[FFmpeg-devel,2/2] lavf/vf_scale_amf: AMF scaler/format converter filter implementation

---
Sending updated patch based on Mark's review
1) added RGBA/BGRA
2) in case is device_ctx is set there is only the device hw format will be allowed as input and output
3) extended amf properties removed for now to have usual for ffmpeg scaler&format converter interface
4) input frame colorspace is set as color profile to select conversion matrix in case YUV->RGB, user setting removed.
5) misc bugs fixed


 configure                  |   1 +
 libavfilter/Makefile       |   1 +
 libavfilter/allfilters.c   |   1 +
 libavfilter/vf_scale_amf.c | 623 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 626 insertions(+)
 create mode 100644 libavfilter/vf_scale_amf.c

2018-07-10 16:30 GMT+02:00, Alexander Kravchenko <akravchenko188@gmail.com>:

> 1) added RGBA/BGRA

> +    static const enum AVPixelFormat input_pix_fmts[] = {
> +        AV_PIX_FMT_NV12,
> +        AV_PIX_FMT_0RGB,
> +        AV_PIX_FMT_BGR0,
> +        AV_PIX_FMT_RGB0,
> +        AV_PIX_FMT_GRAY8,
> +        AV_PIX_FMT_YUV420P,
> +        AV_PIX_FMT_YUYV422,
> +        AV_PIX_FMT_NONE,
> +    };

Do I miss it or did you forget to add RGBA?

Can you confirm that the alpha plane is really scaled?
(Unless I misunderstand, you haven't tested it.)

Carl Eugen

On Tue, Jul 10, 2018 at 05:30:15PM +0300, Alexander Kravchenko wrote:
> ---
> Sending updated patch based on Mark's review
> 1) added RGBA/BGRA
> 2) in case is device_ctx is set there is only the device hw format will be allowed as input and output
> 3) extended amf properties removed for now to have usual for ffmpeg scaler&format converter interface
> 4) input frame colorspace is set as color profile to select conversion matrix in case YUV->RGB, user setting removed.
> 5) misc bugs fixed
> 

not specific to this patch but more general, it would be nice if 
hw scaling/colorspa converting would be in our lib that is intended for 
scaling/colorspace converting (libswscale)

Thanks

[...]

> +    static const enum AVPixelFormat input_pix_fmts[] = {
    > +        AV_PIX_FMT_NV12,
    > +        AV_PIX_FMT_0RGB,
    > +        AV_PIX_FMT_BGR0,
    > +        AV_PIX_FMT_RGB0,
    > +        AV_PIX_FMT_GRAY8,
    > +        AV_PIX_FMT_YUV420P,
    > +        AV_PIX_FMT_YUYV422,
    > +        AV_PIX_FMT_NONE,
    > +    };
    
    Do I miss it or did you forget to add RGBA?
    
    Can you confirm that the alpha plane is really scaled?
    (Unless I misunderstand, you haven't tested it.)
   


Hi Carl. Thanks for your feedback.

Yes, sure, alpha is also scaled in AMF
It look like I did mistake preparing patch and missed RGBA formats here. 
I will update it

Thanks,
Alexander

Hi Mark, 
did you have a chance to review the patches?
I did send them 10 July, and no feedback from you
Could you please review it and send feedback to continue process?

