提交 98637be7 编写于 作者: M Martin Vignali

avcodec/prores_enc : not calculate dct a each quantif search step

Improve encoding speed by 2% (using prores input)
上级 cea5e90b
...@@ -219,7 +219,7 @@ static const uint8_t run_to_cb[16] = { 0x06, 0x06, 0x05, 0x05, 0x04, 0x29, ...@@ -219,7 +219,7 @@ static const uint8_t run_to_cb[16] = { 0x06, 0x06, 0x05, 0x05, 0x04, 0x29,
static const uint8_t lev_to_cb[10] = { 0x04, 0x0A, 0x05, 0x06, 0x04, 0x28, static const uint8_t lev_to_cb[10] = { 0x04, 0x0A, 0x05, 0x06, 0x04, 0x28,
0x28, 0x28, 0x28, 0x4C }; 0x28, 0x28, 0x28, 0x4C };
static void encode_ac_coeffs(AVCodecContext *avctx, PutBitContext *pb, static void encode_ac_coeffs(PutBitContext *pb,
int16_t *in, int blocks_per_slice, int *qmat) int16_t *in, int blocks_per_slice, int *qmat)
{ {
int prev_run = 4; int prev_run = 4;
...@@ -268,16 +268,10 @@ static void fdct_get(FDCTDSPContext *fdsp, uint8_t *pixels, int stride, int16_t* ...@@ -268,16 +268,10 @@ static void fdct_get(FDCTDSPContext *fdsp, uint8_t *pixels, int stride, int16_t*
fdsp->fdct(block); fdsp->fdct(block);
} }
static int encode_slice_plane(AVCodecContext *avctx, int mb_count, static void calc_plane_dct(FDCTDSPContext *fdsp, uint8_t *src, int16_t * blocks, int src_stride, int mb_count, int chroma)
uint8_t *src, int src_stride, uint8_t *buf, unsigned buf_size,
int *qmat, int chroma)
{ {
ProresContext* ctx = avctx->priv_data;
FDCTDSPContext *fdsp = &ctx->fdsp;
LOCAL_ALIGNED(16, int16_t, blocks, [DEFAULT_SLICE_MB_WIDTH << 8]);
int16_t *block; int16_t *block;
int i, blocks_per_slice; int i;
PutBitContext pb;
block = blocks; block = blocks;
for (i = 0; i < mb_count; i++) { for (i = 0; i < mb_count; i++) {
...@@ -291,37 +285,41 @@ static int encode_slice_plane(AVCodecContext *avctx, int mb_count, ...@@ -291,37 +285,41 @@ static int encode_slice_plane(AVCodecContext *avctx, int mb_count,
block += (256 >> chroma); block += (256 >> chroma);
src += (32 >> chroma); src += (32 >> chroma);
} }
}
static int encode_slice_plane(int16_t *blocks, int mb_count, uint8_t *buf, unsigned buf_size, int *qmat, int chroma)
{
int blocks_per_slice;
PutBitContext pb;
blocks_per_slice = mb_count << (2 - chroma); blocks_per_slice = mb_count << (2 - chroma);
init_put_bits(&pb, buf, buf_size); init_put_bits(&pb, buf, buf_size);
encode_dc_coeffs(&pb, blocks, blocks_per_slice, qmat); encode_dc_coeffs(&pb, blocks, blocks_per_slice, qmat);
encode_ac_coeffs(avctx, &pb, blocks, blocks_per_slice, qmat); encode_ac_coeffs(&pb, blocks, blocks_per_slice, qmat);
flush_put_bits(&pb); flush_put_bits(&pb);
return put_bits_ptr(&pb) - pb.buf; return put_bits_ptr(&pb) - pb.buf;
} }
static av_always_inline unsigned encode_slice_data(AVCodecContext *avctx, static av_always_inline unsigned encode_slice_data(AVCodecContext *avctx,
