Let user decide which cupy version to use

README.md

@@ -16,6 +16,8 @@ Interface for using cupy in xarray, providing convenience accessors.
 
 ## Installation
 
+> `cupy-xarray` will use an existing cupy installation, hence cupy needs to be installed manually! Please follow cupy's install instructions at <https://docs.cupy.dev/en/stable/install.html> or use the extras install
+
 From anaconda:
 
 ```console
@@ -34,6 +36,14 @@ The latest version from Github:
 pip install git+https://github.com/xarray-contrib/cupy-xarray.git
 ```
 
+Install cupy with extras:
+
+```console
+pip install cupy-xarray["source"] # will install the cupy package, which will build cupy from source
+pip install cupy-xarray["cuda11"] # will install the prebuild cupy-cuda11x package
+pip install cupy-xarray["cuda12"] # will install the prebuild cupy-cuda12x package
+```
+
 ## Usage
 
 ```python

cupy_xarray/accessors.py

@@ -1,6 +1,18 @@
+import warnings
 from typing import TYPE_CHECKING, Any
 
-import cupy as cp
+try:
+    import cupy as cp
+except ImportError as e:
+    warnings.warn(
+        "Cupy is not installed. cupy-xarray expects cupy to be manually installed. Please install "
+        "cupy either by following the instructions at https://docs.cupy.dev/en/stable/install.html "
+        "or by isntalling cupy-xaray with extras, e.g. `pip install cupy-xarray['source']`. More "
+        "information can be found in the Readme or at the cupy-xarray documentation.",
+        ImportWarning,
+        stacklevel=2,
+    )
+    raise e
 from xarray import (
     DataArray,
     Dataset,

docs/index.md

@@ -16,6 +16,8 @@ CuPy-Xarray is a Python library that leverages [CuPy](https://cupy.dev/), a GPU
 
 ## Installation
 
+> `cupy-xarray` will use an existing cupy installation, hence cupy needs to be installed manually! Please follow cupy's install instructions at <https://docs.cupy.dev/en/stable/install.html> or use the extras install
+
 CuPy-Xarray can be installed using `pip` or `conda`:
 
 From Conda Forge:
@@ -37,6 +39,14 @@ The latest version from Github:
 pip install git+https://github.com/xarray-contrib/cupy-xarray.git
 ```
 
+Install cupy with extras:
+
+```console
+pip install cupy-xarray["source"] # will install the cupy package, which will build cupy from source
+pip install cupy-xarray["cuda11"] # will install the prebuild cupy-cuda11x package
+pip install cupy-xarray["cuda12"] # will install the prebuild cupy-cuda12x package
+```
+
 ## Acknowledgements
 
 Large parts of this documentations comes from [SciPy 2023 Xarray on GPUs tutorial](https://negin513.github.io/cupy-xarray-tutorials/README.html) and [this NCAR tutorial to GPUs](https://github.com/NCAR/GPU_workshop/tree/workshop/13_CuPyAndLegate).

pyproject.toml

@@ -15,7 +15,6 @@ classifiers = [
 ]
 dynamic = ["version"]
 dependencies = [
-    "cupy",
     "xarray>=2024.02.0",
 ]
 
@@ -24,6 +23,16 @@ test = [
     "dask",
     "pytest",
 ]
+source = [
+    "cupy"
+]
+cuda12 = [
+    "cupy-cuda12x",
+]
+cuda11 = [
+    "cupy-cuda11x",
+]
+
 
 [tool.ruff]
 line-length = 100  # E501 (line-too-long)