RE: Patch: Omit frame pointers on x86

["Boehm, Hans" <hans dot boehm at hp dot com>]

I'm not sure what impact this has on a build with --enable-gc-debug,
which will try to save a partial stack trace in each allocated object.
I suspect that in this context it usually uses the backtrace() function,
which may just be atrociously slow either way.

Hans

> -----Original Message-----
> From: java-patches-owner@gcc.gnu.org 
> [mailto:java-patches-owner@gcc.gnu.org] On Behalf Of Bryce McKinlay
> Sent: Wednesday, June 01, 2005 3:53 PM
> To: java-patches@gcc.gnu.org
> Subject: Patch: Omit frame pointers on x86
> 
> 
> Previously, stack traces on i386 required a frame pointer to be 
> available as a fallback mechansim when tracing the stack. The 
> new stack 
> trace infrastructure uses DWARF2 unwind info exclusively, so frame 
> pointers are never required on systems that support DWARF2 unwind. 
> Omitting frame pointers frees up a register for general purpose use, 
> which should improve peformance and reduce code size slightly.
> 
> -fno-omit-frame-pointer is actually still the default on 
> 32-bit x86, so 
> this patch will now explicitly specify -fomit-frame-pointer for that 
> platform, except on Windows. On 64-bit x86, no frame pointer is the 
> default, so we don't need to specify anything.
> 
> I'm checking this in to HEAD.
> 
> Bryce
> 
>