github/concrete
1
1
Fork 0
mirror of https://github.com/zama-ai/concrete.git synced 2026-02-08 19:44:57 -05:00
Code Issues Packages Projects Releases Wiki Activity
Files
5c1a15c514b8d293066f2b19500389083473bd53
concrete/compilers/concrete-compiler/compiler/lib/Runtime/wrappers.cpp

794 lines
34 KiB
C++
Raw Blame History

// Part of the Concrete Compiler Project, under the BSD3 License with Zama
// Exceptions. See
// https://github.com/zama-ai/concrete-compiler-internal/blob/main/LICENSE.txt
// for license information.
#include "concretelang/Runtime/wrappers.h"
#include "concrete-cpu.h"
#include "concretelang/Common/Error.h"
#include <assert.h>
#include <bitset>
#include <cmath>
#include <functional>
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <vector>
#include "concretelang/ClientLib/CRT.h"
#include "concretelang/Runtime/wrappers.h"
#ifdef CONCRETELANG_CUDA_SUPPORT
// CUDA memory utils function /////////////////////////////////////////////////
void *memcpy_async_bsk_to_gpu(mlir::concretelang::RuntimeContext *context,
uint32_t input_lwe_dim, uint32_t poly_size,
uint32_t level, uint32_t glwe_dim,
uint32_t gpu_idx, void *stream) {
return context->get_bsk_gpu(input_lwe_dim, poly_size, level, glwe_dim,
gpu_idx, stream);
}
void *memcpy_async_ksk_to_gpu(mlir::concretelang::RuntimeContext *context,
uint32_t level, uint32_t input_lwe_dim,
uint32_t output_lwe_dim, uint32_t gpu_idx,
void *stream) {
return context->get_ksk_gpu(level, input_lwe_dim, output_lwe_dim, gpu_idx,
stream);
}
void *alloc_and_memcpy_async_to_gpu(uint64_t *buf_ptr, uint64_t buf_offset,
uint64_t buf_size, uint32_t gpu_idx,
void *stream) {
size_t buf_size_ = buf_size * sizeof(uint64_t);
void *ct_gpu = cuda_malloc_async(buf_size_, (cudaStream_t *)stream, gpu_idx);
cuda_memcpy_async_to_gpu(ct_gpu, buf_ptr + buf_offset, buf_size_,
(cudaStream_t *)stream, gpu_idx);
return ct_gpu;
}
void memcpy_async_to_cpu(uint64_t *buf_ptr, uint64_t buf_offset,
uint64_t buf_size, void *buf_gpu, uint32_t gpu_idx,
void *stream) {
cuda_memcpy_async_to_cpu(buf_ptr + buf_offset, buf_gpu,
buf_size * sizeof(uint64_t), (cudaStream_t *)stream,
gpu_idx);
}
void free_from_gpu(void *gpu_ptr, uint32_t gpu_idx = 0) {
cuda_drop(gpu_ptr, gpu_idx);
}
// Single ciphertext CUDA functions ///////////////////////////////////////////
void memref_keyswitch_lwe_cuda_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size, uint64_t out_stride, uint64_t *ct0_allocated,
uint64_t *ct0_aligned, uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride, uint32_t level, uint32_t base_log,
uint32_t input_lwe_dim, uint32_t output_lwe_dim,
mlir::concretelang::RuntimeContext *context) {
assert(out_stride == 1);
assert(ct0_stride == 1);
memref_batched_keyswitch_lwe_cuda_u64(
// Output 1D memref as 2D memref
out_allocated, out_aligned, out_offset, 1, out_size, out_size, out_stride,
// Output 1D memref as 2D memref
ct0_allocated, ct0_aligned, ct0_offset, 1, ct0_size, ct0_size, ct0_stride,
// Keyswitch additional arguments
level, base_log, input_lwe_dim, output_lwe_dim, context);
}
void memref_bootstrap_lwe_cuda_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size, uint64_t out_stride, uint64_t *ct0_allocated,
uint64_t *ct0_aligned, uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride, uint64_t *tlu_allocated, uint64_t *tlu_aligned,
uint64_t tlu_offset, uint64_t tlu_size, uint64_t tlu_stride,
uint32_t input_lwe_dim, uint32_t poly_size, uint32_t level,
uint32_t base_log, uint32_t glwe_dim,
mlir::concretelang::RuntimeContext *context) {
memref_batched_bootstrap_lwe_cuda_u64(
// Output 1D memref as 2D memref
out_allocated, out_aligned, out_offset, 1, out_size, out_size, out_stride,
// Input 1D memref as 2D memref
ct0_allocated, ct0_aligned, ct0_offset, 1, ct0_size, ct0_size, ct0_stride,
