RESOLVED FIXED 56913
Mapping booleans the same way as integers 
https://bugs.webkit.org/show_bug.cgi?id=56913
Summary Mapping booleans the same way as integers
Zoltan Herczeg
Reported 2011-03-23 03:54:58 PDT
There was a discussion about this in the SquirrelFish mailing list. In short, the booleans store their logical values in the payload part when using JSValue32_64 representation. Performance: TEST COMPARISON FROM TO DETAILS ============================================================================= ** TOTAL **: 1.007x as fast 501.7ms +/- 0.3% 498.3ms +/- 0.3% significant ============================================================================= 3d: ?? 68.7ms +/- 0.9% 68.9ms +/- 1.1% not conclusive: might be *1.003x as slow* cube: - 21.5ms +/- 1.8% 21.5ms +/- 1.8% morph: - 21.9ms +/- 1.0% 21.8ms +/- 1.4% raytrace: ?? 25.3ms +/- 1.4% 25.6ms +/- 2.4% not conclusive: might be *1.012x as slow* access: - 59.5ms +/- 0.8% 59.5ms +/- 0.6% binary-trees: - 7.5ms +/- 6.7% 7.5ms +/- 5.0% fannkuch: - 24.0ms +/- 0.0% 24.0ms +/- 0.0% nbody: - 20.0ms +/- 0.0% 20.0ms +/- 0.0% nsieve: - 8.0ms +/- 0.0% 8.0ms +/- 0.0% bitops: 1.031x as fast 33.0ms +/- 0.0% 32.0ms +/- 0.0% significant 3bit-bits-in-byte: - 6.0ms +/- 0.0% 6.0ms +/- 0.0% bits-in-byte: 1.111x as fast 10.0ms +/- 0.0% 9.0ms +/- 0.0% significant bitwise-and: - 6.0ms +/- 0.0% 6.0ms +/- 0.0% nsieve-bits: - 11.0ms +/- 0.0% 11.0ms +/- 0.0% controlflow: - 4.0ms +/- 0.0% 4.0ms +/- 0.0% recursive: - 4.0ms +/- 0.0% 4.0ms +/- 0.0% crypto: - 31.1ms +/- 0.7% 31.0ms +/- 0.0% aes: - 19.1ms +/- 1.2% 19.0ms +/- 0.0% md5: - 7.0ms +/- 0.0% 7.0ms +/- 0.0% sha1: - 5.0ms +/- 0.0% 5.0ms +/- 0.0% date: 1.031x as fast 62.4ms +/- 1.0% 60.5ms +/- 1.1% significant format-tofte: 1.040x as fast 31.4ms +/- 1.2% 30.2ms +/- 1.5% significant format-xparb: 1.023x as fast 31.0ms +/- 1.5% 30.3ms +/- 1.6% significant math: - 51.6ms +/- 0.7% 51.2ms +/- 0.6% cordic: - 16.0ms +/- 0.0% 16.0ms +/- 0.0% partial-sums: - 23.1ms +/- 1.0% 23.0ms +/- 0.0% spectral-norm: - 12.5ms +/- 3.0% 12.2ms +/- 2.5% regexp: 1.023x as fast 22.0ms +/- 0.0% 21.5ms +/- 1.8% significant dna: 1.023x as fast 22.0ms +/- 0.0% 21.5ms +/- 1.8% significant string: ?? 169.4ms +/- 0.4% 169.7ms +/- 0.4% not conclusive: might be *1.002x as slow* base64: ?? 17.0ms +/- 0.0% 17.1ms +/- 1.3% not conclusive: might be *1.006x as slow* fasta: - 24.0ms +/- 0.0% 23.7ms +/- 1.5% tagcloud: - 40.6ms +/- 1.2% 40.2ms +/- 0.7% unpack-code: *1.017x as slow* 65.0ms +/- 0.0% 66.1ms +/- 0.6% significant validate-input: - 22.8ms +/- 1.3% 22.6ms +/- 1.6%
Attachments
patch (15.12 KB, patch)
2011-03-23 03:59 PDT, Zoltan Herczeg 		barraclough: review+
DetailsFormatted DiffDiff
View All Add attachment proposed patch, testcase, etc.
Zoltan Herczeg
Comment 1 2011-03-23 03:59:58 PDT
Created attachment 86600 [details] patch
Zoltan Herczeg
Comment 2 2011-03-23 04:19:37 PDT
TEST COMPARISON FROM TO DETAILS ============================================================================= ** TOTAL **: 1.016x as fast 3496.6ms +/- 1.3% 3442.0ms +/- 0.7% significant ============================================================================= v8: 1.016x as fast 3496.6ms +/- 1.3% 3442.0ms +/- 0.7% significant crypto: - 391.4ms +/- 0.2% 391.7ms +/- 0.1% deltablue: 1.013x as fast 651.1ms +/- 0.7% 642.8ms +/- 0.6% significant earley-boyer: 1.045x as fast 563.4ms +/- 1.7% 539.0ms +/- 1.8% significant raytrace: - 204.3ms +/- 17.1% 192.0ms +/- 14.2% regexp: ?? 355.2ms +/- 0.5% 355.8ms +/- 0.7% not conclusive: might be *1.002x as slow* richards: 1.017x as fast 635.9ms +/- 0.5% 625.1ms +/- 0.5% significant splay: - 695.3ms +/- 0.4% 695.6ms +/- 0.4%
Zoltan Herczeg
Comment 3 2011-03-23 04:23:34 PDT
(measured on an AMD x86-32 system)
Gavin Barraclough
Comment 4 2011-04-07 16:49:29 PDT
Hey Zoltan, looks great, tidy little perf win!, very nice. Fits nicely with my change for JSVALUE64 (no noticeable improvement there, but good to remove the pointless shift). Hope you don't mind, took the liberty to land this because I wanted the documentation of the immediate formats up to date, fixed in r83230.
Note You need to log in before you can comment on or make changes to this bug.