From 1e799e6d026243e194bcb5ece1822ebb8c276121 Mon Sep 17 00:00:00 2001
From: Keck Jean-Baptiste <jean-baptiste.keck@imag.fr>
Date: Thu, 21 Sep 2017 19:47:50 +0200
Subject: [PATCH] fixed transposition, kernel launchers, opencl kernels
---
.../taylor_green/taylor_green_monoscale.py | 14 +-
hysop/__init__.py | 2 +-
.../codegen/kernels/tests/test_transpose.py | 4 +-
.../device/codegen/kernels/transpose.py | 4 +-
.../opencl/autotunable_kernels/transpose.py | 2 +-
.../backend/device/opencl/opencl_allocator.py | 2 +-
.../device/opencl/opencl_array_backend.py | 5 +-
.../opencl/opencl_autotunable_kernel.py | 4 +-
hysop/backend/device/opencl/opencl_copy.py | 1 +
hysop/backend/device/opencl/opencl_kernel.py | 13 +-
.../device/opencl/opencl_kernel_launcher.py | 7 +-
.../backend/device/opencl/opencl_operator.py | 330 +++++++++---------
.../device/opencl/operator/transpose.py | 7 +-
hysop/core/arrays/array.py | 4 +-
hysop/core/arrays/array_backend.py | 3 +-
hysop/core/memory/memory_request.py | 5 +-
hysop/fields/cartesian_discrete_field.py | 3 +-
hysop/operator/base/advection_dir.py | 3 +-
hysop/operator/base/transpose_operator.py | 3 +-
hysop/operator/tests/test_transpose.py | 26 +-
hysop/operator/transpose.py | 1 -
hysop/tools/misc.py | 11 +-
hysop/tools/types.py | 3 +-
hysop/topology/cartesian_topology.py | 8 +-
24 files changed, 244 insertions(+), 221 deletions(-)
diff --git a/examples/taylor_green/taylor_green_monoscale.py b/examples/taylor_green/taylor_green_monoscale.py
index fcfb69787..5b248d6d4 100644
--- a/examples/taylor_green/taylor_green_monoscale.py
+++ b/examples/taylor_green/taylor_green_monoscale.py
@@ -48,12 +48,14 @@ def init_vorticity(data, coords, dim):
data[1][...] = - sin(x) * cos(y) * sin(z)
data[2][...] = 2. * sin(x) * sin(y) * cos(z)
elif dim==2:
- #data[0][...] = cos(x)*cos(y)
(x,y) = coords
- (nx,ny) = (3,3)
- i = np.floor((nx*x)/(2*pi))
- j = np.floor((ny*y)/(2*pi))
- data[0][...] = (-1)**(i+j)
+ #COS-COS
+ data[0][...] = cos(x)*cos(y)
+ #CHECKERBOARD
+ #(nx,ny) = (3,3)
+ #i = np.floor((nx*x)/(2*pi))
+ #j = np.floor((ny*y)/(2*pi))
+ #data[0][...] = (-1)**(i+j)
else:
raise NotImplementedError('dimension {}'.format(dim))
@@ -76,7 +78,7 @@ def run(npts=64+1, viscosity=1./1600., lcfl=0.125, cfl=0.5):
advected_fields = (vorti,),
velocity_cfl = cfl,
variables = {velo: d3d, vorti: d3d},
- method = {TimeIntegrator: RK4, Remesh: Remesh.L2_1},
+ method = {TimeIntegrator: Euler, Remesh: Remesh.L2_1},
)
if dim==3:
stretch = DirectionalStretching(
diff --git a/hysop/__init__.py b/hysop/__init__.py
index ed9bf2cc3..3afa33583 100644
--- a/hysop/__init__.py
+++ b/hysop/__init__.py
@@ -16,7 +16,7 @@ __FFTW_ENABLED__ = "ON" is "ON"
__SCALES_ENABLED__ = "ON" is "ON"
__OPTIMIZE__ = not __debug__
-__VERBOSE__ = True
+__VERBOSE__ = False
__DEBUG__ = False
__TRACE__ = False
__KERNEL_DEBUG__ = False
diff --git a/hysop/backend/device/codegen/kernels/tests/test_transpose.py b/hysop/backend/device/codegen/kernels/tests/test_transpose.py
index cbc15ec7b..16ae23bcd 100644
--- a/hysop/backend/device/codegen/kernels/tests/test_transpose.py
+++ b/hysop/backend/device/codegen/kernels/tests/test_transpose.py
@@ -3,7 +3,7 @@ import copy, math, sys, os, tempfile
from hysop import __ENABLE_LONG_TESTS__
from hysop.deps import np, it
-from hysop.tools.misc import upper_pow2_or_3
+from hysop.tools.misc import upper_pow2_or_3, prod
from hysop.tools.types import check_instance
from hysop.tools.numerics import is_integer
from hysop.backend.device.opencl import cl, clTools
@@ -248,7 +248,7 @@ class TestTranspose(object):
for i in dak.workload_indexes:
local_work_size[i] = 1
max_work = device.max_work_group_size
- while (np.prod(local_work_size) > max_work):
+ while (prod(local_work_size) > max_work):
for i in xrange(work_dim-1,-1,-1):
if local_work_size[i] > 1:
break
diff --git a/hysop/backend/device/codegen/kernels/transpose.py b/hysop/backend/device/codegen/kernels/transpose.py
