Class SimdHorizontalOperations

Namespace

DotCompute.Backends.CPU.Kernels.Simd

Assembly

DotCompute.Backends.CPU.dll

Horizontal operations for SIMD vectors - operations that combine elements within a single vector. Provides optimized horizontal sum, min, max, and product operations.

public static class SimdHorizontalOperations

Inheritance

object

SimdHorizontalOperations

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Methods

HorizontalMax(Vector128<float>)

Performs horizontal max of a 128-bit vector.

public static float HorizontalMax(Vector128<float> vector)

Parameters

vector Vector128<float>

Returns

float

HorizontalMax(Vector256<float>)

Performs horizontal max of a 256-bit vector.

public static float HorizontalMax(Vector256<float> vector)

Parameters

vector Vector256<float>

Returns

float

HorizontalMax(Vector512<float>)

Performs horizontal max of a 512-bit vector.

public static float HorizontalMax(Vector512<float> vector)

Parameters

vector Vector512<float>

Returns

float

HorizontalMaxNeon(Vector128<float>)

Performs horizontal max of a 128-bit float vector using ARM NEON.

public static float HorizontalMaxNeon(Vector128<float> vector)

Parameters

vector Vector128<float>

Returns

float

HorizontalMin(Vector128<float>)

Performs horizontal min of a 128-bit vector.

public static float HorizontalMin(Vector128<float> vector)

Parameters

vector Vector128<float>

Returns

float

HorizontalMin(Vector256<float>)

Performs horizontal min of a 256-bit vector.

public static float HorizontalMin(Vector256<float> vector)

Parameters

vector Vector256<float>

Returns

float

HorizontalMin(Vector512<float>)

Performs horizontal min of a 512-bit vector.

public static float HorizontalMin(Vector512<float> vector)

Parameters

vector Vector512<float>

Returns

float

HorizontalMinNeon(Vector128<float>)

Performs horizontal min of a 128-bit float vector using ARM NEON.

public static float HorizontalMinNeon(Vector128<float> vector)

Parameters

vector Vector128<float>

Returns

float

HorizontalProduct(Vector128<float>)

Performs horizontal product of a 128-bit vector.

public static float HorizontalProduct(Vector128<float> vector)

Parameters

vector Vector128<float>

Returns

float

HorizontalProduct(Vector256<float>)

Performs horizontal product of a 256-bit vector.

public static float HorizontalProduct(Vector256<float> vector)

Parameters

vector Vector256<float>

Returns

float

HorizontalProduct(Vector512<float>)

Performs horizontal product of a 512-bit vector.

public static float HorizontalProduct(Vector512<float> vector)

Parameters

vector Vector512<float>

Returns

float

HorizontalSum(Vector128<double>)

Performs horizontal sum of a 128-bit double vector.

public static double HorizontalSum(Vector128<double> vector)

Parameters

vector Vector128<double>

Returns

double

HorizontalSum(Vector128<int>)

Performs horizontal sum of a 128-bit integer vector.

public static int HorizontalSum(Vector128<int> vector)

Parameters

vector Vector128<int>

Returns

int

HorizontalSum(Vector128<float>)

Performs horizontal sum of a 128-bit vector.

public static float HorizontalSum(Vector128<float> vector)

Parameters

vector Vector128<float>

Returns

float

HorizontalSum(Vector256<double>)

Performs horizontal sum of a 256-bit double vector.

public static double HorizontalSum(Vector256<double> vector)

Parameters

vector Vector256<double>

Returns

double

HorizontalSum(Vector256<int>)

Performs horizontal sum of a 256-bit integer vector.

public static int HorizontalSum(Vector256<int> vector)

Parameters

vector Vector256<int>

Returns

int

HorizontalSum(Vector256<float>)

Performs horizontal sum of a 256-bit vector.

public static float HorizontalSum(Vector256<float> vector)

Parameters

vector Vector256<float>

Returns

float

HorizontalSum(Vector512<float>)

Performs horizontal sum of a 512-bit vector.

public static float HorizontalSum(Vector512<float> vector)

Parameters

vector Vector512<float>

Returns

float

HorizontalSumCrossPlatform(Vector128<float>)

Performs horizontal sum using the best available SIMD implementation for the current platform. Automatically selects AVX/SSE on x86 or NEON on ARM.

public static float HorizontalSumCrossPlatform(Vector128<float> vector)

Parameters

vector Vector128<float>

Returns

float

HorizontalSumNeon(Vector128<double>)

Performs horizontal sum of a 128-bit double vector using ARM NEON.

public static double HorizontalSumNeon(Vector128<double> vector)

Parameters

vector Vector128<double>

Returns

double

HorizontalSumNeon(Vector128<int>)

Performs horizontal sum of a 128-bit integer vector using ARM NEON.

public static int HorizontalSumNeon(Vector128<int> vector)

Parameters

vector Vector128<int>

Returns

int

HorizontalSumNeon(Vector128<float>)

Performs horizontal sum of a 128-bit float vector using ARM NEON. Uses Vector64 pairwise operations for maximum compatibility.

public static float HorizontalSumNeon(Vector128<float> vector)

Parameters

vector Vector128<float>

Returns

float

HorizontalSumPortable<T>(Vector<T>)

Performs horizontal sum using Vector<T> which auto-selects SIMD width. Portable across all platforms with automatic SIMD selection.

public static T HorizontalSumPortable<T>(Vector<T> vector) where T : struct, INumber<T>

Parameters

vector Vector<T>

Returns

T

Type Parameters

T

ParallelTreeReduction(float[])

Performs a parallel tree reduction on an array using SIMD operations. This is the most efficient way to reduce large arrays.

public static float ParallelTreeReduction(float[] data)

Parameters

data float[]

The input array to reduce.

Returns

float

The sum of all elements.

Remarks

Uses a three-phase approach for optimal performance:

1. SIMD reduction phase: processes Vector256 chunks in parallel
2. Horizontal reduction: combines SIMD results per thread
3. Final reduction: combines thread results

ParallelTreeReductionDouble(double[])

Performs a parallel tree reduction for double precision.

public static double ParallelTreeReductionDouble(double[] data)

Parameters

data double[]

Returns

double

ParallelTreeReductionInt(int[])

Performs a parallel tree reduction for integers.

public static long ParallelTreeReductionInt(int[] data)

Parameters

data int[]

Returns

long