Add optional caching #247
Add optional caching #247
Conversation
Also cleaned up the if statements
This uses the `can_prealloc` flag to be able to generate a sparse Jacobian.
| tₑ = 20 | ||
| @info("Solving") | ||
| prob = ODEProblem(f, u₀, (0, tₑ), constants_and_parameters) | ||
| soln = solve(prob, FBDF(linsolve = KLUFactorization()), progress=true, progress_steps=1, saveat=0.1); |
There was a problem hiding this comment.
Leave a comment if you expect this to take longer than other solvers and that this is just for demonstration purposes.
There was a problem hiding this comment.
I've shrunk it down a bit and it only takes 22 seconds to solve.
There was a problem hiding this comment.
What is the relative time to the other solver
There was a problem hiding this comment.
Quite a bit longer, but I want to showcase how to use implicit methods so I've tagged this as demonstration code.
src/simulation.jl
Outdated
|
|
||
| Generates the entire code body for the simulation function. `user_d` is the user passed Decapodes which will be left unmodified and 'input_vars' is the collection of | ||
| state variables and literals in the Decapode. | ||
|
|
||
| Optional keyword arguments are `dimension`, which is the dimension of the problem and defaults to 2D, `stateeltype`, which is the element type of the state forms and | ||
| defaults to Float64 and `code_target`, which is the intended architecture target for the generated code, defaulting to regular CPU compatible code. | ||
| defaults to Float64, `code_target`, which is the intended architecture target for the generated code, defaulting to regular CPU compatible code, and 'can_prealloc' which | ||
| is set to `true` by default and determines if intermediate results can be preallocated. |
There was a problem hiding this comment.
Is setting this flag necessary for using implicit solvers? And if so can that be documented here?
There was a problem hiding this comment.
No, but if you want to find a sparse Jacobian pattern, plus some other more involved implicit solver things I forget exactly what, then you do need this on.
There was a problem hiding this comment.
But this is more so a "safe" mode, where there's less optimization stuff going on that could impact solving, such as needing to find a sparse Jacobian.
There was a problem hiding this comment.
Can you document that either in this docstring or in a blurb in the Docs page
There was a problem hiding this comment.
Yeah, I can do that when I make a docs page on Brusselator.
There was a problem hiding this comment.
Ok. For the time being can you put something in this doctoring then.
There was a problem hiding this comment.
Done and I've also cleaned up the whole docstring since this'll be important to the user.
There was a problem hiding this comment.
I mean can you mention in the docstring the "why" of this flag (Get JacSparsity to work etc.). The "what" is excellently answered.
Also make dimension check more explicit
Now we just need the who, when, where, and how.
|
This is a great feature! The name can_prealloc is a little weird I would like preallocation or a standardized name from the SciML ecosystem if there is one. |
|
I couldn't find anything exactly from SciML but they seem to favor using whole words for keywords. I think |
This adds a keyword
can_preallocthat when set to false turns off preallocations and only uses out of place versions of functions.