Bug when deal with phred 64 reads

[asdcid]

Hi,

I found if I use fastp to process phred 64 reads (with -6), if the data amount is large (such as >20,000 reads, haven't tested the exact number),

1. fastp cannot correct calculate the Q20/Q30 base and the quality_curves is still >60 (haven't changed it to phred 33 to show).
2. the result cannot be repeated (the results were slightly different each time with the same parameters).

However, if the data amount is small (say <15000), everything is fine.

This situation only happens in the version > 0.20.0. In v 0.20.0, everything is fine (both the read quality and result between different runs).

I am not sure whether it is related the speed up after 0.20.0?

Many thanks, Raymond