Adding gpu timer feature to actually the GPU cost of rendering

libraries/gpu/src/gpu/GLBackendQuery.cpp

@@ -100,6 +100,7 @@ void GLBackend::do_getQuery(Batch& batch, size_t paramOffset) {
         #else
             glGetQueryObjectui64v(glquery->_qo, GL_QUERY_RESULT, &glquery->_result);
         #endif
+            query->triggerReturnHandler(glquery->_result);
         (void)CHECK_GL_ERROR();
     }
 }

libraries/gpu/src/gpu/Query.cpp

@@ -10,11 +10,13 @@
 //
 #include "Query.h"
 
-#include <QDebug>
+#include "GPULogging.h"
+#include "Batch.h"
 
 using namespace gpu;
 
-Query::Query()
+Query::Query(const Handler& returnHandler) :
+    _returnHandler(returnHandler)
 {
 }
 
@@ -22,6 +24,45 @@ Query::~Query()
 {
 }
 
-double Query::getElapsedTime() {
+double Query::getElapsedTime() const {
-    return 0.0;
+    return ((double) _queryResult) * 0.000001;
 }
+
+void Query::triggerReturnHandler(uint64_t queryResult) {
+    _queryResult = queryResult;
+    if (_returnHandler) {
+        _returnHandler(*this);
+    }
+}
+
+
+Timer::Timer() {
+    _timerQueries.resize(6, std::make_shared<gpu::Query>([&] (const Query& query) {
+        auto elapsedTime = query.getElapsedTime();
+        _movingAverage.addSample(elapsedTime);
+
+        static int frameNum = 0;
+        frameNum++;
+        frameNum %= 10;
+        if (frameNum == 0) {
+            auto value = _movingAverage.average;
+            qCDebug(gpulogging) << "AO on gpu = " << value;
+        }
+    }));
+
+}
+
+void Timer::begin(gpu::Batch& batch) {
+    batch.beginQuery(_timerQueries[_currentTimerQueryIndex]);
+}
+void Timer::end(gpu::Batch& batch) {
+    batch.endQuery(_timerQueries[_currentTimerQueryIndex]);
+    _currentTimerQueryIndex = (_currentTimerQueryIndex + 1) % _timerQueries.size();
+
+    auto returnQuery = _timerQueries[_currentTimerQueryIndex];
+    batch.getQuery(returnQuery);
+}
+
+void Timer::onQueryReturned(const Query& query) {
+
+}

libraries/gpu/src/gpu/Query.h

@@ -13,26 +13,56 @@
 
 #include <assert.h>
 #include <memory>
+#include <functional>
 #include <vector>
+#include <SimpleMovingAverage.h>
 
 #include "Format.h"
 
 namespace gpu {
 
+    class Batch;
+
     class Query {
     public:
-        Query();
+        using Handler = std::function<void(const Query&)>;
+
+        Query(const Handler& returnHandler);
         ~Query();
 
-        uint32 queryResult;
+        double getElapsedTime() const;
-
-        double getElapsedTime();
 
         const GPUObjectPointer gpuObject {};
+        void triggerReturnHandler(uint64_t queryResult);
+    protected:
+        Handler _returnHandler;
+
+        uint64_t _queryResult = 0;
     };
     
     typedef std::shared_ptr<Query> QueryPointer;
     typedef std::vector< QueryPointer > Queries;
+
+
+    // gpu timer is just returning an estimate of the time taken by a chunck of work delimited by the 
+    // begin and end calls repeated for several times.
+    // The result is always a late average of the time spent for that same task a few cycles ago.
+    class Timer {
+    public:
+        Timer();
+        void begin(gpu::Batch& batch);
+        void end(gpu::Batch& batch);
+        
+        double getAverage() const;
+
+    protected:
+            
+        void onQueryReturned(const Query& query);
+
+        gpu::Queries _timerQueries;
+        int _currentTimerQueryIndex = 0;
+        MovingAverage<double, 5> _movingAverage;
+    };
 };
 
 #endif

libraries/render-utils/src/AmbientOcclusionEffect.cpp

@@ -19,6 +19,7 @@
 #include <SharedUtil.h>
 #include <gpu/Context.h>
 #include <gpu/StandardShaderLib.h>
+#include "RenderUtilsLogging.h"
 
 #include "AmbientOcclusionEffect.h"
 #include "TextureCache.h"
@@ -240,8 +241,6 @@ void AmbientOcclusionEffect::run(const render::SceneContextPointer& sceneContext
         args->_context->getStereoProjections(projMats);
         args->_context->getStereoViews(eyeViews);
 
-        float halfWidth = 0.5f * sWidth;
-
         for (int i = 0; i < 2; i++) {
             // Compose the mono Eye space to Stereo clip space Projection Matrix
             auto sideViewMat = projMats[i] * eyeViews[i];
@@ -260,6 +259,8 @@ void AmbientOcclusionEffect::run(const render::SceneContextPointer& sceneContext
     gpu::doInBatch(args->_context, [=](gpu::Batch& batch) {
         batch.enableStereo(false);
 
+        _gpuTimer.begin(batch);
+
         batch.setViewportTransform(args->_viewport);
         batch.setProjectionTransform(glm::mat4());
         batch.setViewTransform(Transform());
@@ -302,5 +303,7 @@ void AmbientOcclusionEffect::run(const render::SceneContextPointer& sceneContext
         batch.setResourceTexture(AmbientOcclusionEffect_OcclusionMapSlot, occlusionBlurredFBO->getRenderBuffer(0));
         batch.draw(gpu::TRIANGLE_STRIP, 4);
 
+        _gpuTimer.end(batch);
+
     });
 }

libraries/render-utils/src/AmbientOcclusionEffect.h

@@ -16,6 +16,7 @@
 
 #include "render/DrawTask.h"
 
+
 class AmbientOcclusionEffect {
 public:
 
@@ -85,6 +86,8 @@ private:
     gpu::PipelinePointer _occlusionPipeline;
     gpu::PipelinePointer _hBlurPipeline;
     gpu::PipelinePointer _vBlurPipeline;
+
+    gpu::Timer _gpuTimer;
 };
 
 #endif // hifi_AmbientOcclusionEffect_h

libraries/render-utils/src/RenderDeferredTask.cpp

@@ -68,7 +68,6 @@ void RenderDeferred::run(const SceneContextPointer& sceneContext, const RenderCo
 }
 
 void ToneMappingDeferred::run(const SceneContextPointer& sceneContext, const RenderContextPointer& renderContext) {
-    PerformanceTimer perfTimer("ToneMappingDeferred");
     _toneMappingEffect.render(renderContext->getArgs());
 }
 

libraries/render-utils/src/RenderDeferredTask.h

@@ -156,10 +156,6 @@ public:
     int getToneMappingToneCurve() const;
 
     virtual void run(const render::SceneContextPointer& sceneContext, const render::RenderContextPointer& renderContext);
-
-
-    gpu::Queries _timerQueries;
-    int _currentTimerQueryIndex = 0;
 };
 
 

libraries/render-utils/src/ssao_makeOcclusion.slf

@@ -32,7 +32,6 @@ vec3 evalEyePositionFromZeye(ivec2 side, float Zeye, vec2 texcoord) {
     return vec3(Xe, Ye, Zeye);
 }
 
-in vec2 varTexCoord0;
 out vec4 outFragColor;
 
 uniform sampler2D normalMap;