Use rust builtins for round_upto_multiple_of_64 and ceil

[psvri]

Which issue does this PR close?

Closes #.

Rationale for this change

Now that we have updated msrv, we can use rust built ins for round_upto_multiple_of_64 and div_ceil

What changes are included in this PR?

Are there any user-facing changes?

Yes, minor edits to error message

[alamb]

Thanks @psvri - I think we should run some benchmarks to ensure we aren't introducing some performance regressions by accident

[mbutrovich]

Thanks @psvri - I think we should run some benchmarks to ensure we aren't introducing some performance regressions by accident

To investigate round_upto_multiple_of_64 I dumped the disassembly of MutableBuffer::shrink_to_fit...

pub fn shrink_to_fit(&mut self) {
    let new_capacity = bit_util::round_upto_multiple_of_64(self.len);
    if new_capacity < self.layout.size() {
        self.reallocate(new_capacity)
    }
}

...from a release build for main and this PR:

Online diff.

The generated instructions are slightly different, but general number of instructions, look the same. I would be stunned if you measured anything other than noise in a release build. I'll take a look at ceil next.

[mbutrovich]

For ceil I disassembled MutableBuffer::new_null since all it does is use ceil and then allocate:

/// creates a new [MutableBuffer] with capacity and length capable of holding `len` bits.
/// This is useful to create a buffer for packed bitmaps.
pub fn new_null(len: usize) -> Self {
    let num_bytes = bit_util::ceil(len, 8);
    MutableBuffer::from_len_zeroed(num_bytes)
}

The generated code is the same between main this PR.

[mbutrovich]

For ceil what might be interesting is to find somewhere where the compiler can't use a shift immediate because it's a compile-time constant power of 2 (like in new_null above). I'll poke around....

[mbutrovich]

So, almost the entire codebase uses compile time constant args that are powers of two to ceil so those will generate equivalent shifts. The exception is arrow_row::variable::padded_length. That gets inlined in encoded_len, which in turn gets inlined a bunch of times in row_lengths. So let's dive into that!

Online diff.

Darn. Aside from label names, they're exactly the same. I really thought I'd find a difference. Long way of saying, there should not be measurable performance difference on this PR.

[alamb]

Thank you @mbutrovich and @psvri for your contribution and analysis

[alamb]

🚀 !