v3.7.0
NRD v3.7.0:
HIGHLIGHTS:
- NRD: introduced NRD_NORMAL_ENCODING and NRD_ROUGHNESS_ENCODING parameters (controlled via CMake) to cover more encoding variants out of the box
- NRD: introduced read-only "normalEncoding" and "roughnessEncoding" in "LibraryDesc" describing which encoding NRD has been compiled with
- NRD: improved curvature estimation and usage for any type of supported normal encoding
- NRD: fixed wrong curvature direction (was Y-flipped)
- NRD: CLEAR_AND_RESTART now clears only "history" resources, including user-provided outputs
- RELAX: fixed dynamic resolution scale related issue for specular (regression)
- REBLUR: improved specular tracking
- REBLUR: removed "enableAdvancedPrepass"
- REBLUR: reduced potential chroma shifts
- REBLUR_SH: fixed bias (increased lightness)
- REBLUR_DIFFUSE_SPECULAR_SH: fixed wrong resource bindings if pre-pass is off
- NRD INTEGRATION: descriptor caching is now based on native resource pointers instead of NRI pointers
DETAILS:
- NRD: introduced NRD_NORMAL_ENCODING and NRD_ROUGHNESS_ENCODING
- NRD: improved clears on CLEAR_AND_RESTART
- NRD: removed shared NRD_NORMAL_ENCODING_ERROR
- NRD: curvature error compensation made normal encoding dependent
- NRD: fixed numerical instabilities in curvature estimation
- NRD: fixed how thin lens equation interacts with concave surfaces
- NRD: fixed wrong calculation of curvature direction
- RELAX: NRD_NORMAL_ENCODING_ERROR => RELAX_NORMAL_ENCODING_ERROR (no functional changes)
- RELAX: fixed resolution scale regression in virtual motion based reprojection for specular
- RELAX: moving to using "ApplyThinLensEquation"
- REBLUR: NORMAL_ULP increased from 0.0009 to 0.004 (1/255)
- REBLUR: consistent REBLUR_NORMAL_ULP usage
- REBLUR: removed / resolved some TODOs
- REBLUR: prev-prev test improvements
- REBLUR: changing luma doesn't allow energy boost
- REBLUR: removed "enableAdvancedPrepass" (low IQ gains, adds bias in SH mode)
- REBLUR_DIFFUSE_SPECULAR_SH: fixed wrong resource bindings if pre-pass is off
- NRD INTEGRATION: descriptor caching is now based on native resource pointers instead of NRI pointers
- NRD INTEGRATION: improved debug output
- NRD INTEGRATION: NRI updated to v1.86
- updated MathLib