[API Proposal]: Arm64: FEAT_I8MM

[a74nh]

namespace System.Runtime.Intrinsics.Arm;

/// VectorT Summary
public abstract partial class SveI8mm : AdvSimd /// Feature: FEAT_I8MM
{

  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3); // USDOT // MOVPRFX

  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3, ulong imm_index); // SUDOT // MOVPRFX

  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3); // USDOT // MOVPRFX

  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3, ulong imm_index); // USDOT // MOVPRFX

  /// T: [int, sbyte], [uint, byte]
  public static unsafe Vector<T> MatrixMultiplyAccumulate(Vector<T> op1, Vector<T2> op2, Vector<T2> op3); // SMMLA or UMMLA // MOVPRFX

  public static unsafe Vector<int> MatrixMultiplyAccumulateUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3); // USMMLA // MOVPRFX

  /// total method signatures: 6


  /// Optional Entries:

  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, byte op3); // USDOT // MOVPRFX

  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, sbyte op3); // USDOT // MOVPRFX

  /// total optional method signatures: 2

}


/// Full API
public abstract partial class SveI8mm : AdvSimd /// Feature: FEAT_I8MM
{
    /// DotProductSignedUnsigned : Dot product (signed × unsigned)

    /// svint32_t svsudot[_s32](svint32_t op1, svint8_t op2, svuint8_t op3) : "USDOT Ztied1.S, Zop3.B, Zop2.B" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop3.B, Zop2.B"
  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3);

    /// svint32_t svsudot_lane[_s32](svint32_t op1, svint8_t op2, svuint8_t op3, uint64_t imm_index) : "SUDOT Ztied1.S, Zop2.B, Zop3.B[imm_index]" or "MOVPRFX Zresult, Zop1; SUDOT Zresult.S, Zop2.B, Zop3.B[imm_index]"
  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3, ulong imm_index);


    /// DotProductUnsignedSigned : Dot product (unsigned × signed)

    /// svint32_t svusdot[_s32](svint32_t op1, svuint8_t op2, svint8_t op3) : "USDOT Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3);

    /// svint32_t svusdot_lane[_s32](svint32_t op1, svuint8_t op2, svint8_t op3, uint64_t imm_index) : "USDOT Ztied1.S, Zop2.B, Zop3.B[imm_index]" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop2.B, Zop3.B[imm_index]"
  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3, ulong imm_index);


    /// MatrixMultiplyAccumulate : Matrix multiply-accumulate

    /// svint32_t svmmla[_s32](svint32_t op1, svint8_t op2, svint8_t op3) : "SMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; SMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> MatrixMultiplyAccumulate(Vector<int> op1, Vector<sbyte> op2, Vector<sbyte> op3);

    /// svuint32_t svmmla[_u32](svuint32_t op1, svuint8_t op2, svuint8_t op3) : "UMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; UMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<uint> MatrixMultiplyAccumulate(Vector<uint> op1, Vector<byte> op2, Vector<byte> op3);


    /// MatrixMultiplyAccumulateUnsignedSigned : Matrix multiply-accumulate (unsigned × signed)

    /// svint32_t svusmmla[_s32](svint32_t op1, svuint8_t op2, svint8_t op3) : "USMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; USMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> MatrixMultiplyAccumulateUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3);


  /// total method signatures: 7
  /// total method names:      4
}

  /// Optional Entries:
  ///   public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, byte op3); // svsudot[_n_s32]
  ///   public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, sbyte op3); // svusdot[_n_s32]
  ///   Total Maybe: 2

  /// Total ACLE covered across API:      9

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

Issue Details

---

namespace System.Runtime.Intrinsics.Arm

/// VectorT Summary
public abstract class Sve : AdvSimd /// Feature: FEAT_I8MM
{
  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3);

  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3, ulong imm_index);

  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3);

  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3, ulong imm_index);

