Improve performance of large local arrays

[nars1]

Final Release Note

Local arrays with large number of subscripts scale much better. When the number of nodes in a local array is in the millions, node creation time is now noticeably faster (in the example given in this Issue Description, the average $zut column improved from 18 seconds to 0.1 second when the Array Size was 16Mi). [#80]

Description

Cycle ; Local Array Scalability Test.
    new Cycle,Iter,Size,Then,Span,Average
    write !,"Cy Array Size Avg $zut",!
    write "-- ---------- ------",!
    for Cycle=1:1:24 do
    .   new Array
    .   set Size=2**Cycle
    .   set Then=$zut
    .   for Iter=1:1:Size set Array(Iter)=Iter
    .   set Span=$zut-Then
    .   set Average=Span/Size
    .   write $justify(Cycle,2)," ",$justify($fnumber(Size,","),10)," ",$justify(Average,6,3),!
    quit

The above M program creates progressively larger local arrays (according to powers of two) and reports the average duration of the process per array node.

Here are the results it prints out:

Cy Array Size Avg $zut
-- ---------- ------
 1          2  0.500
 2          4  0.000
 3          8  0.125
 4         16  0.063
 5         32  0.063
 6         64  0.078
 7        128  0.078
 8        256  0.059
 9        512  0.063
10      1,024  0.062
11      2,048  0.062
12      4,096  0.061
13      8,192  0.062
14     16,384  0.064
15     32,768  0.066
16     65,536  0.071
17    131,072  0.083
18    262,144  0.137
19    524,288  0.260
20  1,048,576  0.466
21  2,097,152  0.857
22  4,194,304  1.696
23  8,388,608  4.620
24 16,777,216 18.002

The average $zut column increases exponentially as the # of nodes in the array doubles. We instead expect the average time column to be more or less the same.

Draft Release Note

Local arrays with large number of subscripts scale a lot better. When the # of nodes in a local array are in the millions, node creation time is now noticeably faster (in the example given in this Issue Description, the average $zut column improved from 18 seconds to 0.1 second when the Array Size was 16Mi).