Table of Contents

Class UnmanagedStorage

Namespace
NumSharp.Backends
Assembly
NumSharp.dll

Serves as a typed storage for an array.

public class UnmanagedStorage : ICloneable
Inheritance
UnmanagedStorage
Implements
Inherited Members

Remarks

Responsible for :

  • store data type, elements, Shape
  • offers methods for accessing elements depending on shape
  • offers methods for casting elements
  • offers methods for change tensor order
  • GetData always return reference object to the true storage
  • GetData{T} and SetData{T} change dtype and cast storage
  • CloneData always create a clone of storage and return this as reference object
  • CloneData{T} clone storage and cast this clone

Constructors

UnmanagedStorage(IArraySlice, Shape)

Wraps given arraySlice in UnmanagedStorage.

public UnmanagedStorage(IArraySlice arraySlice, Shape shape)

Parameters

arraySlice IArraySlice

The slice to wrap

shape Shape

UnmanagedStorage(DType)

Creates an empty storage described by descr — NumPy's PyArray_NewFromDescr entry: the descriptor IS the array's dtype (Descr), its class supplies the storage lane (TypeCode) and CLR element type (DType).

public UnmanagedStorage(DType descr)

Parameters

descr DType

The dtype descriptor; a non-native byte order is canonicalised (NumSharp buffers are always native).

Remarks

Usually Allocate(Shape, DType, bool) is called after this constructor.

Exceptions

ArgumentNullException

descr is null.

NotSupportedException

The descriptor's class has no storage lane yet (a datetime64/timedelta64 descriptor before Stage C).

UnmanagedStorage(NPTypeCode)

Creates an empty storage of type typeCode.

public UnmanagedStorage(NPTypeCode typeCode)

Parameters

typeCode NPTypeCode

The type of this storage

Remarks

Usually Allocate(Shape, Type) is called after this constructor.

UnmanagedStorage(bool)

public UnmanagedStorage(bool scalar)

Parameters

scalar bool

UnmanagedStorage(bool[])

public UnmanagedStorage(bool[] values)

Parameters

values bool[]

UnmanagedStorage(byte)

public UnmanagedStorage(byte scalar)

Parameters

scalar byte

UnmanagedStorage(byte[])

public UnmanagedStorage(byte[] values)

Parameters

values byte[]

UnmanagedStorage(char)

public UnmanagedStorage(char scalar)

Parameters

scalar char

UnmanagedStorage(char[])

public UnmanagedStorage(char[] values)

Parameters

values char[]

UnmanagedStorage(decimal)

public UnmanagedStorage(decimal scalar)

Parameters

scalar decimal

UnmanagedStorage(decimal[])

public UnmanagedStorage(decimal[] values)

Parameters

values decimal[]

UnmanagedStorage(double)

public UnmanagedStorage(double scalar)

Parameters

scalar double

UnmanagedStorage(double[])

public UnmanagedStorage(double[] values)

Parameters

values double[]

UnmanagedStorage(short)

public UnmanagedStorage(short scalar)

Parameters

scalar short

UnmanagedStorage(short[])

public UnmanagedStorage(short[] values)

Parameters

values short[]

UnmanagedStorage(int)

public UnmanagedStorage(int scalar)

Parameters

scalar int

UnmanagedStorage(int[])

public UnmanagedStorage(int[] values)

Parameters

values int[]

UnmanagedStorage(long)

public UnmanagedStorage(long scalar)

Parameters

scalar long

UnmanagedStorage(long[])

public UnmanagedStorage(long[] values)

Parameters

values long[]

UnmanagedStorage(float)

public UnmanagedStorage(float scalar)

Parameters

scalar float

UnmanagedStorage(float[])

public UnmanagedStorage(float[] values)

Parameters

values float[]

UnmanagedStorage(Type)

Creates an empty storage of type dtype.

public UnmanagedStorage(Type dtype)

Parameters

dtype Type

The type of this storage

Remarks

Usually Allocate(Shape, Type) is called after this constructor.

UnmanagedStorage(ushort)

public UnmanagedStorage(ushort scalar)

Parameters

scalar ushort

UnmanagedStorage(ushort[])

public UnmanagedStorage(ushort[] values)

Parameters

values ushort[]

UnmanagedStorage(uint)

public UnmanagedStorage(uint scalar)

Parameters

scalar uint

UnmanagedStorage(uint[])

public UnmanagedStorage(uint[] values)

Parameters

values uint[]

UnmanagedStorage(ulong)

public UnmanagedStorage(ulong scalar)

Parameters

scalar ulong

UnmanagedStorage(ulong[])

public UnmanagedStorage(ulong[] values)

Parameters

values ulong[]

Fields

Address

public byte* Address

Field Value

byte*

Count

public long Count

Field Value

long

InternalArray

public IArraySlice InternalArray

Field Value

IArraySlice

_descr

The dtype DESCRIPTOR of this storage (NumPy's PyArrayObject.descr) — see Descr. Lazily resolved from _typecode for the storage-backed builtins (each has exactly one canonical descriptor, the class singleton), and reset whenever the storage's lane changes (_Allocate(Shape, IArraySlice), ReplaceData). A parametric descriptor (a future datetime64[ns], whose unit lives on the instance rather than in the enum) is stored here explicitly and travels with aliases/clones.

protected DType _descr

Field Value

DType

_dtype

protected Type _dtype

Field Value

Type

_shape

protected Shape _shape

Field Value

Shape

_typecode

protected NPTypeCode _typecode

Field Value

NPTypeCode

Properties

BaseStorage

Gets the original storage this is a view of, or null if this storage owns its data.

public UnmanagedStorage? BaseStorage { get; }

Property Value

UnmanagedStorage

The ultimate owner storage for views, or null for owned data.

Remarks

NumPy-compatible: All views chain to the ultimate owner (not intermediate views).

Example:

var a = np.arange(10);           // a.Storage.BaseStorage == null (owns data)
var b = a["2:5"];                // b.Storage.BaseStorage == a.Storage
var c = b["1:2"];                // c.Storage.BaseStorage == a.Storage (chains to original!)

Note: This property is read-only by design. Allowing external modification would risk breaking the memory ownership chain and could lead to use-after-free bugs.

See Also

DType

The data type of internal storage array.

public Type DType { get; }

Property Value

Type

numpys equal dtype

Remarks

Has to be compliant with NPTypeCode.

DTypeSize

The size in bytes of a single value of DType as stored in the unmanaged buffer.

public int DTypeSize { get; }

Property Value

int

Remarks

Returns the in-memory element stride, not the marshaling size. For bool that is 1, not SizeOf(object)'s 4 (bool is marshaled to win32 BOOL = int). All pointer arithmetic over Address uses this value, so the in-memory layout is the only correct reference.

Exceptions

NotSupportedException

This storage holds NEP 55 StringDType text. Every caller of this property moves or reinterprets elements as raw bytes of this width, which is never valid for a StringDType cell (an arena offset or a heap pointer that only its own buffer's allocator can read), so the property refuses: a route without its StringDType form fails loudly instead of copying cells. Code that only needs the descriptor's item size (strides in bytes, nbytes, overlap bounds) reads ItemBytes.

Descr

The dtype DESCRIPTOR of this storage — NumPy's PyArrayObject.descr, the object ndarray.dtype returns. For the storage-backed builtins this is the class's canonical instance (np.dtype("int32") for an Int32 lane — the same object every time, so a.dtype == b.dtype is a cheap structural compare); a parametric descriptor is carried explicitly. DType (the CLR element Type) and TypeCode stay the kernel currency — the descriptor is the identity, they are its storage keys.

public DType Descr { get; }

Property Value

DType

Engine

The engine that was used to create this UnmanagedStorage.

public TensorEngine Engine { get; protected set; }

Property Value

TensorEngine

ExternalBase

Set when this storage's memory belongs to a FOREIGN owner even though no NumSharp.Backends.UnmanagedStorage._baseStorage exists — a memory-mapped file (every mmap_mode, the file mapping is the real owner), or an external buffer wrap. It is the NumPy analog of a non-array base object: such arrays report flags.owndata == False (NumPy's memmap does, in every mode) and np.require(…, "O") copies them.

public bool ExternalBase { get; }

Property Value

bool

Remarks

Orthogonal to WriteProtected: an 'r+' memmap is externally based yet writable; an 'r' memmap is both. Readable by everyone, set only inside NumSharp when the storage is created: the shape's cached OWNDATA bit is computed from it then, so a later change would leave flags.owndata stale.

