upd: trying to optmize

This commit is contained in:
2023-10-31 18:00:33 +08:00
parent 4fbd4a579a
commit db53bfdcff
4 changed files with 10 additions and 5 deletions

View File

@ -1,6 +1,6 @@
set(PROJECT_NAME ${CMAKE_PROJECT_NAME})
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_FLAGS "-O2 -pg")
set(CMAKE_CXX_FLAGS "-g")
set(ENV{MAKEFLAGS} "-j16")
include_directories(${PROJECT_SOURCE_DIR}/include)
link_directories(${PROJECT_SOURCE_DIR}/src)

View File

@ -52,6 +52,7 @@ root= 6
void RightMoveBy(int);
void ProcessHalfBlock();
void RestoreHalfBlock();
public:
int2048();
int2048(long long);
@ -86,7 +87,7 @@ root= 6
int2048 &operator-=(const int2048 &);
friend int2048 operator-(int2048, const int2048 &);
inline friend void UnsignedMultiply(int2048 &, const int2048 *, bool);
inline friend void UnsignedMultiply(int2048 &, const int2048 *, bool, int);
int2048 &Multiply(const int2048 &);
friend int2048 Multiply(int2048, const int2048 &);

View File

@ -1,5 +1,5 @@
set(PROJECT_NAME ${CMAKE_PROJECT_NAME})
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_FLAGS "-O2 -pg")
set(CMAKE_CXX_FLAGS "-g")
include_directories(${PROJECT_SOURCE_DIR}/include)
add_library(int2048 STATIC int2048.cpp)

View File

@ -501,10 +501,14 @@ void int2048::NTTTransform(__int128_t *a, int NTT_blocks,
}
}
inline void UnsignedMultiply(int2048 &A, const int2048 *pB,
bool inverse = false) {
bool inverse = false, int lenght_limit = 0) {
if (&A == pB) throw "UnsignedMultiply: A and B are the same object";
int blocks_of_A = ((A.num_length + int2048::kNum - 1) / int2048::kNum);
int blocks_of_B = ((pB->num_length + int2048::kNum - 1) / int2048::kNum);
if (inverse) {
lenght_limit = std::min(lenght_limit, pB->num_length);
blocks_of_B = lenght_limit;
}
int max_blocks = blocks_of_A + blocks_of_B;
int NTT_blocks = 2;
while (NTT_blocks < (max_blocks << 1)) NTT_blocks <<= 1;
@ -704,7 +708,7 @@ inline void UnsignedDivide(int2048 &A, const int2048 *pB) {
}
assert(tmp_x.num_length % int2048::kNum == 0);
assert(inverse_B.num_length % int2048::kNum == 0);
UnsignedMultiply(tmp_x, &inverse_B, true);
UnsignedMultiply(tmp_x, &inverse_B, true, inverse_B.num_length);
for (int i = 0; i < tmp_x_error + inverseB_error; i++)
tmp_x.RestoreHalfBlock();
UnsignedMinus(inverse_two, &tmp_x, true);