lkml.org 
[lkml]   [2009]   [Nov]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH 1/4] perf bench: Add new subsystem and new suite, bench/mem-memcpy.c
Date
This patch adds bench/mem-memcpy.c.
This file provides new subsystem "mem": evaluating for memory performance,
and new suite "memcpy": measurements performance of memcpy(2) like function.

bench/mem-memcpy.c will be start point for comparing
different algorithms of memcpy() on same CPU
or same memcpy() on different CPUs.
Current supported memcpy() is memcpy() provided by glibc.

Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ling Ma <ling.ma@intel.com>

| % perf bench mem memcpy -l 1GB -c # Measure in clock
| # Running mem/memcpy benchmark...
| # Copying 1GB Bytes from 0x7f53f8c25010 to 0x7f5438c26010 ...
|
| 2.156751 Clock/Byte
| % perf bench mem memcpy -l 1GB # Measure in time(default)
| # Running mem/memcpy benchmark...
| # Copying 1GB Bytes from 0x7f2cffefb010 to 0x7f2d3fefc010 ...
|
| 1.415502 GB/Sec
---
tools/perf/bench/mem-memcpy.c | 258 +++++++++++++++++++++++++++++++++++++++++
1 files changed, 258 insertions(+), 0 deletions(-)
create mode 100644 tools/perf/bench/mem-memcpy.c

