Problems with casting

[numba]

Problems with casting

[jriehl]

The float() built-in is not currently handled by the compiler, and we don't currently have code to cleanly identify when a built-in function is not supported.

Numba does handle implicit casts, however. You should only have to say result[i] += avg / N, and Numba should cast N to a floating-point value.

I can't currently speak to the segmentation fault, but I'll definitely add a test case and see if I can reproduce all of this.

[ovillellas]

Related to this: I've run in several problems when working with number literals. I've got problems when trying to generate code for functions that do literals. For example, the following function works fine when generating code with d[:], d[:] input and output, but will fail with f[:], f[:] ones.

def twice(X, D):
    M = X.shape[0]
    for i in range(M):
        D[i] = 2.0 * X[i]

ValueError: Unsupported cast (from <Variable(val=<llvm.core.Instruction object at 0x10984aa50>, _llvm=<llvm.core.Instruction object at 0x10984aa50>, typ='f32')>.typ to 'f64')

Note that this does not happen if I use 2 instead of 2.0.

Something similar happens with the following function:

def pow2(X,D):
    M = X.shape[X]
    for i in range(M):
        D[i] = X[i] ** 2

No matter what combination of input types I try to use, code-generation fails with an "Unsupported Cast".

[markflorisson]

This issue has been fixed for some time now, so I'm closing it.