`div_ceil` does not yield optimal code for unsigned const divisor

Let's say we have a function:

fn foo(begin: usize) -> usize {
    (begin + 128 - 1) / 128
}

This compiles to:

foo:
        leaq    127(%rdi), %rax
        shrq    $7, %rax
        retq

But clippy suggests that I should use div_ceil:

fn foo(begin: usize) -> usize {
    begin.div_ceil(128)
}

... which compiles to:

foo:
        movq    %rdi, %rax
        shrq    $7, %rax
        andl    $127, %edi
        cmpq    $1, %rdi
        sbbq    $-1, %rax
        retq

It seems that the original version produces better code on x86_64 with only two instructions (without the retq), but 5 instructions (without the retq) for the div_ceil version.

Godbolt

The issue names no repository files or tests. Start with the Rust example and linked Godbolt comparison, then inspect how Clippy produces the div_ceil suggestion for unsigned constant divisors. Done means deciding the expected behavior and verifying it with an appropriate regression test and codegen comparison.