Improved performance of list iterator (- 10-20%) #441

ritchie46 · 2021-09-23T08:36:48Z

This PR elides some of the bounds checks done while iteration list values. The slicing still does bound checks and IMO should be one of the invariants of the offsets (being in bounds). I want to propose a slice_unchecked on the Array trait. Let me know what you think.

This optimization might also work for FixedSizeList if we add a value_unchecked to the array.

Gnuplot not found, using plotters backend
list: iter_values 2^10  time:   [15.214 us 15.217 us 15.220 us]                                    
                        change: [-12.125% -12.063% -11.995%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 11 outliers among 100 measurements (11.00%)
  1 (1.00%) low severe
  2 (2.00%) high mild
  8 (8.00%) high severe

list: iter 2^10         time:   [15.274 us 15.277 us 15.280 us]                             
                        change: [-12.382% -12.065% -11.801%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  1 (1.00%) high mild
  1 (1.00%) high severe

list: iter_values 2^12  time:   [60.861 us 60.869 us 60.878 us]                                   
                        change: [-11.760% -9.6523% -7.8631%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 11 outliers among 100 measurements (11.00%)
  1 (1.00%) low severe
  7 (7.00%) high mild
  3 (3.00%) high severe

list: iter 2^12         time:   [61.928 us 61.995 us 62.076 us]                            
                        change: [-5.0382% -4.4427% -3.8211%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  1 (1.00%) high mild
  1 (1.00%) high severe

list: iter_values 2^14  time:   [244.38 us 244.41 us 244.44 us]                                   
                        change: [-8.0274% -7.9989% -7.9730%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  1 (1.00%) low severe
  1 (1.00%) high mild
  5 (5.00%) high severe

list: iter 2^14         time:   [245.20 us 245.23 us 245.26 us]                            
                        change: [-21.663% -18.963% -16.142%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 4 outliers among 100 measurements (4.00%)
  1 (1.00%) high mild
  3 (3.00%) high severe

Benchmarking list: iter_values 2^16: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 5.4s, enable flat sampling, or reduce sample count to 60.
list: iter_values 2^16  time:   [1.0023 ms 1.0034 ms 1.0051 ms]                                    
                        change: [-3.7442% -3.3126% -2.7332%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 14 outliers among 100 measurements (14.00%)
  5 (5.00%) high mild
  9 (9.00%) high severe

Benchmarking list: iter 2^16: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 5.2s, enable flat sampling, or reduce sample count to 60.
list: iter 2^16         time:   [1.0243 ms 1.0254 ms 1.0266 ms]                             
                        change: [-0.3685% -0.1351% +0.1820%] (p = 0.35 > 0.05)
                        No change in performance detected.
Found 19 outliers among 100 measurements (19.00%)
  8 (8.00%) low mild
  8 (8.00%) high mild
  3 (3.00%) high severe

list: iter_values 2^18  time:   [3.9003 ms 3.9005 ms 3.9008 ms]                                    
                        change: [-11.058% -10.852% -10.638%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  1 (1.00%) low mild
  2 (2.00%) high mild
  4 (4.00%) high severe

list: iter 2^18         time:   [3.9241 ms 3.9244 ms 3.9248 ms]                             
                        change: [-6.2696% -6.1584% -6.0561%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 4 outliers among 100 measurements (4.00%)
  1 (1.00%) high mild
  3 (3.00%) high severe

list: iter_values 2^20  time:   [15.604 ms 15.606 ms 15.607 ms]                                   
                        change: [-7.4029% -7.2619% -7.1323%] (p = 0.00 < 0.05)
                        Performance has improved.

list: iter 2^20         time:   [15.923 ms 16.046 ms 16.204 ms]                            
                        change: [-4.1817% -3.4015% -2.4924%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 5 outliers among 100 measurements (5.00%)

codecov · 2021-09-23T08:49:37Z

Codecov Report

Merging #441 (1a1425d) into main (688e979) will decrease coverage by 0.14%.
The diff coverage is 78.07%.