IsDataAligned

Is every element of this storage's array aligned for its dtype — the memory half of NumPy's ALIGNED flag (raw_array_is_aligned, numpy/_core/src/common/array_assign.c), which NumPy recomputes for every new array object. flags reports aligned only when this holds AND the shape's ALIGNED bit (cleared by setflags(align=False)) is set; setflags(align=True) refuses when it does not hold.

public bool IsDataAligned { get; }

Property Value

bool

Remarks

NumPy ORs the data pointer with the byte stride of every axis longer than 1 and tests the low bits. NumSharp's offsets and strides are whole elements, and every dtype's itemsize is a multiple of its alignment, so those terms never add a misaligned bit: the storage's base address alone decides. An empty array is always aligned, as in NumPy.

NumSharp's own allocations are always aligned. Misaligned data comes from a caller's buffer at an odd byte offset (np.ndarray(shape, buffer=b, offset=3), np.frombuffer(b, offset=3)) or a field view at an odd byte offset (getfield), and every view of such an array shares its address, so the flag follows to every view without each view path computing it.

Recomputed on every read (one address test), so it follows the storage as it is now.

IsFixedString

True when this storage holds fixed-width strings (S or U, a StringArraySlice).

public bool IsFixedString { get; }

Property Value

bool

IsString

True when this storage holds TEXT — the String lane, shared by the fixed-width strings (S, U: a StringArraySlice) and NEP 55 StringDType (T: a VStringArraySlice). The two layouts have nothing in common below the lane: test IsFixedString / IsVString (or the descriptor's class) before reading elements as bytes.

public bool IsString { get; }

Property Value

bool

Remarks

A route that branches on this property and then treats the elements as fixed-width units reaches StringSlice, which refuses a StringDType storage with a clear NotSupportedException: every text operation has its own StringDType route (tested with IsVString first), so a T storage that gets here is a route that skipped that test, and it fails loudly instead of copying cells that are meaningless without their own allocator.

IsVString

True when this storage holds NEP 55 StringDType text (T, a VStringArraySlice).

public bool IsVString { get; }

Property Value

bool

IsView

Gets a value indicating whether this storage is a view of another storage.

public bool IsView { get; }

Property Value

bool

true if this storage shares memory with another storage (does not own its data); false if this storage owns its data.

Remarks

Equivalent to checking BaseStorage != null.

Use cases:

  • Determine if an array can be safely modified without affecting other arrays
  • Optimize copy-on-write patterns
  • Debug memory sharing issues

ItemBytes

The descriptor's item size in bytes (NumPy's dtype.itemsize) for every storage: DTypeSize for the numeric and fixed-width string lanes, and 16 (one packed cell) for a NEP 55 StringDType storage.

public int ItemBytes { get; }

Property Value

int

Remarks

Use it for byte ARITHMETIC that never moves an element (byte strides, nbytes, the extent of a view for an overlap test). A route that copies or reinterprets elements must read DTypeSize, which refuses a StringDType storage.

OwnsData

Does this storage own the memory it addresses — NumPy's OWNDATA: no base storage (NumSharp.Backends.UnmanagedStorage._baseStorage) and no foreign owner (ExternalBase). Read by ndarray.flags.owndata and np.require("O").

public bool OwnsData { get; }

Property Value

bool

Remarks

Answers for the storage object itself: a view's storage reports false even when the array it views owns its buffer.

Shape

The shape representing the data in this storage.

public Shape Shape { get; set; }

Property Value

Shape

ShapeReference

The shape representing the data in this storage.

public ref Shape ShapeReference { get; }

Property Value

Shape

Remarks

It is dangerous to set Shape by reference. use Reshape(Shape) instead.

StringSlice

The StringArraySlice of a fixed-width string storage (null for a numeric lane) — the bytes plus the element width they are read with.

public StringArraySlice StringSlice { get; }

Property Value

StringArraySlice

Remarks

The slice covers the whole buffer: a view's elements start at its Shape's offset and follow its strides, both counted in elements of the slice's width.

Exceptions

NotSupportedException

This storage holds StringDType text: every caller of this property is a fixed-width route (unit views, the byte codec, width arithmetic), and StringDType text takes its own routes (VStringSlice), so it refuses here rather than reading the 16-byte cells as fixed-width text.

TypeCode

The NPTypeCode of DType.

public NPTypeCode TypeCode { get; }

Property Value

NPTypeCode

VStringSlice

The VStringArraySlice of a StringDType storage (null for every other storage).

public VStringArraySlice VStringSlice { get; }

Property Value

VStringArraySlice

Remarks

The route from an array to its texts: .Cells are the 16-byte cells of the whole buffer (a view's elements start at its Shape's offset), and .Allocator is the one allocator that may read or write them.

WriteProtected

Set when this storage wraps FOREIGN memory that must never become writeable — a read-only memory-mapped file (np.load(mmap_mode: "r")), or a buffer whose exporter declared it read-only (np.frombuffer over a read-only MemoryView). It is the gate NumPy expresses through the buffer protocol (its _IsWriteable asks the non-array base for a writable buffer and a read-only exporter refuses): without it, setflags(write: true) on an 'r' memmap would re-enable WRITEABLE and the next write would reach PROT_READ pages — an access violation, not an exception.

public bool WriteProtected { get; }

Property Value

bool

Remarks

Ordinary owned allocations and ordinary views never set this; a cleared WRITEABLE flag alone (e.g. a.flags.writeable = false on your own array) is reversible and does NOT imply it.

It answers only that question: whether a cleared WRITEABLE flag may be set again. It does not block a write while the flag is set. So np.nditer's views carry it too, writeable ones included (NDIterRef.CurrentView): their NumPy base is the iterator, an object that exports no buffer at all, so a write flag cleared on one cannot be set again.

Readable by everyone, set only inside NumSharp (where the storage over the foreign memory is created): clearing it on a read-only mapping would let the next write reach PROT_READ pages.

Methods

Alias()

Creates an alias (view) of this storage that shares the same underlying memory.

public UnmanagedStorage Alias()

Returns

UnmanagedStorage

A new UnmanagedStorage that shares memory with this storage. The returned storage's NumSharp.Backends.UnmanagedStorage._baseStorage points to the ultimate owner.

Remarks

Memory Sharing: The alias shares the same InternalArray and underlying memory. Modifications through the alias affect the original data.

Base Tracking: Sets _baseStorage to chain to the ultimate owner:

  • If this storage owns its data: alias._baseStorage = this
  • If this storage is a view: alias._baseStorage = this._baseStorage

This ensures all views in a chain point to the original owner, not intermediate views.

See Also

Alias(Shape)

Creates an alias (view) of this storage with a different shape.

public UnmanagedStorage Alias(Shape shape)

Parameters

shape Shape

The shape for the alias. Should be compatible with the storage size (not validated).

Returns

UnmanagedStorage

A new UnmanagedStorage that shares memory with this storage but has the specified shape. The returned storage's NumSharp.Backends.UnmanagedStorage._baseStorage points to the ultimate owner.

Remarks

Memory Sharing: The alias shares the same InternalArray and underlying memory. Modifications through the alias affect the original data.

Shape Compatibility: This method does NOT validate that the shape is compatible with the storage size. Use with caution.

Base Tracking: Sets _baseStorage to chain to the ultimate owner.

See Also

Alias(ref Shape)

Creates an alias (view) of this storage with a different shape (by reference).

public UnmanagedStorage Alias(ref Shape shape)

Parameters

shape Shape

The shape for the alias. Should be compatible with the storage size (not validated).

Returns

UnmanagedStorage

A new UnmanagedStorage that shares memory with this storage but has the specified shape. The returned storage's NumSharp.Backends.UnmanagedStorage._baseStorage points to the ultimate owner.

Remarks

Memory Sharing: The alias shares the same InternalArray and underlying memory. Modifications through the alias affect the original data.

Shape Compatibility: This method does NOT validate that the shape is compatible with the storage size. Use with caution.

Base Tracking: Sets _baseStorage to chain to the ultimate owner.

See Also

AliasAs(NPTypeCode)

Creates an alias (view) of this storage with a different dtype, reinterpreting bytes.

public UnmanagedStorage AliasAs(NPTypeCode typeCode)

Parameters

typeCode NPTypeCode

The new dtype to interpret the bytes as.

Returns

UnmanagedStorage

A view with reinterpreted bytes.

AliasAs(Type)

Creates an alias (view) of this storage with a different dtype, reinterpreting bytes.

public UnmanagedStorage AliasAs(Type dtype)

Parameters

dtype Type

The new dtype to interpret the bytes as.

Returns

UnmanagedStorage

A view with reinterpreted bytes.

AliasAs<T>()

Creates an alias (view) of this storage with a different dtype, reinterpreting bytes — the storage half of NumPy's ndarray.view(dtype).

public UnmanagedStorage AliasAs<T>() where T : unmanaged

Returns

UnmanagedStorage

A new UnmanagedStorage that shares memory with this storage but interprets the bytes as a different type. Shape is adjusted if type sizes differ.

