diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index ba7898d89434dbd0ca93c6dfe4dc6ef74209d575..65b6e27da23c55e2a55149b7236de01c47b5d4a8 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -1,52 +1,35 @@
+
stages:
- configure
- build
- test
+before_script:
+ script: "sh ci/scripts/version.sh > /tmp/host_info"
+ atrifacts:
+ paths:
+ - /tmp/host_info
-
-config:debian:
- image: fperignon/hysop:debian
- stage: configure
- script: "sh ci/config.sh"
- artifacts:
- paths:
- - build/
-
-build:debian:
- image: fperignon/hysop:debian
- stage: build
- script: "sh ci/build_install_clean.sh"
- dependencies:
- - config:debian
-
-test:debian:
- image: fperignon/hysop:debian
- stage: test
- script: "sh ci/run_tests.sh"
- dependencies:
- - build:debian
-
-config:ubuntu:
- image: fperignon/hysop:ubuntu
+config:ubuntu_zesty:
+ image: keckj/hysop:ubuntu_zesty
stage: configure
- script: "sh ci/config.sh"
+ script: "sh ci/scripts/config.sh"
artifacts:
paths:
- build/
-build:ubuntu:
- image: fperignon/hysop:ubuntu
+build:ubuntu_zesty:
+ image: keckj/hysop:ubuntu_zesty
stage: build
- script: "sh ci/build_install_clean.sh"
+ script: "sh ci/scripts/build_install_clean.sh"
dependencies:
- - config:ubuntu
+ - config:ubuntu_zesty
-test:ubuntu:
- image: fperignon/hysop:ubuntu
+test:ubuntu_zesty:
+ image: keckj/hysop:ubuntu_zesty
stage: test
script:
- useradd "hysoprunner"
- - su "hysoprunner" -c "sh ci/run_tests.sh"
+ - su "hysoprunner" -c "sh ci/scripts/run_tests.sh"
dependencies:
- - build:ubuntu
+ - build:ubuntu_zesty
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 25e5e54b022aa6ddb33b5991e2d3d1062d593941..8c4b38fd7e0a24f0f90c009305fe76701ace6c28 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -38,11 +38,12 @@ option(WITH_SCALES "compile/create scales lib and link it with HySoP. Default =
option(WITH_FFTW "Link with fftw library (required for some HySoP solvers), default = ON" ON)
option(WITH_EXTRAS "Link with some extra fortran libraries (like arnoldi solver), default = OFF" OFF)
option(WITH_GPU "Use of GPU (required for some HySoP solvers), default = ON" ON)
-option(WITH_MAIN_FORTRAN "Create an executable (test purpose) from fortran sources in src/main, linked with libhysop, default = ON" OFF)
-option(WITH_MAIN_CXX "Create an executable (test purpose) from cxx sources in src/hysop++/main, linked with libhysop, default = ON" ON)
-option(DEBUG "Enable debug mode for HySoP (0:disabled, 1:verbose, 2:trace, 3:verbose+trace). Default = 0" 1)
+option(WITH_MAIN_FORTRAN "Create an executable (test purpose) from fortran sources in src/main, linked with libhysop, default = OFF" OFF)
+option(WITH_MAIN_CXX "Create an executable (test purpose) from cxx sources in src/hysop++/main, linked with libhysop, default = OFF" OFF)
+option(PROFILE "Enable profiling mode for HySoP. Default = ON" ON)
+option(VERBOSE "Enable verbose mode for HySoP. Default = ON" ON)
+option(DEBUG "Enable debug mode for HySoP. Default = OFF" OFF)
option(FULL_TEST "Enable all test options (pep8, mpi ...) - Default = OFF" OFF)