index 893ebe6c5..83049193b 100644
--- a/hysop/backend/device/codegen/kernels/transpose.py
+++ b/hysop/backend/device/codegen/kernels/transpose.py
@@ -2,7 +2,7 @@ import operator
import numpy as np
from contextlib import contextmanager, nested
-from hysop.tools.misc import upper_pow2_or_3
+from hysop.tools.misc import upper_pow2_or_3, prod
from hysop.tools.decorators import static_vars
from hysop.tools.numpywrappers import npw
from hysop.tools.types import check_instance
@@ -136,7 +136,7 @@ class TransposeKernelGenerator(KernelCodeGenerator):
nbytes = np.dtype(dtype).itemsize
tile_shape = self.tile_shape
- tile_bytes = np.prod(tile_shape) * nbytes
+ tile_bytes = prod(tile_shape) * nbytes
if self.contiguous_permutation:
count = tile_bytes
diff --git a/hysop/backend/device/opencl/autotunable_kernels/transpose.py b/hysop/backend/device/opencl/autotunable_kernels/transpose.py
index 19eb967bf..7c5e175e3 100644
--- a/hysop/backend/device/opencl/autotunable_kernels/transpose.py
+++ b/hysop/backend/device/opencl/autotunable_kernels/transpose.py
@@ -36,7 +36,7 @@ class OpenClAutotunableTransposeKernel(OpenClAutotunableKernel):
check_instance(axes, tuple, values=int)
check_instance(is_inplace, bool)
- self._check_cartesian_fields(input_field, output_field,
+ self.check_cartesian_fields(input_field, output_field,
check_res=False, check_size=True)
dim = input_field.domain.dim
diff --git a/hysop/backend/device/opencl/opencl_allocator.py b/hysop/backend/device/opencl/opencl_allocator.py
index 79da1e550..0dbdb782a 100644
--- a/hysop/backend/device/opencl/opencl_allocator.py
+++ b/hysop/backend/device/opencl/opencl_allocator.py
@@ -107,4 +107,4 @@ class OpenClImmediateAllocator(OpenClAllocator):
if isinstance(self, MemoryPool):
msg='allocator is already a memory pool.'
raise RuntimeError(msg)
- return OpenClMemoryPool(allocator=self, name=name, verbose=True, **kwds)
+ return OpenClMemoryPool(allocator=self, name=name, verbose=None, **kwds)
diff --git a/hysop/backend/device/opencl/opencl_array_backend.py b/hysop/backend/device/opencl/opencl_array_backend.py
index e65e3fc62..26d570a99 100644
--- a/hysop/backend/device/opencl/opencl_array_backend.py
+++ b/hysop/backend/device/opencl/opencl_array_backend.py
@@ -3,6 +3,7 @@ import warnings
from hysop.deps import re, np, os
from hysop.tools.types import check_instance, to_tuple
from hysop.tools.hash import hash_id
+from hysop.tools.misc import prod
from hysop.tools.numerics import is_complex, get_dtype, float_to_complex_dtype, \
complex_to_float_dtype, find_common_dtype
from hysop.constants import Backend
@@ -1366,8 +1367,8 @@ class OpenClArrayBackend(ArrayBackend):
if min_alignment < self.allocator.device.mem_base_addr_align:
alignment=1
dtype = np.dtype(dtype)
- size = int(np.prod(shape)*dtype.itemsize)
- # np.prod( shape=(,) ) will return 1.0, so we cast to int for scalars
+ size = int(prod(shape)*dtype.itemsize)
+ # prod( shape=(,) ) will return 1.0, so we cast to int for scalars
else:
(size,nbytes,alignment) = self.get_alignment_and_size(shape=shape,
dtype=dtype, min_alignment=min_alignment)
diff --git a/hysop/backend/device/opencl/opencl_autotunable_kernel.py b/hysop/backend/device/opencl/opencl_autotunable_kernel.py
index 952ceb7e1..67c50ce87 100644
--- a/hysop/backend/device/opencl/opencl_autotunable_kernel.py
+++ b/hysop/backend/device/opencl/opencl_autotunable_kernel.py
@@ -21,7 +21,7 @@ class OpenClAutotunableKernel(AutotunableKernel):
autotuner_config=autotuner_config,
build_opts=build_opts, **kwds)
- self._check_cl_env(cl_env)
+ self.check_cl_env(cl_env)
self.cl_env = cl_env
self.usable_cache_bytes_per_wg = clCharacterize.usable_local_mem_size(cl_env.device)
@@ -126,7 +126,7 @@ class OpenClAutotunableKernel(AutotunableKernel):
"""
return self.cl_env.device.max_work_item_sizes
- def check_cache(required_cache_size):
+ def check_cache(self, required_cache_size):
"""Check that required_cache_size bytes can fit in workgroup cache."""
