Re: [RFC] sparse SSA construction

[Christopher Li <sparse@chrisli.org>]

On Wed, Aug 16, 2017 at 8:57 AM, Dibyendu Majumdar
<mobile@majumdar.org.uk> wrote:
>> Not sure I understand the question.
>
> I meant that it would nice to save a function like SSA(linear input)
> -> SSA output - because then you can rerun the SSA conversion several
> times during optimization.

OK, in that regard. I agree.
But it is not relative to the discussion we have there.
Before promoting memory variable to pseudo, it is still
valid SSA form.  After the prompting, we of course want
to stay in SSA.


>
> I think there is still an advantage as the implementation is bottom
> up, uses memoization / dynamic programming techniques, and hence might
> actually be quite efficient. I think Luc is better placed to comment
> on this.

May be. I haven't look at the code yet. The paper did not describe how
to play with "goto", so I don't no clue what Luc have to do to make the
goto work yet.

>
> But Luc's implementation handles gotos fine without the extra phase -
> hence please please try out the implementation before arriving at
> conclusions.

As I state very clear in the title, my thought as regarding the paper,
and paper only. I will take some time to read Luc's implementation
next.

I just want to get a confirm my thought on the paper is not too
far off. If I understand the paper incorrectly, it certainly will have negtive
impact on how I am going to read the code.

Thanks

Chris