diff --git a/mlir/lib/Dialect/Linalg/Transforms/ConvertToDestinationStyle.cpp b/mlir/lib/Dialect/Linalg/Transforms/ConvertToDestinationStyle.cpp
--- a/mlir/lib/Dialect/Linalg/Transforms/ConvertToDestinationStyle.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/ConvertToDestinationStyle.cpp
@@ -180,9 +180,12 @@
       createAllocationForTensor(rewriter, loc, padOp.getResult(), memorySpace);
   rewriter.setInsertionPointAfter(alloc.getDefiningOp());
 
-  // Create linalg.fill or linalg.generic.
-  Operation *fillOp = movePaddingToFillOrGenericOp(rewriter, loc, padOp, alloc);
-  rewriter.setInsertionPointAfter(fillOp);
+  if (!padOp.hasZeroLowPad() || !padOp.hasZeroHighPad()) {
+    // Create linalg.fill or linalg.generic. Not needed if there is no padding.
+    Operation *fillOp =
+        movePaddingToFillOrGenericOp(rewriter, loc, padOp, alloc);
+    rewriter.setInsertionPointAfter(fillOp);
+  }
 
   // Create memref.tensor_store.
   SmallVector<OpFoldResult> sizes =
diff --git a/mlir/lib/Dialect/Tensor/Utils/Utils.cpp b/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
--- a/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
+++ b/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
@@ -36,8 +36,7 @@
     bindDims(b.getContext(), d0);
     OpFoldResult sz = tensor::getMixedSize(b, loc, source, en.index());
     high[en.index()] =
-        affine::makeComposedAffineApply(b, loc, en.value() - d0, {sz})
-            .getResult();
+        affine::makeComposedFoldedAffineApply(b, loc, en.value() - d0, {sz});
   }
   return b.create<PadOp>(loc, type, source, low, high, pad, nofold);
 }
diff --git a/mlir/test/Dialect/Linalg/generalize-tensor-pack.mlir b/mlir/test/Dialect/Linalg/generalize-tensor-pack.mlir
--- a/mlir/test/Dialect/Linalg/generalize-tensor-pack.mlir
+++ b/mlir/test/Dialect/Linalg/generalize-tensor-pack.mlir
@@ -28,9 +28,7 @@
 // CHECK-SAME:    %[[DEST:[a-zA-Z0-9]+]]
 // CHECK-SAME:    %[[PAD_VAL:[a-zA-Z0-9]+]]
 // CHECK-DAG:     %[[C0:.+]] = arith.constant 0 : index
-// CHECK-DAG:     %[[C1:.+]] = arith.constant 1 : index
-// CHECK-DAG:     %[[C3:.+]] = arith.constant 3 : index
-// CHECK:         %[[PAD:.+]] = tensor.pad %[[SRC]] low[%[[C0]], %[[C0]]] high[%[[C3]], %[[C1]]]
+// CHECK:         %[[PAD:.+]] = tensor.pad %[[SRC]] low[%[[C0]], %[[C0]]] high[3, 1]
 // CHECK:           tensor.yield %[[PAD_VAL]]
 // CHECK:         %[[EMPTY:.+]] = tensor.empty() : tensor<8x2xf32>
 // CHECK:         %[[TRANSP:.+]] = linalg.transpose
diff --git a/mlir/test/Dialect/Linalg/pad-to-specific-memory-space.mlir b/mlir/test/Dialect/Linalg/pad-to-specific-memory-space.mlir
--- a/mlir/test/Dialect/Linalg/pad-to-specific-memory-space.mlir
+++ b/mlir/test/Dialect/Linalg/pad-to-specific-memory-space.mlir
@@ -32,7 +32,7 @@
   // CHECK: memref.copy %[[s1]], %[[alloc1_view]]
 
   // CHECK: %[[alloc2:.*]] = memref.alloc() : memref<4x5xf32, 3>
-  // CHECK: linalg.fill {{.*}} outs(%[[alloc2]]
+  // CHECK-NOT: linalg.fill {{.*}} outs(%[[alloc2]]
   // No subview because there is 0 padding
   // CHECK: memref.copy %[[s2]], %[[alloc2]]
 
diff --git a/mlir/test/Dialect/Linalg/transform-op-pad.mlir b/mlir/test/Dialect/Linalg/transform-op-pad.mlir
--- a/mlir/test/Dialect/Linalg/transform-op-pad.mlir
+++ b/mlir/test/Dialect/Linalg/transform-op-pad.mlir
@@ -188,7 +188,6 @@
 // linalg op is not produced by an empty op or an extract_slice op.
 
 // CHECK-DAG: #[[$MAP_MIN:.*]] = affine_map<(d0) -> (-d0 + 2044, 16)>
-// CHECK-DAG: #[[$MAP_C0:.*]] = affine_map<() -> (0)>
 // CHECK-DAG: #[[$MAP_TO_16:.*]] = affine_map<(d0) -> (-d0 + 16)>
 // CHECK-LABEL: @outs_not_produced_by_empty_or_extract_slice(
 // CHECK-SAME: %[[A:[^: ]*]]: tensor<128x2044xf32>,
@@ -212,16 +211,13 @@
     // CHECK-DAG: %[[CST:.*]] = arith.constant 0.
     // CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
 
-    // CHECK-DAG: %[[ZERO:.*]] = affine.apply #[[$MAP_C0]]()
     // CHECK-DAG: %[[TO_16:.*]] = affine.apply #[[$MAP_TO_16]](%[[MIN]])
-    // CHECK: %[[PADDED_A_SLICE:.*]] = tensor.pad %[[A_SLICE]] nofold low[%[[C0]], %[[C0]]] high[%[[ZERO]], %[[TO_16]]]
+    // CHECK: %[[PADDED_A_SLICE:.*]] = tensor.pad %[[A_SLICE]] nofold low[%[[C0]], %[[C0]]] high[0, %[[TO_16]]]
     // CHECK: tensor.yield %[[CST]]
     // CHECK: %[[PADDED_B_SLICE:.*]] = tensor.pad %[[B_SLICE]] nofold
     // The output shape is already padded, so actually we shouldn't
     // add anything to the upper bound.
-    // CHECK: %[[ZERO0:.*]] = affine.apply #[[$MAP_C0]]()
-    // CHECK: %[[ZERO1:.*]] = affine.apply #[[$MAP_C0]]()
-    // CHECK: %[[PADDED_ARG4:.*]] = tensor.pad %[[ARG4]] nofold low[{{.*}}] high[%[[ZERO0]], %[[ZERO1]]]
+    // CHECK: %[[PADDED_ARG4:.*]] = tensor.pad %[[ARG4]] nofold low[{{.*}}] high[0, 0]
 
     //      CHECK: %[[T5:.*]] = linalg.matmul
     // CHECK-SAME:              ins(%[[PADDED_A_SLICE]], %[[PADDED_B_SLICE]] : tensor<128x16xf32>, tensor<16x128xf32>)