Re: ptr vs ref performance?? (test case)

[dewar at gnat dot com (Robert Dewar)]

<<Oh, well, then let me tell you that you aren't likely to see anyone do
anything about such reports.  One cannot, and will not get either
fast, or tight code with no optimizations.  We have no plans of ever
changing this.  Hence, there is little reason to examine a report like
this.
>>

It is worth noting for those not familiar with gcc, that -O0 in gcc
REALLY REALLY REALLY means no optimization. Most other compilers do
a fair amount of optimization even with "no optimization" set (at
least compared to gcc). So if you look at the performance of gcc
with -O0 you often get results MUCH worse that other compilers with
optimization off. Looking at ANY performance issue at -O0 is 
a completely useless excercise.

<<I kinda see where your coming from.  There should be a rough
equivalence between them...  its just that there are more pressing
issues that are more important.
>>

I would say there should be an equivalence at -O1, but it is not at
all surprising to find a difference at -O0.

A comment here is that I have often wished for a -O0.5 setting that
would do all optimizations possible that do not affect debugging.
This would really help the problem of -O0 executables being huge.

It is one thing to have a -O2 executable of 1 meg that is 2 megs in
-O0 mode, quite anther to have an executable of 200 megs that is
500 megs in -O0 mode (we have some users of GNAT with executables
this large, resulting from programs with many millions of lines of
code). 

I know some people react that no one should write programs that large,
but people definitely DO, and the results are generally successful,
but gimendous executable size is a problem :-)

Robert Dewar