std::unique_ptr<OpQuantScaleSpec> GetStableHloQuantConstraints(Operation* op) {
  auto scale_spec = std::make_unique<OpQuantScaleSpec>();
  if (llvm::isa<mlir::stablehlo::BroadcastInDimOp,
                mlir::stablehlo::ConcatenateOp,
                mlir::stablehlo::DynamicReshapeOp,
                mlir::stablehlo::DynamicSliceOp, mlir::stablehlo::GatherOp,
                mlir::stablehlo::PadOp, mlir::stablehlo::ReduceWindowOp,

  // Insert the concat op.
  auto concat_out_type =
      RankedTensorType::get(concat_out_shape, input_type.getElementType());
  auto concat_op = rewriter.create<TFL::ConcatenationOp>(
      pack_op.getLoc(), concat_out_type, pack_inputs, pack_op.getAxis(),
      "NONE");

  auto reshape_op =
      InsertReshapeOp(pack_op.getLoc(), concat_op, input_type.getElementType(),

       2},
      builder.getIntegerType(32));
  // Concatenate the batch dimensions, flattened out dimension and flattened
  // contracting dimension.
  return builder.create<TFL::ConcatenationOp>(
      concat_result_type, flattend_shape_values, /*axis*/ 0,
      /*fused_activation_function*/ "NONE");
}
}  // namespace

Value ConvertDot(PatternRewriter& rewriter, Value lhs, Value rhs,

\"object.x__dash__prop > 0\"}\n  - Expression accessing a property named \"redact__d\": {\"Expression\": \"object.redact__underscores__d > 0\"}\n\nEquality on arrays with list type of 'set' or 'map' ignores element order, i.e. [1, 2] == [2, 1]. Concatenation on arrays with x-kubernetes-list-type use the semantics of the list type:\n  - 'set': `X + Y` performs a union where the array positions of all elements in `X` are preserved and\n    non-intersecting elements in `Y` are appended, retaining their...

\"object.x__dash__prop > 0\"}\n  - Expression accessing a property named \"redact__d\": {\"Expression\": \"object.redact__underscores__d > 0\"}\n\nEquality on arrays with list type of 'set' or 'map' ignores element order, i.e. [1, 2] == [2, 1]. Concatenation on arrays with x-kubernetes-list-type use the semantics of the list type:\n  - 'set': `X + Y` performs a union where the array positions of all elements in `X` are preserved and\n    non-intersecting elements in `Y` are appended, retaining their...

func.func @fuseUnpackAndConcatToReshape(%arg0: tensor<1x3x2xf32>) -> tensor<1x6xf32> {
  %0:3 = "tfl.unpack"(%arg0) {axis = 1 : i32, num = 3 : i32} : (tensor<1x3x2xf32>) -> (tensor<1x2xf32>, tensor<1x2xf32>, tensor<1x2xf32>)
  %1 = "tfl.concatenation"(%0#0, %0#1, %0#2) {axis = -1 : i32, fused_activation_function = "NONE"} : (tensor<1x2xf32>, tensor<1x2xf32>, tensor<1x2xf32>) -> tensor<1x6xf32>
  func.return %1 : tensor<1x6xf32>

Assumes that all arguments of the function are Tensors which will be batched
along their first dimension.

Arguments that are captured, are not batched. The session.run call which does
the concatenation, will use the values of the captured tensors available to it.
Therefore, typical uses of captured tensors should involve values which remain
unchanged across session.run calls. Inference is a good example of this.

     * is the only one in use, but if we encounter nested concatenations, we start a deque of
     * meta-iterators rather than letting the nesting get arbitrarily deep.  This keeps each
     * operation O(1).
     */

    @CheckForNull private Iterator<? extends Iterator<? extends T>> topMetaIterator;

    // Only becomes nonnull if we encounter nested concatenations.

     * is the only one in use, but if we encounter nested concatenations, we start a deque of
     * meta-iterators rather than letting the nesting get arbitrarily deep.  This keeps each
     * operation O(1).
     */

    @CheckForNull private Iterator<? extends Iterator<? extends T>> topMetaIterator;

    // Only becomes nonnull if we encounter nested concatenations.

namespace mlir::quant::stablehlo {

namespace {

using ::mlir::quant::FindUserOfType;
using ::mlir::quant::TryCast;
using ::mlir::stablehlo::AddOp;
using ::mlir::stablehlo::BroadcastInDimOp;
using ::mlir::stablehlo::ConcatenateOp;
using ::mlir::stablehlo::ConvolutionOp;
using ::mlir::stablehlo::DotGeneralOp;
using ::mlir::stablehlo::DynamicBroadcastInDimOp;
using ::mlir::stablehlo::GatherOp;
using ::mlir::stablehlo::GetDimensionSizeOp;