> From: Andi Kleen <[email protected]> > Sent: Friday, August 29, 2025 5:13 AM > > > > > > with GCC configured with > > > > > > ../../gcc/configure --prefix=/export/users3/haochenj/src/gcc- > bisect/master/master/r16-3364/usr --enable-clocale=gnu --with-system-zlib - > -with-demangler-in-ld --with-fpmath=sse --enable-languages=c,c++,fortran -- > enable-cet --without-isl --enable-libmpx x86_64-linux --disable-bootstrap > > > > > > To reproduce: > > > > > > $ cd {build_dir}/gcc && make check > RUNTESTFLAGS="i386.exp=gcc.target/i386/shift-gf2p8affine-5.c -- > target_board='unix{-m32\ -march=cascadelake}'" > > > Ah never mind it's there. The test breaks if you override > -march=... because it changes the vector cost model and then > the number of generated instructions is different. > I've seen that with other tuning options too. > > I'll see how to fix that. Perhaps it just needs to specify its own > -march=generic which might take preference. >
It seems to me caused by more 256 bit vectorization enabled. One of the simplest way is to strict to prefer-vector-width=128 and adjust the testcase. But I am not sure if it meets the original need for the testcase. Thx, Haochen.
