Little improvments on the performance side to get the Stencil pass and the masking of the background under reasonable performances

libraries/gpu/src/gpu/DrawUnitQuadTexcoord.slv

@@ -0,0 +1,28 @@
+<@include gpu/Config.slh@>
+<$VERSION_HEADER$>
+//  Generated on <$_SCRIBE_DATE$>
+//
+//  Draw the unit quad [-1,-1 -> 1,1] amd pass along the unit texcoords [0, 0 -> 1, 1]. Not transform used.
+//  Simply draw a Triangle_strip of 2 triangles, no input buffers or index buffer needed
+//
+//  Created by Sam Gateau on 6/22/2015
+//  Copyright 2015 High Fidelity, Inc.
+//
+//  Distributed under the Apache License, Version 2.0.
+//  See the accompanying file LICENSE or http://www.apache.org/licenses/LICENSE-2.0.html
+//
+out vec2 varTexCoord0;
+
+void main(void) {
+    const vec4 UNIT_QUAD[4] = vec4[4](
+        vec4(-1.0, -1.0, 0.0, 1.0),
+        vec4(1.0, -1.0, 0.0, 1.0),
+        vec4(-1.0, 1.0, 0.0, 1.0),
+        vec4(1.0, 1.0, 0.0, 1.0)
+    );
+    vec4 pos = UNIT_QUAD[gl_VertexID];
+
+    varTexCoord0 = (pos.xy + 1) * 0.5;
+
+    gl_Position = pos;
+}

libraries/gpu/src/gpu/GLBackend.cpp

@@ -132,20 +132,26 @@ void GLBackend::renderPassTransfer(Batch& batch) {
     const size_t numCommands = batch.getCommands().size();
     const Batch::Commands::value_type* command = batch.getCommands().data();
     const Batch::CommandOffsets::value_type* offset = batch.getCommandOffsets().data();
-    
+
-    for (auto& cached : batch._buffers._items) {
+    { // Sync all the buffers
-        if (cached._data) {
+        PROFILE_RANGE("syncGPUBuffer");
-            syncGPUObject(*cached._data);
+
+        for (auto& cached : batch._buffers._items) {
+            if (cached._data) {
+                syncGPUObject(*cached._data);
+            }
         }
     }
-    // Reset the transform buffers
-    _transform._cameras.resize(0);
-    _transform._cameraOffsets.clear();
-    _transform._objects.resize(0);
-    _transform._objectOffsets.clear();
 
-    for (_commandIndex = 0; _commandIndex < numCommands; ++_commandIndex) {
+    { // Sync all the buffers
-        switch (*command) {
+        PROFILE_RANGE("syncCPUTransform");
+        _transform._cameras.resize(0);
+        _transform._cameraOffsets.clear();
+        _transform._objects.resize(0);
+        _transform._objectOffsets.clear();
+
+        for (_commandIndex = 0; _commandIndex < numCommands; ++_commandIndex) {
+            switch (*command) {
             case Batch::COMMAND_draw:
             case Batch::COMMAND_drawIndexed:
             case Batch::COMMAND_drawInstanced:
@@ -164,11 +170,16 @@ void GLBackend::renderPassTransfer(Batch& batch) {
 
             default:
                 break;
+            }
+            command++;
+            offset++;
         }
-        command++;
-        offset++;
     }
-    _transform.transfer();
+
+    { // Sync the transform buffers
+        PROFILE_RANGE("syncGPUTransform");
+        _transform.transfer();
+    }
 }
 
 void GLBackend::renderPassDraw(Batch& batch) {

libraries/gpu/src/gpu/GLBackendState.cpp

@@ -642,8 +642,13 @@ void GLBackend::do_setStateStencil(State::StencilActivation activation, State::S
 
         if (activation.isEnabled()) {
             glEnable(GL_STENCIL_TEST);
-            glStencilMaskSeparate(GL_FRONT, activation.getWriteMaskFront());
+
-            glStencilMaskSeparate(GL_BACK, activation.getWriteMaskBack());
+            if (activation.getWriteMaskFront() != activation.getWriteMaskBack()) {
+                glStencilMaskSeparate(GL_FRONT, activation.getWriteMaskFront());
+                glStencilMaskSeparate(GL_BACK, activation.getWriteMaskBack());
+            } else {
+                glStencilMask(activation.getWriteMaskFront());
+            }
 
             static GLenum STENCIL_OPS[] = {
                 GL_KEEP,
@@ -655,12 +660,16 @@ void GLBackend::do_setStateStencil(State::StencilActivation activation, State::S
                 GL_INCR,
                 GL_DECR };
 
