Am/test/noise check compression

[IceTDrinker]

added a pfail estimate script in the scripts dir if people want to have a quick tool to check and see what some results might be when doing some noise measurements

added tests for the compression atomic pattern, had to update some structs to be more dynamic for noise simulation (like being able to handle ciphertext moduli on u64 or u128, notably for the packing ksk)

one change compared to other tests is the iteration count which are adapted here to generate the same number of keys as the classic PBS ones for the pfail tests, noise checks are still lower in overall keys being generated because some confidence interval estimation functions can return NaN for large amounts of samples

lifted restrictions on the confidence interval, if the computation does not return finite f64 values we panic with an indication of what should be done to the user

---

This change is 

[mayeul-zama]

@mayeul-zama reviewed 14 of 15 files at r1, all commit messages.
Reviewable status: 14 of 15 files reviewed, 4 unresolved discussions

---

-- commits line 12 at r1:
empty

---

scripts/pfail_estimate.py line 0 at r1 (raw file):
What is this file needed for?

---

tfhe/src/shortint/server_key/tests/noise_distribution/br_dp_packingks_ms.rs line 166 at r1 (raw file):

    sanity_check_encrypt_br_dp_packing_ks_ms(
        TEST_PARAM_MESSAGE_2_CARRY_2_KS_PBS_TUNIFORM_2M128,
        TEST_COMP_PARAM_MESSAGE_2_CARRY_2_KS_PBS_TUNIFORM_2M128,

We could make sanity_check_encrypt_br_dp_packing_ks_ms take a single metaparam struct here instead of 2 parameter sets which may not correspond to each other

Same for other similar test functions

---

tfhe/src/shortint/server_key/tests/noise_distribution/br_dp_packingks_ms.rs line 642 at r1 (raw file):

        // - so that the effective encoding after the storage is the one we used to evaluate the
        //   pfail
        // TODO: do something about this

Should this be fixed now?

[IceTDrinker]

@IceTDrinker reviewed 1 of 15 files at r1, all commit messages.
Reviewable status: 14 of 15 files reviewed, 4 unresolved discussions (waiting on @mayeul-zama)

---

-- commits line 12 at r1:

Previously, mayeul-zama wrote…

empty

yep, I opened the issue and updated the commit, will repush with the other changes

---

scripts/pfail_estimate.py line at r1 (raw file):

Previously, mayeul-zama wrote…

What is this file needed for?

It's not needed but it's a useful python tool I regularly use for pfail questions, seemed useful to commit to have in the repo

---

tfhe/src/shortint/server_key/tests/noise_distribution/br_dp_packingks_ms.rs line 166 at r1 (raw file):

Previously, mayeul-zama wrote…

We could make sanity_check_encrypt_br_dp_packing_ks_ms take a single metaparam struct here instead of 2 parameter sets which may not correspond to each other

Same for other similar test functions

fair, is it ok with you if it's done in a follow up PR where I would do it for the various noise tests instead of making this PR even bigger than it is ?

---

tfhe/src/shortint/server_key/tests/noise_distribution/br_dp_packingks_ms.rs line 642 at r1 (raw file):

Previously, mayeul-zama wrote…

Should this be fixed now?

I could add a test primitive to be able to update the message mod yes

[mayeul-zama]

Reviewable status: 14 of 15 files reviewed, 3 unresolved discussions (waiting on @IceTDrinker)

---

-- commits line 12 at r1:

Previously, IceTDrinker wrote…

yep, I opened the issue and updated the commit, will repush with the other changes

Ok

---

tfhe/src/shortint/server_key/tests/noise_distribution/br_dp_packingks_ms.rs line 166 at r1 (raw file):

Previously, IceTDrinker wrote…

fair, is it ok with you if it's done in a follow up PR where I would do it for the various noise tests instead of making this PR even bigger than it is ?

Sure

---

tfhe/src/shortint/server_key/tests/noise_distribution/br_dp_packingks_ms.rs line 642 at r1 (raw file):

Previously, IceTDrinker wrote…

I could add a test primitive to be able to update the message mod yes

Ok

[mayeul-zama]

Thanks!

@mayeul-zama reviewed 1 of 2 files at r2, all commit messages.
Reviewable status: 14 of 15 files reviewed, all discussions resolved