uint8_t *dest_y, uint8_t *dest_u, uint8_t *dest_v, int luma_stride, int16_t * blocks_y, int16_t * blocks_u, int16_t * blocks_v,
int chroma_stride, unsigned mb_count, uint8_t *buf, unsigned data_size, unsigned mb_count, uint8_t *buf, unsigned data_size,
unsigned* y_data_size, unsigned* u_data_size, unsigned* v_data_size, unsigned* y_data_size, unsigned* u_data_size, unsigned* v_data_size,
int qp) int qp)
{ {
ProresContext* ctx = avctx->priv_data; ProresContext* ctx = avctx->priv_data;
*y_data_size = encode_slice_plane(avctx, mb_count, dest_y, luma_stride, *y_data_size = encode_slice_plane(blocks_y, mb_count,
buf, data_size, ctx->qmat_luma[qp - 1], 0); buf, data_size, ctx->qmat_luma[qp - 1], 0);
if (!(avctx->flags & AV_CODEC_FLAG_GRAY)) { if (!(avctx->flags & AV_CODEC_FLAG_GRAY)) {
*u_data_size = encode_slice_plane(avctx, mb_count, dest_u, *u_data_size = encode_slice_plane(blocks_u, mb_count, buf + *y_data_size, data_size - *y_data_size,
chroma_stride, buf + *y_data_size, data_size - *y_data_size, ctx->qmat_chroma[qp - 1], 1);
ctx->qmat_chroma[qp - 1], 1);
*v_data_size = encode_slice_plane(blocks_v, mb_count, buf + *y_data_size + *u_data_size,
*v_data_size = encode_slice_plane(avctx, mb_count, dest_v, data_size - *y_data_size - *u_data_size,
chroma_stride, buf + *y_data_size + *u_data_size, ctx->qmat_chroma[qp - 1], 1);
data_size - *y_data_size - *u_data_size,
ctx->qmat_chroma[qp - 1], 1);
} }
return *y_data_size + *u_data_size + *v_data_size; return *y_data_size + *u_data_size + *v_data_size;
...@@ -366,10 +364,15 @@ static int encode_slice(AVCodecContext *avctx, const AVFrame *pic, int mb_x, ...@@ -366,10 +364,15 @@ static int encode_slice(AVCodecContext *avctx, const AVFrame *pic, int mb_x,
uint8_t *dest_y, *dest_u, *dest_v; uint8_t *dest_y, *dest_u, *dest_v;
unsigned y_data_size = 0, u_data_size = 0, v_data_size = 0; unsigned y_data_size = 0, u_data_size = 0, v_data_size = 0;
ProresContext* ctx = avctx->priv_data; ProresContext* ctx = avctx->priv_data;
FDCTDSPContext *fdsp = &ctx->fdsp;
int tgt_bits = (mb_count * bitrate_table[avctx->profile]) >> 2; int tgt_bits = (mb_count * bitrate_table[avctx->profile]) >> 2;
int low_bytes = (tgt_bits - (tgt_bits >> 3)) >> 3; // 12% bitrate fluctuation int low_bytes = (tgt_bits - (tgt_bits >> 3)) >> 3; // 12% bitrate fluctuation
int high_bytes = (tgt_bits + (tgt_bits >> 3)) >> 3; int high_bytes = (tgt_bits + (tgt_bits >> 3)) >> 3;
LOCAL_ALIGNED(16, int16_t, blocks_y, [DEFAULT_SLICE_MB_WIDTH << 8]);
LOCAL_ALIGNED(16, int16_t, blocks_u, [DEFAULT_SLICE_MB_WIDTH << 8]);
LOCAL_ALIGNED(16, int16_t, blocks_v, [DEFAULT_SLICE_MB_WIDTH << 8]);
luma_stride = pic->linesize[0]; luma_stride = pic->linesize[0];
chroma_stride = pic->linesize[1]; chroma_stride = pic->linesize[1];
...@@ -389,32 +392,40 @@ static int encode_slice(AVCodecContext *avctx, const AVFrame *pic, int mb_x, ...@@ -389,32 +392,40 @@ static int encode_slice(AVCodecContext *avctx, const AVFrame *pic, int mb_x,