Thanks,
Alexander

On 10/07/18 15:30, Alexander Kravchenko wrote:
> ---
>  configure                  |   1 +
>  libavfilter/Makefile       |   1 +
>  libavfilter/allfilters.c   |   1 +
>  libavfilter/vf_scale_amf.c | 623 +++++++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 626 insertions(+)
>  create mode 100644 libavfilter/vf_scale_amf.c
> 
> diff --git a/configure b/configure
> index b1a4dcfc42..321e9bdb70 100755
> --- a/configure
> +++ b/configure
> @@ -3385,6 +3385,7 @@ rubberband_filter_deps="librubberband"
>  sab_filter_deps="gpl swscale"
>  scale2ref_filter_deps="swscale"
>  scale_filter_deps="swscale"
> +scale_amf_filter_deps="amf"
>  scale_qsv_filter_deps="libmfx"
>  select_filter_select="pixelutils"
>  sharpness_vaapi_filter_deps="vaapi VAProcPipelineParameterBuffer"
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index 7735c26529..1b35c9dd5e 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -317,6 +317,7 @@ OBJS-$(CONFIG_ROBERTS_OPENCL_FILTER)         += vf_convolution_opencl.o opencl.o
>  OBJS-$(CONFIG_ROTATE_FILTER)                 += vf_rotate.o
>  OBJS-$(CONFIG_SAB_FILTER)                    += vf_sab.o
>  OBJS-$(CONFIG_SCALE_FILTER)                  += vf_scale.o scale.o
> +OBJS-$(CONFIG_SCALE_AMF_FILTER)              += vf_scale_amf.o scale.o
>  OBJS-$(CONFIG_SCALE_CUDA_FILTER)             += vf_scale_cuda.o vf_scale_cuda.ptx.o
>  OBJS-$(CONFIG_SCALE_NPP_FILTER)              += vf_scale_npp.o scale.o
>  OBJS-$(CONFIG_SCALE_QSV_FILTER)              += vf_scale_qsv.o
> diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
> index 0ded83ede2..7c7eb1526a 100644
> --- a/libavfilter/allfilters.c
> +++ b/libavfilter/allfilters.c
> @@ -303,6 +303,7 @@ extern AVFilter ff_vf_roberts_opencl;
>  extern AVFilter ff_vf_rotate;
>  extern AVFilter ff_vf_sab;
>  extern AVFilter ff_vf_scale;
> +extern AVFilter ff_vf_scale_amf;
>  extern AVFilter ff_vf_scale_cuda;
>  extern AVFilter ff_vf_scale_npp;
>  extern AVFilter ff_vf_scale_qsv;
> diff --git a/libavfilter/vf_scale_amf.c b/libavfilter/vf_scale_amf.c
> new file mode 100644
> index 0000000000..49250281e5
> --- /dev/null
> +++ b/libavfilter/vf_scale_amf.c
> @@ -0,0 +1,623 @@
> +/*
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * FFmpeg is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with FFmpeg; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> + */
> +
> +/**
> + * @file
> + * scale video filter - AMF
> + */
> +
> +#include <stdio.h>
> +#include <string.h>
> +
> +#include "libavutil/avassert.h"
> +#include "libavutil/imgutils.h"
> +#include "libavutil/opt.h"
> +#include "libavutil/pixdesc.h"
> +#include "libavutil/time.h"
> +
> +#include "libavutil/hwcontext.h"
> +#include "libavutil/hwcontext_amf.h"
> +
> +#include "AMF/components/VideoConverter.h"

Probably angle brackets.

> +
> +#include "avfilter.h"
> +#include "formats.h"
> +#include "internal.h"
> +#include "video.h"
> +#include "scale.h"
> +
> +#if CONFIG_DXVA2
> +#include <d3d9.h>
> +#endif
> +
> +#if CONFIG_D3D11VA
> +#include <d3d11.h>
> +#endif

Already included by hwcontext_amf.h.

> +
> +#define AMFAV_RETURN_IF_FALSE(avctx, exp, ret_value, /*message,*/ ...) \
> +    if (!(exp)) { \
> +        av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
> +        return ret_value; \
> +    }
> +
> +#define AMFAV_GOTO_FAIL_IF_FALSE(avctx, exp, ret_value, /*message,*/ ...) \
> +    if (!(exp)) { \
> +        av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
> +        ret = ret_value; \
> +        goto fail; \
> +    }
> +
> +typedef struct FormatMap {
> +    enum AVPixelFormat       av_format;
> +    enum AMF_SURFACE_FORMAT  amf_format;
> +} FormatMap;

There is no reason to name this structure when you're only defining one of them.  Just do "static const struct { } format_map[] = { };".

> +
> +static const FormatMap format_map[] =
> +{
> +    { AV_PIX_FMT_NV12,       AMF_SURFACE_NV12 },
> +
> +    { AV_PIX_FMT_BGR0,       AMF_SURFACE_BGRA },
> +    { AV_PIX_FMT_BGRA,       AMF_SURFACE_BGRA },
> +
> +    { AV_PIX_FMT_RGB0,       AMF_SURFACE_RGBA },
> +    { AV_PIX_FMT_RGBA,       AMF_SURFACE_RGBA },
> +
> +    { AV_PIX_FMT_0RGB,       AMF_SURFACE_ARGB },
> +    { AV_PIX_FMT_ARGB,       AMF_SURFACE_ARGB },
> +
> +    { AV_PIX_FMT_GRAY8,      AMF_SURFACE_GRAY8 },
> +    { AV_PIX_FMT_YUV420P,    AMF_SURFACE_YUV420P },
> +    { AV_PIX_FMT_YUYV422,    AMF_SURFACE_YUY2 },
> +};
> +
> +static enum AMF_SURFACE_FORMAT amf_av_to_amf_format(enum AVPixelFormat fmt)
> +{
> +    int i;
> +    for (i = 0; i < amf_countof(format_map); i++) {
> +        if (format_map[i].av_format == fmt) {
> +            return format_map[i].amf_format;
> +        }
> +    }
> +    return AMF_SURFACE_UNKNOWN;
> +}
> +
> +typedef struct AMFScaleContext {
> +    const AVClass *class;
> +
> +    int width, height;
> +    enum AVPixelFormat format;
> +    int scale_type;
> +
> +    char *w_expr;
> +    char *h_expr;
> +    char *format_str;
> +
> +    AMFComponent        *converter;
> +    AVBufferRef         *amf_device_ref;
> +
> +    AVBufferRef         *hwframes_in_ref;
> +    AVBufferRef         *hwframes_out_ref;
> +    AVBufferRef         *hwdevice_ref;
> +
> +    AMFContext          *context;
> +    AMFFactory          *factory;
> +
> +} AMFScaleContext;
> +
> +
> +static int amf_copy_surface(AVFilterContext *avctx, const AVFrame *frame,
> +    AMFSurface* surface)
> +{
> +    AMFPlane *plane;
> +    uint8_t  *dst_data[4];
> +    int       dst_linesize[4];
> +    int       planes;
> +    int       i;
> +
> +    planes = surface->pVtbl->GetPlanesCount(surface);
> +    av_assert0(planes < FF_ARRAY_ELEMS(dst_data));