-option(PROFILE "Enable profiling mode for HySoP. 0:disabled, 1: enabled. Default = 0" 0)
option(OPTIM "To allow python -OO run, some packages must be deactivated. Set this option to 'ON' to do so. Default = OFF" OFF)
option(WITH_MPI_TESTS "Enable mpi tests. Default = ON if USE_MPI is ON." ON)
option(WITH_GOOGLE_TESTS "Enable google tests (c++). Default = OFF." OFF)
@@ -131,14 +132,14 @@ find_package(PythonFull REQUIRED)
include(FindPythonModule)
# - python packages -
find_python_module(numpy REQUIRED)
-find_python_module(scipy)
-find_python_module(matplotlib)
-if(NOT matplotlib_FOUND)
- find_python_module(Gnuplot)
-endif()
-find_python_module(scitools)
+find_python_module(scipy REQUIRED)
+find_python_module(scitools REQUIRED)
find_python_module(h5py REQUIRED)
-find_python_module(sympy)
+find_python_module(sympy REQUIRED)
+find_python_module(psutil REQUIRED)
+find_python_module(cpuinfo REQUIRED)
+find_python_module(gmpy2 REQUIRED)
+find_python_module(graph_tool REQUIRED)
# --- OpenCL ---
find_python_module(pyopencl REQUIRED)
# --- MPI ---
@@ -146,6 +147,11 @@ if(USE_MPI)
find_package(MPI REQUIRED)
find_python_module(mpi4py REQUIRED)
endif()
+# --- PLOT --
+find_python_module(matplotlib)
+if(NOT matplotlib_FOUND)
+ find_python_module(Gnuplot)
+endif()
# --- Wheel, required for a proper build/install process ---
find_python_module(wheel REQUIRED)
if(USE_CXX)
diff --git a/ci/docker_images/ubuntu/xenial/Dockerfile b/ci/docker_images/ubuntu/xenial/Dockerfile
index daae74853c3eb048429373184b588165381b06f7..9dafb4408f0216eca86e15b77413f6bf616c1da3 100644
--- a/ci/docker_images/ubuntu/xenial/Dockerfile
+++ b/ci/docker_images/ubuntu/xenial/Dockerfile
@@ -38,9 +38,9 @@ RUN apt-get install -y libsparsehash-dev
RUN apt-get install -y python
RUN apt-get install -y python-dev
RUN apt-get install -y python-pip
-RUN apt-get install -y python-scitools
RUN apt-get install -y opencl-headers
-RUN apt-get install -y ocl-icd-opencl-dev
+RUN apt-get install -y ocl-icd-libopencl1
+RUN apt-get install -y clinfo
# python packages
RUN pip install --upgrade pip
@@ -60,16 +60,13 @@ RUN pip install psutil
RUN pip install py-cpuinfo
RUN pip install Mako
-# pyopencl
+# scitools (python-scitools does not exist on ubuntu:xenial)
RUN cd /tmp \
-&& git clone https://github.com/inducer/pyopencl \
-&& cd pyopencl \
-&& git submodule update --init \
-&& ./configure.py \
-&& make \
-&& make install \
-&& cd - \
-&& rm -Rf /tmp/pyopencl
+ && git clone https://github.com/hplgit/scitools \
+ && cd scitools \
+ && pip install . \
+ && cd - \
+ && rm -Rf /tmp/scitools
# patchelf
RUN cd /tmp \
@@ -82,30 +79,56 @@ RUN cd /tmp \
&& cd - \
&& rm -Rf /tmp/patchelf
-# clang 6.0
+# precompiled python graphtools (need to be patched with patchelf)
+RUN echo 'deb http://downloads.skewed.de/apt/xenial xenial universe' >> /etc/apt/sources.list \
+ && echo 'deb-src http://downloads.skewed.de/apt/xenial xenial universe' >> /etc/apt/sources.list \
+ && apt-get update \
+ && apt-get install --assume-yes --allow-unauthenticated python-graph-tool \
