arnaucube/poulpy
1
0
Fork 0
mirror of https://github.com/arnaucube/poulpy.git synced 2026-02-10 13:16:44 +01:00
Code Issues Packages Projects Releases Wiki Activity

Added LWE-GLWE conversion & LWE Keyswitch, improved LUT generation

Browse Source
This commit is contained in:
Jean-Philippe Bossuat
2025-07-07 11:09:04 +02:00
parent c4a517e9c3
commit 5234c3fc63
28 changed files with 979 additions and 782 deletions
Download Patch File Download Diff File Expand all files Collapse all files

216
core/src/blind_rotation/ccgi.rs
View File

@@ -1,13 +1,12 @@
use std::time::Instant;
use backend::{
MatZnxDftOps, MatZnxDftScratch, Module, ScalarZnx, ScalarZnxAlloc, ScalarZnxDft, ScalarZnxDftAlloc, ScalarZnxDftOps, Scratch, VecZnxBigAlloc, VecZnxBigOps, VecZnxBigScratch, VecZnxDftOps, VecZnxOps, ZnxInfos, ZnxView, ZnxViewMut, ZnxZero, FFT64
FFT64, MatZnxDftOps, MatZnxDftScratch, Module, ScalarZnx, ScalarZnxAlloc, ScalarZnxDft, ScalarZnxDftAlloc, ScalarZnxDftOps,
Scratch, VecZnxAlloc, VecZnxBigAlloc, VecZnxBigOps, VecZnxBigScratch, VecZnxDftAlloc, VecZnxDftOps, VecZnxOps, ZnxView,
ZnxViewMut, ZnxZero,
};
use itertools::izip;
use crate::{
FourierGLWECiphertext, FourierGLWESecret, GLWECiphertext, GLWECiphertextToMut, GLWEPlaintext, Infos, LWECiphertext,
ScratchCore,
GLWECiphertext, GLWECiphertextToMut, Infos, LWECiphertext,
blind_rotation::{key::BlindRotationKeyCGGI, lut::LookUpTable},
lwe::ciphertext::LWECiphertextToRef,
};
@@ -21,25 +20,31 @@ pub fn cggi_blind_rotate_scratch_space(
rows: usize,
rank: usize,
) -> usize {
let lut_size: usize = k_lut.div_ceil(basek);
let cols: usize = rank + 1;
let brk_size: usize = k_brk.div_ceil(basek);
let acc_dft: usize = FourierGLWECiphertext::bytes_of(module, basek, k_brk, rank) * extension_factor;
let acc_big: usize = module.bytes_of_vec_znx_big(rank + 1, brk_size);
let acc_dft_add: usize = FourierGLWECiphertext::bytes_of(module, basek, k_brk, rank) * extension_factor;
let vmp_res: usize = FourierGLWECiphertext::bytes_of(module, basek, k_brk, rank) * extension_factor;
let acc_dft: usize = module.bytes_of_vec_znx_dft(cols, rows) * extension_factor;
let acc_big: usize = module.bytes_of_vec_znx_big(1, brk_size);
let vmp_res: usize = module.bytes_of_vec_znx_dft(cols, brk_size) * extension_factor;
let acc_dft_add: usize = vmp_res;
let xai_plus_y: usize = module.bytes_of_scalar_znx(1);
let xai_plus_y_dft: usize = module.bytes_of_scalar_znx_dft(1);
let vmp: usize = module.vmp_apply_tmp_bytes(lut_size, lut_size, rows, 2, 2, brk_size); // GGSW product: (1 x 2) x (2 x 2)
let vmp: usize = module.vmp_apply_tmp_bytes(brk_size, rows, rows, 2, 2, brk_size); // GGSW product: (1 x 2) x (2 x 2)
let acc: usize;
if extension_factor > 1 {
acc = GLWECiphertext::bytes_of(module, basek, k_lut, rank) * extension_factor;
acc = module.bytes_of_vec_znx(cols, k_lut.div_ceil(basek)) * extension_factor;
} else {
acc = 0;
}
return acc + acc_big + acc_dft + acc_dft_add + vmp_res + xai_plus_y + xai_plus_y_dft + (vmp | module.vec_znx_big_normalize_tmp_bytes());
