mirror of
https://github.com/TREX-CoE/Sherman-Morrison.git
synced 2024-12-25 13:53:56 +01:00
Merge pull request #49 from fmgjcoppens/optimise_wb3
Woodbury 3x3 optimisations
This commit is contained in:
commit
05473c81d4
@ -98,7 +98,8 @@ template <typename T> void matMul(T *A, T *B, T *C, unsigned int M) {
|
||||
}
|
||||
|
||||
template <typename T1, typename T2, typename T3>
|
||||
void matMul2(T1 *A, T2 *B, T3 *C, unsigned int M, unsigned int N, unsigned int P) {
|
||||
void matMul2(T1 *A, T2 *B, T3 *C, unsigned int M, unsigned int N,
|
||||
unsigned int P) {
|
||||
for (unsigned int i = 0; i < M; i++) {
|
||||
for (unsigned int j = 0; j < P; j++) {
|
||||
C[i * P + j] = 0;
|
||||
@ -265,7 +266,6 @@ template <typename T> T residual_frobenius2(T *A, unsigned int Dim) {
|
||||
return res;
|
||||
}
|
||||
|
||||
|
||||
template <typename T> T residual2(T *A, unsigned int Dim) {
|
||||
double res = 0.0;
|
||||
for (unsigned int i = 0; i < Dim; i++) {
|
||||
|
@ -1,7 +1,8 @@
|
||||
// Sherman-Morrison-Woodbury kernel 1
|
||||
// WB2, WB3, SM2 mixing scheme
|
||||
void SMWB1(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
double *Updates, unsigned int *Updates_index);
|
||||
void SMWB1(double *Slater_inv, const unsigned int Dim,
|
||||
const unsigned int N_updates, double *Updates,
|
||||
unsigned int *Updates_index);
|
||||
|
||||
// // Sherman-Morrison-Woodbury kernel 2
|
||||
// // WB2, WB3, SM3 mixing scheme
|
||||
@ -15,5 +16,6 @@ void SMWB1(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
|
||||
// Sherman-Morrison-Woodbury kernel 4
|
||||
// WB2, SM2 mixing scheme
|
||||
void SMWB4(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
double *Updates, unsigned int *Updates_index);
|
||||
void SMWB4(double *Slater_inv, const unsigned int Dim,
|
||||
const unsigned int N_updates, double *Updates,
|
||||
unsigned int *Updates_index);
|
||||
|
@ -8,7 +8,9 @@ void SM2(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
double *Updates, unsigned int *Updates_index);
|
||||
|
||||
void SM2star(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
double *Updates, unsigned int *Updates_index, double *later_updates, unsigned int* later_index, unsigned int *later);
|
||||
double *Updates, unsigned int *Updates_index,
|
||||
double *later_updates, unsigned int *later_index,
|
||||
unsigned int *later);
|
||||
|
||||
// Sherman Morrison, leaving zero denominators for later
|
||||
void SM3(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
|
@ -1,7 +1,7 @@
|
||||
// Woodbury 2x2 kernel
|
||||
bool WB2(double *Slater_inv, unsigned int Dim,
|
||||
double *Updates, unsigned int *Updates_index);
|
||||
bool WB2(double *Slater_inv, unsigned int Dim, double *Updates,
|
||||
unsigned int *Updates_index);
|
||||
|
||||
// Woodbury 3x3 kernel
|
||||
bool WB3(double *Slater_inv, unsigned int Dim,
|
||||
double *Updates, unsigned int *Updates_index);
|
||||
bool WB3(double *Slater_inv, unsigned int Dim, double *Updates,
|
||||
unsigned int *Updates_index);
|
||||
|
51
src/SMWB.cpp
51
src/SMWB.cpp
@ -1,16 +1,19 @@
|
||||
#include "SMWB.hpp"
|
||||
#include "Helpers.hpp"
|
||||
#include "SM_Standard.hpp"
|
||||
#include "Woodbury.hpp"
