8. 高階微分の計算
² 2階微分(snap)まで使って6次精度
² 3階微分(crackle)まで使って8次精度
³ 次の微分にはpopと名前が付いている
² 高次のものほど
³ レジスタ消費は多い
³ 積和比率が高い(逆数平方根は一回)
³ 同期オーバーヘッドが相対的に小さい
38
ops
60
ops
97
ops
144
ops
Acceleration:
Jerk:
Snap:
Crackle:
.2.2 Direct calculation of higher order derivatives
The gravitational acceleration from a particle j on a particle i and its first three
me derivatives are expressed as
Aij = mj
rij
r3
ij
, (2.1)
Jij = mj
vij
r3
ij
− 3αAij, (2.2)
Sij = mj
aij
r3
ij
− 6αJij − 3βAij, (2.3)
Cij = mj
jij
r3
ij
− 9αSij − 9βJij − 3γAij. (2.4)
Here, we call the first four time derivatives of the acceleration jerk, snap, crackle and
op, and α, β and γ are given by
α =
rij · vij
r2
ij
, (2.5)
β =
|vij|2
+ rij · aij
r2
ij
+ α2
, (2.6)
γ =
3vij · aij + rij · jij
2 + α(3β − 4α2
), (2.7)
al acceleration from a particle j on a particle i and its first three
re expressed as
Aij = mj
rij
r3
ij
, (2.1)
Jij = mj
vij
r3
ij
− 3αAij, (2.2)
Sij = mj
aij
r3
ij
− 6αJij − 3βAij, (2.3)
Cij = mj
jij
r3
ij
− 9αSij − 9βJij − 3γAij. (2.4)
first four time derivatives of the acceleration jerk, snap, crackle and
γ are given by
α =
rij · vij
r2
ij
, (2.5)
β =
|vij|2
+ rij · aij
r2
ij
+ α2
, (2.6)
γ =
3vij · aij + rij · jij
r2
ij
+ α(3β − 4α2
), (2.7)
i and mi are the position, velocity, total acceleration, total jerk and
Aij = mj
rij
r3
ij
, (2.1)
Jij = mj
vij
r3
ij
− 3αAij, (2.2)
Sij = mj
aij
r3
ij
− 6αJij − 3βAij, (2.3)
Cij = mj
jij
r3
ij
− 9αSij − 9βJij − 3γAij. (2.4)
rst four time derivatives of the acceleration jerk, snap, crackle and
γ are given by
α =
rij · vij
r2
ij
, (2.5)
β =
|vij|2
+ rij · aij
r2
ij
+ α2
, (2.6)
γ =
3vij · aij + rij · jij
r2
ij
+ α(3β − 4α2
), (2.7)
and mi are the position, velocity, total acceleration, total jerk and
and rij = rj − ri, vij = vj − vi, aij = aj − ai and jij = jj − ji
divとsqrtをそれぞれ
10演算と数えてある
34. indexの生成
² 座標は適当にスケールしておく
² s = 2.0f + r2を計算
² 指数部の下位4-bitと仮数部の上位6-bitを用いる
³ 17-bit右シフトだけでいい
³ あとは1次補間で必要な精度に
smax À 2 smax À 2
’
1
ð2F
þ bFÞ1=2
2bEþ1
=2Fþ2
smax À 2
!1=2
; ð5Þ
where we also assume bE ) 1 and F ) 1 for the last approximation.
Therefore, the sampling points with the same fraction bits are dis-
tributed uniformly in logarithmic scale, and those with the same
exponent bits are aligned uniformly in linear scale unless the frac-
tion bit is small.
As an example, we illustrate how the sampling points of the
look-up table depend on the pre-defined integers E and F in
Fig. 4. We first see the cases in which either of E and F is zero, in
Table 4
s-values, their exponent and fraction bits in the IEEE754 expressions, and their indices
in the table for r ¼ 0, rcut=2 and rcut in the case of E ¼ 4 and F ¼ 6 (underlined portion
of exponent and fraction bits).
r s Exponent
bits
Fraction bits Index
0 2 (smin) 10000000 00000000000000000000000 0
rcut=2 3:2514 Â 104
10001101 11111100000001100000000 895
rcut 1:3005 Â 105
10001111 11111100000000000000000 1023
(smax)
10
-2
10
-1
10
0
10
1
10
2
10
3
10
4
f(r)rcut
3
/r
Conventional
10
-2
10
-1
10
0
10
1
10
2
10
3
10
4
10
-3
10
-2
10
-1
10
0
f(r)rcut
3
/r
r / rcut
Presented
A. Tanikawa et al