usable_cache_bytes_per_wg = self.usable_cache_bytes_per_wg
if (required_cache_size > usable_cache_bytes_per_wg):
diff --git a/hysop/backend/device/opencl/opencl_copy.py b/hysop/backend/device/opencl/opencl_copy.py
index 7feceb58a..b75dd13ef 100644
--- a/hysop/backend/device/opencl/opencl_copy.py
+++ b/hysop/backend/device/opencl/opencl_copy.py
@@ -48,6 +48,7 @@ class OpenClCopyKernelLauncher(OpenClKernelLauncher):
queue = first_not_None(queue, self._default_queue)
check_instance(queue, cl.CommandQueue)
evt = cl.enqueue_copy(queue=queue, **self._enqueue_copy_kwds)
+ return evt
def global_size_configured(self):
return True
diff --git a/hysop/backend/device/opencl/opencl_kernel.py b/hysop/backend/device/opencl/opencl_kernel.py
index aa89b36ac..beb71d8c4 100644
--- a/hysop/backend/device/opencl/opencl_kernel.py
+++ b/hysop/backend/device/opencl/opencl_kernel.py
@@ -58,8 +58,8 @@ class OpenClKernel(object):
self._default_local_work_size = default_local_work_size
self._kid = 0
- assert len(program.all_kernels)==1
- self._kernel = program.all_kernels[0]
+ assert len(program.all_kernels())==1
+ self._kernel = program.all_kernels()[0]
default_args = default_args or {}
self.default_args = default_args
@@ -89,7 +89,6 @@ class OpenClKernel(object):
def _set_default_args(self, default_args):
"""Set default arguments."""
args_mapping = self._args_mapping
- default_args = self._default_args
nargs = len(args_mapping)
positions = ()
for argname in default_args:
@@ -154,18 +153,18 @@ class OpenClKernel(object):
args_list=args_list, parameters_map=parameters_map,
default_global_work_size=global_work_size,
default_local_work_size=local_work_size,
- default_queue=default_queue)
+ default_queue=queue)
else:
klauncher = OpenClKernelLauncher(name=name, kernel=self._program,
args_list=args_list,
default_global_work_size=global_work_size,
default_local_work_size=local_work_size,
- default_queue=default_queue)
+ default_queue=queue)
self._kid += 1
return klauncher
- def _compute_args_list(**kwds):
+ def _compute_args_list(self, **kwds):
"""
Compute argument list from default arguments and input keywords.
If all arguments are not specified, also return a parameter_map
@@ -182,7 +181,7 @@ class OpenClKernel(object):
args_per_index = (len(parameters_map)>0)
args_list = ()
- for (arg_name, arg_value) in arguments:
+ for (arg_name, arg_value) in arguments.iteritems():
if (arg_name not in args_mapping):
msg='Unknown argument {}, valid ones are {}.'
msg=msg.format(arg_name, ', '.join(args_mapping.keys()))
diff --git a/hysop/backend/device/opencl/opencl_kernel_launcher.py b/hysop/backend/device/opencl/opencl_kernel_launcher.py
index 6e43f5bb7..c744d101c 100644
--- a/hysop/backend/device/opencl/opencl_kernel_launcher.py
+++ b/hysop/backend/device/opencl/opencl_kernel_launcher.py
@@ -100,6 +100,11 @@ class OpenClKernelListLauncher(object):
raise TypeError(msg)
return self
+ def __iadd__(self, kernel):
+ """Push a kernel into the list."""
+ self.push_kernels(kernel)
+ return self
+
def __call__(self, queue, wait_for=None, **kwds):
"""
Enqueue all kernels on the given queue in order.
@@ -347,7 +352,7 @@ class OpenClKernelLauncher(object):
dprint(self._apply_msg.format(global_work_size, local_work_size))
- kernel = self.set_kernel_args(**kwds)
+ kernel = self._set_kernel_args(**kwds)
evt = cl.enqueue_nd_range_kernel(queue=queue, kernel=kernel,
global_work_size=global_work_size,
diff --git a/hysop/backend/device/opencl/opencl_operator.py b/hysop/backend/device/opencl/opencl_operator.py
index 2e21e228a..569cfa878 100644
--- a/hysop/backend/device/opencl/opencl_operator.py
+++ b/hysop/backend/device/opencl/opencl_operator.py
@@ -35,190 +35,190 @@ class OpenClOperator(ComputationalGraphOperator):
@classmethod
def default_method(cls):
- dm = super(DirectionalAdvectionBase, cls).default_method()
+ dm = super(OpenClOperator, cls).default_method()
dm.update(cls.__default_method)
return dm
@classmethod
def available_methods(cls):
- am = super(DirectionalAdvectionBase, cls).available_methods()
+ am = super(OpenClOperator, cls).available_methods()
am.update(cls.__available_methods)
return am
-@debug
-def __init__(self, cl_env=None, mpi_params=None, **kwds):
- """
- Create the common attributes of all OpenCL operators.
- See handle_method() and setup().
-
- All input and output variable topologies should be of kind
- Backend.OPENCL and share the same OpenClEnvironment.
-
- Attributes
- ----------
- cl_env: OpenClEnvironment
- OpenCL environment shared accross all topologies.
-
- Notes
- -----
- About method keys:
- OpenClKernelConfig: user build options, defines, precision
- and autotuner configuration
- """
- check_instance(cl_env, OpenClEnvironment, allow_none=True)
- check_instance(mpi_params, MPIParams, allow_none=True)
-
- msg='mpi_params was {} and cl_env was {}.'
