//    Multiplication    //
// -------------------- //

func Pow2Muls(n1, n2 int) (int, int) {
	// amd64:"SHLQ\t[$]5",-"IMULQ"
	// 386:"SHLL\t[$]5",-"IMULL"
	// arm:"SLL\t[$]5",-"MUL"
	// arm64:"LSL\t[$]5",-"MUL"
	// ppc64x:"SLD\t[$]5",-"MUL"
	a := n1 * 32

	// amd64:"SHLQ\t[$]6",-"IMULQ"
	// 386:"SHLL\t[$]6",-"IMULL"
	// arm:"SLL\t[$]6",-"MUL"
	// arm64:`NEG\sR[0-9]+<<6,\sR[0-9]+`,-`LSL`,-`MUL`

	IMUL3Q $0x13, AX, AX
	MULQ   24(BX)
	ADDQ   AX, R9
	ADCQ   DX, R8

	// r1 += 19×a4×b2
	MOVQ   32(CX), AX
	IMUL3Q $0x13, AX, AX
	MULQ   16(BX)
	ADDQ   AX, R9
	ADCQ   DX, R8

	// r2 = a0×b2
	MOVQ (CX), AX
	MULQ 16(BX)
	MOVQ AX, R11
	MOVQ DX, R10

	// r2 += a1×b1
	MOVQ 8(CX), AX
	MULQ 8(BX)
	ADDQ AX, R11
	ADCQ DX, R10

	// r2 += a2×b0
	MOVQ 16(CX), AX
	MULQ (BX)
	ADDQ AX, R11

	MOVQ (8*0)(x_ptr), t0

	MOVQ (8*1)(x_ptr), AX
	MULQ t0
	MOVQ AX, acc1
	MOVQ DX, acc2

	MOVQ (8*2)(x_ptr), AX
	MULQ t0
	ADDQ AX, acc2
	ADCQ $0, DX
	MOVQ DX, acc3

	MOVQ (8*3)(x_ptr), AX
	MULQ t0
	ADDQ AX, acc3
	ADCQ $0, DX
	MOVQ DX, acc4
	// y[2:] * y[1]
	MOVQ (8*1)(x_ptr), t0

	MOVQ (8*2)(x_ptr), AX
	MULQ t0
	ADDQ AX, acc3
	ADCQ $0, DX
	MOVQ DX, t1

	"FCVTDLU",
	"FCVTSD",
	"FCVTDS",
	"FSGNJD",
	"FSGNJND",
	"FSGNJXD",
	"FMVXD",
	"FMVDX",
	"FEQD",
	"FLTD",
	"FLED",
	"FCLASSD",
	"FLQ",
	"FSQ",
	"FADDQ",
	"FSUBQ",
	"FMULQ",
	"FDIVQ",
	"FMINQ",
	"FMAXQ",
	"FSQRTQ",
	"FMADDQ",
	"FMSUBQ",
	"FNMADDQ",
	"FNMSUBQ",
	"FCVTWQ",
	"FCVTLQ",
	"FCVTSQ",
	"FCVTDQ",
	"FCVTQW",
	"FCVTQL",
	"FCVTQS",

		Load(aX, RAX)
	case 2:
		Comment(fmt.Sprintf("%s = 2×%s×%s", r, aX, bX))
		Load(aX, RAX)
		SHLQ(Imm(1), RAX)
	default:
		panic("unsupported i value")
	}
	MULQ(mustAddr(bX)) // RDX, RAX = RAX * bX
	MOVQ(RAX, r.lo)
	MOVQ(RDX, r.hi)
}

// addMul64 sets r to r + i * aX * bX.
func addMul64(r uint128, i uint64, aX, bX namedComponent) {
	switch i {
	case 1:

// set tls base to DI
TEXT runtime·settls(SB),NOSPLIT,$0
	MOVQ	runtime·tls_g(SB), CX
	MOVQ	DI, 0(CX)(GS)
	RET

TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
	MOVQ	$_INTERRUPT_TIME, DI
	MOVQ	time_lo(DI), AX
	IMULQ	$100, AX
	MOVQ	AX, ret+0(FP)
	RET

// func osSetupTLS(mp *m)
// Setup TLS. for use by needm on Windows.
TEXT runtime·osSetupTLS(SB),NOSPLIT,$0-8
	MOVQ	mp+0(FP), AX
	LEAQ	m_tls(AX), DI

//    bits.Mul*    //
// --------------- //

func Mul(x, y uint) (hi, lo uint) {
	// amd64:"MULQ"
	// arm64:"UMULH","MUL"
	// ppc64x:"MULHDU","MULLD"
	// s390x:"MLGR"
	// mips64: "MULVU"
	// riscv64:"MULHU","MUL"
	return bits.Mul(x, y)
}

func Mul64(x, y uint64) (hi, lo uint64) {
	// amd64:"MULQ"
	// arm64:"UMULH","MUL"
	// ppc64x:"MULHDU","MULLD"
	// s390x:"MLGR"
	// mips64: "MULVU"

// (ANDQconst [0xFFFFFFFF] x) => (MOVLQZX x)

// strength reduction
// Assumes that the following costs from https://gmplib.org/~tege/x86-timing.pdf:
//    1 - addq, shlq, leaq, negq, subq
//    3 - imulq
// This limits the rewrites to two instructions.
// Note that negq always operates in-place,
// which can require a register-register move
// to preserve the original value,
// so it must be used with care.

	{AMPSADBW, yxshuf, Pq, opBytes{0x3a, 0x42, 0}},
	{AMULB, ydivb, Pb, opBytes{0xf6, 04}},
	{AMULL, ydivl, Px, opBytes{0xf7, 04}},
	{AMULPD, yxm, Pe, opBytes{0x59}},
	{AMULPS, yxm, Ym, opBytes{0x59}},
	{AMULQ, ydivl, Pw, opBytes{0xf7, 04}},
	{AMULSD, yxm, Pf2, opBytes{0x59}},
	{AMULSS, yxm, Pf3, opBytes{0x59}},
	{AMULW, ydivl, Pe, opBytes{0xf7, 04}},
	{ANEGB, yscond, Pb, opBytes{0xf6, 03}},

		commutative:  true,
		clobberFlags: true,
		asm:          x86.AMULQ,
		reg: regInfo{
			inputs: []inputInfo{
				{0, 1},     // AX
				{1, 49151}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15
			},
			clobbers: 4, // DX
			outputs: []outputInfo{
				{1, 0},
				{0, 1}, // AX
			},
		},
	},
	{
		name:         "HMULQ",
		argLen:       2,
		clobberFlags: true,