Class StrideDetector

Namespace

NumSharp.Backends.Kernels

Assembly

NumSharp.dll

Stride-based pattern detection for selecting optimal SIMD execution paths. All methods are aggressively inlined for minimal dispatch overhead.

public static class StrideDetector

Inheritance

object

StrideDetector

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Methods

CanSimdChunk<T>(int*, int*, int*, int)

Check if inner dimension is suitable for SIMD chunking. Returns true if both operands have inner stride of 1 (contiguous) or 0 (broadcast).

public static bool CanSimdChunk<T>(int* lhsStrides, int* rhsStrides, int* shape, int ndim) where T : unmanaged

Parameters

lhsStrides int*

rhsStrides int*

shape int*

ndim int

Returns

bool

Type Parameters

T

Classify<T>(int*, int*, int*, int)

Classify the binary operation into an execution path based on stride analysis. Classification priority:

1. SimdFull - both fully contiguous (fastest)
2. SimdScalarRight/Left - one operand is scalar
3. SimdChunk - inner dimension is contiguous/broadcast
4. General - fallback for arbitrary strides

public static ExecutionPath Classify<T>(int* lhsStrides, int* rhsStrides, int* shape, int ndim) where T : unmanaged

Parameters

lhsStrides int*

rhsStrides int*

shape int*

ndim int

Returns

ExecutionPath

Type Parameters

T

IsContiguous(int*, int*, int)

Check if array is fully contiguous (C-order). An array is contiguous if strides match expected C-order values: strides[n-1] = 1, strides[i] = strides[i+1] * shape[i+1]

public static bool IsContiguous(int* strides, int* shape, int ndim)

Parameters

strides int*

shape int*

ndim int

Returns

bool

IsScalar(int*, int)

Check if array is a scalar (all strides are zero). A scalar is broadcast to any shape - each element accesses the same value.

public static bool IsScalar(int* strides, int ndim)

Parameters

strides int*

ndim int

Returns

bool