Matt Arsenault
Matt Arsenault
> In the new commit I created a separate function just to handle copy from physical reg to VCC. However, I'm not sure this is an improvement over the previous...
> For the problem we are dealing with here, ie. copy from a phys reg to a VCC reg, if you don't query the membership of the phys reg, how...
> The GPU should use `__builtin_fmin` and `__builtin_fmax`. No, these are different operations. Not sure if we have a wired up builtin for these
This was a recent change in 363b05944f9212511ee6811d0eb1af841c177226
> AFAICT, this PR is specifically about whether fmin and fmax must return a sNaN when an input is a sNaN Not exactly, the result should be a quiet nan....
> Don't do this. It is a trap of code coverage. We need to strike a balance between optimizations, compilation time and maintainability. I think just about any of these...
> @arsenm Is the recommended approach still to have a separate X86O0PreLegalizerCombiner as well? I don't know why -O0 would get any combiner
Is this still relevant?
> 1) if we are dealing with scalable vector constants as there is no immediate representation for a negative scalable zeroinitializer. I thought there was a splat constant for these,...
I thought the way to achieve this in new pass manager was to have an analysis printer pass