From patchwork Sun Jun 2 22:37:30 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Andreas Rheinhardt X-Patchwork-Id: 13385 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id 04F824485A8 for ; Mon, 3 Jun 2019 01:39:07 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id E337D689C60; Mon, 3 Jun 2019 01:39:06 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mail-wr1-f68.google.com (mail-wr1-f68.google.com [209.85.221.68]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 5A8A2689BEF for ; Mon, 3 Jun 2019 01:39:00 +0300 (EEST) Received: by mail-wr1-f68.google.com with SMTP id c2so10055911wrm.8 for ; Sun, 02 Jun 2019 15:39:00 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id:in-reply-to:references :mime-version:content-transfer-encoding; bh=6hngLBvd14dZhqhiekVy5oq+idBKTH0zrDTF5bvHqGQ=; b=S1Ak9j9pNSn1Xy4gQWKpFguw2yD6SLRpGimRmiQWfObw4xtztYTD8WXkS8J4mpng9U gtd2aT779qM8BWruuLJF+izyXRPBIqmjJY1ymADELzARCq0px5VxtKtef7+CQV/U6pgy 2bOFDiB4hAD2kUqSpQ7eVqTwqt8vh8UO985sZ0lAjn9HVoeWOXaOXudqddI/mN/7cDh7 cINNj8xeEekKh4RCR8bgVC5c9RtO+Y3Xv+OhDO/l1dXIrH0QV9cS8P6P4iD+IKKPZSen sbyntSMuQhJbItSm9OmHlc0MGHndXdjcm8miolSaaFHshpNUJqeDfDEVXsnPmql3Ew3U BmYA== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=6hngLBvd14dZhqhiekVy5oq+idBKTH0zrDTF5bvHqGQ=; b=EwJg1iY2Sikgr/h0GQNwZzaGtqoLhzggch8NJDgiRoXOwhJ6I3xy7tmmSE/2YixL91 6yf8jaWfxGkXki/b2y+sNUkBjYZQwCqYGOZ7kzJOtW2gDKJOEbS65tEGXC2S4ZNwFgtC aAmaxM8cslS/dzLv7PwBWfQntkGrAFr1hpgbIvM9A8afvjsKHOXUZN4s18qzEsFtG06c ZOi0WLuIIF54PYFFDrlljG0XyHdtX8W8JxIxeNZtsYxtDehETvWerC4KB4n7A2aLswH5 CQtnC142P4PMZ0QC3t7C/s4H7LHkwxT7pjV5GGCbTXgMVs8dIZd/iQVMG+O0lxS3oE++ p7uQ== X-Gm-Message-State: APjAAAXGUOzpkvgsFMA7mGpBfMYtdM915ouK/08mfhCGvuvQ5rdConhc d3553AQAjm7T7FUWqVme6q1ojngfP2g= X-Google-Smtp-Source: APXvYqyTP5w5BMDNYJjvXqI41D6ZIv4fImLq/mmW6pWTpGwL60T6QCV3bBVtpVapBEXyBx5e4iIvfA== X-Received: by 2002:adf:df8f:: with SMTP id z15mr14492625wrl.140.1559515139645; Sun, 02 Jun 2019 15:38:59 -0700 (PDT) Received: from localhost.localdomain (ipbcc063db.dynamic.kabel-deutschland.de. [188.192.99.219]) by smtp.gmail.com with ESMTPSA id o1sm23913501wre.76.2019.06.02.15.38.58 (version=TLS1_3 cipher=AEAD-AES256-GCM-SHA384 bits=256/256); Sun, 02 Jun 2019 15:38:59 -0700 (PDT) From: Andreas Rheinhardt To: ffmpeg-devel@ffmpeg.org Date: Mon, 3 Jun 2019 00:37:30 +0200 Message-Id: <20190602223730.10992-4-andreas.rheinhardt@gmail.com> X-Mailer: git-send-email 2.21.0 In-Reply-To: <20190602223730.10992-1-andreas.rheinhardt@gmail.com> References: <20190602223730.10992-1-andreas.rheinhardt@gmail.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH 4/4] cbs_mpeg2: Remove zero byte stuffing X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Andreas Rheinhardt Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Remove superfluous trailing zeros from slices. Because MPEG-2 slices can end with zero bits a safe number of trailing zero bits is always kept. More explicitly, 6 + max{f_code[i][1] - 1, i = 0,1, f_code[i][1] != 0xf} is an upper bound for the number of possible trailing zeros that are part of the slice. Here f_code[i][1] is the relevant value of the picture coding extension the slice belongs to and the maximum of the empty set is zero. It is this number of trailing zero bits that is actually kept. That this is really an upper bound can be seen as follows: a) Every slice actually ends with a macroblock. b) If the last macroblock of a slice ends with a block(i) structure with pattern_code[i] != 0, then the slice ends with an "End of block" VLC code (namely the "End of block" code of the last block with pattern_code[i] != 0). These codes are 10 and 0110, so that in this case there is exactly one trailing zero bit. c) Otherwise, all pattern_code[i] are zero. In this case, if macroblock_pattern is set for the last macroblock of the slice, then by the definition of pattern_code[i] in 6.3.17.4 cbp (derived according to table B.9) must be zero and also the coded_block_pattern_1/2 (if existing) must consist of zeros alone. The value zero for cbp is coded by 0000 0000 1 so that the maximum number of trailing zeros in this case is the length of coded_block_pattern_1/2 which have a length of two resp. six bits. So six trailing zero bits at most. d) Otherwise, if the slice actually ends with the marker bit of the last macroblock, then there are certainly no trailing zero bits at all. e) Otherwise, if the slice ends with a motion_vectors(s) structure (with s = 0 or 1 -- it doesn't matter which one), then it ends with a motion_vector(r,s) (r, s = 0, 1 -- it doesn't matter) structure. This structure ends with motion_code[r][s][1] (always existing) potentially followed by motion_residual[r][s][1] and dmvector[1]. If dmvector[1] exists, and contains a bit different from 0, there is at most one trailing zero bit; if dmvector[1] consists of zeros alone, its length is one according to B.11. motion_residual[r][s][1] (if it exists) has a length of f_code[s][1] - 1 bits and can consist of zero bits alone. Given that the value 0xf for f_code indicates that there is no motion vector of the mentioned type, the length of motion_residual[r][s][1] is bounded by max{f_code[i][1] - 1, i=1,2, f_code[i][1] != 0xf}. The motion_code[r][s][1] can end with at most five zero bits (see B.10) and always contains a bit set to one, so that in this case there are at most 5 + max{f_code[i][1] - 1, i=1,2, f_code[i][1] != 0xf} + 1 zero trailing bits. f) Otherwise, if the last macroblock of the slice ends with a quantiser_scale_code, then there are at most four trailing zero bits, because quantiser_scale_code has a length of five bits and must not attain the value zero. g) Otherwise, the last macroblock ends with the macroblock_modes syntax structure. The potentially existing dct_type at the end might be a zero bit; the frame/field_motion_type isn't present here, because otherwise we would have a motion_vectors(i) (i = 0 or 1 or both) syntax structure, so that e) (or b)-d)) would have applied. spatial_temporal_weight_code might entirely consist of two zero bits. The macroblock_type VLC code always contains a 1 bit and ends with two zero bits at most (see B.2-B.8 for this), so we have maximally 2+2+1 trailing zero bits. The fate test cbs-mpeg2-sony-ct3 had to be adapted because the input file contains trailing zeros that were stripped away; the filesize is reduced from 135 KB to 117 KB. Of course, decoding the smaller output still produces the same frames. Most of these savings happen in between slices rather than after the last slice: The chomp bitstream filter can only reduce the filesize by 50 bytes. Signed-off-by: Andreas Rheinhardt --- libavcodec/cbs_mpeg2.c | 26 ++++++++++++++++++++++++-- tests/ref/fate/cbs-mpeg2-sony-ct3 | 2 +- 2 files changed, 25 insertions(+), 3 deletions(-) diff --git a/libavcodec/cbs_mpeg2.c b/libavcodec/cbs_mpeg2.c index 3aa003286e..b56fecd6fa 100644 --- a/libavcodec/cbs_mpeg2.c +++ b/libavcodec/cbs_mpeg2.c @@ -170,7 +170,7 @@ static int cbs_mpeg2_split_fragment(CodedBitstreamContext *ctx, uint8_t *unit_data; uint32_t start_code = -1, next_start_code = -1; size_t unit_size; - int err, i, unit_type; + int err, i, unit_type, max_trailing_bits = 14; start = avpriv_find_start_code(frag->data, frag->data + frag->data_size, &start_code); @@ -187,10 +187,32 @@ static int cbs_mpeg2_split_fragment(CodedBitstreamContext *ctx, unit_size = end - (start - 1); } else { // Unit runs from start to the beginning of the start code - // pointed to by end (including any padding zeroes). + // pointed to by end (preliminarily including any padding zeroes). unit_size = (end - 4) - (start - 1); } + if (unit_type == MPEG2_START_EXTENSION && unit_size >= 4 && + *start >> 4 == MPEG2_EXTENSION_PICTURE_CODING) { + // The values f_code[0][1], f_code[1][1] are used to improve + // the upper bound for the number of trailing zero bits. + // 6 + max{f_code[i][1] - 1, i = 0,1, f_code[i][1] != 0xf} is + // an upper bound. An f_code value of 0xf means that there is + // no motion vector of the respective type. + max_trailing_bits = start[1] >> 4 == 0xf ? 0 : (start[1] >> 4) - 1; + max_trailing_bits = FFMAX(start[2] >> 4 == 0xf ? + 0 : (start[2] >> 4) - 1, + max_trailing_bits) + 6; + } + + if (MPEG2_START_IS_SLICE(unit_type)) { + const uint8_t *tmp = start + unit_size - 2; + + while (tmp > start && *tmp == 0) + tmp--; + unit_size = FFMIN(unit_size, tmp - start + max_trailing_bits / 8 + + !!(*tmp & 0xff >> 8 - max_trailing_bits % 8) + 2); + } + unit_data = (uint8_t *)start - 1; err = ff_cbs_insert_unit_data(ctx, frag, i, unit_type, diff --git a/tests/ref/fate/cbs-mpeg2-sony-ct3 b/tests/ref/fate/cbs-mpeg2-sony-ct3 index b5b4b12f07..ab226794d5 100644 --- a/tests/ref/fate/cbs-mpeg2-sony-ct3 +++ b/tests/ref/fate/cbs-mpeg2-sony-ct3 @@ -1 +1 @@ -b1e15a09cfffbad801810af0928736ab +033db56a4632c7bbcf729eddd7d5300b