[FFmpeg-devel] [PATCH v2] avfilter/vpp_qsv: fix regression on older api versions (e.g. 1.11)

[Soft Works]

> -----Original Message-----
> From: Xiang, Haihao <haihao.xiang at intel.com>
> Sent: Tuesday, January 4, 2022 3:16 AM
> To: ffmpeg-devel at ffmpeg.org
> Cc: softworkz at hotmail.com
> Subject: Re: [FFmpeg-devel] [PATCH v2] avfilter/vpp_qsv: fix regression on
> older api versions (e.g. 1.11)
> 
> On Sun, 2022-01-02 at 03:41 +0000, ffmpegagent wrote:
> > From: softworkz <softworkz at hotmail.com>
> >
> > Commit 8b83dad82512a6948b63408f964463b063ad24c9 introduced a
> > regression in a way that scaling via vpp_qsv doesn't work any longer
> > for devices with an MSDK runtime version lower than 1.19. This is true
> > for older CPUs which are stuck at 1.11.
> > The commit added checks for the compile-sdk version but it didn't test
> > for the runtime version.
> >
> > Signed-off-by: softworkz <softworkz at hotmail.com>
> > ---
> >     avfilter/vpp_qsv: fix regression on older api versions (e.g. 1.11)
> >
> >     Fix a recently introduced regression when using QSV VPP.
> >
> >     v2: Fixed commit message wrapping
> >
> > Published-As:
> > https://github.com/ffstaging/FFmpeg/releases/tag/pr-ffstaging-
> 15%2Fsoftworkz%2Fqsv_vpp_regression-v2
> > Fetch-It-Via: git fetch https://github.com/ffstaging/FFmpeg pr-ffstaging-
> > 15/softworkz/qsv_vpp_regression-v2
> > Pull-Request: https://github.com/ffstaging/FFmpeg/pull/15
> >
> > Range-diff vs v1:
> >
> >  1:  eea966f702 ! 1:  ccede9d840 avfilter/vpp_qsv: fix regression on older
> api
> > versions (e.g. 1.11)
> >      @@ Metadata
> >        ## Commit message ##
> >           avfilter/vpp_qsv: fix regression on older api versions (e.g.
> 1.11)
> >
> >      -    Commit 8b83dad82512a6948b63408f964463b063ad24c9 introduced a
> > regression in a way
> >      -    that scaling via vpp_qsv didn't work any longer for devices with
> an
> > MSDK runtime
> >      -    version lower than 1.19. This is true for older CPUs which are
> stuck
> > at 1.11.
> >      -    The commit added checks for the compile-sdk version but it didn't
> > test for the
> >      -    runtime version.
> >      +    Commit 8b83dad82512a6948b63408f964463b063ad24c9 introduced a
> >      +    regression in a way that scaling via vpp_qsv doesn't work any
> longer
> >      +    for devices with an MSDK runtime version lower than 1.19. This is
> > true
> >      +    for older CPUs which are stuck at 1.11.
> >      +    The commit added checks for the compile-sdk version but it didn't
> > test
> >      +    for the runtime version.
> >
> >           Signed-off-by: softworkz <softworkz at hotmail.com>
> >
> >
> >
> >  libavfilter/vf_vpp_qsv.c | 32 ++++++++++++++++++++------------
> >  1 file changed, 20 insertions(+), 12 deletions(-)
> >
> > diff --git a/libavfilter/vf_vpp_qsv.c b/libavfilter/vf_vpp_qsv.c
> > index ab58a5777e..09590157e3 100644
> > --- a/libavfilter/vf_vpp_qsv.c
> > +++ b/libavfilter/vf_vpp_qsv.c
> > @@ -139,8 +139,9 @@ static const AVOption options[] = {
> >      { "height", "Output video height", OFFSET(oh), AV_OPT_TYPE_STRING, {
> > .str="w*ch/cw" }, 0, 255, .flags = FLAGS },
> >      { "format", "Output pixel format", OFFSET(output_format_str),
> > AV_OPT_TYPE_STRING, { .str = "same" }, .flags = FLAGS },
> >      { "async_depth", "Internal parallelization depth, the higher the value
