[FFmpeg-devel] [PATCH] checkasm: improve hevc_sao test

[Martin Storsjö]

On Wed, 25 May 2022, J. Dekker wrote:

>
>
> On 24 May 2022, at 22:27, Martin Storsjö wrote:
>
>> On Tue, 17 May 2022, J. Dekker wrote:
>>
>>> The HEVC decoder can call these functions with smaller widths than the
>>> functions themselves are designed to operate on so we should only check
>>> the relevant output
>>>
>>> Signed-off-by: J. Dekker <jdek at itanimul.li>
>>> ---
>>> tests/checkasm/hevc_sao.c | 51 ++++++++++++++++++++++++---------------
>>> 1 file changed, 31 insertions(+), 20 deletions(-)
>>>
>>> diff --git a/tests/checkasm/hevc_sao.c b/tests/checkasm/hevc_sao.c
>>> index 6b750758e2..72cdb87dd1 100644
>>> --- a/tests/checkasm/hevc_sao.c
>>> +++ b/tests/checkasm/hevc_sao.c
>>> @@ -78,20 +78,26 @@ static void check_sao_band(HEVCDSPContext h, int bit_depth)
>>>
>>>     for (i = 0; i <= 4; i++) {
>>>         int block_size = sao_size[i];
>>> +        int prev_size = i > 0 ? sao_size[i - 1] : 0;
>>>         ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
>>>         declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *dst, uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride,
>>>                           int16_t *sao_offset_val, int sao_left_class, int width, int height);
>>>
>>> -        randomize_buffers(src0, src1, BUF_SIZE);
>>> -        randomize_buffers2(offset_val, OFFSET_LENGTH);
>>> -        memset(dst0, 0, BUF_SIZE);
>>> -        memset(dst1, 0, BUF_SIZE);
>>> -
>>> -        if (check_func(h.sao_band_filter[i], "hevc_sao_band_%dx%d_%d", block_size, block_size, bit_depth)) {
>>> -            call_ref(dst0, src0, stride, stride, offset_val, left_class, block_size, block_size);
>>> -            call_new(dst1, src1, stride, stride, offset_val, left_class, block_size, block_size);
>>> -            if (memcmp(dst0, dst1, BUF_SIZE))
>>> -                fail();
>>> +        if (check_func(h.sao_band_filter[i], "hevc_sao_band_%d_%d", block_size, bit_depth)) {
>>> +
>>> +            for (int w = prev_size + 4; w <= block_size; w += 4) {
>>> +                randomize_buffers(src0, src1, BUF_SIZE);
>>> +                randomize_buffers2(offset_val, OFFSET_LENGTH);
>>> +                memset(dst0, 0, BUF_SIZE);
>>> +                memset(dst1, 0, BUF_SIZE);
>>> +
>>> +                call_ref(dst0, src0, stride, stride, offset_val, left_class, w, block_size);
>>> +                call_new(dst1, src1, stride, stride, offset_val, left_class, w, block_size);
>>> +                for (int j = 0; j < block_size; j++) {
>>> +                    if (memcmp(dst0 + j*MAX_PB_SIZE*2, dst1 + j*MAX_PB_SIZE*2, w))
>>
>> I'm not quite sure about the MAX_PB_SIZE*2 part here - shouldn't that be just the 'stride' variable instead? And for the compared length ('w'), shouldn't that be multiplied by SIZEOF_PIXEL?
>>
>> Other than that, this looks good to me!
>
> Pushed with this fix. Rest of the set as-is. I have an extra SAO patch 
> but don't want to delay this set further.

Thanks! Yes, it's better to land the previously reviewed and good parts, 
to keep the focus in the reviews.

// Martin