self.assertNotEmpty(function_aliases)
    self.assertCountEqual(function_aliases.values(), {'conv_func'})

    # Test that the aliased function contains a quantized op.
    for func_name, alias in function_aliases.items():
      if alias == 'conv_func':
        for func in meta_graph_def.graph_def.library.function:
          if func.signature.name == func_name:

// RUN: stablehlo-quant-opt %s -split-input-file -stablehlo-lift-quantizable-spots-as-functions | FileCheck %s

// CHECK-LABEL: @conv_fn(
// CHECK-SAME:          %[[ARG_0:.*]]: tensor<1x3x3x4xf32>
func.func @conv_fn(%arg0: tensor<1x3x3x4xf32>) -> tensor<1x3x3x4xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<3x3x4x4xf32>

                                   updated_argument_types[arg.getArgNumber()]);
  }
  rewriter.applySignatureConversion(&entry.front(), signature_conversion);
}

// Changes the function type of `cond_func` and `body_func` for the given While
// op.
LogicalResult UpdateFunctionTypesForWhileOp(
    ConversionPatternRewriter &rewriter, TF::WhileOp op, ValueRange operands,
    const llvm::SmallSet<int, 4> &tensor_list_args,

// This case covers for the following quantization patterns because
// activation clipping ranges take affect in scale and zp of the final
// `stablehlo.uniform_quantize`. See more details in b/319168201.
// * conv_fn
// * conv_with_relu_fn
// * conv_with_relu6_fn

func.func @conv_srq(%arg0: tensor<1x5x5x2x!quant.uniform<i8:f32, 2.000000e+00:0>>) -> (tensor<1x4x4x4x!quant.uniform<i8:f32, 8.000000e+00:-128>>) {