[Webkit-unassigned] [Bug 200412] New: Jitted code gives different floating point result

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=200412

            Bug ID: 200412
           Summary: Jitted code gives different floating point result
           Product: WebKit
           Version: Safari Technology Preview
          Hardware: All
                OS: All
            Status: NEW
          Severity: Critical
          Priority: P2
         Component: JavaScriptCore
          Assignee: webkit-unassigned at lists.webkit.org
          Reporter: bert at freudenbergs.de

Created attachment 375469

  --> https://bugs.webkit.org/attachment.cgi?id=375469&action=review

html and js demonstrating the bug

Example code (runnable at https://croquet.studio/static/math-bug.html):

    const pow = require("@stdlib/math/base/special/pow");
    window.StdlibMath = { pow };

    const func = () => window.StdlibMath.pow( 8.96096203115875, -0.6403121474522467 );

    console.log(func());

    let x = 0; for (let i = 0; i < 1000000; i++) x += func();

    console.log(func());

This code prints the same result twice if running in the Safari debugger, and on every other browser (Chrome, Firefox):
  0.24558156189439845
  0.24558156189439845
But the second result is different when run without debugger in Safari on Mac and iOS.
  0.24558156189439845
  0.24558156189439848
Re-running the code also sometimes prints a different second result randomly.

(we're using stdlib-js to avoid the built-in transcendental Math functions because we need deterministic results)

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20190803/84e24c50/attachment.html>