https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85190
--- Comment #4 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> --- > --- Comment #3 from amker at gcc dot gnu.org --- > (In reply to Rainer Orth from comment #0) [...] > Hmm, what options did you use? Nothing special in the Solaris/x86 (i386-pc-solaris2.11) case: $ configure --with-as=/usr/ccs/bin/as --without-gnu-as --enable-languages=all,ada,obj-c++,go > Then compile the test case with below options: > ./powerpc64le-unknown-linux-gnu-gcc -S -O2 -ftree-vectorize pr81196.c -o > pr81196.S -fdump-tree-ifcvt -fdump-tree-vect-details -flto -ffat-lto-objects > -mcpu=power6/7/8 > > power6/power7 would fail, but on power8 loops are vectorized. > > Problem is, on power6/power7, the case would fail even for r258856 as you > bisected. I don't have access to power systems, but for i386 it's $ gcc pr81196.c -mclear-hwcap -msse2 -ftree-vectorize -fno-vect-cost-model -fno-common -O2 -fdump-tree-vect-details or $ gcc pr81196.c -m64 -mclear-hwcap -msse2 -ftree-vectorize -fno-vect-cost-model -fno-common -O2 -fdump-tree-vect-details for amd64.
