This is the mail archive of the
mailing list for the GCC project.
Re: [LTO] patch: new CALL_EXPR abstractions in builtins.c
- From: Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>
- To: Roger Sayle <roger at eyesopen dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Wed, 28 Jun 2006 20:52:37 +0200
- Subject: Re: [LTO] patch: new CALL_EXPR abstractions in builtins.c
- References: <Pine.LNX.firstname.lastname@example.org>
[huge snip](because for a full reply I need to do some experiments and read some code)
Didn't I make a good catch in the review/suggestion?From the incremental point of view, sure you did.
On the other hand, I was running (too?) fast and trying to see the
picture at the end of Sandra's work. My understanding was that instead
of speculatively building an arglist and a CALL_EXPR, she'd end up
building the CALL_EXPR directly with the arglist embedded into it: thus
the break-even analysis in my first reply.
Of course, maybe this understanding is true, maybe it's not. Sandra
obviously knows better what she is doing, so if she accepted your
suggestion, I guess it also fits in her big picture. However, I do
think that it would have been better to present a more complete plan of
the work, that experienced people in this area of the compiler could
evaluate, with approximate estimation of the memory usage at each step.
Note that I'm not at all dismissing CodeSourcery's attempt to clean up
TREE_CHAIN etc. as part of the LTO effort.