Type Parameters

T

The new dtype to interpret the bytes as.

Remarks

Byte Reinterpretation: This does NOT convert values. It reinterprets the raw bytes as a different type, like NumPy's view(). For example, viewing float64 as int64 will show the IEEE 754 bit patterns, not converted values.

Same itemsize: The dtype tag changes and the shape, strides and offset are kept verbatim, so ANY layout works (C/F-contiguous, sliced, strided, transposed, negative-stride, broadcast, 0-d). A read-only source (broadcast) stays read-only.

Different itemsize (NumPy 2.x rule): only the LAST axis must be contiguous (byte-stride == old itemsize) — the whole array need NOT be C-contiguous. That last axis is resized by the size ratio while every outer axis keeps its byte stride, so a[::2].view(...) and other outer-strided/offset views are supported, matching NumPy. Read-only (broadcast) sources stay read-only.

Exceptions

ValueError

NumPy's errors with NumPy's texts (ReinterpretShape(int, int, out long)): a 0-d array whose itemsize changes, a non-contiguous last axis, a smaller dtype that does not divide the old itemsize, a larger one that does not divide the last axis' byte length.

InvalidOperationException

NumSharp stores strides/offset in ELEMENTS rather than bytes, so the rare strided+offset view whose byte offset/stride is not a whole multiple of the new itemsize is refused (NumPy, which carries a byte pointer, would allow it).

AliasComplexLane(bool)

Creates a double (float64) VIEW onto one lane — real or imaginary — of a Complex storage, reproducing NumPy's a.real / a.imag: a strided float64 view that SHARES memory (and writeability) with the Complex base.

public UnmanagedStorage AliasComplexLane(bool imaginary)

Parameters

imaginary bool

false selects the real lane, true the imaginary lane.

Returns

UnmanagedStorage

A float64 UnmanagedStorage aliasing this storage's chosen lane. Writes through to the Complex base; the base's buffer is kept alive by the lane's counted reference on the base's block (WrapShared<T>(void*, long, IArraySlice)) — disposing the base while the lane lives does not free it — and the base object by _baseStorage.

Remarks

A Complex is two consecutive float64s (real, then imaginary), laid out exactly like NumPy's complex128. A lane is therefore a plain strided float64 view: every Complex element stride is doubled (a Complex spans two float64s) and the base offset is doubled, plus one for the imaginary lane. This works for EVERY layout (contiguous / F-contiguous / transposed / strided / negative-stride / sliced-offset / broadcast), because it operates on the raw contiguous backing buffer and mirrors whatever strides/offset the Complex shape carries — the same strided-lane read the FFT driver uses.

A lane of a read-only Complex (a broadcast view, or an 'r' memmap) stays read-only.

Exceptions

InvalidOperationException

If this storage is not a Complex storage.

Allocate(IArraySlice, Shape, bool)

Makes values this storage's data and installs shape's layout over it AS GIVEN — strides, offset and flags included. This is the analog of NumPy's C API (PyArray_NewFromDescr with a data pointer and strides, where the caller is responsible for correctness), not of the Python-level np.ndarray(shape, buffer=...).

public void Allocate(IArraySlice values, Shape shape, bool copy = false)

Parameters

values IArraySlice

The buffer to adopt as this storage's data.

shape Shape

The exact layout to install. Every element it addresses must lie inside values; nothing here checks that.

copy bool

True to adopt a clone of values instead of the slice itself.

Remarks

Does not copy values unless copy is true.

A view's layout installed here reads past the buffer: for x = arange(20)["::4"], Allocate(fiveElements, x.Shape) reads element i at i * 4. The public wrap constructors (new NDArray(IArraySlice, Shape, char), new NDArray(Array, Shape, char)) therefore lay out the dimensions afresh first (ForWrappedBuffer(char)), as NumPy's ndarray does, and only pass a layout they built.

Exceptions

ArgumentNullException

shape is uninitialized (default(Shape)).

ArgumentException

The element count of values is not the size of shape.

Allocate(Shape, DType, bool)

Allocates a new Array into memory described by a dtype DESCRIPTOR (NumPy's PyArray_NewFromDescr): the descriptor becomes Descr, its class supplies the lane.

public void Allocate(Shape shape, DType dtype, bool fillZeros)

Parameters

shape Shape

The dimensions to allocate. Only a fresh layout (C, F or a dense permutation at offset 0) is kept; a view's layout is re-laid out in C order (NumSharp.Shape.ForFreshAllocation()), which is what makes new NDArray(dtype, x.Shape), np.empty(x.Shape) and np.zeros(x.Shape) safe for a view x.

dtype DType

The dtype descriptor of the array; a non-native byte order is canonicalised (NumSharp buffers are always native).

fillZeros bool

Zero the fresh buffer, else leave the allocation noise.

Exceptions

ArgumentNullException

shape is uninitialized, or dtype is null.

NotSupportedException

The descriptor's class has no storage lane yet (a datetime64/timedelta64 descriptor before Stage C).

ValueError

A dimension is negative, or the byte count overflows (NumSharp.AllocationGuard).

Allocate(Shape, NPTypeCode, bool)

Allocates a new buffer for shape's dimensions and makes it this storage's data — NumPy's PyArray_NewFromDescr with no data pointer.

public void Allocate(Shape shape, NPTypeCode dtype, bool fillZeros)

Parameters

shape Shape

The dimensions to allocate. Only a fresh layout (C, F or a dense permutation at offset 0) is kept; a view's layout is re-laid out in C order (NumSharp.Shape.ForFreshAllocation()).

dtype NPTypeCode

The storage lane to allocate.

fillZeros bool

Zero the fresh buffer, else leave the allocation noise (the caller writes every element).

Exceptions

ArgumentNullException

shape is uninitialized, or dtype is Empty.

NotSupportedException

dtype is String and this storage has no string descriptor to keep (the lane alone names no width; see From(NPTypeCode)).

ValueError

A dimension is negative, or the byte count overflows (NumSharp.AllocationGuard).

Allocate(Shape, Type)

Allocates a new, zero-filled buffer for shape's dimensions and makes it this storage's data — NumPy's PyArray_NewFromDescr with no data pointer.

public void Allocate(Shape shape, Type dtype = null)

Parameters

shape Shape

The dimensions to allocate. Only a fresh layout (C, F or a dense permutation at offset 0) is kept; a view's layout (stepped, offset, reversed, broadcast) is re-laid out in C order (NumSharp.Shape.ForFreshAllocation()), so Allocate(x.Shape) of a view never addresses past the new buffer.

dtype Type

The type of the Array, if null DType is used.

Exceptions

ArgumentNullException

shape is uninitialized (default(Shape)).

ValueError

A dimension is negative, or the byte count overflows (NumSharp.AllocationGuard).

Allocate(Shape, Type, bool)

Allocates a new buffer for shape's dimensions and makes it this storage's data — NumPy's PyArray_NewFromDescr with no data pointer.

public void Allocate(Shape shape, Type dtype, bool fillZeros)

Parameters

shape Shape

The dimensions to allocate. Only a fresh layout (C, F or a dense permutation at offset 0) is kept; a view's layout is re-laid out in C order (NumSharp.Shape.ForFreshAllocation()).

dtype Type

The type of the Array, if null DType is used.

fillZeros bool

Zero the fresh buffer, else leave the allocation noise (the caller writes every element).

Exceptions

ArgumentNullException

shape is uninitialized (default(Shape)).

ValueError

A dimension is negative, or the byte count overflows (NumSharp.AllocationGuard).

Allocate(Array)

Allocate array into memory.

public void Allocate(Array array)

Parameters

array Array

The array to set as internal data storage

Remarks

Does not copy array

Allocate(Array, Shape)

Allocate values into memory.

public void Allocate(Array values, Shape shape)

Parameters

values Array

The array to set as internal data storage

shape Shape

The shape of given array

Remarks

Does not copy values

Allocate<T>(ArraySlice<T>, Shape, bool)

Assign this ArraySlice<T> as the internal array storage and assign shape to it.

public void Allocate<T>(ArraySlice<T> values, Shape shape, bool copy = false) where T : unmanaged

Parameters

values ArraySlice<T>

The array to set as internal data storage

shape Shape

The shape of the array.

copy bool

Should perform a copy of values

Type Parameters

T

Remarks

Does not copy values

Allocate<T>(T[])

Allocate values into memory.

public void Allocate<T>(T[] values) where T : unmanaged

Parameters

values T[]

The array to set as internal data storage

Type Parameters

T

Remarks

Does not copy values

AsSpan<T>()

Returns an UnmanagedSpan representing this storage's memory.

public Span<T> AsSpan<T>() where T : unmanaged

Returns

Span<T>

Type Parameters

T

Remarks

This ignores completely slicing. Supports long indexing for arrays > 2B elements.

