Re: elf32 support for IA64 (patch 3)

[Jim Wilson <wilson at cygnus dot com>]

+ Mon Oct 30 08:19:10 PST 2000  Steve Ellcey <sje@cup.hp.com>
+ 
+         * tc-ia64.c: (md_shortopts, md_parse_option, md_show_usage)
+         Change M to m for -milp32 or -mlp64 to match gcc.
+         (dot_endp) Modify various calls to change args based on
+         elf32 vs. elf64.
+	  ...

The code you wrote doesn't follow the GNU style conventions.  After all of Kazu
Hirata's formatting fixes, we should try not to mess that up again.  This is
primarily a problem in ia64_hpux_init and ia64_target_format.  There is also
a hunk of code in md_begin.  The indentation needs to be two spaces instead of
tabs.  Open/close braces should be on a line by themselves and indented.
Functions have to have a comment before them explaining what they do, and
what the arguments/return values mean if any.

The TE_HPUX stuff looks a little ugly to me, but it does seem to be the
right gas convention, so I guess I have to live with it.

In md_begin, the local variable ok is only used when TE_HPUX is true, so it
should only be defined when TE_HPUX is true.

There is a comment that needs updating:
    /* ??? This overrides the -M options, but they aren't working anyways.  */
There aren't any -M options anymore, they are -m, and the -m options do work.
By the way, the reason -M options are used here is because this is what is
specified in Intel's assembler documentation.  Unfortunately, -M can't work
as a target dependent option because there is a target independent -M option
that takes precedence.   So using -m instead is reasonable.  Ideally, I think
the code in md_begin setting md.flags ABI should check to see if either ABI
option was used, and if not, then set a default.  The code you added for hpux
looks like it might handle this right.

It looks like both ia64_hpux_init and ia64_target_format don't have to be
HPUX specific.  ia64_hpux_init would be correct for linux also, except for
the default endianness, and there is already a TARGET_BYTES_BIG_ENDIAN macro
for that.  Using this unconditionally would eliminate the need for hpux
specific code in md_begin, and would fix the bug documented by the comment I
mentioned above.  Similarly, ia64_targert_format is correct for linux also.
This would eliminate all of the HPUX specific changes to tc-ia64.c, and thus
would make the changes a lot cleaner in my opinion.  The only HPUX specific
changes we need are a couple of lines in tc-ia64.h to handle the different
default endianness.

I think I can make these cleanups myself, but I have no way to test them
under hpux, as I have no access to ia64-hpux machines.

Jim