diff --git a/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp b/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp
index dfd4e154c9275..c2d4e20efd364 100644
--- a/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp
+++ b/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp
@@ -622,6 +622,18 @@ OpFoldResult CastOp::fold(ArrayRef<Attribute> operands) {
   if (getInput().getType() == getType())
     return getInput();
 
+  // cast-to-iN(cast-to-iM(x)) -> cast-to-iN(x) when N <= M
+  if (auto cast = getInput().getDefiningOp<CastOp>()) {
+    auto intermediateElTy = cast.getType().getElementType().dyn_cast<IntegerType>();
+    auto finalElTy = getType().getElementType().dyn_cast<IntegerType>();
+    if (intermediateElTy && finalElTy &&
+        intermediateElTy.getSignedness() == finalElTy.getSignedness() &&
+        intermediateElTy.getWidth() >= finalElTy.getWidth()) {
+      getInputMutable().assign(cast.getInput());
+      return getResult();
+    }
+  }
+
   auto operand = operands[0].dyn_cast_or_null<ElementsAttr>();
   if (!operand)
     return {};
diff --git a/mlir/test/Dialect/Tosa/canonicalize.mlir b/mlir/test/Dialect/Tosa/canonicalize.mlir
index a52764e2cc038..c1f87d83892dd 100644
--- a/mlir/test/Dialect/Tosa/canonicalize.mlir
+++ b/mlir/test/Dialect/Tosa/canonicalize.mlir
@@ -39,6 +39,32 @@ func.func @cast_nofold(%arg0: tensor<?x1xf32>) -> tensor<?x1xi32> {
   return %0 : tensor<?x1xi32>
 }
 
+// CHECK-LABEL: @cast_fold_double
+func.func @cast_fold_double(%arg0: tensor<?x1xf32>) -> tensor<?x1xi8> {
+  // CHECK: "tosa.cast"{{.*}} (tensor<?x1xf32>) -> tensor<?x1xi8>
+  %0 = "tosa.cast"(%arg0) : (tensor<?x1xf32>) -> tensor<?x1xi16>
+  %1 = "tosa.cast"(%0) : (tensor<?x1xi16>) -> tensor<?x1xi8>
+  return %1 : tensor<?x1xi8>
+}
+
+// CHECK-LABEL: @cast_no_fold_double1
+func.func @cast_no_fold_double1(%arg0: tensor<?x1xf32>) -> tensor<?x1xi8> {
+  // CHECK: "tosa.cast"{{.*}} (tensor<?x1xf32>) -> tensor<?x1xui16>
+    // CHECK: "tosa.cast"{{.*}} (tensor<?x1xui16>) -> tensor<?x1xi8>
+  %0 = "tosa.cast"(%arg0) : (tensor<?x1xf32>) -> tensor<?x1xui16>
+  %1 = "tosa.cast"(%0) : (tensor<?x1xui16>) -> tensor<?x1xi8>
+  return %1 : tensor<?x1xi8>
+}
+
+// CHECK-LABEL: @cast_no_fold_double2
+func.func @cast_no_fold_double2(%arg0: tensor<?x1xf32>) -> tensor<?x1xi16> {
+  // CHECK: "tosa.cast"{{.*}} (tensor<?x1xf32>) -> tensor<?x1xi8>
+  // CHECK: "tosa.cast"{{.*}} (tensor<?x1xi8>) -> tensor<?x1xi16>
+  %0 = "tosa.cast"(%arg0) : (tensor<?x1xf32>) -> tensor<?x1xi8>
+  %1 = "tosa.cast"(%0) : (tensor<?x1xi8>) -> tensor<?x1xi16>
+  return %1 : tensor<?x1xi16>
+}
+
 // CHECK-LABEL: @clamp_not_noop
 func.func @clamp_not_noop(%arg0: tensor<4xi32>) -> tensor<4xi32> {
   // CHECK: "tosa.clamp"