This is the mail archive of the
mailing list for the GCC project.
Re: [RFC] Run fast RTL DCE pass at -O0
- From: Richard Guenther <richard dot guenther at gmail dot com>
- To: Eric Botcazou <ebotcazou at adacore dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Sun, 19 Apr 2009 09:41:01 +0200
- Subject: Re: [RFC] Run fast RTL DCE pass at -O0
- References: <email@example.com>
On Sat, Apr 18, 2009 at 5:22 PM, Eric Botcazou <firstname.lastname@example.org> wrote:
> flow.c used to run (integrated) DCE at -O0, this was ditched with the switch
> to full DF. ?As a result, we measured significant code size increases at -O0,
> for example 10 MB on a big Ada application (85 MB).
> This patch adds a fast RTL DCE pass at -O0; in our 4.3-based compiler, this
> entirely fixed the code size regression with no compilation time increases.
> Tested (GCC, GDB) on i586-suse-linux. ?Thoughts?
Is running SSA DCE maybe cheaper? Doesn't this affect the ability to debug?
> 2009-04-18 ?Eric Botcazou ?<email@example.com>
> ? ? ? ?* tree-pass.h (pass_fast_rtl_dce_no_opt): Declare.
> ? ? ? ?* dce.c (insn_is_nop): New function.
> ? ? ? ?(delete_unmarked_insns): Do not delete NOPs at -O0.
> ? ? ? ?(gate_fast_dce_no_opt): New gate function.
> ? ? ? ?(pass_fast_rtl_dce_no_opt): New pass.
> ? ? ? ?* passes.c (init_optimization_passes): Register it.
> Eric Botcazou