mirror of
https://github.com/zama-ai/concrete.git
synced 2026-02-09 20:25:34 -05:00
feat(cuda): introduce cuda acceleration for the pbs and keyswitch
- a new crate concrete-cuda is added to the repository, containing some Cuda implementations for the bootstrap and keyswitch and a Rust wrapping to call them - a new backend_cuda is added to concrete-core, with dedicated entities whose memory is located on the GPU and engines that call the Cuda accelerated functions
This commit is contained in:
Agnes Leroy
94
include/bootstrap.h
Normal file
94
include/bootstrap.h
Normal file
@@ -0,0 +1,94 @@
|
||||
#ifndef CUDA_BOOTSTRAP_H
|
||||
#define CUDA_BOOTSTRAP_H
|
||||
|
||||
#include <cstdint>
|
||||
|
||||
extern "C" {
|
||||
|
||||
void cuda_initialize_twiddles(uint32_t polynomial_size, uint32_t gpu_index);
|
||||
|
||||
void cuda_convert_lwe_bootstrap_key_32(void *dest, void *src, void *v_stream,
|
||||
uint32_t gpu_index, uint32_t input_lwe_dim, uint32_t glwe_dim,
|
||||
uint32_t l_gadget, uint32_t polynomial_size);
|
||||
|
||||
void cuda_convert_lwe_bootstrap_key_64(void *dest, void *src, void *v_stream,
|
||||
uint32_t gpu_index, uint32_t input_lwe_dim, uint32_t glwe_dim,
|
||||
uint32_t l_gadget, uint32_t polynomial_size);
|
||||
|
||||
void cuda_bootstrap_amortized_lwe_ciphertext_vector_32(
|
||||
void *v_stream,
|
||||
void *lwe_out,
|
||||
void *test_vector,
|
||||
void *test_vector_indexes,
|
||||
void *lwe_in,
|
||||
void *bootstrapping_key,
|
||||
uint32_t input_lwe_dimension,
|
||||
uint32_t polynomial_size,
|
||||
uint32_t base_log,
|
||||
uint32_t l_gadget,
|
||||
uint32_t num_samples,
|
||||
uint32_t num_test_vectors,
|
||||
uint32_t lwe_idx,
|
||||
uint32_t max_shared_memory);
|
||||
|
||||
void cuda_bootstrap_amortized_lwe_ciphertext_vector_64(
|
||||
void *v_stream,
|
||||
void *lwe_out,
|
||||
void *test_vector,
|
||||
void *test_vector_indexes,
|
||||
void *lwe_in,
|
||||
void *bootstrapping_key,
|
||||
uint32_t input_lwe_dimension,
|
||||
uint32_t polynomial_size,
|
||||
uint32_t base_log,
|
||||
uint32_t l_gadget,
|
||||
uint32_t num_samples,
|
||||
uint32_t num_test_vectors,
|
||||
uint32_t lwe_idx,
|
||||
uint32_t max_shared_memory);
|
||||
|
||||
void cuda_bootstrap_low_latency_lwe_ciphertext_vector_32(
|
||||
void *v_stream,
|
||||
void *lwe_out,
|
||||
void *test_vector,
|
||||
void *test_vector_indexes,
|
||||
void *lwe_in,
|
||||
void *bootstrapping_key,
|
||||
uint32_t input_lwe_dimension,
|
||||
uint32_t polynomial_size,
|
||||
uint32_t base_log,
|
||||
uint32_t l_gadget,
|
||||
uint32_t num_samples,
|
||||
uint32_t num_test_vectors,
|
||||
uint32_t lwe_idx,
|
||||
uint32_t max_shared_memory);
|
||||
|
||||
void cuda_bootstrap_low_latency_lwe_ciphertext_vector_64(
|
||||
void *v_stream,
|
||||
void *lwe_out,
|
||||
void *test_vector,
|
||||
void *test_vector_indexes,
|
||||
void *lwe_in,
|
||||
void *bootstrapping_key,
|
||||
uint32_t input_lwe_dimension,
|
||||
uint32_t polynomial_size,
|
||||
uint32_t base_log,
|
||||
uint32_t l_gadget,
|
||||
uint32_t num_samples,
|
||||
uint32_t num_test_vectors,
|
||||
uint32_t lwe_idx,
|
||||
uint32_t max_shared_memory);
|
||||
};
|
||||
|
||||
__device__ inline int get_start_ith_ggsw(int i, uint32_t polynomial_size,
|
||||
int glwe_dimension,
|
||||
uint32_t l_gadget);
|
||||
|
||||
__device__ double2*
|
||||
get_ith_mask_kth_block(double2* ptr, int i, int k, int level, uint32_t polynomial_size,
|
||||
int glwe_dimension, uint32_t l_gadget);
|
||||
|
||||
__device__ double2*
|
||||
get_ith_body_kth_block(double2 *ptr, int i, int k, int level, uint32_t polynomial_size,
|
||||
int glwe_dimension, uint32_t l_gadget);
|
||||
#endif // CUDA_BOOTSTRAP_H
|
||||
30
include/device.h
Normal file
30
include/device.h
Normal file
@@ -0,0 +1,30 @@
|
||||
#include <cstdint>
|
||||
|
||||
extern "C" {
|
||||
void *cuda_create_stream(uint32_t gpu_index);
|
||||
|
||||
int cuda_destroy_stream(void *v_stream, uint32_t gpu_index);
|
||||
|
||||
void *cuda_malloc(uint64_t size, uint32_t gpu_index);
|
||||
|
||||
int cuda_check_valid_malloc(uint64_t size, uint32_t gpu_index);
|
||||
|
||||
int cuda_memcpy_to_cpu(void *dest, const void *src, uint64_t size,
|
||||
uint32_t gpu_index);
|
||||
|
||||
int cuda_memcpy_async_to_gpu(void *dest, void *src, uint64_t size,
|
||||
void *v_stream, uint32_t gpu_index);
|
||||
|
||||
