On Wed, 1 Feb 2017, Jakub Jelinek wrote: > > Yes; I imagine the approach taken in patch 2/5 can be extended to achieve > > this. > > That is, instead of just storing a flag 'bool in_simtreg' in struct loop, > > store > > pointers to corresponding SIMT_ENTER/EXIT gimple statements, use a similar > > upwards walk on loop tree to discover if we're inlining into a SIMT region, > > and > > if yes, adjust their argument lists. Does this sound ok? > > I'd prefer the abnormal edges and flags on the vars, if it can work, but > won't fight for that hard.
Sorry, I'm uncomfortable with that because introducing abnormal edges seems like a big hammer, e.g. it constrains non-privatized variables too. And they don't seem to be rigorously defined, so to me that leaves some uncertainty -- as opposed to an approach I seek that makes constraints obvious in the IR. > That said, I think pointers to gimple stmts in struct loop or something > similar is problematic, you'd need to adjust those whenever something would > remove those stmts, or e.g. duplicate the loop and stmts, handle those > during inlining (if you inline some function with SIMT_ENTER/EXIT in them) > etc. Trying to find those stmts on preheader or in exit block from the > marked loop might be easier. Ah, sorry, so I'd need to keep the bool flag, and for SIMT_ENTER walk the dominator tree upwards, scanning each bb until I find it (and likewise on postdominator tree for SIMT_EXIT). Alternatively, if simduid is already properly remapped, we could assign to it when calling SIMT_ENTER, and then just look up its defining statement? Alexander
