Add support for Python3.5, More bispectrum work

This commit is contained in:
Guilhem Lavaux 2017-05-15 16:03:17 +02:00
parent 23a8b07229
commit 019480c0e0
6 changed files with 172 additions and 130 deletions

View File

@ -65,10 +65,10 @@ endif()
# Discover where to put packages
if (NOT PYTHON_SITE_PACKAGES)
execute_process (COMMAND ${PYTHON_EXECUTABLE} -c "from distutils.sysconfig import get_python_lib; print get_python_lib()" OUTPUT_VARIABLE internal_PYTHON_SITE_PACKAGES OUTPUT_STRIP_TRAILING_WHITESPACE)
execute_process (COMMAND ${PYTHON_EXECUTABLE} -c "from distutils.sysconfig import get_python_lib; print(get_python_lib())" OUTPUT_VARIABLE internal_PYTHON_SITE_PACKAGES OUTPUT_STRIP_TRAILING_WHITESPACE)
SET(SYSTEM_PYTHON_SITE_PACKAGES ${internal_PYTHON_SITE_PACKAGES} CACHE PATH "Path to the target system-wide site-package where to install python modules")
execute_process (COMMAND ${PYTHON_EXECUTABLE} -c "from site import USER_SITE; print USER_SITE" OUTPUT_VARIABLE internal_PYTHON_SITE_PACKAGES OUTPUT_STRIP_TRAILING_WHITESPACE)
execute_process (COMMAND ${PYTHON_EXECUTABLE} -c "from site import USER_SITE; print(USER_SITE)" OUTPUT_VARIABLE internal_PYTHON_SITE_PACKAGES OUTPUT_STRIP_TRAILING_WHITESPACE)
SET(USER_PYTHON_SITE_PACKAGES ${internal_PYTHON_SITE_PACKAGES} CACHE PATH "Path to the target user site-package where to install python modules")
mark_as_advanced(USER_PYTHON_SITE_PACKAGES SYSTEM_PYTHON_SITE_PACKAGES)

