|
From: Julian S. <js...@ac...> - 2013-01-30 08:36:26
|
>> pretty good insn set tests, and I will incrementally verify it using >> GSL over the next week or so. That should shake out any badness. Looks good for x86, amd64, arm, at least. >> in some way, by comparing generated code size before/after for (eg) >> perf/bz2 and perf/ffbench. > Sure, here you go. Basically identical: Good, thanks. >> Also, maybe rename Mxxx to Ixxx or something. > > Yes, definitely. I left it this way on purpose, because it minimises > the changes in that area. So, a post patch should clean this up. > Also, eX -> e1 seems appropriate etc. Yup. ---- So, with all that in place it's possible to get on and explain fun facts to iropt, like ITE(c,x,ITE(!c,y,z)) -> ITE(c,x,y), etc. J |