AsUnmanagedSpan<T>()

Returns an UnmanagedSpan representing this storage's memory.

public UnmanagedSpan<T> AsUnmanagedSpan<T>() where T : unmanaged

Returns

UnmanagedSpan<T>

Type Parameters

T

Remarks

This ignores completely slicing. Supports long indexing for arrays > 2B elements.

Cast(NPTypeCode)

Return a casted UnmanagedStorage to a specific dtype.

public UnmanagedStorage Cast(NPTypeCode typeCode)

Parameters

typeCode NPTypeCode

The dtype to convert to

Returns

UnmanagedStorage

A copy of this UnmanagedStorage casted to a specific dtype.

Remarks

Always copies, If dtype==typeof(T) then a Clone() is returned.

Cast(Type)

Return a casted UnmanagedStorage to a specific dtype.

public UnmanagedStorage Cast(Type dtype)

Parameters

dtype Type

The dtype to convert to

Returns

UnmanagedStorage

A copy of this UnmanagedStorage casted to a specific dtype.

Remarks

Always copies, If dtype==typeof(T) then a Clone() is returned.

CastIfNecessary(NPTypeCode)

Return a casted UnmanagedStorage to a specific dtype only if necessary

public UnmanagedStorage CastIfNecessary(NPTypeCode typeCode)

Parameters

typeCode NPTypeCode

The dtype to convert to

Returns

UnmanagedStorage

A copy of this UnmanagedStorage casted to a specific dtype.

Remarks

Copies only if dtypes does not match typeCode

CastIfNecessary(Type)

Return a casted UnmanagedStorage to a specific dtype.

public UnmanagedStorage CastIfNecessary(Type dtype)

Parameters

dtype Type

The dtype to convert to

Returns

UnmanagedStorage

A copy of this UnmanagedStorage casted to a specific dtype.

Remarks

Copies only if dtypes does not match typeCode

CastIfNecessary<T>()

Return a casted UnmanagedStorage to a specific dtype only if necessary.

public UnmanagedStorage CastIfNecessary<T>() where T : unmanaged

Returns

UnmanagedStorage

A copy of this UnmanagedStorage casted to a specific dtype.

Type Parameters

T

The dtype to convert to

Remarks

Copies only if dtypes does not match T

Cast<T>()

Return a casted UnmanagedStorage to a specific dtype.

public UnmanagedStorage Cast<T>() where T : unmanaged

Returns

UnmanagedStorage

A copy of this UnmanagedStorage casted to a specific dtype.

Type Parameters

T

The dtype to convert to

Remarks

Always copies, If dtype==typeof(T) then a Clone() is returned.

Clone()

Perform a complete copy of this UnmanagedStorage and InternalArray.

public UnmanagedStorage Clone()

Returns

UnmanagedStorage

Remarks

If shape is sliced, discards any slicing properties but copies only the sliced data

CloneData()

Clone internal storage and return an owning IArraySlice sized to _shape.size (NOT InternalArray.Count).

public IArraySlice CloneData()

Returns

IArraySlice

A freshly-allocated IArraySlice whose Count == _shape.size. For contiguous shapes the elements come from InternalArray[_shape.offset.._shape.offset + _shape.size) via Slice(long, long) + Clone; for strided / broadcast / transposed shapes they are materialised via Copy(NDArray, NDArray).

Remarks

Subtle: the C-contig branch must always slice to _shape.size when _shape.bufferSize > _shape.size, even when offset == 0. A 1-D slice like arr[0:4] on a 5-element source has offset 0 yet covers only the first 4 elements; a previous version unconditionally cloned the entire InternalArray in the offset == 0 branch, then handed the 5-element clone to UnmanagedStorage(IArraySlice, Shape) paired with a (4,) shape, tripping IncorrectShapeException.

CloneData<T>()

Get all elements from cloned storage as ArraySlice<T> and cast if necessary.

public ArraySlice<T> CloneData<T>() where T : unmanaged

Returns

ArraySlice<T>

reference to cloned storage and casted (if necessary) as ArraySlice<T>

Type Parameters

T

cloned storgae dtype

CopyTo(IMemoryBlock)

Copies the entire contents of this storage to given address (using Count).

public void CopyTo(IMemoryBlock block)

Parameters

block IMemoryBlock

The block to copy to.

CopyTo(nint)

Copies the entire contents of this storage to given address.

public void CopyTo(nint ptr)

Parameters

ptr nint

CopyTo(void*)

Copies the entire contents of this storage to given address.

public void CopyTo(void* address)

Parameters

address void*

The address to copy to; it must hold Count * DTypeSize bytes.

Remarks

A fixed-width string storage copies its raw element bytes (DTypeSize each: NUL-padded ASCII for S, UCS-4 code units for U) in C order — the bytes tobytes() returns.

Exceptions

ArgumentNullException

address is null.

CopyTo<T>(IMemoryBlock<T>)

Copies the entire contents of this storage to given address (using Count).

public void CopyTo<T>(IMemoryBlock<T> block) where T : unmanaged

Parameters

block IMemoryBlock<T>

The block to copy to.

Type Parameters

T

CopyTo<T>(T*)

Copies the entire contents of this storage to given address.

public void CopyTo<T>(T* address) where T : unmanaged

Parameters

address T*

The address to copy to.

Type Parameters

T

CopyTo<T>(T[])

Copies the entire contents of this storage to given array.

public void CopyTo<T>(T[] array) where T : unmanaged

Parameters

array T[]

The array to copy to.

Type Parameters

T

CreateBroadcastedUnsafe(IArraySlice, Shape)

Creates a new storage with a broadcasted shape from an array slice.

public static UnmanagedStorage CreateBroadcastedUnsafe(IArraySlice arraySlice, Shape shape)

Parameters

arraySlice IArraySlice

The array slice to wrap.

shape Shape

The broadcasted shape to represent this storage.

Returns

UnmanagedStorage

A new UnmanagedStorage that owns the data (not a view). The returned storage's NumSharp.Backends.UnmanagedStorage._baseStorage is null.

Remarks

Named "Unsafe": This method does not validate that the shape is compatible with the array slice size.

Ownership: This overload creates owned storage (not a view) because it receives raw data without storage context. Compare with the CreateBroadcastedUnsafe(UnmanagedStorage, Shape) overload which preserves base tracking.

CreateBroadcastedUnsafe(UnmanagedStorage, Shape)

Creates a broadcasted view of an existing storage with a new shape.

public static UnmanagedStorage CreateBroadcastedUnsafe(UnmanagedStorage storage, Shape shape)

Parameters

storage UnmanagedStorage

The source storage to take InternalArray from.

shape Shape

The broadcasted shape to represent this storage.

Returns

UnmanagedStorage

A new UnmanagedStorage that shares memory with the source storage. The returned storage's NumSharp.Backends.UnmanagedStorage._baseStorage points to the ultimate owner.

Remarks

Named "Unsafe": This method does not validate that the shape is compatible with the storage size.

Base Tracking: Sets _baseStorage to chain to the ultimate owner:

  • If source storage owns its data: result._baseStorage = storage
  • If source storage is a view: result._baseStorage = storage._baseStorage

Used By: np.broadcast_to(), np.broadcast_arrays(), and internal broadcasting operations.

See Also

ExpandDimension(int)

protected void ExpandDimension(int axis)

Parameters

axis int

GetAtIndex(long)

Reads the element at logical flat index index (C order) boxed as the dtype's own CLR type, after NumPy's flat-index check (a.item(i), a.flat[i]).

public object GetAtIndex(long index)

Parameters

index long

The flat index; a negative one counts from the end of the array.

Returns

object

The element, boxed.

Exceptions

IndexError

index is outside the array (NumPy's item texts).

NotSupportedException

The storage has no element type.

GetAtIndex<T>(long)

Reads the element at logical flat index index (C order) as T: the stored value for the dtype's own type, NumPy's cast of it for any other T.

public T GetAtIndex<T>(long index) where T : unmanaged

Parameters

index long

The flat index; a negative one counts from the end of the array.

Returns

T

The element as T.

Type Parameters

T

The type to read the element as.

Remarks

Before parity QA-12 / QA-33 this loaded sizeof(T) bytes at an unchecked offset (a Debug.Assert was the only type test), so a wrong T or index read neighbouring elements or memory past the buffer in Release.

Exceptions

IndexError

index is outside the array (NumPy's item texts).

NotSupportedException

The storage has no element type.

