// RUN: tf-opt %s -split-input-file -tfl-legalize-hashtables-tf --cse | FileCheck %s

// Test for case with string -> int64 hashtable.
func.func @hashtable_string_to_int64(%arg0: tensor<i64>) -> tensor<*xi64> {
  %cst = arith.constant dense<"f"> : tensor<!tf_type.string>

// RUN: tf-opt -convert-tf-quant-types -quant-convert-tf-quant-ops-to-mhlo -canonicalize "-xla-legalize-tf=legalize-chlo=false" -split-input-file %s | FILECHECK_OPTS="" FileCheck %s


//===----------------------------------------------------------------------===//
// tf.UniformQuantizedDotHybrid legalization
//===----------------------------------------------------------------------===//

// CHECK-LABEL: func @quantized_matmul_fn

        continue;
      if (auto size_op = llvm::dyn_cast<TF::LookupTableSizeV2Op>(user))
        continue;

      return false;
    }
  }
  return true;
}

// Pass which legalizes TF hash tables only when they are covered by the
// TensorFlow Lite hash table kernels.
class LegalizeHashTablesPass
    : public impl::LegalizeHashTablesPassBase<LegalizeHashTablesPass> {
 public:

                                         llvm::SmallVector<Value>& outputs);

  // Tries to legalize the specified TensorFlow op, if supported.
  //
  // Emits an error and returns failure if an error is encountered during
  // conversion. Note that success return value doesn't mean successful
  // legalization.
  mlir::LogicalResult LegalizeOp();

  // Converts the given operand to expression of kind kConstant or kXlaOp.

// RUN: tf-opt %s -tfl-legalize-tf --cse -split-input-file| FileCheck %s --dump-input=fail

func.func @add(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<1xf32> {
  %0 = "tf.Add"(%arg0, %arg1) : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
  func.return %0: tensor<1xf32>

// CHECK-LABEL: add
// CHECK:  tfl.add %arg0, %arg1 {fused_activation_function = "NONE"} : tensor<1xf32>
// CHECK:  return
}

  ];
}

def LegalizeHloToTfPass : Pass<"tf-legalize-hlo", "ModuleOp"> {
  let summary = "Legalize from MHLO to the TF dialect";
  let dependentDialects = ["TF::TensorFlowDialect"];
  let constructor = "mlir::odml::CreateLegalizeHloToTfPass()";
}

def LegalizeHloToTfLitePass : Pass<"tfl-legalize-hlo", "mlir::ModuleOp"> {
  let summary = "Legalize from MHLO to the TFLite dialect";

    name = "all_tests",
    data = [":test_utilities"],
    driver = "@llvm-project//mlir:run_lit.sh",
    size_override = {
        "legalize-tf-binary-elementwise.mlir": "medium",
        "legalize-tf-include-tf2xla-fallback.mlir": "medium",
        "legalize-tf-prefer-tf2xla.mlir": "medium",
        "legalize-tf.mlir": "medium",
    },
    test_file_exts = [
        "mlir",
        "hlotxt",
    ],
)

After serialization, this pass removes the `_entry_function` attribute from
`tf.XlaCallModule`, and removes all the serialized stablehlo functions
from the top-level module.
### `-tfe-legalize-tfg`

// RUN: tf-opt -split-input-file -verify-diagnostics -xla-legalize-tf-communication %s | FileCheck %s

// Test legalization of `tf._XlaHostComputeMlir` expands into individual
// `mhlo.send` per operand and `mhlo.recv` per result. Channel Id's are uniquely
// assigned per mhlo communcation op, and frontend attributes (modified keys)
// and op shardings are added. Sink tokens are created
// if there are more than one operand or more than one result.
//

// -----

// ============================================================================
// The following functions tests example quantization patterns outputted from
// StableHLO Quantizer. These patterns should be legalized early directly
// to fused tflite ops.
// ============================================================================

// Tests that a simple per-channel quantized `stablehlo.dot_general` is properly