def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
platform = kwargs['platform']
sregistry = kwargs['sregistry']
# Flush denormal numbers
avoid_denormals(graph)
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# Lower IncrDimensions so that blocks of arbitrary shape may be used
relax_incr_dimensions(graph, sregistry=sregistry)
# Parallelism
parizer = cls._Target.Parizer(sregistry, options, platform)
parizer.make_simd(graph)
parizer.make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
cls._Target.DataManager(sregistry).process(graph)
# Initialize the target-language runtime
parizer.initialize(graph)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
platform = kwargs['platform']
sregistry = kwargs['sregistry']
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# Lower IncrDimensions so that blocks of arbitrary shape may be used
relax_incr_dimensions(graph, sregistry=sregistry)
# SIMD-level parallelism
ompizer = Ompizer(sregistry, options)
ompizer.make_simd(graph, simd_reg_size=platform.simd_reg_size)
# Shared-memory parallelism
ompizer.make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
DataManager(sregistry).process(graph)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
platform = kwargs['platform']
# Flush denormal numbers
avoid_denormals(graph)
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# Lower IncrDimensions so that blocks of arbitrary shape may be used
relax_incr_dimensions(graph, counter=generator())
# SIMD-level parallelism
ompizer = Ompizer()
ompizer.make_simd(graph, simd_reg_size=platform.simd_reg_size)
# Shared-memory parallelism
ompizer.make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
data_manager = DataManager()
data_manager.place_definitions(graph)
data_manager.place_casts(graph)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
sregistry = kwargs['sregistry']
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# GPU parallelism via OpenMP offloading
DeviceOmpizer(sregistry, options).make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
data_manager = DeviceOpenMPDataManager(sregistry)
data_manager.place_ondevice(graph)
data_manager.place_definitions(graph)
data_manager.place_casts(graph)
# Initialize OpenMP environment
initialize(graph)
# TODO: This should be moved right below the `mpiize` pass, but currently calling
# `mpi_gpu_direct` before Symbol definitions` block would create Blocks before
# creating C variables. That would lead to MPI_Request variables being local to
# their blocks. This way, it would generate incorrect C code.
if options['gpu-direct']:
mpi_gpu_direct(graph)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
platform = kwargs['platform']
sregistry = kwargs['sregistry']
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# GPU parallelism
parizer = cls._Target.Parizer(sregistry, options, platform)
parizer.make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
cls._Target.DataManager(sregistry, options).process(graph)
# Initialize the target-language runtime
parizer.initialize(graph)
# TODO: This should be moved right below the `mpiize` pass, but currently calling
# `make_gpudirect` before Symbol definitions` block would create Blocks before
# creating C variables. That would lead to MPI_Request variables being local to
# their blocks. This way, it would generate incorrect C code.
if options['gpu-direct']:
parizer.make_gpudirect(graph)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
platform = kwargs['platform']
compiler = kwargs['compiler']
sregistry = kwargs['sregistry']
# Flush denormal numbers
avoid_denormals(graph, platform=platform)
# Distributed-memory parallelism
mpiize(graph, sregistry=sregistry, options=options)
# Lower BlockDimensions so that blocks of arbitrary shape may be used
relax_incr_dimensions(graph)
# Parallelism
parizer = cls._Target.Parizer(sregistry, options, platform, compiler)
parizer.make_simd(graph)
parizer.make_parallel(graph)
parizer.initialize(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
cls._Target.DataManager(sregistry).process(graph)
# Linearize n-dimensional Indexeds
linearize(graph, mode=options['linearize'], sregistry=sregistry)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
sregistry = kwargs['sregistry']
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# GPU parallelism via OpenACC offloading
DeviceAccizer(sregistry).make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
data_manager = DeviceOpenACCDataManager(sregistry)
data_manager.place_ondevice(graph)
data_manager.place_definitions(graph)
data_manager.place_casts(graph)
# Initialize OpenACC environment
if options['mpi']:
initialize(graph)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
language = kwargs['language']
platform = kwargs['platform']
sregistry = kwargs['sregistry']
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph,
mode=options['mpi'],
language=language,
sregistry=sregistry)
# GPU parallelism
parizer = cls._Target.Parizer(sregistry, options, platform)
parizer.make_parallel(graph)
parizer.initialize(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
cls._Target.DataManager(sregistry, options).process(graph)
# Linearize n-dimensional Indexeds
if options['linearize']:
linearize(graph, sregistry=sregistry)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
platform = kwargs['platform']
# Flush denormal numbers
avoid_denormals(graph)
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# Tiling
blocker = Blocker(options['blockinner'], options['blocklevels']
or cls.BLOCK_LEVELS)
blocker.make_blocking(graph)
# Shared-memory and SIMD-level parallelism
ompizer = Ompizer()
ompizer.make_simd(graph, simd_reg_size=platform.simd_reg_size)
if options['openmp']:
ompizer.make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
data_manager = DataManager()
data_manager.place_definitions(graph)
data_manager.place_casts(graph)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
platform = kwargs['platform']
compiler = kwargs['compiler']
sregistry = kwargs['sregistry']
# Distributed-memory parallelism
mpiize(graph, sregistry=sregistry, options=options)
# Loop tiling
relax_incr_dimensions(graph)
# GPU parallelism
parizer = cls._Target.Parizer(sregistry, options, platform, compiler)
parizer.make_parallel(graph)
parizer.initialize(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
cls._Target.DataManager(sregistry, options).process(graph)
# Linearize n-dimensional Indexeds
linearize(graph, mode=options['linearize'], sregistry=sregistry)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# GPU parallelism via OpenMP offloading
DeviceOmpizer().make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
data_manager = DeviceDataManager()
data_manager.place_ondevice(graph, efuncs=list(graph.efuncs.values()))
data_manager.place_definitions(graph)
data_manager.place_casts(graph)
return graph
def _specialize_iet(cls, graph, **kwargs):
options = kwargs['options']
sregistry = kwargs['sregistry']
# Distributed-memory parallelism
optimize_halospots(graph)
if options['mpi']:
mpiize(graph, mode=options['mpi'])
# Device and host parallelism via OpenACC offloading
accizer = DeviceAccizer(sregistry, options)
accizer.make_parallel(graph)
# Misc optimizations
hoist_prodders(graph)
# Symbol definitions
DeviceOpenACCDataManager(sregistry, options).process(graph)
# Initialize OpenACC environment
initialize(graph)
return graph