View File

@ -4,54 +4,31 @@
#include <boost/multi_array.hpp>
#include <sys/types.h>
#include <cmath>
#include "symbol_visible.hpp"
#include "algo.hpp"
using std::cout;
using std::endl;
using boost::format;
#if defined _WIN32 || defined __CYGWIN__
#ifdef BUILDING_DLL
#ifdef __GNUC__
#define DLL_PUBLIC __attribute__ ((dllexport))
#else
#define DLL_PUBLIC __declspec(dllexport) // Note: actually gcc seems to also supports this syntax.
#endif
#else
#ifdef __GNUC__
#define DLL_PUBLIC __attribute__ ((dllimport))
#else
#define DLL_PUBLIC __declspec(dllimport) // Note: actually gcc seems to also supports this syntax.
#endif
#endif
#define DLL_LOCAL
#else
#if __GNUC__ >= 4
#define DLL_PUBLIC __attribute__ ((visibility ("default")))
#define DLL_LOCAL __attribute__ ((visibility ("hidden")))
#else
#define DLL_PUBLIC
#define DLL_LOCAL
#endif
#endif
using CosmoTool::square;
struct ModeSet
{
size_t N1, N2, N3;
bool omp;
ssize_t N1, N2, N3;
bool half_copy;
struct TriangleIterator
{
ssize_t i1, i2, i3;
size_t N1, N2, N3;
ssize_t N1, N2, N3;
ssize_t first_iteration;
TriangleIterator& operator++() {
i3++;
if (i3==N3) { i3 = 0; i2++; }
if (i2==N2) { i2 = 0; i1++; }
if (i3==(N3/2+1)) { i3 = first_iteration; i2++; }
if (i2==(N2/2+1)) { i2 = -N2/2; i1++; }
return *this;
}
@ -62,7 +39,7 @@ struct ModeSet
bool in_box() const {
ssize_t hN1 = N1/2, hN2 = N2/2, hN3 = N3/2;
return (i1 >= -hN1) && (i1 < hN1) && (i2 >= -hN2) && (i2 < hN2) && (i3 >= -hN3) && (i3 < hN3);
return (i1 >= -hN1) && (i1 <= hN1) && (i2 >= -hN2) && (i2 <= hN2) && (i3 >= -hN3) && (i3 <= hN3);
}
TriangleIterator operator+(const TriangleIterator& other_t) const {
@ -74,6 +51,23 @@ struct ModeSet
return t;
}
TriangleIterator& inp_array() {
if (i1 < 0)
i1 += N1;
if (i2 < 0)
i2 += N2;
if (i3 < 0)
i3 += N3;
return *this;
}
TriangleIterator array() const {
TriangleIterator t = *this;
t.inp_array();
return t;
}
TriangleIterator real() const {
TriangleIterator t = *this;
if (t.i1 >= N1/2)
@ -86,7 +80,7 @@ struct ModeSet
}
double norm() const {
double r1 = i1, r2 = i3, r3 = i3;
double r1 = i1, r2 = i2, r3 = i3;
return std::sqrt(r1*r1 + r2*r2 + r3*r3);
}
void reverse() { i1=-i1; i2=-i2; i3=-i3; }
@ -94,23 +88,30 @@ struct ModeSet
TriangleIterator& operator*() { return *this; }
};
ModeSet(size_t N1_, size_t N2_, size_t N3_, bool do_openmp = false)
: N1(N1_), N2(N2_), N3(N3_),omp(do_openmp) {
ModeSet(size_t N1_, size_t N2_, size_t N3_, bool _half_copy = false)
: N1(N1_), N2(N2_), N3(N3_),half_copy(_half_copy) {
}
TriangleIterator begin() const {
TriangleIterator t;
t.i1 = t.i2 = t.i3 = 0;
t.i1 = -N1/2;
t.i2 = -N2/2;
if (half_copy)
t.first_iteration = t.i3 = 0;
else
t.first_iteration = t.i3 = -N3/2;
t.N1 = N1;
t.N2 = N2;
t.N3 = N3;
t.N1 = N1;
return t;
}
TriangleIterator end() const {
TriangleIterator t;
t.i2 = t.i3 = 0;
t.i1 = N1;
t.first_iteration = (half_copy ? 0 : (-N3/2));
t.i3 = t.first_iteration;
t.i2 = -N2/2;
t.i1 = N1/2+1;
t.N1 = N1;
t.N2 = N2;
t.N3 = N3;
@ -119,10 +120,50 @@ struct ModeSet
};
std::ostream& operator<<(std::ostream& o, const ModeSet::TriangleIterator& t)
{
o << t.i1 << "," << t.i2 << "," << t.i3;
return o;
}
template<typename T>
static T no_conj(const T& a) { return a; }
extern "C" DLL_PUBLIC
template<typename SubArrayB,typename SubArrayCnt,typename Delta>
static inline void accum_bispec(const Delta& delta_mirror, SubArrayB b_Nt, SubArrayCnt b_B,
const typename Delta::element& v1, const typename Delta::element& v2,
const ModeSet::TriangleIterator& rm1,
const ModeSet::TriangleIterator& rm2,
const ModeSet::TriangleIterator& rm3,
double delta_k,
size_t Nk)
{
typedef std::complex<double> CType;
size_t q1 = std::floor(rm1.norm()/delta_k);
if (q1 >= Nk)
return;
size_t q2 = std::floor(rm2.norm()/delta_k);
