def prepare_inputs(*inputs, device):
out = [bh.array(k) for k in inputs]
for o in out:
# force allocation on target device
tmp = o * 1 # noqa: F841
bh.flush()
return out
def gauss(a):
"""
Performe Gausian elimination on matrix a without pivoting
"""
dw = False
if dw:
def loop_body(a):
c = get_iterator(1)
x1 = a[c:, c - 1:]
# x2 = (a[c:, c - 1, None] / a[c - 1, c - 1:c, None])#[:, None]
x2 = a[c:, c - 1] / a[c - 1, c - 1:c]
# x2 = a[c:, c - 1] / a[c - 1, c - 1:c]
# x2 = (a[c:, c - 1] / a[c - 1, c - 1:c])[:, None]
x3 = a[c - 1, c - 1:]
x1 = x1 - x2 * x3
loop.do_while(loop_body, a.shape[0] - 2, a)
a /= np.diagonal(a)[:, None]
return a
else:
for c in range(1, a.shape[0]):
x1 = a[c:, c - 1:]
x2 = (a[c:, c - 1] / a[c - 1, c - 1:c])[:, None]
x3 = a[c - 1, c - 1:]
x1 = x1 - x2 * x3
# a[c:, c - 1:] = a[c:, c - 1:] - (a[c:, c - 1] / a[c - 1, c - 1:c]) * a[c - 1, c - 1:]
np.flush()
a /= np.diagonal(a)[:, None]
return a
def sign_is(z):
bh.flush()
x = np.real(z)
y = np.imag(z)
out = z / (np.sqrt(x*x+y*y)+(z==0))
bh.flush()
return out
def lu(a):
"""
Performe LU decomposition on the matrix a so A = L*U
"""
dw = True
if dw:
def loop_body(l, u):
c = get_iterator()
l[c:, c - 1] = (u[c:, c - 1] / u[c - 1, c - 1:c])
u[c:, c -
1:] = u[c:, c - 1:] - l[c:, c - 1][:, None] * u[c - 1, c - 1:]
u = a.copy()
l = np.zeros_like(a)
np.diagonal(l)[:] = 1.0
loop.do_while(loop_body, u.shape[0] - 2, l, u)
return (l, u)
else:
u = a.copy()
l = np.zeros_like(a)
np.diagonal(l)[:] = 1.0
for c in range(1, u.shape[0]):
l[c:, c - 1] = (u[c:, c - 1] / u[c - 1, c - 1:c])
u[c:, c -
1:] = u[c:, c - 1:] - l[c:, c - 1][:, None] * u[c - 1, c - 1:]
np.flush()
return (l, u)
def gauss(a):
"""
Performe Gausian elimination on matrix a without pivoting
"""
for c in xrange(1,a.shape[0]):
a[c:,c-1:] = a[c:,c-1:] - (a[c:,c-1]/a[c-1,c-1:c])[:,None] * a[c-1,c-1:]
np.flush(a)
a /= np.diagonal(a)[:,None]
return a
def gauss(a):
"""
Performe Gausian elimination on matrix a without pivoting
"""
for c in xrange(1,a.shape[0]):
a[c:,c-1:] = a[c:,c-1:] - (a[c:,c-1]/a[c-1,c-1:c])[:,None] * a[c-1,c-1:]
np.flush(a)
a /= np.diagonal(a)[:,None]
return a
def thesign(z):
bh.flush()
z_abs = np.absolute(z)
z_zero = (z==0)
divisor = z_abs + z_zero
del z_abs
del z_zero
sign = z / divisor
del divisor
bh.flush()
return sign
def lu(a):
"""
Performe LU decomposition on the matrix a so A = L*U
"""
u = a.copy()
l = np.zeros_like(a)
np.diagonal(l)[:] = 1.0
for c in xrange(1,u.shape[0]):
l[c:,c-1] = (u[c:,c-1]/u[c-1,c-1:c])
u[c:,c-1:] = u[c:,c-1:] - l[c:,c-1][:,None] * u[c-1,c-1:]
np.flush(u)
return (l,u)
def lu(a):
"""
Performe LU decomposition on the matrix a so A = L*U
"""
u = a.copy()
l = np.zeros_like(a)
np.diagonal(l)[:] = 1.0
for c in xrange(1,u.shape[0]):
l[c:,c-1] = (u[c:,c-1]/u[c-1,c-1:c])
u[c:,c-1:] = u[c:,c-1:] - l[c:,c-1][:,None] * u[c-1,c-1:]
np.flush(u)
return (l,u)
def simulate(m, x, y, z, vx, vy, vz, timesteps):
temporaries = (np.ones(
(size, size), dtype="float64"), np.ones((size, size), dtype="float64"),
np.ones((size, size), dtype="float64"),
np.ones((size, size), dtype="float64"))
start = time.time()
for i in range(timesteps):
ret = move(m, x, y, z, vx, vy, vz, dt, temporaries)
np.flush()
print(x, y, z)
end = time.time()
print("Simulation time:", end - start)
def solve(a, b):
"""
Solve a linear matrix equation, or system of linear scalar equations
using Gausian elimination.