@@            Coverage Diff             @@
##             main     #441      +/-   ##
==========================================
- Coverage   80.77%   80.63%   -0.15%     
==========================================
  Files         372      372              
  Lines       22637    22726      +89     
==========================================
+ Hits        18286    18324      +38     
- Misses       4351     4402      +51

Impacted Files	Coverage Δ
src/array/mod.rs	`55.55% <ø> (-3.42%)`	⬇️
src/array/null.rs	`35.71% <0.00%> (-2.75%)`	⬇️
src/array/union/mod.rs	`74.31% <0.00%> (-6.69%)`	⬇️
src/array/fixed_size_binary/mod.rs	`45.33% <70.00%> (+2.04%)`	⬆️
src/array/list/iterator.rs	`62.50% <75.00%> (ø)`
src/array/fixed_size_list/mod.rs	`51.42% <76.92%> (+3.97%)`	⬆️
src/array/list/mod.rs	`69.76% <83.33%> (-2.03%)`	⬇️
src/array/struct_.rs	`34.06% <88.88%> (-3.29%)`	⬇️
src/array/binary/mod.rs	`77.77% <90.00%> (-1.91%)`	⬇️
src/array/boolean/mod.rs	`67.34% <90.00%> (+3.93%)`	⬆️
... and 11 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 688e979...1a1425d. Read the comment docs.

jorgecarleitao · 2021-09-23T17:28:53Z

Thanks!

Yeap, unchecked slicing makes sense 👍

benches/iter_list.rs

Also adds a slice unchecked and dispatches slice method to the unchecked. This also reduces a double assert to a single assert in safe code. As the bounds were checked at the array buffer and at the validity.

ritchie46 · 2021-09-23T19:04:08Z

@jorgecarleitao I also added the unchecked_slice here. This reduces the bounds checking in safe code from two checks to a single one.

Now the bounds are asserted array::slice and the implementation is dispatched to the unchecked function.
Before the bounds were asserted in buffer::slice and in validity::slice.

jorgecarleitao

Went through it carefully and all checks great. Left some nits to apply, but it is good to go.

Thanks a lot, @ritchie46 ! 💯

src/array/dictionary/mod.rs

src/buffer/immutable.rs

src/bitmap/immutable.rs

src/buffer/immutable.rs

src/array/fixed_size_list/mod.rs

src/array/fixed_size_binary/mod.rs

src/array/boolean/mod.rs

src/array/binary/mod.rs

Co-authored-by: Jorge Leitao <[email protected]>

jorgecarleitao reviewed Sep 23, 2021

View reviewed changes

benches/iter_list.rs Outdated Show resolved Hide resolved

ritchie46 force-pushed the improve_list_iterator branch from 16ff535 to 4ed6808 Compare September 23, 2021 19:00

Elide bounds checks in (fixedsize)list-iter

189b424

Also adds a slice unchecked and dispatches slice method to the unchecked. This also reduces a double assert to a single assert in safe code. As the bounds were checked at the array buffer and at the validity.

ritchie46 force-pushed the improve_list_iterator branch from 4ed6808 to 189b424 Compare September 23, 2021 19:02

jorgecarleitao reviewed Sep 23, 2021

View reviewed changes

Apply suggestions from code review

1a1425d

Co-authored-by: Jorge Leitao <[email protected]>

jorgecarleitao merged commit fa11752 into jorgecarleitao:main Sep 25, 2021

jorgecarleitao changed the title ~~elide partial bounds check in list iterator~~ Improved performance of list iterator (-10%) Sep 25, 2021

jorgecarleitao added the enhancement An improvement to an existing feature label Sep 25, 2021

jorgecarleitao changed the title ~~Improved performance of list iterator (-10%)~~ Improved performance of list iterator (- 10-20%) Sep 25, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Improved performance of list iterator (- 10-20%) #441

Improved performance of list iterator (- 10-20%) #441

ritchie46 commented Sep 23, 2021

codecov bot commented Sep 23, 2021 •

edited

Loading

jorgecarleitao commented Sep 23, 2021

ritchie46 commented Sep 23, 2021

jorgecarleitao left a comment

Improved performance of list iterator (- 10-20%) #441

Improved performance of list iterator (- 10-20%) #441

Conversation

ritchie46 commented Sep 23, 2021

codecov bot commented Sep 23, 2021 • edited Loading

Codecov Report

jorgecarleitao commented Sep 23, 2021

ritchie46 commented Sep 23, 2021

jorgecarleitao left a comment

Choose a reason for hiding this comment

codecov bot commented Sep 23, 2021 •

edited

Loading