search

dotnet / runtime

.NET is a cross-platform runtime for cloud, mobile, desktop, and IoT apps.
https://docs.microsoft.com/dotnet/core/
MIT License
15.15k stars 4.71k forks source link

[API Proposal]: Expose remaining AVX512-VBMI2 hardware instructions #88946

Open MadProbe opened 1 year ago

MadProbe commented 1 year ago

Background and motivation

There are approved and soon to be added AVX512-VBMI2 Compress & Expand intrinsics as part of new vector mask proposal. There is little reason if at all to not to add the left-over instructions from AVX512-VBMI2 instruction set. Notice:

  1. LeftLower and RightUpeer versions don't exist and can be replaced by lower << count and upper >> count, which produce the same result and the other operand is unused.

API Proposal

namespace System.Runtime.Intrinsics.X86
{
    [Intrinsic]
    public abstract class Avx512Vbmi2 : Avx512BW
    {
        public static new bool IsSupported { get => IsSupported; }

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

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

            /// <summary>
            /// __m128i _mm512_shldi_epi16 (__m128i a, __m128i b, int imm8)
            /// VPSHLDW xmm, xmm, xmm/m128, imm8
            /// </summary>
            public static Vector128<ushort> ConcatenateShiftLeftUpper(Vector128<ushort> upper, Vector128<ushort> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shldi_epi16 (__m256i a, __m256i b, int imm8)
            /// VPSHLDW ymm, ymm, ymm/m256, imm8
            /// </summary>
            public static Vector256<ushort> ConcatenateShiftLeftUpper(Vector256<ushort> upper, Vector256<ushort> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shldi_epi32 (__m128i a, __m128i b, int imm8)
            /// VPSHLDD xmm, xmm, xmm/m128, imm8
            /// </summary>
            public static Vector128<uint> ConcatenateShiftLeftUpper(Vector128<uint> upper, Vector128<uint> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shldi_epi32 (__m256i a, __m256i b, int imm8)
            /// VPSHLDD ymm, ymm, ymm/m256, imm8
            /// </summary>
            public static Vector256<uint> ConcatenateShiftLeftUpper(Vector256<uint> upper, Vector256<uint> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shldi_epi64 (__m128i a, __m128i b, int imm8)
            /// VPSHLDQ xmm, xmm, xmm/m128, imm8
            /// </summary>
            public static Vector128<ulong> ConcatenateShiftLeftUpper(Vector128<ulong> upper, Vector128<ulong> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shldi_epi64 (__m256i a, __m256i b, int imm8)
            /// VPSHLDQ ymm, ymm, ymm/m256, imm8
            /// </summary>
            public static Vector256<ulong> ConcatenateShiftLeftUpper(Vector256<ulong> upper, Vector256<ulong> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shrdi_epi16 (__m128i a, __m128i b, int imm8)
            /// VPSHRDW xmm, xmm, xmm/m128, imm8
            /// </summary>
            public static Vector128<ushort> ConcatenateShiftRightLower(Vector128<ushort> upper, Vector128<ushort> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shrdi_epi16 (__m256i a, __m256i b, int imm8)
            /// VPSHRDW ymm, ymm, ymm/m256, imm8
            /// </summary>
            public static Vector256<ushort> ConcatenateShiftRightLower(Vector256<ushort> upper, Vector256<ushort> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shrdi_epi32 (__m128i a, __m128i b, int imm8)
            /// VPSHRDD xmm, xmm, xmm/m128, imm8
            /// </summary>
            public static Vector128<uint> ConcatenateShiftRightLower(Vector128<uint> upper, Vector128<uint> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shrdi_epi32 (__m256i a, __m256i b, int imm8)
            /// VPSHRDD ymm, ymm, ymm/m256, imm8
            /// </summary>
