Expose general purpose Crc32 APIs

[tannergooding]

Rationale

Computing a Cyclic Redundancy Check (CRC) is a common algorithm used for error detection in things like networks or storage. Additionally, modern hardware provides instruction level support for computing these values. It would be beneficial if we exposed a set of general-purpose methods which allow iterative CRC32 computation.

Proposed API

namespace System.Numerics
{
    public static class BitOperations
    {
        public static uint Crc32(uint crc, byte data);
        public static uint Crc32(uint crc, ushort data);
        public static uint Crc32(uint crc, uint data);
        public static uint Crc32(uint crc, ulong data);
    }
}

Open Questions

The output is generally treated as unsigned, however .NET considers unsigned integers (other than byte) as non-CLS compliant. It would be possible to expose both versions (those that take/return int and those that take/return uint). It would likewise be good on the input data to determine if they should be signed, unsigned, or both.

[Wraith2]

Is this using a specific polynomial? and if so which one? and can there be a way for the user to choose?

[tannergooding]

can there be a way for the user to choose?

Not if we want these accelerated by the underlying hardware.

• X86 uses: 0x1EDC6F41
• ARM uses: 0x04C11DB7 or 0x1EDC6F41 (depending on the instruction)

[Wraith2]

So if someone is dealing with zip or something using the old polynomial they'll still need to make their own. That's a bit of a shame.

Given that at least half of the time you want to calculate a crc you will be validating incoming data it might make sense to have a ReadOnlySpan<byte> overload (that works on a fixed length, leave looping to users).

[GrabYourPitchforks]

If we were to expose a CRC32 API over a buffer rather than an integral type I'd probably push for it not to be on the BitOperations class.

[Wraith2]

If BitOperations and these methods only exist to provide access to the hardware intrinsics I agree.
It would be useful to have a CRC32HashAlgorithm available which uses these operations if they are available.

[tannergooding]

If BitOperations and these methods only exist to provide access to the hardware intrinsics I agree

Right, this is namely meant to be a "cross-platform" version of the hardware intrinsics (much like the other functions on System.Numerics.BitOperations) and should allow users to easily get:

if (X86.Sse42.IsSupported) { }
else if (Arm.Crc32.IsSupported) { }
else { /* software fallback */ }

I agree that having a more general purpose class that allows polynomial customization and trivial operation over spans of data would be useful, but that isn't the immediate goal of this API.