Files
poulpy/core/src/blind_rotation/ccgi.rs
2025-06-20 11:00:04 +02:00

365 lines
12 KiB
Rust

use std::time::Instant;
use backend::{
FFT64, MatZnxDftOps, MatZnxDftScratch, Module, ScalarZnxAlloc, ScalarZnxDftAlloc, ScalarZnxDftOps, Scratch, VecZnxDftOps,
VecZnxOps, ZnxView, ZnxViewMut, ZnxZero,
};
use itertools::izip;
use crate::{
FourierGLWECiphertext, GGSWCiphertext, GLWECiphertext, GLWECiphertextToMut, Infos, LWECiphertext, ScratchCore,
blind_rotation::{key::BlindRotationKeyCGGI, lut::LookUpTable},
lwe::ciphertext::LWECiphertextToRef,
};
pub fn cggi_blind_rotate_scratch_space(
module: &Module<FFT64>,
basek: usize,
k_lut: usize,
k_brk: usize,
rows: usize,
rank: usize,
) -> usize {
let size = k_brk.div_ceil(basek);
GGSWCiphertext::<Vec<u8>, FFT64>::bytes_of(module, basek, k_brk, rows, 1, rank)
+ (module.mat_znx_dft_mul_x_pow_minus_one_scratch_space(size, rank + 1)
| GLWECiphertext::external_product_inplace_scratch_space(module, basek, k_lut, k_brk, 1, rank))
}
pub fn cggi_blind_rotate<DataRes, DataIn>(
module: &Module<FFT64>,
res: &mut GLWECiphertext<DataRes>,
lwe: &LWECiphertext<DataIn>,
lut: &LookUpTable,
brk: &BlindRotationKeyCGGI<FFT64>,
scratch: &mut Scratch,
) where
DataRes: AsRef<[u8]> + AsMut<[u8]>,
DataIn: AsRef<[u8]>,
{
if lut.data.len() > 1 {
cggi_blind_rotate_block_binary_exnteded(module, res, lwe, lut, brk, scratch);
} else if brk.block_size() > 1 {
cggi_blind_rotate_block_binary(module, res, lwe, lut, brk, scratch);
} else {
todo!("implement this case")
}
}
pub(crate) fn cggi_blind_rotate_block_binary_exnteded<DataRes, DataIn>(
module: &Module<FFT64>,
res: &mut GLWECiphertext<DataRes>,
lwe: &LWECiphertext<DataIn>,
lut: &LookUpTable,
brk: &BlindRotationKeyCGGI<FFT64>,
scratch: &mut Scratch,
) where
DataRes: AsRef<[u8]> + AsMut<[u8]>,
DataIn: AsRef<[u8]>,
{
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 basek: usize = out_mut.basek();
let cols: usize = out_mut.rank() + 1;
mod_switch_2n(
2 * module.n() * lut.extension_factor(),
&mut lwe_2n,
&lwe_ref,
);
let extension_factor: i64 = lut.extension_factor() as i64;
let mut acc: Vec<GLWECiphertext<Vec<u8>>> = Vec::with_capacity(lut.extension_factor());
for _ in 0..extension_factor {
acc.push(GLWECiphertext::alloc(
module,
basek,
out_mut.k(),
out_mut.rank(),
));
}
let a: &[i64] = &lwe_2n[1..];
let b: i64 = lwe_2n[0];
let b_inner: i64 = b / extension_factor;
let b_outer: i64 = b % extension_factor;
for (i, j) in (0..b_outer).zip(extension_factor - b_outer..extension_factor) {
module.vec_znx_rotate(
b_inner + 1,
&mut acc[j as usize].data,
0,
&lut.data[i as usize],
0,
);
}
for (i, j) in (b_outer..extension_factor).zip(0..extension_factor - b_outer) {
module.vec_znx_rotate(
b_inner,
&mut acc[j as usize].data,
0,
&lut.data[i as usize],
0,
);
}
let block_size: usize = brk.block_size();
let mut acc_dft: Vec<FourierGLWECiphertext<Vec<u8>, FFT64>> = Vec::with_capacity(lut.extension_factor());
for _ in 0..extension_factor {
acc_dft.push(FourierGLWECiphertext::alloc(
module,
basek,
out_mut.k(),
out_mut.rank(),
));
