[FFmpeg-devel] 回复: [PATCH v2 10/20] avcodec/h274: add ff_h274_hash functions
Wu Jianhua
toqsxw at outlook.com
Sun Jun 1 00:46:30 EEST 2025
Andreas Rheinhardt <andreas.rheinhardt at outlook.com>:
> toqsxw at gmail.com:
>> From: Wu Jianhua <toqsxw at outlook.com>
>>
>> Signed-off-by: Wu Jianhua <toqsxw at outlook.com>
>> ---
>> libavcodec/h274.c | 155 ++++++++++++++++++++++++++++++++++++++++++++++
>> libavcodec/h274.h | 7 +++
>> 2 files changed, 162 insertions(+)
>>
>> diff --git a/libavcodec/h274.c b/libavcodec/h274.c
>> index 5709200322..e46926e4cc 100644
>> --- a/libavcodec/h274.c
>> +++ b/libavcodec/h274.c
>> @@ -26,7 +26,11 @@
>> */
>>
>> #include "libavutil/avassert.h"
>> +#include "libavutil/bswap.h"
>> +#include "libavutil/crc.h"
>> #include "libavutil/imgutils.h"
>> +#include "libavutil/md5.h"
>> +#include "libavutil/mem.h"
>>
>> #include "h274.h"
>>
>> @@ -790,3 +794,154 @@ static const int8_t R64T[64][64] = {
>> 17, -16, 15, -14, 13, -12, 11, -10, 9, -8, 7, -6, 4, -3, 2, -1,
>> }
>> };
>> +
>> +static int verify_plane_md5(struct AVMD5 *ctx,
>> + const uint8_t *src, const int w, const int h, const int stride,
>> + const uint8_t *expected)
>> +{
>> +#define MD5_SIZE 16
>> + uint8_t md5[MD5_SIZE];
>> + av_md5_init(ctx);
>> + for (int j = 0; j < h; j++) {
>> + av_md5_update(ctx, src, w);
>> + src += stride;
>> + }
>> + av_md5_final(ctx, md5);
>> +
>> + if (memcmp(md5, expected, MD5_SIZE))
>> + return AVERROR_INVALIDDATA;
>> +
>> + return 0;
>> +}
>> +
>> +static int verify_plane_crc(const uint8_t *src, const int w, const int h, const int stride,
>> + uint16_t expected)
>> +{
>> + uint32_t crc = 0x0F1D; // CRC-16-CCITT-AUG
>> + const AVCRC *ctx = av_crc_get_table(AV_CRC_16_CCITT);
>> +
>> + expected = av_le2ne32(expected);
>> + for (int j = 0; j < h; j++) {
>> + crc = av_crc(ctx, crc, src, w);
>
> Did you test this with >8 bit content? For this type of content, it can
> only work on either LE or BE (which one depends upon whether the spec
> wants the CRC applied on LE or BE data), but not both. Same for md5.
>
Hi Andreas,
I tested the >8 bits on LE, but you are right. There may be some issues
on BE. I didn't test them on BE for the reason that I don't have a BE machine.
If it is endian-related, I may need to add the bswap16_buf before I generate
the hash.
>> + src += stride;
>> + }
>> + crc = av_bswap16(crc);
>> +
>> + if (crc != expected)
>> + return AVERROR_INVALIDDATA;
>> +
>> + return 0;
>> +}
>> +
>> +#define CAL_CHECKSUM(pixel) ((pixel) ^ xor_mask)
>> +static int verify_plane_checksum(const uint8_t *src, const int w, const int h, const int stride, const int ps,
>> + uint32_t expected)
>> +{
>> + uint32_t checksum = 0;
>> + expected = av_le2ne32(expected);
>> +
>> + for (int y = 0; y < h; y++) {
>> + for (int x = 0; x < w; x++) {
>> + const int xor_mask = (x & 0xFF) ^ (y & 0xFF) ^ (x >> 8) ^ (y >> 8);
>> + checksum += CAL_CHECKSUM(src[x << ps]);
>> + if (ps)
>> + checksum += CAL_CHECKSUM(src[(x << ps) + 1]);
>
> It seems to me that you think that ps != 0 is equivalent to a >8 bit
> pixel format like AV_PIX_FMT_YUV420P10. This assumption may be true now,
> but it is not true for e.g. NV12, which may be used by future hwaccels.
> Maybe we should not verify checksums when hardware acceleration is in use?
>
Yeah. It needs to download the hardware acceleration frame and add some
changes for reading uv interleaved format. The current hash generation
only works for the YUV planar format. I'm not sure if the developer of
hardware acceleration needs it, but it's a good point. I'll keep it in
mind and see what we can do when the hardware acceleration is introduced
for VVC.
