From f1ee64b9e2b800ff6e4f58efd161b9d3a6bcdefe Mon Sep 17 00:00:00 2001
From: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Date: Sun, 2 Mar 2025 04:01:24 +0100
Subject: [PATCH 24/77] avcodec/mpegvideo: Move ratecontrol-adjacent fields to
MPVMainEncContext
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
---
libavcodec/mpegvideo.h | 6 ------
libavcodec/mpegvideo_enc.c | 40 ++++++++++++++++++++------------------
libavcodec/mpegvideoenc.h | 15 ++++++++++----
libavcodec/ratecontrol.c | 18 +++++++++--------
libavcodec/snowenc.c | 4 ++--
5 files changed, 44 insertions(+), 39 deletions(-)
diff --git a/libavcodec/mpegvideo.h b/libavcodec/mpegvideo.h
index 3259c1117c..6c71d9bc01 100644
--- a/libavcodec/mpegvideo.h
+++ b/libavcodec/mpegvideo.h
@@ -101,7 +101,6 @@ typedef struct MpegEncContext {
int h263_flv; ///< use flv H.263 header
enum AVCodecID codec_id; /* see AV_CODEC_ID_xxx */
- int fixed_qscale; ///< fixed qscale if non zero
int encoding; ///< true if we are encoding (vs decoding)
int max_b_frames; ///< max number of B-frames for encoding
int luma_elim_threshold;
@@ -478,10 +477,6 @@ typedef struct MpegEncContext {
me_cmp_func sse_cmp[2];
int (*sum_abs_dctelem)(const int16_t *block);
- float border_masking;
- int lmin, lmax;
- int vbv_ignore_qmax;
-
/// Bitfield containing information which frames to reconstruct.
int frame_reconstruction_bitfield;
@@ -493,7 +488,6 @@ typedef struct MpegEncContext {
int error_rate;
- int scenechange_threshold;
int noise_reduction;
int intra_penalty;
diff --git a/libavcodec/mpegvideo_enc.c b/libavcodec/mpegvideo_enc.c
index cc3c7a745f..06314dc602 100644
--- a/libavcodec/mpegvideo_enc.c
+++ b/libavcodec/mpegvideo_enc.c
@@ -195,8 +195,10 @@ void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64],
}
}
-static inline void update_qscale(MpegEncContext *s)
+static inline void update_qscale(MPVMainEncContext *const m)
{
+ MpegEncContext *const s = &m->s;
+
if (s->q_scale_type == 1 && 0) {
int i;
int bestdiff=INT_MAX;
@@ -205,7 +207,7 @@ static inline void update_qscale(MpegEncContext *s)
for (i = 0 ; i<FF_ARRAY_ELEMS(ff_mpeg2_non_linear_qscale); i++) {
int diff = FFABS((ff_mpeg2_non_linear_qscale[i]<<(FF_LAMBDA_SHIFT + 6)) - (int)s->lambda * 139);
if (ff_mpeg2_non_linear_qscale[i] < s->avctx->qmin ||
- (ff_mpeg2_non_linear_qscale[i] > s->avctx->qmax && !s->vbv_ignore_qmax))
+ (ff_mpeg2_non_linear_qscale[i] > s->avctx->qmax && !m->vbv_ignore_qmax))
continue;
if (diff < bestdiff) {
bestdiff = diff;
@@ -216,7 +218,7 @@ static inline void update_qscale(MpegEncContext *s)
} else {
s->qscale = (s->lambda * 139 + FF_LAMBDA_SCALE * 64) >>
(FF_LAMBDA_SHIFT + 7);
- s->qscale = av_clip(s->qscale, s->avctx->qmin, s->vbv_ignore_qmax ? 31 : s->avctx->qmax);
+ s->qscale = av_clip(s->qscale, s->avctx->qmin, m->vbv_ignore_qmax ? 31 : s->avctx->qmax);
}
s->lambda2 = (s->lambda * s->lambda + FF_LAMBDA_SCALE / 2) >>
@@ -507,16 +509,16 @@ av_cold int ff_mpv_encode_init(AVCodecContext *avctx)
}
/* Fixed QSCALE */
- s->fixed_qscale = !!(avctx->flags & AV_CODEC_FLAG_QSCALE);
+ m->fixed_qscale = !!(avctx->flags & AV_CODEC_FLAG_QSCALE);
s->adaptive_quant = (avctx->lumi_masking ||
avctx->dark_masking ||
avctx->temporal_cplx_masking ||
avctx->spatial_cplx_masking ||
avctx->p_masking ||
- s->border_masking ||
+ m->border_masking ||
(s->mpv_flags & FF_MPV_FLAG_QP_RD)) &&
- !s->fixed_qscale;
+ !m->fixed_qscale;
s->loop_filter = !!(avctx->flags & AV_CODEC_FLAG_LOOP_FILTER);
@@ -580,7 +582,7 @@ av_cold int ff_mpv_encode_init(AVCodecContext *avctx)
return AVERROR(EINVAL);
}
- if (!s->fixed_qscale &&
+ if (!m->fixed_qscale &&
