https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
Jakub Jelinek changed:
What|Removed |Added
Status|UNCONFIRMED |RESOLVED
Resolution|---
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #13 from Flössie ---
Thanks for the thorough analysis. We'll discuss in our team how to cope with
it.
Best regards,
Flössie
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #12 from Jakub Jelinek ---
The problem goes away even in the original unmodified dcp.ii with -msse2
-mfpmath=sse or with -ffloat-store, which means most likely the source just
relies on excess precision not happening.
If so, this is
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #11 from Jakub Jelinek ---
Adding __attribute__((optimize (0))) to DCPProfile::apply also does help, which
means the problem is likely in that method or something that is inline into it.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #10 from Flössie ---
> Have you tried -fsanitize=address -fsanitize=undefined on the code?
Not yet on i686 with 7.3. We run RT occasionally with sanitizers when there are
hard to debug problems, but I must admit, not yet on this
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #9 from Jakub Jelinek ---
Leaving -ftree-vectorize out doesn't help either, -fno-inline helps though.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #8 from Jakub Jelinek ---
Commenting out the OpenMP pragmas and/or leaving out -fopenmp for dcp.ii
doesn't help either.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #7 from Jakub Jelinek ---
Removing the 4 __builtin_expect uses doesn't help either though.
--- dcp.ii 2018-02-09 21:21:16.675015748 +0100
+++ dcp.ii 2018-02-09 21:38:18.795348083 +0100
@@ -284532,7 +284532,7 @@ void
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
Jakub Jelinek changed:
What|Removed |Added
CC||jakub at gcc dot gnu.org
--- Comment #6
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #5 from Flössie ---
Created attachment 43278
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43278=edit
Complete -v output
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #4 from Flössie ---
Created attachment 43276
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43276=edit
cc1 command line
Here it is.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
--- Comment #3 from Jakub Jelinek ---
-march=native is ambiguous, please run g++ with the given options with
additional -v and attach what it emits as cc1plus command line.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84077
Richard Biener changed:
What|Removed |Added
Keywords||wrong-code
Target|
13 matches
Mail list logo