ignoring zero diagonals in diagonal_ct_vector_matmul

[mrader1248]

Description

As explained in #398, matrices with diagonals that contain nothing but zeros cannot be multiplied with a vector, as this results in ValueError: result ciphertext is transparent. This can be fixed by simply ignoring zero diagonals, which in addition potentially saves some unnecessary computations.

Affected Dependencies

None

How has this been tested?

I used the following code to test, which previously resulted in the error explained above.

import numpy as np
import tenseal as ts

n = 1000
x = np.random.rand(n)
i = np.eye(n)
poly_mod_degree = 4096
coeff_mod_bit_sizes = [40, 20, 40]
ctx = ts.context(ts.SCHEME_TYPE.CKKS, poly_mod_degree, -1, coeff_mod_bit_sizes)
ctx.generate_galois_keys()
ctx.global_scale = 2 ** 20
x_enc = ts.ckks_vector(ctx, x)
x_enc @ i

Checklist

• I have followed the Contribution Guidelines and Code of Conduct
• I have commented my code following the OpenMined Styleguide
• I have labeled this PR with the relevant Type labels
• My changes are covered by tests

[mrader1248]

Is it possible that the test runner had a problem or _almost_equal had a bad day? I cannot reproduce the reported failure of

tests/python/tenseal/tensors/test_ckks_tensor.py::test_subscript[data0-slices0-new_shape0]

locally.

[mrader1248]

@bcebere, could you help me here?

[mrader1248]

If make is called with -j the number of processes for the the build process is not limited and it crashes unless there is enough RAM. For this reason, I modified setup.py such that -j is only added if the environment variable MAKEFLAGS is not set. With this, setup.py respects a user's default make flags, e.g. defined in .zshrc or .bashrc, or MAKEFLAGS="-j4" pip install ... can be called to limit the number of processes directly. The default behaviour is unchanged (unlimited number of processes if MAKEFLAGS is not set).

[MartinNoc]

@youben11
Does anybody have time to review this PR? Thanks 🙏

[bcebere]

LGTM!
Thank you for your contribution, @mrader1248 !