[FFmpeg-devel] [PATCH 3/9] avcodec/av1dec: support setup shear process
Wang, Fei W
fei.w.wang at intel.com
Fri Jul 2 10:07:37 EEST 2021
On Thu, 2021-07-01 at 09:39 -0300, James Almer wrote:
> On 6/17/2021 3:10 AM, Fei Wang wrote:
> > Defined in spec 7.11.3.6/7.11.3.7.
> >
> > Signed-off-by: Fei Wang <fei.w.wang at intel.com>
> > ---
> > libavcodec/av1dec.c | 97
> > +++++++++++++++++++++++++++++++++++++++++++++
> > libavcodec/av1dec.h | 1 +
> > 2 files changed, 98 insertions(+)
> >
> > diff --git a/libavcodec/av1dec.c b/libavcodec/av1dec.c
> > index 1dda0f9160..3fca17e84b 100644
> > --- a/libavcodec/av1dec.c
> > +++ b/libavcodec/av1dec.c
> > @@ -28,6 +28,33 @@
> > #include "internal.h"
> > #include "profiles.h"
> >
> > +static const uint16_t div_lut[AV1_DIV_LUT_NUM] = {
> > + 16384, 16320, 16257, 16194, 16132, 16070, 16009, 15948, 15888,
> > 15828, 15768,
> > + 15709, 15650, 15592, 15534, 15477, 15420, 15364, 15308, 15252,
> > 15197, 15142,
> > + 15087, 15033, 14980, 14926, 14873, 14821, 14769, 14717, 14665,
> > 14614, 14564,
> > + 14513, 14463, 14413, 14364, 14315, 14266, 14218, 14170, 14122,
> > 14075, 14028,
> > + 13981, 13935, 13888, 13843, 13797, 13752, 13707, 13662, 13618,
> > 13574, 13530,
> > + 13487, 13443, 13400, 13358, 13315, 13273, 13231, 13190, 13148,
> > 13107, 13066,
> > + 13026, 12985, 12945, 12906, 12866, 12827, 12788, 12749, 12710,
> > 12672, 12633,
> > + 12596, 12558, 12520, 12483, 12446, 12409, 12373, 12336, 12300,
> > 12264, 12228,
> > + 12193, 12157, 12122, 12087, 12053, 12018, 11984, 11950, 11916,
> > 11882, 11848,
> > + 11815, 11782, 11749, 11716, 11683, 11651, 11619, 11586, 11555,
> > 11523, 11491,
> > + 11460, 11429, 11398, 11367, 11336, 11305, 11275, 11245, 11215,
> > 11185, 11155,
> > + 11125, 11096, 11067, 11038, 11009, 10980, 10951, 10923, 10894,
> > 10866, 10838,
> > + 10810, 10782, 10755, 10727, 10700, 10673, 10645, 10618, 10592,
> > 10565, 10538,
> > + 10512, 10486, 10460, 10434, 10408, 10382, 10356, 10331, 10305,
> > 10280, 10255,
> > + 10230, 10205, 10180, 10156, 10131, 10107, 10082, 10058, 10034,
> > 10010, 9986,
> > + 9963, 9939, 9916, 9892, 9869, 9846, 9823, 9800, 9777, 9
> > 754, 9732,
> > + 9709, 9687, 9664, 9642, 9620, 9598, 9576, 9554, 9533, 9
> > 511, 9489,
> > + 9468, 9447, 9425, 9404, 9383, 9362, 9341, 9321, 9300, 9
> > 279, 9259,
> > + 9239, 9218, 9198, 9178, 9158, 9138, 9118, 9098, 9079, 9
> > 059, 9039,
> > + 9020, 9001, 8981, 8962, 8943, 8924, 8905, 8886, 8867, 8
> > 849, 8830,
> > + 8812, 8793, 8775, 8756, 8738, 8720, 8702, 8684, 8666, 8
> > 648, 8630,
> > + 8613, 8595, 8577, 8560, 8542, 8525, 8508, 8490, 8473, 8
> > 456, 8439,
> > + 8422, 8405, 8389, 8372, 8355, 8339, 8322, 8306, 8289, 8
> > 273, 8257,
> > + 8240, 8224, 8208, 8192
> > +};
> > +
> > static uint32_t inverse_recenter(int r, uint32_t v)
> > {
> > if (v > 2 * r)
> > @@ -97,6 +124,70 @@ static void read_global_param(AV1DecContext *s,
> > int type, int ref, int idx)
> > -mx, mx + 1, r) <<
> > prec_diff) + round;
> > }
> >
> > +static uint64_t round_two(uint64_t x, uint16_t n)
> > +{
> > + if (n == 0)
> > + return x;
> > + return ((x + (1 << (n - 1))) >> n);
> > +}
> > +
> > +static int64_t round_two_signed(int64_t x, uint16_t n)
> > +{
> > + return ((x<0) ? -round_two(-x, n) : round_two(x, n));
> > +}
> > +
> > +/**
> > + * Resolve divisor process.
> > + * see spec 7.11.3.7
> > + */
> > +static int16_t resolve_divisor(uint32_t d, uint16_t *shift)
> > +{
> > + int32_t e, f;
> > +
> > + *shift = av_log2(d);
> > + e = d - (1 << (*shift));
> > + if (*shift > AV1_DIV_LUT_BITS)
> > + f = round_two(e, *shift - AV1_DIV_LUT_BITS);
> > + else
> > + f = e << (AV1_DIV_LUT_BITS - (*shift));
> > +
> > + *shift += AV1_DIV_LUT_PREC_BITS;
> > +
> > + return div_lut[f];
> > +}
> > +
> > +/**
> > + * check if global motion params is valid.
