* @see http://wiki.multimedia.cx/index.php?title=Apple_ProRes
*/
-#define A32_BITSTREAM_READER // some ProRes vlc codes require up to 28 bits to be read at once
+#define LONG_BITSTREAM_READER // some ProRes vlc codes require up to 28 bits to be read at once
#include <stdint.h>
#include "libavutil/intmath.h"
#include "avcodec.h"
-#include "dsputil.h"
+#include "proresdsp.h"
#include "get_bits.h"
-#define BITS_PER_SAMPLE 10 ///< output precision of that decoder
-#define BIAS (1 << (BITS_PER_SAMPLE - 1)) ///< bias value for converting signed pixels into unsigned ones
-#define CLIP_MIN (1 << (BITS_PER_SAMPLE - 8)) ///< minimum value for clipping resulting pixels
-#define CLIP_MAX (1 << BITS_PER_SAMPLE) - CLIP_MIN - 1 ///< maximum value for clipping resulting pixels
-
+typedef struct {
+ const uint8_t *index; ///< pointers to the data of this slice
+ int slice_num;
+ int x_pos, y_pos;
+ int slice_width;
+ DECLARE_ALIGNED(16, DCTELEM, blocks[8 * 4 * 64]);
+} ProresThreadData;
typedef struct {
- DSPContext dsp;
+ ProresDSPContext dsp;
AVFrame picture;
ScanTable scantable;
int scantable_type; ///< -1 = uninitialized, 0 = progressive, 1/2 = interlaced
int prev_slice_sf; ///< scalefactor of the previous decoded slice
DECLARE_ALIGNED(16, int16_t, qmat_luma_scaled[64]);
DECLARE_ALIGNED(16, int16_t, qmat_chroma_scaled[64]);
- DECLARE_ALIGNED(16, DCTELEM, blocks[8 * 4 * 64]);
int total_slices; ///< total number of slices in a picture
- const uint8_t **slice_data_index; ///< array of pointers to the data of each slice
+ ProresThreadData *slice_data;
+ int pic_num;
int chroma_factor;
int mb_chroma_factor;
int num_chroma_blocks; ///< number of chrominance blocks in a macroblock
int slice_height_factor;
int num_x_mbs;
int num_y_mbs;
+ int alpha_info;
} ProresContext;
ProresContext *ctx = avctx->priv_data;
ctx->total_slices = 0;
- ctx->slice_data_index = 0;
-
- avctx->pix_fmt = PIX_FMT_YUV422P10; // set default pixel format
+ ctx->slice_data = NULL;
- avctx->bits_per_raw_sample = BITS_PER_SAMPLE;
- dsputil_init(&ctx->dsp, avctx);
+ avctx->bits_per_raw_sample = PRORES_BITS_PER_SAMPLE;
+ ff_proresdsp_init(&ctx->dsp);
avctx->coded_frame = &ctx->picture;
avcodec_get_frame_defaults(&ctx->picture);
hdr_size = AV_RB16(buf);
if (hdr_size > data_size) {
av_log(avctx, AV_LOG_ERROR, "frame data too small\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
version = AV_RB16(buf + 2);
if (version >= 2) {
av_log(avctx, AV_LOG_ERROR,
"unsupported header version: %d\n", version);
- return -1;
+ return AVERROR_INVALIDDATA;
}
width = AV_RB16(buf + 8);
av_log(avctx, AV_LOG_ERROR,
"picture dimension changed: old: %d x %d, new: %d x %d\n",
avctx->width, avctx->height, width, height);
- return -1;
+ return AVERROR_INVALIDDATA;
}
ctx->frame_type = (buf[12] >> 2) & 3;
if (ctx->frame_type > 2) {
av_log(avctx, AV_LOG_ERROR,
"unsupported frame type: %d\n", ctx->frame_type);
- return -1;
+ return AVERROR_INVALIDDATA;
}
ctx->chroma_factor = (buf[12] >> 6) & 3;
default:
av_log(avctx, AV_LOG_ERROR,
"unsupported picture format: %d\n", ctx->pic_format);
- return -1;
+ return AVERROR_INVALIDDATA;
}
if (ctx->scantable_type != ctx->frame_type) {
ctx->picture.top_field_first = ctx->frame_type & 1;
}
+ ctx->alpha_info = buf[17] & 0xf;
+ if (ctx->alpha_info)
+ av_log_missing_feature(avctx, "alpha channel", 0);
+
ctx->qmat_changed = 0;
ptr = buf + 20;
flags = buf[19];
if (flags & 2) {
if (ptr - buf > hdr_size - 64) {
