Paul Mackerras 23ff954059 core: Change bperm to a simpler and slower implementation ...

This does bperm in the bitsort unit instead of the logical unit, and
no longer tries to do it in a single cycle with eight 64-to-1
multiplexers.  Instead it is now a state machine in the bitsort unit,
takes 8 cycles, and only has one 64-to-1 multiplexer.  This helps
improve timing and reduces LUT usage.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

2025-01-17 19:36:20 +11:00

4.9 KiB

Raw Blame History

View Raw