GetBoolean(int[])

Reads the element at indices as a bool: the stored value itself in a bool array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public bool GetBoolean(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

bool

The element as a bool.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetBoolean(params long[])

Reads the element at indices as a bool: the stored value itself in a bool array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public bool GetBoolean(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

bool

The element as a bool.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetByte(int[])

Reads the element at indices as a byte: the stored value itself in a uint8 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public byte GetByte(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

byte

The element as a byte.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetByte(params long[])

Reads the element at indices as a byte: the stored value itself in a uint8 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public byte GetByte(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

byte

The element as a byte.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetChar(int[])

Reads the element at indices as a char: the stored value itself in a char array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public char GetChar(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

char

The element as a char.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetChar(params long[])

Reads the element at indices as a char: the stored value itself in a char array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public char GetChar(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

char

The element as a char.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetComplex(int[])

Reads the element at indices as a Complex: the stored value itself in a complex128 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public Complex GetComplex(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

Complex

The element as a Complex.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetComplex(params long[])

Reads the element at indices as a Complex: the stored value itself in a complex128 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public Complex GetComplex(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

Complex

The element as a Complex.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetData()

Get reference to internal data storage

public IArraySlice GetData()

Returns

IArraySlice

reference to internal storage as System.Array

GetData(int*, int)

Gets a sub-array based on the given indices (pointer version), returning a view that shares memory.

public UnmanagedStorage GetData(int* dims, int ndims)

Parameters

dims int*

Pointer to an array of dimension indices.

ndims int

The number of indices in the array.

Returns

UnmanagedStorage

A new UnmanagedStorage representing the sub-array. This is a view that shares memory with the original storage. The returned storage's NumSharp.Backends.UnmanagedStorage._baseStorage points to the ultimate owner.

Remarks

Memory Sharing: The returned storage shares memory with this storage. Modifications through either storage affect the same underlying data.

Base Tracking: Sets _baseStorage to chain to the ultimate owner, ensuring all views in a chain point to the original storage, not intermediate views.

Code Paths:

  • Broadcasted shapes: Creates a broadcasted view with base tracking
  • Non-contiguous shapes: Delegates to GetView(params Slice[]) which uses Alias(Shape)
  • Contiguous shapes: Creates a direct memory slice with base tracking
See Also

GetData(int[])

Gets a sub-array based on the given indices, returning a view that shares memory.

public UnmanagedStorage GetData(int[] indices)

Parameters

indices int[]

The indices specifying which dimensions to select. Negative indices are supported and are converted to positive indices relative to the dimension size.

Returns

UnmanagedStorage

A new UnmanagedStorage representing the sub-array. This is a view that shares memory with the original storage. The returned storage's NumSharp.Backends.UnmanagedStorage._baseStorage points to the ultimate owner.

Remarks

Memory Sharing: The returned storage shares memory with this storage. Modifications through either storage affect the same underlying data.

Base Tracking: Sets _baseStorage to chain to the ultimate owner, ensuring all views in a chain point to the original storage, not intermediate views.

Code Paths:

  • Broadcasted shapes: Creates a broadcasted view with base tracking
  • Non-contiguous shapes: Delegates to GetView(params Slice[]) which uses Alias(Shape)
  • Contiguous shapes: Creates a direct memory slice with base tracking
See Also

GetData(long*, int)

Gets a sub-array based on the given indices (long pointer version), returning a view that shares memory.

public UnmanagedStorage GetData(long* dims, int ndims)

Parameters

dims long*

Pointer to an array of long dimension indices.

ndims int

The number of indices in the array.

Returns

UnmanagedStorage

A new UnmanagedStorage representing the sub-array.

See Also

GetData(params long[])

Gets a sub-array based on the given indices (long array version), returning a view that shares memory.

public UnmanagedStorage GetData(params long[] indices)

Parameters

indices long[]

The dimension indices specifying the sub-array.

Returns

UnmanagedStorage

A new UnmanagedStorage representing the sub-array. This is a view that shares memory with the original storage. The returned storage's NumSharp.Backends.UnmanagedStorage._baseStorage points to the ultimate owner.

See Also

GetData<T>()

Get reference to internal data storage and cast (also copies) elements to new dtype if necessary

public ArraySlice<T> GetData<T>() where T : unmanaged

Returns

ArraySlice<T>

reference to internal (casted) storage as T[]

Type Parameters

T

new storage data type

Remarks

Copies if T does not equal to DType or if Shape is sliced.

GetDecimal(int[])

Reads the element at indices as a decimal: the stored value itself in a decimal array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public decimal GetDecimal(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

decimal

The element as a decimal.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetDecimal(params long[])

Reads the element at indices as a decimal: the stored value itself in a decimal array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public decimal GetDecimal(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

decimal

The element as a decimal.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetDouble(int[])

Reads the element at indices as a double: the stored value itself in a float64 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public double GetDouble(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

double

The element as a double.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetDouble(params long[])

Reads the element at indices as a double: the stored value itself in a float64 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public double GetDouble(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

double

The element as a double.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetFieldAlias(NPTypeCode, int)

Dtype-code dispatcher for GetFieldAlias<T>(int) — mirrors AliasAs(NPTypeCode).

public UnmanagedStorage GetFieldAlias(NPTypeCode typeCode, int offset)

Parameters

typeCode NPTypeCode

The field dtype to read.

offset int

Byte offset of the field within each element.

Returns

UnmanagedStorage

GetFieldAlias<T>(int)

Builds a byte-reinterpreting VIEW of one field of each element — the storage half of NumPy's ndarray.getfield(dtype, offset) (numpy/_core/src/multiarray/methods.c, PyArray_GetField). Unlike AliasAs<T>() (view(dtype), which rescales the last axis), this keeps EVERY byte-stride and dimension verbatim and simply reads T out of the offset-th byte of each element.

public UnmanagedStorage GetFieldAlias<T>(int offset) where T : unmanaged

Parameters

offset int

Byte offset of the field within each element; [0, oldItemsize - newItemsize].

Returns

UnmanagedStorage

A new UnmanagedStorage of dtype T that SHARES memory with this storage (writes through, unless the source is read-only). The base is rooted through _baseStorage so the shared buffer outlives the view.

Type Parameters

T

The field dtype to read; its itemsize must be ≤ this storage's itemsize.

Remarks

Byte-strides are preserved. NumPy carries strides in bytes; the getfield view keeps them unchanged, so a C-contiguous int32[4] field-viewed as int16 stays strided (byte-stride 4, itemsize 2 → non-contiguous), reading the low or high half of each int32. NumSharp stores strides in ELEMENTS, so each is recomputed as oldByteStride / newItemsize — always exact, because every NumSharp itemsize (1/2/4/8/16) divides every larger one, so an old byte-stride (a whole multiple of the old itemsize) is always a whole multiple of the new.

Sub-element byte offset. When offset is not a multiple of the new itemsize (e.g. int32.getfield(int16, 1)), the logical element start falls mid-element, which the element-granular offset cannot express. The remainder r = offset % newItemsize is absorbed into the WRAP POINTER (shifted by r bytes) while offset carries the aligned part — so EVERY offset in range is representable, matching NumPy's raw byte pointer.

Contiguity flags are recomputed from the resulting strides (NumPy does the same — a same-size field of a C-contiguous array stays C-contiguous, a narrower one does not). A read-only source (broadcast view, 'r' memmap) yields a read-only field view.

Exceptions

ValueError

new type is larger than original type (newItemsize > oldItemsize), offset is negative, or new type plus offset is larger than original type (offset > oldItemsize − newItemsize) — the three verbatim NumPy ValueError texts, in NumPy's check order.