            public static Vector256<uint> ConcatenateShiftRightLower(Vector256<uint> upper, Vector256<uint> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shrdi_epi64 (__m128i a, __m128i b, int imm8)
            /// VPSHRDQ xmm, xmm, xmm/m128, imm8
            /// </summary>
            public static Vector128<ulong> ConcatenateShiftRightLower(Vector128<ulong> upper, Vector128<ulong> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shrdi_epi64 (__m256i a, __m256i b, int imm8)
            /// VPSHRDQ ymm, ymm, ymm/m256, imm8
            /// </summary>
            public static Vector256<ulong> ConcatenateShiftRightLower(Vector256<ulong> upper, Vector256<ulong> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shldv_epi16 (__m128i a, __m128i b, __m128i c)
            /// VPSHLDVW xmm, xmm, xmm/m128
            /// </summary>
            public static Vector128<ushort> ConcatenateShiftLeftUpperVariable(Vector128<ushort> upper, Vector128<ushort> lower, Vector128<ushort> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shldv_epi16 (__m256i a, __m256i b, __m256i c)
            /// VPSHLDVW ymm, ymm, ymm/m256
            /// </summary>
            public static Vector256<ushort> ConcatenateShiftLeftUpperVariable(Vector256<ushort> upper, Vector256<ushort> lower, Vector256<ushort> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shldv_epi32 (__m128i a, __m128i b, __m128i c)
            /// VPSHLDVD xmm, xmm, xmm/m128
            /// </summary>
            public static Vector128<uint> ConcatenateShiftLeftUpperVariable(Vector128<uint> upper, Vector128<uint> lower, Vector128<uint> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shldv_epi32 (__m256i a, __m256i b, __m256i c)
            /// VPSHLDVD ymm, ymm, ymm/m256
            /// </summary>
            public static Vector256<uint> ConcatenateShiftLeftUpperVariable(Vector256<uint> upper, Vector256<uint> lower, Vector256<uint> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shldv_epi64 (__m128i a, __m128i b, __m128i c)
            /// VPSHLDVQ xmm, xmm, xmm/m128
            /// </summary>
            public static Vector128<ulong> ConcatenateShiftLeftUpperVariable(Vector128<ulong> upper, Vector128<ulong> lower, Vector128<ulong> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shldv_epi64 (__m256i a, __m256i b, __m256i c)
            /// VPSHLDVQ ymm, ymm, ymm/m256
            /// </summary>
            public static Vector256<ulong> ConcatenateShiftLeftUpperVariable(Vector256<ulong> upper, Vector256<ulong> lower, Vector256<ulong> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shrdv_epi16 (__m128i a, __m128i b, __m128i c)
            /// VPSHLDVW xmm, xmm, xmm/m128
            /// </summary>
            public static Vector128<ushort> ConcatenateShiftRightLowerVariable(Vector128<ushort> upper, Vector128<ushort> lower, Vector128<ushort> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shrdv_epi16 (__m256i a, __m256i b, __m256i c)
            /// VPSHLDVW ymm, ymm, ymm/m256
            /// </summary>
            public static Vector256<ushort> ConcatenateShiftRightLowerVariable(Vector256<ushort> upper, Vector256<ushort> lower, Vector256<ushort> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shrdv_epi32 (__m128i a, __m128i b, __m128i c)
            /// VPSHLDVD xmm, xmm, xmm/m128
            /// </summary>
            public static Vector128<uint> ConcatenateShiftRightLowerVariable(Vector128<uint> upper, Vector128<uint> lower, Vector128<uint> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shrdv_epi32 (__m256i a, __m256i b, __m256i c)
            /// VPSHLDVD ymm, ymm, ymm/m256
            /// </summary>
            public static Vector256<uint> ConcatenateShiftRightLowerVariable(Vector256<uint> upper, Vector256<uint> lower, Vector256<uint> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);

