// CHECK-LABEL: @quantize_constant
// CHECK-SAME: %[[ARG0:.*]]: tensor<1x3xf32>
func.func @quantize_constant(%arg0: tensor<1x3xf32>) -> tensor<1x2xf32> {
  // CHECK-DAG: %[[QCST:.*]] = stablehlo.constant() <{value = dense<-78> : tensor<3x2xi8>}> : () -> tensor<3x2x!quant.uniform<i8<-127:127>:f32, 5.000000e-03>>
  // CHECK-DAG: %[[Q1:.*]] = stablehlo.uniform_quantize %[[ARG0]]
  // CHECK-NOT: "quantfork.qcast"
  // CHECK: %[[DOT:.*]] = stablehlo.dot %[[Q1]], %[[QCST]]

    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>>

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>>

  module attributes {tf.versions = {bad_consumers = [], min_consumer = 0 : i32, producer = 268 : i32}} {
  func.func @main() -> () {
    %cst0 = "tf.Const"(){ value = dense<0> : tensor<3x5xi1>} : () -> tensor<3x5xi1>
    %0 = "tf.Where"(%cst0) : (tensor<3x5xi1>) -> tensor<?x2xi64>
    func.return
  }
})";

absl::StatusOr<XlaCompiler::CompilationResult> CompileMlirModule(
    const char* mlir_module_str,

}

// -----

func.func @batchMatMulV3MatrixInt8(%arg0: tensor<4x5xi8>, %arg1: tensor<5x6xi8>) -> tensor<4x6xi32> {
  %0 = "tf.BatchMatMulV3"(%arg0, %arg1) : (tensor<4x5xi8>, tensor<5x6xi8>) -> tensor<4x6xi32>
  func.return %0 : tensor<4x6xi32>

  // CHECK-LABEL: batchMatMulV3MatrixInt8
  // CHECK: %0 = "tf.BatchMatMulV3"(%arg0, %arg1) : (tensor<4x5xi8>, tensor<5x6xi8>) -> tensor<4x6xi32>
  // CHECK: return %0 : tensor<4x6xi32>

    func.return %1 : tensor<2x2xf32>
  }

  // CHECK-LABEL: dynamic_update_slice
  // CHECK-SAME: (%[[ARG0:.*]]: tensor<3x4xi32>, %[[ARG1:.*]]: tensor<2x2xi32>, %[[ARG2:.*]]: tensor<2xi32>
  func.func @dynamic_update_slice(%arg0: tensor<3x4xi32>, %arg1: tensor<2x2xi32>, %arg2: tensor<2xi32>) -> tensor<3x4xi32> {

    // CHECK: %[[SLICE0:.*]] = "mhlo.slice"(%[[ARG2]])
    // CHECK-DAG-SAME: start_indices = dense<0> : tensor<1xi64>

func.func @no_pushdown_gather_with_no_reduction(%arg0: tensor<2xi32>) -> tensor<2x2xf32> {
  %w = "tfl.pseudo_qconst"() {qtype = tensor<2x2x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>, value = dense<127> : tensor<2x2xi8>} : () -> tensor<2x2x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>
  %dq_w = "tfl.dequantize"(%w) : (tensor<2x2x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>) -> tensor<2x2xf32>

// TF-DAG: %[[cc_0:.*]] = "tf.ConcatV2"(%[[pc_1]], %[[pc_5]], %[[cst_0]]) : (tensor<2x2xi8>, tensor<2x2xi8>, tensor<i32>) -> tensor<2x4xi8>
// TF-DAG: %[[pc_7:.*]] = "tf.PartitionedCall"(%[[cc_0]], %[[cst_6]], %[[cst_5]]) <{config = "", config_proto = "", executor_type = "", f = @dequantize_i8}
// TF-DAG: return %[[pc_7]]

  %weight_zps = "tf.Const"() { value = dense<3> : tensor<i32> } : () -> tensor<i32>

  // CHECK: "tf.AddV2"
  // CHECK: mhlo.constant
  // CHECK-SAME{LITERAL}: dense<[[1, 2], [3, 4]]> : tensor<2x2xi8>
  %0 = "tf.AddV2"(%input, %input) : (tensor<?x?xf32>, tensor<?x?xf32>) -> tensor<?x?xf32>

  // CHECK: "mhlo.dot"
  // CHECK-SAME: (tensor<?x?xf32>, tensor<2x2x!quant.uniform<i8:f32, 1.000000e+00:3>>) -> tensor<?x?xf32>

  // CHECK: %[[OPT_BARRIER:.*]]:2 = mhlo.optimization_barrier %arg0, %arg1
  // CHECK-NEXT: return %[[OPT_BARRIER]]#0, %[[OPT_BARRIER]]#1
  %0, %1 = "tf.XlaOptimizationBarrier"(%arg0, %arg1) : (tensor<4x4xf32>, tensor<3x4xi32>) -> (tensor<4x4xf32>, tensor<3x4xi32>)
  func.return %0, %1 : tensor<4x4xf32>, tensor<3x4xi32>
}

// CHECK-LABEL: @ifRegion