def __init__(self, device_id=None, comm=None, **kwargs):
super(GPUTransformer, self).__init__(**kwargs)
GPUTransformer.gpu_transformers.add(self)
layout_domain_pass = GenerateLayoutDomains(self)
layout_constraints_pass = GenerateLayoutConstraints(self)
layout_assign_pass = AssignLayouts(layout_domain_pass,
layout_constraints_pass)
layout_convert_pass = AddLayoutConversions(layout_assign_pass)
self.graph_passes = [
SimplePrune(),
PruneContiguousPass(),
GPUSubstitution(), layout_domain_pass, layout_constraints_pass,
layout_assign_pass, layout_convert_pass
] # , VizPass(show_metadata="layout")]
self.buffer_allocators = []
self.kernel_groups = dict()
self.tensors = dict()
self.argmax_tensors = dict()
self.finished_transform = False
self.current_buffer = None
self.device_id = device_id
self.runtime = None
self.comm = comm
def __init__(self, **kwargs):
super(CPUTransformer, self).__init__(**kwargs)
self.current_computation = None
self.conv_engine = CPUConvEngine()
self.init_code = CPUCodeGenerator(self)
self.allocate_storage_code = CPUCodeGenerator(self)
self.allocate_code = CPUCodeGenerator(self)
self.compute_code = CPUCodeGenerator(self)
self.code = CPUCodeGenerator(self)
self.globals = self.code.globals
self.n_computations = 0
self.use_pinned_mem = False
self.rng_seed = None
self.initialize_mkldnn()
add_layout_conversion = AddLayoutConversions(None)
self.graph_passes = [
CPUFusion(),
CPUTensorLayout(),
SimplePrune(),
RequiredTensorShaping(),
CPUTensorShaping()
]
if self.mkldnn.enabled:
self.graph_passes.append(MklCreateOpDescriptors(self.mkldnn)),
self.graph_passes.append(
MklAddLayoutConversions(self.mkldnn, add_layout_conversion))
def __init__(self, **kwargs):
super(CPUTransformer, self).__init__(**kwargs)
self.device_computation = None
self.conv_engine = CPUConvEngine()
self.init_code = CPUCodeGenerator(self)
self.allocate_storage_code = CPUCodeGenerator(self)
self.allocate_code = CPUCodeGenerator(self)
self.code = CPUCodeGenerator(self)
self.globals = PyModule(prefix="op")
self.initialize_module(self.globals)
self.n_computations = 0
self.use_pinned_mem = False
self.rng_seed = None
self.exop_codegen_pools = CPUCodeGenerator(self)
self.exop_codegen_tensor = CPUCodeGenerator(self)
self.exop_codegen_tensor_view = CPUCodeGenerator(self)
self.exop_codegen = CPUCodeGenerator(self)
self.exop_codegen_define_length = 0
self.prefix = ''
# from ngraph.transformers.passes.exnviz import ExVizPass
# from ngraph.transformers.passes.verify import VerifyPass
# from ngraph.transformers.passes.visualizemem import VisualizeMemPass
# from ngraph.transformers.passes.dumpgraphpass import DumpGraphPass
self.graph_passes = []
if self.mkldnn.enabled:
self.graph_passes.append(CPUFusion())
self.graph_passes += [
# ExVizPass(view=True, filename="initial"),
CPUTensorLayout(),
SimplePrune(),
RequiredTensorShaping(),
CPUTensorShaping(),
DeadCodeEliminationPass(),
]
add_layout_conversion = AddLayoutConversions(None)
if self.mkldnn.enabled:
self.graph_passes.append(
MklCreateOpDescriptors(mkldnn=self.mkldnn)),
DeadCodeEliminationPass(),
self.graph_passes.append(
MklAddLayoutConversions(mkldnn=self.mkldnn,
layoutpass=add_layout_conversion)),
DeadCodeEliminationPass()
self.graph_passes += [
SSAConversion(),
IndexElision(),
# DCE here eliminates return values. Need to figure out why.
# DeadCodeEliminationPass(),
LivenessPass(),
MemOptimizePass(),
LivenessPass(),
MemLayoutPass()
]
def __init__(self, fusion=None, **kwargs):
super(Transformer, self).__init__(**kwargs)
self.computations = OrderedSet()
self.all_results = OrderedSet()
self.finalized = False
self.allocated = False
self.initialized = False
self.opids = dict()
self.fusion = fusion
self.device_buffers = OrderedSet()
self.cpu_initializations = []
self.init_computation = None
self.graph_passes = [SimplePrune(), RequiredTensorShaping()]
def __init__(self, **kwargs):
super(GPUTransformer, self).__init__(**kwargs)
self.graph_passes = [
DerivPass(),
SimplePrune(),
RequiredTensorShaping(),
GPUTensorShaping(),
GPUTensorLayout(),
GPUContiguousPrune()
]
self.buffer_allocators = []
self.kernel_groups = dict()
self.tensors = dict()
self.argmax_tensors = dict()
self.finished_transform = False
self.current_buffer = None
def __init__(self, **kwargs):
super(CPUTransformer, self).__init__(**kwargs)
self.conv_engine = CPUConvEngine()
self.init_code = CPUCodeGenerator()
self.allocate_storage_code = CPUCodeGenerator()
self.allocate_code = CPUCodeGenerator()
self.compute_code = CPUCodeGenerator()
self.code = CPUCodeGenerator()
self.globals = self.code.globals
self.n_computations = 0
self.use_pinned_mem = False
self.rng_seed = None
self.graph_passes = [FusionPass(),
CPUTensorLayout(),
SimplePrune(),
RequiredTensorShaping(),
CPUTensorShaping()]
def test_simpleprune_graph_pass():
base_op, simple_graph = get_simple_graph()
output_graph = SimplePrune().do_pass([simple_graph])
assert output_graph.pop() is base_op
def test_simpleprune_graph_pass():
base_op, simple_graph = get_simple_graph()
SimplePrune().do_pass(ops=[simple_graph])
assert simple_graph.forwarded is base_op
def test_simpleprune_graph_pass():
transformer = StubTransformer()
base_op, simple_graph = get_simple_graph()
SimplePrune().do_pass([simple_graph], transformer)
assert simple_graph.forwarded is base_op