|
|
|
@ -44,6 +44,7 @@ |
|
|
|
|
typedef struct USPPContext { |
|
|
|
|
const AVClass *av_class; |
|
|
|
|
int log2_count; |
|
|
|
|
int count; |
|
|
|
|
int hsub, vsub; |
|
|
|
|
int qp; |
|
|
|
|
char *codec_name; |
|
|
|
@ -55,12 +56,13 @@ typedef struct USPPContext { |
|
|
|
|
uint8_t *outbuf; |
|
|
|
|
AVCodecContext *avctx_enc[BLOCK*BLOCK]; |
|
|
|
|
AVCodecContext *avctx_dec[BLOCK*BLOCK]; |
|
|
|
|
AVPacket *pkt; |
|
|
|
|
AVFrame *frame; |
|
|
|
|
AVFrame *frame_dec; |
|
|
|
|
AVPacket *pkt [BLOCK*BLOCK]; |
|
|
|
|
AVFrame *frame [BLOCK*BLOCK]; |
|
|
|
|
AVFrame *frame_dec [BLOCK*BLOCK]; |
|
|
|
|
int8_t *non_b_qp_table; |
|
|
|
|
int non_b_qp_stride; |
|
|
|
|
int use_bframe_qp; |
|
|
|
|
int quality; |
|
|
|
|
} USPPContext; |
|
|
|
|
|
|
|
|
|
#define OFFSET(x) offsetof(USPPContext, x) |
|
|
|
@ -188,13 +190,87 @@ static void store_slice_c(uint8_t *dst, const uint16_t *src, |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
static void filter(USPPContext *p, uint8_t *dst[3], uint8_t *src[3], |
|
|
|
|
static int filter_1phase(AVFilterContext *ctx, void *arg, int i, int nb_jobs) |
|
|
|
|
{ |
|
|
|
|
USPPContext *p = ctx->priv; |
|
|
|
|
int ret, x, y; |
|
|
|
|
int width = ctx->inputs[0]->w; |
|
|
|
|
int height = ctx->inputs[0]->h; |
|
|
|
|
|
|
|
|
|
const int x1 = offset[i+nb_jobs-1][0]; |
|
|
|
|
const int y1 = offset[i+nb_jobs-1][1]; |
|
|
|
|
const int x1c = x1 >> p->hsub; |
|
|
|
|
const int y1c = y1 >> p->vsub; |
|
|
|
|
const int BLOCKc = BLOCK >> p->hsub; |
|
|
|
|
int offset; |
|
|
|
|
AVPacket *pkt = p->pkt[i]; |
|
|
|
|
|
|
|
|
|
av_packet_unref(pkt); |
|
|
|
|
pkt->data = p->outbuf; |
|
|
|
|
pkt->size = p->outbuf_size; |
|
|
|
|
|
|
|
|
|
p->frame[i]->linesize[0] = p->temp_stride[0]; |
|
|
|
|
p->frame[i]->linesize[1] = p->temp_stride[1]; |
|
|
|
|
p->frame[i]->linesize[2] = p->temp_stride[2]; |
|
|
|
|
p->frame[i]->height = height + BLOCK; |
|
|
|
|
p->frame[i]->width = width + BLOCK; |
|
|
|
|
p->frame[i]->data[0] = p->src[0] + x1 + y1 * p->frame[i]->linesize[0]; |
|
|
|
|
p->frame[i]->data[1] = p->src[1] + x1c + y1c * p->frame[i]->linesize[1]; |
|
|
|
|
p->frame[i]->data[2] = p->src[2] + x1c + y1c * p->frame[i]->linesize[2]; |
|
|
|
|
p->frame[i]->format = p->avctx_enc[i]->pix_fmt; |
|
|
|
|
p->frame[i]->quality = p->quality; |
|
|
|
|
|
|
|
|
|
ret = avcodec_send_frame(p->avctx_enc[i], p->frame[i]); |
|
|
|
|
if (ret < 0) { |
|
|
|
|
av_log(p->avctx_enc[i], AV_LOG_ERROR, "Error sending a frame for encoding\n"); |
|
|
|
|
return ret; |
|
|
|
|
} |
|
|
|
|
ret = avcodec_receive_packet(p->avctx_enc[i], pkt); |
|
|
|
|
if (ret < 0) { |
|
|
|
|
av_log(p->avctx_enc[i], AV_LOG_ERROR, "Error receiving a packet from encoding\n"); |
|
|
|
|
return ret; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
ret = avcodec_send_packet(p->avctx_dec[i], pkt); |
|
|
|
|
av_packet_unref(pkt); |
|
|
|
|
if (ret < 0) { |
|
|
|
|
av_log(p->avctx_dec[i], AV_LOG_ERROR, "Error sending a packet for decoding\n"); |
|
|
|
|
return ret; |
|
|
|
|
} |
|
|
|
|
