This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PR47036] sel-sched: don't attempt to move up unconditional jumps
On 12/22/2010 04:02 PM, Andrey Belevantsev wrote:
On 22.12.2010 20:57, Alexander Monakov wrote:
Hi,
This patch drops bits of support for moving up unconditional jumps.
The rest
of the scheduler does not handle it anyway, and most of the time
dependencies
would prohibit such motion (in the testcase it's possible because
other insns
are nops).
While I'm happy with this patch, I would like to note that I'm rather
tired of fixing various corner cases that are only seen in the
scheduler when running it with -O0. I do believe that we need to
force cleanup_cfg and DCE when not optimizing and stop caring.
I don't think it is a good solution, Andrey. It is practically the same
solution as preventing insn scheduling for -O0. While the passes are
separate and have options to switch them on/off, the pass should work
independently of the other passes work.
I've verified that the patch does not affect code generation on
several cc1 .i
files at -O3 (targeting ia64). Bootstrapped on x86-64-linux, ia64-linux
(testsuite running). OK for trunk?
I would also suggest to go further and check codegen differences on
SPEC2k.
Yes, it would be interesting to see differences.