[Webkit-unassigned] [Bug 162980] New: [DOMJIT] Add DOMJITSignature

bugzilla-daemon at webkit.org bugzilla-daemon at webkit.org
Wed Oct 5 10:56:54 PDT 2016


https://bugs.webkit.org/show_bug.cgi?id=162980

            Bug ID: 162980
           Summary: [DOMJIT] Add DOMJITSignature
    Classification: Unclassified
           Product: WebKit
           Version: WebKit Nightly Build
          Hardware: Unspecified
                OS: Unspecified
            Status: NEW
          Severity: Normal
          Priority: P2
         Component: JavaScriptCore
          Assignee: webkit-unassigned at lists.webkit.org
          Reporter: utatane.tea at gmail.com

We should have two type of DOMJIT functionality.

1. DOMJIT::GetterSetter based inlining system

We can define all the DOMJIT operations in DOMJIT::Patchpoint. It is great since we have a chance to inline it completely!
While it's quite good for performance, ASM is typically written by hand.
I explored the way to generate ASM automatically from IDL compiler, but the accessors we can do so is still limited.

For example, even in Node::firstChild, it has special check path for ContainerNode, while nextSibling, parentNode, and previousSibling do not have such a path.
And this fact is not exposed in IDL.

We manually write ASM in this case. It has scalability & complexity (But CSS Selector JIT already does the similar things for parentNode, siblings) problem.
So we will write this for super hot accessors. Like parentNode, firstChild etc.

2. DOMJIT::Signature based type checking elimination

On the other hand, this can be applied to many bindings automatically.
We support DOMJIT::Signature. That is the way to express DOM function's signature to DFG/FTL.
DFG / FTL recognizes and emit Checks instead of binding code. And in the middle of code, we will invoke the thin binding code that does not have any checks.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20161005/6907d7a1/attachment-0001.html>


More information about the webkit-unassigned mailing list