Re: IRA reload review.

[Vladimir Makarov <vmakarov at redhat dot com>]

Bernd Schmidt wrote:
> Richard Henderson wrote:
> > Vladimir Makarov wrote:
>
> > > The last peace I needed was only approval for the above patches in 
> > > the reload part (reload1.c and caller-saves.c).  Bernd aggreed to do 
> > > review and requested me to break reload changes for IRA on several 
> > > patches.  Unfortunately, he disappeared (I guess he is on vacation).
>
> I was on vacation for a few weeks.
>
> Before that, my request to do a _real_ split up, i.e. either provide 
> pieces that can be committed independently, or explain why this is 
> impossible, was ignored.
Sorry, Bernd for our miscommunication.  I really was not going to ignore 
you.  The first I sent descriptions of reload changes when I asked you 
first time to review them.

http://gcc.gnu.org/ml/gcc-patches/2008-06/msg00485.html

After you asked me split them up, I sent the parts to make easy to 
review the reload changes.  They were not real patches because they were 
extracted from IRA patch.  That was probably my mistake, I should have 
sent the real patches.

http://gcc.gnu.org/ml/gcc-patches/2008-06/msg00535.html

You wrote that you need real patches and I promised to do this and 
started to work on them

http://gcc.gnu.org/ml/gcc-patches/2008-07/msg00148.html

I finished this work and sent the real patches

http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01446.html
http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01447.html
http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01448.html
http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01449.html
http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01450.html
http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01451.html
http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01452.html
http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01453.html

I split up all reload changes on 8 pieces which can be committed 
separately but mostly in order of their enumeration.  Some of them could 
be applied independently but unfortunately not all of them.  Probably it 
was happened after your vacation start.

What I did not do was benchmarking all changes as you asked.  
Unfortunatly I had no time to do this because it would have taken a lot 
of time (probably 1-2 week) and I was and am still busy fixing IRA 
bugs.  But I did all these changes to improve the code (mostly SPEC) for 
my 2 years work on IRA project if there were a degradation I would not 
put the change on the branch.  So all this changes improves the code in 
some way.

Once again, I am really sorry for the miscommunication.  It is really 
hard to do what you are asked right when you are under time pressure.  I 
was really too busy -- a lot of my time took reviewing of 
machine-independent part of selective scheduling (900KB patch).  The 
irony is that this review could have been postponed because selective 
scheduling probably will not be merged into 4.4 as another part (ia64 
machine dependent part) of it is not reviewed yet.