x0, x1 = x1, x0
	}
	if y0 > y1 {
		y0, y1 = y1, y0
	}
	return Rectangle{Point{x0, y0}, Point{x1, y1}}
}

// mul3NonNeg returns (x * y * z), unless at least one argument is negative or
// if the computation overflows the int type, in which case it returns -1.
func mul3NonNeg(x int, y int, z int) int {
	if (x < 0) || (y < 0) || (z < 0) {
		return -1
	}
	hi, lo := bits.Mul64(uint64(x), uint64(y))
	if hi != 0 {

	// 		Let F ← log₂(d) and c = 1.
	// 	else
	// 		Let F ← N + L where L is the smallest integer
	// 		such that d ≤ (2^(N+L) mod d) + 2^L.
	// 	end if
	//
	// [1] "Faster Remainder by Direct Computation: Applications to
	// Compilers and Software Libraries" Daniel Lemire, Owen Kaser,
	// Nathan Kurz arXiv:1902.01961
	//
	// To minimize the risk of introducing errors, we implement the

// in the GraphDef.
// Currently, it is used by eager runtime. FunctionLibraryRuntime creates
// this kernel when asked to create a kernel for an XLA-compiled function.
//
// `has_ref_vars`: whether the input computation can have reference variables.
// TODO(cheshire): instead derive this information from the input graph.
class XlaLocalLaunchBase : public AsyncOpKernel {
 public:
  XlaLocalLaunchBase(OpKernelConstruction* ctx,

//   1. Normalization
//      Scale x to y in [1,4) with even powers of 2:
//      find an integer k such that  1 <= (y=x*2**(2k)) < 4, then
//              sqrt(x) = 2**k * sqrt(y)
//   2. Bit by bit computation
//      Let q  = sqrt(y) truncated to i bit after binary point (q = 1),
//           i                                                   0
//                                     i+1         2

      for (int i = 0; i < c->num_outputs(); ++i) {
        c->set_output(i, c->input(0));
      }
      return absl::OkStatus();
    })
    .Doc(
        "Operator that connects the output of an XLA computation to other "
        "consumer graph nodes.");

REGISTER_OP("_XlaCompile")
    .Input("constants: Tconstants")
    .Attr("Tconstants: list(type) >= 0")
    .Attr("must_compile: bool")
    .Input("args: Targs")

// Replaces aten.avg_pool2d with ceil mode with (T -> tfl.pad -> tfl.average_pool_2d -> mul -> T). Multiplies by a constant
// which corrects the overcounting of divisors that would occur if doing this computation on a padded tensor with ceil mode off.
def LegalizeAvgPoolCeilModeTrue: Pat<
                    (MHLO_CompositeOp:$old_val
                    (variadic $a_input), 

// the quantized inputs, performing the operation in the expressed type, then
// requantizing if a quantized output is required.
//
// The motivation behind these changes is for Dialects that assume only float
// or quantized computation, and do not support a mixture of these types on
// dense operations. Decomposition allows TFLite to be compiled to these
// dialects, such as TOSA.

#include <utility>

 *
 * @author Pete Gillin
 */
public class QuantilesTest extends TestCase {

  /*
   * Since Quantiles provides a fluent-style API, each test covers a chain of methods resulting in
   * the computation of one or more quantiles (or in an error) rather than individual methods. The
   * tests are divided into three sections:
   * 1. Tests on a hardcoded dataset for chains starting with median(), quartiles(), and scale(10);

// license that can be found in the LICENSE file.

#include "textflag.h"

// The method is based on a paper by Naoki Shibata: "Efficient evaluation
// methods of elementary functions suitable for SIMD computation", Proc.
// of International Supercomputing Conference 2010 (ISC'10), pp. 25 -- 32
// (May 2010). The paper is available at
// https://link.springer.com/article/10.1007/s00450-010-0108-2
//

   * computation is {@linkplain java.util.concurrent.Future#isDone() complete} or, if the
   * computation is already complete, immediately.
   *
   * <p>The callback is run on {@code executor}. There is no guaranteed ordering of execution of
   * callbacks, but any callback added through this method is guaranteed to be called once the
   * computation is complete.
   *
   * <p>Example:
   *