// CHECK-LABEL: testDotToDotGeneralMatrixMatrix
func.func @testDotToDotGeneralMatrixMatrix(%arg0: tensor<2x3072xf32>, %arg1: tensor<3072x512xf32>) -> tensor<2x512xf32> {
  %0 = "mhlo.dot"(%arg0, %arg1) : (tensor<2x3072xf32>, tensor<3072x512xf32>) -> tensor<2x512xf32>
  func.return %0 : tensor<2x512xf32>

// CHECK:      %[[RES:.*]] = "mhlo.dot_general"(%arg0, %arg1) <{
// CHECK-SAME:   dot_dimension_numbers = #mhlo.dot<

  %1 = "tfl.add"(%cst_2, %cst_1) {fused_activation_function = "NONE"} : (tensor<2x2x2xf32>, tensor<  2x2xf32>) -> tensor<2x2x2xf32>
  %2 = "tfl.add"(%cst_0, %cst_2) {fused_activation_function = "NONE"} : (tensor<    2xf32>, tensor<2x2x2xf32>) -> tensor<2x2x2xf32>

  func.return %0, %1, %2 : tensor<2x2xf32>, tensor<2x2x2xf32>, tensor<2x2x2xf32>

  module {
    func.func @main(%arg0: tensor<2x2x2xf32>, %arg1: tensor<2x2x2xf32>) -> tensor<2x2x2xf32> attributes {_from_xla_call_module} {
      %0 = stablehlo.dot_general %arg0, %arg1,
        batching_dims = [0] x [0],
        contracting_dims = [2] x [1],
        precision = [DEFAULT, DEFAULT]
      : (tensor<2x2x2xf32>, tensor<2x2x2xf32>) -> tensor<2x2x2xf32>
        return %0 : tensor<2x2x2xf32>
    }
  }
)mlir";

}

// CHECK-LABEL: prepareAdd
func.func @prepareAdd(%arg0: tensor<2x2xf32>) -> tensor<2x2xf32> {
  %cst = arith.constant dense<[[0.0, 1.0], [2.0, 255.0]]> : tensor<2x2xf32>
  %add = "tfl.add"(%arg0, %cst) {fused_activation_function="NONE"} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
  func.return %add : tensor<2x2xf32>

}

func.func @einsum_reduceddim(%arg0: tensor<2x5x7xf32>, %arg1: tensor<2x5x7x3xf32>) -> tensor<2x5x3xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "bin,binj->bij"}: (tensor<2x5x7xf32>, tensor<2x5x7x3xf32>) -> tensor<2x5x3xf32>
  func.return %0 : tensor<2x5x3xf32>
  // CHECK-LABEL: einsum_reduceddim
  // CHECK-DAG: %[[cst:.*]] = arith.constant dense<[2, 5, 1, 7]> : tensor<4xi64>

    %7 = stablehlo.broadcast_in_dim %6, dims = [2, 1] : (tensor<1x3xf32>) -> tensor<2x3x2xf32>
    %8 = "quantfork.qcast"(%7) {volatile} : (tensor<2x3x2xf32>) -> tensor<2x3x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>
    %9 = "quantfork.dcast"(%8) : (tensor<2x3x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>) -> tensor<2x3x2xf32>
    return %9 : tensor<2x3x2xf32>
  }

  // CHECK: quantized_dot_general_fn_1

func.func @NotQuantizeFloatConst() -> tensor<2x2xf32> {
  %0 = arith.constant dense<-0.1> : tensor<2x2xf32>
  %1 = "tfl.quantize"(%0) {qtype = tensor<2x2x!quant.uniform<u8:f32, 7.8431372549019615E-4:128>>} : (tensor<2x2xf32>) -> tensor<2x2x!quant.uniform<u8:f32, 7.8431372549019615E-4:128>>
  %2 = "tfl.dequantize"(%1) : (tensor<2x2x!quant.uniform<u8:f32, 7.8431372549019615E-4:128>>) -> tensor<2x2xf32>
  func.return %2 : tensor<2x2xf32>

  %flow_1_out = "tf.TensorArrayWriteV3"(%handle_1, %index, %y, %flow_1) {device = "/job:localhost/replica:0/task:0/device:CPU:0"} : (tensor<2x!tf_type.resource<tensor<?x512xf32>>>, tensor<i32>, tensor<?x512xf32>, tensor<f32>) -> tensor<f32>

  func.func @matmul(%arg0: tensor<2x12xf32>) -> (tensor<*xf32>) {
    %cst_0 = "tf.Const"() {value = dense<0.000000e+00> : tensor<12x2xf32>} : () -> tensor<12x2xf32>
    %1 = "tf.PartitionedCall"(%arg0, %cst_0) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_matmul_fn_1} : (tensor<2x12xf32>, tensor<12x2xf32>) -> tensor<*xf32>
    func.return %1: tensor<*xf32>
  }

  func.func @matmul(%arg0: tensor<2x12xf32>) -> (tensor<*xf32>) {
    %cst_0 = "tf.Const"() {value = dense<0.000000e+00> : tensor<12x2xf32>} : () -> tensor<12x2xf32>
    %1 = "tf.PartitionedCall"(%arg0, %cst_0) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_matmul_fn_1} : (tensor<2x12xf32>, tensor<12x2xf32>) -> tensor<*xf32>
    func.return %1: tensor<*xf32>
  }