LMMP 4.3.0
Lamina High-Precision Arithmetic Library
载入中...
搜索中...
未找到
perf.h
浏览该文件的文档.
1/**
2 * Copyright (C) 2026 HJimmyK (Jericho Knox)
3 *
4 * This file is part of LMMP.
5 *
6 * LMMP is free software: you can redistribute it and/or modify it
7 * under the terms of the GNU Lesser General Public License as
8 * published by the Free Software Foundation; either version 3 of
9 * the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library. If not, see
18 * <https://www.gnu.org/licenses/>.
19 *
20 * SPDX-License-Identifier: LGPL-3.0-or-later
21 */
22
23#ifndef LMMP_PERF_H
24#define LMMP_PERF_H
25
26#include <stdbool.h>
27#include <stdio.h>
28#include <stdlib.h>
29#include <string.h>
30#include <time.h>
31
32#include "lmmp/lmmp.h"
33#include "lmmp/mprand.h"
34#include "lmmp/numth.h"
35
36#if defined(LMMP_WINDOWS)
37#include <windows.h>
38#endif
39
40#define ARRAY_SIZE(a) (sizeof(a) / sizeof((a)[0]))
41
43
51
52static const bench_options bench_default = {BENCH_MIN, 5, 0.02, 0.05, 1ull << 26};
53static volatile mp_limb_t bench_sink;
54
55static inline double bench_clock(void) {
56#if defined(LMMP_WINDOWS)
57 static double ticks_to_seconds;
58 LARGE_INTEGER ticks;
59 if (ticks_to_seconds == 0) {
60 QueryPerformanceFrequency(&ticks);
61 ticks_to_seconds = 1.0 / (double)ticks.QuadPart;
62 }
63 QueryPerformanceCounter(&ticks);
64 return (double)ticks.QuadPart * ticks_to_seconds;
65#else
66 struct timespec now;
67 clock_gettime(CLOCK_MONOTONIC, &now);
68 return (double)now.tv_sec + 1e-9 * (double)now.tv_nsec;
69#endif
70}
71
72static inline int bench_compare_times(const void *lhs, const void *rhs) {
73 double a = *(const double *)lhs, b = *(const double *)rhs;
74 return (a > b) - (a < b);
75}
76
77static inline double bench_aggregate(double *samples, unsigned n, bench_aggregation agg) {
78 if (agg == BENCH_MEDIAN) {
79 qsort(samples, n, sizeof(*samples), bench_compare_times);
80 return (samples[(n - 1) / 2] + samples[n / 2]) * 0.5;
81 }
82 double result = samples[0];
83 for (unsigned i = 1; i < n; ++i) {
84 if (agg == BENCH_MEAN)
85 result += samples[i];
86 else if (samples[i] < result)
87 result = samples[i];
88 }
89 return agg == BENCH_MEAN ? result / n : result;
90}
91
92// 调用表达式在计时循环内直接展开。
93#define BENCH_MEASURE(result, options, ...) \
94 do { \
95 const bench_options *bench_opts_ = (options); \
96 { __VA_ARGS__ } \
97 mp_size_t bench_iters_ = 1; \
98 double bench_elapsed_; \
99 for (;;) { \
100 double bench_start_ = bench_clock(); \
101 for (mp_size_t bench_i_ = 0; bench_i_ < bench_iters_; ++bench_i_) { __VA_ARGS__ } \
102 bench_elapsed_ = bench_clock() - bench_start_; \
103 if (bench_elapsed_ >= bench_opts_->min_block_sec || bench_iters_ >= bench_opts_->max_iters) \
104 break; \
105 bench_iters_ = LMMP_MIN(2 * bench_iters_, bench_opts_->max_iters); \
106 } \
107 unsigned bench_count_ = bench_opts_->max_samples ? bench_opts_->max_samples : 1; \
108 if (bench_elapsed_ >= bench_opts_->slow_block_sec && bench_count_ > 2) \
109 bench_count_ = 2; \
110 double bench_samples_[bench_count_]; \
