[v5] benchtests: Add memset zero fill benchtest
Checks
Context |
Check |
Description |
dj/TryBot-apply_patch |
success
|
Patch applied to master at the time it was sent
|
dj/TryBot-32bit |
success
|
Build for i686
|
Commit Message
Memset takes 0 as the second parameter in most cases.
However, we cannot measure the zero fill performance by
bench-memset.c, bench-memset-large.c and bench-memset-walk.c
precisely.
X86_64 micro-architecture has some zero-over-zero optimization, and
AArch64 micro-architecture also has some optimization for DC ZVA
instruction.
This patch provides bench-memset-zerofill.c which is suitable to
analyze the zero fill performance by comparing among 4 patterns,
zero-over-zero, zero-over-one, one-over-zero and one-over-one through
L1, L2 and L3 caches.
The following commands are examples to analyze a JSON output,
bench-memset-zerofill.out, by 'jq' and 'plot_strings.py'.
1) compare zero-over-zero performance
$ cat bench-memset-zerofill.out | \
jq -r '
.functions.memset."bench-variant"="zerofill-0o0" |
del(.functions.memset.results[] | select(.char1 != 0 or .char2 != 0))
' | \
plot_strings.py -l -p thru -v -
2) compare zero paformance
$ cat bench-memset-zerofill.out | \
jq -r '
.functions.memset."bench-variant"="zerofill-zero" |
del(.functions.memset.results[] | select(.char2 != 0))
' | \
plot_strings.py -l -p thru -v -
3) compare nonzero paformance
$ cat bench-memset-zerofill.out | \
jq -r '
.functions.memset."bench-variant"="zerofill-nonzero" |
del(.functions.memset.results[] | select(.char2 == 0))
' | \
plot_strings.py -l -p thru -v -
Reviewed-by: Lucas A. M. Magalhaes <lamm@linux.ibm.com>
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
---
benchtests/Makefile | 2 +-
benchtests/bench-memset-zerofill.c | 140 +++++++++++++++++++++++++++++
2 files changed, 141 insertions(+), 1 deletion(-)
create mode 100644 benchtests/bench-memset-zerofill.c
Comments
On 9/21/21 21:05, Naohiro Tamura via Libc-alpha wrote:
> Memset takes 0 as the second parameter in most cases.
> However, we cannot measure the zero fill performance by
> bench-memset.c, bench-memset-large.c and bench-memset-walk.c
> precisely.
> X86_64 micro-architecture has some zero-over-zero optimization, and
> AArch64 micro-architecture also has some optimization for DC ZVA
> instruction.
> This patch provides bench-memset-zerofill.c which is suitable to
> analyze the zero fill performance by comparing among 4 patterns,
> zero-over-zero, zero-over-one, one-over-zero and one-over-one through
> L1, L2 and L3 caches.
As I aim to drive the SLI to 90-days for patch review... this is next on my list.
We have 3 reviewed-by aggregations here, but we have not yet committed this patch.
Where did we land here? Is this ready to be included in the microbenchmark?
> The following commands are examples to analyze a JSON output,
> bench-memset-zerofill.out, by 'jq' and 'plot_strings.py'.
>
> 1) compare zero-over-zero performance
>
> $ cat bench-memset-zerofill.out | \
> jq -r '
> .functions.memset."bench-variant"="zerofill-0o0" |
> del(.functions.memset.results[] | select(.char1 != 0 or .char2 != 0))
> ' | \
> plot_strings.py -l -p thru -v -
>
> 2) compare zero paformance
>
> $ cat bench-memset-zerofill.out | \
> jq -r '
> .functions.memset."bench-variant"="zerofill-zero" |
> del(.functions.memset.results[] | select(.char2 != 0))
> ' | \
> plot_strings.py -l -p thru -v -
>
> 3) compare nonzero paformance
>
> $ cat bench-memset-zerofill.out | \
> jq -r '
> .functions.memset."bench-variant"="zerofill-nonzero" |
> del(.functions.memset.results[] | select(.char2 == 0))
> ' | \
> plot_strings.py -l -p thru -v -
>
> Reviewed-by: Lucas A. M. Magalhaes <lamm@linux.ibm.com>
> Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
> Reviewed-by: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
> ---
> benchtests/Makefile | 2 +-
> benchtests/bench-memset-zerofill.c | 140 +++++++++++++++++++++++++++++
> 2 files changed, 141 insertions(+), 1 deletion(-)
> create mode 100644 benchtests/bench-memset-zerofill.c
>
> diff --git a/benchtests/Makefile b/benchtests/Makefile
> index 1530939a8ce8..21b95c736190 100644
> --- a/benchtests/Makefile
> +++ b/benchtests/Makefile
> @@ -53,7 +53,7 @@ string-benchset := 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 memset-zerofill
>
> # Build and run locale-dependent benchmarks only if we're building natively.
