| 1 | #pragma once
|
|---|
| 2 |
|
|---|
| 3 | #include <cassert>
|
|---|
| 4 | #include <cstddef>
|
|---|
| 5 | #include <atomic>
|
|---|
| 6 | #include <chrono>
|
|---|
| 7 | #include <fstream>
|
|---|
| 8 | #include <iostream>
|
|---|
| 9 |
|
|---|
| 10 | #include <unistd.h>
|
|---|
| 11 | #include <sys/sysinfo.h>
|
|---|
| 12 |
|
|---|
| 13 | #include <x86intrin.h>
|
|---|
| 14 |
|
|---|
| 15 | // Barrier from
|
|---|
| 16 | class barrier_t {
|
|---|
| 17 | public:
|
|---|
| 18 | barrier_t(size_t total)
|
|---|
| 19 | : waiting(0)
|
|---|
| 20 | , total(total)
|
|---|
| 21 | {}
|
|---|
| 22 |
|
|---|
| 23 | void wait(unsigned) {
|
|---|
| 24 | size_t target = waiting++;
|
|---|
| 25 | target = (target - (target % total)) + total;
|
|---|
| 26 | while(waiting < target)
|
|---|
| 27 | asm volatile("pause");
|
|---|
| 28 |
|
|---|
| 29 | assert(waiting < (1ul << 60));
|
|---|
| 30 | }
|
|---|
| 31 |
|
|---|
| 32 | private:
|
|---|
| 33 | std::atomic<size_t> waiting;
|
|---|
| 34 | size_t total;
|
|---|
| 35 | };
|
|---|
| 36 |
|
|---|
| 37 | class Random {
|
|---|
| 38 | private:
|
|---|
| 39 | unsigned int seed;
|
|---|
| 40 | public:
|
|---|
| 41 | Random(int seed) {
|
|---|
| 42 | this->seed = seed;
|
|---|
| 43 | }
|
|---|
| 44 |
|
|---|
| 45 | /** returns pseudorandom x satisfying 0 <= x < n. **/
|
|---|
| 46 | unsigned int next() {
|
|---|
| 47 | seed ^= seed << 6;
|
|---|
| 48 | seed ^= seed >> 21;
|
|---|
| 49 | seed ^= seed << 7;
|
|---|
| 50 | return seed;
|
|---|
| 51 | }
|
|---|
| 52 | };
|
|---|
| 53 |
|
|---|
| 54 | static inline long long rdtscl(void) {
|
|---|
| 55 | unsigned int lo, hi;
|
|---|
| 56 | __asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi));
|
|---|
| 57 | return ( (unsigned long long)lo)|( ((unsigned long long)hi)<<32 );
|
|---|
| 58 | }
|
|---|
| 59 |
|
|---|
| 60 | static inline void affinity(int tid) {
|
|---|
| 61 | static int cpus = get_nprocs();
|
|---|
| 62 |
|
|---|
| 63 | cpu_set_t mask;
|
|---|
| 64 | CPU_ZERO(&mask);
|
|---|
| 65 | int cpu = cpus - tid; // Set CPU affinity to tid, starting from the end
|
|---|
| 66 | CPU_SET(cpu, &mask);
|
|---|
| 67 | auto result = sched_setaffinity(0, sizeof(mask), &mask);
|
|---|
| 68 | if(result != 0) {
|
|---|
| 69 | std::cerr << "Affinity set failed with " << result<< ", wanted " << cpu << std::endl;
|
|---|
| 70 | }
|
|---|
| 71 | }
|
|---|
| 72 |
|
|---|
| 73 | static const constexpr std::size_t cache_line_size = 64;
|
|---|
| 74 | static inline void check_cache_line_size() {
|
|---|
| 75 | std::cout << "Checking cache line size" << std::endl;
|
|---|
| 76 | const std::string cache_file = "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size";
|
|---|
| 77 |
|
|---|
| 78 | std::ifstream ifs (cache_file, std::ifstream::in);
|
|---|
| 79 |
|
|---|
| 80 | if(!ifs.good()) {
|
|---|
| 81 | std::cerr << "Could not open file to check cache line size" << std::endl;
