946426 - While loop using "count-- != 0" as its condition is not optimized very well

Status: RESOLVED DUPLICATE of bug 946425

(Core :: JavaScript Engine: JIT, defect)

Description

[Boris Zbarsky [:bzbarsky]]

Consider this script:

  (function(count) {
    var start = new Date;
    while (count-- != 0);
    var stop = new Date;
    print((stop - start) / 10000000 * 1e6);
  })(10000000);

This prints about 2.6ns per iteration for me in the shell.  With --ion-eager it's more like .9ns, which is what I would expect given bug 946422.

Below is what JIT inspector says we're doing.  We seem to be doing some sort of ValueToInt32 conversion (why?) and overflow checks (why?).

This may be a duplicate of bug 946425...

Block #3 -> #3 -> #4 :: 9998904 hits
[Label]
[InterruptCheckImplicit]
[OsiPoint]
[ValueToInt32:Normal]
movq       %rax, %r11
shrq       $47, %r11
cmpl       $0x1fff1, %r11d
je         ((690))
cmpl       $0x1fff6, %r11d
jne        ((703))
xorl       %ebx, %ebx
jmp        ((710))
##link     ((690)) jumps to ((710))
movl       %eax, %ebx
##link     ((710)) jumps to ((712))
[OsiPoint]
[MoveGroup]
movq       %rbx, %rbp
[SubI:OverflowCheck]
subl       $0x1, %ebp
jo         ((724))
[Box:Int32]
movabsq    $0xfff8800000000000, %rax
orq        %rbp, %rax
[CompareAndBranch:ne]
testl      %ebx, %ebx
jne        ((745))
##link     ((745)) jumps to ((745))

Comment 1

[Jan de Mooij [:jandem]]

(In reply to Boris Zbarsky [:bz] from comment #0)
> This may be a duplicate of bug 946425...

Yup, this looks fine now.