Performance of Scope.getSymbolsByName()

Jan Lahoda jan.lahoda at oracle.com
Thu Jun 6 14:35:49 UTC 2019


Hi Ron,

I think we should run it through the perf tests, just to be sure (and I 
didn't have time for that yet), but otherwise, I think that simple 
change looks useful to me.

Thanks,
     Jan

On 06. 06. 19 3:41, Ron Shapiro wrote:
> Friendly ping - is the first webrev ready for submission?
> 
> On Thu, May 30, 2019, 7:09 PM Ron Shapiro <ronshapiro at google.com 
> <mailto:ronshapiro at google.com>> wrote:
> 
>     Do you have any other comments on the first webrev? Is there
>     anything else we need to do to submit that?
> 
>     On Wed, May 22, 2019 at 11:00 AM Ron Shapiro <ronshapiro at google.com
>     <mailto:ronshapiro at google.com>> wrote:
> 
>         I think there still would be benefit in that as well, as I'm
>         seeing that come up in other contexts (as referenced in the bug).
> 
>         On Wed, May 22, 2019 at 9:06 AM Jan Lahoda
>         <jan.lahoda at oracle.com <mailto:jan.lahoda at oracle.com>> wrote:
> 
>             Sorry, I was too busy last few days.
> 
>             I was peeking at some possible improvements, but I think I
>             like Ron's
>             first (.00) patch - that caches what can be cached nicely.
> 
>             Looking at the testcase generated by Ron's reproducer using:
>             python test.py 7
> 
>             and the (biggest) size of the outcome of:
>             types.membersClosure(site, false).getSymbolsByName(sym.name
>             <http://sym.name>, cf)
> 
>             seems to be 13700 elements - which means the Scope lookup
>             and iteration
>             runs ~13700 times, so avoiding these additional lookup costs
>             seems like
>             a clear win.
> 
>             I have an idea that might speed up the iterations through
>             deeply nested
>             CompoundScopes, although the effect of that in combination
>             with Ron's is
>             going to be fairly limited, if any, I think.
> 
>             Jan
> 
>             On 22. 05. 19 12:24, Maurizio Cimadamore wrote:
>              > This doesn't work. You are basically relying on the order
>             in which
>              > symbols are entered in the members closure scope.
>              >
>              > In simple example like these:
>              >
>              > class A {
>              >      int m(List<String> l) {return 0;}
>              > }
>              >
>              > class B extends A {
>              >     int m(List<Integer> l) {return 0;}
>              > }
>              >
>              >
>              > The logic you proposed will not work. That's because we
>             first see B::m -
>              > and 'symbolByName' is empty at that stage; so we add it
>             there. Then we
>              > do another round and see A::m - but we don't really look
>             there - given
>              > that we first check to see if the symbol we're
>             considering (sym) is
>              > override-equivalent with B::m (the only symbol in
>             symbolByName). And
>              > that happens to be the case, since they are the same
>             symbol. So we exit
>              > the loop, w/o having found any clash.
>              >
>              > In other words, symbolByName would need to also contain
>             A::m for the
>              > code to see the clash - but that never happens; by the
>             time A::m is
>              > added, is already too late.
>              >
>              >
>              > I think caching the result of
>              >
>              > types.membersClosure(site,
>             false).getSymbolsByName(sym.name <http://sym.name>, cf)
>              >
>              > is a good measure.
>              >
>              > I'm a bit surprised that iteration is so slow
>             (membersClosure is slow to
>              > set up, but once you do it the results are cached). So,
>             rather than
>              > tweaking the algorithm, I think it'd be better to
>             investigate the reason
>              > was to why asking a compound scope iterator is so slow,
>             which then would
>              > yield dividends for the rest of the code as well.
>              >
>              > Maurizio
>              >
>              >
>              > On 21/05/2019 21:21, Maurizio Cimadamore wrote:
>              >>
>              >> I see what you have done - I have to think about it a
>             bit to see if I
>              >> can come up with some counter example.
>              >>
>              >> Thanks
>              >> Maurizio
>              >>
>              >> On 21/05/2019 17:39, Ron Shapiro wrote:
>              >>> Are the checks of the inner loop symmetrical?
>              >>>
>              >>> Currently it's checking m_i against (m_0..n - m_i ).
>             This second
>              >>> webrev below would check it against just (m_0..i-1 ),
>             which albeit
>              >>> still n^2, it divides by a factor of 2.
>              >>>
>              >>> (sorry if the subscripting here doesn't display correctly)
>              >>>
>              >>> http://cr.openjdk.java.net/~ronsh/8224161/webrev.01/
>              >>>
>              >>> This feels conceptually logical to me, but I'm not
>             seeing a
>              >>> measurable change by it. It looks a little bit cleaner
>             to me, but I'm
>              >>> fine with either webrev given the benefits they both bring.
>              >>>
>              >>> I can take a look in another thread about speeding up
>             CompoundScope
>              >>> iteration.
>              >>>
>              >>> On Tue, May 21, 2019 at 8:05 AM Maurizio Cimadamore
>              >>> <maurizio.cimadamore at oracle.com
>             <mailto:maurizio.cimadamore at oracle.com>
>              >>> <mailto:maurizio.cimadamore at oracle.com
>             <mailto:maurizio.cimadamore at oracle.com>>> wrote:
>              >>>
>              >>>
>              >>>     On 21/05/2019 12:16, Ron Shapiro wrote:
>              >>>>     I still think that something to optimize the
>             actual ScopeImpl
>              >>>>     Iterable is a worthwhile endeavor, as that would
>             alleviate the
>              >>>>     need to materialize here (and solve hopefully the
>             other issues
>              >>>>     I'm seeing), but I was having trouble figuring out
>             how to do
>              >>>>     that. This may be a good interim option without
>             much cost.
>              >>>
>              >>>     Sure - I'm not opposed to optimizing the iteration
>             process - I
>              >>>     was expressing my skepticism w.r.t. making
>             checkOverrideClash
>              >>>     simpler/non quadratic.
>              >>>
>              >>>     Maurizio
>              >>>
>              >>>>
>              >>>>
>              >>>>
>              >>>>     On Tue, May 21, 2019, 5:59 AM Maurizio Cimadamore
>              >>>>     <maurizio.cimadamore at oracle.com
>             <mailto:maurizio.cimadamore at oracle.com>
>              >>>>     <mailto:maurizio.cimadamore at oracle.com
>             <mailto:maurizio.cimadamore at oracle.com>>> wrote:
>              >>>>
>              >>>>         I think your fix is a good one. We spent some
>             cycles
>              >>>>         optimizing this, a bit odd we have missed this :-)
>              >>>>
>              >>>>         I'm very skeptical you can collapse into a
>             single loop, as
>              >>>>         this implement the logic in JLS 8.4.8.3 [1]
>             which, as you
>              >>>>         can see, is inherently quadratic (for each
>             method, we have
>              >>>>         to scan all methods with same name in
>             supertypes to see if
>              >>>>         there is an override clash). The algorithm
>             that was there
>              >>>>         before wasn't - and it lead to the wrong
>             answers in tricky
>              >>>>         cases - so while you can get 80% there with a
>             non-quadratic
>              >>>>         algorithm, you will miss issues by doing so.
>              >>>>
>              >>>>         One thing that would help would be, instead,
>             to limit the
>              >>>>         analysis only in cases where it adds value -
>             for instance,
>              >>>>         if your hierarchy is just non-generic classes
>             (as in your
>              >>>>         example), then there's no way for you to
>             accidentally
>              >>>>         override a 'bridge' method, since no bridges
>             will be
>              >>>>         generated! But when looking at this, I
>             couldn't find great
>              >>>>         ways to detect these conditions w/o spending
>             more time than
>              >>>>         the check itself.
>              >>>>
>              >>>>         [1] -
>              >>>>
>             https://docs.oracle.com/javase/specs/jls/se12/html/jls-8.html#jls-8.4.8.3
>              >>>>
>              >>>>         Maurizio
>              >>>>
>              >>>>         On 20/05/2019 21:58, Ron Shapiro wrote:
>              >>>>>         In the real world example, I'm seeing the 40s
>             that was
>              >>>>>         previously spent in
>             Check.checkOverrideClashes drop to to
>              >>>>>         9.5s when using this patch. Of that 9.5, 8.9
>             is spent in
>              >>>>>         iterating through the CompoundIterator and
>             calling
>              >>>>>         getSymbolsByName.
>              >>>>>
>              >>>>>         On Mon, May 20, 2019 at 4:34 PM Ron Shapiro
>              >>>>>         <ronshapiro at google.com
>             <mailto:ronshapiro at google.com> <mailto:ronshapiro at google.com
>             <mailto:ronshapiro at google.com>>> wrote:
>              >>>>>
>              >>>>>             This patch, which materializes the
>             duplicate outer and
>              >>>>>             inner Iterables first into a list. It
>             removes the
>              >>>>>             entire section of the CompoundIterator
>             iteration from
>              >>>>>             the profile.
>              >>>>>
>              >>>>>             webrev:
>              >>>>>
>             http://cr.openjdk.java.net/~ronsh/8224161/webrev.00/src/jdk.compiler/share/classes/com/sun/tools/javac/comp/Check.java.sdiff.html
>              >>>>>
>              >>>>>             I'm not sure it's the absolutely correct
>             solution as it
>              >>>>>             possibly masks an underlying issue.
>              >>>>>
>              >>>>>             I'm still seeing some time spent in
>              >>>>>             MethodSymbol.overrides,
>             Types.isSubSignature, and
>              >>>>>             Types.memberType, all of which happen in
>             the inner
>              >>>>>             loop. If we can remove those and collapse
>             the nested
>              >>>>>             loops into one, then this solution isn't
