fix alloca (don't get_op_results twice)

mlir/extras/dialects/ext/memref.py

@@ -45,9 +45,7 @@ def alloca(
 ):
     if loc is None:
         loc = get_user_code_loc()
-    return get_op_result_or_op_results(
-        _alloc(AllocaOp, sizes, element_type, loc=loc, ip=ip)
-    )
+    return _alloc(AllocaOp, sizes, element_type, loc=loc, ip=ip)
 
 
 def load(mem: Value, indices: Sequence[Union[Value, int]], *, loc=None, ip=None):

tests/test_memref.py

@@ -7,7 +7,13 @@
 import mlir.extras.types as T
 from mlir.extras.ast.canonicalize import canonicalize
 from mlir.extras.dialects.ext.arith import Scalar, constant
-from mlir.extras.dialects.ext.memref import alloc, S
+from mlir.extras.dialects.ext.memref import (
+    alloc,
+    alloca,
+    S,
+    alloca_scope,
+    alloca_scope_return,
+)
 from mlir.extras.dialects.ext.scf import (
     range_,
     yield_,
@@ -43,6 +49,32 @@ def test_simple_literal_indexing(ctx: MLIRContext):
     filecheck(correct, ctx.module)
 
 
+def test_simple_literal_indexing_alloca(ctx: MLIRContext):
+    @alloca_scope([])
+    def demo_scope2():
+        mem = alloca((10, 22, 333, 4444), T.i32())
+
+        w = mem[2, 4, 6, 8]
+        assert isinstance(w, Scalar)
+        alloca_scope_return([])
+
+    correct = dedent(
+        """\
+    module {
+      memref.alloca_scope  {
+        %alloca = memref.alloca() : memref<10x22x333x4444xi32>
+        %c2 = arith.constant 2 : index
+        %c4 = arith.constant 4 : index
+        %c6 = arith.constant 6 : index
+        %c8 = arith.constant 8 : index
+        %0 = memref.load %alloca[%c2, %c4, %c6, %c8] : memref<10x22x333x4444xi32>
+      }
+    }
+    """
+    )
+    filecheck(correct, ctx.module)
+
+
 def test_ellipsis_and_full_slice(ctx: MLIRContext):
     mem = alloc((10, 22, 333, 4444), T.i32())