func.return %0 : tensor<3x4x4xf32>
}

func.func @batchmatmulv2_lhs_batch(%arg0: tensor<3x4x2xf32>, %arg1: tensor<2x4xf32>) -> tensor<3x4x4xf32> {
// CHECK-LABEL:   func @batchmatmulv2_lhs_batch
// CHECK:           "mhlo.dynamic_broadcast_in_dim"({{.*}}, {{.*}}) <{broadcast_dimensions = dense<[1, 2]> : tensor<2xi64>}>
// CHECK:           "mhlo.dot_general"({{.*}}, {{.*}}) <{

    // CHECK: mhlo.reduce
    // CHECK: mhlo.dot_general
    // CHECK: mhlo.transpose
    %0 = "tf.BatchMatMulV2"(%arg0, %arg1) {T = f32, adj_x = false, adj_y = false, grad_x = false, grad_y = false, device = ""} : (tensor<1x4x2xf32>, tensor<3x2x4xf32>) -> tensor<3x4x4xf32>
    func.return %0 : tensor<3x4x4xf32>
  }

  // CHECK-LABEL: approx_topk

// CHECK-NEXT:    return %2 : tensor<3x2xf32>
}

func.func @transpose_3d(%arg0: tensor<1x2x3xf32>) -> tensor<3x2x1xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[2, 1, 0]> : tensor<3xi64>}> : (tensor<1x2x3xf32>) -> tensor<3x2x1xf32>
  func.return %0 : tensor<3x2x1xf32>

// CHECK-LABEL:   transpose_3d

  %lookup0 = "tfl.embedding_lookup"(%index0, %value0) : (tensor<5xi32>, tensor<3x2xf32>) -> tensor<5x2xf32>
  %lookup1 = "tfl.embedding_lookup"(%index1, %value1) : (tensor<3xi32>, tensor<2x2x2xf32>) -> tensor<3x2x2xf32>
  func.return %lookup0, %lookup1 : tensor<5x2xf32>, tensor<3x2x2xf32>

// CHECK:           %12 = "tf.Reshape"(%0, %11) : (tensor<2x?x3x4xf32>, tensor<4xi32>) -> tensor<2x?x3x4xf32>
// CHECK:           %13 = "tf.BatchMatMulV3"(%6, %12) <{adj_x = false, adj_y = false, grad_x = false, grad_y = false}> : (tensor<2x?x2x3xf32>, tensor<2x?x3x4xf32>) -> tensor<2x?x2x4xf32>
// CHECK:           %14 = "tf.Shape"(%arg0) : (tensor<2x?x2x3xf32>) -> tensor<4xi32>

    %2 = "quantfork.stats"(%1) {layerStats = dense<[5.00000000e-6, 7.00000000e-1]> : tensor<2xf32>} : (tensor<2x2x3xf32>) -> tensor<2x2x3xf32>
    return %2 : tensor<2x2x3xf32>
  }
// CHECK: func.func private @quantize_dot_general_batch_per_tensor_quantized_fn(%[[ARG_0:.+]]: tensor<2x2x2xf32>) -> tensor<2x2x3xf32> attributes {tf._original_func_name = "main_0"}

        none, none, none, none) -> tensor<1x2x3xf32>
  %17 = "quantfork.stats"(%16) {
    layerStats = dense<[0.0, 1.0]> : tensor<2xf32>,
    axisStats = dense<[
      [-1.0, 1.0],
      [-8.0, 8.0],
      [-0.5, 0.5]
    ]> : tensor<3x2xf32>, axis = 2 : i64
  } : (tensor<1x2x3xf32>) -> tensor<1x2x3xf32>
  func.return %17 : tensor<1x2x3xf32>

  func.return %17 : tensor<1x2x3xf32>

  // CHECK: %[[NONE:.*]] = "tfl.no_value"() <{value}> : () -> none
  // CHECK: %[[DQ_1:.*]] = "tfl.dequantize"({{.*}}) : (tensor<1x1xf16>) -> tensor<1x1xf32>
  // CHECK: %[[DQ_2:.*]] = "tfl.dequantize"({{.*}}) : (tensor<1x1xf16>) -> tensor<1x1xf32>
  // CHECK: %[[DQ_3:.*]] = "tfl.dequantize"({{.*}}) : (tensor<1x1xf16>) -> tensor<1x1xf32>

  
  %40 = "tfl.read_variable"(%4) : (tensor<!tf_type.resource>) -> tensor<1x2x3xf32>
  %41 = "quantfork.stats"(%40) {layerStats = dense<[0.0, 1.0]> : tensor<2xf32>} : (tensor<1x2x3xf32>) -> tensor<1x2x3xf32>
  %42 = "tfl.concatenation"(%41, %0) {axis = 1 : i32, fused_activation_function = "NONE"} : (tensor<1x2x3xf32>, tensor<1x2x3xf32>) -> tensor<1x4x3xf32>

  module {
    func.func @constant_add() -> (tensor<3x2xf32>) {
      %cst1 = stablehlo.constant dense<2.4> : tensor<3x2xf32>
      %cst2 = stablehlo.constant dense<5.7> : tensor<3x2xf32>
      %add = stablehlo.add %cst1, %cst2 : (tensor<3x2xf32>, tensor<3x2xf32>) -> tensor<3x2xf32>
      func.return %add : tensor<3x2xf32>
    }
  }
)mlir";