- if (mpi_params is None):
- if (cl_env is None):
- _vars = kwds.get('input_vars', None) or kwds.get('output_vars', None)
- assert (_vars is not None), 'No input or output variables.'
- domain = _vars.keys()[0].domain
- mpi_params = MPIParams(comm=domain.task_comm,
- task_id=domain.current_task())
- cl_env = get_or_create_opencl_env(mpi_params)
- msg=msg.format('None', 'None')
- else:
- mpi_params = cl_env.mpi_params
- msg=msg.format('None', 'given')
- else:
- if (cl_env is None):
- cl_env = get_or_create_opencl_env(mpi_params)
- msg=msg.format('given', 'None')
+ @debug
+ def __init__(self, cl_env=None, mpi_params=None, **kwds):
+ """
+ Create the common attributes of all OpenCL operators.
+ See handle_method() and setup().
+
+ All input and output variable topologies should be of kind
+ Backend.OPENCL and share the same OpenClEnvironment.
+
+ Attributes
+ ----------
+ cl_env: OpenClEnvironment
+ OpenCL environment shared accross all topologies.
+
+ Notes
+ -----
+ About method keys:
+ OpenClKernelConfig: user build options, defines, precision
+ and autotuner configuration
+ """
+ check_instance(cl_env, OpenClEnvironment, allow_none=True)
+ check_instance(mpi_params, MPIParams, allow_none=True)
+
+ msg='mpi_params was {} and cl_env was {}.'
+ if (mpi_params is None):
+ if (cl_env is None):
+ _vars = kwds.get('input_vars', None) or kwds.get('output_vars', None)
+ assert (_vars is not None), 'No input or output variables.'
+ domain = _vars.keys()[0].domain
+ mpi_params = MPIParams(comm=domain.task_comm,
+ task_id=domain.current_task())
+ cl_env = get_or_create_opencl_env(mpi_params)
+ msg=msg.format('None', 'None')
+ else:
+ mpi_params = cl_env.mpi_params
+ msg=msg.format('None', 'given')
else:
- msg=msg.format('given', 'given')
- pass
-
- super(OpenClOperator, self).__init__(mpi_params=mpi_params, **kwds)
- self.cl_env = cl_env
-
- if (cl_env.mpi_params is not self.mpi_params):
- msg0='MPI Communicators do not match between OpenClEnvironment and MPIParams.'
- msg0+='\n => {}'.format(msg)
- raise RuntimeError(msg0)
+ if (cl_env is None):
+ cl_env = get_or_create_opencl_env(mpi_params)
+ msg=msg.format('given', 'None')
+ else:
+ msg=msg.format('given', 'given')
+ pass
+
+ super(OpenClOperator, self).__init__(mpi_params=mpi_params, **kwds)
+ self.cl_env = cl_env
+
+ if (cl_env.mpi_params is not self.mpi_params):
+ msg0='MPI Communicators do not match between OpenClEnvironment and MPIParams.'
+ msg0+='\n => {}'.format(msg)
+ raise RuntimeError(msg0)
-def supported_backends(self):
- """
- Return the backends that this operator's topologies can support.
- """
- return set([Backend.OPENCL])
+ def supported_backends(self):
+ """
+ Return the backends that this operator's topologies can support.
+ """
+ return set([Backend.OPENCL])
-@debug
-def handle_method(self, method):
- """
- Extract device configuration and precision from OpenClKernelConfig.
- """
- super(OpenClOperator,self).handle_method(method)
-
- assert OpenClKernelConfig in method
+ @debug
+ def handle_method(self, method):
+ """
+ Extract device configuration and precision from OpenClKernelConfig.
+ """
+ super(OpenClOperator,self).handle_method(method)
- kernel_config = method.pop(OpenClKernelConfig)
- autotuner_config = kernel_config.autotuner_config
-
- precision = kernel_config.precision
-
- if precision == Precision.SAME:
- precision = self.cl_env.precision
- elif precision in [Precision.LONG_DOUBLE, Precision.QUAD]:
- msg='Precision {} is not supported for OpenCl environment.'
- msg=msg.format(precision)
- raise ValueError(msg)
- else:
- from hysop.backend.device.opencl.opencl_tools import convert_precision
- precision = convert_precision(precision)
-
- self.precision = precision
- self.autotuner_config = autotuner_config
-
- self._initialize_cl_build_options(kernel_config.user_build_options)
- self._initialize_cl_size_constants(kernel_config.user_size_constants)
-
-def check(self):
- super(OpenClOperator, self).check()
- self._check_cl_env()
+ assert OpenClKernelConfig in method
+
+ kernel_config = method.pop(OpenClKernelConfig)
+ autotuner_config = kernel_config.autotuner_config
+
+ precision = kernel_config.precision
+
+ if precision == Precision.SAME:
+ precision = self.cl_env.precision
+ elif precision in [Precision.LONG_DOUBLE, Precision.QUAD]:
+ msg='Precision {} is not supported for OpenCl environment.'
+ msg=msg.format(precision)
+ raise ValueError(msg)
+ else:
+ from hysop.backend.device.opencl.opencl_tools import convert_precision
+ precision = convert_precision(precision)
+
+ self.precision = precision
+ self.autotuner_config = autotuner_config
+
+ self._initialize_cl_build_options(kernel_config.user_build_options)
+ self._initialize_cl_size_constants(kernel_config.user_size_constants)
+
+ def check(self):
+ super(OpenClOperator, self).check()
+ self._check_cl_env()
-@debug
-def get_field_requirements(self):
- """
- called just after handle_method(), ie self.method has been set.
- topology requirements are:
- 1) min and max ghosts for each input and output variables
- 2) allowed splitting directions for cartesian topologies
- 3) required local and global transposition state, if any.
- and more
- they are stored in self.input_field_requirements and
- self.output_field_requirements.
-
- keys are continuous fields and values are of type
- hysop.fields.field_requirement.discretefieldrequirements
-
- default is backend.opencl, no min or max ghosts, basis.cartesian and no specific
- transposition state for each input and output variables.
- """
+ @debug
+ def get_field_requirements(self):
+ """
+ called just after handle_method(), ie self.method has been set.
+ topology requirements are:
+ 1) min and max ghosts for each input and output variables
+ 2) allowed splitting directions for cartesian topologies
+ 3) required local and global transposition state, if any.
+ and more
+ they are stored in self.input_field_requirements and
+ self.output_field_requirements.
+
+ keys are continuous fields and values are of type
+ hysop.fields.field_requirement.discretefieldrequirements
+
+ default is backend.opencl, no min or max ghosts, basis.cartesian and no specific
+ transposition state for each input and output variables.
+ """
- # by default we create OPENCL (gpu) TopologyDescriptors
- for field, topo_descriptor in self.input_vars.iteritems():
- topo_descriptor = TopologyDescriptor.build_descriptor(
- backend=Backend.OPENCL,
- operator=self,
- field=field,
- handle=topo_descriptor,
- cl_env=self.cl_env)
- self.input_vars[field] = topo_descriptor
-
- for field, topo_descriptor in self.output_vars.iteritems():
- topo_descriptor = TopologyDescriptor.build_descriptor(
- backend=Backend.OPENCL,
- operator=self,
- field=field,
- handle=topo_descriptor,
- cl_env=self.cl_env)
- self.output_vars[field] = topo_descriptor
-
- return super(OpenClOperator, self).get_field_requirements()
-
-def build_options(self):
- """
- Build and return opencl build option string from
- self._cl_build_options and self._cl_defines.
- """
- build_options = self._cl_build_options
- defines = set()
- for define,value in self._cl_defines.iteritems():
- if (value is not None):
- define = '{}={}'.format(define.strip(), value.strip())
- else:
- define = define.strip()
- defines.update(define)
- return ' '.join(build_options) + ' -D'.join(defines)
+ # by default we create OPENCL (gpu) TopologyDescriptors
+ for field, topo_descriptor in self.input_vars.iteritems():
+ topo_descriptor = TopologyDescriptor.build_descriptor(
+ backend=Backend.OPENCL,
+ operator=self,
+ field=field,
+ handle=topo_descriptor,
+ cl_env=self.cl_env)
+ self.input_vars[field] = topo_descriptor
+
+ for field, topo_descriptor in self.output_vars.iteritems():
+ topo_descriptor = TopologyDescriptor.build_descriptor(
+ backend=Backend.OPENCL,
+ operator=self,
+ field=field,
+ handle=topo_descriptor,
+ cl_env=self.cl_env)
+ self.output_vars[field] = topo_descriptor
+
+ return super(OpenClOperator, self).get_field_requirements()
+
+ def build_options(self):
+ """
+ Build and return opencl build option string from
+ self._cl_build_options and self._cl_defines.
+ """
+ build_options = self._cl_build_options
+ defines = set()
+ for define,value in self._cl_defines.iteritems():
+ if (value is not None):
+ define = '{}={}'.format(define.strip(), value.strip())
+ else:
+ define = define.strip()
+ defines.update(define)
+ return ' '.join(build_options) + ' -D'.join(defines)
-@debug
-def _check_cl_env(self):
- """
- Check if all topologies are on OpenCL backend and check that all opencl environments
- match.
- """
- topo = (self.input_vars.values()+self.output_vars.values())[0]
- assert isinstance(topo, Topology)
- assert topo.backend.kind == Backend.OPENCL
- ref_env = self.cl_env
-
- for topo in set(self.input_vars.values()+self.output_vars.values()):
+ @debug
+ def _check_cl_env(self):
+ """
+ Check if all topologies are on OpenCL backend and check that all opencl environments
+ match.
+ """
+ topo = (self.input_vars.values()+self.output_vars.values())[0]
assert isinstance(topo, Topology)
assert topo.backend.kind == Backend.OPENCL
- assert topo.cl_env == ref_env
+ ref_env = self.cl_env
+
+ for topo in set(self.input_vars.values()+self.output_vars.values()):
+ assert isinstance(topo, Topology)
+ assert topo.backend.kind == Backend.OPENCL
+ assert topo.cl_env == ref_env
-@debug
-def _initialize_cl_build_options(self, user_options):
- """
- Initialize OpenCl build options.
