// FMUL <Sd>, <Sn>, <Sm>
	{0xffe0fc00, 0x1e200800, FMUL, instArgs{arg_Sd, arg_Sn, arg_Sm}, nil},
	// FMUL <Dd>, <Dn>, <Dm>
	{0xffe0fc00, 0x1e600800, FMUL, instArgs{arg_Dd, arg_Dn, arg_Dm}, nil},
	// FMUL <Vd>.<t>, <Vn>.<t>, <Vm>.<t>

  // CHECK: %[[CONST:.*]] = "tf.Const"()
  // CHECK-DAG: %[[MUL1:.*]] = "tf.Mul"(%arg0, %arg2)
  // CHECK-DAG: %[[MUL2:.*]] = "tf.Mul"(%arg1, %arg3)
  // CHECK: "tf.ConcatV2"(%[[MUL1]], %[[MUL2]], %[[CONST]])
  %0 = "tf.Const"() { value = dense<1> : tensor<i32> } : () -> tensor<i32>
  %1 = "tf.Mul"(%arg0, %arg2) : (tensor<?x2xf32>, tensor<f32>) -> tensor<?x2xf32>
  %2 = "tf.Mul"(%arg1, %arg3) : (tensor<?x2xf32>, tensor<f32>) -> tensor<?x2xf32>

  (Add32 (Const32 <t> [c*d]) (Mul32 <t> (Const32 <t> [c]) x))

// Rewrite x*y ± x*z  to  x*(y±z)
(Add(64|32|16|8) <t> (Mul(64|32|16|8) x y) (Mul(64|32|16|8) x z))
	=> (Mul(64|32|16|8) x (Add(64|32|16|8) <t> y z))
(Sub(64|32|16|8) <t> (Mul(64|32|16|8) x y) (Mul(64|32|16|8) x z))
	=> (Mul(64|32|16|8) x (Sub(64|32|16|8) <t> y z))

// rewrite shifts of 8/16/32 bit consts into 64 bit consts to reduce

  // CHECK: %[[MUL:.*]] = tfl.mul %arg0, %[[BROADCAST_TO]] {fused_activation_function = "NONE"} : tensor<8x7x6x5x4x3x2x1xf32>
  // CHECK: return %[[MUL]] : tensor<8x7x6x5x4x3x2x1xf32>
}

func.func @mul_with_high_dims_dynamic_shape_both_sides(%arg0: tensor<8x7x6x5x?x3x2x1xf32>, %arg1: tensor<?x3x2x1xf32>) -> tensor<8x7x6x5x?x3x2x1xf32> {

  // CHECK: tfl.mul %arg0, %arg1 {fused_activation_function = "RELU6"}
  %0 = tfl.mul %arg0, %arg1 {fused_activation_function = "RELU6"} : tensor<? x i16>
  func.return %0#0 : tensor<? x i16>
}

// CHECK-LABEL: testMulComplex
func.func @testMulComplex(tensor<? x complex<f32>>, tensor<? x complex<f32>>) -> tensor<? x complex<f32>> {
^bb0(%arg0: tensor<? x complex<f32>>, %arg1: tensor<? x complex<f32>>):

{"Name":"FMUL (scalar)","Bits":"0|0|0|1|1|1|1|0|00:2|1|Rm:5|0|0|0|0|1|0|Rn:5|Rd:5","Arch":"Single-precision variant","Syntax":"FMUL <Sd>, <Sn>, <Sm>","Code":"","Alias":""},
{"Name":"FMUL (scalar)","Bits":"0|0|0|1|1|1|1|0|01:2|1|Rm:5|0|0|0|0|1|0|Rn:5|Rd:5","Arch":"Double-precision variant","Syntax":"FMUL <Dd>, <Dn>, <Dm>","Code":"","Alias":""},

  std::vector<string> expected_nodes = {"cluster1", "cluster2", "mul", "x"};
  EXPECT_EQ(expected_nodes, GraphNodes(*graph));

  std::vector<std::pair<string, string>> expected_edges = {
      {"cluster1:0", "cluster2:0"},
      {"cluster1:0", "mul:0"},
      {"cluster2:0", "mul:1"},
      {"x:0", "cluster1:0"}};
  EXPECT_EQ(expected_edges, GraphEdges(*graph));
}

  %15 = "tf.AddV2"(%14, %3) {device = ""} : (tensor<2x16xi32>, tensor<i32>) -> tensor<2x16xi32>
  %16 = "tf.Mul"(%15, %8) {device = ""} : (tensor<2x16xi32>, tensor<i32>) -> tensor<2x16xi32>
  %17 = "tf.AddV2"(%16, %6) {device = ""} : (tensor<2x16xi32>, tensor<2x1xi32>) -> tensor<2x16xi32>
  %18 = "tf.Mul"(%14, %8) {device = ""} : (tensor<2x16xi32>, tensor<i32>) -> tensor<2x16xi32>

int64_t GetElementIndex(const std::vector<int64_t>& shape,
                        const std::vector<int64_t>& current_index) {
  int64_t ind = 0;
  int64_t mul = 1;
  for (int i = shape.size() - 1; i >= 0; --i) {
    ind += (current_index[i] % shape[i]) * mul;
    mul *= shape[i];
  }
  return ind;
}

// Helper method that increment index represented in 'current_index_ptr'
// in the shape of 'result_shape'.

  func.return %0 : tensor<4x1xcomplex<f32>>
}

// -----

// CHECK-LABEL: func @testMul
func.func @testMul(%arg0: tensor<2xui16>) -> (tensor<2xui16>) {
  %0 = "tf.Mul"(%arg0, %arg0) {T = "tfdtype$DT_UINT16", device = "/device:CPU:0", name = "Mul"} : (tensor<2xui16>, tensor<2xui16>) -> tensor<2xui16>
  func.return %0 : tensor<2xui16>
}

// -----

// Test error message for incompatible element types.