2 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2014 Arwa Arif <arwaarif1994@gmail.com>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License along
18 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
19 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * Ultra Slow/Simple Post-processing filter.
26 * Originally written by Michael Niedermayer for the MPlayer project, and
27 * ported by Arwa Arif for FFmpeg.
30 #include "libavutil/avassert.h"
31 #include "libavutil/imgutils.h"
32 #include "libavutil/opt.h"
33 #include "libavutil/pixdesc.h"
35 #include "libavcodec/avcodec.h"
38 #define MAX_LEVEL 8 /* quality levels */
42 const AVClass *av_class;
52 AVCodecContext *avctx_enc[BLOCK*BLOCK];
55 uint8_t *non_b_qp_table;
56 int non_b_qp_alloc_size;
60 #define OFFSET(x) offsetof(USPPContext, x)
61 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
62 static const AVOption uspp_options[] = {
63 { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, FLAGS },
64 { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS },
65 { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, FLAGS },
69 AVFILTER_DEFINE_CLASS(uspp);
71 DECLARE_ALIGNED(8, static const uint8_t, dither)[8][8] = {
72 { 0*4, 48*4, 12*4, 60*4, 3*4, 51*4, 15*4, 63*4, },
73 { 32*4, 16*4, 44*4, 28*4, 35*4, 19*4, 47*4, 31*4, },
74 { 8*4, 56*4, 4*4, 52*4, 11*4, 59*4, 7*4, 55*4, },
75 { 40*4, 24*4, 36*4, 20*4, 43*4, 27*4, 39*4, 23*4, },
76 { 2*4, 50*4, 14*4, 62*4, 1*4, 49*4, 13*4, 61*4, },
77 { 34*4, 18*4, 46*4, 30*4, 33*4, 17*4, 45*4, 29*4, },
78 { 10*4, 58*4, 6*4, 54*4, 9*4, 57*4, 5*4, 53*4, },
79 { 42*4, 26*4, 38*4, 22*4, 41*4, 25*4, 37*4, 21*4, },
82 static const uint8_t offset[511][2] = {
84 { 0, 0}, { 8, 8}, // quality 1
85 { 0, 0}, { 4, 4}, {12, 8}, { 8,12}, // quality 2
86 { 0, 0}, {10, 2}, { 4, 4}, {14, 6}, { 8, 8}, { 2,10}, {12,12}, { 6,14}, // quality 3
88 { 0, 0}, {10, 2}, { 4, 4}, {14, 6}, { 8, 8}, { 2,10}, {12,12}, { 6,14},
89 { 5, 1}, {15, 3}, { 9, 5}, { 3, 7}, {13, 9}, { 7,11}, { 1,13}, {11,15}, // quality 4
91 { 0, 0}, { 8, 0}, { 0, 8}, { 8, 8}, { 5, 1}, {13, 1}, { 5, 9}, {13, 9},
92 { 2, 2}, {10, 2}, { 2,10}, {10,10}, { 7, 3}, {15, 3}, { 7,11}, {15,11},
93 { 4, 4}, {12, 4}, { 4,12}, {12,12}, { 1, 5}, { 9, 5}, { 1,13}, { 9,13},
94 { 6, 6}, {14, 6}, { 6,14}, {14,14}, { 3, 7}, {11, 7}, { 3,15}, {11,15}, // quality 5
96 { 0, 0}, { 8, 0}, { 0, 8}, { 8, 8}, { 4, 0}, {12, 0}, { 4, 8}, {12, 8},
97 { 1, 1}, { 9, 1}, { 1, 9}, { 9, 9}, { 5, 1}, {13, 1}, { 5, 9}, {13, 9},
98 { 3, 2}, {11, 2}, { 3,10}, {11,10}, { 7, 2}, {15, 2}, { 7,10}, {15,10},