|
||||
#include "Helpers.hpp"
|
||||
|
||||
// #define DEBUG1
|
||||
// #define DEBUG2
|
||||
|
||||
// Sherman-Morrison-Woodbury kernel 1
|
||||
// WB2, WB3, SM2 mixing scheme
|
||||
void SMWB1(double *Slater_inv, unsigned int Dim, unsigned int N_updates, double *Updates, unsigned int *Updates_index) {
|
||||
void SMWB1(double *Slater_inv, const unsigned int Dim,
|
||||
const unsigned int N_updates, double *Updates,
|
||||
unsigned int *Updates_index) {
|
||||
#ifdef DEBUG2
|
||||
std::cerr << "Called Sherman-Morrison-Woodbury kernel 1 with " << N_updates << " updates" << std::endl;
|
||||
std::cerr << "Called Sherman-Morrison-Woodbury kernel 1 with " << N_updates
|
||||
<< " updates" << std::endl;
|
||||
showMatrix2(Updates_index, 1, N_updates, "Updates_index");
|
||||
showMatrix2(Updates, N_updates, Dim, "Updates");
|
||||
#endif
|
||||
@ -21,7 +24,8 @@ void SMWB1(double *Slater_inv, unsigned int Dim, unsigned int N_updates, double
|
||||
unsigned int length_2block = 2 * Dim;
|
||||
unsigned int length_1block = 1 * Dim;
|
||||
|
||||
// Apply first 3*n_of_3blocks updates in n_of_3blocks blocks of 3 updates with Woodbury 3x3 kernel
|
||||
// Apply first 3*n_of_3blocks updates in n_of_3blocks blocks of 3 updates with
|
||||
// Woodbury 3x3 kernel
|
||||
double later_updates[Dim * N_updates];
|
||||
unsigned int later_index[N_updates];
|
||||
unsigned int later = 0;
|
||||
@ -33,49 +37,57 @@ void SMWB1(double *Slater_inv, unsigned int Dim, unsigned int N_updates, double
|
||||
ok = WB3(Slater_inv, Dim, Updates_3block, Updates_index_3block);
|
||||
if (!ok) { // Send the entire block to SM2
|
||||
#ifdef DEBUG2
|
||||
std::cerr << "Woodbury 3x3 kernel failed! Sending block to SM2" << std::endl;
|
||||
std::cerr << "Woodbury 3x3 kernel failed! Sending block to SM2"
|
||||
<< std::endl;
|
||||
showMatrix2(Updates_3block, 3, Dim, "Updates_3block");
|
||||
showMatrix2(Updates_index_3block, 1, 3, "Updates_index_3block");
|
||||
#endif
|
||||
unsigned int l = 0;
|
||||
SM2star(Slater_inv, Dim, 3, Updates_3block, Updates_index_3block, later_updates + (Dim * later), later_index + later, &l);
|
||||
SM2star(Slater_inv, Dim, 3, Updates_3block, Updates_index_3block,
|
||||
later_updates + (Dim * later), later_index + later, &l);
|
||||
later = later + l;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (remainder == 2) { // Apply last remaining block of 2 updates with Woodbury 2x2 kernel
|
||||
if (remainder ==
|
||||
2) { // Apply last remaining block of 2 updates with Woodbury 2x2 kernel
|
||||
double *Updates_2block = &Updates[n_of_3blocks * length_3block];
|
||||
unsigned int *Updates_index_2block = &Updates_index[3 * n_of_3blocks];
|
||||
bool ok;
|
||||
ok = WB2(Slater_inv, Dim, Updates_2block, Updates_index_2block);
|
||||
if (!ok) { // Send the entire block to SM2
|
||||
#ifdef DEBUG2
|
||||
std::cerr << "Woodbury 2x2 kernel failed! Sending block to SM2" << std::endl;
|
||||
std::cerr << "Woodbury 2x2 kernel failed! Sending block to SM2"
|
||||
<< std::endl;
|
||||
#endif
|
||||
unsigned int l = 0;
|
||||
SM2star(Slater_inv, Dim, 2, Updates_2block, Updates_index_2block, later_updates+(Dim * later), later_index + later, &l);
