Some tweaks for stacked nufft (#173)

* fix(test): make stacked-gpu tests run.

* feat: make z_index default to "auto"

* docs: z-index computation.

src/mrinufft/operators/stacked.py

@@ -40,7 +40,8 @@ class MRIStackedNUFFT(FourierOperatorBase):
     shape: tuple
         Shape of the image.
     z_index: array-like
-        Cartesian z index of masked plan.
+        Cartesian z index of masked plan. if "auto" the z_index is computed from the
+        samples, if they are 3D, using the last coordinate.
     backend: str or FourierOperatorBase
         Backend to use.
         If str, a NUFFT operator is initialized with str being a registered backend.
@@ -74,9 +75,9 @@ def __init__(
         self,
         samples,
         shape,
-        z_index,
         backend,
         smaps,
+        z_index="auto",
         n_coils=1,
         n_batchs=1,
         squeeze_dims=False,
@@ -353,7 +354,8 @@ class MRIStackedNUFFTGPU(MRIStackedNUFFT):
     shape: tuple
         Shape of the image.
     z_index: array-like
-        Cartesian z index of masked plan.
+        Cartesian z index of masked plan. if "auto" the z_index is computed from the
+        samples, if they are 3D, using the last coordinate.
     smaps: array-like
         Sensitivity maps.
     n_coils: int
@@ -371,11 +373,11 @@ def __init__(
         self,
         samples,
         shape,
-        z_index,
         smaps,
         n_coils=1,
         n_batchs=1,
         n_trans=1,
+        z_index="auto",
         squeeze_dims=False,
         smaps_cached=False,
         density=False,

tests/helpers/factories.py

@@ -102,6 +102,7 @@ def wrapper(operator, array_interface, *args, **kwargs):
                 "gpunufft",
                 "torchkbnufft-gpu",
                 "tensorflow",
+                "stacked-cufinufft",
             ]:
                 pytest.skip("Uncompatible backend and array")
         if array_interface in ["torch-cpu", "numpy"]: