Vecxt API
API
Generated with
Vecxt API/vecxt

vecxt

package vecxt

Members list

Packages

package vecxt.dimensionExtender
package vecxt.rangeExtender
package vecxt.reinsurance

Type members

Classlikes

object BooleanArrays

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object Cholesky

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
Cholesky.type
enum ComparisonOp

Attributes

Supertypes
trait Enum
trait Serializable
trait Product
trait Equals
class Object
trait Matchable
class Any
Show all
object Determinant

JVM-specific determinant implementation with SIMD optimizations

JVM-specific determinant implementation with SIMD optimizations

This implementation uses partial pivoting but no scaling, so it’s not suitable for ill-conditioned matrices.

Performance optimizations:

  1. Direct array access for row-major matrices (2-5x faster than matrix element access)
  2. SIMD vectorization for row swaps (3-7x faster, uses vector loads/stores)
  3. SIMD FMA for row elimination inner loop (5-8x faster, dominates O(n³) compute time)

SIMD uses jdk.incubator.vector API with SPECIES_PREFERRED (AVX-512 or AVX2 depending on CPU)

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object DoubleArraysX

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object DoubleMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object Eigenvalues

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object FloatArraysX

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object IntArraysX

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
IntArraysX.type
case class InvalidDimensionException(dim: Int) extends Exception

DimensionExtender (Int | Dimension) accepts any Int, not just the two meaningful values (Rows/0, Cols/1) - so reduceAlongDimension's dim can't be validated at compile time and a caller can genuinely reach this at runtime, e.g. mat.sum(2).

DimensionExtender (Int | Dimension) accepts any Int, not just the two meaningful values (Rows/0, Cols/1) - so reduceAlongDimension's dim can't be validated at compile time and a caller can genuinely reach this at runtime, e.g. mat.sum(2).

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
case class InvalidMatrix(cols: Int, rows: Int, data: Int) extends Exception

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
case class InvalidNDArray(message: String) extends Exception

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
object JsDoubleMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object JsFloatMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object JsNativeBooleanArrays

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object JsNativeDoubleArrays

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object JsNativeFloatArrays

Cross compilation shim

Cross compilation shim

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object JvmDoubleMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object JvmFloatMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object JvmIntMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object JvmNativeDoubleMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object LU

LU decomposition with partial pivoting using LAPACK.

LU decomposition with partial pivoting using LAPACK.

Decomposes a matrix A into the product PA = LU where:

  • P is a permutation matrix
  • L is a lower triangular matrix with unit diagonal
  • U is an upper triangular matrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
LU.type
object LongArrays

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
LongArrays.type
case class MatrixDimensionMismatch(aCols: Int, aRows: Int, bCols: Int, bRows: Int) extends Exception

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
case class MatrixEmptyException() extends Exception

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
object MatrixHelper

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object MatrixInstance

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
case class MatrixNotSquareException(rows: Int, cols: Int) extends Exception

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
case class MatrixNotSymmetricException(rows: Int, cols: Int, i: Int, j: Int, valueIJ: Double, valueJI: Double) extends Exception

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
final class MeanAndVariance(val mean: Double, val variance: Double)

The result of a mean-and-variance pass over Array[Double] or Array[Int].

The result of a mean-and-variance pass over Array[Double] or Array[Int].

Replaces the named tuple (mean: Double, variance: Double), and the reason is measured rather than assumed (vecxt/issues/105):

  • A named tuple erases to scala.Tuple2, so reading one field costs an unbox. Check C3 measured variance(mode) — whose whole body is meanAndVariance(mode).variance — at 37 bytes against a 35-byte MaxInlineSize, roughly 30 of them being the destructuring.
  • Check D1 measured 59.33 bytes/op allocated when a caller reads the pair. Notably it measured zero for variance(mode), where the pair is dead and escape analysis scalarizes it — so the allocation is real only for callers who actually want both numbers, which is every caller of meanAndVariance.

Two primitive double fields is one flat object with nothing nested for escape analysis to chase, and a field read is an invokevirtual on a final class rather than an unbox.

final class, not case class, for the reason recorded on vecxt.matrix.Layout: case class synthesises productElement: Int => Object, which boxes.

No unapply, deliberately. val (m, v) = arr.meanAndVariance was the old idiom and is now .mean / .variance; val (m, v) = x is a Tuple2 pattern and cannot match a non-tuple whatever the companion offers, so restoring destructuring would mean case MeanAndVariance(m, v) => at every site — new API surface to buy back a spelling.