ret = avcodec_receive_frame(p->avctx_dec[i], p->frame_dec[i]); |
|
|
|
|
if (ret < 0) { |
|
|
|
|
av_log(p->avctx_dec[i], AV_LOG_ERROR, "Error receiving a frame from decoding\n"); |
|
|
|
|
return ret; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
offset = (BLOCK-x1) + (BLOCK-y1) * p->frame_dec[i]->linesize[0]; |
|
|
|
|
|
|
|
|
|
for (y = 0; y < height; y++) |
|
|
|
|
for (x = 0; x < width; x++) |
|
|
|
|
p->temp[0][x + y * p->temp_stride[0]] += p->frame_dec[i]->data[0][x + y * p->frame_dec[i]->linesize[0] + offset]; |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
if (!p->frame_dec[i]->data[2] || !p->temp[2]) |
|
|
|
|
return 0; |
|
|
|
|
|
|
|
|
|
offset = (BLOCKc-x1c) + (BLOCKc-y1c) * p->frame_dec[i]->linesize[1]; |
|
|
|
|
|
|
|
|
|
for (y = 0; y < AV_CEIL_RSHIFT(height, p->vsub); y++) { |
|
|
|
|
for (x = 0; x < AV_CEIL_RSHIFT(width, p->hsub); x++) { |
|
|
|
|
p->temp[1][x + y * p->temp_stride[1]] += p->frame_dec[i]->data[1][x + y * p->frame_dec[i]->linesize[1] + offset]; |
|
|
|
|
p->temp[2][x + y * p->temp_stride[2]] += p->frame_dec[i]->data[2][x + y * p->frame_dec[i]->linesize[2] + offset]; |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
return 0; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
static void filter(AVFilterContext *ctx, uint8_t *dst[3], uint8_t *src[3], |
|
|
|
|
int dst_stride[3], int src_stride[3], int width, |
|
|
|
|
int height, uint8_t *qp_store, int qp_stride) |
|
|
|
|
{ |
|
|
|
|
USPPContext *p = ctx->priv; |
|
|
|
|
int x, y, i, j; |
|
|
|
|
const int count = 1<<p->log2_count; |
|
|
|
|
int ret; |
|
|
|
|
|
|
|
|
|
for (i = 0; i < 3; i++) { |
|
|
|
|
int is_chroma = !!i; |
|
|
|
@ -219,12 +295,11 @@ static void filter(USPPContext *p, uint8_t *dst[3], uint8_t *src[3], |
|
|
|
|
memcpy(p->src[i] + (h+block +y) * stride, p->src[i] + (h-y+block-1) * stride, stride); |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
p->frame->linesize[i] = stride; |
|
|
|
|
memset(p->temp[i], 0, (h + 2 * block) * stride * sizeof(int16_t)); |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
if (p->qp) |
|
|
|
|
p->frame->quality = p->qp * FF_QP2LAMBDA; |
|
|
|
|
p->quality = p->qp * FF_QP2LAMBDA; |
|
|
|
|
else { |
|
|
|
|
int qpsum=0; |
|
|
|
|
int qpcount = (height>>4) * (height>>4); |
|
|
|
@ -233,71 +308,11 @@ static void filter(USPPContext *p, uint8_t *dst[3], uint8_t *src[3], |
|
|
|
|
for (x = 0; x < (width>>4); x++) |
|
|
|
|
qpsum += qp_store[x + y * qp_stride]; |
|
|
|
|
} |
|
|
|
|
p->frame->quality = ff_norm_qscale((qpsum + qpcount/2) / qpcount, p->qscale_type) * FF_QP2LAMBDA; |
|
|
|
|
p->quality = ff_norm_qscale((qpsum + qpcount/2) / qpcount, p->qscale_type) * FF_QP2LAMBDA; |
|
|
|
|
} |
|
|
|
|
// init per MB qscale stuff FIXME
|
|
|
|
|
p->frame->height = height + BLOCK; |
|
|
|
|
p->frame->width = width + BLOCK; |
|
|
|
|
|
|
|
|
|
for (i = 0; i < count; i++) { |
|
|
|
|
const int x1 = offset[i+count-1][0]; |
|
|
|
|
const int y1 = offset[i+count-1][1]; |
|
|
|
|
const int x1c = x1 >> p->hsub; |
|
|
|
|
