Piña is a Python-to-OpenCL translator. And a fruit.

Import the static decorator to turn any Python function to OpenCL kernel code:

from pina import static

@static
def add(x, y):
    return 2.0 * cos(x) + sin(y)

Any return statement will generate an implicit write to the hidden output kernel argument.

Executing the Python code requires a run-time system environment. This is based on PyOpenCL so make sure to install it first:

from pina.ext.pycl import Runtime

r = Runtime()

@r.jit
def add(x, y):
    return 2.0 * cos(x) + sin(y)

Omitting square brackets reading and writing values will be local to the executed kernel. By using square brackets, adjacent data can be accessed in a relative way:

@static
def foo(x):
    left = x[-1]
    right = x[+1]
    top_left = x[-1, -1]
    bottom_left = x[-1, +1]

Static type annotations are written as simple decorator arguments. All integral builtin types as well as all NumPy types are supported::

@static(float, np.uint8)
def bar(x, y):
    pass

OpenCL address space qualifiers can be specified by using the appropriate qualifier classes::

from pina import Global, Constant

@static(Global(np.float32), Constant)
def qux(x, y):
    pass

Some Python features cannot be mapped reasonably onto OpenCL, e.g. import, del, lambda, try, except and finally , with, assert and class.