if (q2 >= Nk)
return;
size_t q3 = std::floor(rm3.norm()/delta_k);
if (q3 >= Nk)
return;
CType prod = v1*v2;
ModeSet::TriangleIterator m3 = rm3;
// We use hermitic symmetry to get -m3, it is just the mode in m3 but conjugated.
m3.reverse();
m3.inp_array();
prod *= delta_mirror[m3.i1][m3.i2][m3.i3];
b_Nt[q1][q2][q3] ++;
b_B[q1][q2][q3] += prod;
}
extern "C" CTOOL_DLL_PUBLIC
void CosmoTool_compute_bispectrum(
double *delta_hat, size_t Nx, size_t Ny, size_t Nz,
size_t *Ntriangles,
@ -137,71 +178,68 @@ void CosmoTool_compute_bispectrum(
boost::multi_array<std::complex<double>, 4> b_B(boost::extents[Ntasks][Nk][Nk][Nk]);
boost::multi_array<size_t, 4> b_Nt(boost::extents[Ntasks][Nk][Nk][Nk]);
typedef std::complex<double> CType;
boost::multi_array<std::complex<double>, 3> delta_mirror(boost::extents[Nx][Ny][Nz]);
// Add hermiticity
for (auto m : ModeSet(Nx, Ny, Nz, true)) {
auto n1 = m;
auto n2 = m.array();
n1.reverse();
n1.inp_array();
delta_mirror[n2.i1][n2.i2][n2.i3] = (a_delta[n2.i1][n2.i2][n2.i3]);
delta_mirror[n1.i1][n1.i2][n1.i3] = std::conj(delta_mirror[n2.i1][n2.i2][n2.i3]);
}
// First loop over m1
#pragma omp parallel
{
{
#pragma omp single
{
for (auto m1 : ModeSet(Nx, Ny, Nz, true)) {
int tid = omp_get_thread_num();
{
for (auto m1 : ModeSet(Nx, Ny, Nz)) {
auto am1 = m1.array();
CType v1 = delta_mirror[am1.i1][am1.i2][am1.i3];
int tid = omp_get_thread_num();
#pragma omp task
{
CType v1 = (m1.i3 >= kNz) ? std::conj(a_delta[m1.i1][m1.i2][(m1.N3-m1.i3)%m1.N3]) : a_delta[m1.i1][m1.i2][m1.i3];
{
auto rm1 = m1.real();
// Second mode m2
for (auto m2 : ModeSet(Nx, Ny, Nz)) {
// Now derive m3
CType v2 = (m2.i3 >= kNz) ? std::conj(a_delta[m2.i1][m2.i2][(m2.N3-m2.i3)%m2.N3]) : a_delta[m2.i1][m2.i2][m2.i3];
auto rm2 = m2.real();
auto m3 = (rm1+rm2);
auto am2 = m2.array();
auto m3 = (m1+m2);
CType v2 = delta_mirror[am2.i1][am2.i2][am2.i3];
// Not in Fourier box, stop here
if (!m3.in_box())
continue;
size_t q1 = std::floor(rm1.norm()/delta_k);
size_t q2 = std::floor(rm2.norm()/delta_k);
size_t q3 = std::floor(m3.norm()/delta_k);
accum_bispec(delta_mirror, b_Nt[tid], b_B[tid], v1, v2, m1, m2, m3, delta_k, Nk);
if (q1 >= Nk || q2 >= Nk || q3 >= Nk)
continue;
CType prod = v1*v2;
bool do_conj = false;
// We use hermitic symmetry to get -m3, it is just the mode in m3 but conjugated.
m3.reverse();
if (m3.i3 > 0) {
assert(m3.i3 < kNz);
} else {
m3.i3 = -m3.i3;
do_conj = !do_conj;
}
m3.i1 = (m3.N1 - m1.i1)%m3.N1;
m3.i2 = (m3.N2 - m3.i2)%m3.N2;
}
if (do_conj)
prod *= std::conj(a_delta[m3.i1][m3.i2][m3.i3]);
else
prod *= (a_delta[m3.i1][m3.i2][m3.i3]);
}
}
}
b_Nt[tid][q1][q2][q3] ++;
b_B[tid][q1][q2][q3] += prod;
#pragma omp taskwait
for (int tid = 0; tid < Ntasks; tid++) {
size_t *b_p = b_Nt[tid].origin();
size_t *a_p = a_Nt.data();
std::complex<double> *b_B_p = b_B[tid].origin();
std::complex<double> *a_B_p = a_B.origin();
//#pragma omp simd
#pragma omp parallel for
for (size_t q = 0; q < Nk*Nk*Nk; q++) {
a_p[q] += b_p[q];
a_B_p[q] += b_B_p[q];
}
}
}
}
}
for (int tid = 0; tid < Ntasks; tid++)
for (auto m1 : ModeSet(Nk, Nk, Nk)) {
a_Nt[m1.i1][m1.i2][m1.i3] += b_Nt[tid][m1.i1][m1.i2][m1.i3];
a_B[m1.i1][m1.i2][m1.i3] += b_B[tid][m1.i1][m1.i2][m1.i3];
}
}
extern "C" DLL_PUBLIC
extern "C" CTOOL_DLL_PUBLIC
void CosmoTool_compute_powerspectrum(
double *delta_hat, size_t Nx, size_t Ny, size_t Nz,
size_t *Ncounts,
@ -215,10 +253,11 @@ void CosmoTool_compute_powerspectrum(
typedef std::complex<double> CType;
// First loop over m1
for (auto m1 : ModeSet(Nx, Ny, kNz)) {
for (auto m : ModeSet(Nx, Ny, kNz)) {
auto m1 = m.array();
CType& v1 = a_delta[m1.i1][m1.i2][m1.i3];
size_t q1 = std::floor(m1.norm()/delta_k);
size_t q1 = std::floor(m.norm()/delta_k);
if (q1 >= Nk)
continue;

