4 * memset: Simple memory set in various ways
6 * Trivial clone of mem-memcpy.c.
10 #include "../util/util.h"
11 #include "../util/parse-options.h"
12 #include "../util/header.h"
13 #include "../util/cloexec.h"
15 #include "mem-memset-arch.h"
25 static const char *length_str = "1MB";
26 static const char *routine = "default";
27 static int iterations = 1;
28 static bool use_cycle;
30 static bool only_prefault;
31 static bool no_prefault;
33 static const struct option options[] = {
34 OPT_STRING('l', "length", &length_str, "1MB",
35 "Specify length of memory to set. "
36 "Available units: B, KB, MB, GB and TB (upper and lower)"),
37 OPT_STRING('r', "routine", &routine, "default",
38 "Specify routine to set"),
39 OPT_INTEGER('i', "iterations", &iterations,
40 "repeat memset() invocation this number of times"),
41 OPT_BOOLEAN('c', "cycle", &use_cycle,
42 "Use cycles event instead of gettimeofday() for measuring"),
43 OPT_BOOLEAN('o', "only-prefault", &only_prefault,
44 "Show only the result with page faults before memset()"),
45 OPT_BOOLEAN('n', "no-prefault", &no_prefault,
46 "Show only the result without page faults before memset()"),
50 typedef void *(*memset_t)(void *, int, size_t);
58 static const struct routine routines[] = {
60 "Default memset() provided by glibc",
62 #ifdef HAVE_ARCH_X86_64_SUPPORT
64 #define MEMSET_FN(fn, name, desc) { name, desc, fn },
65 #include "mem-memset-x86-64-asm-def.h"
75 static const char * const bench_mem_memset_usage[] = {
76 "perf bench mem memset <options>",
80 static struct perf_event_attr cycle_attr = {
81 .type = PERF_TYPE_HARDWARE,
82 .config = PERF_COUNT_HW_CPU_CYCLES
85 static void init_cycle(void)
87 cycle_fd = sys_perf_event_open(&cycle_attr, getpid(), -1, -1,
88 perf_event_open_cloexec_flag());
90 if (cycle_fd < 0 && errno == ENOSYS)
91 die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
96 static u64 get_cycle(void)
101 ret = read(cycle_fd, &clk, sizeof(u64));
102 BUG_ON(ret != sizeof(u64));
107 static double timeval2double(struct timeval *ts)
109 return (double)ts->tv_sec +
110 (double)ts->tv_usec / (double)1000000;
113 static void alloc_mem(void **dst, size_t length)
115 *dst = zalloc(length);
117 die("memory allocation failed - maybe length is too large?\n");
120 static u64 do_memset_cycle(memset_t fn, size_t len, bool prefault)
122 u64 cycle_start = 0ULL, cycle_end = 0ULL;
126 alloc_mem(&dst, len);
131 cycle_start = get_cycle();
132 for (i = 0; i < iterations; ++i)
134 cycle_end = get_cycle();
137 return cycle_end - cycle_start;
140 static double do_memset_gettimeofday(memset_t fn, size_t len, bool prefault)
142 struct timeval tv_start, tv_end, tv_diff;
146 alloc_mem(&dst, len);
151 BUG_ON(gettimeofday(&tv_start, NULL));
152 for (i = 0; i < iterations; ++i)
154 BUG_ON(gettimeofday(&tv_end, NULL));
156 timersub(&tv_end, &tv_start, &tv_diff);
159 return (double)((double)len / timeval2double(&tv_diff));
162 #define pf (no_prefault ? 0 : 1)
164 #define print_bps(x) do { \
166 printf(" %14lf B/Sec", x); \
167 else if (x < K * K) \
168 printf(" %14lfd KB/Sec", x / K); \
169 else if (x < K * K * K) \
170 printf(" %14lf MB/Sec", x / K / K); \
172 printf(" %14lf GB/Sec", x / K / K / K); \
175 int bench_mem_memset(int argc, const char **argv,
176 const char *prefix __maybe_unused)
180 double result_bps[2];
183 argc = parse_options(argc, argv, options,
184 bench_mem_memset_usage, 0);
186 if (no_prefault && only_prefault) {
187 fprintf(stderr, "Invalid options: -o and -n are mutually exclusive\n");
194 len = (size_t)perf_atoll((char *)length_str);
196 result_cycle[0] = result_cycle[1] = 0ULL;
197 result_bps[0] = result_bps[1] = 0.0;
200 fprintf(stderr, "Invalid length:%s\n", length_str);
204 /* same to without specifying either of prefault and no-prefault */
205 if (only_prefault && no_prefault)
206 only_prefault = no_prefault = false;
208 for (i = 0; routines[i].name; i++) {
209 if (!strcmp(routines[i].name, routine))
212 if (!routines[i].name) {
213 printf("Unknown routine:%s\n", routine);
214 printf("Available routines...\n");
215 for (i = 0; routines[i].name; i++) {
216 printf("\t%s ... %s\n",
217 routines[i].name, routines[i].desc);
222 if (bench_format == BENCH_FORMAT_DEFAULT)
223 printf("# Copying %s Bytes ...\n\n", length_str);
225 if (!only_prefault && !no_prefault) {
226 /* show both of results */
229 do_memset_cycle(routines[i].fn, len, false);
231 do_memset_cycle(routines[i].fn, len, true);
234 do_memset_gettimeofday(routines[i].fn,
237 do_memset_gettimeofday(routines[i].fn,
243 do_memset_cycle(routines[i].fn,
247 do_memset_gettimeofday(routines[i].fn,
252 switch (bench_format) {
253 case BENCH_FORMAT_DEFAULT:
254 if (!only_prefault && !no_prefault) {
256 printf(" %14lf Cycle/Byte\n",
257 (double)result_cycle[0]
259 printf(" %14lf Cycle/Byte (with prefault)\n ",
260 (double)result_cycle[1]
263 print_bps(result_bps[0]);
265 print_bps(result_bps[1]);
266 printf(" (with prefault)\n");
270 printf(" %14lf Cycle/Byte",
271 (double)result_cycle[pf]
274 print_bps(result_bps[pf]);
276 printf("%s\n", only_prefault ? " (with prefault)" : "");
279 case BENCH_FORMAT_SIMPLE:
280 if (!only_prefault && !no_prefault) {
283 (double)result_cycle[0] / (double)len,
284 (double)result_cycle[1] / (double)len);
287 result_bps[0], result_bps[1]);
291 printf("%lf\n", (double)result_cycle[pf]
294 printf("%lf\n", result_bps[pf]);
298 /* reaching this means there's some disaster: */
299 die("unknown format: %d\n", bench_format);