Merge pull request #14915 from samcake/brown

Case 21190: Introducing msaa to forward renderer

libraries/gpu/src/gpu/State.h

@@ -246,7 +246,7 @@ public:
 
     struct Flags {
         Flags() :
-            frontFaceClockwise(false), depthClampEnable(false), scissorEnable(false), multisampleEnable(false),
+            frontFaceClockwise(false), depthClampEnable(false), scissorEnable(false), multisampleEnable(true),
             antialisedLineEnable(true), alphaToCoverageEnable(false), _spare1(0) {}
         bool frontFaceClockwise : 1;
         bool depthClampEnable : 1;

libraries/render-utils/src/RenderCommonTask.cpp

@@ -101,7 +101,7 @@ void DrawLayered3D::run(const RenderContextPointer& renderContext, const Inputs&
     }
 }
 
-void CompositeHUD::run(const RenderContextPointer& renderContext) {
+void CompositeHUD::run(const RenderContextPointer& renderContext, const gpu::FramebufferPointer& inputs) {
     assert(renderContext->args);
     assert(renderContext->args->_context);
 
@@ -119,6 +119,9 @@ void CompositeHUD::run(const RenderContextPointer& renderContext) {
         renderContext->args->getViewFrustum().evalViewTransform(viewMat);
         batch.setProjectionTransform(projMat);
         batch.setViewTransform(viewMat, true);
+        if (inputs) {
+            batch.setFramebuffer(inputs);
+        }
         if (renderContext->args->_hudOperator) {
             renderContext->args->_hudOperator(batch, renderContext->args->_hudTexture, renderContext->args->_renderMode == RenderArgs::RenderMode::MIRROR_RENDER_MODE);
         }
@@ -197,7 +200,73 @@ void Blit::run(const RenderContextPointer& renderContext, const gpu::Framebuffer
     });
 }
 
-void ExtractFrustums::run(const render::RenderContextPointer& renderContext, const Inputs& inputs, Outputs& output) {
+
+void ResolveFramebuffer::run(const render::RenderContextPointer& renderContext, const Inputs& inputs, Outputs& outputs) {
+    RenderArgs* args = renderContext->args;
+    auto srcFbo = inputs.get0();
+    auto destFbo = inputs.get1();
+
+    if (!destFbo) {
+        destFbo = args->_blitFramebuffer;
+    }
+    outputs = destFbo;
+
+    // Check valid src and dest
+    if (!srcFbo || !destFbo) {
+        return;
+    }
+    
+    // Check valid size for sr and dest
+    auto frameSize(srcFbo->getSize());
+    if (destFbo->getSize() != frameSize) {
+        return;
+    }
+
+    gpu::Vec4i rectSrc;
+    rectSrc.z = frameSize.x;
+    rectSrc.w = frameSize.y;
+    gpu::doInBatch("Resolve", args->_context, [&](gpu::Batch& batch) { 
+        batch.blit(srcFbo, rectSrc, destFbo, rectSrc);
+    });
+}
+
+void ResolveNewFramebuffer::run(const render::RenderContextPointer& renderContext, const Inputs& inputs, Outputs& outputs) {
+    RenderArgs* args = renderContext->args;
+    auto srcFbo = inputs;
+    outputs.reset();
+
+    // Check valid src
+    if (!srcFbo) {
+        return;
+    }
+
+    // Check valid size for sr and dest
+    auto frameSize(srcFbo->getSize());
+
+    // Resizing framebuffers instead of re-building them seems to cause issues with threaded rendering
+    if (_outputFramebuffer && _outputFramebuffer->getSize() != frameSize) {
+        _outputFramebuffer.reset();
+    }
+
+    if (!_outputFramebuffer) {
+        _outputFramebuffer = gpu::FramebufferPointer(gpu::Framebuffer::create("resolvedNew.out"));
+        auto colorFormat = gpu::Element::COLOR_SRGBA_32;
+        auto defaultSampler = gpu::Sampler(gpu::Sampler::FILTER_MIN_MAG_LINEAR);
+        auto colorTexture = gpu::Texture::createRenderBuffer(colorFormat, frameSize.x, frameSize.y, gpu::Texture::SINGLE_MIP, defaultSampler);
+        _outputFramebuffer->setRenderBuffer(0, colorTexture);
+    }
+
+    gpu::Vec4i rectSrc;
+    rectSrc.z = frameSize.x;
+    rectSrc.w = frameSize.y;
+    gpu::doInBatch("ResolveNew", args->_context, [&](gpu::Batch& batch) { batch.blit(srcFbo, rectSrc, _outputFramebuffer, rectSrc); });
+
+    outputs = _outputFramebuffer;
+}
+
+
+
+ void ExtractFrustums::run(const render::RenderContextPointer& renderContext, const Inputs& inputs, Outputs& output) {
     assert(renderContext->args);
     assert(renderContext->args->_context);
 