- """
- check_instance(user_options, list)
- build_options = set()
- build_options.update(self.cl_env.default_build_opts)
- build_options.update(user_options)
- self._cl_build_options = build_options
+ @debug
+ def _initialize_cl_build_options(self, user_options):
+ """
+ Initialize OpenCl build options.
+ """
+ check_instance(user_options, list)
+ build_options = set()
+ build_options.update(self.cl_env.default_build_opts)
+ build_options.update(user_options)
+ self._cl_build_options = build_options
@debug
def _initialize_cl_size_constants(self, user_size_constants):
diff --git a/hysop/backend/device/opencl/operator/transpose.py b/hysop/backend/device/opencl/operator/transpose.py
index 1f89e509c..518d88423 100644
--- a/hysop/backend/device/opencl/operator/transpose.py
+++ b/hysop/backend/device/opencl/operator/transpose.py
@@ -7,9 +7,14 @@ from hysop.tools.decorators import debug
from hysop.operator.base.transpose_operator import TransposeOperatorBase
from hysop.backend.device.opencl.opencl_operator import OpenClOperator, op_apply
from hysop.backend.device.opencl.autotunable_kernels.transpose import OpenClAutotunableTransposeKernel
+from hysop.backend.device.opencl.opencl_kernel_launcher import OpenClKernelListLauncher
class OpenClTranspose(TransposeOperatorBase, OpenClOperator):
+ @debug
+ def __init__(self, **kwds):
+ super(OpenClTranspose, self).__init__(**kwds)
+
@debug
def discretize(self):
super(OpenClTranspose,self).discretize()
@@ -38,7 +43,7 @@ class OpenClTranspose(TransposeOperatorBase, OpenClOperator):
(transpose, compute_inplace) = kernel.autotune(axes=axes,
is_inplace=is_inplace, input_field=input_field, output_field=output_field)
- launcher = OpenClKernelListLauncher(name=kernel.name)
+ launcher = OpenClKernelListLauncher(name=transpose.name)
for i in xrange(self.nb_components):
if compute_inplace:
launcher += transpose.build_launcher(inout=input_field[i].data)
diff --git a/hysop/core/arrays/array.py b/hysop/core/arrays/array.py
index 3a5f850f8..d5fd4369e 100644
--- a/hysop/core/arrays/array.py
+++ b/hysop/core/arrays/array.py
@@ -3,9 +3,11 @@ import numpy as np
from abc import ABCMeta, abstractmethod
from hysop.constants import MemoryOrdering
from hysop.constants import DirectionLabels, default_order
+from hysop.tools.misc import prod
from hysop.tools.types import check_instance
from hysop.tools.numpywrappers import slices_empty
from hysop.tools.decorators import required_property, optional_property
+
class Array(object):
"""
@@ -237,7 +239,7 @@ class Array(object):
"""
Number of elements in the array.
"""
- return np.prod(self.get_shape())
+ return prod(self.get_shape())
def get_itemsize(self):
"""
diff --git a/hysop/core/arrays/array_backend.py b/hysop/core/arrays/array_backend.py
index eabbac65f..9f6c9d78b 100644
--- a/hysop/core/arrays/array_backend.py
+++ b/hysop/core/arrays/array_backend.py
@@ -4,6 +4,7 @@ from hysop.deps import np, sys
from hysop.constants import default_order, MemoryOrdering
from hysop.constants import HYSOP_REAL, HYSOP_COMPLEX
from hysop.constants import HYSOP_INTEGER, HYSOP_INDEX, HYSOP_DIM, HYSOP_BOOL
+from hysop.tools.misc import prod
from hysop.tools.types import check_instance, to_tuple, to_list
from hysop.tools.numerics import is_fp, is_complex, match_float_type, \
match_complex_type, complex_to_float_dtype
@@ -93,7 +94,7 @@ class ArrayBackend(object):
"""
bytes_per_elem = dtype.itemsize
min_alignment = min_alignment or 1
- min_alloc_bytes = np.prod(shape) * bytes_per_elem
+ min_alloc_bytes = prod(shape) * bytes_per_elem
msg0='min_alignment is not a power of two, got {}.'
msg1='bytes_per_elem is not a power of two, got {}.'
diff --git a/hysop/core/memory/memory_request.py b/hysop/core/memory/memory_request.py
index 3efc91348..2facdf55d 100644
--- a/hysop/core/memory/memory_request.py
+++ b/hysop/core/memory/memory_request.py
@@ -5,6 +5,7 @@ from hysop.deps import np, copy
from hysop.tools.types import check_instance, first_not_None
from hysop.tools.units import bytes2str
from hysop.tools.numerics import get_dtype
+from hysop.tools.misc import prod
from hysop.core.arrays.array_backend import ArrayBackend
from hysop.constants import HYSOP_BOOL, Backend
@@ -78,9 +79,9 @@ class MemoryRequest(object):
msg += '\nSpecify one either explicitely or through a numpy.dtype.'
raise ValueError(msg)
- if np.prod(shape) != size:
+ if prod(shape) != size:
msg='Shape does not match size (size={}, prod(shape)={}).'
