On Feb 18, 2011, at 2:19 PM, Christian Thalinger wrote:
> On Feb 18, 2011, at 1:28 PM, Charles Oliver Nutter wrote:
>> On Fri, Feb 18, 2011 at 6:17 AM, Christian Thalinger
>> <[email protected]> wrote:
>>> Is there a switch to turn direct ruby-to-ruby calls on and off? I'd like
>>> to compare inlining trees and maybe code output.
>>
>> Just pushed a couple for you in a4a0802:
>>
>> jruby.compile.invokedynamic.rubyDirect=true|false (Ruby targets
>> directly bind; defaults to true)
>> jruby.compile.invokedynamic.nativeDirect=true|false (Native (Java)
>> targets directly bind; defaults to true)
>>
>> Not to put any pressure on, but I had hoped to see this
>> directly-binding Ruby calls would *improve* perf :) With this in
>> place, the only thing dynopt does that my indy logic doesn't do is
>> dispatch literal fixnums directly through as long values (rather than
>> as RubyFixnum objects...and yes, this is only in place for literals).
>> Ideally, I'd expect optimized invokedynamic to perform as well as
>> dynopt in this situation.
>
>
> What I can tell so far is that setting rubyDirect=true inlines a whole lot
> more stuff than without. I looks like it wants to inline all recursive calls
> of fib_ruby into a single method, which doesn't work very well, hitting two
> limits: NodeCountInliningCutoff and DesiredMethodLimit.
>
> I tried to bump these limits but that is a dead end for the compiler since at
> one point it skips the compile because it gets too many nodes:
>
> 43 COMPILE SKIPPED: out of nodes parsing method (not retryable)
>
> I don't know why it wants to inline all these recursive calls. Is that
> something JRuby tries to do?
Here are two (plain) inlining trees with
rubyDirect=false:
@ 51 java.dyn.MethodHandle::invokeExact inline (hot)
@ 8 sun.dyn.MethodHandleImpl$GuardWithTest::invoke_L5
inline (hot)
@ 11 java.dyn.MethodHandle::invokeExact inline (hot)
@ 15
org.jruby.runtime.invokedynamic.InvokeDynamicSupport::test inline (hot)
@ 5 org.jruby.RubyBasicObject::getMetaClass inline
(hot)
@ 8 org.jruby.RubyModule::getCacheToken inline (hot)
@ 28 java.dyn.MethodHandle::invokeExact inline (hot)
@ 23 sun.dyn.FilterGeneric$F6::invoke_F6 inline (hot)
@ 17 java.dyn.MethodHandle::invokeExact inline (hot)
@ 20
org.jruby.runtime.invokedynamic.InvokeDynamicSupport::pollAndGetClass inline
(hot)
@ 1 org.jruby.runtime.ThreadContext::callThreadPoll
inline (hot)
@ 19
org.jruby.runtime.ThreadContext::pollThreadEvents inline (hot)
@ 5 org.jruby.RubyThread::pollThreadEvents
inline (hot)
@ 5 org.jruby.RubyBasicObject::getMetaClass inline
(hot)
@ 5 org.jruby.RubyBasicObject::getMetaClass inline
(hot)
@ 29 java.dyn.MethodHandle::invokeExact inline (hot)
@ 35 sun.dyn.FilterGeneric$F7::invoke_F7 inline
(hot)
@ 19 java.dyn.MethodHandle::invokeExact inline
(hot)
@ 23
org.jruby.runtime.invokedynamic.InvokeDynamicSupport::getMethod inline (hot)
@ 33 java.dyn.MethodHandle::invokeExact inline
(hot)
@ 43 java.dyn.MethodHandle::invokeExact call site not
reached
@ 64
org.jruby.ast.executable.AbstractScript::getCallSite2 inline (hot)
rubyDirect=true:
@ 51 java.dyn.MethodHandle::invokeExact inline (hot)
@ 8 sun.dyn.MethodHandleImpl$GuardWithTest::invoke_L5
inline (hot)
@ 11 java.dyn.MethodHandle::invokeExact inline (hot)
@ 15
org.jruby.runtime.invokedynamic.InvokeDynamicSupport::test inline (hot)
@ 5 org.jruby.RubyBasicObject::getMetaClass inline
(hot)
@ 8 org.jruby.RubyModule::getCacheToken inline (hot)
@ 28 java.dyn.MethodHandle::invokeExact inline (hot)
@ 20
bench.bench_fib_recursive::method__0$RUBY$fib_ruby inline (hot)
@ 4
org.jruby.ast.executable.AbstractScript::getCallSite0 inline (hot)
@ 14 org.jruby.runtime.callsite.CachingCallSite::call
inline (hot)
@ 5 org.jruby.runtime.ThreadContext::getRuntime
inline (hot)
@ 10 org.jruby.RubyFixnum::newFixnum inline (hot)
@ 1 org.jruby.RubyFixnum::isInCacheRange inline
(hot)
@ 13
org.jruby.runtime.callsite.CachingCallSite::call inline (hot)
@ 2
org.jruby.runtime.callsite.CachingCallSite::pollAndGetClass inline (hot)
@ 1
org.jruby.runtime.ThreadContext::callThreadPoll inline (hot)
@ 19
org.jruby.runtime.ThreadContext::pollThreadEvents executed <
MinInliningThreshold times
@ 5 org.jruby.RubyBasicObject::getMetaClass
inline (hot)
@ 17
org.jruby.runtime.callsite.CacheEntry::typeOk inline (hot)
@ 5 org.jruby.RubyModule::getCacheToken inline
(hot)
@ 38 org.jruby.RubyFixnum$i$1$0$op_minus::call
inline (hot)
@ 38 org.jruby.RubyFixnum$i$1$0$op_lt::call
inline (hot)
You can see that at @ 28 the inokeExact starts another round of inlining
bench.bench_fib_recursive::method__0$RUBY$fib_ruby. And that repeats until the
inlining bails out.
-- Christian
_______________________________________________
mlvm-dev mailing list
[email protected]
http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
