def run_ode( FORMULA, FUNCTIONS, INPUTS, inits, params, T , fs, inputs = None, decimate=1 ,variable_length = False, stochastic = False, Tterm = 0, gpu = False, nthreads = 4 , dtype = np.float32, debug = False,seed =None,threads_per_block=32 ):
from gpuODE import ode, param_grid, funcs2code
PARAMETERS = params.keys()
odeRK4 = ode("ode")
odeRK4.setup(FORMULA, PARAMETERS, INPUTS)
Tt = T + Tterm
dt = 1.0/np.array(fs)
N = np.int32(Tt*fs)
Nterm = np.int32(Tterm*fs)
extra_funcs = funcs2code(FUNCTIONS, gpu = gpu)
odeRK4.extra_func = extra_funcs
odeRK4.generate_code(debug=debug, gpu = gpu,variable_length = variable_length, stochastic=stochastic, dtype = dtype)
odeRK4.compile()
if gpu==True:
time, out = odeRK4.run(inits, param_grid(**params) , dt, decimate = decimate, inputs=inputs, N = N, Nterm = Nterm,seed = seed,threads_per_block=threads_per_block)
return time, out
else:
import distributed_exploration as de
import timeit
from random import randint
seed_off = 0
#Modifications are needed to change the seed on differnet threads
def func(**args):
seedr = seed + randint(0,10000)
time,out = odeRK4.run(inits, args, dt ,decimate=decimate, inputs=inputs, N = N , Nterm = Nterm,seed = seedr)
return time, out
tic = timeit.default_timer()
out0 = de.explore_thread(func, param_grid(**params) ,nthreads=nthreads)
toc = timeit.default_timer()
Nm = int(T*fs)/decimate
M = len(out0)
out = {}
for j,k in enumerate(FORMULA.keys()):
out[k] = np.zeros((Nm,M))
for i,o in enumerate(out0):
for j,k in enumerate(FORMULA.keys()):
out[k][:,i] = o['out'][1][k].T[0,:N]
return toc-tic, out
M = 32
mu = np.linspace(-1,10,M)
params = {'mu': mu}
Tterm = 0
T = 100 + Tterm
fs = 100
dt = 1/fs
N = int(T*fs)
Nterm = int(Tterm*fs)
t = np.arange(N-Nterm)/fs
pgrid = param_grid(**params)
time, out = vdp.run(x0, pgrid ,dt, N = N, Nterm = Nterm)
x = out['x']
print x.dtype
pl.figure(56)
pl.clf()
pl.plot( x[:,-1] ,'o-',label="float32")
vdp.generate_code(debug=False, stochastic = False, gpu = True, dtype = np.float64)
vdp.compile()
time, out = vdp.run(x0, pgrid ,dt, N = N, Nterm = Nterm)
x = out['x']
T = T + Tterm
dt = 1.0/float(fs)
N = np.int32(T*fs)
Nterm = np.int32(Tterm*fs).min()
dtype = np.float32
variable_length = False
extra_funcs = funcs2code(FUNCTION, gpu = gpu)
odeRK4.extra_func = extra_funcs
odeRK4.generate_code(debug=debug, gpu = gpu,variable_length = variable_length, stochastic=stochastic, dtype = dtype )
odeRK4.compile()
if gpu==True:
time, out = odeRK4.run(x0, param_grid(**params) , dt, decimate = decimate, inputs=inputs, N = N, Nterm = Nterm)
w = out['w']
else:
time,out = odeRK4.run(x0, param_grid(**params) , dt,decimate=decimate, inputs=inputs, N = N , Nterm = Nterm,seed = 1234)
w = out['w']
#
#
#pl.figure()
