mirror of
https://github.com/kdave/btrfs-progs
synced 2024-12-25 07:32:22 +00:00
btrfs-progs: crypto: add perf support to speed test
Use perf events to read the cycle count, this should work on all
architectures. Enabled by option --perf and the sysctl
kernel.perf_event_paranoid must be 0 or 1.
The results are roughly the same as for raw cycles on x86_64 but worse
because of the additional overhead (read, context switch):
Block size: 4096
Iterations: 100000
Implementation: builtin
Units: CPU cycles
NULL-NOP: cycles: 42719688, cycles/i 427
NULL-MEMCPY: cycles: 72941208, cycles/i 729, 18670.314 MiB/s
CRC32C: cycles: 183709926, cycles/i 1837, 7413.009 MiB/s
XXHASH: cycles: 136727614, cycles/i 1367, 9960.264 MiB/s
SHA256: cycles: 10711594532, cycles/i 107115, 127.137 MiB/s
BLAKE2: cycles: 2256957529, cycles/i 22569, 603.398 MiB/s
Block size: 4096
Iterations: 100000
Implementation: builtin
Units: perf event: CPU cycles
NULL-NOP: perf_c: 29649530, perf_c/i 296
NULL-MEMCPY: perf_c: 59954062, perf_c/i 599, 15137.464 MiB/s
CRC32C: perf_c: 179009071, perf_c/i 1790, 6929.460 MiB/s
XXHASH: perf_c: 136413509, perf_c/i 1364, 9982.950 MiB/s
SHA256: perf_c: 10997356664, perf_c/i 109973, 127.046 MiB/s
BLAKE2: perf_c: 2379077576, perf_c/i 23790, 588.780 MiB/s
Signed-off-by: David Sterba <dsterba@suse.com>
This commit is contained in:
David Sterba
parent
1d4bab875a
commit
9527bc0649
@ -60,6 +60,9 @@ AC_CHECK_FUNCS([reallocarray])
|
||||
|
||||
AC_CHECK_FUNCS([clock_gettime])
|
||||
|
||||
AC_CHECK_HEADERS([linux/perf_event.h])
|
||||
AC_CHECK_HEADERS([linux/hw_breakpoint.h])
|
||||
|
||||
m4_ifndef([PKG_PROG_PKG_CONFIG],
|
||||
[m4_fatal([Could not locate the pkg-config autoconf
|
||||
macros. These are usually located in /usr/share/aclocal/pkg.m4.
|
||||
|
||||
@ -1,6 +1,13 @@
|
||||
#include "../kerncompat.h"
|
||||
#include <time.h>
|
||||
#include <getopt.h>
|
||||
#include <unistd.h>
|
||||
#if HAVE_LINUX_PERF_EVENT_H == 1 && HAVE_LINUX_HW_BREAKPOINT_H == 1
|
||||
#include <linux/perf_event.h>
|
||||
#include <linux/hw_breakpoint.h>
|
||||
#include <sys/syscall.h>
|
||||
#define HAVE_PERF
|
||||
#endif
|
||||
#include "crypto/hash.h"
|
||||
#include "crypto/crc32c.h"
|
||||
#include "crypto/sha.h"
|
||||
@ -12,6 +19,12 @@ static const int cycles_supported = 1;
|
||||
static const int cycles_supported = 0;
|
||||
#endif
|
||||
|
||||
enum {
|
||||
UNITS_CYCLES,
|
||||
UNITS_TIME,
|
||||
UNITS_PERF,
|
||||
};
|
||||
|
||||
const int blocksize = 4096;
|
||||
int iterations = 100000;
|
||||
|
||||
@ -31,14 +44,56 @@ static inline u64 read_tsc(void)
|
||||
return rdtsc();
|
||||
}
|
||||
|
||||
#define get_cycles() read_tsc()
|
||||
#define cpu_cycles() read_tsc()
|
||||
|
||||
#else
|
||||
|
||||
#define get_cycles() (0)
|
||||
#define cpu_cycles() (0)
|
||||
|
||||
#endif
|
||||
|
||||
#ifdef HAVE_PERF
|
||||
|
||||
static int perf_fd = -1;
|
||||
static int perf_init(void)
|
||||
{
|
||||
static struct perf_event_attr attr = {
|
||||
.type = PERF_TYPE_HARDWARE,
|
||||
.config = PERF_COUNT_HW_CPU_CYCLES
|
||||
};
|
||||
|
||||
perf_fd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
|
||||
return perf_fd;
|
||||
}
|
||||
|
||||
static void perf_finish(void)
|
||||
{
|
||||
close(perf_fd);
|
||||
}
|
||||
|
||||
static long long perf_cycles(void)
|
||||
{
|
||||
long long cycles;
|
||||
int ret;
|
||||
|
||||
ret = read(perf_fd, &cycles, sizeof(cycles));
|
||||
if (ret != sizeof(cycles))
|
||||
return 0;
|
||||
return cycles;
|
||||
}
|
||||
|
||||
#else
|
||||
static int perf_init()
|
||||
{
|
||||
errno = EOPNOTSUPP;
|
||||
return -1;
|
||||
}
|
||||
static void perf_finish() {}
|
||||
static long long perf_cycles() {
|
||||
return 0;
|
||||
}
|
||||
#endif
|
||||
|
||||
static inline u64 get_time(void)
|
||||
{
|
||||
struct timespec ts;
|
||||
@ -47,6 +102,16 @@ static inline u64 get_time(void)
|
||||
return ts.tv_sec * 1000 * 1000 * 1000 + ts.tv_nsec;
|
||||
}
|
||||
|