>             necessary and
>              >>>>>             it would also solve that performance issue.
>              >>>>>
>              >>>>>             On Fri, May 17, 2019 at 5:55 PM Ron Shapiro
>              >>>>>             <ronshapiro at google.com
>             <mailto:ronshapiro at google.com> <mailto:ronshapiro at google.com
>             <mailto:ronshapiro at google.com>>>
>              >>>>>             wrote:
>              >>>>>
>              >>>>>                 I still have more to investigate to
>             fully wrap my
>              >>>>>                 head around it, but I finally found a
>             sample
>              >>>>>                 program that exhibits this. Filed a
>             bug here:
>              >>>>> https://bugs.openjdk.java.net/browse/JDK-8224161
>              >>>>>
>              >>>>>                 On Fri, May 17, 2019 at 11:21 AM Jan
>             Lahoda
>              >>>>>                 <jan.lahoda at oracle.com
>             <mailto:jan.lahoda at oracle.com>
>              >>>>>                 <mailto:jan.lahoda at oracle.com
>             <mailto:jan.lahoda at oracle.com>>> wrote:
>              >>>>>
>              >>>>>                     Hi Ron,
>              >>>>>
>              >>>>>                     I am afraid it is hard to guess
>             what is the
>              >>>>>                     problem without some
>              >>>>>                     testcase. So, at least to me,
>             having a sample
>              >>>>>                     would be helpful.
>              >>>>>
>              >>>>>                     Thanks,
>              >>>>>                          Jan
>              >>>>>
>              >>>>>                     On 17. 05. 19 0:41, Ron Shapiro
>             wrote:
>              >>>>>                     > Hi,
>              >>>>>                     >
>              >>>>>                     > I'm observing a particularly
>             bizarre
>              >>>>>                     compilation. It's a single file
>              >>>>>                     > with annotation processing, and
>             the type that
>              >>>>>                     is being compiled and
>              >>>>>                     > processed has ~1000 declared
>             and inherited
>              >>>>>                     methods combined. The total
>              >>>>>                     > compilation is 3 minutes, but
>             65% of the
>              >>>>>                     entire compilation is spent in
>              >>>>>                     > 3 methods:
>              >>>>>                     >
>              >>>>>                   
>               Check.checkOverrideClashes(), Resolve.findInheritedMemberType(),
>              >>>>>
>              >>>>>                     > and Resolve.findField().
>              >>>>>                     >
>              >>>>>                     > Looking at profiles, it looks like
>              >>>>>                     getSymbolsByName() is the major
>              >>>>>                     > culprit here. I initially
>             thought the reason
>              >>>>>                     was that there were far too
>              >>>>>                     > many overloads (this type had >600
>              >>>>>                     overloads...) and that that was
>              >>>>>                     > causing a bad regression for the
>              >>>>>                     pseudo-hashmap that ScopeImpl uses.
>              >>>>>                     > However, renaming the methods
>             did not
>              >>>>>                     alleviate the build pain and these
>              >>>>>                     > methods continue to be taking
>             long amounts of
>              >>>>>                     time.
>              >>>>>                     >
>              >>>>>                     > I was wondering what could be
>             done to improve
>              >>>>>                     the performance of this
>              >>>>>                     > code. It seemed to me that
>             something like a
>              >>>>>                     Map<Name, List<Symbol>>
>              >>>>>                     > could be a reasonable+modern
>             replacement for
>              >>>>>                     this table, which would
>              >>>>>                     > naturally have a fast
>             getSymbolsForName()
>              >>>>>                     implementation. I'm having
>              >>>>>                     > some trouble implementing it
>             correctly, and I
>              >>>>>                     believe it's partially
>              >>>>>                     > related to not fully
>             understanding some of
>              >>>>>                     the semantics of the class.
>              >>>>>                     >
>              >>>>>                     > Does what I wrote make sense to
>             anyone, and
>              >>>>>                     maybe spark a lightbulb?
>              >>>>>                     >
>              >>>>>                     > I'm trying to put together a
>             repro in case
>              >>>>>                     that helps, but I'm not 100%
>              >>>>>                     > sure I even understand what the
>             regression
>              >>>>>                     case is.
>              >>>>>                     >
>              >>>>>                     > Thanks for you help!
>              >>>>>
> 


More information about the compiler-dev mailing list