956402 - OdinMonkey: refactoring: add Float to NumLit

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Luke Wagner [:luke]]

Attachment: factor-f32-literal

I just noticed that we could simplify the float logic by having float literals use the same NumLit/IsNumericLiteral/ExtractNumericLiteral machinery as everything else.  Doing this caused a cascade of minor simplifications which altogether remove 40 lines.  The patch would've been shorter if I hadn't had to move NumLit et al to be under ModuleCompiler (b/c they now depend on the definition of ModuleCompiler).

Comment 1

[Benjamin Bouvier [:bbouvier] (inactive)]

Comment on attachment 8355647 [details] [diff] [review]
factor-f32-literal

Review of attachment 8355647 [details] [diff] [review]:
-----------------------------------------------------------------

Thanks! That's a great refactoring, it integrates way cleaner with the pre-existing code.

::: js/src/jit/AsmJS.cpp
@@ +1672,5 @@
> +
> +// Represents the type and value of an asm.js numeric literal.
> +//
> +// A literal is a double iff the literal contains an exponent or decimal point
> +// (even if the fractional part is 0). Otherwise, integers may be classified:

This comment asks for an update.

@@ +1716,5 @@
> +    }
> +
> +    float toFloat() const {
> +        JS_ASSERT(which_ == Float);
> +        return (float)v_.toDouble();

nit: wouldn't you prefer a C++-er conversion, like
  return float(v_.toDouble());

though it probably doesn't make a difference for the compiler.

Comment 2

[Luke Wagner [:luke]]

I would prefer that :)

https://hg.mozilla.org/integration/mozilla-inbound/rev/5568609fa907

Comment 3

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/5568609fa907