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
@@ -1861,8 +1861,12 @@
   AffineMap candidateLayout;
   if (candidateReduced.getAffineMaps().empty())
     candidateLayout = getStridedLinearLayoutMap(candidateReduced);
-  else
-    candidateLayout = candidateReduced.getAffineMaps().front();
+  else {
+    auto affineMaps = candidateReduced.getAffineMaps();
+    candidateLayout = affineMaps.back();
+    for (size_t i = affineMaps.size() - 1; i > 0; --i)
+      candidateLayout = candidateLayout.compose(affineMaps[i - 1]);
+  }
   assert(inferredType.getNumResults() == 1 &&
          candidateLayout.getNumResults() == 1);
   if (inferredType.getNumSymbols() != candidateLayout.getNumSymbols() ||
diff --git a/mlir/test/Dialect/MemRef/ops.mlir b/mlir/test/Dialect/MemRef/ops.mlir
--- a/mlir/test/Dialect/MemRef/ops.mlir
+++ b/mlir/test/Dialect/MemRef/ops.mlir
@@ -211,3 +211,15 @@
 //      CHECK: func @collapse_shape_to_dynamic
 //      CHECK:   memref.collapse_shape
 // CHECK-SAME:    [0], [1], [2, 3, 4]
+
+#map1 = affine_map<(d0, d1, d2, d3) -> (d0, d2, d3, d1)>
+#map2 = affine_map<(d0, d1, d2, d3) -> (64 * d0 + 16 * d1 + 2 * d2 + d3)>
+#map3 = affine_map<(d0, d1, d2, d3) -> (64 * d0 + 16 * d1 + 2 * d2 + d3 + 39)>
+
+// CHECK-LABEL: func @subview_with_affine_maps
+func @subview_with_affine_maps(%arg0: memref<1x2x4x8xf32, #map1, #map2>)
+    -> memref<1x1x2x3xf32, #map1, #map3> {
+  %0 = memref.subview %arg0 [0, 1, 2, 3][1, 1, 2, 3][1, 1, 1, 1] :
+    memref<1x2x4x8xf32, #map1, #map2> to memref<1x1x2x3xf32, #map1, #map3>
+  return %0 : memref<1x1x2x3xf32, #map1, #map3>
+}