sqr_toom4.c 文件参考

#include "../../include/lammp/impl/toom_interp.h"

sqr_toom4.c 的引用(Include)关系图:

浏览源代码.

宏定义
#define	a0   numa
#define	a1   (numa + n)
#define	a2   (numa + 2 * n)
#define	a3   (numa + 3 * n)
#define	amx   (dst + 4 * n + 2) /* n+1 */
#define	apx   dst /* n+1 */
#define	tp   (scratch + 8 * n + 5)
#define	v0   dst /* 2n */
#define	v1   (dst + 2 * n) /* 2n+1 */
#define	v2   scratch /* 2n+1 */
#define	vh   (scratch + 4 * n + 2) /* 2n+1 */
#define	vinf   (dst + 6 * n) /* s+t */
#define	vm1   (scratch + 6 * n + 3) /* 2n+1 */
#define	vm2   (scratch + 2 * n + 1) /* 2n+1 */

函数
void	lmmp_sqr_toom4_ (mp_ptr restrict dst, mp_srcptr restrict numa, mp_size_t na)

宏定义说明

a0

#define a0   numa

a1

#define a1   (numa + n)

a2

#define a2   (numa + 2 * n)

a3

#define a3   (numa + 3 * n)

amx

#define amx   (dst + 4 * n + 2) /* n+1 */

apx

#define apx   dst /* n+1 */

tp

#define tp   (scratch + 8 * n + 5)

v0

#define v0   dst /* 2n */

v1

#define v1   (dst + 2 * n) /* 2n+1 */

v2

#define v2   scratch /* 2n+1 */

vh

#define vh   (scratch + 4 * n + 2) /* 2n+1 */

vinf

#define vinf   (dst + 6 * n) /* s+t */

vm1

#define vm1   (scratch + 6 * n + 3) /* 2n+1 */

vm2

#define vm2   (scratch + 2 * n + 1) /* 2n+1 */

函数说明

lmmp_sqr_toom4_()

void lmmp_sqr_toom4_	(	mp_ptr restrict	dst,
		mp_srcptr restrict	numa,
		mp_size_t	na
	)

在文件 sqr_toom4.c 第 24 行定义.

                                                                                 {
    lmmp_param_assert(na > 0);
    lmmp_param_assert(dst != NULL);
    lmmp_param_assert(numa != NULL);
    mp_size_t n, s;
    mp_limb_t cy;
 
#define a0 numa
#define a1 (numa + n)
#define a2 (numa + 2 * n)
#define a3 (numa + 3 * n)
 
    n = (na + 3) >> 2;
    TEMP_S_DECL;
    mp_ptr restrict scratch = SALLOC_TYPE(8 * n + 8, mp_limb_t);
 
    s = na - 3 * n;
 
    lmmp_debug_assert(0 < s && s <= n);
 
    /* NOTE: The multiplications to v2, vm2, vh and vm1 overwrites the
     * following limb, so these must be computed in order, and we need a
     * one limb gap to tp. */
#define v0 dst                     /* 2n */
#define v1 (dst + 2 * n)           /* 2n+1 */
#define vinf (dst + 6 * n)         /* s+t */
#define v2 scratch                 /* 2n+1 */
#define vm2 (scratch + 2 * n + 1)  /* 2n+1 */
#define vh (scratch + 4 * n + 2)   /* 2n+1 */
#define vm1 (scratch + 6 * n + 3)  /* 2n+1 */
#define tp (scratch + 8 * n + 5)
 
    /* No overlap with v1 */
#define apx dst               /* n+1 */
#define amx (dst + 4 * n + 2) /* n+1 */
 
    /* Compute apx = a0 + 2 a1 + 4 a2 + 8 a3 and amx = a0 - 2 a1 + 4 a2 - 8 a3.  */
    lmmp_toom_eval_dgr3_pm2_(apx, amx, numa, n, s, tp);
 
    lmmp_sqr_(v2, apx, n + 1);  /* v2,  2n+1 limbs */
    lmmp_sqr_(vm2, amx, n + 1); /* vm2,  2n+1 limbs */
 
    /* Compute apx = 8 a0 + 4 a1 + 2 a2 + a3 = (((2*a0 + a1) * 2 + a2) * 2 + a3 */
    cy = lmmp_addshl1_n_(apx, a1, a0, n);
    cy = 2 * cy + lmmp_addshl1_n_(apx, a2, apx, n);
    if (s < n) {
        mp_limb_t cy2;
        cy2 = lmmp_addshl1_n_(apx, a3, apx, s);
        apx[n] = 2 * cy + lmmp_shl_(apx + s, apx + s, n - s, 1);
        lmmp_inc_1(apx + s, cy2);
    } else
        apx[n] = 2 * cy + lmmp_addshl1_n_(apx, a3, apx, n);
 
    lmmp_debug_assert(apx[n] < 15);
 
    lmmp_sqr_(vh, apx, n + 1); /* vh,  2n+1 limbs */
 
    /* Compute apx = a0 + a1 + a2 + a3 and amx = a0 - a1 + a2 - a3.  */
    lmmp_toom_eval_dgr3_pm1_(apx, amx, numa, n, s, tp);
 
    lmmp_sqr_(v1, apx, n + 1);  /* v1,  2n+1 limbs */
    lmmp_sqr_(vm1, amx, n + 1); /* vm1,  2n+1 limbs */
 
    lmmp_sqr_(v0, a0, n);
    lmmp_sqr_(vinf, a3, s); /* vinf, 2s limbs */
 
    lmmp_toom_interp7_(dst, n, (enum toom7_flags)0, vm2, vm1, v2, vh, 2 * s, tp);
    TEMP_S_FREE;
}

引用了 a0, a1, a2, a3, amx, apx, lmmp_addshl1_n_(), lmmp_debug_assert, lmmp_inc_1, lmmp_param_assert, lmmp_shl_(), lmmp_sqr_(), lmmp_toom_eval_dgr3_pm1_(), lmmp_toom_eval_dgr3_pm2_(), lmmp_toom_interp7_(), SALLOC_TYPE, scratch, TEMP_S_DECL, TEMP_S_FREE, tp, v0, v1, v2, vh, vinf, vm1 , 以及 vm2.

函数调用图: