[b35ab2d] | 1 | #include <clock.hfa> |
---|
| 2 | #include <kernel.hfa> |
---|
| 3 | #include <parseargs.hfa> |
---|
| 4 | #include <stdio.h> |
---|
| 5 | #include <stdlib.hfa> |
---|
| 6 | #include <thread.hfa> |
---|
| 7 | #include <time.hfa> |
---|
| 8 | #include <unistd.h> |
---|
| 9 | |
---|
| 10 | volatile bool stop = false; |
---|
| 11 | bool clock_mode; |
---|
| 12 | double duration = -1; |
---|
| 13 | unsigned long long stop_count = 0; |
---|
| 14 | unsigned nprocs = 1; |
---|
| 15 | unsigned nthreads = 1; |
---|
| 16 | |
---|
| 17 | volatile unsigned long long threads_left; |
---|
| 18 | |
---|
[0b84b15] | 19 | #define thread_loop for(this.count = 0;; this.count++) |
---|
[b35ab2d] | 20 | |
---|
| 21 | #define BENCH_OPT \ |
---|
| 22 | {'d', "duration", "Duration of the experiments in seconds", duration }, \ |
---|
| 23 | {'i', "iterations", "Number of iterations of the experiments", stop_count }, \ |
---|
| 24 | {'t', "nthreads", "Number of threads to use", nthreads }, \ |
---|
| 25 | {'p', "nprocs", "Number of processors to use", nprocs } |
---|
| 26 | |
---|
| 27 | #define BENCH_OPT_PARSE(name) \ |
---|
| 28 | { \ |
---|
| 29 | int opt_cnt = sizeof(opt) / sizeof(cfa_option); \ |
---|
| 30 | char **left; \ |
---|
| 31 | parse_args( argc, argv, opt, opt_cnt, "[OPTIONS]...\n" name, left ); \ |
---|
| 32 | if(duration > 0 && stop_count > 0) { \ |
---|
| 33 | fprintf(stderr, "--duration and --iterations cannot be used together\n"); \ |
---|
| 34 | print_args_usage(argc, argv, opt, opt_cnt, "[OPTIONS]...\n" name, true); \ |
---|
| 35 | } else if(duration > 0) { \ |
---|
| 36 | clock_mode = true; \ |
---|
| 37 | stop_count = 0xFFFFFFFFFFFFFFFF; \ |
---|
[0b84b15] | 38 | printf("Running for %lf seconds\n", duration); \ |
---|
[b35ab2d] | 39 | } else if(stop_count > 0) { \ |
---|
| 40 | clock_mode = false; \ |
---|
[0b84b15] | 41 | printf("Running for %lu iterations\n", stop_count); \ |
---|
[b35ab2d] | 42 | } else { \ |
---|
| 43 | duration = 5; clock_mode = true;\ |
---|
[0b84b15] | 44 | printf("Running for %lf seconds\n", duration); \ |
---|
[b35ab2d] | 45 | } \ |
---|
| 46 | } |
---|
| 47 | |
---|
| 48 | struct cluster & bench_cluster; |
---|
| 49 | |
---|
| 50 | struct BenchCluster { |
---|
| 51 | cluster cl; |
---|
| 52 | processor * procs; |
---|
| 53 | unsigned nprocs; |
---|
| 54 | }; |
---|
| 55 | |
---|
| 56 | void ?{}( BenchCluster & this, unsigned nprocs ) { |
---|
| 57 | (this.cl){ "Benchmark Cluster" }; |
---|
| 58 | &bench_cluster = &this.cl; |
---|
| 59 | this.nprocs = nprocs; |
---|
| 60 | this.procs = alloc( this.nprocs ); |
---|
| 61 | for(i; this.nprocs){ |
---|
| 62 | processor * p = &this.procs[i]; |
---|
| 63 | (*p){ "Benchmark Processor", this.cl }; |
---|
| 64 | } |
---|
| 65 | } |
---|
| 66 | |
---|
| 67 | void ^?{}( BenchCluster & this ) { |
---|
[2c7eee0] | 68 | adelete( this.procs ); |
---|
[b35ab2d] | 69 | ^(this.cl){}; |
---|
| 70 | } |
---|
| 71 | |
---|
[0b84b15] | 72 | void wait(const Time & start, bool is_tty) { |
---|
[b35ab2d] | 73 | for() { |
---|
| 74 | sleep(100`ms); |
---|
[0b84b15] | 75 | Time end = getTimeNsec(); |
---|
[b35ab2d] | 76 | Duration delta = end - start; |
---|
| 77 | if(is_tty) { |
---|
| 78 | printf(" %.1f\r", delta`ds); |
---|
| 79 | fflush(stdout); |
---|
| 80 | } |
---|
| 81 | if( clock_mode && delta >= duration`s ) { |
---|
| 82 | break; |
---|
| 83 | } |
---|
| 84 | else if( !clock_mode && threads_left == 0 ) { |
---|
| 85 | break; |
---|
| 86 | } |
---|
| 87 | } |
---|
[2c7eee0] | 88 | } |
---|
| 89 | |
---|
[b5d51b0] | 90 | struct __attribute__((aligned(128))) bench_sem { |
---|
[2c7eee0] | 91 | struct $thread * volatile ptr; |
---|
| 92 | }; |
---|
| 93 | |
---|
| 94 | static inline { |
---|
| 95 | void ?{}(bench_sem & this) { |
---|
| 96 | this.ptr = 0p; |
---|
| 97 | } |
---|
| 98 | |
---|
| 99 | void ^?{}(bench_sem & this) {} |
---|
| 100 | |
---|
| 101 | bool wait(bench_sem & this) { |
---|
| 102 | for() { |
---|
| 103 | struct $thread * expected = this.ptr; |
---|
| 104 | if(expected == 1p) { |
---|
| 105 | if(__atomic_compare_exchange_n(&this.ptr, &expected, 0p, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) { |
---|
| 106 | return false; |
---|
| 107 | } |
---|
| 108 | } |
---|
| 109 | else { |
---|
| 110 | /* paranoid */ verify( expected == 0p ); |
---|
| 111 | if(__atomic_compare_exchange_n(&this.ptr, &expected, active_thread(), false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) { |
---|
| 112 | park(); |
---|
| 113 | return true; |
---|
| 114 | } |
---|
| 115 | } |
---|
| 116 | |
---|
| 117 | } |
---|
| 118 | } |
---|
| 119 | |
---|
| 120 | bool post(bench_sem & this) { |
---|
| 121 | for() { |
---|
| 122 | struct $thread * expected = this.ptr; |
---|
| 123 | if(expected == 1p) return false; |
---|
| 124 | if(expected == 0p) { |
---|
| 125 | if(__atomic_compare_exchange_n(&this.ptr, &expected, 1p, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) { |
---|
| 126 | return false; |
---|
| 127 | } |
---|
| 128 | } |
---|
| 129 | else { |
---|
| 130 | if(__atomic_compare_exchange_n(&this.ptr, &expected, 0p, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) { |
---|
| 131 | unpark( expected ); |
---|
| 132 | return true; |
---|
| 133 | } |
---|
| 134 | } |
---|
| 135 | } |
---|
| 136 | } |
---|
[b35ab2d] | 137 | } |
---|