[FFmpeg-devel] [PATCH 1/2] checkasm/flacdsp: add LPC test

[James Almer]

On 11/15/2023 4:19 PM, Rémi Denis-Courmont wrote:
> Le keskiviikkona 15. marraskuuta 2023, 21.14.26 EET James Almer a écrit :
>> On 11/15/2023 3:02 PM, Rémi Denis-Courmont wrote:
>>> ---
>>>
>>>    tests/checkasm/flacdsp.c | 32 ++++++++++++++++++++++++++++++++
>>>    1 file changed, 32 insertions(+)
>>>
>>> diff --git a/tests/checkasm/flacdsp.c b/tests/checkasm/flacdsp.c
>>> index 51a0e0060b..4d69cbe507 100644
>>> --- a/tests/checkasm/flacdsp.c
>>> +++ b/tests/checkasm/flacdsp.c
>>> @@ -54,6 +54,28 @@ static void check_decorrelate(uint8_t **ref_dst,
>>> uint8_t **ref_src, uint8_t **ne>
>>>        bench_new(new_dst, (int32_t **)new_src, channels, BUF_SIZE /
>>>        sizeof(int32_t), 8);>
>>>    }
>>>
>>> +static void check_lpc(FLACDSPContext *c, int pred_order)
>>
>> c is unused.
>>
>>> +{
>>> +    int qlevel = rnd() % 16;
>>> +    LOCAL_ALIGNED_16(int32_t, coeffs, [32]);
>>> +    LOCAL_ALIGNED_16(int32_t, dst0, [BUF_SIZE]);
>>> +    LOCAL_ALIGNED_16(int32_t, dst1, [BUF_SIZE]);
>>> +
>>> +    declare_func(void, int32_t *, const int[32], int, int, int);
>>> +
>>> +    for (int i = 0; i < 32; i++)
>>> +        coeffs[i] = rnd();
>>> +    for (int i = 0; i < BUF_SIZE; i++)
>>> +        dst0[i] = rnd();
>>> +
>>> +    memcpy(dst1, dst0, BUF_SIZE * sizeof (int32_t));
>>> +    call_ref(dst0, coeffs, pred_order, qlevel, BUF_SIZE);
>>> +    call_new(dst1, coeffs, pred_order, qlevel, BUF_SIZE);
>>> +    if (memcmp(dst0, dst1, BUF_SIZE * sizeof (int32_t)) != 0)
>>> +       fail();
>>> +    bench_new(dst1, coeffs, pred_order, qlevel, BUF_SIZE);
>>
>> Not sure if it matters, but dst1 is already trashed by call_new().
> 
> Yeah I know. I could allocate a third buffer. AFAICT, the only parameter that
> should affect the benchmarks is pred-order (which indeed affects the result on
> both x86 and RVV). So that the extra code to preserve dst seemed pointless?

I guess it's pointless in this case, yeah, but I know that some other 
dsp functions ended up with different benchmark results if the contents 
of a buffer were bogus (Which afaik is why when being filled with rnd() 
they are also clipped with a mask).

> 
>>
>>> +}
>>> +
>>>
>>>    void checkasm_check_flacdsp(void)
>>>    {
>>>    
>>>        LOCAL_ALIGNED_16(uint8_t, ref_dst, [BUF_SIZE*MAX_CHANNELS]);
>>>
>>> @@ -72,6 +94,7 @@ void checkasm_check_flacdsp(void)
>>>
>>>            { AV_SAMPLE_FMT_S16, 16 },
>>>            { AV_SAMPLE_FMT_S32, 32 },
>>>        
>>>        };
>>>
>>> +    static const signed char pred_orders[] = { 13, 16, 29, 32 };
>>>
>>>        FLACDSPContext h;
>>>        int i, j;
>>>
>>> @@ -88,4 +111,13 @@ void checkasm_check_flacdsp(void)
>>>
>>>        }
>>>        
>>>        report("decorrelate");
>>>
>>> +
>>> +    for (int i = 0; i < sizeof (pred_orders); i++) {
>>
>> i is already defined. Also, use FF_ARRAY_ELEMS(pred_orders), so it
>> doesn't depend on char being 1 byte.
>>
>>> +        if (check_func(h.lpc16, "flac_lpc_16_%d", pred_orders[i]))
>>> +            check_lpc(&h, pred_orders[i]);
>>> +        if (check_func(h.lpc32, "flac_lpc_32_%d", pred_orders[i]))
>>> +            check_lpc(&h, pred_orders[i]);
>>> +    }
>>> +
>>> +    report("lpc");
>>>
>>>    }
>>
>> LGTM otherwise.
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
> 
>