Is there a fast way to evaluate the exp() function in Julia?

[nilshg]

Isn’t this the same question you asked a few hours ago in

Accelerate pairwise damage computation

Can someone help me to accelerate the pairwise damage computation? This problem is very similar to the one in Accelerate pairwise L-J force computation. And I adopted the conclusion that the pairlist should be sorted first to have a better performance. The following codes are modified based on the one in Accelerate pairwise Lennard-Jones force computation given by @ lmiq using BenchmarkTools using StaticArrays const λ = 1.0E-6 @inline @fastmath function damage_pair!(f,x,y,i,j,d) new_d = 0.0…

Please don’t double post

[maphdze]

I am sorry. I have flagged this to the admin to delate it.

[goerch]

There exists Base.Math.exp_fast.

Compare for example @code_native Base.Math.exp_fast.([1., 2., 3., 4.]) with @code_native exp.([1., 2., 3., 4.])…

[maphdze]

Thanks for your reply. Is the Base.Math.exp_fast the same as using @fastmath which I already done in the posted code?

[maphdze]

Well, I have just tried it and the performance is the same as using @fastmath.

[giordano]

@maphdze

Is the Base.Math.exp_fast the same as using @fastmath

Yes

[FrancisKing]

I don’t know how accurate the calculation needs to be. But you could pre-compute an array of values - then it’s just a question of a lookup. That should be quite fast.

[maphdze]

Thanks for your suggestion! Do you mean interplation? Actually the x in exp(x) is not known beforehand, and if we pre-cmpute an array we have to look up in the array and then do some interplation. I guess the looking up process maybe time-consuming.