avctx->bit_rate * av_q2d(avctx->time_base) >
avctx->bit_rate_tolerance) {
double nbt = avctx->bit_rate * av_q2d(avctx->time_base) * 5;
@@ -675,7 +677,7 @@ av_cold int ff_mpv_encode_init(AVCodecContext *avctx)
return AVERROR(EINVAL);
}
- if (s->scenechange_threshold < 1000000000 &&
+ if (m->scenechange_threshold < 1000000000 &&
(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP)) {
av_log(avctx, AV_LOG_ERROR,
"closed gop with scene change detection are not supported yet, "
@@ -890,9 +892,9 @@ av_cold int ff_mpv_encode_init(AVCodecContext *avctx)
s->frame_reconstruction_bitfield = 0;
}
- if (s->lmin > s->lmax) {
- av_log(avctx, AV_LOG_WARNING, "Clipping lmin value to %d\n", s->lmax);
- s->lmin = s->lmax;
+ if (m->lmin > m->lmax) {
+ av_log(avctx, AV_LOG_WARNING, "Clipping lmin value to %d\n", m->lmax);
+ m->lmin = m->lmax;
}
/* init */
@@ -1876,7 +1878,7 @@ int ff_mpv_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
ff_mpv_unref_picture(&s->cur_pic);
- s->vbv_ignore_qmax = 0;
+ m->vbv_ignore_qmax = 0;
m->picture_in_gop_number++;
@@ -1935,7 +1937,7 @@ vbv_retry:
int min_step = hq ? 1 : (1<<(FF_LAMBDA_SHIFT + 7))/139;
if (put_bits_count(&s->pb) > max_size &&
- s->lambda < s->lmax) {
+ s->lambda < m->lmax) {
m->next_lambda = FFMAX(s->lambda + min_step, s->lambda *
(s->qscale + 1) / s->qscale);
if (s->adaptive_quant) {
@@ -1955,7 +1957,7 @@ vbv_retry:
s->time_base = s->last_time_base;
s->last_non_b_time = s->time - s->pp_time;
}
- s->vbv_ignore_qmax = 1;
+ m->vbv_ignore_qmax = 1;
av_log(avctx, AV_LOG_VERBOSE, "reencoding frame due to VBV\n");
goto vbv_retry;
}
@@ -3616,7 +3618,7 @@ static int estimate_qp(MPVMainEncContext *const m, int dry_run)
if (m->next_lambda){
s->cur_pic.ptr->f->quality = m->next_lambda;
if(!dry_run) m->next_lambda= 0;
- } else if (!s->fixed_qscale) {
+ } else if (!m->fixed_qscale) {
int quality = ff_rate_estimate_qscale(m, dry_run);
s->cur_pic.ptr->f->quality = quality;
if (s->cur_pic.ptr->f->quality < 0)
@@ -3643,7 +3645,7 @@ static int estimate_qp(MPVMainEncContext *const m, int dry_run)
//FIXME broken
}else
s->lambda = s->cur_pic.ptr->f->quality;
- update_qscale(s);
+ update_qscale(m);
return 0;
}
@@ -3700,7 +3702,7 @@ static int encode_picture(MPVMainEncContext *const m, const AVPacket *pkt)
s->lambda= s->last_lambda_for[s->pict_type];
else
s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
- update_qscale(s);
+ update_qscale(m);
}
ff_me_init_pic(s);
@@ -3742,7 +3744,7 @@ static int encode_picture(MPVMainEncContext *const m, const AVPacket *pkt)
for(i=0; i<s->mb_stride*s->mb_height; i++)
s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
- if(!s->fixed_qscale){
+ if (!m->fixed_qscale) {
/* finding spatial complexity for I-frame rate control */
s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
}
@@ -3754,7 +3756,7 @@ static int encode_picture(MPVMainEncContext *const m, const AVPacket *pkt)
s->mb_var_sum = s->me. mb_var_sum_temp;
emms_c();
- if (s->me.scene_change_score > s->scenechange_threshold &&
+ if (s->me.scene_change_score > m->scenechange_threshold &&
s->pict_type == AV_PICTURE_TYPE_P) {
s->pict_type= AV_PICTURE_TYPE_I;
for(i=0; i<s->mb_stride*s->mb_height; i++)
diff --git a/libavcodec/mpegvideoenc.h b/libavcodec/mpegvideoenc.h
index a76d17f63e..1fbe5590df 100644
--- a/libavcodec/mpegvideoenc.h
+++ b/libavcodec/mpegvideoenc.h
@@ -39,6 +39,8 @@
typedef struct MPVMainEncContext {
MpegEncContext s; ///< The main slicecontext
+ int scenechange_threshold;
+
int intra_only; ///< if true, only intra pictures are generated
int gop_size;
int picture_in_gop_number; ///< 0-> first pic in gop, ...
