def project(XK, XV, LorY, surfSrc, surfTar, K_diag, V_diag, IorE, self, param,
ind0, timing, kernel):
if param.GPU == 1:
tic = cuda.Event()
toc = cuda.Event()
else:
tic = Event()
toc = Event()
REAL = param.REAL
Ns = len(surfSrc.triangle)
Nt = len(surfTar.triangle)
L = numpy.sqrt(2 * surfSrc.Area) # Representative length
tic.record()
K = param.K
w = getWeights(K)
X_V = numpy.zeros(Ns * K)
X_Kx = numpy.zeros(Ns * K)
X_Ky = numpy.zeros(Ns * K)
X_Kz = numpy.zeros(Ns * K)
X_Kc = numpy.zeros(Ns * K)
X_Vc = numpy.zeros(Ns * K)
NsK = numpy.arange(Ns * K)
X_V[:] = XV[NsK / K] * w[NsK % K] * surfSrc.Area[NsK / K]
X_Kx[:] = XK[NsK / K] * w[NsK % K] * surfSrc.Area[NsK /
K] * surfSrc.normal[NsK /
K, 0]
X_Ky[:] = XK[NsK / K] * w[NsK % K] * surfSrc.Area[NsK /
K] * surfSrc.normal[NsK /
K, 1]
X_Kz[:] = XK[NsK / K] * w[NsK % K] * surfSrc.Area[NsK /
K] * surfSrc.normal[NsK /
K, 2]
X_Kc[:] = XK[NsK / K]
X_Vc[:] = XV[NsK / K]
toc.record()
toc.synchronize()
timing.time_mass += tic.time_till(toc) * 1e-3
tic.record()
C = 0
getMultipole(surfSrc.tree, C, surfSrc.xj, surfSrc.yj, surfSrc.zj, X_V,
X_Kx, X_Ky, X_Kz, ind0, param.P, param.NCRIT)
toc.record()
toc.synchronize()
timing.time_P2M += tic.time_till(toc) * 1e-3
tic.record()
for C in reversed(range(1, len(surfSrc.tree))):
PC = surfSrc.tree[C].parent
upwardSweep(surfSrc.tree, C, PC, param.P, ind0.II, ind0.JJ, ind0.KK,
ind0.index, ind0.combII, ind0.combJJ, ind0.combKK,
ind0.IImii, ind0.JJmjj, ind0.KKmkk, ind0.index_small,
ind0.index_ptr)
toc.record()
toc.synchronize()
timing.time_M2M += tic.time_till(toc) * 1e-3
tic.record()
X_V = X_V[surfSrc.sortSource]
X_Kx = X_Kx[surfSrc.sortSource]
X_Ky = X_Ky[surfSrc.sortSource]
X_Kz = X_Kz[surfSrc.sortSource]
X_Kc = X_Kc[surfSrc.sortSource]
X_Vc = X_Vc[surfSrc.sortSource]
toc.record()
toc.synchronize()
timing.time_sort += tic.time_till(toc) * 1e-3
param.Nround = len(surfTar.twig) * param.NCRIT
K_aux = numpy.zeros(param.Nround)
V_aux = numpy.zeros(param.Nround)
AI_int = 0
### CPU code
if param.GPU == 0:
K_aux, V_aux = M2P_sort(surfSrc, surfTar, K_aux, V_aux, self,
ind0.index_large, param, LorY, timing)
K_aux, V_aux = P2P_sort(surfSrc, surfTar, X_V, X_Kx, X_Ky, X_Kz, X_Kc,
X_Vc, K_aux, V_aux, self, LorY, K_diag, V_diag,
IorE, L, w, param, timing)
### GPU code
elif param.GPU == 1:
K_gpu = cuda.to_device(K_aux.astype(REAL))
V_gpu = cuda.to_device(V_aux.astype(REAL))
if surfTar.offsetMlt[self, len(surfTar.twig)] > 0:
K_gpu, V_gpu = M2P_gpu(surfSrc, surfTar, K_gpu, V_gpu, self, ind0,
param, LorY, timing, kernel)
K_gpu, V_gpu = P2P_gpu(surfSrc, surfTar, X_V, X_Kx, X_Ky, X_Kz, X_Kc,
X_Vc, K_gpu, V_gpu, self, LorY, K_diag, IorE, L,
w, param, timing, kernel)
tic.record()
K_aux = cuda.from_device(K_gpu, len(K_aux), dtype=REAL)
V_aux = cuda.from_device(V_gpu, len(V_aux), dtype=REAL)
toc.record()
toc.synchronize()
timing.time_trans += tic.time_till(toc) * 1e-3
tic.record()
K_lyr = K_aux[surfTar.unsort]
V_lyr = V_aux[surfTar.unsort]
toc.record()