libraries/render-utils/src/RenderCommonTask.h

@@ -77,10 +77,12 @@ protected:
 
 class CompositeHUD {
 public:
-    using JobModel = render::Job::Model<CompositeHUD>;
+    // IF specified the input Framebuffer is actively set by the batch of this job before calling the HUDOperator.
+    // If not, the current Framebuffer is left unchanged.
+    //using Inputs = gpu::FramebufferPointer;
+    using JobModel = render::Job::ModelI<CompositeHUD, gpu::FramebufferPointer>;
 
-    CompositeHUD() {}
-    void run(const render::RenderContextPointer& renderContext);
+    void run(const render::RenderContextPointer& renderContext, const gpu::FramebufferPointer& inputs);
 };
 
 class Blit {
@@ -90,6 +92,28 @@ public:
     void run(const render::RenderContextPointer& renderContext, const gpu::FramebufferPointer& srcFramebuffer);
 };
 
+
+class ResolveFramebuffer {
+public:
+    using Inputs = render::VaryingSet2<gpu::FramebufferPointer, gpu::FramebufferPointer>;
+    using Outputs = gpu::FramebufferPointer;
+    using JobModel = render::Job::ModelIO<ResolveFramebuffer, Inputs, Outputs>;
+
+    void run(const render::RenderContextPointer& renderContext, const Inputs& source, Outputs& dest);
+};
+
+class ResolveNewFramebuffer {
+public:
+    using Inputs = gpu::FramebufferPointer;
+    using Outputs = gpu::FramebufferPointer;
+    using JobModel = render::Job::ModelIO<ResolveNewFramebuffer, Inputs, Outputs>;
+
+    void run(const render::RenderContextPointer& renderContext, const Inputs& source, Outputs& dest);
+private:
+    gpu::FramebufferPointer _outputFramebuffer;
+};
+
+
 class ExtractFrustums {
 public:
 

libraries/render-utils/src/RenderDeferredTask.cpp

@@ -257,8 +257,8 @@ void RenderDeferredTask::build(JobModel& task, const render::Varying& input, ren
     // Upscale to finale resolution
     const auto primaryFramebuffer = task.addJob<render::Upsample>("PrimaryBufferUpscale", scaledPrimaryFramebuffer);
 
-    // Composite the HUD and HUD layered objects
-    task.addJob<CompositeHUD>("HUD");
+    // Composite the HUD and HUD overlays
+    task.addJob<CompositeHUD>("HUD", primaryFramebuffer);
 
     const auto nullJitter = Varying(glm::vec2(0.0f, 0.0f));
     const auto hudOpaquesInputs = DrawLayered3D::Inputs(hudOpaque, lightingModel, nullJitter).asVarying();

libraries/render-utils/src/RenderForwardTask.cpp

@@ -126,14 +126,24 @@ void RenderForwardTask::build(JobModel& task, const render::Varying& input, rend
         task.addJob<DebugZoneLighting>("DrawZoneStack", debugZoneInputs);
     }
 
+    // Just resolve the msaa
+    const auto resolveInputs =
+        ResolveFramebuffer::Inputs(framebuffer, static_cast<gpu::FramebufferPointer>(nullptr)).asVarying();
+    const auto resolvedFramebuffer = task.addJob<ResolveFramebuffer>("Resolve", resolveInputs);
+    //auto resolvedFramebuffer = task.addJob<ResolveNewFramebuffer>("Resolve", framebuffer);
+
+#if defined(Q_OS_ANDROID)
+#else
     // Lighting Buffer ready for tone mapping
     // Forward rendering on GLES doesn't support tonemapping to and from the same FBO, so we specify 
     // the output FBO as null, which causes the tonemapping to target the blit framebuffer
-    const auto toneMappingInputs = ToneMappingDeferred::Inputs(framebuffer, static_cast<gpu::FramebufferPointer>(nullptr) ).asVarying();
+    const auto toneMappingInputs = ToneMappingDeferred::Inputs(resolvedFramebuffer, static_cast<gpu::FramebufferPointer>(nullptr)).asVarying();
     task.addJob<ToneMappingDeferred>("ToneMapping", toneMappingInputs);
+#endif
 