diff --git a/tools/perf/bench/mem-memcpy.c b/tools/perf/bench/mem-memcpy.c
new file mode 100644
index 0000000..dd9cbc6
--- /dev/null
+++ b/tools/perf/bench/mem-memcpy.c
@@ -0,1 +1,258 @@
+/*
+ *
+ * mem-memcpy.c
+ *
+ * memcpy: Simple memory copy in various ways
+ *
+ * Based on memcpy.c by Ma Ling <ling.ma@intel.com>
+ * http://marc.info/?l=linux-kernel&m=125792321123782&w=2
+ * This memcpy.c is posted to LKML by Ma Ling for comparing
+ * two ways of memory copying.
+ * The thread is started from
+ * http://marc.info/?l=linux-kernel&m=125750023424093&w=2
+ *
+ * Ported to perf by Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
+ *
+ */
+
+#include <ctype.h>
+
+#include "../perf.h"
+#include "../util/util.h"
+#include "../util/parse-options.h"
+#include "../builtin.h"
+#include "bench.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/time.h>
+#include <assert.h>
+
+#define K 1024
+static char *length_str = (char *)"1MB";
+static char *routine = (char *)"default";
+static int use_clockcycle = 0;
+
+typedef unsigned long int clockcycle_t;
+
+#ifdef x86_64
+
+static inline clockcycle_t get_clock(void)
+{
+ long int ret;
+
+ asm("rdtsc; shlq $32, %%rdx;"
+ "orq %%rdx, %%rax;"
+ "movq %%rax, %0;"
+ : "=r" (ret));
+
+ return ret;
+}
+
+#endif /* x86_64 */
+
+static const struct option options[] = {
+ OPT_STRING('l', "length", &length_str, "1MB",
+ "Specify length of memory to copy. "
+ "available unit: B, MB, GB (upper and lower)"),
+ OPT_STRING('r', "routine", &routine, "default",
+ "Specify routine to copy"),
+#ifdef x86_64
+ /*
+ * TODO: This should be expanded to any architecuture
+ * perf supports
+ */
+ OPT_BOOLEAN('c', "clockcycle", &use_clockcycle,
+ "Use CPU's clock cycle for measurement"),
+#endif /* x86_64 */
+ OPT_END()
+};
+
+struct routine {
+ const char *name;
+ void * (*fn)(void *dst, const void *src, size_t len);
+ const char *desc;
+};
+
+struct routine routines[] = {
+ { "default",
+ memcpy,
+ "Default memcpy() provided by glibc" },
+ { NULL,
+ NULL,
+ NULL }
+};
+
+static const char * const bench_mem_memcpy_usage[] = {
+ "perf bench mem memcpy <options>",
+ NULL
+};
+
+static size_t str2length(char *_str)
+{
+ int i, unit = 1;
+ char *str;
+ size_t length = -1;
+
+ str = calloc(strlen(_str) + 1, sizeof(char));
+ assert(str);
+ strcpy(str, _str);
+
+ if (!isdigit(str[0]))
+ goto err;
+
+ for (i = 1; i < (int)strlen(str); i++) {
+ switch ((int)str[i]) {
+ case 'B':
+ case 'b':
+ str[i] = '\0';
+ break;
+ case 'K':
+ case 'k':
+ if (str[i + 1] != 'B' && str[i + 1] != 'b')
+ goto err;
+ unit = K;
+ str[i] = '\0';
+ break;
+ case 'M':
+ case 'm':
+ if (str[i + 1] != 'B' && str[i + 1] != 'b')
+ goto err;
+ unit = K * K;
+ str[i] = '\0';
+ break;
+ case 'G':
+ case 'g':
+ if (str[i + 1] != 'B' && str[i + 1] != 'b')
+ goto err;
+ unit = K * K * K;
+ str[i] = '\0';
+ break;
+ case '\0': /* only specified figures */
+ unit = 1;
+ break;
+ default:
+ if (!isdigit(str[i]))
+ goto err;
+ break;
+ }
+ }
+
+ length = atoi(str) * unit;
+ goto end;
+
+err:
+ fprintf(stderr, "Invalid length:%s\n", str);
+end:
+ free(str);
+ return length;
+}
+
+static double timeval2double(struct timeval *ts)
+{
+ return (double)ts->tv_sec +
+ (double)ts->tv_usec / (double)1000000;
+}
+
+int bench_mem_memcpy(int argc, const char **argv,
+ const char *prefix __used)
+{
+ int i;
+ void *dst, *src;
+ struct timeval start, stop, diff;
+ clockcycle_t clock_start = 0, clock_diff = 0;
+ size_t length;
+ double bps = 0.0;
+
+ argc = parse_options(argc, argv, options,
+ bench_mem_memcpy_usage, 0);
+
+ /*
+ * Caution!
+ * Without the statement
+ * gettimeofday(&diff, NULL);
+ * compiler warns (and build environment of perf regards it as error)
+ * like this,
+ * bench/mem-memcpy.c:93: error: ‘diff.tv_sec’ may be\
+ * used uninitialized in this function
+ * bench/mem-memcpy.c:93: error: ‘diff.tv_usec’ may be\
+ * used uninitialized in this function
+ *
+ * hmm...
+ */
+ gettimeofday(&diff, NULL);
+
+ length = str2length(length_str);
+ if ((int)length < 0)
+ return 1;
+
+ for (i = 0; routines[i].name; i++)
+ if (!strcmp(routines[i].name, routine))
+ break;
+ if (!routines[i].name) {
+ printf("Unknown routine:%s\n", routine);
+ printf("Available routines...\n");
+ for (i = 0; routines[i].name; i++)
+ printf("\t%s ... %s\n",
+ routines[i].name, routines[i].desc);
+ return 1;
+ }
+
+ dst = calloc(length, sizeof(char));
+ assert(dst);
+ src = calloc(length, sizeof(char));
+ assert(src);
+
+ if (bench_format == BENCH_FORMAT_DEFAULT)
+ printf("# Copying %s Bytes from %p to %p ...\n\n",
+ length_str, src, dst);
+
+ if (use_clockcycle) {
+ clock_start = get_clock();
+ } else {
+ gettimeofday(&start, NULL);
+ }
+
+ routines[i].fn(dst, src, length);
+
+ if (use_clockcycle) {
+ clock_diff = get_clock() - clock_start;
+ } else {
+ gettimeofday(&stop, NULL);
+ timersub(&stop, &start, &diff);
+ bps = (double)((double)length / timeval2double(&diff));
+ }
+
+ switch (bench_format) {
+ case BENCH_FORMAT_DEFAULT:
+ if (use_clockcycle)
+ printf(" %14lf Clock/Byte\n",
+ (double)clock_diff / (double)length);
+ else
+ if (bps < K)
+ printf(" %14lf B/Sec\n", bps);
+ else if (bps < K * K)
+ printf(" %14lfd KB/Sec\n", bps / 1024);
+ else if (bps < K * K * K)
+ printf(" %14lf MB/Sec\n", bps / 1024 / 1024);
+ else
+ printf(" %14lf GB/Sec\n",
+ bps / 1024 / 1024 / 1024);
+ break;
+ case BENCH_FORMAT_SIMPLE:
+ if (use_clockcycle)
+ printf("%lf\n",
+ (double)clock_diff / (double)length);
+ else
+ printf("%lf\n", bps);
+ break;
+ default:
+ /* reaching here is something disaster */
+ fprintf(stderr, "Unknown format:%d\n", bench_format);
+ exit(1);
+ break;
+ }
+
+ return 0;
+}
--
1.6.5.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2009-11-13 05:27    [from the cache]
©2003-2011 Jasper Spaans