Add Type Change Pass

[creavin]

This change adds a new DaCE Pass that can replace one simple DaCe dtype with another. This change was successfully tested on a complex ICON sdfg. Here is a sample usage:

import dace
import numpy as np

from dace.transformation import pass_pipeline as ppl
from dace.transformation.passes.type_change import TypeChange

N = dace.symbol('N')

@dace.program
def simple(a: dace.float64[N], b: dace.float64[1]):
    for i in range(N):
        a[i] = a[i] * 2.0 + b[0]

sdfg = simple.to_sdfg()
tc = TypeChange(dace.float64, dace.float32)

type_change_pipeline = ppl.Pipeline([tc])
print("Pipeline created")
results = type_change_pipeline.apply_pass(sdfg, {})

print(results) # {'TypeChange': 6}

A = np.ones(10, dtype=np.float32)
B = np.ones(1, dtype=np.float64)
sdfg(A, B, N=10)
print(A)

[tbennun]

I’m not sure this is a good candidate for a pass. Maybe if it checked for reductions and, e.g., transcendental ops, for higher precision it would be more appropriate. Otherwise, as it is right now it should be a utility function.

Additionally, please avoid type(x) == y checks

[phschaad]

Some comments for code improvements. Additionally, I agree with the point raised by @tbennun

[creavin]

Thank you both for your comments

I’m not sure this is a good candidate for a pass. Maybe if it checked for reductions and, e.g., transcendental ops, for higher precision it would be more appropriate. Otherwise, as it is right now it should be a utility function.

I'm happy to move this into a utility function then. Where is the correct place to put it / do you have a similar function I can mimic

Additionally, please avoid type(x) == y checks

Yes, good point. Will use isinstance instead

[creavin]

Changes:

• moved to helper function file
• uses isinstance checks
• ran linter first
• I keep literal == checks for the direct because I know that behaviour to work correctly. I don't have the time to extensively test for new behaviour with isinstance

[phschaad]

The change looks good to me, however the PR is missing unit tests. This is a new feature and as such is certainly not covered by existing tests. Please ensure there are at least 2-3 unit tests to check for correct type changing behavior w.r.t. the data types and places the utility function modifies.

[creavin]

Perfect, will do