This is an automated email from the git hooks/post-receive script. Git pushed a commit to branch master in repository ffmpeg.
commit 4ff36f83eb5c644de7a694ffe8060b19d63b0c4f Author: Charly Morgand-Poyac <[email protected]> AuthorDate: Sun Jul 5 22:45:39 2026 +0200 Commit: charlymp <[email protected]> CommitDate: Sat Jul 11 14:23:12 2026 +0000 lavc/hevcdec: remove data dependent branches from residual decoding Freshly decoded bins are close to equiprobable, so branching on them mispredicts half of the time. Apply the coefficient signs and the clipping with the xor/subtract idiom and FFMIN/FFMAX, store the significant coefficient positions unconditionally while advancing the count by the flag value, and derive the greater1 contexts and the first greater1 index arithmetically from a flag mask. About 7% faster single threaded decoding on high bitrate streams. Signed-off-by: Charly Morgand-Poyac <[email protected]> --- libavcodec/hevc/cabac.c | 54 ++++++++++++++++++++++++------------------------- 1 file changed, 26 insertions(+), 28 deletions(-) diff --git a/libavcodec/hevc/cabac.c b/libavcodec/hevc/cabac.c index cbf774fae4..57f43be06f 100644 --- a/libavcodec/hevc/cabac.c +++ b/libavcodec/hevc/cabac.c @@ -1347,6 +1347,7 @@ void ff_hevc_hls_residual_coding(HEVCLocalContext *lc, const HEVCPPS *pps, }; const uint8_t *ctx_idx_map_p; int scf_offset = 0; + int nb0; if (sps->transform_skip_context_enabled && (transform_skip_flag || lc->cu.cu_transquant_bypass_flag)) { ctx_idx_map_p = &ctx_idx_map[scan_idx][4 * 16]; @@ -1378,13 +1379,15 @@ void ff_hevc_hls_residual_coding(HEVCLocalContext *lc, const HEVCPPS *pps, } } } + // the flag is a fresh bin: store always, advance conditionally + nb0 = nb_significant_coeff_flag; for (n = n_end; n > 0; n--) { - if (significant_coeff_flag_decode(lc, n, scf_offset, ctx_idx_map_p)) { - significant_coeff_flag_idx[nb_significant_coeff_flag] = n; - nb_significant_coeff_flag++; - implicit_non_zero_coeff = 0; - } + int sig = significant_coeff_flag_decode(lc, n, scf_offset, ctx_idx_map_p); + significant_coeff_flag_idx[nb_significant_coeff_flag] = n; + nb_significant_coeff_flag += sig; } + if (nb_significant_coeff_flag != nb0) + implicit_non_zero_coeff = 0; if (implicit_non_zero_coeff == 0) { if (sps->transform_skip_context_enabled && (transform_skip_flag || lc->cu.cu_transquant_bypass_flag)) { @@ -1403,10 +1406,9 @@ void ff_hevc_hls_residual_coding(HEVCLocalContext *lc, const HEVCPPS *pps, scf_offset = 2 + scf_offset; } } - if (significant_coeff_flag_decode_0(lc, c_idx, scf_offset) == 1) { - significant_coeff_flag_idx[nb_significant_coeff_flag] = 0; - nb_significant_coeff_flag++; - } + significant_coeff_flag_idx[nb_significant_coeff_flag] = 0; + nb_significant_coeff_flag += + significant_coeff_flag_decode_0(lc, c_idx, scf_offset); } else { significant_coeff_flag_idx[nb_significant_coeff_flag] = 0; nb_significant_coeff_flag++; @@ -1426,6 +1428,7 @@ void ff_hevc_hls_residual_coding(HEVCLocalContext *lc, const HEVCPPS *pps, int sum_abs = 0; int sign_hidden; int sb_type; + int gt1_mask = 0; // initialize first elem of coeff_bas_level_greater1_flag @@ -1444,18 +1447,16 @@ void ff_hevc_hls_residual_coding(HEVCLocalContext *lc, const HEVCPPS *pps, greater1_ctx = 1; last_nz_pos_in_cg = significant_coeff_flag_idx[0]; + // fresh bins: branchless context update, first set index from the mask for (m = 0; m < (n_end > 8 ? 8 : n_end); m++) { - int inc = (ctx_set << 2) + greater1_ctx; - coeff_abs_level_greater1_flag[m] = - coeff_abs_level_greater1_flag_decode(lc, c_idx, inc); - if (coeff_abs_level_greater1_flag[m]) { - greater1_ctx = 0; - if (first_greater1_coeff_idx == -1) - first_greater1_coeff_idx = m; - } else if (greater1_ctx > 0 && greater1_ctx < 3) { - greater1_ctx++; - } + int inc = (ctx_set << 2) + greater1_ctx; + int flag = coeff_abs_level_greater1_flag_decode(lc, c_idx, inc); + coeff_abs_level_greater1_flag[m] = flag; + gt1_mask |= flag << m; + greater1_ctx = (greater1_ctx + (greater1_ctx - 1U < 2)) & (flag - 1); } + if (gt1_mask) + first_greater1_coeff_idx = ff_ctz(gt1_mask); first_nz_pos_in_cg = significant_coeff_flag_idx[n_end - 1]; if (lc->cu.cu_transquant_bypass_flag || @@ -1477,6 +1478,8 @@ void ff_hevc_hls_residual_coding(HEVCLocalContext *lc, const HEVCPPS *pps, } for (m = 0; m < n_end; m++) { + int64_t sign; + n = significant_coeff_flag_idx[m]; GET_COORD(offset, n); if (m < 8) { @@ -1518,8 +1521,9 @@ void ff_hevc_hls_residual_coding(HEVCLocalContext *lc, const HEVCPPS *pps, if (n == first_nz_pos_in_cg && (sum_abs&1)) trans_coeff_level = -trans_coeff_level; } - if (coeff_sign_flag >> 15) - trans_coeff_level = -trans_coeff_level; + // equiprobable sign: apply it branchless + sign = -(int64_t)(coeff_sign_flag >> 15); + trans_coeff_level = (trans_coeff_level ^ sign) - sign; coeff_sign_flag <<= 1; if(!lc->cu.cu_transquant_bypass_flag) { if (sps->scaling_list_enabled && !(transform_skip_flag && log2_trafo_size > 2)) { @@ -1536,13 +1540,7 @@ void ff_hevc_hls_residual_coding(HEVCLocalContext *lc, const HEVCPPS *pps, } } trans_coeff_level = (trans_coeff_level * (int64_t)scale * (int64_t)scale_m + add) >> shift; - if(trans_coeff_level < 0) { - if((~trans_coeff_level) & 0xFffffffffff8000) - trans_coeff_level = -32768; - } else { - if(trans_coeff_level & 0xffffffffffff8000) - trans_coeff_level = 32767; - } + trans_coeff_level = FFMAX(FFMIN(trans_coeff_level, 32767), -32768); } coeffs[y_c * trafo_size + x_c] = trans_coeff_level; } _______________________________________________ ffmpeg-cvslog mailing list -- [email protected] To unsubscribe send an email to [email protected]