toc.synchronize()
timing.time_sort += tic.time_till(toc) * 1e-3
return K_lyr, V_lyr
def project(XK, XV, LorY, surfSrc, surfTar, K_diag, V_diag, IorE,
self, param, ind0, timing, kernel):
if param.GPU==1:
tic = cuda.Event()
toc = cuda.Event()
else:
tic = Event()
toc = Event()
REAL = param.REAL
Ns = len(surfSrc.triangle)
Nt = len(surfTar.triangle)
L = numpy.sqrt(2*surfSrc.Area) # Representative length
tic.record()
K = param.K
w = getWeights(K)
X_V = numpy.zeros(Ns*K)
X_Kx = numpy.zeros(Ns*K)
X_Ky = numpy.zeros(Ns*K)
X_Kz = numpy.zeros(Ns*K)
X_Kc = numpy.zeros(Ns*K)
X_Vc = numpy.zeros(Ns*K)
NsK = numpy.arange(Ns*K)
X_V[:] = XV[NsK/K]*w[NsK%K]*surfSrc.Area[NsK/K]
X_Kx[:] = XK[NsK/K]*w[NsK%K]*surfSrc.Area[NsK/K]*surfSrc.normal[NsK/K,0]
X_Ky[:] = XK[NsK/K]*w[NsK%K]*surfSrc.Area[NsK/K]*surfSrc.normal[NsK/K,1]
X_Kz[:] = XK[NsK/K]*w[NsK%K]*surfSrc.Area[NsK/K]*surfSrc.normal[NsK/K,2]
X_Kc[:] = XK[NsK/K]
X_Vc[:] = XV[NsK/K]
toc.record()
toc.synchronize()
timing.time_mass += tic.time_till(toc)*1e-3
tic.record()
C = 0
getMultipole(surfSrc.tree, C, surfSrc.xj, surfSrc.yj, surfSrc.zj,
X_V, X_Kx, X_Ky, X_Kz, ind0, param.P, param.NCRIT)
toc.record()
toc.synchronize()
timing.time_P2M += tic.time_till(toc)*1e-3
tic.record()
for C in reversed(range(1,len(surfSrc.tree))):
PC = surfSrc.tree[C].parent
upwardSweep(surfSrc.tree, C, PC, param.P, ind0.II, ind0.JJ, ind0.KK, ind0.index, ind0.combII, ind0.combJJ,
ind0.combKK, ind0.IImii, ind0.JJmjj, ind0.KKmkk, ind0.index_small, ind0.index_ptr)
toc.record()
toc.synchronize()
timing.time_M2M += tic.time_till(toc)*1e-3
tic.record()
X_V = X_V[surfSrc.sortSource]
X_Kx = X_Kx[surfSrc.sortSource]
X_Ky = X_Ky[surfSrc.sortSource]
X_Kz = X_Kz[surfSrc.sortSource]
X_Kc = X_Kc[surfSrc.sortSource]
X_Vc = X_Vc[surfSrc.sortSource]
toc.record()
toc.synchronize()
timing.time_sort += tic.time_till(toc)*1e-3
param.Nround = len(surfTar.twig)*param.NCRIT
K_aux = numpy.zeros(param.Nround)
V_aux = numpy.zeros(param.Nround)
AI_int = 0
### CPU code
if param.GPU==0:
K_aux, V_aux = M2P_sort(surfSrc, surfTar, K_aux, V_aux, self,
ind0.index_large, param, LorY, timing)
K_aux, V_aux = P2P_sort(surfSrc, surfTar, X_V, X_Kx, X_Ky, X_Kz, X_Kc, X_Vc,
K_aux, V_aux, self, LorY, K_diag, V_diag, IorE, L, w, param, timing)
### GPU code
elif param.GPU==1:
K_gpu = cuda.to_device(K_aux.astype(REAL))
V_gpu = cuda.to_device(V_aux.astype(REAL))
if surfTar.offsetMlt[self,len(surfTar.twig)]>0:
K_gpu, V_gpu = M2P_gpu(surfSrc, surfTar, K_gpu, V_gpu, self,
ind0, param, LorY, timing, kernel)
K_gpu, V_gpu = P2P_gpu(surfSrc, surfTar, X_V, X_Kx, X_Ky, X_Kz, X_Kc, X_Vc,
K_gpu, V_gpu, self, LorY, K_diag, IorE, L, w, param, timing, kernel)
tic.record()
K_aux = cuda.from_device(K_gpu, len(K_aux), dtype=REAL)
V_aux = cuda.from_device(V_gpu, len(V_aux), dtype=REAL)
toc.record()
toc.synchronize()
timing.time_trans += tic.time_till(toc)*1e-3
tic.record()
K_lyr = K_aux[surfTar.unsort]
V_lyr = V_aux[surfTar.unsort]
toc.record()