- msg.format(size, np.prod(shape))
+ msg.format(size, prod(shape))
raise ValueError(msg)
if alignment <= 0:
msg = 'Alignment should be positive (got {}).'.format(alignment)
diff --git a/hysop/fields/cartesian_discrete_field.py b/hysop/fields/cartesian_discrete_field.py
index 5f216f2ca..add6b4226 100644
--- a/hysop/fields/cartesian_discrete_field.py
+++ b/hysop/fields/cartesian_discrete_field.py
@@ -12,6 +12,7 @@ from hysop.constants import Backend
from hysop.tools.decorators import debug
from hysop.tools.types import check_instance, to_tuple
from hysop.tools.numpywrappers import npw
+from hysop.tools.misc import prod
from hysop.fields.discrete_field import DiscreteField, DiscreteFieldView
from hysop.topology.cartesian_topology import CartesianTopologyState
@@ -219,7 +220,7 @@ class CartesianDiscreteFieldView(DiscreteFieldView):
def _get_size(self):
"""Size of the underlying contiguous data arrays."""
- return np.prod(self.shape, dtype=np.int32)
+ return prod(self.shape)
def _get_shape(self):
"""Alias for resolution."""
return self.mesh.local_resolution
diff --git a/hysop/operator/base/advection_dir.py b/hysop/operator/base/advection_dir.py
index 769705b5c..dc476007d 100644
--- a/hysop/operator/base/advection_dir.py
+++ b/hysop/operator/base/advection_dir.py
@@ -208,6 +208,7 @@ class DirectionalAdvectionBase(object):
request, mesh = MemoryRequest.dfield_like(a=f, ghosts=0, nb_components=1)
requests.push_mem_request('position', request)
assert all(self.dadvected_fields_in.values()[0].shape == mesh.local_resolution)
+ self.position_mesh = mesh
return requests
@debug
@@ -222,7 +223,7 @@ class DirectionalAdvectionBase(object):
"""
if (work is None):
raise ValueError('work is None.')
- self.dposition, self.position_mesh = work.get_buffer(self, 'position')
+ self.dposition, = work.get_buffer(self, 'position')
## Backend methods
# ComputationalNode
diff --git a/hysop/operator/base/transpose_operator.py b/hysop/operator/base/transpose_operator.py
index 22bc6bcac..46779dc5a 100644
--- a/hysop/operator/base/transpose_operator.py
+++ b/hysop/operator/base/transpose_operator.py
@@ -54,8 +54,7 @@ class TransposeOperatorBase(object):
input_vars = { input_field: variables[input_field] }
output_vars = { output_field: variables[output_field] }
- print kwds
- super(TransposeOperatorBase,self).__init__(input_vars=input_vars,
+ super(TransposeOperatorBase, self).__init__(input_vars=input_vars,
output_vars=output_vars, **kwds)
self.input_field = input_field
diff --git a/hysop/operator/tests/test_transpose.py b/hysop/operator/tests/test_transpose.py
index 315cd25ee..bd78960ef 100644
--- a/hysop/operator/tests/test_transpose.py
+++ b/hysop/operator/tests/test_transpose.py
@@ -21,7 +21,7 @@ class TestTransposeOperator(object):
IO.set_default_path('/tmp/hysop_tests/test_transpose')
if enable_debug_mode:
- cls.size_min = 3
+ cls.size_min = 2
cls.size_max = 6
else:
cls.size_min = 2
@@ -195,9 +195,9 @@ class TestTransposeOperator(object):
np.uint8, np.uint16, np.uint32, np.uint64,
np.float32, np.float64]
random.shuffle(types)
- for i in xrange(5,17):
+ for i in xrange(5,9):
self._test(dim=i, dtype=types[i%len(types)], is_inplace=False,
- size_min=2, size_max=4, naxes=1)
+ size_min=3, size_max=4, naxes=1)
def test_2d_int_inplace(self):
self._test(dim=2, dtype=np.int32, is_inplace=True)
@@ -210,14 +210,20 @@ class TestTransposeOperator(object):
def test_4d_int_inplace(self):
self._test(dim=4, dtype=np.int32, is_inplace=True)
def test_upper_dimensions_inplace(self):
- raise NotImplementedError()
+ types = [np.int8, np.int16, np.int32, np.int64,
+ np.uint8, np.uint16, np.uint32, np.uint64,
+ np.float32, np.float64]
+ random.shuffle(types)
+ for i in xrange(5,9):
+ self._test(dim=i, dtype=types[i%len(types)], is_inplace=True,
+ size_min=3, size_max=4, naxes=1)
def perform_tests(self):
- #self.test_2d_int_out_of_place()
- #self.test_2d_float_out_of_place()
- #self.test_3d_int_out_of_place()
- #self.test_3d_float_out_of_place()
- #self.test_4d_int_out_of_place()
+ self.test_2d_int_out_of_place()
+ self.test_2d_float_out_of_place()
+ self.test_3d_int_out_of_place()
+ self.test_3d_float_out_of_place()
+ self.test_4d_int_out_of_place()
self.test_upper_dimensions_out_of_place()
self.test_2d_int_inplace()
@@ -229,7 +235,7 @@ class TestTransposeOperator(object):
if __name__ == '__main__':
TestTransposeOperator.setup_class(enable_extra_tests=False,
- enable_debug_mode=True)
+ enable_debug_mode=False)
test = TestTransposeOperator()
test.perform_tests()
diff --git a/hysop/operator/transpose.py b/hysop/operator/transpose.py
index 348951ac0..4c3cd8ba9 100644
--- a/hysop/operator/transpose.py
+++ b/hysop/operator/transpose.py
@@ -56,7 +56,6 @@ class Transpose(ComputationalGraphNodeGenerator):
@debug
def __init__(self, fields, variables, axes,
output_fields=None,
- #components_in=None, components_out=None,
implementation=None,
name=None,
base_kwds=None,
diff --git a/hysop/tools/misc.py b/hysop/tools/misc.py
index 6f3e2cd8a..8e10e0618 100644
--- a/hysop/tools/misc.py
+++ b/hysop/tools/misc.py
@@ -10,11 +10,11 @@
from hysop.deps import inspect, np, functools, operator
from hysop.constants import HYSOP_REAL, HYSOP_INTEGER
-def prod(values, neutral=1):
+def prod(values):
"""
Like sum but for products.
