From: funny.falcon@... Date: 2016-03-07T05:42:25+00:00 Subject: [ruby-core:74192] [Ruby trunk Feature#12142] Hash tables with open addressing Issue #12142 has been updated by Yura Sokolov. File 0001-st.c-use-array-for-storing-st_table_entry.patch added First: `make benchmark-each ITEM=bm_hash OPTS='-r 3 -v' COMPARE_RUBY=''` is broken :-( it shows speedup even if compares with same ruby after `make install`. I suppose, it measures startup time, which increases after install, or it measures miniruby vs ruby. As you will see, only creating huge Hashes are improved by both mine and Vladimir's patches. (ok, Vladimir's one also has better iteration time) Second: I've made my version of patch: - closed addressing + double linked list for insertion order - 32bit indices instead of pointers and 32bit hashes - some tricks to reduce size of st_table, and fit array into sizes comfortable for allocators (patch attached, and also here https://github.com/ruby/ruby/compare/trunk...funny-falcon:st_table_with_array_storage.patch ) First number - trunk, second - Vladimirs, third - mine ```` hash_aref_dsym 0.558 0.556 0.563 hash_aref_dsym_long 12.049 7.784 6.841 hash_aref_fix 0.592 0.621 0.602 hash_aref_flo 0.15 0.13 0.135 hash_aref_miss 0.804 0.772 0.806 hash_aref_str 0.738 0.725 0.748 hash_aref_sym 0.551 0.552 0.556 hash_aref_sym_long 0.792 0.777 0.785 hash_flatten 0.546 0.498 0.546 hash_ident_flo 0.113 0.11 0.114 hash_ident_num 0.533 0.564 0.548 hash_ident_obj 0.523 0.529 0.53 hash_ident_str 0.527 0.529 0.543 hash_ident_sym 0.558 0.552 0.561 hash_keys 0.533 0.311 0.423 hash_shift 0.08 0.077 0.075 hash_shift_u16 0.176 0.202 0.148 hash_shift_u24 0.175 0.212 0.143 hash_shift_u32 0.172 0.203 0.143 hash_to_proc 0.066 0.059 0.064 hash_values 0.46 0.309 0.441 vm2_bighash 6.196 3.555 2.524 ```` Memory usage is lesser than Vladimir's, and comparable to trunk (sometimes lesser than trunk, sometimes bigger). Usage may be bigger cause array is preallocated with room of empty elements. My patch has ability to specify more intermediate steps of increasing capacities, but allocator behavior should be considered. So, as I said, main advantage is from storing `st_table_entry` as an array, so less calls to `malloc` performed, less `TLB` misses and less memory used. Open addressing gives almost nothing to performance in this case, cause it is not a case where open addressing plays well. Open addressing plays well when you whole key-value structure is small and stored inside of hash-array. But in case of Ruby's Hash we store st_table_entry outside of open-addressing array, so jump is performed, and main benefit (cache locality) is lost. Vladimir's proposal for storing insertion order by position in array can still benefit in memory usage, if carefully designed. By the way, PHP's array do exactly this (but uses closed addressing). Also, Vladimir's patch has better time for iterating Hash (hash_flatten, hash_keys and hash_values benchmarks). ---------------------------------------- Feature #12142: Hash tables with open addressing https://bugs.ruby-lang.org/issues/12142#change-57333 * Author: Vladimir Makarov * Status: Open * Priority: Normal * Assignee: ---------------------------------------- ~~~ Hello, the following patch contains a new implementation of hash tables (major files st.c and include/ruby/st.h). Modern processors have several levels of cache. Usually,the CPU reads one or a few lines of the cache from memory (or another level of cache). So CPU is much faster at reading data stored close to each other. The current implementation of Ruby hash tables does not fit well to modern processor cache organization, which requires better data locality for faster program speed. The new hash table implementation achieves a better data locality mainly by o switching to open addressing hash tables for access by keys. Removing hash collision lists lets us avoid *pointer chasing*, a common problem that produces bad data locality. I see a tendency to move from chaining hash tables to open addressing hash tables due to their better fit to modern CPU memory organizations. CPython recently made such switch (https://hg.python.org/cpython/file/ff1938d12240/Objects/dictobject.c). PHP did this a bit earlier https://nikic.github.io/2014/12/22/PHPs-new-hashtable-implementation.html. GCC has widely-used such hash tables (https://gcc.gnu.org/svn/gcc/trunk/libiberty/hashtab.c) internally for more than 15 years. o removing doubly linked lists and putting the elements into an array for accessing to elements by their inclusion order. That also removes pointer chaising on the doubly linked lists used for traversing elements by their inclusion order. A more detailed description of the proposed implementation can be found in the top comment of the file st.c. The new implementation was benchmarked on 21 MRI hash table benchmarks for two most widely used targets x86-64 (Intel 4.2GHz i7-4790K) and ARM (Exynos 5410 - 1.6GHz Cortex-A15): make benchmark-each ITEM=bm_hash OPTS='-r 3 -v' COMPARE_RUBY='' Here the results for x86-64: hash_aref_dsym 1.094 hash_aref_dsym_long 1.383 hash_aref_fix 1.048 hash_aref_flo 1.860 hash_aref_miss 1.107 hash_aref_str 1.107 hash_aref_sym 1.191 hash_aref_sym_long 1.113 hash_flatten 1.258 hash_ident_flo 1.627 hash_ident_num 1.045 hash_ident_obj 1.143 hash_ident_str 1.127 hash_ident_sym 1.152 hash_keys 2.714 hash_shift 2.209 hash_shift_u16 1.442 hash_shift_u24 1.413 hash_shift_u32 1.396 hash_to_proc 2.831 hash_values 2.701 The average performance improvement is more 50%. ARM results are analogous -- no any benchmark performance degradation and about the same average improvement. The patch can be seen as https://github.com/vnmakarov/ruby/compare/trunk...hash_tables_with_open_addressing.patch or in a less convenient way as pull request changes https://github.com/ruby/ruby/pull/1264/files This is my first patch for MRI and may be my proposal and implementation have pitfalls. But I am keen to learn and work on inclusion of this code into MRI. ~~~ ---Files-------------------------------- 0001-st.c-use-array-for-storing-st_table_entry.patch (46.7 KB) -- https://bugs.ruby-lang.org/ Unsubscribe: