NumericAL
A header-only C++20 lexer built on REAL
Loading...
Searching...
No Matches
numerical Directory Reference
Directory dependency graph for numerical:

Files

 amx.hpp
 The matrix coprocessor of Apple's M-series chips (AMX), for the double matrix product.
 
 banded.hpp
 Structured systems: banded matrices, in LAPACK's diagonal-ordered storage, the general band by its LU factorisation with partial pivoting (gbsv, gtsv for three diagonals), the Hermitian positive-definite band by its Cholesky factor (pbsv, ptsv's L D Lᴴ for three); Toeplitz matrices by Levinson's recursion; circulant matrices by the Fourier transform. And the structured matrices themselves, as scipy.linalg forms them: Toeplitz, circulant, Hankel, block diagonal.
 
 blas.hpp
 The column-major semantics of the BLAS and LAPACK routines NumericAL stands in for (ddot, dgemm, dsyrk, dgemv, dtrsm, dgesv, dgetrf, dgetrs, dpotrf, dpotrs, dsyevd, dgesdd, dgeqrf, dorgqr, dgelsd, dgeev), as their reference implementations define them, over NumericAL's kernels: what the optional numerical_blas library exports with the Fortran calling convention (blas/numerical_blas.cpp), so that a program linked against BLAS and LAPACK, or a switch such as FlexiBLAS, can run on NumericAL.
 
 buffer.hpp
 numerical::buffer, the storage of the library's vectors, matrices and tensors: a std::vector whose elements constructed with no value are left uninitialised.
 
 complex_math.hpp
 The elementary functions of a complex number of float or double as the GNU C library computes them (NumPy's on Linux, to the bit): C's Annex G for the infinities and NaNs, scaled against overflow, the logarithm's real part near the unit circle from an almost exact x² + y² − 1.
 
 concepts.hpp
 The scalar types NumericAL's factorisations and solvers take: a field of floating-point numbers, real or complex. An integer matrix multiplies exactly, but a decomposition of one would divide in integers.
 
 decomposition.hpp
 Matrix factorisations and the algorithms built on them.
 
 dqds.hpp
 The singular values of a real bidiagonal by the dqds algorithm (Fernando and Parlett's differential quotient-difference with shifts, as LAPACK's dlasq1–dlasq6): the eigenvalues of BᵀB from the qd array of the squares of B's entries, every transform's numbers non-negative, so that each value is found to high relative accuracy. The routines keep LAPACK's one-based indices into the array Z of 4n entries, so that each line reads against its original.
 
 eigen.hpp
 The eigenvalues and eigenvectors of a general square matrix.
 
 elementary.hpp
 exp, exp2, expm1, tanh, sinh and cosh of double by a table of 2^(j/128) and a polynomial, atan by a table of atan(k/64) and a series, log1p by a table of ln(k/128) and a series, branch-free so that a loop of them runs in vectors: within 0.506 ulp of the exact value for a normal result of exp and exp2, 0.503 ulp for expm1 and sinh; the same bits from a vector lane as from a call.
 
 factorizations.hpp
 The LU, Cholesky and QR factorisations and the symmetric and Hermitian eigen-decompositions.
 
 fft.hpp
 The discrete Fourier transform of a tensor along an axis, as numpy.fft.fft and numpy.fft.ifft compute it.
 
 fft_kernels.hpp
 The transform's radix-4 and radix-16 passes in registers of complex numbers, written once for every unit (AVX2, AVX-512, NEON): each operation overloaded on a register of the unit (x86.hpp's on x86-64, NEON's here) and the passes' bodies inlined into a function of each, so that every unit runs the same operations lane by lane, whatever its width: the same bits.
 
 gemm.hpp
 The matrix product kernel behind matrix * matrix and the einsum matmul path.
 
 generalized.hpp
 The generalized Schur decomposition (QZ) of a pencil A − λ B, as LAPACK's zgges and dgges: the operands scaled into range, the reduction to Hessenberg-triangular form (zgghrd, dgghrd), the QZ iteration (zhgeqz single-shift, dhgeqz double-shift with 2 × 2 blocks for conjugate pairs), and the reordering of the eigenvalues by exchanges of adjacent blocks (ztgexc/ztgsen, dtgexc/dtgex2/dtgsen).
 
 ldl.hpp
 The symmetric indefinite factorisation A = L D Lᵀ (or L D Lᴴ) of a symmetric or Hermitian matrix with Bunch and Kaufman's diagonal pivoting, D block diagonal with blocks of order 1 and 2, as LAPACK's sytrf and hetrf (by panels, lasyf and lahef, from order 384; sytf2 and hetf2 below) and as scipy.linalg.ldl returns it.
 
 matrix.hpp
 A dense, row-major, value-semantic matrix over a scalar type.
 
 matrix_functions.hpp
 Functions of a square matrix and the matrix equations solved through Schur forms: the exponential, by scaling and squaring of a Padé approximant; the principal square root, by the Schur method; the Sylvester equation A X + X B = Q, as Bartels and Stewart.
 
 memory.hpp
 The workspaces NumericAL keeps between calls, given back (numerical::release_memory).
 
 numerical.hpp
 Single public header for NumericAL — include this to use the library.
 
 parallel.hpp
 A minimal std::thread parallel-for for the hot kernels.
 
 simd.hpp
 The native vector of float and double the kernels hold their lanes in (GNU vector extensions, under GCC and clang; a kernel keeps a scalar form beside it for other compilers).
 
 svd.hpp
 The singular value decomposition and what it answers: the singular values, least squares, the pseudo-inverse and the numerical rank. The bidiagonal reduction and its iterations, divide and conquer and one-sided Jacobi live here, apart from the factorisations they build on (factorizations.hpp).
 
 tensor.hpp
 A dense, row-major, N-dimensional tensor and Einstein summation.
 
 tensor_ops.hpp
 Element-wise arithmetic with NumPy's broadcasting, reductions, reshaping and creation of tensors.
 
 vector.hpp
 A dense, value-semantic vector over a scalar type.
 
 x86.hpp
 The x86-64 tiles of the double matrix product: AVX-512 and AVX2 with FMA, chosen at run time.