D3D12RaytracingRealTimeDenoisedAmbientOcclusion - large per frame upload #681
Description
m_hemisphereSamplesGPUBuffer.CopyStagingToGpu(frameIndex) in RTAO:Run takes about 8ms of CPU time on i74790K 4.4GHz 32GB RAM 2070 RTX driver 442.92 Win10 1909. The function performs a memcpy of 87031808 bytes per frame. This big upload is not even required. It is should be enough to upload it once on AO samples recreation. If I modify the source code to upload the buffer just for 3 first (in flight) frames then my framerate increases from 100 to 300fps.