99 { 2, 3}, {10, 3}, { 2,11}, {10,11}, { 6, 3}, {14, 3}, { 6,11}, {14,11},
100 { 0, 4}, { 8, 4}, { 0,12}, { 8,12}, { 4, 4}, {12, 4}, { 4,12}, {12,12},
101 { 1, 5}, { 9, 5}, { 1,13}, { 9,13}, { 5, 5}, {13, 5}, { 5,13}, {13,13},
102 { 3, 6}, {11, 6}, { 3,14}, {11,14}, { 7, 6}, {15, 6}, { 7,14}, {15,14},
103 { 2, 7}, {10, 7}, { 2,15}, {10,15}, { 6, 7}, {14, 7}, { 6,15}, {14,15}, // quality 6
105 { 0, 0}, { 8, 0}, { 0, 8}, { 8, 8}, { 0, 2}, { 8, 2}, { 0,10}, { 8,10},
106 { 0, 4}, { 8, 4}, { 0,12}, { 8,12}, { 0, 6}, { 8, 6}, { 0,14}, { 8,14},
107 { 1, 1}, { 9, 1}, { 1, 9}, { 9, 9}, { 1, 3}, { 9, 3}, { 1,11}, { 9,11},
108 { 1, 5}, { 9, 5}, { 1,13}, { 9,13}, { 1, 7}, { 9, 7}, { 1,15}, { 9,15},
109 { 2, 0}, {10, 0}, { 2, 8}, {10, 8}, { 2, 2}, {10, 2}, { 2,10}, {10,10},
110 { 2, 4}, {10, 4}, { 2,12}, {10,12}, { 2, 6}, {10, 6}, { 2,14}, {10,14},
111 { 3, 1}, {11, 1}, { 3, 9}, {11, 9}, { 3, 3}, {11, 3}, { 3,11}, {11,11},
112 { 3, 5}, {11, 5}, { 3,13}, {11,13}, { 3, 7}, {11, 7}, { 3,15}, {11,15},
113 { 4, 0}, {12, 0}, { 4, 8}, {12, 8}, { 4, 2}, {12, 2}, { 4,10}, {12,10},
114 { 4, 4}, {12, 4}, { 4,12}, {12,12}, { 4, 6}, {12, 6}, { 4,14}, {12,14},
115 { 5, 1}, {13, 1}, { 5, 9}, {13, 9}, { 5, 3}, {13, 3}, { 5,11}, {13,11},
116 { 5, 5}, {13, 5}, { 5,13}, {13,13}, { 5, 7}, {13, 7}, { 5,15}, {13,15},
117 { 6, 0}, {14, 0}, { 6, 8}, {14, 8}, { 6, 2}, {14, 2}, { 6,10}, {14,10},
118 { 6, 4}, {14, 4}, { 6,12}, {14,12}, { 6, 6}, {14, 6}, { 6,14}, {14,14},
119 { 7, 1}, {15, 1}, { 7, 9}, {15, 9}, { 7, 3}, {15, 3}, { 7,11}, {15,11},
120 { 7, 5}, {15, 5}, { 7,13}, {15,13}, { 7, 7}, {15, 7}, { 7,15}, {15,15}, // quality 7
122 { 0, 0}, { 8, 0}, { 0, 8}, { 8, 8}, { 4, 4}, {12, 4}, { 4,12}, {12,12},
123 { 0, 4}, { 8, 4}, { 0,12}, { 8,12}, { 4, 0}, {12, 0}, { 4, 8}, {12, 8},
124 { 2, 2}, {10, 2}, { 2,10}, {10,10}, { 6, 6}, {14, 6}, { 6,14}, {14,14},
125 { 2, 6}, {10, 6}, { 2,14}, {10,14}, { 6, 2}, {14, 2}, { 6,10}, {14,10},
126 { 0, 2}, { 8, 2}, { 0,10}, { 8,10}, { 4, 6}, {12, 6}, { 4,14}, {12,14},
127 { 0, 6}, { 8, 6}, { 0,14}, { 8,14}, { 4, 2}, {12, 2}, { 4,10}, {12,10},
128 { 2, 0}, {10, 0}, { 2, 8}, {10, 8}, { 6, 4}, {14, 4}, { 6,12}, {14,12},
129 { 2, 4}, {10, 4}, { 2,12}, {10,12}, { 6, 0}, {14, 0}, { 6, 8}, {14, 8},
130 { 1, 1}, { 9, 1}, { 1, 9}, { 9, 9}, { 5, 5}, {13, 5}, { 5,13}, {13,13},
131 { 1, 5}, { 9, 5}, { 1,13}, { 9,13}, { 5, 1}, {13, 1}, { 5, 9}, {13, 9},
132 { 3, 3}, {11, 3}, { 3,11}, {11,11}, { 7, 7}, {15, 7}, { 7,15}, {15,15},
133 { 3, 7}, {11, 7}, { 3,15}, {11,15}, { 7, 3}, {15, 3}, { 7,11}, {15,11},