// Table lookup memref
tlu_allocated, tlu_aligned, tlu_offset, tlu_size, tlu_stride,
// Bootstrap additional arguments
input_lwe_dim, poly_size, level, base_log, glwe_dim, context);
}
// Batched CUDA function //////////////////////////////////////////////////////
void memref_batched_keyswitch_lwe_cuda_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size0, uint64_t out_size1, uint64_t out_stride0,
uint64_t out_stride1, uint64_t *ct0_allocated, uint64_t *ct0_aligned,
uint64_t ct0_offset, uint64_t ct0_size0, uint64_t ct0_size1,
uint64_t ct0_stride0, uint64_t ct0_stride1, uint32_t level,
uint32_t base_log, uint32_t input_lwe_dim, uint32_t output_lwe_dim,
mlir::concretelang::RuntimeContext *context) {
assert(out_size0 == ct0_size0);
assert(out_size1 == output_lwe_dim + 1);
assert(ct0_size1 == input_lwe_dim + 1);
// TODO: Multi GPU
uint32_t gpu_idx = 0;
uint32_t num_samples = out_size0;
uint64_t ct0_batch_size = ct0_size0 * ct0_size1;
uint64_t out_batch_size = out_size0 * out_size1;
// Create the cuda stream
// TODO: Should be created by the compiler codegen
void *stream = cuda_create_stream(gpu_idx);
// Get the pointer on the keyswitching key on the GPU
void *ksk_gpu = memcpy_async_ksk_to_gpu(context, level, input_lwe_dim,
output_lwe_dim, gpu_idx, stream);
// Move the input and output batch of ciphertexts to the GPU
// TODO: The allocation should be done by the compiler codegen
void *ct0_gpu = alloc_and_memcpy_async_to_gpu(
ct0_aligned, ct0_offset, ct0_batch_size, gpu_idx, (cudaStream_t *)stream);
void *out_gpu = cuda_malloc_async(out_batch_size * sizeof(uint64_t),
(cudaStream_t *)stream, gpu_idx);
// Run the keyswitch kernel on the GPU
cuda_keyswitch_lwe_ciphertext_vector_64(
stream, gpu_idx, out_gpu, ct0_gpu, ksk_gpu, input_lwe_dim, output_lwe_dim,
base_log, level, num_samples);
// Copy the output batch of ciphertext back to CPU
memcpy_async_to_cpu(out_aligned, out_offset, out_batch_size, out_gpu, gpu_idx,
stream);
cuda_synchronize_device(gpu_idx);
// free memory that we allocated on gpu
cuda_drop(ct0_gpu, gpu_idx);
cuda_drop(out_gpu, gpu_idx);
cuda_destroy_stream((cudaStream_t *)stream, gpu_idx);
}
void memref_batched_bootstrap_lwe_cuda_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size0, uint64_t out_size1, uint64_t out_stride0,
uint64_t out_stride1, uint64_t *ct0_allocated, uint64_t *ct0_aligned,
uint64_t ct0_offset, uint64_t ct0_size0, uint64_t ct0_size1,
uint64_t ct0_stride0, uint64_t ct0_stride1, uint64_t *tlu_allocated,
uint64_t *tlu_aligned, uint64_t tlu_offset, uint64_t tlu_size,
uint64_t tlu_stride, uint32_t input_lwe_dim, uint32_t poly_size,
uint32_t level, uint32_t base_log, uint32_t glwe_dim,
mlir::concretelang::RuntimeContext *context) {
assert(out_size0 == ct0_size0);
assert(out_size1 == glwe_dim * poly_size + 1);
// TODO: Multi GPU
uint32_t gpu_idx = 0;
uint32_t num_samples = out_size0;
uint64_t ct0_batch_size = ct0_size0 * ct0_size1;
uint64_t out_batch_size = out_size0 * out_size1;
int8_t *pbs_buffer = nullptr;
// Create the cuda stream
// TODO: Should be created by the compiler codegen
void *stream = cuda_create_stream(gpu_idx);
// Get the pointer on the bootstraping key on the GPU
void *fbsk_gpu = memcpy_async_bsk_to_gpu(context, input_lwe_dim, poly_size,
level, glwe_dim, gpu_idx, stream);
// Move the input and output batch of ciphertext to the GPU
// TODO: The allocation should be done by the compiler codegen
void *ct0_gpu = alloc_and_memcpy_async_to_gpu(
ct0_aligned, ct0_offset, ct0_batch_size, gpu_idx, (cudaStream_t *)stream);
void *out_gpu = cuda_malloc_async(out_batch_size * sizeof(uint64_t),
(cudaStream_t *)stream, gpu_idx);
// Construct the glwe accumulator (on CPU)
// TODO: Should be done outside of the bootstrap call, compile time if
// possible. Refactor in progress
