Hi Sven, Please see CallStubCompiler::CompileMathAbsCall() in both stub-cache-ia32/x64.cc file. The branchless abs is used there originally.
After discussion, I think there is no need to change current patch. HI Jakob, Would you please have a review. Thanks a lot~ On Wed, Aug 7, 2013 at 2:56 PM, Sven Panne <[email protected]> wrote: > On Wed, Aug 7, 2013 at 2:50 AM, Weiliang Lin <[email protected]>wrote: > >> Yes, I agree with you, especially on IA32 which does not have much >> registers. >> > > That platform is my main concern, too, especially because it is still the > most important platform we have. > > >> So could I have a conclusion here. The branchless code is suitable for >> the stub call where there is no register pressure, but not suitable for >> lithium. Right? >> > > I jumped into the discussion a little bit late, so I don't know exactly > which stub is meant, but in general, yes: For the Lithium code generators > using an additional register just to avoid a tiny branch is a no-go, but if > you have a spare register in some stub you can of course use it. Adding a > comment plus a URL describing the bit twiddling hack would be good, too, > because if you haven't seen such code before, it is totally unclear what it > is doing and how. > > -- > -- > v8-dev mailing list > [email protected] > http://groups.google.com/group/v8-dev > --- > You received this message because you are subscribed to the Google Groups > "v8-dev" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/groups/opt_out. > > > -- Thanks -Weiliang -- -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev --- You received this message because you are subscribed to the Google Groups "v8-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/groups/opt_out.
