//tensorflow/cc/saved_model:saved_model_half_plus_two
//tensorflow:no_tensorflow_py_deps
//tensorflow/tools/pip_package:win_pip_package_marker
//tensorflow/core:image_testdata
//tensorflow/core/lib/lmdb:lmdb_testdata
//tensorflow/core/lib/lmdb/testdata:lmdb_testdata
//tensorflow/core/kernels/cloud:bigquery_reader_ops
//tensorflow/python:extra_py_tests_deps
//tensorflow/python:mixed_precision
//tensorflow/python:tf_optimizer

vulnerabilities.

## Security properties of execution modes

TensorFlow has several execution modes, with Eager-mode being the default in v2.
Eager mode lets users write imperative-style statements that can be easily
inspected and debugged and it is intended to be used during the development
phase.

As part of the differences that make Eager mode easier to debug, the [shape
inference

#define GEN_PASS_DEF_FUSEDKERNELMATCHERPASS
#include "tensorflow/compiler/mlir/tensorflow/transforms/tf_passes.h.inc"

// Optimizes TF computations by fusing subgraphs/nodes onto more efficient
// implementations to decrease the number of operations needed to perform a
// computation.
struct FusedKernelMatcherPass
    : public impl::FusedKernelMatcherPassBase<FusedKernelMatcherPass> {

  let summary = "Updates inputs to TPU embedding enqueue ops depending on whether graph is in training mode or in evaluation mode.";

  let description = [{
    Updates inputs to TPU embedding enqueue ops depending on whether graph
    is in training mode or in evaluation mode.
  }];

  let constructor = "TFTPU::CreateTPUUpdateEmbeddingEnqueueOpInputsPass()";
}

#include "tensorflow/c/c_test_util.h"
#include "tensorflow/core/framework/common_shape_fns.h"
#include "tensorflow/core/framework/function.h"
#include "tensorflow/core/framework/function.pb.h"
#include "tensorflow/core/framework/op_def.pb.h"
#include "tensorflow/core/lib/hash/hash.h"
#include "tensorflow/core/lib/strings/proto_serialization.h"
#include "tensorflow/core/platform/logging.h"
#include "tensorflow/core/platform/status.h"

      // TODO(b/172517537): check if other tensors should go through this
      // check too.
      op.emitError() << "Input tensor [" << input_index
                     << "] is a state tensor, but has more than one use.";
      return failure();
    }
    auto stats = mlir::dyn_cast<DenseFPElementsAttr>(stats_op.getLayerStats());
    if (!stats || stats.getNumElements() != 2) {

// `base_loc` is used for error reporting and debug info.
// If ordered_output_arrays is not empty, then the imported mlir function will
// only return nodes in ordered_output_arrays in the same order. Returns nullptr
// on failure, and more specific errors will be emitted via the context.
// If `use_external_constant` is true, it will create `tfl.external_const`
// instead of `tfl.const`.
// If `experimental_prune_unreachable_nodes_unconditionally` is true, nodes that

    int64_t dim_val = dim.value().getSExtValue();
    split_sizes.push_back(dim_val);
    if (dim_val == tensorflow::kTFDynamicSize) {
      // We cannot have more than one dynamic dimension.
      if (dynamic_dim_index)
        return op.emitOpError(
            "cannot have more than one dynamic dimension in split sizes");
      dynamic_dim_index = dim.index();
    } else {
      total_dim_size += dim_val;
    }
  }

    if (llvm::isa<FixedOutputRangeInterface, SameScalesOpInterface>(def) ||
        !def->hasTrait<OpTrait::quant::QuantizableResult>()) {
      return failure();
    }

    // This op should not clobber def, if more than one requant of this value.
    if (!pre_quantized.hasOneUse()) {
      return failure();
    }

    op.emitWarning("Remove trivial `rescale` op. Please fix the source graph.");

          kStablehloModuleDefaultEntryFuncName);
      // TODO(b/291988976): move enforcement of this variable outside of this
      // rewrite pattern such that it's only checked once. Currently, this
      // approach results in duplicate error messages as this pattern executes
      // more than once.
      if (!IsShloMainFuncOp(main_func_op)) {
        auto error_msg =