int cuda_memcpy_to_gpu(void *dest, void *src, uint64_t size,
|
||||
uint32_t gpu_index);
|
||||
|
||||
int cuda_memcpy_async_to_cpu(void *dest, const void *src, uint64_t size,
|
||||
void *v_stream, uint32_t gpu_index);
|
||||
int cuda_get_number_of_gpus();
|
||||
|
||||
int cuda_synchronize_device(uint32_t gpu_index);
|
||||
|
||||
int cuda_drop(void *ptr, uint32_t gpu_index);
|
||||
|
||||
int cuda_get_max_shared_memory(uint32_t gpu_index);
|
||||
}
|
||||
1154
include/helper_cuda.h
Normal file
1154
include/helper_cuda.h
Normal file
File diff suppressed because it is too large
Load Diff
100
include/helper_debug.cuh
Normal file
100
include/helper_debug.cuh
Normal file
@@ -0,0 +1,100 @@
|
||||
#include "cuComplex.h"
|
||||
#include "thrust/complex.h"
|
||||
#include <iostream>
|
||||
#include <string>
|
||||
#include <type_traits>
|
||||
|
||||
#define PRINT_VARS
|
||||
#ifdef PRINT_VARS
|
||||
#define PRINT_DEBUG_5(var, begin, end, step, cond) \
|
||||
_print_debug(var, #var, begin, end, step, cond, "", false)
|
||||
#define PRINT_DEBUG_6(var, begin, end, step, cond, text) \
|
||||
_print_debug(var, #var, begin, end, step, cond, text, true)
|
||||
#define CAT(A, B) A##B
|
||||
#define PRINT_SELECT(NAME, NUM) CAT(NAME##_, NUM)
|
||||
#define GET_COUNT(_1, _2, _3, _4, _5, _6, COUNT, ...) COUNT
|
||||
#define VA_SIZE(...) GET_COUNT(__VA_ARGS__, 6, 5, 4, 3, 2, 1)
|
||||
#define PRINT_DEBUG(...) \
|
||||
PRINT_SELECT(PRINT_DEBUG, VA_SIZE(__VA_ARGS__))(__VA_ARGS__)
|
||||
#else
|
||||
#define PRINT_DEBUG(...)
|
||||
#endif
|
||||
|
||||
template <typename T>
|
||||
__device__ typename std::enable_if<std::is_unsigned<T>::value, void>::type
|
||||
_print_debug(T *var, const char *var_name, int start, int end, int step,
|
||||
bool cond, const char *text, bool has_text) {
|
||||
__syncthreads();
|
||||
if (cond) {
|
||||
if (has_text)
|
||||
printf("%s\n", text);
|
||||
for (int i = start; i < end; i += step) {
|
||||
printf("%s[%u]: %u\n", var_name, i, var[i]);
|
||||
}
|
||||
}
|
||||
__syncthreads();
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
__device__ typename std::enable_if<std::is_signed<T>::value, void>::type
|
||||
_print_debug(T *var, const char *var_name, int start, int end, int step,
|
||||
bool cond, const char *text, bool has_text) {
|
||||
__syncthreads();
|
||||
if (cond) {
|
||||
if (has_text)
|
||||
printf("%s\n", text);
|
||||
for (int i = start; i < end; i += step) {
|
||||
printf("%s[%u]: %d\n", var_name, i, var[i]);
|
||||
}
|
||||
}
|
||||
__syncthreads();
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
__device__ typename std::enable_if<std::is_floating_point<T>::value, void>::type
|
||||
_print_debug(T *var, const char *var_name, int start, int end, int step,
|
||||
bool cond, const char *text, bool has_text) {
|
||||
__syncthreads();
|
||||
if (cond) {
|
||||
if (has_text)
|
||||
printf("%s\n", text);
|
||||
for (int i = start; i < end; i += step) {
|
||||
printf("%s[%u]: %.15f\n", var_name, i, var[i]);
|
||||
}
|
||||
}
|
||||
__syncthreads();
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
__device__
|
||||
typename std::enable_if<std::is_same<T, thrust::complex<double>>::value,
|
||||
void>::type
|
||||
_print_debug(T *var, const char *var_name, int start, int end, int step,
|
||||
bool cond, const char *text, bool has_text) {
|
||||
__syncthreads();
|
||||
if (cond) {
|
||||
if (has_text)
|
||||
printf("%s\n", text);
|
||||
for (int i = start; i < end; i += step) {
|
||||
printf("%s[%u]: %.15f , %.15f\n", var_name, i, var[i].real(),
|
||||
var[i].imag());
|
||||
}
|
||||
}
|
||||
__syncthreads();
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
__device__
|
||||
typename std::enable_if<std::is_same<T, cuDoubleComplex>::value, void>::type
|
||||
_print_debug(T *var, const char *var_name, int start, int end, int step,
|
||||
bool cond, const char *text, bool has_text) {
|
||||
__syncthreads();
|
||||
if (cond) {
|
||||
if (has_text)
|
||||
printf("%s\n", text);
|
||||
for (int i = start; i < end; i += step) {
|
||||
printf("%s[%u]: %.15f , %.15f\n", var_name, i, var[i].x, var[i].y);
|
||||
}
|
||||
}
|
||||
__syncthreads();
|
||||
}
|
||||
664
include/helper_string.h
Normal file
664
include/helper_string.h
Normal file
@@ -0,0 +1,664 @@
|
||||
/**
|
||||
* Copyright 1993-2013 NVIDIA Corporation. All rights reserved.
