mirror of
https://github.com/arnaucube/poulpy.git
synced 2026-02-10 05:06:44 +01:00
115 lines
4.1 KiB
Rust
115 lines
4.1 KiB
Rust
use poulpy_hal::{
|
|
api::{
|
|
ModuleN, ScratchAvailable, SvpApplyDftToDftInplace, SvpPPolAlloc, SvpPPolBytesOf, SvpPrepare, VecZnxAddInplace, VecZnxAddNormal,
|
|
VecZnxAddScalarInplace, VecZnxBigNormalize, VecZnxDftApply, VecZnxDftBytesOf, VecZnxFillUniform, VecZnxIdftApplyConsume,
|
|
VecZnxNormalize, VecZnxNormalizeInplace, VecZnxNormalizeTmpBytes, VecZnxSub, VecZnxSubInplace, VecZnxSwitchRing,
|
|
ScratchTakeBasic,
|
|
},
|
|
layouts::{Backend, DataMut, DataRef, Module, ScalarZnx, Scratch},
|
|
source::Source,
|
|
};
|
|
|
|
use crate::{
|
|
ScratchTakeCore,
|
|
layouts::{
|
|
GGLWE, GGLWEInfos, GLWEInfos, GLWESecret, GLWESwitchingKey, LWEInfos, RingDegree, prepared::GLWESecretPrepared,
|
|
},
|
|
};
|
|
|
|
impl GLWESwitchingKey<Vec<u8>> {
|
|
pub fn encrypt_sk_tmp_bytes<B: Backend, A>(module: &Module<B>, infos: &A) -> usize
|
|
where
|
|
A: GGLWEInfos,
|
|
Module<B>: ModuleN + SvpPPolBytesOf + VecZnxNormalizeTmpBytes + VecZnxDftBytesOf + VecZnxNormalizeTmpBytes + SvpPPolAlloc<B>,
|
|
{
|
|
(GGLWE::encrypt_sk_tmp_bytes(module, infos) | ScalarZnx::bytes_of(module.n(), 1))
|
|
+ ScalarZnx::bytes_of(module.n(), infos.rank_in().into())
|
|
+ GLWESecretPrepared::bytes_of_from_infos(module, &infos.glwe_layout())
|
|
}
|
|
|
|
pub fn encrypt_pk_tmp_bytes<B: Backend, A>(module: &Module<B>, _infos: &A) -> usize
|
|
where
|
|
A: GGLWEInfos,
|
|
{
|
|
GGLWE::encrypt_pk_tmp_bytes(module, _infos)
|
|
}
|
|
}
|
|
|
|
impl<DataSelf: DataMut> GLWESwitchingKey<DataSelf> {
|
|
#[allow(clippy::too_many_arguments)]
|
|
pub fn encrypt_sk<DataSkIn: DataRef, DataSkOut: DataRef, B: Backend>(
|
|
&mut self,
|
|
module: &Module<B>,
|
|
sk_in: &GLWESecret<DataSkIn>,
|
|
sk_out: &GLWESecret<DataSkOut>,
|
|
source_xa: &mut Source,
|
|
source_xe: &mut Source,
|
|
scratch: &mut Scratch<B>,
|
|
) where
|
|
Module<B>: ModuleN
|
|
+ VecZnxAddScalarInplace
|
|
+ VecZnxDftBytesOf
|
|
+ VecZnxBigNormalize<B>
|
|
+ VecZnxDftApply<B>
|
|
+ SvpApplyDftToDftInplace<B>
|
|
+ VecZnxIdftApplyConsume<B>
|
|
+ VecZnxNormalizeTmpBytes
|
|
+ VecZnxFillUniform
|
|
+ VecZnxSubInplace
|
|
+ VecZnxAddInplace
|
|
+ VecZnxNormalizeInplace<B>
|
|
+ VecZnxAddNormal
|
|
+ VecZnxNormalize<B>
|
|
+ VecZnxSub
|
|
+ SvpPrepare<B>
|
|
+ VecZnxSwitchRing
|
|
+ SvpPPolBytesOf
|
|
+ SvpPPolAlloc<B>,
|
|
Scratch<B>: ScratchAvailable + ScratchTakeBasic + ScratchTakeCore<B>,
|
|
{
|
|
#[cfg(debug_assertions)]
|
|
{
|
|
assert!(sk_in.n().0 <= module.n() as u32);
|
|
assert!(sk_out.n().0 <= module.n() as u32);
|
|
assert!(
|
|
scratch.available() >= GLWESwitchingKey::encrypt_sk_tmp_bytes(module, self),
|
|
"scratch.available()={} < GLWESwitchingKey::encrypt_sk_tmp_bytes={}",
|
|
scratch.available(),
|
|
GLWESwitchingKey::encrypt_sk_tmp_bytes(module, self)
|
|
)
|
|
}
|
|
|
|
let n: usize = sk_in.n().max(sk_out.n()).into();
|
|
|
|
let (mut sk_in_tmp, scratch_1) = scratch.take_scalar_znx(module, sk_in.rank().into());
|
|
(0..sk_in.rank().into()).for_each(|i| {
|
|
module.vec_znx_switch_ring(
|
|
&mut sk_in_tmp.as_vec_znx_mut(),
|
|
i,
|
|
&sk_in.data.as_vec_znx(),
|
|
i,
|
|
);
|
|
});
|
|
|
|
let (mut sk_out_tmp, scratch_2) = scratch_1.take_glwe_secret_prepared(module, sk_out.rank());
|
|
{
|
|
let (mut tmp, _) = scratch_2.take_scalar_znx(module, 1);
|
|
(0..sk_out.rank().into()).for_each(|i| {
|
|
module.vec_znx_switch_ring(&mut tmp.as_vec_znx_mut(), 0, &sk_out.data.as_vec_znx(), i);
|
|
module.svp_prepare(&mut sk_out_tmp.data, i, &tmp, 0);
|
|
});
|
|
}
|
|
|
|
self.key.encrypt_sk(
|
|
module,
|
|
&sk_in_tmp,
|
|
&sk_out_tmp,
|
|
source_xa,
|
|
source_xe,
|
|
scratch_2,
|
|
);
|
|
self.sk_in_n = sk_in.n().into();
|
|
self.sk_out_n = sk_out.n().into();
|
|
}
|
|
}
|