Again, the performance is terrible (~3500 us).Īre there texture formats that allow for a faster copy or performance tips for Intel UHD that address this issue? So I tried with a shader that samples from the ms depth texture and writes only the 0th sample to depth. However, in d3d11 it's not possible to resolve a resource with BIND_DEPTH_STENCIL flag, so this approach is not viable for the depth copy. This approach is significantly faster than the full ms texture copy. I also tried a slightly different approach: instead of copying the entire ms texture, I downsampled (resolved) the ms color texture and copied it to a single sample resource (multi-sample is important for the source of the copy, the destination could be a single-sampled texture without loosing too much graphics quality). Since I need to copy both resources in the render loop, those performances make the app almost unusable when the graphics adapter in use is the Intel UHD. ![]() Profiling with both Graphics Frame Analyzer and d3d time queries, it looks like a single copy can take up to 5000 us for a 1200x800 R8G8B8A8_UNORM (x8 samples) resource and 6000 us for a D24_UNORM_S8_UINT (x8) of the same size. ![]() I noticed that ID3D11DeviceContext::CopyResource is really slow on Intel UHD 630: I need to perform a copy of multisample textures (used as render targets as well as shader resources) in the render loop of my application.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |