sqrt.c 文件参考

#include "../../include/lammp/impl/mparam.h"
#include "../../include/lammp/impl/tmp_alloc.h"
#include "../../include/lammp/lmmpn.h"

sqrt.c 的引用(Include)关系图:

浏览源代码.

函数
static void	lmmp_invsqrt_newton_ (mp_ptr dstis, mp_size_t ns, mp_srcptr numa, mp_size_t na)
void	lmmp_sqrt_ (mp_ptr dsts, mp_ptr dstr, mp_srcptr numa, mp_size_t na, mp_size_t nf)
	大数平方根和取余操作
static mp_limb_t	lmmp_sqrt_1_ (mp_ptr dsts, mp_limb_t x)
static mp_limb_t	lmmp_sqrt_2_ (mp_ptr dsts, mp_ptr dstr, mp_srcptr numa)
static mp_limb_t	lmmp_sqrt_divide_ (mp_ptr dsts, mp_ptr numa, mp_size_t ns, int nsh)
static void	lmmp_sqrt_newton_ (mp_ptr dsts, mp_srcptr numa, mp_size_t na, mp_size_t nf)

变量
static const mp_byte_t	lmmp_invsqrt_table_ []

函数说明

lmmp_invsqrt_newton_()

static void lmmp_invsqrt_newton_ ( mp_ptr  dstis, mp_size_t  ns, mp_srcptr  numa, mp_size_t  na  )

static

在文件 sqrt.c 第 147 行定义.

                                                                                           {
    lmmp_param_assert(ns >= 3);
    lmmp_param_assert(na > 0);
    lmmp_param_assert(numa[na - 1] >= LIMB_B_4);
    mp_size_t nr = ns, namax = na, mn;
    mp_size_t sizes[LIMB_BITS], *sizp = sizes;
 
    do {
        *sizp = nr;
        nr = (nr >> 1) + 1;
        ++sizp;
    } while (nr > 2);
 
    numa += na;
    dstis += ns;
 
    // nr=2
    // i2=floor((B^5-1)/(1+floor(sqrt(x*B^4))))
    mp_limb_t numa2[6], sval[3];
    lmmp_zero(numa2, 4);
    numa2[5] = numa[-1];
    if (na > 1)
        numa2[4] = numa[-2];
    else
        numa2[4] = 0;
    lmmp_sqrt_divide_(sval, numa2, 3, 0);
    lmmp_inc(sval);
    for (mp_size_t i = 0; i < 5; ++i) numa2[i] = LIMB_MAX;
    dstis[0] = lmmp_div_s_(dstis - 2, numa2, 5, sval, 3);
 
    TEMP_DECL;
    mp_limb_t alloc_size = na + 2 * ns + 6;
    mp_ptr xp = TALLOC_TYPE(alloc_size, mp_limb_t);
    do {
        na = *--sizp;
 
        // ar = 0:[numa-nr,nr]
        // an = 0:[numa-na,na]
        // ir = 1:[dst-nr,nr] = floor(B^(3*nr/2)/sqrt(ar)) - [0|1]
        //  d = B^(na+2*nr)-an*ir*ir
        //  -4*B^(na+nr) < d < 4*B^(na+nr)
 
        mp_size_t naz = LMMP_MIN(na, namax);
        //mp_size_t zeros = na - naz;
        mp_size_t nsqr, nres = naz + nr + 1;
        mp_ptr dp = xp + 2 * nr + 1, dip = xp + nr + 1;
        int cmod;  // 1=mod b^mn-1, 0=mod b^(naz+nr+1)
        int sign;  // 1:d<0, 0:d>=0
        mn = lmmp_fft_next_size_(nres);
 
        // ir^2
        if (2 * SQRT_NEWTON_MODM_THRESHOLD + mn >= nr * 2 + 1) {
            cmod = 0;
            lmmp_sqr_(xp, dstis - nr, nr + 1);
            nsqr = 2 * nr + 1;
        } else {
            cmod = 1;
            lmmp_mul_mersenne_(xp, mn, dstis - nr, nr + 1, dstis - nr, nr + 1);
            nsqr = mn;
        }
 
