MOVQ acc0, acc4
	MOVQ acc1, acc5
	MOVQ acc2, y_ptr
	MOVQ acc3, t1
	// Subtract p256
	SUBQ $-1, acc0
	SBBQ p256const0<>(SB) ,acc1
	SBBQ $0, acc2
	SBBQ p256const1<>(SB), acc3
	SBBQ $0, t0

	CMOVQCS acc4, acc0
	CMOVQCS acc5, acc1
	CMOVQCS y_ptr, acc2
	CMOVQCS t1, acc3

	MOVQ acc0, (8*0)(res_ptr)
	MOVQ acc1, (8*1)(res_ptr)
	MOVQ acc2, (8*2)(res_ptr)
	MOVQ acc3, (8*3)(res_ptr)

	ADCS	t0, acc3
	ADCS	t1, acc0
	ADC	$0, acc1
	// Third reduction step
	ADDS	acc2<<32, acc3, acc3
	LSR	$32, acc2, t0
	MUL	acc2, const1, t1
	UMULH	acc2, const1, acc2
	ADCS	t0, acc0
	ADCS	t1, acc1
	ADC	$0, acc2
	// Last reduction step
	ADDS	acc3<<32, acc0, acc0
	LSR	$32, acc3, t0
	MUL	acc3, const1, t1
	UMULH	acc3, const1, acc3
	ADCS	t0, acc1
	ADCS	t1, acc2
	ADC	$0, acc3

	SUBS	$-1, acc0, t0

#define polyMulReduceStage MOVQ t0, acc0; MOVQ t1, acc1; MOVQ t2, acc2; ANDQ $3, acc2; MOVQ t2, t0; ANDQ $-4, t0; MOVQ t3, t1; SHRQ $2, t3, t2; SHRQ $2, t3; ADDQ t0, acc0; ADCQ t1, acc1; ADCQ $0, acc2; ADDQ t2, acc0; ADCQ t3, acc1; ADCQ $0, acc2

#define polyMulStage1_AVX2 MOVQ (0*8)(BP), DX; MOVQ DX, t2; MULXQ acc0, t0, t1; IMULQ acc2, t2; MULXQ acc1, AX, DX; ADDQ AX, t1; ADCQ DX, t2

		}
		if got := x.Prec(); got != prec {
			t.Errorf("%s.SetPrec(%d).Prec() == %d; want %d", test.x, test.prec, got, prec)
		}
		if got, acc := x.String(), x.Acc(); got != test.want || acc != test.acc {
			t.Errorf("%s.SetPrec(%d) = %s (%s); want %s (%s)", test.x, test.prec, got, acc, test.want, test.acc)
		}
	}
}

func TestFloatMinPrec(t *testing.T) {
	const max = 100
	for _, test := range []struct {
		x    string

			continue
		}

		// Test against expectation.
		f, acc := i.Float64()
		if f != test.f || acc != test.acc {
			t.Errorf("%s: got %f (%s); want %f (%s)", test.istr, f, acc, test.f, test.acc)
		}

		// Cross-check the fast path against the big.Float implementation.
		f2, acc2 := new(Float).SetInt(i).Float64()
		if f != f2 || acc != acc2 {

	PXOR ACC0, ACCM
	PXOR ACC1, ACCM
	MOVOU ACCM, T0
	PSRLDQ $8, ACCM
	PSLLDQ $8, T0
	PXOR ACCM, ACC1
	PXOR T0, ACC0

	MOVOU POLY, T0
	PCLMULQDQ $0x01, ACC0, T0
	PSHUFD $78, ACC0, ACC0
	PXOR T0, ACC0

	MOVOU POLY, T0
	PCLMULQDQ $0x01, ACC0, T0
	PSHUFD $78, ACC0, ACC0
	PXOR T0, ACC0

	PXOR ACC1, ACC0

	PSHUFB BSWAP, ACC0
	PXOR T2, ACC0
	MOVOU ACC0, (tPtr)

#define KLAST V31

#define reduce() \
	VEOR	ACC0.B16, ACCM.B16, ACCM.B16     \
	VEOR	ACC1.B16, ACCM.B16, ACCM.B16     \
	VEXT	$8, ZERO.B16, ACCM.B16, T0.B16   \
	VEXT	$8, ACCM.B16, ZERO.B16, ACCM.B16 \
	VEOR	ACCM.B16, ACC0.B16, ACC0.B16     \
	VEOR	T0.B16, ACC1.B16, ACC1.B16       \
	VPMULL	POLY.D1, ACC0.D1, T0.Q1          \
	VEXT	$8, ACC0.B16, ACC0.B16, ACC0.B16 \
	VEOR	T0.B16, ACC0.B16, ACC0.B16       \

	// As a matter of fact, this benchmark still triggers some spinning in the mutex.
	var m Mutex
	var acc0, acc1 uint64
	b.SetParallelism(4)
	b.RunParallel(func(pb *testing.PB) {
		c := make(chan bool)
		var data [4 << 10]uint64
		for i := 0; pb.Next(); i++ {
			if i%4 == 0 {
				m.Lock()
				acc0 -= 100
				acc1 += 100
				m.Unlock()
			} else {
				for i := 0; i < len(data); i += 4 {
					data[i]++
				}

	m := NewDRWMutex(ds, "")
	var acc0, acc1 uint64
	b.SetParallelism(4)
	b.RunParallel(func(pb *testing.PB) {
		c := make(chan bool)
		var data [4 << 10]uint64
		for i := 0; pb.Next(); i++ {
			if i%4 == 0 {
				m.Lock(id, source)
				acc0 -= 100
				acc1 += 100
				m.Unlock(context.Background())
			} else {
				for i := 0; i < len(data); i += 4 {

		}
	}
	B interface {
		b() interface {
			ABC2
		}
	}
	C interface {
		c() interface {
			ABC3
		}
	}

	AB interface {
		A
		B
	}
	BC interface {
		B
		C
	}

	ABC1 interface {
		A
		B
		C
	}
	ABC2 interface {
		AB
		C
	}
	ABC3 interface {
		A
		BC
	}
)

var (
	x1 ABC1
	x2 ABC2
	x3 ABC3
)

func _() {