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"
14 #include "mem-memset-arch.h"
24 static const char *length_str = "1MB";
25 static const char *routine = "default";
26 static int iterations = 1;
27 static bool use_cycle;
29 static bool only_prefault;
30 static bool no_prefault;
32 static const struct option options[] = {
33 OPT_STRING('l', "length", &length_str, "1MB",
34 "Specify length of memory to set. "
35 "Available units: B, KB, MB, GB and TB (upper and lower)"),
36 OPT_STRING('r', "routine", &routine, "default",
37 "Specify routine to set"),
38 OPT_INTEGER('i', "iterations", &iterations,
39 "repeat memset() invocation this number of times"),
40 OPT_BOOLEAN('c', "cycle", &use_cycle,
41 "Use cycles event instead of gettimeofday() for measuring"),
42 OPT_BOOLEAN('o', "only-prefault", &only_prefault,
43 "Show only the result with page faults before memset()"),
44 OPT_BOOLEAN('n', "no-prefault", &no_prefault,
45 "Show only the result without page faults before memset()"),
49 typedef void *(*memset_t)(void *, int, size_t);
57 static const struct routine routines[] = {
59 "Default memset() provided by glibc",
61 #ifdef HAVE_ARCH_X86_64_SUPPORT
63 #define MEMSET_FN(fn, name, desc) { name, desc, fn },
64 #include "mem-memset-x86-64-asm-def.h"
74 static const char * const bench_mem_memset_usage[] = {
75 "perf bench mem memset <options>",
79 static struct perf_event_attr cycle_attr = {
80 .type = PERF_TYPE_HARDWARE,
81 .config = PERF_COUNT_HW_CPU_CYCLES
84 static void init_cycle(void)
86 cycle_fd = sys_perf_event_open(&cycle_attr, getpid(), -1, -1, 0);
88 if (cycle_fd < 0 && errno == ENOSYS)
89 die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
94 static u64 get_cycle(void)
99 ret = read(cycle_fd, &clk, sizeof(u64));
100 BUG_ON(ret != sizeof(u64));
105 static double timeval2double(struct timeval *ts)
107 return (double)ts->tv_sec +
108 (double)ts->tv_usec / (double)1000000;
111 static void alloc_mem(void **dst, size_t length)
113 *dst = zalloc(length);
115 die("memory allocation failed - maybe length is too large?\n");
118 static u64 do_memset_cycle(memset_t fn, size_t len, bool prefault)
120 u64 cycle_start = 0ULL, cycle_end = 0ULL;
124 alloc_mem(&dst, len);
129 cycle_start = get_cycle();
130 for (i = 0; i < iterations; ++i)
132 cycle_end = get_cycle();
135 return cycle_end - cycle_start;
138 static double do_memset_gettimeofday(memset_t fn, size_t len, bool prefault)
140 struct timeval tv_start, tv_end, tv_diff;
144 alloc_mem(&dst, len);
149 BUG_ON(gettimeofday(&tv_start, NULL));
150 for (i = 0; i < iterations; ++i)
152 BUG_ON(gettimeofday(&tv_end, NULL));
154 timersub(&tv_end, &tv_start, &tv_diff);
157 return (double)((double)len / timeval2double(&tv_diff));
160 #define pf (no_prefault ? 0 : 1)
162 #define print_bps(x) do { \
164 printf(" %14lf B/Sec", x); \
165 else if (x < K * K) \
166 printf(" %14lfd KB/Sec", x / K); \
167 else if (x < K * K * K) \
168 printf(" %14lf MB/Sec", x / K / K); \
170 printf(" %14lf GB/Sec", x / K / K / K); \
173 int bench_mem_memset(int argc, const char **argv,
174 const char *prefix __maybe_unused)
178 double result_bps[2];
181 argc = parse_options(argc, argv, options,
182 bench_mem_memset_usage, 0);
184 if (no_prefault && only_prefault) {
185 fprintf(stderr, "Invalid options: -o and -n are mutually exclusive\n");
192 len = (size_t)perf_atoll((char *)length_str);
194 result_cycle[0] = result_cycle[1] = 0ULL;
195 result_bps[0] = result_bps[1] = 0.0;
198 fprintf(stderr, "Invalid length:%s\n", length_str);
202 /* same to without specifying either of prefault and no-prefault */
203 if (only_prefault && no_prefault)
204 only_prefault = no_prefault = false;
206 for (i = 0; routines[i].name; i++) {
207 if (!strcmp(routines[i].name, routine))
210 if (!routines[i].name) {
211 printf("Unknown routine:%s\n", routine);
212 printf("Available routines...\n");
213 for (i = 0; routines[i].name; i++) {
214 printf("\t%s ... %s\n",
215 routines[i].name, routines[i].desc);
220 if (bench_format == BENCH_FORMAT_DEFAULT)
221 printf("# Copying %s Bytes ...\n\n", length_str);
223 if (!only_prefault && !no_prefault) {
224 /* show both of results */
227 do_memset_cycle(routines[i].fn, len, false);
229 do_memset_cycle(routines[i].fn, len, true);
232 do_memset_gettimeofday(routines[i].fn,
235 do_memset_gettimeofday(routines[i].fn,
241 do_memset_cycle(routines[i].fn,
245 do_memset_gettimeofday(routines[i].fn,
250 switch (bench_format) {
251 case BENCH_FORMAT_DEFAULT:
252 if (!only_prefault && !no_prefault) {
254 printf(" %14lf Cycle/Byte\n",
255 (double)result_cycle[0]
257 printf(" %14lf Cycle/Byte (with prefault)\n ",
258 (double)result_cycle[1]
261 print_bps(result_bps[0]);
263 print_bps(result_bps[1]);
264 printf(" (with prefault)\n");
268 printf(" %14lf Cycle/Byte",
269 (double)result_cycle[pf]
272 print_bps(result_bps[pf]);
274 printf("%s\n", only_prefault ? " (with prefault)" : "");
277 case BENCH_FORMAT_SIMPLE:
278 if (!only_prefault && !no_prefault) {
281 (double)result_cycle[0] / (double)len,
282 (double)result_cycle[1] / (double)len);
285 result_bps[0], result_bps[1]);
289 printf("%lf\n", (double)result_cycle[pf]
292 printf("%lf\n", result_bps[pf]);
296 /* reaching this means there's some disaster: */
297 die("unknown format: %d\n", bench_format);