| 1 | /** @file kmp_stats_timing.cpp |
| 2 | * Timing functions |
| 3 | */ |
| 4 | |
| 5 | //===----------------------------------------------------------------------===// |
| 6 | // |
| 7 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 8 | // See https://llvm.org/LICENSE.txt for license information. |
| 9 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #include <stdlib.h> |
| 14 | #include <unistd.h> |
| 15 | |
| 16 | #include <iomanip> |
| 17 | #include <iostream> |
| 18 | #include <sstream> |
| 19 | |
| 20 | #include "kmp.h" |
| 21 | #include "kmp_stats_timing.h" |
| 22 | |
| 23 | using namespace std; |
| 24 | |
| 25 | #if KMP_HAVE_TICK_TIME |
| 26 | #if KMP_MIC |
| 27 | double tsc_tick_count::tick_time() { |
| 28 | // pretty bad assumption of 1GHz clock for MIC |
| 29 | return 1 / ((double)1000 * 1.e6); |
| 30 | } |
| 31 | #elif KMP_ARCH_X86 || KMP_ARCH_X86_64 |
| 32 | #include <string.h> |
| 33 | // Extract the value from the CPUID information |
| 34 | double tsc_tick_count::tick_time() { |
| 35 | static double result = 0.0; |
| 36 | |
| 37 | if (result == 0.0) { |
| 38 | kmp_cpuid_t cpuinfo; |
| 39 | char brand[256]; |
| 40 | |
| 41 | __kmp_x86_cpuid(0x80000000, 0, &cpuinfo); |
| 42 | memset(brand, 0, sizeof(brand)); |
| 43 | int ids = cpuinfo.eax; |
| 44 | |
| 45 | for (unsigned int i = 2; i < (ids ^ 0x80000000) + 2; i++) |
| 46 | __kmp_x86_cpuid(i | 0x80000000, 0, |
| 47 | (kmp_cpuid_t *)(brand + (i - 2) * sizeof(kmp_cpuid_t))); |
| 48 | |
| 49 | char *start = &brand[0]; |
| 50 | for (; *start == ' '; start++) |
| 51 | ; |
| 52 | |
| 53 | char *end = brand + KMP_STRLEN(brand) - 3; |
| 54 | uint64_t multiplier; |
| 55 | |
| 56 | if (*end == 'M') |
| 57 | multiplier = 1000LL * 1000LL; |
| 58 | else if (*end == 'G') |
| 59 | multiplier = 1000LL * 1000LL * 1000LL; |
| 60 | else if (*end == 'T') |
| 61 | multiplier = 1000LL * 1000LL * 1000LL * 1000LL; |
| 62 | else { |
| 63 | cout << "Error determining multiplier '" << *end << "'\n" ; |
| 64 | exit(-1); |
| 65 | } |
| 66 | *end = 0; |
| 67 | while (*end != ' ') |
| 68 | end--; |
| 69 | end++; |
| 70 | |
| 71 | double freq = strtod(end, &start); |
| 72 | if (freq == 0.0) { |
| 73 | cout << "Error calculating frequency " << end << "\n" ; |
| 74 | exit(-1); |
| 75 | } |
| 76 | |
| 77 | result = ((double)1.0) / (freq * multiplier); |
| 78 | } |
| 79 | return result; |
| 80 | } |
| 81 | #endif |
| 82 | #endif |
| 83 | |
| 84 | static bool useSI = true; |
| 85 | |
| 86 | // Return a formatted string after normalising the value into |
| 87 | // engineering style and using a suitable unit prefix (e.g. ms, us, ns). |
| 88 | std::string formatSI(double interval, int width, char unit) { |
| 89 | std::stringstream os; |
| 90 | |
| 91 | if (useSI) { |
| 92 | // Preserve accuracy for small numbers, since we only multiply and the |
| 93 | // positive powers of ten are precisely representable. |
| 94 | static struct { |
| 95 | double scale; |
| 96 | char prefix; |
| 97 | } ranges[] = {{.scale: 1.e21, .prefix: 'y'}, {.scale: 1.e18, .prefix: 'z'}, {.scale: 1.e15, .prefix: 'a'}, {.scale: 1.e12, .prefix: 'f'}, |
| 98 | {.scale: 1.e9, .prefix: 'p'}, {.scale: 1.e6, .prefix: 'n'}, {.scale: 1.e3, .prefix: 'u'}, {.scale: 1.0, .prefix: 'm'}, |
| 99 | {.scale: 1.e-3, .prefix: ' '}, {.scale: 1.e-6, .prefix: 'k'}, {.scale: 1.e-9, .prefix: 'M'}, {.scale: 1.e-12, .prefix: 'G'}, |
| 100 | {.scale: 1.e-15, .prefix: 'T'}, {.scale: 1.e-18, .prefix: 'P'}, {.scale: 1.e-21, .prefix: 'E'}, {.scale: 1.e-24, .prefix: 'Z'}, |
| 101 | {.scale: 1.e-27, .prefix: 'Y'}}; |
| 102 | |
| 103 | if (interval == 0.0) { |
| 104 | os << std::setw(width - 3) << std::right << "0.00" << std::setw(3) |
| 105 | << unit; |
| 106 | return os.str(); |
| 107 | } |
| 108 | |
| 109 | bool negative = false; |
| 110 | if (interval < 0.0) { |
| 111 | negative = true; |
| 112 | interval = -interval; |
| 113 | } |
| 114 | |
| 115 | for (int i = 0; i < (int)(sizeof(ranges) / sizeof(ranges[0])); i++) { |
| 116 | if (interval * ranges[i].scale < 1.e0) { |
| 117 | interval = interval * 1000.e0 * ranges[i].scale; |
| 118 | os << std::fixed << std::setprecision(2) << std::setw(width - 3) |
| 119 | << std::right << (negative ? -interval : interval) << std::setw(2) |
| 120 | << ranges[i].prefix << std::setw(1) << unit; |
| 121 | |
| 122 | return os.str(); |
| 123 | } |
| 124 | } |
| 125 | } |
| 126 | os << std::setprecision(2) << std::fixed << std::right << std::setw(width - 3) |
| 127 | << interval << std::setw(3) << unit; |
| 128 | |
| 129 | return os.str(); |
| 130 | } |
| 131 | |