updated repo for publishing (#74)

This commit is contained in:
Jean-Philippe Bossuat
2025-08-17 14:57:39 +02:00
committed by GitHub
parent 0be569eca0
commit 62eb87cc07
244 changed files with 374 additions and 539 deletions

View File

@@ -0,0 +1,489 @@
use std::collections::HashMap;
use poulpy_backend::hal::{
api::{
ScratchAvailable, TakeMatZnx, TakeVecZnx, TakeVecZnxBig, TakeVecZnxDft, TakeVecZnxDftSlice, TakeVecZnxSlice,
VecZnxAddInplace, VecZnxAutomorphismInplace, VecZnxBigAddSmallInplace, VecZnxBigAllocBytes, VecZnxBigAutomorphismInplace,
VecZnxBigNormalize, VecZnxBigNormalizeTmpBytes, VecZnxBigSubSmallBInplace, VecZnxCopy, VecZnxDftAddInplace,
VecZnxDftAllocBytes, VecZnxDftCopy, VecZnxDftFromVecZnx, VecZnxDftToVecZnxBigConsume, VecZnxDftToVecZnxBigTmpA,
VecZnxNegateInplace, VecZnxNormalizeInplace, VecZnxNormalizeTmpBytes, VecZnxRotate, VecZnxRotateInplace,
VecZnxRshInplace, VecZnxSub, VecZnxSubABInplace, VecZnxSwithcDegree, VmpApply, VmpApplyAdd, VmpApplyTmpBytes,
},
layouts::{Backend, DataMut, DataRef, Module, Scratch, ToOwnedDeep},
oep::{ScratchOwnedAllocImpl, ScratchOwnedBorrowImpl},
};
use poulpy_core::{GLWEOperations, TakeGGLWE, TakeGLWECt, layouts::Infos};
use poulpy_core::layouts::{GGSWCiphertext, GLWECiphertext, LWECiphertext, prepared::GGLWEAutomorphismKeyPrepared};
use crate::tfhe::{
blind_rotation::{
BlincRotationExecute, BlindRotationAlgo, BlindRotationKeyPrepared, LookUpTable, LookUpTableRotationDirection,
},
circuit_bootstrapping::{CircuitBootstrappingKeyPrepared, CirtuitBootstrappingExecute},
};
impl<D: DataRef, BRA: BlindRotationAlgo, B> CirtuitBootstrappingExecute<B> for CircuitBootstrappingKeyPrepared<D, BRA, B>
where
Module<B>: VecZnxRotateInplace
+ VecZnxNormalizeInplace<B>
+ VecZnxNormalizeTmpBytes
+ VecZnxSwithcDegree
+ VecZnxBigAutomorphismInplace<B>
+ VecZnxRshInplace
+ VecZnxDftCopy<B>
+ VecZnxDftToVecZnxBigTmpA<B>
+ VecZnxSub
+ VecZnxAddInplace
+ VecZnxNegateInplace
+ VecZnxCopy
+ VecZnxSubABInplace
+ VecZnxDftAllocBytes
+ VmpApplyTmpBytes
+ VecZnxBigNormalizeTmpBytes
+ VmpApply<B>
+ VmpApplyAdd<B>
+ VecZnxDftFromVecZnx<B>
+ VecZnxDftToVecZnxBigConsume<B>
+ VecZnxBigAddSmallInplace<B>
+ VecZnxBigNormalize<B>
+ VecZnxAutomorphismInplace
+ VecZnxBigSubSmallBInplace<B>
+ VecZnxBigAllocBytes
+ VecZnxDftAddInplace<B>
+ VecZnxRotate,
B: Backend + ScratchOwnedAllocImpl<B> + ScratchOwnedBorrowImpl<B>,
Scratch<B>: TakeVecZnx
+ TakeVecZnxDftSlice<B>
+ TakeVecZnxBig<B>
+ TakeVecZnxDft<B>
+ TakeMatZnx
+ ScratchAvailable
+ TakeVecZnxSlice,
BlindRotationKeyPrepared<D, BRA, B>: BlincRotationExecute<B>,
{
fn execute_to_constant<DM: DataMut, DR: DataRef>(
&self,
module: &Module<B>,
res: &mut GGSWCiphertext<DM>,
lwe: &LWECiphertext<DR>,
log_domain: usize,
extension_factor: usize,
scratch: &mut Scratch<B>,
) {
circuit_bootstrap_core(
false,
module,
0,
res,
lwe,
log_domain,
extension_factor,
self,
scratch,
);
}
fn execute_to_exponent<DM: DataMut, DR: DataRef>(
&self,
module: &Module<B>,
log_gap_out: usize,
res: &mut GGSWCiphertext<DM>,
lwe: &LWECiphertext<DR>,
log_domain: usize,
extension_factor: usize,
scratch: &mut Scratch<B>,
) {
circuit_bootstrap_core(
true,
module,
log_gap_out,
res,
lwe,
log_domain,
extension_factor,
self,
scratch,
);
}
}
#[allow(clippy::too_many_arguments)]
pub fn circuit_bootstrap_core<DRes, DLwe, DBrk, BRA: BlindRotationAlgo, B>(
to_exponent: bool,
module: &Module<B>,
log_gap_out: usize,
res: &mut GGSWCiphertext<DRes>,
lwe: &LWECiphertext<DLwe>,
log_domain: usize,
extension_factor: usize,
key: &CircuitBootstrappingKeyPrepared<DBrk, BRA, B>,
scratch: &mut Scratch<B>,
) where
DRes: DataMut,
DLwe: DataRef,
DBrk: DataRef,
Module<B>: VecZnxRotateInplace
+ VecZnxNormalizeInplace<B>
+ VecZnxNormalizeTmpBytes
+ VecZnxSwithcDegree
+ VecZnxBigAutomorphismInplace<B>
+ VecZnxRshInplace
+ VecZnxDftCopy<B>
+ VecZnxDftToVecZnxBigTmpA<B>
+ VecZnxSub
+ VecZnxAddInplace
+ VecZnxNegateInplace
+ VecZnxCopy
+ VecZnxSubABInplace
+ VecZnxDftAllocBytes
+ VmpApplyTmpBytes
+ VecZnxBigNormalizeTmpBytes
+ VmpApply<B>
+ VmpApplyAdd<B>
+ VecZnxDftFromVecZnx<B>
+ VecZnxDftToVecZnxBigConsume<B>
+ VecZnxBigAddSmallInplace<B>
+ VecZnxBigNormalize<B>
+ VecZnxAutomorphismInplace
+ VecZnxBigSubSmallBInplace<B>
+ VecZnxBigAllocBytes
+ VecZnxDftAddInplace<B>
+ VecZnxRotate,
B: Backend + ScratchOwnedAllocImpl<B> + ScratchOwnedBorrowImpl<B>,
Scratch<B>: TakeVecZnxDftSlice<B>
+ TakeVecZnxBig<B>
+ TakeVecZnxDft<B>
+ TakeVecZnx
+ ScratchAvailable
+ TakeVecZnxSlice
+ TakeMatZnx,
BlindRotationKeyPrepared<DBrk, BRA, B>: BlincRotationExecute<B>,
{
#[cfg(debug_assertions)]
{
assert_eq!(res.n(), key.brk.n());
assert_eq!(lwe.basek(), key.brk.basek());
assert_eq!(res.basek(), key.brk.basek());
}
let n: usize = res.n();
let basek: usize = res.basek();
let rows: usize = res.rows();
let rank: usize = res.rank();
let k: usize = res.k();
let alpha: usize = rows.next_power_of_two();
let mut f: Vec<i64> = vec![0i64; (1 << log_domain) * alpha];
if to_exponent {
(0..rows).for_each(|i| {
f[i] = 1 << (basek * (rows - 1 - i));
});
} else {
(0..1 << log_domain).for_each(|j| {
(0..rows).for_each(|i| {
f[j * alpha + i] = j as i64 * (1 << (basek * (rows - 1 - i)));
});
});
}
// Lut precision, basically must be able to hold the decomposition power basis of the GGSW
let mut lut: LookUpTable = LookUpTable::alloc(n, basek, basek * rows, extension_factor);
