Discuss Scratch

gtoal
Scratcher
1000+ posts

explicit parallelism

I suggest a block which contains 3 C-block sections, to be called “simultaneously”…

When you execute it, it will start executing the first two C-blocks at the same time, and then when both are finished the ‘simultaneously’ block exits.

The third-leg of the block is for when there are insufficient resources to run simultaneously. In this case neither of the first two blocks executes and only the third block does.

I'm assuming that parallel executions would be limited in the same way that clones are.

This block would effectively add threads/clones to any code, with a rendezvous after execution. In systems with multiple cores it would execute in genuine parallel otherwise it would simulate parallel execution the way that clones currently do.

G
ChocolatePi
Scratcher
1000+ posts

explicit parallelism

Why three in particular? Why not just a block like this?
launch thread and run :: control cstart
...
end
And then you could join as many as you wanted together
ChocolatePi
Scratcher
1000+ posts

explicit parallelism

Upon reading this a second time I see what you mean about three branches. I now support your suggestion as it is.
gtoal
Scratcher
1000+ posts

explicit parallelism

ChocolatePi wrote:

Upon reading this a second time I see what you mean about three branches. I now support your suggestion as it is.

You saved me a lot of typing there :-)

Look on it as a fork + try/catch block if you prefer that model. I hesitated to include the part for a failure but it's really necessary I'm afraid. As is waiting for both parts of the fork to complete before continuing. Otherwise the multi-threading has to be handled by the entire Scratch run-time - whereas with this not only is the user code clearer but the new code to add to Scratch to implement it can be localised to one block (I hope).
liam48D
Scratcher
1000+ posts

explicit parallelism

Sounds rather complicated for Scratch, but support.

Would it look like this?
launch :: control cstart
...
and :: celse
...
if out of resources :: celse // better wording anybody?
...
:: cend

202e-202e-202e-202e-202e UNI-CODE~~~~~
gtoal
Scratcher
1000+ posts

explicit parallelism

liam48D wrote:

Sounds rather complicated for Scratch, but support.

Would it look like this?
launch :: control cstart
...
and :: celse
...
if out of resources :: celse // better wording anybody?
...
:: cend

yep, that was exactly what I meant. Wording to be determined.
PrincessPanda_test_
Scratcher
1000+ posts

explicit parallelism

No support, new scratchers and scratchers who do not know advanced code will be extremely confused.

Support, although it may be confusing.

Last edited by PrincessPanda_test_ (Jan. 9, 2016 02:45:35)















Click the image for something special!

Wanna collab on something Sonic?
Chibi-Matoran
Scratcher
1000+ posts

explicit parallelism

PrincessPanda_test_ wrote:

No support, new scratchers and scratchers who do not know advanced code will be extremely confused.
Many Scratch features are confusing to beginners. Gtoal's suggestion would be very helpful for many Scratchers. Merely being too “advanced” is not a valid reason to not support suggestion; one must also consider the potential benefits.

If you see this signature while scrolling, you have been visited by ʎ++ haskal. Lazy evaluation, monads, function purity, lambdas, currying, partial application, algebraic data types, pattern matching, and typeclasses will come to you, but only if you comment, “Thanks, ʎ++ haskal” on my profile.

Cassian && ||
gtoal
Scratcher
1000+ posts

explicit parallelism

Chibi-Matoran wrote:

PrincessPanda_test_ wrote:

No support, new scratchers and scratchers who do not know advanced code will be extremely confused.
Many Scratch features are confusing to beginners. Gtoal's suggestion would be very helpful for many Scratchers. Merely being too “advanced” is not a valid reason to not support suggestion; one must also consider the potential benefits.

