Will the impulse response graph be same no matter what kind of variables I use in DYNARE(For example, x(level variable) or exp(x)(log variable))? If I use linear model(do log-linearization by myself), will the impulse response graph be same with graph generated by using level variables or log variables in DYNARE?

Of course not. In one case (variable in levels), the IRF will be for the total/absolute change in the level. If you use log variables through exp() substitution or log-linearize yourself, your IRF will be the percentage change in a variable relative from steady state. Thus, the two will only coincide for variables of steady state 1 as the normalization with the steady state does not do anything.

That is what I think so. However, no matter percent change or level change, they should have the same direction. I mean if there is an increment for certain variable in level, there should be an increment in percent change in log. Am I correct?

Yes, comovement must be the same.