Adds mem_pattern needed for SoA and AoS types #944
Conversation
|
@rok-cesnovar would you mind reviewing this? |
|
But of course, would love to. This is exciting. |
|
🙏 🙏 🙏 Much appreciated! Do you happen to know what would cause the error on jenkins with the |
This reverts commit db8f4d5.
|
I tried bumping it but no dice. Paging @serban-nicusor-toptal :) |
|
Checking Edit: Fixed now, there was a cache issue on that machine. Sorry for the trouble! |
|
Np and thank you! |
# Conflicts: # test/integration/good/code-gen/transformed_mir.expected # test/integration/good/compiler-optimizations/cpp.expected
|
There seems to be an issue on master with 2 models: https://jenkins.mc-stan.org/blue/organizations/jenkins/stanc3/detail/master/868/pipeline/185 We need to fix those first. Given the error I would say they are related to stan-dev/math#2421 but not super sure. |
|
@rok-cesnovar I think your right I just opened a revert on that PR and know what the issue is and can open up a revert revert after the revert happens |
|
@rok-cesnovar it looks like cmdstan and Stan hasn't update in the last 5 days so the changes in math haven't propogated upwards. Is there something goofed with the Jenkins stuff? |
|
Math develop has failed and thus not propagated up. Should pass now after stan-dev/math#2564 was merged. |
…recompiled header error
|
I really gotta figure out this precompiled header issue. Maybe we just need to add a |
|
Its annoying. I think we arent cleaning up something from previous runs or something along those lines. |
|
Yeah I'm looking at this right now in cmdstan and will post up a PR once I get something sorted, but going to merge this for now |
This is a subset of #898. There's a lot of stuff going on in that PR so I think it's smarter to do it in smaller pieces.
This PR adds the
Common.Helpers.mem_patterntype and uses it in the necessary types that we need for optimization of the tmir and printing the C++ the usesSoA(Struct of Arrays). The largest change is toStan_math_signatures, where theStanLibtypes there now have an extra space in their tuple for amem_pattern.Everything here is set to
AoS(Array of Structs) and so while the mir and tmir have changed, the C++ should not. That said, the C++ from the optimizer has changed slightly, though only the names of the symbol ticks have changed while the actual code has not changed at all.Submission Checklist
Release notes
Adds types used for deducing memory patterns
Copyright and Licensing
By submitting this pull request, the copyright holder is agreeing to
license the submitted work under the BSD 3-clause license (https://opensource.org/licenses/BSD-3-Clause)