av_log(avctx, AV_LOG_ERROR, "header data too small\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
if (memcmp(ctx->qmat_luma, ptr, 64)) {
memcpy(ctx->qmat_luma, ptr, 64);
hdr_size = data_size > 0 ? buf[0] >> 3 : 0;
if (hdr_size < 8 || hdr_size > data_size) {
av_log(avctx, AV_LOG_ERROR, "picture header too small\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
pic_data_size = AV_RB32(buf + 1);
if (pic_data_size > data_size) {
av_log(avctx, AV_LOG_ERROR, "picture data too small\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
slice_width_factor = buf[7] >> 4;
av_log(avctx, AV_LOG_ERROR,
"unsupported slice dimension: %d x %d\n",
1 << slice_width_factor, 1 << slice_height_factor);
- return -1;
+ return AVERROR_INVALIDDATA;
}
ctx->slice_width_factor = slice_width_factor;
num_slices = num_x_slices * ctx->num_y_mbs;
if (num_slices != AV_RB16(buf + 5)) {
av_log(avctx, AV_LOG_ERROR, "invalid number of slices\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
if (ctx->total_slices != num_slices) {
- av_freep(&ctx->slice_data_index);
- ctx->slice_data_index = av_malloc((num_slices + 1) * sizeof(uint8_t*));
- if (!ctx->slice_data_index)
+ av_freep(&ctx->slice_data);
+ ctx->slice_data = av_malloc((num_slices + 1) * sizeof(ctx->slice_data[0]));
+ if (!ctx->slice_data)
return AVERROR(ENOMEM);
ctx->total_slices = num_slices;
}
if (hdr_size + num_slices * 2 > data_size) {
av_log(avctx, AV_LOG_ERROR, "slice table too small\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
/* parse slice table allowing quick access to the slice data */
data_ptr = index_ptr + num_slices * 2;
for (i = 0; i < num_slices; i++) {
- ctx->slice_data_index[i] = data_ptr;
+ ctx->slice_data[i].index = data_ptr;
data_ptr += AV_RB16(index_ptr + i * 2);
}
- ctx->slice_data_index[i] = data_ptr;
+ ctx->slice_data[i].index = data_ptr;
if (data_ptr > buf + data_size) {
av_log(avctx, AV_LOG_ERROR, "out of slice data\n");
lev_cb_index = lev_to_cb_index[FFMIN(level, 9)];
bits_left = get_bits_left(gb);
- if (bits_left <= 8 && !show_bits(gb, bits_left))
+ if (bits_left <= 0 || (bits_left <= 8 && !show_bits(gb, bits_left)))
return;
run = decode_vlc_codeword(gb, ac_codebook[run_cb_index]);
bits_left = get_bits_left(gb);
- if (bits_left <= 8 && !show_bits(gb, bits_left))
+ if (bits_left <= 0 || (bits_left <= 8 && !show_bits(gb, bits_left)))
return;
level = decode_vlc_codeword(gb, ac_codebook[lev_cb_index]) + 1;
}
-#define CLIP_AND_BIAS(x) (av_clip((x) + BIAS, CLIP_MIN, CLIP_MAX))
-
-/**
- * Add bias value, clamp and output pixels of a slice
- */
-static void put_pixels(const DCTELEM *in, uint16_t *out, int stride,
- int mbs_per_slice, int blocks_per_mb)
-{
- int mb, x, y, src_offset, dst_offset;
- const DCTELEM *src1, *src2;
- uint16_t *dst1, *dst2;
-
- src1 = in;
- src2 = in + (blocks_per_mb << 5);
- dst1 = out;
- dst2 = out + (stride << 3);
-
- for (mb = 0; mb < mbs_per_slice; mb++) {
- for (y = 0, dst_offset = 0; y < 8; y++, dst_offset += stride) {
- for (x = 0; x < 8; x++) {
- src_offset = (y << 3) + x;
-
- dst1[dst_offset + x] = CLIP_AND_BIAS(src1[src_offset]);
- dst2[dst_offset + x] = CLIP_AND_BIAS(src2[src_offset]);
-
- if (blocks_per_mb > 2) {
- dst1[dst_offset + x + 8] =
- CLIP_AND_BIAS(src1[src_offset + 64]);
- dst2[dst_offset + x + 8] =
- CLIP_AND_BIAS(src2[src_offset + 64]);
- }
- }
- }
-
- src1 += blocks_per_mb << 6;
- src2 += blocks_per_mb << 6;
- dst1 += blocks_per_mb << 2;
- dst2 += blocks_per_mb << 2;
- }
-}
-
-
/**
* Decode a slice plane (luma or chroma).
