On Thursday, February 13, 2020 at 7:02:37 PM UTC+1, Robert Engels wrote: > > @atom ... I changed the cpp code to use unordered_map (cc and h) and it > made the code slower... (iteration over most hash map implementations is > slower than sorted maps since often you need to skip the unused key slots). >
It is more likely that if hash-map iteration is slower than sorted-map it is because of an increase in LLC (last-level cache) misses. > So I would be very curious to see your code changes where this brought the > performance to the level of Java ? > It depends on the CPU used to run the benchmarks. In particular, a large L3 cache seems to help a lot. On a notebook CPU with 2 MB of LLC, x86-64 mode, the performance with std::unordered_map (849e6 LLC misses) indeed isn't better than std::map (605e6 misses). However, on a desktop CPU with 32 MB of LLC, x86-64 mode, std::unordered_map has (if I am reading it correctly) 2.0e6 misses and std::map has 2.2e6 misses. Changing the compiler target to x86-32 (gcc switch: -m32) improves performance on both CPUs. The complete git-diff in directory multi-language-bench/src/havlak/cpp, except #include directives, is: - typedef std::map<BasicBlock*, int> BasicBlockMap; + typedef std::unordered_map<BasicBlock*, int> BasicBlockMap; - typedef std::map<int, BasicBlock*> NodeMap; + typedef std::unordered_map<int, BasicBlock*> NodeMap; > > This is not my experience. Did you actually do this? > It is probable some CPU performance counter can explain the difference in our experiences. > > > -----Original Message----- > From: Robert Engels > Sent: Feb 13, 2020 11:33 AM > To: ⚛ <0xe2.0...@gmail.com <javascript:>>, golang-nuts > Subject: Re: [go-nuts] Go without garbage collector > > I won't dispute that, but at least this particular case, it requires > on-going maintenance by the developer (company). In this case of Java very > few performance improvements have required code changes. > > Using a different structure entirely (hash map vs. tree map) is a > different issue - it touches on the robustness of the stdlib, dynamic > runtime class replacement, etc. I can only assume that std::map was used > because it made the code easier/portable to write in C++ rather than adding > an outside dependency or custom hash map. implementation (In the Java code, > they use a custom IntegerSet, IntegerList to avoid boxing...) > > -----Original Message----- > From: ⚛ <0xe2.0...@gmail.com <javascript:>> > Sent: Feb 13, 2020 9:57 AM > To: golang-nuts > Subject: Re: [go-nuts] Go without garbage collector > > On Thursday, February 13, 2020 at 3:05:45 PM UTC+1, Robert Engels wrote: >> >> The code hasn’t been changed. The performance numbers have changed >> dramatically with no developer intervention. No “hand optimizing” required. >> > > C++ evolves over time also. Hashmaps have been added to C++ in C++11, > which from today's viewpoint invalidates most benchmarks published before > year 2011 that translated Java's HashMap to C++'s std::map. > > -- > You received this message because you are subscribed to the Google Groups > "golang-nuts" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to golan...@googlegroups.com <javascript:>. > To view this discussion on the web visit > https://groups.google.com/d/msgid/golang-nuts/f5a4a3cf-e4ce-4fb4-af03-fa5f32a5ce62%40googlegroups.com > > <https://groups.google.com/d/msgid/golang-nuts/f5a4a3cf-e4ce-4fb4-af03-fa5f32a5ce62%40googlegroups.com?utm_medium=email&utm_source=footer> > . > > > > > -- > You received this message because you are subscribed to the Google Groups > "golang-nuts" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to golan...@googlegroups.com <javascript:>. > To view this discussion on the web visit > https://groups.google.com/d/msgid/golang-nuts/1149226989.3184.1581615195727%40wamui-sophie.atl.sa.earthlink.net > > <https://groups.google.com/d/msgid/golang-nuts/1149226989.3184.1581615195727%40wamui-sophie.atl.sa.earthlink.net?utm_medium=email&utm_source=footer> > . > > > > > -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/golang-nuts/b4dadd10-e44c-458c-9d27-fbd3d7e6ba93%40googlegroups.com.
