Index: benchmark/readyQ/yield.cfa =================================================================== --- benchmark/readyQ/yield.cfa (revision db89655bc5956ac9b93d6b72963a2fda2172d2a6) +++ benchmark/readyQ/yield.cfa (revision 9cf2b0f4dc8ecca0d2753969d3132960ce3071c6) @@ -1,19 +1,3 @@ -#include -#include -#include -#include - -extern "C" { - #include - #include -} - -#include - -#include -#include -#include - -#include "../benchcltr.hfa" +#include "rq_bench.hfa" extern bool traceHeapOn(); @@ -28,5 +12,5 @@ void ?{}( Yielder & this ) { this.counter = 0; - ((thread&)this){ "Yielder Thread", *the_benchmark_cluster }; + ((thread&)this){ "Yielder Thread" }; } @@ -43,14 +27,8 @@ int main(int argc, char * argv[]) { - unsigned num_io = 1; - io_context_params params; - cfa_option opt[] = { - BENCH_OPT_CFA + BENCH_OPT }; - int opt_cnt = sizeof(opt) / sizeof(cfa_option); - - char **left; - parse_args( argc, argv, opt, opt_cnt, "[OPTIONS]...\ncforall yield benchmark", left ); + BENCH_OPT_PARSE("cforall yield benchmark"); { @@ -58,35 +36,33 @@ Time start, end; - BenchCluster cl = { num_io, params, CFA_STATS_READY_Q }; + BenchCluster bc = { nprocs }; { - BenchProc procs[nprocs]; - { - Yielder threads[nthreads]; - printf("Starting\n"); + Yielder threads[nthreads]; + printf("Starting\n"); - bool is_tty = isatty(STDOUT_FILENO); - start = timeHiRes(); - run = true; + bool is_tty = isatty(STDOUT_FILENO); + start = timeHiRes(); + run = true; - for(i; nthreads) { - unpark( threads[i] ); - } - wait(duration, start, end, is_tty); + for(i; nthreads) { + unpark( threads[i] ); + } + wait(start, is_tty); - run = false; - end = timeHiRes(); - printf("\nDone\n"); - } + run = false; + end = timeHiRes(); + printf("\nDone\n"); } - printf("Duration (ms) : %'ld\n", (end - start)`dms); - printf("Number of processors: %'d\n", nprocs); - printf("Number of threads : %'d\n", nthreads); - printf("Total yields : %'15llu\n", global_counter); - printf("Yields per second : %'18.2lf\n", ((double)global_counter) / (end - start)`s); - printf("ns per yields : %'18.2lf\n", ((double)(end - start)`ns) / global_counter); - printf("Yields per procs : %'15llu\n", global_counter / nprocs); - printf("Yields/sec/procs : %'18.2lf\n", (((double)global_counter) / nprocs) / (end - start)`s); - printf("ns per yields/procs : %'18.2lf\n", ((double)(end - start)`ns) / (global_counter / nprocs)); + printf("Duration (ms) : %'ld\n", (end - start)`dms); + printf("Number of processors : %'d\n", nprocs); + printf("Number of threads : %'d\n", nthreads); + printf("Total Operations(ops): %'15llu\n", global_counter); + printf("Ops per second : %'18.2lf\n", ((double)global_counter) / (end - start)`s); + printf("ns per ops : %'18.2lf\n", (end - start)`dns / global_counter); + printf("Ops per threads : %'15llu\n", global_counter / nthreads); + printf("Ops per procs : %'15llu\n", global_counter / nprocs); + printf("Ops/sec/procs : %'18.2lf\n", (((double)global_counter) / nprocs) / (end - start)`s); + printf("ns per ops/procs : %'18.2lf\n", (end - start)`dns / (global_counter / nprocs)); fflush(stdout); } Index: benchmark/readyQ/yield.cpp =================================================================== --- benchmark/readyQ/yield.cpp (revision db89655bc5956ac9b93d6b72963a2fda2172d2a6) +++ benchmark/readyQ/yield.cpp (revision 9cf2b0f4dc8ecca0d2753969d3132960ce3071c6) @@ -1,167 +1,67 @@ -#include -#include -#include -#include -#include - -extern "C" { - #include - #include -} - -#include - -#include - -using Clock = std::chrono::high_resolution_clock; -using duration_t = std::chrono::duration; -using std::chrono::nanoseconds; - - -template -T duration_cast(T seconds) { - return std::chrono::duration_cast>(std::chrono::duration(seconds)).count(); -} +#include "rq_bench.hpp" +#include volatile bool run = false; volatile unsigned long long global_counter; -#include "libfibre/fibre.h" -FredBarrier * barrier; -struct __attribute__((aligned(128))) counter_t { - int value = 0; -}; +void fibre_main() { + fibre_park(); + unsigned long long count = 0; + for(;;) { + Fibre::forceYield(); + count++; + if( clock_mode && stop) break; + if(!clock_mode && count >= stop_count) break; + } -void fibre_main( counter_t * counter ) { - barrier->wait(); - // /* paranoid */ assert( true == __atomic_load_n(&run, __ATOMIC_RELAXED) ); - - while(__atomic_load_n(&run, __ATOMIC_RELAXED)) { - Fibre::forceYield(); - // fibre_yield(); - counter->value++; - } - __atomic_fetch_add(&global_counter, counter->value, __ATOMIC_SEQ_CST); + __atomic_fetch_add(&global_counter, count, __ATOMIC_SEQ_CST); + __atomic_fetch_add(&threads_left, -1, __ATOMIC_SEQ_CST); } int main(int argc, char * argv[]) { - double duration = 5; - int nprocs = 1; - int nthreads = 1; - - std::cout.imbue(std::locale("")); - setlocale(LC_ALL, ""); - - for(;;) { - static struct option options[] = { - {"duration", required_argument, 0, 'd'}, - {"nprocs", required_argument, 0, 'p'}, - {"nthreads", required_argument, 0, 't'}, - {0, 0, 0, 0} - }; - - int idx = 0; - int opt = getopt_long(argc, argv, "d:p:t:", options, &idx); - - const char * arg = optarg ? optarg : ""; - char * end; - switch(opt) { - case -1: - goto run; - // Numeric Arguments - case 'd': - duration = strtod(arg, &end); - if(*end != '\0') { - fprintf(stderr, "Duration must be a valid double, was %s\n", arg); - goto usage; - } - break; - case 't': - nthreads = strtoul(arg, &end, 10); - if(*end != '\0' || nthreads < 1) { - fprintf(stderr, "Number of threads must be a positive integer, was %s\n", arg); - goto usage; - } - break; - case 'p': - nprocs = strtoul(arg, &end, 10); - if(*end != '\0' || nprocs < 1) { - fprintf(stderr, "Number of processors must be a positive integer, was %s\n", arg); - goto usage; - } - break; - // Other cases - default: /* ? */ - fprintf( stderr, "Unkown option '%c'\n", opt); - usage: - fprintf( stderr, "Usage: %s [options]\n", argv[0]); - fprintf( stderr, "\n" ); - fprintf( stderr, " -d, --duration=DURATION Duration of the experiment, in seconds\n" ); - fprintf( stderr, " -t, --nthreads=NTHREADS Number of kernel threads\n" ); - fprintf( stderr, " -q, --nqueues=NQUEUES Number of queues per threads\n" ); - exit(1); - } - } - run: + option_t opt[] = { + BENCH_OPT + }; + BENCH_OPT_PARSE("libfibre yield benchmark"); { printf("Running %d threads on %d processors for %lf seconds\n", nthreads, nprocs, duration); - FibreInit(); - barrier = new FredBarrier(nthreads + 1); + FibreInit(1, nprocs); + uint64_t start, end; { - Context::CurrCluster().addWorkers(nprocs); - { - counter_t counters[nthreads]; - Fibre threads[nthreads]; - for(int i = 0; i < nthreads; i++) { - threads[i].run(fibre_main, &counters[i]); - } - printf("Starting\n"); - bool is_tty = isatty(STDOUT_FILENO); - auto before = Clock::now(); - run = true; + threads_left = nthreads; + Fibre * threads[nthreads]; + for(unsigned i = 0; i < nthreads; i++) { + threads[i] = new Fibre( reinterpret_cast(fibre_main), nullptr ); + } + printf("Starting\n"); + bool is_tty = isatty(STDOUT_FILENO); + start = timeHiRes(); - barrier->wait(); - for(;;) { - usleep(500'000); - auto now = Clock::now(); - duration_t durr = now - before; - if( durr.count() > duration ) { - break; - } - if(is_tty) { - std::cout << "\r" << std::setprecision(4) << durr.count(); - std::cout.flush(); - } - } + for(unsigned i = 0; i < nthreads; i++ ) { + fibre_unpark( threads[i] ); + } + wait(start, is_tty); - auto after = Clock::now(); - duration_t durr = after - before; - duration = durr.count(); - run = false; - printf("\nDone\n"); - for(auto & thread : threads) { - thread.join(); - } - - // for(const auto & counter : counters) { - // std::cout << counter.value << std::endl; - // } + stop = true; + end = timeHiRes(); + for(unsigned i = 0; i < nthreads; i++ ) { + fibre_join( threads[i], nullptr ); } } - auto dur_nano = duration_cast(duration); - auto dur_dms = duration_cast(duration); - - printf("Duration (ms) : %'.2lf\n", dur_dms ); - printf("Total yields : %'15llu\n", global_counter ); - printf("Yields per procs : %'15llu\n", global_counter / nprocs ); - printf("Yields per second : %'18.2lf\n", ((double)global_counter) / duration ); - printf("Yields/sec/procs : %'18.2lf\n", (((double)global_counter) / nprocs) / duration ); - printf("ns per yields : %'18.2lf\n", dur_nano / global_counter ); - printf("ns per yields/procs : %'18.2lf\n", dur_nano / (global_counter / nprocs) ); - + printf("Duration (ms) : %'ld\n", to_miliseconds(end - start)); + printf("Number of processors : %'d\n", nprocs); + printf("Number of threads : %'d\n", nthreads); + printf("Total Operations(ops): %'15llu\n", global_counter); + printf("Ops per second : %'18.2lf\n", ((double)global_counter) / to_fseconds(end - start)); + printf("ns per ops : %'18.2lf\n", ((double)(end - start)) / global_counter); + printf("Ops per threads : %'15llu\n", global_counter / nthreads); + printf("Ops per procs : %'15llu\n", global_counter / nprocs); + printf("Ops/sec/procs : %'18.2lf\n", (((double)global_counter) / nprocs) / to_fseconds(end - start)); + printf("ns per ops/procs : %'18.2lf\n", ((double)(end - start)) / (global_counter / nprocs)); + fflush(stdout); } } Index: benchmark/readyQ/yield.rs =================================================================== --- benchmark/readyQ/yield.rs (revision db89655bc5956ac9b93d6b72963a2fda2172d2a6) +++ benchmark/readyQ/yield.rs (revision 9cf2b0f4dc8ecca0d2753969d3132960ce3071c6) @@ -90,13 +90,13 @@ }); - println!("Duration (ms) : {}", (duration.as_millis()).to_formatted_string(&Locale::en)); - println!("Number of processors: {}", (nprocs).to_formatted_string(&Locale::en)); - println!("Number of threads : {}", (nthreads).to_formatted_string(&Locale::en)); - println!("Total yields : {:>15}", (global_counter).to_formatted_string(&Locale::en)); - println!("Yields per second : {:>15}", (((global_counter as f64) / duration.as_secs() as f64) as u64).to_formatted_string(&Locale::en)); - println!("ns per yields : {:>15}", ((duration.as_nanos() as f64 / global_counter as f64) as u64).to_formatted_string(&Locale::en)); - println!("Yields per threads : {:>15}", (global_counter / nthreads as u64).to_formatted_string(&Locale::en)); - println!("Yields per procs : {:>15}", (global_counter / nprocs as u64).to_formatted_string(&Locale::en)); - println!("Yields/sec/procs : {:>15}", ((((global_counter as f64) / nprocs as f64) / duration.as_secs() as f64) as u64).to_formatted_string(&Locale::en)); - println!("ns per yields/procs : {:>15}", ((duration.as_nanos() as f64 / (global_counter as f64 / nprocs as f64)) as u64).to_formatted_string(&Locale::en)); + println!("Duration (ms) : {}", (duration.as_millis()).to_formatted_string(&Locale::en)); + println!("Number of processors : {}", (nprocs).to_formatted_string(&Locale::en)); + println!("Number of threads : {}", (nthreads).to_formatted_string(&Locale::en)); + println!("Total Operations(ops): {:>15}", (global_counter).to_formatted_string(&Locale::en)); + println!("Ops per second : {:>15}", (((global_counter as f64) / duration.as_secs() as f64) as u64).to_formatted_string(&Locale::en)); + println!("ns per ops : {:>15}", ((duration.as_nanos() as f64 / global_counter as f64) as u64).to_formatted_string(&Locale::en)); + println!("Ops per threads : {:>15}", (global_counter / nthreads as u64).to_formatted_string(&Locale::en)); + println!("Ops per procs : {:>15}", (global_counter / nprocs as u64).to_formatted_string(&Locale::en)); + println!("Ops/sec/procs : {:>15}", ((((global_counter as f64) / nprocs as f64) / duration.as_secs() as f64) as u64).to_formatted_string(&Locale::en)); + println!("ns per ops/procs : {:>15}", ((duration.as_nanos() as f64 / (global_counter as f64 / nprocs as f64)) as u64).to_formatted_string(&Locale::en)); }