+ && patchelf --add-needed libboost_context.so $(find /usr/ -name 'libgraph_tool_search.so')
+
RUN echo 'deb http://apt.llvm.org/xenial/ llvm-toolchain-xenial main' >> /etc/apt/sources.list \
&& echo 'deb-src http://apt.llvm.org/xenial/ llvm-toolchain-xenial main' >> /etc/apt/sources.list \
- && apt-get update \
- && apt-get install --assume-yes --allow-unauthenticated clang-6.0 libclang-6.0-dev
+ && apt-get update \
+ && apt-get install --assume-yes --allow-unauthenticated llvm-3.8 clang-3.8 libllvm3.8 libclang-3.8-dev
-# oclgrind
+# POCL
+RUN cd /tmp \
+&& git clone http://github.com/pocl/pocl \
+&& cd pocl \
+&& mkdir build \
+&& cd build \
+&& cmake -DCMAKE_BUILD_TYPE=Release .. \
+&& make \
+&& make install \
+&& cd -- \
+&& rm -Rf /tmp/pocl
+
+# pyopencl
RUN cd /tmp \
- && git clone https://github.com/jrprice/Oclgrind \
- && cd Oclgrind \
- && mkdir build \
- && cd build \
- && cmake -DCMAKE_BUILD_TYPE=Release .. \
- && make \
- && make install \
- && cd - \
- && rm -Rf /tmp/Oclgrind
+&& rm -f /usr/lib/x86_64-linux-gnu/libOpenCL.so.* \
+&& ldconfig \
+&& git clone https://github.com/inducer/pyopencl \
+&& cd pyopencl \
+&& git submodule update --init \
+&& ./configure.py \
+&& echo 'CL_PRETEND_VERSION="1.2"' >> siteconf.py \
+&& make \
+&& pip install --upgrade . \
+&& cd - \
+&& rm -Rf /tmp/pyopencl
-# precompiled python graphtools (need to be patched with patchelf)
-RUN echo 'deb http://downloads.skewed.de/apt/xenial xenial universe' >> /etc/apt/sources.list \
- && echo 'deb-src http://downloads.skewed.de/apt/xenial xenial universe' >> /etc/apt/sources.list \
- && apt-get update \
- && apt-get install --assume-yes --allow-unauthenticated python-graph-tool \
- && patchelf --add-needed libboost_context.so $(find /usr/ -name 'libgraph_tool_search.so')
+# oclgrind
+RUN apt-get install --assume-yes --allow-unauthenticated llvm-6.0 clang-6.0 libllvm6.0 libclang-6.0-dev
+RUN cd /tmp \
+ && git clone https://github.com/jrprice/Oclgrind \
+ && cd Oclgrind \
+ && mkdir build \
+ && cd build \
+ && cmake -DCMAKE_BUILD_TYPE=Release .. \
+ && make \
+ && make install \
+ && cd - \
+ && rm -Rf /tmp/Oclgrind
# clean cached packages
RUN rm -rf /var/lib/apt/lists/*
diff --git a/ci/docker_images/ubuntu/zesty/Dockerfile b/ci/docker_images/ubuntu/zesty/Dockerfile
index 501321cf88fb7a005f98dc609dea1b4ceda369b3..a04d4e5087131c5b0e28ee2cc94aa6f1a783fc2b 100644
--- a/ci/docker_images/ubuntu/zesty/Dockerfile
+++ b/ci/docker_images/ubuntu/zesty/Dockerfile
@@ -38,11 +38,14 @@ RUN apt-get install -y libsparsehash-dev
RUN apt-get install -y python
RUN apt-get install -y python-dev
RUN apt-get install -y python-pip
-RUN apt-get install -y python-pyopencl
+RUN apt-get install -y opencl-headers
+RUN apt-get install -y ocl-icd-libopencl1
+RUN apt-get install -y clinfo
# python packages
RUN pip install --upgrade pip
RUN pip install --upgrade setuptools
+RUN pip install cffi
RUN pip install wheel
RUN pip install pytest
RUN pip install numpy
@@ -55,6 +58,7 @@ RUN pip install h5py
RUN pip install gmpy2
RUN pip install psutil
RUN pip install py-cpuinfo
+RUN pip install Mako