lut.set(module, &f, basek * rows);
if to_exponent {
lut.set_rotation_direction(LookUpTableRotationDirection::Right);
}
// TODO: separate GGSW k from output of blind rotation k
let (mut res_glwe, scratch1) = scratch.take_glwe_ct(n, basek, k, rank);
let (mut tmp_gglwe, scratch2) = scratch1.take_gglwe(n, basek, k, rows, 1, rank.max(1), rank);
key.brk.execute(module, &mut res_glwe, lwe, &lut, scratch2);
let gap: usize = 2 * lut.drift / lut.extension_factor();
let log_gap_in: usize = (usize::BITS - (gap * alpha - 1).leading_zeros()) as _;
(0..rows).for_each(|i| {
let mut tmp_glwe: GLWECiphertext<&mut [u8]> = tmp_gglwe.at_mut(i, 0);
if to_exponent {
// Isolates i-th LUT and moves coefficients according to requested gap.
post_process(
module,
&mut tmp_glwe,
&res_glwe,
log_gap_in,
log_gap_out,
log_domain,
&key.atk,
scratch2,
);
} else {
tmp_glwe.trace(module, 0, module.log_n(), &res_glwe, &key.atk, scratch2);
}
if i < rows {
res_glwe.rotate_inplace(module, -(gap as i64));
}
});
// Expands GGLWE to GGSW using GGLWE(s^2)
res.from_gglwe(module, &tmp_gglwe, &key.tsk, scratch2);
}
#[allow(clippy::too_many_arguments)]
fn post_process<DataRes, DataA, B: Backend>(
module: &Module<B>,
res: &mut GLWECiphertext<DataRes>,
a: &GLWECiphertext<DataA>,
log_gap_in: usize,
log_gap_out: usize,
log_domain: usize,
auto_keys: &HashMap<i64, GGLWEAutomorphismKeyPrepared<Vec<u8>, B>>,
scratch: &mut Scratch<B>,
) where
DataRes: DataMut,
DataA: DataRef,
Module<B>: VecZnxRotateInplace
+ VecZnxNormalizeInplace<B>
+ VecZnxNormalizeTmpBytes
+ VecZnxSwithcDegree
+ VecZnxBigAutomorphismInplace<B>
+ VecZnxRshInplace
+ VecZnxDftCopy<B>
+ VecZnxDftToVecZnxBigTmpA<B>
+ VecZnxSub
+ VecZnxAddInplace
+ VecZnxNegateInplace
+ VecZnxCopy
+ VecZnxSubABInplace
+ VecZnxDftAllocBytes
+ VmpApplyTmpBytes
+ VecZnxBigNormalizeTmpBytes
+ VmpApply<B>
+ VmpApplyAdd<B>
+ VecZnxDftFromVecZnx<B>
+ VecZnxDftToVecZnxBigConsume<B>
+ VecZnxBigAddSmallInplace<B>
+ VecZnxBigNormalize<B>
+ VecZnxAutomorphismInplace
+ VecZnxBigSubSmallBInplace<B>
+ VecZnxRotate,
Scratch<B>: TakeVecZnxDft<B> + ScratchAvailable + TakeVecZnx,
{
let log_n: usize = module.log_n();
let mut cts: HashMap<usize, GLWECiphertext<Vec<u8>>> = HashMap::new();
// First partial trace, vanishes all coefficients which are not multiples of gap_in
// [1, 1, 1, 1, 0, 0, 0, ..., 0, 0, -1, -1, -1, -1] -> [1, 0, 0, 0, 0, 0, 0, ..., 0, 0, 0, 0, 0, 0]
res.trace(
module,
module.log_n() - log_gap_in + 1,
log_n,
a,
auto_keys,
scratch,
);
// TODO: optimize with packing and final partial trace
// If gap_out < gap_in, then we need to repack, i.e. reduce the cap between coefficients.
if log_gap_in != log_gap_out {
let steps: i32 = 1 << log_domain;
(0..steps).for_each(|i| {
if i != 0 {
res.rotate_inplace(module, -(1 << log_gap_in));
}
cts.insert(i as usize * (1 << log_gap_out), res.to_owned_deep());
});
pack(module, &mut cts, log_gap_out, auto_keys, scratch);
let packed: GLWECiphertext<Vec<u8>> = cts.remove(&0).unwrap();
res.trace(
module,
log_n - log_gap_out,
log_n,
&packed,
auto_keys,
scratch,
);
}
}
pub fn pack<D: DataMut, B: Backend>(
module: &Module<B>,
cts: &mut HashMap<usize, GLWECiphertext<D>>,
log_gap_out: usize,
auto_keys: &HashMap<i64, GGLWEAutomorphismKeyPrepared<Vec<u8>, B>>,
scratch: &mut Scratch<B>,
) where
Module<B>: VecZnxRotateInplace
+ VecZnxNormalizeInplace<B>
+ VecZnxNormalizeTmpBytes
+ VecZnxSwithcDegree
+ VecZnxBigAutomorphismInplace<B>
+ VecZnxRshInplace
+ VecZnxDftCopy<B>
+ VecZnxDftToVecZnxBigTmpA<B>
+ VecZnxSub
+ VecZnxAddInplace
+ VecZnxNegateInplace
+ VecZnxCopy
+ VecZnxSubABInplace
+ VecZnxDftAllocBytes
+ VmpApplyTmpBytes
+ VecZnxBigNormalizeTmpBytes
+ VmpApply<B>
+ VmpApplyAdd<B>
+ VecZnxDftFromVecZnx<B>
+ VecZnxDftToVecZnxBigConsume<B>
+ VecZnxBigAddSmallInplace<B>
+ VecZnxBigNormalize<B>
+ VecZnxAutomorphismInplace
+ VecZnxBigSubSmallBInplace<B>
+ VecZnxRotate,
Scratch<B>: TakeVecZnx + TakeVecZnxDft<B> + ScratchAvailable,
{
let log_n: usize = module.log_n();
let basek: usize = cts.get(&0).unwrap().basek();
let k: usize = cts.get(&0).unwrap().k();
let rank: usize = cts.get(&0).unwrap().rank();
(0..log_n - log_gap_out).for_each(|i| {
let t: usize = 16.min(1 << (log_n - 1 - i));
let auto_key: &GGLWEAutomorphismKeyPrepared<Vec<u8>, B> = if i == 0 {
auto_keys.get(&-1).unwrap()
} else {
auto_keys.get(&module.galois_element(1 << (i - 1))).unwrap()
};
(0..t).for_each(|j| {
let mut a: Option<GLWECiphertext<D>> = cts.remove(&j);
let mut b: Option<GLWECiphertext<D>> = cts.remove(&(j + t));
combine(
module,
basek,
k,
rank,
a.as_mut(),
b.as_mut(),
i,
auto_key,
scratch,
);
if let Some(a) = a {
cts.insert(j, a);
} else if let Some(b) = b {
cts.insert(j, b);
}
});
});
}
#[allow(clippy::too_many_arguments)]
fn combine<A: DataMut, D: DataMut, DataAK: DataRef, B: Backend>(
module: &Module<B>,
basek: usize,
k: usize,
rank: usize,
a: Option<&mut GLWECiphertext<A>>,
b: Option<&mut GLWECiphertext<D>>,
i: usize,
auto_key: &GGLWEAutomorphismKeyPrepared<DataAK, B>,
scratch: &mut Scratch<B>,
) where
Module<B>: VecZnxRotateInplace
+ VecZnxNormalizeInplace<B>
