This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: CPROP and spec2000 slowdown in mid february
- To: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- Subject: Re: CPROP and spec2000 slowdown in mid february
- From: Joern Rennecke <amylaar at redhat dot com>
- Date: Mon, 4 Jun 2001 19:38:24 +0100 (BST)
- Cc: jh at suse dot cz, gcc at gcc dot gnu dot org
> In my previous experiments this turned out to be neccesary, since
> certain backends require insns to be in non-canonical form. This may
> be considered as bug in the backend tought.
>
> Yes, they are.
Sometimes it is necessary to write insns in a non-canonical form,
because constants are involved, and you need to give the right
predicates and/or constraints for these constants.
> Yes, but that's only a very minor change in the way substitution is
> done. Given how quickly we can nnow allocate an rtx, I think the
> performance difference will be negligable and made up for by removing
> the code that deals with the subst table.
You have to amortise the cost of garbage collection over the rtx allocation.
AFAIK garbage collection is still slower at than just using malloc / free
consistently.