-    // Composite the HUD and HUD layered objects
-    task.addJob<CompositeHUD>("HUD");
+    // Layered Overlays
+    // Composite the HUD and HUD overlays
+    task.addJob<CompositeHUD>("HUD", resolvedFramebuffer);
 
     const auto hudOpaquesInputs = DrawLayered3D::Inputs(hudOpaque, lightingModel, nullJitter).asVarying();
     const auto hudTransparentsInputs = DrawLayered3D::Inputs(hudTransparent, lightingModel, nullJitter).asVarying();
@@ -145,26 +155,32 @@ void RenderForwardTask::build(JobModel& task, const render::Varying& input, rend
     // task.addJob<Blit>("Blit", framebuffer);
 }
 
+void PrepareFramebuffer::configure(const Config& config) {
+    _numSamples = config.getNumSamples();
+}
+
 void PrepareFramebuffer::run(const RenderContextPointer& renderContext, gpu::FramebufferPointer& framebuffer) {
     glm::uvec2 frameSize(renderContext->args->_viewport.z, renderContext->args->_viewport.w);
 
     // Resizing framebuffers instead of re-building them seems to cause issues with threaded rendering
-    if (_framebuffer && _framebuffer->getSize() != frameSize) {
+    if (_framebuffer && (_framebuffer->getSize() != frameSize || _framebuffer->getNumSamples() != _numSamples)) {
         _framebuffer.reset();
     }
 
     if (!_framebuffer) {
         _framebuffer = gpu::FramebufferPointer(gpu::Framebuffer::create("forward"));
 
+        int numSamples = _numSamples;
+
         auto colorFormat = gpu::Element::COLOR_SRGBA_32;
-        auto defaultSampler = gpu::Sampler(gpu::Sampler::FILTER_MIN_MAG_POINT);
+        auto defaultSampler = gpu::Sampler(gpu::Sampler::FILTER_MIN_MAG_LINEAR);
         auto colorTexture =
-            gpu::Texture::createRenderBuffer(colorFormat, frameSize.x, frameSize.y, gpu::Texture::SINGLE_MIP, defaultSampler);
+            gpu::Texture::createRenderBufferMultisample(colorFormat, frameSize.x, frameSize.y, numSamples, defaultSampler);
         _framebuffer->setRenderBuffer(0, colorTexture);
 
         auto depthFormat = gpu::Element(gpu::SCALAR, gpu::UINT32, gpu::DEPTH_STENCIL);  // Depth24_Stencil8 texel format
         auto depthTexture =
-            gpu::Texture::createRenderBuffer(depthFormat, frameSize.x, frameSize.y, gpu::Texture::SINGLE_MIP, defaultSampler);
+           gpu::Texture::createRenderBufferMultisample(depthFormat, frameSize.x, frameSize.y, numSamples, defaultSampler);
         _framebuffer->setDepthStencilBuffer(depthTexture, depthFormat);
     }
 

libraries/render-utils/src/RenderForwardTask.h

@@ -27,16 +27,37 @@ public:
     void build(JobModel& task, const render::Varying& input, render::Varying& output);
 };
 
+
+class PrepareFramebufferConfig : public render::Job::Config {
+    Q_OBJECT
+    Q_PROPERTY(int numSamples WRITE setNumSamples READ getNumSamples NOTIFY dirty)
+public:
+    int getNumSamples() const { return numSamples; }
+    void setNumSamples(int num) {
+        numSamples = std::max(1, std::min(32, num));
+        emit dirty();
+    }
+
+signals:
+    void dirty();
+
+protected:
+    int numSamples{ 4 };
+};
+
 class PrepareFramebuffer {
 public:
     using Inputs = gpu::FramebufferPointer;
-    using JobModel = render::Job::ModelO<PrepareFramebuffer, Inputs>;
+    using Config = PrepareFramebufferConfig;
+    using JobModel = render::Job::ModelO<PrepareFramebuffer, Inputs, Config>;
 
+    void configure(const Config& config);
     void run(const render::RenderContextPointer& renderContext,
             gpu::FramebufferPointer& framebuffer);
 
 private:
     gpu::FramebufferPointer _framebuffer;
+    int _numSamples;
 };
 
 class PrepareForward {