On Wed, Sep 5, 2012 at 5:34 PM, Rik <address@hidden> wrote:
> On 09/05/2012 11:42 AM, Michael D Godfrey wrote:
>
>
> 9/5/12
>
> Michael,
>
> This is something for the configuration and build system and I believe it
> is worth doing. Eventually, everyone will want to have code acceleration,
> but as it is still fairly experimental we should have an easy way to
> disable it with a '--disable-jit' option.
>
> --Rik
>
> If it is feasible, --disable-jit would be very useful, not just while JIT is
> still
> experimental. It would make it quite a lot easier to measure how much good
> the JIT is doing. If --disable-jit is feasible, it seems to me that an
> Octave
> command to turn it on/off could (should) also be implemented.
>
> Will this work?
>
> 9/5/12
>
> Michael,
>
> I just committed a change
> (http://hg.savannah.gnu.org/hgweb/octave/rev/df7ab06ff541) which, among
> other things, adds a '--disable-jit' option to configure. This is a
> configure time decision. Eventually we could try and build a run-time
> capability for turning JIT on or off, but that is another project down the
> road.
>
> --Rik
>
>
> Michael
>
>
I think the idea was to include JIT in the next release as an
experimental feature disabled by default. Currently, if you do not
pass LLVM_CONFIG, JIT is automatically disabled. Maybe instead we
should have a --enable-jit option and automatically try to find
llvm-config if it is specified?
I would like to get a few bugs fixed before the release. Mainly, the
32 bit windows issue and JIT skipping breakpoints. Adding a runtime
option to enable/disable JIT is very easy. I'm not sure if it is
necessary though. Ideally, we should fix JIT related bugs.
--
Max Brister