return acc
+ acc_dft
+ acc_dft_add
+ vmp_res
+ xai_plus_y
+ xai_plus_y_dft
+ (vmp | (acc_big + (module.vec_znx_big_normalize_tmp_bytes() | module.vec_znx_idft_tmp_bytes())));
}
pub fn cggi_blind_rotate<DataRes, DataIn>(
@@ -62,6 +67,7 @@ pub fn cggi_blind_rotate<DataRes, DataIn>(
}
}
// TODO: ENSURE DOMAIN EXTENSION AS
pub(crate) fn cggi_blind_rotate_block_binary_extended<DataRes, DataIn>(
module: &Module<FFT64>,
res: &mut GLWECiphertext<DataRes>,
@@ -75,27 +81,25 @@ pub(crate) fn cggi_blind_rotate_block_binary_extended<DataRes, DataIn>(
{
let extension_factor: usize = lut.extension_factor();
let basek: usize = res.basek();
let rows: usize = brk.rows();
let cols: usize = res.rank() + 1;
let (mut acc, scratch1) = scratch.tmp_vec_glwe_ct(extension_factor, module, basek, res.k(), res.rank());
let (mut acc_dft, scratch2) = scratch1.tmp_vec_fourier_glwe_ct(extension_factor, module, basek, brk.k(), res.rank());
let (mut vmp_res, scratch3) = scratch2.tmp_vec_fourier_glwe_ct(extension_factor, module, basek, brk.k(), res.rank());
let (mut acc_add_dft, scratch4) = scratch3.tmp_vec_fourier_glwe_ct(extension_factor, module, basek, brk.k(), res.rank());
(0..extension_factor).for_each(|i| {
acc[i].data.zero();
});
let (mut acc, scratch1) = scratch.tmp_slice_vec_znx(extension_factor, module, cols, res.size());
let (mut acc_dft, scratch2) = scratch1.tmp_slice_vec_znx_dft(extension_factor, module, cols, rows);
let (mut vmp_res, scratch3) = scratch2.tmp_slice_vec_znx_dft(extension_factor, module, cols, brk.size());
let (mut acc_add_dft, scratch4) = scratch3.tmp_slice_vec_znx_dft(extension_factor, module, cols, brk.size());
let (mut xai_plus_y, scratch5) = scratch4.tmp_scalar_znx(module, 1);
let (mut xai_plus_y_dft, scratch6) = scratch5.tmp_scalar_znx_dft(module, 1);
let (mut acc_add_big, scratch7) = scratch6.tmp_vec_znx_big(module, 1, brk.size());
(0..extension_factor).for_each(|i| {
acc[i].zero();
});
let mut lwe_2n: Vec<i64> = vec![0i64; lwe.n() + 1]; // TODO: from scratch space
let lwe_ref: LWECiphertext<&[u8]> = lwe.to_ref();
let two_n_ext: usize = 2 * lut.domain_size();
let cols: usize = res.rank() + 1;
negate_and_mod_switch_2n(two_n_ext, &mut lwe_2n, &lwe_ref);
let a: &[i64] = &lwe_2n[1..];
@@ -105,10 +109,10 @@ pub(crate) fn cggi_blind_rotate_block_binary_extended<DataRes, DataIn>(
let b_lo: usize = b_pos % extension_factor;
for (i, j) in (0..b_lo).zip(extension_factor - b_lo..extension_factor) {
module.vec_znx_rotate(b_hi as i64 + 1, &mut acc[i].data, 0, &lut.data[j], 0);
module.vec_znx_rotate(b_hi as i64 + 1, &mut acc[i], 0, &lut.data[j], 0);
}
for (i, j) in (b_lo..extension_factor).zip(0..extension_factor - b_lo) {
module.vec_znx_rotate(b_hi as i64, &mut acc[i].data, 0, &lut.data[j], 0);
module.vec_znx_rotate(b_hi as i64, &mut acc[i], 0, &lut.data[j], 0);
}
let block_size: usize = brk.block_size();
@@ -121,9 +125,9 @@ pub(crate) fn cggi_blind_rotate_block_binary_extended<DataRes, DataIn>(
.for_each(|(i, (ai, ski))| {