@@ -61,6 +63,10 @@ typedef struct MPVMainEncContext {
int b_sensitivity;
int brd_scale;
+ float border_masking;
+ int lmin, lmax;
+ int vbv_ignore_qmax;
+
/* frame skip options */
int frame_skip_threshold;
int frame_skip_factor;
@@ -73,6 +79,7 @@ typedef struct MPVMainEncContext {
int frame_bits; ///< bits used for the current frame
int stuffing_bits; ///< bits used for stuffing
int next_lambda; ///< next lambda used for retrying to encode a frame
+ int fixed_qscale; ///< fixed qscale if non zero
RateControlContext rc_context; ///< contains stuff only accessed in ratecontrol.c
} MPVMainEncContext;
@@ -154,14 +161,14 @@ FF_MPV_OPT_CMP_FUNC, \
FF_RC_OFFSET(rc_eq), AV_OPT_TYPE_STRING, .flags = FF_MPV_OPT_FLAGS }, \
{"rc_init_cplx", "initial complexity for 1-pass encoding", FF_RC_OFFSET(initial_cplx), AV_OPT_TYPE_FLOAT, {.dbl = 0 }, -FLT_MAX, FLT_MAX, FF_MPV_OPT_FLAGS}, \
{"rc_buf_aggressivity", "currently useless", FF_RC_OFFSET(buffer_aggressivity), AV_OPT_TYPE_FLOAT, {.dbl = 1.0 }, -FLT_MAX, FLT_MAX, FF_MPV_OPT_FLAGS}, \
-{"border_mask", "increase the quantizer for macroblocks close to borders", FF_MPV_OFFSET(border_masking), AV_OPT_TYPE_FLOAT, {.dbl = 0 }, -FLT_MAX, FLT_MAX, FF_MPV_OPT_FLAGS}, \
-{"lmin", "minimum Lagrange factor (VBR)", FF_MPV_OFFSET(lmin), AV_OPT_TYPE_INT, {.i64 = 2*FF_QP2LAMBDA }, 0, INT_MAX, FF_MPV_OPT_FLAGS }, \
-{"lmax", "maximum Lagrange factor (VBR)", FF_MPV_OFFSET(lmax), AV_OPT_TYPE_INT, {.i64 = 31*FF_QP2LAMBDA }, 0, INT_MAX, FF_MPV_OPT_FLAGS }, \
+{"border_mask", "increase the quantizer for macroblocks close to borders", FF_MPV_MAIN_OFFSET(border_masking), AV_OPT_TYPE_FLOAT, {.dbl = 0 }, -FLT_MAX, FLT_MAX, FF_MPV_OPT_FLAGS}, \
+{"lmin", "minimum Lagrange factor (VBR)", FF_MPV_MAIN_OFFSET(lmin), AV_OPT_TYPE_INT, {.i64 = 2*FF_QP2LAMBDA }, 0, INT_MAX, FF_MPV_OPT_FLAGS }, \
+{"lmax", "maximum Lagrange factor (VBR)", FF_MPV_MAIN_OFFSET(lmax), AV_OPT_TYPE_INT, {.i64 = 31*FF_QP2LAMBDA }, 0, INT_MAX, FF_MPV_OPT_FLAGS }, \
{"skip_threshold", "Frame skip threshold", FF_MPV_MAIN_OFFSET(frame_skip_threshold), AV_OPT_TYPE_INT, {.i64 = 0 }, INT_MIN, INT_MAX, FF_MPV_OPT_FLAGS }, \
{"skip_factor", "Frame skip factor", FF_MPV_MAIN_OFFSET(frame_skip_factor), AV_OPT_TYPE_INT, {.i64 = 0 }, INT_MIN, INT_MAX, FF_MPV_OPT_FLAGS }, \
{"skip_exp", "Frame skip exponent", FF_MPV_MAIN_OFFSET(frame_skip_exp), AV_OPT_TYPE_INT, {.i64 = 0 }, INT_MIN, INT_MAX, FF_MPV_OPT_FLAGS }, \
{"skip_cmp", "Frame skip compare function", FF_MPV_MAIN_OFFSET(frame_skip_cmp), AV_OPT_TYPE_INT, {.i64 = FF_CMP_DCTMAX }, INT_MIN, INT_MAX, FF_MPV_OPT_FLAGS, .unit = "cmp_func" }, \
-{"sc_threshold", "Scene change threshold", FF_MPV_OFFSET(scenechange_threshold), AV_OPT_TYPE_INT, {.i64 = 0 }, INT_MIN, INT_MAX, FF_MPV_OPT_FLAGS }, \
+{"sc_threshold", "Scene change threshold", FF_MPV_MAIN_OFFSET(scenechange_threshold), AV_OPT_TYPE_INT, {.i64 = 0 }, INT_MIN, INT_MAX, FF_MPV_OPT_FLAGS }, \
