Right now, support MPI, OMP, CUDA and Backtrace is via compile time macro such that those features are built into library. We might want to consider to support runtime configuration of those features such that the library supports all the features, but it is up to runtime to decide which feature to use, and this can be further refined to support enabling/disabling specific feature for specific lexgion. To do this, we will need to define multiple lttng_ust tracepoints (e.g. for MPI/OMP,etc) for each tracepoint/location. This needs more thinking to come up an efficient implementation.
Right now, support MPI, OMP, CUDA and Backtrace is via compile time macro such that those features are built into library. We might want to consider to support runtime configuration of those features such that the library supports all the features, but it is up to runtime to decide which feature to use, and this can be further refined to support enabling/disabling specific feature for specific lexgion. To do this, we will need to define multiple lttng_ust tracepoints (e.g. for MPI/OMP,etc) for each tracepoint/location. This needs more thinking to come up an efficient implementation.