|
||||
SM2star(Slater_inv, Dim, 2, Updates_2block, Updates_index_2block,
|
||||
later_updates + (Dim * later), later_index + later, &l);
|
||||
later = later + l;
|
||||
}
|
||||
} else if (remainder == 1) { // Apply last remaining update with SM2
|
||||
double *Updates_1block = &Updates[n_of_3blocks * length_3block];
|
||||
unsigned int *Updates_index_1block = &Updates_index[3 * n_of_3blocks];
|
||||
unsigned int l = 0;
|
||||
SM2star(Slater_inv, Dim, 1, Updates_1block, Updates_index_1block, later_updates+(Dim * later), later_index + later, &l);
|
||||
SM2star(Slater_inv, Dim, 1, Updates_1block, Updates_index_1block,
|
||||
later_updates + (Dim * later), later_index + later, &l);
|
||||
later = later + l;
|
||||
}
|
||||
|
||||
if (later > 0) {
|
||||
SM2(Slater_inv, Dim, later, later_updates, later_index);
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
// Sherman-Morrison-Woodbury kernel 4
|
||||
// WB2, SM2 mixing scheme
|
||||
void SMWB4(double *Slater_inv, const unsigned int Dim, const unsigned int N_updates, double *Updates, unsigned int *Updates_index) {
|
||||
void SMWB4(double *Slater_inv, const unsigned int Dim,
|
||||
const unsigned int N_updates, double *Updates,
|
||||
unsigned int *Updates_index) {
|
||||
#ifdef DEBUG2
|
||||
std::cerr << "Called Sherman-Morrison-Woodbury kernel 1 with " << N_updates << " updates" << std::endl;
|
||||
std::cerr << "Called Sherman-Morrison-Woodbury kernel 1 with " << N_updates
|
||||
<< " updates" << std::endl;
|
||||
showMatrix2(Updates_index, 1, N_updates, "Updates_index");
|
||||
showMatrix2(Updates, N_updates, Dim, "Updates");
|
||||
#endif
|
||||
@ -85,7 +97,8 @@ void SMWB4(double *Slater_inv, const unsigned int Dim, const unsigned int N_upda
|
||||
unsigned int length_2block = 2 * Dim;
|
||||
unsigned int length_1block = 1 * Dim;
|
||||
|
||||
// Apply first 2*n_of_2blocks updates in n_of_2blocks blocks of 2 updates with Woodbury 2x2 kernel
|
||||
// Apply first 2*n_of_2blocks updates in n_of_2blocks blocks of 2 updates with
|
||||
// Woodbury 2x2 kernel
|
||||
double later_updates[Dim * N_updates];
|
||||
unsigned int later_index[N_updates];
|
||||
unsigned int later = 0;
|
||||
@ -97,14 +110,16 @@ void SMWB4(double *Slater_inv, const unsigned int Dim, const unsigned int N_upda
|
||||
ok = WB2(Slater_inv, Dim, Updates_2block, Updates_index_2block);
|
||||
if (!ok) { // Send the entire block to SM2
|
||||
#ifdef DEBUG1
|
||||
std::cerr << "Woodbury 2x2 kernel failed! Sending block to SM2star" << std::endl;
|
||||
std::cerr << "Woodbury 2x2 kernel failed! Sending block to SM2star"
|
||||
<< std::endl;
|
||||
#endif
|
||||
#ifdef DEBUG2
|
||||
showMatrix2(Updates_2block, 2, Dim, "Updates_2block");
|
||||
showMatrix2(Updates_index_2block, 1, 2, "Updates_index_2block");
|
||||
#endif
|
||||
unsigned int l = 0;
|
||||
SM2star(Slater_inv, Dim, 2, Updates_2block, Updates_index_2block, later_updates + (Dim * later), later_index + later, &l);
|
||||
SM2star(Slater_inv, Dim, 2, Updates_2block, Updates_index_2block,
|
||||
later_updates + (Dim * later), later_index + later, &l);
|
||||
later = later + l;
|
||||
}
|
||||
}
|
||||
