func.return %4 : tensor<4x72x72xf32>

// CHECK:      %[[R0:.*]] = "mhlo.concatenate"(%arg0, %arg2, %arg4, %arg6) <{dimension = 0 : i64}> : (tensor<1x72x128xf32>, tensor<1x72x128xf32>, tensor<1x72x128xf32>, tensor<1x72x128xf32>) -> tensor<4x72x128xf32>

  // CHECK-NEXT: [[RESULT:%.*]] = "tf.BiasAdd"([[SQUEEZE]], [[BIAS]]) : (tensor<1x128x128xf32>, tensor<128xf32>) -> tensor<1x128x128xf32>
  // CHECK-NEXT: return [[RESULT]] : tensor<1x128x128xf32>
}

func.func @testDilatedDepthWiseConvWithExpandSqueeze1(%arg0: tensor<1x128x128xf32>, %arg1: tensor<5x5x1x1xf32>, %arg2: tensor<128xf32>) -> tensor<1x128x128xf32> {

  %7 = "tfl.add"(%1, %6) {tac.device = "GPU", tac.inference_type = "FLOAT", fused_activation_function = "NONE"} : (tensor<1x128x128xf32>, tensor<1x128x128xf32>) -> tensor<1x128x128xf32>
  func.return %7 : tensor<1x128x128xf32>
}

// CHECK:   func @norm1(%[[VAL_0:.*]]: tensor<1x128x128xf32>) -> tensor<1x128x128xf32> {

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-NOT: "tfl.batch_matmul"
func.func @Batchmatmul2FullyconnectedAdjx(%arg0: tensor<4x2x128xf32>) -> (tensor<4x128x1xf32>) {
  %0 = arith.constant dense<[[1.0], [2.0]]> : tensor<2x1xf32>
  %1 = "tfl.batch_matmul"(%arg0, %0) {adj_x = true, adj_y = false, asymmetric_quantize_inputs = false} : (tensor<4x2x128xf32>, tensor<2x1xf32>) -> tensor<4x128x1xf32>
  func.return %1 : tensor<4x128x1xf32>

  // CHECK-DAG: %[[V21:.*]] = mhlo.add %[[V13]], %[[V20]] : tensor<1x22x128xi32>
  // CHECK-DAG: %[[V22:.*]] = mhlo.maximum %[[V18]], %[[V5]] : tensor<1x22x128xi32>
  // CHECK-DAG: %[[V23:.*]] = mhlo.subtract %[[V15]], %[[V22]] : tensor<1x22x128xi32>
  // CHECK-DAG: %[[V24:.*]] = mhlo.minimum %[[V21]], %[[V23]] : tensor<1x22x128xi32>

// CHECK: return %[[v4]] : tensor<?x?x8x?xf32>
}

func.func @einsum_ellipsis(%arg0: tensor<1x512x128xf32>, %arg1: tensor<128x256xf32>) -> tensor<1x512x256xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {device = "", equation = "...x,xy->...y"} : (tensor<1x512x128xf32>, tensor<128x256xf32>) -> tensor<1x512x256xf32>
  func.return %0 : tensor<1x512x256xf32>
// CHECK-LABEL: einsum_ellipsis

  %403 = "tfl.pseudo_qconst"() {qtype = tensor<128x64x!quant.uniform<i8:f32, 0.0047710379585623741>>, value = dense<9> : tensor<128x64xi8>} : () -> tensor<128x64x!quant.uniform<i8:f32, 0.0047710379585623741>>