Kenny Weiss
Kenny Weiss
@tomstitt reported a compiler error with `fruit` when using the XL compiler on LLNL's blueos clusters. ``` .../blt-ddd5a0c/thirdparty_builtin/fruit-3.4.1/fruit.f90, line 292.53: 1515-009 (E) Null literal string is not permitted. A single...
BLT currently has a host-config for Cray's CCE + hip: https://github.com/LLNL/blt/blob/develop/host-configs/llnl/toss_4_x86_64_ib_cray/cce%4012.0.0_clang_hip.cmake We should add a few more host-configs for relevant host/device compiler combinations on this platform. Combinations of interest: *...
We should add tables to our user docs/tutorial (https://llnl-blt.readthedocs.io/en/develop/tutorial/common_hpc_dependencies.html) indicating which configuration flags and variables are available/supported/expected to set up `hip` and `cuda`.
When debugging configuration issues, I often find that it would be helpful to print out a subset of CMake's variables and their values. I'd like to add a new utility...
It might be useful to add a new utility macro to set a variable using a ternary operator ```cmake blt_set(TO IF THEN ELSE [CACHE TRUE|FALSE] ) ``` This would replace...
Should we create a new macro to help blt users to gracefully migrate configuration options? This macro could: * produce a custom warning message * indicate which option should be...
CMake uses the mixed-case spelling `Fortran` for fortran-related variables, such as `CMAKE_Fortran_FLAGS`. For self-consistency, BLT uses all upper case spellings, such as `BLT_FORTRAN_FLAGS`. Should we add some checks around the...
@mclarsen came across a problem with the openmp generator expressions that were introduced in #190. In his BLT-based code (ascent), the main library was built with Fortran support (``ENABLE_FORTRAN=ON``) and...
A recent change to LLNL's toss3 slurm has caused problems with our PGI-18 host-config. It looks like we can do one or both of the following: * fix the PGI-18...
One of our user codes recently had an issue in an application that uses ``MPI`` and ``OpenMP``. The issue was related a change in the default ``mpibind`` policy on an...