}
let mut vmp_res: Vec<FourierGLWECiphertext<Vec<u8>, FFT64>> = Vec::with_capacity(lut.extension_factor());
for _ in 0..extension_factor {
vmp_res.push(FourierGLWECiphertext::alloc(
module,
basek,
out_mut.k(),
out_mut.rank(),
));
}
let mut acc_add_dft: Vec<FourierGLWECiphertext<Vec<u8>, FFT64>> = Vec::with_capacity(lut.extension_factor());
for _ in 0..extension_factor {
acc_add_dft.push(FourierGLWECiphertext::alloc(
module,
basek,
out_mut.k(),
out_mut.rank(),
));
}
let mut xai_minus_one: backend::ScalarZnx<Vec<u8>> = module.new_scalar_znx(1);
let mut xai_minus_one_dft: backend::ScalarZnxDft<Vec<u8>, FFT64> = module.new_scalar_znx_dft(1);
izip!(
a.chunks_exact(block_size),
brk.data.chunks_exact(block_size)
)
.enumerate()
.for_each(|(i, (ai, ski))| {
(0..lut.extension_factor()).for_each(|i| {
acc[i].dft(module, &mut acc_dft[i]);
acc_add_dft[i].data.zero();
});
izip!(ai.iter(), ski.iter()).for_each(|(aii, skii)| {
let aii_inner: i64 = aii / extension_factor;
let aii_outer: i64 = aii % extension_factor;
// vmp_res = DFT(acc) * BRK[i]
(0..lut.extension_factor()).for_each(|i| {
module.vmp_apply(&mut vmp_res[i].data, &acc_dft[i].data, &skii.data, scratch);
});
if aii_outer == 0 {
xai_minus_one.zero();
xai_minus_one.at_mut(0, 0)[0] = 1;
module.vec_znx_rotate_inplace(aii_inner, &mut xai_minus_one, 0);
xai_minus_one.at_mut(0, 0)[0] -= 1;
module.svp_prepare(&mut xai_minus_one_dft, 0, &xai_minus_one, 0);
(0..lut.extension_factor()).for_each(|j| {
(0..cols).for_each(|i| {
module.svp_apply_inplace(&mut vmp_res[j].data, i, &xai_minus_one_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft[j].data, i, &vmp_res[j].data, i);
});
})
} else {
xai_minus_one.zero();
xai_minus_one.at_mut(0, 0)[0] = 1;
module.vec_znx_rotate_inplace(aii_inner + 1, &mut xai_minus_one, 0);
xai_minus_one.at_mut(0, 0)[0] -= 1;
module.svp_prepare(&mut xai_minus_one_dft, 0, &xai_minus_one, 0);
for (i, j) in (0..aii_outer).zip(extension_factor - aii_outer..extension_factor) {
module.vec_znx_rotate(
b_inner + 1,
&mut acc[j as usize].data,
0,
&lut.data[i as usize],
0,
);
(0..cols).for_each(|k| {
module.svp_apply_inplace(&mut vmp_res[i as usize].data, k, &xai_minus_one_dft, 0);
module.vec_znx_dft_add_inplace(
&mut acc_add_dft[j as usize].data,
k,
&vmp_res[i as usize].data,
k,
);
});
}
xai_minus_one.zero();
xai_minus_one.at_mut(0, 0)[0] = 1;
module.vec_znx_rotate_inplace(aii_inner, &mut xai_minus_one, 0);
xai_minus_one.at_mut(0, 0)[0] -= 1;
module.svp_prepare(&mut xai_minus_one_dft, 0, &xai_minus_one, 0);
for (i, j) in (aii_outer..extension_factor).zip(0..extension_factor - aii_outer) {
module.vec_znx_rotate(
b_inner,
&mut acc[j as usize].data,
0,
&lut.data[i as usize],
0,
);
(0..cols).for_each(|k| {
module.svp_apply_inplace(&mut vmp_res[i as usize].data, k, &xai_minus_one_dft, 0);
module.vec_znx_dft_add_inplace(
&mut acc_add_dft[j as usize].data,
k,
&vmp_res[i as usize].data,
k,
);
});
}
}
});
if i == lwe.n() - block_size {
(0..cols).for_each(|i| {
module.vec_znx_dft_add_inplace(&mut acc_dft[0].data, i, &acc_add_dft[0].data, i);
