Update AST and Shape() implementation (tlc-pack#5)

* Update AST.

* ShapeOf.

* ShapeOf.

* Address comment.

Author: ZihengJiang

include/tvm/ir/expr.h

@@ -154,12 +154,13 @@ class RelayExprNode : public BaseExprNode {
   mutable Type checked_type_ = Type(nullptr);
 
   /*!
-  * \brief Stores the result of static shape analysis.
+  * \brief Stores the result of static shape analysis. It must be a RelayExpr
+  * and ObjectRef is used here to avoid cyclic typing.
   *
   * \note The value will be optional if a static shape can not be inferred.
   * use .shape() instead to acesss an always defined shape expression.
   */
-  Optional<Array<PrimExpr>> shape_ = Optional<Array<PrimExpr>>();
+  mutable Optional<ObjectRef> shape_ = Optional<ObjectRef>();
 
   /*!
    * \return The checked_type
@@ -171,7 +172,7 @@ class RelayExprNode : public BaseExprNode {
    *
    * Only valid when the expression's type is a Tensor.
    */
-  inline RelayExpr shape() const;
+  RelayExpr shape() const;
 
   /*!
    * \brief Check if the inferred(checked) type of the Expr

include/tvm/relax/expr.h

@@ -31,12 +31,12 @@
 namespace tvm {
 namespace relax {
 
+using Expr = RelayExpr;
+using ExprNode = RelayExprNode;
 using relay::Id;
 using relay::Call;
 using relay::Tuple;
 using relay::TupleGetItem;
-using ExprNode = RelayExprNode;
-using Expr = RelayExpr;
 
 /*! \brief A shape expression which allows users to construct a shape containing PrimExpr.
  */
@@ -121,13 +121,13 @@ class VarNode : public ExprNode {
 class Var : public Expr {
  public:
   TVM_DLL Var(String name_hint,
-              runtime::Optional<Array<PrimExpr>> shape_annotation,
+              runtime::Optional<Expr> shape_annotation,
               runtime::Optional<Type> type_annotation,
               Span span = Span())
     : Var(Id(name_hint), shape_annotation, type_annotation, span) {}
 
   TVM_DLL Var(Id vid,
-              runtime::Optional<Array<PrimExpr>> shape_annotation,
+              runtime::Optional<Expr> shape_annotation,
               runtime::Optional<Type> type_annotation,
               Span span = Span());
   TVM_DEFINE_OBJECT_REF_METHODS(Var, Expr, VarNode);

python/tvm/ir/expr.py

@@ -50,6 +50,18 @@ def checked_type(self):
             raise ValueError("The type checker has not populated" " the checked_type for this node")
         return ret
 
+    @property
+    def shape(self):
+        """Get the shape of tvm.relay.Expr.
+
+        Returns
+        -------
+        shape : tvm.ir.RelayExpr
+            The expression that represents the shape.
+        """
+        return _ffi_api.RelayExprShape(self) 
+
+
 
 @tvm._ffi.register_object("GlobalVar")
 class GlobalVar(RelayExpr):

python/tvm/relax/expr.py

@@ -37,15 +37,31 @@ class ShapeExpr(Expr):
     def __init__(self, values: List[PrimExpr]) -> None:
         self.__init_handle_by_constructor__(_ffi_api.ShapeExpr, values)
 
+    def __getitem__(self, index):
+        if index >= len(self):
+            raise IndexError("Tuple index out of range")
+        return self.values[index]
+
+    def __len__(self):
+        return len(self.values)
+
+def make_shape(shape: List[PrimExpr]) -> ShapeExpr:
+    if isinstance(shape, (list, tuple)):
+        return ShapeExpr(shape)
+    else:
+        raise ValueError
+
 
 @tvm._ffi.register_object("relax.expr.Var")
 class Var(Expr):
     id: Id
     type_annotation: Optional[Type]
 
     def __init__(self, name_hint: str,
-                 shape_annotation: Optional[List[Type]] = None,
+                 shape_annotation: Optional[Expr] = None,
                  type_annotation: Optional[Type] = None) -> None:
+        if shape_annotation is not None:
+            shape_annotation = make_shape(shape_annotation)
         self.__init_handle_by_constructor__(_ffi_api.Var, name_hint,
                                             shape_annotation,
                                             type_annotation)

src/relax/expr.cc

@@ -19,11 +19,25 @@
 #include <tvm/relax/expr.h>
 
 namespace tvm {
+
+RelayExpr RelayExprNode::shape() const {
+  if (this->shape_.defined()) {
+    return Downcast<RelayExpr>(this->shape_);
+  }
+  static const Op& op = Op::Get("relax.shape_of");
+  RelayExpr self = GetRef<RelayExpr>(this);
+  return relay::Call(op, {self}, {}, {});
+}
+
+TVM_REGISTER_GLOBAL("ir.RelayExprShape")
+.set_body_typed([](RelayExpr expr) {
+  return expr->shape();
+});
+
 namespace relax {
 
 using tvm::runtime::Optional;
 