:param a: Coefficient matrix
:type a: array_like, shape (M, M)
:param b: Ordinate or "dependent variable" values
:type b: array_like, shape (M,) or (M, N)
:return: Solution to the system a x = b
:rtype: ndarray, shape (M,) or (M, N) depending on b
:raises: :py:exc:`LinAlgError` If `a` is singular or not square.
**Examples:**
Solve the system of equations ``3 * x0 + x1 = 9`` and ``x0 + 2 * x1 = 8``:
>>> import bohrium as np
>>> a = np.array([[3.,1.], [1.,2.]])
>>> b = np.array([9.,8.])
>>> x = np.linalg.solve(a, b)
>>> x
array([ 2., 3.])
Check that the solution is correct:
>>> (np.dot(a, x) == b).all()
True
"""
if not (len(a.shape) == 2 and a.shape[0] == a.shape[1]):
raise la.LinAlgError("a is not square")
w = gauss(np.hstack((a,b[:,np.newaxis])))
lc = w.shape[1]-1
x = w[:,lc].copy()
for c in xrange(lc-1,0,-1):
x[:c] -= w[:c,c] * x[c:c+1]
np.flush(x)
return x
def solve(a, b):
"""
Solve a linear matrix equation, or system of linear scalar equations
using Gausian elimination.
:param a: Coefficient matrix
:type a: array_like, shape (M, M)
:param b: Ordinate or "dependent variable" values
:type b: array_like, shape (M,) or (M, N)
:return: Solution to the system a x = b
:rtype: ndarray, shape (M,) or (M, N) depending on b
:raises: :py:exc:`LinAlgError` If `a` is singular or not square.
**Examples:**
Solve the system of equations ``3 * x0 + x1 = 9`` and ``x0 + 2 * x1 = 8``:
>>> import bohrium as np
>>> a = np.array([[3.,1.], [1.,2.]])
>>> b = np.array([9.,8.])
>>> x = np.linalg.solve(a, b)
>>> x
array([ 2., 3.])
Check that the solution is correct:
>>> (np.dot(a, x) == b).all()
True
"""
if not (len(a.shape) == 2 and a.shape[0] == a.shape[1]):
raise la.LinAlgError("a is not square")
w = gauss(np.hstack((a, b[:, np.newaxis])))
lc = w.shape[1] - 1
x = w[:, lc].copy()
for c in range(lc - 1, 0, -1):
x[:c] -= w[:c, c] * x[c:c + 1]
np.flush()
return x
def main():
B = util.Benchmark()
nx = B.size[0]
ny = B.size[1]
nz = B.size[2]
ITER = B.size[3]
NO_OBST = 1
omega = 1.0
density = 1.0
deltaU = 1e-7
t1 = 1/3.0
t2 = 1/18.0
t3 = 1/36.0
B.start()
F = np.ones((19, nx, ny, nz), dtype=np.float64)
F[:] = density/19.0
FEQ = np.ones((19, nx, ny, nz), dtype=np.float64)
FEQ[:] = density/19.0
T = np.zeros((19, nx, ny, nz), dtype=np.float64)
#Create the scenery.