# scitools (python-scitools does not exist on ubuntu:zesty)
RUN cd /tmp \
@@ -75,24 +79,6 @@ RUN cd /tmp \
&& cd - \
&& rm -Rf /tmp/patchelf
-# clang
-RUN echo 'deb http://apt.llvm.org/zesty/ llvm-toolchain-zesty main' >> /etc/apt/sources.list \
- && echo 'deb-src http://apt.llvm.org/zesty/ llvm-toolchain-zesty main' >> /etc/apt/sources.list \
- && apt-get update \
- && apt-get install --assume-yes --allow-unauthenticated clang-6.0 libclang-6.0-dev
-
-# oclgrind
-RUN cd /tmp \
- && git clone https://github.com/jrprice/Oclgrind \
- && cd Oclgrind \
- && mkdir build \
- && cd build \
- && cmake -DCMAKE_BUILD_TYPE=Release .. \
- && make \
- && make install \
- && cd - \
- && rm -Rf /tmp/Oclgrind
-
# precompiled python graphtools (need to be patched with patchelf)
RUN echo 'deb http://downloads.skewed.de/apt/zesty zesty universe' >> /etc/apt/sources.list \
&& echo 'deb-src http://downloads.skewed.de/apt/zesty zesty universe' >> /etc/apt/sources.list \
@@ -100,6 +86,50 @@ RUN echo 'deb http://downloads.skewed.de/apt/zesty zesty universe' >> /etc/a
&& apt-get install --assume-yes --allow-unauthenticated python-graph-tool \
&& patchelf --add-needed libboost_context.so $(find /usr/ -name 'libgraph_tool_search.so')
+RUN echo 'deb http://apt.llvm.org/zesty/ llvm-toolchain-zesty main' >> /etc/apt/sources.list \
+ && echo 'deb-src http://apt.llvm.org/zesty/ llvm-toolchain-zesty main' >> /etc/apt/sources.list \
+ && apt-get update \
+ && apt-get install --assume-yes --allow-unauthenticated llvm-3.8 clang-3.8 libllvm3.8 libclang-3.8-dev
+
+# POCL
+RUN cd /tmp \
+&& git clone http://github.com/pocl/pocl \
+&& cd pocl \
+&& mkdir build \
+&& cd build \
+&& cmake -DCMAKE_BUILD_TYPE=Release .. \
+&& make \
+&& make install \
+&& cd -- \
+&& rm -Rf /tmp/pocl
+
+# pyopencl
+RUN cd /tmp \
+&& rm -f /usr/lib/x86_64-linux-gnu/libOpenCL.so.* \
+&& ldconfig \
+&& git clone https://github.com/inducer/pyopencl \
+&& cd pyopencl \
+&& git submodule update --init \
+&& ./configure.py \
+&& echo 'CL_PRETEND_VERSION="1.2"' >> siteconf.py \
+&& make \
+&& pip install --upgrade . \
+&& cd - \
+&& rm -Rf /tmp/pyopencl
+
+# oclgrind
+RUN apt-get install --assume-yes --allow-unauthenticated llvm-6.0 clang-6.0 libllvm6.0 libclang-6.0-dev
+RUN cd /tmp \
+ && git clone https://github.com/jrprice/Oclgrind \
+ && cd Oclgrind \
+ && mkdir build \
+ && cd build \
+ && cmake -DCMAKE_BUILD_TYPE=Release .. \
+ && make \
+ && make install \
+ && cd - \
+ && rm -Rf /tmp/Oclgrind
+
# clean cached packages
RUN rm -rf /var/lib/apt/lists/*
RUN rm -rf $HOME/.cache/pip/*
diff --git a/ci/build_install_clean.sh b/ci/scripts/build_install_clean.sh
similarity index 90%
rename from ci/build_install_clean.sh
rename to ci/scripts/build_install_clean.sh
index d2304fb9c62a7cdd5fb81028bee54fa39ab07483..a0663d04a595d0431d4a4add64a9df9df33f31b5 100644
--- a/ci/build_install_clean.sh
+++ b/ci/scripts/build_install_clean.sh
@@ -2,11 +2,10 @@
cd build
make
make install
-cd ..
+cd -
python -c 'import hysop'
cd build
echo $'y\n'|make uninstall
make clean
-cd ..
