#include "tensorflow/compiler/mlir/tensorflow/ir/tf_dialect.h"
#include "tensorflow/compiler/mlir/tensorflow/ir/tf_ops_a_m.h"
#include "tensorflow/compiler/mlir/tensorflow/ir/tf_ops_n_z.h"
#include "xla/mlir_hlo/mhlo/IR/hlo_ops.h"

namespace mlir {
namespace odml {

// Broadcasts the 1D value tensor 'value_1d' to the shape of 'result_type'. If
// 'shape_value' is initialized, creates a dynamic broadcast, otherwise creates

#include "tensorflow/compiler/mlir/tensorflow/utils/stablehlo_custom_call.h"
#include "tensorflow/compiler/mlir/tensorflow/utils/xla_call_module_attrs.h"
#include "tensorflow/compiler/tf2xla/kernels/xla_call_module_loader.h"
#include "xla/mlir_hlo/mhlo/IR/hlo_ops.h"  // IWYU pragma: keep
#include "tsl/platform/statusor.h"

namespace mlir {
namespace TF {
namespace {

#define GEN_PASS_DEF_XLACALLMODULEDESERIALIZATIONPASS

#include "absl/container/flat_hash_set.h"
#include "absl/container/inlined_vector.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/str_join.h"
#include "absl/strings/str_split.h"
#include "llvm/ADT/STLExtras.h"
#include "xla/status_macros.h"
#include "tensorflow/core/framework/tensor_shape.pb.h"
#include "tensorflow/core/framework/types.h"
#include "tensorflow/core/framework/types.pb.h"
#include "tensorflow/core/lib/core/errors.h"

#include "tensorflow/compiler/mlir/tensorflow/ir/tf_types.h"
#include "tensorflow/compiler/mlir/tensorflow/utils/convert_tensor.h"
#include "tensorflow/compiler/mlir/tensorflow/utils/tpu_rewrite_device_util.h"
#include "xla/mlir_hlo/mhlo/IR/hlo_ops.h"
#include "tensorflow/core/framework/tensor.h"
#include "tensorflow/core/framework/tensor_shape.pb.h"
#include "tensorflow/core/framework/types.pb.h"
#include "tensorflow/core/platform/random.h"

        "@llvm-project//mlir:Rewrite",
        "@llvm-project//mlir:Support",
        "@llvm-project//mlir:TransformUtils",
        "@local_xla//xla:xla_data_proto_cc",
        "@local_xla//xla/client:sharding_builder",
    ],
)

cc_library(
    name = "hoist_broadcast_read",
    srcs = ["hoist_broadcast_read.cc"],
    textual_hdrs = [
        "clustering_passes.h.inc",
    ],

#include "absl/container/inlined_vector.h"
#include "absl/strings/match.h"
#include "absl/strings/numbers.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/str_join.h"
#include "tensorflow/compiler/jit/flags.h"
#include "xla/status_macros.h"
#include "tensorflow/core/common_runtime/function.h"
#include "tensorflow/core/framework/bounds_check.h"
#include "tensorflow/core/framework/node_def.pb.h"
#include "tensorflow/core/graph/control_flow.h"

// (unlike the rest), since this is the primary use case for such ops and
// verification of shapes and broadcasts is desired.
// RUN: tf-opt "-xla-legalize-tf=legalize-chlo=true" -canonicalize %s | FileCheck %s
// RUN: tf-opt "-xla-legalize-tf=legalize-chlo=false" %s | FileCheck --check-prefix CHLO %s

//===----------------------------------------------------------------------===//
// Binary op legalizations.

// width)) tensor format to HWIO. This is commonly used to transpose convolution
// weights represented as OIHW format to HWIO, which is more desirable for
// certain downstream optimization passes (e.g. XLA).
inline constexpr std::array<int64_t, 4> kOihwToHwioPermutation = {2, 3, 1, 0};

// Returns true if the value has static shape.
bool HasStaticShape(Value value);

#include "tensorflow/cc/ops/control_flow_ops_internal.h"
#include "tensorflow/cc/ops/function_ops.h"
#include "tensorflow/cc/ops/functional_ops.h"
#include "tensorflow/cc/ops/standard_ops.h"
#include "xla/status_macros.h"
#include "tensorflow/core/common_runtime/graph_constructor.h"
#include "tensorflow/core/common_runtime/process_function_library_runtime.h"
#include "tensorflow/core/framework/function_testlib.h"

#include "mlir/Support/LogicalResult.h"  // from @llvm-project
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"  // from @llvm-project
#include "tensorflow/compiler/mlir/lite/stablehlo/transforms/passes.h"
#include "xla/mlir_hlo/mhlo/IR/hlo_ops.h"

namespace mlir {
namespace odml {

static const APFloat &addSign(const APFloat &v, Type) { return v; }
static APSInt addSign(const APInt &v, Type t) {