Closed PHHargrove closed 10 years ago
It must be that NUMA was not detected OR/AND it is different on freebsd.
I am rebuilding my freebds host for harness testing and will take a look.
On 7/8/2014 11:29 AM, Paul H. Hargrove wrote:
Just saw the output below on FreeBSD.
I short: I get the message "Scheduling affinity not supported or configured" followed by a usage summary that includes the /unsupported/ |-sched-policy| option.
I've not verified, but I suspect |-sched-cpu-avoid| and |-mem-policy| have the same issue.
{phargrov@freebsd10-i386 ~}$ ./a.out -n 4 -sched-policy auto Scheduling affinity not supported or configured usage: ./a.out [UPC switches] ... where the possible UPC switches are: -fupc-threads-N or -n N N is number of threads to run (N must be in the range 1..4096) -fupc-heap-N or -heap N N is the maximum per-thread memory allocation heap size The value of N may be followed by a scale factor of K, M, or G (N must be in the range 1..1073741824) -sched-policy [cpu,strict,node,auto] UPC scheduling policy cpu - bind to CPU strict - bind to CPU (one thread per CPU) node - bind to node (if NUMA available) auto - let kernel schedule -sched-cpu-avoid n1,n2,.. List of CPUs to avoid schedulig on 0 to max CPUs -mem-policy [node,strict,auto] UPC memory allocation policy node - allocate on local node first strict - only allocate on local node -g Turn on UPC source code debugging — Reply to this email directly or view it on GitHub https://github.com/Intrepid/clang-upc/issues/63.
It must be that NUMA was not detected OR/AND it is different on freebsd.
That is a separate issue entirely, and I was holding off on reporting it until I'd had a closer look. This particular issue was just a request to prune unsupported options from the usage message.
Just saw the output below on FreeBSD.
I short: I get the message "Scheduling affinity not supported or configured" followed by a usage summary that includes the unsupported
-sched-policy
option.I've not verified, but I suspect
-sched-cpu-avoid
and-mem-policy
have the same issue.