+cd -
rm -rf build
-
diff --git a/ci/config.sh b/ci/scripts/config.sh
similarity index 100%
rename from ci/config.sh
rename to ci/scripts/config.sh
diff --git a/ci/run_tests.sh b/ci/scripts/run_tests.sh
similarity index 100%
rename from ci/run_tests.sh
rename to ci/scripts/run_tests.sh
diff --git a/ci/scripts/version.sh b/ci/scripts/version.sh
new file mode 100644
index 0000000000000000000000000000000000000000..2e8fdc77edc8a85f9d22a1c2e1d8e8183886bc76
--- /dev/null
+++ b/ci/scripts/version.sh
@@ -0,0 +1,16 @@
+echo "HOST"
+uname -a
+echo
+echo "BUILD TOOLS:"
+echo $(cmake --version | head -1)
+echo $(make --version | head -1)
+echo $(gcc --version | head -1)
+echo $(swig -version | head -2 | tail -1)
+echo "f2py version $(f2py -v)"
+echo
+echo "PYTHON PACKAGES:"
+pip list --format=columns
+echo
+echo "CL_INFO"
+clinfo
+echo
diff --git a/examples/scalar_advection/scalar_advection.py b/examples/scalar_advection/scalar_advection.py
index 6580c2a12cf9e8ad6558ecbdf12eeff5a6bc17ae..5ed1d7cfa87d5e5fe3f71823013541e9e230adb8 100644
--- a/examples/scalar_advection/scalar_advection.py
+++ b/examples/scalar_advection/scalar_advection.py
@@ -40,8 +40,8 @@ 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=True, verbose=0)
+ autotuner_flag=AutotunerFlags.ESTIMATE, max_candidates=10,
+ prune_threshold=1.2, override_cache=True, verbose=0,)
kernel_config = OpenClKernelConfig(autotuner_config=autotuner_config)
method = { OpenClKernelConfig : kernel_config }
else:
@@ -60,11 +60,10 @@ def run(npts=64+1, cfl=0.5):
)
splitting = StrangSplitting(splitting_dim=dim,
- order=StrangOrder.STRANG_SECOND_ORDER,
- method=method)
+ order=StrangOrder.STRANG_SECOND_ORDER)
splitting.push_operators(advec)
- problem = Problem()
+ problem = Problem(method=method)
problem.insert(splitting)
io_params = IOParams(filename='S0', frequency=1)
diff --git a/hysop/__init__.py b/hysop/__init__.py
index 279e86abeb02be2bfb61b0e3665f0412e9c2564c..da7a5a741cc7440e3168119b618ab1f056fa6148 100644
--- a/hysop/__init__.py
+++ b/hysop/__init__.py
@@ -16,12 +16,12 @@ __FFTW_ENABLED__ = "ON" is "ON"
__SCALES_ENABLED__ = "ON" is "ON"
__OPTIMIZE__ = not __debug__
-__VERBOSE__ = False
-__DEBUG__ = False
+__VERBOSE__ = True
+__DEBUG__ = "OFF" is "ON"
+__PROFILE__ = "ON" is "ON"
__TRACE__ = False
__TRACE_WARNINGS__ = False
-__KERNEL_DEBUG__ = True
-__PROFILE__ = True
+__KERNEL_DEBUG__ = False
__ENABLE_LONG_TESTS__ = "OFF" is "ON"
diff --git a/hysop/__init__.py.in b/hysop/__init__.py.in
index 1a7bd799e2c874ef17bb64e939dba42a170db50e..6666b268fe21d75d24d131e72e641a000d472d31 100644
--- a/hysop/__init__.py.in
+++ b/hysop/__init__.py.in
@@ -16,12 +16,12 @@ __FFTW_ENABLED__ = "@WITH_FFTW@" is "ON"
__SCALES_ENABLED__ = "@WITH_SCALES@" is "ON"
__OPTIMIZE__ = not __debug__
-__VERBOSE__ = "@DEBUG@" in ["1", "3"]
-__DEBUG__ = "@DEBUG@" in ["2", "3"]
+__VERBOSE__ = "@VERBOSE@" is "ON"
+__DEBUG__ = "@DEBUG@" is "ON"
+__PROFILE__ = "@PROFILE@" is "ON"
__TRACE__ = False
__TRACE_WARNINGS__ = False
__KERNEL_DEBUG__ = False
-__PROFILE__ = True
