This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Thoughts about semantics of BB_VISITED


> On Friday 07 January 2005 21:44, Zdenek Dvorak wrote:
> > what about "dfs_enumerate_from does not work unless the marks are clear"
> > is not clear?  Either you call dfs_enumerate_from in situation where
> > all marks are clear, and then it does not change anything.
> What you are basically saying is that no pass should ever use
> I would have no problems with that, the uses of BB_VISITED that
> are in mainline (tree-ssa-pre.c, regrename.c, sched-rgn.c) are
> easily replaced with an sbitmap, and in fact that is probably
> even faster than using the flag because you need the extra pass
> over all basic blocks to clean BB_VISITED again.
> On the other hand, even if it is clear that "dfs_enumerate_from
> not work unless the marks are clear", that does not make it
> right for that function to set and clear that flag if BB_VISITED
> is supposed to be available to passes.
> > Or you call 
> > it when some of the marks are set, and then you have created a bug in
> > gcc.
> If there is a bug in gcc, then it is there because *you* have
> taken an existing BB flag and turned it into one that is basically
> private to, without documenting that.

please check Changelog before stating accusations like this.  I
*created* BB_VISITED specially for purpose of *dfs_enumerate_from*.
I made mistake of not documenting that it should be kept cleared
and used only inside utility functions, however, and someone
missunderstood its meaning and started using it inside optimization


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]