https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102294
--- Comment #24 from Mateusz Guzik <mjguzik at gmail dot com> ---
I got the thing compiled against top of git.
with this as a testcase:
void zero(char *buf)
{
__builtin_memset(buf, 0, SIZE);
}
compiled like so:
./xgcc -O2 -DSIZE=128 -mno-sse -c ~/zero.c && objdump --disassemble=zero zero.o
The compiler emits completely unrolled stores for sizes up to 128, which raises
an eye-brow but is perhaps fine.
However, for 129 and higher I see the code going back to rep, which is not the
expected state.
The expected behavior is unrolled loops, 32 bytes per iteration, of up to 256.