134 { 1, 3}, { 9, 3}, { 1,11}, { 9,11}, { 5, 7}, {13, 7}, { 5,15}, {13,15},
135 { 1, 7}, { 9, 7}, { 1,15}, { 9,15}, { 5, 3}, {13, 3}, { 5,11}, {13,11}, // quality 8
136 { 3, 1}, {11, 1}, { 3, 9}, {11, 9}, { 7, 5}, {15, 5}, { 7,13}, {15,13},
137 { 3, 5}, {11, 5}, { 3,13}, {11,13}, { 7, 1}, {15, 1}, { 7, 9}, {15, 9},
138 { 0, 1}, { 8, 1}, { 0, 9}, { 8, 9}, { 4, 5}, {12, 5}, { 4,13}, {12,13},
139 { 0, 5}, { 8, 5}, { 0,13}, { 8,13}, { 4, 1}, {12, 1}, { 4, 9}, {12, 9},
140 { 2, 3}, {10, 3}, { 2,11}, {10,11}, { 6, 7}, {14, 7}, { 6,15}, {14,15},
141 { 2, 7}, {10, 7}, { 2,15}, {10,15}, { 6, 3}, {14, 3}, { 6,11}, {14,11},
142 { 0, 3}, { 8, 3}, { 0,11}, { 8,11}, { 4, 7}, {12, 7}, { 4,15}, {12,15},
143 { 0, 7}, { 8, 7}, { 0,15}, { 8,15}, { 4, 3}, {12, 3}, { 4,11}, {12,11},
144 { 2, 1}, {10, 1}, { 2, 9}, {10, 9}, { 6, 5}, {14, 5}, { 6,13}, {14,13},
145 { 2, 5}, {10, 5}, { 2,13}, {10,13}, { 6, 1}, {14, 1}, { 6, 9}, {14, 9},
146 { 1, 0}, { 9, 0}, { 1, 8}, { 9, 8}, { 5, 4}, {13, 4}, { 5,12}, {13,12},
147 { 1, 4}, { 9, 4}, { 1,12}, { 9,12}, { 5, 0}, {13, 0}, { 5, 8}, {13, 8},
148 { 3, 2}, {11, 2}, { 3,10}, {11,10}, { 7, 6}, {15, 6}, { 7,14}, {15,14},
149 { 3, 6}, {11, 6}, { 3,14}, {11,14}, { 7, 2}, {15, 2}, { 7,10}, {15,10},
150 { 1, 2}, { 9, 2}, { 1,10}, { 9,10}, { 5, 6}, {13, 6}, { 5,14}, {13,14},
151 { 1, 6}, { 9, 6}, { 1,14}, { 9,14}, { 5, 2}, {13, 2}, { 5,10}, {13,10},
152 { 3, 0}, {11, 0}, { 3, 8}, {11, 8}, { 7, 4}, {15, 4}, { 7,12}, {15,12},
153 { 3, 4}, {11, 4}, { 3,12}, {11,12}, { 7, 0}, {15, 0}, { 7, 8}, {15, 8},
156 static void store_slice_c(uint8_t *dst, const uint16_t *src,
157 int dst_stride, int src_stride,
158 int width, int height, int log2_scale)
162 #define STORE(pos) do { \
163 temp = ((src[x + y * src_stride + pos] << log2_scale) + d[pos]) >> 8; \
164 if (temp & 0x100) temp = ~(temp >> 31); \
165 dst[x + y * dst_stride + pos] = temp; \
168 for (y = 0; y < height; y++) {
169 const uint8_t *d = dither[y&7];
170 for (x = 0; x < width; x += 8) {
184 static inline int norm_qscale(int qscale, int type)
187 case FF_QSCALE_TYPE_MPEG1: return qscale;
188 case FF_QSCALE_TYPE_MPEG2: return qscale >> 1;
189 case FF_QSCALE_TYPE_H264: return qscale >> 2;
190 case FF_QSCALE_TYPE_VP56: return (63 - qscale + 2) >> 2;
191 default: av_assert0(0);
196 static void filter(USPPContext *p, uint8_t *dst[3], uint8_t *src[3],
197 int dst_stride[3], int src_stride[3], int width,
198 int height, uint8_t *qp_store, int qp_stride)
201 const int count = 1<<p->log2_count;
203 for (i = 0; i < 3; i++) {
205 int w = width >> (is_chroma ? p->hsub : 0);
206 int h = height >> (is_chroma ? p->vsub : 0);
