2011/9/2 Richard Guenther <[email protected]>: > On Fri, Sep 2, 2011 at 3:45 PM, Ilya Enkovich <[email protected]> wrote: >> 2011/9/2 Richard Guenther <[email protected]>: >>>On Fri, Sep 2, 2011 at 2:52 PM, Uros Bizjak <[email protected]> wrote: >>>> >>>> I assume that you need to split tune attribute to int and FP part to >>>> handle reassociation for other targets, since Atom handles both in the >>>> same way. >>>> >>>> Please also describe function return value in the comment (and perhaps >>>> in documentation, too). >>>> >>>> OK with this addition. >>> >>> Btw, I would expect integer add and integer multiply to have different >>> settings for some targets which would mean splitting this up even >>> further ... >> >> Which tune attributes are meant here? Is it X86_TUNE_REASSOC_* flags >> or new command line param? > > The X86_TUNE_REASSOC_* flags. The setting surely depends on the > number of available execution units and/or whether the instructions > are pipelined or not.
Now I'm not sure it is a good idea to use these tune flags at all because we may need a plenty of them. We have a lot of types including vector ones and a variety of opcodes. Any combination may require delicate tuning. Some sort of separate table would suit better here. Though I doubt we should introduce it right now filled with 1s. Ilya > > Richard. >
