>>>>> "Bernd" == Bernd Paysan <[email protected]> writes:
>> If somebody wants to do Mips64-specific tuning to Gforth, I can hand >> out login access to the machine in question. > Look at some benchmarks with see-code in gforth-fast, if it doesn't > look too bad, there's no problem. Yes, benchmarks look fine. Without --dynamic there's no noticeable performance impact. With --dynamic, gforth-fast is slightly faster, which might be due to the mips-specific primitive relocatability checks having no 64-bit opcodes in the opcode blacklist. Not that any checks would be needed. Everything new enough to have 64-bit opcodes at all, should not have any of the delay slots and instruction order constraints these checks enforce. Will have a look at that again, maybe relaxing some checks we can speed 32-bit gforth-fast somewhat up (AFAIK Debian nowadays compiles with -march=mips2, so if we detected cpu >= mips2 we may disable some checks). David -- GnuPG public key: http://dvdkhlng.users.sourceforge.net/dk2.gpg Fingerprint: B63B 6AF2 4EEB F033 46F7 7F1D 935E 6F08 E457 205F
pgp3XduqFzEd5.pgp
Description: PGP signature