> ifeq (no,$(cross-compiling))
> diff --git a/benchtests/bench-memset-zerofill.c b/benchtests/bench-memset-zerofill.c
> new file mode 100644
> index 000000000000..2e146e157e25
> --- /dev/null
> +++ b/benchtests/bench-memset-zerofill.c
> @@ -0,0 +1,140 @@
> +/* Measure memset functions with zero fill data.
> + Copyright (C) 2021 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
> + <https://www.gnu.org/licenses/>. */
> +
> +#define TEST_MAIN
> +#define TEST_NAME "memset"
> +#define START_SIZE (16 * 1024)
> +#define MIN_PAGE_SIZE (getpagesize () + 16 * 1024 * 1024)
> +#define TIMEOUT (20 * 60)
> +#include "bench-string.h"
> +
> +#include "json-lib.h"
> +
> +void *generic_memset (void *, int, size_t);
> +typedef void *(*proto_t) (void *, int, size_t);
> +
> +IMPL (MEMSET, 1)
> +IMPL (generic_memset, 0)
> +
> +static void
> +__attribute__((noinline, noclone))
> +do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s,
> + int c1 __attribute ((unused)), int c2 __attribute ((unused)),
> + size_t n)
> +{
> + size_t i, iters = 32;
> + timing_t start, stop, cur, latency = 0;
> +
> + CALL (impl, s, c2, n); // warm up
> +
> + for (i = 0; i < iters; i++)
> + {
> + memset (s, c1, n); // alternation
> +
> + TIMING_NOW (start);
> +
> + CALL (impl, s, c2, n);
> +
> + TIMING_NOW (stop);
> + TIMING_DIFF (cur, start, stop);
> + TIMING_ACCUM (latency, cur);
> + }
> +
> + json_element_double (json_ctx, (double) latency / (double) iters);
> +}
> +
> +static void
> +do_test (json_ctx_t *json_ctx, size_t align, int c1, int c2, size_t len)
> +{
> + align &= getpagesize () - 1;
> + if ((align + len) * sizeof (CHAR) > page_size)
> + return;
> +
> + json_element_object_begin (json_ctx);
> + json_attr_uint (json_ctx, "length", len);
> + json_attr_uint (json_ctx, "alignment", align);
> + json_attr_int (json_ctx, "char1", c1);
> + json_attr_int (json_ctx, "char2", c2);
> + json_array_begin (json_ctx, "timings");
> +
> + FOR_EACH_IMPL (impl, 0)
> + {
> + do_one_test (json_ctx, impl, (CHAR *) (buf1) + align, c1, c2, len);
> + alloc_bufs ();
> + }
> +
> + json_array_end (json_ctx);
> + json_element_object_end (json_ctx);
> +}
> +
> +int
> +test_main (void)
> +{
> + json_ctx_t json_ctx;
> + size_t i;
> + int c1, c2;
> +
> + 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", "zerofill");
> +
> + 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");
> +
> + // To analyze zero fill performance by comparing among the following 4
> + // patterns through L1, L2 and L3 caches.
> + // - zero-over-zero: c1=0, c2=0
> + // - zero-over-one: c1=0, c2=1
> + // - one-over-zero: c1=1, c2=0
> + // - one-over-one: c1=1, c2=1
> + for (c1 = 0; c1 < 2; c1++)
> + for (c2 = 0; c2 < 2; c2++)
> + for (i = START_SIZE; i <= MIN_PAGE_SIZE; i <<= 1)
> + {
> + do_test (&json_ctx, 0, c1, c2, i);
> + do_test (&json_ctx, 3, c1, c2, 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>
> +
> +#define libc_hidden_builtin_def(X)
> +#define libc_hidden_def(X)
> +#define libc_hidden_weak(X)
> +#define weak_alias(X,Y)
> +#undef MEMSET
> +#define MEMSET generic_memset
> +#include <string/memset.c>
On Thu, Feb 9, 2023 at 12:23 PM Carlos O'Donell <carlos@redhat.com> wrote:
>
> On 9/21/21 21:05, Naohiro Tamura via Libc-alpha wrote:
> > Memset takes 0 as the second parameter in most cases.
> > However, we cannot measure the zero fill performance by
> > bench-memset.c, bench-memset-large.c and bench-memset-walk.c
> > precisely.
