Lynne via ffmpeg-devel <ffmpeg-devel@ffmpeg.org> writes:
> On 13/08/2024 16:03, J. Dekker wrote:
>> Port dav1d's checkasm output format to FFmpeg's checkasm, includes
>> relative speedups and aligns results.
>> Signed-off-by: J. Dekker <j...@itanimul.li>
>> ---
>> tests/checkasm/checkasm.c | 53 +++++++++++++++++++++++++++++++++++----
>> 1 file changed, 48 insertions(+), 5 deletions(-)
>> diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
>> index f82ee0864f..0095758268 100644
>> --- a/tests/checkasm/checkasm.c
>> +++ b/tests/checkasm/checkasm.c
>> @@ -18,6 +18,31 @@
>> * You should have received a copy of the GNU General Public License along
>> * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
>> * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
>> + *
>> + * Copyright © 2018, VideoLAN and dav1d authors
>> + * Copyright © 2018, Two Orioles, LLC
>> + * All rights reserved.
>> + *
>> + * Redistribution and use in source and binary forms, with or without
>> + * modification, are permitted provided that the following conditions are
>> met:
>> + *
>> + * 1. Redistributions of source code must retain the above copyright
>> notice, this
>> + * list of conditions and the following disclaimer.
>> + *
>> + * 2. Redistributions in binary form must reproduce the above copyright
>> notice,
>> + * this list of conditions and the following disclaimer in the
>> documentation
>> + * and/or other materials provided with the distribution.
>> + *
>> + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
>> IS" AND
>> + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
>> IMPLIED
>> + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
>> + * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
>> LIABLE FOR
>> + * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
>> DAMAGES
>> + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
>> SERVICES;
>> + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
>> AND
>> + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR
>> TORT
>> + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
>> THIS
>> + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
>> */
>> #include "config.h"
>> @@ -575,6 +600,16 @@ static int measure_nop_time(void)
>> return nop_sum / 500;
>> }
>> +static inline double avg_cycles_per_call(const CheckasmPerf *const p)
>> +{
>> + if (p->iterations) {
>> + const double cycles = (double)(10 * p->cycles) / p->iterations -
>> state.nop_time;
>> + if (cycles > 0.0)
>> + return cycles / 4.0; /* 4 calls per iteration */
>> + }
>> + return 0.0;
>> +}
>> +
>> /* Print benchmark results */
>> static void print_benchs(CheckasmFunc *f)
>> {
>> @@ -584,17 +619,25 @@ static void print_benchs(CheckasmFunc *f)
>> /* Only print functions with at least one assembly version */
>> if (f->versions.cpu || f->versions.next) {
>> CheckasmFuncVersion *v = &f->versions;
>> + const CheckasmPerf *p = &v->perf;
>> + const double baseline = avg_cycles_per_call(p);
>> + double decicycles;
>> do {
>> - CheckasmPerf *p = &v->perf;
>> if (p->iterations) {
>> - int decicycles = (10*p->cycles/p->iterations -
>> state.nop_time) / 4;
>> + p = &v->perf;
>> + decicycles = avg_cycles_per_call(p);
>> if (state.csv) {
>> const char sep = state.tsv ? '\t' : ',';
>> - printf("%s%c%s%c%d.%d\n", f->name, sep,
>> + printf("%s%c%s%c%.1f\n", f->name, sep,
>> cpu_suffix(v->cpu), sep,
>> - decicycles / 10, decicycles % 10);
>> + decicycles / 10.0);
>> } else {
>> - printf("%s_%s: %d.%d\n", f->name,
>> cpu_suffix(v->cpu), decicycles/10, decicycles%10);
>> + const int pad_length = 10 + 50 -
>> + printf("%s_%s:", f->name, cpu_suffix(v->cpu));
>> + const double ratio = decicycles ?
>> + baseline / decicycles : 0.0;
>> + printf("%*.1f (%5.2fx)\n", FFMAX(pad_length, 0),
>> + decicycles / 10.0, ratio);
>> }
>> }
>> } while ((v = v->next));
>
> How does it improve it?
>
> You're only interested in the last X iterations, after cache has fully warmed
> up and is out of the equation. Averaging all results from all iteration would
> be also benchmarking the memory layout of the system, but only the cycles are
> of interest.
The semantics of how the benchmark is calculated is essentially
unchanged by this commit. The only intention is to follow dav1d's
checkasm format and include a relative speedup for easier at-a-glance
comparisons.
--
jd
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
