Re: GNAT Optimization of Constant Expressions

[Duncan Sands <baldrick@free.fr>]

PS: If you disable fold, then the function calls are hoisted
out of the loop if you mark sin etc with attribute "const",
but not if you mark them as "pure".  Now why aren't they
being marked "const" in the Ada case?  The gcc notion "const"
corresponds pretty much to Ada's notion of Pure.  In decl.c,
there is logic to mark a function as "const" if Is_Pure (an
internal Ada attribute) is true.  However Is_Pure is returning
false for the math functions.  Why?  I will investigate - but
whatever the reason it's basically a front-end bug.

Ciao,

Duncan.