diff --git a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
--- a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
+++ b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
@@ -767,6 +767,11 @@
     if (!reshape)
       return failure();
 
+    // We need to make sure that the index dominates the new load. Since we
+    // can't check that here, restrict the transformation to constant indices.
+    if (!matchPattern(dim.index(), matchConstantIndex()))
+      return failure();
+
     // Place the load directly after the reshape to ensure that the shape memref
     // was not mutated.
     rewriter.setInsertionPointAfter(reshape);
diff --git a/mlir/test/Dialect/Standard/canonicalize.mlir b/mlir/test/Dialect/Standard/canonicalize.mlir
--- a/mlir/test/Dialect/Standard/canonicalize.mlir
+++ b/mlir/test/Dialect/Standard/canonicalize.mlir
@@ -142,6 +142,21 @@
 
 // -----
 
+// Test case: Folding of memref.dim(memref.reshape %v %shp, %idx) -> memref.load %shp[%idx]
+// CHECK-LABEL: func @dim_of_memref_reshape_not_dominating(
+//   CHECK:   memref.dim
+func @dim_of_memref_reshape_not_dominating(%arg0: memref<*xf32>, %arg1: memref<?xi32>, %arg2: memref<?xindex>)
+    -> index {
+  %c3 = constant 3 : index
+  %0 = memref.reshape %arg0(%arg1)
+      : (memref<*xf32>, memref<?xi32>) -> memref<*xf32>
+  %idx = memref.load %arg2[%c3] : memref<?xindex>
+  %1 = memref.dim %0, %idx : memref<*xf32>
+  return %1 : index
+}
+
+// -----
+
 // Test case: Folding memref.dim(tensor.cast %0, %idx) -> memref.dim %0, %idx
 // CHECK-LABEL: func @fold_dim_of_tensor.cast
 //  CHECK-SAME:   %[[ARG0:.[a-z0-9A-Z_]+]]: tensor<4x?xf32>