Copy depth buffer if texture barrier unavailable

If the texture barrier GL extension is enabled, make a read-only copy of
the depth buffer to use for u_DepthMap sampling, thereby avoiding
texture feedback loops in another way. Configurable by setting
r_readonlyDepthBuffer 0 (skip the copy even if there is no texture
barrier - useful for low-performance systems) or r_readonlyDepthBuffer 2
(always make the copy - useful for testing that code path).

Fixes #1783 (broken depth fade on Apple Silicon).

Author: slipher

src/engine/renderer/Material.cpp

@@ -904,7 +904,7 @@ void BindShaderGeneric3D( Material* material ) {
 	gl_genericShaderMaterial->SetUniform_ModelMatrix( backEnd.orientation.transformMatrix );
 	gl_genericShaderMaterial->SetUniform_ModelViewProjectionMatrix( glState.modelViewProjectionMatrix[glState.stackIndex] );
 
-	gl_genericShaderMaterial->SetUniform_DepthMapBindless( GL_BindToTMU( 1, tr.currentDepthImage ) );
+	gl_genericShaderMaterial->SetUniform_DepthMapBindless( GL_BindToTMU( 1, tr.depthSamplerImage ) );
 
 	// u_DeformGen
 	gl_genericShaderMaterial->SetUniform_Time( backEnd.refdef.floatTime - backEnd.currentEntity->e.shaderTime );
@@ -1080,7 +1080,7 @@ void BindShaderLiquid( Material* material ) {
 	gl_liquidShaderMaterial->SetUniform_ModelViewProjectionMatrix( glState.modelViewProjectionMatrix[glState.stackIndex] );
 
 	// depth texture
-	gl_liquidShaderMaterial->SetUniform_DepthMapBindless( GL_BindToTMU( 2, tr.currentDepthImage ) );
+	gl_liquidShaderMaterial->SetUniform_DepthMapBindless( GL_BindToTMU( 2, tr.depthSamplerImage ) );
 
 	// bind u_PortalMap
 	gl_liquidShaderMaterial->SetUniform_PortalMapBindless( GL_BindToTMU( 1, tr.portalRenderImage ) );

src/engine/renderer/tr_backend.cpp

@@ -1213,6 +1213,11 @@ void RB_RunVisTests( )
 
 void RB_PrepareForSamplingDepthMap()
 {
+	if ( glConfig.usingReadonlyDepth )
+	{
+		return;
+	}
+
 	if ( !glConfig.textureBarrierAvailable )
 	{
 		return;
@@ -1413,7 +1418,7 @@ void RB_RenderGlobalFog()
 
 	// bind u_DepthMap
 	gl_fogGlobalShader->SetUniform_DepthMapBindless(
-		GL_BindToTMU( 1, tr.currentDepthImage )
+		GL_BindToTMU( 1, tr.depthSamplerImage )
 	);
 
 	Tess_InstantScreenSpaceQuad();
@@ -1535,7 +1540,7 @@ void RB_RenderMotionBlur()
 	gl_motionblurShader->SetUniform_blurVec(tr.refdef.blurVec);
 
 	gl_motionblurShader->SetUniform_DepthMapBindless(
-		GL_BindToTMU( 1, tr.currentDepthImage )
+		GL_BindToTMU( 1, tr.depthSamplerImage )
 	);
 
 	Tess_InstantScreenSpaceQuad();
@@ -1586,7 +1591,7 @@ void RB_RenderSSAO()
 	gl_ssaoShader->SetUniform_UnprojectionParams( unprojectionParams );
 
 	gl_ssaoShader->SetUniform_DepthMapBindless(
-		GL_BindToTMU( 0, tr.currentDepthImage )
+		GL_BindToTMU( 0, tr.depthSamplerImage )
 	);
 
 	Tess_InstantScreenSpaceQuad();
@@ -2699,6 +2704,17 @@ static void RB_RenderView( bool depthPass )
 		return;
 	}
 
