[Bug tree-optimization/47770] wrong code -O2 -ftree-loop-if-convert-stores -fno-tree-reassoc

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47770

Andrew Pinski  changed:

   What|Removed |Added

   Target Milestone|--- |4.6.0

[Bug tree-optimization/47770] wrong code -O2 -ftree-loop-if-convert-stores -fno-tree-reassoc

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47770

Richard Biener  changed:

   What|Removed |Added

 Status|ASSIGNED|RESOLVED
  Known to work||4.6.0
 Resolution|--- |FIXED
  Known to fail|4.6.0   |

--- Comment #5 from Richard Biener  ---
Original testcase fixed during development of GCC 4.6.0.

[Bug tree-optimization/47770] wrong code -O2 -ftree-loop-if-convert-stores -fno-tree-reassoc

[egallager at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47770

Eric Gallager  changed:

   What|Removed |Added

 CC||irar at il dot ibm.com,
   ||rguenth at gcc dot gnu.org

--- Comment #4 from Eric Gallager  ---
(In reply to Zdenek Sojka from comment #1)
> "-ftree-loop-if-convert -fno-tree-reassoc -fno-tree-vectorize" gives about
> 316 exec failures on current trunk. I should try the fix for PR46029 from
> http://gcc.gnu.org/ml/gcc-patches/2010-11/msg01613.html

cc-ing Richard from that thread

(In reply to Sebastian Pop from comment #2)
> I do not think that 0001-Fix-PR46029-reimplement-if-convert-stores.patch
> is the right way to fix this: I discussed with Ira Rosen

cc-ing Ira

[Bug tree-optimization/47770] wrong code -O2 -ftree-loop-if-convert-stores -fno-tree-reassoc

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47770

--- Comment #3 from Zdenek Sojka zsojka at seznam dot cz 2011-02-24 02:20:56 
UTC ---
It seems most of the exec failures disappeared between r170436 and r170443 -
that is, probably by the fix for PR47849.

Running testsuite (languages=c,c++,lto,fortran) with 
RUNTESTFLAGS=--target_board=unix/-ftree-loop-if-convert-stores/-fno-tree-reassoc/-fno-tree-vectorize
gives only one ICE - PR47738 and two exec failures:

ext/pb_ds/example/tree_intervals.cc (PR46029)
gfortran.dg/allocate_deferred_char_scalar_1.f03 (fails even with no
optimisation flags)

Also, the attachment in comment #0 doesn't fail anymore.

[Bug tree-optimization/47770] wrong code -O2 -ftree-loop-if-convert-stores -fno-tree-reassoc

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47770

--- Comment #1 from Zdenek Sojka zsojka at seznam dot cz 2011-02-16 22:36:17 
UTC ---
-ftree-loop-if-convert -fno-tree-reassoc -fno-tree-vectorize gives about 316
exec failures on current trunk. I should try the fix for PR46029 from
http://gcc.gnu.org/ml/gcc-patches/2010-11/msg01613.html, but I am failing to
apply it (I can't find the revision it applies to, and the mentioned
0001-Fix-PR46029-reimplement-if-convert-stores.patch in general).

[Bug tree-optimization/47770] wrong code -O2 -ftree-loop-if-convert-stores -fno-tree-reassoc

[spop at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47770

Sebastian Pop spop at gcc dot gnu.org changed:

   What|Removed |Added

 Status|UNCONFIRMED |ASSIGNED
   Last reconfirmed||2011.02.17 00:36:38
 AssignedTo|unassigned at gcc dot   |spop at gcc dot gnu.org
   |gnu.org |
 Ever Confirmed|0   |1

--- Comment #2 from Sebastian Pop spop at gcc dot gnu.org 2011-02-17 00:36:38 
UTC ---
I do not think that 0001-Fix-PR46029-reimplement-if-convert-stores.patch
is the right way to fix this: I discussed with Ira Rosen about the
vectorization
of the code produced by this patch and we ended up realizing that the
vectorizer would have to answer the same hard questions (memory accesses
out of bounds and writes into read-only memory) that are now analyzed in
-ftree-loop-if-convert-stores.

I will look at how to fix these miscompiles.