From 78938143da029dceeccdc3e5dc25bbbe53dce0dc Mon Sep 17 00:00:00 2001
From: Jean-Baptiste Keck <Jean-Baptiste.Keck@imag.fr>
Date: Fri, 23 Dec 2016 16:09:23 +0100
Subject: [PATCH] working advection, periodic non cached
---
.../codegen/kernels/directional_advection.py | 81 ++++++++++++-------
1 file changed, 53 insertions(+), 28 deletions(-)
diff --git a/hysop/codegen/kernels/directional_advection.py b/hysop/codegen/kernels/directional_advection.py
index ada57d960..c2f36e6fd 100644
--- a/hysop/codegen/kernels/directional_advection.py
+++ b/hysop/codegen/kernels/directional_advection.py
@@ -44,15 +44,16 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
ftype[0],nparticles,min_ghosts)
@staticmethod
- def min_ghosts(dx, max_velocity):
- return math.ceil(max_velocity / dx)
+ def min_ghosts(max_dt, velocity_dx, max_velocity):
+ return int(math.ceil(max_dt*umax/float(velocity_dx)))
def __init__(self, typegen, work_dim, ftype,
is_cached, rk_scheme,
boundary, nparticles,
min_ghosts=0,
symbolic_mode = False,
- known_vars = None):
+ debug_mode = False,
+ known_vars = None):
assert work_dim>0 and work_dim<=3
assert nparticles in [1,2,4,8,16]
@@ -79,10 +80,10 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
name = DirectionalAdvectionKernel.codegen_name(ftype,is_cached,rk_scheme,nparticles,min_ghosts)
kernel_reqs = self.build_requirements(typegen, work_dim, itype, ftype, is_cached,
- rk_scheme, boundary, nparticles, symbolic_mode, storage, is_periodic)
+ rk_scheme, boundary, nparticles, symbolic_mode, storage, is_periodic, known_vars)
kernel_args = self.gen_kernel_arguments(typegen, work_dim, itype, ftype, kernel_reqs, is_cached,
- local_size_known)
+ local_size_known, debug_mode)
super(DirectionalAdvectionKernel,self).__init__(
name=name,
@@ -127,9 +128,11 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
#return global_work_size from effective work_size without
# taking into account local_work_size alignment
@staticmethod
- def get_max_global_size(work_size, work_load):
+ def get_max_global_size(work_size, work_load, nparticles):
work_size = np.asarray(work_size)
work_load = np.asarray(work_load)
+
+ assert work_load[0] == 1
global_size = work_size.copy()
global_size = ((global_size+work_load-1)/work_load)
@@ -137,17 +140,28 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
#return global_work_size from effective work_size and given local_work_size
# global_work_size will be a multiple of local_work_size
- def get_global_size(self, work_size, local_work_size, work_load=[1,1,1]):
+ def get_global_size(self, work_size, local_work_size, work_load=None):
+ work_dim = self.work_dim
+ work_load = [1]*work_dim if (work_load is None) else work_load
+
work_size = np.asarray(work_size)
work_load = np.asarray(work_load)
local_work_size = np.asarray(local_work_size)
+
+ nparticles = self.nparticles
- assert (local_work_size[1] == 1) and (local_work_size[2]==1)
+ for var in [work_size, work_load, local_work_size]:
+ msg = 'Dimension mismatch for variable {}, expected work_dim is {}.'
+ msg = msg.format(var, work_dim)
+ if var.size != work_dim:
+ raise ValueError(msg)
+ for i in xrange(1,work_dim):
+ assert (local_work_size[i] == 1), 'local_work_size error!'
if 'local_size' in self.known_vars:
- assert (self.known_vars['local_size'] == local_work_size).all()
+ assert (self.known_vars['local_size'] == local_work_size).all(), 'local_work_size mismatch!'