            /// <summary>
            /// __m128i _mm512_shrdv_epi64 (__m128i a, __m128i b, __m128i c)
            /// VPSHLDVQ xmm, xmm, xmm/m128
            /// </summary>
            public static Vector128<ulong> ConcatenateShiftRightLowerVariable(Vector128<ulong> upper, Vector128<ulong> lower, Vector128<ulong> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);

            /// <summary>
            /// __m256i _mm512_shrdv_epi64 (__m256i a, __m256i b, __m256i c)
            /// VPSHLDVQ ymm, ymm, ymm/m256
            /// </summary>
            public static Vector256<ulong> ConcatenateShiftRightLowerVariable(Vector256<ulong> upper, Vector256<ulong> lower, Vector256<ulong> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);
        }

        /// <summary>
        /// __m512i _mm512_shldi_epi16 (__m512i a, __m512i b, int imm8)
        /// VPSHLDW zmm, zmm, zmm/m512, imm8
        /// </summary>
        public static Vector512<ushort> ConcatenateShiftLeftUpper(Vector512<ushort> upper, Vector512<ushort> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shldi_epi32 (__m512i a, __m512i b, int imm8)
        /// VPSHLDD zmm, zmm, zmm/m512, imm8
        /// </summary>
        public static Vector512<uint> ConcatenateShiftLeftUpper(Vector512<uint> upper, Vector512<uint> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shldi_epi64 (__m512i a, __m512i b, int imm8)
        /// VPSHLDQ zmm, zmm, zmm/m512, imm8
        /// </summary>
        public static Vector512<ulong> ConcatenateShiftLeftUpper(Vector512<ulong> upper, Vector512<ulong> lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shrdi_epi16 (__m512i a, __m512i b, int imm8)
        /// VPSHRDW zmm, zmm, zmm/m512, imm8
        /// </summary>
        public static Vector512<ushort> ConcatenateShiftRightLower(Vector512<ushort> upper, Vector512<ushort> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shrdi_epi32 (__m512i a, __m512i b, int imm8)
        /// VPSHRDD zmm, zmm, zmm/m512, imm8
        /// </summary>
        public static Vector512<uint> ConcatenateShiftRightLower(Vector512<uint> upper, Vector512<uint> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shrdi_epi64 (__m512i a, __m512i b, int imm8)
        /// VPSHRDQ zmm, zmm, zmm/m512, imm8
        /// </summary>
        public static Vector512<ulong> ConcatenateShiftRightLower(Vector512<ulong> upper, Vector512<ulong> lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shldv_epi16 (__m512i a, __m512i b, __m512i c)
        /// VPSHLDVW zmm, zmm, zmm/m512
        /// </summary>
        public static Vector512<ushort> ConcatenateShiftLeftUpperVariable(Vector512<ushort> upper, Vector512<ushort> lower, Vector512<ushort> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shldv_epi32 (__m512i a, __m512i b, __m512i c)
        /// VPSHLDVD zmm, zmm, zmm/m512
        /// </summary>
        public static Vector512<uint> ConcatenateShiftLeftUpperVariable(Vector512<uint> upper, Vector512<uint> lower, Vector512<uint> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shldv_epi64 (__m512i a, __m512i b, __m512i c)
        /// VPSHLDVQ zmm, zmm, zmm/m512
        /// </summary>
        public static Vector512<ulong> ConcatenateShiftLeftUpperVariable(Vector512<ulong> upper, Vector512<ulong> lower, Vector512<ulong> count) => ConcatenateShiftLeftUpperVariable(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shrdv_epi16 (__m512i a, __m512i b, __m512i c)
        /// VPSHLDVW zmm, zmm, zmm/m512
        /// </summary>
        public static Vector512<ushort> ConcatenateShiftRightLowerVariable(Vector512<ushort> upper, Vector512<ushort> lower, Vector512<ushort> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shrdv_epi32 (__m512i a, __m512i b, __m512i c)
        /// VPSHLDVD zmm, zmm, zmm/m512
        /// </summary>
        public static Vector512<uint> ConcatenateShiftRightLowerVariable(Vector512<uint> upper, Vector512<uint> lower, Vector512<uint> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);

