[CakeML-dev] Evaluating the compiler in the logic
Ramana.Kumar at cl.cam.ac.uk
Wed Feb 15 23:01:52 UTC 2017
The parmap code is already generic and in miscLib. I think it would be good
to define a compset fragment for each phase of the backend, e.g.
compiler/backend/stack_to_labComputeLib.sml, so we can proceed
incrementally in breaking up the giant compilerComputeLib.sml while also
making the individual libs smarter in how they evaluate the compiler (using
custom conversions and parallelism for example).
On 16 Feb 2017 9:08 am, "Yong Kiam" <tanyongkiam at gmail.com> wrote:
I'm now done with the env-refactor translation (at least, until parsing
gets fixed). I'll try and pull out your parallel code tomorrow if you
haven't already got to it by then.
On Wed, Feb 15, 2017 at 12:53 AM, Yong Kiam <tanyongkiam at gmail.com> wrote:
> Yeah, I wasn't planning on doing the move until env-refactor was done.
> Also, I was going to say that we definitely should factor out the parallel
> evaluation from the bootstrap to use here instead of my existing
> compilation machinery.
> It takes upwards of 10mins to compile a single program with the basis
> I'm still on the parsing PR for HOL, but I expect to finish fixing the
> translation tomorrow, then I can switch to basis and help with the move and
> On Wed, Feb 15, 2017 at 12:47 AM, Ramana Kumar <Ramana.Kumar at cl.cam.ac.uk>
>> Hi Yong Kiam,
>> As I mentioned, Connor is looking into evaluating the compiler in the
>> logic on the echo program on the basis branch. We just merged master
>> into basis and will start work on this in
>> characteristic/examples/compilation. The plan is to copy your to_bytes
>> function from benchmarks and tweak as necessary. Does this sound
>> There are a few improvements one could imagine trying too:
>> - rather than copy to_bytes, put it in a reusable library
>> - make the compset a bit smarter, so it does some evaluations in
>> parallel (like the bootstrap)
>> I know we had planned to reorganise (and delete) the eval directory.
>> I'm a little hesitant to do this while we're still waiting on
>> env-refactor to merge, in case it makes the merge harder than it needs
>> to be.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Developers