) { float v = r_lightScale.GetFloat() * vLight->shaderRegisters[ stage->color.registers[j] ]; if ( v > max ) { max = v; } } } } backEnd.pc.maxLightValue = max; if ( max <= tr.backEndRendererMaxLight ) { backEnd.lightScale = r_lightScale.GetFloat(); backEnd.overBright = 1.0; } else { backEnd.lightScale = r_lightScale.GetFloat() * tr.backEndRendererMaxLight / max; backEnd.overBright = max / tr.backEndRendererMaxLight; } } /* ================= 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) { const viewDef_t* viewDef = backEnd.viewDef; // set the modelview matrix for the viewer qglMatrixMode(GL_PROJECTION); qglLoadMatrixf( viewDef->projectionMatrix ); qglMatrixMode(GL_MODELVIEW); // set the window clipping qglViewport( tr.viewportOffset[0] + viewDef->viewport.x1, tr.viewportOffset[1] + viewDef->viewport.y1, viewDef->viewport.x2 + 1 - viewDef->viewport.x1, viewDef->viewport.y2 + 1 - viewDef->viewport.y1 ); // the scissor may be smaller than the viewport for subviews qglScissor( tr.viewportOffset[0] + viewDef->viewport.x1 + viewDef->scissor.x1, tr.viewportOffset[1] + viewDef->viewport.y1 + viewDef->scissor.y1, viewDef->scissor.x2 + 1 - viewDef->scissor.x1, viewDef->scissor.y2 + 1 - viewDef->scissor.y1 ); backEnd.currentScissor = viewDef->scissor; // ensures that depth writes are enabled for the depth clear GL_State( GLS_DEFAULT ); // we don't have to clear the depth / stencil buffer for 2D rendering if ( backEnd.viewDef->viewEntitys ) { qglStencilMask( 0xff ); // some cards may have 7 bit stencil buffers, so don't assume this // should be 128 qglClearStencil( 1<<(glConfig.stencilBits-1) ); qglClear( GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT ); qglEnable( GL_DEPTH_TEST ); } else { qglDisable( GL_DEPTH_TEST ); qglDisable( GL_STENCIL_TEST ); } backEnd.glState.faceCulling = -1; // force face culling to set next time GL_Cull( CT_FRONT_SIDED ); } /* ================== R_SetDrawInteractions ================== */ void R_SetDrawInteraction( const shaderStage_t *surfaceStage, const float *surfaceRegs, idImage **image, idVec4 matrix[2], float color[4] ) { *image = surfaceStage->texture.image; if ( surfaceStage->texture.hasMatrix ) { matrix[0][0] = surfaceRegs[surfaceStage->texture.matrix[0][0]]; matrix[0][1] = surfaceRegs[surfaceStage->texture.matrix[0][1]]; matrix[0][2] = 0; matrix[0][3] = surfaceRegs[surfaceStage->texture.matrix[0][2]]; matrix[1][0] = surfaceRegs[surfaceStage->texture.matrix[1][0]]; matrix[1][1] = surfaceRegs[surfaceStage->texture.matrix[1][1]]; matrix[1][2] = 0; matrix[1][3] = surfaceRegs[surfaceStage->texture.matrix[1][2]]; // we attempt to keep scrolls from generating incredibly large texture values, but // center rotations and center scales can still generate offsets that need to be > 1 if ( matrix[0][3] < -40 || matrix[0][3] > 40 ) { matrix[0][3] -= (int)matrix[0][3]; } if ( matrix[1][3] < -40 || matrix[1][3] > 40 ) { matrix[1][3] -= (int)matrix[1][3]; } } else { matrix[0][0] = 1; matrix[0][1] = 0; matrix[0][2] = 0; matrix[0][3] = 0; matrix[1][0] = 0; matrix[1][1] = 1; matrix[1][2] = 0; matrix[1][3] = 0; } if ( color ) { for ( int i = 0 ; i < 4 ; i++ ) { color[i] = surfaceRegs[surfaceStage->color.registers[i]]; // clamp here, so card with greater range don't look different. // we could perform overbrighting like we do for lights, but // it doesn't currently look worth it. if ( color[i] < 0 ) { color[i] = 0; } else if ( color[i] > 1.0 ) { color[i] = 1.0; } } } } /* ================= RB_SubmittInteraction ================= */ static void RB_SubmittInteraction( drawInteraction_t *din, void (*DrawInteraction)(const drawInteraction_t *) ) { if ( !din->bumpImage ) { return; } if ( !din->diffuseImage || r_skipDiffuse.GetBool() ) { din->diffuseImage = globalImages->blackImage; } if ( !din->specularImage || r_skipSpecular.GetBool() || din->ambientLight ) { din->specularImage = globalImages->blackImage; } if ( !din->bumpImage || r_skipBump.GetBool() ) { din->bumpImage = globalImages->flatNormalMap; } // if we wouldn't draw anything, don't call the Draw function if ( ( ( din->diffuseColor[0] > 0 || din->diffuseColor[1] > 0 || din->diffuseColor[2] > 0 ) && din->diffuseImage != globalImages->blackImage ) || ( ( din->specularColor[0] > 0 || din->specularColor[1] > 0 || din->specularColor[2] > 0 ) && din->specularImage != globalImages->blackImage ) ) { DrawInteraction( din ); } } /* ============= RB_CreateSingleDrawInteractions This can be used by different draw_* backends to decompose a complex light / surface interaction into primitive interactions ============= */ void RB_CreateSingleDrawInteractions( const drawSurf_t *surf, void (*DrawInteraction)(const drawInteraction_t *) ) { const idMaterial *surfaceShader = surf->material; const float *surfaceRegs = surf->shaderRegisters; const viewLight_t *vLight = backEnd.vLight; const idMaterial *lightShader = vLight->lightShader; const float *lightRegs = vLight->shaderRegisters; drawInteraction_t inter; inter.diffuseMatrix[0].Zero(); inter.diffuseMatrix[1].Zero(); inter.specularMatrix[0].Zero(); inter.specularMatrix[1].Zero(); if ( r_skipInteractions.GetBool() || !surf->geo || !surf->geo->ambientCache ) { return; } // change the matrix and light projection vectors if needed if ( surf->space != backEnd.currentSpace ) { backEnd.currentSpace = surf->space; qglLoadMatrixf( surf->space->modelViewMatrix ); } // change the scissor if needed if ( r_useScissor.GetBool() && !backEnd.currentScissor.Equals( surf->scissorRect ) ) { backEnd.currentScissor = surf->scissorRect; qglScissor( backEnd.viewDef->viewport.x1 + backEnd.currentScissor.x1, backEnd.viewDef->viewport.y1 + backEnd.currentScissor.y1, backEnd.currentScissor.x2 + 1 - backEnd.currentScissor.x1, backEnd.currentScissor.y2 + 1 - backEnd.currentScissor.y1 ); } // hack depth range if needed if ( surf->space->weaponDepthHack ) { RB_EnterWeaponDepthHack(); } if ( surf->space->modelDepthHack ) { RB_EnterModelDepthHack( surf->space->modelDepthHack ); } inter.surf = surf; inter.lightFalloffImage = vLight->falloffImage; R_GlobalPointToLocal( surf->space->modelMatrix, vLight->globalLightOrigin, inter.localLightOrigin.ToVec3() ); R_GlobalPointToLocal( surf->space->modelMatrix, backEnd.viewDef->renderView.vieworg, inter.localViewOrigin.ToVec3() ); inter.localLightOrigin[3] = 0; inter.localViewOrigin[3] = 1; inter.ambientLight = lightShader->IsAmbientLight(); // the base projections may be modified by texture matrix on light stages idPlane lightProject[4]; for ( int i = 0 ; i < 4 ; i++ ) { R_GlobalPlaneToLocal( surf->space->modelMatrix, backEnd.vLight->lightProject[i], lightProject[i] ); } for ( int lightStageNum = 0 ; lightStageNum < lightShader->GetNumStages() ; lightStageNum++ ) { const shaderStage_t *lightStage = lightShader->GetStage( lightStageNum ); // ignore stages that fail the condition if ( !lightRegs[ lightStage->conditionRegister ] ) { continue; } inter.lightImage = lightStage->texture.image; memcpy( inter.lightProjection, lightProject, sizeof( inter.lightProjection ) ); // now multiply the texgen by the light texture matrix if ( lightStage->texture.hasMatrix ) { RB_GetShaderTextureMatrix( lightRegs, &lightStage->texture, backEnd.lightTextureMatrix ); RB_BakeTextureMatrixIntoTexgen( reinterpret_cast(inter.lightProjection), backEnd.lightTextureMatrix ); } inter.bumpImage = NULL; inter.specularImage = NULL; inter.diffuseImage = NULL; inter.diffuseColor[0] = inter.diffuseColor[1] = inter.diffuseColor[2] = inter.diffuseColor[3] = 0; inter.specularColor[0] = inter.specularColor[1] = inter.specularColor[2] = inter.specularColor[3] = 0; float lightColor[4]; // backEnd.lightScale is calculated so that lightColor[] will never exceed // tr.backEndRendererMaxLight lightColor[0] = backEnd.lightScale * lightRegs[ lightStage->color.registers[0] ]; lightColor[1] = backEnd.lightScale * lightRegs[ lightStage->color.registers[1] ]; lightColor[2] = backEnd.lightScale * lightRegs[ lightStage->color.registers[2] ]; lightColor[3] = lightRegs[ lightStage->color.registers[3] ]; // go through the individual stages for ( int surfaceStageNum = 0 ; surfaceStageNum < surfaceShader->GetNumStages() ; surfaceStageNum++ ) { const shaderStage_t *surfaceStage = surfaceShader->GetStage( surfaceStageNum ); switch( surfaceStage->lighting ) { case SL_AMBIENT: { // ignore ambient stages while drawing interactions break; } case SL_BUMP: { // ignore stage that fails the condition if ( !surfaceRegs[ surfaceStage->conditionRegister ] ) { break; } // draw any previous interaction RB_SubmittInteraction( &inter, DrawInteraction ); inter.diffuseImage = NULL; inter.specularImage = NULL; R_SetDrawInteraction( surfaceStage, surfaceRegs, &inter.bumpImage, inter.bumpMatrix, NULL ); break; } case SL_DIFFUSE: { // ignore stage that fails the condition if ( !surfaceRegs[ surfaceStage->conditionRegister ] ) { break; } if ( inter.diffuseImage ) { RB_SubmittInteraction( &inter, DrawInteraction ); } R_SetDrawInteraction( surfaceStage, surfaceRegs, &inter.diffuseImage, inter.diffuseMatrix, inter.diffuseColor.ToFloatPtr() ); inter.diffuseColor[0] *= lightColor[0]; inter.diffuseColor[1] *= lightColor[1]; inter.diffuseColor[2] *= lightColor[2]; inter.diffuseColor[3] *= lightColor[3]; inter.vertexColor = surfaceStage->vertexColor; break; } case SL_SPECULAR: { // ignore stage that fails the condition if ( !surfaceRegs[ surfaceStage->conditionRegister ] ) { break; } if ( inter.specularImage ) { RB_SubmittInteraction( &inter, DrawInteraction ); } R_SetDrawInteraction( surfaceStage, surfaceRegs, &inter.specularImage, inter.specularMatrix, inter.specularColor.ToFloatPtr() ); inter.specularColor[0] *= lightColor[0]; inter.specularColor[1] *= lightColor[1]; inter.specularColor[2] *= lightColor[2]; inter.specularColor[3] *= lightColor[3]; inter.vertexColor = surfaceStage->vertexColor; break; } } } // draw the final interaction RB_SubmittInteraction( &inter, DrawInteraction ); } // unhack depth range if needed if ( surf->space->weaponDepthHack || surf->space->modelDepthHack != 0.0f ) { RB_LeaveDepthHack(); } } /* ============= RB_DrawView ============= */ void RB_DrawView( const void *data ) { const drawSurfsCommand_t *cmd; cmd = (const drawSurfsCommand_t *)data; // with r_lockSurfaces enabled, we set the locked render view // for the primary viewDef for all the "what should be drawn" calculations. // now it must be reverted to the real render view so the scene gets rendered // from the actual current players point of view if(r_lockSurfaces.GetBool() && tr.primaryView == cmd->viewDef) { viewDef_t* parms = cmd->viewDef; const viewDef_t origParms = *parms; *parms = tr.lockSurfacesRealViewDef; // actual current player/camera position parms->renderWorld = origParms.renderWorld; parms->floatTime = origParms.floatTime; parms->drawSurfs = origParms.drawSurfs; parms->numDrawSurfs = origParms.numDrawSurfs; parms->maxDrawSurfs = origParms.maxDrawSurfs; parms->viewLights = origParms.viewLights; parms->viewEntitys = origParms.viewEntitys; parms->connectedAreas = origParms.connectedAreas; for( viewEntity_t* vModel = parms->viewEntitys ; vModel ; vModel = vModel->next ) { myGlMultMatrix( vModel->modelMatrix, parms->worldSpace.modelViewMatrix, vModel->modelViewMatrix ); } } backEnd.viewDef = cmd->viewDef; // we will need to do a new copyTexSubImage of the screen // when a SS_POST_PROCESS material is used backEnd.currentRenderCopied = false; // if there aren't any drawsurfs, do nothing if ( !backEnd.viewDef->numDrawSurfs ) { return; } // skip render bypasses everything that has models, assuming // them to be 3D views, but leaves 2D rendering visible if ( r_skipRender.GetBool() && backEnd.viewDef->viewEntitys ) { return; } // skip render context sets the gl context to NULL, // which should factor out the API cost, under the assumption // that all gl calls just return if the context isn't valid if ( r_skipRenderContext.GetBool() && backEnd.viewDef->viewEntitys ) { GLimp_DeactivateContext(); } backEnd.pc.c_surfaces += backEnd.viewDef->numDrawSurfs; RB_ShowOverdraw(); // render the scene, jumping to the hardware specific interaction renderers RB_STD_DrawView(); // restore the context for 2D drawing if we were stubbing it out if ( r_skipRenderContext.GetBool() && backEnd.viewDef->viewEntitys ) { GLimp_ActivateContext(); RB_SetDefaultGLState(); } }