|
|---|
| 82 | std::cerr << "Looking for: " << cache_file << std::endl;
|
|---|
| 83 | std::exit(2);
|
|---|
| 84 | }
|
|---|
| 85 |
|
|---|
| 86 | size_t got;
|
|---|
| 87 | ifs >> got;
|
|---|
| 88 |
|
|---|
| 89 | ifs.close();
|
|---|
| 90 |
|
|---|
| 91 | if(cache_line_size != got) {
|
|---|
| 92 | std::cerr << "Cache line has incorrect size : " << got << std::endl;
|
|---|
| 93 | std::exit(1);
|
|---|
| 94 | }
|
|---|
| 95 |
|
|---|
| 96 | std::cout << "Done" << std::endl;
|
|---|
| 97 | }
|
|---|
| 98 |
|
|---|
| 99 | using Clock = std::chrono::high_resolution_clock;
|
|---|
| 100 | using duration_t = std::chrono::duration<double>;
|
|---|
| 101 | using std::chrono::nanoseconds;
|
|---|
| 102 |
|
|---|
| 103 | template<typename Ratio, typename T>
|
|---|
| 104 | T duration_cast(T seconds) {
|
|---|
| 105 | return std::chrono::duration_cast<std::chrono::duration<T, Ratio>>(std::chrono::duration<T>(seconds)).count();
|
|---|
| 106 | }
|
|---|
| 107 |
|
|---|
| 108 | static inline unsigned rand_bit(unsigned rnum, size_t mask) {
|
|---|
| 109 | unsigned bit = mask ? rnum % __builtin_popcountl(mask) : 0;
|
|---|
| 110 | #if !defined(__BMI2__)
|
|---|
| 111 | uint64_t v = mask; // Input value to find position with rank r.
|
|---|
| 112 | unsigned int r = bit + 1;// Input: bit's desired rank [1-64].
|
|---|
| 113 | unsigned int s; // Output: Resulting position of bit with rank r [1-64]
|
|---|
| 114 | uint64_t a, b, c, d; // Intermediate temporaries for bit count.
|
|---|
| 115 | unsigned int t; // Bit count temporary.
|
|---|
| 116 |
|
|---|
| 117 | // Do a normal parallel bit count for a 64-bit integer,
|
|---|
| 118 | // but store all intermediate steps.
|
|---|
| 119 | a = v - ((v >> 1) & ~0UL/3);
|
|---|
| 120 | b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
|
|---|
| 121 | c = (b + (b >> 4)) & ~0UL/0x11;
|
|---|
| 122 | d = (c + (c >> 8)) & ~0UL/0x101;
|
|---|
| 123 |
|
|---|
| 124 |
|
|---|
| 125 | t = (d >> 32) + (d >> 48);
|
|---|
| 126 | // Now do branchless select!
|
|---|
| 127 | s = 64;
|
|---|
| 128 | s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
|
|---|
| 129 | t = (d >> (s - 16)) & 0xff;
|
|---|
| 130 | s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
|
|---|
| 131 | t = (c >> (s - 8)) & 0xf;
|
|---|
| 132 | s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
|
|---|
| 133 | t = (b >> (s - 4)) & 0x7;
|
|---|
| 134 | s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
|
|---|
| 135 | t = (a >> (s - 2)) & 0x3;
|
|---|
| 136 | s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
|
|---|
| 137 | t = (v >> (s - 1)) & 0x1;
|
|---|
| 138 | s -= ((t - r) & 256) >> 8;
|
|---|
| 139 | return s - 1;
|
|---|
| 140 | #else
|
|---|
| 141 | uint64_t picked = _pdep_u64(1ul << bit, mask);
|
|---|
| 142 | return picked ? __builtin_ctzl(picked) : 0;
|
|---|
| 143 | #endif
|
|---|
| 144 | }
|
|---|
