Re: Excessive copies of set elements in GMPL

[Andrew Makhorin]

On Fri, 2020-07-24 at 09:12 +0200, Domingo Alvarez Duarte wrote:
> And here is the output of the the script bellow:
> 
> Memory usage:
> 
> ubuntu glpsol package : 1,422,160 KB => 4.20s elapsed
> 
> glpk with my changes: 429,000 KB => 1.37s elapsed
> 
> 

As I explained in my previous message, the MathProg translator performs
 all computations directly as specified by corresponding expressions,
i.e. *no code optimization* is used. 

In most cases the user may perform some "optimization" replacing
non-efficient expressions by more efficient ones. For example,

  set D := { (a,b,c) in A cross B cross C : b=a+1 and c=b+2 };

can be replaced by

  set D := setof{a in A, b in B, c in C: b=a+1 and c=b+2} (a,b,c);

in order not to compute "A cross B cross C".

I agree that the translator could be more smart to recognize such
cases, but necessary analysis in a general case is a non-trivial thing,
and there was no attempt to implement anything of this kind.

The issue can be illustrated by the following example:

  for (i = 0; i < 1000000; i++)
  for (j = 0; j < 1000000; j++)
  for (k = 0; k < 1000000; k++)
    if (j == i+1 && j == j+2)
      foo(i, j, k);

Would you expect the C compiler to optimize this fragment in order not
to perform obvious excessive computations?


Andrew Makhorin