From e8a0dbacd60556adea7287f2b97ddd117503c576 Mon Sep 17 00:00:00 2001
From: Chichi Lalescu <Cristian.Lalescu@ds.mpg.de>
Date: Sat, 26 Dec 2015 17:55:58 +0100
Subject: [PATCH] preliminary optimization of interpolation
---
bfps/cpp/rFFTW_interpolator.cpp | 48 +++++++++------------------------
bfps/cpp/rFFTW_interpolator.hpp | 7 ++---
done.txt | 2 ++
todo.txt | 1 -
4 files changed, 19 insertions(+), 39 deletions(-)
diff --git a/bfps/cpp/rFFTW_interpolator.cpp b/bfps/cpp/rFFTW_interpolator.cpp
index be8546f6..8310be04 100644
--- a/bfps/cpp/rFFTW_interpolator.cpp
+++ b/bfps/cpp/rFFTW_interpolator.cpp
@@ -43,39 +43,17 @@ rFFTW_interpolator<rnumber, interp_neighbours>::rFFTW_interpolator(
this->field = (rnumber*)((void*)fftw_alloc_real(this->field_size));
// compute dx, dy, dz;
- this->dx = 4*acos(0) / (fs->dkx*fs->rd->sizes[2]);
- this->dy = 4*acos(0) / (fs->dky*fs->rd->sizes[1]);
- this->dz = 4*acos(0) / (fs->dkz*fs->rd->sizes[0]);
+ this->dx = 4*acos(0) / (fs->dkx*this->descriptor->sizes[2]);
+ this->dy = 4*acos(0) / (fs->dky*this->descriptor->sizes[1]);
+ this->dz = 4*acos(0) / (fs->dkz*this->descriptor->sizes[0]);
- // compute lower and upper bounds
- this->lbound = new double[nprocs];
- this->ubound = new double[nprocs];
- double *tbound = new double[nprocs];
- std::fill_n(tbound, nprocs, 0.0);
- tbound[this->descriptor->myrank] = fs->rd->starts[0]*this->dz;
- MPI_Allreduce(
- tbound,
- this->lbound,
- nprocs,
- MPI_DOUBLE,
- MPI_SUM,
- this->descriptor->comm);
- std::fill_n(tbound, this->descriptor->nprocs, 0.0);
- tbound[this->descriptor->myrank] = (fs->rd->starts[0] + fs->rd->subsizes[0])*this->dz;
- MPI_Allreduce(
- tbound,
- this->ubound,
- nprocs,
- MPI_DOUBLE,
- MPI_SUM,
- this->descriptor->comm);
- delete[] tbound;
- //for (int r = 0; r<nprocs; r++)
- // DEBUG_MSG(
- // "lbound[%d] = %lg, ubound[%d] = %lg\n",
- // r, this->lbound[r],
- // r, this->ubound[r]
- // );
+ // generate compute array
+ this->compute = new bool[this->descriptor->sizes[0]];
+ std::fill_n(this->compute, this->descriptor->sizes[0], false);
+ for (int iz = this->descriptor->starts[0]-interp_neighbours-1;
+ iz <= this->descriptor->starts[0]+this->descriptor->subsizes[0]+interp_neighbours+1;
+ iz++)
+ this->compute[((iz + this->descriptor->sizes[0]) % this->descriptor->sizes[0])] = true;
}
template <class rnumber, int interp_neighbours>
@@ -85,8 +63,7 @@ rFFTW_interpolator<rnumber, interp_neighbours>::~rFFTW_interpolator()
fftwf_free((float*)((void*)this->field));
else if (sizeof(rnumber) == 8)
fftw_free((double*)((void*)this->field));
- delete[] this->lbound;
- delete[] this->ubound;
+ delete[] this->compute;
}
template <class rnumber, int interp_neighbours>
@@ -139,7 +116,8 @@ void rFFTW_interpolator<rnumber, interp_neighbours>::sample(
this->get_grid_coordinates(nparticles, pdimension, x, xg, xx);
/* perform interpolation */
for (int p=0; p<nparticles; p++)
- this->operator()(xg + p*3, xx + p*3, yy + p*3, deriv);
+ if (this->compute[xg[p*3+2]])
+ this->operator()(xg + p*3, xx + p*3, yy + p*3, deriv);
MPI_Allreduce(
yy,
y,
diff --git a/bfps/cpp/rFFTW_interpolator.hpp b/bfps/cpp/rFFTW_interpolator.hpp
index 1b07bcb2..0864bd96 100644
--- a/bfps/cpp/rFFTW_interpolator.hpp
+++ b/bfps/cpp/rFFTW_interpolator.hpp
@@ -60,9 +60,10 @@ class rFFTW_interpolator
/* physical parameters of field */
double dx, dy, dz;
- /* precomputed boundaries of process's domain */
- double *lbound;
- double *ubound;
+ /* compute[iz] is true if .
+ * local_zstart - neighbours <= iz <= local_zend + 1 + neighbours
+ * */
+ bool *compute;
rFFTW_interpolator(
fluid_solver_base<rnumber> *FSOLVER,
diff --git a/done.txt b/done.txt
index 8dd974c1..344832ae 100644
--- a/done.txt
+++ b/done.txt
@@ -1,3 +1,5 @@
x 2015-12-04 make code py3 compatible @python3
x 2015-12-23 decide on versioning system +merge0
x 2015-12-24 move get_grid coords to interpolator @optimization +v1.0
+x 2015-12-25 get rid of temporal interpolation @optimization +v1.0
+x 2015-12-26 call interpolation only when needed @optimization +v1.0
diff --git a/todo.txt b/todo.txt
index dcc58366..0ec54254 100644
--- a/todo.txt
+++ b/todo.txt
@@ -1,6 +1,5 @@
(B) FFTW interpolator doesn't need its own field @optimization +v1.0
(B) compute z polynomials only when needed @optimization +v1.0
-(B) get rid of temporal interpolation @optimization +v1.0
(B) use less memory @optimization
(C) clean up machine_settings mess @design @documentation +v2.0
(C) code overview @documentation
--
GitLab