(0..extension_factor).for_each(|i| {
(0..cols).for_each(|j| {
module.vec_znx_dft(1, 0, &mut acc_dft[i].data, j, &acc[i].data, j);
module.vec_znx_dft(1, 0, &mut acc_dft[i], j, &acc[i], j);
});
acc_add_dft[i].data.zero();
acc_add_dft[i].zero();
});
// TODO: first & last iterations can be optimized
@@ -134,25 +138,19 @@ pub(crate) fn cggi_blind_rotate_block_binary_extended<DataRes, DataIn>(
// vmp_res = DFT(acc) * BRK[i]
(0..extension_factor).for_each(|i| {
module.vmp_apply(&mut vmp_res[i].data, &acc_dft[i].data, &skii.data, scratch7);
module.vmp_apply(&mut vmp_res[i], &acc_dft[i], &skii.data, scratch6);
});
// Trivial case: no rotation between polynomials, we can directly multiply with (X^{-ai} - 1)
if ai_lo == 0 {
// DFT X^{-ai}
set_xai_plus_y(
module,
ai_hi as i64,
-1,
&mut xai_plus_y_dft,
&mut xai_plus_y,
);
set_xai_plus_y(module, ai_hi, -1, &mut xai_plus_y_dft, &mut xai_plus_y);
// Sets acc_add_dft[i] = (acc[i] * sk) * (X^{-ai} - 1)
(0..extension_factor).for_each(|j| {
(0..cols).for_each(|i| {
module.svp_apply_inplace(&mut vmp_res[j].data, i, &xai_plus_y_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft[j].data, i, &vmp_res[j].data, i);
module.svp_apply_inplace(&mut vmp_res[j], i, &xai_plus_y_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft[j], i, &vmp_res[j], i);
});
});
// Non trivial case: rotation between polynomials
@@ -163,74 +161,83 @@ pub(crate) fn cggi_blind_rotate_block_binary_extended<DataRes, DataIn>(
// Sets acc_add_dft[i] = acc[i] * sk
(0..extension_factor).for_each(|i| {
(0..cols).for_each(|k| {
module.vec_znx_dft_sub_ab_inplace(&mut acc_add_dft[i].data, k, &vmp_res[i].data, k);
module.vec_znx_dft_sub_ab_inplace(&mut acc_add_dft[i], k, &vmp_res[i], k);
})
});
// DFT X^{-ai+1}
set_xai_plus_y(
module,
ai_hi as i64 + 1,
0,
&mut xai_plus_y_dft,
&mut xai_plus_y,
);
// DFT X^{-ai}
set_xai_plus_y(module, ai_hi + 1, 0, &mut xai_plus_y_dft, &mut xai_plus_y);
// Sets acc_add_dft[0..ai_lo] += (acc[extension_factor - ai_lo..extension_factor] * sk) * X^{-ai+1}
for (i, j) in (0..ai_lo).zip(extension_factor - ai_lo..extension_factor) {
module.vec_znx_rotate(b_hi as i64 + 1, &mut acc[i].data, 0, &lut.data[j], 0);
(0..cols).for_each(|k| {
module.svp_apply_inplace(&mut vmp_res[j].data, k, &xai_plus_y_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft[i].data, k, &vmp_res[j].data, k);
module.svp_apply_inplace(&mut vmp_res[j], k, &xai_plus_y_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft[i], k, &vmp_res[j], k);
});
}
// DFT X^{-ai}
set_xai_plus_y(
module,
ai_hi as i64,
0,
&mut xai_plus_y_dft,
&mut xai_plus_y,
);
set_xai_plus_y(module, ai_hi, 0, &mut xai_plus_y_dft, &mut xai_plus_y);
// Sets acc_add_dft[ai_lo..extension_factor] += (acc[0..extension_factor - ai_lo] * sk) * X^{-ai}
for (i, j) in (ai_lo..extension_factor).zip(0..extension_factor - ai_lo) {
module.vec_znx_rotate(b_hi as i64, &mut acc[i].data, 0, &lut.data[j], 0);
(0..cols).for_each(|k| {
