Created
March 15, 2020 20:38
-
-
Save omar-3/1d30e5a1ae8ffda09e1170e0a9c50bca to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
use Sort; | |
var sentialPermutation = [42]; | |
// return true if the two arrays are equal | |
proc isEqual(a, b) : bool { | |
for (i, j) in zip(a,b) { | |
if i != j { | |
return false; | |
} | |
} | |
return true; | |
} | |
// we need to get ALL the permutations of [1,2,3,4,5] so we know | |
// that if we generate them in lexicographical order we would have | |
// [1, 2, 3, 4, 5] as the first permutation and [5, 4, 3, 2, 1] as the last permutation | |
// the partition function do the following: | |
// assume we have 4 tasks we need to distribute the whole lexicographical range of all permutations | |
// into 4 intervals. so assume we insert arr to be [1,2,3,4,5] and numOfTasks to 4 tasks | |
// we have in return the following array of arrays | |
// [1,2,3,4,5]--- | |
// ---- for task number 1 to generate | |
// [1,2,3,5,4]--- | |
// ---- for task number 2 to generate | |
// [1,2,5,4,3]--- | |
// ---- for task number 3 to generate | |
// [1,5,2,4,3]--- | |
// ---- for task number 4 to generate | |
// [5,1,2,4,3]--- | |
// ---- for task number 5 to generate | |
// [5,4,3,2,1]--- | |
// it works for arbitrary length of arrays and number of tasks | |
proc partition(arr, numOfTasks) { | |
var length = arr.numElements; | |
var step = length / numOfTasks; | |
var partitions: [1..numOfTasks+1] [1..length] int; | |
var j = length; | |
var i = 1; | |
partitions[i] = arr; | |
while j > 1 && i < numOfTasks { | |
arr[j] <=> arr[j - step]; | |
j -= 1; | |
i += 1; | |
partitions[i] = arr; | |
} | |
i += 1; | |
sort(arr, comparator=reverseComparator); | |
partitions[i] = arr; | |
return partitions; | |
} | |
// you feed this function arr to get the next lexicographic permutation | |
// we pass arrFinal to be able know when to stop | |
proc next_permutation(in arr, in arrFinal) { | |
// non-increasing suffix | |
var i = arr.numElements; | |
while i > 1 && arr[i - 1] >= arr[i] { | |
i = i - 1; | |
} | |
if isEqual(arr, arrFinal) { | |
return sentialPermutation; | |
} | |
var j = arr.numElements; | |
while arr[j] <= arr[i - 1] { | |
j = j - 1; | |
} | |
arr[i - 1] <=> arr[j]; | |
// sort the suffix | |
j = arr.numElements; | |
while i < j { | |
arr[i] <=> arr[j]; | |
i = i + 1; | |
j = j - 1; | |
} | |
return arr; | |
} | |
// for serial permutation ... works very well | |
iter permute(in arr) { | |
var firstPermutation = arr; | |
sort(firstPermutation); | |
var lastPermutation = arr; | |
sort(lastPermutation, comparator=reverseComparator); | |
yield firstPermutation; | |
var nowPermutation = next_permutation(firstPermutation, lastPermutation); | |
while !isEqual(nowPermutation, sentialPermutation) { | |
yield nowPermutation; | |
nowPermutation = next_permutation(nowPermutation, lastPermutation); | |
} | |
} | |
// here we the first permutation and assume it is sorted and the job of the leader iterator | |
// is to "partition" and pass the appropiate permutations to the followers through followThis | |
// tuple. | |
iter permute(param tag: iterKind, arr) | |
where tag == iterKind.leader { | |
var numOfTasks = here.maxTaskPar; // it works in forall loop if we make numOfTasks 1 | |
var partitions = partition(arr, numOfTasks); // otherwise the behavior of forall loop is undeterminstric and change for every run | |
coforall tid in 1..#numOfTasks { | |
var firstPermutation = partitions[tid]; | |
var secondPermutation = partitions[tid+1]; | |
yield (firstPermutation, secondPermutation); // now every follower needs to generate permutations | |
} // between these two limits | |
} | |
iter permute(param tag: iterKind, arr, followThis) | |
where tag == iterKind.follower && followThis.size == 2 { | |
var startPermutation = followThis(1); // this is the starting permutation | |
var finishPermutation = followThis(2); // this is the last permutation | |
yield startPermutation; | |
startPermutation = next_permutation(startPermutation, finishPermutation); | |
while true { | |
yield startPermutation; | |
startPermutation = next_permutation(startPermutation, finishPermutation); | |
if isEqual(startPermutation, sentialPermutation) { return; } | |
} | |
} | |
forall i in permute([1,2,3,4,5]) { // if we made numOfTasks equals to 1, it works as if we used | |
writeln(i); // serial iterators. otherwise it behavior changes every time | |
} | |
// for i in permute([1,2,3,4]) { ====> this works | |
// writeln(i); | |
// } | |
// var partitions = partition([1,2,3,4,5], 4) ===> behaves as expected |
oh thank you very much.
I will try to fix that first, and hopefully, things get fixed in parallel iterators. I think that It explains that non-deterministic output of it. Thank you again.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
for serial permutation where you are
if I use last permutation I dont reach an error halt in that case , Since it is trying to assigning differnt types of arrays , maybe is is linked to
issues chapel-lang/chapel#10497
if I try the your code of serial with --no-bounds , It inturn falls into segmentation falt.