@ -114,14 +129,14 @@ void SMWB4(double *Slater_inv, const unsigned int Dim, const unsigned int N_upda
|
||||
double *Updates_1block = &Updates[n_of_2blocks * length_2block];
|
||||
unsigned int *Updates_index_1block = &Updates_index[2 * n_of_2blocks];
|
||||
unsigned int l = 0;
|
||||
SM2star(Slater_inv, Dim, 1, Updates_1block, Updates_index_1block, later_updates + (Dim * later), later_index + later, &l);
|
||||
SM2star(Slater_inv, Dim, 1, Updates_1block, Updates_index_1block,
|
||||
later_updates + (Dim * later), later_index + later, &l);
|
||||
later = later + l;
|
||||
}
|
||||
|
||||
if (later > 0) {
|
||||
SM2(Slater_inv, Dim, later, later_updates, later_index);
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
extern "C" {
|
||||
|
@ -123,7 +123,9 @@ void SM2(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
// Sherman Morrison, with J. Slagel splitting
|
||||
// http://hdl.handle.net/10919/52966
|
||||
void SM2star(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
double *Updates, unsigned int *Updates_index, double *later_updates, unsigned int* later_index, unsigned int *later) {
|
||||
double *Updates, unsigned int *Updates_index,
|
||||
double *later_updates, unsigned int *later_index,
|
||||
unsigned int *later) {
|
||||
#ifdef DEBUG1
|
||||
std::cerr << "Called SM2* with " << N_updates << " updates" << std::endl;
|
||||
#endif
|
||||
@ -179,7 +181,6 @@ void SM2star(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
// Sherman Morrison, leaving zero denominators for later
|
||||
void SM3(double *Slater_inv, unsigned int Dim, unsigned int N_updates,
|
||||
double *Updates, unsigned int *Updates_index) {
|
||||
|
@ -52,7 +52,8 @@ bool WB2(double *Slater_inv, unsigned int Dim, double *Updates,
|
||||
double det = B[0] * B[3] - B[1] * B[2];
|
||||
if (std::fabs(det) < threshold()) {
|
||||
#ifdef DEBUG1
|
||||
std::cerr << "Determinant too close to zero! No inverse found." << std::endl;
|
||||
std::cerr << "Determinant too close to zero! No inverse found."
|
||||
<< std::endl;
|
||||
std::cerr << "Determinant = " << det << std::endl;
|
||||
#endif
|
||||
return false;
|
||||
@ -96,23 +97,10 @@ bool WB3(double *Slater_inv, unsigned int Dim, double *Updates,
|
||||
#ifdef DEBUG1
|
||||
std::cerr << "Called Woodbury 3x3 kernel" << std::endl;
|
||||
#endif
|
||||
#ifdef DEBUG2
|
||||
showMatrix2(Slater_inv, Dim, Dim, "Slater_inv BEFORE update");
|
||||
showMatrix2(Updates, 3, Dim, "Updates");
|
||||
showMatrix2(Updates_index, 1, 3, "Updates_index");
|
||||
#endif
|
||||
|
||||
// Compute D = V * S^{-1}
|
||||
double D[3 * Dim];
|
||||
unsigned int row1, row2, row3;
|
||||
row1 = Updates_index[0] - 1;
|
||||
row2 = Updates_index[1] - 1;
|
||||
row3 = Updates_index[2] - 1;
|
||||
for (unsigned int i = 0; i < Dim; i++) {
|
||||
D[i] = Slater_inv[row1 * Dim + i];
|
||||
D[Dim + i] = Slater_inv[row2 * Dim + i];
|
||||
D[2 * Dim + i] = Slater_inv[row3 * Dim + i];
|
||||
}
|
||||
const unsigned int row1 = (Updates_index[0] - 1);
|
||||
const unsigned int row2 = (Updates_index[1] - 1);
|
||||
const unsigned int row3 = (Updates_index[2] - 1);
|
||||
|
||||
// Compute C = S_inv * U !! NON-STANDARD MATRIX MULTIPLICATION BECAUSE
|
||||
// OF LAYOUT OF 'Updates' !!