__ENABLE_LONG_TESTS__ = "@ENABLE_LONG_TESTS@" is "ON"
diff --git a/hysop/backend/device/kernel_autotuner.py b/hysop/backend/device/kernel_autotuner.py
index a5e0fb0efae54999a66ffff340eda6b979cc6495..d92437fc6c97929dd1210fb3939b3ee9270a33ed 100644
--- a/hysop/backend/device/kernel_autotuner.py
+++ b/hysop/backend/device/kernel_autotuner.py
@@ -112,7 +112,8 @@ class KernelAutotuner(object):
force_verbose, force_debug)
result_keys = ('extra_parameters', 'work_load', 'global_work_size', 'local_work_size',
- 'program', 'kernel', 'kernel_statistics', 'kernel_src', 'kernel_name', 'src_hash')
+ 'program', 'kernel', 'kernel_statistics', 'kernel_src', 'kernel_name',
+ 'src_hash')
assert len(result_keys) == len(best_candidate)
return dict(zip(result_keys, best_candidate))
@@ -180,6 +181,7 @@ class KernelAutotuner(object):
force_verbose, force_debug):
autotuner_config = self.autotuner_config
nruns = autotuner_config.nruns
+ max_candidates = autotuner_config.max_candidates
bench_results = {}
best_stats = None
@@ -190,6 +192,7 @@ class KernelAutotuner(object):
with Timer() as timer:
params = tkernel.compute_parameters(extra_kwds=extra_kwds)
total_count, pruned_count, kept_count, failed_count = 0,0,0,0
+ abort = False
for extra_parameters in params.iter_parameters():
work_bounds = tkernel.compute_work_bounds(extra_parameters=extra_parameters,
@@ -287,9 +290,20 @@ class KernelAutotuner(object):
statistics = None
from_cache=False
total_count += 1
-
- self._print_candidate((statistics is None), from_cache, total_count)
+ abort = (max_candidates is not None) and \
+ ((pruned_count + kept_count) >= max_candidates)
+ self._print_candidate((statistics is None), from_cache, total_count, abort)
+ if abort:
+ break
+ if abort:
+ break
self._dump_cache(silent=True)
+ if abort:
+ break
+ if abort:
+ msg='>Achieved maximum number of configured candidates: {}'.format(max_candidates)
+ if self.verbose>1:
+ print msg
assert total_count == (kept_count+pruned_count+failed_count)
if (kept_count == 0):
@@ -480,7 +494,7 @@ class KernelAutotuner(object):
self._print_separator()
print msg.upper()
- def _print_candidate(self, failed, from_cache, total_count):
+ def _print_candidate(self, failed, from_cache, total_count, abort):
if self.verbose==2:
if total_count==1:
sys.stdout.write(self.indent(2))
@@ -490,6 +504,8 @@ class KernelAutotuner(object):
sys.stdout.write(':')
else:
sys.stdout.write('.')
+ if abort:
+ sys.stdout.write('|')
if total_count % 40 == 0:
sys.stdout.write('\n'+self.indent(2))
elif total_count % 5 == 0:
diff --git a/hysop/backend/device/kernel_autotuner_config.py b/hysop/backend/device/kernel_autotuner_config.py
index 0cb7733313d4be6ff58d60e055be7f468a973fc0..6fbf1a015dc64bf2f7b61a538cd0d6829d6941bd 100644
--- a/hysop/backend/device/kernel_autotuner_config.py
+++ b/hysop/backend/device/kernel_autotuner_config.py
@@ -19,6 +19,7 @@ class KernelAutotunerConfig(object):
def __init__(self, dump_folder = None,
autotuner_flag = DEFAULT_AUTOTUNER_FLAG,
prune_threshold = DEFAULT_AUTOTUNER_PRUNE_THRESHOLD,
+ max_candidates = None,
verbose = None,
debug = __KERNEL_DEBUG__,
