Urgent development needs

[jmellorcrummey]

• spack recipes for intel
  • separate gtpin from level0 and opencl
• replace inter-thread communication data structures with multiple-producer, single consumer data structures
  • avoid serialization of intel instrumentation processing
• add support for threaded memory allocator (for use with inter-thread communication outside signal handler)
  • use this with new multi-producer, single consumer inter-thread communication
• improve thread view in hpcviewer
  • improve titles
  • scalable display of execution contexts
  • support both hardware and software contexts
• collect and analyze summary metrics for machine contexts, e.g. GPU, core
  • especially GPU
• suppress display of excess GPU metrics in hpcviewer by default

[Jokeren]

Hi John, can you point me a reference of the "new multi-producer, single consumer" queue you plan to implement? If events can be unordered, can we just use the triber stack?