Thread overview
how to print progress of a long running parallel() loop?
Dec 07, 2020
mw
Dec 07, 2020
Dukc
Dec 07, 2020
Ali Çehreli
December 07, 2020
https://dlang.org/phobos/std_parallelism.html#.parallel

auto logs = new double[1_000_000];

foreach (i, ref elem; parallel(logs))
{
    elem = log(i + 1.0);
}


In Python, using joblib, I can use `verbose` level like this:

r = Parallel(n_jobs=2, verbose=10)(delayed(sleep)(.2) for _ in range(10))

to print out the progress.

How to do this in D's parallel loop?

thanks.

December 07, 2020
On Monday, 7 December 2020 at 08:16:50 UTC, mw wrote:
> r = Parallel(n_jobs=2, verbose=10)(delayed(sleep)(.2) for _ in range(10))
>
> to print out the progress.
>
> How to do this in D's parallel loop?
>
> thanks.

Allocate a `shared int` before the foreach loop. In the loop when, let's say `!(i & 0xFFF)`,  atomically increment the shared variable and print it's number.

Disclaimer: no experience about using `shared`, but this is what I'd try.
December 07, 2020
On 12/7/20 4:48 AM, Dukc wrote:
> On Monday, 7 December 2020 at 08:16:50 UTC, mw wrote:
>> r = Parallel(n_jobs=2, verbose=10)(delayed(sleep)(.2) for _ in range(10))
>>
>> to print out the progress.
>>
>> How to do this in D's parallel loop?
>>
>> thanks.
> 
> Allocate a `shared int` before the foreach loop. In the loop when, let's say `!(i & 0xFFF)`,  atomically increment the shared variable and print it's number.
> 
> Disclaimer: no experience about using `shared`, but this is what I'd try.

Yes, that would be the way but I don't know the effect of using shared or not. But because the progress needs to be somehow reported, an atomic variable may not be sufficient (unless a separate thread is reading it, etc.)

This issue came up during one of my recent presentations at this point:

  https://www.youtube.com/watch?v=dRORNQIB2wA&feature=youtu.be&t=1527

I used a synchronized block in that code because it did not affect performance in my use case.

Ali