On Thu, May 09, 2019 at 09:59:55AM -0600, Jeff Law wrote: > THe main reason it's discouraged is because some targets have insns > where the count would be truncated and others where it would not. So > for example normal shifts might truncate, but vector shifts might or > (mips) or shifts might truncate but bit tests do not (x86).
And don't forget about zero_extract, on targets that are unforunate enough to have that. > I don't know enough about the s390 architecture to know if there's any > corner cases. You'd have to look at ever pattern in your machine > description with a shift and verify that it's going to DTRT if the count > hasn't been truncated. Yeah. > It would really help if you could provide testcases which show the > suboptimal code and any analysis you've done. I'd love to see this, too! Segher