const int y1c = y1 >> p->vsub; |
|
|
|
|
const int BLOCKc = BLOCK >> p->hsub; |
|
|
|
|
int offset; |
|
|
|
|
AVPacket *pkt = p->pkt; |
|
|
|
|
|
|
|
|
|
av_packet_unref(pkt); |
|
|
|
|
pkt->data = p->outbuf; |
|
|
|
|
pkt->size = p->outbuf_size; |
|
|
|
|
|
|
|
|
|
p->frame->data[0] = p->src[0] + x1 + y1 * p->frame->linesize[0]; |
|
|
|
|
p->frame->data[1] = p->src[1] + x1c + y1c * p->frame->linesize[1]; |
|
|
|
|
p->frame->data[2] = p->src[2] + x1c + y1c * p->frame->linesize[2]; |
|
|
|
|
p->frame->format = p->avctx_enc[i]->pix_fmt; |
|
|
|
|
|
|
|
|
|
ret = avcodec_send_frame(p->avctx_enc[i], p->frame); |
|
|
|
|
if (ret < 0) { |
|
|
|
|
av_log(p->avctx_enc[i], AV_LOG_ERROR, "Error sending a frame for encoding\n"); |
|
|
|
|
continue; |
|
|
|
|
} |
|
|
|
|
ret = avcodec_receive_packet(p->avctx_enc[i], pkt); |
|
|
|
|
if (ret < 0) { |
|
|
|
|
av_log(p->avctx_enc[i], AV_LOG_ERROR, "Error receiving a packet from encoding\n"); |
|
|
|
|
continue; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
ret = avcodec_send_packet(p->avctx_dec[i], pkt); |
|
|
|
|
av_packet_unref(pkt); |
|
|
|
|
if (ret < 0) { |
|
|
|
|
av_log(p->avctx_dec[i], AV_LOG_ERROR, "Error sending a packet for decoding\n"); |
|
|
|
|
continue; |
|
|
|
|
} |
|
|
|
|
ret = avcodec_receive_frame(p->avctx_dec[i], p->frame_dec); |
|
|
|
|
if (ret < 0) { |
|
|
|
|
av_log(p->avctx_dec[i], AV_LOG_ERROR, "Error receiving a frame from decoding\n"); |
|
|
|
|
continue; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
offset = (BLOCK-x1) + (BLOCK-y1) * p->frame_dec->linesize[0]; |
|
|
|
|
|
|
|
|
|
for (y = 0; y < height; y++) |
|
|
|
|
for (x = 0; x < width; x++) |
|
|
|
|
p->temp[0][x + y * p->temp_stride[0]] += p->frame_dec->data[0][x + y * p->frame_dec->linesize[0] + offset]; |
|
|
|
|
|
|
|
|
|
if (!src[2] || !dst[2]) |
|
|
|
|
continue; |
|
|
|
|
|
|
|
|
|
offset = (BLOCKc-x1c) + (BLOCKc-y1c) * p->frame_dec->linesize[1]; |
|
|
|
|
|
|
|
|
|
for (y = 0; y < AV_CEIL_RSHIFT(height, p->vsub); y++) { |
|
|
|
|
for (x = 0; x < AV_CEIL_RSHIFT(width, p->hsub); x++) { |
|
|
|
|
p->temp[1][x + y * p->temp_stride[1]] += p->frame_dec->data[1][x + y * p->frame_dec->linesize[1] + offset]; |
|
|
|
|
p->temp[2][x + y * p->temp_stride[2]] += p->frame_dec->data[2][x + y * p->frame_dec->linesize[2] + offset]; |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
ff_filter_execute(ctx, filter_1phase, NULL, NULL, p->count); |
|
|
|
|
|
|
|
|
|
for (j = 0; j < 3; j++) { |
|
|
|
|
int is_chroma = !!j; |
|
|
|
@ -342,6 +357,7 @@ static int config_input(AVFilterLink *inlink) |
|
|
|
|
|
|
|
|
|
uspp->hsub = desc->log2_chroma_w; |
|
|
|
|
uspp->vsub = desc->log2_chroma_h; |
|
|
|
|
uspp->count = 1<<uspp->log2_count; |
|
|
|
|
|
|
|
|
|
for (i = 0; i < 3; i++) { |
|
|
|
|
int is_chroma = !!i; |
|
|
|
@ -360,7 +376,7 @@ static int config_input(AVFilterLink *inlink) |
|
|
|
|
return AVERROR(ENOMEM); |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
for (i = 0; i < (1<<uspp->log2_count); i++) { |
|
|
|
|
for (i = 0; i < uspp->count; i++) { |
