Re: [C++ PATCH] Serious time and code quality regression withinlining on trees

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Jason" == Jason Merrill <jason@redhat.com> writes:

    >> If I understand correctly, this case occurs because of our
    >> destructive modification of functions.  When we output a
    >> function, we actually transform it to contain the inlining of
    >> the functions it calls, and then we forget about the inlining
    >> we've already done...

    Jason> FWIW, the RTL inliner's solution to this is to disable
    Jason> inlining once the function has been emitted.

True.

Do you think that's a better solution?  To me, it seems simpler, but
inferior.  I always figured that was just because emitting RTL
actually made it hard to inline the function again; a problem we don't
have on trees.

Thanks,

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com