> > + * see spec 7.11.3.6
> > + */
> > +static uint8_t get_shear_params_valid(AV1DecContext *s, int idx)
> > +{
> > + int16_t alpha, beta, gamma, delta, divf, divs;
> > + int64_t v, w;
> > + int32_t *param = &s->cur_frame.gm_params[idx][0];
> > + if (param[2] < 0)
> > + return 0;
> > +
> > + alpha = av_clip_int16(param[2] - (1 <<
> > AV1_WARPEDMODEL_PREC_BITS));
> > + beta = av_clip_int16(param[3]);
> > + divf = resolve_divisor(abs(param[2]), &divs);
> > + v = param[4] * (1 << AV1_WARPEDMODEL_PREC_BITS);
> > + w = param[3] * param[4];
> > + gamma = av_clip_int16(round_two_signed((v * divf), divs));
> > + delta = av_clip_int16(param[5] - round_two_signed((w * divf),
> > divs) - (1 << AV1_WARPEDMODEL_PREC_BITS));
> > +
> > + alpha = round_two_signed(alpha, AV1_WARP_PARAM_REDUCE_BITS) <<
> > AV1_WARP_PARAM_REDUCE_BITS;
> > + beta = round_two_signed(beta, AV1_WARP_PARAM_REDUCE_BITS) <<
> > AV1_WARP_PARAM_REDUCE_BITS;
> > + gamma = round_two_signed(gamma, AV1_WARP_PARAM_REDUCE_BITS) <<
> > AV1_WARP_PARAM_REDUCE_BITS;
> > + delta = round_two_signed(delta, AV1_WARP_PARAM_REDUCE_BITS) <<
> > AV1_WARP_PARAM_REDUCE_BITS;
> > +
> > + if ((4 * abs(alpha) + 7 * abs(beta)) >= (1 <<
> > AV1_WARPEDMODEL_PREC_BITS) ||
> > + (4 * abs(gamma) + 4 * abs(delta)) >= (1 <<
> > AV1_WARPEDMODEL_PREC_BITS))
> > + return 0;
> > +
> > + return 1;
> > +}
> > +
> > /**
> > * update gm type/params, since cbs already implemented part of
> > this funcation,
> > * so we don't need to full implement spec.
> > @@ -144,6 +235,9 @@ static void global_motion_params(AV1DecContext
> > *s)
> > read_global_param(s, type, ref, 0);
> > read_global_param(s, type, ref, 1);
> > }
> > + if (type <= AV1_WARP_MODEL_AFFINE) {
> > + s->cur_frame.gm_invalid[ref] =
> > !get_shear_params_valid(s, ref);
>
> Section 7.11.3.1 says
> "If (YMode == GLOBALMV || YMode == GLOBAL_GLOBALMV) and
> GmType[refFrame]
> > TRANSLATION, the setup shear process specified in section 7.11.3.6
> is
> invoked with gm_params[refFrame] as input, and the output warpValid
> is
> assigned to globalValid", but you're checking <= AFFINE.
This is guaranteed inside get_shear_params_valid, if gm_type >
TRANSLATION, then gm_param[2] should be more than 0 according to
read_global_param;
> Also, what about those YMode values? They are not coded in the
> bitstream, so they need to be derived.
YMode is block level variable, we can deal with it here. We can provide
the gm_valid without considering it. When the bottom level hardware
decode block, it can directly use it with YMode.
>
> Is this patchset fixing real world files, for that matter? Timo
> mentioned that at least with nvdec he hasn't seen reports of files
> that
> decoded wrong.
Yes, this patch do fixed very few clips in my side. Most clips with gm
params are valid params, which can be decode correctly without doing
shear check.
Thanks
Fei
>
> > + }
> > }
> > }
> >
> > @@ -509,6 +603,9 @@ static int av1_frame_ref(AVCodecContext *avctx,
> > AV1Frame *dst, const AV1Frame *s
> >
> > dst->spatial_id = src->spatial_id;
> > dst->temporal_id = src->temporal_id;
> > + memcpy(dst->gm_invalid,
> > + src->gm_invalid,
> > + AV1_NUM_REF_FRAMES * sizeof(uint8_t));
> > memcpy(dst->gm_type,
> > src->gm_type,
> > AV1_NUM_REF_FRAMES * sizeof(uint8_t));
> > diff --git a/libavcodec/av1dec.h b/libavcodec/av1dec.h
> > index 248a68750f..4e140588b9 100644
> > --- a/libavcodec/av1dec.h
> > +++ b/libavcodec/av1dec.h
> > @@ -42,6 +42,7 @@ typedef struct AV1Frame {
> > int temporal_id;
> > int spatial_id;
> >
> > + uint8_t gm_invalid[AV1_NUM_REF_FRAMES];
> > uint8_t gm_type[AV1_NUM_REF_FRAMES];
> > int32_t gm_params[AV1_NUM_REF_FRAMES][6];
> >
> >
>
> _______________________________________________
> 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".
More information about the ffmpeg-devel
mailing list