uint64_t glwe_ct_size = poly_size * (glwe_dim + 1);
uint64_t *glwe_ct = (uint64_t *)malloc(glwe_ct_size * sizeof(uint64_t));
auto tlu = tlu_aligned + tlu_offset;
// Glwe trivial encryption
for (size_t i = 0; i < poly_size * glwe_dim; i++) {
glwe_ct[i] = 0;
}
for (size_t i = 0; i < poly_size; i++) {
glwe_ct[poly_size * glwe_dim + i] = tlu[i];
}
// Move the glwe accumulator to the GPU
void *glwe_ct_gpu = alloc_and_memcpy_async_to_gpu(
glwe_ct, 0, glwe_ct_size, gpu_idx, (cudaStream_t *)stream);
// Move test vector indexes to the GPU, the test vector indexes is set of 0
uint32_t num_test_vectors = 1, lwe_idx = 0,
test_vector_idxes_size = num_samples * sizeof(uint64_t);
void *test_vector_idxes = malloc(test_vector_idxes_size);
memset(test_vector_idxes, 0, test_vector_idxes_size);
void *test_vector_idxes_gpu = cuda_malloc_async(
test_vector_idxes_size, (cudaStream_t *)stream, gpu_idx);
cuda_memcpy_async_to_gpu(test_vector_idxes_gpu, test_vector_idxes,
test_vector_idxes_size, (cudaStream_t *)stream,
gpu_idx);
// Allocate PBS buffer on GPU
scratch_cuda_bootstrap_amortized_64(
stream, gpu_idx, &pbs_buffer, glwe_dim, poly_size, num_samples,
cuda_get_max_shared_memory(gpu_idx), true);
// Run the bootstrap kernel on the GPU
cuda_bootstrap_amortized_lwe_ciphertext_vector_64(
stream, gpu_idx, out_gpu, glwe_ct_gpu, test_vector_idxes_gpu, ct0_gpu,
fbsk_gpu, pbs_buffer, input_lwe_dim, glwe_dim, poly_size, base_log, level,
num_samples, num_test_vectors, lwe_idx,
cuda_get_max_shared_memory(gpu_idx));
cleanup_cuda_bootstrap_amortized(stream, gpu_idx, &pbs_buffer);
// Copy the output batch of ciphertext back to CPU
memcpy_async_to_cpu(out_aligned, out_offset, out_batch_size, out_gpu, gpu_idx,
stream);
// free memory that we allocated on gpu
cuda_drop_async(ct0_gpu, (cudaStream_t *)stream, gpu_idx);
cuda_drop_async(out_gpu, (cudaStream_t *)stream, gpu_idx);
cuda_drop_async(glwe_ct_gpu, (cudaStream_t *)stream, gpu_idx);
cuda_drop_async(test_vector_idxes_gpu, (cudaStream_t *)stream, gpu_idx);
cudaStreamSynchronize(*(cudaStream_t *)stream);
// Free the glwe accumulator (on CPU)
free(glwe_ct);
cuda_destroy_stream((cudaStream_t *)stream, gpu_idx);
}
#endif
void memref_encode_plaintext_with_crt(
uint64_t *output_allocated, uint64_t *output_aligned,
uint64_t output_offset, uint64_t output_size, uint64_t output_stride,
uint64_t input, uint64_t *mods_allocated, uint64_t *mods_aligned,
uint64_t mods_offset, uint64_t mods_size, uint64_t mods_stride,
uint64_t mods_product) {
assert(output_stride == 1 && "Runtime: stride not equal to 1, check "
"memref_encode_plaintext_with_crt");
assert(mods_stride == 1 && "Runtime: stride not equal to 1, check "
"memref_encode_plaintext_with_crt");
for (size_t i = 0; i < (size_t)mods_size; ++i) {
output_aligned[output_offset + i] =
encode_crt(input, mods_aligned[mods_offset + i], mods_product);
}
return;
}
void memref_encode_expand_lut_for_bootstrap(
uint64_t *output_lut_allocated, uint64_t *output_lut_aligned,
uint64_t output_lut_offset, uint64_t output_lut_size,
uint64_t output_lut_stride, uint64_t *input_lut_allocated,
uint64_t *input_lut_aligned, uint64_t input_lut_offset,
uint64_t input_lut_size, uint64_t input_lut_stride, uint32_t poly_size,
uint32_t out_MESSAGE_BITS, bool is_signed) {
assert(input_lut_stride == 1 && "Runtime: stride not equal to 1, check "
"memref_encode_expand_lut_bootstrap");
assert(output_lut_stride == 1 && "Runtime: stride not equal to 1, check "
"memref_encode_expand_lut_bootstrap");
size_t mega_case_size = output_lut_size / input_lut_size;
assert((mega_case_size % 2) == 0);
// When the bootstrap is executed on encrypted signed integers, the lut must
// be half-rotated. This map takes care about properly indexing into the input
// lut depending on what bootstrap gets executed.
