* any of the public methods of ListenableFuture.
   */
  interface Trusted<V extends @Nullable Object> extends ListenableFuture<V> {}

  /**
   * A less abstract subclass of AbstractFuture. This can be used to optimize setFuture by ensuring
   * that {@link #get} calls exactly the implementation of {@link AbstractFuture#get}.
   */
  abstract static class TrustedFuture<V extends @Nullable Object> extends AbstractFuture<V>

   * any of the public methods of ListenableFuture.
   */
  interface Trusted<V extends @Nullable Object> extends ListenableFuture<V> {}

  /**
   * A less abstract subclass of AbstractFuture. This can be used to optimize setFuture by ensuring
   * that {@link #get} calls exactly the implementation of {@link AbstractFuture#get}.
   */
  abstract static class TrustedFuture<V extends @Nullable Object> extends AbstractFuture<V>

// Lowering pointer arithmetic
(OffPtr [off] ptr:(SP)) => (MOVDaddr [int32(off)] ptr)
(OffPtr [off] ptr) && is32Bit(off) => (ADDconst [int32(off)] ptr)
(OffPtr [off] ptr) => (ADD (MOVDconst [off]) ptr)

// TODO: optimize these cases?
(Ctz64NonZero ...) => (Ctz64 ...)
(Ctz32NonZero ...) => (Ctz32 ...)

// Ctz(x) = 64 - findLeftmostOne((x-1)&^x)
(Ctz64 <t> x) => (SUB (MOVDconst [64]) (FLOGR (AND <t> (SUBconst <t> [1] x) (NOT <t> x))))

		//
		// We use a local static variable to get the zeroed
		// value of the argument type. This avoids including
		// string.h for memset, and is also robust to C++
		// types with constructors. Both GCC and LLVM optimize
		// this into just zeroing _cgo_a.
		fmt.Fprintf(fgcc, "\ttypedef %s %v _cgo_argtype;\n", ctype, p.packedAttribute())
		fmt.Fprintf(fgcc, "\tstatic _cgo_argtype _cgo_zero;\n")

(MOVDaddr {sym} [n] p:(ADD x y)) && sym == nil && n == 0 => p
(MOVDaddr {sym} [n] ptr) && sym == nil && n == 0 && (ptr.Op == OpArgIntReg || ptr.Op == OpPhi) => ptr

// TODO: optimize these cases?
(Ctz32NonZero ...) => (Ctz32 ...)
(Ctz64NonZero ...) => (Ctz64 ...)

(Ctz64 x) && buildcfg.GOPPC64<=8 => (POPCNTD (ANDN <typ.Int64> (ADDconst <typ.Int64> [-1] x) x))
(Ctz64 x) => (CNTTZD x)

  /**
   * Maximum number of entries to be drained in a single cleanup run. This applies independently to
   * the cleanup queue and both reference queues.
   */
  // TODO(fry): empirically optimize this
  static final int DRAIN_MAX = 16;

  // Fields

  /**
   * Mask value for indexing into segments. The upper bits of a key's hash code are used to choose
   * the segment.
   */

  /**
   * Maximum number of entries to be drained in a single cleanup run. This applies independently to
   * the cleanup queue and both reference queues.
   */
  // TODO(fry): empirically optimize this
  static final int DRAIN_MAX = 16;

  // Fields

  /**
   * Mask value for indexing into segments. The upper bits of a key's hash code are used to choose
   * the segment.
   */

// unary ops
(Neg(32|16|8) x) => (RSBconst [0] x)
(Neg(32|64)F ...) => (NEG(F|D) ...)

(Com(32|16|8) ...) => (MVN ...)

(Sqrt ...) => (SQRTD ...)
(Sqrt32 ...) => (SQRTF ...)
(Abs ...) => (ABSD ...)

// TODO: optimize this for ARMv5 and ARMv6
(Ctz32NonZero ...) => (Ctz32 ...)
(Ctz16NonZero ...) => (Ctz32 ...)
(Ctz8NonZero ...) => (Ctz32 ...)

// count trailing zero for ARMv5 and ARMv6
// 32 - CLZ(x&-x - 1)

(Select1 (Sub64borrow x y c)) =>
	(NEGQ <typ.UInt64> (SBBQcarrymask <typ.UInt64> (Select1 <types.TypeFlags> (SBBQ x y (Select1 <types.TypeFlags> (NEGLflags c))))))

// Optimize ADCQ and friends
(ADCQ x (MOVQconst [c]) carry) && is32Bit(c) => (ADCQconst x [int32(c)] carry)
(ADCQ x y (FlagEQ)) => (ADDQcarry x y)
(ADCQconst x [c] (FlagEQ)) => (ADDQconstcarry x [c])

   *
   * The alternative natural refactorings into void or Exception-returning methods are much slower.
   * This is a big deal - we're talking factors of 2-8 in microbenchmarks, not just 10-20%. (This is
   * a hotspot optimizer bug, which should be fixed, but that's a separate, big project).
   *
   * The coding pattern above is heavily used in java.util, e.g. in ArrayList. There is a