}

// -----

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>
}

func.func @testPadUnknownPaddings(tensor<2x1x3xf32>, tensor<*xi32>) -> tensor<? x f32> {
^bb0(%arg0: tensor<2x1x3xf32>, %arg1: tensor<*xi32>):
  %0 = "tfl.pad"(%arg0, %arg1) : (tensor<2x1x3xf32>, tensor<*xi32>) -> tensor<? x f32>
  func.return %0#0 : tensor<? x f32>

  // CHECK-LABEL: testPadUnknownPaddings
  // CHECK:  "tfl.pad"(%arg0, %arg1) : (tensor<2x1x3xf32>, tensor<*xi32>) -> tensor<?xf32>
  // CHECK:  return
}

// -----

    %12 = "tf.Mul"(%11, %cst) {device = ""} : (tensor<2x1x3xf32>, tensor<f32>) -> tensor<2x1x3xf32>
    %13 = "tf.Identity"(%12) {device = ""} : (tensor<2x1x3xf32>) -> tensor<2x1x3xf32>
    %14 = "tf.Identity"(%13) {device = ""} : (tensor<2x1x3xf32>) -> tensor<2x1x3xf32>
    return %14 : tensor<2x1x3xf32>
  }

// CHECK-LABEL: func @broadcasting_weight_batch_matmul

func.func @QuantizePad2(tensor<2x1x3x!quant.uniform<u8:f32, 0.1>>, tensor<2x1x3xf32>, tensor<3x2xi32>) -> (tensor<?xf32>, tensor<?xf32>) {
^bb0(%arg0: tensor<2x1x3x!quant.uniform<u8:f32, 0.1>>, %arg1: tensor<2x1x3xf32>, %arg2: tensor<3x2xi32>):
  %0 = "tfl.dequantize"(%arg0) : (tensor<2x1x3x!quant.uniform<u8:f32, 0.1>>) -> tensor<2x1x3xf32>
  %1 = "tfl.pad"(%arg1, %arg2) : (tensor<2x1x3xf32>, tensor<3x2xi32>) -> tensor<?xf32>

  // tf.XlaReduceScatter legalization
  //===--------------------------------------------------------------------===//
  // CHECK-LABEL: func @xla_reduce_scatter
  func.func @xla_reduce_scatter(%arg0: tensor<128x128xf32>) -> tensor<64x128xf32> {
      %cst = "tf.Const"() {value = dense<0> : tensor<i32>} : () -> tensor<i32>
      %cst_0 = "tf.Const"() {value = dense<[[0, 4], [1, 5], [2, 6], [3, 7]]> : tensor<4x2xi32>} : () -> tensor<4x2xi32>

  func.return %0#0 : tensor<8x8x8x8xf32>
}

// -----

// Test invalid tf.FusedBatchNorm
func.func @testFusedBatchNormWrongXType(tensor<8x8x8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>) -> tensor<8x8x8xf32> {
^bb0(%arg0: tensor<8x8x8xf32>, %arg1: tensor<8xf32>, %arg2: tensor<8xf32>, %arg3: tensor<8xf32>, %arg4: tensor<8xf32>):
  // expected-error @+1 {{requires x to be a 4D float tensor}}

}

// CHECK-LABEL: testConvertPackToReshapeAxis1
func.func @testConvertPackToReshapeAxis1(%arg0: tensor<2x3xf32>) -> tensor<2x1x3xf32> {
  %0 = "tf.Pack"(%arg0) {axis = 1 : i64, device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<2x3xf32>) -> tensor<2x1x3xf32>
  func.return %0 : tensor<2x1x3xf32>
  // CHECK: %[[SHAPE:.*]] = "tf.Const"() <{value = dense<[2, 1, 3]> : tensor<3xi32>}> : () -> tensor<3xi32>

  func.return %0 : tensor<8xf32>
}

// -----

// CHECK-LABEL: strided_slice_grad_ellipsis_mask
// CHECK-SAME: [[GRAD:%.*]]: tensor<2x4x8xf32>
func.func @strided_slice_grad_ellipsis_mask(%grad: tensor<2x4x8xf32>) -> tensor<4x4x8xf32> {
  // Input to StridedSlice was of shape 4x4x8xf32
  // Strided slice gets input[2:4, ...]

// CHECK-SAME:                         %[[VAL_1:.*]]: tensor<8x8x8x64xf32>) -> tensor<?x3x3x64xf32> {
// CHECK:           %[[VAL_2:.*]] = "tf.Conv2D"(%[[VAL_0]], %[[VAL_1]]) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [0, 0, 1, 1, 1, 1, 0, 0], padding = "EXPLICIT", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true}> : (tensor<?x8x8x8xf32>, tensor<8x8x8x64xf32>) -> tensor<?x3x3x64xf32>