Re: [ET Trac] [Einstein Toolkit] #437: Correct autoconfiguration of restrict keyword; add support for builtin_expect
#437: Correct autoconfiguration of restrict keyword; add support for builtin_expect -----------------------------------------------------+---------------------- Reporter: Erik Schnetter <schnetter@…> | Owner: Type: defect | Status: review Priority: minor | Milestone: Component: Cactus | Version: Resolution: | Keywords: -----------------------------------------------------+----------------------
Comment (by barry.wardell):
I also add support for the gcc built-in function "builtin_expect"
This sounds like potentially quite a nice addition. How difficult would it be to take this further and enable the use of profile guided optimization (eg. http://stackoverflow.com/questions/2738835/learning-sample-of-likely- and-unlikely-compiler-hints). Do you think it would be likely to help much?
On Sun, Jun 5, 2011 at 2:59 PM, Einstein Toolkit trac-noreply@einsteintoolkit.org wrote:
This sounds like potentially quite a nice addition. How difficult would it be to take this further and enable the use of profile guided optimization (eg. http://stackoverflow.com/questions/2738835/learning-sample-of-likely- and-unlikely-compiler-hints). Do you think it would be likely to help much?
Profile-guided optimisations need to be handled differently, namely via compiler options, not via source code annotations. One also needs a representative test case. I assume the main problem with this is to ensure that the compiler can make the necessary connections between executable, object files, and source files, which often fails since Cactus places object files in libraries.
Apart from this, PGO would just involve experimenting with compiler options to see whether one can obtain a performance benefit.
-erik
trac@lists.einsteintoolkit.org
-
Einstein Toolkit -
Erik Schnetter