On Fri, Aug 13, 2021 at 12:08 PM Segher Boessenkool <seg...@kernel.crashing.org> wrote: > > On Fri, Aug 13, 2021 at 11:15:21AM -0400, David Edelsohn wrote: > > On Fri, Aug 13, 2021 at 10:49 AM Segher Boessenkool > > <seg...@kernel.crashing.org> wrote: > > > > > > On Fri, Aug 13, 2021 at 12:14:14AM -0400, Michael Meissner wrote: > > > > I noticed that the xxeval built-in function used the > > > > altivec_register_operand > > > > predicate. Since it takes vsx registers, this might force the register > > > > allocate to issue a move when it could use a traditional floating point > > > > register. This patch fixes that. > > > > > > Why register_operand instead of gpc_reg_operand? The former allows > > > subregs of memory, likely not what you want here (and not in other > > > rs6000 pattern that currently use it, either). > > > > Because it's consistent with the other patterns. > > Not with the vast majority of other patterns, no.
For vector operands in altivec.md, if the predicates do not use altivec_register_operand, they use register_operand, not gpc_reg_operand. Mike's change is consistent with the rest of the altivec.md file. Maybe the predicates for those patterns should be changed. Maybe that change would be an improvement. Your statement about the vast majority of patterns is wrong with respect to altivec.md. Thanks, David