chroma_stride, avctx->width >> 1, avctx->height, chroma_stride, avctx->width >> 1, avctx->height,
(uint16_t *) ctx->fill_v, mb_count << 3, 16); (uint16_t *) ctx->fill_v, mb_count << 3, 16);
encode_slice_data(avctx, ctx->fill_y, ctx->fill_u, ctx->fill_v, calc_plane_dct(fdsp, ctx->fill_y, blocks_y, mb_count << 5, mb_count, 0);
mb_count << 5, mb_count << 4, mb_count, buf + hdr_size, calc_plane_dct(fdsp, ctx->fill_u, blocks_u, mb_count << 4, mb_count, 1);
data_size - hdr_size, &y_data_size, &u_data_size, &v_data_size, calc_plane_dct(fdsp, ctx->fill_v, blocks_v, mb_count << 4, mb_count, 1);
*qp);
encode_slice_data(avctx, blocks_y, blocks_u, blocks_v,
mb_count, buf + hdr_size, data_size - hdr_size,
&y_data_size, &u_data_size, &v_data_size,
*qp);
} else { } else {
slice_size = encode_slice_data(avctx, dest_y, dest_u, dest_v, calc_plane_dct(fdsp, dest_y, blocks_y, luma_stride, mb_count, 0);
luma_stride, chroma_stride, mb_count, buf + hdr_size, calc_plane_dct(fdsp, dest_u, blocks_u, chroma_stride, mb_count, 1);
data_size - hdr_size, &y_data_size, &u_data_size, &v_data_size, calc_plane_dct(fdsp, dest_v, blocks_v, chroma_stride, mb_count, 1);
*qp);
slice_size = encode_slice_data(avctx, blocks_y, blocks_u, blocks_v,
mb_count, buf + hdr_size, data_size - hdr_size,
&y_data_size, &u_data_size, &v_data_size,
*qp);
if (slice_size > high_bytes && *qp < qp_end_table[avctx->profile]) { if (slice_size > high_bytes && *qp < qp_end_table[avctx->profile]) {
do { do {
*qp += 1; *qp += 1;
slice_size = encode_slice_data(avctx, dest_y, dest_u, dest_v, slice_size = encode_slice_data(avctx, blocks_y, blocks_u, blocks_v,
luma_stride, chroma_stride, mb_count, buf + hdr_size, mb_count, buf + hdr_size, data_size - hdr_size,
data_size - hdr_size, &y_data_size, &u_data_size, &y_data_size, &u_data_size, &v_data_size,
&v_data_size, *qp); *qp);
} while (slice_size > high_bytes && *qp < qp_end_table[avctx->profile]); } while (slice_size > high_bytes && *qp < qp_end_table[avctx->profile]);
} else if (slice_size < low_bytes && *qp } else if (slice_size < low_bytes && *qp
> qp_start_table[avctx->profile]) { > qp_start_table[avctx->profile]) {
do { do {
*qp -= 1; *qp -= 1;
slice_size = encode_slice_data(avctx, dest_y, dest_u, dest_v, slice_size = encode_slice_data(avctx, blocks_y, blocks_u, blocks_v,
luma_stride, chroma_stride, mb_count, buf + hdr_size, mb_count, buf + hdr_size, data_size - hdr_size,
data_size - hdr_size, &y_data_size, &u_data_size, &y_data_size, &u_data_size, &v_data_size,
&v_data_size, *qp); *qp);
} while (slice_size < low_bytes && *qp > qp_start_table[avctx->profile]); } while (slice_size < low_bytes && *qp > qp_start_table[avctx->profile]);
} }
} }
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册