BOUND = np.zeros((nx, ny, nz), dtype=np.float64)
BOUNDi = np.ones((nx, ny, nz), dtype=np.float64)
"""
if not NO_OBST:
for i in xrange(nx):
for j in xrange(ny):
for k in xrange(nz):
if ((i-4)**2+(j-5)**2+(k-6)**2) < 6:
BOUND[i,j,k] += 1.0
BOUNDi[i,j,k] += 0.0
BOUND[:,0,:] += 1.0
BOUNDi[:,0,:] *= 0.0
"""
if util.Benchmark().bohrium:
np.flush()
for ts in xrange(0, ITER):
##Propagate / Streaming step
T[:] = F
#nearest-neighbours
F[1,:,:,0] = T[1,:,:,-1]
F[1,:,:,1:] = T[1,:,:,:-1]
F[2,:,:,:-1] = T[2,:,:,1:]
F[2,:,:,-1] = T[2,:,:,0]
F[3,:,0,:] = T[3,:,-1,:]
F[3,:,1:,:] = T[3,:,:-1,:]
F[4,:,:-1,:] = T[4,:,1:,:]
F[4,:,-1,:] = T[4,:,0,:]
F[5,0,:,:] = T[5,-1,:,:]
F[5,1:,:,:] = T[5,:-1,:,:]
F[6,:-1,:,:] = T[6,1:,:,:]
F[6,-1,:,:] = T[6,0,:,:]
#next-nearest neighbours
F[7,0 ,0 ,:] = T[7,-1 , -1,:]
F[7,0 ,1:,:] = T[7,-1 ,:-1,:]
F[7,1:,0 ,:] = T[7,:-1, -1,:]
F[7,1:,1:,:] = T[7,:-1,:-1,:]
F[8,0 ,:-1,:] = T[8,-1 ,1:,:]
F[8,0 , -1,:] = T[8,-1 ,0 ,:]
F[8,1:,:-1,:] = T[8,:-1,1:,:]
F[8,1:, -1,:] = T[8,:-1,0 ,:]
F[9,:-1,0 ,:] = T[9,1:, -1,:]
F[9,:-1,1:,:] = T[9,1:,:-1,:]
F[9,-1 ,0 ,:] = T[9,0 , 0,:]
F[9,-1 ,1:,:] = T[9,0 ,:-1,:]
F[10,:-1,:-1,:] = T[10,1:,1:,:]
F[10,:-1, -1,:] = T[10,1:,0 ,:]
F[10,-1 ,:-1,:] = T[10,0 ,1:,:]
F[10,-1 , -1,:] = T[10,0 ,0 ,:]
F[11,0 ,:,0 ] = T[11,0 ,:, -1]
F[11,0 ,:,1:] = T[11,0 ,:,:-1]
F[11,1:,:,0 ] = T[11,:-1,:, -1]
F[11,1:,:,1:] = T[11,:-1,:,:-1]
F[12,0 ,:,:-1] = T[12, -1,:,1:]
F[12,0 ,:, -1] = T[12, -1,:,0 ]
F[12,1:,:,:-1] = T[12,:-1,:,1:]
F[12,1:,:, -1] = T[12,:-1,:,0 ]
F[13,:-1,:,0 ] = T[13,1:,:, -1]
F[13,:-1,:,1:] = T[13,1:,:,:-1]
F[13, -1,:,0 ] = T[13,0 ,:, -1]
F[13, -1,:,1:] = T[13,0 ,:,:-1]
F[14,:-1,:,:-1] = T[14,1:,:,1:]
F[14,:-1,:, -1] = T[14,1:,:,0 ]
F[14,-1 ,:,:-1] = T[14,0 ,:,1:]
F[14,-1 ,:, -1] = T[14,0 ,:,0 ]
F[15,:,0 ,0 ] = T[15,:, -1, -1]
F[15,:,0 ,1:] = T[15,:, -1,:-1]
F[15,:,1:,0 ] = T[15,:,:-1, -1]
F[15,:,1:,1:] = T[15,:,:-1,:-1]
F[16,:,0 ,:-1] = T[16,:, -1,1:]
