Speculative prefetching: report

[Dale Johannesen]

On Oct 15, 2004, at 1:56 PM, Zdenek Dvorak wrote:
> Hello,
>
>>>> 2.  In the profile-generation phase, there is a huge code bloat
>>>> resutling from
>>>> instrumenting all the loads and stores.  This badly stresses the 
>>>> rest
>>>> of the BE,
>>>> causing out-of-memory ICEs or unacceptably slow compile times 
>>>> (hours)
>>>> on
>>>> several SPECmarks.  (With IMA to be sure.)
>>>
>>> this is bad.  I guess it should be possible to prune the references 
>>> we
>>> try to prefetch somehow.
>>
>> Implementing the instrumentation as function calls rather than
>> inline code might help.
>
> it is not possible to insert new function calls at rtl level,
> unfortunately.

I'm not sure I believe that, but in any case it's certainly possible 
with
tree-based profiling.

Which reminds me, could somebody look at this tree-profiling-branch
patch to implement value-based profiling in trees?
http://gcc.gnu.org/ml/gcc-patches/2004-09/msg00439.html