  /// T: [int, sbyte], [uint, byte]
  public static unsafe Vector<T> MatrixMultiplyAccumulate(Vector<T> op1, Vector<T2> op2, Vector<T2> op3); // SMMLA or UMMLA (MOVPRFX)

  public static unsafe Vector<int> MatrixMultiplyAccumulateUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3);

  /// total method signatures: 6
}


/// Full API
public abstract class Sve : AdvSimd /// Feature: FEAT_I8MM
{
    /// DotProductSignedUnsigned : Dot product (signed × unsigned)

    /// svint32_t svsudot[_s32](svint32_t op1, svint8_t op2, svuint8_t op3) : "USDOT Ztied1.S, Zop3.B, Zop2.B" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop3.B, Zop2.B"
  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3);

    /// svint32_t svsudot_lane[_s32](svint32_t op1, svint8_t op2, svuint8_t op3, uint64_t imm_index) : "SUDOT Ztied1.S, Zop2.B, Zop3.B[imm_index]" or "MOVPRFX Zresult, Zop1; SUDOT Zresult.S, Zop2.B, Zop3.B[imm_index]"
  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3, ulong imm_index);


    /// DotProductUnsignedSigned : Dot product (unsigned × signed)

    /// svint32_t svusdot[_s32](svint32_t op1, svuint8_t op2, svint8_t op3) : "USDOT Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3);

    /// svint32_t svusdot_lane[_s32](svint32_t op1, svuint8_t op2, svint8_t op3, uint64_t imm_index) : "USDOT Ztied1.S, Zop2.B, Zop3.B[imm_index]" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop2.B, Zop3.B[imm_index]"
  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3, ulong imm_index);


    /// MatrixMultiplyAccumulate : Matrix multiply-accumulate

    /// svint32_t svmmla[_s32](svint32_t op1, svint8_t op2, svint8_t op3) : "SMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; SMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> MatrixMultiplyAccumulate(Vector<int> op1, Vector<sbyte> op2, Vector<sbyte> op3);

    /// svuint32_t svmmla[_u32](svuint32_t op1, svuint8_t op2, svuint8_t op3) : "UMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; UMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<uint> MatrixMultiplyAccumulate(Vector<uint> op1, Vector<byte> op2, Vector<byte> op3);


    /// MatrixMultiplyAccumulateUnsignedSigned : Matrix multiply-accumulate (unsigned × signed)

    /// svint32_t svusmmla[_s32](svint32_t op1, svuint8_t op2, svint8_t op3) : "USMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; USMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> MatrixMultiplyAccumulateUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3);


  /// total method signatures: 7
  /// total method names:      4
}

  /// Rejected:
  ///   public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, byte op3); // svsudot[_n_s32]
  ///   public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, sbyte op3); // svusdot[_n_s32]
  ///   Total Rejected: 787

  /// Total ACLE covered across API:      9

Author:	a74nh
Assignees:	-
Labels:	area-System.Numerics
Milestone:	-

[a74nh]

This contributes to #93095

It covers all of the instructions in FEAT_I8MM. This an optional 8.2 feature and from 8.6 is mandatory in any hardware with SVE.

This list was auto generated from the C ACLE for SVE, and is in three parts:

The methods list reduced down to Vector versions. All possible varaints of T are given above the method.
The complete list of all methods. The corresponding ACLE methods and SVE instructions are given above the method.
All rejected ACLE methods. These are methods we have agreed that do not need including in C#.
Where possible, existing C# naming conventions have been matched.

Many of the C functions include predicate argument(s), of type svbool_t as the first argument. These are missing from the C# method. It is expected that the Jit will create predicates where required, or combine with uses of conditionalSelect(). For more discussion see #88140 comment.

[a74nh]

Updated to reflect review comments from other API proposals.

