Re: Writing fast compilers...

[From comp.arch -John]


In article <PCG.91Aug11154854@aberdb.aber.ac.uk> pcg@aber.ac.uk (Piercarlo Grandi) writes:


| Now, now, let's not get carried away. He has done a clone of Turbo
| Pascal, or of the small/fast Pascal compiler from ETH. It is good
| implementation architecture, yes, but it shines only in comparison with
| the awful things we usually get and accept.


    Single pass compilers are hardly that recent, and small compilers are
not new. The original B compiler (two pass because of memory limits) was
19 pages of code for one pass and either 22 or 29 for the other (that's
going back a way).


    Note that true one pass compilers can not generate best possible code
for some machines, notably those which have a short branch instruction,
since forward jumps must be coded as long jumps. I believe the plan 9
compiler delays code generation for a bit, so peephole optimization can
be done. That permits a little code motion, better register use, etc.
--
bill davidsen (davidsen@crdos1.crd.GE.COM -or- uunet!crdgw1!crdos1!davidsen)
    GE Corp R&D Center, Information Systems Operation, tech support group
    Moderator comp.binaries.ibm.pc and 386-users digest.
--

Return to the comp.compilers page.
Search the comp.compilers archives again.