119737 – [sh4] DFG_JIT implementation for sh4 architecture

RESOLVED FIXED119737

[sh4] DFG_JIT implementation for sh4 architecture

https://bugs.webkit.org/show_bug.cgi?id=119737

Summary [sh4] DFG_JIT implementation for sh4 architecture

Julien Brianceau

Reported 2013-08-13 03:17:19 PDT

I'll submit a DFG_JIT implementation proposal for sh4 architecture soon.

Attachments
DFG_JIT implementation for sh4 architecture (86.56 KB, patch) 2013-08-13 04:52 PDT, Julien Brianceau	jbriance: commit-queue-	Details Formatted Diff Diff
DFG_JIT implementation for sh4 architecture (87.62 KB, patch) 2013-08-13 09:39 PDT, Julien Brianceau	no flags	Details Formatted Diff Diff
DFG_JIT implementation for sh4 architecture (after split) (37.74 KB, patch) 2013-08-14 02:14 PDT, Julien Brianceau	no flags	Details Formatted Diff Diff
Show Obsolete (2) View All Add attachment proposed patch, testcase, etc.

Julien Brianceau

Comment 1 2013-08-13 04:52:21 PDT

Created attachment 208618 [details] DFG_JIT implementation for sh4 architecture I've tested this patch on my sh4 board using r153882 and I didn't see regressions when running - Tools/Scripts/run-javascriptcore-tests - Tools/Scripts/run-fast-jsc - SunSpider 1.0 SunSpider 1.0 results on r153882 without this patch: ============================================ RESULTS (means and 95% confidence intervals) -------------------------------------------- Total: 13166.3ms +/- 0.2% -------------------------------------------- 3d: 2706.3ms +/- 0.5% cube: 837.8ms +/- 0.6% morph: 539.2ms +/- 0.7% raytrace: 1329.3ms +/- 0.7% access: 1905.7ms +/- 0.1% binary-trees: 114.2ms +/- 2.7% fannkuch: 1148.7ms +/- 0.4% nbody: 455.2ms +/- 1.0% nsieve: 187.7ms +/- 0.3% bitops: 1031.2ms +/- 0.4% 3bit-bits-in-byte: 78.2ms +/- 0.5% bits-in-byte: 111.2ms +/- 0.4% bitwise-and: 196.0ms +/- 0.0% nsieve-bits: 645.8ms +/- 0.6% controlflow: 74.0ms +/- 0.0% recursive: 74.0ms +/- 0.0% crypto: 1165.7ms +/- 1.3% aes: 666.2ms +/- 1.9% md5: 317.5ms +/- 1.1% sha1: 182.0ms +/- 2.0% date: 2085.0ms +/- 1.1% format-tofte: 925.7ms +/- 1.2% format-xparb: 1159.3ms +/- 1.3% math: 850.2ms +/- 1.1% cordic: 248.8ms +/- 3.8% partial-sums: 343.8ms +/- 2.7% spectral-norm: 257.5ms +/- 1.2% regexp: 531.0ms +/- 0.8% dna: 531.0ms +/- 0.8% string: 2817.3ms +/- 0.5% base64: 218.2ms +/- 0.5% fasta: 446.3ms +/- 3.2% tagcloud: 560.0ms +/- 1.1% unpack-code: 1255.2ms +/- 1.0% validate-input: 337.7ms +/- 2.2% SunSpider 1.0 results on r153882 with this patch: ============================================ RESULTS (means and 95% confidence intervals) -------------------------------------------- Total: 10122.6ms +/- 0.2% -------------------------------------------- 3d: 2440.8ms +/- 0.2% cube: 899.0ms +/- 0.3% morph: 278.5ms +/- 0.6% raytrace: 1263.3ms +/- 0.2% access: 882.0ms +/- 0.3% binary-trees: 106.5ms +/- 1.6% fannkuch: 289.3ms +/- 0.5% nbody: 335.9ms +/- 0.7% nsieve: 150.3ms +/- 0.7% bitops: 335.5ms +/- 0.6% 3bit-bits-in-byte: 29.2ms +/- 1.0% bits-in-byte: 74.6ms +/- 1.8% bitwise-and: 79.2ms +/- 0.4% nsieve-bits: 152.5ms +/- 0.9% controlflow: 96.1ms +/- 0.2% recursive: 96.1ms +/- 0.2% crypto: 886.4ms +/- 0.3% aes: 464.9ms +/- 0.6% md5: 251.0ms +/- 0.7% sha1: 170.5ms +/- 0.7% date: 1512.6ms +/- 0.4% format-tofte: 780.6ms +/- 0.7% format-xparb: 732.0ms +/- 0.7% math: 589.3ms +/- 0.4% cordic: 148.5ms +/- 0.3% partial-sums: 314.9ms +/- 0.6% spectral-norm: 125.9ms +/- 0.2% regexp: 529.1ms +/- 0.4% dna: 529.1ms +/- 0.4% string: 2850.8ms +/- 0.7% base64: 189.2ms +/- 0.9% fasta: 402.9ms +/- 0.9% tagcloud: 541.1ms +/- 0.6% unpack-code: 1323.1ms +/- 1.0% validate-input: 394.5ms +/- 1.5%

Julien Brianceau

Comment 2 2013-08-13 06:28:05 PDT

Comment on attachment 208618 [details] DFG_JIT implementation for sh4 architecture I cq- because I just saw that this patch introduces a regression (crash) in v8-v4 test suite. However, I'd still be happy to get feedbacks on this patch, especially the "SH4_32BIT_DUMMY_ARG" part introduced in DFGSpeculativeJIT.h.

