157121 – [JSC] Unify Math.pow() accross all tiers

RESOLVED FIXED 157121

[JSC] Unify Math.pow() accross all tiers

https://bugs.webkit.org/show_bug.cgi?id=157121

Summary [JSC] Unify Math.pow() accross all tiers

Benjamin Poulain

Reported 2016-04-27 22:53:52 PDT

[JSC] Unify Math.pow() accross all tiers

Attachments
Patch (34.93 KB, patch) 2016-04-27 23:12 PDT, Benjamin Poulain	no flags	Details Formatted Diff Diff
Patch for landing (34.83 KB, patch) 2016-04-28 11:05 PDT, Benjamin Poulain	no flags	Details Formatted Diff Diff
Archive of layout-test-results from webkit-cq-03 for mac-yosemite (712.04 KB, application/zip) 2016-04-28 12:57 PDT, WebKit Commit Bot	no flags	Details
Show Obsolete (1) View All Add attachment proposed patch, testcase, etc.

Benjamin Poulain

Comment 1 2016-04-27 23:12:16 PDT

Created attachment 277596 [details] Patch

Benjamin Poulain

Comment 2 2016-04-27 23:24:11 PDT

~1% on Sunspider. The extra branches do not seem to affect anything else much. The full pow() is ridiculously slow, the few extra instructions are likely well amortized. Conf#1 Conf#2 SunSpider: 3d-cube 4.9506+-0.1445 ? 4.9932+-0.2252 ? 3d-morph 5.0139+-0.2810 4.9960+-0.1840 3d-raytrace 5.1359+-0.1969 ? 5.2358+-0.2849 ? might be 1.0195x slower access-binary-trees 2.1585+-0.1651 2.1107+-0.0095 might be 1.0226x faster access-fannkuch 5.6708+-0.6388 ? 5.8778+-0.6348 ? might be 1.0365x slower access-nbody 2.4800+-0.0116 ? 2.6102+-0.3954 ? might be 1.0525x slower access-nsieve 3.0219+-0.0840 3.0193+-0.0377 bitops-3bit-bits-in-byte 1.1414+-0.0584 1.1205+-0.0171 might be 1.0186x faster bitops-bits-in-byte 2.7360+-0.0404 2.7241+-0.0787 bitops-bitwise-and 2.0010+-0.0350 ? 2.1057+-0.1545 ? might be 1.0523x slower bitops-nsieve-bits 3.1046+-0.0643 ? 3.1172+-0.2807 ? controlflow-recursive 2.3163+-0.0644 ? 2.3289+-0.0888 ? crypto-aes 4.0208+-0.0619 4.0052+-0.0458 crypto-md5 2.5262+-0.3326 2.4612+-0.0498 might be 1.0264x faster crypto-sha1 2.3273+-0.0832 2.2680+-0.0685 might be 1.0261x faster date-format-tofte 6.6755+-0.6119 6.6704+-0.4869 date-format-xparb 4.5273+-0.0604 ? 4.6358+-0.2428 ? might be 1.0240x slower math-cordic 2.9507+-0.4597 2.8063+-0.0811 might be 1.0515x faster math-partial-sums 4.7396+-0.1258 ^ 3.6521+-0.1642 ^ definitely 1.2978x faster math-spectral-norm 2.0932+-0.3824 1.9695+-0.0330 might be 1.0628x faster regexp-dna 6.1671+-0.0361 ? 6.2686+-0.1767 ? might be 1.0164x slower string-base64 4.5476+-0.1204 ? 4.5972+-0.0732 ? might be 1.0109x slower string-fasta 5.6635+-0.0100 ? 5.6834+-0.0651 ? string-tagcloud 8.0394+-0.3858 7.9290+-0.0602 might be 1.0139x faster string-unpack-code 19.3922+-1.5464 19.3517+-1.6633 string-validate-input 4.5809+-0.7342 4.3083+-0.0684 might be 1.0633x faster <arithmetic> 4.5378+-0.0860 4.4941+-0.0769 might be 1.0097x faster Conf#1 Conf#2 Octane: encrypt 0.15838+-0.00072 0.15772+-0.00284 decrypt 2.81682+-0.03935 2.80042+-0.02260 deltablue x2 0.12986+-0.00310 0.12827+-0.00191 might be 1.0124x faster earley 0.28448+-0.00241 ? 0.28698+-0.00550 ? boyer 5.07231+-0.08016 5.03594+-0.10381 navier-stokes x2 4.93060+-0.02064 ? 4.95492+-0.05826 ? raytrace x2 0.78993+-0.00486 0.78717+-0.00452 richards x2 0.08233+-0.00187 0.08210+-0.00235 splay x2 0.33413+-0.00374 0.33177+-0.00705 regexp x2 15.41254+-0.60013 15.03767+-0.15336 might be 1.0249x faster pdfjs x2 39.16134+-0.62180 38.99780+-0.27454 mandreel x2 42.23144+-0.52180 41.86420+-0.18079 gbemu x2 24.01304+-0.19162 23.93201+-0.30454 closure 0.51968+-0.00444 0.51721+-0.00431 jquery 6.46755+-0.11489 6.44485+-0.05603 box2d x2 9.21245+-0.17873 ? 9.30133+-0.17537 ? zlib x2 356.78636+-2.08378 353.51636+-5.41979 typescript x2 613.39075+-10.82888 609.56531+-14.21036 <geometric> 4.93713+-0.02300 4.91222+-0.00804 might be 1.0051x faster Conf#1 Conf#2 Kraken: ai-astar 87.966+-5.488 86.503+-2.024 might be 1.0169x faster audio-beat-detection 39.605+-2.884 38.418+-0.079 might be 1.0309x faster audio-dft 97.632+-2.143 ? 99.324+-4.857 ? might be 1.0173x slower audio-fft 30.895+-0.097 30.818+-0.126 audio-oscillator 47.858+-0.390 ? 48.985+-3.866 ? might be 1.0236x slower imaging-darkroom 59.703+-0.109 59.646+-0.067 imaging-desaturate 46.749+-3.836 44.568+-0.589 might be 1.0489x faster imaging-gaussian-blur 64.555+-4.752 63.194+-5.675 might be 1.0215x faster json-parse-financial 34.710+-0.553 ? 35.017+-0.761 ? json-stringify-tinderbox 24.279+-0.450 23.369+-3.174 might be 1.0389x faster stanford-crypto-aes 37.494+-0.235 ? 40.321+-4.959 ? might be 1.0754x slower stanford-crypto-ccm 32.016+-2.402 ? 32.616+-1.158 ? might be 1.0188x slower stanford-crypto-pbkdf2 93.685+-0.636 ? 94.828+-2.803 ? might be 1.0122x slower stanford-crypto-sha256-iterative 35.799+-0.403 ? 37.014+-2.550 ? might be 1.0339x slower <arithmetic> 52.353+-0.677 ? 52.473+-0.623 ? might be 1.0023x slower Conf#1 Conf#2 AsmBench: bigfib.cpp 441.0201+-11.7104 439.5991+-7.0323 cray.c 369.8727+-8.4264 365.0147+-3.0296 might be 1.0133x faster dry.c 453.0382+-103.5874 425.1428+-6.2480 might be 1.0656x faster FloatMM.c 722.0267+-15.3660 717.2737+-17.3113 gcc-loops.cpp 3704.7706+-140.2455 3662.5536+-39.3203 might be 1.0115x faster n-body.c 798.3393+-4.0115 ? 801.9427+-8.8173 ? Quicksort.c 392.9307+-1.6659 ? 395.0790+-3.5237 ? stepanov_container.cpp 3260.5988+-32.5583 ? 3298.3015+-57.1320 ? might be 1.0116x slower Towers.c 272.2712+-4.1250 270.7480+-0.7336 <geometric> 724.0382+-19.3567 717.9682+-2.1488 might be 1.0085x faster Conf#1 Conf#2 Geomean of preferred means: <scaled-result> 30.3555+-0.1724 30.1983+-0.0471 might be 1.0052x faster

