linux/tools/perf/bench/mem-functions.c
Vincent Whitchurch 1beaef29c3 perf bench mem: Always memset source before memcpy
For memcpy, the source pages are memset to zero only when --cycles is
used.  This leads to wildly different results with or without --cycles,
since all sources pages are likely to be mapped to the same zero page
without explicit writes.

Before this fix:

$ export cmd="./perf stat -e LLC-loads -- ./perf bench \
  mem memcpy -s 1024MB -l 100 -f default"
$ $cmd

         2,935,826      LLC-loads
       3.821677452 seconds time elapsed

$ $cmd --cycles

       217,533,436      LLC-loads
       8.616725985 seconds time elapsed

After this fix:

$ $cmd

       214,459,686      LLC-loads
       8.674301124 seconds time elapsed

$ $cmd --cycles

       214,758,651      LLC-loads
       8.644480006 seconds time elapsed

Fixes: 47b5757bac ("perf bench mem: Move boilerplate memory allocation to the infrastructure")
Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: kernel@axis.com
Link: http://lore.kernel.org/lkml/20200810133404.30829-1-vincent.whitchurch@axis.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2020-08-13 09:34:26 -03:00

376 lines
8.7 KiB
C

// SPDX-License-Identifier: GPL-2.0
/*
* mem-memcpy.c
*
* Simple memcpy() and memset() benchmarks
*
* Written by Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
*/
#include "debug.h"
#include "../perf-sys.h"
#include <subcmd/parse-options.h>
#include "../util/header.h"
#include "../util/cloexec.h"
#include "../util/string2.h"
#include "bench.h"
#include "mem-memcpy-arch.h"
#include "mem-memset-arch.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/time.h>
#include <errno.h>
#include <linux/time64.h>
#include <linux/zalloc.h>
#define K 1024
static const char *size_str = "1MB";
static const char *function_str = "all";
static int nr_loops = 1;
static bool use_cycles;
static int cycles_fd;
static const struct option options[] = {
OPT_STRING('s', "size", &size_str, "1MB",
"Specify the size of the memory buffers. "
"Available units: B, KB, MB, GB and TB (case insensitive)"),
OPT_STRING('f', "function", &function_str, "all",
"Specify the function to run, \"all\" runs all available functions, \"help\" lists them"),
OPT_INTEGER('l', "nr_loops", &nr_loops,
"Specify the number of loops to run. (default: 1)"),
OPT_BOOLEAN('c', "cycles", &use_cycles,
"Use a cycles event instead of gettimeofday() to measure performance"),
OPT_END()
};
typedef void *(*memcpy_t)(void *, const void *, size_t);
typedef void *(*memset_t)(void *, int, size_t);
struct function {
const char *name;
const char *desc;
union {
memcpy_t memcpy;
memset_t memset;
} fn;
};
static struct perf_event_attr cycle_attr = {
.type = PERF_TYPE_HARDWARE,
.config = PERF_COUNT_HW_CPU_CYCLES
};
static int init_cycles(void)
{
cycles_fd = sys_perf_event_open(&cycle_attr, getpid(), -1, -1, perf_event_open_cloexec_flag());
if (cycles_fd < 0 && errno == ENOSYS) {
pr_debug("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
return -1;
}
return cycles_fd;
}
static u64 get_cycles(void)
{
int ret;
u64 clk;
ret = read(cycles_fd, &clk, sizeof(u64));
BUG_ON(ret != sizeof(u64));
return clk;
}
static double timeval2double(struct timeval *ts)
{
return (double)ts->tv_sec + (double)ts->tv_usec / (double)USEC_PER_SEC;
}
#define print_bps(x) do { \
if (x < K) \
printf(" %14lf bytes/sec\n", x); \
else if (x < K * K) \
printf(" %14lfd KB/sec\n", x / K); \
else if (x < K * K * K) \
printf(" %14lf MB/sec\n", x / K / K); \
else \
printf(" %14lf GB/sec\n", x / K / K / K); \
} while (0)
struct bench_mem_info {
const struct function *functions;
u64 (*do_cycles)(const struct function *r, size_t size, void *src, void *dst);
double (*do_gettimeofday)(const struct function *r, size_t size, void *src, void *dst);
const char *const *usage;
bool alloc_src;
};
static void __bench_mem_function(struct bench_mem_info *info, int r_idx, size_t size, double size_total)
{
const struct function *r = &info->functions[r_idx];
double result_bps = 0.0;
u64 result_cycles = 0;
void *src = NULL, *dst = zalloc(size);
printf("# function '%s' (%s)\n", r->name, r->desc);
if (dst == NULL)
goto out_alloc_failed;
if (info->alloc_src) {
src = zalloc(size);
if (src == NULL)
goto out_alloc_failed;
}
if (bench_format == BENCH_FORMAT_DEFAULT)
printf("# Copying %s bytes ...\n\n", size_str);
if (use_cycles) {
result_cycles = info->do_cycles(r, size, src, dst);
} else {
result_bps = info->do_gettimeofday(r, size, src, dst);
}
switch (bench_format) {
case BENCH_FORMAT_DEFAULT:
if (use_cycles) {
printf(" %14lf cycles/byte\n", (double)result_cycles/size_total);
} else {
print_bps(result_bps);
}
break;
case BENCH_FORMAT_SIMPLE:
if (use_cycles) {
printf("%lf\n", (double)result_cycles/size_total);
} else {
printf("%lf\n", result_bps);
}
break;
default:
