Conversation
…of silently skipping the operation addPt, subPt, multPt, add, sub and mult adjust the operand's level and scaling degree to the ciphertext's when they differ. When that adjustment fails (for example a plaintext encoded with fewer RNS limbs than the ciphertext, or a NoiseLevel-2 plaintext at the ciphertext's level) the code hit assert(false) and returned, leaving the ciphertext unmodified. In Release builds the assert is compiled out, so the operation is silently skipped and the caller gets the un-multiplied (un-added) ciphertext back. Report the failure with std::runtime_error naming the operation, both (level, NoiseLevel) pairs and what the operand needs. Also fix subPt's adjust path, which called addPt on the adjusted plaintext and therefore computed ct + pt instead of ct - pt whenever the levels differed. Valid operands are unchanged. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…tment step runs Under FIXEDMANUAL (and with ignore_scale) the level-adjustment branch is skipped, so a plaintext with fewer RNS limbs than the ciphertext reached the kernels and was read past its end (illegal memory access). Check the limb count before the kernels and throw the same exception as the adjustment failure. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This was referenced Sep 5, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Ciphertext::multPt(andaddPt,subPt,add,sub,mult) adjust the operand's level and scaling degree to the ciphertext's when they differ. When the adjustment fails, for example because the plaintext was encoded with fewer RNS limbs than the ciphertext has, the code reachesassert(false)and returns. In a Release build the assert is compiled out, so the multiplication is silently skipped and the caller receives the unmodified ciphertext. We chased this for a while as "multPt does nothing above level 16": the boundary was simply where our plaintexts had fewer limbs than the ciphertext.This PR replaces the six
assert(false)sites with astd::runtime_errorthat names the operation, both (level, NoiseLevel) pairs and what the operand needs. It also fixessubPt's adjust path, which calledaddPton the adjusted plaintext and therefore computed ct + pt instead of ct - pt whenever the levels differed (already fixed onOpenFHECompatTests; included here formain). Nothing changes for operands that adjust successfully.Tested on v2.1.3 (786c760), CUDA 13.0, H200 with N=2^16 / depth 25 / FLEXIBLEAUTO and a depth-12 variant: a plaintext encoded at the same level or at full level multiplies correctly at every ciphertext level (max error ~2e-10, unchanged); a plaintext with fewer limbs than the ciphertext used to return the input unchanged and now throws, leaving the ciphertext untouched;
subPtwith a full-level plaintext now gives ct - pt (before: ct + pt). A stand-alone single-file repro is available if useful.The commit also applies cleanly on
OpenFHECompatTests.Update: a second commit adds the same check before the kernels in
addPt,subPtandmultPtfor the paths where no adjustment step runs (FIXEDMANUAL,ignore_scale). There a plaintext with fewer limbs than the ciphertext was read past its end (illegal memory access); it now throws the same exception. The FIXEDMANUAL top-limb bug found alongside is #39.Fable 5.1 on behalf of Seyfal