dump_kernels = True,
@@ -38,6 +39,7 @@ class KernelAutotunerConfig(object):
check_instance(verbose, int)
check_instance(debug, int)
check_instance(nruns, int)
+ check_instance(max_candidates, int, allow_none=True)
self.autotuner_flag = autotuner_flag
self.prune_threshold = prune_threshold
@@ -47,6 +49,7 @@ class KernelAutotunerConfig(object):
self.nruns = nruns
self.dump_folder = dump_folder
self.dump_kernels = dump_kernels
+ self.max_candidates = max_candidates
@abstractmethod
def default_dump_folder(self):
diff --git a/hysop/backend/device/opencl/opencl_env.py b/hysop/backend/device/opencl/opencl_env.py
index da81b51c678b5868afc9324a6e372b66f40d8ad8..da75933a9eac55eba84dd765f790ade19caa05af 100644
--- a/hysop/backend/device/opencl/opencl_env.py
+++ b/hysop/backend/device/opencl/opencl_env.py
@@ -573,7 +573,7 @@ Dumped OpenCL Kernel '{}'
build_opts = ' '.join(set(self.default_build_opts).update(build_options))
else:
build_opts = ' '.join(self.default_build_opts) + ' ' + build_options
- VERBOSE = __VERBOSE__ if (force_verbose is None) else force_verbose
+ VERBOSE = False if (force_verbose is None) else force_verbose
DEBUG = __KERNEL_DEBUG__ if (force_debug is None) else force_debug
gpu_src = src
diff --git a/hysop/backend/device/opencl/opencl_operator.py b/hysop/backend/device/opencl/opencl_operator.py
index 1419a725f8728ab6f958a8798cfc5a32a5f4306e..ff074ab31ab0f459df9f0fd956535d418cab06e7 100644
--- a/hysop/backend/device/opencl/opencl_operator.py
+++ b/hysop/backend/device/opencl/opencl_operator.py
@@ -126,7 +126,7 @@ class OpenClOperator(ComputationalGraphOperator):
super(OpenClOperator,self).handle_method(method)
assert OpenClKernelConfig in method
-
+
kernel_config = method.pop(OpenClKernelConfig)
autotuner_config = kernel_config.autotuner_config
diff --git a/hysop/backend/device/opencl/operator/transpose.py b/hysop/backend/device/opencl/operator/transpose.py
index 6c20bda7d6a45d0810f0a5b9b9a1d395c0417b2d..238ec7a35db2e885ed0485c27a654c786c4be315 100644
--- a/hysop/backend/device/opencl/operator/transpose.py
+++ b/hysop/backend/device/opencl/operator/transpose.py
@@ -40,7 +40,7 @@ class OpenClTranspose(TransposeOperatorBase, OpenClOperator):
compute_inplace = False
hardcode_arrays = (compute_inplace or not is_inplace)
- transpose, _ = kernel.autotune(axes=axes, force_debug=True,
+ transpose, _ = kernel.autotune(axes=axes,
hardcode_arrays=hardcode_arrays,
is_inplace=compute_inplace, input_field=input_field, output_field=output_field)
diff --git a/hysop/constants.py.in b/hysop/constants.py.in
index 13d8a48574a322af63d820d50e73e086262f8bb7..a402100c3a3bedcfd5bf86d17a63554ef3c6af0a 100644
--- a/hysop/constants.py.in
+++ b/hysop/constants.py.in
@@ -152,7 +152,7 @@ SpaceDiscretization = EnumFactory.create('SpaceDiscretization',
AutotunerFlags = EnumFactory.create('AutotunerFlags',
- ['ESTIMATE', 'MEASURE', 'PATIENT', 'EXHAUSTIVE' ])
+ ['ESTIMATE', 'MEASURE', 'PATIENT', 'EXHAUSTIVE'])
"""Configuration flags for kernel autotuner
(automatic runtime parameters tuning for cuda and opencl)."""
