diff mbox series

New benchmark strlen-walk

Message ID 20180810192555.11657-1-siddhesh@sourceware.org
State New
Headers show
Series New benchmark strlen-walk | expand

Commit Message

Siddhesh Poyarekar Aug. 10, 2018, 7:25 p.m. UTC
Hi,

This is a second take at a strlen benchmark and it takes a different
approach from the previous linked list idea.

A comment in the test provides the rationale for the benchmark; to
summarize it focuses on testing strlen with small to medium sized
inputs with different sizes mixed in and walking backwards to try and
trick the prefetcher.  The numbers are kinda stable; I'm not super happy
but they're close enough to make out a general performance
characteristic.

	* benchtests/bench-strlen-walk.c: New benchmark.
	* benchtests/Makefile (string-benchset): Add it.
    
CC: Wilco.Dijkstra@arm.com
CC: carlos@redhat.com

---
 benchtests/Makefile            |   2 +-
 benchtests/bench-strlen-walk.c | 217 +++++++++++++++++++++++++++++++++
 2 files changed, 218 insertions(+), 1 deletion(-)
 create mode 100644 benchtests/bench-strlen-walk.c

Comments

Wilco Dijkstra Aug. 15, 2018, 2:58 p.m. UTC | #1
Siddhesh Poyarekar wrote:

> A comment in the test provides the rationale for the benchmark; to
> summarize it focuses on testing strlen with small to medium sized
> inputs with different sizes mixed in and walking backwards to try and
> trick the prefetcher.  The numbers are kinda stable; I'm not super happy
> but they're close enough to make out a general performance
> characteristic.

I don't believe walking backwards will fool a modern prefetcher. I would just
shuffle the array a bit (eg. swap arr[i] with arr[rand()%n] for all indices). Doing
that significantly increases runtime, showing that it is harder to prefetch.

Another issue is this:

+      cnt--;
+      if (cnt == 0)
+       {
+         cur_cnt = cur_cnt << 1;
+         cnt = cur_cnt;
+         maxlen >>= 1;
+         if (maxlen == 0)
+           maxlen = orig_maxlen;
+       }
+      len = maxlen + rand () % 8;

The result is that strings in length 1-16 are hugely overrepresented. If we
have N entries when maxlen is 16, we get N values in the range 16-24.
However we will get 30N values in the range 1-16 rather than the expected
2N. So the if should check maxlen==4 and reset (or something similar).

Wilco
Siddhesh Poyarekar Aug. 15, 2018, 5:29 p.m. UTC | #2
On 08/15/2018 08:28 PM, Wilco Dijkstra wrote:
> Siddhesh Poyarekar wrote:
> 
>> A comment in the test provides the rationale for the benchmark; to
>> summarize it focuses on testing strlen with small to medium sized
>> inputs with different sizes mixed in and walking backwards to try and
>> trick the prefetcher.  The numbers are kinda stable; I'm not super happy
>> but they're close enough to make out a general performance
>> characteristic.
> 
> I don't believe walking backwards will fool a modern prefetcher. I would just
> shuffle the array a bit (eg. swap arr[i] with arr[rand()%n] for all indices). Doing
> that significantly increases runtime, showing that it is harder to prefetch.

Let me try this and get back to you.

> Another issue is this:
> 
> +      cnt--;
> +      if (cnt == 0)
> +       {
> +         cur_cnt = cur_cnt << 1;
> +         cnt = cur_cnt;
> +         maxlen >>= 1;
> +         if (maxlen == 0)
> +           maxlen = orig_maxlen;
> +       }
> +      len = maxlen + rand () % 8;
> 
> The result is that strings in length 1-16 are hugely overrepresented. If we
> have N entries when maxlen is 16, we get N values in the range 16-24.
> However we will get 30N values in the range 1-16 rather than the expected
> 2N. So the if should check maxlen==4 and reset (or something similar).

I didn't work too hard to control this because of the assumption that < 
16 bytes is a very common size for strings passed to strlen() and a bit 
more bias towards it won't hurt.  I can try to control it a bit more.

Thanks,
Siddhesh
diff mbox series

Patch

diff --git a/benchtests/Makefile b/benchtests/Makefile
index bcd6a9c26d..31cacef373 100644
--- a/benchtests/Makefile
+++ b/benchtests/Makefile
@@ -43,7 +43,7 @@  string-benchset := bcopy bzero memccpy memchr memcmp memcpy memmem memmove \
 		   strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
 		   strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
 		   strcoll memcpy-large memcpy-random memmove-large memset-large \
-		   memcpy-walk memset-walk memmove-walk
+		   memcpy-walk memset-walk memmove-walk strlen-walk
 
