diff --git a/CMakeLists.txt b/CMakeLists.txt
index 04b08f506b621747f83bd4f18387f5f516dd1a1f..7203581db508b915940a0b2f4c467785ba75d215 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -197,6 +197,10 @@ if(WITH_FFTW)
add_definitions(${FFTW_DEFINES})
endif()
+if(WITH_LIB_CXX)
+ compile_with(Boost REQUIRED)
+endif()
+
if(WITH_EXTRAS)
# Arnoldi solver needs zgeev, which means lapack
compile_with(LAPACK)
@@ -317,7 +321,7 @@ endif()
if(WITH_LIB_CXX)
#C++ variables used by setup.py.in for swig
- set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -W -Wall -Wextra -Wno-unused-variable -Wno-unused-but-set-variable -Wno-unused-parameter ${FFTW_COMPILE_FLAGS} -fPIC -std=c++11")
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -W -Wall -Wextra -Wno-unused-variable -Wno-unused-but-set-variable -Wno-unused-parameter -Wno-unused-local-typedefs ${FFTW_COMPILE_FLAGS} -fPIC -std=c++11")
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG}")
set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO}")
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}")
@@ -325,11 +329,14 @@ if(WITH_LIB_CXX)
set(CXX_FLAGS "${CMAKE_CXX_FLAGS}")
set(CXX_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS}")
- set(CXX_EXT_INCLUDES ${PYTHON_INCLUDE_DIR} ${FFTW_INCLUDE_DIRS})
- set(CXX_EXT_LIBS ${PYTHON_LIBRARIES} ${FFTW_LIBRARIES})
- set(CXX_EXT_LIB_DIRS ${FFTW_LIBRARY_DIRS})
+ set(CXX_EXT_INCLUDES ${Boost_INCLUDE_DIR} ${PYTHON_INCLUDE_DIR} ${FFTW_INCLUDE_DIRS})
+ set(CXX_EXT_LIBS ${PYTHON_LIBRARIES} ${FFTW_LIBRARIES} ${Boost_LIBRARIES})
+ set(CXX_EXT_LIB_DIRS ${FFTW_LIBRARY_DIRS} ${Boost_LIBRARY_DIRS})
set(CXX_EXTRA_DEFINES ${FFTW_DEFINES} -DHAS_EXTERN_TEMPLATES)
+ set(CMAKE_INCLUDE_SYSTEM_FLAG_C "-isystem ")
+ set(CMAKE_INCLUDE_SYSTEM_FLAG_CXX "-isystem ")
+
#swig package name (lib name generated by swig)
set(CPP_2_HYSOP "cpp2hysop")
endif()
diff --git a/hysop/codegen/kernels/stretching.py b/hysop/codegen/kernels/stretching.py
index 128777239d66329f3b3815310452d6785a08872b..72ed5136420629c4b121bd27dff8f98df0c8d0c2 100644
--- a/hysop/codegen/kernels/stretching.py
+++ b/hysop/codegen/kernels/stretching.py
@@ -19,9 +19,10 @@ class CachedStretchingKernel(KernelCodeGenerator):
def codegen_name(ftype,work_dim):
return 'cached_stretching_{}_{}d'.format(ftype,work_dim)
- def __init__(self, typegen, dim, order=2,
+ def __init__(self, typegen, dim,
+ device,context,
+ order=2,
ftype=None,
- device=None,context=None,
known_vars = None,
symbolic_mode=True):
@@ -29,7 +30,8 @@ class CachedStretchingKernel(KernelCodeGenerator):
ftype = ftype if ftype is not None else typegen.fbtype
work_dim=3
- kernel_reqs = self.build_requirements(typegen,work_dim, order, cached)
+ kernel_reqs = self.build_requirements(typegen=typegen,device=device,context=context,
+ work_dim=work_dim, order=order, cached=cached)
kernel_args = self.gen_kernel_arguments(typegen, work_dim, ftype, kernel_reqs)
name = CachedStretchingKernel.codegen_name(ftype, dim)
@@ -55,16 +57,16 @@ class CachedStretchingKernel(KernelCodeGenerator):
return reduce(operator.mul, local_size+order, 1)*self.typegen.FLT_BYTES[self.ftype]
- def build_requirements(self,typegen,work_dim,order,cached):
+ def build_requirements(self,typegen,device,context,work_dim,order,cached):
reqs = WriteOnceDict()
compute_id = ComputeIndexFunction(typegen=typegen, dim=work_dim, itype='int', wrap=False)
reqs['compute_id'] = compute_id
- mesh_base_struct = MeshBaseStruct(typegen=typegen, typedef='MeshBaseStruct_s')
+ mesh_base_struct = MeshBaseStruct(device=device,context=context,typegen=typegen, typedef='MeshBaseStruct_s')
reqs['MeshBaseStruct'] = mesh_base_struct
- mesh_info_struct = MeshInfoStruct(typegen=typegen, typedef='MeshInfoStruct_s')
