Callback on Internal Stages of TSIT5

ChrisRackauckas · August 5, 2023, 1:55pm

You’d make modifications right here:

SciML/OrdinaryDiffEq.jl/blob/master/src/perform_step/low_order_rk_perform_step.jl#L770-L823


      
              integrator.k[2] = k2
              integrator.k[3] = k3
              integrator.k[4] = k4
              integrator.k[5] = k5
              integrator.k[6] = k6
              integrator.k[7] = k7
              integrator.u = u
          end
          
          function initialize!(integrator, cache::Tsit5Cache)
              integrator.kshortsize = 7
              integrator.fsalfirst = cache.k1
              integrator.fsallast = cache.k7 # setup pointers
              resize!(integrator.k, integrator.kshortsize)
              # Setup k pointers
              integrator.k[1] = cache.k1
              integrator.k[2] = cache.k2
              integrator.k[3] = cache.k3
              integrator.k[4] = cache.k4
              integrator.k[5] = cache.k5

This file has been truncated. show original

If you want a version that’s more plyable for playing with, you may be interested in SimpleDiffEq.jl

github.com

SciML/SimpleDiffEq.jl/blob/master/src/tsit5/gpuatsit5.jl#L104-L223


      
          @muladd function DiffEqBase.solve(prob::ODEProblem,
              alg::GPUSimpleATsit5;
              dt = 0.1f0, saveat = nothing,
              save_everystep = true,
              abstol = 1.0f-6, reltol = 1.0f-3)
              @assert !isinplace(prob)
              u0 = prob.u0
              tspan = prob.tspan
              f = prob.f
              p = prob.p
              beta1, beta2, qmax, qmin, gamma, qoldinit, _ = build_adaptive_controller_cache(eltype(u0))
          
              t = tspan[1]
              tf = prob.tspan[2]
          
              if saveat === nothing
                  ts = Vector{eltype(dt)}(undef, 1)
                  ts[1] = prob.tspan[1]
                  us = Vector{typeof(u0)}(undef, 0)
                  push!(us, recursivecopy(u0))

This file has been truncated. show original

The barebones version for GPU kernels is a rather nice version for hacking to do weird things.

Topic		Replies	Views
Callback at every 3 integrator steps in DIfferential equations.jl New to Julia	4	699	October 8, 2022
Callback problem with DifferentialEquations Modelling & Simulations diffeq	10	1110	September 11, 2020
Modeling Step function in ODEs Modelling & Simulations	4	242	February 29, 2024
Significant allocations with Callbacks (Tsit5) Modelling & Simulations diffeq	4	580	August 16, 2021
Making a callback event at a single timepoint Modelling & Simulations diffeq	4	784	February 13, 2018

Callback on Internal Stages of TSIT5

Related topics