Re: quality / availability of gnat

[mgk25@cl.cam.ac.uk (Markus Kuhn)]

|> Martin Gangkofer <mttg@my-deja.com>:
|> > I am planning to use gnat as cross compiler for an embedded application
|> > based on LynxOS operating system. The application  demands high
|> > performance, this means the compiler shall produce good quality code.

|> > (1) Do you know wether there is a complete (and supported) port of gnat
|> >     to LynxOS.

I don't know about LynxOS, but ACT does offer commercial support for
a couple of embedded real-time operating systems, including for example
VxWorks, and some avionics applications of significant complexity built
with VxWorks and GNAT are already flying on board of various US military
aircraft.

|> > (2) How good is the quality of gnat in terms of performance of the
|> >     generated code?

It is usually possible and relatively straight forward to produce
with GNAT the exact same sequence of machine instructions that
the GNU C compiler would produce from the corresponding C program.
Sometimes GNAT does implement things internally in a way that
surprises beginners a bit (especially related to how variable-length
objects are treated on the stack, where a LISP/Perl/Java/Python
programmer might intuitively expect a heap allocation). These things
can however usually easily be spotted by inspecting the "-gnatdg" output,
and the intermediate (low-level Ada-style) code that you see there
is converted into machine code very similar to how the same C code
would be compiled. If you use very basic Ada features that are
equivalent to C's features, you get the same code. If you use high-level
features of Ada (exceptions, unbounded arrays, tasking, etc.) that are not
available in C, you have to test whether the code produced from these
high-level features meets your performance requirements.

I have done encryption and image processing applications with GNAT and
the performance was very comparable to what I get with similar code
in C (or at least not more than 10-20% worse, and that is just a few
months in Moore's law).

|> > (4) Is gnat suitable for use in a big projects 100,000 ... 500,000 SLOC?

I hear that it is successfully being used in several projects of this
size by ACT and ACT-Europe customers.

Markus

-- 
Markus G. Kuhn, Computer Laboratory, University of Cambridge, UK
Email: mkuhn at acm.org,  WWW: <http://www.cl.cam.ac.uk/~mgk25/>