-
 TVM_REGISTER_NODE_TYPE(ShapeExprNode);
 
 ShapeExpr::ShapeExpr(Array<PrimExpr> values) {
@@ -41,7 +55,7 @@ TVM_REGISTER_GLOBAL("relax.ShapeExpr")
 TVM_REGISTER_NODE_TYPE(VarNode);
 
 Var::Var(Id vid,
-         Optional<Array<PrimExpr>> shape_annotation,
+         Optional<Expr> shape_annotation,
          Optional<Type> type_annotation,
          Span span) {
   ObjectPtr<VarNode> n = make_object<VarNode>();
@@ -54,7 +68,7 @@ Var::Var(Id vid,
 
 TVM_REGISTER_GLOBAL("relax.Var")
 .set_body_typed([](String name_hint,
-                   Optional<Array<PrimExpr>> shape_annotation,
+                   Optional<Expr> shape_annotation,
                    Optional<Type> type_annotation) {
   return Var(name_hint, shape_annotation, type_annotation);
 });
@@ -64,7 +78,7 @@ TVM_REGISTER_NODE_TYPE(DataflowVarNode);
 
 TVM_REGISTER_GLOBAL("relax.DataflowVar")
 .set_body_typed([](String name_hint,
-                   Optional<Array<PrimExpr>> shape_annotation,
+                   Optional<Expr> shape_annotation,
                    Optional<Type> type_annotation) {
   return DataflowVar(name_hint, shape_annotation, type_annotation);
 });

src/relax/op.cc

@@ -22,19 +22,34 @@
 namespace tvm {
 namespace relax {
 
+// call_dps
+
+RELAY_REGISTER_OP("relax.call_dps")
+.set_num_inputs(3)
+.add_argument("shape", "ShapeExpr", "The output shape.")
+.add_argument("func", "Expr", "The destination-passing-style function.")
+.add_argument("args", "Tuple", "The input arguments.");
+
 Expr MakeCallDPS(ShapeExpr shape, Expr func, Tuple args) {
-  static const Op& op = Op::Get("call_dps");
+  static const Op& op = Op::Get("relax.call_dps");
   return Call(op, {shape, func, args}, {}, {});
 }
 
 TVM_REGISTER_GLOBAL("relax.op.call_dps")
 .set_body_typed(MakeCallDPS);
 
-RELAY_REGISTER_OP("call_dps")
-.set_num_inputs(3)
-.add_argument("shape", "ShapeExpr", "The output shape.")
-.add_argument("func", "Expr", "The destination-passing-style function.")
-.add_argument("args", "Tuple", "The input arguments.");
+// shape_of
+
+RELAY_REGISTER_OP("relax.shape_of")
+.set_num_inputs(1)
+.add_argument("input", "Expr", "The input expression");
+
+Expr MakeShapeOf(Expr expr) {
+  static const Op& op = Op::Get("relax.shape_of");
+  return Call(op, {expr}, {}, {});
+}
 
+TVM_REGISTER_GLOBAL("relax.op.shape_of")
+.set_body_typed(MakeShapeOf);
 } // namespace relax
 } // namespace tvm

tests/python/relax/test_ast.py renamed to tests/python/relax/test_expr.py

@@ -113,6 +113,19 @@ def test_func():
     assert func.name.name_hint == "func"
 
 
+def test_shape_of():
+    v0 = rx.Var("v0")
+    s0 = v0.shape
+    assert isinstance(s0, tvm.relay.Call)
+    assert s0.op.name == "relax.shape_of"
+
+    shape_anno = [96, 54]
+    v1 = rx.Var("v1", shape_anno)
+    s1 = v1.shape
+    for x, y in zip(shape_anno, s1):
+        assert x == y
+
+
 if __name__ == "__main__":
     test_var()
     test_dataflow_var()
@@ -123,3 +136,4 @@ def test_func():
     test_seq_expr()
     test_shape_expr()
     test_func()
+    test_shape_of()