GetHalf(int[])

Reads the element at indices as a Half: the stored value itself in a float16 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public Half GetHalf(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

Half

The element as a Half.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetHalf(params long[])

Reads the element at indices as a Half: the stored value itself in a float16 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public Half GetHalf(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

Half

The element as a Half.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetInt16(int[])

Reads the element at indices as a short: the stored value itself in a int16 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public short GetInt16(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

short

The element as a short.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetInt16(params long[])

Reads the element at indices as a short: the stored value itself in a int16 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public short GetInt16(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

short

The element as a short.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetInt32(int[])

Reads the element at indices as a int: the stored value itself in a int32 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public int GetInt32(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

int

The element as a int.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetInt32(params long[])

Reads the element at indices as a int: the stored value itself in a int32 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public int GetInt32(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

int

The element as a int.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetInt64(int[])

Reads the element at indices as a long: the stored value itself in a int64 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public long GetInt64(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

long

The element as a long.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetInt64(params long[])

Reads the element at indices as a long: the stored value itself in a int64 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public long GetInt64(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

long

The element as a long.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetSByte(int[])

Reads the element at indices as a sbyte: the stored value itself in a int8 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public sbyte GetSByte(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

sbyte

The element as a sbyte.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetSByte(params long[])

Reads the element at indices as a sbyte: the stored value itself in a int8 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public sbyte GetSByte(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

sbyte

The element as a sbyte.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetSingle(int[])

Reads the element at indices as a float: the stored value itself in a float32 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public float GetSingle(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

float

The element as a float.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetSingle(params long[])

Reads the element at indices as a float: the stored value itself in a float32 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public float GetSingle(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

float

The element as a float.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetUInt16(int[])

Reads the element at indices as a ushort: the stored value itself in a uint16 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public ushort GetUInt16(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

ushort

The element as a ushort.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetUInt16(params long[])

Reads the element at indices as a ushort: the stored value itself in a uint16 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public ushort GetUInt16(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

ushort

The element as a ushort.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetUInt32(int[])

Reads the element at indices as a uint: the stored value itself in a uint32 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public uint GetUInt32(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

uint

The element as a uint.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetUInt32(params long[])

Reads the element at indices as a uint: the stored value itself in a uint32 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public uint GetUInt32(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

uint

The element as a uint.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetUInt64(int[])

Reads the element at indices as a ulong: the stored value itself in a uint64 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public ulong GetUInt64(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

ulong

The element as a ulong.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetUInt64(params long[])

Reads the element at indices as a ulong: the stored value itself in a uint64 array, NumPy's cast of it (the cast astype runs) in an array of any other dtype.

public ulong GetUInt64(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

ulong

The element as a ulong.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

GetValue(int[])

Reads the element at indices boxed as the dtype's own CLR type (a text element as its decoded C# value), after NumPy's coordinate check — the read behind item(i, j, …).

public object GetValue(int[] indices)

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

object

The element, boxed.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

NotSupportedException

The storage has no element type.

GetValue(params long[])

Reads the element at indices boxed as the dtype's own CLR type (a text element as its decoded C# value), after NumPy's coordinate check — the read behind item(i, j, …).

public object GetValue(params long[] indices)

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

object

The element, boxed.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

NotSupportedException

The storage has no element type.

GetValue<T>(int[])

Reads the element at indices as T: the stored value for the dtype's own type, NumPy's cast of it (the cast astype runs) for any other T.

public T GetValue<T>(int[] indices) where T : unmanaged

Parameters

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

T

The element as T.

Type Parameters

T

The type to read the element as.

Remarks

Before parity QA-12 / QA-33 this loaded sizeof(T) bytes at an unchecked offset: GetValue<double>(1) of a float32 array read elements 1 and 2 as one double, and a wrong coordinate read another row or memory outside the array in Release.

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

NotSupportedException

The storage has no element type.

GetValue<T>(params long[])

Reads the element at indices as T: the stored value for the dtype's own type, NumPy's cast of it (the cast astype runs) for any other T.

public T GetValue<T>(params long[] indices) where T : unmanaged

Parameters

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out read their first element.

Returns

T

The element as T.

Type Parameters

T

The type to read the element as.

Remarks

Exceptions

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is read).

NotSupportedException

The storage has no element type.

GetView(params Slice[])

public UnmanagedStorage GetView(params Slice[] slices)

Parameters

slices Slice[]

Returns

UnmanagedStorage

GetView(string)

public UnmanagedStorage GetView(string slicing_notation)

Parameters

slicing_notation string

Returns

UnmanagedStorage

OnReshaped()

Post-transition hook — run after this storage's _shape is (re)assigned or its ownership links (NumSharp.Backends.UnmanagedStorage._baseStorage / ExternalBase) change. Reconciles the shape's OWNDATA bit with OwnsData — the NumSharp analog of NumPy's PyArray_NewFromDescr_int (ctors.c: fa->flags |= NPY_ARRAY_OWNDATA when it allocates, fa->flags &= ~NPY_ARRAY_OWNDATA when data is passed in) plus the invariant NumPy asserts in _IsWriteable (common.c): base != NULL ⟹ !OWNDATA.

protected void OnReshaped()

Remarks

Must be called wherever ownership or _shape transitions: at the end of every allocating constructor / _Allocate(Shape, IArraySlice), after every NumSharp.Backends.UnmanagedStorage._baseStorage assignment (views borrow the parent's shape, which carries the parent's bit), after ExternalBase is raised, and in SetShapeUnsafe(ref Shape) / the ReplaceData family (freshly built shapes never carry the bit, which would silently strip it from an owner).

ReplaceData(IArraySlice)

Sets values as the internal data source and changes the internal storage data type to values type.

public void ReplaceData(IArraySlice values)

Parameters

values IArraySlice

Remarks

Does not copy values and doesn't change shape.

ReplaceData(IArraySlice, Shape)

Sets values as the internal data source and changes the internal storage data type to values type.

public void ReplaceData(IArraySlice values, Shape shape)

Parameters

values IArraySlice
shape Shape

The shape to set in this storage. (without checking if shape matches storage)

Remarks

Does not copy values and doesn't change shape. Doesn't check if shape size matches.

ReplaceData(IArraySlice, Type)

Sets values as the internal data source and changes the internal storage data type to values type.

public void ReplaceData(IArraySlice values, Type dtype)

Parameters

values IArraySlice
dtype Type

Remarks

Does not copy values and doesn't change shape.

ReplaceData(IArraySlice, Type, Shape)

Sets values as the internal data source and changes the internal storage data type to values type.

public void ReplaceData(IArraySlice values, Type dtype, Shape shape)

Parameters

values IArraySlice
dtype Type
shape Shape

The shape to set in this storage. (without checking if shape matches storage)

Remarks

Does not copy values and doesn't change shape. Doesn't check if shape size matches.

ReplaceData(NDArray)

Sets nd as the internal data storage and changes the internal storage data type to nd type.

public void ReplaceData(NDArray nd)

Parameters

nd NDArray

Remarks

Does not copy values and does change shape and dtype.

ReplaceData(NDArray, Shape)

Sets nd as the internal data storage and changes the internal storage data type to nd type.

public void ReplaceData(NDArray nd, Shape shape)

Parameters

nd NDArray
shape Shape

The shape to set in this storage. (without checking if shape matches storage)

Remarks

Does not copy values and does change shape and dtype. Doesn't check if shape size matches.

ReplaceData(Array)

Sets values as the internal data source and changes the internal storage data type to values type.

public void ReplaceData(Array values)

Parameters

values Array

Remarks

Copies values only if values type does not match DType and doesn't change shape.

ReplaceData(Array, NPTypeCode)

Set an Array to internal storage, cast it to new dtype and if necessary change dtype

public void ReplaceData(Array values, NPTypeCode typeCode)

Parameters

values Array
typeCode NPTypeCode

Remarks

Does not copy values unless cast is necessary and doesn't change shape.

ReplaceData(Array, NPTypeCode, Shape)

Set an Array to internal storage, cast it to new dtype and if necessary change dtype

public void ReplaceData(Array values, NPTypeCode typeCode, Shape shape)

Parameters

values Array
typeCode NPTypeCode
shape Shape

The shape to set in this storage. (without checking if shape matches storage)

Remarks

Does not copy values unless cast is necessary and doesn't change shape. Doesn't check if shape size matches.

ReplaceData(Array, Shape)

Sets values as the internal data source and changes the internal storage data type to values type.

public void ReplaceData(Array values, Shape shape)

Parameters

values Array
shape Shape

The shape to set in this storage. (without checking if shape matches storage)

Remarks

Copies values only if values type does not match DType and doesn't change shape. Doesn't check if shape size matches.

ReplaceData(Array, Type)

Set an Array to internal storage, cast it to new dtype and change dtype

public void ReplaceData(Array values, Type dtype)

Parameters

values Array
dtype Type

Remarks

Does not copy values unless cast in necessary and doesn't change shape.

ReplaceData(Array, Type, Shape)

Set an Array to internal storage, cast it to new dtype and change dtype

public void ReplaceData(Array values, Type dtype, Shape shape)

Parameters

values Array
dtype Type
shape Shape

The shape to set in this storage. (without checking if shape matches storage)

Remarks

Does not copy values unless cast in necessary and doesn't change shape. Doesn't check if shape size matches.

Reshape(Shape, bool)

Changes the shape representing this storage.

public void Reshape(Shape newShape, bool @unsafe = false)

Parameters

newShape Shape
unsafe bool

Exceptions

IncorrectShapeException

If shape's size mismatches current shape size.

ArgumentException

If newShape's size == 0

Reshape(ref Shape, bool)

Changes the shape representing this storage.

public void Reshape(ref Shape newShape, bool @unsafe = false)

Parameters

newShape Shape
unsafe bool

Exceptions

IncorrectShapeException

If shape's size mismatches current shape size.

ArgumentException

If newShape's size == 0

Reshape(params long[])

Changes the shape representing this storage.

public void Reshape(params long[] dimensions)

Parameters

dimensions long[]

Exceptions

IncorrectShapeException

If shape's size mismatches current shape size.

