Bug 190264

Summary:

[BigInt] Implement ValueBitXor into DFG

Product:

WebKit

Reporter:

Caio Lima <ticaiolima>

Component:

JavaScriptCore

Assignee:

Caio Lima <ticaiolima>

Status:

RESOLVED FIXED

Severity:

Normal

CC:

commit-queue, ews-watchlist, keith_miller, mark.lam, msaboff, saam, webkit-bug-importer, ysuzuki

Priority:

Keywords:

InRadar

Version:

WebKit Nightly Build

Hardware:

Unspecified

OS:

Unspecified

Bug Depends on:

186235

Bug Blocks:

186173

Attachments:

Description	Flags
WIP - Patch	none
WIP - Patch	none
WIP - Patch	none
Patch	none
Patch	none
Benchmarks	none
Patch	none
Patch	none
Patch	none
Benchmarks	none
Patch	none
Patch	none
Patch	none
Patch	none
Patch	none
Patch	none
Patch	none
Patch	none

Description Caio Lima 2018-10-03 14:58:50 PDT

...

Comment 1 Caio Lima 2018-10-06 08:27:15 PDT

Created attachment 351727 [details]
WIP - Patch

Comment 2 Caio Lima 2018-10-06 11:52:11 PDT

Created attachment 351730 [details]
WIP - Patch

Patch doesn't compile.

Comment 3 Caio Lima 2018-10-09 19:49:24 PDT

Created attachment 351934 [details]
WIP - Patch

Comment 4 Caio Lima 2018-10-21 18:48:51 PDT

Created attachment 352878 [details]
Patch

Comment 5 Caio Lima 2018-10-21 20:05:44 PDT

Created attachment 352880 [details]
Patch

Comment 6 Caio Lima 2018-10-22 04:55:14 PDT

Created attachment 352885 [details]
Benchmarks

Changes are perf neutral.

Comment 7 Caio Lima 2018-10-31 18:54:30 PDT

Created attachment 353569 [details]
Patch

Comment 8 Caio Lima 2018-10-31 19:04:06 PDT

Created attachment 353570 [details]
Patch

Comment 9 Caio Lima 2018-11-01 04:41:53 PDT

Comment on attachment 353570 [details]
Patch

Investigating performance regression

Comment 10 Caio Lima 2018-11-01 09:49:23 PDT

Created attachment 353609 [details]
Patch

Comment 11 Caio Lima 2018-11-01 09:50:56 PDT

Created attachment 353611 [details]
Benchmarks

This patch is perf neutral.

Comment 12 Caio Lima 2018-11-05 06:03:48 PST

Ping Review

Comment 13 Caio Lima 2018-11-08 06:15:55 PST

Created attachment 354235 [details]
Patch

Comment 14 EWS Watchlist 2018-11-08 07:43:09 PST

Comment on attachment 354235 [details]
Patch

Attachment 354235 [details] did not pass jsc-ews (mac):
Output: https://webkit-queues.webkit.org/results/9908802

New failing tests:
stress/ftl-put-by-id-setter-exception.js.dfg-eager-no-cjit-validate
apiTests

Comment 15 Caio Lima 2018-11-12 07:59:56 PST

Created attachment 354559 [details]
Patch

Comment 16 Caio Lima 2018-11-13 10:41:52 PST

Created attachment 354681 [details]
Patch

Comment 17 Caio Lima 2018-11-14 08:47:07 PST

Created attachment 354813 [details]
Patch

Comment 18 Caio Lima 2018-11-18 16:22:45 PST

Created attachment 355243 [details]
Patch

Comment 19 Caio Lima 2018-11-19 16:32:35 PST

Ping review

Comment 20 Caio Lima 2018-11-26 03:11:59 PST

Created attachment 355629 [details]
Patch

Comment 21 Caio Lima 2018-11-26 03:21:59 PST

Created attachment 355630 [details]
Patch

Comment 22 Caio Lima 2018-11-28 04:47:05 PST

Created attachment 355863 [details]
Patch

Comment 23 Yusuke Suzuki 2018-11-29 17:53:17 PST

Comment on attachment 355863 [details]
Patch

r=me

Comment 24 WebKit Commit Bot 2018-11-30 05:25:01 PST

Comment on attachment 355863 [details]
Patch

Clearing flags on attachment: 355863

Committed r238732: <https://trac.webkit.org/changeset/238732>

Comment 25 WebKit Commit Bot 2018-11-30 05:25:03 PST

All reviewed patches have been landed.  Closing bug.

Comment 26 Radar WebKit Bug Importer 2018-11-30 05:26:39 PST

<rdar://problem/46371247>

Comment 27 Saam Barati 2019-01-20 13:03:40 PST

Comment on attachment 355863 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=355863&action=review

> Source/JavaScriptCore/dfg/DFGAbstractInterpreterInlines.h:398
> +    case ValueBitXor:
>      case ValueBitAnd:
> +    case ValueBitOr:

Can we add constant folding for this in the future if the inputs are constant?

> Source/JavaScriptCore/dfg/DFGFixupPhase.cpp:167
> +                case ArithBitXor:
> +                    node->setOpAndDefaultFlags(ValueBitXor);
> +                    break;

Did we end up switching this back to be the opposite direction? First emit ValueBitXor, then optimize to ArithBitXor.