-            glStencilOpSeparate(GL_FRONT, STENCIL_OPS[frontTest.getFailOp()], STENCIL_OPS[frontTest.getPassOp()], STENCIL_OPS[frontTest.getDepthFailOp()]);
+            if (frontTest != backTest) {
-            glStencilFuncSeparate(GL_FRONT, GL_COMPARISON_FUNCTIONS[frontTest.getFunction()], frontTest.getReference(), frontTest.getReadMask());
+                glStencilOpSeparate(GL_FRONT, STENCIL_OPS[frontTest.getFailOp()], STENCIL_OPS[frontTest.getPassOp()], STENCIL_OPS[frontTest.getDepthFailOp()]);
-
+                glStencilFuncSeparate(GL_FRONT, GL_COMPARISON_FUNCTIONS[frontTest.getFunction()], frontTest.getReference(), frontTest.getReadMask());
-            glStencilOpSeparate(GL_BACK, STENCIL_OPS[backTest.getFailOp()], STENCIL_OPS[backTest.getPassOp()], STENCIL_OPS[backTest.getDepthFailOp()]);
-            glStencilFuncSeparate(GL_BACK, GL_COMPARISON_FUNCTIONS[backTest.getFunction()], backTest.getReference(), backTest.getReadMask());
 
+                glStencilOpSeparate(GL_BACK, STENCIL_OPS[backTest.getFailOp()], STENCIL_OPS[backTest.getPassOp()], STENCIL_OPS[backTest.getDepthFailOp()]);
+                glStencilFuncSeparate(GL_BACK, GL_COMPARISON_FUNCTIONS[backTest.getFunction()], backTest.getReference(), backTest.getReadMask());
+            } else {
+                glStencilOp(STENCIL_OPS[frontTest.getFailOp()], STENCIL_OPS[frontTest.getPassOp()], STENCIL_OPS[frontTest.getDepthFailOp()]);
+                glStencilFunc(GL_COMPARISON_FUNCTIONS[frontTest.getFunction()], frontTest.getReference(), frontTest.getReadMask());
+            }
         } else {
             glDisable(GL_STENCIL_TEST);
         }

libraries/gpu/src/gpu/GLBackendTransform.cpp

@@ -130,19 +130,27 @@ void GLBackend::TransformStageState::preUpdate(size_t commandIndex, const Stereo
 