+	if ( glConfig.usingReadonlyDepth )
+	{
+		FBO_t *currentDrawFBO = glState.currentFBO;
+		R_BindFBO( GL_READ_FRAMEBUFFER, currentDrawFBO );
+		R_BindFBO( GL_DRAW_FRAMEBUFFER, tr.readonlyDepthFBO );
+		int x0 = backEnd.viewParms.viewportX, x1 = x0 + backEnd.viewParms.viewportWidth;
+		int y0 = backEnd.viewParms.viewportY, y1 = y0 + backEnd.viewParms.viewportHeight;
+		GL_fboShim.glBlitFramebuffer( x0, y0, x1, y1, x0, y0, x1, y1, GL_DEPTH_BUFFER_BIT, GL_NEAREST );
+		R_BindFBO( currentDrawFBO );
+	}
+
 	if( tr.refdef.blurVec[0] != 0.0f ||
 			tr.refdef.blurVec[1] != 0.0f ||
 			tr.refdef.blurVec[2] != 0.0f )

src/engine/renderer/tr_fbo.cpp

@@ -263,6 +263,14 @@ void R_InitFBOs()
 	R_AttachFBOTexturePackedDepthStencil( tr.currentDepthImage->texnum );
 	R_CheckFBO( tr.mainFBO[1] );
 
+	if ( glConfig.usingReadonlyDepth )
+	{
+		tr.readonlyDepthFBO = R_CreateFBO( "_depthReadonly", width, height );
+		R_BindFBO( tr.readonlyDepthFBO );
+		GL_fboShim.glFramebufferTexture2D( GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, tr.depthSamplerImage->texnum, 0 );
+		glConfig.usingReadonlyDepth = R_CheckFBO( tr.readonlyDepthFBO );
+	}
+
 	if ( glConfig.realtimeLighting )
 	{
 		/* It's only required to create frame buffers only used by the

src/engine/renderer/tr_image.cpp

@@ -2473,6 +2473,12 @@ static void R_CreateCurrentRenderImage()
 
 	tr.currentDepthImage = R_CreateImage( "*currentDepth", nullptr, width, height, 1, imageParams );
 
+	if ( glConfig.usingReadonlyDepth )
+	{
+		imageParams.bits = IF_NOPICMIP | IF_DEPTH24;
+		tr.depthSamplerImage = R_CreateImage( "*readonlyDepth", nullptr, width, height, 1, imageParams );
+	}
+
 	if ( glConfig.usingMaterialSystem ) {
 		materialSystem.GenerateDepthImages( width, height, imageParams );
 	}

src/engine/renderer/tr_init.cpp

@@ -92,6 +92,9 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 	Cvar::Cvar<bool> r_overbrightIgnoreMapSettings("r_overbrightIgnoreMapSettings", "force usage of r_overbrightDefaultClamp / r_overbrightDefaultExponent, ignoring worldspawn", Cvar::NONE, false);
 	Cvar::Range<Cvar::Cvar<int>> r_lightMode("r_lightMode", "lighting mode: 0: fullbright (cheat), 1: vertex light, 2: grid light (cheat), 3: light map", Cvar::NONE, Util::ordinal(lightMode_t::MAP), Util::ordinal(lightMode_t::FULLBRIGHT), Util::ordinal(lightMode_t::MAP));
 	Cvar::Cvar<bool> r_colorGrading( "r_colorGrading", "Use color grading", Cvar::NONE, true );
+	static Cvar::Range<Cvar::Cvar<int>> r_readonlyDepthBuffer(
+		"r_readonlyDepthBuffer", "sample depth from a copy of the depth texture: 0 = no (unsafe), 1 = if necessary depending on GL features, 2 = yes",
+		Cvar::NONE, 1, 0, 2);
 	Cvar::Cvar<bool> r_preferBindlessTextures( "r_preferBindlessTextures", "use bindless textures even when material system is off", Cvar::NONE, false );
 	Cvar::Cvar<bool> r_materialSystem( "r_materialSystem", "Use Material System", Cvar::NONE, false );
 	Cvar::Cvar<bool> r_gpuFrustumCulling( "r_gpuFrustumCulling", "Use frustum culling on the GPU for the Material System", Cvar::NONE, true );
@@ -1193,6 +1196,7 @@ ScreenshotCmd screenshotPNGRegistration("screenshotPNG", ssFormat_t::SSF_PNG, "p
 
 		Cvar::Latch( r_realtimeLighting );
 		Cvar::Latch( r_realtimeLightLayers );
+		Cvar::Latch( r_readonlyDepthBuffer );
 		Cvar::Latch( r_preferBindlessTextures );
 		Cvar::Latch( r_materialSystem );
 
@@ -1407,12 +1411,37 @@ ScreenshotCmd screenshotPNGRegistration("screenshotPNG", ssFormat_t::SSF_PNG, "p
 			backEndData[ 1 ] = nullptr;
 		}
 
+		switch ( r_readonlyDepthBuffer.Get() )
+		{
+		case 0:
+			glConfig.usingReadonlyDepth = false;
+			break;
+		case 1:
+			glConfig.usingReadonlyDepth = !glConfig.textureBarrierAvailable;
+			break;
+		case 2:
+			glConfig.usingReadonlyDepth = true;
+			break;
+		}
+
+		if ( glConfig.usingReadonlyDepth && !r_depthShaders.Get() )
+		{
+			Log::Warn( "Disabling read-only depth buffer because depth pre-pass is disabled" );
+			glConfig.usingReadonlyDepth = false;
+		}
+
 		R_ToggleSmpFrame();
 
 		R_InitImages();
 
 		R_InitFBOs();
 
+		// This is here in case creating the depth-only FBO failed.
+		if ( !glConfig.usingReadonlyDepth )
+		{
+			tr.depthSamplerImage = tr.currentDepthImage;
+		}
+
 		R_InitVBOs();
 
 		R_InitShaders();

src/engine/renderer/tr_local.h

@@ -2454,6 +2454,7 @@ enum
 		image_t    *bloomRenderFBOImage[ 2 ];
 		image_t    *currentRenderImage[ 2 ];
 		image_t    *currentDepthImage;
+		image_t    *depthSamplerImage;
 		image_t    *depthtile1RenderImage;
 		image_t    *depthtile2RenderImage;
 		image_t    *lighttileRenderImage;
@@ -2465,6 +2466,7 @@ enum
 
 		// framebuffer objects
 		FBO_t *mainFBO[ 2 ];
+		FBO_t *readonlyDepthFBO;
 		FBO_t *depthtile1FBO;
 		FBO_t *depthtile2FBO;
 		FBO_t *lighttileFBO;

src/engine/renderer/tr_public.h

@@ -111,6 +111,7 @@ struct GLConfig
 	bool computeShaderAvailable;
 	bool bindlessTexturesAvailable; // do the driver/hardware support it
 	bool usingBindlessTextures; // are we using them right now
+	bool usingReadonlyDepth;
 	bool shaderDrawParametersAvailable;
 	bool SSBOAvailable;
 	bool multiDrawIndirectAvailable;

src/engine/renderer/tr_shade.cpp

@@ -951,7 +951,7 @@ void Render_generic3D( shaderStage_t *pStage )
 	if ( needDepthMap )
 	{
 		gl_genericShader->SetUniform_DepthMapBindless(
-			GL_BindToTMU( 1, tr.currentDepthImage )
+			GL_BindToTMU( 1, tr.depthSamplerImage )
 		);
 	}
 
@@ -1550,7 +1550,7 @@ void Render_liquid( shaderStage_t *pStage )
 	gl_liquidShader->SetUniform_PortalMapBindless( GL_BindToTMU( 1, tr.portalRenderImage ) );
 
 	// depth texture
-	gl_liquidShader->SetUniform_DepthMapBindless( GL_BindToTMU( 2, tr.currentDepthImage ) );
+	gl_liquidShader->SetUniform_DepthMapBindless( GL_BindToTMU( 2, tr.depthSamplerImage ) );
 
 	// bind u_HeightMap u_depthScale u_reliefOffsetBias
 	if ( pStage->enableReliefMapping )