2.28 Varmat issue

[bbbales2]

Summary:

This issue is to track what is yet to be done for varmat to go out in 2.27. Edit at will!

This is a resurrected version of the 2.26 Issue: #958

I will try to keep it updated with what I know about:

• Stanc3 [Early WIP]: var_value<Eigen::matrix> stanc3#755
  • Create a list of function signatures in Math that support varmat arguments. Work in progress here
  • Figure out a language implementation - A version of the compiler that makes a variable a varmat if the name ends in _varmat is available here
    • Finish the language parts of the design doc. Finished design doc here
    • scalar read/writes of varmat variables might be slow (it's come up when we talk implementation). Results here
• Math
  • Finish the varmat parts of the design doc. Finished design doc here
  • Document varmat at the C++ level, from the perspective of someone coming to Math to write a function. This is higher level than Doxygen and lower level than Stan user docs. Work in progress here
  • Update all distributions to work with varmat
    • Non-multivariate distributions. Work in progress here and here
    • Multivariate distributions. Finished here and here
  • Write varmat implementations for every function that had custom matvar autodiff, work in progress here
  • Write varmat implementations for all the functions used in the constrain functions.
    • Matrix constrains here
    • Upper/lower boundshere

Current Version:

v2.26.0

[bbbales2]

Here is the list of fully compatible functions (any signature that could varmat, does varmat, 320 functions): fully_compatible.txt

Here is the list of functions which have at least one signature that doesn't support varmat (197 functions, 44 of which are rngs which won't ever be autodiffed): any_incompatible.txt