Index: benchmark/readyQ/locality.cfa =================================================================== --- benchmark/readyQ/locality.cfa (revision c4241b65cfa6efe7a40dc30450a6a74a14491a08) +++ benchmark/readyQ/locality.cfa (revision e112a2422e4e6bf9a9a4884df0f64f67835cc8ff) @@ -184,7 +184,9 @@ unsigned wsize = 2; unsigned wcnt = 2; + unsigned nspots = 0; bool share = false; cfa_option opt[] = { BENCH_OPT, + { 'n', "nspots", "Number of spots where threads sleep (nthreads - nspots are active at the same time)", nspots}, { 'w', "worksize", "Size of the array for each threads, in words (64bit)", wsize}, { 'c', "workcnt" , "Number of words to touch when working (random pick, cells can be picked more than once)", wcnt }, @@ -196,4 +198,6 @@ unsigned long long global_gmigs = 0; unsigned long long global_dmigs = 0; + + if( nspots == 0 ) { nspots = nthreads - nprocs; } Time start, end; @@ -205,6 +209,6 @@ } - MySpot * spots[nthreads - nprocs]; - for(i; nthreads - nprocs) { + MySpot * spots[nspots]; + for(i; nspots) { spots[i] = malloc(); (*spots[i]){ i }; @@ -220,5 +224,5 @@ data_arrays[i], spots, - nthreads - nprocs, + nspots, wcnt, share, @@ -258,5 +262,5 @@ } - for(i; nthreads - nprocs) { + for(i; nspots) { ^( *spots[i] ){}; free( spots[i] ); Index: benchmark/readyQ/locality.cpp =================================================================== --- benchmark/readyQ/locality.cpp (revision c4241b65cfa6efe7a40dc30450a6a74a14491a08) +++ benchmark/readyQ/locality.cpp (revision e112a2422e4e6bf9a9a4884df0f64f67835cc8ff) @@ -201,5 +201,5 @@ void thread_main( MyCtx & ctx ) { - uint64_t state; + uint64_t state = ctx.id; // Wait for start @@ -237,7 +237,9 @@ unsigned wsize = 2; unsigned wcnt = 2; + unsigned nspots = 0; bool share = false; option_t opt[] = { BENCH_OPT, + { 'n', "nspots", "Number of spots where threads sleep (nthreads - nspots are active at the same time)", nspots}, { 'w', "worksize", "Size of the array for each threads, in words (64bit)", wsize}, { 'c', "workcnt" , "Number of words to touch when working (random pick, cells can be picked more than once)", wcnt }, @@ -252,4 +254,6 @@ unsigned long long global_gmigs = 0; unsigned long long global_dmigs = 0; + + if( nspots == 0 ) { nspots = nthreads - nprocs; } uint64_t start, end; @@ -261,10 +265,10 @@ } - MySpot * spots[nthreads - nprocs]; - for(size_t i = 0; i < (nthreads - nprocs); i++) { + MySpot * spots[nspots]; + for(unsigned i = 0; i < nspots; i++) { spots[i] = new MySpot{ i }; } - threads_left = nthreads; + threads_left = nthreads - nspots; Fibre * threads[nthreads]; MyCtx * thddata[nthreads]; @@ -274,5 +278,5 @@ data_arrays[i], spots, - nthreads - nprocs, + nspots, wcnt, share, @@ -307,5 +311,5 @@ } - for(size_t i = 0; i < (nthreads - nprocs); i++) { + for(size_t i = 0; i < nspots; i++) { delete( spots[i] ); } @@ -315,15 +319,15 @@ printf("Number of processors : %'d\n", nprocs); printf("Number of threads : %'d\n", nthreads); - printf("Total Operations(ops) : %'15llu\n", global_count); + printf("Number of spots : %'d\n", nspots); printf("Work size (64bit words): %'15u\n", wsize); printf("Total Operations(ops) : %'15llu\n", global_count); printf("Total G Migrations : %'15llu\n", global_gmigs); printf("Total D Migrations : %'15llu\n", global_dmigs); - printf("Ops per second : %'18.2lf\n", ((double)global_count) / to_fseconds(end - start)); - printf("ns per ops : %'18.2lf\n", ((double)(end - start)) / global_count); - printf("Ops per threads : %'15llu\n", global_count / nthreads); - printf("Ops per procs : %'15llu\n", global_count / nprocs); - printf("Ops/sec/procs : %'18.2lf\n", (((double)global_count) / nprocs) / to_fseconds(end - start)); - printf("ns per ops/procs : %'18.2lf\n", ((double)(end - start)) / (global_count / nprocs)); + printf("Ops per second : %'18.2lf\n", ((double)global_count) / to_fseconds(end - start)); + printf("ns per ops : %'18.2lf\n", ((double)(end - start)) / global_count); + printf("Ops per threads : %'15llu\n", global_count / nthreads); + printf("Ops per procs : %'15llu\n", global_count / nprocs); + printf("Ops/sec/procs : %'18.2lf\n", (((double)global_count) / nprocs) / to_fseconds(end - start)); + printf("ns per ops/procs : %'18.2lf\n", ((double)(end - start)) / (global_count / nprocs)); fflush(stdout); } Index: benchmark/readyQ/locality.go =================================================================== --- benchmark/readyQ/locality.go (revision c4241b65cfa6efe7a40dc30450a6a74a14491a08) +++ benchmark/readyQ/locality.go (revision e112a2422e4e6bf9a9a4884df0f64f67835cc8ff) @@ -221,4 +221,5 @@ func main() { // Benchmark specific command line arguments + nspotsOpt := flag.Int ("n", 0 , "Number of spots where threads sleep (nthreads - nspots are active at the same time)") work_sizeOpt := flag.Uint64("w", 2 , "Size of the array for each threads, in words (64bit)") countOpt := flag.Uint64("c", 2 , "Number of words to touch when working (random pick, cells can be picked more than once)") @@ -229,7 +230,10 @@ // Eval command line arguments + nspots:= *nspotsOpt size := *work_sizeOpt cnt := *countOpt share := *shareOpt + + if nspots == 0 { nspots = nthreads - nprocs; } // Check params @@ -241,7 +245,7 @@ // Make global data barrierStart := make(chan struct{}) // Barrier used at the start - threads_left = int64(nprocs) // Counter for active threads (not 'nthreads' because at all times 'nthreads - nprocs' are blocked) + threads_left = int64(nthreads - nspots) // Counter for active threads (not 'nthreads' because at all times 'nthreads - nprocs' are blocked) result := make(chan Result) // Channel for results - channels := make([]Spot, nthreads - nprocs) // Number of spots + channels := make([]Spot, nspots) // Number of spots for i := range channels { channels[i] = Spot{[16]uint64{0},uintptr(0), i,[16]uint64{0}} // init spots Index: benchmark/readyQ/locality.rs =================================================================== --- benchmark/readyQ/locality.rs (revision c4241b65cfa6efe7a40dc30450a6a74a14491a08) +++ benchmark/readyQ/locality.rs (revision e112a2422e4e6bf9a9a4884df0f64f67835cc8ff) @@ -257,7 +257,8 @@ let options = App::new("Locality Tokio") .args(&bench::args()) - .arg(Arg::with_name("size") .short("w").long("worksize").takes_value(true).default_value("2").help("Size of the array for each threads, in words (64bit)")) - .arg(Arg::with_name("work") .short("c").long("workcnt") .takes_value(true).default_value("2").help("Number of words to touch when working (random pick, cells can be picked more than once)")) - .arg(Arg::with_name("share").short("s").long("share") .takes_value(true).default_value("true").help("Pass the work data to the next thread when blocking")) + .arg(Arg::with_name("nspots").short("n").long("nspots") .takes_value(true).default_value("0").help("Number of spots where threads sleep (nthreads - nspots are active at the same time)")) + .arg(Arg::with_name("size") .short("w").long("worksize").takes_value(true).default_value("2").help("Size of the array for each threads, in words (64bit)")) + .arg(Arg::with_name("work") .short("c").long("workcnt") .takes_value(true).default_value("2").help("Number of words to touch when working (random pick, cells can be picked more than once)")) + .arg(Arg::with_name("share") .short("s").long("share") .takes_value(true).default_value("true").help("Pass the work data to the next thread when blocking")) .get_matches(); @@ -267,4 +268,12 @@ let wcnt = options.value_of("work").unwrap().parse::().unwrap(); let share = options.value_of("share").unwrap().parse::().unwrap(); + let nspots = { + let val = options.value_of("nspots").unwrap().parse::().unwrap(); + if val != 0 { + val + } else { + nthreads - nprocs + } + }; // Check params @@ -282,5 +291,5 @@ let mut data_arrays : Vec = (0..nthreads).map(|i| MyData::new(i, wsize)).rev().collect(); - let spots : Arc> = Arc::new((0..nthreads - nprocs).map(|i| MySpot::new(i)).rev().collect()); + let spots : Arc> = Arc::new((0..nspots).map(|i| MySpot::new(i)).rev().collect()); let barr = Arc::new(sync::Barrier::new(nthreads + 1));