module.svp_apply_inplace(&mut vmp_res[j].data, k, &xai_plus_y_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft[i].data, k, &vmp_res[j].data, k);
module.svp_apply_inplace(&mut vmp_res[j], k, &xai_plus_y_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft[i], k, &vmp_res[j], k);
});
}
}
});
(0..extension_factor).for_each(|j| {
(0..cols).for_each(|i| {
module.vec_znx_dft_add_inplace(&mut acc_dft[j].data, i, &acc_add_dft[j].data, i);
module.vec_znx_idft(&mut acc_add_big, 0, &acc_dft[j].data, i, scratch7);
module.vec_znx_big_normalize(basek, &mut acc[j].data, i, &acc_add_big, 0, scratch7);
{
let (mut acc_add_big, scratch7) = scratch6.tmp_vec_znx_big(module, 1, brk.size());
(0..extension_factor).for_each(|j| {
(0..cols).for_each(|i| {
module.vec_znx_idft(&mut acc_add_big, 0, &acc_add_dft[j], i, scratch7);
module.vec_znx_big_add_small_inplace(&mut acc_add_big, 0, &acc[j], i);
module.vec_znx_big_normalize(basek, &mut acc[j], i, &acc_add_big, 0, scratch7);
});
});
});
}
});
(0..cols).for_each(|i| {
module.vec_znx_copy(&mut res.data, i, &acc[0].data, i);
module.vec_znx_copy(&mut res.data, i, &acc[0], i);
});
}
fn set_xai_plus_y(
module: &Module<FFT64>,
k: i64,
ai: usize,
y: i64,
res: &mut ScalarZnxDft<&mut [u8], FFT64>,
buf: &mut ScalarZnx<&mut [u8]>,
) {
buf.zero();
buf.at_mut(0, 0)[0] = 1;
module.vec_znx_rotate_inplace(k, buf, 0);
buf.at_mut(0, 0)[0] += y;
let n: usize = module.n();
{
let raw: &mut [i64] = buf.at_mut(0, 0);
if ai < n {
raw[ai] = 1;
} else {
raw[(ai - n) & (n - 1)] = -1;
}
raw[0] += y;
}
module.svp_prepare(res, 0, buf, 0);
{
let raw: &mut [i64] = buf.at_mut(0, 0);
if ai < n {
raw[ai] = 0;
} else {
raw[(ai - n) & (n - 1)] = 0;
}
raw[0] = 0;
}
}
pub(crate) fn cggi_blind_rotate_block_binary<DataRes, DataIn>(
@@ -244,11 +251,12 @@ pub(crate) fn cggi_blind_rotate_block_binary<DataRes, DataIn>(
DataRes: AsRef<[u8]> + AsMut<[u8]>,
DataIn: AsRef<[u8]>,
{
let basek: usize = res.basek();
let mut lwe_2n: Vec<i64> = vec![0i64; lwe.n() + 1]; // TODO: from scratch space
let mut out_mut: GLWECiphertext<&mut [u8]> = res.to_mut();
let lwe_ref: LWECiphertext<&[u8]> = lwe.to_ref();
let two_n: usize = module.n() << 1;
let basek: usize = brk.basek();
let rows = brk.rows();
let cols: usize = out_mut.rank() + 1;
@@ -260,48 +268,59 @@ pub(crate) fn cggi_blind_rotate_block_binary<DataRes, DataIn>(
out_mut.data.zero();
// Initialize out to X^{b} * LUT(X)
module.vec_znx_rotate(-b, &mut out_mut.data, 0, &lut.data[0], 0);
module.vec_znx_rotate(b, &mut out_mut.data, 0, &lut.data[0], 0);
let block_size: usize = brk.block_size();
// ACC + [sum DFT(X^ai -1) * (DFT(ACC) x BRKi)]
let (mut acc_dft, scratch1) = scratch.tmp_fourier_glwe_ct(module, brk.basek(), out_mut.k(), out_mut.rank());
let (mut acc_add_dft, scratch2) = scratch1.tmp_fourier_glwe_ct(module, brk.basek(), out_mut.k(), out_mut.rank());
let (mut vmp_res, scratch3) = scratch2.tmp_fourier_glwe_ct(module, basek, out_mut.k(), out_mut.rank());
let (mut acc_dft, scratch1) = scratch.tmp_vec_znx_dft(module, cols, rows);