|
||||
*
|
||||
* Please refer to the NVIDIA end user license agreement (EULA) associated
|
||||
* with this source code for terms and conditions that govern your use of
|
||||
* this software. Any use, reproduction, disclosure, or distribution of
|
||||
* this software and related documentation outside the terms of the EULA
|
||||
* is strictly prohibited.
|
||||
*
|
||||
*/
|
||||
|
||||
// These are helper functions for the SDK samples (string parsing, timers, etc)
|
||||
#ifndef STRING_HELPER_H
|
||||
#define STRING_HELPER_H
|
||||
|
||||
#include <fstream>
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include <string>
|
||||
|
||||
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
|
||||
#ifndef _CRT_SECURE_NO_DEPRECATE
|
||||
#define _CRT_SECURE_NO_DEPRECATE
|
||||
#endif
|
||||
#ifndef STRCASECMP
|
||||
#define STRCASECMP _stricmp
|
||||
#endif
|
||||
#ifndef STRNCASECMP
|
||||
#define STRNCASECMP _strnicmp
|
||||
#endif
|
||||
#ifndef STRCPY
|
||||
#define STRCPY(sFilePath, nLength, sPath) strcpy_s(sFilePath, nLength, sPath)
|
||||
#endif
|
||||
|
||||
#ifndef FOPEN
|
||||
#define FOPEN(fHandle, filename, mode) fopen_s(&fHandle, filename, mode)
|
||||
#endif
|
||||
#ifndef FOPEN_FAIL
|
||||
#define FOPEN_FAIL(result) (result != 0)
|
||||
#endif
|
||||
#ifndef SSCANF
|
||||
#define SSCANF sscanf_s
|
||||
#endif
|
||||
#ifndef SPRINTF
|
||||
#define SPRINTF sprintf_s
|
||||
#endif
|
||||
#else // Linux Includes
|
||||
#include <string.h>
|
||||
#include <strings.h>
|
||||
|
||||
#ifndef STRCASECMP
|
||||
#define STRCASECMP strcasecmp
|
||||
#endif
|
||||
#ifndef STRNCASECMP
|
||||
#define STRNCASECMP strncasecmp
|
||||
#endif
|
||||
#ifndef STRCPY
|
||||
#define STRCPY(sFilePath, nLength, sPath) strcpy(sFilePath, sPath)
|
||||
#endif
|
||||
|
||||
#ifndef FOPEN
|
||||
#define FOPEN(fHandle, filename, mode) (fHandle = fopen(filename, mode))
|
||||
#endif
|
||||
#ifndef FOPEN_FAIL
|
||||
#define FOPEN_FAIL(result) (result == NULL)
|
||||
#endif
|
||||
#ifndef SSCANF
|
||||
#define SSCANF sscanf
|
||||
#endif
|
||||
#ifndef SPRINTF
|
||||
#define SPRINTF sprintf
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#ifndef EXIT_WAIVED
|
||||
#define EXIT_WAIVED 2
|
||||
#endif
|
||||
|
||||
// CUDA Utility Helper Functions
|
||||
inline int stringRemoveDelimiter(char delimiter, const char *string) {
|
||||
int string_start = 0;
|
||||
|
||||
while (string[string_start] == delimiter) {
|
||||
string_start++;
|
||||
}
|
||||
|
||||
if (string_start >= (int)strlen(string) - 1) {
|
||||
return 0;
|
||||
}
|
||||
|
||||
return string_start;
|
||||
}
|
||||
|
||||
inline int getFileExtension(char *filename, char **extension) {
|
||||
int string_length = (int)strlen(filename);
|
||||
|
||||
while (filename[string_length--] != '.') {
|
||||
if (string_length == 0)
|
||||
break;
|
||||
}
|
||||
|
||||
if (string_length > 0)
|
||||
string_length += 2;
|
||||
|
||||
if (string_length == 0)
|
||||
*extension = NULL;
|
||||
else
|
||||
*extension = &filename[string_length];
|
||||
|
||||
return string_length;
|
||||
}
|
||||
|
||||
inline bool checkCmdLineFlag(const int argc, const char **argv,
|
||||
const char *string_ref) {
|
||||
bool bFound = false;
|
||||
|
||||
if (argc >= 1) {
|
||||
for (int i = 1; i < argc; i++) {
|
||||
int string_start = stringRemoveDelimiter('-', argv[i]);
|
||||
const char *string_argv = &argv[i][string_start];
|
||||
|
||||
const char *equal_pos = strchr(string_argv, '=');
|
||||
int argv_length =
|
||||
(int)(equal_pos == 0 ? strlen(string_argv) : equal_pos - string_argv);
|
||||
|
||||
int length = (int)strlen(string_ref);
|
||||
|
||||
if (length == argv_length &&
