[FFmpeg-devel] [PATCH 1/3] checkasm/hevc_add_res: add 12bit test

Martin Storsjö martin at martin.st
Tue Aug 9 14:02:00 EEST 2022


On Thu, 23 Jun 2022, J. Dekker wrote:

> Signed-off-by: J. Dekker <jdek at itanimul.li>
> ---
> tests/checkasm/hevc_add_res.c | 15 ++++++++-------
> 1 file changed, 8 insertions(+), 7 deletions(-)
>
> diff --git a/tests/checkasm/hevc_add_res.c b/tests/checkasm/hevc_add_res.c
> index 0c896adaca..f17d121939 100644
> --- a/tests/checkasm/hevc_add_res.c
> +++ b/tests/checkasm/hevc_add_res.c
> @@ -36,14 +36,14 @@
>         }                                       \
>     } while (0)
>
> -#define randomize_buffers2(buf, size)             \
> +#define randomize_buffers2(buf, size, mask)       \
>     do {                                          \
>         int j;                                    \
>         for (j = 0; j < size; j++)                \
> -            AV_WN16A(buf + j * 2, rnd() & 0x3FF); \
> +            AV_WN16A(buf + j * 2, rnd() & mask); \
>     } while (0)
>
> -static void compare_add_res(int size, ptrdiff_t stride, int overflow_test)
> +static void compare_add_res(int size, ptrdiff_t stride, int overflow_test, int mask)
> {
>     LOCAL_ALIGNED_32(int16_t, res0, [32 * 32]);
>     LOCAL_ALIGNED_32(int16_t, res1, [32 * 32]);
> @@ -53,7 +53,7 @@ static void compare_add_res(int size, ptrdiff_t stride, int overflow_test)
>     declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *dst, int16_t *res, ptrdiff_t stride);
>
>     randomize_buffers(res0, size);
> -    randomize_buffers2(dst0, size);
> +    randomize_buffers2(dst0, size, mask);
>     if (overflow_test)
>         res0[0] = 0x8000;
>     memcpy(res1, res0, sizeof(*res0) * size);
> @@ -69,6 +69,7 @@ static void compare_add_res(int size, ptrdiff_t stride, int overflow_test)
> static void check_add_res(HEVCDSPContext h, int bit_depth)
> {
>     int i;
> +    int mask = bit_depth == 8 ? 0xFFFF : bit_depth == 10 ? 0x03FF : 0x07FF;

Previously we always used the mask 0x03FF, while we now use 0xFFFF for 8 
bit. I presume that means that for 8 bit, we mask two pixels with one 
0xFFFF (and keep all bits), and previously we accidentally masked out 
everything but the lowest two bits, from every other pixel, in 8 bit mode?

The patch LGTM, but it'd be good to acknowledge this existing issue in the 
commit message.

// Martin



More information about the ffmpeg-devel mailing list