This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: RFA: MIPS paired single support


> OK.  Sounds reasonable, but it isn't very well specified.

I know, I don't think we have a spec for what goes in the
target-specific subdirectories.  It varies with each target what kind
of things we're looking for and trying to test.  For example, for
xstormy16 I need to check for the presense of certain opcodes in the
assembler *and* for proper execution, so I had to do something a
little custom.

> I think it would make more sense if I could see an example.  Are you
> using the dg-* commands?

Half use c-torture, half use gcc-dg.

> Do you still have target tests in individual testcase .c files?

Some do, some don't.  The xstormy16 test has a directory tree full of
individual dg tests, plus some common sources.  Another I have in the
works has one executable built of a couple of sources (each compiled a
different way to test compatibility) that internally runs many tests,
with a custom driver (we'd want a second .exp if we added dg to that
directory).

> Do you have multiple .exp files in each <cpu> directory?

Not at the moment, but I don't see why it wouldn't be allowed, if only
to make maintenance easier.  IMHO being target-specific directories,
it's understandable that you'd need to do unique things sometimes.

I'm thinking of the gas testsuite, which has multiple .exp's per
target.  It's (arguably) a bit messier than the gcc testsuite, but it
demonstrates that different targets have different testsuite needs.

> Is there an established convention for which one is better?  Are
> there established conventions on how to write the <foo>.exp files?

No to both, because so far each target has had different needs.

> If Red Hat adds the first one, then I could make the mips one work the
> same way.

Since our aren't consistent, I suggest just using plain gcc-dg or
c-torture, whichever makes the most sense.  Or pick one (dg?) and make
that the rule for new targets.

> There is also the question of whether Red Hat might already have a
> gcc.target/mips directory.  If so, I don't want to conflict with it.

[checks] Nope.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]