+ mesh_info_struct = MeshInfoStruct(device=device, context=context, typegen=typegen, typedef='MeshInfoStruct_s')
reqs['MeshInfoStruct'] = mesh_info_struct
gradient = GradientFunction(typegen=typegen, dim=work_dim, order=order,
@@ -163,11 +165,22 @@ class CachedStretchingKernel(KernelCodeGenerator):
if __name__ == '__main__':
-
- tg = OpenClTypeGen('float', 'dec')
- ek = CachedStretchingKernel(typegen=tg, order=16, dim=1 ,ftype=tg.fbtype,
- known_vars=dict(local_size=(1024,1,1)))
- ek.edit()
- ek.test_compile()
- #print ek
- print
+
+ import pyopencl as cl
+
+ devices = []
+ contexts = {}
+ for plat in cl.get_platforms():
+ devices += plat.get_devices()
+ for dev in devices:
+ ctx = cl.Context([dev])
+ contexts[dev] = ctx
+
+ tg = OpenClTypeGen('float', 'dec')
+ for dev,ctx in contexts.iteritems():
+ ek = CachedStretchingKernel(typegen=tg, context=ctx, device=dev,
+ order=16, dim=1 ,ftype=tg.fbtype,
+ known_vars=dict(local_size=(1024,1,1)))
+ ek.edit()
+ ek.test_compile()
+ break
diff --git a/hysop/constants.py b/hysop/constants.py
index 5894a663591087c1a71e00828d9ccc7103acc216..3f9323cdccf16d799ecf9efb70974b5df9d9fddd 100755
--- a/hysop/constants.py
+++ b/hysop/constants.py
@@ -111,10 +111,10 @@ def debugdecorator(f):
debug = debugdecorator
# redefine profile decorator
-if __PROFILE__:
- from memory_profiler import profile
- prof = profile
-else:
- def prof(f):
- # Nothing ...
- return f
+# if __PROFILE__:
+ # from memory_profiler import profile
+ # prof = profile
+# else:
+def prof(f):
+ # Nothing ...
+ return f
diff --git a/hysop/gpu/static_gpu_particle_advection_dir.py b/hysop/gpu/static_gpu_particle_advection_dir.py
index d14e76332eb0064d65da3ecbce0abdbecb7dc665..33e3508f88cfd9d1daae19883a3cf83fac46ceec 100644
--- a/hysop/gpu/static_gpu_particle_advection_dir.py
+++ b/hysop/gpu/static_gpu_particle_advection_dir.py
@@ -44,6 +44,9 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
'transpose_xz':[],
'transpose_zx':[],
'stretching':[],
+ 'advec':[],
+ 'remesh':[],
+ 'advec_remesh':[]
}
# Additional method and configuration checks
@@ -418,7 +421,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
mesh_info = self._fields_mesh_info_var
dt=0.1
- nruns=16
+ nruns=4
force_renew_cache=True
(kernel_launcher, kernel_args, kernel_args_mapping, cached_bytes) = \
@@ -490,6 +493,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
)
evt = self._advec_and_remesh[nbc](*args, wait_for=wait_evts)
+ self.bench['advec_remesh'].append(evt)
fg.events.append(evt)
velocity.events.append(evt)
@@ -506,6 +510,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
self._cl_mesh_info
])
advec_evt = self._advec(*args,wait_for=velocity.events)
+ self.bench['advec'].append(advec_evt)
velocity.events.append(advec_evt)
for (fg,fp) in self.fields_on_part.iteritems():
@@ -518,6 +523,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
+ [self._cl_mesh_info]
)
remesh_evt = self._remesh[nbc](*args, wait_for=[advec_evt])
+ self.bench['remesh'].append(remesh_evt)
fg.events.append(remesh_evt)
if self._has_stretching:
@@ -608,7 +614,7 @@ if __name__=='__main__':
dim = 3
GHOSTS = 0
NSCALARS = 0
- f_resolution = (65,65,65)[:dim]
+ f_resolution = (513,513,129)[:dim]
v_resolution = f_resolution
#v_resolution = (33,33,33)[:dim]
ghosts = (GHOSTS,)*dim
@@ -750,4 +756,15 @@ if __name__=='__main__':
simu.advance()
i+=1
+
+ from hysop.gpu.kernel_autotuner import OpenClKernelStatistics
+ bench = A._advec_dir[0].bench
+ for i in xrange(1,dim):
+ for k,v in A._advec_dir[i].bench.iteritems():
+ bench[k] += v
+
+ for name,evts in bench.iteritems():
+ if len(evts)>=1:
+ print name, OpenClKernelStatistics(events=evts,nruns=i)
+