Noob question: how to manually free the memory of an array

[TsXor]

For example, I want to cauculate:
m = n**2 + p**2
When I execute this line of code, I can see the GPU memory usage increasing (because a new array is assigned to m, right?).
And when I want to free the memory used by n and p, what should I do? I tried del, but it won't help.

Can ary.data.release free the memory used by ary?

[TsXor]

For another example:

a = somewhatarray
b = somewhatarray
x = a**2
x = b**2

Look at the task manager, 4 increase in memory comsumption can be observed (by adding breakpoint() between every line), but no decrease.
Expected: after x = b**2, the memory for a**2 should be freed.

[inducer]

Memory will be freed after the last reference to an object goes away. At x = b**2, the local variables a and b are still holding references to the input arrays.

To manually free memory (technically: release the reference to the buffer, because OpenCL has its own refcounting): https://documen.tician.de/pyopencl/runtime_memory.html#pyopencl.MemoryObject.release

i.e.

a.base_data.release()

[TsXor]

However, I think a**2 is not a, and it is just a thing returned by the magic __pow__, and x is the only reference to it.

[TsXor]

If a**2 is sonewhat a, then memory usage shouldn't increase (at least visibly) when x=a**2 is executed.