Use the least significant beat to determine if int/uint is even

[flaper87]

This implementation should be a bit more optimal than calling self.is_multiple_of(&2)

[alexcrichton]

I'm curious, but have you verified that the generated code is actually better? I would expect LLVM to do proper inlining and figure everything out.

Regardless, the documentation shouldn't change, just the implementation, but thanks for the patch!

[flaper87]

I didn't verify the generated code, I could check with --emit-llvm, though.

Why shouldn't the documentation change? It's not verifying the number is divisible by 2 anymore :/

[flaper87]

Maybe not the best test. TBH, not sure what the best test is.

define i64 @main(i64, i8**) unnamed_addr {
top:
  %2 = call i64 @_ZN10lang_start19hdc66d049ef36cac4ay4v0.9E({ i64, %tydesc*, i8*, i8*, i8 }* null, i8* bitcast (void ({ i64, %tydesc*, i8*, i8*, i8 }*)* @_ZN4main19h2320e51607f4f569ai4v0.0E to i8*), i64 %0, i8** %
1)
  ret i64 %2
}

use std::num;

fn main() {
    2u & 1 == 0;
    2u.is_even();
}

[adrientetar]

@flaper87 It still checks that the number is divisible by 2. Looking at the least significant bit is just how it is implemented.

[flaper87]

Interesting! Then, let me trash this patch! Thanks! 😄

[flaper87]

I misunderstood @adridu59's comment. He was referring to the documentation.

[jdm]

@flaper87 It looks like llvm eliminated all of your test since the results were unused. You probably need to print the output or something.

[flaper87]

This is the full output of the emit. http://pastebin.mozilla.org/4022063

And the test

use std::num;

fn main() {
    let a = 2u & 1 == 0;
    let b = 2u.is_even();

    println!("{}", a);
    println!("{}", b);
}

[jdm]

  store i8 1, i8* %a
  store i64 2, i64* %1
  %6 = call i8 @"_ZN4uint12Integer$uint7is_even21h93c2e9593d9b8cb6Ijak4v0.0E"(i64* %1)
  store i8 %6, i8* %b

Still not convinced that this test is representative of the comparison we're looking for.

[flaper87]

Before: rustc -O -S --emit-llvm test.rs http://pastebin.mozilla.org/4022199

After: rustc -O -S --emit-llvm test.rs http://pastebin.mozilla.org/4022178

[alexcrichton]

From the tests it appears they optimize to exactly the same thing, so LLVM is definitely doing its job. When running tests at O0, however, I imagine that and-ing with 1 is faster, so I'm gonna go ahead and r+ this.

Thanks for the patch and the good investigation!