[Bug target/15968] [3.5 Regression] gcc.c-torture/execute/va-arg-22.c fails on hppa64-hp-hpux11.11

["dave at hiauly1 dot hia dot nrc dot ca" <gcc-bugzilla at gcc dot gnu dot org>]

------- Additional Comments From dave at hiauly1 dot hia dot nrc dot ca  2004-06-15 00:16 -------
Subject: Re:  [3.5 Regression] gcc.c-torture/execute/va-arg-

> Want to take a crack at implementing TARGET_GIMPLIFY_VA_ARG_EXPR, to see if
> that
> fixes it?

I'm working on a small contract at the moment but after it completes
this is definitely something that should be looked at, even if it's
only for the learning experience.

I think the current problem is with va_start.  I see the following
error message with the testscase and a simplified version xxx.c:

# ../xgcc -B../ -O1 -s -o xyz.s xxx.c
xxx.c: In function `foo':
xxx.c:37: warning: second parameter of `va_start' not last named argument

xxx.c contains just D(0) and D(1) from the original test.  Looking at
the assembler output, I see that we are loading the high order byte
from the argument passed for D(0) instead of that for D(1).

I still don't know why this is happening.  The same code works in 3.3
and 3.4, and for some reason nearly identical code seems to work on the
32-bit port.  In both cases, empty structs are passed by reference.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15968