|
||||
!STRNCASECMP(string_argv, string_ref, length)) {
|
||||
bFound = true;
|
||||
continue;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return bFound;
|
||||
}
|
||||
|
||||
// This function wraps the CUDA Driver API into a template function
|
||||
template <class T>
|
||||
inline bool getCmdLineArgumentValue(const int argc, const char **argv,
|
||||
const char *string_ref, T *value) {
|
||||
bool bFound = false;
|
||||
|
||||
if (argc >= 1) {
|
||||
for (int i = 1; i < argc; i++) {
|
||||
int string_start = stringRemoveDelimiter('-', argv[i]);
|
||||
const char *string_argv = &argv[i][string_start];
|
||||
int length = (int)strlen(string_ref);
|
||||
|
||||
if (!STRNCASECMP(string_argv, string_ref, length)) {
|
||||
if (length + 1 <= (int)strlen(string_argv)) {
|
||||
int auto_inc = (string_argv[length] == '=') ? 1 : 0;
|
||||
*value = (T)atoi(&string_argv[length + auto_inc]);
|
||||
}
|
||||
|
||||
bFound = true;
|
||||
i = argc;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return bFound;
|
||||
}
|
||||
|
||||
inline int getCmdLineArgumentInt(const int argc, const char **argv,
|
||||
const char *string_ref) {
|
||||
bool bFound = false;
|
||||
int value = -1;
|
||||
|
||||
if (argc >= 1) {
|
||||
for (int i = 1; i < argc; i++) {
|
||||
int string_start = stringRemoveDelimiter('-', argv[i]);
|
||||
const char *string_argv = &argv[i][string_start];
|
||||
int length = (int)strlen(string_ref);
|
||||
|
||||
if (!STRNCASECMP(string_argv, string_ref, length)) {
|
||||
if (length + 1 <= (int)strlen(string_argv)) {
|
||||
int auto_inc = (string_argv[length] == '=') ? 1 : 0;
|
||||
value = atoi(&string_argv[length + auto_inc]);
|
||||
} else {
|
||||
value = 0;
|
||||
}
|
||||
|
||||
bFound = true;
|
||||
continue;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (bFound) {
|
||||
return value;
|
||||
} else {
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
inline float getCmdLineArgumentFloat(const int argc, const char **argv,
|
||||
const char *string_ref) {
|
||||
bool bFound = false;
|
||||
float value = -1;
|
||||
|
||||
if (argc >= 1) {
|
||||
for (int i = 1; i < argc; i++) {
|
||||
int string_start = stringRemoveDelimiter('-', argv[i]);
|
||||
const char *string_argv = &argv[i][string_start];
|
||||
int length = (int)strlen(string_ref);
|
||||
|
||||
if (!STRNCASECMP(string_argv, string_ref, length)) {
|
||||
if (length + 1 <= (int)strlen(string_argv)) {
|
||||
int auto_inc = (string_argv[length] == '=') ? 1 : 0;
|
||||
value = (float)atof(&string_argv[length + auto_inc]);
|
||||
} else {
|
||||
value = 0.f;
|
||||
}
|
||||
|
||||
bFound = true;
|
||||
continue;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (bFound) {
|
||||
return value;
|
||||
} else {
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
inline bool getCmdLineArgumentString(const int argc, const char **argv,
|
||||
const char *string_ref,
|
||||
char **string_retval) {
|
||||
bool bFound = false;
|
||||
|
||||
if (argc >= 1) {
|
||||
for (int i = 1; i < argc; i++) {
|
||||
int string_start = stringRemoveDelimiter('-', argv[i]);
|
||||
char *string_argv = (char *)&argv[i][string_start];
|
||||
int length = (int)strlen(string_ref);
|
||||
|
||||
if (!STRNCASECMP(string_argv, string_ref, length)) {
|
||||
*string_retval = &string_argv[length + 1];
|
||||
bFound = true;
|
||||
continue;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (!bFound) {
|
||||
*string_retval = NULL;
|
||||
}
|
||||
|
||||
return bFound;
|
||||
}
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////////
|
||||
//! Find the path for a file assuming that
|
||||
//! files are found in the searchPath.
|
||||
//!
