Hi
I've read some code of compiler (try to understand how it work)
and I notice, that tNode.isequal is invoked very often but has not
optimal (by me) implementation
my proposition is (one assigned less):
function tnode.isequal(p : tnode) : boolean;
begin
if assigned(self) then begin
result:=(assigned(p) and
(p.classtype=classtype) and
(p.nodetype=nodetype) and
(flags*flagsequal=p.flags*flagsequal) and
docompare(p));
end else
result:=not assigned(p);
end;
(* *original*
isequal:= (not assigned(self) and not assigned(p)) or
(assigned(self) and assigned(p) and
{ optimized subclasses have the same nodetype as their }
{ superclass (for compatibility), so also check the classtype
(JM) }
(p.classtype=classtype) and
(p.nodetype=nodetype) and
(flags*flagsequal=p.flags*flagsequal) and
docompare(p));
*)
and second:
make isequal inlined, then in function
function tunarynode.docompare(p : tnode) : boolean;
begin
docompare:=(inherited docompare(p) and
(not assigned(left) or left.isequal(tunarynode(p).left))
);
end;
compiler can fold two equal node : not assigned(left) or
((assigned(p) and
{ optimized subclasses have the same nodetype as their }
{ superclass (for compatibility), so also check the classtype
(JM) }
(p.classtype=classtype) and
(p.nodetype=nodetype) and
(flags*flagsequal=p.flags*flagsequal) and
docompare(p));
Will those optimization give better code?
Can compiler remove unnecessary nodes in boolean operation after inline
unfolding?
--
Darek
_______________________________________________
fpc-devel maillist - fpc-devel@lists.freepascal.org
http://lists.freepascal.org/mailman/listinfo/fpc-devel
