[FFmpeg-devel] [PATCH 2/7] proresdec2: store precomputed EC parameters
Andreas Rheinhardt
andreas.rheinhardt at outlook.com
Fri Sep 8 11:39:10 EEST 2023
Christophe Gisquet:
> Having the various orders and offsets stored in a codebook is compact
> but causes additional computations. Using instead a table for the
> precomputed results achieve some speedups at the cost of ~132 bytes.
>
> Around 5% speedup.
> ---
> libavcodec/proresdec2.c | 54 +++++++++++++++++++++++++++++++++++------
> 1 file changed, 47 insertions(+), 7 deletions(-)
>
> diff --git a/libavcodec/proresdec2.c b/libavcodec/proresdec2.c
> index 6e243cfc17..65e8b01755 100644
> --- a/libavcodec/proresdec2.c
> +++ b/libavcodec/proresdec2.c
> @@ -427,6 +427,7 @@ static int decode_picture_header(AVCodecContext *avctx, const uint8_t *buf, cons
> # define READ_BITS get_bits
> #endif
>
> +/* Kept for reference and because clearer for first DC */
> #define DECODE_CODEWORD(val, codebook) \
> do { \
> unsigned int rice_order, exp_order, switch_bits; \
> @@ -454,18 +455,41 @@ static int decode_picture_header(AVCodecContext *avctx, const uint8_t *buf, cons
> } \
> } while (0)
>
> +/* number of bits to switch between rice and exp golomb */
> +#define DECODE_CODEWORD2(val, switch_bits, rice_order, diff, offset) \
> + do { \
> + unsigned int q, buf, bits; \
> + \
> + buf = show_bits(gb, 14); \
> + q = 13 - av_log2(buf); \
> + \
> + if (q > switch_bits) { /* exp golomb */ \
> + bits = (q<<1) + (int)diff; \
> + val = READ_BITS(gb, bits) + (int)offset; \
> + } else if (rice_order) { \
> + skip_remaining(gb, q+1); \
> + val = (q << rice_order) + get_bits(gb, rice_order); \
> + } else { \
> + val = q; \
> + skip_remaining(gb, q+1); \
> + } \
> + } while (0)
> +
> +
> #define TOSIGNED(x) (((x) >> 1) ^ (-((x) & 1)))
>
> #define FIRST_DC_CB 0xB8
>
> -static const uint8_t dc_codebook[7] = { 0x04, 0x28, 0x28, 0x4D, 0x4D, 0x70, 0x70};
> +static const char dc_codebook[7][4] = {
> + { 0, 0, 1, -1 }, { 0, 1, 2, -2 }, { 0, 1, 2, -2 },
> + { 1, 2, 2, 0 }, { 1, 2, 2, 0 }, { 0, 3, 4, -8 }, { 0, 3, 4, -8 }
> +};
>
> static av_always_inline int decode_dc_coeffs(GetBitContext *gb, int16_t *out,
> int blocks_per_slice)
> {
> int16_t prev_dc;
> int code, i, sign;
> -
> DECODE_CODEWORD(code, FIRST_DC_CB);
> prev_dc = TOSIGNED(code);
> out[0] = prev_dc;
> @@ -475,7 +499,9 @@ static av_always_inline int decode_dc_coeffs(GetBitContext *gb, int16_t *out,
> code = 5;
> sign = 0;
> for (i = 1; i < blocks_per_slice; i++, out += 64) {
> - DECODE_CODEWORD(code, dc_codebook[FFMIN(code, 6U)]);
> + unsigned int dccb = FFMIN(code, 6U);
> + DECODE_CODEWORD2(code, dc_codebook[dccb][0], dc_codebook[dccb][1],
> + dc_codebook[dccb][2], dc_codebook[dccb][3]);
> if(code) sign ^= -(code & 1);
> else sign = 0;
> prev_dc += (((code + 1) >> 1) ^ sign) - sign;
> @@ -485,8 +511,18 @@ static av_always_inline int decode_dc_coeffs(GetBitContext *gb, int16_t *out,
> }
>
> // adaptive codebook switching lut according to previous run/level values
> -static const uint8_t run_to_cb[16] = { 0x06, 0x06, 0x05, 0x05, 0x04, 0x29, 0x29, 0x29, 0x29, 0x28, 0x28, 0x28, 0x28, 0x28, 0x28, 0x4C };
> -static const uint8_t lev_to_cb[10] = { 0x04, 0x0A, 0x05, 0x06, 0x04, 0x28, 0x28, 0x28, 0x28, 0x4C };
> +static const char run_to_cb[16][4] = {
> + { 2, 0, -1, 1 }, { 2, 0, -1, 1 }, { 1, 0, 0, 0 }, { 1, 0, 0, 0 }, { 0, 0, 1, -1 },
> + { 1, 1, 1, 0 }, { 1, 1, 1, 0 }, { 1, 1, 1, 0 }, { 1, 1, 1, 0 },
> + { 0, 1, 2, -2 }, { 0, 1, 2, -2 }, { 0, 1, 2, -2 }, { 0, 1, 2, -2 }, { 0, 1, 2, -2 }, { 0, 1, 2, -2 },
> + { 0, 2, 3, -4 }
> +};
> +
> +static const char lev_to_cb[10][4] = {
You seem to require signed chars here, but this is not a given on many
platforms. Better use int8_t.
> + { 0, 0, 1, -1 }, { 2, 0, 0, -1 }, { 1, 0, 0, 0 }, { 2, 0, -1, 1 }, { 0, 0, 1, -1 },
> + { 0, 1, 2, -2 }, { 0, 1, 2, -2 }, { 0, 1, 2, -2 }, { 0, 1, 2, -2 },
> + { 0, 2, 3, -4 }
> +};
>
> static av_always_inline int decode_ac_coeffs(AVCodecContext *avctx, GetBitContext *gb,
> int16_t *out, int blocks_per_slice)
> @@ -504,18 +540,22 @@ static av_always_inline int decode_ac_coeffs(AVCodecContext *avctx, GetBitContex
> block_mask = blocks_per_slice - 1;
>
> for (pos = block_mask;;) {
> + unsigned int runcb = FFMIN(run, 15);
> + unsigned int levcb = FFMIN(level, 9);
> bits_rem = get_bits_left(gb);
> if (!bits_rem || (bits_rem < 16 && !show_bits(gb, bits_rem)))
> break;
>
> - DECODE_CODEWORD(run, run_to_cb[FFMIN(run, 15)]);
> + DECODE_CODEWORD2(run, run_to_cb[runcb][0], run_to_cb[runcb][1],
> + run_to_cb[runcb][2], run_to_cb[runcb][3]);
> pos += run + 1;
> if (pos >= max_coeffs) {
> av_log(avctx, AV_LOG_ERROR, "ac tex damaged %d, %d\n", pos, max_coeffs);
> return AVERROR_INVALIDDATA;
> }
>
> - DECODE_CODEWORD(level, lev_to_cb[FFMIN(level, 9)]);
> + DECODE_CODEWORD2(level, lev_to_cb[levcb][0], lev_to_cb[levcb][1],
> + lev_to_cb[levcb][2], lev_to_cb[levcb][3]);
> level += 1;
>
> i = pos >> log2_block_count;
More information about the ffmpeg-devel
mailing list