// Folding will infer that: Pow(%mul, 1.0) -> %mul
    // However we don't have the actual value for the mul, but we can use the
    // mul type!
    // CHECK: tf.Pow
    // CHECK-SAME: -> tensor<f32>
    %pow = "tf.Pow"(%mul, %cst1) : (tensor<f32>, tensor<f32>) -> tensor<*xf32>
    func.return %pow : tensor<*xf32>
  }

  // CHECK:  return
}

func.func @pow(%arg0: tensor<2x1x3xf32>, %arg1: tensor<2x1x1xf32>) -> tensor<2x1x3xf32> {
  %0 = "tf.Pow"(%arg0, %arg1) : (tensor<2x1x3xf32>, tensor<2x1x1xf32>) -> tensor<2x1x3xf32>
  func.return %0 : tensor<2x1x3xf32>

  // CHECK-LABEL: pow
  // CHECK:  %[[pow:.*]] = tfl.pow(%arg0, %arg1) : (tensor<2x1x3xf32>, tensor<2x1x1xf32>) -> tensor<2x1x3xf32>
  // CHECK:  return

// and then does an indirect call to that value.
//
// Here is what a direct reference might look like:
//
//	     137: e9 20 06 00 00               	jmp	0x75c <pow+0x75c>
//	     13c: e8 00 00 00 00               	callq	0x141 <pow+0x141>
//			000000000000013d:  IMAGE_REL_AMD64_REL32	_errno
//
// The assembly below dispenses with the import symbol and just makes
// a direct call to _errno.
//

}

// CHECK-LABEL: testPow
func.func @testPow(tensor<? x i32>, tensor<? x i32>) -> tensor<? x i32> {
^bb0(%arg0: tensor<? x i32>, %arg1: tensor<? x i32>):
  // CHECK: tfl.pow %arg0, %arg1
  %0 = tfl.pow %arg0, %arg1 : tensor<? x i32>
  func.return %0#0 : tensor<? x i32>
}

// CHECK-LABEL: testAtan2
func.func @testAtan2(%arg0: tensor<?xf32>, %arg1: tensor<?xf32>) -> tensor<?xf32> {

      // CHECK: tf.RandomStandardNormal{{.*}}(%[[SHAPE]])
      %9 = "tf.RandomStandardNormal"(%arg4) {device = "", seed = 87654321 : i64, seed2 = 0 : i64} : (tensor<1xi32>) -> tensor<?xf32>
      %10 = "tf.Pow"(%9, %5) {device = ""} : (tensor<?xf32>, tensor<f32>) -> tensor<?xf32>
      %11 = "tf.AddV2"(%arg3, %10) {device = ""} : (tensor<?xf32>, tensor<?xf32>) -> tensor<?xf32>

  let results = (outs Variadic<AnyTensor>:$output);

  let regions = (region VariadicRegion<SizedRegion<1>>:$calls);

  let hasCanonicalizer = 1;
}


def TFL_PowOp : TFL_Op<"pow", [
    ResultsBroadcastableShape,
    Pure,
    TFL_OperandsHaveSameShapesOrBroadcastableShape<[0, 1], 4>]> {
  let summary = "Power operator";

  let description = [{
    Element-wise power operation.