jqj@cornell.UUCP (01/21/85)
From: jqj (J Q Johnson) Now that the flames about optimizing C compilers are dieing down, perhaps we could return to the subject that spurred the furious debate: the merits and demerits of the Tartan C compiler. Specifically, does anyone yet have any hands-on experience with it? Lets grant that the Tartan compiler will change the behavior of some programs (at least make them run faster!). Let's grant that it would NOT be possible to simply recompile the 4.2BSD kernel and expect it to run correctly. But would it be possible to recompile the (vast) majority of user programs distributed as part of 4.2BSD? Tartan reports quite substantial performance improvements for typical C programs. For example, Baskett's "puzzle" program runs 3 times faster and is 22% smaller in code size compared to PCC. On the other hand, none of the sample programs Tartan benchmarks have been hand optimized (they don't even have any "register" declarations); this suggests that real Tartan C improvements might be much less than advertised. I would be most interested in anyone's actual handson experience; speculation on possible performance is much less useful.
Ron Natalie <ron@BRL-TGR> (01/27/85)
Tartan will give you a 30 day free trial on their compiler. -Ron
dgp@lanl.ARPA (01/31/85)
> Tartan will give you a 30 day free trial on their compiler. > > -Ron *** REPLACE THIS LINE WITH YOUR MESSAGE ***
whm@arizona.UUCP (Bill Mitchell) (02/16/85)
I stopped by the Tartan booth at UniForum and picked up a trial license
form, hoping to be able to get a copy of Tartan C for the Vax for
purposes of evaluation. I guess I didn't read the fine print or something
because I was under the impression that the 30-day trial license was
*free*. Anyway, I sent the form in a few days ago and today I got a
call from Tartan saying that we have send in a purchase order in for
the full purchase price in order to get the 30-day trial. They say
they won't cash it unless we keep the system longer than 30 days, but
it seems like what they've got is closer to "money-back guarantee"
than "free trial". Sigh.
I wonder if Tartan will try and retarget the Vax system for the CCI 6/32
(aka Sperry 7000/40) since the 6/32's instruction set seems to be a very
large subset of the Vax. If they could get the same performance gains as
they claim for the Vax, one might end up with a system that's an average
of 6-7 times faster than a 780 with pcc. An interesting possibility to
say the least.
Bill Mitchell
whm.arizona@csnet-relay
{ihnp4,noao,mcnc,utah-cs}!arizona!whmwls@astrovax.UUCP (William L. Sebok) (10/24/85)
My Tartan C tape arrived here yesterday. The first thing I've done is to
recompile /lib/libc.a. I substituted tcc for cc in everything except the
stuff in /usr/src/lib/libc/vax, where assembler stuff resides. There were
no compilation errors.
I will say more when I get more experience. I intend to recompile the
utilities a bit at a time. In the compiler's documentaion they list 4.2 BSD
programs with which they have had trouble. The problems are mostly bugs in the
sources being compiled and read like articles in net.bugs.4bsd or entries in
the MtXinu bugs tape.
Things that they say to watch out for:
1) their C preprocessor only works on source with legal C tokens and cannot be
used to massage assembler programs.
2) although they recognize the "asm" construction, such constructions often
contain compiler dependent assumptions that are not valid with tcc. In
particular the first item declared "register" will not necessarily be in r11,
the 2nd in r10, and so on, as is the case in "cc".
3) The same considerations as in 2) also apply to sed scripts which massage
assembler output.
4) the C optimizer c2 can not usefully be applied to tcc assembler output.
--
Bill Sebok Princeton University, Astrophysics
{allegra,akgua,burl,cbosgd,decvax,ihnp4,noao,princeton,vax135}!astrovax!wls