[(IsNotTFStyleBroadcast $broadcast_dimensions, $output)]>;


def : Pat<(MHLO_TransposeOp $arg, $permutation),
          (TF_TransposeOp $arg, (TF_ConstOp $permutation))>;
def : Pat<(MHLO_ReverseOp $op, $dims), (TF_ReverseV2Op $op, (TF_ConstOp $dims))>;
def : Pat<(MHLO_ReshapeOp:$output $input),
          (TF_ReshapeOp $input, (ShapeToConst $output))>;

// into 26-bit big-endian limbs and places the results into
// the first and second lane of d₂₆[0:4] respectively.
//
// The EX0, EX1 and EX2 constants are arrays of byte indices
// for permutation. The permutation both reverses the bytes
// in the input and ensures the bytes are copied into the
// destination limb ready to be shifted into their final
// position.
#define EXPAND(in0, in1, d0, d1, d2, d3, d4) \

TF::TransposeOp GetTransposeOpForConv2DFilter(OpBuilder* builder, Value input) {
  SmallVector<int32_t, 6> permutation = {0, 2, 1, 3, 4, 5};
  auto permute_type = RankedTensorType::get({6}, builder->getIntegerType(32));
  auto permute_attr = DenseIntElementsAttr::get(permute_type, permutation);
  auto permute_value =
      builder->create<TF::ConstOp>(input.getLoc(), permute_attr);

    // In practice, wildcards are always in the leftmost position. For now, this implementation
    // cheats and does not attempt every possible permutation. Instead, it only considers wildcards
    // in the leftmost position. We assert this fact when we generate the public suffix file. If
    // this assertion ever fails we'll need to refactor this implementation.

// CHECK-LABEL: main
// CHECK: return %arg0 : tensor<5x7xf32>
}

// - transpose
//
func.func @transpose_2d(%arg0: tensor<2x3xf32>) -> tensor<3x2xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[1, 0]> : tensor<2xi64>}> : (tensor<2x3xf32>) -> tensor<3x2xf32>
  func.return %0 : tensor<3x2xf32>

// CHECK-LABEL:   transpose_2d

Value Transpose(OpBuilder* builder, Value value_to_transpose,
                SmallVector<int32_t, 4> perm, RankedTensorType original_type,
                mlir::Location location) {
  // Create a constant op for transpose permutation.
  auto perm_op = CreateI32DenseConst(builder, perm, location);

  // Create tensor type for the transpose result.
  auto transpose_type = original_type;
  auto transpose_shape =

//
// Example: compilation using XLA (MHLO) lowering
//
//   %0 = "tf.Transpose"(%input, %perm)
//        : (tensor<?x?xf32>, tensor<2xi32>) -> tensor<?x?xf32>
//
//   XLAs `mhlo.transpose` operation requires permutation to be an attribute
//   (compile time value), so it means that if we want to put `tf.Transpose`
//   into a cluster that will be compiled with XLA, the `%perm` operand must

	// Convert the user state into an internal state
	NOTQ _be(rpState)
	NOTQ _bi(rpState)
	NOTQ _go(rpState)
	NOTQ _ki(rpState)
	NOTQ _mi(rpState)
	NOTQ _sa(rpState)

	// Execute the KeccakF permutation
	MOVQ _ba(rpState), rCa
	MOVQ _be(rpState), rCe
	MOVQ _bu(rpState), rCu

	XORQ _ga(rpState), rCa
	XORQ _ge(rpState), rCe
	XORQ _gu(rpState), rCu

	XORQ _ka(rpState), rCa
	XORQ _ke(rpState), rCe

  %0 = "vhlo.transpose_v1"(%arg0) <{permutation = #vhlo.tensor_v1<dense<[2, 1, 0]> : tensor<3xi64>>}> : (tensor<2x3x2xi32>) -> tensor<2x3x2xi32>
  return %0 : tensor<2x3x2xi32>
}

// CHECK:func.func private @transpose(%arg0: tensor<2x3x2xi32>) -> tensor<2x3x2xi32> {

struct FoldTransposeOp : public OpRewritePattern<TransposeOp> {
  explicit FoldTransposeOp(MLIRContext* context)
      : OpRewritePattern<TransposeOp>(context, 1) {}

  // Computes the permutation of a constant `input_tensor` according to `perm`.
  // The function recursively traverses the dimensions of the output tensor in
  // a row-major order and writes the value in the output tensor into
  // `new_values`.