// PerTensor-SAME: f = @quantized_matmul_fn_0}> : (tensor<2x12xf32>, tensor<12x2xi8>, tensor<f32>, tensor<i32>) -> tensor<*xf32>
// PerTensor: return %[[out]]

// PerChannel-LABEL: func @matmul
// PerChannel-DAG: %[[q_w:.*]] = "tf.Const"() <{value = dense<0> : tensor<12x2xi8>}> : () -> tensor<12x2xi8>
// PerChannel-DAG: %[[scale:.*]] = "tf.Const"() <{value = dense<3.93700805E-9> : tensor<f32>}> : () -> tensor<f32>

}

func.func @RemoveTrivialSubInt8(%arg0: tensor<2x2xi8>) -> tensor<2x2xi8> {
  %cst = arith.constant dense<0> : tensor<2x2xi8>
  %0 = "tf.Sub"(%arg0, %cst) : (tensor<2x2xi8>, tensor<2x2xi8>) -> tensor<2x2xi8>
  func.return %0 : tensor<2x2xi8>

  // CHECK-LABEL: RemoveTrivialSubInt8
  // CHECK-NEXT: return %arg0 : tensor<2x2xi8>
}

func.func @RemoveTrivialMul(%arg0: tensor<2x2xf32>) -> tensor<2x2xf32> {

// CHECK: %[[quant:.*]] = "tfr.cast"(%arg0) : (tensor<1x10xi8>) -> !tfr.tensor
// CHECK: %[[scale_cast:.*]] = "tfr.cast"(%[[scale]])
// CHECK: %[[zp_cast:.*]] = "tfr.cast"(%[[zp]])
// CHECK: %[[requant:.*]] = tfr.call @tf__requantize(%[[quant]], %[[scale_cast]], %[[zp_cast]])
// CHECK: %[[result:.*]] = "tfr.cast"(%[[requant]])
// CHECK-NOT: quant.uniform
// CHECK: return %[[result]], %[[float_resunt:.*]] : tensor<1x10xi8>, tensor<1x5xf32>
}

// WHOLE-PASSES-NEXT:  %[[q2:.*]] = "tfl.quantize"(%[[cc1]]) <{qtype = tensor<1x4x3x!quant.uniform<u8:f32, {{.*}}>>}> : (tensor<1x4x3x!quant.uniform<u8:f32, {{.*}}>>) -> tensor<1x4x3x!quant.uniform<u8:f32, {{.*}}>>

  func.return %0 : tensor<1x2x!quant.uniform<i8:f32, 0.0078431372549019607:-128>>

// LEGACY:  "tfl.pseudo_qconst"() <{qtype = tensor<1x2x!quant.uniform<i8:f32, 0.0078431372549019607:-128>>, value = dense<{{\[\[}}-1, 127]]> : tensor<1x2xi8>}>
}

func.func private @testIfThen(tensor<*xf32>) -> tensor<*xf32>
func.func private @testIfElse(tensor<*xf32>) -> tensor<*xf32>

// CHECK-LABEL: NotQuantizeIf
func.func @NotQuantizeIf(%arg0: tensor<i1>,

  // CHECK-NEXT:     %[[GROUP:.*]] = "tf.Const"()
  // CHECK-SAME:       [0, 1, 2, 3]
  // CHECK-NEXT:     %[[REDUCED:.*]] = "tf.XlaAllReduce"(%[[ID]], %[[GROUP]]) <{mode = "CrossReplica", reduce_op = "Add"}> : (tensor<f32>, tensor<1x4xi32>) -> tensor<f32>
  // CHECK-NEXT:     "tf.OpA"(%[[REDUCED]]) : (tensor<f32>) -> ()
  tf_device.replicate {n = 4 : i32} {
    "tf_device.cluster"() ({
      "tf.OpA"(%arg0) : (tensor<f32>) -> ()
      tf_device.return

TEST_F(QuantizationReportTest, InitializeWithModuleOp) {
  constexpr absl::string_view kQuantizedDotGeneral = R"mlir(
    func.func @main(%arg0: tensor<1x2xf32>) -> tensor<1x3xf32> {
      %0 = stablehlo.constant() {value = dense<127> : tensor<2x3xi8>} : () -> tensor<2x3x!quant.uniform<i8<-127:127>:f32:1, {1.000000e+0,2.000000e+0,3.000000e+0}>>
      %1 = stablehlo.uniform_quantize %arg0 : (tensor<1x2xf32>) -> tensor<1x2x!quant.uniform<i8:f32, 4.000000e+0>>

    return %0 : tensor<1x3xf32>
  }
}

// CHECK-LABEL: quantize_dot_general_per_tensor
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x2xf32>
// CHECK: %[[CST:.+]] = stablehlo.constant() <{value = dense<127> : tensor<2x3xi8>}> : () -> tensor<2x3x!quant.uniform<i8<-127:127>:f32, 0.0023622048182750312>>

  func.return %0 : tensor<4xf64>
}

func.func @i8() -> tensor<4xi8> {
  // CHECK-LABEL: @i8
  // CHECK: value = dense<[1, 2, 3, 4]> : tensor<4xi8>
  %0 = "tfl.pseudo_const" () { value = dense<[1, 2, 3, 4]> : tensor<4xi8> } : () -> tensor<4xi8>
  func.return %0 : tensor<4xi8>
}

func.func @i16() -> tensor<4xi16> {
  // CHECK-LABEL: @i16

  func.return %0 : tensor<4xf64>
}

func.func @i8() -> tensor<4xi8> {
  // CHECK-LABEL: @i8
  // CHECK: value = dense<[1, 2, 3, 4]> : tensor<4xi8>
  %0 = "tfl.pseudo_const" () { value = dense<[1, 2, 3, 4]> : tensor<4xi8> } : () -> tensor<4xi8>
  func.return %0 : tensor<4xi8>
}

func.func @i16() -> tensor<4xi16> {
  // CHECK-LABEL: @i16