|
||||
//! @return the path if succeeded, otherwise 0
|
||||
//! @param filename name of the file
|
||||
//! @param executable_path optional absolute path of the executable
|
||||
//////////////////////////////////////////////////////////////////////////////
|
||||
inline char *sdkFindFilePath(const char *filename,
|
||||
const char *executable_path) {
|
||||
// <executable_name> defines a variable that is replaced with the name of the
|
||||
// executable
|
||||
|
||||
// Typical relative search paths to locate needed companion files (e.g. sample
|
||||
// input data, or JIT source files) The origin for the relative search may be
|
||||
// the .exe file, a .bat file launching an .exe, a browser .exe launching the
|
||||
// .exe or .bat, etc
|
||||
const char *searchPath[] = {
|
||||
"./", // same dir
|
||||
"./common/", // "/common/" subdir
|
||||
"./common/data/", // "/common/data/" subdir
|
||||
"./data/", // "/data/" subdir
|
||||
"./src/", // "/src/" subdir
|
||||
"./src/<executable_name>/data/", // "/src/<executable_name>/data/" subdir
|
||||
"./inc/", // "/inc/" subdir
|
||||
"./0_Simple/", // "/0_Simple/" subdir
|
||||
"./1_Utilities/", // "/1_Utilities/" subdir
|
||||
"./2_Graphics/", // "/2_Graphics/" subdir
|
||||
"./3_Imaging/", // "/3_Imaging/" subdir
|
||||
"./4_Finance/", // "/4_Finance/" subdir
|
||||
"./5_Simulations/", // "/5_Simulations/" subdir
|
||||
"./6_Advanced/", // "/6_Advanced/" subdir
|
||||
"./7_CUDALibraries/", // "/7_CUDALibraries/" subdir
|
||||
"./8_Android/", // "/8_Android/" subdir
|
||||
"./samples/", // "/samples/" subdir
|
||||
|
||||
"../", // up 1 in tree
|
||||
"../common/", // up 1 in tree, "/common/" subdir
|
||||
"../common/data/", // up 1 in tree, "/common/data/" subdir
|
||||
"../data/", // up 1 in tree, "/data/" subdir
|
||||
"../src/", // up 1 in tree, "/src/" subdir
|
||||
"../inc/", // up 1 in tree, "/inc/" subdir
|
||||
|
||||
"../0_Simple/<executable_name>/data/", // up 1 in tree,
|
||||
// "/0_Simple/<executable_name>/"
|
||||
// subdir
|
||||
"../1_Utilities/<executable_name>/data/", // up 1 in tree,
|
||||
// "/1_Utilities/<executable_name>/"
|
||||
// subdir
|
||||
"../2_Graphics/<executable_name>/data/", // up 1 in tree,
|
||||
// "/2_Graphics/<executable_name>/"
|
||||
// subdir
|
||||
"../3_Imaging/<executable_name>/data/", // up 1 in tree,
|
||||
// "/3_Imaging/<executable_name>/"
|
||||
// subdir
|
||||
"../4_Finance/<executable_name>/data/", // up 1 in tree,
|
||||
// "/4_Finance/<executable_name>/"
|
||||
// subdir
|
||||
"../5_Simulations/<executable_name>/data/", // up 1 in tree,
|
||||
// "/5_Simulations/<executable_name>/"
|
||||
// subdir
|
||||
"../6_Advanced/<executable_name>/data/", // up 1 in tree,
|
||||
// "/6_Advanced/<executable_name>/"
|
||||
// subdir
|
||||
"../7_CUDALibraries/<executable_name>/data/", // up 1 in tree,
|
||||
// "/7_CUDALibraries/<executable_name>/"
|
||||
// subdir
|
||||
"../8_Android/<executable_name>/data/", // up 1 in tree,
|
||||
// "/8_Android/<executable_name>/"
|
||||
// subdir
|
||||
"../samples/<executable_name>/data/", // up 1 in tree,
|
||||
// "/samples/<executable_name>/"
|
||||
// subdir
|
||||
"../../", // up 2 in tree
|
||||
"../../common/", // up 2 in tree, "/common/" subdir
|
||||
"../../common/data/", // up 2 in tree, "/common/data/" subdir
|
||||
"../../data/", // up 2 in tree, "/data/" subdir
|
||||
"../../src/", // up 2 in tree, "/src/" subdir
|
||||
"../../inc/", // up 2 in tree, "/inc/" subdir
|
||||
"../../sandbox/<executable_name>/data/", // up 2 in tree,
|
||||
// "/sandbox/<executable_name>/"
|
||||
// subdir
|
||||
"../../0_Simple/<executable_name>/data/", // up 2 in tree,
|
||||
// "/0_Simple/<executable_name>/"
|
||||
// subdir
|
||||
"../../1_Utilities/<executable_name>/data/", // up 2 in tree,
|
||||
// "/1_Utilities/<executable_name>/"
|
||||
// subdir
|
||||
"../../2_Graphics/<executable_name>/data/", // up 2 in tree,
|
||||
// "/2_Graphics/<executable_name>/"
|
||||
// subdir
|
||||
"../../3_Imaging/<executable_name>/data/", // up 2 in tree,
|
||||
// "/3_Imaging/<executable_name>/"
|
||||
// subdir
|
||||
"../../4_Finance/<executable_name>/data/", // up 2 in tree,
|
||||
// "/4_Finance/<executable_name>/"
|
||||
// subdir
|
||||
"../../5_Simulations/<executable_name>/data/", // up 2 in tree,
|
||||
// "/5_Simulations/<executable_name>/"
|
||||
// subdir
|
||||
"../../6_Advanced/<executable_name>/data/", // up 2 in tree,
|
||||
// "/6_Advanced/<executable_name>/"
