// [-512, 512], instead of [-32767, 32767].
    // For now this behavior is specific for SVDF, where 6 bits are reserved for
    // the reduce operation after element-wise multiplication between state and
    // time weights.
    if (tensor_property.number_of_bits == 10) {
      SmallVector<double, 4> mins(1, std::numeric_limits<double>::max());

    assertEquals(1, future.get(-1, SECONDS).intValue());
  }

  @J2ktIncompatible
  @GwtIncompatible // threads
  public void testOverflowTimeout() throws Exception {
    // First, sanity check that naive multiplication would really overflow to a negative number:
    long nanosPerSecond = NANOSECONDS.convert(1, SECONDS);
    assertThat(nanosPerSecond * Long.MAX_VALUE).isLessThan(0L);

  }

  const int64_t rows = lhs_shape[lhs_dims - 2];
  const int64_t cols = rhs_shape[rhs_dims - 1];

  if (lhs_shape[lhs_dims - 1] != rhs_shape[rhs_dims - 2]) {
    // Input dimensions must be compatible for multiplication.
    return failure();
  }

  const auto matmul_type = RankedTensorType::get({rows, cols}, element_type);

  if (lhs_dims == 2 && rhs_dims == 2) {

    assertEquals(1, future.get(-1, SECONDS).intValue());
  }

  @J2ktIncompatible
  @GwtIncompatible // threads
  public void testOverflowTimeout() throws Exception {
    // First, sanity check that naive multiplication would really overflow to a negative number:
    long nanosPerSecond = NANOSECONDS.convert(1, SECONDS);
    assertThat(nanosPerSecond * Long.MAX_VALUE).isLessThan(0L);

// The following assembly functions are implemented in p256_asm_*.s

// Montgomery multiplication. Sets res = in1 * in2 * R⁻¹ mod p.
//
//go:noescape
func p256Mul(res, in1, in2 *p256Element)

// Montgomery square, repeated n times (n >= 1).
//
//go:noescape
func p256Sqr(res, in *p256Element, n int)

// Montgomery multiplication by R⁻¹, or 1 outside the domain.

	//	R = Y' + 1.40200*(Cr-128)
	//	G = Y' - 0.34414*(Cb-128) - 0.71414*(Cr-128)
	//	B = Y' + 1.77200*(Cb-128)
	// https://www.w3.org/Graphics/JPEG/jfif3.pdf says Y but means Y'.
	//
	// Those formulae use non-integer multiplication factors. When computing,
	// integer math is generally faster than floating point math. We multiply
	// all of those factors by 1<<16 and round to the nearest integer:
	//	 91881 = roundToNearestInteger(1.40200 * 65536).

	AADDIW
	ASLLIW
	ASRLIW
	ASRAIW
	AADDW
	ASLLW
	ASRLW
	ASUBW
	ASRAW

	// 5.3: Load and Store Instructions (RV64I)
	ALD
	ASD

	// 7.1: Multiplication Operations
	AMUL
	AMULH
	AMULHU
	AMULHSU
	AMULW
	ADIV
	ADIVU
	AREM
	AREMU
	ADIVW
	ADIVUW
	AREMW
	AREMUW

	// 8.2: Load-Reserved/Store-Conditional Instructions

	//
	// We want to compute
	// 	hi, lo := bits.Mul64(r.Uint64(), n)
	// In terms of 32-bit halves, this is:
	// 	x1:x0 := r.Uint64()
	// 	0:hi, lo1:lo0 := bits.Mul64(x1:x0, 0:n)
	// Writing out the multiplication in terms of bits.Mul32 allows
	// using direct hardware instructions and avoiding
	// the computations involving these zeros.
	x := r.Uint64()
	lo1a, lo0 := bits.Mul32(uint32(x), n)

	SRAW	$1, X6					// 1b531340

	// 5.3: Load and Store Instructions (RV64I)
	LD	(X5), X6				// 03b30200
	LD	4(X5), X6				// 03b34200
	SD	X5, (X6)				// 23305300
	SD	X5, 4(X6)				// 23325300

	// 7.1: Multiplication Operations
	MUL	X5, X6, X7				// b3035302
	MULH	X5, X6, X7				// b3135302
	MULHU	X5, X6, X7				// b3335302
	MULHSU	X5, X6, X7				// b3235302
	MULW	X5, X6, X7				// bb035302
	DIV	X5, X6, X7				// b3435302

    // Strip off 2s from this value.
    int shift = Long.numberOfTrailingZeros(product);
    product >>= shift;

    // Use floor(log2(num)) + 1 to prevent overflow of multiplication.
    int productBits = LongMath.log2(product, FLOOR) + 1;
    int bits = LongMath.log2(startingNumber, FLOOR) + 1;
    // Check for the next power of two boundary, to save us a CLZ operation.