Runtime options

The runtime options below set how PIANISTpy runs. Every option except the platform leaves every result identical bit for bit. CPU-only and GPU-enabled results differ by less than \(10^{-10}\) in relative terms.

pp.set_verbosity("info")   # "quiet" (the default), "info" or "debug"
pp.set_timing(True)        # print the wall-clock time of each step of the computation
pp.set_cores("half")       # "auto" (the default), "full", "half" or a number of cores
pp.set_compilation_cache(False)  # the compilation cache is on by default; a path moves it
pp.set_platform("cpu")     # "auto" (the default), "cpu" (the CPU-only mode) or "gpu", set before the first computation

The CPU-only mode gives the same numbers on every machine. The environment variable PIANISTPY_PLATFORM=cpu, set before the package is imported, also selects the CPU-only mode. Each result file records the platform (CPU or GPU) and the devices that computed the result.

On shared clusters, a GPU in the Exclusive_Process compute mode serves one process at a time. If another process holds the GPU, JAX stops with Unable to initialize backend 'cuda'. Free the GPU, share the GPU through an MPS daemon (nvidia-cuda-mps-control -d), or use the CPU-only mode.

set_cores sets how many cores the package uses for parallel work. "auto" uses every core the process may use inside a batch job, and half of those cores, at least one, anywhere else. No value of set_cores changes a computed number. Each parallel process of the package runs one thread, so the number of cores is also the largest number of parallel processes.

Environment variables set the verbosity, the timing and the number of cores at import: PIANISTPY_VERBOSITY, PIANISTPY_TIMING and PIANISTPY_CORES. The values of the three variables are read in any letter case. PIANISTPY_TIMING takes 1, true, yes or on to switch the printing on, and 0, false, no or off to switch the printing off. Any other value of PIANISTPY_TIMING raises InputError when the package is imported, and the message gives the variable.

The compilation cache keeps JAX’s compiled kernels between sessions, in one folder. The folder is ~/.cache/pianistpy/jax by default, or the folder the environment variable PIANISTPY_COMPILATION_CACHE sets. Setting PIANISTPY_COMPILATION_CACHE to off, 0, false or no turns the cache off. set_compilation_cache returns the folder in use. The cache changes no number.