Loop isn't optimized in pow() with constant power

[eefriedman]

Testcase:

pub fn f(x: i32) -> i32 { x.pow(5) }

Produces the following asm in release mode:

_ZN8rust_out1f17hc690d9db51fd1099E:
    .cfi_startproc
    movl    $5, %ecx
    movl    $1, %r8d
    movl    $1, %eax
    .align  16, 0x90
.LBB0_1:
    movl    %ecx, %esi
    testb   $1, %sil
    movl    %edi, %edx
    cmovel  %r8d, %edx
    imull   %edx, %eax
    shrl    %ecx
    imull   %edi, %edi
    cmpl    $3, %esi
    ja  .LBB0_1
    cmpl    $1, %ecx
    movl    $1, %ecx
    cmovel  %edi, %ecx
    imull   %ecx, %eax
    retq

I would expect it to generate straight-line code with just three multiplies.

It's possible this will be fixed by upgrading LLVM; there have been some recent changes to unrolling heuristics.

[Mark-Simulacrum]

Assembly produced today is correct, and as expected. Closing.

0000000000000000 <_ZN4test1f17h2e3ee7790982a020E>:
   0:	89 f8                	mov    %edi,%eax
   2:	0f af c0             	imul   %eax,%eax
   5:	0f af f8             	imul   %eax,%edi
   8:	0f af f8             	imul   %eax,%edi
   b:	89 f8                	mov    %edi,%eax
   d:	c3                   	retq

[Mark-Simulacrum]

Reopening as E-needstest. This should get a codegen test.