On Fri, Apr 07, 2017 at 02:38:50PM -0500, Aaron Sawdey wrote: > Turns out we get passed const -1 for the length arg from this code. > ROUND_UP adds load_mode_size to that resulting in a small positive > number, hilarity ensues.
Glad you liked it as well ;-) > Fixed by computing a sensible limit and using > IN_RANGE instead, which won't overflow in this way. > > OK for trunk if bootstrap/regtest in progress passes? Yes, looks good. Thanks, Segher > 2017-04-07 Aaron Sawdey <acsaw...@linux.vnet.ibm.com> > > PR target/80358 > * config/rs6000/rs6000.c (expand_block_compare): Fix boundary check. > > Index: gcc/config/rs6000/rs6000.c > =================================================================== > --- gcc/config/rs6000/rs6000.c (revision 246771) > +++ gcc/config/rs6000/rs6000.c (working copy) > @@ -19672,8 +19672,9 @@ > unsigned int load_mode_size = GET_MODE_SIZE (load_mode); > > /* We don't want to generate too much code. */ > - if (ROUND_UP (bytes, load_mode_size) / load_mode_size > - > (unsigned HOST_WIDE_INT) rs6000_block_compare_inline_limit) > + unsigned HOST_WIDE_INT max_bytes = > + load_mode_size * (unsigned HOST_WIDE_INT) > rs6000_block_compare_inline_limit; > + if (!IN_RANGE (bytes, 1, max_bytes)) > return false; > > bool generate_6432_conversion = false;
