8x shader compilation optimization and watchdog disable

addPlumberPipeline in RenderPipelines.cpp would recompile
a single program 7 more times then necessary.  This was
causing startup times on Mac OS X to trigger the deadlock
watchdog.

Even with this fix, compiling shaders is still too slow,
As a workaround, we disable the  watchdog thread during
rendering pipeline setup.

interface/src/Application.cpp

@@ -378,6 +378,7 @@ public:
         setObjectName("Deadlock Watchdog");
         // Give the heartbeat an initial value
         _heartbeat = usecTimestampNow();
+        _paused = false;
         connect(qApp, &QCoreApplication::aboutToQuit, [this] {
             _quit = true;
         });
@@ -395,11 +396,20 @@ public:
         *crashTrigger = 0xDEAD10CC;
     }
 
+    static void pause() {
+        _paused = true;
+    }
+
+    static void resume() {
+        _paused = false;
+        updateHeartbeat();
+    }
+
     void run() override {
         while (!_quit) {
             QThread::sleep(HEARTBEAT_UPDATE_INTERVAL_SECS);
             // Don't do heartbeat detection under nsight
-            if (nsightActive()) {
+            if (nsightActive() || _paused) {
                 continue;
             }
             uint64_t lastHeartbeat = _heartbeat; // sample atomic _heartbeat, because we could context switch away and have it updated on us
@@ -455,6 +465,7 @@ public:
         }
     }
 
+    static std::atomic<bool> _paused;
     static std::atomic<uint64_t> _heartbeat;
     static std::atomic<uint64_t> _maxElapsed;
     static std::atomic<int> _maxElapsedAverage;
@@ -463,6 +474,7 @@ public:
     bool _quit { false };
 };
 
+std::atomic<bool> DeadlockWatchdogThread::_paused;
 std::atomic<uint64_t> DeadlockWatchdogThread::_heartbeat;
 std::atomic<uint64_t> DeadlockWatchdogThread::_maxElapsed;
 std::atomic<int> DeadlockWatchdogThread::_maxElapsedAverage;
@@ -2206,6 +2218,11 @@ void Application::initializeGL() {
     initDisplay();
     qCDebug(interfaceapp, "Initialized Display.");
 
+#ifdef Q_OS_OSX
+    // FIXME: on mac os the shaders take up to 1 minute to compile, so we pause the deadlock watchdog thread.
+    DeadlockWatchdogThread::pause();
+#endif
+
     // Set up the render engine
     render::CullFunctor cullFunctor = LODManager::shouldRender;
     static const QString RENDER_FORWARD = "HIFI_RENDER_FORWARD";
@@ -2213,6 +2230,11 @@ void Application::initializeGL() {
     _renderEngine->addJob<UpdateSceneTask>("UpdateScene");
     _renderEngine->addJob<SecondaryCameraRenderTask>("SecondaryCameraJob", cullFunctor);
     _renderEngine->addJob<RenderViewTask>("RenderMainView", cullFunctor, isDeferred);
+
+#ifdef Q_OS_OSX
+    DeadlockWatchdogThread::resume();
+#endif
+
     _renderEngine->load();
     _renderEngine->registerScene(_main3DScene);
 

libraries/render/src/render/ShapePipeline.cpp

@@ -70,26 +70,30 @@ void ShapePlumber::addPipeline(const Key& key, const gpu::ShaderPointer& program
 