std::function<size_t(size_t)> indexMap;
if (is_signed) {
size_t halfInputSize = input_lut_size / 2;
indexMap = [=](size_t idx) {
if (idx < halfInputSize) {
return idx + halfInputSize;
} else {
return idx - halfInputSize;
}
};
} else {
indexMap = [=](size_t idx) { return idx; };
}
// The first lut value should be centered over zero. This means that half of
// it should appear at the beginning of the output lut, and half of it at the
// end (but negated).
for (size_t idx = 0; idx < mega_case_size / 2; ++idx) {
output_lut_aligned[output_lut_offset + idx] =
input_lut_aligned[input_lut_offset + indexMap(0)]
<< (64 - out_MESSAGE_BITS - 1);
}
for (size_t idx = (input_lut_size - 1) * mega_case_size + mega_case_size / 2;
idx < output_lut_size; ++idx) {
output_lut_aligned[output_lut_offset + idx] =
-(input_lut_aligned[input_lut_offset + indexMap(0)]
<< (64 - out_MESSAGE_BITS - 1));
}
// Treats the other ut values.
for (size_t lut_idx = 1; lut_idx < input_lut_size; ++lut_idx) {
uint64_t lut_value = input_lut_aligned[input_lut_offset + indexMap(lut_idx)]
<< (64 - out_MESSAGE_BITS - 1);
size_t start = mega_case_size * (lut_idx - 1) + mega_case_size / 2;
for (size_t output_idx = start; output_idx < start + mega_case_size;
++output_idx) {
output_lut_aligned[output_lut_offset + output_idx] = lut_value;
}
}
return;
}
void memref_encode_lut_for_crt_woppbs(
// Output encoded/expanded lut
uint64_t *output_lut_allocated, uint64_t *output_lut_aligned,
uint64_t output_lut_offset, uint64_t output_lut_size0,
uint64_t output_lut_size1, uint64_t output_lut_stride0,
uint64_t output_lut_stride1,
// Input lut
uint64_t *input_lut_allocated, uint64_t *input_lut_aligned,
uint64_t input_lut_offset, uint64_t input_lut_size,
uint64_t input_lut_stride,
// Crt coprimes
uint64_t *crt_decomposition_allocated, uint64_t *crt_decomposition_aligned,
uint64_t crt_decomposition_offset, uint64_t crt_decomposition_size,
uint64_t crt_decomposition_stride,
// Crt number of bits
uint64_t *crt_bits_allocated, uint64_t *crt_bits_aligned,
uint64_t crt_bits_offset, uint64_t crt_bits_size, uint64_t crt_bits_stride,
// Crypto parameters
uint32_t modulus_product, bool is_signed) {
assert(input_lut_stride == 1 && "Runtime: stride not equal to 1, check "
"memref_encode_lut_woppbs");
assert(output_lut_stride0 == output_lut_size1 &&
"Runtime: out dim stride not equal to in_dim size, check "
"memref_encode_lut_woppbs");
assert(output_lut_stride1 == 1 && "Runtime: stride not equal to 1, check "
"memref_encode_lut_woppbs");
assert(modulus_product >= input_lut_size);
// Initialize lut cases not supposed to be reached
for (uint64_t i = 0; i < output_lut_size0 * output_lut_size1; i++) {
output_lut_aligned[output_lut_offset + i] = 0;
}
// When the woppbs is executed on encrypted signed integers, the index of the
// lut elements must be adapted to fit the way signed are encrypted in CRT
// (to ensure the lookup falls into the proper case).
// This map takes care about properly indexing into the output lut depending
// on what bootstrap gets executed.
std::function<uint64_t(uint64_t)> indexMap;
if (!is_signed) {
// When not signed, the integer values are encoded in increasing order. That
// is (example of 9 bits values, using crt decomposition [5,7,16]):
//
// |0 511|
// |---------|
// |0 511|
//
// is encoded as
//
// |0 511| INVALID |
// |-------|-----------|
// |0 511|512 559|
//
// Where on top are represented the semantic values, and below, the actual
// encoding of values, either on uint64_t or as increasing crt values.
