Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[API Proposal]: AVX-512 VPOPCNTDQ and BITALG Intrinsics #96162

Open
MineCake147E opened this issue Dec 19, 2023 · 4 comments
Open

[API Proposal]: AVX-512 VPOPCNTDQ and BITALG Intrinsics #96162

MineCake147E opened this issue Dec 19, 2023 · 4 comments
Labels
api-approved API was approved in API review, it can be implemented area-System.Runtime.Intrinsics avx512 Related to the AVX-512 architecture
Milestone
10.0.0

Comments

@MineCake147E
Copy link
Contributor

MineCake147E commented Dec 19, 2023 •
edited by tannergooding
Loading

Background and motivation

Both VPOPCNTDQ and BITALG are supported by Intel in the Ice Lake and newer architectures, and by AMD in Zen 4.
VPOPCNTDQ allows for parallel popcnt in either Vector128, Vector256, or Vector512 for ulong and uint.
BITALG expands parallel popcnt for ushort and byte, and it also adds VPSHUFBITQMB instruction, which performs a bit gather select.
VPOPCNTW is highly beneficial for my current project, allowing me for counting filled blocks row-by-row for a bit-board of block games.

API Proposal

namespace System.Runtime.Intrinsics.X86;

[Intrinsic]
[CLSCompliant(false)]
public abstract class Avx512VPopcntDQ : Avx512DQ
{
    public static new bool IsSupported { get; }

    [Intrinsic]
    public new abstract class X64 : Avx512DQ.X64
    {
        internal X64() { }
        public static new bool IsSupported { get; }
    }

    public static Vector512<int> PopCount(Vector512<int> value);

    public static Vector512<uint> PopCount(Vector512<uint> value);

    public static Vector512<long> PopCount(Vector512<long> value);

    public static Vector512<ulong> PopCount(Vector512<ulong> value);

    public abstract class VL : Avx512DQ.VL
    {
        public static new bool IsSupported { get; }

        public static Vector256<int> PopCount(Vector256<int> value);

        public static Vector256<uint> PopCount(Vector256<uint> value);

        public static Vector256<long> PopCount(Vector256<long> value);

        public static Vector256<ulong> PopCount(Vector256<ulong> value);

        public static Vector128<int> PopCount(Vector128<int> value);

        public static Vector128<uint> PopCount(Vector128<uint> value);

        public static Vector128<long> PopCount(Vector128<long> value);

        public static Vector128<ulong> PopCount(Vector128<ulong> value);
    }
}

[Intrinsic]
[CLSCompliant(false)]
public abstract class Avx512BitAlg : Avx512BW
{
    public static new bool IsSupported { get; }

    [Intrinsic]
    public new abstract class X64 : Avx512BW.X64
    {
        internal X64() { }
        public static new bool IsSupported { get; }
    }

    public static Vector512<short> PopCount(Vector512<short> value);

    public static Vector512<ushort> PopCount(Vector512<ushort> value);

    public static Vector512<byte> PopCount(Vector512<byte> value);

    public static Vector512<sbyte> PopCount(Vector512<sbyte> value);

    public static Vector512<byte> ShuffleBits(Vector512<ulong> value, Vector512<byte> control);

    public static Vector512<sbyte> ShuffleBits(Vector512<long> value, Vector512<sbyte> control);

    public static Vector512<byte> MaskShuffleBits(Vector512<byte> mask, Vector512<ulong> value, Vector512<byte> control);

    public static Vector512<sbyte> MaskShuffleBits(Vector512<sbyte> mask, Vector512<long> value, Vector512<sbyte> control);

    public abstract class VL : Avx512BW.VL
    {
        public static new bool IsSupported { get; }

        public static Vector256<short> PopCount(Vector256<short> value);

        public static Vector256<ushort> PopCount(Vector256<ushort> value);

        public static Vector256<byte> PopCount(Vector256<byte> value);

        public static Vector256<sbyte> PopCount(Vector256<sbyte> value);

        public static Vector128<short> PopCount(Vector128<short> value);

        public static Vector128<ushort> PopCount(Vector128<ushort> value);

        public static Vector128<byte> PopCount(Vector128<byte> value);

        public static Vector128<sbyte> PopCount(Vector128<sbyte> value);
        