||||
static inline u64 get_cycles(int units)
|
||||
{
|
||||
switch (units) {
|
||||
case UNITS_CYCLES: return cpu_cycles();
|
||||
case UNITS_TIME: return get_time();
|
||||
case UNITS_PERF: return perf_cycles();
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
|
||||
/* Read the input and copy last bytes as the hash */
|
||||
static int hash_null_memcpy(const u8 *buf, size_t length, u8 *out)
|
||||
{
|
||||
@ -68,11 +133,22 @@ static int hash_null_nop(const u8 *buf, size_t length, u8 *out)
|
||||
return 0;
|
||||
}
|
||||
|
||||
const char *units_to_str(int units)
|
||||
static const char *units_to_desc(int units)
|
||||
{
|
||||
switch (units) {
|
||||
case 0: return "cycles";
|
||||
case 1: return "nsecs";
|
||||
case UNITS_CYCLES: return "CPU cycles";
|
||||
case UNITS_TIME: return "time: ns";
|
||||
case UNITS_PERF: return "perf event: CPU cycles";
|
||||
}
|
||||
return "unknown";
|
||||
}
|
||||
|
||||
static const char *units_to_str(int units)
|
||||
{
|
||||
switch (units) {
|
||||
case UNITS_CYCLES: return "cycles";
|
||||
case UNITS_TIME: return "nsecs";
|
||||
case UNITS_PERF: return "perf_c";
|
||||
}
|
||||
return "unknown";
|
||||
}
|
||||
@ -96,18 +172,19 @@ int main(int argc, char **argv) {
|
||||
{ .name = "SHA256", .digest = hash_sha256, .digest_size = 32 },
|
||||
{ .name = "BLAKE2", .digest = hash_blake2b, .digest_size = 32 },
|
||||
};
|
||||
int units = 0;
|
||||
int units = UNITS_CYCLES;
|
||||
|
||||
optind = 0;
|
||||
while (1) {
|
||||
static const struct option long_options[] = {
|
||||
{ "cycles", no_argument, NULL, 'c' },
|
||||
{ "time", no_argument, NULL, 't' },
|
||||
{ "perf", no_argument, NULL, 'p' },
|
||||
{ NULL, 0, NULL, 0}
|
||||
};
|
||||
int c;
|
||||
|
||||
c = getopt_long(argc, argv, "ct", long_options, NULL);
|
||||
c = getopt_long(argc, argv, "ctp", long_options, NULL);
|
||||
if (c < 0)
|
||||
break;
|
||||
switch (c) {
|
||||
@ -117,10 +194,18 @@ int main(int argc, char **argv) {
|
||||
"ERROR: cannot measure cycles on this arch, use --time\n");
|
||||
return 1;
|
||||
}
|
||||
units = 0;
|
||||
units = UNITS_CYCLES;
|
||||
break;
|
||||
case 't':
|
||||
units = 1;
|
||||
units = UNITS_TIME;
|
||||
break;
|
||||
case 'p':
|
||||
if (perf_init() == -1) {
|
||||
fprintf(stderr,
|
||||
"ERROR: cannot initialize perf, please check sysctl kernel.perf_event_paranoid: %m\n");
|
||||
return 1;
|
||||
}
|
||||
units = UNITS_PERF;
|
||||
break;
|
||||
default:
|
||||
fprintf(stderr, "ERROR: unknown option\n");
|
||||
@ -140,33 +225,33 @@ int main(int argc, char **argv) {
|
||||
printf("Block size: %d\n", blocksize);
|
||||
printf("Iterations: %d\n", iterations);
|
||||
printf("Implementation: %s\n", CRYPTOPROVIDER);
|
||||
printf("Units: %s\n", units_to_str(units));
|
||||
printf("Units: %s\n", units_to_desc(units));
|
||||
printf("\n");
|
||||
|
||||
for (idx = 0; idx < ARRAY_SIZE(contestants); idx++) {
|
||||
struct contestant *c = &contestants[idx];
|
||||
u64 start, end;
|
||||
u64 tstart, tend;
|
||||
u64 total;
|
||||
u64 total = 0;
|
||||
|
||||
printf("%12s: ", c->name);
|
||||
fflush(stdout);
|
||||
|
||||
tstart = get_time();
|
||||
start = get_cycles();
|
||||
start = get_cycles(units);
|
||||
for (iter = 0; iter < iterations; iter++) {
|
||||
memset(buf, iter & 0xFF, blocksize);
|
||||
memset(hash, 0, 32);
|
||||
c->digest(buf, blocksize, hash);
|
||||
}
|
||||
end = get_cycles();
|
||||
end = get_cycles(units);
|
||||
tend = get_time();
|
||||
c->cycles = end - start;
|
||||
c->time = tend - tstart;
|
||||
|
||||
if (units == 0)
|
||||
if (units == UNITS_CYCLES || units == UNITS_PERF)
|
||||
total = c->cycles;
|
||||
if (units == 1)
|
||||
if (units == UNITS_TIME)
|
||||
total = c->time;
|
||||
|
||||
printf("%s: %12llu, %s/i %8llu",
|
||||
@ -182,6 +267,7 @@ int main(int argc, char **argv) {
|
||||
}
|
||||
putchar('\n');
|
||||
}
|
||||
perf_finish();
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user