*/
-static void decode_slice_plane(ProresContext *ctx, const uint8_t *buf,
+static void decode_slice_plane(ProresContext *ctx, ProresThreadData *td,
+ const uint8_t *buf,
int data_size, uint16_t *out_ptr,
int linesize, int mbs_per_slice,
int blocks_per_mb, int plane_size_factor,
{
GetBitContext gb;
DCTELEM *block_ptr;
- int i, blk_num, blocks_per_slice;
+ int mb_num, blocks_per_slice;
blocks_per_slice = mbs_per_slice * blocks_per_mb;
- memset(ctx->blocks, 0, 8 * 4 * 64 * sizeof(*ctx->blocks));
+ memset(td->blocks, 0, 8 * 4 * 64 * sizeof(*td->blocks));
init_get_bits(&gb, buf, data_size << 3);
- decode_dc_coeffs(&gb, ctx->blocks, blocks_per_slice);
+ decode_dc_coeffs(&gb, td->blocks, blocks_per_slice);
- decode_ac_coeffs(&gb, ctx->blocks, blocks_per_slice,
+ decode_ac_coeffs(&gb, td->blocks, blocks_per_slice,
plane_size_factor, ctx->scantable.permutated);
/* inverse quantization, inverse transform and output */
- block_ptr = ctx->blocks;
-
- for (blk_num = 0; blk_num < blocks_per_slice; blk_num++, block_ptr += 64) {
- /* TODO: the correct solution shoud be (block_ptr[i] * qmat[i]) >> 1
- * and the input of the inverse transform should be scaled by 2
- * in order to avoid rounding errors.
- * Due to the fact the existing Libav transforms are incompatible with
- * that input I temporally introduced the coarse solution below... */
- for (i = 0; i < 64; i++)
- block_ptr[i] = (block_ptr[i] * qmat[i]) >> 2;
-
- ctx->dsp.idct(block_ptr);
+ block_ptr = td->blocks;
+
+ for (mb_num = 0; mb_num < mbs_per_slice; mb_num++, out_ptr += blocks_per_mb * 4) {
+ ctx->dsp.idct_put(out_ptr, linesize, block_ptr, qmat);
+ block_ptr += 64;
+ if (blocks_per_mb > 2) {
+ ctx->dsp.idct_put(out_ptr + 8, linesize, block_ptr, qmat);
+ block_ptr += 64;
+ }
+ ctx->dsp.idct_put(out_ptr + linesize * 4, linesize, block_ptr, qmat);
+ block_ptr += 64;
+ if (blocks_per_mb > 2) {
+ ctx->dsp.idct_put(out_ptr + linesize * 4 + 8, linesize, block_ptr, qmat);
+ block_ptr += 64;
+ }
}
-
- put_pixels(ctx->blocks, out_ptr, linesize >> 1, mbs_per_slice,
- blocks_per_mb);
}
-static int decode_slice(ProresContext *ctx, int pic_num, int slice_num,
- int mb_x_pos, int mb_y_pos, int mbs_per_slice,
- AVCodecContext *avctx)
+static int decode_slice(AVCodecContext *avctx, void *tdata)
{
+ ProresThreadData *td = tdata;
+ ProresContext *ctx = avctx->priv_data;
+ int mb_x_pos = td->x_pos;
+ int mb_y_pos = td->y_pos;
+ int pic_num = ctx->pic_num;
+ int slice_num = td->slice_num;
+ int mbs_per_slice = td->slice_width;
const uint8_t *buf;
uint8_t *y_data, *u_data, *v_data;
AVFrame *pic = avctx->coded_frame;
int slice_data_size, hdr_size, y_data_size, u_data_size, v_data_size;
int y_linesize, u_linesize, v_linesize;
- buf = ctx->slice_data_index[slice_num];
- slice_data_size = ctx->slice_data_index[slice_num + 1] - buf;
+ buf = ctx->slice_data[slice_num].index;
+ slice_data_size = ctx->slice_data[slice_num + 1].index - buf;
slice_width_factor = av_log2(mbs_per_slice);
if (slice_data_size < 6) {
av_log(avctx, AV_LOG_ERROR, "slice data too small\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
/* parse slice header */
hdr_size = buf[0] >> 3;
y_data_size = AV_RB16(buf + 2);
u_data_size = AV_RB16(buf + 4);
- v_data_size = slice_data_size - y_data_size - u_data_size - hdr_size;
+ v_data_size = hdr_size > 7 ? AV_RB16(buf + 6) :
+ slice_data_size - y_data_size - u_data_size - hdr_size;
- if (v_data_size < 0 || hdr_size < 6) {
