This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [Patch] Factor self-referential size trees into functions

> This should be separately tested/committed.  The PROCESS_ARG
> change, build3_stat and the process_call_operands change are ok.

OK, will do.

> Hm, do you really need to avoid unsharing trees or is this an
> optimization?

This is required because of some complex mechanism in Gigi.

> I don't like the tree-inline.c dependency you create for tree.c here - I
> don't have really a better place, but I wonder if we can move some of this
> stuff into gigi via a langhook (all the substitute-in-placeholder stuff)?

You can do a grep "tree-inline.h" and see that many files already depend on 
tree-inline.c.  As for the PLACEHOLDER machinery, given that the node is 
generic, I don't think that it would be logical to move the machinery 
handling it in Gigi.  I don't know if privatizing the node itself is doable.

FWIW the size of cc1 on x86 grows from 54015658 to 54070225 bytes (+0.1%).

> I guess if we would do gimplification unit-at-a-time then the
> special-handling of size functions could go.

Yes, probably.

Eric Botcazou

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]