// CHECK: %[[ARG1:.*]]: tensor<!quant.uniform<i8:f32, 3.000000e-01:1>>, %[[ARG2:.*]]: tensor<!quant.uniform<i8:f32, 3.000000e-01:1>>
    // CHECK: %[[MAX:.*]] = stablehlo.maximum %[[ARG1]], %[[ARG2]] : tensor<!quant.uniform<i8:f32, 3.000000e-01:1>>
    // CHECK: stablehlo.return %[[MAX]] : tensor<!quant.uniform<i8:f32, 3.000000e-01:1>>

  %5 = "tfl.add"(%4, %arg3) {tac.device = "CPU", tac.inference_type = "QUANTIZED_INT8", fused_activation_function = "NONE"} : (tensor<1x384x128x!quant.uniform<i8:f32, 0.3:3>>, tensor<128x!quant.uniform<i8:f32, 0.2:-4>>) -> tensor<1x384x128x!quant.uniform<i8:f32, 0.3:-3>>
  func.return %5 : tensor<1x384x128x!quant.uniform<i8:f32, 0.3:-3>>

  constexpr absl::string_view kFullyQuantizedAdd = R"mlir(
    func.func @fully_quantized_add(%arg0: tensor<2x!quant.uniform<i8:f32, 1.000000e+00:0>>) -> tensor<2x!quant.uniform<i8:f32, 1.000000e+00:0>> {
      %0 = stablehlo.add %arg0, %arg0 : tensor<2x!quant.uniform<i8:f32, 1.000000e+00:0>>
      return %0 : tensor<2x!quant.uniform<i8:f32, 1.000000e+00:0>>
    }
  )mlir";

// RUN: tf-quant-opt %s -split-input-file -quant-prepare-quantize | FileCheck %s

module {
  func.func @same_scale_test(%arg0: tensor<*xf32>) -> tensor<*xf32> {
    %cst = arith.constant dense<[-1, 144]> : tensor<2xi32>
    %cst_1 = arith.constant dense<1.0> : tensor<144x10xf32>
    %cst_2 = arith.constant dense<0.1> : tensor<10xf32>
    %0 = "quantfork.qcast"(%arg0) : (tensor<*xf32>) -> tensor<*x!quant.uniform<i8:f32, 0.05:-10>>

  %0 = "tfl.pseudo_qconst"() { qtype = tensor<3x3x!quant.uniform<i32:f32:0, {1.0, 0.5:1, 0.25:1}>>, value = dense<1> : tensor<3x3xi32>} : () -> tensor<3x3x!quant.uniform<i32:f32:0, {1.0, 0.5:1, 0.25:1}>>
  func.return %0 : tensor<3x3x!quant.uniform<i32:f32:0, {1.0, 0.5:1, 0.25:1}>>
}

func.func @qu8() -> tensor<3x!quant.uniform<u8<1:255>:f32, 1.0>> {
  // CHECK-LABEL: @qu8

// UINT8-NEXT: %[[dq:.*]] = "tfl.quantize"(%[[softmax]]) <{qtype = tensor<1x401408x!quant.uniform<u8:f32, 3.906250e-03:128>>}> : (tensor<1x401408x!quant.uniform<i8:f32, 3.906250e-03>>) -> tensor<1x401408x!quant.uniform<u8:f32, 3.906250e-03:128>>
// UINT8-NEXT: return %[[dq]] : tensor<1x401408x!quant.uniform<u8:f32, 3.906250e-03:128>>
}

// CHECK: %[[dq_w:.*]] = "tfl.dequantize"(%[[q_w]]) : (tensor<64x3x3x3x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>) -> tensor<64x3x3x3xf32>
// CHECK: %[[emb:.*]] = "tfl.gather"(%[[dq_w]], %arg0)
// CHECK: return %[[emb:.*]]

// PerTensor: %[[q_w:.*]] = "tfl.pseudo_qconst"() <{qtype = tensor<64x3x3x3x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>

  %0 = "tfl.quantize"(%arg0) {qtype = tensor<2x1x!quant.uniform<i16:f32, 1.0>>} : (tensor<2x1xf32>) -> tensor<2x1x!quant.uniform<i16:f32, 1.0>>
  %1 = "tfl.dequantize"(%0) : (tensor<2x1x!quant.uniform<i16:f32, 1.0>>) -> (tensor<2x1xf32>)
  %2 = "tfl.quantize"(%arg1) {qtype = tensor<2x3x!quant.uniform<i16:f32, 1.0>>} : (tensor<2x3xf32>) -> tensor<2x3x!quant.uniform<i16:f32, 1.0>>
  %3 = "tfl.dequantize"(%2) : (tensor<2x3x!quant.uniform<i16:f32, 1.0>>) -> (tensor<2x3xf32>)

  func.return %2 : tensor<1x32x32x16x!quant.uniform<i8:f32, 1.0>>
}

// -----

// CHECK-LABEL: @test_conv2d_qi16

    :  ( tensor<1x28x28x!quant.uniform<i8:f32, 0.003:-128>>,
         tensor<20x28x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x28x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x28x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x28x!quant.uniform<i8:f32, 0.006:-34>>,