Reshape(long[], bool)

Changes the shape representing this storage.

public void Reshape(long[] dimensions, bool @unsafe)

Parameters

dimensions long[]
unsafe bool

Exceptions

IncorrectShapeException

If shape's size mismatches current shape size.

ArgumentException

If dimensions's size == 0

Scalar(object)

public static UnmanagedStorage Scalar(object value)

Parameters

value object

Returns

UnmanagedStorage

Scalar(object, NPTypeCode)

public static UnmanagedStorage Scalar(object value, NPTypeCode typeCode)

Parameters

value object
typeCode NPTypeCode

Returns

UnmanagedStorage

Scalar<T>(T)

public static UnmanagedStorage Scalar<T>(T value) where T : unmanaged

Parameters

value T

Returns

UnmanagedStorage

Type Parameters

T

SetAtIndex(object, long)

Writes one boxed element at a logical flat index (resolved through the shape's strides and offset). The value must be the dtype's exact CLR type — an explicit unboxing, never a conversion — except null, which is Python's None and stores what NumPy's setitem stores for it.

public void SetAtIndex(object value, long index)

Parameters

value object

The element, boxed as this storage's exact type; null is None (NaN for a float dtype, nan+nanj for complex, False for bool, 'None' or a StringDType NA object for text).

index long

The logical flat index (C order); a negative one counts from the end of the array.

Exceptions

ValueError

The storage is read-only (assignment destination is read-only), or None into a StringDType(coerce=False) storage.

TypeError

None into an integer or decimal storage (Python's int(None)).

InvalidCastException

A non-null value that is not the dtype's exact CLR type.

IndexError

index is outside the array (NumPy's item texts; nothing is written).

SetAtIndexUnsafe(object, long)

Performs a set of index without calling TransformOffset(long).

public void SetAtIndexUnsafe(object value, long index)

Parameters

value object
index long

SetAtIndexUnsafe<T>(T, long)

Performs a set of index without calling TransformOffset(long).

public void SetAtIndexUnsafe<T>(T value, long index) where T : unmanaged

Parameters

value T
index long

Type Parameters

T

SetAtIndex<T>(T, long)

Stores value at logical flat index index (C order): as it is for the dtype's own type, cast to the dtype with NumPy's cast for any other T.

public void SetAtIndex<T>(T value, long index) where T : unmanaged

Parameters

value T

The value to store.

index long

The flat index; a negative one counts from the end of the array.

Type Parameters

T

The value's type.

Remarks

Before parity QA-12 / QA-33 this stored sizeof(T) raw bytes at an unchecked offset: a double into a float32 array overwrote two elements, and its last element four bytes past the buffer, in Release.

Exceptions

ValueError

The storage is read-only (assignment destination is read-only).

IndexError

index is outside the array (NumPy's item texts; nothing is written).

NotSupportedException

The storage has no element type.

SetBoolean(bool, int[])

Stores value at indices: as it is in a bool array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a bool scalar).

public void SetBoolean(bool value, int[] indices)

Parameters

value bool

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetBoolean(bool, params long[])

Stores value at indices: as it is in a bool array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a bool scalar).

public void SetBoolean(bool value, params long[] indices)

Parameters

value bool

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetByte(byte, int[])

Stores value at indices: as it is in a uint8 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a uint8 scalar).

public void SetByte(byte value, int[] indices)

Parameters

value byte

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetByte(byte, params long[])

Stores value at indices: as it is in a uint8 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a uint8 scalar).

public void SetByte(byte value, params long[] indices)

Parameters

value byte

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetChar(char, int[])

Stores value at indices: as it is in a char array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a char scalar).

public void SetChar(char value, int[] indices)

Parameters

value char

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetChar(char, params long[])

Stores value at indices: as it is in a char array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a char scalar).

public void SetChar(char value, params long[] indices)

Parameters

value char

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetComplex(Complex, int[])

Stores value at indices: as it is in a complex128 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a complex128 scalar).

public void SetComplex(Complex value, int[] indices)

Parameters

value Complex

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetComplex(Complex, params long[])

Stores value at indices: as it is in a complex128 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a complex128 scalar).

public void SetComplex(Complex value, params long[] indices)

Parameters

value Complex

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetData(IArraySlice, int[])

Set a IArraySlice at given indices.

public void SetData(IArraySlice value, int[] indices)

Parameters

value IArraySlice

The value to set

indices int[]

The

Remarks

Does not change internal storage data type.
If value does not match DType, value will be converted.

SetData(IArraySlice, params long[])

Set a IArraySlice at given indices (long version).

public void SetData(IArraySlice value, params long[] indices)

Parameters

value IArraySlice
indices long[]

SetData(NDArray, int[])

Set a NDArray at given indices.

public void SetData(NDArray value, int[] indices)

Parameters

value NDArray

The value to set

indices int[]

The

Remarks

Does not change internal storage data type.
If value does not match DType, value will be converted: an array is CAST (NumPy's unsafe assignment cast), a C# literal (NumSharp.NDArray.WasPythonLiteral) is PACKED as the Python scalar it stands for (PackForStore(object)).

Exceptions

OverflowError

A C# literal out of this dtype's range (NumPy's OverflowError texts).

ValueError

A float NaN literal into an integer dtype, a sequence into one element, or a value that does not broadcast to the selection.

TypeError

A complex literal into a real dtype; None (a null value) into an integer or decimal storage — Python's int(None) text, raised after the coordinates' bounds check (a[10] = None on a 4-element array is NumPy's IndexError) and for an empty selection too. A null is Python's None everywhere: a float storage stores NaN (NumPy's positive NPY_NAN), complex nan+nanj, bool False, a NEP 55 StringDType storage its NA object or 'None', and a fixed-width text storage 'None', truncated to its width.

ValueError

None into a StringDType(coerce=False) storage.

SetData(NDArray, params long[])

Set a NDArray at given indices (long version).

public void SetData(NDArray value, params long[] indices)

Parameters

value NDArray

The value; null is Python's None: a float storage stores NaN, complex nan+nanj, bool False, a NEP 55 StringDType storage its NA object or 'None', a fixed-width one 'None', truncated; an integer storage raises int(None)'s TypeError.

indices long[]

The leading coordinates.

Exceptions

OverflowException

A coordinate outside the int range.

Exception

The SetData(NDArray, int[]) overload's errors.

SetData(object, int[])

Sets the element or sub-array at indices from value: an array-like value goes through SetData(NDArray, int[]), a boxed scalar is stored at the element the coordinates address.

public void SetData(object value, int[] indices)

Parameters

value object

An NDArray, an IArraySlice, a C# array, a boxed scalar of the dtype's exact CLR type, or null — Python's None.

indices int[]

The leading coordinates; a negative coordinate counts from the end of its axis.

Remarks

Does not change the storage's dtype. A boxed scalar is unboxed as the dtype's exact type, never converted (see SetValue(object, int[])).

Exceptions

ValueError

The storage is read-only, or None into a StringDType(coerce=False) storage.

TypeError

None (a null value) into an integer or decimal storage, raised after the coordinates' bounds check (SetData(NDArray, int[])).

InvalidCastException

A boxed scalar that is not the dtype's exact CLR type.

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing is written).

SetData(object, params long[])

Set a value at given indices (long version).

public void SetData(object value, params long[] indices)

Parameters

value object

An NDArray, an IArraySlice, a C# array, a boxed scalar, or null — Python's None (see SetData(NDArray, int[])).

indices long[]

The leading coordinates.

Exceptions

ValueError

The storage is read-only, or None into a StringDType(coerce=False) storage.

TypeError

None into an integer or decimal storage.

InvalidCastException

A boxed scalar that is not the dtype's exact CLR type.

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing is written).