|
||||
@ -128,9 +116,6 @@ bool WB3(double *Slater_inv, unsigned int Dim, double *Updates,
|
||||
|
||||
#ifdef DEBUG2
|
||||
showMatrix2(C, Dim, 3, "C = S_inv * U");
|
||||
#endif
|
||||
|
||||
#ifdef DEBUG2
|
||||
showMatrix2(D, 3, Dim, "D = V * S_inv");
|
||||
#endif
|
||||
|
||||
@ -154,14 +139,14 @@ bool WB3(double *Slater_inv, unsigned int Dim, double *Updates,
|
||||
// Check if determinant of B is not too close to zero
|
||||
double det;
|
||||
det = B[0] * (B[4] * B[8] - B[5] * B[7]) -
|
||||
B[1] * (B[3] * B[8] - B[5] * B[6]) +
|
||||
B[2] * (B[3] * B[7] - B[4] * B[6]);
|
||||
B[1] * (B[3] * B[8] - B[5] * B[6]) + B[2] * (B[3] * B[7] - B[4] * B[6]);
|
||||
#ifdef DEBUG2
|
||||
std::cerr << "Determinant of B = " << det << std::endl;
|
||||
#endif
|
||||
if (std::fabs(det) < threshold()) {
|
||||
#ifdef DEBUG1
|
||||
std::cerr << "Determinant too close to zero! No inverse found." << std::endl;
|
||||
std::cerr << "Determinant too close to zero! No inverse found."
|
||||
<< std::endl;
|
||||
std::cerr << "Determinant = " << det << std::endl;
|
||||
#endif
|
||||
return false;
|
||||
@ -180,29 +165,32 @@ bool WB3(double *Slater_inv, unsigned int Dim, double *Updates,
|
||||
Binv[8] = (B[0] * B[4] - B[3] * B[1]) * idet;
|
||||
|
||||
#ifdef DEBUG2
|
||||
std::cerr << "Conditioning number of B = " << condition1(B, Binv, 3) << std::endl;
|
||||
std::cerr << "Conditioning number of B = " << condition1(B, Binv, 3)
|
||||
<< std::endl;
|
||||
showMatrix2(Binv, 3, 3, "Binv");
|
||||
#endif
|
||||
|
||||
// Compute B^{-1} x D
|
||||
// Compute tmp = B^{-1} x (V.S^{-1})
|
||||
double tmp[3 * Dim];
|
||||
matMul2(Binv, D, tmp, 3, 3, Dim);
|
||||
for (unsigned int i = 0; i < 3; i++) {
|
||||
for (unsigned int j = 0; j < Dim; j++) {
|
||||
tmp[i * Dim + j] = Binv[i * 3] * Slater_inv[row1 * Dim + j];
|
||||
tmp[i * Dim + j] += Binv[i * 3 + 1] * Slater_inv[row2 * Dim + j];
|
||||
tmp[i * Dim + j] += Binv[i * 3 + 2] * Slater_inv[row3 * Dim + j];
|
||||
}
|
||||
}
|
||||
|
||||
#ifdef DEBUG2
|
||||
showMatrix2(tmp, 3, Dim, "tmp = Binv * D");
|
||||
#endif
|
||||
|
||||
// Compute C x B^{-1} x D
|
||||
double tmp2[Dim * Dim];
|
||||
matMul2(C, tmp, tmp2, Dim, 3, Dim);
|
||||
|
||||
#ifdef DEBUG2
|
||||
showMatrix2(tmp2, Dim, Dim, "tmp2 = C * tmp");
|
||||
#endif
|
||||
|
||||
// Compute (S + U V)^{-1} = S^{-1} - C B^{-1} D
|
||||
for (unsigned int i = 0; i < Dim * Dim; i++) {
|
||||
Slater_inv[i] -= tmp2[i];
|
||||
// Compute (S + U V)^{-1} = S^{-1} - C x tmp
|
||||
for (unsigned int i = 0; i < Dim; i++) {
|
||||
for (unsigned int j = 0; j < Dim; j++) {
|
||||
Slater_inv[i * Dim + j] -= C[i * 3] * tmp[j];
|
||||
Slater_inv[i * Dim + j] -= C[i * 3 + 1] * tmp[Dim + j];
|
||||
Slater_inv[i * Dim + j] -= C[i * 3 + 2] * tmp[2 * Dim + j];
|
||||
}
|
||||
}
|
||||
|
||||
#ifdef DEBUG2
|
||||
|
@ -2,10 +2,10 @@
|
||||
#include "hdf5/serial/hdf5.h"
|
||||
|
||||
#include "Helpers.hpp"
|
||||
#include "SMWB.hpp"
|
||||
#include "SM_Maponi.hpp"
|
||||
#include "SM_Standard.hpp"
|
||||
#include "Woodbury.hpp"
|
||||
#include "SMWB.hpp"
|
||||
#include <fstream>
|
||||
#include <vector>