Thanks,
Jianhua
>> + }
>> + src += stride;
>> + }
>> +
>> + if (checksum != expected)
>> + return AVERROR_INVALIDDATA;
>> +
>> + return 0;
>> +}
>> +
>> +enum {
>> + HASH_MD5SUM,
>> + HASH_CRC,
>> + HASH_CHECKSUM,
>> + HASH_LAST = HASH_CHECKSUM,
>> +};
>> +
>> +struct H274HashContext {
>> + int type;
>> + struct AVMD5 *ctx;
>> +};
>> +
>> +void ff_h274_hash_freep(H274HashContext **ctx)
>> +{
>> + if (*ctx) {
>> + H274HashContext *c = *ctx;
>> + if (c->ctx)
>> + av_free(c->ctx);
>> + av_freep(ctx);
>> + }
>> +}
>> +
>> +int ff_h274_hash_init(H274HashContext **ctx, const int type)
>> +{
>> + H274HashContext *c;
>> +
>> + if (type > HASH_LAST || !ctx)
>> + return AVERROR(EINVAL);
>> +
>> + c = *ctx;
>> + if (c) {
>> + if (c->type != type) {
>> + if (c->type == HASH_MD5SUM)
>> + av_freep(&c->ctx);
>> + c->type = type;
>> + }
>> + } else {
>> + c = av_mallocz(sizeof(H274HashContext));
>> + if (!c)
>> + return AVERROR(ENOMEM);
>> + c->type = type;
>> + *ctx = c;
>> + }
>> +
>> + if (type == HASH_MD5SUM && !c->ctx) {
>> + c->ctx = av_md5_alloc();
>> + if (!c->ctx)
>> + return AVERROR(ENOMEM);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +int ff_h274_hash_verify(H274HashContext *c, const H274SEIPictureHash *hash,
>> + const AVFrame *frame, const int coded_width, const int coded_height)
>> +{
>> + const AVPixFmtDescriptor *desc;
>> + int err = 0;
>> +
>> + if (!c || !hash || !frame)
>> + return AVERROR(EINVAL);
>> +
>> + if (c->type != hash->hash_type)
>> + return AVERROR(EINVAL);
>> +
>> + desc = av_pix_fmt_desc_get(frame->format);
>> + if (!desc)
>> + return AVERROR(EINVAL);
>> +
>> + for (int i = 0; i < desc->nb_components; i++) {
>> + const int w = i ? (coded_width >> desc->log2_chroma_w) : coded_width;
>> + const int h = i ? (coded_height >> desc->log2_chroma_h) : coded_height;
>> + const int ps = desc->comp[i].step - 1;
>> + const uint8_t *src = frame->data[i];
>> + const int stride = frame->linesize[i];
>> +
>> + if (c->type == HASH_MD5SUM)
>> + err = verify_plane_md5(c->ctx, src, w << ps, h, stride, hash->md5[i]);
>> + else if (c->type == HASH_CRC)
>> + err = verify_plane_crc(src, w << ps, h, stride, hash->crc[i]);
>> + else if (c->type == HASH_CHECKSUM)
>> + err = verify_plane_checksum(src, w, h, stride, ps, hash->checksum[i]);
>> + if (err < 0)
>> + goto fail;
>> + }
>> +
>> +fail:
>> + return err;
>> +}
>> diff --git a/libavcodec/h274.h b/libavcodec/h274.h
>> index e1803edaf3..0ca5f0cbab 100644
>> --- a/libavcodec/h274.h
>> +++ b/libavcodec/h274.h
>> @@ -64,6 +64,8 @@ int ff_h274_apply_film_grain(AVFrame *out, const AVFrame *in,
>> H274FilmGrainDatabase *db,
>> const AVFilmGrainParams *params);
>>
>> +typedef struct H274HashContext H274HashContext;
>> +
>> typedef struct H274SEIPictureHash {
>> int present;
>> union {
>> @@ -74,4 +76,9 @@ typedef struct H274SEIPictureHash {
>> uint8_t hash_type;
>> } H274SEIPictureHash;
>>
>> +int ff_h274_hash_init(H274HashContext **c, int type);
>> +int ff_h274_hash_verify(H274HashContext *c, const H274SEIPictureHash *hash,
>> + const AVFrame *frame, int coded_width, int coded_height);
>> +void ff_h274_hash_freep(H274HashContext **c);
>> +
>> #endif /* AVCODEC_H274_H */
More information about the ffmpeg-devel
mailing list