        // ir^2*an
        if (naz < SQRT_NEWTON_MODM_THRESHOLD || naz * 8 < nsqr || mn >= nsqr + naz) {
            if (cmod == 0)
                nsqr = LMMP_MIN(nsqr, nres);
            lmmp_mul_(dp, xp, nsqr, numa - naz, naz);
            if (cmod == 1) {
                if (lmmp_add_(dp, dp, mn, dp + mn, naz))
                    lmmp_inc(dp);
            }
        } else {
            if (nsqr > mn) {  // cmod==0
                if (lmmp_add_(xp, xp, mn, xp + mn, nsqr - mn))
                    lmmp_inc(xp);
            }
            lmmp_mul_mersenne_(dp, mn, xp, nsqr, numa - naz, naz);
            cmod = 1;
        }
 
        if (cmod == 1) {
            // naz+nr < mn <= naz+2*nr
            //[dp,mn] -= B^(naz+2*nr) mod (B^mn-1)
            dp[mn] = 1;
            lmmp_dec(dp + naz + 2 * nr - mn);
            if (dp[mn] == 0)
                lmmp_dec(dp);
        }
 
        if (dp[nres - 1] > 3) {  //-d<0
            if (cmod == 0)
                lmmp_dec(dp);  // for neg to not
            // else (neg to not) compensate (mod transfer)
            dp += naz;
            lmmp_shlnot_(xp, dp + 1, nr, LIMB_BITS - 1);
            xp[0] ^= dp[0] >> 1;
            xp[nr] = ~dp[nr] >> 1;
            sign = 0;
        } else {  //-d>0
            lmmp_shr_(xp, dp + naz, nr + 1, 1);
            if ((dp[naz] & 1) || !lmmp_zero_q_(dp, naz))
                lmmp_inc(xp);
            sign = 1;
        }
 
        lmmp_mul_n_(dip, xp, dstis - nr, nr + 1);
 
        if (sign) {
            if (lmmp_zero_q_(dip, 3 * nr - na)) {
                // a limit for dec
                dip[2 * nr + 1] = 1;
                lmmp_dec(dip + 3 * nr - na);
            }
            lmmp_not_(dstis - na, dip + 3 * nr - na, na - nr);
            lmmp_dec_1(dstis - nr, dip[2 * nr] + 1);
        } else {
            lmmp_copy(dstis - na, dip + 3 * nr - na, na - nr);
            lmmp_inc_1(dstis - nr, dip[2 * nr]);
        }
 
        nr = na;
    } while (sizp != sizes);
    TEMP_FREE;
}

引用了 LIMB_B_4, LIMB_BITS, LIMB_MAX, lmmp_add_(), lmmp_copy, lmmp_dec, lmmp_dec_1, lmmp_div_s_(), lmmp_fft_next_size_(), lmmp_inc, lmmp_inc_1, LMMP_MIN, lmmp_mul_(), lmmp_mul_mersenne_(), lmmp_mul_n_(), lmmp_not_(), lmmp_param_assert, lmmp_shlnot_(), lmmp_shr_(), lmmp_sqr_(), lmmp_sqrt_divide_(), lmmp_zero, lmmp_zero_q_(), SQRT_NEWTON_MODM_THRESHOLD, TALLOC_TYPE, TEMP_DECL , 以及 TEMP_FREE.

被这些函数引用 lmmp_sqrt_newton_().

函数调用图:

这是这个函数的调用关系图:

lmmp_sqrt_()

void lmmp_sqrt_	(	mp_ptr	dsts,
		mp_ptr	dstr,
		mp_srcptr	numa,
		mp_size_t	na,
		mp_size_t	nf
	)

大数平方根和取余操作

注解

如果dstr不为NULL: [dsts,nf+na/2+1], [dstr,nf+na/2+1] = sqrtrem([numa,na]*B^(2*nf)) 也即 [numa,na] × B^(2×nf) = [dsts,nf+na/2+1]^2 + [dstr,nf+na/2+1] 且 0 <= [dstr,nf+na/2+1] < 2 * [dsts,nf+na/2+1] + 1 如果dstr为NULL: [dsts,nf+na/2+1] = [round|floor](sqrt([numa,na]*B^(2*nf)))

警告

na>0, numa[na-1]!=0, eqsep(dsts,numa), eqsep(dstr,numa)

参数

dsts	平方根结果输出指针
dstr	余数结果输出指针（NULL表示不计算余数）
numa	源操作数指针
na	操作数的 limb 长度
nf	精度因子

在文件 sqrt.c 第 323 行定义.