F[16,:,0 , -1] = T[16,:, -1,0 ]
F[16,:,1:,:-1] = T[16,:,:-1,1:]
F[16,:,1:, -1] = T[16,:,:-1,0 ]
F[17,:,:-1,0 ] = T[17,:,1:, -1]
F[17,:,:-1,1:] = T[17,:,1:,:-1]
F[17,:, -1,0 ] = T[17,:,0 , -1]
F[17,:, -1,1:] = T[17,:,0 ,:-1]
F[18,:,:-1,:-1] = T[18,:,1:,1:]
F[18,:,:-1, -1] = T[18,:,1:,0 ]
F[18,:,-1 ,:-1] = T[18,:,0 ,1:]
F[18,:,-1 , -1] = T[18,:,0 ,0 ]
#Densities bouncing back at next timestep
BB = np.empty(F.shape)
T[:] = F
T[1:,:,:,:] *= BOUND[np.newaxis,:,:,:]
BB[2 ,:,:,:] += T[1 ,:,:,:]
BB[1 ,:,:,:] += T[2 ,:,:,:]
BB[4 ,:,:,:] += T[3 ,:,:,:]
BB[3 ,:,:,:] += T[4 ,:,:,:]
BB[6 ,:,:,:] += T[5 ,:,:,:]
BB[5 ,:,:,:] += T[6 ,:,:,:]
BB[10,:,:,:] += T[7 ,:,:,:]
BB[9 ,:,:,:] += T[8 ,:,:,:]
BB[8 ,:,:,:] += T[9 ,:,:,:]
BB[7 ,:,:,:] += T[10,:,:,:]
BB[14,:,:,:] += T[11,:,:,:]
BB[13,:,:,:] += T[12,:,:,:]
BB[12,:,:,:] += T[13,:,:,:]
BB[11,:,:,:] += T[14,:,:,:]
BB[18,:,:,:] += T[15,:,:,:]
BB[17,:,:,:] += T[16,:,:,:]
BB[16,:,:,:] += T[17,:,:,:]
BB[15,:,:,:] += T[18,:,:,:]
# Relax calculate equilibrium state (FEQ) with equivalent speed and density to F
DENSITY = np.add.reduce(F)
#UX = F[5,:,:,:].copy()
UX = np.ones(F[5,:,:,:].shape, dtype=np.float64)
UX[:,:,:] = F[5,:,:,:]
UX += F[7,:,:,:]
UX += F[8,:,:,:]
UX += F[11,:,:,:]
UX += F[12,:,:,:]
UX -= F[6,:,:,:]
UX -= F[9,:,:,:]
UX -= F[10,:,:,:]
UX -= F[13,:,:,:]
UX -= F[14,:,:,:]
UX /=DENSITY
#UY = F[3,:,:,:].copy()
UY = np.ones(F[3,:,:,:].shape, dtype=np.float64)
UY[:,:,:] = F[3,:,:,:]
UY += F[7,:,:,:]
UY += F[9,:,:,:]
UY += F[15,:,:,:]
UY += F[16,:,:,:]
UY -= F[4,:,:,:]
UY -= F[8,:,:,:]
UY -= F[10,:,:,:]
UY -= F[17,:,:,:]
UY -= F[18,:,:,:]
UY /=DENSITY
#UZ = F[1,:,:,:].copy()
UZ = np.ones(F[1,:,:,:].shape, dtype=np.float64)
UZ[:,:,:] = F[1,:,:,:]
UZ += F[11,:,:,:]
UZ += F[13,:,:,:]
UZ += F[15,:,:,:]
UZ += F[17,:,:,:]
UZ -= F[2,:,:,:]
UZ -= F[12,:,:,:]
UZ -= F[14,:,:,:]
UZ -= F[16,:,:,:]
UZ -= F[18,:,:,:]
UZ /=DENSITY
UX[0,:,:] += deltaU #Increase inlet pressure
#Set bourderies to zero.