207 int stride = p->temp_stride[i];
208 int block = BLOCK >> (is_chroma ? p->hsub : 0);
210 if (!src[i] || !dst[i])
212 for (y = 0; y < h; y++) {
213 int index = block + block * stride + y * stride;
215 memcpy(p->src[i] + index, src[i] + y * src_stride[i], w );
216 for (x = 0; x < block; x++) {
217 p->src[i][index - x - 1] = p->src[i][index + x ];
218 p->src[i][index + w + x ] = p->src[i][index + w - x - 1];
221 for (y = 0; y < block; y++) {
222 memcpy(p->src[i] + ( block-1-y) * stride, p->src[i] + ( y+block ) * stride, stride);
223 memcpy(p->src[i] + (h+block +y) * stride, p->src[i] + (h-y+block-1) * stride, stride);
226 p->frame->linesize[i] = stride;
227 memset(p->temp[i], 0, (h + 2 * block) * stride * sizeof(int16_t));
231 p->frame->quality = p->qp * FF_QP2LAMBDA;
234 int qpcount = (height>>4) * (height>>4);
236 for (y = 0; y < (height>>4); y++) {
237 for (x = 0; x < (width>>4); x++)
238 qpsum += qp_store[x + y * qp_stride];
240 p->frame->quality = norm_qscale((qpsum + qpcount/2) / qpcount, p->qscale_type) * FF_QP2LAMBDA;
242 // init per MB qscale stuff FIXME
243 p->frame->height = height;
244 p->frame->width = width;
246 for (i = 0; i < count; i++) {
247 const int x1 = offset[i+count-1][0];
248 const int y1 = offset[i+count-1][1];
249 const int x1c = x1 >> p->hsub;
250 const int y1c = y1 >> p->vsub;
251 const int BLOCKc = BLOCK >> p->hsub;
256 av_init_packet(&pkt);
257 pkt.data = p->outbuf;
258 pkt.size = p->outbuf_size;
260 p->frame->data[0] = p->src[0] + x1 + y1 * p->frame->linesize[0];
261 p->frame->data[1] = p->src[1] + x1c + y1c * p->frame->linesize[1];
262 p->frame->data[2] = p->src[2] + x1c + y1c * p->frame->linesize[2];
263 p->frame->format = p->avctx_enc[i]->pix_fmt;
265 avcodec_encode_video2(p->avctx_enc[i], &pkt, p->frame, &got_pkt_ptr);
266 p->frame_dec = p->avctx_enc[i]->coded_frame;
268 offset = (BLOCK-x1) + (BLOCK-y1) * p->frame_dec->linesize[0];
270 for (y = 0; y < height; y++)
271 for (x = 0; x < width; x++)
272 p->temp[0][x + y * p->temp_stride[0]] += p->frame_dec->data[0][x + y * p->frame_dec->linesize[0] + offset];
274 if (!src[2] || !dst[2])
277 offset = (BLOCKc-x1c) + (BLOCKc-y1c) * p->frame_dec->linesize[1];
279 for (y = 0; y < height>>p->vsub; y++) {
280 for (x = 0; x < width>>p->hsub; x++) {
281 p->temp[1][x + y * p->temp_stride[1]] += p->frame_dec->data[1][x + y * p->frame_dec->linesize[1] + offset];
282 p->temp[2][x + y * p->temp_stride[2]] += p->frame_dec->data[2][x + y * p->frame_dec->linesize[2] + offset];
287 for (j = 0; j < 3; j++) {
291 store_slice_c(dst[j], p->temp[j], dst_stride[j], p->temp_stride[j],
