Fix #15: do not require pycuda for OpenCL execution

pystencils/transformations.py

@@ -1312,13 +1312,16 @@ def implement_interpolations(ast_node: ast.Node,
         substitutions = {i: to_texture_map[i.symbol.interpolator].at(
             [o for o in i.offsets]) for i in interpolation_accesses}
 
-        import pycuda.driver as cuda
-        for texture in substitutions.values():
-            if can_use_hw_interpolation(texture):
-                texture.filter_mode = cuda.filter_mode.LINEAR
-            else:
-                texture.filter_mode = cuda.filter_mode.POINT
-                texture.read_as_integer = True
+        try:
+            import pycuda.driver as cuda
+            for texture in substitutions.values():
+                if can_use_hw_interpolation(texture):
+                    texture.filter_mode = cuda.filter_mode.LINEAR
+                else:
+                    texture.filter_mode = cuda.filter_mode.POINT
+                    texture.read_as_integer = True
+        except Exception:
+            pass
 
         if isinstance(ast_node, AssignmentCollection):
             ast_node = ast_node.subs(substitutions)

pystencils_tests/test_opencl.py

 import numpy as np
 import pytest
-import sympy as sp
 
 import pystencils
+import sympy as sp
 from pystencils.backends.cuda_backend import CudaBackend
 from pystencils.backends.opencl_backend import OpenClBackend
 from pystencils.opencl.opencljit import make_python_function
@@ -195,5 +195,35 @@ def test_opencl_jit_with_parameter():
     assert np.allclose(result_cuda, result_opencl)
 
 
-if __name__ == '__main__':
-    test_opencl_jit()
+@pytest.mark.skipif(not HAS_OPENCL, reason="Test requires pyopencl")
+def test_without_cuda():
+    z, y, x = pystencils.fields("z, y, x: [20,30]")
+
+    assignments = pystencils.AssignmentCollection({
+        z[0, 0]: x[0, 0] * sp.log(x[0, 0] * y[0, 0])
+    })
+
+    print(assignments)
+
+    ast = pystencils.create_kernel(assignments, target='gpu')
+
+    print(ast)
+
+    opencl_code = pystencils.show_code(ast, custom_backend=OpenClBackend())
+    print(opencl_code)
+
+    x_cpu = np.random.rand(20, 30)
+    y_cpu = np.random.rand(20, 30)
+    z_cpu = np.random.rand(20, 30)
+
+    import pyopencl.array as array
+    ctx = cl.create_some_context(0)
+    queue = cl.CommandQueue(ctx)
+
+    x = array.to_device(queue, x_cpu)
+    y = array.to_device(queue, y_cpu)
+    z = array.to_device(queue, z_cpu)
+
+    opencl_kernel = make_python_function(ast, queue, ctx)
+    assert opencl_kernel is not None
+    opencl_kernel(x=x, y=y, z=z)