Julien Brianceau

Comment 3 2013-08-13 09:39:58 PDT

Created attachment 208642 [details] DFG_JIT implementation for sh4 architecture I've tested this patch on my sh4 board using r153999 and I didn't see regressions when running - Tools/Scripts/run-javascriptcore-tests - Tools/Scripts/run-fast-jsc - SunSpider 1.0 - v8-v4 test suite v8-v4 test suite results on r153999 without this patch: ============================================ RESULTS (means and 95% confidence intervals) -------------------------------------------- Total: 74696.0ms +/- 11.7% -------------------------------------------- v8: 74696.0ms +/- 11.7% crypto: 13296.7ms +/- 1.4% deltablue: 21649.7ms +/- 17.6% earley-boyer: 6108.0ms +/- 7.4% raytrace: 7092.7ms +/- 7.3% regexp: 6166.3ms +/- 1.2% richards: 15396.0ms +/- 43.9% splay: 4986.7ms +/- 6.6% v8-v4 test suite results on r153999 with this patch: ============================================ RESULTS (means and 95% confidence intervals) -------------------------------------------- Total: 36705.7ms +/- 0.7% -------------------------------------------- v8: 36705.7ms +/- 0.7% crypto: 4592.0ms +/- 0.5% deltablue: 7425.7ms +/- 1.6% earley-boyer: 4113.3ms +/- 0.2% raytrace: 5562.7ms +/- 1.1% regexp: 7486.7ms +/- 0.1% richards: 4316.7ms +/- 1.4% splay: 3208.7ms +/- 9.5%

Oliver Hunt

Comment 4 2013-08-13 09:40:09 PDT

Comment on attachment 208618 [details] DFG_JIT implementation for sh4 architecture Separate out the llint changes into a separate patch (it will jus take it easier to be sure a regression was not introduced by those vs. the dfg. Similarly there's a lot of assembler changes that look like they should be safe on there own (separate tom this patch)

Julien Brianceau

Comment 5 2013-08-13 09:59:11 PDT

Comment on attachment 208642 [details] DFG_JIT implementation for sh4 architecture (In reply to comment #4) > Separate out the llint changes into a separate patch (it will jus take it easier to be sure a regression was not introduced by those vs. the dfg. Similarly there's a lot of assembler changes that look like they should be safe on there own (separate tom this patch) Thanks. So I'm going to separate this patch in 3 parts: LLINT, MacroAssembler not related to the DFG then the DFG.

Julien Brianceau

Comment 6 2013-08-13 10:16:26 PDT

LLINT part: https://bugs.webkit.org/show_bug.cgi?id=119755

Julien Brianceau

Comment 7 2013-08-13 10:54:55 PDT

Baseline JIT part: https://bugs.webkit.org/show_bug.cgi?id=119758

Julien Brianceau

Comment 8 2013-08-14 02:14:01 PDT

Created attachment 208707 [details] DFG_JIT implementation for sh4 architecture (after split)

WebKit Commit Bot

Comment 9 2013-08-14 09:24:11 PDT

Comment on attachment 208707 [details] DFG_JIT implementation for sh4 architecture (after split) Rejecting attachment 208707 [details] from commit-queue. Failed to run "['/Volumes/Data/EWS/WebKit/Tools/Scripts/webkit-patch', '--status-host=webkit-queues.appspot.com', '--bot-id=webkit-cq-03', 'build', '--no-clean', '--no-update', '--build-style=release', '--port=mac']" exit_code: 2 cwd: /Volumes/Data/EWS/WebKit Last 500 characters of output: ilure) Failed to run "['Tools/Scripts/build-webkit', '--release']" exit_code: 65 Process.sb make: *** wait: No child processes. Stop. make: *** Waiting for unfinished jobs.... make: *** wait: No child processes. Stop. Command /bin/sh failed with exit code 2 ** BUILD FAILED ** The following build commands failed: PhaseScriptExecution "Generate Derived Sources" "/Volumes/Data/EWS/WebKit/WebKitBuild/WebKit2.build/Release/Derived Sources.build/Script-C0CE72841247E66800BC0EC4.sh" (1 failure) Full output: http://webkit-queues.appspot.com/results/1464096

Julien Brianceau

Comment 10 2013-08-14 09:27:47 PDT

Comment on attachment 208707 [details] DFG_JIT implementation for sh4 architecture (after split) Mac bot looks sick today :-/

Allan Sandfeld Jensen

Comment 11 2013-08-14 09:50:03 PDT

Comment on attachment 208707 [details] DFG_JIT implementation for sh4 architecture (after split) Clearing flags on attachment: 208707 Committed r154052: <http://trac.webkit.org/changeset/154052>

Allan Sandfeld Jensen

Comment 12 2013-08-14 09:50:10 PDT

All reviewed patches have been landed. Closing bug.

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 528+ (Nightly build)

Hardware Unspecified

OS Unspecified

Product WebKit

Component JavaScriptCore

Assignee

Nobody

Reported

2013-08-13 03:17 PDT

Modified

2013-08-14 09:50 PDT History

CC List

6 users Show

URL

Keywords

Depends on

119755 119758

Blocks

Dependencies

tree graph