SPEED problem with pointers to structure elements

[GillesDuvert]

There is a huge loss of efficiency in GDL between accessing directly structure members, like !X.TICKV and indirectly via a pointer to it. See this:
toto=replicate(!X,100000L)

GDL> tic & a=max(toto.tickv) & toc
% Time elapsed : 0.038621187 seconds

IDL> tic & a=max(toto.tickv) & toc
% Time elapsed: 0.038397074 seconds.

so far so good.

z=toto.tickv

GDL> tic & a=max(z) & toc 
% Time elapsed : 0.0026879311 seconds.

IDL> tic & a=max(z) & toc         
% Time elapsed: 0.0063710213 seconds.

ok again.
p=ptr_new(toto)

IDL> tic & a=max((*p).tickv) & toc
% Time elapsed: 0.030293941 seconds.

GDL> tic & a=max((*p).tickv) & toc
% Time elapsed : 0.49356413 seconds.

that is, GDL is SIX TIMES slower.

[GillesDuvert]

So probably GDL is a bit incapacited when it comes to ptr-to-struct (aka OO 😃 ) programming (widely used in NASA's SSW package --- hence HESSI performances issues) ...

[GillesDuvert]

hm this will need a bit of ANTLR alas.
max((*p).tickv) is interpreted as (*p) -> creates a copy of (*p) , in this case a huge amount of memory,
then .tickv -> creates a new variable corresponding to this array and apply max() on it to get result.

Speed difference can be largely worse than example if tag size on which we operate is small (say, one byte) and the rest of the structure is huge!

The only possibility IMHO is to get ANTLR to register (*p).tickv as a special kind of node, and have it treat (*p).tickv in one pass.