Misc #15800
closed
Reduce ONIG_NREGION from 10 to 4: power of 2 and testing revealed most pattern matches are less than or equal to 4 results
Description
References PR https://github.com/ruby/ruby/pull/2135 - it's a very small change, but runnin due diligence past the list too for discussion.
I noticed onig_region_resize (called from onig_region_copy) would default to allocating a 10 * 8 bytes block on 64bit for both the beg and end members of OnigRegion.
Preliminary testing with Rails and the benchmark suite suggests that most pattern matches are <= 4 results.
Due diligence with debug counters¶
Few requests on a blank redmine instance:
[RUBY_DEBUG_COUNTER] obj_match_under4 10650 <<<<<<<<<<
[RUBY_DEBUG_COUNTER] obj_match_ge4 1589 <<<<<<<<<<
[RUBY_DEBUG_COUNTER] obj_match_ge8 66
[RUBY_DEBUG_COUNTER] obj_match_ptr 12305
single match 1000000.times { 'haystack'.match(/hay/) }
[RUBY_DEBUG_COUNTER] obj_match_under4 999366 <<<<<<<<<<
[RUBY_DEBUG_COUNTER] obj_match_ge4 473 <<<<<<<<<<
[RUBY_DEBUG_COUNTER] obj_match_ge8 0
[RUBY_DEBUG_COUNTER] obj_match_ptr 999839
multiple matches > 4 1000000.times { /(.)(.)(\d+)(\d)/.match("THX1138.") }
[RUBY_DEBUG_COUNTER] obj_match_under4 353 <<<<<<<<<<
[RUBY_DEBUG_COUNTER] obj_match_ge4 997579 <<<<<<<<<<
[RUBY_DEBUG_COUNTER] obj_match_ge8 0
[RUBY_DEBUG_COUNTER] obj_match_ptr 997932
Memory and ips benchmarks, MatchData specific¶
lourens@CarbonX1:~/src/ruby/ruby$ /usr/local/bin/ruby --disable=gems -rrubygems -I./benchmark/lib ./benchmark/benchmark-driver/exe/benchmark-driver --executables="compare-ruby::~/src/ruby/trunk/ruby --disable=gems -I.ext/common --disable-gem" --executables="built-ruby::./miniruby -I./lib -I. -I.ext/common -r./prelude --disable-gem" -v --repeat-count=24 -r memory $(ls ./benchmark/*match*.{yml,rb} 2>/dev/null)
compare-ruby: ruby 2.7.0dev (2019-04-19 trunk 67619) [x86_64-linux]
built-ruby: ruby 2.7.0dev (2019-04-19 reduce-onig-de.. 67619) [x86_64-linux]
last_commit=Reduce ONIG_NREGION from 10 to 4: power of 2 and testing revealed most pattern matches are less than or equal to 4 results
Calculating -------------------------------------
compare-ruby built-ruby
match_gt4 11.936M 11.600M bytes - 1.000 times
match_small 11.848M 11.608M bytes - 1.000 times
Comparison:
match_gt4
built-ruby: 11600000.0 bytes
compare-ruby: 11936000.0 bytes - 1.03x larger
match_small
built-ruby: 11608000.0 bytes
compare-ruby: 11848000.0 bytes - 1.02x larger
lourens@CarbonX1:~/src/ruby/ruby$ /usr/local/bin/ruby --disable=gems -rrubygems -I./benchmark/lib ./benchmark/benchmark-driver/exe/benchmark-driver --executables="compare-ruby::~/src/ruby/trunk/ruby --disable=gems -I.ext/common --disable-gem" --executables="built-ruby::./miniruby -I./lib -I. -I.ext/common -r./prelude --disable-gem" -v --repeat-count=24 -r ips $(ls ./benchmark/*match*.{yml,rb} 2>/dev/null)
compare-ruby: ruby 2.7.0dev (2019-04-19 trunk 67619) [x86_64-linux]
built-ruby: ruby 2.7.0dev (2019-04-19 reduce-onig-de.. 67619) [x86_64-linux]
last_commit=Reduce ONIG_NREGION from 10 to 4: power of 2 and testing revealed most pattern matches are less than or equal to 4 results
Calculating -------------------------------------
compare-ruby built-ruby
match_gt4 1.664 1.754 i/s - 1.000 times in 0.600793s 0.570031s
match_small 1.856 2.047 i/s - 1.000 times in 0.538838s 0.488407s
Comparison:
match_gt4
built-ruby: 1.8 i/s
compare-ruby: 1.7 i/s - 1.05x slower
match_small
built-ruby: 2.0 i/s
compare-ruby: 1.9 i/s - 1.10x slower
I am fine with removing the debug counters and committed them for now as it's easier for reviewers to also reproduce locally.
For additional context I noticed that character offsets are bounded by the num_regs member as per https://github.com/ruby/ruby/blob/trunk/re.c#L989-L1005 and therefore investigated converging allocated and num_regs to be less divergent for the common cases
And some more of the 80 byte allocs from strscan with only the first chunk referenced:
==24182== -------------------- 283 of 1000 --------------------
==24182== max-live: 19,520 in 244 blocks
==24182== tot-alloc: 30,480 in 381 blocks (avg size 80.00)
==24182== deaths: 381, at avg age 423,950,747 (3.96% of prog lifetime)
==24182== acc-ratios: 1.95 rd, 4.98 wr (59,728 b-read, 151,920 b-written)
==24182== at 0x4C2DECF: malloc (in /usr/lib/valgrind/vgpreload_exp-dhat-amd64-linux.so)
==24182== by 0x2561E6: onig_region_resize (regexec.c:260)
==24182== by 0x2561E6: onig_region_resize_clear (regexec.c:298)
==24182== by 0x2561E6: onig_match (regexec.c:3882)
==24182== by 0xA4C376B: strscan_do_scan (strscan.c:472)
==24182== by 0xA4C376B: strscan_skip (strscan.c:570)
==24182== by 0x2E5B4E: vm_call_cfunc_with_frame (vm_insnhelper.c:2207)
==24182== by 0x2E5B4E: vm_call_cfunc (vm_insnhelper.c:2225)
==24182==
==24182== Aggregated access counts by offset:
==24182==
==24182== [ 0] 26456 26456 26456 26456 26456 26456 26456 26456 0 0 0 0 0 0 0 0
==24182== [ 16] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <<<<<<<<<<
==24182== [ 32] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <<<<<<<<<<
==24182== [ 48] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <<<<<<<<<<
==24182== [ 64] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <<<<<<<<<<
Updated by 
Updated by