        public static Vector256<byte> ShuffleBits(Vector256<ulong> value, Vector256<byte> control);

        public static Vector256<sbyte> ShuffleBits(Vector256<long> value, Vector256<sbyte> control);

        public static Vector256<byte> MaskShuffleBits(Vector256<byte> mask, Vector256<ulong> value, Vector256<byte> control);

        public static Vector256<sbyte> MaskShuffleBits(Vector256<sbyte> mask, Vector256<long> value, Vector256<sbyte> control);

        public static Vector128<byte> ShuffleBits(Vector128<ulong> value, Vector128<byte> control);

        public static Vector128<sbyte> ShuffleBits(Vector128<long> value, Vector128<sbyte> control);

        public static Vector128<byte> MaskShuffleBits(Vector128<byte> mask, Vector128<ulong> value, Vector128<byte> control);

        public static Vector128<sbyte> MaskShuffleBits(Vector128<sbyte> mask, Vector128<long> value, Vector128<sbyte> control);

    }
}

API Usage

var blocksPerRows = Avx512BitAlg.PopCount(board);

Alternative Designs

  • Avx512Vpopcntdq could have a different name.
  • MaskBitShuffle could have a different name and/or parameter/return types (e.g. mask and return type could be ulong instead of Vector512<byte>).

Risks

None
@MineCake147E MineCake147E added the api-suggestion Early API idea and discussion, it is NOT ready for implementation label Dec 19, 2023
@ghost ghost added the untriaged New issue has not been triaged by the area owner label Dec 19, 2023
@ghost
Copy link

ghost commented Dec 19, 2023

Tagging subscribers to this area: @dotnet/area-system-runtime-intrinsics
See info in area-owners.md if you want to be subscribed.

Issue Details

Background and motivation

Both VPOPCNTDQ and BITALG are supported by Intel in the Ice Lake and newer architectures, and by AMD in Zen 4.
VPOPCNTDQ allows for parallel popcnt in either Vector128, Vector256, or Vector512 for ulong and uint.
BITALG expands parallel popcnt for ushort and byte, and it also adds VPSHUFBITQMB instruction, which performs a bit gather select.
VPOPCNTW is highly beneficial for my current project, allowing me for counting filled blocks row-by-row for a bit-board of block games.

API Proposal

namespace System.Runtime.Intrinsics.X86;

[Intrinsic]
[CLSCompliant(false)]
public abstract class Avx512Vpopcntdq : Avx512DQ
{
    public static new bool IsSupported { get; }

    [Intrinsic]
    public new abstract class X64 : Avx512DQ.X64
    {
        internal X64() { }
        public static new bool IsSupported { get; }
    }

    public static Vector512<int> PopCount(Vector512<int> value);

    public static Vector512<uint> PopCount(Vector512<uint> value);

    public static Vector512<long> PopCount(Vector512<long> value);

    public static Vector512<ulong> PopCount(Vector512<ulong> value);

    public abstract class VL : Avx512DQ.VL
    {
        public static new bool IsSupported { get; }

        public static Vector256<int> PopCount(Vector256<int> value);

        public static Vector256<uint> PopCount(Vector256<uint> value);

        public static Vector256<long> PopCount(Vector256<long> value);

        public static Vector256<ulong> PopCount(Vector256<ulong> value);

        public static Vector128<int> PopCount(Vector128<int> value);

        public static Vector128<uint> PopCount(Vector128<uint> value);

        public static Vector128<long> PopCount(Vector128<long> value);

        public static Vector128<ulong> PopCount(Vector128<ulong> value);
    }
}

[Intrinsic]
[CLSCompliant(false)]
public abstract class Avx512BitAlg : Avx512BW
{
    public static new bool IsSupported { get; }