UX[:,:,:] *= BOUNDi
UY[:,:,:] *= BOUNDi
UZ[:,:,:] *= BOUNDi
DENSITY[:,:,:] *= BOUNDi
U_SQU = UX**2 + UY**2 + UZ**2
# Calculate equilibrium distribution: stationary
FEQ[0,:,:,:] = (t1*DENSITY)*(1.0-3.0*U_SQU/2.0)
# nearest-neighbours
T1 = 3.0/2.0*U_SQU
tDENSITY = t2*DENSITY
FEQ[1,:,:,:]=tDENSITY*(1.0 + 3.0*UZ + 9.0/2.0*UZ**2 - T1)
FEQ[2,:,:,:]=tDENSITY*(1.0 - 3.0*UZ + 9.0/2.0*UZ**2 - T1)
FEQ[3,:,:,:]=tDENSITY*(1.0 + 3.0*UY + 9.0/2.0*UY**2 - T1)
FEQ[4,:,:,:]=tDENSITY*(1.0 - 3.0*UY + 9.0/2.0*UY**2 - T1)
FEQ[5,:,:,:]=tDENSITY*(1.0 + 3.0*UX + 9.0/2.0*UX**2 - T1)
FEQ[6,:,:,:]=tDENSITY*(1.0 - 3.0*UX + 9.0/2.0*UX**2 - T1)
# next-nearest neighbours
T1 = 3.0*U_SQU/2.0
tDENSITY = t3*DENSITY
U8 = UX+UY
FEQ[7,:,:,:] =tDENSITY*(1.0 + 3.0*U8 + 9.0/2.0*(U8)**2 - T1)
U9 = UX-UY
FEQ[8,:,:,:] =tDENSITY*(1.0 + 3.0*U9 + 9.0/2.0*(U9)**2 - T1)
U10 = -UX+UY
FEQ[9,:,:,:] =tDENSITY*(1.0 + 3.0*U10 + 9.0/2.0*(U10)**2 - T1)
U8 *= -1.0
FEQ[10,:,:,:]=tDENSITY*(1.0 + 3.0*U8 + 9.0/2.0*(U8)**2 - T1)
U12 = UX+UZ
FEQ[11,:,:,:]=tDENSITY*(1.0 + 3.0*U12 + 9.0/2.0*(U12)**2 - T1)
U12 *= 1.0
FEQ[14,:,:,:]=tDENSITY*(1.0 + 3.0*U12 + 9.0/2.0*(U12)**2 - T1)
U13 = UX-UZ
FEQ[12,:,:,:]=tDENSITY*(1.0 + 3.0*U13 + 9.0/2.0*(U13)**2 - T1)
U13 *= -1.0
FEQ[13,:,:,:]=tDENSITY*(1.0 + 3.0*U13 + 9.0/2.0*(U13)**2 - T1)
U16 = UY+UZ
FEQ[15,:,:,:]=tDENSITY*(1.0 + 3.0*U16 + 9.0/2.0*(U16)**2 - T1)
U17 = UY-UZ
FEQ[16,:,:,:]=tDENSITY*(1.0 + 3.0*U17 + 9.0/2.0*(U17)**2 - T1)
U17 *= -1.0
FEQ[17,:,:,:]=tDENSITY*(1.0 + 3.0*U17 + 9.0/2.0*(U17)**2 - T1)
U16 *= -1.0
FEQ[18,:,:,:]=tDENSITY*(1.0 + 3.0*U16 + 9.0/2.0*(U16)**2 - T1)
F *= (1.0-omega)
F += omega * FEQ
#Densities bouncing back at next timestep
F[1:,:,:,:] *= BOUNDi[np.newaxis,:,:,:]
F[1:,:,:,:] += BB[1:,:,:,:]
del BB
del T1
del UX, UY, UZ
del U_SQU
del DENSITY, tDENSITY
del U8, U9, U10, U12, U13, U16, U17
if util.Benchmark().bohrium:
np.flush()
B.stop()
B.pprint()
if B.outputfn:
B.tofile(B.outputfn, {'res': UX})
"""
def run(*inputs, device='cpu'):
isoneutral_diffusion_pre(*inputs)
bh.flush()
return inputs[-7:]
def sor_compute(full):
cells = full[1:-1, 1:-1]
black1 = full[1:-1:2, 1:-1:2]
black2 = full[2:-1:2, 2:-1:2]
red1 = full[1:-1:2, 2:-1:2]
red2 = full[2:-1:2, 1:-1:2]
black1_up = full[0:-3:2, 1:-1:2]
black2_up = red1
black1_right = red1
black2_right = full[2:-1:2, 3::2]
black1_left = full[1:-2:2, 0:-2:2]
black2_left = red2
black1_down = red2
black2_down = full[3::2, 2:-1:2]
red1_up = full[0:-2:2, 2:-1:2]
red2_up = black1
red1_right = full[1:-2:2, 3::2]
red2_right = black2
red1_left = black1
red2_left = full[2:-1:2, 0:-3:2]
red1_down = black2
red2_down = full[3::2, 1:-2:2]
epsilon = full.shape[0]**2 * 0.002
delta = epsilon + 1
