// RUN: tf-opt %s -tf-optimize | FileCheck %s

// CHECK-LABEL: @fuseMulIntoConv2d
func.func @fuseMulIntoConv2d(%arg0: tensor<1x112x112x3xf32>) -> tensor<1x28x23x2xf32> {
  %cst0 = arith.constant dense<[[[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], [[7.0, 8.0], [9.0, 10.0], [11.0, 12.0]], [[13.0, 14.0], [15.0, 16.0], [17.0, 18.0]]]]> : tensor<1x3x3x2xf32>
  %cst2 = arith.constant dense<[1.0, 2.0]> : tensor<2xf32>

  %6 = "tfl.quantize"(%5) {qtype = tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>} : (tensor<1x112x112x32xf32>) -> tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>
  func.return %6 : tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>

  func.return %3 : tensor<1x112x112x32xf32>

  func.return %3 : tensor<1x112x112x32xf32>

  %1 = "tfl.mul"(%0, %cst2) {fused_activation_function = "NONE"} : (tensor<1x112x112x2xf32>, tensor<1x112x112x2xf32>) -> tensor<1x112x112x2xf32>

  func.return %1 : tensor<1x112x112x2xf32>

  func.return  %3, %6 : tensor<?xf32>, tensor<?xf32>
}

// -----

// CHECK-LABEL: func @skip_nan_inf_constant
// CHECK-SAME: (%[[ARG_0:.*]]: tensor<?x112x112x64xf32>) -> tensor<?x56x56x64xf32>
func.func @skip_nan_inf_constant(%arg0: tensor<?x112x112x64xf32>) -> tensor<?x56x56x64xf32> {
  // CHECK-DAG: %[[cst0:.*]] = stablehlo.constant dense<0xFF800000> : tensor<f32

  %6 = "tfl.quantize"(%5) {qtype = tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>} : (tensor<1x112x112x32xf32>) -> tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>
  func.return %6 : tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>

    %15 = "tf.Conv2D"(%11, %arg2) {data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 2, 2, 1], use_cudnn_on_gpu = true} : (tensor<2x230x230x3xf32>, tensor<7x7x3x64xf32>) -> tensor<2x112x112x64xf32>
    %16 = "tf.Mean"(%15, %8) {keep_dims = false} : (tensor<2x112x112x64xf32>, tensor<2xi32>) -> tensor<2x64xf32>

  func.return %2 : tensor<1x112x112x32xf32>
}

// -----

// CHECK-LABEL: @test_fullyconnected_replace_float
func.func @test_fullyconnected_replace_float(%arg0: tensor<4x256x6x6xf32>) -> tensor<4x256x36xf32> {

               %filter: tensor<7x7x3x64xf32>) {
     %6 = "tf.Conv2D"(%input, %filter)  {strides = [1, 2, 2, 1]}:
                 (tensor<2x230x230x3xf32>, tensor<7x7x3x64xf32>) ->
      tensor<2x112x112x64xf32>
   }
}

// With this pass, the program will be transformed into:
module {
   func @while_body {
     %input = "tf.IteratorGetNext"(...) {device = "/CPU:0"}
               -> tensor<2x224x224x3xf32>