    [Intrinsic]
    public new abstract class X64 : Avx512BW.X64
    {
        internal X64() { }
        public static new bool IsSupported { get; }
    }

    public static Vector512<short> PopCount(Vector512<short> value);

    public static Vector512<ushort> PopCount(Vector512<ushort> value);

    public static Vector512<byte> PopCount(Vector512<byte> value);

    public static Vector512<sbyte> PopCount(Vector512<sbyte> value);

    public static Vector512<byte> BitShuffle(Vector512<ulong> value, Vector512<byte> control);

    public static Vector512<sbyte> BitShuffle(Vector512<long> value, Vector512<sbyte> control);

    public static Vector512<byte> MaskBitShuffle(Vector512<byte> mask, Vector512<ulong> value, Vector512<byte> control);

    public static Vector512<sbyte> MaskBitShuffle(Vector512<sbyte> mask, Vector512<long> value, Vector512<sbyte> control);

    public abstract class VL : Avx512BW.VL
    {
        public static new bool IsSupported { get; }

        public static Vector256<short> PopCount(Vector256<short> value);

        public static Vector256<ushort> PopCount(Vector256<ushort> value);

        public static Vector256<byte> PopCount(Vector256<byte> value);

        public static Vector256<sbyte> PopCount(Vector256<sbyte> value);

        public static Vector128<short> PopCount(Vector128<short> value);

        public static Vector128<ushort> PopCount(Vector128<ushort> value);

        public static Vector128<byte> PopCount(Vector128<byte> value);

        public static Vector128<sbyte> PopCount(Vector128<sbyte> value);
        
        public static Vector256<byte> BitShuffle(Vector256<ulong> value, Vector256<byte> control);

        public static Vector256<sbyte> BitShuffle(Vector256<long> value, Vector256<sbyte> control);

        public static Vector256<byte> MaskBitShuffle(Vector256<byte> mask, Vector256<ulong> value, Vector256<byte> control);

        public static Vector256<sbyte> MaskBitShuffle(Vector256<sbyte> mask, Vector256<long> value, Vector256<sbyte> control);

        public static Vector128<byte> BitShuffle(Vector128<ulong> value, Vector128<byte> control);

        public static Vector128<sbyte> BitShuffle(Vector128<long> value, Vector128<sbyte> control);

        public static Vector128<byte> MaskBitShuffle(Vector128<byte> mask, Vector128<ulong> value, Vector128<byte> control);

        public static Vector128<sbyte> MaskBitShuffle(Vector128<sbyte> mask, Vector128<long> value, Vector128<sbyte> control);

    }
}

API Usage

var blocksPerRows = Avx512BitAlg.PopCount(board);

Alternative Designs

  • Avx512Vpopcntdq could have a different name.
  • MaskBitShuffle could have a different name and/or parameter/return types (e.g. mask and return type could be ulong instead of Vector512<byte>).

Risks

None

Author: MineCake147E
Assignees: -
Labels:

api-suggestion, area-System.Runtime.Intrinsics
Milestone: -

@tannergooding tannergooding added api-ready-for-review API is ready for review, it is NOT ready for implementation and removed api-suggestion Early API idea and discussion, it is NOT ready for implementation untriaged New issue has not been triaged by the area owner labels Jan 2, 2024
@tannergooding
Copy link
Member

ShuffleBits might be a better name than BitShuffle and better matches the instruction name of VPSHUFBITQMB. Notable _mm_bitshuffle_epi64_mask is the name of the C API, however.

Returning a vector is consistent with how the other APIs that "return masks" work and should mesh with the existing pattern recognition allowing it to be consumed directly as a mask.

@MichalPetryka MichalPetryka mentioned this issue Feb 11, 2024
Merged
@terrajobst
Copy link
Contributor

terrajobst commented Feb 29, 2024 •
edited by dotnet-api-review bot
Loading

Video

  • Looks good as proposed
namespace System.Runtime.Intrinsics.X86;

[Intrinsic]
[CLSCompliant(false)]
public abstract class Avx512VPopcntDQ : Avx512DQ
{
    public static new bool IsSupported { get; }

    [Intrinsic]
    public new abstract class X64 : Avx512DQ.X64
    {
        public static new bool IsSupported { get; }
    }

    public static Vector512<int> PopCount(Vector512<int> value);
    public static Vector512<uint> PopCount(Vector512<uint> value);
    public static Vector512<long> PopCount(Vector512<long> value);
    public static Vector512<ulong> PopCount(Vector512<ulong> value);

    public abstract class VL : Avx512DQ.VL
    {
        public static new bool IsSupported { get; }

        public static Vector256<int> PopCount(Vector256<int> value);
        public static Vector256<uint> PopCount(Vector256<uint> value);
        public static Vector256<long> PopCount(Vector256<long> value);
        public static Vector256<ulong> PopCount(Vector256<ulong> value);
        public static Vector128<int> PopCount(Vector128<int> value);
        public static Vector128<uint> PopCount(Vector128<uint> value);
        public static Vector128<long> PopCount(Vector128<long> value);
        public static Vector128<ulong> PopCount(Vector128<ulong> value);
    }
}