//
// As a consequence, there is nothing particular to do to map the index of
// the input lut to an index of the output lut.
indexMap = [=](uint64_t plaintext) { return plaintext; };
} else {
// When signed, the integer values are encoded in a way that resembles 2s
// complement. That is (example of 9 bits values, using crt decomposition
// [5,7,16]):
//
// |0 255|-256 -1|
// |---------|----------|
// |0 255|256 511|
//
// is encoded as
//
// |0 255| INVALID |-256 -1|
// |---------|-------------|----------|
// |0 255|256 303|304 559|
//
// Where on top are represented the semantic values, and below, the actual
// encoding of values, either on uint64_t or as increasing crt values.
//
// As a consequence, to map the index of the input lut to an index of the
// output lut we must take care of crossing the invalid range in between
// positive values and negative values.
indexMap = [=](uint64_t plaintext) {
if (plaintext >= (input_lut_size / 2)) {
plaintext += modulus_product - input_lut_size;
}
return plaintext;
};
}
uint64_t log_lut_crt_size = 0;
for (size_t in_block = 0; in_block < crt_decomposition_size; in_block++) {
auto bits_count = crt_bits_aligned[crt_bits_offset + in_block];
log_lut_crt_size += bits_count;
}
uint64_t lut_crt_size = 1 << log_lut_crt_size;
assert(lut_crt_size == output_lut_size1);
assert(crt_decomposition_size == output_lut_size0);
for (uint64_t in_index = 0; in_index < input_lut_size; in_index++) {
uint64_t out_index = 0;
{
uint64_t total_bit_count = 0;
for (size_t in_block = 0; in_block < crt_decomposition_size; in_block++) {
auto in_base =
crt_decomposition_aligned[crt_decomposition_offset + in_block];
auto bits_count = crt_bits_aligned[crt_bits_offset + in_block];
out_index += (((indexMap(in_index) % in_base) << bits_count) / in_base)
<< total_bit_count;
total_bit_count += bits_count;
}
}
for (size_t out_block = 0; out_block < crt_decomposition_size;
out_block++) {
auto out_base =
crt_decomposition_aligned[crt_decomposition_offset + out_block];
auto v = encode_crt(input_lut_aligned[input_lut_offset + in_index],
out_base, modulus_product);
output_lut_aligned[output_lut_offset + out_block * lut_crt_size +
out_index] = v;
}
}
}
void memref_add_lwe_ciphertexts_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size, uint64_t out_stride, uint64_t *ct0_allocated,
uint64_t *ct0_aligned, uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride, uint64_t *ct1_allocated, uint64_t *ct1_aligned,
uint64_t ct1_offset, uint64_t ct1_size, uint64_t ct1_stride) {
assert(out_size == ct0_size && out_size == ct1_size &&
"size of lwe buffer are incompatible");
size_t lwe_dimension = out_size - 1;
concrete_cpu_add_lwe_ciphertext_u64(out_aligned + out_offset,
ct0_aligned + ct0_offset,
ct1_aligned + ct1_offset, lwe_dimension);
}
void memref_add_plaintext_lwe_ciphertext_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size, uint64_t out_stride, uint64_t *ct0_allocated,
uint64_t *ct0_aligned, uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride, uint64_t plaintext) {
assert(out_size == ct0_size && "size of lwe buffer are incompatible");
size_t lwe_dimension = out_size - 1;
concrete_cpu_add_plaintext_lwe_ciphertext_u64(out_aligned + out_offset,
ct0_aligned + ct0_offset,
plaintext, lwe_dimension);
}
void memref_mul_cleartext_lwe_ciphertext_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size, uint64_t out_stride, uint64_t *ct0_allocated,
uint64_t *ct0_aligned, uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride, uint64_t cleartext) {
assert(out_size == ct0_size && "size of lwe buffer are incompatible");
size_t lwe_dimension = out_size - 1;
concrete_cpu_mul_cleartext_lwe_ciphertext_u64(out_aligned + out_offset,
ct0_aligned + ct0_offset,
cleartext, lwe_dimension);
}
void memref_negate_lwe_ciphertext_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size, uint64_t out_stride, uint64_t *ct0_allocated,
uint64_t *ct0_aligned, uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride) {
assert(out_size == ct0_size && "size of lwe buffer are incompatible");
size_t lwe_dimension = {out_size - 1};
concrete_cpu_negate_lwe_ciphertext_u64(
out_aligned + out_offset, ct0_aligned + ct0_offset, lwe_dimension);
}
void memref_keyswitch_lwe_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size, uint64_t out_stride, uint64_t *ct0_allocated,
uint64_t *ct0_aligned, uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride, uint32_t decomposition_level_count,
uint32_t decomposition_base_log, uint32_t input_dimension,
uint32_t output_dimension, mlir::concretelang::RuntimeContext *context) {
assert(out_stride == 1 && ct0_stride == 1);
// Get keyswitch key - TODO Give a non hardcoded keyID
const uint64_t *keyswitch_key = context->keyswitch_key_buffer(0);
// Get stack parameter
concrete_cpu_keyswitch_lwe_ciphertext_u64(
out_aligned + out_offset, ct0_aligned + ct0_offset, keyswitch_key,
decomposition_level_count, decomposition_base_log, input_dimension,
output_dimension);
}
void memref_batched_keyswitch_lwe_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size0, uint64_t out_size1, uint64_t out_stride0,
uint64_t out_stride1, uint64_t *ct0_allocated, uint64_t *ct0_aligned,
uint64_t ct0_offset, uint64_t ct0_size0, uint64_t ct0_size1,
uint64_t ct0_stride0, uint64_t ct0_stride1, uint32_t level,
uint32_t base_log, uint32_t input_lwe_dim, uint32_t output_lwe_dim,
mlir::concretelang::RuntimeContext *context) {
for (size_t i = 0; i < ct0_size0; i++) {
memref_keyswitch_lwe_u64(
out_allocated + i * out_size1, out_aligned + i * out_size1, out_offset,
out_size1, out_stride1, ct0_allocated + i * ct0_size1,
ct0_aligned + i * ct0_size1, ct0_offset, ct0_size1, ct0_stride1, level,
base_log, input_lwe_dim, output_lwe_dim, context);
}
}
void memref_bootstrap_lwe_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size, uint64_t out_stride, uint64_t *ct0_allocated,
uint64_t *ct0_aligned, uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride, uint64_t *tlu_allocated, uint64_t *tlu_aligned,
uint64_t tlu_offset, uint64_t tlu_size, uint64_t tlu_stride,
uint32_t input_lwe_dimension, uint32_t polynomial_size,
uint32_t decomposition_level_count, uint32_t decomposition_base_log,
uint32_t glwe_dimension, mlir::concretelang::RuntimeContext *context) {
uint64_t glwe_ct_size = polynomial_size * (glwe_dimension + 1);
uint64_t *glwe_ct = (uint64_t *)malloc(glwe_ct_size * sizeof(uint64_t));
auto tlu = tlu_aligned + tlu_offset;
// Glwe trivial encryption
for (size_t i = 0; i < polynomial_size * glwe_dimension; i++) {
glwe_ct[i] = 0;
}
for (size_t i = 0; i < polynomial_size; i++) {
glwe_ct[polynomial_size * glwe_dimension + i] = tlu[i];
}
// Get fourrier bootstrap key - TODO Give a non hardcoded keyID
size_t keyId = 0;
const auto &fft = context->fft(keyId);
auto bootstrap_key = context->fourier_bootstrap_key_buffer(keyId);
// Get stack parameter
size_t scratch_size;
size_t scratch_align;
concrete_cpu_bootstrap_lwe_ciphertext_u64_scratch(
&scratch_size, &scratch_align, glwe_dimension, polynomial_size, fft);
// Allocate scratch
auto scratch = (uint8_t *)aligned_alloc(scratch_align, scratch_size);
// Bootstrap
concrete_cpu_bootstrap_lwe_ciphertext_u64(
out_aligned + out_offset, ct0_aligned + ct0_offset, glwe_ct,
bootstrap_key, decomposition_level_count, decomposition_base_log,
glwe_dimension, polynomial_size, input_lwe_dimension, fft, scratch,
scratch_size);
free(glwe_ct);
free(scratch);
}
void memref_batched_bootstrap_lwe_u64(
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size0, uint64_t out_size1, uint64_t out_stride0,
uint64_t out_stride1, uint64_t *ct0_allocated, uint64_t *ct0_aligned,
uint64_t ct0_offset, uint64_t ct0_size0, uint64_t ct0_size1,
uint64_t ct0_stride0, uint64_t ct0_stride1, uint64_t *tlu_allocated,
uint64_t *tlu_aligned, uint64_t tlu_offset, uint64_t tlu_size,
uint64_t tlu_stride, uint32_t input_lwe_dim, uint32_t poly_size,
uint32_t level, uint32_t base_log, uint32_t glwe_dim,
mlir::concretelang::RuntimeContext *context) {
for (size_t i = 0; i < out_size0; i++) {
memref_bootstrap_lwe_u64(
out_allocated + i * out_size1, out_aligned + i * out_size1, out_offset,
out_size1, out_stride1, ct0_allocated, ct0_aligned + i * ct0_size1,
ct0_offset, ct0_size1, ct0_stride1, tlu_allocated, tlu_aligned,
tlu_offset, tlu_size, tlu_stride, input_lwe_dim, poly_size, level,
base_log, glwe_dim, context);
}
}
uint64_t encode_crt(int64_t plaintext, uint64_t modulus, uint64_t product) {
return concretelang::clientlib::crt::encode(plaintext, modulus, product);
}
void memref_wop_pbs_crt_buffer(
// Output 2D memref
uint64_t *out_allocated, uint64_t *out_aligned, uint64_t out_offset,
uint64_t out_size_0, uint64_t out_size_1, uint64_t out_stride_0,
uint64_t out_stride_1,
// Input 2D memref
uint64_t *in_allocated, uint64_t *in_aligned, uint64_t in_offset,
uint64_t in_size_0, uint64_t in_size_1, uint64_t in_stride_0,
uint64_t in_stride_1,
// clear text lut 1D memref
uint64_t *lut_ct_allocated, uint64_t *lut_ct_aligned,
uint64_t lut_ct_offset, uint64_t lut_ct_size0, uint64_t lut_ct_size1,
uint64_t lut_ct_stride0, uint64_t lut_ct_stride1,
// CRT decomposition 1D memref
uint64_t *crt_decomp_allocated, uint64_t *crt_decomp_aligned,
uint64_t crt_decomp_offset, uint64_t crt_decomp_size,
uint64_t crt_decomp_stride,
// Additional crypto parameters
uint32_t lwe_small_size, uint32_t cbs_level_count, uint32_t cbs_base_log,
uint32_t ksk_level_count, uint32_t ksk_base_log, uint32_t bsk_level_count,
uint32_t bsk_base_log, uint32_t fpksk_level_count, uint32_t fpksk_base_log,
uint32_t polynomial_size,
// runtime context that hold evluation keys
mlir::concretelang::RuntimeContext *context) {
// The compiler should only generates 2D memref<BxS>, where B is the number of
// ciphertext block and S the lweSize.
// Check for the strides
assert(out_stride_1 == 1);
assert(in_stride_0 == in_size_1 && in_stride_0 == in_size_1);
// Check for the size B
assert(out_size_0 == in_size_0 && out_size_0 == crt_decomp_size);
// Check for the size S
assert(out_size_1 == in_size_1);
uint64_t lwe_small_dim = lwe_small_size - 1;
assert(out_size_1 == in_size_1);
uint64_t lwe_big_size = in_size_1;
uint64_t lwe_big_dim = lwe_big_size - 1;
assert(lwe_big_dim % polynomial_size == 0);
assert(lwe_big_dim % polynomial_size == 0);
uint64_t glwe_dim = lwe_big_dim / polynomial_size;
// Compute the numbers of bits to extract for each block and the total one.
uint64_t total_number_of_bits_per_block = 0;
auto number_of_bits_per_block = new uint64_t[crt_decomp_size]();
for (uint64_t i = 0; i < crt_decomp_size; i++) {
uint64_t modulus = crt_decomp_aligned[i + crt_decomp_offset];
uint64_t nb_bit_to_extract =
static_cast<uint64_t>(ceil(log2(static_cast<double>(modulus))));
number_of_bits_per_block[i] = nb_bit_to_extract;
total_number_of_bits_per_block += nb_bit_to_extract;
}
// Create the buffer of ciphertexts for storing the total number of bits to
// extract.
// The extracted bit should be in the following order:
//
// [msb(m%crt[n-1])..lsb(m%crt[n-1])...msb(m%crt[0])..lsb(m%crt[0])] where n
// is the size of the crt decomposition
auto extract_bits_output_buffer =
new uint64_t[lwe_small_size * total_number_of_bits_per_block]{0};
// We make a private copy to apply a subtraction on the body
auto first_ciphertext = in_aligned + in_offset;
auto copy_size = crt_decomp_size * lwe_big_size;
std::vector<uint64_t> in_copy(first_ciphertext, first_ciphertext + copy_size);
// Extraction of each bit for each block
size_t fftKeyId = 0;
const auto &fft = context->fft(fftKeyId);
size_t bskKeyId = 0;
auto bootstrap_key = context->fourier_bootstrap_key_buffer(bskKeyId);
size_t kskKeyId = 0;
auto keyswicth_key = context->keyswitch_key_buffer(kskKeyId);
for (int64_t i = crt_decomp_size - 1, extract_bits_output_offset = 0; i >= 0;