+ VecZnxNormalizeTmpBytes
+ VecZnxSwithcDegree
+ VecZnxBigAutomorphismInplace<B>
+ VecZnxRshInplace
+ VecZnxDftCopy<B>
+ VecZnxDftToVecZnxBigTmpA<B>
+ VecZnxSub
+ VecZnxAddInplace
+ VecZnxNegateInplace
+ VecZnxCopy
+ VecZnxSubABInplace
+ VecZnxDftAllocBytes
+ VmpApplyTmpBytes
+ VecZnxBigNormalizeTmpBytes
+ VmpApply<B>
+ VmpApplyAdd<B>
+ VecZnxDftFromVecZnx<B>
+ VecZnxDftToVecZnxBigConsume<B>
+ VecZnxBigAddSmallInplace<B>
+ VecZnxBigNormalize<B>
+ VecZnxAutomorphismInplace
+ VecZnxBigSubSmallBInplace<B>
+ VecZnxRotate,
Scratch<B>: TakeVecZnx + TakeVecZnxDft<B> + ScratchAvailable,
{
// Goal is to evaluate: a = a + b*X^t + phi(a - b*X^t))
// We also use the identity: AUTO(a * X^t, g) = -X^t * AUTO(a, g)
// where t = 2^(log_n - i - 1) and g = 5^{2^(i - 1)}
// Different cases for wether a and/or b are zero.
//
// Implicite RSH without modulus switch, introduces extra I(X) * Q/2 on decryption.
// Necessary so that the scaling of the plaintext remains constant.
// It however is ok to do so here because coefficients are eventually
// either mapped to garbage or twice their value which vanishes I(X)
// since 2*(I(X) * Q/2) = I(X) * Q = 0 mod Q.
if let Some(a) = a {
let n: usize = a.n();
let log_n: usize = (u64::BITS - (n - 1).leading_zeros()) as _;
let t: i64 = 1 << (log_n - i - 1);
if let Some(b) = b {
let (mut tmp_b, scratch_1) = scratch.take_glwe_ct(n, basek, k, rank);
// a = a * X^-t
a.rotate_inplace(module, -t);
// tmp_b = a * X^-t - b
tmp_b.sub(module, a, b);
tmp_b.rsh(module, 1);
// a = a * X^-t + b
a.add_inplace(module, b);
a.rsh(module, 1);
tmp_b.normalize_inplace(module, scratch_1);
// tmp_b = phi(a * X^-t - b)
tmp_b.automorphism_inplace(module, auto_key, scratch_1);
// a = a * X^-t + b - phi(a * X^-t - b)
a.sub_inplace_ab(module, &tmp_b);
a.normalize_inplace(module, scratch_1);
// a = a + b * X^t - phi(a * X^-t - b) * X^t
// = a + b * X^t - phi(a * X^-t - b) * - phi(X^t)
// = a + b * X^t + phi(a - b * X^t)
a.rotate_inplace(module, t);
} else {
a.rsh(module, 1);
// a = a + phi(a)
a.automorphism_add_inplace(module, auto_key, scratch);
}
} else if let Some(b) = b {
let n: usize = b.n();
let log_n: usize = (u64::BITS - (n - 1).leading_zeros()) as _;
let t: i64 = 1 << (log_n - i - 1);
let (mut tmp_b, scratch_1) = scratch.take_glwe_ct(n, basek, k, rank);
tmp_b.rotate(module, t, b);
tmp_b.rsh(module, 1);
// a = (b* X^t - phi(b* X^t))
b.automorphism_sub_ba(module, &tmp_b, auto_key, scratch_1);
}
}