On 2015-07-17 07:38, Richard Henderson wrote: > On 07/15/2015 12:03 PM, Aurelien Jarno wrote: > >Now that we have real size changing ops, we don't need to marked high > >bits of the destination as garbage. The goal of the optimizer is to > >predict the value of the temps (and not of the registers) and do > >simplifications when possible. The problem there is therefore not the > >fact that those bits are not counted as garbage, but that a size > >changing op is replaced by a move. > > > >This patch is basically a revert of 24666baf, including the changes that > >have been made since then. > > > >Cc: Paolo Bonzini <pbonz...@redhat.com> > >Cc: Richard Henderson <r...@twiddle.net> > >Signed-off-by: Aurelien Jarno <aurel...@aurel32.net> > > What we're missing here is whether the omitted size changing op is extu or > exts. Mask should be extended to match. Which means keeping most of this > code.
I am afraid your correct. Unfortunately one of my goal is to remove this part in the optimizer, as I need that in a patch series I am preparing. I have also tried to check the temp type directly from the optimizer (it is accessible), but it has some performance impact. Propagating the extu/exts as real opcode means propagating the information about size changing up to the optimizer or the register allocator, without having to recreate it from other available information. For now I do wonder if we shouldn't get the size changing extu/exts mandatory instead of reusing the 64-bit only version. This doesn't change the generated code, at least on x86. -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://www.aurel32.net
