RESOLVED FIXED 181991
JSC should use a speculation fence on VM entry/exit
https://bugs.webkit.org/show_bug.cgi?id=181991
Summary JSC should use a speculation fence on VM entry/exit
Filip Pizlo
Reported 2018-01-23 10:09:59 PST
Patch forthcoming.
Attachments
the patch (3.67 KB, patch)
2018-01-23 10:12 PST, Filip Pizlo
jfbastien: review+
Filip Pizlo
Comment 1 2018-01-23 10:12:04 PST
Created attachment 332035 [details] the patch
Filip Pizlo
Comment 2 2018-01-23 10:12:11 PST
Radar WebKit Bug Importer
Comment 3 2018-01-23 10:12:48 PST
JF Bastien
Comment 4 2018-01-23 10:16:46 PST
Comment on attachment 332035 [details] the patch r=me
Mark Lam
Comment 5 2018-01-23 10:17:21 PST
Comment on attachment 332035 [details] the patch View in context: https://bugs.webkit.org/attachment.cgi?id=332035&action=review > Source/JavaScriptCore/ChangeLog:10 > + For a microbenchmark that just calls a native function (supplied via a Objective-C block) in a /a Objective-C/an Objective-C/. > Source/JavaScriptCore/ChangeLog:11 > + tight loop from JS is a 0% regression on x86 and a 11% regression on ARM64. I suggest /JS is a 0%/JS, performance shows a 0%/.
Filip Pizlo
Comment 6 2018-01-23 10:30:13 PST
Saam Barati
Comment 7 2018-01-23 12:39:24 PST
Comment on attachment 332035 [details] the patch View in context: https://bugs.webkit.org/attachment.cgi?id=332035&action=review > Source/WTF/wtf/Atomics.h:285 > +#if !OS(WINDOWS) Should we have someone who works on Windows implement this for them?
Saam Barati
Comment 8 2018-01-23 12:42:31 PST
Comment on attachment 332035 [details] the patch View in context: https://bugs.webkit.org/attachment.cgi?id=332035&action=review > Source/JavaScriptCore/runtime/JSLock.cpp:126 > + WTF::speculationFence(); Do we not want this for entry/exit in and out of other native code. E.g, WebCore bindings?
Note You need to log in before you can comment on or make changes to this bug.