Merge pull request #119 from xdslproject/emilien/return-type

Implement stencil.return operand type promotion

devito/ir/xdsl_iet/cluster_to_ssa.py

@@ -287,7 +287,7 @@ def _visit_math_nodes(self, dim: SteppingDimension, node: Expr,
             SSAargs = (self._visit_math_nodes(dim, arg, output_indexed)
                        for arg in node.args)
             return reduce(lambda x, y : arith.AndI(x, y).result, SSAargs)
-        
+
         # Trigonometric functions
         elif isinstance(node, sin):
             assert len(node.args) == 1, "Expected single argument for sin."
@@ -298,13 +298,13 @@ def _visit_math_nodes(self, dim: SteppingDimension, node: Expr,
             assert len(node.args) == 1, "Expected single argument for cos."           
             return math.CosOp(self._visit_math_nodes(dim, node.args[0],
                               output_indexed)).result
-        
+
         elif isinstance(node, tan):
             assert len(node.args) == 1, "Expected single argument for TanOp."
-            
+
             return math.TanOp(self._visit_math_nodes(dim, node.args[0],
                               output_indexed)).result
-                   
+
         elif isinstance(node, Relational):
             if isinstance(node, GreaterThan):
                 mnemonic = "sge"
@@ -382,7 +382,20 @@ def build_stencil_step(self, dim: SteppingDimension, eq: LoweredEq) -> None:
         self.function_values |= self.apply_temps
 
         with ImplicitBuilder(apply.region.block):
-            stencil.ReturnOp.get([self._visit_math_nodes(dim, eq.rhs, eq.lhs)])
+            result = self._visit_math_nodes(dim, eq.rhs, eq.lhs)
+            expected_type = apply.res[0].type.get_element_type()
+            match expected_type:
+                case result.type:
+                    pass
+                case builtin.f32:
+                    if result.type == IndexType():
+                        result = arith.IndexCastOp(result, builtin.i64).result
+                    result = arith.SIToFPOp(result, builtin.f32).result
+                case builtin.IndexType:
+                    result = arith.IndexCastOp(result, IndexType()).result
+                case _:
+                    raise Exception(f"Unexpected result type {type(result)}")
+            stencil.ReturnOp.get([result])
 
         lb = stencil.IndexAttr.get(*([0] * len(shape)))
         ub = stencil.IndexAttr.get(*shape)
@@ -439,7 +452,6 @@ def build_condition(self, dim: SteppingDimension, eq: BooleanFunction):
             self.build_generic_step_expression(dim, eq)
             scf.Yield()
 
-
     def build_time_loop(
         self, eqs: list[Any], step_dim: SteppingDimension, **kwargs
     ):
@@ -450,7 +462,7 @@ def build_time_loop(
         ub = iet_ssa.LoadSymbolic.get(
             step_dim.symbolic_max._C_name, IndexType()
         )
-        
+
         one = arith.Constant.from_int_and_width(1, IndexType())
 
         # Devito iterates from time_m to time_M *inclusive*, MLIR only takes
@@ -497,7 +509,7 @@ def build_time_loop(
             for i, (f, t) in enumerate(self.time_buffers)
         }
         self.function_values |= self.block_args
-        
+
         # Name the block argument for debugging
         for (f, t), arg in self.block_args.items():
             arg.name_hint = f"{f.name}_t{t}"
@@ -513,8 +525,7 @@ def build_time_loop(
 
     def lower_devito_Eqs(self, eqs: list[Any], **kwargs):
         # Lower devito Equations to xDSL
-
-
+
         for eq in eqs:
             lowered = self.operator._lower_exprs(as_tuple(eq), **kwargs)
             if isinstance(eq, Eq):
@@ -546,7 +557,7 @@ def _lower_injection(self, eqs: list[LoweredEq]):
                 lb = arith.Constant.from_int_and_width(int(lower), IndexType())
             else:
                 raise NotImplementedError(f"Lower bound of type {type(lower)} not supported")
-            
+
             try:
                 name = interval.dim.symbolic_min.name
             except:
@@ -633,7 +644,7 @@ def convert(self, eqs: Iterable[Eq], **kwargs) -> ModuleOp:
         # Instantiate the module.
         self.function_values: dict[tuple[Function, int], SSAValue] = {}
         self.symbol_values: dict[str, SSAValue] = {}
-        
+
         module = ModuleOp(Region([block := Block([])]))
         with ImplicitBuilder(block):
             # Get all functions used in the equations
@@ -647,7 +658,7 @@ def convert(self, eqs: Iterable[Eq], **kwargs) -> ModuleOp:
                         functions.add(f.function)
 
                 elif isinstance(eq, Injection):
-                    
+
                     functions.add(eq.field.function)
                     for f in retrieve_functions(eq.expr):
                         if isinstance(f, PointSource):

tests/test_xdsl_base.py

@@ -972,6 +972,20 @@ def test_function_IV():
     assert np.isclose(norm(u), devito_norm_u)
 
 
+def test_function_V():
+    grid = Grid(shape=(5, 5))
+    x, y = grid.dimensions
+
+    f = Function(name="f", grid=grid)
+
+    eqns = [Eq(f, 2)]
+
+    op = Operator(eqns, opt="xdsl")
+    op.apply()
+
+    assert np.all(f.data == 2)
+
+
 class TestTrigonometric(object):
 
     @pytest.mark.parametrize('deg, exp', ([90.0, 3.5759869], [30.0, 3.9521265],
@@ -1028,37 +1042,20 @@ def test_tan(self, deg, exp):
         assert np.isclose(norm(u), exp, rtol=1e-4)
 
 
-class TestOperatorUnsupported(object):
+def test_forward_assignment():
+    # simple forward assignment
 
-    @pytest.mark.xfail(reason="stencil.return operation does not verify for i64")
-    def test_forward_assignment(self):
-        # simple forward assignment
-
-        grid = Grid(shape=(4, 4))
-        u = TimeFunction(name="u", grid=grid, space_order=2)
-        u.data[:, :, :] = 0
-
-        eq0 = Eq(u.forward, 1)
-
-        op = Operator([eq0], opt='xdsl')
-
-        op.apply(time_M=1)
-
-        assert np.isclose(norm(u), 5.6584, rtol=0.001)
-
-    @pytest.mark.xfail(reason="stencil.return operation does not verify for i64")
-    def test_function(self):
-        grid = Grid(shape=(5, 5))
-        x, y = grid.dimensions
+    grid = Grid(shape=(4, 4))
+    u = TimeFunction(name="u", grid=grid, space_order=2)
+    u.data[:, :, :] = 0
 
-        f = Function(name="f", grid=grid)
+    eq0 = Eq(u.forward, 1)
 
-        eqns = [Eq(f, 2)]
+    op = Operator([eq0], opt='xdsl')
 
-        op = Operator(eqns, opt='xdsl')
-        op.apply()
+    op.apply(time_M=1)
 
-        assert np.all(f.data == 4)
+    assert np.isclose(norm(u), 5.6584, rtol=0.001)
 
 
 class TestElastic():