 void GLBackend::TransformStageState::transfer() const {
     static QByteArray bufferData;
-    glBindBuffer(GL_UNIFORM_BUFFER, _cameraBuffer);
+    if (!_cameras.empty()) {
-    bufferData.resize(_cameraUboSize * _cameras.size());
+        glBindBuffer(GL_UNIFORM_BUFFER, _cameraBuffer);
-    for (size_t i = 0; i < _cameras.size(); ++i) {
+        bufferData.resize(_cameraUboSize * _cameras.size());
-        memcpy(bufferData.data() + (_cameraUboSize * i), &_cameras[i], sizeof(TransformCamera));
+        for (size_t i = 0; i < _cameras.size(); ++i) {
+            memcpy(bufferData.data() + (_cameraUboSize * i), &_cameras[i], sizeof(TransformCamera));
+        }
+        glBufferData(GL_UNIFORM_BUFFER, bufferData.size(), bufferData.data(), GL_DYNAMIC_DRAW);
     }
-    glBufferData(GL_UNIFORM_BUFFER, bufferData.size(), bufferData.data(), GL_DYNAMIC_DRAW);
+
-    glBindBuffer(GL_UNIFORM_BUFFER, _objectBuffer);
+    if (!_objects.empty()) {
-    bufferData.resize(_objectUboSize * _objects.size());
+        glBindBuffer(GL_UNIFORM_BUFFER, _objectBuffer);
-    for (size_t i = 0; i < _objects.size(); ++i) {
+        bufferData.resize(_objectUboSize * _objects.size());
-        memcpy(bufferData.data() + (_objectUboSize * i), &_objects[i], sizeof(TransformObject));
+        for (size_t i = 0; i < _objects.size(); ++i) {
+            memcpy(bufferData.data() + (_objectUboSize * i), &_objects[i], sizeof(TransformObject));
+        }
+        glBufferData(GL_UNIFORM_BUFFER, bufferData.size(), bufferData.data(), GL_DYNAMIC_DRAW);
+    }
+
+    if (!_cameras.empty() || !_objects.empty()) {
+        glBindBuffer(GL_UNIFORM_BUFFER, 0);
     }
-    glBufferData(GL_UNIFORM_BUFFER, bufferData.size(), bufferData.data(), GL_DYNAMIC_DRAW);
-    glBindBuffer(GL_UNIFORM_BUFFER, 0);
     CHECK_GL_ERROR();
 }
 

libraries/gpu/src/gpu/StandardShaderLib.cpp

@@ -12,6 +12,7 @@
 //
 #include "StandardShaderLib.h"
 
+#include "DrawUnitQuadTexcoord_vert.h"
 #include "DrawTransformUnitQuad_vert.h"
 #include "DrawTexcoordRectTransformUnitQuad_vert.h"
 #include "DrawViewportQuadTransformTexcoord_vert.h"
@@ -21,6 +22,7 @@
 
 using namespace gpu;
 
+ShaderPointer StandardShaderLib::_drawUnitQuadTexcoordVS;
 ShaderPointer StandardShaderLib::_drawTransformUnitQuadVS;
 ShaderPointer StandardShaderLib::_drawTexcoordRectTransformUnitQuadVS;
 ShaderPointer StandardShaderLib::_drawViewportQuadTransformTexcoordVS;
@@ -55,6 +57,12 @@ ShaderPointer StandardShaderLib::getProgram(GetShader getVS, GetShader getPS) {
 }
 
 
+ShaderPointer StandardShaderLib::getDrawUnitQuadTexcoordVS() {
+    if (!_drawUnitQuadTexcoordVS) {
+        _drawUnitQuadTexcoordVS = gpu::ShaderPointer(gpu::Shader::createVertex(std::string(DrawUnitQuadTexcoord_vert)));
+    }
+    return _drawUnitQuadTexcoordVS;
+}
 
 ShaderPointer StandardShaderLib::getDrawTransformUnitQuadVS() {
     if (!_drawTransformUnitQuadVS) {

libraries/gpu/src/gpu/StandardShaderLib.h

@@ -23,6 +23,9 @@ namespace gpu {
 class StandardShaderLib {
 public:
 
+    // Shader draws the unit quad in the full viewport clipPos = ([(-1,-1),(1,1)]) and the unit texcoord = [(0,0),(1,1)].
+    static ShaderPointer getDrawUnitQuadTexcoordVS();
+
     // Shader draw the unit quad objectPos = ([(-1,-1),(1,1)]) and transform it by the full model transform stack (Model, View, Proj). 
     // A texcoord attribute is also generated texcoord = [(0,0),(1,1)]
     static ShaderPointer getDrawTransformUnitQuadVS();
@@ -44,6 +47,7 @@ public:
 
 protected:
 
+    static ShaderPointer _drawUnitQuadTexcoordVS;
     static ShaderPointer _drawTransformUnitQuadVS;
     static ShaderPointer _drawTexcoordRectTransformUnitQuadVS;
     static ShaderPointer _drawViewportQuadTransformTexcoordVS;

libraries/render-utils/src/RenderDeferredTask.cpp

@@ -37,15 +37,19 @@ void SetupDeferred::run(const SceneContextPointer& sceneContext, const RenderCon
     RenderArgs* args = renderContext->args;
     gpu::doInBatch(args->_context, [=](gpu::Batch& batch) {
 