[Intrinsic]
[CLSCompliant(false)]
public abstract class Avx512BitAlg : Avx512BW
{
    public static new bool IsSupported { get; }

    [Intrinsic]
    public new abstract class X64 : Avx512BW.X64
    {
        public static new bool IsSupported { get; }
    }

    public static Vector512<short> PopCount(Vector512<short> value);
    public static Vector512<ushort> PopCount(Vector512<ushort> value);
    public static Vector512<byte> PopCount(Vector512<byte> value);
    public static Vector512<sbyte> PopCount(Vector512<sbyte> value);
    public static Vector512<byte> ShuffleBits(Vector512<ulong> value, Vector512<byte> control);
    public static Vector512<sbyte> ShuffleBits(Vector512<long> value, Vector512<sbyte> control);
    public static Vector512<byte> MaskShuffleBits(Vector512<byte> mask, Vector512<ulong> value, Vector512<byte> control);
    public static Vector512<sbyte> MaskShuffleBits(Vector512<sbyte> mask, Vector512<long> value, Vector512<sbyte> control);

    public abstract class VL : Avx512BW.VL
    {
        public static new bool IsSupported { get; }

        public static Vector256<short> PopCount(Vector256<short> value);
        public static Vector256<ushort> PopCount(Vector256<ushort> value);
        public static Vector256<byte> PopCount(Vector256<byte> value);
        public static Vector256<sbyte> PopCount(Vector256<sbyte> value);
        public static Vector128<short> PopCount(Vector128<short> value);
        public static Vector128<ushort> PopCount(Vector128<ushort> value);
        public static Vector128<byte> PopCount(Vector128<byte> value);
        public static Vector128<sbyte> PopCount(Vector128<sbyte> value);       
        public static Vector256<byte> ShuffleBits(Vector256<ulong> value, Vector256<byte> control);
        public static Vector256<sbyte> ShuffleBits(Vector256<long> value, Vector256<sbyte> control);
        public static Vector256<byte> MaskShuffleBits(Vector256<byte> mask, Vector256<ulong> value, Vector256<byte> control);
        public static Vector256<sbyte> MaskShuffleBits(Vector256<sbyte> mask, Vector256<long> value, Vector256<sbyte> control);
        public static Vector128<byte> ShuffleBits(Vector128<ulong> value, Vector128<byte> control);
        public static Vector128<sbyte> ShuffleBits(Vector128<long> value, Vector128<sbyte> control);
        public static Vector128<byte> MaskShuffleBits(Vector128<byte> mask, Vector128<ulong> value, Vector128<byte> control);
        public static Vector128<sbyte> MaskShuffleBits(Vector128<sbyte> mask, Vector128<long> value, Vector128<sbyte> control);
    }
}

@terrajobst terrajobst added api-approved API was approved in API review, it can be implemented and removed api-ready-for-review API is ready for review, it is NOT ready for implementation labels Feb 29, 2024
@BruceForstall BruceForstall added the avx512 Related to the AVX-512 architecture label Mar 18, 2024
@Mrnikbobjeff
Copy link

This would be wonderful for our internal algorithm library as well. We need to compute hamming distances on larger number sets, and using the popcount x64 intrinsic is a limiting factor in our implementation.

@tannergooding tannergooding mentioned this issue Jun 11, 2024
Closed
@stephentoub stephentoub added this to the 10.0.0 milestone Jul 19, 2024
@BruceForstall BruceForstall mentioned this issue Dec 12, 2024
Open
40 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
api-approved API was approved in API review, it can be implemented area-System.Runtime.Intrinsics avx512 Related to the AVX-512 architecture
Projects
None yet
Milestone
10.0.0
Development

No branches or pull requests
6 participants
@stephentoub @terrajobst @Mrnikbobjeff @tannergooding @BruceForstall @MineCake147E