});
acc_dft[0].idft(module, &mut out_mut, scratch);
} else {
(0..lut.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);
});
acc_dft[j].idft(module, &mut acc[j], scratch);
})
}
});
}
pub(crate) fn cggi_blind_rotate_block_binary<DataRes, DataIn>(
module: &Module<FFT64>,
res: &mut GLWECiphertext<DataRes>,
lwe: &LWECiphertext<DataIn>,
lut: &LookUpTable,
brk: &BlindRotationKeyCGGI<FFT64>,
scratch: &mut Scratch,
) where
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 cols: usize = out_mut.rank() + 1;
mod_switch_2n(2 * module.n(), &mut lwe_2n, &lwe_ref);
let a: &[i64] = &lwe_2n[1..];
let b: i64 = lwe_2n[0];
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);
let block_size: usize = brk.block_size();
// ACC + [sum DFT(X^ai -1) * (DFT(ACC) x BRKi)]
let (mut acc_dft, scratch1) = scratch.tmp_glwe_fourier(module, brk.basek(), out_mut.k(), out_mut.rank());
let (mut acc_add_dft, scratch2) = scratch1.tmp_glwe_fourier(module, brk.basek(), out_mut.k(), out_mut.rank());
let (mut vmp_res, scratch3) = scratch2.tmp_glwe_fourier(module, basek, out_mut.k(), out_mut.rank());
let (mut xai_minus_one, scratch4) = scratch3.tmp_scalar_znx(module, 1);
let (mut xai_minus_one_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();
izip!(ai.iter(), ski.iter()).for_each(|(aii, skii)| {
// vmp_res = DFT(acc) * BRK[i]
module.vmp_apply(&mut vmp_res.data, &acc_dft.data, &skii.data, scratch5);
// DFT(X^ai -1)
xai_minus_one.zero();
xai_minus_one.at_mut(0, 0)[0] = 1;
module.vec_znx_rotate_inplace(*aii, &mut xai_minus_one, 0);
xai_minus_one.at_mut(0, 0)[0] -= 1;
module.svp_prepare(&mut xai_minus_one_dft, 0, &xai_minus_one, 0);
// DFT(X^ai -1) * (DFT(acc) * BRK[i])
(0..cols).for_each(|i| {
module.svp_apply_inplace(&mut vmp_res.data, i, &xai_minus_one_dft, 0);
module.vec_znx_dft_add_inplace(&mut acc_add_dft.data, i, &vmp_res.data, i);
});
});
(0..cols).for_each(|i| {
module.vec_znx_dft_add_inplace(&mut acc_dft.data, i, &acc_add_dft.data, i);
});
acc_dft.idft(module, &mut out_mut, scratch5);
});
let duration: std::time::Duration = start.elapsed();
println!("external products: {} us", duration.as_micros());
}
pub(crate) fn mod_switch_2n(n: usize, res: &mut [i64], lwe: &LWECiphertext<&[u8]>) {
let basek: usize = lwe.basek();
let log2n: usize = usize::BITS as usize - (n - 1).leading_zeros() as usize + 1;
res.copy_from_slice(&lwe.data.at(0, 0));
if basek > log2n {
let diff: usize = basek - log2n;
res.iter_mut().for_each(|x| {
*x = div_signed_by_pow2(x, diff);
})
} else {
let rem: usize = basek - (log2n % basek);
let size: usize = log2n.div_ceil(basek);
(1..size).for_each(|i| {
if i == size - 1 && rem != basek {
let k_rem: usize = basek - rem;
izip!(lwe.data.at(0, i).iter(), res.iter_mut()).for_each(|(x, y)| {
*y = (*y << k_rem) + (x >> rem);
});
} else {
izip!(lwe.data.at(0, i).iter(), res.iter_mut()).for_each(|(x, y)| {
*y = (*y << basek) + x;
});
}
})
}
}
#[inline(always)]
fn div_signed_by_pow2(x: &i64, k: usize) -> i64 {
let bias: i64 = (1 << k) - 1;
(x + ((x >> 63) & bias)) >> k
}