|
||||
// subdir
|
||||
"../../7_CUDALibraries/<executable_name>/data/", // up 2 in tree,
|
||||
// "/7_CUDALibraries/<executable_name>/"
|
||||
// subdir
|
||||
"../../8_Android/<executable_name>/data/", // up 2 in tree,
|
||||
// "/8_Android/<executable_name>/"
|
||||
// subdir
|
||||
"../../samples/<executable_name>/data/", // up 2 in tree,
|
||||
// "/samples/<executable_name>/"
|
||||
// subdir
|
||||
"../../../", // up 3 in tree
|
||||
"../../../src/<executable_name>/", // up 3 in tree,
|
||||
// "/src/<executable_name>/" subdir
|
||||
"../../../src/<executable_name>/data/", // up 3 in tree,
|
||||
// "/src/<executable_name>/data/"
|
||||
// subdir
|
||||
"../../../src/<executable_name>/src/", // up 3 in tree,
|
||||
// "/src/<executable_name>/src/"
|
||||
// subdir
|
||||
"../../../src/<executable_name>/inc/", // up 3 in tree,
|
||||
// "/src/<executable_name>/inc/"
|
||||
// subdir
|
||||
"../../../sandbox/<executable_name>/", // up 3 in tree,
|
||||
// "/sandbox/<executable_name>/"
|
||||
// subdir
|
||||
"../../../sandbox/<executable_name>/data/", // up 3 in tree,
|
||||
// "/sandbox/<executable_name>/data/"
|
||||
// subdir
|
||||
"../../../sandbox/<executable_name>/src/", // up 3 in tree,
|
||||
// "/sandbox/<executable_name>/src/"
|
||||
// subdir
|
||||
"../../../sandbox/<executable_name>/inc/", // up 3 in tree,
|
||||
// "/sandbox/<executable_name>/inc/"
|
||||
// subdir
|
||||
"../../../0_Simple/<executable_name>/data/", // up 3 in tree,
|
||||
// "/0_Simple/<executable_name>/"
|
||||
// subdir
|
||||
"../../../1_Utilities/<executable_name>/data/", // up 3 in tree,
|
||||
// "/1_Utilities/<executable_name>/"
|
||||
// subdir
|
||||
"../../../2_Graphics/<executable_name>/data/", // up 3 in tree,
|
||||
// "/2_Graphics/<executable_name>/"
|
||||
// subdir
|
||||
"../../../3_Imaging/<executable_name>/data/", // up 3 in tree,
|
||||
// "/3_Imaging/<executable_name>/"
|
||||
// subdir
|
||||
"../../../4_Finance/<executable_name>/data/", // up 3 in tree,
|
||||
// "/4_Finance/<executable_name>/"
|
||||
// subdir
|
||||
"../../../5_Simulations/<executable_name>/data/", // up 3 in tree,
|
||||
// "/5_Simulations/<executable_name>/"
|
||||
// subdir
|
||||
"../../../6_Advanced/<executable_name>/data/", // up 3 in tree,
|
||||
// "/6_Advanced/<executable_name>/"
|
||||
// subdir
|
||||
"../../../7_CUDALibraries/<executable_name>/data/", // up 3 in tree,
|
||||
// "/7_CUDALibraries/<executable_name>/"
|
||||
// subdir
|
||||
"../../../8_Android/<executable_name>/data/", // up 3 in tree,
|
||||
// "/8_Android/<executable_name>/"
|
||||
// subdir
|
||||
"../../../0_Simple/<executable_name>/", // up 3 in tree,
|
||||
// "/0_Simple/<executable_name>/"
|
||||
// subdir
|
||||
"../../../1_Utilities/<executable_name>/", // up 3 in tree,
|
||||
// "/1_Utilities/<executable_name>/"
|
||||
// subdir
|
||||
"../../../2_Graphics/<executable_name>/", // up 3 in tree,
|
||||
// "/2_Graphics/<executable_name>/"
|
||||
// subdir
|
||||
"../../../3_Imaging/<executable_name>/", // up 3 in tree,
|
||||
// "/3_Imaging/<executable_name>/"
|
||||
// subdir
|
||||
"../../../4_Finance/<executable_name>/", // up 3 in tree,
|
||||
// "/4_Finance/<executable_name>/"
|
||||
// subdir
|
||||
"../../../5_Simulations/<executable_name>/", // up 3 in tree,
|
||||
// "/5_Simulations/<executable_name>/"
|
||||
// subdir
|
||||
"../../../6_Advanced/<executable_name>/", // up 3 in tree,
|
||||
// "/6_Advanced/<executable_name>/"
|
||||
// subdir
|
||||
"../../../7_CUDALibraries/<executable_name>/", // up 3 in tree,
|
||||
// "/7_CUDALibraries/<executable_name>/"
|
||||
// subdir
|
||||
"../../../8_Android/<executable_name>/", // up 3 in tree,
|
||||
// "/8_Android/<executable_name>/"
|
||||
// subdir
|
||||
"../../../samples/<executable_name>/data/", // up 3 in tree,
|
||||
// "/samples/<executable_name>/"
|
||||
// subdir
|
||||
"../../../common/", // up 3 in tree, "../../../common/" subdir
|
||||
"../../../common/data/", // up 3 in tree, "../../../common/data/" subdir
|
||||
"../../../data/", // up 3 in tree, "../../../data/" subdir
|
||||
"../../../../", // up 4 in tree
|
||||
"../../../../src/<executable_name>/", // up 4 in tree,
|
||||
// "/src/<executable_name>/" subdir
|
||||
"../../../../src/<executable_name>/data/", // up 4 in tree,
|
||||
// "/src/<executable_name>/data/"
|
||||
// subdir
|
||||
"../../../../src/<executable_name>/src/", // up 4 in tree,
|
||||
// "/src/<executable_name>/src/"
|
||||