 # Build and run locale-dependent benchmarks only if we're building natively.
 ifeq (no,$(cross-compiling))
diff --git a/benchtests/bench-strlen-walk.c b/benchtests/bench-strlen-walk.c
new file mode 100644
index 0000000000..1ac0ae6fdf
--- /dev/null
+++ b/benchtests/bench-strlen-walk.c
@@ -0,0 +1,217 @@ 
+/* Measure STRLEN functions - walk through a list of elements and measure
+   string lengths.
+   Copyright (C) 2018 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+/* RATIONALE
+   ---------
+
+   The following assumptions are made in this test about strlen usage in the
+   wild:
+
+   - Target strings are small or medium in size, rarely (if ever) very large
+   - In well written code, the target string is not in cache since strlen is
+     among the earliest operations on it.
+
+   This test measures the latency of strlen with a mix of sizes up to a maximum
+   length, for a set of maximum lengths.  The distribution of lengths in each
+   set is logarithmic, with the longest length having one occurrence, it's half
+   having 2, quarter having 4 and so on.  The lengths are further randomized by
+   adding a jitter of up to 8 bytes.  This makes the small string sets
+   completely arbitrary.
+
+   Further, The buffer and its index are reallocated at every run to ensure
+   that there is no cross talk between implementation.  Finally, the list of
+   strings is walked through backwards to try and trick the hardware
+   prefetcher.  There is an issue with this access too though, which is that
+   the most recently touched string ends up getting measured first when we go
+   from front to back, so that introduces some cache side effect.  */
+
+#define TEST_MAIN
+#define MIN_PAGE_SIZE (getpagesize () * 4096)
+#ifndef WIDE
+# define TEST_NAME "strlen"
+#else
+# define TEST_NAME "wcslen"
+#endif
+#include "bench-string.h"
+
+#ifndef WIDE
+# define STRLEN strlen
+# define CHAR char
+# define MAX_CHAR CHAR_MAX
+#else
+# include <wchar.h>
+# define STRLEN wcslen
+# define CHAR wchar_t
+# define MAX_CHAR WCHAR_MAX
+#endif
+
+#include "json-lib.h"
+
+typedef size_t (*proto_t) (const CHAR *);
+
+size_t
+simple_STRLEN (const CHAR *s)
+{
+  const CHAR *p;
+
+  for (p = s; *p; ++p);
+  return p - s;
+}
+
+#ifndef WIDE
+size_t
+builtin_strlen (const CHAR *p)
+{
+  return __builtin_strlen (p);
+}
+IMPL (builtin_strlen, 0)
+#endif
+
+IMPL (simple_STRLEN, 0)
+IMPL (STRLEN, 1)
+
+static unsigned char **str_index;
+
+static void
+do_one_test (json_ctx_t *json_ctx, impl_t *impl, size_t last_str_index)
+{
+  timing_t start, stop, cur;
+
+  TIMING_NOW (start);
+  for (int i = last_str_index - 1; i >= 0; i--)
+    CALL (impl, (char *) str_index[i]);
+  TIMING_NOW (stop);
+
+  TIMING_DIFF (cur, start, stop);
+
+  json_element_double (json_ctx, (double) cur / (double) last_str_index);
+}
+
+/* Split the buffer into strings and populate an str_index.  Return
+   the size of the str_index so that it can be iterated backwards   */
+static size_t
+setup_strings (size_t maxlen)
+{
+  unsigned char *p = buf1;
+  size_t orig_maxlen = maxlen, i = 0;
+  int cur_cnt, cnt;
+  size_t logn = 0, m = maxlen;
+
+  while ((m>>=1) > 0)
+    logn++;
+
+  /*Size of the index is buf_size*(2*M-1)/(M*ln(M)) where M is the max len and
+     we have a logarithmic distribution of string sizes, i.e. 1 of maxlen, 2 of
+     maxlen/2, 4 of maxlen/4 and so on.  Round up to avoid buffer overflows.  */
+  size_t index_size = (2 * buf1_size - buf1_size / orig_maxlen) / (logn + 1);
+  index_size *= sizeof (unsigned char *);
+
+  str_index = malloc (index_size);
+
+  if (str_index == NULL)
+    error (1, ENOMEM, "Out of memory\n");
+
+  srand (42);
+  cur_cnt = cnt = 1;
+
+  size_t len = maxlen + rand () % 8;
+
+  while (p < buf1 + buf1_size - len - 1)
+    {
+      str_index[i++] = p;
+      memset (p, 'a', len);
+      p[len] = '\0';
+      p += len + 1;
+
+      cnt--;
+      if (cnt == 0)
+	{
+	  cur_cnt = cur_cnt << 1;
+	  cnt = cur_cnt;
+	  maxlen >>= 1;
+	  if (maxlen == 0)
+	    maxlen = orig_maxlen;
+	}
+      len = maxlen + rand () % 8;
+    }
+
+  return i;
+}
+
+
+static void
+do_test (json_ctx_t *json_ctx, size_t len)
+{
+  json_element_object_begin (json_ctx);
+  json_attr_uint (json_ctx, "length", len);
+  json_array_begin (json_ctx, "timings");
+
+  /* Rebuild everything for each implementation so that we don't have cache
+     side effects across implementations.  */
+  FOR_EACH_IMPL (impl, 0)
+    {
+      size_t i = setup_strings (len);
+      do_one_test (json_ctx, impl, i);
+      alloc_bufs ();
+      free (str_index);
+    }
+
+  json_array_end (json_ctx);
+  json_element_object_end (json_ctx);
+}
+
+int
+test_main (void)
+{
+  json_ctx_t json_ctx;
+  size_t i;
+
+  test_init ();
+
+  json_init (&json_ctx, 0, stdout);
+
+  json_document_begin (&json_ctx);
+  json_attr_string (&json_ctx, "timing_type", TIMING_TYPE);
+
+  json_attr_object_begin (&json_ctx, "functions");
+  json_attr_object_begin (&json_ctx, TEST_NAME);
+  json_attr_string (&json_ctx, "bench-variant", "random");
+
+  json_array_begin (&json_ctx, "ifuncs");
+  FOR_EACH_IMPL (impl, 0)
+    json_element_string (&json_ctx, impl->name);
+  json_array_end (&json_ctx);
+
+  json_array_begin (&json_ctx, "results");
+
+  /* The maximum sizes to test.  These are arbitrary.  */
+  const size_t size_ranges[] = {64, 128, 512, 2048, 8192};
+
+  for (i = 0; i < sizeof (size_ranges) / sizeof (size_t); i++)
+    do_test (&json_ctx, size_ranges[i]);
+
+  json_array_end (&json_ctx);
+  json_attr_object_end (&json_ctx);
+  json_attr_object_end (&json_ctx);
+  json_document_end (&json_ctx);
+
+  return ret;
+}
+
+#include <support/test-driver.c>