diff --git a/include/Transforms/Passes.h b/include/Transforms/Passes.h
index 41da49a..75120f1 100644
--- a/include/Transforms/Passes.h
+++ b/include/Transforms/Passes.h
@@ -17,6 +17,7 @@ namespace scalehls {
 
 std::unique_ptr<mlir::Pass> createPragmaDSEPass();
 std::unique_ptr<mlir::Pass> createAffineLoopPerfectionPass();
+std::unique_ptr<mlir::Pass> createPartialAffineLoopTilePass();
 
 void registerTransformsPasses();
 
diff --git a/include/Transforms/Passes.td b/include/Transforms/Passes.td
index 5491d64..e2621aa 100644
--- a/include/Transforms/Passes.td
+++ b/include/Transforms/Passes.td
@@ -23,13 +23,32 @@ def PragmaDSE : Pass<"pragma-dse", "ModuleOp"> {
 def AffineLoopPerfection : Pass<"affine-loop-perfection", "ModuleOp"> {
   let summary = "Try to perfect a nested loop";
   let description = [{
-    This affine-loop-perfection pass will try to perfect affine loops.
+    This affine-loop-perfection pass will try to perfect all affine loops.
     Specifically, this pass will move operations under non-innermost loops into
-    innermost loop and create an affine if region to ensure the correctness of
-    the transformation.
+    innermost loop and create affine if regions to ensure the correctness of the
+    movement.
   }];
 
   let constructor = "mlir::scalehls::createAffineLoopPerfectionPass()";
 }
 
+def PartialAffineLoopTile : Pass<"partial-affine-loop-tile", "ModuleOp"> {
+  let summary = "Try to perfect a nested loop";
+  let description = [{
+    This partial-affine-loop-tile pass will try to tile the nested loops. The
+    difference with the official affine-loop-tile pass is this pass will only
+    tile the first "tileLevel" outermost loop levels rather than all loops
+    levels.
+  }];
+
+  let constructor = "mlir::scalehls::createPartialAffineLoopTilePass()";
+
+    let options = [
+    Option<"tileLevel", "tile-level", "unsigned", /*default=*/"1", 
+           "Positive number: the level of loops to be tiles">,
+    Option<"tileSize", "tile-size", "unsigned", /*default=*/"2", 
+           "Positive number: the size of tiling">
+  ];
+}
+
 #endif // SCALEHLS_TRANSFORMS_PASSES_TD
diff --git a/lib/Transforms/PartialAffineLoopTile.cpp b/lib/Transforms/PartialAffineLoopTile.cpp
new file mode 100644
index 0000000..30b29fc
--- /dev/null
+++ b/lib/Transforms/PartialAffineLoopTile.cpp
@@ -0,0 +1,71 @@
+//===------------------------------------------------------------*- C++ -*-===//
+//
+//===----------------------------------------------------------------------===//
+
+#include "Transforms/Passes.h"
+#include "mlir/Analysis/LoopAnalysis.h"
+#include "mlir/Dialect/Affine/IR/AffineOps.h"
+#include "mlir/Dialect/Affine/Passes.h"
+#include "mlir/IR/Builders.h"
+#include "mlir/Transforms/LoopUtils.h"
+
+using namespace std;
+using namespace mlir;
+using namespace scalehls;
+
+namespace {
+struct PartialAffineLoopTile
+    : public PartialAffineLoopTileBase<PartialAffineLoopTile> {
+  void runOnOperation() override;
+  void getTileSizes(ArrayRef<mlir::AffineForOp> band,
+                    SmallVectorImpl<unsigned> *tileSizes);
+};
+} // namespace
+
+void PartialAffineLoopTile::runOnOperation() {
+  // Walk through all functions and loops.
+  for (auto func : getOperation().getOps<FuncOp>()) {
+    // Bands of loops to tile.
+    std::vector<SmallVector<mlir::AffineForOp, 6>> bands;
+    getTileableBands(func, &bands);
+
+    // Tile each band.
+    for (auto &band : bands) {
+      // Truncate band and only keep first tileLevel loops.
+      size_t realTileLevel = band.size();
+      if (realTileLevel > tileLevel) {
+        band.resize(tileLevel);
+        realTileLevel = tileLevel;
+      }
+
+      // Set up tile sizes; fill missing tile sizes at the end with default tile
+      // size or tileSize if one was provided.
+      SmallVector<unsigned, 6> tileSizes;
+      tileSizes.assign(band.size(), tileSize);
+
+      SmallVector<mlir::AffineForOp, 6> tiledNest;
+      if (failed(tilePerfectlyNested(band, tileSizes, &tiledNest)))
+        return signalPassFailure();
+
+      // Permute loop order to move the tiled loop to the innermost of the
+      // perfect nested loop.
+      SmallVector<mlir::AffineForOp, 4> nestedLoops;
+      getPerfectlyNestedLoops(nestedLoops, tiledNest.front());
+
+      SmallVector<unsigned, 4> permMap;
+      for (size_t i = 0, e = nestedLoops.size(); i < e; ++i) {
+        if (i < realTileLevel)
+          permMap.push_back(i);
+        else if (i < 2 * realTileLevel)
+          permMap.push_back(e + i - 2 * realTileLevel);
+        else
+          permMap.push_back(i - realTileLevel);
+      }
+      permuteLoops(nestedLoops, permMap);
+    }
+  }
+}
+
+std::unique_ptr<mlir::Pass> scalehls::createPartialAffineLoopTilePass() {
+  return std::make_unique<PartialAffineLoopTile>();
+}
diff --git a/test/Transforms/PartialAffineLoopTile/test_for.mlir b/test/Transforms/PartialAffineLoopTile/test_for.mlir
new file mode 100644
index 0000000..5ab9474
--- /dev/null
+++ b/test/Transforms/PartialAffineLoopTile/test_for.mlir
@@ -0,0 +1,6 @@
+// RUN: scalehls-opt -partial-affine-loop-tile %s | FileCheck %s
+
+// CHECK-LABEL: func @test_for
+func @test_for() {
+  return
+}