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

Issue Details

---

namespace System.Runtime.Intrinsics.Arm

/// VectorT Summary
public abstract class SveI8mm : AdvSimd /// Feature: FEAT_I8MM
{

  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3); // USDOT // MOVPRFX

  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3, ulong imm_index); // SUDOT // MOVPRFX

  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3); // USDOT // MOVPRFX

  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3, ulong imm_index); // USDOT // MOVPRFX

  /// T: [int, sbyte], [uint, byte]
  public static unsafe Vector<T> MatrixMultiplyAccumulate(Vector<T> op1, Vector<T2> op2, Vector<T2> op3); // SMMLA or UMMLA // MOVPRFX

  public static unsafe Vector<int> MatrixMultiplyAccumulateUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3); // USMMLA // MOVPRFX

  /// total method signatures: 6


  /// Optional Entries:

  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, byte op3); // USDOT // MOVPRFX

  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, sbyte op3); // USDOT // MOVPRFX

  /// total optional method signatures: 2

}


/// Full API
public abstract class SveI8mm : AdvSimd /// Feature: FEAT_I8MM
{
    /// DotProductSignedUnsigned : Dot product (signed × unsigned)

    /// svint32_t svsudot[_s32](svint32_t op1, svint8_t op2, svuint8_t op3) : "USDOT Ztied1.S, Zop3.B, Zop2.B" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop3.B, Zop2.B"
  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3);

    /// svint32_t svsudot_lane[_s32](svint32_t op1, svint8_t op2, svuint8_t op3, uint64_t imm_index) : "SUDOT Ztied1.S, Zop2.B, Zop3.B[imm_index]" or "MOVPRFX Zresult, Zop1; SUDOT Zresult.S, Zop2.B, Zop3.B[imm_index]"
  public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, Vector<byte> op3, ulong imm_index);


    /// DotProductUnsignedSigned : Dot product (unsigned × signed)

    /// svint32_t svusdot[_s32](svint32_t op1, svuint8_t op2, svint8_t op3) : "USDOT Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3);

    /// svint32_t svusdot_lane[_s32](svint32_t op1, svuint8_t op2, svint8_t op3, uint64_t imm_index) : "USDOT Ztied1.S, Zop2.B, Zop3.B[imm_index]" or "MOVPRFX Zresult, Zop1; USDOT Zresult.S, Zop2.B, Zop3.B[imm_index]"
  public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3, ulong imm_index);


    /// MatrixMultiplyAccumulate : Matrix multiply-accumulate

    /// svint32_t svmmla[_s32](svint32_t op1, svint8_t op2, svint8_t op3) : "SMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; SMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> MatrixMultiplyAccumulate(Vector<int> op1, Vector<sbyte> op2, Vector<sbyte> op3);

    /// svuint32_t svmmla[_u32](svuint32_t op1, svuint8_t op2, svuint8_t op3) : "UMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; UMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<uint> MatrixMultiplyAccumulate(Vector<uint> op1, Vector<byte> op2, Vector<byte> op3);


    /// MatrixMultiplyAccumulateUnsignedSigned : Matrix multiply-accumulate (unsigned × signed)

    /// svint32_t svusmmla[_s32](svint32_t op1, svuint8_t op2, svint8_t op3) : "USMMLA Ztied1.S, Zop2.B, Zop3.B" or "MOVPRFX Zresult, Zop1; USMMLA Zresult.S, Zop2.B, Zop3.B"
  public static unsafe Vector<int> MatrixMultiplyAccumulateUnsignedSigned(Vector<int> op1, Vector<byte> op2, Vector<sbyte> op3);


  /// total method signatures: 7
  /// total method names:      4
}

  /// Optional Entries:
  ///   public static unsafe Vector<int> DotProductSignedUnsigned(Vector<int> op1, Vector<sbyte> op2, byte op3); // svsudot[_n_s32]
  ///   public static unsafe Vector<int> DotProductUnsignedSigned(Vector<int> op1, Vector<byte> op2, sbyte op3); // svusdot[_n_s32]
  ///   Total Maybe: 2

  /// Total ACLE covered across API:      9

Author:	a74nh
Assignees:	-
Labels:	area-System.Runtime.Intrinsics, untriaged, api-ready-for-review
Milestone:	-

[tannergooding]

Fixed various method names to match those used in already exposed API surface and fixed the class to inherit from Sve