[BUG] Count leading zeros can't be used at compile time

[martinvuyk]

Bug description

Count leading zeros can't be used at compile time. This is actually part of #933. This blocks a lot of other work to fully support unicode due to it needing this function.

part of the error: note: LLVM intrinsic operand has unknown value: #pop<simd 15> : !pop.scalar<ui8>

CC: @lattner

Steps to reproduce

from bit import count_leading_zeros

fn _utf8_first_byte_sequence_length(b: Byte) -> Int:
    """Get the length of the sequence starting with given byte. Do note that
    this does not work correctly if given a continuation byte."""

    debug_assert(
        (b & 0b1100_0000) != 0b1000_0000,
        "Function does not work correctly if given a continuation byte.",
    )
    return int(count_leading_zeros(~b)) + int(b < 0b1000_0000)

fn main():
    print(_utf8_first_byte_sequence_length(ord("a")) == 1)
    print(_utf8_first_byte_sequence_length("🔥".as_bytes()[0]) == 4)
    alias `🔥` = "🔥".as_bytes()[0]
    alias length = _utf8_first_byte_sequence_length(`🔥`)
    print(length == 4)

System information

- Provide version information for Mojo by pasting the output of `mojo -v`
`mojo 24.6.0.dev2024120805`

[martinvuyk]

Why would this not be planned? we need Unicode support in StringLiteral. And the StringSlice iterator uses this function, and many other things that users will want to run at compile time.

Unless we get a way to branch when the code is running at compile time, this is needed

PS: I'm not saying I want this tomorrow, but it should be on the roadmap to at least provide alternatives (like a boolean when running code at compile time) for these cases.