On 18 December 2015 at 12:13, James Greenhalgh <james.greenha...@arm.com> wrote:
> Looking back at the patch just before I hit commit, the 4.9 backport was > a little different (as we still have a CANNOT_CHANGE_MODE_CLASS there). > We can drop the aarch64-protos.h and aarch64.h changes, and we need to > change the sense of the new check, such that we can return true for the > case added by this patch, and false for the limited number of other safe > cases in 4.9. > > Bootstrapped on aarch64-none-linux-gnu. > > OK? > > Thanks, > James > > --- > gcc/ > > 2015-12-14 James Greenhalgh <james.greenha...@arm.com> > > Backport from mainline. > 2015-12-09 James Greenhalgh <james.greenha...@arm.com> > > PR rtl-optimization/67609 > * config/aarch64/aarch64.c > (aarch64_cannot_change_mode_class): Don't permit word_mode > subregs of full vector modes. > * config/aarch64/aarch64.md (aarch64_movdi_<mode>low): Use > zero_extract rather than truncate. > (aarch64_movdi_<mode>high): Likewise. > > gcc/testsuite/ > > 2015-12-14 James Greenhalgh <james.greenha...@arm.com> > > Backport from mainline. > 2015-12-09 James Greenhalgh <james.greenha...@arm.com> > > PR rtl-optimization/67609 > * gcc.dg/torture/pr67609.c: New. > OK /Marcus
