Implement all the arithmetic and logical instructions in WebAssembly

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Sukolsak Sakshuwong]

Implement all the arithmetic and logical instructions for WebAssembly files generated by pack-asmjs <https://github.com/WebAssembly/polyfill-prototype-1>.

[Sukolsak Sakshuwong]

Created attachment 260675 [details]
Patch

[Sukolsak Sakshuwong]

Created attachment 260730 [details]
Patch

[Mark Lam]

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.

> 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?

[Sukolsak Sakshuwong]

(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.

> > 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>

[Mark Lam]

(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.

[WebKit Commit Bot]

Comment on attachment 260730 [details]
Patch

Clearing flags on attachment: 260730

Committed r189499: <http://trac.webkit.org/changeset/189499>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.