This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Mips profiling
Is this the one you're talking about? I had already seen your patch to take
the .set noreorder off - I guess this one went in at the same time. Now I'm
wondering if your mcount works the same way - we have a architecture
dependent _mcount() stub for x86, mips, sh, arm and ppc which basically
calculates the arguments and calls the generic mcount(). Thank you for
explaining the calling conventions to mcount earlier - that might help me
iron out some of the issues I'm having with getting sh working as well.
I've only been looking at profiling (and mips/sh4 assembly) for a week or
two now so I'm still mostly wallowing in ignorance.
Incidentally, for anyone interested, I've got sh4 mcount profiling working
in a limited way - I'll be posting the diffs here at some point when I get
the kinks ironed out. (actually, I think most of the kinks are in our
mcount, not gcc, but that's another story ;-).
cheers,
Kris
> RCS file: /cvs/gcc/gcc/gcc/config/mips/linux.h,v
> retrieving revision 1.33
> diff -u -r1.33 linux.h
> --- linux.h 2001/11/19 11:28:52 1.33
> +++ linux.h 2002/01/17 00:59:53
> @@ -283,3 +283,9 @@
> /* Tell function_prologue in mips.c that we have already output the
.ent/.end
> pseudo-ops. */
> #define FUNCTION_NAME_ALREADY_DECLARED
> +
> +/* The glibc _mcount stub will save $v0 for us. Don't mess with saving
> + it, since ASM_OUTPUT_REG_PUSH/ASM_OUTPUT_REG_POP do not work in the
> + presence of $gp-relative calls. */
> +#undef ASM_OUTPUT_REG_PUSH
> +#undef ASM_OUTPUT_REG_POP
> Index: mips.h
----- Original Message -----
From: "Daniel Jacobowitz" <drow@mvista.com>
To: "Kris Warkentin" <kewarken@qnx.com>
Cc: <gcc@gcc.gnu.org>
Sent: Thursday, March 21, 2002 5:06 PM
Subject: Re: Mips profiling
> On Thu, Mar 21, 2002 at 05:08:19PM -0500, Kris Warkentin wrote:
> > Well, that possibly explains the ra issue...I'll look at our code to see
> > what happens to it. Seems odd to me that it fails on optimized code
only.
> > We're also seeing some problems with bus errors when we compile C++ with
> > profiling but I haven't followed that up yet.
>
> Search the gcc-patches (I think) archives for posts about MIPS
> profiling. It's probably the PUSH/POP macros (which can destroy $gp)
> causing this. I fixed it for Linux several months ago.
>
> --
> Daniel Jacobowitz Carnegie Mellon University
> MontaVista Software Debian GNU/Linux Developer
>