// subdir
|
||||
"../../../../src/<executable_name>/inc/", // up 4 in tree,
|
||||
// "/src/<executable_name>/inc/"
|
||||
// subdir
|
||||
"../../../../sandbox/<executable_name>/", // up 4 in tree,
|
||||
// "/sandbox/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../sandbox/<executable_name>/data/", // up 4 in tree,
|
||||
// "/sandbox/<executable_name>/data/"
|
||||
// subdir
|
||||
"../../../../sandbox/<executable_name>/src/", // up 4 in tree,
|
||||
// "/sandbox/<executable_name>/src/"
|
||||
// subdir
|
||||
"../../../../sandbox/<executable_name>/inc/", // up 4 in tree,
|
||||
// "/sandbox/<executable_name>/inc/"
|
||||
// subdir
|
||||
"../../../../0_Simple/<executable_name>/data/", // up 4 in tree,
|
||||
// "/0_Simple/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../1_Utilities/<executable_name>/data/", // up 4 in tree,
|
||||
// "/1_Utilities/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../2_Graphics/<executable_name>/data/", // up 4 in tree,
|
||||
// "/2_Graphics/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../3_Imaging/<executable_name>/data/", // up 4 in tree,
|
||||
// "/3_Imaging/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../4_Finance/<executable_name>/data/", // up 4 in tree,
|
||||
// "/4_Finance/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../5_Simulations/<executable_name>/data/", // up 4 in tree,
|
||||
// "/5_Simulations/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../6_Advanced/<executable_name>/data/", // up 4 in tree,
|
||||
// "/6_Advanced/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../7_CUDALibraries/<executable_name>/data/", // up 4 in tree,
|
||||
// "/7_CUDALibraries/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../8_Android/<executable_name>/data/", // up 4 in tree,
|
||||
// "/8_Android/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../0_Simple/<executable_name>/", // up 4 in tree,
|
||||
// "/0_Simple/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../1_Utilities/<executable_name>/", // up 4 in tree,
|
||||
// "/1_Utilities/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../2_Graphics/<executable_name>/", // up 4 in tree,
|
||||
// "/2_Graphics/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../3_Imaging/<executable_name>/", // up 4 in tree,
|
||||
// "/3_Imaging/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../4_Finance/<executable_name>/", // up 4 in tree,
|
||||
// "/4_Finance/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../5_Simulations/<executable_name>/", // up 4 in tree,
|
||||
// "/5_Simulations/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../6_Advanced/<executable_name>/", // up 4 in tree,
|
||||
// "/6_Advanced/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../7_CUDALibraries/<executable_name>/", // up 4 in tree,
|
||||
// "/7_CUDALibraries/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../8_Android/<executable_name>/", // up 4 in tree,
|
||||
// "/8_Android/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../samples/<executable_name>/data/", // up 4 in tree,
|
||||
// "/samples/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../common/", // up 4 in tree, "../../../common/" subdir
|
||||
"../../../../common/data/", // up 4 in tree, "../../../common/data/"
|
||||
// subdir
|
||||
"../../../../data/", // up 4 in tree, "../../../data/" subdir
|
||||
"../../../../../", // up 5 in tree
|
||||
"../../../../../src/<executable_name>/", // up 5 in tree,
|
||||
// "/src/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../src/<executable_name>/data/", // up 5 in tree,
|
||||
// "/src/<executable_name>/data/"
|
||||
// subdir
|
||||
"../../../../../src/<executable_name>/src/", // up 5 in tree,
|
||||
// "/src/<executable_name>/src/"
|
||||
// subdir
|
||||
"../../../../../src/<executable_name>/inc/", // up 5 in tree,
|
||||
// "/src/<executable_name>/inc/"
|
||||
// subdir
|
||||
"../../../../../sandbox/<executable_name>/", // up 5 in tree,
|
||||
// "/sandbox/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../sandbox/<executable_name>/data/", // up 5 in tree,
|
||||
// "/sandbox/<executable_name>/data/"
|
||||
// subdir
|
||||
"../../../../../sandbox/<executable_name>/src/", // up 5 in tree,
|
||||
// "/sandbox/<executable_name>/src/"
|
||||
// subdir
|
||||
"../../../../../sandbox/<executable_name>/inc/", // up 5 in tree,
|
||||
// "/sandbox/<executable_name>/inc/"
|
||||
// subdir
|
||||
"../../../../../0_Simple/<executable_name>/data/", // up 5 in tree,
|
||||
// "/0_Simple/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../1_Utilities/<executable_name>/data/", // up 5 in tree,
|
||||
// "/1_Utilities/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../2_Graphics/<executable_name>/data/", // up 5 in tree,
|
||||
// "/2_Graphics/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../3_Imaging/<executable_name>/data/", // up 5 in tree,
|
||||
// "/3_Imaging/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../4_Finance/<executable_name>/data/", // up 5 in tree,
|
||||
// "/4_Finance/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../5_Simulations/<executable_name>/data/", // up 5 in tree,
|
||||
// "/5_Simulations/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../6_Advanced/<executable_name>/data/", // up 5 in tree,
|
||||
// "/6_Advanced/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../7_CUDALibraries/<executable_name>/data/", // up 5 in tree,
|
||||
// "/7_CUDALibraries/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../8_Android/<executable_name>/data/", // up 5 in tree,
|
||||
// "/8_Android/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../samples/<executable_name>/data/", // up 5 in tree,
|
||||
// "/samples/<executable_name>/"
|
||||
// subdir
|
||||
"../../../../../common/", // up 5 in tree, "../../../common/" subdir
|
||||
"../../../../../common/data/", // up 5 in tree, "../../../common/data/"
|
||||
// subdir
|
||||
};
|
||||
|
||||
// Extract the executable name
|
||||
std::string executable_name;
|
||||
|
||||
if (executable_path != 0) {
|
||||
executable_name = std::string(executable_path);
|
||||
|
||||
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
|
||||
// Windows path delimiter
|
||||
size_t delimiter_pos = executable_name.find_last_of('\\');
|
||||
executable_name.erase(0, delimiter_pos + 1);
|
||||
|
||||
if (executable_name.rfind(".exe") != std::string::npos) {
|
||||
// we strip .exe, only if the .exe is found
|
||||
executable_name.resize(executable_name.size() - 4);
|
||||
}
|
||||
|
||||
#else
|
||||
// Linux & OSX path delimiter
|
||||
size_t delimiter_pos = executable_name.find_last_of('/');
|
||||
executable_name.erase(0, delimiter_pos + 1);
|
||||
#endif
|
||||
}
|
||||
|
||||
// Loop over all search paths and return the first hit
|
||||
for (unsigned int i = 0; i < sizeof(searchPath) / sizeof(char *); ++i) {
|
||||
std::string path(searchPath[i]);
|
||||
size_t executable_name_pos = path.find("<executable_name>");
|
||||
|
||||
// If there is executable_name variable in the searchPath
|
||||
// replace it with the value
|
||||
if (executable_name_pos != std::string::npos) {
|
||||
if (executable_path != 0) {
|
||||
path.replace(executable_name_pos, strlen("<executable_name>"),
|
||||
executable_name);
|
||||
} else {
|
||||
// Skip this path entry if no executable argument is given
|
||||
continue;
|
||||
}
|
||||
}
|
||||
|
||||
#ifdef _DEBUG
|
||||
printf("sdkFindFilePath <%s> in %s\n", filename, path.c_str());
|
||||
#endif
|
||||
|
||||
// Test if the file exists
|
||||
path.append(filename);
|
||||
FILE *fp;
|
||||
FOPEN(fp, path.c_str(), "rb");
|
||||
|
||||
if (fp != NULL) {
|
||||
fclose(fp);
|
||||
// File found
|
||||
// returning an allocated array here for backwards compatibility reasons
|
||||
char *file_path = (char *)malloc(path.length() + 1);
|
||||
STRCPY(file_path, path.length() + 1, path.c_str());
|
||||
return file_path;
|
||||
}
|
||||
|
||||
if (fp) {
|
||||
fclose(fp);
|
||||
}
|
||||
}
|
||||
|
||||
// File not found
|
||||
return 0;
|
||||
}
|
||||
|
||||
#endif
|
||||
24
include/keyswitch.h
Normal file
24
include/keyswitch.h
Normal file
@@ -0,0 +1,24 @@
|
||||
#ifndef CNCRT_KS_H_
|
||||
#define CNCRT_KS_H_
|
||||
|
||||
#include <cstdint>
|
||||
|
||||
extern "C" {
|
||||
|
||||
void cuda_keyswitch_lwe_ciphertext_vector_32(void *v_stream, void *lwe_out, void *lwe_in,
|
||||
void *ksk,
|
||||
uint32_t lwe_dimension_before,
|
||||
uint32_t lwe_dimension_after,
|
||||
uint32_t base_log, uint32_t l_gadget,
|
||||
uint32_t num_samples);
|
||||
|
||||
void cuda_keyswitch_lwe_ciphertext_vector_64(void *v_stream, void *lwe_out, void *lwe_in,
|
||||
void *ksk,
|
||||
uint32_t lwe_dimension_before,
|
||||
uint32_t lwe_dimension_after,
|
||||
uint32_t base_log, uint32_t l_gadget,
|
||||
uint32_t num_samples);
|
||||
|
||||
}
|
||||
|
||||
#endif // CNCRT_KS_H_
|
||||
Reference in New Issue
Block a user