Ye Luo

> I think QMC_CUDA2HIP is just a terrible flag and without reading the cmake again I couldn't tell you what parts of the native CUDA code it effects. I know...

> > for the code above, -DQMC_GPU=NVIDIA -DENABLE_CUDA=OFF ends up with ENABLE_CUDA being OFF. > > And that is exactly the behavior I want. The explicit option overrides the general....

The lab does spin polarized calculation and doesn't get affected.

Revision needed in a few tests https://cdash.qmcpack.org/CDash/queryTests.php?project=QMCPACK&date=2022-08-20&filtercount=2&showfilters=1&filtercombine=and&field1=testname&compare1=63&value1=excited&field2=testname&compare2=64&value2=C4

@quantumsteve could you paste your cmake line here? I mean on a regular workstation not spock. I got a strange error ``` cmake -DCMAKE_C_COMPILER=/opt/rocm/llvm/bin/clang -DCMAKE_CXX_COMPILER=/opt/rocm/llvm/bin/clang++ -DQMC_MPI=0 -DENABLE_CUDA=ON -DQMC_CUDA2HIP=ON .. ......

Had no luck with gcc ``` cmake -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DENABLE_CUDA=ON -DQMC_CUDA2HIP=ON -DQMC_MPI=OFF .. ``` it tries to link programs with clang instead of gcc. Got messages like ``` [100%] Linking...

HIP is only part of the program. Having HIP compiler (and its flags) to take over linking is just so wrong.

Unfortunately, I found the current ROCm 4.5 release is bad in many ways for adopting HIP as a language. I have to park this PR for the moment. https://github.com/RadeonOpenCompute/ROCm/issues/1259#issuecomment-991753189

> No defaulting nameless estimators to name = "LocalEnergy". If there is no name and no type, abort If there is either name or type, assign one to the other....

nexus carries a similar problem.