|
|
|
|
AVCodecContext *avctx_enc, *avctx_dec; |
|
|
|
|
AVDictionary *opts = NULL; |
|
|
|
|
int ret; |
|
|
|
@ -383,6 +399,8 @@ static int config_input(AVFilterLink *inlink) |
|
|
|
|
avctx_enc->flags = AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_LOW_DELAY; |
|
|
|
|
avctx_enc->strict_std_compliance = FF_COMPLIANCE_EXPERIMENTAL; |
|
|
|
|
avctx_enc->global_quality = 123; |
|
|
|
|
avctx_dec->thread_count = |
|
|
|
|
avctx_enc->thread_count = 1; // We do threading in the filter with muiltiple codecs
|
|
|
|
|
ret = avcodec_open2(avctx_enc, enc, &opts); |
|
|
|
|
av_dict_free(&opts); |
|
|
|
|
if (ret < 0) |
|
|
|
@ -394,15 +412,15 @@ static int config_input(AVFilterLink *inlink) |
|
|
|
|
if (ret < 0) |
|
|
|
|
return ret; |
|
|
|
|
|
|
|
|
|
if (!(uspp->frame[i] = av_frame_alloc())) |
|
|
|
|
return AVERROR(ENOMEM); |
|
|
|
|
if (!(uspp->frame_dec[i] = av_frame_alloc())) |
|
|
|
|
return AVERROR(ENOMEM); |
|
|
|
|
if (!(uspp->pkt[i] = av_packet_alloc())) |
|
|
|
|
return AVERROR(ENOMEM); |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
uspp->outbuf_size = (width + BLOCK) * (height + BLOCK) * 10; |
|
|
|
|
if (!(uspp->frame = av_frame_alloc())) |
|
|
|
|
return AVERROR(ENOMEM); |
|
|
|
|
if (!(uspp->frame_dec = av_frame_alloc())) |
|
|
|
|
return AVERROR(ENOMEM); |
|
|
|
|
if (!(uspp->pkt = av_packet_alloc())) |
|
|
|
|
return AVERROR(ENOMEM); |
|
|
|
|
if (!(uspp->outbuf = av_malloc(uspp->outbuf_size))) |
|
|
|
|
return AVERROR(ENOMEM); |
|
|
|
|
|
|
|
|
@ -464,7 +482,7 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in) |
|
|
|
|
out->height = in->height; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
filter(uspp, out->data, in->data, out->linesize, in->linesize, |
|
|
|
|
filter(ctx, out->data, in->data, out->linesize, in->linesize, |
|
|
|
|
inlink->w, inlink->h, qp_table, qp_stride); |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
@ -492,15 +510,16 @@ static av_cold void uninit(AVFilterContext *ctx) |
|
|
|
|
av_freep(&uspp->src[i]); |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
for (i = 0; i < (1 << uspp->log2_count); i++) { |
|
|
|
|
for (i = 0; i < uspp->count; i++) { |
|
|
|
|
avcodec_free_context(&uspp->avctx_enc[i]); |
|
|
|
|
avcodec_free_context(&uspp->avctx_dec[i]); |
|
|
|
|
av_frame_free(&uspp->frame[i]); |
|
|
|
|
av_frame_free(&uspp->frame_dec[i]); |
|
|
|
|
av_packet_free(&uspp->pkt[i]); |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
av_freep(&uspp->non_b_qp_table); |
|
|
|
|
av_freep(&uspp->outbuf); |
|
|
|
|
av_packet_free(&uspp->pkt); |
|
|
|
|
av_frame_free(&uspp->frame); |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
static const AVFilterPad uspp_inputs[] = { |
|
|
|
@ -528,5 +547,5 @@ const AVFilter ff_vf_uspp = { |
|
|
|
|
FILTER_OUTPUTS(uspp_outputs), |
|
|
|
|
FILTER_PIXFMTS_ARRAY(pix_fmts), |
|
|
|
|
.priv_class = &uspp_class, |
|
|
|
|
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL, |
|
|
|
|
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS, |
|
|
|
|
}; |
|
|
|
|