[x265] [PATCH] slicetype: do not re-calculate AQ cost of B-frames

[Steve Borho]

On Tue, Mar 24, 2015 at 5:18 PM, Steve Borho <steve at borho.org> wrote:
> # HG changeset patch
> # User Steve Borho <steve at borho.org>
> # Date 1427235333 18000
> #      Tue Mar 24 17:15:33 2015 -0500
> # Node ID 59a1cce104ac15d855f2b17aa5f9a997be642b69
> # Parent  41904c8881a20ff502cc55898343f8a757ad36fb
> slicetype: do not re-calculate AQ cost of B-frames

for some compilers, this changes one smoke-test output. for others it
doesn't. I'm holding it up until the last determinism bug is squashed.

> diff -r 41904c8881a2 -r 59a1cce104ac source/encoder/slicetype.cpp
> --- a/source/encoder/slicetype.cpp      Tue Mar 24 17:15:13 2015 -0500
> +++ b/source/encoder/slicetype.cpp      Tue Mar 24 17:15:33 2015 -0500
> @@ -1753,9 +1753,12 @@
>   * re-running lookahead. */
>  int64_t Lookahead::frameCostRecalculate(Lowres** frames, int p0, int p1, int b)
>  {
> +    if (frames[b]->sliceType == X265_TYPE_B)
> +        return frames[b]->costEstAq[b - p0][p1 - b];
> +
>      int64_t score = 0;
>      int *rowSatd = frames[b]->rowSatds[b - p0][p1 - b];
> -    double *qp_offset = (frames[b]->sliceType == X265_TYPE_B) ? frames[b]->qpAqOffset : frames[b]->qpCuTreeOffset;
> +    double *qp_offset = frames[b]->qpCuTreeOffset;
>
>      x265_emms();
>      for (int cuy = m_8x8Height - 1; cuy >= 0; cuy--)



-- 
Steve Borho