Question on arithmetic overflow detection support in C2 (original) (raw)

Krystal Mok rednaxelafx at gmail.com
Wed Jun 27 08:05:32 PDT 2012

On Wed, Jun 27, 2012 at 9:52 PM, Rémi Forax <forax at univ-mlv.fr> wrote:

On 06/27/2012 03:37 PM, Krystal Mok wrote:

Hi all,

Just FYI, I've posted an updated version of the patch here: https://gist.github.com/db03ab15ef8b76246b84#file checkedaddprototype.ver3.**patch<https://gist.github.com/db03ab15ef8b76246b84#filecheckedaddprototype.ver3.patch> It pretty much implements what John suggested in a previous email (quoted below). This version still suffers from a couple of problem mentioned before: 1. It's emitting a jmpConU instead of a jmpCon during instruction selection. Is there a way to force it use jmpCon here? I'm may be wrong but the idea is to put 0 (or a very small value) to the frequency of the basic block that follow the jump on overflow. No, that's a different problem. The problem here is that the Matcher decided to use an unsigned version of conditional jump, instead of a normal (signed) version. It doesn't really affect the actual emitted code because I've made both cases do the same thing. But it just doesn't feel right to be using the unsigned version here.

2. I had to use a fixed register for the sum projection of CheckedAddI, otherwise I couldn't find a way to specify this projection should use the same register as "dst".

- Kris Rémi -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20120627/386fa7e9/attachment.html

More information about the hotspot-compiler-dev mailing list