Anyway Scratchers already handle multithreading/parallelism, through clones. This just makes it a little more explicit and easier to understand. And control.
Chibi-Matoran
Scratcher
1000+ posts

explicit parallelism

gtoal wrote:

Chibi-Matoran wrote:

PrincessPanda_test_ wrote:

No support, new scratchers and scratchers who do not know advanced code will be extremely confused.
Many Scratch features are confusing to beginners. Gtoal's suggestion would be very helpful for many Scratchers. Merely being too “advanced” is not a valid reason to not support suggestion; one must also consider the potential benefits.

Anyway Scratchers already handle multithreading/parallelism, through clones. This just makes it a little more explicit and easier to understand. And control.
Yeah, creating a clone is inefficient. However, I'm a bit confused. If I'm understanding the suggestion correctly, except for the third part, can't it also be worked around with broadcasts (still, this suggestion would be useful because it would allow blocks to be executed simultaneously in the same script)?
...
broadcast [launch threads v] and wait
...
when I receive [launch threads v]
...
when I receive [launch threads v]
...
Am I misunderstanding the suggestion?

Last edited by Chibi-Matoran (Jan. 9, 2016 03:41:04)


If you see this signature while scrolling, you have been visited by ʎ++ haskal. Lazy evaluation, monads, function purity, lambdas, currying, partial application, algebraic data types, pattern matching, and typeclasses will come to you, but only if you comment, “Thanks, ʎ++ haskal” on my profile.

Cassian && ||
gtoal
Scratcher
1000+ posts

explicit parallelism

Chibi-Matoran wrote:

gtoal wrote:

Chibi-Matoran wrote:

PrincessPanda_test_ wrote:

No support, new scratchers and scratchers who do not know advanced code will be extremely confused.
Many Scratch features are confusing to beginners. Gtoal's suggestion would be very helpful for many Scratchers. Merely being too “advanced” is not a valid reason to not support suggestion; one must also consider the potential benefits.

Anyway Scratchers already handle multithreading/parallelism, through clones. This just makes it a little more explicit and easier to understand. And control.
Yeah, creating a clone is inefficient. However, I'm a bit confused. If I'm understanding the suggestion correctly, except for the third part, can't it also be worked around with broadcasts (still, this suggestion would be useful because it would allow blocks to be executed simultaneously in the same script)?
...
broadcast [launch threads v] and wait
...
when I receive [launch threads v]
...
when I receive [launch threads v]
...
Am I misunderstanding the suggestion?

Interesting, I didn't realise you could trigger the same script twice outside of clones. You're right, this is a similar effect and may suffice, though what I was suggesting would have a much stronger guarantee of actual parallelism if possible. The idea was to take a mechanism that Scratchers were to some extent familiar with (clones) but supply a stronger version of it that could be used to do some real computation and not only speed up heavy Scratch computations, but give Scratchers something that is a little bit closer to the model of parallelism that is used in multi-processor systems (eg OpenMP, pthreads, fork)

Question: does the broadcast and wait actually wait until both receive blocks have completed?

G
Failord
Scratcher
1000+ posts

explicit parallelism

gtoal wrote:

Chibi-Matoran wrote:

-snip-

…can't it also be worked around with broadcasts (still, this suggestion would be useful because it would allow blocks to be executed simultaneously in the same script)?
...
broadcast [launch threads v] and wait
...
when I receive [launch threads v]
...
when I receive [launch threads v]
...
-snip-

Question: does the broadcast and wait actually wait until both receive blocks have completed?

Actually, it does! That's an interesting and generally poorly understood function of Scratch that is actually quite useful.

Last edited by Failord (Jan. 9, 2016 14:38:38)


Thanks for the ride, Scratch Team and community! Unfortunately, the time has come for me to move on from Scratch, so I'll no longer be using the forums. Thanks again, and farewell!
Invisible text! This will show you if a profile I claim to be mine IS mine. Just remember to visit the profile and see if I confirmed in the comments.
For old time's sake, see the revolutionary Laser Battle 2.0! Also see Unstoppable Game!

And no, the evil kumquats didn't alter my signature…

Powered by DjangoBB

Standard | Mobile