{"noise_reduction", "Noise reduction", FF_MPV_OFFSET(noise_reduction), AV_OPT_TYPE_INT, {.i64 = 0 }, INT_MIN, INT_MAX, FF_MPV_OPT_FLAGS }, \
{"ps", "RTP payload size in bytes", FF_MPV_OFFSET(rtp_payload_size), AV_OPT_TYPE_INT, {.i64 = 0 }, INT_MIN, INT_MAX, FF_MPV_OPT_FLAGS }, \
diff --git a/libavcodec/ratecontrol.c b/libavcodec/ratecontrol.c
index 4aa358a4e5..f4df67aab0 100644
--- a/libavcodec/ratecontrol.c
+++ b/libavcodec/ratecontrol.c
@@ -142,10 +142,11 @@ static double get_diff_limited_q(MPVMainEncContext *m, const RateControlEntry *r
/**
* Get the qmin & qmax for pict_type.
*/
-static void get_qminmax(int *qmin_ret, int *qmax_ret, MpegEncContext *s, int pict_type)
+static void get_qminmax(int *qmin_ret, int *qmax_ret, MPVMainEncContext *const m, int pict_type)
{
- int qmin = s->lmin;
- int qmax = s->lmax;
+ MpegEncContext *const s = &m->s;
+ int qmin = m->lmin;
+ int qmax = m->lmax;
av_assert0(qmin <= qmax);
@@ -182,7 +183,7 @@ static double modify_qscale(MPVMainEncContext *const m, const RateControlEntry *
const int pict_type = rce->new_pict_type;
int qmin, qmax;
- get_qminmax(&qmin, &qmax, s, pict_type);
+ get_qminmax(&qmin, &qmax, m, pict_type);
/* modulation */
if (rcc->qmod_freq &&
@@ -784,15 +785,16 @@ static void update_predictor(Predictor *p, double q, double var, double size)
}
static void adaptive_quantization(RateControlContext *const rcc,
- MpegEncContext *const s, double q)
+ MPVMainEncContext *const m, double q)
{
+ MpegEncContext *const s = &m->s;
int i;
const float lumi_masking = s->avctx->lumi_masking / (128.0 * 128.0);
const float dark_masking = s->avctx->dark_masking / (128.0 * 128.0);
const float temp_cplx_masking = s->avctx->temporal_cplx_masking;
const float spatial_cplx_masking = s->avctx->spatial_cplx_masking;
const float p_masking = s->avctx->p_masking;
- const float border_masking = s->border_masking;
+ const float border_masking = m->border_masking;
float bits_sum = 0.0;
float cplx_sum = 0.0;
float *cplx_tab = rcc->cplx_tab;
@@ -932,7 +934,7 @@ float ff_rate_estimate_qscale(MPVMainEncContext *const m, int dry_run)
const int pict_type = s->pict_type;
emms_c();
- get_qminmax(&qmin, &qmax, s, pict_type);
+ get_qminmax(&qmin, &qmax, m, pict_type);
fps = get_fps(s->avctx);
/* update predictors */
@@ -1068,7 +1070,7 @@ float ff_rate_estimate_qscale(MPVMainEncContext *const m, int dry_run)
q = qmax;
if (s->adaptive_quant)
- adaptive_quantization(rcc, s, q);
+ adaptive_quantization(rcc, m, q);
else
q = (int)(q + 0.5);
diff --git a/libavcodec/snowenc.c b/libavcodec/snowenc.c
index 830f6f5096..d974a12f6b 100644
--- a/libavcodec/snowenc.c
+++ b/libavcodec/snowenc.c
@@ -228,8 +228,8 @@ static av_cold int encode_init(AVCodecContext *avctx)
mpv->avctx = avctx;
mpv->bit_rate= avctx->bit_rate;
- mpv->lmin = avctx->mb_lmin;
- mpv->lmax = avctx->mb_lmax;
+ enc->m.lmin = avctx->mb_lmin;
+ enc->m.lmax = avctx->mb_lmax;
mpv->mb_num = (avctx->width * avctx->height + 255) / 256; // For ratecontrol
mpv->me.temp =
--
2.45.2