[tree-ssa] Memory usage on the rise again

[tree-ssa] Memory usage on the rise again

[law]

So I go away for 2 days and when I return, memory usage on Gerald's
testcase for the tree-ssa branch has risen 5% (both in terms of 
total memory allocated by the GC system and in terms of persistent
memory allocated by the GC system).

I would like to ask that folks start looking closely at their changes
to see how they impact the memory system.  It's frustrating to spend
considerable time to get a few percent back to have all the gains
disappear in a few days.

jeff

Re: [tree-ssa] Memory usage on the rise again

[Daniel Berlin]

On Dec 11, 2003, at 10:14 PM, law@redhat.com wrote:

>
> So I go away for 2 days and when I return, memory usage on Gerald's
> testcase for the tree-ssa branch has risen 5% (both in terms of
> total memory allocated by the GC system and in terms of persistent
> memory allocated by the GC system).
>
> I would like to ask that folks start looking closely at their changes
> to see how they impact the memory system.  It's frustrating to spend
> considerable time to get a few percent back to have all the gains
> disappear in a few days.

It's highly likely this is due to GC'ifying of the CFG.

>
> jeff
>
>
>

Re: [tree-ssa] Memory usage on the rise again

[Steven Bosscher]

On Friday 12 December 2003 04:14, law@redhat.com wrote:
> So I go away for 2 days and when I return, memory usage on Gerald's
> testcase for the tree-ssa branch has risen 5% (both in terms of
> total memory allocated by the GC system and in terms of persistent
> memory allocated by the GC system).
>
> I would like to ask that folks start looking closely at their changes
> to see how they impact the memory system.  It's frustrating to spend
> considerable time to get a few percent back to have all the gains
> disappear in a few days.

This is in part because of the garbage collectable CFG.  We did not produce 
memory statistics for that since alloc-pool allocated objects were not 
included in the memory stats.

Gr.
Steven

Re: [tree-ssa] Memory usage on the rise again

[Jan Hubicka]

> 
> On Dec 11, 2003, at 10:14 PM, law@redhat.com wrote:
> 
> >
> >So I go away for 2 days and when I return, memory usage on Gerald's
> >testcase for the tree-ssa branch has risen 5% (both in terms of
> >total memory allocated by the GC system and in terms of persistent
> >memory allocated by the GC system).
> >
> >I would like to ask that folks start looking closely at their changes
> >to see how they impact the memory system.  It's frustrating to spend
> >considerable time to get a few percent back to have all the gains
> >disappear in a few days.
> 
> It's highly likely this is due to GC'ifying of the CFG.
Except for GGCized CFG, there is also latent bug in the inliner worked
around by duplicatin inline pass before tree-optimize and copying
resulting body around to avoid gimple sharing (ouch).  That interacts
badly with new non-transitive inline plan code (I have RFC patch on
fixing the problem in question).  Depending on your setup it is possible
that it cause some rise of memory consumption.

However in my tests before commiting the patch (done on Gerald's
testcase) the memory overhead lowered instead of rised due to patch.  It
will reduce again once I get past this major mess.  It got caught by new
sanity checking code for cgraph checking that all functions are properly
deallocated I am hoping to manage to send out this or next week (it
finds really many positives). my current branch is again 12% behind
mainline consumption.  

Honza
> 
> >
> >jeff
> >
> >
> >