On 28/02/17 00:31, wm4 wrote:
> On Tue, 28 Feb 2017 00:11:26 +0000
> Mark Thompson <[email protected]> wrote:
> 
>> ---
>> Don't laugh.
>>
>> Thanks to wm4 for debugging this issue.
>>
>>
>>  libavfilter/vf_deinterlace_vaapi.c | 15 +++++++++++++++
>>  libavutil/hwcontext_vaapi.c        |  3 ++-
>>  libavutil/hwcontext_vaapi.h        |  5 +++++
>>  3 files changed, 22 insertions(+), 1 deletion(-)
>>
>> diff --git a/libavfilter/vf_deinterlace_vaapi.c 
>> b/libavfilter/vf_deinterlace_vaapi.c
>> index 526ba6c87..6a983e1d5 100644
>> --- a/libavfilter/vf_deinterlace_vaapi.c
>> +++ b/libavfilter/vf_deinterlace_vaapi.c
>> @@ -22,6 +22,7 @@
>>  #include <va/va_vpp.h>
>>  
>>  #include "libavutil/avassert.h"
>> +#include "libavutil/common.h"
>>  #include "libavutil/hwcontext.h"
>>  #include "libavutil/hwcontext_vaapi.h"
>>  #include "libavutil/mem.h"
>> @@ -210,6 +211,12 @@ static int 
>> deint_vaapi_build_filter_params(AVFilterContext *avctx)
>>          return AVERROR(EIO);
>>      }
>>  
>> +    if (ctx->hwctx->driver_quirks &
>> +        AV_VAAPI_DRIVER_QUIRK_REVERSED_FILTER_REFERENCES) {
>> +        FFSWAP(unsigned int, ctx->pipeline_caps.num_backward_references,
>> +               ctx->pipeline_caps.num_forward_references);
>> +    }
>> +
>>      ctx->queue_depth = ctx->pipeline_caps.num_backward_references +
>>                         ctx->pipeline_caps.num_forward_references + 1;
>>      if (ctx->queue_depth > MAX_REFERENCES) {
>> @@ -488,6 +495,14 @@ static int deint_vaapi_filter_frame(AVFilterLink 
>> *inlink, AVFrame *input_frame)
>>              params.num_backward_references =
>>                  ctx->pipeline_caps.num_backward_references;
>>  
>> +            if (ctx->hwctx->driver_quirks &
>> +                AV_VAAPI_DRIVER_QUIRK_REVERSED_FILTER_REFERENCES) {
>> +                FFSWAP(VASurfaceID*, params.forward_references,
>> +                       params.backward_references);
>> +                FFSWAP(unsigned int, params.num_forward_references,
>> +                       params.num_backward_references);
>> +            }
>> +
>>          } else {
>>              params.filters     = NULL;
>>              params.num_filters = 0;
>> diff --git a/libavutil/hwcontext_vaapi.c b/libavutil/hwcontext_vaapi.c
>> index 23bce7660..bfe4d078a 100644
>> --- a/libavutil/hwcontext_vaapi.c
>> +++ b/libavutil/hwcontext_vaapi.c
>> @@ -265,7 +265,8 @@ static const struct {
>>      {
>>          "Intel i965 (Quick Sync)",
>>          "i965",
>> -        AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS,
>> +        AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS |
>> +        AV_VAAPI_DRIVER_QUIRK_REVERSED_FILTER_REFERENCES,
>>      },
>>      {
>>          "Intel iHD",
>> diff --git a/libavutil/hwcontext_vaapi.h b/libavutil/hwcontext_vaapi.h
>> index 529d26e07..4b17f3f78 100644
>> --- a/libavutil/hwcontext_vaapi.h
>> +++ b/libavutil/hwcontext_vaapi.h
>> @@ -58,6 +58,11 @@ enum {
>>       * and the results of the vaQuerySurfaceAttributes() call will be faked.
>>       */
>>      AV_VAAPI_DRIVER_QUIRK_SURFACE_ATTRIBUTES = (1 << 3),
>> +
>> +    /**
>> +     * The driver requires past frames as forward references.
>> +     */
>> +    AV_VAAPI_DRIVER_QUIRK_REVERSED_FILTER_REFERENCES = (1 << 4),
>>  };
>>  
>>  /**
> 
> Yeah, that's apparently the result of my debugging sessions. Could you
> confirm this, and does it really happen only with this driver?

Yes.  I tested all of this on Skylake with the i965 driver, Skylake with the 
Media SDK iHD driver, and Polaris with the Mesa/Gallium driver.

> I also asked a kodi person to confirm. (Apparently they've been
> successfully using it with switched references.)

_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to