mirror of
https://github.com/SheffieldML/GPy.git
synced 2026-05-21 14:05:14 +02:00
Made openmp switch in only dependent on potion in rbf.py and linear.py
This commit is contained in:
Neil Lawrence
parent
169e81702d
commit
2256127130
7 changed files with 122 additions and 48 deletions
|
|
@ -12,6 +12,7 @@ from ...core.parameterization.transformations import Logexp
|
|||
from ...util.caching import Cache_this
|
||||
from ...core.parameterization import variational
|
||||
from psi_comp import linear_psi_comp
|
||||
from ...util.config import *
|
||||
|
||||
class Linear(Kern):
|
||||
"""
|
||||
|
|
@ -224,12 +225,23 @@ class Linear(Kern):
|
|||
AZZA = ZA.T[:, None, :, None] * ZA[None, :, None, :]
|
||||
AZZA = AZZA + AZZA.swapaxes(1, 2)
|
||||
AZZA_2 = AZZA/2.
|
||||
if config.getboolean('parallel', 'openmp'):
|
||||
pragma_string = '#pragma omp parallel for private(m,mm,q,qq,factor,tmp)'
|
||||
header_string = '#include <omp.h>'
|
||||
weave_options = {'headers' : ['<omp.h>'],
|
||||
'extra_compile_args': ['-fopenmp -O3'],
|
||||
'extra_link_args' : ['-lgomp'],
|
||||
'libraries': ['gomp']}
|
||||
else:
|
||||
pragma_string = ''
|
||||
header_string = ''
|
||||
weave_options = {'extra_compile_args': ['-O3']}
|
||||
|
||||
#Using weave, we can exploit the symmetry of this problem:
|
||||
code = """
|
||||
int n, m, mm,q,qq;
|
||||
double factor,tmp;
|
||||
#pragma omp parallel for private(m,mm,q,qq,factor,tmp)
|
||||
%s
|
||||
for(n=0;n<N;n++){
|
||||
for(m=0;m<num_inducing;m++){
|
||||
for(mm=0;mm<=m;mm++){
|
||||
|
|
@ -253,26 +265,36 @@ class Linear(Kern):
|
|||
}
|
||||
}
|
||||
}
|
||||
"""
|
||||
""" % pragma_string
|
||||
support_code = """
|
||||
#include <omp.h>
|
||||
%s
|
||||
#include <math.h>
|
||||
"""
|
||||
weave_options = {'headers' : ['<omp.h>'],
|
||||
'extra_compile_args': ['-fopenmp -O3'], #-march=native'],
|
||||
'extra_link_args' : ['-lgomp']}
|
||||
""" % header_string
|
||||
mu = vp.mean
|
||||
N,num_inducing,input_dim,mu = mu.shape[0],Z.shape[0],mu.shape[1],param_to_array(mu)
|
||||
weave.inline(code, support_code=support_code, libraries=['gomp'],
|
||||
weave.inline(code, support_code=support_code,
|
||||
arg_names=['N','num_inducing','input_dim','mu','AZZA','AZZA_2','target_mu','target_S','dL_dpsi2'],
|
||||
type_converters=weave.converters.blitz,**weave_options)
|
||||
|
||||
|
||||
def _weave_dpsi2_dZ(self, dL_dpsi2, Z, vp, target):
|
||||
AZA = self.variances*self._ZAinner(vp, Z)
|
||||
|
||||
if config.getboolean('parallel', 'openmp'):
|
||||
pragma_string = '#pragma omp parallel for private(n,mm,q)'
|
||||
header_string = '#include <omp.h>'
|
||||
weave_options = {'headers' : ['<omp.h>'],
|
||||
'extra_compile_args': ['-fopenmp -O3'],
|
||||
'extra_link_args' : ['-lgomp'],
|
||||
'libraries': ['gomp']}
|
||||
else:
|
||||
pragma_string = ''
|
||||
header_string = ''
|
||||
weave_options = {'extra_compile_args': ['-O3']}
|
||||
|
||||
code="""
|
||||
int n,m,mm,q;
|
||||
#pragma omp parallel for private(n,mm,q)
|
||||
%s
|
||||
for(m=0;m<num_inducing;m++){
|
||||
for(q=0;q<input_dim;q++){
|
||||
for(mm=0;mm<num_inducing;mm++){
|
||||
|
|
@ -282,18 +304,15 @@ class Linear(Kern):
|
|||
}
|
||||
}
|
||||
}
|
||||
"""
|
||||
""" % pragma_string
|
||||
support_code = """
|
||||
#include <omp.h>
|
||||
%s
|
||||
#include <math.h>
|
||||
"""
|
||||
weave_options = {'headers' : ['<omp.h>'],
|
||||
'extra_compile_args': ['-fopenmp -O3'], #-march=native'],
|
||||
'extra_link_args' : ['-lgomp']}
|
||||
""" % header_string
|
||||
|
||||
N,num_inducing,input_dim = vp.mean.shape[0],Z.shape[0],vp.mean.shape[1]
|
||||
mu = param_to_array(vp.mean)
|
||||
weave.inline(code, support_code=support_code, libraries=['gomp'],
|
||||
