diff --git a/examples/scalar_advection/scalar_advection.py b/examples/scalar_advection/scalar_advection.py
index 45610c6462490f350e31382f059e9d15c9dbe280..6580c2a12cf9e8ad6558ecbdf12eeff5a6bc17ae 100644
--- a/examples/scalar_advection/scalar_advection.py
+++ b/examples/scalar_advection/scalar_advection.py
@@ -41,7 +41,7 @@ def run(npts=64+1, cfl=0.5):
if impl is Implementation.OPENCL_CODEGEN:
autotuner_config = OpenClKernelAutotunerConfig(
autotuner_flag=AutotunerFlags.ESTIMATE,
- prune_threshold=1.2, override_cache=False, verbose=0)
+ prune_threshold=1.2, override_cache=True, verbose=0)
kernel_config = OpenClKernelConfig(autotuner_config=autotuner_config)
method = { OpenClKernelConfig : kernel_config }
else:
diff --git a/hysop/__init__.py b/hysop/__init__.py
index 0d496006a739c48dd5c623e40d511b8b20d04b1f..279e86abeb02be2bfb61b0e3665f0412e9c2564c 100644
--- a/hysop/__init__.py
+++ b/hysop/__init__.py
@@ -16,11 +16,11 @@ __FFTW_ENABLED__ = "ON" is "ON"
__SCALES_ENABLED__ = "ON" is "ON"
__OPTIMIZE__ = not __debug__
-__VERBOSE__ = True
+__VERBOSE__ = False
__DEBUG__ = False
__TRACE__ = False
__TRACE_WARNINGS__ = False
-__KERNEL_DEBUG__ = False
+__KERNEL_DEBUG__ = True
__PROFILE__ = True
__ENABLE_LONG_TESTS__ = "OFF" is "ON"
diff --git a/hysop/backend/device/codegen/functions/directional_remesh.py b/hysop/backend/device/codegen/functions/directional_remesh.py
index 3cdb730856289ba43b64c2aa595ca2934351d7d9..78b618a0b5e9b4e27ae641455cd748304af60d86 100644
--- a/hysop/backend/device/codegen/functions/directional_remesh.py
+++ b/hysop/backend/device/codegen/functions/directional_remesh.py
@@ -229,7 +229,7 @@ class DirectionalRemeshFunction(OpenClFunctionCodeGenerator):
y = CodegenVectorClBuiltin(name='y', btype=ftype, dim=nparticles, typegen=tg)
ind = CodegenVectorClBuiltin(name='ind', btype=itype, dim=nparticles, typegen=tg)
find = CodegenVectorClBuiltin(name='find', btype=ftype, dim=nparticles, typegen=tg)
- vone = CodegenVectorClBuiltin(name='one', btype=ftype, dim=nparticles, typegen=tg)
+ vone = CodegenVectorClBuiltin(name='one', btype=ftype, dim=nparticles, typegen=tg, value=(1,)*nparticles)
if poly_splitted:
wl = CodegenVectorClBuiltin(name='Wl', btype=ftype, dim=nparticles, typegen=tg)
@@ -246,7 +246,8 @@ class DirectionalRemeshFunction(OpenClFunctionCodeGenerator):
eps.declare(s)
s.jumpline()
- s.decl_aligned_vars(find, ind, y, vone)
+ s.decl_aligned_vars(find, ind, y)
+ vone.declare(s, const=True)
s.decl_vars(*weights)
s.jumpline()
diff --git a/hysop/backend/device/kernel_autotuner.py b/hysop/backend/device/kernel_autotuner.py
index 4ce75bb7d453bb91e3c08463cce1e4741e0eb4ff..5c8b95438875556f7e94f7885c0c75bcf37afba5 100644
--- a/hysop/backend/device/kernel_autotuner.py
+++ b/hysop/backend/device/kernel_autotuner.py
@@ -151,14 +151,16 @@ class KernelAutotuner(object):
if (kernel_name != cached_kernel_name):
msg='\nCached kernel name did not match the benched one:\n {}\n {}\n'
- msg+='\nThis might be due to a faulty implementation of {}.hash_extra_kwds().'