|
||||
|
||||
@ -86,48 +86,57 @@ int test_cycle(H5File file, int cycle, std::string version, double tolerance) {
|
||||
double *slater_inverse_nonpersistent = new double[dim * dim];
|
||||
if (version == "sm1") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
SM1(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
}
|
||||
} else if (version == "sm2") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
SM2(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
}
|
||||
} else if (version == "sm3") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
SM3(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
}
|
||||
} else if (version == "sm4") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
SM4(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
}
|
||||
} else if (version == "wb2") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
WB2(slater_inverse_nonpersistent, dim, u, col_update_index);
|
||||
}
|
||||
} else if (version == "wb3") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
WB3(slater_inverse_nonpersistent, dim, u, col_update_index);
|
||||
}
|
||||
} else if (version == "smwb1") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
SMWB1(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
}
|
||||
} else if (version == "smwb4") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
SMWB4(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
}
|
||||
#ifdef MKL
|
||||
} else if (version == "lapack") {
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_matrix, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_matrix,
|
||||
dim * dim * sizeof(double));
|
||||
inverse(slater_inverse_nonpersistent, dim);
|
||||
}
|
||||
#endif // MKL
|
||||
@ -135,7 +144,8 @@ int test_cycle(H5File file, int cycle, std::string version, double tolerance) {
|
||||
std::cerr << "Unknown version " << version << std::endl;
|
||||
exit(1);
|
||||
}
|
||||
std::memcpy(slater_inverse, slater_inverse_nonpersistent, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse, slater_inverse_nonpersistent,
|
||||
dim * dim * sizeof(double));
|
||||
delete[] slater_inverse_nonpersistent;
|
||||
#else // No performance measurements repetition
|
||||
if (version == "maponia3") {
|
||||
@ -210,8 +220,7 @@ int main(int argc, char **argv) {
|
||||
#else
|
||||
if (argc != 4) {
|
||||
std::cerr << "Execute from within 'datasets/'" << std::endl;
|
||||
std::cerr
|
||||
<< "usage: test_h5 <version> <cycle file> <tolerance>"
|
||||
std::cerr << "usage: test_h5 <version> <cycle file> <tolerance>"
|
||||
<< std::endl;
|
||||
return 1;
|
||||
}
|
||||
@ -221,7 +230,8 @@ int main(int argc, char **argv) {
|
||||
std::ifstream cyclefile(cyclefile_name);
|
||||
std::vector<int> cycles;
|
||||
unsigned int cycle;
|
||||
while (cyclefile >> cycle) cycles.push_back(cycle);
|
||||
while (cyclefile >> cycle)
|
||||
cycles.push_back(cycle);
|
||||
double tolerance = std::stod(argv[3]);
|
||||
H5File file(FILE_NAME, H5F_ACC_RDONLY);
|
||||
|
||||
|
@ -2,10 +2,10 @@
|
||||
#include "hdf5/serial/hdf5.h"
|
||||
|
||||
#include "Helpers.hpp"
|
||||
#include "SMWB.hpp"
|
||||
