This is the mail archive of the
mailing list for the GCC project.
Re: converting wide-int so that it uses its own type rather than hwi.
- From: Richard Biener <richard dot guenther at gmail dot com>
- To: Kenneth Zadeck <zadeck at naturalbridge dot com>
- Cc: Richard Biener <rguenther at suse dot de>, Mike Stump <mikestump at comcast dot net>, Richard Sandiford <r dot sandiford at uk dot ibm dot com>, gcc-patches <gcc-patches at gcc dot gnu dot org>
- Date: Mon, 2 Sep 2013 11:22:46 +0200
- Subject: Re: converting wide-int so that it uses its own type rather than hwi.
- Authentication-results: sourceware.org; auth=none
- References: <5220B866 dot 2070800 at naturalbridge dot com>
On Fri, Aug 30, 2013 at 5:21 PM, Kenneth Zadeck
> on further thought, this is going to be a huge task. The problem is at the
> edges. right now we share the rep of the array with the tree-csts and rtl
> (though the rtl sharing may go away to support canonization). So if the hwi
> rep inside of wide int changes then we will have to implement copying with
> reblocking at that interface or push the type into there and change all of
> the fits_*hwi and to_*hwi interfaces to fit this different type.
Obviously the reps of RTL and tree would need to change as well. Yes,
I am aware of the explicit HWI references in the API - those functions would
be more complicated.
> i think i can get at least as good and perhaps better test coverage by
> changing the rep of hwi for a port. There will also be fallout work
> there, but it will be productive, in that it is just changing code from only
> doing the hwi case to supporting all precisions.
Well, I'm not sure the fallout will be exactly small ;)
What also could improve testing coverage is to skip the fast path, thus
always go the out-of-line path even for len == 1. Not sure if the ool path
is prepared to handle len == 1 though.