        /// <summary>
        /// __m512i _mm512_shrdv_epi64 (__m512i a, __m512i b, __m512i c)
        /// VPSHLDVQ zmm, zmm, zmm/m512
        /// </summary>
        public static Vector512<ulong> ConcatenateShiftRightLowerVariable(Vector512<ulong> upper, Vector512<ulong> lower, Vector512<ulong> count) => ConcatenateShiftRightLowerVariable(upper, lower, count);
    }
}

API Usage

// Avx512Vbmi2.ConcatenateShift(LeftUpper/RightLower)(...) example
Vector512<ushort> some_data = GetData(), some_data2 = GetData();
// result consists of lower 8 bit part of ushort data in upper_data and upper 8 bit of ushort data in lower_data
// shift count can be changed to get count of lower bits of ushort data in upper_data and upper 16 - count bit of ushort data in lower_data
var result = Avx512Vbmi2.ConcatenateShiftLeftUpper(upper_data, lower_data, 8);

Alternative Designs

N/A

Risks

N/A

ghost commented 1 year ago

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 There is already present [AVX512-VBMI PermuteVar64x8(x2) intrinsics](https://learn.microsoft.com/en-us/dotnet/api/system.runtime.intrinsics.x86.avx512vbmi.permutevar64x8x2?view=net-8.0) and approved and soon to be added [AVX512-VBMI2 Compress & Expand intrinsics](https://github.com/dotnet/runtime/issues/87097) as part of new vector mask proposal. There is little reason if at all to not to add the left-over instructions from aforementioned instruction sets. Notice: 1. LeftLower and RightUpeer versions don't exist and can be replaced by `lower << count` and `upper >> count`, which produce the same result and the other operand is unused. 2. I saw MultipleShift discussed in API Review stream, but I can't find it anywhere: Am I not searching hard enough or is it lost into oblivion? ### API Proposal ```csharp namespace System.Runtime.Intrinsics.X86 { [Intrinsic] public abstract class Avx512Vbmi : Avx512BW { public static new bool IsSupported { get => IsSupported; } [Intrinsic] public new abstract class X64 : Avx512BW.X64 { public static new bool IsSupported { get => IsSupported; } } [Intrinsic] public new abstract class VL : Avx512BW.VL { public static new bool IsSupported { get => IsSupported; } /// /// __m128i _mm128_multishift_epi64_epi8 (__m128i a, __m128i b) /// VPMULTISHIFTQB xmm, xmm, xmm /// public static Vector128 MultipleShift(Vector128 control, Vector128 source) => MultipleShift(control, source); /// /// __m256i _mm256_multishift_epi64_epi8 (__m256i a, __m256i b) /// VPMULTISHIFTQB ymm, ymm, ymm /// public static Vector256 MultipleShift(Vector256 control, Vector256 source) => MultipleShift(control, source); } /// /// __m512i _mm512_multishift_epi64_epi8 (__m512i a, __m512i b) /// VPMULTISHIFTQB zmm, zmm, zmm /// public static Vector512 MultipleShift(Vector512 control, Vector512 source) => MultipleShift(control, source); } [Intrinsic] public abstract class Avx512Vbmi2 : Avx512BW { public static new bool IsSupported { get => IsSupported; } [Intrinsic] public new abstract class X64 : Avx512BW.X64 { public static new bool IsSupported { get => IsSupported; } } [Intrinsic] public new abstract class VL : Avx512BW.VL { public static new bool IsSupported { get => IsSupported; } /// /// __m128i _mm512_shldi_epi16 (__m128i a, __m128i b, int imm8) /// VPSHLDW xmm, xmm, xmm, imm8 /// public static Vector128 ConcatenateShiftLeftUpper(Vector128 upper, Vector128 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m256i _mm512_shldi_epi16 (__m256i a, __m256i b, int