MOVW	R5, R13
	MOVW	R6, R14
	MOVW	R7, R15

	MOVW	(0*4)(R1), R8
	MOVW	R7, R9

#define ROUND1(a, b, c, d, index, const, shift) \
	ADDW	$const, a; \
	ADDW	R8, a; \
	MOVW	(index*4)(R1), R8; \
	EORW	c, R9; \
	ANDW	b, R9; \
	EORW	d, R9; \
	ADDW	R9, a; \
	RORW	$(32-shift), a; \
	MOVW	c, R9; \
	ADDW	b, a

	ROUND1(R4,R5,R6,R7, 1,0xd76aa478, 7);
	ROUND1(R7,R4,R5,R6, 2,0xe8c7b756,12);

if($ARGV[0] eq "-b32") {
	$_32bit = "big-endian";
	shift;
} elsif($ARGV[0] eq "-l32") {
	$_32bit = "little-endian";
	shift;
}
if($ARGV[0] eq "-plan9") {
	$plan9 = 1;
	shift;
}
if($ARGV[0] eq "-darwin") {
	$darwin = 1;
	$libc = 1;
	shift;
}
if($ARGV[0] eq "-openbsd") {
	$openbsd = 1;
	shift;
}
if($ARGV[0] eq "-netbsd") {
	$netbsd = 1;
	shift;
}
if($ARGV[0] eq "-dragonfly") {
	$dragonfly = 1;

	case isPowerOfTwo(et.Size_):
		var shift uintptr
		if goarch.PtrSize == 8 {
			// Mask shift for better code generation.
			shift = uintptr(sys.TrailingZeros64(uint64(et.Size_))) & 63
		} else {
			shift = uintptr(sys.TrailingZeros32(uint32(et.Size_))) & 31
		}
		lenmem = uintptr(oldLen) << shift
		newlenmem = uintptr(newLen) << shift
		capmem = roundupsize(uintptr(newcap)<<shift, noscan)

#define ROUND1(a, b, c, d, index, const, shift) \
	ADD	$const, index, R9; \
	ADD	R9, a; \
	AND     b, c, R9; \
	ANDN    b, d, R31; \
	OR	R9, R31, R9; \
	ADD	R9, a; \
	ROTLW	$shift, a; \
	ADD	b, a;

#define ROUND2(a, b, c, d, index, const, shift) \
	ADD	$const, index, R9; \
	ADD	R9, a; \
	AND	b, d, R31; \
	ANDN	d, c, R9; \
	OR	R9, R31; \
	ADD	R31, a; \
	ROTLW	$shift, a; \
	ADD	b, a;

		{name: "SLLIW", argLength: 1, reg: gp11, asm: "SLLIW", aux: "Int64"}, // arg0 << auxint, shift amount 0-31, logical left shift of 32 bit value, sign extended to 64 bits
		{name: "SRAI", argLength: 1, reg: gp11, asm: "SRAI", aux: "Int64"},   // arg0 >> auxint, shift amount 0-63, arithmetic right shift

      sqrt0 = sqrtApproxWithDoubles(x);
    } else {
      int shift = (log2 - DoubleUtils.SIGNIFICAND_BITS) & ~1; // even!
      /*
       * We have that x / 2^shift < 2^54. Our initial approximation to sqrtFloor(x) will be
       * 2^(shift/2) * sqrtApproxWithDoubles(x / 2^shift).
       */
      sqrt0 = sqrtApproxWithDoubles(x.shiftRight(shift)).shiftLeft(shift >> 1);
    }

(Max(64|32)F <t> x y) => (Neg(64|32)F <t> (Min(64|32)F <t> (Neg(64|32)F <t> x) (Neg(64|32)F <t> y)))

(CvtBoolToUint8 ...) => (Copy ...)

// Lowering shifts
// Unsigned shifts need to return 0 if shift amount is >= width of shifted value.
//   result = (arg << shift) & (shift >= argbits ? 0 : 0xffffffffffffffff)
(Lsh64x(64|32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDQ (SHLQ <t> x y) (SBBQcarrymask <t> (CMP(Q|L|W|B)const y [64])))

      sqrt0 = sqrtApproxWithDoubles(x);
    } else {
      int shift = (log2 - DoubleUtils.SIGNIFICAND_BITS) & ~1; // even!
      /*
       * We have that x / 2^shift < 2^54. Our initial approximation to sqrtFloor(x) will be
       * 2^(shift/2) * sqrtApproxWithDoubles(x / 2^shift).
       */
      sqrt0 = sqrtApproxWithDoubles(x.shiftRight(shift)).shiftLeft(shift >> 1);
    }

	if mant == 0 {
		exp = 0
	}

	// Shift digits so leading 1 (if any) is at bit 1<<60.
	mant <<= 60 - flt.mantbits
	for mant != 0 && mant&(1<<60) == 0 {
		mant <<= 1
		exp--
	}

	// Round if requested.
	if prec >= 0 && prec < 15 {
		shift := uint(prec * 4)
		extra := (mant << shift) & (1<<60 - 1)
		mant >>= 60 - shift
		if extra|(mant&1) > 1<<59 {
			mant++
		}

(SRLI <t> [x] (MOVWUreg y)) && x >= 0 && x <= 31 => (SRLIW <t> [int64(x)] y)

// Replace right shifts that exceed size of signed type.
(SRAI <t> [x] (MOVBreg y)) && x >=  8 => (SRAI  [63] (SLLI <t> [56] y))
(SRAI <t> [x] (MOVHreg y)) && x >= 16 => (SRAI  [63] (SLLI <t> [48] y))
(SRAI <t> [x] (MOVWreg y)) && x >= 32 => (SRAIW [31] y)

// Eliminate right shifts that exceed size of unsigned type.
(SRLI <t> [x] (MOVBUreg y)) && x >=  8 => (MOVDconst <t> [0])