Divide large constant buffer into subsets and implement push constants for performance

neo/compileshaders.cmake

@@ -120,6 +120,7 @@ function(compile_shaders)
                                    --platform=SPIRV
                                    --binaryBlob
                                    --outputExt=.bin
+                                   --retryCount=20				# SRS - allow 20 retries to recover from any macOS/linux command shell failures during multi-threaded compilation
                                    -I ${SHADER_INCLUDE_DIR}
                                    -D SPIRV
                                    ${CFLAGS}

neo/renderer/ImmediateMode.cpp

@@ -138,7 +138,7 @@ void fhImmediateMode::End()
 	vertexBuffer.Update( drawVerts, drawVertsUsed * sizeof( idDrawVert ), 0, false, commandList );
 	indexBuffer.Update( lineIndices, drawVertsUsed * sizeof( triIndex_t ), 0, false, commandList );
 
-	renderProgManager.CommitConstantBuffer( commandList, true );
+	renderProgManager.CommitConstantBuffer( commandList, false );
 
 	int bindingLayoutType = renderProgManager.BindingLayoutType();
 
@@ -147,7 +147,7 @@ void fhImmediateMode::End()
 
 	for( int i = 0; i < layouts->Num(); i++ )
 	{
-		if( !tr.backend.currentBindingSets[i] || *tr.backend.currentBindingSets[i]->getDesc() != tr.backend.pendingBindingSetDescs[bindingLayoutType][i] )
+		if( !tr.backend.currentBindingSets[i] || *tr.backend.currentBindingSets[i]->getDesc() != tr.backend.pendingBindingSetDescs[bindingLayoutType][i] || bindingLayoutType != tr.backend.prevBindingLayoutType )
 		{
 			tr.backend.currentBindingSets[i] = tr.backend.bindingCache.GetOrCreateBindingSet( tr.backend.pendingBindingSetDescs[bindingLayoutType][i], ( *layouts )[i] );
 		}
@@ -182,6 +182,8 @@ void fhImmediateMode::End()
 		state.viewport.addScissorRect( nvrhi::Rect( viewport ) );
 
 		commandList->setGraphicsState( state );
+
+		renderProgManager.CommitPushConstants( commandList, bindingLayoutType );
 	}
 
 	nvrhi::DrawArguments args;

neo/renderer/Material.cpp

@@ -2174,12 +2174,15 @@ void idMaterial::ParseStage( idLexer& src, const textureRepeat_t trpDefault )
 			ss->conditionRegister = ParseExpression( src );
 			continue;
 		}
+
+		auto usePushConstants = renderProgManager.layoutTypeAttributes[BINDING_LAYOUT_POST_PROCESS_INGAME].pcEnabled ? "1" : "0";
+
 		if( !token.Icmp( "program" ) )
 		{
 			if( src.ReadTokenOnLine( &token ) )
 			{
 #if !defined( DMAP )
-				idList<shaderMacro_t> macros = { { "USE_GPU_SKINNING", "0" } };
+				idList<shaderMacro_t> macros = { { "USE_GPU_SKINNING", "0" }, { "USE_PUSH_CONSTANTS", usePushConstants } };
 				newStage.vertexProgram = renderProgManager.FindShader( token.c_str(), SHADER_STAGE_VERTEX, "", macros, false );
 				newStage.fragmentProgram = renderProgManager.FindShader( token.c_str(), SHADER_STAGE_FRAGMENT, "", macros, false );
 #endif
@@ -2191,7 +2194,7 @@ void idMaterial::ParseStage( idLexer& src, const textureRepeat_t trpDefault )
 			if( src.ReadTokenOnLine( &token ) )
 			{
 #if !defined( DMAP )
-				idList<shaderMacro_t> macros = { { "USE_GPU_SKINNING", "0" } };
+				idList<shaderMacro_t> macros = { { "USE_GPU_SKINNING", "0" }, { "USE_PUSH_CONSTANTS", usePushConstants } };
 				newStage.fragmentProgram = renderProgManager.FindShader( token.c_str(), SHADER_STAGE_FRAGMENT, "", macros, false );
 #endif
 			}
@@ -2202,7 +2205,7 @@ void idMaterial::ParseStage( idLexer& src, const textureRepeat_t trpDefault )
 			if( src.ReadTokenOnLine( &token ) )
 			{
 #if !defined( DMAP )
-				idList<shaderMacro_t> macros = { { "USE_GPU_SKINNING", "0" } };
+				idList<shaderMacro_t> macros = { { "USE_GPU_SKINNING", "0" }, { "USE_PUSH_CONSTANTS", usePushConstants } };
 				newStage.vertexProgram = renderProgManager.FindShader( token.c_str(), SHADER_STAGE_VERTEX, "", macros, false );
 #endif
 			}

neo/renderer/NVRHI/BufferObject_NVRHI.cpp

@@ -729,13 +729,9 @@ bool idUniformBuffer::AllocBufferObject( const void* data, int allocSize, buffer
 	// This buffer is a shader resource as opposed to a constant buffer due to
 	// constant buffers not being able to be sub-ranged.
 	nvrhi::BufferDesc bufferDesc;
-	//bufferDesc.initialState = nvrhi::ResourceStates::ConstantBuffer;		// SRS - shouldn't this be initialized to CopyDest?
-	bufferDesc.initialState = nvrhi::ResourceStates::CopyDest;
-	bufferDesc.canHaveTypedViews = true;
-	bufferDesc.canHaveRawViews = true;
 	bufferDesc.byteSize = numBytes;
-	bufferDesc.structStride = sizeof( idVec4 );
-	bufferDesc.isConstantBuffer = true;
+	bufferDesc.structStride = sizeof( idVec4 );						// SRS - this defines a structured storage buffer vs. a constant buffer
+	bufferDesc.initialState = nvrhi::ResourceStates::Common;
 
 	if( usage == BU_DYNAMIC )
 	{