DEFAULT_AUTOTUNER_FLAG = AutotunerFlags.MEASURE
diff --git a/hysop/core/graph/computational_graph.py b/hysop/core/graph/computational_graph.py
index 17a40b22191e4e2e37fc4bb03295cd353b17a8e2..a8a409edc699581c43800af4009e1d9041fdb46a 100644
--- a/hysop/core/graph/computational_graph.py
+++ b/hysop/core/graph/computational_graph.py
@@ -240,7 +240,7 @@ class ComputationalGraph(ComputationalGraphNode):
if is_root:
field_requirements = self.get_and_set_field_requirements()
field_requirements.build_topologies()
-
+
self._build_graph(outputs_are_inputs=outputs_are_inputs, current_level=0)
# fix for auto generated nodes
@@ -291,7 +291,7 @@ class ComputationalGraph(ComputationalGraphNode):
builder = GraphBuilder(node=self)
builder.configure(current_level=current_level,
outputs_are_inputs=outputs_are_inputs, **kwds)
-
+
builder.build_graph()
# keep variables
diff --git a/hysop/core/graph/computational_operator.py b/hysop/core/graph/computational_operator.py
index e5ec4ec59303a087f567b05ff21a0b21805b1cd2..53493d07cae0b03878027c79bf77a81cb74b35a5 100644
--- a/hysop/core/graph/computational_operator.py
+++ b/hysop/core/graph/computational_operator.py
@@ -495,7 +495,7 @@ class ComputationalGraphOperator(ComputationalGraphNode):
fields, io_params, op_kwds = self._input_fields_to_dump
variables = {k: self.input_vars[k] for k in fields}
op = HDF_Writer(io_params=io_params, variables=variables, **op_kwds)
- op.initialize()
+ op.initialize(topgraph_method=self.method)
op.get_and_set_field_requirements()
ops.append(op)
@@ -505,7 +505,7 @@ class ComputationalGraphOperator(ComputationalGraphNode):
fields, io_params, op_kwds = self._output_fields_to_dump
variables = {k: self.output_vars[k] for k in fields}
op = HDF_Writer(io_params=io_params, variables=variables, **op_kwds)
- op.initialize()
+ op.initialize(topgraph_method=self.method)
op.get_and_set_field_requirements()
ops.append(op)
diff --git a/hysop/core/graph/graph_builder.py b/hysop/core/graph/graph_builder.py
index 4fa00ddb91221b441be214f4191b5cb1f18f2193..565ddc46d5be584f1c1ccd6d4e6a99200f09592f 100644
--- a/hysop/core/graph/graph_builder.py
+++ b/hysop/core/graph/graph_builder.py
@@ -133,7 +133,8 @@ class GraphBuilder(object):
return self.__ContinuousFieldState(field,
self.op_input_topology_states,
self.op_output_topology_states,
- self.target_node._input_fields_to_dump)
+ self.target_node._input_fields_to_dump,
+ self.target_node.method)
def new_node(self, opname, op, subgraph,
current_level, node, node_id,
@@ -415,7 +416,7 @@ class GraphBuilder(object):
io_leader=io_params.io_leader)
op = HDF_Writer(io_params=io_params, variables=variables,
**op_kwds)
- op.initialize()
+ op.initialize(topgraph_method=self.target_node.method)
op.get_and_set_field_requirements()
opnode = self.new_node(opname, op, None,
current_level, None, None, None, None)
@@ -640,7 +641,7 @@ class GraphBuilder(object):
def __init__(self, field,
op_input_topology_states, op_output_topology_states,
- input_fields_to_dump):
+ input_fields_to_dump, topgraph_method):
# all states are related to this continuous field
self.field = field
@@ -669,6 +670,8 @@ class GraphBuilder(object):
# dictionnary (topology -> TopologyState)
self.discrete_topology_states = {}
+
+ self.method = topgraph_method
def add_vertex(self, graph, vertex_properties, operator):
vertex = graph.add_vertex()
@@ -708,7 +711,7 @@ class GraphBuilder(object):
op.name = '{}_{}'.format(op_name_prefix, field.name)
if len(op_generator.nodes)>1:
op.name += '__{}'.format(i)
- op.initialize()
+ op.initialize(topgraph_method=self.method)
assert len(op.input_vars) == 1
assert len(op.output_vars) == 1
@@ -910,7 +913,7 @@ class GraphBuilder(object):
variables = {ifield: target_topo}
writer_op = HDF_Writer(io_params=io_params, variables=variables,
**op_kwds)
- writer_op.initialize()
+ writer_op.initialize(topgraph_method=self.method)
writer_op.get_and_set_field_requirements()
writer_opnode = self.add_vertex(graph, vertex_properties, writer_op)
self.add_edge(graph, edge_properties,