[Bug c++/82505] g++ -O3 -funroll-loops generates weird code

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82505

--- Comment #6 from Jonathan Wakely  ---
Ah no, because we have global pointers here, with potentially unknowable
aliasing, as Alexander said.

[Bug c++/82505] g++ -O3 -funroll-loops generates weird code

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82505

--- Comment #5 from Jonathan Wakely  ---
dup of PR 82394?

[Bug c++/82505] g++ -O3 -funroll-loops generates weird code

[amonakov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82505

Alexander Monakov  changed:

   What|Removed |Added

 Status|UNCONFIRMED |RESOLVED
 CC||amonakov at gcc dot gnu.org
 Resolution|--- |INVALID

--- Comment #4 from Alexander Monakov  ---
t2 needs more code simply because it does more work: in C you can read
representations of objects byte-by-byte via 'char *' pointers. Consider what
happens if src is initialized like 'src = (void *)' prior to invoking t2
(then the loop in t2 modifies src on each iteration, and also reads a byte of
its representation).

[Bug c++/82505] g++ -O3 -funroll-loops generates weird code

[barto at cambridgesemantics dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82505

--- Comment #3 from David Barto  ---
I understand that it can't optimize much, my major concern is that the pointer
version of the code is so much worse than the index version of the code.

[Bug c++/82505] g++ -O3 -funroll-loops generates weird code

[glisse at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82505

--- Comment #2 from Marc Glisse  ---
dest/src might alias anything (even themselves), so the compiler can't really
optimize much.

[Bug c++/82505] g++ -O3 -funroll-loops generates weird code

[barto at cambridgesemantics dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82505

--- Comment #1 from David Barto  ---
This is gcc-mp-7 pre-built object from MacPorts.
Should have mentioned that.