Location: FermiLab Date: June 27th-29th

• There are a number of outstanding branches that need to be merged into develop for the 0.8 release. Ensuring these are done should be a high-priority task for the meeting.
• Once this is done, we should target getting 0.8 out the door, and merge the multigrid branch into develop.

• Between use, we could hammer out a multi-src solver interface, associated dslash operators and hook this into the deflation algorithms.
• Define an interface for this. This is easy to prototype with a serial loop over calls to invertQuda
• What might we want to support in the future (deflation / mg / simple 5-d solver / block solver)?

• There is huge scope for large speed ups in MILC HMC. A factor of 2 over the current implementation is likely achievable by removing unnecessary PCIe traffic and using gauge field compression in the solver.
• Merge in the MILC profiling branch intro develop
• Enabled automated build testing for MILC - setting Makefile variables through environment variables
• Test the feature/clover-force branch with MILC RHMC
• Interface cleanup and expose feature
• use reconstruction by default or tune ?
• cleanup HISQ force (number of kernels)

• using google test for all tests
• setting up a build server (Jenkins ?)

• switch to cmake (y)
• add a build dir option
• use auto-generated dependencies
• other ways to improve build time
• hierarchical directories
• more separation of files for faster builds (reduce_quda.cu)

• feedback on the current model
• QUDA calls

• Top priority to get this merged into develop
• Does ARM-32 build problem arise with QUDA in 32-bit mode on x86?
• Merge of plaquette codes
• Google Test for gauge fixing and evolution test

• With Justin gone, this is suffering from code rot

• Optimization of clover term reading in the asym_dslash

• gauge force, fat link, HISQ force etc. need to be rewritten using gauge::accessors
• QUDA matrix class should use quda::complex
• remove invertMDQuda