[PATCH] D65737: [InstCombine] mempcpy(d,s,n) to memcpy(d,s,n) + n

[Dávid Bolvanský via Phabricator via llvm-commits]

xbolva00 added a comment.

mempcpy is quite rare, so hard to measure it on codebases.

I wrote small C “benchmarks” and no perf changes (backend produced memcpy/inlined anyway).

Another motivation is our recent attribute annotation work so we would get all good things from memcpy. Otherwise, I would have to annotate it individually.

It seems (some gcc discussion) only x86-64 has fast mempcpy libc impl. - so I really think backend should create mempcpy under target hook.

Not sure, maybe @craig.topper has more info about mempcpy perf on x86-64.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D65737/new/

https://reviews.llvm.org/D65737