BUG_ON(1);
break;
}
out_free:
free(src);
free(dst);
return;
out_alloc_failed:
printf("# Memory allocation failed - maybe size (%s) is too large?\n", size_str);
goto out_free;
}
static int bench_mem_common(int argc, const char **argv, struct bench_mem_info *info)
{
int i;
size_t size;
double size_total;
argc = parse_options(argc, argv, options, info->usage, 0);
if (use_cycles) {
i = init_cycles();
if (i < 0) {
fprintf(stderr, "Failed to open cycles counter\n");
return i;
}
}
size = (size_t)perf_atoll((char *)size_str);
size_total = (double)size * nr_loops;
if ((s64)size <= 0) {
fprintf(stderr, "Invalid size:%s\n", size_str);
return 1;
}
if (!strncmp(function_str, "all", 3)) {
for (i = 0; info->functions[i].name; i++)
__bench_mem_function(info, i, size, size_total);
return 0;
}
for (i = 0; info->functions[i].name; i++) {
if (!strcmp(info->functions[i].name, function_str))
break;
}
if (!info->functions[i].name) {
if (strcmp(function_str, "help") && strcmp(function_str, "h"))
printf("Unknown function: %s\n", function_str);
printf("Available functions:\n");
for (i = 0; info->functions[i].name; i++) {
printf("\t%s ... %s\n",
info->functions[i].name, info->functions[i].desc);
}
return 1;
}
__bench_mem_function(info, i, size, size_total);
return 0;
}
static void memcpy_prefault(memcpy_t fn, size_t size, void *src, void *dst)
{
/* Make sure to always prefault zero pages even if MMAP_THRESH is crossed: */
memset(src, 0, size);
/*
* We prefault the freshly allocated memory range here,
* to not measure page fault overhead:
*/
fn(dst, src, size);
}
static u64 do_memcpy_cycles(const struct function *r, size_t size, void *src, void *dst)
{
u64 cycle_start = 0ULL, cycle_end = 0ULL;
memcpy_t fn = r->fn.memcpy;
int i;
memcpy_prefault(fn, size, src, dst);
cycle_start = get_cycles();
for (i = 0; i < nr_loops; ++i)
fn(dst, src, size);
cycle_end = get_cycles();
return cycle_end - cycle_start;
}
static double do_memcpy_gettimeofday(const struct function *r, size_t size, void *src, void *dst)
{
struct timeval tv_start, tv_end, tv_diff;
memcpy_t fn = r->fn.memcpy;
int i;
memcpy_prefault(fn, size, src, dst);
BUG_ON(gettimeofday(&tv_start, NULL));
for (i = 0; i < nr_loops; ++i)
fn(dst, src, size);
BUG_ON(gettimeofday(&tv_end, NULL));
timersub(&tv_end, &tv_start, &tv_diff);
return (double)(((double)size * nr_loops) / timeval2double(&tv_diff));
}
struct function memcpy_functions[] = {
{ .name = "default",
.desc = "Default memcpy() provided by glibc",
.fn.memcpy = memcpy },
#ifdef HAVE_ARCH_X86_64_SUPPORT
# define MEMCPY_FN(_fn, _name, _desc) {.name = _name, .desc = _desc, .fn.memcpy = _fn},
# include "mem-memcpy-x86-64-asm-def.h"
# undef MEMCPY_FN
#endif
{ .name = NULL, }
};
static const char * const bench_mem_memcpy_usage[] = {
"perf bench mem memcpy <options>",
NULL
};
int bench_mem_memcpy(int argc, const char **argv)
{
struct bench_mem_info info = {
.functions = memcpy_functions,
.do_cycles = do_memcpy_cycles,
.do_gettimeofday = do_memcpy_gettimeofday,
.usage = bench_mem_memcpy_usage,
.alloc_src = true,
};
return bench_mem_common(argc, argv, &info);
}
static u64 do_memset_cycles(const struct function *r, size_t size, void *src __maybe_unused, void *dst)
{
u64 cycle_start = 0ULL, cycle_end = 0ULL;
memset_t fn = r->fn.memset;
int i;
/*
* We prefault the freshly allocated memory range here,
* to not measure page fault overhead:
*/
fn(dst, -1, size);
cycle_start = get_cycles();
for (i = 0; i < nr_loops; ++i)
fn(dst, i, size);
cycle_end = get_cycles();
return cycle_end - cycle_start;
}
static double do_memset_gettimeofday(const struct function *r, size_t size, void *src __maybe_unused, void *dst)
{
struct timeval tv_start, tv_end, tv_diff;
memset_t fn = r->fn.memset;
int i;
/*
* We prefault the freshly allocated memory range here,
* to not measure page fault overhead:
*/
fn(dst, -1, size);
BUG_ON(gettimeofday(&tv_start, NULL));
for (i = 0; i < nr_loops; ++i)
fn(dst, i, size);
BUG_ON(gettimeofday(&tv_end, NULL));
timersub(&tv_end, &tv_start, &tv_diff);
return (double)(((double)size * nr_loops) / timeval2double(&tv_diff));
}
static const char * const bench_mem_memset_usage[] = {
"perf bench mem memset <options>",
NULL
};
static const struct function memset_functions[] = {
{ .name = "default",
.desc = "Default memset() provided by glibc",
.fn.memset = memset },
#ifdef HAVE_ARCH_X86_64_SUPPORT
# define MEMSET_FN(_fn, _name, _desc) { .name = _name, .desc = _desc, .fn.memset = _fn },
# include "mem-memset-x86-64-asm-def.h"
# undef MEMSET_FN
#endif
{ .name = NULL, }
};
int bench_mem_memset(int argc, const char **argv)
{
struct bench_mem_info info = {
.functions = memset_functions,
.do_cycles = do_memset_cycles,
.do_gettimeofday = do_memset_gettimeofday,
.usage = bench_mem_memset_usage,
};
return bench_mem_common(argc, argv, &info);
}