+ if (hdr_size + y_data_size + u_data_size + v_data_size > slice_data_size ||
+ v_data_size < 0 || hdr_size < 6) {
av_log(avctx, AV_LOG_ERROR, "invalid data size\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
sf = av_clip(buf[1], 1, 224);
sf = sf > 128 ? (sf - 96) << 2 : sf;
/* scale quantization matrixes according with slice's scale factor */
- /* TODO: this can be SIMD-optimized alot */
+ /* TODO: this can be SIMD-optimized a lot */
if (ctx->qmat_changed || sf != ctx->prev_slice_sf) {
ctx->prev_slice_sf = sf;
for (i = 0; i < 64; i++) {
- ctx->qmat_luma_scaled[i] = ctx->qmat_luma[i] * sf;
- ctx->qmat_chroma_scaled[i] = ctx->qmat_chroma[i] * sf;
+ ctx->qmat_luma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_luma[i] * sf;
+ ctx->qmat_chroma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_chroma[i] * sf;
}
}
/* decode luma plane */
- decode_slice_plane(ctx, buf + hdr_size, y_data_size,
+ decode_slice_plane(ctx, td, buf + hdr_size, y_data_size,
(uint16_t*) (y_data + (mb_y_pos << 4) * y_linesize +
(mb_x_pos << 5)), y_linesize,
mbs_per_slice, 4, slice_width_factor + 2,
ctx->qmat_luma_scaled);
/* decode U chroma plane */
- decode_slice_plane(ctx, buf + hdr_size + y_data_size, u_data_size,
+ decode_slice_plane(ctx, td, buf + hdr_size + y_data_size, u_data_size,
(uint16_t*) (u_data + (mb_y_pos << 4) * u_linesize +
(mb_x_pos << ctx->mb_chroma_factor)),
u_linesize, mbs_per_slice, ctx->num_chroma_blocks,
ctx->qmat_chroma_scaled);
/* decode V chroma plane */
- decode_slice_plane(ctx, buf + hdr_size + y_data_size + u_data_size,
+ decode_slice_plane(ctx, td, buf + hdr_size + y_data_size + u_data_size,
v_data_size,
(uint16_t*) (v_data + (mb_y_pos << 4) * v_linesize +
(mb_x_pos << ctx->mb_chroma_factor)),
slice_num = 0;
+ ctx->pic_num = pic_num;
for (y_pos = 0; y_pos < ctx->num_y_mbs; y_pos++) {
slice_width = 1 << ctx->slice_width_factor;
while (ctx->num_x_mbs - x_pos < slice_width)
slice_width >>= 1;
- if (decode_slice(ctx, pic_num, slice_num, x_pos, y_pos,
- slice_width, avctx) < 0)
- return -1;
+ ctx->slice_data[slice_num].slice_num = slice_num;
+ ctx->slice_data[slice_num].x_pos = x_pos;
+ ctx->slice_data[slice_num].y_pos = y_pos;
+ ctx->slice_data[slice_num].slice_width = slice_width;
slice_num++;
}
}
- return 0;
+ return avctx->execute(avctx, decode_slice,
+ ctx->slice_data, NULL, slice_num,
+ sizeof(ctx->slice_data[0]));
}
if (buf_size < 28 || buf_size < AV_RB32(buf) ||
AV_RB32(buf + 4) != FRAME_ID) {
av_log(avctx, AV_LOG_ERROR, "invalid frame\n");
- return -1;
+ return AVERROR_INVALIDDATA;
}
MOVE_DATA_PTR(8);
frame_hdr_size = decode_frame_header(ctx, buf, buf_size, avctx);
if (frame_hdr_size < 0)
- return -1;
+ return AVERROR_INVALIDDATA;
MOVE_DATA_PTR(frame_hdr_size);
for (pic_num = 0; ctx->picture.interlaced_frame - pic_num + 1; pic_num++) {
pic_data_size = decode_picture_header(ctx, buf, buf_size, avctx);
if (pic_data_size < 0)
- return -1;
+ return AVERROR_INVALIDDATA;
if (decode_picture(ctx, pic_num, avctx))
return -1;
if (ctx->picture.data[0])
avctx->release_buffer(avctx, &ctx->picture);
- av_freep(&ctx->slice_data_index);
+ av_freep(&ctx->slice_data);
return 0;
}
.init = decode_init,
.close = decode_close,
.decode = decode_frame,
- .capabilities = CODEC_CAP_DR1,
+ .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,
.long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)")
};
projects / ffmpeg / blobdiff