This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Make CSE path following use the CFG

Richard Sandiford wrote:
"Steven Bosscher" <> writes:

On 12/11/06, Steven Bosscher <> wrote:

Fortunately, there appears to be a more serious problem than the typo.

And another one, where I intended to traverse the CFG in topological sort, and I actually asked for a DFS sort.

For CC0-targets, we clear prev_insn_cc0 too early now.  I've hacked
together the attached patch, but I can't test it right now (0:00 AM,
work tomorrow, etc.). I wanted to post it anyway because maybe it
helps people forward, and maybe someone can help me by testing this on
a CC0-target;-)

The patch below *should* fix both the CC0 problems and the HPPA/MIPSEL problems. I have built an AVR cross-compiler with this, and I have verified that the test case from PR30149 does not ICE anymore with this patch. I have asked for some assistance to have this patch bootstrapped&tested properly.

FWIW, the original cc0 patch seemed to work great on a locally-modified
m68k tree.  Thanks for the quick fix.


Just to reiterate: This patch seems to work well for mipsel-linux.

I just finished a successful bootstrap (testsuite still in progress), with it. Unless someone can think of a good reason not to, I think Seven should commit the patch.

David Daney

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]