                                                                                      {
    lmmp_debug_assert(na > 0);
    lmmp_debug_assert(numa[na - 1] > 0);
    mp_limb_t high = numa[na - 1];
    int nsh = lmmp_leading_zeros_(high) / 2;
    mp_size_t nl = na + 2 * nf;
    if (nl == 1) {
        mp_limb_t srt;
        lmmp_sqrt_1_(&srt, high << nsh * 2);
        srt >>= nsh;
        dsts[0] = srt;
        if (dstr)
            dstr[0] = high - srt * srt;
    } else if (!dstr && nf >= 10 * na + SQRT_NEWTON_THRESHOLD) {
        lmmp_sqrt_newton_(dsts, numa, na, nf);
    } else {
        TEMP_DECL;
        mp_limb_t ns = (nl + 1) / 2;
        mp_ptr numa2 = TALLOC_TYPE(2 * ns, mp_limb_t);
        if (nf)
            lmmp_zero(numa2, 2 * nf);
        if (nsh)
            lmmp_shl_(numa2 + 2 * ns - na, numa, na, nsh * 2);
        else
            lmmp_copy(numa2 + 2 * ns - na, numa, na);
        if (nl & 1) {
            numa2[2 * nf] = 0;
            nsh += LIMB_BITS / 2;
        } else {
            dsts[ns] = 0;
        }
        mp_limb_t rh = lmmp_sqrt_divide_(dsts, numa2, ns, dstr ? 0 : nsh);
        if (nsh) {
            if (dstr) {
                mp_limb_t ds = dsts[0] & (((mp_limb_t)1 << nsh) - 1);
                rh += lmmp_addmul_1_(numa2, dsts, ns, 2 * ds);
                mp_limb_t b = lmmp_submul_1_(numa2, &ds, 1, ds);
                if (ns == 1)
                    rh -= b;
                else
                    rh -= lmmp_sub_1_(numa2 + 1, numa2 + 1, ns - 1, b);
            }
            lmmp_shr_(dsts, dsts, ns, nsh);
        }
        if (dstr) {
            numa2[ns] = rh;
            nsh *= 2;
            if (nsh >= LIMB_BITS) {
                nsh -= LIMB_BITS;
                ++numa2;
            } else
                ++ns;
            if (nsh)
                lmmp_shr_(dstr, numa2, ns, nsh);
            else
                lmmp_copy(dstr, numa2, ns);
        }
 
        TEMP_FREE;
    }
}

引用了 LIMB_BITS, lmmp_addmul_1_(), lmmp_copy, lmmp_debug_assert, lmmp_leading_zeros_(), lmmp_shl_(), lmmp_shr_(), lmmp_sqrt_1_(), lmmp_sqrt_divide_(), lmmp_sqrt_newton_(), lmmp_sub_1_(), lmmp_submul_1_(), lmmp_zero, SQRT_NEWTON_THRESHOLD, TALLOC_TYPE, TEMP_DECL , 以及 TEMP_FREE.

函数调用图:

lmmp_sqrt_1_()

static mp_limb_t lmmp_sqrt_1_ ( mp_ptr  dsts, mp_limb_t  x  )

static

在文件 sqrt.c 第 38 行定义.

                                                        {
    lmmp_param_assert(x >= LIMB_B_4);
    mp_limb_t v, xh = x >> 24, s, s2;
    mp_slimb_t t;
 
    // round(sqrt(2^25/(1/2+floor(x/2^55))))
    v = 256 + lmmp_invsqrt_table_[(x >> 55) - 128];
 
    t = (((mp_limb_t)1 << 48) - ((x >> 32) + 1) * v * v) * v;
    v = (v << 16) + (t >> 33);
 
    s = v * xh;
    s2 = (s >> 28) + 1;
    t = (xh << 32) - s2 * s2;
    s = s + v * (t >> 33);
 
    // we proved that -0.616 < s/2^32 - sqrt(x) < 0
    // so (s>>32) will be either floor(sqrt(x)), or 1 too small
    s >>= 32;
    x -= s * s;
 
    if (x >= 2 * s + 1) {
        x -= 2 * s + 1;
        ++s;
    }
 
    *dsts = s;
    return x;
}

引用了 LIMB_B_4, lmmp_invsqrt_table_, lmmp_param_assert , 以及 s2.

被这些函数引用 lmmp_sqrt_() , 以及 lmmp_sqrt_2_().

这是这个函数的调用关系图:

lmmp_sqrt_2_()

static mp_limb_t lmmp_sqrt_2_ ( mp_ptr  dsts, mp_ptr  dstr, mp_srcptr  numa  )

static

在文件 sqrt.c 第 70 行定义.

                                                                        {
    lmmp_param_assert(numa[1] >= LIMB_B_4);
    mp_limb_t rl, s, q, al, u;
    mp_slimb_t rh;
 
    rl = lmmp_sqrt_1_(&s, numa[1]);
    al = numa[0];
 
    //(r:alh)/2
    rl = rl << 31 | al >> 33;
    q = rl / s;
    q -= q >> 32;
 
    u = rl - s * q;
    s = s << 32 | q;
    rh = u >> 31;
    rl = (u << 33) | (al & (((mp_limb_t)1 << 33) - 1));
 
    q *= q;
    rh -= rl < q;
    rl -= q;
    if (rh < 0) {
        rl += s;
        rh += rl < s;
        --s;
        rl += s;
        rh += rl < s;
    }
 
    dsts[0] = s;
    dstr[0] = rl;
    return rh;
}

引用了 LIMB_B_4, lmmp_param_assert , 以及 lmmp_sqrt_1_().

被这些函数引用 lmmp_sqrt_divide_().

函数调用图:

这是这个函数的调用关系图:

lmmp_sqrt_divide_()

static mp_limb_t lmmp_sqrt_divide_ ( mp_ptr  dsts, mp_ptr  numa, mp_size_t  ns, int  nsh  )

static

在文件 sqrt.c 第 107 行定义.

                                                                                    {
    lmmp_param_assert(ns > 0);
    lmmp_param_assert(nsh >= 0 && nsh < LIMB_BITS);
    lmmp_param_assert(numa[2 * ns - 1] >= LIMB_B_4);
    mp_slimb_t rh;
    if (ns == 1) {
        rh = lmmp_sqrt_2_(dsts, numa, numa);
    } else {
        mp_size_t lo = ns / 2, hi = ns - lo;
        mp_limb_t qh = lmmp_sqrt_divide_(dsts + lo, numa + 2 * lo, hi, 0);
        if (qh)
            lmmp_sub_n_(numa + 2 * lo, numa + 2 * lo, dsts + lo, hi);
        qh += lmmp_div_s_(dsts, numa + lo, ns, dsts + lo, hi);
        rh = lmmp_shr_c_(dsts, dsts, lo, 1, qh << (LIMB_BITS - 1));
        // now dsts is either correct or 1 too big,
        // if nsh-LSBs are non-zero, subtracting 1
        // will not affect anything after de-normalization
        if (dsts[0] & (((mp_limb_t)1 << nsh) - 1))
            return 1;
        if (rh)
            rh = lmmp_add_n_(numa + lo, numa + lo, dsts + lo, hi);
        qh >>= 1;
        lmmp_sqr_(numa + ns, dsts, lo);
        mp_limb_t b = qh + lmmp_sub_n_(numa, numa, numa + ns, lo * 2);
        if (lo == hi)
            rh -= b;
        else
            rh -= lmmp_sub_1_(numa + 2 * lo, numa + 2 * lo, 1, b);
        if (rh < 0) {
            qh = lmmp_add_1_(dsts + lo, dsts + lo, hi, qh);
            rh += 2 * qh + lmmp_addshl1_n_(numa, numa, dsts, ns);
            rh -= lmmp_sub_1_(numa, numa, ns, 1);
            qh -= lmmp_sub_1_(dsts, dsts, ns, 1);
        }
    }
    return rh;
}

引用了 LIMB_B_4, LIMB_BITS, lmmp_add_1_(), lmmp_add_n_(), lmmp_addshl1_n_(), lmmp_div_s_(), lmmp_param_assert, lmmp_shr_c_(), lmmp_sqr_(), lmmp_sqrt_2_(), lmmp_sqrt_divide_(), lmmp_sub_1_() , 以及 lmmp_sub_n_().

被这些函数引用 lmmp_invsqrt_newton_(), lmmp_sqrt_() , 以及 lmmp_sqrt_divide_().

函数调用图:

这是这个函数的调用关系图:

lmmp_sqrt_newton_()

static void lmmp_sqrt_newton_ ( mp_ptr  dsts, mp_srcptr  numa, mp_size_t  na, mp_size_t  nf  )

static

在文件 sqrt.c 第 275 行定义.

                                                                                       {
    lmmp_param_assert(na > 0);
    lmmp_param_assert(nf >= 2);
    mp_limb_t high = numa[na - 1];
    int nsh = lmmp_leading_zeros_(high) / 2;
    mp_size_t ns = na / 2 + 1 + nf;
 
    TEMP_DECL;
    mp_limb_t alloc_size = (nsh ? na : 0) + ns + 1;
    mp_ptr tp = TALLOC_TYPE(alloc_size, mp_limb_t), numa2;
    if (nsh) {
        numa2 = tp;
        lmmp_shl_(numa2, numa, na, nsh * 2);
        tp += na;
    } else
        numa2 = (mp_ptr)numa;
 
    lmmp_invsqrt_newton_(tp, ns, numa2, na);
 
    mp_ptr msqr = TALLOC_TYPE(na + ns + 1, mp_limb_t);
 
    if (ns + 1 > na)
        lmmp_mul_(msqr, tp, ns + 1, numa2, na);
    else
        lmmp_mul_(msqr, numa2, na, tp, ns + 1);
 
    mp_limb_t cceil;
    if (na & 1) {
        nsh += LIMB_BITS / 2;
        lmmp_shr_(dsts, msqr + na, ns, nsh);
        cceil = msqr[na] >> (nsh - 1);
    } else {
        if (nsh) {
            lmmp_shr_(dsts, msqr + na + 1, ns - 1, nsh);
            cceil = msqr[na + 1] >> (nsh - 1);
        } else {
            lmmp_copy(dsts, msqr + na + 1, ns - 1);
            cceil = msqr[na] >> (LIMB_BITS - 1);
        }
        dsts[ns - 1] = 0;
    }
 
    if (cceil & 1)
        lmmp_inc(dsts);
 
    TEMP_FREE;
}

引用了 LIMB_BITS, lmmp_copy, lmmp_inc, lmmp_invsqrt_newton_(), lmmp_leading_zeros_(), lmmp_mul_(), lmmp_param_assert, lmmp_shl_(), lmmp_shr_(), TALLOC_TYPE, TEMP_DECL, TEMP_FREE , 以及 tp.

被这些函数引用 lmmp_sqrt_().

函数调用图:

这是这个函数的调用关系图:

变量说明

lmmp_invsqrt_table_

const mp_byte_t lmmp_invsqrt_table_[]

static

初始值:

= {
    0xff, 0xfd, 0xfb, 0xf9, 0xf7, 0xf5, 0xf3, 0xf2, 0xf0, 0xee, 0xec, 0xea, 0xe9, 0xe7, 0xe5, 0xe4, 0xe2, 0xe0, 0xdf,
    0xdd, 0xdb, 0xda, 0xd8, 0xd7, 0xd5, 0xd4, 0xd2, 0xd1, 0xcf, 0xce, 0xcc, 0xcb, 0xc9, 0xc8, 0xc6, 0xc5, 0xc4, 0xc2,
    0xc1, 0xc0, 0xbe, 0xbd, 0xbc, 0xba, 0xb9, 0xb8, 0xb7, 0xb5, 0xb4, 0xb3, 0xb2, 0xb0, 0xaf, 0xae, 0xad, 0xac, 0xaa,
    0xa9, 0xa8, 0xa7, 0xa6, 0xa5, 0xa4, 0xa3, 0xa2, 0xa0, 0x9f, 0x9e, 0x9d, 0x9c, 0x9b, 0x9a, 0x99, 0x98, 0x97, 0x96,
    0x95, 0x94, 0x93, 0x92, 0x91, 0x90, 0x8f, 0x8e, 0x8d, 0x8c, 0x8c, 0x8b, 0x8a, 0x89, 0x88, 0x87, 0x86, 0x85, 0x84,
    0x83, 0x83, 0x82, 0x81, 0x80, 0x7f, 0x7e, 0x7e, 0x7d, 0x7c, 0x7b, 0x7a, 0x79, 0x79, 0x78, 0x77, 0x76, 0x76, 0x75,
    0x74, 0x73, 0x72, 0x72, 0x71, 0x70, 0x6f, 0x6f, 0x6e, 0x6d, 0x6d, 0x6c, 0x6b, 0x6a, 0x6a, 0x69, 0x68, 0x68, 0x67,
    0x66, 0x66, 0x65, 0x64, 0x64, 0x63, 0x62, 0x62, 0x61, 0x60, 0x60, 0x5f, 0x5e, 0x5e, 0x5d, 0x5c, 0x5c, 0x5b, 0x5a,
    0x5a, 0x59, 0x59, 0x58, 0x57, 0x57, 0x56, 0x56, 0x55, 0x54, 0x54, 0x53, 0x53, 0x52, 0x52, 0x51, 0x50, 0x50, 0x4f,
    0x4f, 0x4e, 0x4e, 0x4d, 0x4d, 0x4c, 0x4b, 0x4b, 0x4a, 0x4a, 0x49, 0x49, 0x48, 0x48, 0x47, 0x47, 0x46, 0x46, 0x45,
    0x45, 0x44, 0x44, 0x43, 0x43, 0x42, 0x42, 0x41, 0x41, 0x40, 0x40, 0x3f, 0x3f, 0x3e, 0x3e, 0x3d, 0x3d, 0x3c, 0x3c,
    0x3b, 0x3b, 0x3a, 0x3a, 0x39, 0x39, 0x39, 0x38, 0x38, 0x37, 0x37, 0x36, 0x36, 0x35, 0x35, 0x35, 0x34, 0x34, 0x33,
    0x33, 0x32, 0x32, 0x32, 0x31, 0x31, 0x30, 0x30, 0x2f, 0x2f, 0x2f, 0x2e, 0x2e, 0x2d, 0x2d, 0x2d, 0x2c, 0x2c, 0x2b,
    0x2b, 0x2b, 0x2a, 0x2a, 0x29, 0x29, 0x29, 0x28, 0x28, 0x27, 0x27, 0x27, 0x26, 0x26, 0x26, 0x25, 0x25, 0x24, 0x24,
    0x24, 0x23, 0x23, 0x23, 0x22, 0x22, 0x21, 0x21, 0x21, 0x20, 0x20, 0x20, 0x1f, 0x1f, 0x1f, 0x1e, 0x1e, 0x1e, 0x1d,
    0x1d, 0x1d, 0x1c, 0x1c, 0x1b, 0x1b, 0x1b, 0x1a, 0x1a, 0x1a, 0x19, 0x19, 0x19, 0x18, 0x18, 0x18, 0x18, 0x17, 0x17,
    0x17, 0x16, 0x16, 0x16, 0x15, 0x15, 0x15, 0x14, 0x14, 0x14, 0x13, 0x13, 0x13, 0x12, 0x12, 0x12, 0x12, 0x11, 0x11,
    0x11, 0x10, 0x10, 0x10, 0x0f, 0x0f, 0x0f, 0x0f, 0x0e, 0x0e, 0x0e, 0x0d, 0x0d, 0x0d, 0x0c, 0x0c, 0x0c, 0x0c, 0x0b,
    0x0b, 0x0b, 0x0a, 0x0a, 0x0a, 0x0a, 0x09, 0x09, 0x09, 0x09, 0x08, 0x08, 0x08, 0x07, 0x07, 0x07, 0x07, 0x06, 0x06,
    0x06, 0x06, 0x05, 0x05, 0x05, 0x04, 0x04, 0x04, 0x04, 0x03, 0x03, 0x03, 0x03, 0x02, 0x02, 0x02, 0x02, 0x01, 0x01,
    0x01, 0x01, 0x00, 0x00}

在文件 sqrt.c 第 12 行定义.

