include "mlir/IR/PatternBase.td"
include "mlir/Dialect/Arith/IR/ArithOps.td"
include "tensorflow/compiler/mlir/lite/ir/tfl_ops.td"
include "tensorflow/compiler/mlir/lite/utils/utils.td"

// Returns Squeezed shape of a ranked-tensor.
// Squeezed, here, means eliminating any 1s' in the
// dimensions of the tensor
def GetSqueezedShape: NativeCodeCall<"GetSqueezedShape($0)">;

	for len(out) > 0 {
		n := copy(out, d.storage[d.i:d.n])
		d.i += n
		out = out[n:]

		// Apply the permutation if we've squeezed the sponge dry.
		if d.i == d.rate {
			d.permute()
		}
	}

	return
}

// Sum applies padding to the hash state and then squeezes out the desired
// number of output bytes. It panics if any output has already been read.
func (d *state) Sum(in []byte) []byte {

// is then "full" and the permutation is applied to "empty" it. This process is
// repeated until all the input has been "absorbed". The input is then padded.
// The digest is "squeezed" from the sponge in the same way, except that output
// is copied out instead of input being XORed in.
//
// A sponge is parameterized by its generic security strength, which is equal

  TF::ExpandDimsOp expand_op;
  TF::SqueezeOp squeeze_op;
  int64_t expand_axis = -1;
  // Expand + Squeeze op.
  if (llvm::isa<TF::ExpandDimsOp>(producer_op)) {
    if (!llvm::isa<TF::SqueezeOp>(consumer_op)) {
      // Expand/Squeeze op must come in pair.
      return rewriter.notifyMatchFailure(
          op, "ExpandDimsOp and SqueezeOp should come in pair");
    }

// TFL_TransposeOp when the tensor has some dimensions with value==1
// Example- "tfl.transpose"(tensor<56x8x56x1x1x1x7xf32>, [4, 5, 1, 2, 0, 6, 3])
// Permutation before squeese is [4, 5, 1, 2, 0, 6, 3] becomes [1, 2, 0, 3]
// after squeeze is perfomed to retain the relative ordering of the non-1 dims.
DenseElementsAttr GetSqueezedPermutation(Value input_value,
                                         Value input_permutation) {

  // CHECK-NEXT: [[SQUEEZE:%.*]] = "tf.Squeeze"([[CONV]]) <{squeeze_dims = [3]}> : (tensor<1x128x128x1xf32>) -> tensor<1x128x128xf32>
  // CHECK-NEXT: [[RESULT:%.*]] = "tf.BiasAdd"([[SQUEEZE]], [[BIAS]]) : (tensor<1x128x128xf32>, tensor<128xf32>) -> tensor<1x128x128xf32>
  // CHECK-NEXT: return [[RESULT]] : tensor<1x128x128xf32>
}

 Mean 1
 Mul 164
 Pad 1
 ReadVariableOp 646
 Relu 49
 ReluGrad 49
 Reshape 2
 ResourceApplyKerasMomentum 161
 ShapeN 50
 Softmax 1
 SparseSoftmaxCrossEntropyWithLogits 1
 Square 55
 Squeeze 1
 Sub 106
 Sum 57
 Tile 1

 MaxPoolGrad 1
 Mean 1
 Mul 218
 Pad 2
 ReadVariableOp 538
 Relu 49
 ReluGrad 49
 Reshape 2
 ResourceApplyKerasMomentum 161
 Slice 1
 Softmax 1
 SparseSoftmaxCrossEntropyWithLogits 1
 Squeeze 1
 Sum 1
 Tile 1
 Transpose 1
cluster 1 size 815
 AddN 1
 AssignAddVariableOp 1
 AssignSubVariableOp 106
 Const 220
 DivNoNan 1
 Identity 1
 Mul 161
 ReadVariableOp 106
 Square 55
 Sub 106

    "Shape",
    "ShapeN",
    "Size",
    "Slice",
    "Snapshot",
    "SpaceToBatch",
    "SpaceToBatchND",
    "SpaceToDepth",
    "Split",
    "SplitV",
    "Squeeze",
    "StopGradient",
    "StridedSlice",
    "StridedSliceGrad",
    "TensorScatterAdd",
    "TensorScatterMax",
    "TensorScatterMin",
    "TensorScatterSub",
    "TensorStridedSliceUpdate",

is, but I shall remember it in a bit.'

  `Perhaps it hasn't one,' Alice ventured to remark.

  `Tut, tut, child!' said the Duchess.  `Everything's got a
moral, if only you can find it.'  And she squeezed herself up
closer to Alice's side as she spoke.

  Alice did not much like keeping so close to her:  first,
because the Duchess was VERY ugly; and secondly, because she was