"STXVW4X",
	"STXVH8X",
	"STXVB16X",
	"STXVX",
	"LXSDX",
	"STXSDX",
	"LXSIWAX",
	"LXSIWZX",
	"STXSIWX",
	"MFVSRD",
	"MFFPRD",
	"MFVRD",
	"MFVSRWZ",
	"MFVSRLD",
	"MTVSRD",
	"MTFPRD",
	"MTVRD",
	"MTVSRWA",
	"MTVSRWZ",
	"MTVSRDD",
	"MTVSRWS",
	"XXLAND",
	"XXLANDC",
	"XXLEQV",
	"XXLNAND",
	"XXLOR",
	"XXLORC",
	"XXLNOR",
	"XXLORQ",
	"XXLXOR",

	VSEL     V14, V13, V31, V7  // final merge
	VCLZD    V7, V18            // Find first index for each half
	MFVSRD   V18, R25           // Isolate value
	CMP      R25, $64           // If < 64, found
	BLT      foundR25           // Return found index
	VSLDOI   $8, V18, V18, V18  // Move for MFVSRD
	MFVSRD   V18, R25           // Isolate other value
	CMP      R25, $64           // If < 64, found

	STXSIWX VS1, (R3)               // 7c201918
	MFVSRD VS1, R3                  // 7c230066
	MTFPRD R3, F0                   // 7c030166
	MFVRD V0, R3                    // 7c030067
	MFVSRLD VS63,R4                 // 7fe40267
	MFVSRLD V31,R4                  // 7fe40267
	MFVSRWZ VS33,R4                 // 7c2400e7
	MFVSRWZ V1,R4                   // 7c2400e7
	MTVSRD R3, VS1                  // 7c230166

	LXSIWAX:        "lxsiwax",
	LXSIWZX:        "lxsiwzx",
	LXSSPX:         "lxsspx",
	MFBHRBE:        "mfbhrbe",
	MFVSRD:         "mfvsrd",
	MFVSRWZ:        "mfvsrwz",
	MSGCLR:         "msgclr",
	MSGCLRP:        "msgclrp",
	MSGSND:         "msgsnd",
	MSGSNDP:        "msgsndp",
	MTVSRD:         "mtvsrd",
	MTVSRWA:        "mtvsrwa",
	MTVSRWZ:        "mtvsrwz",
	RFEBB:          "rfebb",
	STQCXCC:        "stqcx.",

			opset(ALXSIWZX, r0)

		case ASTXSIWX: /* stxsiwx */
			opset(ASTXSIWX, r0)

		case AMFVSRD: /* mfvsrd, mfvsrwz (and extended mnemonics), mfvsrld */
			opset(AMFFPRD, r0)
			opset(AMFVRD, r0)
			opset(AMFVSRWZ, r0)
			opset(AMFVSRLD, r0)

		case AMTVSRD: /* mtvsrd, mtvsrwa, mtvsrwz (and extended mnemonics), mtvsrdd, mtvsrws */
			opset(AMTFPRD, r0)
			opset(AMTVRD, r0)
			opset(AMTVSRWA, r0)

	VADDUDM	V0, V4, V4	// Accumulate the popcounts. They are 8x the count.
	VADDUDM	V2, V5, V5	// The count will be fixed up afterwards.
	ADD	$32, R3
	BDNZ	cmploop

	VADDUDM	V4, V5, V5
	MFVSRD	V5, R18
	VSLDOI	$8, V5, V5, V5
	MFVSRD	V5, R21
	ADD	R21, R18, R18
	ANDCC	$31, R4, R4
	// Skip the tail processing if no bytes remaining.
	BEQ	tail_0

#ifdef GOPPC64_power10

	VPERM	V3,V3,SWAP,V3
	VPERM	V4,V4,SWAP,V4
#endif

	MFVSRD	VS35,R16	// move upper doublewords of A and B into GPR for comparison
	MFVSRD	VS36,R10

	CMPU	R16,R10
	BEQ	lower
	SETB_CR0_NE(R3)
	RET

	PCALIGN $16
lower:
	VSLDOI	$8,V3,V3,V3	// move lower doublewords of A and B into GPR for comparison
	MFVSRD	VS35,R16
	VSLDOI	$8,V4,V4,V4
	MFVSRD	VS36,R10

	CMPU	R16,R10
	SETB_CR0_NE(R3)
	RET

func fnma(x, y, z float64) float64 {
	// riscv64:"FNMADDD",-"FNMSUBD"
	return math.FMA(x, -y, -z)
}

func fromFloat64(f64 float64) uint64 {
	// amd64:"MOVQ\tX.*, [^X].*"
	// arm64:"FMOVD\tF.*, R.*"
	// ppc64x:"MFVSRD"
	// mips64/hardfloat:"MOVV\tF.*, R.*"
	return math.Float64bits(f64+1) + 1
}

func fromFloat32(f32 float32) uint32 {
	// amd64:"MOVL\tX.*, [^X].*"
	// arm64:"FMOVS\tF.*, R.*"

(MOVDload [off] {sym} ptr (FMOVDstore [off] {sym} ptr x _)) => (MFVSRD x)
(FMOVDload [off] {sym} ptr (MOVDstore [off] {sym} ptr x _)) => (MTVSRD x)

(FMOVDstore [off] {sym} ptr (MTVSRD x) mem) => (MOVDstore [off] {sym} ptr x mem)
(MOVDstore [off] {sym} ptr (MFVSRD x) mem) => (FMOVDstore [off] {sym} ptr x mem)

	ALXVDSX
	ASTXV
	ASTXVL
	ASTXVLL
	ASTXVD2X
	ASTXVW4X
	ASTXVH8X
	ASTXVB16X
	ASTXVX
	ALXSDX
	ASTXSDX
	ALXSIWAX
	ALXSIWZX
	ASTXSIWX
	AMFVSRD
	AMFFPRD
	AMFVRD
	AMFVSRWZ
	AMFVSRLD
	AMTVSRD
	AMTFPRD
	AMTVRD
	AMTVSRWA
	AMTVSRWZ
	AMTVSRDD
	AMTVSRWS
	AXXLAND
	AXXLANDC
	AXXLEQV
	AXXLNAND
	AXXLOR
	AXXLORC
	AXXLNOR