> > X86_64 micro-architecture has some zero-over-zero optimization, and
> > AArch64 micro-architecture also has some optimization for DC ZVA
> > instruction.
> > This patch provides bench-memset-zerofill.c which is suitable to
> > analyze the zero fill performance by comparing among 4 patterns,
> > zero-over-zero, zero-over-one, one-over-zero and one-over-one through
> > L1, L2 and L3 caches.
>
> As I aim to drive the SLI to 90-days for patch review... this is next on my list.
>
> We have 3 reviewed-by aggregations here, but we have not yet committed this patch.
>
> Where did we land here? Is this ready to be included in the microbenchmark?
I dug around a bit and found that the conversation for this v5
continued in the v3 thread[1] and AFAICT, the conclusion was to drop
this patch.
Sid
[1] https://patchwork.sourceware.org/project/glibc/patch/20210805075053.433538-1-naohirot@fujitsu.com/
@@ -53,7 +53,7 @@ string-benchset := 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 memset-zerofill
# Build and run locale-dependent benchmarks only if we're building natively.
ifeq (no,$(cross-compiling))
new file mode 100644
@@ -0,0 +1,140 @@
+/* Measure memset functions with zero fill data.
+ Copyright (C) 2021 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
+ <https://www.gnu.org/licenses/>. */
+
+#define TEST_MAIN
+#define TEST_NAME "memset"
+#define START_SIZE (16 * 1024)
+#define MIN_PAGE_SIZE (getpagesize () + 16 * 1024 * 1024)
+#define TIMEOUT (20 * 60)
+#include "bench-string.h"
+
+#include "json-lib.h"
+
+void *generic_memset (void *, int, size_t);
+typedef void *(*proto_t) (void *, int, size_t);
+
+IMPL (MEMSET, 1)
+IMPL (generic_memset, 0)
+
+static void
+__attribute__((noinline, noclone))
+do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s,
+ int c1 __attribute ((unused)), int c2 __attribute ((unused)),
+ size_t n)
+{
+ size_t i, iters = 32;
+ timing_t start, stop, cur, latency = 0;
+
+ CALL (impl, s, c2, n); // warm up
+
+ for (i = 0; i < iters; i++)
+ {
+ memset (s, c1, n); // alternation
+
+ TIMING_NOW (start);
+
+ CALL (impl, s, c2, n);
+
+ TIMING_NOW (stop);
+ TIMING_DIFF (cur, start, stop);
+ TIMING_ACCUM (latency, cur);
+ }
+
+ json_element_double (json_ctx, (double) latency / (double) iters);
+}
+
+static void
+do_test (json_ctx_t *json_ctx, size_t align, int c1, int c2, size_t len)
+{
+ align &= getpagesize () - 1;
+ if ((align + len) * sizeof (CHAR) > page_size)
+ return;
+
+ json_element_object_begin (json_ctx);
+ json_attr_uint (json_ctx, "length", len);
+ json_attr_uint (json_ctx, "alignment", align);
+ json_attr_int (json_ctx, "char1", c1);
+ json_attr_int (json_ctx, "char2", c2);
+ json_array_begin (json_ctx, "timings");
+
+ FOR_EACH_IMPL (impl, 0)
+ {
+ do_one_test (json_ctx, impl, (CHAR *) (buf1) + align, c1, c2, len);
+ alloc_bufs ();
+ }
+
+ json_array_end (json_ctx);
+ json_element_object_end (json_ctx);
+}
+
+int
+test_main (void)
+{
+ json_ctx_t json_ctx;
+ size_t i;
+ int c1, c2;
+
+ 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", "zerofill");
+
+ 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");
+
+ // To analyze zero fill performance by comparing among the following 4
+ // patterns through L1, L2 and L3 caches.
+ // - zero-over-zero: c1=0, c2=0
+ // - zero-over-one: c1=0, c2=1
+ // - one-over-zero: c1=1, c2=0
+ // - one-over-one: c1=1, c2=1
+ for (c1 = 0; c1 < 2; c1++)
+ for (c2 = 0; c2 < 2; c2++)
+ for (i = START_SIZE; i <= MIN_PAGE_SIZE; i <<= 1)
+ {
+ do_test (&json_ctx, 0, c1, c2, i);
+ do_test (&json_ctx, 3, c1, c2, 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>
+
+#define libc_hidden_builtin_def(X)
+#define libc_hidden_def(X)
+#define libc_hidden_weak(X)
+#define weak_alias(X,Y)
+#undef MEMSET
+#define MEMSET generic_memset
+#include <string/memset.c>