View File

@ -1,8 +1,8 @@
from _cosmotool import *
from _project import *
from _cosmo_power import *
from _cosmo_cic import *
from _fast_interp import *
from ._cosmotool import *
from ._project import *
from ._cosmo_power import *
from ._cosmo_cic import *
from ._fast_interp import *
from .grafic import writeGrafic, writeWhitePhase, readGrafic, readWhitePhase
from .borg import read_borg_vol
from .cic import cicParticles

View File

@ -124,4 +124,4 @@ if __name__=="__main__":
delta[3,2,1]=1
b = powerspectrum(delta, 1, 16, fourier=False)
a = bispectrum(delta, 1, 16, fourier=False)
print a[0].max()
print(a[0].max())

View File

@ -26,7 +26,6 @@ def build_filelist(fdir):
fname_0=fname_0+glob.glob(fd+'initial_density_*')
fname_1=fname_1+glob.glob(fd+'final_density_*')
return fname_0, fname_1
def read_borg_vol(BORGFILE):
@ -57,13 +56,14 @@ def read_borg_vol(BORGFILE):
if size(r)==5 :
if r[0] =="define":
if r[1]=="Lattice" : N0=int(r[2])
if r[1]=="Lattice" :
N0=int(r[2])
N1=int(r[3])
N2=int(r[4])
if size(r)==11 :
if r[4] =="BoundingBox": xmin=float(r[5])
if r[4] =="BoundingBox":
xmin=float(r[5])
xmax=float(r[6])
ymin=float(r[7])
ymax=float(r[8])

View File

@ -3,23 +3,26 @@ import numpy as np
import cosmotool as ct
def myfun(N):
f=0.10
d=np.random.normal(size=(N,)*3)
f=0.001
L = 1.0
d=np.random.normal(size=(N,)*3) * np.sqrt(float(N))**3 / L**3
rho = d + f *(d*d - np.average(d*d))
delta = (L/N)**3
B = ct.bispectrum(rho, 1, N, fourier=False)
P = ct.powerspectrum(rho, 1, N, fourier=False)
PP = P[1]/P[0]/N**3
B = ct.bispectrum(rho * delta, 1, N, fourier=False)
P = ct.powerspectrum(rho * delta, 1, N, fourier=False)
PP = P[1]/P[0] / L**3
x = PP[:,None,None] * PP[None,:,None] + PP[:,None,None]*PP[None,None,:] + PP[None,:,None]*PP[None,None,:]
BB = B[1]/B[0]/N**6
BB = B[1]/B[0] / L**3
y = BB/x
np.savez("bispec_%d.npz" % N, y=y, B_nt=B[0], B_r=B[1], P_n=P[0], P=P[1], rho=rho);
np.savez("bispec_%d.npz" % N, x=x, y=y, d=d,B_nt=B[0], B_r=B[1], P_n=P[0], P=P[1], BB=BB, rho=rho, PP=PP);
print( timeit.timeit('from __main__ import myfun; myfun(16)', number=1) )
#print( timeit.timeit('from __main__ import myfun; myfun(16)', number=1) )
#print( timeit.timeit('from __main__ import myfun; myfun(24)', number=1) )
print( timeit.timeit('from __main__ import myfun; myfun(32)', number=1) )
print( timeit.timeit('from __main__ import myfun; myfun(64)', number=1) )
#print( timeit.timeit('from __main__ import myfun; myfun(64)', number=1) )