[Bug target/14650] Testsuite failures with -mabi=64 in compile/20031023-1.c, 20031023-2.c and 20031023-3.c

[rsandifo at redhat dot com]

------- Additional Comments From rsandifo at redhat dot com  2004-04-14 19:35 -------
Subject: Re:  Testsuite failures with -mabi=64 in
 compile/20031023-1.c, 20031023-2.c and 20031023-3.c

"ghazi at caip dot rutgers dot edu" <gcc-bugzilla@gcc.gnu.org> writes:
> Richard - Can you suggest a minimal C testcase that passes with GAS
> but fails with the MIPSpro n64 assembler to check for >4GB frames?
> I'll try adding that to the testsuite harness.

Would an assembler testcase be OK?  A minimal-ish example is:

	.ent	foo
foo:
	.frame	$fp,5000000000,$31
	.end	foo

assembled with "as -64".  For MIPSpro, you get:

as: Error: foo.s, line 3: Overflow: 5000000000
      .frame $fp,5000000000,$31

but it works OK with binutils.

> If that's overkill, perhaps I could just bypass these tests for n64
> regardless of whether we're using GAS.

Well, the tests only pass for other ABIs because they're a no-op when
sizeof (long) < 8.  If we're going to disable them for n64, we might as
well go the whole hog and disable them for mips-sgi-irix*, or perhaps
for all MIPS targets.

And maybe that's a reasonable thing to do.  The tests did show up a bug
in the MIPS backend ("long" being used instead of HOST_WIDE_INT), but it
seems very unlikely that any real MIPS code would want to use stack
frames >4Gb.

Richard


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14650