i = 0
while epsilon < delta:
i += 1
diff = black1.copy()
black1 += black1_up
black1 += black1_right
black1 += black1_left
black1 += black1_down
black1 *= 0.2
diff -= black1
np.absolute(diff, diff)
delta = np.add.reduce(np.add.reduce(diff))
diff = black2.copy()
black2 += black2_up
black2 += black2_right
black2 += black2_left
black2 += black2_down
black2 *= 0.2
diff -= black2
np.absolute(diff, diff)
delta += np.add.reduce(np.add.reduce(diff))
diff = red1.copy()
red1 += red1_up
red1 += red1_right
red1 += red1_left
red1 += red1_down
red1 *= 0.2
diff -= red1
delta += np.add.reduce(np.add.reduce(diff))
diff = red2.copy()
red2 += red2_up
red2 += red2_right
red2 += red2_left
red2 += red2_down
red2 *= 0.2
diff -= red2
delta += np.add.reduce(np.add.reduce(diff))
np.flush()
return cells
def run(sa, ct, p, device='cpu'):
out = gsw_dHdT(sa, ct, p)
bh.flush()
return out
def sor_compute(full):
cells = full[1:-1, 1:-1]
black1 = full[1:-1:2, 1:-1:2]
black2 = full[2:-1:2, 2:-1:2]
red1 = full[1:-1:2, 2:-1:2]
red2 = full[2:-1:2, 1:-1:2]
black1_up = full[0:-3:2, 1:-1:2]
black2_up = red1
black1_right = red1
black2_right = full[2:-1:2, 3::2]
black1_left = full[1:-2:2, 0:-2:2]
black2_left = red2
black1_down = red2
black2_down = full[3::2, 2:-1:2]
red1_up = full[0:-2:2, 2:-1:2]
red2_up = black1
red1_right = full[1:-2:2, 3::2]
red2_right = black2
red1_left = black1
red2_left = full[2:-1:2, 0:-3:2]
red1_down = black2
red2_down = full[3::2, 1:-2:2]
epsilon=full.shape[0]**2*0.002
delta=epsilon+1
i=0
while epsilon<delta:
i+=1
diff = black1.copy()
black1 += black1_up
black1 += black1_right
black1 += black1_left
black1 += black1_down
black1 *= 0.2
diff -= black1
np.absolute(diff, diff)
delta = np.add.reduce(np.add.reduce(diff))
diff = black2.copy()
black2 += black2_up
black2 += black2_right
black2 += black2_left
black2 += black2_down
black2 *= 0.2
diff -= black2
np.absolute(diff, diff)
delta += np.add.reduce(np.add.reduce(diff))
diff = red1.copy()
red1 += red1_up
red1 += red1_right
red1 += red1_left
red1 += red1_down
red1 *= 0.2
diff -= red1
delta += np.add.reduce(np.add.reduce(diff))
diff = red2.copy()
red2 += red2_up
red2 += red2_right
red2 += red2_left
red2 += red2_down
red2 *= 0.2
diff -= red2
delta += np.add.reduce(np.add.reduce(diff))
np.flush()
return cells
def run(*inputs, device='cpu'):
outputs = integrate_tke(*inputs)
bh.flush()
return outputs
def flush(self, ignore_no_flush_arg=False):
"""Executes the queued instructions when running through Bohrium. Set `ignore_no_flush_arg=True` to flush
even when the --no-flush argument is used"""
if bh_is_loaded_as_np:
if ignore_no_flush_arg or not self.args.no_flush:
bohrium.flush()
