dace fails to compile a function with nested loops

[rohanrayan]

Describe the bug
I have a simple program below:

N = dace.symbol('N')
@dace.program(auto_optimize=True)
def error_fn_dace(L:dace.float64[N, N], index:dace.uint64, active_size:dace.uint64):
    for i in range(index, active_size - 1):
        L[i + 1][i] = 1.0

        for j in range(i, active_size-1):
            L[j + 1][i] = 2.0

When trying to compile this function, I get an error, but if I change the j for loop to:
for j in range(i, active_size)
it compiles.

Could you please let me know if I am doing something wrong?

To Reproduce
Steps to reproduce the behavior:

1. I am using dace v0.15.1
2. I have attached the python code which produces the error
   error_report.txt

Expected behavior
The compilation should not fail

Desktop (please complete the following information):

• OS: [Linux]
• DaCe v0.15.1

[tbennun]

Thanks! This is an auto-optimize related failure, which is now fixed in #1746.

There was actually another issue with the code: using uint64 for the bounds and subtracting one would actually convert it to a float64, according to NumPy 1 rules:

>>> np.uint64(1) + 1
2.0

So the correct code (for OpenMP to work) would be:

    @dace.program(auto_optimize=True)
    def tester(L: dace.float64[N, N], index: dace.uint64, active_size: dace.uint64):
        for i in range(index, active_size - 1):
            L[i + 1][i] = 1.0

            for j in range(i, dace.int64(active_size - 1)):
                L[j + 1][i] = 2.0

If you change both arguments to dace.int64, the code runs even faster.