- max_global_size = self.get_max_global_size(work_size, work_load)
+ max_global_size = self.get_max_global_size(work_size, work_load, nparticles)
global_size = ((max_global_size+local_work_size-1)/local_work_size) * local_work_size
return global_size
@@ -178,20 +192,15 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
return (sc,dc,tc)
def build_requirements(self,typegen,work_dim,itype,ftype,is_cached,rk_scheme,
- boundary,nparticles,force_symbolic,storage,is_periodic):
+ boundary,nparticles,force_symbolic,storage,is_periodic,known_vars):
tg=typegen
reqs = WriteOnceDict()
- # compute_id = ComputeIndexFunction(typegen=typegen, dim=work_dim, itype='int',
- # wrap=is_periodic)
- # reqs['compute_id'] = compute_id
-
mesh_base_struct = MeshBaseStruct(typegen=typegen, typedef='MeshBase_s')
reqs['MeshBaseStruct'] = mesh_base_struct
mesh_info_struct = MeshInfoStruct(typegen=typegen, typedef='MeshInfo_s')
reqs['MeshInfoStruct'] = mesh_info_struct
-
advection_rhs = DirectionalAdvectionRhsFunction(typegen=typegen, work_dim=work_dim,
ftype=ftype, is_cached=is_cached,
@@ -212,7 +221,7 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
return reqs
def gen_kernel_arguments(self, typegen, work_dim, itype, ftype, requirements,is_cached,
- local_size_known):
+ local_size_known, debug_mode):
kargs = ArgDict()
kargs['dt'] = CodegenVariable(ctype=ftype,name='dt',typegen=typegen,
@@ -225,12 +234,18 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
restrict=True,ptr=True,const=True,add_impl_const=True)
kargs['position'] = CodegenVariable(storage=_global,name='position',ctype=ftype,typegen=typegen,
restrict=True,ptr=True,const=False,add_impl_const=True)
+
+ if debug_mode:
+ kargs['dbg0'] = CodegenVariable(storage=_global,name='dbg0',ctype=itype,typegen=typegen,
+ restrict=True,ptr=True,const=False,add_impl_const=True)
+ kargs['dbg1'] = CodegenVariable(storage=_global,name='dbg1',ctype=itype,typegen=typegen,
+ restrict=True,ptr=True,const=False,add_impl_const=True)
kargs['mesh_info'] = requirements['MeshInfoStruct'].build_codegen_variable(const=True,name='mesh_info')
if is_cached and not local_size_known:
- kargs['buffer'] = CodegenVariable(storage=_local,ctype=ftype, add_impl_const=True,
- name='buffer', ptr=True, restrict=True, typegen=typegen, nl=False)
+ kargs['Vc'] = CodegenVariable(storage=_local,ctype=ftype, add_impl_const=True,
+ name='Vc', ptr=True, restrict=True, typegen=typegen, nl=False)
return kargs
@@ -268,8 +283,6 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
velocity = s.vars['velocity']
position = s.vars['position']
- if is_cached:
- shared = s.vars['buffer']
grid_size = mesh_info['local_mesh']['resolution'].view(
'grid_size', slice(None,work_dim))
@@ -291,6 +304,16 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
pid = CodegenVectorClBuiltin('pid', itype, nparticles, typegen=tg)
poffset = CodegenVectorClBuiltin('poffset', itype, nparticles, typegen=tg)
+ if is_cached:
+ cache_ghosts = CodegenVariable('cache_ghosts', itype, typegen=tg, value=min_ghosts)
+ cache_width = CodegenVariable('cache_width', itype, typegen=tg)
+ if local_size_known:
+ local_size = s.known_vars['local_size']
+ Vc = CodegenArray(name='Vc',dim=1,ctype=ftype,typegen=tg,
+ shape=(local_size.value[0]+2*min_ghosts,), storage='__local')
+ else:
+ Vc = s.vars['Vc']
+
part_ftype = X.ctype
part_itype = pid.ctype
@@ -377,13 +400,15 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
s.append(code)
s.append('{} += {};'.format(X(), xmin()))
-
- if nparticles>1:
- code = 'vstore{}({}, {}, {}+{});'.format(nparticles,X(),
- global_id[0], position(), line_index())
- else:
- code = '{} = {};'.format(position['{}+{}'.format(line_index(),pid())],X())
- s.append(code)
+
+ with s._if_(active()):
+ if nparticles>1:
+ code = 'vstore{}({}, {}, {}+{});'.format(nparticles,X(),
+ global_id[0], position(), line_index())
+ s.append(code)
+ else:
+ code = '{} = {};'.format(position['{}+{}'.format(line_index(),pid())],X())
+ s.append(code)
--
GitLab