let (mut vmp_res, scratch2) = scratch1.tmp_vec_znx_dft(module, cols, brk.size());
let (mut acc_add_dft, scratch3) = scratch2.tmp_vec_znx_dft(module, cols, brk.size());
let (mut xai_plus_y, scratch4) = scratch3.tmp_scalar_znx(module, 1);
let (mut xai_plus_y_dft, scratch5) = scratch4.tmp_scalar_znx_dft(module, 1);
let start: Instant = Instant::now();
izip!(
a.chunks_exact(block_size),
brk.data.chunks_exact(block_size)
)
.for_each(|(ai, ski)| {
out_mut.dft(module, &mut acc_dft);
acc_add_dft.data.zero();
(0..cols).for_each(|j| {
module.vec_znx_dft(1, 0, &mut acc_dft, j, &out_mut.data, j);
});
acc_add_dft.zero();
izip!(ai.iter(), ski.iter()).for_each(|(aii, skii)| {
let ai_pos: usize = ((aii + two_n as i64) % two_n as i64) as usize;
// vmp_res = DFT(acc) * BRK[i]
module.vmp_apply(&mut vmp_res.data, &acc_dft.data, &skii.data, scratch5);
module.vmp_apply(&mut vmp_res, &acc_dft, &skii.data, scratch5);
// DFT(X^ai -1)
set_xai_plus_y(module, *aii, -1, &mut xai_plus_y_dft, &mut xai_plus_y);
set_xai_plus_y(module, ai_pos, -1, &mut xai_plus_y_dft, &mut xai_plus_y);
// DFT(X^ai -1) * (DFT(acc) * BRK[i])
(0..cols).for_each(|i| {
module.svp_apply_inplace(&mut vmp_res.data, i, &xai_plus_y_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft.data, i, &vmp_res.data, i);
module.svp_apply_inplace(&mut vmp_res, i, &xai_plus_y_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft, i, &vmp_res, i);
});
});
(0..cols).for_each(|i| {
module.vec_znx_dft_add_inplace(&mut acc_dft.data, i, &acc_add_dft.data, i);
module.vec_znx_dft_add_inplace(&mut acc_dft, i, &acc_add_dft, i);
});
acc_dft.idft(module, &mut out_mut, scratch5);
{
let (mut acc_add_big, scratch6) = scratch5.tmp_vec_znx_big(module, 1, brk.size());
(0..cols).for_each(|i| {
module.vec_znx_idft(&mut acc_add_big, 0, &acc_add_dft, i, scratch6);
module.vec_znx_big_add_small_inplace(&mut acc_add_big, 0, &out_mut.data, i);
module.vec_znx_big_normalize(basek, &mut out_mut.data, i, &acc_add_big, 0, scratch6);
});
}
});
let duration: std::time::Duration = start.elapsed();
}
pub(crate) fn negate_and_mod_switch_2n(n: usize, res: &mut [i64], lwe: &LWECiphertext<&[u8]>) {
@@ -315,7 +334,7 @@ pub(crate) fn negate_and_mod_switch_2n(n: usize, res: &mut [i64], lwe: &LWECiphe
if basek > log2n {
let diff: usize = basek - log2n;
res.iter_mut().for_each(|x| {
*x = div_signed_by_pow2(x, diff);
*x = div_ceil_signed_by_pow2(x, diff);
})
} else {
let rem: usize = basek - (log2n % basek);
@@ -336,7 +355,22 @@ pub(crate) fn negate_and_mod_switch_2n(n: usize, res: &mut [i64], lwe: &LWECiphe
}
#[inline(always)]
fn div_signed_by_pow2(x: &i64, k: usize) -> i64 {
fn div_round_by_pow2(x: &i64, k: usize) -> i64 {
if x >= &0 {
(x + (1 << (k - 1))) >> k
} else {
(x + (-1 << (k - 1))) >> k
}
}
// #[inline(always)]
// fn div_floor_signed_by_pow2(x: &i64, k: usize) -> i64{
// let bias: i64 = (1 << k) - 1;
// (x + ((x >> 63) & bias)) >> k
// }
#[inline(always)]
fn div_ceil_signed_by_pow2(x: &i64, k: usize) -> i64 {
let bias: i64 = (1 << k) - 1;
(x + ((x >> 63) & bias)) >> k
}