petsc-master 2020-11-30
Report Typos and Errors

PetscTimSortWithArray

Sorts an array in place in increasing order using Tim Peters adaptive sorting algorithm and reorders a second array to match the first. The arrays need not be the same type.

Synopsis

#include "petscsys.h"  
PetscErrorCode PetscTimSortWithArray(PetscInt n, void *arr, size_t asize, void *barr, size_t bsize, int (*cmp)(const void *, const void *, void *), void *ctx)
Not Collective

Input Parameters

n - number of values
arr - array to be sorted
asize - size in bytes of the datatype held in arr
barr - array to be reordered
asize - size in bytes of the datatype held in barr
cmp - function pointer to comparison function
ctx - optional context to be passed to comparison function, NULL if not needed

Output Parameters

arr - sorted array
barr - reordered array

Notes

The arrays need not be of the same type, however barr MUST contain at least as many elements as arr and the two CANNOT overlap.

Timsort makes the assumption that input data is already likely partially ordered, or that it contains contiguous sections (termed 'runs') where the data is locally ordered (but not necessarily globally ordered). It therefore aims to select slices of the array in such a way that resulting mergesorts operate on near perfectly length-balanced arrays. To do so it repeatedly triggers attempts throughout to merge adjacent runs.

Should one run continuously "win" a comparison the algorithm begins the "gallop" phase. It will aggressively search the "winner" for the location of the "losers" next entry (and vice versa) to copy all preceding elements into place in bulk. However if the data is truly unordered (as is the case with random data) the immense gains possible from these searches are expected __not__ to repay their costs. While adjacent arrays are almost all nearly the same size, they likely all contain similar data.

Sample usage

The comparison function must follow the qsort() comparison function paradigm, returning the sign of the difference between its arguments. If left < right : return -1, if left == right : return 0, if left > right : return 1. The user may also change or reverse the order of the sort by flipping the above. Note that stability of the sort is only guaranteed if the comparison function forms a valid trigraph. For example when sorting an array of type "my_type" in increasing order

  int my_increasing_comparison_function(const void *left, const void *right, void *ctx) {
    my_type l = *(my_type *) left, r = *(my_type *) right;
    return (l < r) ? -1 : (l > r);
  }
Note the context is unused here but you may use it to pass and subsequently access whatever information required inside the comparison function. The context pointer will unaltered except for any changes made inside the comparison function. Then pass the function
  PetscTimSortWithArray(n, arr, sizeof(arr[0]), barr, sizeof(barr[0]), my_increasing_comparison_function, ctx)

Fortran Notes

To use this from fortran you must write a comparison subroutine with 4 arguments which accepts left, right, ctx and returns result. For example
 subroutine CompareIntegers(left,right,ctx,result)
   implicit none

   PetscInt,intent(in) :: left, right
   type(UserCtx)       :: ctx
   integer,intent(out) :: result

   if (left < right) then
     result = -1
   else if (left == right) then
     result = 0
   else
     result = 1
   end if
   return
 end subroutine CompareIntegers

References

1. - Tim Peters. https://bugs.python.org/file4451/timsort.txt

See Also

PetscTimSort(), PetscIntSortSemiOrderedWithArray(), PetscRealSortSemiOrderedWithArrayInt(), PetscMPIIntSortSemiOrderedWithArray()

Level

developer

Location

src/sys/utils/sortso.c
Index of all Sys routines
Table of Contents for all manual pages
Index of all manual pages