292 width >> (is_chroma ? p->hsub : 0),
293 height >> (is_chroma ? p->vsub : 0),
298 static int query_formats(AVFilterContext *ctx)
300 static const enum PixelFormat pix_fmts[] = {
309 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
313 static int config_input(AVFilterLink *inlink)
316 AVFilterContext *ctx = inlink->dst;
317 USPPContext *uspp = ctx->priv;
318 const int height = inlink->h;
319 const int width = inlink->w;
320 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
323 AVCodec *enc = avcodec_find_encoder(AV_CODEC_ID_SNOW);
325 av_log(ctx, AV_LOG_ERROR, "SNOW encoder not found.\n");
326 return AVERROR(EINVAL);
329 uspp->hsub = desc->log2_chroma_w;
330 uspp->vsub = desc->log2_chroma_h;
332 for (i = 0; i < 3; i++) {
334 int w = ((width + 4 * BLOCK-1) & (~(2 * BLOCK-1))) >> (is_chroma ? uspp->hsub : 0);
335 int h = ((height + 4 * BLOCK-1) & (~(2 * BLOCK-1))) >> (is_chroma ? uspp->vsub : 0);
337 uspp->temp_stride[i] = w;
338 if (!(uspp->temp[i] = av_malloc_array(uspp->temp_stride[i], h * sizeof(int16_t))))
339 return AVERROR(ENOMEM);
340 if (!(uspp->src [i] = av_malloc_array(uspp->temp_stride[i], h * sizeof(uint8_t))))
341 return AVERROR(ENOMEM);
344 for (i = 0; i < (1<<uspp->log2_count); i++) {
345 AVCodecContext *avctx_enc;
346 AVDictionary *opts = NULL;
349 if (!(uspp->avctx_enc[i] = avcodec_alloc_context3(NULL)))
350 return AVERROR(ENOMEM);
352 avctx_enc = uspp->avctx_enc[i];
353 avctx_enc->width = width + BLOCK;
354 avctx_enc->height = height + BLOCK;
355 avctx_enc->time_base = (AVRational){1,25}; // meaningless
356 avctx_enc->gop_size = 300;
357 avctx_enc->max_b_frames = 0;
358 avctx_enc->pix_fmt = inlink->format;
359 avctx_enc->flags = CODEC_FLAG_QSCALE | CODEC_FLAG_LOW_DELAY;
360 avctx_enc->strict_std_compliance = FF_COMPLIANCE_EXPERIMENTAL;
361 avctx_enc->global_quality = 123;
362 av_dict_set(&opts, "no_bitstream", "1", 0);
363 ret = avcodec_open2(avctx_enc, enc, &opts);
367 av_assert0(avctx_enc->codec);
370 uspp->outbuf_size = (width + BLOCK) * (height + BLOCK) * 10;
371 if (!(uspp->frame = av_frame_alloc()))
372 return AVERROR(ENOMEM);
373 if (!(uspp->outbuf = av_malloc(uspp->outbuf_size)))
374 return AVERROR(ENOMEM);
379 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
381 AVFilterContext *ctx = inlink->dst;
382 USPPContext *uspp = ctx->priv;
383 AVFilterLink *outlink = ctx->outputs[0];
387 uint8_t *qp_table = NULL;
389 /* if we are not in a constant user quantizer mode and we don't want to use
390 * the quantizers from the B-frames (B-frames often have a higher QP), we