toc.synchronize()
timing.time_sort += tic.time_till(toc)*1e-3
return K_lyr, V_lyr
def project_Kt(XKt, LorY, surfSrc, surfTar, Kt_diag, self, param, ind0, timing,
kernel):
if param.GPU == 1:
tic = cuda.Event()
toc = cuda.Event()
else:
tic = Event()
toc = Event()
REAL = param.REAL
Ns = len(surfSrc.triangle)
Nt = len(surfTar.triangle)
L = numpy.sqrt(2 * surfSrc.Area) # Representative length
tic.record()
K = param.K
w = getWeights(K)
X_Kt = numpy.zeros(Ns * K)
X_Ktc = numpy.zeros(Ns * K)
NsK = numpy.arange(Ns * K)
X_Kt[:] = XKt[NsK / K] * w[NsK % K] * surfSrc.Area[NsK / K]
X_Ktc[:] = XKt[NsK / K]
toc.record()
toc.synchronize()
timing.time_mass += tic.time_till(toc) * 1e-3
tic.record()
C = 0
X_aux = numpy.zeros(Ns * K)
getMultipole(surfSrc.tree, C, surfSrc.xj, surfSrc.yj, surfSrc.zj, X_Kt,
X_aux, X_aux, X_aux, ind0, param.P, param.NCRIT)
toc.record()
toc.synchronize()
timing.time_P2M += tic.time_till(toc) * 1e-3
tic.record()
for C in reversed(range(1, len(surfSrc.tree))):
PC = surfSrc.tree[C].parent
upwardSweep(surfSrc.tree, C, PC, param.P, ind0.II, ind0.JJ, ind0.KK,
ind0.index, ind0.combII, ind0.combJJ, ind0.combKK,
ind0.IImii, ind0.JJmjj, ind0.KKmkk, ind0.index_small,
ind0.index_ptr)
toc.record()
toc.synchronize()
timing.time_M2M += tic.time_till(toc) * 1e-3
tic.record()
X_Kt = X_Kt[surfSrc.sortSource]
X_Ktc = X_Ktc[surfSrc.sortSource]
toc.record()
toc.synchronize()
timing.time_sort += tic.time_till(toc) * 1e-3
param.Nround = len(surfTar.twig) * param.NCRIT
Ktx_aux = numpy.zeros(param.Nround)
Kty_aux = numpy.zeros(param.Nround)
Ktz_aux = numpy.zeros(param.Nround)
AI_int = 0
### CPU code
if param.GPU == 0:
if surfTar.offsetMlt[self, len(surfTar.twig)] > 0:
Ktx_aux, Kty_aux, Ktz_aux = M2PKt_sort(surfSrc, surfTar, Ktx_aux,
Kty_aux, Ktz_aux, self,
ind0.index_large, param,
LorY, timing)
Ktx_aux, Kty_aux, Ktz_aux = P2PKt_sort(surfSrc, surfTar, X_Kt, X_Ktc,
Ktx_aux, Kty_aux, Ktz_aux, self,
LorY, w, param, timing)
### GPU code
elif param.GPU == 1:
Ktx_gpu = cuda.to_device(Ktx_aux.astype(REAL))
Kty_gpu = cuda.to_device(Kty_aux.astype(REAL))
Ktz_gpu = cuda.to_device(Ktz_aux.astype(REAL))
if surfTar.offsetMlt[self, len(surfTar.twig)] > 0:
Ktx_gpu, Kty_gpu, Ktz_gpu = M2PKt_gpu(surfSrc, surfTar, Ktx_gpu,
Kty_gpu, Ktz_gpu, self, ind0,
param, LorY, timing, kernel)
Ktx_gpu, Kty_gpu, Ktz_gpu = P2PKt_gpu(surfSrc, surfTar, X_Kt, X_Ktc,
Ktx_gpu, Kty_gpu, Ktz_gpu, self,
LorY, w, param, timing, kernel)
tic.record()
Ktx_aux = cuda.from_device(Ktx_gpu, len(Ktx_aux), dtype=REAL)
Kty_aux = cuda.from_device(Kty_gpu, len(Kty_aux), dtype=REAL)
Ktz_aux = cuda.from_device(Ktz_gpu, len(Ktz_aux), dtype=REAL)
toc.record()
toc.synchronize()
timing.time_trans += tic.time_till(toc) * 1e-3
tic.record()
Kt_lyr = Ktx_aux[surfTar.unsort]*surfTar.normal[:,0] \
+ Kty_aux[surfTar.unsort]*surfTar.normal[:,1] \
+ Ktz_aux[surfTar.unsort]*surfTar.normal[:,2]
if abs(Kt_diag) > 1e-12: # if same surface
Kt_lyr += Kt_diag * XKt
toc.record()
toc.synchronize()