#include "SM_Maponi.hpp"
|
||||
#include "SM_Standard.hpp"
|
||||
#include "Woodbury.hpp"
|
||||
#include "SMWB.hpp"
|
||||
|
||||
#define PERF
|
||||
|
||||
@ -82,11 +82,14 @@ int test_cycle(H5File file, int cycle, std::string version, double tolerance) {
|
||||
std::cout << "# of reps. = " << repetition_number << std::endl;
|
||||
double *slater_inverse_nonpersistent = new double[dim * dim];
|
||||
for (unsigned int i = 0; i < repetition_number; i++) {
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse_nonpersistent, slater_inverse,
|
||||
dim * dim * sizeof(double));
|
||||
if (version == "maponia3") {
|
||||
MaponiA3(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
MaponiA3(slater_inverse_nonpersistent, dim, nupdates, u,
|
||||
col_update_index);
|
||||
} else if (version == "maponia3s") {
|
||||
MaponiA3S(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
MaponiA3S(slater_inverse_nonpersistent, dim, nupdates, u,
|
||||
col_update_index);
|
||||
} else if (version == "sm1") {
|
||||
SM1(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
} else if (version == "sm2") {
|
||||
@ -102,14 +105,17 @@ int test_cycle(H5File file, int cycle, std::string version, double tolerance) {
|
||||
} else if (version == "smwb1") {
|
||||
SMWB1(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
// } else if (version == "smwb2") {
|
||||
// SMWB2(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
// SMWB2(slater_inverse_nonpersistent, dim, nupdates, u,
|
||||
// col_update_index);
|
||||
// } else if (version == "smwb3") {
|
||||
// SMWB3(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
// SMWB3(slater_inverse_nonpersistent, dim, nupdates, u,
|
||||
// col_update_index);
|
||||
} else if (version == "smwb4") {
|
||||
SMWB4(slater_inverse_nonpersistent, dim, nupdates, u, col_update_index);
|
||||
#ifdef MKL
|
||||
} else if (version == "lapack") {
|
||||
memcpy(slater_inverse_nonpersistent, slater_matrix, dim * dim * sizeof(double));
|
||||
memcpy(slater_inverse_nonpersistent, slater_matrix,
|
||||
dim * dim * sizeof(double));
|
||||
inverse(slater_inverse_nonpersistent, dim);
|
||||
#endif // MKL
|
||||
} else {
|
||||
@ -117,7 +123,8 @@ int test_cycle(H5File file, int cycle, std::string version, double tolerance) {
|
||||
exit(1);
|
||||
}
|
||||
}
|
||||
std::memcpy(slater_inverse, slater_inverse_nonpersistent, dim * dim * sizeof(double));
|
||||
std::memcpy(slater_inverse, slater_inverse_nonpersistent,
|
||||
dim * dim * sizeof(double));
|
||||
delete[] slater_inverse_nonpersistent;
|
||||
#else
|
||||
if (version == "maponia3") {
|
||||
|
@ -174,9 +174,7 @@ int main(int argc, char **argv) {
|
||||
} else {
|
||||
std::cerr << "failed -- cycle " << std::to_string(i) << std::endl;
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
vfc_dump_probes(&probes);
|
||||
|
@ -11,5 +11,5 @@ fi
|
||||
|
||||
for ext in c cc cpp h hpp
|
||||
do
|
||||
find $SMROOT -type f -iname "*.${ext}" -exec echo "$FORMATER $STYLE" {} \;
|
||||
find $SMROOT -type f -iname "*.${ext}" -exec $FORMATER $STYLE {} \;
|
||||
done
|
||||
|
Loading…
Reference in New Issue
Block a user