+ msg+='\nThis might be due to an upgrade of the generated code or '
+ msg+='a faulty implementation of {}.hash_extra_kwds().'
msg=msg.format(kernel_name, cached_kernel_name, type(tkernel).__name__)
warnings.warn(msg, CodeGeneratorWarning)
return None
if (src_hash != cached_src_hash):
msg='\nCached kernel source hash did not match the benched one.\n {}\n {}'
- msg+='\nThis might be due to a faulty implementation of {}.hash_extra_kwds().'
+ msg+='\nThis might be due to an upgrade of the generated code or '
+ msg+='a faulty implementation of {}.hash_extra_kwds().'
msg=msg.format(src_hash, cached_src_hash, type(tkernel).__name__)
warnings.warn(msg, CodeGeneratorWarning)
return None
@@ -237,8 +239,9 @@ class KernelAutotuner(object):
(cache_src_hash, cache_stats) = results[run_key]
if cache_src_hash != src_hash:
msg='\nCached parameters candidate did not match the benched one.\n {}\n {}'
- msg+='\nThis might be due to a faulty implementation of {}.hash_extra_kwds().'
- msg=msg.format(src_hash, cached_src_hash, type(tunable_kernel).__name__)
+ msg+='\nThis might be due to an upgrade of the generated code or '
+ msg+='a faulty implementation of {}.hash_extra_kwds().'
+ msg=msg.format(src_hash, cache_src_hash, type(tkernel).__name__)
warnings.warn(msg, CodeGeneratorWarning)
old_stats = None
else:
diff --git a/hysop/backend/device/opencl/opencl_env.py b/hysop/backend/device/opencl/opencl_env.py
index c807744af886bd35fa80af2fe030d611c701452f..da81b51c678b5868afc9324a6e372b66f40d8ad8 100644
--- a/hysop/backend/device/opencl/opencl_env.py
+++ b/hysop/backend/device/opencl/opencl_env.py
@@ -599,7 +599,6 @@ Dumped OpenCL Kernel '{}'
print('Dumping kernel src at \'{}\'.'.format(dump_file))
with open(dump_file, 'w+') as f:
f.write(gpu_src)
- #build_opts += ' '+' '.join(['-g', '-s "{}"'.format(dump_file)])
# Build OpenCL program
try:
diff --git a/hysop/backend/device/opencl/operator/directional/advection_dir.py b/hysop/backend/device/opencl/operator/directional/advection_dir.py
index 4b9950563d3a91fcd2f8bb6993d94f7511bc9da8..10d4e573c91579fb01db26301a2c4056a5f5c2b8 100644
--- a/hysop/backend/device/opencl/operator/directional/advection_dir.py
+++ b/hysop/backend/device/opencl/operator/directional/advection_dir.py
@@ -66,6 +66,9 @@ class OpenClDirectionalAdvection(DirectionalAdvectionBase, OpenClDirectionalOper
self.relax_min_particles = relax_min_particles
self.remesh_criteria_eps = remesh_criteria_eps
+ self._force_autotuner_verbose = True
+ self._force_autotuner_debug = True
+
@debug
def get_work_properties(self):
requests = super(OpenClDirectionalAdvection, self).get_work_properties()
@@ -107,7 +110,8 @@ class OpenClDirectionalAdvection(DirectionalAdvectionBase, OpenClDirectionalOper
kwds['velocity_cfl'] = self.velocity_cfl
kwds['time_integrator'] = self.time_integrator
- (advec_kernel, args_dict) = kernel.autotune(force_debug=True, **kwds)
+ (advec_kernel, args_dict) = kernel.autotune(force_verbose=self._force_autotuner_verbose,
+ force_debug=self._force_autotuner_debug, **kwds)
args_dict.pop('dt')
advec_launcher = advec_kernel.build_launcher(**args_dict)
@@ -145,7 +149,8 @@ class OpenClDirectionalAdvection(DirectionalAdvectionBase, OpenClDirectionalOper
kwds['force_atomics'] = self.force_atomics
kwds['relax_min_particles'] = self.relax_min_particles
- (remesh_kernel, args_dict) = kernel.autotune(**kwds)
+ (remesh_kernel, args_dict) = kernel.autotune(force_verbose=self._force_autotuner_verbose,
+ force_debug=self._force_autotuner_debug, **kwds)
kl = remesh_kernel.build_launcher(**args_dict)
self.remesh_kernel_launcher = kl