weave.inline(code, support_code=support_code,
|
||||
arg_names=['N','num_inducing','input_dim','AZA','target','dL_dpsi2'],
|
||||
type_converters=weave.converters.blitz,**weave_options)
|
||||
|
||||
|
|
|
|||
|
|
@ -10,6 +10,7 @@ from GPy.util.caching import Cache_this
|
|||
from ...core.parameterization import variational
|
||||
from psi_comp import ssrbf_psi_comp
|
||||
from psi_comp.ssrbf_psi_gpucomp import PSICOMP_SSRBF
|
||||
from ...util.config import *
|
||||
|
||||
class RBF(Stationary):
|
||||
"""
|
||||
|
|
@ -231,6 +232,16 @@ class RBF(Stationary):
|
|||
|
||||
@Cache_this(limit=1)
|
||||
def _psi2computations(self, Z, vp):
|
||||
|
||||
if config.getboolean('parallel', 'openmp'):
|
||||
pragma_string = '#pragma omp parallel for private(tmp, exponent_tmp)'
|
||||
header_string = '#include <omp.h>'
|
||||
libraries = ['gomp']
|
||||
else:
|
||||
pragma_string = ''
|
||||
header_string = ''
|
||||
libraries = []
|
||||
|
||||
mu, S = vp.mean, vp.variance
|
||||
|
||||
N, Q = mu.shape
|
||||
|
|
@ -253,8 +264,7 @@ class RBF(Stationary):
|
|||
variance_sq = float(np.square(self.variance))
|
||||
code = """
|
||||
double tmp, exponent_tmp;
|
||||
|
||||
#pragma omp parallel for private(tmp, exponent_tmp)
|
||||
%s
|
||||
for (int n=0; n<N; n++)
|
||||
{
|
||||
for (int m=0; m<M; m++)
|
||||
|
|
@ -278,20 +288,20 @@ class RBF(Stationary):
|
|||
tmp = -Zdist_sq(m,mm,q) - tmp - half_log_denom(n,q);
|
||||
exponent_tmp += tmp;
|
||||
}
|
||||
//compute psi2 by exponontiating
|
||||
//compute psi2 by exponentiating
|
||||
psi2(n,m,mm) = variance_sq * exp(exponent_tmp);
|
||||
psi2(n,mm,m) = psi2(n,m,mm);
|
||||
}
|
||||
}
|
||||
}
|
||||
"""
|
||||
""" % pragma_string
|
||||
|
||||
support_code = """
|
||||
#include <omp.h>
|
||||
%s
|
||||
#include <math.h>
|
||||
"""
|
||||
""" % header_string
|
||||
mu = param_to_array(mu)
|
||||
weave.inline(code, support_code=support_code, libraries=['gomp'],
|
||||
weave.inline(code, support_code=support_code, libraries=libraries,
|
||||
arg_names=['N', 'M', 'Q', 'mu', 'Zhat', 'mudist_sq', 'mudist', 'denom_l2', 'Zdist_sq', 'half_log_denom', 'psi2', 'variance_sq'],
|
||||
type_converters=weave.converters.blitz, **self.weave_options)
|
||||
|
||||
|
|
@ -303,12 +313,20 @@ class RBF(Stationary):
|
|||
#return 2.*np.einsum( 'ijk,ijk,ijkl,il->l', dL_dpsi2, psi2, Zdist_sq * (2.*S[:,None,None,:]/l2 + 1.) + mudist_sq + S[:, None, None, :] / l2, 1./(2.*S + l2))*self.lengthscale
|
||||
|
||||
result = np.zeros(self.input_dim)
|
||||
if config.getboolean('parallel', 'openmp'):
|
||||
pragma_string = '#pragma omp parallel for reduction(+:tmp)'
|
||||
header_string = '#include <omp.h>'
|
||||
libraries = ['gomp']
|
||||
else:
|
||||
pragma_string = ''
|
||||
header_string = ''
|
||||
libraries = []
|
||||
code = """
|
||||
double tmp;
|
||||
for(int q=0; q<Q; q++)
|
||||
{
|
||||
tmp = 0.0;
|
||||
#pragma omp parallel for reduction(+:tmp)
|
||||
%s
|
||||
for(int n=0; n<N; n++)
|
||||
{
|
||||
for(int m=0; m<M; m++)
|
||||
|
|
@ -326,16 +344,16 @@ class RBF(Stationary):
|
|||
result(q) = tmp;
|
||||
}
|
||||
|
||||
"""
|
||||
""" % pragma_string
|
||||
support_code = """
|
||||
#include <omp.h>
|
||||
%s
|
||||
#include <math.h>
|
||||
"""
|
||||
""" % header_string
|
||||
N,Q = S.shape
|
||||
M = psi2.shape[-1]
|
||||
|
||||
S = param_to_array(S)
|
||||
weave.inline(code, support_code=support_code, libraries=['gomp'],
|
||||
weave.inline(code, support_code=support_code, libraries=libraries,
|
||||
arg_names=['psi2', 'dL_dpsi2', 'N', 'M', 'Q', 'mudist_sq', 'l2', 'Zdist_sq', 'S', 'result'],
|
||||
type_converters=weave.converters.blitz, **self.weave_options)
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue