On Tue, Jul 16, 2024 at 07:11:28PM +0200, Anton Khirnov wrote:
> It is a copy of FFV1Context.quant_tables[quant_table_index].
> ---
> libavcodec/ffv1.h | 1 -
> libavcodec/ffv1_template.c | 22 +++++++++++-----------
> libavcodec/ffv1dec.c | 28 ++++++++++++----------------
> libavcodec/ffv1dec_template.c | 14 +++++++++-----
> libavcodec/ffv1enc.c | 24 ++++++++++++------------
> libavcodec/ffv1enc_template.c | 13 ++++++++-----
> 6 files changed, 52 insertions(+), 50 deletions(-)
>
> diff --git a/libavcodec/ffv1.h b/libavcodec/ffv1.h
> index 4d57172d5b..a87c2d2a36 100644
> --- a/libavcodec/ffv1.h
> +++ b/libavcodec/ffv1.h
> @@ -59,7 +59,6 @@ typedef struct VlcState {
> } VlcState;
>
> typedef struct PlaneContext {
> - int16_t quant_table[MAX_CONTEXT_INPUTS][256];
> int quant_table_index;
> int context_count;
> uint8_t (*state)[CONTEXT_SIZE];
> diff --git a/libavcodec/ffv1_template.c b/libavcodec/ffv1_template.c
> index c5f61b0182..d15ad11021 100644
> --- a/libavcodec/ffv1_template.c
> +++ b/libavcodec/ffv1_template.c
> @@ -29,25 +29,25 @@ static inline int RENAME(predict)(TYPE *src, TYPE *last)
> return mid_pred(L, L + T - LT, T);
> }
>
> -static inline int RENAME(get_context)(PlaneContext *p, TYPE *src,
> - TYPE *last, TYPE *last2)
> +static inline int RENAME(get_context)(const int16_t
> quant_table[MAX_CONTEXT_INPUTS][256],
> + TYPE *src, TYPE *last, TYPE *last2)
> {
> const int LT = last[-1];
> const int T = last[0];
> const int RT = last[1];
> const int L = src[-1];
>
> - if (p->quant_table[3][127] || p->quant_table[4][127]) {
> + if (quant_table[3][127] || quant_table[4][127]) {
the data for each decoder task should be together and not scattered around
more than needed, reducing cache efficiency
[...]
> diff --git a/libavcodec/ffv1dec.c b/libavcodec/ffv1dec.c
> index 66d9f63c1a..618020d10f 100644
> --- a/libavcodec/ffv1dec.c
> +++ b/libavcodec/ffv1dec.c
> @@ -117,7 +117,8 @@ static int is_input_end(FFV1Context *s, GetBitContext *gb)
> #define RENAME(name) name ## 32
> #include "ffv1dec_template.c"
>
> -static int decode_plane(FFV1Context *s, FFV1SliceContext *sc,
> +static int decode_plane(FFV1Context *f,
> + FFV1Context *s, FFV1SliceContext *sc,
[...]
> static av_always_inline int
> -RENAME(decode_line)(FFV1Context *s, FFV1SliceContext *sc, GetBitContext *gb,
> +RENAME(decode_line)(FFV1Context *f,
> + FFV1Context *s, FFV1SliceContext *sc,
> + GetBitContext *gb,
> int w, TYPE *sample[2], int plane_index, int bits)
> {
> PlaneContext *const p = &s->plane[plane_index];
> @@ -57,7 +59,8 @@ RENAME(decode_line)(FFV1Context *s, FFV1SliceContext *sc,
> GetBitContext *gb,
> return AVERROR_INVALIDDATA;
> }
>
> - context = RENAME(get_context)(p, sample[1] + x, sample[0] + x,
> sample[1] + x);
> + context = RENAME(get_context)(f->quant_tables[p->quant_table_index],
> + sample[1] + x, sample[0] + x,
> sample[1] + x);
putting all this extra code in the inner per pixel loop is not ok
especially not for the sake of avoiding a memcpy of a few hundread bytes
multiple levels of loops outside
[...]
thx
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
In a rich man's house there is no place to spit but his face.
-- Diogenes of Sinope
signature.asc
Description: PGP signature
_______________________________________________ ffmpeg-devel mailing list [email protected] https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
