示例#1
0
def test_strides_forwarding1():
    grid = Grid(shape=(4, 4))

    a = Array(name='a', dimensions=grid.dimensions, shape=grid.shape)

    bar = Callable('bar',
                   DummyExpr(a[0, 0], 0),
                   'void',
                   parameters=[a.indexed])
    call = Call(bar.name, [a.indexed])
    foo = Callable('foo', call, 'void', parameters=[a])

    # Emulate what the compiler would do
    graph = Graph(foo)
    graph.efuncs['bar'] = bar

    linearize(graph, mode=True, sregistry=SymbolRegistry())

    # Despite `a` is passed via `a.indexed`, and since it's an Array (which
    # have symbolic shape), we expect the stride exprs to be placed in `bar`,
    # and in `bar` only, as `foo` doesn't really use `a`, it just propagates it
    # down to `bar`
    foo = graph.root
    bar = graph.efuncs['bar']

    assert len(foo.body.body) == 1
    assert foo.body.body[0].is_Call

    assert len(bar.body.body) == 5
    assert bar.body.body[0].write.name == 'y_fsz0'
    assert bar.body.body[2].write.name == 'y_stride0'
示例#2
0
def test_strides_forwarding0():
    grid = Grid(shape=(4, 4))

    f = Function(name='f', grid=grid)

    bar = Callable('bar',
                   DummyExpr(f[0, 0], 0),
                   'void',
                   parameters=[f.indexed])
    call = Call(bar.name, [f.indexed])
    foo = Callable('foo', call, 'void', parameters=[f])

    # Emulate what the compiler would do
    graph = Graph(foo)
    graph.efuncs['bar'] = bar

    linearize(graph, mode=True, sregistry=SymbolRegistry())

    # Since `f` is passed via `f.indexed`, we expect the stride exprs to be
    # lifted in `foo` and then passed down to `bar` as arguments
    foo = graph.root
    bar = graph.efuncs['bar']

    assert foo.body.body[0].write.name == 'y_fsz0'
    assert foo.body.body[2].write.name == 'y_stride0'
    assert len(foo.body.body[4].arguments) == 2

    assert len(bar.parameters) == 2
    assert bar.parameters[1].name == 'y_stride0'
    assert len(bar.body.body) == 1
示例#3
0
    def _lower_iet(cls, stree, profiler, **kwargs):
        """
        Iteration/Expression tree lowering:

            * Turn a ScheduleTree into an Iteration/Expression tree;
            * Introduce distributed-memory, shared-memory, and SIMD parallelism;
            * Introduce optimizations for data locality;
            * Finalize (e.g., symbol definitions, array casts)
        """
        name = kwargs.get("name", "Kernel")
        sregistry = kwargs['sregistry']

        # Build an IET from a ScheduleTree
        iet = iet_build(stree)

        # Analyze the IET Sections for C-level profiling
        profiler.analyze(iet)

        # Wrap the IET with an EntryFunction (a special Callable representing
        # the entry point of the generated library)
        parameters = derive_parameters(iet, True)
        iet = EntryFunction(name, iet, 'int', parameters, ())

        # Lower IET to a target-specific IET
        graph = Graph(iet)
        graph = cls._specialize_iet(graph, **kwargs)

        # Instrument the IET for C-level profiling
        # Note: this is postponed until after _specialize_iet because during
        # specialization further Sections may be introduced
        instrument(graph, profiler=profiler, sregistry=sregistry)

        return graph.root, graph
示例#4
0
    def _lower_iet(cls, stree, profiler, **kwargs):
        """
        Iteration/Expression tree lowering:

            * Turn a ScheduleTree into an Iteration/Expression tree;
            * Perform analysis to detect optimization opportunities;
            * Introduce distributed-memory, shared-memory, and SIMD parallelism;
            * Introduce optimizations for data locality;
            * Finalize (e.g., symbol definitions, array casts)
        """
        name = kwargs.get("name", "Kernel")

        iet = iet_build(stree)

        # Instrument the IET for C-level profiling
        iet = profiler.instrument(iet)

        # Wrap the IET with a Callable
        parameters = derive_parameters(iet, True)
        iet = Callable(name, iet, 'int', parameters, ())

        # Lower IET to a target-specific IET
        graph = Graph(iet)
        graph = cls._specialize_iet(graph, **kwargs)

        return graph.root, graph