Re: Add an alternative vector loop iv mechanism

2017-10-19 Thread Richard Biener
On Thu, Oct 19, 2017 at 10:48 AM, Richard Sandiford wrote: > Richard Biener writes: >> On Thu, Oct 19, 2017 at 12:28 AM, Richard Sandiford >> wrote: >>> Richard Biener writes: On Fri, Oct 13, 2017 at 4:10 PM, Richard Sandiford wrote: > Normally we adjust the vector loop so that i

Re: Add an alternative vector loop iv mechanism

2017-10-19 Thread Richard Sandiford
Richard Biener writes: > On Thu, Oct 19, 2017 at 12:28 AM, Richard Sandiford > wrote: >> Richard Biener writes: >>> On Fri, Oct 13, 2017 at 4:10 PM, Richard Sandiford >>> wrote: Normally we adjust the vector loop so that it iterates: (original number of scalar iterations - num

Re: Add an alternative vector loop iv mechanism

2017-10-19 Thread Richard Biener
On Thu, Oct 19, 2017 at 12:28 AM, Richard Sandiford wrote: > Richard Biener writes: >> On Fri, Oct 13, 2017 at 4:10 PM, Richard Sandiford >> wrote: >>> Normally we adjust the vector loop so that it iterates: >>> >>>(original number of scalar iterations - number of peels) / VF >>> >>> times,

Re: Add an alternative vector loop iv mechanism

2017-10-18 Thread Richard Sandiford
Richard Biener writes: > On Fri, Oct 13, 2017 at 4:10 PM, Richard Sandiford > wrote: >> Normally we adjust the vector loop so that it iterates: >> >>(original number of scalar iterations - number of peels) / VF >> >> times, enforcing this using an IV that starts at zero and increments >> by o

Re: Add an alternative vector loop iv mechanism

2017-10-18 Thread Richard Biener
On Fri, Oct 13, 2017 at 4:10 PM, Richard Sandiford wrote: > Normally we adjust the vector loop so that it iterates: > >(original number of scalar iterations - number of peels) / VF > > times, enforcing this using an IV that starts at zero and increments > by one each iteration. However, divid

Add an alternative vector loop iv mechanism

2017-10-13 Thread Richard Sandiford
Normally we adjust the vector loop so that it iterates: (original number of scalar iterations - number of peels) / VF times, enforcing this using an IV that starts at zero and increments by one each iteration. However, dividing by VF would be expensive for variable VF, so this patch adds an a