+        auto primaryFboStencil = DependencyManager::get<FramebufferCache>()->getPrimaryFramebufferStencilColor();
         auto primaryFbo = DependencyManager::get<FramebufferCache>()->getPrimaryFramebufferDepthColor();
 
         batch.enableStereo(false);
-        batch.setFramebuffer(nullptr);
-        batch.setFramebuffer(primaryFbo);
- 
         batch.setViewportTransform(args->_viewport);
         batch.setStateScissorRect(args->_viewport);
 
+        batch.setFramebuffer(primaryFboStencil);
+        batch.clearFramebuffer(
+            gpu::Framebuffer::BUFFER_STENCIL,
+            vec4(vec3(0), 1), 1.0, 0.0, true);
+
+        batch.setFramebuffer(primaryFbo);
         batch.clearFramebuffer(
             gpu::Framebuffer::BUFFER_COLOR0 |
             gpu::Framebuffer::BUFFER_DEPTH,
@@ -68,7 +72,6 @@ void ResolveDeferred::run(const SceneContextPointer& sceneContext, const RenderC
 
 RenderDeferredTask::RenderDeferredTask() : Task() {
     _jobs.push_back(Job(new SetupDeferred::JobModel("SetupFramebuffer")));
- //   _jobs.push_back(Job(new DrawBackground::JobModel("DrawBackground")));
 
     _jobs.push_back(Job(new PrepareDeferred::JobModel("PrepareDeferred")));
     _jobs.push_back(Job(new FetchItems::JobModel("FetchOpaque",
@@ -302,7 +305,7 @@ gpu::PipelinePointer DrawStencilDeferred::_opaquePipeline;
 const gpu::PipelinePointer& DrawStencilDeferred::getOpaquePipeline() {
     if (!_opaquePipeline) {
         const gpu::int8 STENCIL_OPAQUE = 1;
-        auto vs = gpu::StandardShaderLib::getDrawViewportQuadTransformTexcoordVS();
+        auto vs = gpu::StandardShaderLib::getDrawUnitQuadTexcoordVS();
         auto ps = gpu::ShaderPointer(gpu::Shader::createPixel(std::string(drawOpaqueStencil_frag)));
         auto program = gpu::ShaderPointer(gpu::Shader::createProgram(vs, ps));
         
@@ -329,7 +332,6 @@ void DrawStencilDeferred::run(const SceneContextPointer& sceneContext, const Ren
         args->_batch = &batch;
 
         auto primaryFboColorDepthStencil = DependencyManager::get<FramebufferCache>()->getPrimaryFramebufferStencilColor();
-        auto primaryFboFull = DependencyManager::get<FramebufferCache>()->getPrimaryFramebuffer();
         auto primaryDepth = DependencyManager::get<FramebufferCache>()->getPrimaryDepthTexture();
    
         batch.enableStereo(false);
@@ -337,11 +339,6 @@ void DrawStencilDeferred::run(const SceneContextPointer& sceneContext, const Ren
         batch.setFramebuffer(primaryFboColorDepthStencil);
         batch.setViewportTransform(args->_viewport);
         batch.setStateScissorRect(args->_viewport);
-        batch.clearStencilFramebuffer(0, true);
-
-
-        Transform modelMat;
-        batch.setModelTransform(modelMat);
 
         batch.setPipeline(getOpaquePipeline());
         batch.setResourceTexture(0, primaryDepth);
@@ -349,8 +346,6 @@ void DrawStencilDeferred::run(const SceneContextPointer& sceneContext, const Ren
         batch.draw(gpu::TRIANGLE_STRIP, 4);
         batch.setResourceTexture(0, nullptr);
 
-        batch.setFramebuffer(primaryFboFull);
-
     });
     args->_batch = nullptr;
 }