CLA-R Sample Exam Question 22

[crossrulz]

---

@JÞB wrote:

---

@Intaris wrote:

Canyou even parallelise loops which don't have autoindexed inputs?  I thought you could only properly paralellise a for this reason (As the loop is currently written).

Shane

---

Yes, you can. In all cases # of iterations must be determistic (No conditional terminal) but that is true for any for loop without a conditional terminal that does not break the compiler.

---

And there can't be any dependance of previous iterations.  If doing in-place operations, the iterations are not dependant on the previous iterations.

[JÞB]

---

@crossrulz wrote:

---

And there can't be any dependance of previous iterations.  If doing in-place operations, the iterations are not dependant on the previous iterations.

---

With 1 notable exception: You will get a warning if parallizing a for loop containing Random. because the seed is internal to the function there is a dependancy internal to the history of random calls.  You'll still get the same values but in a different order in the output unless you enable debugging. (Forces sequential iterations) Trivia for the masses.

[Darin.K]

---

@JÞB wrote:

With 1 notable exception: You will get a warning if parallizing a for loop containing Random. because the seed is internal to the function there is a dependancy internal to the history of random calls.  You'll still get the same values but in a different order in the output unless you enable debugging. (Forces sequential iterations) Trivia for the masses.

---

More than 1, there are a number of nodes that have side effects in parallel loops, not just Random.  Off the top of my head I know of Get Date/Time in Seconds and Tick Count.

[crossrulz]

---

@JÞB wrote:

---

@crossrulz wrote:

---

And there can't be any dependance of previous iterations.  If doing in-place operations, the iterations are not dependant on the previous iterations.

---

With 1 notable exception: You will get a warning if parallizing a for loop containing Random. because the seed is internal to the function there is a dependancy internal to the history of random calls.  You'll still get the same values but in a different order in the output unless you enable debugging. (Forces sequential iterations) Trivia for the masses.

---

I wouldn't really consider Random to be dependent.  You truely want something random after all.  But it does make sense that it comes out as a warning due to random functions not truely being random.

[herrlin]

Thank you everyone for the explanations.   In the past I have just blindly looked at the shift registers and told myself that iteration parallelism would not be beneficial.  I will be more attentive of the index and replace operation in future.

Thanks