several messages

[Michael Matz]

Hi,

sorry for taking so long to answer, but I was away some time, and now my
mailbox is flooded ;)

On Tue, 1 Aug 2000, Jeffrey A Law wrote:
> Some general notes about the dominator code.
...

Thanks. As I said, it was a first version, so people can look at the
general algorithms. It was by no means ready (style-wise at least ;).

> What is TBB?  You use it in a number of places, but I have no idea what it's

TBB == Type Basicblock ;) (In a former life I was a great fan of Turbo
Pascal, and now I'm still calling my types Txxx sometime ;)

> Low level stuff:
> 
>   The formatting for init_ar, stack_top,stack_pop has a number of problems. 

They were supposed to be throw-away macros, but they survived somehow. So
even such macros should follow the coding standards? No problem.

> Comments should be complete sentences with proper capitalization and
> punctuation.  You have several that start with lower case.

Hmm. Seems so. I first wrote the code, with some comment in them. Then
after some time (the code was done), I wrote comments all over the place,
which was boring ;) It seems I was so tired, that I forgot to change some
to the comment standard. OTOH some of them start with a variable name,
which is spelled lowercase, so what should I do?

> Most of our code could stand for some whitespace at boundaries where you
> transition from one logical task to another.  For example, you have code

I thought, exactly that was the reason, that '{' should be on an extra
line itself (and I use colorizing editors ;) I'll insert some white lines
here and there.

> There's a lot more code to look at, but I'm not going to be able to do it 
> soon.

;-(

> I figured at least getting you some of my initial comments might be
> helpful.

Of course, thanks.


On Thu, 3 Aug 2000, Richard Henderson wrote:
> On Sat, Jun 24, 2000 at 04:22:15AM +0200, Michael Matz wrote:
> >  compute_flow_dominators (dominators, post_dominators)
> > +     sbitmap *dominators;
> > +     sbitmap *post_dominators;
> > +{
> > +  if (dominators)
> > +    {
> > +      calculate_flow_dominators (dominators, 0);
> > +    }
> 
> I'd prefer that we not create such trivial wrapper functions.  Instead,
> just modify all current users of compute_flow_dominators to use your
> new function.

This is the plan, yes. I can only state the reason I've done the above for
now. As I wrote it, it was clear, that it'll take some time, until the
code is included (I hadn't even the assignment ready). So I expected quite
some changes in gcc over time, and in this way my local changes to gcc's
source (with its potential for conflicts) were kept to a minimum.

> Additionally, since the algorithm has the entire dominator tree, then
> immediate dominators are available directly.  You should export an
> interface to get at them, and then replace the existing
> compute_immediate_dominators function.

Yep. I hadn't done that, because (at the time I wrote it at least) that
function wasn't used anywere in gcc, so there would have been no direct
test for correctness (only the indirect one over the dominators).


On Fri, 4 Aug 2000, Jeffrey A Law wrote:
> >From ldc.c:
> /* This file implements an algorithm for calculating dominators in linear
>    time, as presented in ftp://ftp.diku.dk/diku/semantics/papers/D-320.ps.gz
> 
> I wasn't able to connect to that FTP server.  Do you have a copy of this
> paper that you could send me?  It would greatly help in trying to understand
> this code.

Yes, without the paper its not the easiest thing ;) I believe, you have it
now, if not, just tell me again.

Thanks again for your comments, I hope I have some time on the weekend to
change the source accordingly.


Ciao,
Michael.