1
$\begingroup$

In the following Mathematica compiled program, 3 persons take turns in a fair Russian Roulette Game with 1 bullet in the 6th chamber and we find the probabilities of each person being killed after 1,000,000 repetitions of the experiment,

The program:

  p = 3; n = 1000000.; arg = Table[0, n]; Roulette = 
  Compile[{i}, Module[{c = 1}, While[RandomInteger[{1, 6}]
  != 6, c++]; m = Mod[c, p]; If[m != 0, m, p]], 
  RuntimeAttributes -> {Listable}, Parallelization -> True, 
  RuntimeOptions -> "Speed"]; 
  Counts[Roulette[arg]]/n // AbsoluteTiming

with Parallelization -> True, gives nonsense whereas without:

  p = 3; n = 1000000.; arg = Table[0, n]; Roulette = 
  Compile[{i}, Module[{c = 1}, While[RandomInteger[{1, 6}]
  != 6, c++]; m = Mod[c, p]; If[m != 0, m, p]], 
  RuntimeAttributes -> {Listable},  
  RuntimeOptions -> "Speed"]; 
  Counts[Roulette[arg]]/n // AbsoluteTiming

gives correct answer.

By including local variable m in the module as in:

  p = 3; n = 1000000.; arg = Table[0, n]; Roulette = 
  Compile[{i}, Module[{m = 1,c = 1}, While[RandomInteger[{1, 6}]
  != 6, c++]; m = Mod[c, p]; If[m != 0, m, p]], 
  RuntimeAttributes -> {Listable}, Parallelization -> True, 
  RuntimeOptions -> "Speed"]; 
  Counts[Roulette[arg]]/n // AbsoluteTiming

we take the correct answer this time, but parallelization still fails.

$\endgroup$
9
  • $\begingroup$ The function is changing a global variable m and using it in the result, so that could be a problem. $\endgroup$ Commented Jun 23, 2019 at 15:22
  • $\begingroup$ I have tried initialization {c = 1,m = 1} inside module, no avail. Also, do not forget to use first program with Parallelization -> True, not the second one. $\endgroup$ Commented Jun 23, 2019 at 17:11
  • $\begingroup$ In V12 on macOS, the first program gives me an error "Instruction 10 in CompiledFunction[...] calls ordinary code that can be evaluated on only one thread at a time." Also, I get some probability that person 0 will be killed. Is that what you mean by it gives nonsense? $\endgroup$
    – MassDefect
    Commented Jun 23, 2019 at 18:52
  • $\begingroup$ Yes, there is no 0 person by definition of: m = Mod[c, p]; If[m != 0, m, p] $\endgroup$ Commented Jun 23, 2019 at 19:19
  • 1
    $\begingroup$ (1) I cannot get the bad behavior on my office Linux machine. Yet I saw it on my Windows laptop. Very strange. (2) Return Mod[c,p,1] from the Module-- it's faster that way. $\endgroup$ Commented Jun 23, 2019 at 21:18

1 Answer 1

1
$\begingroup$

Program needs all variables p, c, m to be local inside the module:

n = 1000000.; arg = Table[0, n]; Roulette = 
Compile[{i}, Module[{p = 3,m = 1,c = 1}, While[RandomInteger[{1, 6}]
!= 6, c++]; m = Mod[c, p]; If[m != 0, m, p]], 
RuntimeAttributes -> {Listable}, Parallelization -> True, 
RuntimeOptions -> "Speed"]; 
Counts[Roulette[arg]]/n // AbsoluteTiming

There is no bug!

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.