extract_bits_output_offset += number_of_bits_per_block[i--]) {
auto nb_bits_to_extract = number_of_bits_per_block[i];
size_t delta_log = 64 - nb_bits_to_extract;
auto in_block = &in_copy[lwe_big_size * i];
// trick ( ct - delta/2 + delta/2^4 )
uint64_t sub = (uint64_t(1) << (uint64_t(64) - nb_bits_to_extract - 1)) -
(uint64_t(1) << (uint64_t(64) - nb_bits_to_extract - 5));
in_block[lwe_big_size - 1] -= sub;
size_t scratch_size;
size_t scratch_align;
concrete_cpu_extract_bit_lwe_ciphertext_u64_scratch(
&scratch_size, &scratch_align, lwe_small_dim, lwe_big_dim, glwe_dim,
polynomial_size, fft);
// Allocate scratch
auto *scratch = (uint8_t *)aligned_alloc(scratch_align, scratch_size);
concrete_cpu_extract_bit_lwe_ciphertext_u64(
&extract_bits_output_buffer[lwe_small_size *
extract_bits_output_offset],
in_block, bootstrap_key, keyswicth_key, lwe_small_dim,
nb_bits_to_extract, lwe_big_dim, nb_bits_to_extract, delta_log,
bsk_level_count, bsk_base_log, glwe_dim, polynomial_size, lwe_small_dim,
ksk_level_count, ksk_base_log, lwe_big_dim, lwe_small_dim, fft, scratch,
scratch_size);
free(scratch);
}
size_t ct_in_count = total_number_of_bits_per_block;
size_t lut_size = 1 << ct_in_count;
size_t ct_out_count = out_size_0;
size_t lut_count = ct_out_count;
assert(lut_ct_size0 == lut_count);
assert(lut_ct_size1 == lut_size);
// Vertical packing
size_t scratch_size;
size_t scratch_align;
concrete_cpu_circuit_bootstrap_boolean_vertical_packing_lwe_ciphertext_u64_scratch(
&scratch_size, &scratch_align, ct_out_count, lwe_small_dim, ct_in_count,
lut_size, lut_count, glwe_dim, polynomial_size, polynomial_size,
cbs_level_count, fft);
auto *scratch = (uint8_t *)aligned_alloc(scratch_align, scratch_size);
size_t fpkskKeyId = 0;
auto fp_keyswicth_key = context->fp_keyswitch_key_buffer(fpkskKeyId);
concrete_cpu_circuit_bootstrap_boolean_vertical_packing_lwe_ciphertext_u64(
out_aligned + out_offset, extract_bits_output_buffer,
lut_ct_aligned + lut_ct_offset, bootstrap_key, fp_keyswicth_key,
lwe_big_dim, ct_out_count, lwe_small_dim, ct_in_count, lut_size,
lut_count, bsk_level_count, bsk_base_log, glwe_dim, polynomial_size,
lwe_small_dim, fpksk_level_count, fpksk_base_log, lwe_big_dim, glwe_dim,
polynomial_size, glwe_dim + 1, cbs_level_count, cbs_base_log, fft,
scratch, scratch_size);
free(scratch);
}
void memref_copy_one_rank(uint64_t *src_allocated, uint64_t *src_aligned,
uint64_t src_offset, uint64_t src_size,
uint64_t src_stride, uint64_t *dst_allocated,
uint64_t *dst_aligned, uint64_t dst_offset,
uint64_t dst_size, uint64_t dst_stride) {
assert(src_size == dst_size && "memref_copy_one_rank size differs");
if (src_stride == dst_stride) {
memcpy(dst_aligned + dst_offset, src_aligned + src_offset,
src_size * sizeof(uint64_t));
return;
}
for (size_t i = 0; i < src_size; i++) {
dst_aligned[dst_offset + i * dst_stride] =
src_aligned[src_offset + i * src_stride];
}
}
void memref_trace_ciphertext(uint64_t *ct0_allocated, uint64_t *ct0_aligned,
uint64_t ct0_offset, uint64_t ct0_size,
uint64_t ct0_stride, char *message_ptr,
uint32_t message_len, uint32_t msb) {
std::string message{message_ptr, (size_t)message_len};
std::cout << message << " : ";
std::bitset<64> bits{ct0_aligned[ct0_offset + ct0_size - 1]};
std::string bitstring = bits.to_string();
bitstring.insert(msb, 1, ' ');
std::cout << bitstring << std::endl;
}
void memref_trace_plaintext(uint64_t input, uint64_t input_width,
char *message_ptr, uint32_t message_len,
uint32_t msb) {
std::string message{message_ptr, (size_t)message_len};
std::cout << message << " : ";
std::bitset<64> bits{input};
std::string bitstring = bits.to_string();
bitstring.erase(0, 64 - input_width);
bitstring.insert(msb, 1, ' ');
std::cout << bitstring << std::endl;
}
void memref_trace_message(char *message_ptr, uint32_t message_len) {
std::string message{message_ptr, (size_t)message_len};
std::cout << message << std::endl;
}
Reference in New Issue View Git Blame Copy Permalink