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: Checked in : Out Of SSA Rewrite - New TER implementation

> On Tue, 2006-11-21 at 14:04 -0500, Andrew MacLeod wrote:
> > This patch re-implements TER.  The fundamental idea remains the same...
> > step through the function determining tracking which single use
> > expressions are currently available for substitution,  and marking them
> > as replaceable if the use is seen while they are still valid.
> > 
> > The list of "current" values was maintained in a linked list, and the
> > patch in
> > replaced that with a bitmap, which turns out ot be much faster in the
> > more horrible cases.
> > 
> > On top of that a lot of functions have been renamed to make them
> > clearer, a long description of what TER does and how is at the top of
> > the file.  The mechanism for tracking current expressions is different
> > now as well. This allows TER to get a few cases it couldn't get before,
> > addressing a couple of open PRs.  Although it shares similar
> > functionality, this really is a rewrite of TER. 
> > 
> > It has also been moved to its own self contained file, tree-ssa-ter.c.
> > I'd suggest just reading the file rather than the patch :-)
> The latent expand_builtin_memcpy bug has now been fixed, and this patch
> was just checked in.  It bootstraps on i686-pc-linux-gnu and causes no
> new regressions.
> I've attached the patch which applies cleanly after all the other
> changes that have gone in since I originally posted this.

Nice, does it imply that I can drop -fno-tre-ter from memory tester on
bug2.c  now?


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