 void ShapePlumber::addPipeline(const Filter& filter, const gpu::ShaderPointer& program, const gpu::StatePointer& state,
         BatchSetter batchSetter, ItemSetter itemSetter) {
-    gpu::Shader::BindingSet slotBindings;
-    slotBindings.insert(gpu::Shader::Binding(std::string("lightingModelBuffer"), Slot::BUFFER::LIGHTING_MODEL));
-    slotBindings.insert(gpu::Shader::Binding(std::string("skinClusterBuffer"), Slot::BUFFER::SKINNING));
-    slotBindings.insert(gpu::Shader::Binding(std::string("materialBuffer"), Slot::BUFFER::MATERIAL));
-    slotBindings.insert(gpu::Shader::Binding(std::string("texMapArrayBuffer"), Slot::BUFFER::TEXMAPARRAY));
-    slotBindings.insert(gpu::Shader::Binding(std::string("albedoMap"), Slot::MAP::ALBEDO));
-    slotBindings.insert(gpu::Shader::Binding(std::string("roughnessMap"), Slot::MAP::ROUGHNESS));
-    slotBindings.insert(gpu::Shader::Binding(std::string("normalMap"), Slot::MAP::NORMAL));
-    slotBindings.insert(gpu::Shader::Binding(std::string("metallicMap"), Slot::MAP::METALLIC));
-    slotBindings.insert(gpu::Shader::Binding(std::string("emissiveMap"), Slot::MAP::EMISSIVE_LIGHTMAP));
-    slotBindings.insert(gpu::Shader::Binding(std::string("occlusionMap"), Slot::MAP::OCCLUSION));
-    slotBindings.insert(gpu::Shader::Binding(std::string("scatteringMap"), Slot::MAP::SCATTERING));
-    slotBindings.insert(gpu::Shader::Binding(std::string("lightBuffer"), Slot::BUFFER::LIGHT));
-    slotBindings.insert(gpu::Shader::Binding(std::string("lightAmbientBuffer"), Slot::BUFFER::LIGHT_AMBIENT_BUFFER));
-    slotBindings.insert(gpu::Shader::Binding(std::string("skyboxMap"), Slot::MAP::LIGHT_AMBIENT));
-    slotBindings.insert(gpu::Shader::Binding(std::string("fadeMaskMap"), Slot::MAP::FADE_MASK));
-    slotBindings.insert(gpu::Shader::Binding(std::string("fadeParametersBuffer"), Slot::BUFFER::FADE_PARAMETERS));
-    slotBindings.insert(gpu::Shader::Binding(std::string("hazeBuffer"), Slot::BUFFER::HAZE_MODEL));
 
-    gpu::Shader::makeProgram(*program, slotBindings);
+    // don't call makeProgram on shaders that are already made.
+    if (program->getUniformBuffers().empty()) {
+        gpu::Shader::BindingSet slotBindings;
+        slotBindings.insert(gpu::Shader::Binding(std::string("lightingModelBuffer"), Slot::BUFFER::LIGHTING_MODEL));
+        slotBindings.insert(gpu::Shader::Binding(std::string("skinClusterBuffer"), Slot::BUFFER::SKINNING));
+        slotBindings.insert(gpu::Shader::Binding(std::string("materialBuffer"), Slot::BUFFER::MATERIAL));
+        slotBindings.insert(gpu::Shader::Binding(std::string("texMapArrayBuffer"), Slot::BUFFER::TEXMAPARRAY));
+        slotBindings.insert(gpu::Shader::Binding(std::string("albedoMap"), Slot::MAP::ALBEDO));
+        slotBindings.insert(gpu::Shader::Binding(std::string("roughnessMap"), Slot::MAP::ROUGHNESS));
+        slotBindings.insert(gpu::Shader::Binding(std::string("normalMap"), Slot::MAP::NORMAL));
+        slotBindings.insert(gpu::Shader::Binding(std::string("metallicMap"), Slot::MAP::METALLIC));
+        slotBindings.insert(gpu::Shader::Binding(std::string("emissiveMap"), Slot::MAP::EMISSIVE_LIGHTMAP));
+        slotBindings.insert(gpu::Shader::Binding(std::string("occlusionMap"), Slot::MAP::OCCLUSION));
+        slotBindings.insert(gpu::Shader::Binding(std::string("scatteringMap"), Slot::MAP::SCATTERING));
+        slotBindings.insert(gpu::Shader::Binding(std::string("lightBuffer"), Slot::BUFFER::LIGHT));
+        slotBindings.insert(gpu::Shader::Binding(std::string("lightAmbientBuffer"), Slot::BUFFER::LIGHT_AMBIENT_BUFFER));
+        slotBindings.insert(gpu::Shader::Binding(std::string("skyboxMap"), Slot::MAP::LIGHT_AMBIENT));
+        slotBindings.insert(gpu::Shader::Binding(std::string("fadeMaskMap"), Slot::MAP::FADE_MASK));
+        slotBindings.insert(gpu::Shader::Binding(std::string("fadeParametersBuffer"), Slot::BUFFER::FADE_PARAMETERS));
+        slotBindings.insert(gpu::Shader::Binding(std::string("hazeBuffer"), Slot::BUFFER::HAZE_MODEL));
+
+        gpu::Shader::makeProgram(*program, slotBindings);
+    }
 
     auto locations = std::make_shared<Locations>();