clad
clad copied to clipboard
vgvassilev
clad -- automatic differentiation for C/C++
This PR solves #1036. CUDA kernels can now be an arg to a `clad::gradient` function and the derived kernels can be executed successfully. Details on issues faced and the final...
While the differentiation of `std::array`s is likely to be landed in #992 soon, that PR does not cover the reverse mode. I'm opening this issue to keep track of this.
currently, if a function contains a lambda expression, Clad treats it as a special case of a functor. there's no general support for lambdas with captures yet. however, I believe...
currently, if a function contains a lambda expression, Clad treats it as a special case of a functor. there's no general support for lambdas with captures yet. however, I believe...
Currently, Clad has 2 different types of pushforwards/pullbacks for methods / member functions: the automatically generated ones are put directly into the class they belong to, while the user-defined ones...
https://github.com/vgvassilev/clad/blob/69d15a31d11fa38b193d274e2392a086f54006bc/include/clad/Differentiator/DynamicGraph.h#L112 outputs something like: ``` f1__order_1__mode_hessian: #0 (source), (done) f2__order_1__mode_hessian: #1 (source), (done) f3__order_1__mode_hessian: #2 (source), (done) f4__order_1__mode_hessian: #3 (source), (done) f5__order_1__mode_hessian: #4 (source), (done) f6__order_1__mode_hessian: #5 (source), (done) f7__order_1__mode_hessian:...
There's a plenty of examples in Kokkos unittests, for instance. Using Clad with the `generate-source-file` option thus produces invalid code with some nested name qualifiers missing. Here's a reproducer: ```c++...
This is the same as #889 but I cannot login to the bots. Let's try if this will let me use tmate...