Changes:
- new tools: nlmeans, tensorflow, nufftbase, calc, psf
- calc: access some mathematical functions from the command-line
- nlmean: non-local means filter
- tensorflow: load and apply Tensorflow graph
- nufftbase: compute Fourier transform of basis used in nuFFT
- psf: compute PSF
- nufft: various new options to control behavior
- nufft: basis and weights
- nufft: zero and low memory modes
- nufft: use -x for image dimensions
- sim: Bloch-McConnell simulations
- traj: rotation of Cartesian trajectories
- traj: RAGA trajectories
- reconet: MPI support
- reconet: tensorflow graph support
- reconet: export graph for visualization
- pics: mpi and multi-GPU support
- pics: add shared-img-dims option to pics
- pics: add option to specify FISTA
- pics: preconditioning
- nlinv: add option to specifiy non-Cart image size
- phantom: BRAIN geometry
- phantom: 64ch sensitivities
- phantom: ellipsoid simulation
- phantom: support for generic geometries
- mobafit: add basis to mobafit
- mobafit: add IR and IR LL to mobafit
- mobafit: magnitude fitting
- mobafit: scaling and initial values
- mobafit: minimum and maximum bounds
- mobafit: GPU support
- nrmse: add option for scientific notation
- moba: Export and ksp coils in moba
- moba: Various new options
- estdelay: support for looping
- version: show tensorflow and MPI support
- signal: Spin-Echo signal model
- scripts/shell: kspace_precond.sh: Script to compute k-space preconditioner
- scripts/python: bart_tf.py: Export tensorflow graphs for use with BART
- scripts/python: bart.py: Print stdout and stderr output from bart call
- scripts/python: splines_from_svg.py: extract spline control points from SVG files
- library/geom: Brain geometry
- library/misc: better backtraces
- library/num: md_ravel_index, md_select_strides, md_zsmin2, md_unravel
- library/num: library for complex-valued Gaussians
- library/num: cuda kernels: zdot, zfsq2, zfftmod_3d, zsmin, compress, decompress
- library/num: cuda batch kernels: xpay, axpy, dot_bat
- library/num: cuda kernels: (z)add_unfold, (z)mul(c)_unfold, (z)fmac(c)_unfold
- library/num: md_nlmeans(2), md_nlmeans_distance(2)
- library/num: library for MPI sync, broadcast, reductions
- library/num: double precision bessel function
- library/num: special functions: sinc, sincf, jinc
- library/num: library for virtual pointers
- library/num: time-ordered matrix exponentials: mat_to_exp
- library/num: functions for adjoint sensitivity analysis
- library/num: functions for quadratures: trapezoidal, simpson
- library/num: configure fftw wisdom/measure via environment variable
- library/simu: Bloch McConnell simulations
- library/simu: Spin Echo signal model
- library/linops: linop: flip
- library/nlops: nlops: zphsr and zabs
- library/nn: nlops: ardioid, siglog, igaussian
- library/network: tensorflow networks
- library/noncart: nufft inverse gridding on GPU
- library/noncart: RAGA trajectories
- library/iter: weighted least-squares prox function
- library/iter: batched conjgrad
- library/iter: multiple regularizations for primal-dual
- general: webasm support
- general: BART_ prefix to environment variables
- general: generic loop interface to bart
- general: basic MPI support
- general: OpenMPI and GPU for gridding
- general: zero-overhead mode for nuFFT
- general: performance improvements
- general: many GPU improvements, unified GPU initialization
- built: FORTRAN and PNG options
- built: LTO integration test
- built: OMP=0 CUDA=1 compilation
- built: more unit and integration tests
- many other bug fixes and improvements