111 for (unsigned bench_s_ = 0; bench_s_ < bench_count_; ++bench_s_) { \
112 double bench_start_ = bench_clock(); \
113 for (mp_size_t bench_i_ = 0; bench_i_ < bench_iters_; ++bench_i_) { __VA_ARGS__ } \
114 bench_samples_[bench_s_] = (bench_clock() - bench_start_) * 1e6 / bench_iters_; \
115 } \
116 (result) = bench_aggregate(bench_samples_, bench_count_, bench_opts_->agg); \
117 } while (0)
118
120 return lmmp_alloc(n * sizeof(mp_limb_t));
121}
122
123static inline void fill_random(mp_ptr p, mp_size_t n, mp_limb_t seed) {
124 lmmp_seed_random_(p, n, seed, 1);
125 if (n != 0)
126 p[n - 1] |= (mp_limb_t)1 << 63;
127}
128
129static inline void report(const char *name, double us, double bytes) {
130 double ops = us > 0 ? 1e6 / us : 0;
131 printf(" %-40s %14.4f us/op %14.2f ops/s", name, us, ops);
132 if (bytes > 0)
133 printf(" %10.2f MiB/s", bytes * ops / (1024.0 * 1024.0));
134 putchar('\n');
135}
136
137static inline void report_n(const char *name, mp_size_t n, double us, double bytes) {
138 char label[96];
139 snprintf(label, sizeof(label), "%s(%lluL)", name, (unsigned long long)n);
140 report(label, us, bytes);
141}
142
143static inline void report_ab(const char *name, mp_size_t a, mp_size_t b, double us, double bytes) {
144 char label[96];
145 snprintf(label, sizeof(label), "%s(%llux%llu)", name, (unsigned long long)a, (unsigned long long)b);
146 report(label, us, bytes);
147}
148
149static inline void report_bits(const char *name, uint bits, double us) {
150 char label[96];
151 snprintf(label, sizeof(label), "%s(%ub)", name, bits);
152 report(label, us, 0);
153}
154
155#endif
mp_limb_t * mp_ptr
定义 lmmp.h:96
uint64_t mp_size_t
定义 lmmp.h:93
void * lmmp_alloc(size_t size)
内存分配函数(调用lmmp_heap_alloc_fn)
uint64_t mp_limb_t
定义 lmmp.h:92
mp_size_t lmmp_seed_random_(mp_ptr dst, mp_size_t n, mp_limb_t seed, unsigned seed_type)
生成随机大整数(0 - B^n-1 均匀分布)
#define n
uint32_t uint
定义 numth.h:38
unsigned max_samples
定义 perf.h:46
static void report_ab(const char *name, mp_size_t a, mp_size_t b, double us, double bytes)
定义 perf.h:143
bench_aggregation agg
定义 perf.h:45
static mp_ptr alloc_limbs(mp_size_t n)
定义 perf.h:119
static double bench_clock(void)
定义 perf.h:55
static void report(const char *name, double us, double bytes)
定义 perf.h:129
static void report_n(const char *name, mp_size_t n, double us, double bytes)
定义 perf.h:137
double min_block_sec
定义 perf.h:47
double slow_block_sec
定义 perf.h:48
mp_size_t max_iters
定义 perf.h:49
bench_aggregation
定义 perf.h:42
@ BENCH_MEAN
定义 perf.h:42
@ BENCH_MIN
定义 perf.h:42
@ BENCH_MEDIAN
定义 perf.h:42
static void report_bits(const char *name, uint bits, double us)
定义 perf.h:149
static int bench_compare_times(const void *lhs, const void *rhs)
定义 perf.h:72
static volatile mp_limb_t bench_sink
定义 perf.h:53
static double bench_aggregate(double *samples, unsigned n, bench_aggregation agg)
定义 perf.h:77
static void fill_random(mp_ptr p, mp_size_t n, mp_limb_t seed)
定义 perf.h:123
static const bench_options bench_default
定义 perf.h:52