[Webkit-unassigned] [Bug 148882] Implement all the arithmetic and logical instructions in WebAssembly

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=148882

--- Comment #5 from Mark Lam <mark.lam at apple.com> ---
(In reply to comment #4)
> (In reply to comment #3)
> > Comment on attachment 260730 [details]
> > Patch
> > 
> > View in context:
> > https://bugs.webkit.org/attachment.cgi?id=260730&action=review
> > 
> > r=me
> > 
> > > Source/JavaScriptCore/tests/stress/wasm-arithmetic.js:45
> > > +        return (6 * 7) | 0;
> > 
> > What's stopping the WASM compiler from just constant folding this operation
> > and just returning a constant (thereby defeating this test)?  Why not pass
> > the values into the test functions as arguments?  Unless there's a reason to
> > not do so, please pass the args in to the test instead.  I'm fine with doing
> > a follow up patch to update all the tests to do this.
> 
> pack-asmjs does not do constant folding. (I don't think they will ever do
> it. Optimization of asm.js should be done by Emscripten's.) But I will do it
> to be future-proof. It should also make the code easier to read.

OK.  Let's do this in a separate patch since there are other tests (than the ones in this patch) that would benefit from this refactoring as well.

> > > Source/JavaScriptCore/tests/stress/wasm-arithmetic.js:192
> > > +shouldBe(module.multiplyOverflow(), -2147483648);
> > 
> > What does it mean to multiplyOverflow?  Does WASM require that we overflow
> > silently?
> 
> multiplyOverflow = the result from multiplication is larger than INT32_MAX.
> The WASM spec says "Sign-agnostic operations silently wrap overflowing
> results into the result type." It also says that multiplication on 32-bit
> integers is a sign-agnostic operation.
> <https://github.com/WebAssembly/design/blob/master/AstSemantics.md>

OK.

-- 
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/20150908/b1599bc8/attachment-0001.html>