391 * need to save the qp table from the last non B-frame; this is what the
392 * following code block does */
394 qp_table = av_frame_get_qp_table(in, &qp_stride, &uspp->qscale_type);
396 if (qp_table && !uspp->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) {
399 /* if the qp stride is not set, it means the QP are only defined on
402 w = FF_CEIL_RSHIFT(inlink->w, 4);
406 h = FF_CEIL_RSHIFT(inlink->h, 4);
409 if (w * h > uspp->non_b_qp_alloc_size) {
410 int ret = av_reallocp_array(&uspp->non_b_qp_table, w, h);
412 uspp->non_b_qp_alloc_size = 0;
415 uspp->non_b_qp_alloc_size = w * h;
418 av_assert0(w * h <= uspp->non_b_qp_alloc_size);
419 memcpy(uspp->non_b_qp_table, qp_table, w * h);
423 if (uspp->log2_count && !ctx->is_disabled) {
424 if (!uspp->use_bframe_qp && uspp->non_b_qp_table)
425 qp_table = uspp->non_b_qp_table;
427 if (qp_table || uspp->qp) {
429 /* get a new frame if in-place is not possible or if the dimensions
430 * are not multiple of 8 */
431 if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
432 const int aligned_w = FFALIGN(inlink->w, 8);
433 const int aligned_h = FFALIGN(inlink->h, 8);
435 out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
438 return AVERROR(ENOMEM);
440 av_frame_copy_props(out, in);
441 out->width = in->width;
442 out->height = in->height;
445 filter(uspp, out->data, in->data, out->linesize, in->linesize,
446 inlink->w, inlink->h, qp_table, qp_stride);
452 av_image_copy_plane(out->data[3], out->linesize[3],
453 in ->data[3], in ->linesize[3],
454 inlink->w, inlink->h);
457 return ff_filter_frame(outlink, out);
460 static av_cold void uninit(AVFilterContext *ctx)
462 USPPContext *uspp = ctx->priv;
465 for (i = 0; i < 3; i++) {
466 av_freep(&uspp->temp[i]);
467 av_freep(&uspp->src[i]);
470 for (i = 0; i < (1 << uspp->log2_count); i++) {
471 avcodec_close(uspp->avctx_enc[i]);
472 av_freep(&uspp->avctx_enc[i]);
475 av_freep(&uspp->non_b_qp_table);
476 av_freep(&uspp->outbuf);
477 av_frame_free(&uspp->frame);
480 static const AVFilterPad uspp_inputs[] = {
483 .type = AVMEDIA_TYPE_VIDEO,
484 .config_props = config_input,
485 .filter_frame = filter_frame,
490 static const AVFilterPad uspp_outputs[] = {
493 .type = AVMEDIA_TYPE_VIDEO,
498 AVFilter ff_vf_uspp = {
500 .description = NULL_IF_CONFIG_SMALL("Apply Ultra Simple / Slow Post-processing filter."),
501 .priv_size = sizeof(USPPContext),
503 .query_formats = query_formats,
504 .inputs = uspp_inputs,
505 .outputs = uspp_outputs,
506 .priv_class = &uspp_class,
507 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,