                                               {
    0xff, 0xfd, 0xfb, 0xf9, 0xf7, 0xf5, 0xf3, 0xf2, 0xf0, 0xee, 0xec, 0xea, 0xe9, 0xe7, 0xe5, 0xe4, 0xe2, 0xe0, 0xdf,
    0xdd, 0xdb, 0xda, 0xd8, 0xd7, 0xd5, 0xd4, 0xd2, 0xd1, 0xcf, 0xce, 0xcc, 0xcb, 0xc9, 0xc8, 0xc6, 0xc5, 0xc4, 0xc2,
    0xc1, 0xc0, 0xbe, 0xbd, 0xbc, 0xba, 0xb9, 0xb8, 0xb7, 0xb5, 0xb4, 0xb3, 0xb2, 0xb0, 0xaf, 0xae, 0xad, 0xac, 0xaa,
    0xa9, 0xa8, 0xa7, 0xa6, 0xa5, 0xa4, 0xa3, 0xa2, 0xa0, 0x9f, 0x9e, 0x9d, 0x9c, 0x9b, 0x9a, 0x99, 0x98, 0x97, 0x96,
    0x95, 0x94, 0x93, 0x92, 0x91, 0x90, 0x8f, 0x8e, 0x8d, 0x8c, 0x8c, 0x8b, 0x8a, 0x89, 0x88, 0x87, 0x86, 0x85, 0x84,
    0x83, 0x83, 0x82, 0x81, 0x80, 0x7f, 0x7e, 0x7e, 0x7d, 0x7c, 0x7b, 0x7a, 0x79, 0x79, 0x78, 0x77, 0x76, 0x76, 0x75,
    0x74, 0x73, 0x72, 0x72, 0x71, 0x70, 0x6f, 0x6f, 0x6e, 0x6d, 0x6d, 0x6c, 0x6b, 0x6a, 0x6a, 0x69, 0x68, 0x68, 0x67,
    0x66, 0x66, 0x65, 0x64, 0x64, 0x63, 0x62, 0x62, 0x61, 0x60, 0x60, 0x5f, 0x5e, 0x5e, 0x5d, 0x5c, 0x5c, 0x5b, 0x5a,
    0x5a, 0x59, 0x59, 0x58, 0x57, 0x57, 0x56, 0x56, 0x55, 0x54, 0x54, 0x53, 0x53, 0x52, 0x52, 0x51, 0x50, 0x50, 0x4f,
    0x4f, 0x4e, 0x4e, 0x4d, 0x4d, 0x4c, 0x4b, 0x4b, 0x4a, 0x4a, 0x49, 0x49, 0x48, 0x48, 0x47, 0x47, 0x46, 0x46, 0x45,
    0x45, 0x44, 0x44, 0x43, 0x43, 0x42, 0x42, 0x41, 0x41, 0x40, 0x40, 0x3f, 0x3f, 0x3e, 0x3e, 0x3d, 0x3d, 0x3c, 0x3c,
    0x3b, 0x3b, 0x3a, 0x3a, 0x39, 0x39, 0x39, 0x38, 0x38, 0x37, 0x37, 0x36, 0x36, 0x35, 0x35, 0x35, 0x34, 0x34, 0x33,
    0x33, 0x32, 0x32, 0x32, 0x31, 0x31, 0x30, 0x30, 0x2f, 0x2f, 0x2f, 0x2e, 0x2e, 0x2d, 0x2d, 0x2d, 0x2c, 0x2c, 0x2b,
    0x2b, 0x2b, 0x2a, 0x2a, 0x29, 0x29, 0x29, 0x28, 0x28, 0x27, 0x27, 0x27, 0x26, 0x26, 0x26, 0x25, 0x25, 0x24, 0x24,
    0x24, 0x23, 0x23, 0x23, 0x22, 0x22, 0x21, 0x21, 0x21, 0x20, 0x20, 0x20, 0x1f, 0x1f, 0x1f, 0x1e, 0x1e, 0x1e, 0x1d,
    0x1d, 0x1d, 0x1c, 0x1c, 0x1b, 0x1b, 0x1b, 0x1a, 0x1a, 0x1a, 0x19, 0x19, 0x19, 0x18, 0x18, 0x18, 0x18, 0x17, 0x17,
    0x17, 0x16, 0x16, 0x16, 0x15, 0x15, 0x15, 0x14, 0x14, 0x14, 0x13, 0x13, 0x13, 0x12, 0x12, 0x12, 0x12, 0x11, 0x11,
    0x11, 0x10, 0x10, 0x10, 0x0f, 0x0f, 0x0f, 0x0f, 0x0e, 0x0e, 0x0e, 0x0d, 0x0d, 0x0d, 0x0c, 0x0c, 0x0c, 0x0c, 0x0b,
    0x0b, 0x0b, 0x0a, 0x0a, 0x0a, 0x0a, 0x09, 0x09, 0x09, 0x09, 0x08, 0x08, 0x08, 0x07, 0x07, 0x07, 0x07, 0x06, 0x06,
    0x06, 0x06, 0x05, 0x05, 0x05, 0x04, 0x04, 0x04, 0x04, 0x03, 0x03, 0x03, 0x03, 0x02, 0x02, 0x02, 0x02, 0x01, 0x01,
    0x01, 0x01, 0x00, 0x00};

被这些函数引用 lmmp_sqrt_1_().