[FFmpeg-devel] [PATCH 2/7] checkasm: improve print format

[J. Dekker]

Lynne via ffmpeg-devel <ffmpeg-devel at 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 <jdek at 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