imm8) /// VPSHLDW ymm, ymm, ymm, imm8 /// public static Vector256 ConcatenateShiftLeftUpper(Vector256 upper, Vector256 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m128i _mm512_shldi_epi32 (__m128i a, __m128i b, int imm8) /// VPSHLDD xmm, xmm, xmm, imm8 /// public static Vector128 ConcatenateShiftLeftUpper(Vector128 upper, Vector128 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m256i _mm512_shldi_epi32 (__m256i a, __m256i b, int imm8) /// VPSHLDD ymm, ymm, ymm, imm8 /// public static Vector256 ConcatenateShiftLeftUpper(Vector256 upper, Vector256 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m128i _mm512_shldi_epi64 (__m128i a, __m128i b, int imm8) /// VPSHLDQ xmm, xmm, xmm, imm8 /// public static Vector128 ConcatenateShiftLeftUpper(Vector128 upper, Vector128 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m256i _mm512_shldi_epi64 (__m256i a, __m256i b, int imm8) /// VPSHLDQ ymm, ymm, ymm, imm8 /// public static Vector256 ConcatenateShiftLeftUpper(Vector256 upper, Vector256 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m128i _mm512_shrdi_epi16 (__m128i a, __m128i b, int imm8) /// VPSHRDW xmm, xmm, xmm, imm8 /// public static Vector128 ConcatenateShiftRightLower(Vector128 upper, Vector128 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m256i _mm512_shrdi_epi16 (__m256i a, __m256i b, int imm8) /// VPSHRDW ymm, ymm, ymm, imm8 /// public static Vector256 ConcatenateShiftRightLower(Vector256 upper, Vector256 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m128i _mm512_shrdi_epi32 (__m128i a, __m128i b, int imm8) /// VPSHRDD xmm, xmm, xmm, imm8 /// public static Vector128 ConcatenateShiftRightLower(Vector128 upper, Vector128 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m256i _mm512_shrdi_epi32 (__m256i a, __m256i b, int imm8) /// VPSHRDD ymm, ymm, ymm, imm8 /// public static Vector256 ConcatenateShiftRightLower(Vector256 upper, Vector256 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m128i _mm512_shrdi_epi64 (__m128i a, __m128i b, int imm8) /// VPSHRDQ xmm, xmm, xmm, imm8 /// public static Vector128 ConcatenateShiftRightLower(Vector128 upper, Vector128 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m256i _mm512_shrdi_epi64 (__m256i a, __m256i b, int imm8) /// VPSHRDQ ymm, ymm, ymm, imm8 /// public static Vector256 ConcatenateShiftRightLower(Vector256 upper, Vector256 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m128i _mm512_shldv_epi16 (__m128i a, __m128i b, __m128i c) /// VPSHLDVW xmm, xmm, xmm /// public static Vector128 ConcatenateShiftLeftUpperVariable(Vector128 upper, Vector128 lower, Vector128 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m256i _mm512_shldv_epi16 (__m256i a, __m256i b, __m256i c) /// VPSHLDVW ymm, ymm, ymm /// public static Vector256 ConcatenateShiftLeftUpperVariable(Vector256 upper, Vector256 lower, Vector256 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m128i _mm512_shldv_epi32 (__m128i a, __m128i b, __m128i c) /// VPSHLDVD xmm, xmm, xmm /// public static Vector128 ConcatenateShiftLeftUpperVariable(Vector128 upper, Vector128 lower, Vector128 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m256i _mm512_shldv_epi32 (__m256i a, __m256i b, __m256i c) /// VPSHLDVD ymm, ymm, ymm /// public static Vector256 ConcatenateShiftLeftUpperVariable(Vector256 upper, Vector256 lower, Vector256 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m128i _mm512_shldv_epi64 (__m128i a, __m128i b, __m128i c) /// VPSHLDVQ xmm, xmm, xmm /// public static Vector128 ConcatenateShiftLeftUpperVariable(Vector128 upper, Vector128 lower, Vector128 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m256i _mm512_shldv_epi64 (__m256i a, __m256i b, __m256i c) /// VPSHLDVQ ymm, ymm, ymm /// public static Vector256 ConcatenateShiftLeftUpperVariable(Vector256 upper, Vector256 lower, Vector256 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m128i _mm512_shrdv_epi16 (__m128i a, __m128i b, __m128i c) /// VPSHLDVW xmm, xmm, xmm /// public static Vector128 ConcatenateShiftRightLowerVariable(Vector128 upper, Vector128 lower, Vector128 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); /// /// __m256i _mm512_shrdv_epi16 (__m256i a, __m256i b, __m256i c) /// VPSHLDVW ymm, ymm, ymm /// public static Vector256 ConcatenateShiftRightLowerVariable(Vector256 upper, Vector256 lower, Vector256 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); /// /// __m128i _mm512_shrdv_epi32 (__m128i a, __m128i b, __m128i c) /// VPSHLDVD xmm, xmm, xmm /// public static Vector128 ConcatenateShiftRightLowerVariable(Vector128 upper, Vector128 lower, Vector128 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); /// /// __m256i _mm512_shrdv_epi32 (__m256i a, __m256i b, __m256i c) /// VPSHLDVD ymm, ymm, ymm /// public static Vector256 ConcatenateShiftRightLowerVariable(Vector256 upper, Vector256 lower, Vector256 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); /// /// __m128i _mm512_shrdv_epi64 (__m128i a, __m128i b, __m128i c) /// VPSHLDVQ xmm, xmm, xmm /// public static Vector128 ConcatenateShiftRightLowerVariable(Vector128 upper, Vector128 lower, Vector128 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); /// /// __m256i _mm512_shrdv_epi64 (__m256i a, __m256i b, __m256i c) /// VPSHLDVQ ymm, ymm, ymm /// public static Vector256 ConcatenateShiftRightLowerVariable(Vector256 upper, Vector256 lower, Vector256 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); } /// /// __m512i _mm512_shldi_epi16 (__m512i a, __m512i b, int imm8) /// VPSHLDW zmm, zmm, zmm, imm8 /// public static Vector512 ConcatenateShiftLeftUpper(Vector512 upper, Vector512 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m512i _mm512_shldi_epi32 (__m512i a, __m512i b, int imm8) /// VPSHLDD zmm, zmm, zmm, imm8 /// public static Vector512 ConcatenateShiftLeftUpper(Vector512 upper, Vector512 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m512i _mm512_shldi_epi64 (__m512i a, __m512i b, int imm8) /// VPSHLDQ zmm, zmm, zmm, imm8 /// public static Vector512 ConcatenateShiftLeftUpper(Vector512 upper, Vector512 lower, [ConstantExpected] byte count) => ConcatenateShiftLeftUpper(upper, lower, count); /// /// __m512i _mm512_shrdi_epi16 (__m512i a, __m512i b, int imm8) /// VPSHRDW zmm, zmm, zmm, imm8 /// public static Vector512 ConcatenateShiftRightLower(Vector512 upper, Vector512 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m512i _mm512_shrdi_epi32 (__m512i a, __m512i b, int imm8) /// VPSHRDD zmm, zmm, zmm, imm8 /// public static Vector512 ConcatenateShiftRightLower(Vector512 upper, Vector512 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m512i _mm512_shrdi_epi64 (__m512i a, __m512i b, int imm8) /// VPSHRDQ zmm, zmm, zmm, imm8 /// public static Vector512 ConcatenateShiftRightLower(Vector512 upper, Vector512 lower, [ConstantExpected] byte count) => ConcatenateShiftRightLower(upper, lower, count); /// /// __m512i _mm512_shldv_epi16 (__m512i a, __m512i