Assembly output from JRuby 'fib'

[Rémi Forax]

On 04/29/2011 01:09 PM, Christian Thalinger wrote:
> On Apr 28, 2011, at 3:19 PM, Charles Oliver Nutter wrote:
>> On Thu, Apr 28, 2011 at 5:16 AM, Christian Thalinger
>> <christian.thalinger at oracle.com>  wrote:
>>> I took a look at it.  I used 64-bit x86 since the code is a bit smaller than with 32-bit.
>>>
>>> The code is almost identical but three things popped into my eye (the output is from PrintOptoAssembly):
>>>
>>> 1. The obvious one:  the method handle call site guard:
>>>
>>> 1a4   B32: #    B160 B33<- B31 B149 B123  Freq: 0.499969
>>> 1a4     movq    R10, byte[int:>=0]<ciObject ident=770 PERM address=0xe99088>  *  # ptr
>>> 1ae     movq    R10, [R10 + #1576 (32-bit)]     # ptr
>>> 1b5     movq    R11, [R10 + #32 (8-bit)]        # ptr
>>> 1b9     movq    R8, java/lang/invoke/AdapterMethodHandle:exact *        # ptr
>>> 1c3     cmpq    R11, R8 # ptr
>>> 1c6     jne,u  B160  P=0.000000 C=-1.000000
>> I saw in your other email that eliminating this puts indy on par with
>> dynopt, which is spectacular news. Can you elaborate on how that would
>> be possible to do "correctly" (as in not via a hack)? Would it be a
>> lighter-weight check and deopt of some kind (in Hotspot), or is it
>> something I'd need to rig up on my code?
>
> This would be what we referred to in the past as pull-vs-push notification.  I don't have details about that yet but maybe John has already thought this through and has an implementation idea (or even details).  Not sure when get around to implement that.

The API for push notification is 
MutableCallSite.syncAll/SwitchPoint.invalidateAll and is not currently 
implemented in the RI.
The idea is that instead of using a guard to check if there is a meta 
model change, the target of a callsite has
no guard for that but the runtime track all these callsites.
By example for a + callsite in JRuby, you only install a guard that 
check if the current object is a Fixnum
using a getClass() identity check. If someone somewhere open the Fixnum 
class to change something,
the runtime will change all callsites that use Fixnum to set their 
targets to fallbacks that use
a slow path and then call MutableCallSite.syncAll on all these callsites.
syncAll ensures that only the new targets (the fallbacks) are visible 
for the whole program.

Instead of using syncAll you can use SwitchPoint.invalidateAll which 
allow you to create
a special guard on a condition like is my meta model change or not.
This guard is not a real one because instead of doing the check each 
time when it calls,
it does nothing but when the SwitchPoint is invalidated, the guard is 
replaced by its fallback
doing a kind of deoptimization.

Currently in the RI, SwitchPoint checks a volatile field instead of 
doing the optimization/deoptimization.
I'm not far from having this working in the backport (I'm just stuck 
with a memory model issue).

So currently, the push API is here but there is no support from hotspot.
I guess we should to kill any people that will be between John and a 
keyboard
until he implements MutableCallSite.syncAll.

> -- Christian

Rémi