[PATCH v2] Benchtests: Improve large memcpy/memset benchmarks
Adhemerval Zanella Netto
adhemerval.zanella@linaro.org
Tue Mar 4 15:20:34 GMT 2025
On 20/02/25 12:12, Wilco Dijkstra wrote:
>
>
> ping
>
>
> Hi Carlos,
>
>> Would it be possible for you to post a v2 that rolls up the feedback
>> from Noah and HJ here?
>>
>> I know that some of their feedback you have chosen not to integrate,
>> but any other changes you made would be good to get into the tree.
>
> Sure v2 is below. This doesn't include fixes for page_size = getpagesize ()
> since that affects a lot more benchmarks and incorrect range checks
> (there are many uses of 4095 and 4096 for page alignment), so I've kept the
> rest the same as v1.
>
> Cheers,
> Wilco
>
>
> v2: Use getpagesize () as suggested.
>
> Adjust sizes between 64KB and 16MB and iterations based on length.
> Remove incorrect uses of alloc_bufs since we're not interested in measuring
> Linux clear_page time. Use getpagesize() - 1 instead of 4095 when
> aligning within a page.
>
LGTM, thanks.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
> ---
>
> diff --git a/benchtests/bench-bzero-large.c b/benchtests/bench-bzero-large.c
> index 71300e878afbfcc92c378d41487e469fde08093a..2eebd08315d75a23f19a7b750493d1a63ce0319d 100644
> --- a/benchtests/bench-bzero-large.c
> +++ b/benchtests/bench-bzero-large.c
> @@ -22,9 +22,8 @@
> #else
> # define TEST_NAME "bzero"
> #endif
> -#define START_SIZE (128 * 1024)
> -#define MIN_PAGE_SIZE (getpagesize () + 64 * 1024 * 1024)
> -#define TIMEOUT (20 * 60)
> +#define START_SIZE (64 * 1024)
> +#define MIN_PAGE_SIZE (getpagesize () + 16 * 1024 * 1024)
> #include "bench-string.h"
>
> #include "json-lib.h"
> @@ -52,7 +51,7 @@ IMPL (memset_zero, 0)
> static void
> do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s, size_t n)
> {
> - size_t i, iters = 16;
> + size_t i, iters = (MIN_PAGE_SIZE * 64) / n;
> timing_t start, stop, cur;
>
> TIMING_NOW (start);
> @@ -74,20 +73,13 @@ do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s, size_t n)
> static void
> do_test (json_ctx_t *json_ctx, size_t align, size_t len)
> {
> - align &= 63;
> - 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_array_begin (json_ctx, "timings");
>
> FOR_EACH_IMPL (impl, 0)
> - {
> - do_one_test (json_ctx, impl, (CHAR *) (buf1) + align, len);
> - alloc_bufs ();
> - }
> + do_one_test (json_ctx, impl, (CHAR *) (buf1) + align, len);
>
> json_array_end (json_ctx);
> json_element_object_end (json_ctx);
> diff --git a/benchtests/bench-bzero.c b/benchtests/bench-bzero.c
> index d0497dd8b0ab852ab661cac6df82202b0d5969f5..637b4f765eeac2a7bcdac35a785060e8afbbbc35 100644
> --- a/benchtests/bench-bzero.c
> +++ b/benchtests/bench-bzero.c
> @@ -73,7 +73,7 @@ do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s, size_t n)
> static void
> do_test (json_ctx_t *json_ctx, size_t align, size_t len)
> {
> - align &= 4095;
> + align &= getpagesize () - 1;
> if ((align + len) * sizeof (CHAR) > page_size)
> return;
>
> diff --git a/benchtests/bench-memcpy-large.c b/benchtests/bench-memcpy-large.c
> index 713588e8ce8835d382b9e8e781e351ad21b54418..acdc175a7843acade1defa0244b35f003bac8241 100644
> --- a/benchtests/bench-memcpy-large.c
> +++ b/benchtests/bench-memcpy-large.c
> @@ -19,10 +19,9 @@
> #ifndef MEMCPY_RESULT
> # define MEMCPY_RESULT(dst, len) dst
> # define START_SIZE (64 * 1024)
> -# define MIN_PAGE_SIZE (getpagesize () + 32 * 1024 * 1024)
> +# define MIN_PAGE_SIZE (getpagesize () + 16 * 1024 * 1024)
> # define TEST_MAIN
> # define TEST_NAME "memcpy"
> -# define TIMEOUT (20 * 60)
> # include "bench-string.h"
>
> IMPL (memcpy, 1)
> @@ -36,7 +35,7 @@ static void
> do_one_test (json_ctx_t *json_ctx, impl_t *impl, char *dst, const char *src,
> size_t len)
> {
> - size_t i, iters = 16;
> + size_t i, iters = (MIN_PAGE_SIZE * 8) / len;
> timing_t start, stop, cur;
>
> TIMING_NOW (start);
> @@ -58,13 +57,8 @@ do_test (json_ctx_t *json_ctx, size_t align1, size_t align2, size_t len,
> size_t i, j;
> char *s1, *s2;
> size_t repeats;
> - align1 &= 4095;
> - if (align1 + len >= page_size)
> - return;
> -
> - align2 &= 4095;
> - if (align2 + len >= page_size)
> - return;
> + align1 &= getpagesize () - 1;
> + align2 &= getpagesize () - 1;
>
> s1 = (char *) (buf1 + align1);
> s2 = (char *) (buf2 + align2);
> diff --git a/benchtests/bench-memmove-large.c b/benchtests/bench-memmove-large.c
> index cc0145534de690652ce05e07608470ea3cf0efd1..c660288f96d5eaeddc5b83ef53a17efe1c572ae8 100644
> --- a/benchtests/bench-memmove-large.c
> +++ b/benchtests/bench-memmove-large.c
> @@ -16,12 +16,10 @@
> License along with the GNU C Library; if not, see
> <https://www.gnu.org/licenses/>. */
>
> -#define BASE_PAGE_SIZE (1024 * 1024)
> -#define START_SIZE (4 * 1024)
> +#define START_SIZE (64 * 1024)
> #define MIN_PAGE_SIZE (getpagesize () + 16 * 1024 * 1024)
> #define TEST_MAIN
> #define TEST_NAME "memmove"
> -#define TIMEOUT (20 * 60)
> #include "bench-string.h"
> #include "json-lib.h"
>
> @@ -33,7 +31,7 @@ static void
> do_one_test (json_ctx_t *json_ctx, impl_t *impl, char *dst, char *src,
> size_t len)
> {
> - size_t i, iters = 16;
> + size_t i, iters = (MIN_PAGE_SIZE * 8) / len;
> timing_t start, stop, cur;
>
> TIMING_NOW (start);
> @@ -54,13 +52,8 @@ do_test (json_ctx_t *json_ctx, size_t align1, size_t align2, size_t len)
> size_t i, j;
> char *s1, *s2;
>
> - align1 &= 127;
> - if (align1 + len >= page_size)
> - return;
> -
> - align2 &= 127;
> - if (align2 + len >= page_size)
> - return;
> + align1 &= getpagesize () - 1;
> + align2 &= getpagesize () - 1;
>
> s1 = (char *) (buf2 + align1);
> s2 = (char *) (buf2 + align2);
> diff --git a/benchtests/bench-memset-large.c b/benchtests/bench-memset-large.c
> index 22ecea4da8a726c4751056a1cbabc0a505046625..3a5169940b94b7cd118ca2b461c15f5e53fc7b46 100644
> --- a/benchtests/bench-memset-large.c
> +++ b/benchtests/bench-memset-large.c
> @@ -18,9 +18,8 @@
>
> #define TEST_MAIN
> #define TEST_NAME "memset"
> -#define START_SIZE (128 * 1024)
> -#define MIN_PAGE_SIZE (getpagesize () + 64 * 1024 * 1024)
> -#define TIMEOUT (20 * 60)
> +#define START_SIZE (64 * 1024)
> +#define MIN_PAGE_SIZE (getpagesize () + 16 * 1024 * 1024)
> #include "bench-string.h"
>
> #include "json-lib.h"
> @@ -35,7 +34,7 @@ static void
> do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s,
> int c __attribute ((unused)), size_t n)
> {
> - size_t i, iters = 16;
> + size_t i, iters = (MIN_PAGE_SIZE * 64) / n;
> timing_t start, stop, cur;
>
> TIMING_NOW (start);
> @@ -53,10 +52,6 @@ do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s,
> static void
> do_test (json_ctx_t *json_ctx, size_t align, int c, size_t len)
> {
> - align &= 63;
> - 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);
> @@ -64,10 +59,7 @@ do_test (json_ctx_t *json_ctx, size_t align, int c, size_t len)
> json_array_begin (json_ctx, "timings");
>
> FOR_EACH_IMPL (impl, 0)
> - {
> - do_one_test (json_ctx, impl, (CHAR *) (buf1) + align, c, len);
> - alloc_bufs ();
> - }
> + do_one_test (json_ctx, impl, (CHAR *) (buf1) + align, c, len);
>
> json_array_end (json_ctx);
> json_element_object_end (json_ctx);
> diff --git a/benchtests/bench-memset.c b/benchtests/bench-memset.c
> index cbeb6b942b3a0fcf712cefd7c1a75b97336fa677..5682f7298b7fbaaaebb607dead7ea005a3d4a790 100644
> --- a/benchtests/bench-memset.c
> +++ b/benchtests/bench-memset.c
> @@ -61,7 +61,7 @@ do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s,
> static void
> do_test (json_ctx_t *json_ctx, size_t align, int c, size_t len)
> {
> - align &= 4095;
> + align &= getpagesize () - 1;
> if ((align + len) * sizeof (CHAR) > page_size)
> return;
>
More information about the Libc-alpha
mailing list