Atb.py

import copy

import numpy as np
from _Atb import _Atb_ext

from .gpu import GpuIds


def Atb(projections, geo, angles, backprojection_type="FDK", **kwargs):
    if projections.dtype != np.float32:
        raise TypeError("Input data should be float32, not " + str(projections.dtype))
    if not np.isreal(projections).all():
        raise ValueError("Complex types not compatible for back projection.")
    geox = copy.deepcopy(geo)
    geox.check_geo(angles)
    """
    Here we cast all values in geo to single point precision float. This way we know what behavior
    to expect from pytigre to Cuda and can change single parameters accordingly.
    """
    geox.cast_to_single()
    # geox.checknans()

    if projections.shape != (geox.angles.shape[0], geox.nDetector[0], geox.nDetector[1]):
        raise ValueError(
            "Expected data shape: "
            + str((geox.angles.shape[0], geox.nDetector[0], geox.nDetector[1]))
            + " not compatible with: "
            + str(projections.shape)
        )

    if kwargs.get("gpuids", None) is None:
        gpuids = GpuIds()
    else:
        gpuids = kwargs["gpuids"]

    # if we have more GPUs than slices to compute, reduce the amount of GPUs. 
    if geo.nVoxel[0]< len(gpuids):
        gpuids.devices = list(gpuids.devices[0:geo.nVoxel[0]])

    return _Atb_ext(projections, geox, geox.angles, backprojection_type, geox.mode, gpuids=gpuids)