"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",

	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

			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)
			opset(AMTVSRWZ, r0)
			opset(AMTVSRWS, 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

	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
	AXXLORQ
	AXXLXOR
	AXXSEL
	AXXMRGHW

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.*"

// Lowering float <=> int
(Cvt32to(32|64)F x) => ((FCFIDS|FCFID) (MTVSRD (SignExt32to64 x)))
(Cvt64to(32|64)F x) => ((FCFIDS|FCFID) (MTVSRD x))

(Cvt32Fto(32|64) x) => (MFVSRD (FCTI(W|D)Z x))
(Cvt64Fto(32|64) x) => (MFVSRD (FCTI(W|D)Z x))

(Cvt32Fto64F ...) => (Copy ...) // Note v will have the wrong type for patterns dependent on Float32/Float64
(Cvt64Fto32F ...) => (FRSP ...)

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

	VSLDOI	$8,V0,zeroes,V0

#else

	VAND	V0,mask_32bit,V1
	VPMSUMD	V1,const1,V1
	VAND	V1,mask_32bit,V1
	VPMSUMD	V1,const2,V1
	VXOR	V0,V1,V0
	VSLDOI  $4,V0,zeroes,V0

#endif

	MFVSRD	VS32,R3 // VS32 = V0

	NOR	R3,R3,R3 // return ^crc
	MOVW	R3,ret+32(FP)
	RET

first_warm_up_done:

	LVX	(R3),const1
	ADD	$16,R3

	VPMSUMD	V16,const1,V8
	VPMSUMD	V17,const1,V9

		// There are optimizations that should apply -- (Xi2f64 (MOVWload (not-ADD-ptr+offset) ) ) could use
		// the word-load instructions.  (Xi2f64 (MOVDload ptr )) can be (FMOVDload ptr)

		{name: "MFVSRD", argLength: 1, reg: fpgp, asm: "MFVSRD", typ: "Int64"},   // move 64 bits of F register into G register
		{name: "MTVSRD", argLength: 1, reg: gpfp, asm: "MTVSRD", typ: "Float64"}, // move 64 bits of G register into F register