[FFmpeg-devel] [PATCH v10 4/6] libavcodec/dnxucdec: DNxUncompressed decoder

[Michael Niedermayer]

On Fri, Oct 04, 2024 at 11:07:33PM +0200, Martin Schitter wrote:
[...]

> +static int half_add_alpha(AVCodecContext *avctx, AVFrame *frame, const AVPacket *pkt)
> +{
> +    /* ffmpeg doesn't provide RGB half bit depth without alpha channel right now
> +     * we simply add an opaque alpha layer as workaround */
> +
> +    int lw;
> +    const size_t soh = 2;
> +    const uint16_t opaque = 0x3c00;
> +
> +    lw = frame->width;
> +
> +    for(int y = 0; y < frame->height; y++){
> +        for(int x = 0; x < frame->width; x++){
> +            memcpy(&frame->data[0][soh*4*(lw*y + x)], &pkt->data[soh*3*(lw*y + x)], soh*3);
> +            memcpy(&frame->data[0][soh*(4*(lw*y + x) + 3)], &opaque, soh);

opaque is in native byte order within 16bit, pkt->data is a byte array
copying them together cannot be right
also as this is a LE pixfmt, the 2nd part should be wrong



> +        }
> +    }
> +    return pkt->size;
> +}
> +
> +/* DNxUncompressed utilizes a very dense bitpack representation of 10bit and 12bit pixel data.
> +
> +Lines of Image data, which look like in their ordinary 8bit counterpart, contain the most
> +significant upper bits of the pixel data. These sections alternate with shorter segments in
> +which the complementary least significant bits of information get packed in a gapless sequence.
> +
> ++----------------------+ +----------------------+ +------------------------+ +----------~
> +|  8 m.s.bits of R[1]  | |  8 m.s.bits of G[1]  | |  8 m.s.bits of B[1]    | | msb R[2]    ... one line
> ++----------------------+ +----------------------+ +------------------------+ +----------~
> +
> ++---------------------------------------------------------------+  +-----------~
> +| +------------+ +------------+ +------------+ +--------------+ |  | +--------~
> +| | 2 lsb R[2] | | 2 lsb B[1] | | 2 lsb G[1] | | 2 lsb R[1]   | |  | | G[3]lsb    ... LSB bits for line
> +| +------------+ +------------+ +------------+ +--------------+ |  | +--------~
> ++---------------------------- one byte ------------------------ +  +-----------~
> +
> +next line of MSB bytes...   */
> +
> +static int unpack_rg10(AVCodecContext *avctx, AVFrame *frame, const AVPacket *pkt)
> +{
> +    int lw, msp, pack, lsp, p_off;
> +    uint16_t r,g,b;
> +
> +    lw = frame->width;
> +
> +    for(int y = 0; y < frame->height; y++){
> +        for(int x = 0; x < frame->width; x++){
> +            msp = pkt->data[y*3*(lw + lw/4) + 3*x];
> +            p_off = y*(3*(lw + lw/4)) + 3*lw + 3*x/4;
> +            pack = pkt->data[p_off];
> +            lsp = (pack >> (3*x%4)*2) & 0x3;

I would suggest & instead of % because if the compiler doesnt optimize the modulo
then it is the slower operation

thx

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

It is a danger to trust the dream we wish for rather than
the science we have, -- Dr. Kenneth Brown
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20241009/0f57ed36/attachment.sig>