"""
- return functools.reduce(operator.mul, values, neutral)
+ return np.prod(values, dtype=np.int64)
def get_default_args(func):
"""
@@ -213,11 +213,10 @@ class WorkSpaceTools(object):
"""
from hysop.tools.numpywrappers import npw
result = []
- # print subshape, len(subshape), np.prod(subshape[0])
if isinstance(subshape, list):
- subsize = [np.prod(subshape[i]) for i in xrange(len(subshape))]
+ subsize = [prod(subshape[i]) for i in xrange(len(subshape))]
else:
- subsize = [np.prod(subshape), ] * lwork
+ subsize = [prod(subshape), ] * lwork
subshape = [subshape, ] * lwork
if work is None:
for i in xrange(lwork):
@@ -352,7 +351,7 @@ class WorkSpaceTools(object):
for prop in properties:
lp = len(prop)
for i in xrange(lp):
- shapes[i] = tuple(np.maximum(shapes[i], np.prod(prop[i])))
+ shapes[i] = tuple(np.maximum(shapes[i], prod(prop[i])))
work = [npw.zeros(shape) for shape in shapes]
return work
diff --git a/hysop/tools/types.py b/hysop/tools/types.py
index 16fdf053f..0c72c3b5d 100644
--- a/hysop/tools/types.py
+++ b/hysop/tools/types.py
@@ -1,6 +1,7 @@
from hysop.deps import np
from collections import Iterable
+from hysop.tools.misc import prod
class InstanceOf(object):
def __init__(self, cls):
@@ -152,7 +153,7 @@ def check_instance(val, cls, allow_none=False, **kargs):
maxval = kargs.pop('maxval', None)
all_val_cls = kargs.pop('values', None)
if shape and size:
- assert np.prod(shape)==size
+ assert prod(shape)==size
if dtype and (dtype != val.dtype):
msg='np.ndarray dtype does not match, expected {} but got {}.'
msg=msg.format(dtype, val.dtype)
diff --git a/hysop/topology/cartesian_topology.py b/hysop/topology/cartesian_topology.py
index 7a3e6add9..026ed5034 100644
--- a/hysop/topology/cartesian_topology.py
+++ b/hysop/topology/cartesian_topology.py
@@ -15,7 +15,7 @@ from hysop.domain.box import Box, BoxView
from hysop.core.mpi import MPI
from hysop.tools.types import check_instance, to_tuple, first_not_None
from hysop.tools.parameters import Discretization, MPIParams
-from hysop.tools.misc import Utils
+from hysop.tools.misc import Utils, prod
from hysop.tools.decorators import debug, deprecated
from hysop.tools.numpywrappers import npw
from hysop.tools.string_utils import prepend
@@ -688,7 +688,7 @@ class CartesianTopology(CartesianTopologyView, Topology):
msg = 'The given cartesian topology does not fit with the others'
msg += ' input parameters.'
assert (dim is None) or (cartesian_topology.dim == dim), msg
- assert (shape is None) or (cartesian_topology.size == np.prod(shape)), msg
+ assert (shape is None) or (cartesian_topology.size == prod(shape)), msg
assert (shape is None) or (shape ==
npw.asintegerarray(cartesian_topology.dims)).all(), msg
assert (is_periodic is None) or (is_periodic == cartesian_topology.periods), msg
@@ -753,13 +753,13 @@ class CartesianTopology(CartesianTopologyView, Topology):
is_distributed[-1] = True
cart_shape = shape[is_distributed]
cart_dim = cart_shape.size
- cart_size = np.prod(cart_shape)
+ cart_size = prod(cart_shape)
is_periodic = is_periodic * is_distributed
is_distributed = is_distributed
assert (cart_dim>0) and (cart_dim <= domain_dim)
- assert np.prod(cart_shape) == np.prod(shape)
+ assert prod(cart_shape) == prod(shape)
assert (cart_shape <= shape[is_distributed]).all()
assert cart_size <= parent_size
assert is_periodic.size == domain_dim
--
GitLab