wip all of Encryption

This commit is contained in:
Rasoul Akhavan Mahdavi
2025-10-15 21:31:48 -04:00
parent 2f2c7aef00
commit 9f35a92a33
16 changed files with 316 additions and 231 deletions

View File

@@ -1,29 +1,33 @@
use poulpy_hal::{
api::{
SvpApplyDftToDft, SvpApplyDftToDftInplace, SvpPPolBytesOf, SvpPrepare, VecZnxAddInplace, VecZnxAddNormal,
ModuleN, ScratchTakeBasic, SvpApplyDftToDft, SvpApplyDftToDftInplace, SvpPPolAlloc, SvpPPolBytesOf, SvpPrepare, VecZnxAddInplace, VecZnxAddNormal,
VecZnxAddScalarInplace, VecZnxBigBytesOf, VecZnxBigNormalize, VecZnxDftApply, VecZnxDftBytesOf, VecZnxFillUniform,
VecZnxIdftApplyConsume, VecZnxIdftApplyTmpA, VecZnxNormalize, VecZnxNormalizeInplace, VecZnxNormalizeTmpBytes, VecZnxSub,
VecZnxSubInplace, VecZnxSwitchRing,
},
layouts::{Backend, DataMut, DataRef, Module, Scratch},
oep::VecZnxBigAllocBytesImpl,
source::Source,
};
use crate::layouts::{
GGLWEInfos, GLWEInfos, GLWESecret, GLWESwitchingKey, LWEInfos, Rank, RingDegree, TensorKey, prepared::GLWESecretPrepared,
use crate::{
ScratchTakeCore,
layouts::{
GetDist, GGLWEInfos, GLWEInfos, GLWESecret, GLWESwitchingKey, LWEInfos, Rank, TensorKey, prepared::GLWESecretPrepared,
},
};
impl TensorKey<Vec<u8>> {
pub fn encrypt_sk_tmp_bytes<B: Backend, A>(module: &Module<B>, infos: &A) -> usize
where
A: GGLWEInfos,
Module<B>: SvpPPolBytesOf + VecZnxNormalizeTmpBytes + VecZnxDftBytesOf + VecZnxNormalizeTmpBytes + VecZnxBigBytesOf,
Module<B>: ModuleN + SvpPPolBytesOf + SvpPPolAlloc<B> + VecZnxNormalizeTmpBytes + VecZnxDftBytesOf + VecZnxNormalizeTmpBytes + VecZnxBigBytesOf,
{
GLWESecretPrepared::bytes_of(module, infos.rank_out())
+ module.bytes_of_vec_znx_dft(infos.rank_out().into(), 1)
+ module.bytes_of_vec_znx_big(1, 1)
+ module.bytes_of_vec_znx_dft(1, 1)
+ GLWESecret::bytes_of(RingDegree(module.n() as u32), Rank(1))
+ GLWESecret::bytes_of(module, Rank(1))
+ GLWESwitchingKey::encrypt_sk_tmp_bytes(module, infos)
}
}
@@ -37,7 +41,9 @@ impl<DataSelf: DataMut> TensorKey<DataSelf> {
source_xe: &mut Source,
scratch: &mut Scratch<B>,
) where
Module<B>: SvpApplyDftToDft<B>
GLWESecret<DataSk>: GetDist,
Module<B>: ModuleN
+ SvpApplyDftToDft<B>
+ VecZnxIdftApplyTmpA<B>
+ VecZnxAddScalarInplace
+ VecZnxDftBytesOf
@@ -55,8 +61,11 @@ impl<DataSelf: DataMut> TensorKey<DataSelf> {
+ VecZnxSub
+ SvpPrepare<B>
+ VecZnxSwitchRing
+ SvpPPolBytesOf,
Scratch<B>:,
+ SvpPPolBytesOf
+ VecZnxBigAllocBytesImpl<B>
+ VecZnxBigBytesOf
+ SvpPPolAlloc<B>,
Scratch<B>: ScratchTakeBasic + ScratchTakeCore<B>,
{
#[cfg(debug_assertions)]
{
@@ -64,21 +73,21 @@ impl<DataSelf: DataMut> TensorKey<DataSelf> {
assert_eq!(self.n(), sk.n());
}
let n: RingDegree = sk.n();
// let n: RingDegree = sk.n();
let rank: Rank = self.rank_out();
let (mut sk_dft_prep, scratch_1) = scratch.take_glwe_secret_prepared(n, rank);
sk_dft_prep.prepare(module, sk, scratch_1);
let (mut sk_dft_prep, scratch_1) = scratch.take_glwe_secret_prepared(module, rank);
sk_dft_prep.prepare(module, sk);
let (mut sk_dft, scratch_2) = scratch_1.take_vec_znx_dft(n.into(), rank.into(), 1);
let (mut sk_dft, scratch_2) = scratch_1.take_vec_znx_dft(module, rank.into(), 1);
(0..rank.into()).for_each(|i| {
module.vec_znx_dft_apply(1, 0, &mut sk_dft, i, &sk.data.as_vec_znx(), i);
});
let (mut sk_ij_big, scratch_3) = scratch_2.take_vec_znx_big(n.into(), 1, 1);
let (mut sk_ij, scratch_4) = scratch_3.take_glwe_secret(n, Rank(1));
let (mut sk_ij_dft, scratch_5) = scratch_4.take_vec_znx_dft(n.into(), 1, 1);
let (mut sk_ij_big, scratch_3) = scratch_2.take_vec_znx_big(module, 1, 1);
let (mut sk_ij, scratch_4) = scratch_3.take_glwe_secret(module, Rank(1));
let (mut sk_ij_dft, scratch_5) = scratch_4.take_vec_znx_dft(module, 1, 1);
(0..rank.into()).for_each(|i| {
(i..rank.into()).for_each(|j| {