%8 = stablehlo.constant dense<-5> : tensor<1x1x1xi8>  // Output zero point (z3).
    %9 = stablehlo.constant dense<1.250000e+01> : tensor<1x1x1xf32>  // Merged scale (s1 * s2).
    %10 = call @uniform_quantize(%arg0, %1, %2) : (tensor<8x16x16xf32>, tensor<1x1x1xf32>, tensor<1x1x1xi8>) -> tensor<8x16x16xi8>  // q1
    %11 = call @uniform_quantize_0(%arg1, %4, %5) : (tensor<8x16x4xf32>, tensor<1x1x1xf32>, tensor<1x1x1xi8>) -> tensor<8x16x4xi8>  // q2

}

TEST_F(AttrsAndConstraintsTest, HasRankOfReturnsTrueForMatchingRank) {
  constexpr absl::string_view kConstantOpWithRankFour =
      R"mlir(%0 = stablehlo.constant dense<0> : tensor<1x1x1x1xi8>)mlir";
  OwningOpRef<ModuleOp> module_op =
      ParseModuleOpString(kConstantOpWithRankFour);
  ASSERT_TRUE(module_op);

  ASSERT_FALSE(module_op->getBodyRegion().empty());

// RUN: stablehlo-quant-opt -optimize-int-graph -split-input-file %s -verify-diagnostics | FileCheck %s

// CHECK-LABEL: func @convolution_add_add
func.func @convolution_add_add(
    %lhs: tensor<?x3x2x1xi8>, %rhs: tensor<2x1x1x1xi8>,
    %zp_offset: tensor<?x2x2x1xi32>, %bias: tensor<1xi32>
  ) -> tensor<?x2x2x1xi32> {
  // CHECK-DAG: %[[conv:.*]] = mhlo.convolution
  // CHECK-DAG: %[[combined:.*]] = chlo.broadcast_add %[[zp_offset:.*]], %[[bias:.*]]

func.func @test_conv2d_float(%arg0: tensor<1x32x32x8xf32>) -> tensor<1x32x32x16xf32> {
  // CHECK-DAG: %[[VAL0:.+]] = "tfl.pseudo_const"() <{value = dense<42> : tensor<16x1x1x8xi8>}>
  // CHECK-DAG: %[[VAL1:.+]] = "tfl.pseudo_const"() <{value = dense<1> : tensor<16x1x1x8xi8>}>

func.func @mul_with_quantized_i16_five_dim_broadcasting(tensor<1x1x1x1x1x!quant.any<i16:f32>>, tensor<1x!quant.any<i16:f32>>) -> tensor<1x1x1x1x1x!quant.any<i16:f32>> {
^bb0(%arg0: tensor<1x1x1x1x1x!quant.any<i16:f32>>, %arg1: tensor<1x!quant.any<i16:f32>>):
  // CHECK: tfl.mul(%arg0, %arg1) <{fused_activation_function = "RELU6"}>

  %10 = "tfl.reduce_all"(%arg_unknown, %axis_unknown) { keep_dims = true } : (tensor<?xi1>, tensor<?xi32>) -> tensor<?xi1>
  func.return %0, %1, %2, %3, %4, %5, %6, %7, %8, %9, %10 : tensor<i1>, tensor<i1>, tensor<2x1x3xi1>, tensor<1x2x1x3xi1>, tensor<1x1x1x3xi1>, tensor<1x1x3xi1>, tensor<1x2x3xi1>, tensor<1x2x1x3xi1>, tensor<1x2x1xi1>, tensor<1x2x1x1xi1>, tensor<?xi1>

  func.func @simple_folding(%arg0: tensor<1x1x1x1xi32>, %arg1: tensor<1x1x1x1xf32>) -> tensor<?x?x?x?xf32> {
    // CHECK: %[[SHAPE:.*]] = "tf.Shape"
    // CHECK: %[[CONV:.*]] = "tf.Conv2DBackpropInput"(%[[SHAPE]]
    // CHECK-SAME: (tensor<4xi32>, tensor<1x1x1x1xf32>, tensor<1x1x1x1xf32>) -> tensor<1x1x1x1xf32>
    // CHECK: return %[[CONV]] : tensor<1x1x1x1xf32>
    %0 = "tf.Shape"(%arg0) : (tensor<1x1x1x1xi32>) -> tensor<4xi32>