Attributes

Supertypes
class Object
trait Matchable
class Any
final class MeanAndVarianceF(val mean: Float, val variance: Float)

The Array[Float] counterpart of MeanAndVariance, carrying two primitive float fields.

The Array[Float] counterpart of MeanAndVariance, carrying two primitive float fields.

A separate class rather than one generic in the element type: MeanAndVariance[A] would erase A to Object and box both fields, which is the cost being removed. The float kernels accumulate in Double and narrow once at the end, so this describes the result, not the arithmetic.

Attributes

Supertypes
class Object
trait Matchable
class Any
object NDArrayBooleanIndexing

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NDArrayBooleanOps

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NDArrayDoubleOps

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NDArrayFloatOps

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NDArrayFloatReductions

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NDArrayIntOps

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NDArrayIntReductions

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NDArrayReductions

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NDArrayWhere

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NativeDoubleMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NativeFloatMatrix

Cross-compilation stubs — the real implementations live in src-js/floatmatrix.scala and src-native/floatmatrix.scala. all exports both names on every platform, so both have to exist here too. Same arrangement JvmDoubleMatrix/JsDoubleMatrix/NativeDoubleMatrix already use. NativeFloatMatrix was already declared empty here before either had an implementation.

Cross-compilation stubs — the real implementations live in src-js/floatmatrix.scala and src-native/floatmatrix.scala. all exports both names on every platform, so both have to exist here too. Same arrangement JvmDoubleMatrix/JsDoubleMatrix/NativeDoubleMatrix already use. NativeFloatMatrix was already declared empty here before either had an implementation.

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object NativeIntMatrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
trait OneAndZero[A]

Attributes

Companion
object
Supertypes
class Object
trait Matchable
class Any
Known subtypes
object OneAndZero

Attributes

Companion
trait
Supertypes
class Object
trait Matchable
class Any
Self type
OneAndZero.type
object QR

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
QR.type
object Solve

Linear system solver using LAPACK.

Linear system solver using LAPACK.

Provides methods to solve systems of linear equations Ax = b using LAPACK's dgesv routine.

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
Solve.type
object Svd

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
Svd.type
case class UnsupportedLayoutException(message: String) extends Exception

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
enum VarianceMode

Attributes

Companion
object
Supertypes
trait Enum
trait Serializable
trait Product
trait Equals
class Object
trait Matchable
class Any
Show all
object VarianceMode

Attributes

Companion
enum
Supertypes
trait Sum
trait Mirror
class Object
trait Matchable
class Any
Self type
case class VectorDimensionMismatch(givenDimension: Int, requiredDimension: Int) extends Exception

Attributes

Supertypes
trait Product
trait Equals
class Exception
class Throwable
trait Serializable
class Object
trait Matchable
class Any
Show all
object all

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
all.type
object arrayUtil

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
arrayUtil.type
object blasLeadingDimensionCheck

Whether a single BLAS leading dimension can describe m.

Whether a single BLAS leading dimension can describe m.

BLAS addresses a matrix as one block in which one axis is contiguous and the other advances by a constant lda — a(offset + p + q * lda). So an operand is expressible exactly when one of its strides is 1 and the other is a valid leading dimension for the extent BLAS will check it against.

The second half is the part that is easy to drop, and dropping it is not harmless. lda must be at least the block's row count or the routine rejects the call, and layouts satisfying stride == 1 while failing it do occur: a broadcast column has colStride == 0, repeating one column across the matrix, which no leading dimension expresses. Guarding on stride == 1 alone therefore lets a broadcast operand through to BLAS with lda = 0.

Which extent applies follows from which stride is unit, because that also decides the transpose flag: an operand with rowStride == 1 is passed untransposed and its block has rows rows, so colStride must be at least rows; one with colStride == 1 is passed transposed, its block has cols rows, so rowStride must be at least cols.

Callers that need to know which of the two cases holds — to pick the transpose flag and lda — should test the halves directly rather than call this; this is for the guard that decides whether BLAS can be used at all.

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object broadcast

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
broadcast.type
object cosineSimilarity

Compute the cosine similarity between two vectors

Compute the cosine similarity between two vectors

Value parameters

v1

the first vector

v2

the second vector

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object dimMatCheck

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object dimMatDInstantiateCheck

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object dimMatInstantiateCheck

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object dimNDArrayCheck

dimNDArrayCheck validates that the product of shape dimensions equals data.length.

dimNDArrayCheck validates that the product of shape dimensions equals data.length.

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object doublearrays

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object floatarrays

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object indexCheckMat

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object indexNDArrayCheck

indexNDArrayCheck validates element-access indices against the NDArray's shape.

indexNDArrayCheck validates element-access indices against the NDArray's shape.

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object intarrays

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
intarrays.type
object matmulOutputCheck

Validates the output matrix c of a matmulInPlace! call: it must be shaped exactly (m.rows, b.cols) and dense column-major. matmulInPlace! hardcodes ldc = m.rows and always writes (and, when beta != 0, reads) c assuming that layout, so a wrongly-shaped or non-dense-column-major c would otherwise be corrupted or misread silently instead of failing loudly. matmul/@@ always build a conforming c themselves, so this only bites direct callers of the in-place API.

Validates the output matrix c of a matmulInPlace! call: it must be shaped exactly (m.rows, b.cols) and dense column-major. matmulInPlace! hardcodes ldc = m.rows and always writes (and, when beta != 0, reads) c assuming that layout, so a wrongly-shaped or non-dense-column-major c would otherwise be corrupted or misread silently instead of failing loudly. matmul/@@ always build a conforming c themselves, so this only bites direct callers of the in-place API.

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object matrix

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
matrix.type
object matrixUtil

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
matrixUtil.type
object ndarray

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
ndarray.type
object ndarrayOps

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
ndarrayOps.type
object nonEmptyMatCheck

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object sameDenseElementWiseMemoryLayoutCheck

If this is true, then we can use the same memory layout for element-wise operations

If this is true, then we can use the same memory layout for element-wise operations

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object sameDimMatCheck

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object shapeCheck

shapeCheck validates that all dimensions are > 0. A 0-length shape (0-d array) is valid.

shapeCheck validates that all dimensions are > 0. A 0-length shape (0-d array) is valid.

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
shapeCheck.type
object squareMatCheck

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object strideMatInstantiateCheck

strideMatInstantiateCheck performs a set of safety checks when constructing a matrix view with arbitrary strides and offset into a backing array. The checks include:

strideMatInstantiateCheck performs a set of safety checks when constructing a matrix view with arbitrary strides and offset into a backing array. The checks include:

  • Ensuring the number of rows and columns are positive.
  • Ensuring the offset is within the bounds of the backing array.
  • Ensuring both rowStride and colStride are non-zero.
  • Calculating the maximum and minimum indices that could be accessed by the matrix view, given the strides and offset, and ensuring these indices are within the bounds of the array.
  • Throws appropriate exceptions if any check fails.

Validates matrix construction parameters for stride-based layout.

Performs comprehensive bounds checking for matrices with flexible stride patterns, including support for broadcasting (zero strides) and negative strides for flipped views.

Validates:

  • Positive matrix dimensions
  • Valid offset within array bounds
  • Sensible stride values (zero for broadcasting, non-zero otherwise)
  • All matrix elements remain within array bounds
  • Negative strides don't cause negative index access
  • 1x1 matrices have semantically meaningful strides (0 or 1)

Value parameters

colStride

Memory offset between consecutive columns

cols

Number of matrix columns

offset

Starting position in the data array

raw

The underlying data array

rowStride

Memory offset between consecutive rows

rows

Number of matrix rows

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object strideNDArrayCheck

strideNDArrayCheck validates construction of an NDArray with arbitrary strides and offset.

strideNDArrayCheck validates construction of an NDArray with arbitrary strides and offset.

Validates:

  • shape and strides have the same length (rank consistency)
  • All dimensions in shape are > 0
  • Offset is >= 0 and < data.length
  • Strides are non-zero (except 0 for broadcast dims of size 1)
  • All corner combinations of indices stay within [0, data.length)

All three checks in this file are plain defs despite being generic. They read data.size — deliberately, because .size does not route an abstract element type through ScalaRunTime$.array_length the way .length does — and everything else they touch is shape, strides or indices, all Array[Int]. So there is nothing here that inline was protecting, which makes them unlike dimCheck's generic arms: those reach a.length on an Array[A] and stay inline for exactly the reason these do not need to.

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type
object symmetricMatCheck

Attributes

Supertypes
class Object
trait Matchable
class Any
Self type

Types

type Col = Int
type Row = Int
type RowCol = (Row, Col)
Generated with