This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH]: Move DECL_VALUE_EXPR into side table
On Fri, 2005-06-03 at 17:30 -0700, Mark Mitchell wrote:
> Daniel Berlin wrote:
> > This moves DECL_VALUE_EXPR into a side table, which will save 4/8 bytes
> > on struct tree_decl_with_rtl derived trees.
> >
> > Most things actually just want to know whether DECL_VALUE_EXPR is NULL
> > or not, so i used a bit that is kept in sync with whether there is an
> > entry in the hashtable or not.
> >
> > Bootstrapped and regtested on i686-pc-linux-gnu.
> >
> > Timings show this has no speedup or slowdown on cc1-i-files, tramp3d, or
> > any other cases i threw at it.
> >
> > Okay for mainline?
>
> OK. Do we ever clear out DECL_VALUE_EXPR?
Hmmmmm.
I never actually looked.
> We don't really need it
> after the function has been emitted to the .s file.
The same is true of DECL_DEBUG_EXPR, actually, thinking about it.
> I don't know that
> it would be a win to discard, but it might be.
>
Is it sufficient to test TREE_ASM_WRITTEN, or is there some other flags
i need to test to see whether something has been "really written" to the
output file.