// CHECK: return %[[v5]] : tensor<?x?x8xf32>
}

func.func @einsum_no_reshape(%arg0 : tensor<?x?x8x128xf32>, %arg1 : tensor<1x?x8x128xf32>) -> tensor<?x?x8x?xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {device = "", equation = "BQNH,BTNH->BQNT"} : (tensor<?x?x8x128xf32>, tensor<1x?x8x128xf32>) -> tensor<?x?x8x?xf32>
  func.return %0 : tensor<?x?x8x?xf32>
// CHECK-LABEL: einsum_no_reshape

func.func @testConv2dShapeInferenceDynamic(%arg0: tensor<1x?x?x128xf32>, %arg1: tensor<128x3x3x128xf32>, %arg2: tensor<128xf32>) -> tensor<1x?x?x128xf32> {
  // CHECK: "tfl.conv_2d"(%arg0, %arg1, %arg2) <{dilation_h_factor = 2 : i32, dilation_w_factor = 2 : i32, fused_activation_function = "NONE", padding = "VALID", stride_h = 1 : i32, stride_w = 1 : i32}> : (tensor<1x?x?x128xf32>, tensor<128x3x3x128xf32>, tensor<128xf32>) -> tensor<1x?x?x128xf32>

   "tfl.yield"(%4) : (tensor<128x128xf32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>

   "tfl.yield"(%4) : (tensor<128x128xf32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>

  // time constant.
  %0 = "tf.ConcatV2"(%arg0, %arg0, %arg0, %arg0, %axis) : (tensor<16x1xf32>, tensor<16x1xf32>, tensor<16x1xf32>, tensor<16x1xf32>, tensor<i32>) -> tensor<16x4xf32>

  // Just to introduce an extra use for %cst.
  %1 = "tf.AddV2"(%axis, %axis) {device = ""} : (tensor<i32>, tensor<i32>) -> tensor<i32>

  return %0 : tensor<16x4xf32>
}
// Check that the constant is cloned with same value.

}

// -----

module {
func.func private @func_20_GPU_FLOAT(%arg0: tensor<128x128xf32>, %arg1: tensor<3xi32>) -> tensor<1x128x128xf32> attributes {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_20"} {
  %0 = "tfl.reshape"(%arg0, %arg1) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<128x128xf32>, tensor<3xi32>) -> tensor<1x128x128xf32>
  func.return %0 : tensor<1x128x128xf32>
}

      // CHECK-SAME:     scatter_dimension = 0
      //
      %1 = "tf.XlaReduceScatter"(%arg0, %cst_0, %cst) {reduce_op = "Add"} : (tensor<128x128xf32>, tensor<4x2xi32>, tensor<i32>) -> tensor<64x128xf32>
      func.return %1 : tensor<64x128xf32>
  }

  // CHECK-LABEL: func @tf_mod
  func.func @tf_mod(%arg1: tensor<2x2xf32>) -> tensor<2x2xf32> {

// CHECK:}

func.func @reshape(%arg0 : tensor<1x128xi32>) -> tensor<4x32x1xi32>{
  %0 = "vhlo.reshape_v1"(%arg0) : (tensor<1x128xi32>) -> tensor<4x32x1xi32>
  func.return %0 : tensor<4x32x1xi32>
}

//CHECK:func.func private @reshape(%arg0: tensor<1x128xi32>) -> tensor<4x32x1xi32> {
//CHECK-NEXT: %0 = "vhlo.reshape_v1"(%arg0) : (tensor<1x128xi32>) -> tensor<4x32x1xi32>
//CHECK-NEXT: return %0 : tensor<4x32x1xi32>

  %0 = mhlo.reshape %arg1 : (tensor<1x72x128xf32>) -> tensor<72x128xf32>
  %1 = "mhlo.dot_general"(%arg0, %0) {
    dot_dimension_numbers = #mhlo.dot<
      lhs_contracting_dimensions = [2],
      rhs_contracting_dimensions = [0]
    >} : (tensor<1x72x72xf32>, tensor<72x128xf32>) -> tensor<1x72x128xf32>
  func.return %1 : tensor<1x72x128xf32>

// CHECK-SAME: (tensor<1x37xf32>, tensor<40x37xf32>, none) -> tensor<1x40xf32>
  func.return %1 : tensor<1x40xf32>
}

// CHECK-LABEL: RemoveLstmQuantZeroBias
func.func @RemoveLstmQuantZeroBias(
  %arg0: tensor<1x528xf32>,
  %arg1: tensor<2048x528xf32>,
  %arg2: tensor<2048x528xf32>,
  %arg3: tensor<2048x528xf32>,
  %arg4: tensor<2048x528xf32>,
  %arg5: tensor<2048x640xf32>,
  %arg6: tensor<2048x640xf32>,