README.md

Sample Kernels

This directory contains sample kernels implemented using google test and google benchmark for Microsoft SEAL and PALISADE. These sample kernels are for operations that are not full standalone use cases. An example of such an operation is matrix multiplication for which we provide several different sample kernels showing alternative methods of implementing the same algorithm using different approaches, libraries, and schemes. The full set of available sample kernels for each library and a brief description are described in the details section below.

Usage

This directory contains sample kernels which benchmark a set of more complex HE operations in both SEAL and PALISADE.

To run the sample kernels such as dot product, call

# Note, these will take several minutes
$HE_SAMPLES/build/sample-kernels/sample-kernels-seal
$HE_SAMPLES/build/sample-kernels/sample-kernels-palisade

this will output timing results using google benchmark.

For unit test, run

$HE_SAMPLES/build/sample-kernels/test/unit-test

Details

The following tables list and give a brief description of the different sample kernels available. All sample kernels use N = 8192 with 3 Coefficient modulus primes.

Microsoft SEAL Sample Kernels

Sample Kernel Name	Brief Description
DotPlainBatchAxis_CKKS	Ciphertext-plaintext matrix multiplication w/ matrix represented as one element per ciphertext using a collection of adds and multiplications w/ CKKS.
DotCipherBatchAxis_CKKS	Ciphertext-ciphertext matrix multiplication w/ matrix represented as one element per ciphertext using a collection of adds and multiplications w/ CKKS.
MatMulVal_CKKS	Matrix multiplication w/ matrix represented one row per ciphertext and manual computation using a collection of adds, multiplications, and rotations w/ CKKS.
LogisticRegression_CKKS	Computes logistic regression inference using a batch of inputs.
DotPlainBatchAxis_BFV	Ciphertext-plaintext matrix multiplication w/ matrix represented as one element per ciphertext using a collection of adds and multiplications w/ BFV.
DotCipherBatchAxis_BFV	Ciphertext-ciphertext matrix multiplication w/ matrix represented as one element per ciphertext using a collection of adds and multiplications w/ BFV.
MatMulVal_BFV	Matrix multiplication w/ matrix represented one row per ciphertext and manual computation using a collection of adds, multiplications, and rotations w/ BFV.
MatMulRow_BFV	Matrix multiplication w/ one matrix represented fully in a single ciphertext and the other matrix represented two rows at a time w/ BFV.

PALISADE Sample Kernels

Sample Kernel Name	Brief Description
DotPlainBatchAxis_CKKS	Ciphertext-plaintext matrix multiplication w/ matrix represented as one element per ciphertext using a collection of adds and multiplications w/ CKKS.
DotCipherBatchAxis_CKKS	Ciphertext-ciphertext matrix multiplication w/ matrix represented as one element per ciphertext using a collection of adds and multiplications w/ CKKS.
MatMulEIP_CKKS	Matrix multiplication w/ matrix represented one row per ciphertext using Palisade’s internal dot product calculation w/ CKKS.
MatMulVal_CKKS	Matrix multiplication w/ matrix represented one row per ciphertext and manual computation using a collection of adds, multiplications, and rotations w/ CKKS.
MatMulRow_CKKS	Matrix multiplication w/ one matrix represented fully in a single ciphertext and the other matrix represented row by row w/ CKKS.
LogisticRegression_CKKS	Computes logistic regression inference using a batch of inputs.
DotPlainBatchAxis_BFV	Ciphertext-plaintext matrix multiplication w/ matrix represented as one element per ciphertext using a collection of adds and multiplications w/ BFV.
DotCipherBatchAxis_BFV	Ciphertext-ciphertext matrix multiplication w/ matrix represented as one element per ciphertext using a collection of adds and multiplications w/ BFV.
MatMulEIP_BFV	Matrix multiplication w/ matrix represented one row per ciphertext using Palisade’s internal dot product calculation w/ BFV.
MatMulVal_BFV	Matrix multiplication w/ matrix represented one row per ciphertext and manual computation using a collection of adds, multiplications, and rotations w/ BFV.
MatMulRow_BFV	Matrix multiplication w/ one matrix represented fully in a single ciphertext and the other matrix represented row by row w/ BFV.