Description:
When running the lanczos function from ffx_fsr2.cpp clang uses an integer based version of abs and implicitly converts the value argument to an integer format using cvtss2si and back to float with cvtsi2ss afterwards. This causes the full lanczos kernel mask to be basically useless, turning FSR output into a blur.
OS: Arch-Linux Compiler: clang16 (blobbed)
Description: When running the lanczos function from ffx_fsr2.cpp clang uses an integer based version of abs and implicitly converts the value argument to an integer format using cvtss2si and back to float with cvtsi2ss afterwards. This causes the full lanczos kernel mask to be basically useless, turning FSR output into a blur.
Resolution: It should probably use absf instead.