[PATCH] more PR33928: DIY dataflow in fwprop, to enable it at -O1 and speed it up at -O2

[Paolo Bonzini]

Paolo Bonzini wrote:
>>> Bootstrapped/regtested i686-pc-linux-gnu, checked that the time spent in
>>> fwprop even for devilish PR26854 testcases is very small.  Ok for trunk
>>> and after a while 4.4?
>> Err - if this patch doesn't make a difference what critical bug
>> does it fix to support backporting to 4.4?
> 
> It makes a difference at -O1, and fixes a performance and memory
> regression at -O2.

For -O1, the relevant performance regression is PR33928.  After this
patch only -frename-registers is needed to go back to 4.2 performance.

For -O2, the relevant compile-time (not performance) and memory
regression is PR26854.

Paolo