b, __m512i c) /// VPSHLDVW zmm, zmm, zmm /// public static Vector512 ConcatenateShiftLeftUpperVariable(Vector512 upper, Vector512 lower, Vector512 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m512i _mm512_shldv_epi32 (__m512i a, __m512i b, __m512i c) /// VPSHLDVD zmm, zmm, zmm /// public static Vector512 ConcatenateShiftLeftUpperVariable(Vector512 upper, Vector512 lower, Vector512 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m512i _mm512_shldv_epi64 (__m512i a, __m512i b, __m512i c) /// VPSHLDVQ zmm, zmm, zmm /// public static Vector512 ConcatenateShiftLeftUpperVariable(Vector512 upper, Vector512 lower, Vector512 count) => ConcatenateShiftLeftUpperVariable(upper, lower, count); /// /// __m512i _mm512_shrdv_epi16 (__m512i a, __m512i b, __m512i c) /// VPSHLDVW zmm, zmm, zmm /// public static Vector512 ConcatenateShiftRightLowerVariable(Vector512 upper, Vector512 lower, Vector512 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); /// /// __m512i _mm512_shrdv_epi32 (__m512i a, __m512i b, __m512i c) /// VPSHLDVD zmm, zmm, zmm /// public static Vector512 ConcatenateShiftRightLowerVariable(Vector512 upper, Vector512 lower, Vector512 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); /// /// __m512i _mm512_shrdv_epi64 (__m512i a, __m512i b, __m512i c) /// VPSHLDVQ zmm, zmm, zmm /// public static Vector512 ConcatenateShiftRightLowerVariable(Vector512 upper, Vector512 lower, Vector512 count) => ConcatenateShiftRightLowerVariable(upper, lower, count); } } ``` ### API Usage ```csharp // Avx512Vbmi2.ConcatenateShift(LeftUpper/RightLower)(...) example Vector512 some_data = GetData(), some_data2 = GetData(); // result consists of lower 8 bit part of ushort data in upper_data and upper 8 bit of ushort data in lower_data // shift count can be changed to get count of lower bits of ushort data in upper_data and upper 16 - count bit of ushort data in lower_data var result = Avx512Vbmi2.ConcatenateShiftLeftUpper(upper_data, lower_data, 8); ``` ```csharp // Avx512Vbmi.MultipleShift(...) example Vector512 d = GetData(), control = Vector512.Create(0x0B_1B_2B_3B_04_14_24_34); // Given control makes every 64 bit part of result structure is this: (d[59:52], d[43:36], d[27:20], d[11:4], (d[3:0] << 4) | d[63:60], d[51:44], d[35:28], d[19:12]) var result = Avx512Vbmi.MultupleShift(control, d); ``` ### Alternative Designs N/A ### Risks N/A
Author: MadProbe
Assignees: -
Labels: `api-suggestion`, `area-System.Runtime.Intrinsics`
Milestone: -
tannergooding commented 1 year ago

MultiShift is part of https://github.com/dotnet/runtime/issues/86168

ConcatenateShift hasn't gone through any review yet and I need to think more on the general name, etc. It's too late in the cycle for either to land for .NET 8, however. Both will end up being .NET 9 instead.

MadProbe commented 1 year ago

I have deleted the duplicated MultiShift and changed the title and proposal description accordingly.

tannergooding commented 1 year ago

Need to give some consideration around the suggested names, but this will be a .NET 9 change regardless; so marking with needs-further-triage for the moment.

lemire commented 7 months ago

Note that AMD Zen 4 and better as well as Intel Ice Lake and better support VBMI2.

As motivation, this would make it possible support fast Unicode transcoding function in C#, like what is done in the simdutf library (which is part of the Node.js runtime).

See https://arxiv.org/pdf/2212.05098.pdf

cc @EgorBo