chicken-hackers
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Chicken-hackers] merging the "argvector" changes


From: Ivan Raikov
Subject: Re: [Chicken-hackers] merging the "argvector" changes
Date: Fri, 28 Aug 2015 09:51:45 -0700

  I agree that the argvector patch should be merged, as it is a change
that allows Real Work to be done with CHICKEN.
It would be nice to have performance comparisons of course, and I
would be particularly curious to benchmark CPS-style tree traversal
and combinator parsing code. I  think having the patch merged would
make it easier to test and identify potential performance problems.

  -Ivan

On Fri, Aug 28, 2015 at 3:32 AM,  <address@hidden> wrote:
> Hello!
>
> Peter asked me to write a few words about the status of the "argvector" 
> related
> changes (passing arguments to CPS-procedures in a stack-allocated vector
> instead of doing it directly).
>
> As I already wrote, the initial reason for this change was the new ARM64 iOS
> ABI, which rigorously punishes the undefined behaviour we exploit in the way
> CHICKEN translates Scheme to C. At bevuta, we have a large mobile App that
> needed to be ported to arm64 (Apple requires this now for all updates in the
> App Store and of course for all newly submitted apps). The changes in the
> "argvector" branch of the core repo have been used for the new (universal)
> 32/64-bit build and seem to work flawlessly. We are not sure about any
> performance differences yet - one tester say it actually got faster, but that
> is likely to be caused by the 64-bit environment and not by the
> argvector-related differences in the generated code, I guess.
>
> Mario was so kind to run Salmonella on the argvector branch and the results 
> are
> actually quite good: 50 or so eggs break vs. 16 previously, and nearly all of
> the breakages are caused by the new signature of "C_apply" (was "C_do_apply")
> and "C_values", but these are very easy to fix.
>
> So we think this should be merged. Peter already has prepared patches for the
> "numbers" egg and the chicken-5 core.
>
> Opinions and comments are welcome.
>
>
> felix
>
>
> _______________________________________________
> Chicken-hackers mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/chicken-hackers



reply via email to

[Prev in Thread] Current Thread [Next in Thread]