Re: [dataflow] [RFC] Remove many almost useless DCE passes

[Vladimir Makarov <vmakarov at redhat dot com>]

Paolo Bonzini wrote:

> > I know for example that your cfgcleanup.c changes (which are not
> > properly documented in the ChangeLog, btw) will cause us to miss many
> > optimizations (in particular, crossjumping -- we already have PRs
> > showing this is a problem on the dataflow branch).
> >
> > I'd also think that we have to be careful that we do not make it too
> > hard for ourselves to re-order RTL passes and to interpret dump files.
> > Both are harder with your approach.  I'd be perfectly happy to
> > sacrifice 0.5% of compile time for a more/easier maintainable
> > compiler.
>
>
> Agreed.  As I said, I just wanted to find a lower limit, and to 
> understand whether pruning DCE passes is the way to go for the 
> immediate goal of getting below the 5% mark set by the SC.
Sorry, I just don't understand how is this patch (at least removing some 
calls of delete_trivially_dead_insns) related to the new 
df-infrastructure.  You could do the same on the mainline.  Right?

And another question, what benchmarks and platforms do you use to 
measure the compiler speed for the df branch?