* is precisely representable as a {@code double}, its {@code double} value will be returned;
   * otherwise, the rounding will choose between the two nearest representable values with {@code
   * mode}.
   *
   * <p>For the case of {@link RoundingMode#HALF_EVEN}, this implementation uses the IEEE 754

  public void testLog2HalfEven() {
    for (BigInteger x : POSITIVE_BIGINTEGER_CANDIDATES) {
      int halfEven = BigIntegerMath.log2(x, HALF_EVEN);
      // Now figure out what rounding mode we should behave like (it depends if FLOOR was
      // odd/even).
      boolean floorWasEven = (BigIntegerMath.log2(x, FLOOR) & 1) == 0;

  /**
   * Starts the stopwatch.
   *
   * @return this {@code Stopwatch} instance
   * @throws IllegalStateException if the stopwatch is already running.
   */
  @CanIgnoreReturnValue
  public Stopwatch start() {
    checkState(!isRunning, "This stopwatch is already running.");
    isRunning = true;
    startTick = ticker.read();
    return this;
  }

  /**

        boolean isInBounds =
            expected.compareTo(MAX_INT_AS_BIG_DECIMAL) <= 0
                & expected.compareTo(MIN_INT_AS_BIG_DECIMAL) >= 0;

        try {
          assertEquals(
              "Rounding " + d + " with mode " + mode,
              expected.intValue(),
              DoubleMath.roundToInt(d, mode));
          assertTrue(isInBounds);
        } catch (ArithmeticException e) {

    assertThrows(IllegalArgumentException.class, () -> RateLimiter.create(1.0, -1, NANOSECONDS));
  }

  @AndroidIncompatible // difference in String.format rounding?
  public void testWarmUp() {
    RateLimiter limiter = RateLimiter.create(2.0, 4000, MILLISECONDS, 3.0, stopwatch);
    for (int i = 0; i < 8; i++) {
      limiter.acquire(); // #1
    }

    assertThrows(IllegalArgumentException.class, () -> RateLimiter.create(1.0, -1, NANOSECONDS));
  }

  @AndroidIncompatible // difference in String.format rounding?
  public void testWarmUp() {
    RateLimiter limiter = RateLimiter.create(2.0, 4000, MILLISECONDS, 3.0, stopwatch);
    for (int i = 0; i < 8; i++) {
      limiter.acquire(); // #1
    }

        boolean isInBounds =
            expected.compareTo(MAX_INT_AS_BIG_DECIMAL) <= 0
                & expected.compareTo(MIN_INT_AS_BIG_DECIMAL) >= 0;

        try {
          assertEquals(
              "Rounding " + d + " with mode " + mode,
              expected.intValue(),
              DoubleMath.roundToInt(d, mode));
          assertTrue(isInBounds);
        } catch (ArithmeticException e) {

fused operation, possibly across statements, and produce a result that differs
from the value obtained by executing and rounding the instructions individually.
An explicit <a href="#Numeric_types">floating-point type</a> <a href="#Conversions">conversion</a> rounds to
the precision of the target type, preventing fusion that would discard that rounding.
</p>

<p>
For instance, some architectures provide a "fused multiply and add" (FMA) instruction

			if len(operands) == 0 && len(items) == 0 {
				if p.arch.InFamily(sys.ARM, sys.ARM64, sys.AMD64, sys.I386, sys.Loong64, sys.RISCV64) && tok == '.' {
					// Suffixes: ARM conditionals, Loong64 vector instructions, RISCV rounding mode or x86 modifiers.
					tok = p.nextToken()
					str := p.lex.Text()
					if tok != scanner.Ident {
						p.errorf("instruction suffix expected identifier, found %s", str)
					}
					cond = cond + "." + str

        model is called for the first time.
*   `tf.linalg`:
    *   Add `CompositeTensor` as a base class to `LinearOperator`.
*   `tf.lite`:
    *   Fix mean op reference quantization rounding issue.
    *   Added `framework_stable` BUILD target, which links in only the
        non-experimental TF Lite APIs.
    *   Remove deprecated Java `Interpreter` methods: