#682 Add an update rate parameter to help throttle mirror updates.

So far 4 seems like a good balance for performance and quality.  2 is great for quality, with 6 or 8 being for higher performance.  Also bring back the gaussian filter - may end up adding the FXAA filter though instead.

1 files changed, 24 insertions, 0 deletions

diff --git a/indra/newview/llheroprobemanager.cpp b/indra/newview/llheroprobemanager.cpp
index 929364be44..0d9efb860a 100644
--- a/indra/newview/llheroprobemanager.cpp
+++ b/indra/newview/llheroprobemanager.cpp
@@ -264,6 +264,30 @@ void LLHeroProbeManager::updateProbeFace(LLReflectionMap* probe, U32 face, F32 n
         LLRenderTarget *screen_rt = &gPipeline.mHeroProbeRT.screen;
         LLRenderTarget *depth_rt  = &gPipeline.mHeroProbeRT.deferredScreen;
 
+        // perform a gaussian blur on the super sampled render before downsampling
+        {
+            gGaussianProgram.bind();
+            gGaussianProgram.uniform1f(resScale, 1.f / (mProbeResolution * 2));
+            S32 diffuseChannel = gGaussianProgram.enableTexture(LLShaderMgr::DEFERRED_DIFFUSE, LLTexUnit::TT_TEXTURE);
+
+            // horizontal
+            gGaussianProgram.uniform2f(direction, 1.f, 0.f);
+            gGL.getTexUnit(diffuseChannel)->bind(screen_rt);
+            mRenderTarget.bindTarget();
+            gPipeline.mScreenTriangleVB->setBuffer();
+            gPipeline.mScreenTriangleVB->drawArrays(LLRender::TRIANGLES, 0, 3);
+            mRenderTarget.flush();
+
+            // vertical
+            gGaussianProgram.uniform2f(direction, 0.f, 1.f);
+            gGL.getTexUnit(diffuseChannel)->bind(&mRenderTarget);
+            screen_rt->bindTarget();
+            gPipeline.mScreenTriangleVB->setBuffer();
+            gPipeline.mScreenTriangleVB->drawArrays(LLRender::TRIANGLES, 0, 3);
+            screen_rt->flush();
+            gGaussianProgram.unbind();
+        }
+
         S32 mips = log2((F32)mProbeResolution) + 0.5f;
 
         gReflectionMipProgram.bind();