// See chacha8_generic.go for additional details.

// ROL rotates the uint32s in register R left by N bits, using temporary T.
#define ROL(N, R, T) \
	MOVO R, T; PSLLL $(N), T; PSRLL $(32-(N)), R; PXOR T, R

// ROL16 rotates the uint32s in register R left by 16, using temporary T if needed.
#ifdef GOAMD64_v2
#define ROL16(R, T) PSHUFB ·rol16<>(SB), R
#else
#define ROL16(R, T) ROL(16, R, T)
#endif

	"SLLIUW",
	"ANDN",
	"ORN",
	"XNOR",
	"CLZ",
	"CLZW",
	"CTZ",
	"CTZW",
	"CPOP",
	"CPOPW",
	"MAX",
	"MAXU",
	"MIN",
	"MINU",
	"SEXTB",
	"SEXTH",
	"ZEXTH",
	"ROL",
	"ROLW",
	"ROR",
	"RORI",
	"RORIW",
	"RORW",
	"ORCB",
	"REV8",
	"BCLR",
	"BCLRI",
	"BEXT",
	"BEXTI",
	"BINV",
	"BINVI",
	"BSET",
	"BSETI",
	"WORD",
	"BEQZ",

func rot64nc(x uint64, z uint) uint64 {
	var a uint64

	z &= 63

	// amd64:"ROLQ",-"AND"
	// arm64:"ROR","NEG",-"AND"
	// ppc64x:"ROTL",-"NEG",-"AND"
	// loong64: "ROTRV", -"AND"
	// riscv64: "ROL",-"AND"
	a += x<<z | x>>(64-z)

	// amd64:"RORQ",-"AND"
	// arm64:"ROR",-"NEG",-"AND"
	// ppc64x:"ROTL","NEG",-"AND"
	// loong64: "ROTRV", -"AND"
	// riscv64: "ROR",-"AND"
	a += x>>z | x<<(64-z)

	NOT:       true,
	OR:        true,
	OUT:       true,
	POP:       true,
	POPA:      true,
	POPCNT:    true,
	PUSH:      true,
	PUSHA:     true,
	RCL:       true,
	RCR:       true,
	ROL:       true,
	ROR:       true,
	SAR:       true,
	SBB:       true,
	SHL:       true,
	SHLD:      true,
	SHR:       true,
	SHRD:      true,
	SUB:       true,
	TEST:      true,
	XADD:      true,