timing.time_sort += tic.time_till(toc) * 1e-3
return Kt_lyr
def project_Kt(XKt, LorY, surfSrc, surfTar, Kt_diag,
self, param, ind0, timing, kernel):
if param.GPU==1:
tic = cuda.Event()
toc = cuda.Event()
else:
tic = Event()
toc = Event()
REAL = param.REAL
Ns = len(surfSrc.triangle)
Nt = len(surfTar.triangle)
L = numpy.sqrt(2*surfSrc.Area) # Representative length
tic.record()
K = param.K
w = getWeights(K)
X_Kt = numpy.zeros(Ns*K)
X_Ktc = numpy.zeros(Ns*K)
NsK = numpy.arange(Ns*K)
X_Kt[:] = XKt[NsK/K]*w[NsK%K]*surfSrc.Area[NsK/K]
X_Ktc[:] = XKt[NsK/K]
toc.record()
toc.synchronize()
timing.time_mass += tic.time_till(toc)*1e-3
tic.record()
C = 0
X_aux = numpy.zeros(Ns*K)
getMultipole(surfSrc.tree, C, surfSrc.xj, surfSrc.yj, surfSrc.zj,
X_Kt, X_aux, X_aux, X_aux, ind0, param.P, param.NCRIT)
toc.record()
toc.synchronize()
timing.time_P2M += tic.time_till(toc)*1e-3
tic.record()
for C in reversed(range(1,len(surfSrc.tree))):
PC = surfSrc.tree[C].parent
upwardSweep(surfSrc.tree, C, PC, param.P, ind0.II, ind0.JJ, ind0.KK, ind0.index, ind0.combII, ind0.combJJ,
ind0.combKK, ind0.IImii, ind0.JJmjj, ind0.KKmkk, ind0.index_small, ind0.index_ptr)
toc.record()
toc.synchronize()
timing.time_M2M += tic.time_till(toc)*1e-3
tic.record()
X_Kt = X_Kt[surfSrc.sortSource]
X_Ktc = X_Ktc[surfSrc.sortSource]
toc.record()
toc.synchronize()
timing.time_sort += tic.time_till(toc)*1e-3
param.Nround = len(surfTar.twig)*param.NCRIT
Ktx_aux = numpy.zeros(param.Nround)
Kty_aux = numpy.zeros(param.Nround)
Ktz_aux = numpy.zeros(param.Nround)
AI_int = 0
### CPU code
if param.GPU==0:
if surfTar.offsetMlt[self,len(surfTar.twig)]>0:
Ktx_aux, Kty_aux, Ktz_aux = M2PKt_sort(surfSrc, surfTar, Ktx_aux, Kty_aux, Ktz_aux, self,
ind0.index_large, param, LorY, timing)
Ktx_aux, Kty_aux, Ktz_aux = P2PKt_sort(surfSrc, surfTar, X_Kt, X_Ktc,
Ktx_aux, Kty_aux, Ktz_aux, self, LorY, w, param, timing)
### GPU code
elif param.GPU==1:
Ktx_gpu = cuda.to_device(Ktx_aux.astype(REAL))
Kty_gpu = cuda.to_device(Kty_aux.astype(REAL))
Ktz_gpu = cuda.to_device(Ktz_aux.astype(REAL))
if surfTar.offsetMlt[self,len(surfTar.twig)]>0:
Ktx_gpu, Kty_gpu, Ktz_gpu = M2PKt_gpu(surfSrc, surfTar,
Ktx_gpu, Kty_gpu, Ktz_gpu, self,
ind0, param, LorY, timing, kernel)
Ktx_gpu, Kty_gpu, Ktz_gpu = P2PKt_gpu(surfSrc, surfTar, X_Kt, X_Ktc, Ktx_gpu, Kty_gpu, Ktz_gpu,
self, LorY, w, param, timing, kernel)
tic.record()
Ktx_aux = cuda.from_device(Ktx_gpu, len(Ktx_aux), dtype=REAL)
Kty_aux = cuda.from_device(Kty_gpu, len(Kty_aux), dtype=REAL)
Ktz_aux = cuda.from_device(Ktz_gpu, len(Ktz_aux), dtype=REAL)
toc.record()
toc.synchronize()
timing.time_trans += tic.time_till(toc)*1e-3
tic.record()
Kt_lyr = Ktx_aux[surfTar.unsort]*surfTar.normal[:,0] \
+ Kty_aux[surfTar.unsort]*surfTar.normal[:,1] \
+ Ktz_aux[surfTar.unsort]*surfTar.normal[:,2]
if abs(Kt_diag)>1e-12: # if same surface
Kt_lyr += Kt_diag * XKt
toc.record()
toc.synchronize()
timing.time_sort += tic.time_till(toc)*1e-3
return Kt_lyr