http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53616
Vladimir Yakovlev <vbyakovl23 at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |vbyakovl23 at gmail dot com
--- Comment #11 from Vladimir Yakovlev <vbyakovl23 at gmail dot com> 2012-07-23
12:53:13 UTC ---
Miscompare in 416 .games is caused by a wrong transformation of a loop in file
grd2b.f, lines 113-121.
DO 110 M=1,3
P12(M,1)= C(M,IAT)
P12(M,2)= C(M,JAT)
P12(M,3)= P12(M,2)-P12(M,1)
R12= R12+P12(M,3)*P12(M,3)
P34(M,1)= C(M,KAT)
P34(M,2)= C(M,LAT)
P34(M,3)= P34(M,2)-P34(M,1)
110 R34= R34+P34(M,3)*P34(M,3)
After transformation we have
P12(:,1) = C(:,IAT)
P12(:,2) = C(:,jAT)
DO 110 M=1,3
P12(M,3)= P12(M,2)-P12(M,1)
R12= R12+P12(M,3)*P12(M,3)
P34(M,3)= P34(M,2)-P34(M,1)
110 R34= R34+P34(M,3)*P34(M,3)
P34(:,1) = C(:,KAT)
P34(:,2) = C(:,LAT)
That is we changed order of operators in the loop. Right transformation should
be
P12(:,1) = C(:,IAT)
P12(:,2) = C(:,jAT)
DO 110 M=1,3
P12(M,3)= P12(M,2)-P12(M,1)
110 R12= R12+P12(M,3)*P12(M,3)
P34(:,1) = C(:,KAT)
P34(:,2) = C(:,LAT)
DO 111 M=1,3
P34(M,3)= P34(M,2)-P34(M,1)
111 R34= R34+P34(M,3)*P34(M,3)
I attached a reduced test case and dumps with and without transformations.
Command line to compile is
gfortran m.f t.f -O3
The result of run is differed from a result of code compiled with -O0 opt
level.
I used compiler
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc/configure --with-arch=corei7 --with-cpu=corei7
--enable-clocale=gnu --with-system-zlib --enable-shared --with-demangler-in-ld
--enable-cloog-backend=isl --with-fpmath=sse --enable-languages=c,c++,fortran
--enable-bootstrap=no
Thread model: posix
gcc version 4.8.0 20120606 (experimental) (GCC)