> > the higher the latency.", OFFSET(async_depth), AV_OPT_TYPE_INT, { .i64 = 0
> },
> > 0, INT_MAX, .flags = FLAGS },
> > +#ifdef QSV_HAVE_SCALING_CONFIG
> >      { "scale_mode", "scale mode: 0=auto, 1=low power, 2=high quality",
> > OFFSET(scale_mode), AV_OPT_TYPE_INT, { .i64 = MFX_SCALING_MODE_DEFAULT },
> > MFX_SCALING_MODE_DEFAULT, MFX_SCALING_MODE_QUALITY, .flags = FLAGS, "scale
> > mode" },
> > -
> > +#endif
> >      { NULL }
> >  };
> >
> > @@ -304,6 +305,7 @@ static int config_output(AVFilterLink *outlink)
> >      QSVVPPParam     param = { NULL };
> >      QSVVPPCrop      crop  = { 0 };
> >      mfxExtBuffer    *ext_buf[ENH_FILTERS_COUNT];
> > +    mfxVersion      mfx_version;
> >      AVFilterLink    *inlink = ctx->inputs[0];
> >      enum AVPixelFormat in_format;
> >
> > @@ -317,6 +319,11 @@ static int config_output(AVFilterLink *outlink)
> >      param.ext_buf       = ext_buf;
> >      param.async_depth   = vpp->async_depth;
> >
> > +    if (MFXQueryVersion(vpp->qsv->session, &mfx_version) < 0) {
> > +        av_log(ctx, AV_LOG_ERROR, "Failed to query mfx version.\n");
> > +        return AVERROR(EINVAL);
> > +    }
> > +
> >      if (inlink->format == AV_PIX_FMT_QSV) {
> >           if (!inlink->hw_frames_ctx || !inlink->hw_frames_ctx->data)
> >               return AVERROR(EINVAL);
> > @@ -467,19 +474,20 @@ static int config_output(AVFilterLink *outlink)
> >  #endif
> >      }
> >
> > -    if (inlink->w != outlink->w || inlink->h != outlink->h) {
> >  #ifdef QSV_HAVE_SCALING_CONFIG
> > -        memset(&vpp->scale_conf, 0, sizeof(mfxExtVPPScaling));
> > -        vpp->scale_conf.Header.BufferId    = MFX_EXTBUFF_VPP_SCALING;
> > -        vpp->scale_conf.Header.BufferSz    = sizeof(mfxExtVPPScaling);
> > -        vpp->scale_conf.ScalingMode        = vpp->scale_mode;
> > -
> > -        param.ext_buf[param.num_ext_buf++] = (mfxExtBuffer*)&vpp-
> >scale_conf;
> > -#else
> > -        av_log(ctx, AV_LOG_WARNING, "The QSV VPP Scale option is "
> > -            "not supported with this MSDK version.\n");
> > -#endif
> > +    if (inlink->w != outlink->w || inlink->h != outlink->h) {
> > +        if (QSV_RUNTIME_VERSION_ATLEAST(mfx_version, 1, 19)) {
> > +            memset(&vpp->scale_conf, 0, sizeof(mfxExtVPPScaling));
> > +            vpp->scale_conf.Header.BufferId    = MFX_EXTBUFF_VPP_SCALING;
> > +            vpp->scale_conf.Header.BufferSz    = sizeof(mfxExtVPPScaling);
> > +            vpp->scale_conf.ScalingMode        = vpp->scale_mode;
> > +
> > +            param.ext_buf[param.num_ext_buf++] = (mfxExtBuffer*)&vpp-
> > >scale_conf;
> > +        } else
> > +            av_log(ctx, AV_LOG_WARNING, "The QSV VPP Scale option is "
> > +                "not supported with this MSDK version.\n");
> >      }
> > +#endif
> >
> >      if (vpp->use_frc || vpp->use_crop || vpp->deinterlace || vpp->denoise
> ||
> >          vpp->detail || vpp->procamp || vpp->rotate || vpp->hflip ||
> 
> LGTM. I wonder whether you installed libmfx and msdk runtime from
> different packages, so the mfx version is higher than the runtime version?

It's about a user with a Broadwell CPU on Windows. The latest drivers' MSDK 
version is 1.11 and our ffmpeg is compiled against MSDK 1.35.
(a setup that is basically working fine, just the runtime checks are missing 
at several places)

Thanks,
softworkz