[LLVMdev] Tail calls (TCO) in PNaCL | PNaCl Bitcode reference manual
JF Bastien
jfb at google.com
Fri Aug 2 07:12:29 PDT 2013
> That'd be ideal, yes. When the client runs llc, is it going to run it
> at a fixed or client determined optimization level or one specified by
> metadata that follows the code?
You mean for PNaCl? The developer can run LLVM's optimizations on his
machine and generate a pexe. The pexe and manifest are then served to
users, and this manifest specifies which optimization level to use to
translate the pexe to a nexe on user's devices.
You can therefore choose between SelectionDAG and FastISel (fast code or
fast translation) on the user's device without giving up all performance
because the non-target-specific optimizations have been executed already.
This is something that we've benchmarked fairly closely and plan to
improve.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130802/b90205fb/attachment.html>
More information about the llvm-dev
mailing list