https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63271
Richard Biener <rguenth at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |NEW Last reconfirmed| |2014-09-16 Blocks| |53947 Ever confirmed|0 |1 --- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> --- basic-block SLP misses treating a vector constructor as source to search for an SLP opportunity. And yes, it has issues with non-matching SLPs if adds of zero / multiplications of one could be inserted to make the SLP match. Both are missed optimization opportunities there. The SLP vectorizer also requires loads to end the SLP chain which isn't necessary either. We may have duplicate bugreports about each of the three issues (and they should be addressed separately with separate testcases).