edw@ius2.cs.cmu.edu (Eddie Wyatt) (04/12/87)
I haven't seen any other postings about compiler control lines to
control optimization. Something like :
#sacred :-)
a = (b + c) + f;
#endsacred
Seriously I think this has been the best proposal yet. It has no
affect on existing code and seems to allow the most flexiblity
over what parts of the code may be optimized.
I like the fact that the C compiler optimizes my code to a degree
for me. I don't have to rewrite expressions like xbar = x - 0.0 to
xbar = x and angle = (PI/180.0)*x/2.0 to angle = (PI/360.0)*x.
Why would I write code like this in the first place. Take a look
at a previous posting I made with a set of macros that I do use.
I am not totally insensitive to the complaints the numerically analysis
people are giving. I do think that the unitary '+' is an awlful hack to
the problem. In turn though, I do not think the number crunching
community has been listening to the complaints of us speed oriented
programmers. The types of applications that I am dealing with
are autonomous vehicles where every millisecond counts.
--
Eddie Wyatt
They say there are strangers, who threaten us
In our immigrants and infidels
They say there is strangeness, too dangerous
In our theatres and bookstore shelves
Those who know what's best for us-
Must rise and save us from ourselves
Quick to judge ... Quick to anger ... Slow to understand...
Ignorance and prejudice and fear [all] Walk hand in hand.
- RUSH hen@bu-cs.UUCP (04/17/87)
Many people who object to the use of unary plus to preserve order
are advocating some sort of compiler directive such as
#immune2optimization
...
{ code which you think you did real good }
...
#ok2optimize
I don't understand what this is supposed to mean - is the code only
exempt from explicit reordering, but a candidate for common
subexpression elimination? Can the optimizer do code hoisting?
Constant rippling into the immune section? Constant rippling out of
the immune section? Or os the directive supposed to take arguments,
so the code looks like
#immune2optimization reordering,rippling,hoisting,jamming,unscrolling
...
{ code which you think you did real good }
...
#ok2optimize
If the latter is what people have in mind, can these things be nested?
Thanks, folks, but I think I prefer unary plus. I may or may not
think it makes my listing look ugly, but at least I can understand how
any when to use it.
Bill Henneman
Gnurd in Residencejtr485@umich.UUCP (Johnathan Tainter) (04/19/87)
In article <1100@ius2.cs.cmu.edu>, edw@ius2.cs.cmu.edu.UUCP writes: > > I haven't seen any other postings about compiler control lines to > control optimization. Something like : > > #sacred :-) > a = (b + c) + f; > #endsacred I think under Draft Standard #pragma sacrosanct a = (b + c) #pragma ~sacrosanct is intended. My objection to this is that it overloads the preprocessor semantics for something inherently later in the process. Not too serious a problem but I think it could have been avoided. Just introducing pragma as a reserved word and using pragma sacrosanct; a = (b + c) pragma ~sacrosanct; appeals to me. > Eddie Wyatt --j.a.tainter