Re: GNAT compiler switches and optimization

[Jeffrey Creem <jeff@thecreems.com>]

Followup on the bug report.

One of the comments asserted that the two programs were not equivilent 
though I am not yet 100% convinced that I believe it yet.

His recommendation was to remove a level of indirection by changing the 
way the array is declared.

    N : Positive := Positive'Value (Argument (1));
    G : Ada.Numerics.Float_Random.Generator;

    type Real_Matrix is array (1..N, 1..N) of Float;
    type Matrix_Access is access Real_Matrix;

    A,B,C : Matrix_Access;
    Start, Finish : Ada.Calendar.Time;
    Sum : Float := 0.0;
begin
    A := new Real_Matrix;
    B := new Real_Matrix;
    C := new Real_Matrix;

This does indeed substantially improve the performance (still not quite 
to FORTRAN levels). The reason I question the equilivence is that in the 
original version, the array could have been passed to a procedure that 
took in an unconstrained array (which is pretty much what I think the 
FORTRAN version would allow) while this new version would not do that.


A quick check shows the FORTRAN is now only 1.2 times faster (did not do 
the multiple run thing yet). Perhaps tonight.

I'll also attempt to take one of the threaded ones and include that as well.


Can someone that understands FORTRAN better make an argument about the 
"closeness" of this approach v.s. the other?