llvm-gcc4_2

hi

> i fixed some major bugs in the llvm-gcc4_2 patch since my last post.
> this brings me to an interesting question:
>
> what are your criteria to use my patch for further development?

I have no fixed criteria. I think the best solution is for the work to be
checked into the (forthcoming) public SVN server. When it is available,
you can check it in there and we can all hack on it together. Even if it
has serious bugs or known deficiencies, I see no problem in that.

ok, thats fine to me.
i think the worst part of the work is done.

> at the moment i use the llvm-test1.9 suite for testing, because the
> actual llvm-test trunk in the repository seems to be broken.
> in my configuration the official llvm-gcc4 does not run completely error
> free against this suite (or is this my configuration fault?).
> so i guess a 100% error free run against the testsuite will not be the
> criteria.

Hrm, that's strange. llvm-test doesn't pass 100% (e.g. some of the EH
stuff is now working yet), but it should ahve basic functionality. What
is broken?

major error causes:

the sse.* tests produce errors during dagselection (see log)

there is a problem when exception handling is used because
all the __cxa.* functions are not found (bei llc and when compiling c-backend code) (i guess thats what you were refering to)

and a few other failures i did not relate to this 2 groups

for more information:

here is a list of failures (53):
http://geldspeicher.entenhausen.at/dl/failure_list.txt

and here is the log of the above test run :
http://geldspeicher.entenhausen.at/dl/testlog.txt.gz (36K)
or
http://geldspeicher.entenhausen.at/dl/testlog.txt (556K)

(the test run is not complete, just what my computer could calc in
one hour to check the sanity of th test)

kind regards
peter

ok, thats fine to me.
i think the worst part of the work is done.

wow, you look quite far along.

Hrm, that's strange. llvm-test doesn't pass 100% (e.g. some of the EH
stuff is now working yet), but it should ahve basic functionality. What
is broken?

major error causes:

the sse.* tests produce errors during dagselection (see log)

there is a problem when exception handling is used because
all the __cxa.* functions are not found (bei llc and when compiling c-backend code) (i guess thats what you were refering to)

and a few other failures i did not relate to this 2 groups

Right, C++ EH is disabled in llvm-gcc at this point. Some of the failures do look real though, things like pointer_method, BuiltinTypeInfo, ofstream_ctor, etc.

Very nice. Reid/Anton, do you guys have an idea when the svn server will be available?

-Chris

Very nice. Reid/Anton, do you guys have an idea when the svn server will
be available?

Mid April.

The SVN server is getting installed tomorrow. Anton and I are still
testing the automated conversion script (almost there).

We will probably make a beta svn server available (via my machine) for
anyone that wants a preview.

One of the things we have to do is also convert llvm-test and we haven't
tested it much at all.

Very nice. Reid/Anton, do you guys have an idea when the svn server will
be available?

Mid April.

The SVN server is getting installed tomorrow. Anton and I are still
testing the automated conversion script (almost there).

We will probably make a beta svn server available (via my machine) for
anyone that wants a preview.

Good deal.

One of the things we have to do is also convert llvm-test and we haven't
tested it much at all.

One positive thing is that we don't really care about history for most of llvm-test (we do care for the makefiles etc, but not the programs themselves). If some things are lost, noone will be too unhappy. I don't know if this simplifies things or not though :slight_smile:

-Chris

Not at all. Getting rid of the history would be *hard* at this point :slight_smile:

We're actually not going to convert llvm-test for a while. By "convert
llvm-test" I meant "make sure nightly test script can work with an svn
based llvm". It will still check out llvm-test from cvs. We'll migrate
llvm-test, llvm-java, hlvm, etc. once the basic conversion of llvm is
done.

Reid.