SetDecimal(decimal, int[])

Stores value at indices: as it is in a decimal array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a decimal scalar).

public void SetDecimal(decimal value, int[] indices)

Parameters

value decimal

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetDecimal(decimal, params long[])

Stores value at indices: as it is in a decimal array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a decimal scalar).

public void SetDecimal(decimal value, params long[] indices)

Parameters

value decimal

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetDouble(double, int[])

Stores value at indices: as it is in a float64 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a float64 scalar).

public void SetDouble(double value, int[] indices)

Parameters

value double

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetDouble(double, params long[])

Stores value at indices: as it is in a float64 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a float64 scalar).

public void SetDouble(double value, params long[] indices)

Parameters

value double

The value to store.

indices long[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetHalf(Half, int[])

Stores value at indices: as it is in a float16 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a float16 scalar).

public void SetHalf(Half value, int[] indices)

Parameters

value Half

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetHalf(Half, params long[])

Stores value at indices: as it is in a float16 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a float16 scalar).

public void SetHalf(Half value, params long[] indices)

Parameters

value Half

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetInt16(short, int[])

Stores value at indices: as it is in a int16 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a int16 scalar).

public void SetInt16(short value, int[] indices)

Parameters

value short

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetInt16(short, params long[])

Stores value at indices: as it is in a int16 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a int16 scalar).

public void SetInt16(short value, params long[] indices)

Parameters

value short

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetInt32(int, int[])

Stores value at indices: as it is in a int32 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a int32 scalar).

public void SetInt32(int value, int[] indices)

Parameters

value int

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetInt32(int, params long[])

Stores value at indices: as it is in a int32 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a int32 scalar).

public void SetInt32(int value, params long[] indices)

Parameters

value int

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetInt64(long, int[])

Stores value at indices: as it is in a int64 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a int64 scalar).

public void SetInt64(long value, int[] indices)

Parameters

value long

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetInt64(long, params long[])

Stores value at indices: as it is in a int64 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a int64 scalar).

public void SetInt64(long value, params long[] indices)

Parameters

value long

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetInternalArray(IArraySlice)

Replace internal storage array with given array.

protected void SetInternalArray(IArraySlice array)

Parameters

array IArraySlice

The array to set as internal storage

Exceptions

InvalidCastException

When type of array does not match DType of this storage

SetInternalArray(Array)

Replace internal storage array with given array.

protected void SetInternalArray(Array array)

Parameters

array Array

The array to set as internal storage

Exceptions

InvalidCastException

When type of array does not match DType of this storage

SetSByte(sbyte, int[])

Stores value at indices: as it is in a int8 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a int8 scalar).

public void SetSByte(sbyte value, int[] indices)

Parameters

value sbyte

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetSByte(sbyte, params long[])

Stores value at indices: as it is in a int8 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a int8 scalar).

public void SetSByte(sbyte value, params long[] indices)

Parameters

value sbyte

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetShapeUnsafe(Shape)

Set the shape of this storage without checking if sizes match.

protected void SetShapeUnsafe(Shape shape)

Parameters

shape Shape

Remarks

Used during broadcasting

SetShapeUnsafe(ref Shape)

Set the shape of this storage without checking if sizes match.

protected void SetShapeUnsafe(ref Shape shape)

Parameters

shape Shape

Remarks

Used during broadcasting. Uses bufferSize (not size) because broadcast shapes have logical size != physical buffer size.

SetSingle(float, int[])

Stores value at indices: as it is in a float32 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a float32 scalar).

public void SetSingle(float value, int[] indices)

Parameters

value float

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetSingle(float, params long[])

Stores value at indices: as it is in a float32 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a float32 scalar).

public void SetSingle(float value, params long[] indices)

Parameters

value float

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetUInt16(ushort, int[])

Stores value at indices: as it is in a uint16 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a uint16 scalar).

public void SetUInt16(ushort value, int[] indices)

Parameters

value ushort

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetUInt16(ushort, params long[])

Stores value at indices: as it is in a uint16 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a uint16 scalar).

public void SetUInt16(ushort value, params long[] indices)

Parameters

value ushort

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetUInt32(uint, int[])

Stores value at indices: as it is in a uint32 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a uint32 scalar).

public void SetUInt32(uint value, int[] indices)

Parameters

value uint

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetUInt32(uint, params long[])

Stores value at indices: as it is in a uint32 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a uint32 scalar).

public void SetUInt32(uint value, params long[] indices)

Parameters

value uint

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetUInt64(ulong, int[])

Stores value at indices: as it is in a uint64 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a uint64 scalar).

public void SetUInt64(ulong value, int[] indices)

Parameters

value ulong

The value to store.

indices int[]

The element's coordinates; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetUInt64(ulong, params long[])

Stores value at indices: as it is in a uint64 array, cast to the array's dtype with NumPy's cast in an array of any other dtype (NumPy's setitem of a uint64 scalar).

public void SetUInt64(ulong value, params long[] indices)

Parameters

value ulong

The value to store.

indices long[]

The element's coordinates (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Exceptions

ValueError

The array is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing outside the array is written).

SetValue(object, int[])

Stores one boxed element at indices. The value must be the dtype's exact CLR type — an explicit unboxing, never a conversion — except null, which is Python's None and stores what NumPy's setitem stores for it.

public void SetValue(object value, int[] indices)

Parameters

value object

The element, boxed as this storage's exact type; null is None (NaN for a float dtype, nan+nanj for complex, False for bool, 'None' or a StringDType NA object for text).

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Remarks

Does not change the storage's dtype. The generic SetValue<T>(T, int[]) converts a value of another type; this boxed form does not.

Exceptions

ValueError

The storage is read-only, or None into a StringDType(coerce=False) storage.

TypeError

None (a null value) into an integer or decimal storage; a float storage stores NaN, complex nan+nanj, bool False.

InvalidCastException

A non-null value that is not the dtype's exact CLR type.

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing is written).

SetValue(object, params long[])

Stores one boxed element at indices. The value must be the dtype's exact CLR type — an explicit unboxing, never a conversion — except null, which is Python's None and stores what NumPy's setitem stores for it.

public void SetValue(object value, params long[] indices)

Parameters

value object

The element, boxed as this storage's exact type; null is None (NaN for a float dtype, nan+nanj for complex, False for bool, 'None' or a StringDType NA object for text).

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Remarks

Does not change the storage's dtype. The generic SetValue<T>(T, params long[]) converts a value of another type; this boxed form does not.

Exceptions

ValueError

The storage is read-only, or None into a StringDType(coerce=False) storage.

TypeError

None (a null value) into an integer or decimal storage; a float storage stores NaN, complex nan+nanj, bool False.

InvalidCastException

A non-null value that is not the dtype's exact CLR type.

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing is written).

SetValue<T>(T, int[])

Stores value at indices: as it is for the dtype's own type, cast to the dtype with NumPy's cast (the cast astype runs) for any other T — NumPy's setitem of a scalar of another type.

public void SetValue<T>(T value, int[] indices) where T : unmanaged

Parameters

value T

The value to store.

indices int[]

One coordinate per leading axis; a negative coordinate counts from the end of its axis, and axes left out address their first element.

Type Parameters

T

The value's type.

Remarks

Does not change the storage's dtype. Before parity QA-12 / QA-33 this stored sizeof(T) raw bytes at an unchecked offset (a Debug.Assert was the only type test): SetValue(2.5, 0) on a float32 array gave [0, 2.0625, …], the last element wrote past the buffer, and a wrong coordinate wrote into another row or outside the array, in Release.

Exceptions

ValueError

The storage is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing is written).

NotSupportedException

The storage has no element type.

SetValue<T>(T, params long[])

Stores value at indices: as it is for the dtype's own type, cast to the dtype with NumPy's cast (the cast astype runs) for any other T — NumPy's setitem of a scalar of another type.

public void SetValue<T>(T value, params long[] indices) where T : unmanaged

Parameters

value T

The value to store.

indices long[]

One coordinate per leading axis (long version); a negative coordinate counts from the end of its axis, and axes left out address their first element.

Type Parameters

T

The value's type.

Remarks

Exceptions

ValueError

The storage is read-only (assignment destination is read-only).

IndexError

A coordinate outside its axis, or more coordinates than dimensions (NumPy's texts; nothing is written).

NotSupportedException

The storage has no element type.

ToArray<T>()

public T[] ToArray<T>() where T : unmanaged

Returns

T[]

Type Parameters

T

_Allocate(Shape, IArraySlice)

protected void _Allocate(Shape shape, IArraySlice values)

Parameters

shape Shape
values IArraySlice

_ChangeTypeOfArray(Array, Type)

Changes the type of sourceArray to to_dtype if necessary.

[SuppressMessage("ReSharper", "AssignNullToNotNullAttribute")]
protected static Array _ChangeTypeOfArray(Array sourceArray, Type to_dtype)

Parameters

sourceArray Array

The array to change his type

to_dtype Type

The type to change to.

Returns

Array

Returns sourceArray or new array with changed type to to_dtype

Remarks

If the return type is equal to source type, this method does not return a copy.

_ChangeTypeOfArray<TOut>(IArraySlice)

Changes the type of sourceArray to to_dtype if necessary.

[SuppressMessage("ReSharper", "AssignNullToNotNullAttribute")]
protected static ArraySlice<TOut> _ChangeTypeOfArray<TOut>(IArraySlice sourceArray) where TOut : unmanaged

Parameters

sourceArray IArraySlice

The array to change his type

Returns

ArraySlice<TOut>

Returns sourceArray or new array with changed type to to_dtype

Type Parameters

TOut

Remarks

If the return type is equal to source type, this method does not return a copy.