Geoffrey Garen

Comment 3 2016-04-28 11:00:00 PDT

Comment on attachment 277596 [details] Patch r=me

Benjamin Poulain

Comment 4 2016-04-28 11:05:50 PDT

Created attachment 277633 [details] Patch for landing

WebKit Commit Bot

Comment 5 2016-04-28 12:57:41 PDT

Comment on attachment 277633 [details] Patch for landing Rejecting attachment 277633 [details] from commit-queue. New failing tests: imported/w3c/web-platform-tests/fetch/nosniff/stylesheet.html Full output: http://webkit-queues.webkit.org/results/1235761

WebKit Commit Bot

Comment 6 2016-04-28 12:57:43 PDT

Created attachment 277641 [details] Archive of layout-test-results from webkit-cq-03 for mac-yosemite The attached test failures were seen while running run-webkit-tests on the commit-queue. Bot: webkit-cq-03 Port: mac-yosemite Platform: Mac OS X 10.10.5

Benjamin Poulain

Comment 7 2016-04-28 13:48:19 PDT

(In reply to comment #5) > Comment on attachment 277633 [details] > Patch for landing > > Rejecting attachment 277633 [details] from commit-queue. > > New failing tests: > imported/w3c/web-platform-tests/fetch/nosniff/stylesheet.html > Full output: http://webkit-queues.webkit.org/results/1235761 That looks completely unrelated. I'll try landing manually.

Benjamin Poulain

Comment 8 2016-04-28 13:49:36 PDT

Comment on attachment 277633 [details] Patch for landing Clearing flags on attachment: 277633 Committed r200208: <http://trac.webkit.org/changeset/200208>

Benjamin Poulain

Comment 9 2016-04-28 13:49:40 PDT

All reviewed patches have been landed. Closing bug.

Csaba Osztrogonác

Comment 10 2016-04-29 02:26:54 PDT

(In reply to comment #8) > Comment on attachment 277633 [details] > Patch for landing > > Clearing flags on attachment: 277633 > > Committed r200208: <http://trac.webkit.org/changeset/200208> It caused test failures on 32 bit platforms, see bug157168 for details.

Note You need to log in before you can comment on or make changes to this bug.

Status RESOLVED

Resolution FIXED

Priority P2

Severity Normal

Classification Unclassified

Version WebKit Nightly Build

Hardware Unspecified

OS Unspecified

Product WebKit

Component New Bugs

Assignee

Benjamin Poulain

Reported

2016-04-27 22:53 PDT

Modified

2016-04-29 02:27 PDT History

CC List

7 users Show

URL

Keywords

Depends on

157168

Blocks

Dependencies

tree graph