Probably meant to be <=, though currently you don't have any YUVA to be caught out by.

> +
> +    for (i = 0; i < planes; i++) {
> +        plane = surface->pVtbl->GetPlaneAt(surface, i);
> +        dst_data[i] = plane->pVtbl->GetNative(plane);
> +        dst_linesize[i] = plane->pVtbl->GetHPitch(plane);

Presumably none of these functions can fail?  (Everything is already mapped, so you can't run out of memory or address space?)

> +    }
> +    av_image_copy(dst_data, dst_linesize,
> +        (const uint8_t**)frame->data, frame->linesize, frame->format,
> +        frame->width, frame->height);
> +
> +    return 0;
> +}
> +
> +static void amf_free_amfsurface(void *opaque, uint8_t *data)
> +{
> +    AMFSurface *surface = (AMFSurface*)(opaque);
> +    surface->pVtbl->Release(surface);
> +}

This function can be unused, mark it with av_unused to avoid the warning.

> +
> +static AVFrame *amf_amfsurface_to_avframe(AVFilterContext *avctx, AMFSurface* pSurface)
> +{
> +    AVFrame *frame = av_frame_alloc();
> +
> +    if (!frame)
> +        return NULL;
> +
> +    switch (pSurface->pVtbl->GetMemoryType(pSurface))
> +    {
> +#if CONFIG_D3D11VA
> +        case AMF_MEMORY_DX11:
> +        {
> +            AMFPlane *plane0 = pSurface->pVtbl->GetPlaneAt(pSurface, 0);

This feels dubious - it looks like it only works for single-plane surfaces, but it actually works anyway?  A comment explaining that might help.

> +            frame->data[0] = plane0->pVtbl->GetNative(plane0);
> +            frame->data[1] = (uint8_t*)(intptr_t)0;
> +
> +            frame->buf[0] = av_buffer_create(NULL,
> +                                     0,
> +                                     amf_free_amfsurface,
> +                                     pSurface,
> +                                     AV_BUFFER_FLAG_READONLY);

Unchecked allocation.

> +            pSurface->pVtbl->Acquire(pSurface);

Can Acquire ever fail?  It seems to have a return value.

> +        }
> +        break;
> +#endif
> +#if CONFIG_DXVA2
> +        case AMF_MEMORY_DX9:
> +        {
> +            AMFPlane *plane0 = pSurface->pVtbl->GetPlaneAt(pSurface, 0);
> +            frame->data[3] = plane0->pVtbl->GetNative(plane0);
> +
> +            frame->buf[0] = av_buffer_create(NULL,
> +                                     0,
> +                                     amf_free_amfsurface,
> +                                     pSurface,
> +                                     AV_BUFFER_FLAG_READONLY);
> +            pSurface->pVtbl->Acquire(pSurface);

Same comments as for D3D11.

> +        }
> +        break;
> +#endif
> +    default:
> +        {
> +            av_assert0(0);//should not happen
> +        }
> +    }
> +
> +    return frame;
> +}
> +
> +static int amf_avframe_to_amfsurface(AVFilterContext *avctx, const AVFrame *frame, AMFSurface** ppSurface)
> +{
> +    AMFScaleContext *ctx = avctx->priv;
> +    AMFSurface *surface;
> +    AMF_RESULT  res;
> +    int hw_surface = 0;
> +
> +    switch (frame->format) {
> +#if CONFIG_D3D11VA
> +    case AV_PIX_FMT_D3D11:
> +        {
> +            static const GUID AMFTextureArrayIndexGUID = { 0x28115527, 0xe7c3, 0x4b66, { 0x99, 0xd3, 0x4f, 0x2a, 0xe6, 0xb4, 0x7f, 0xaf } };
> +            ID3D11Texture2D *texture = (ID3D11Texture2D*)frame->data[0]; // actual texture
> +            int index = (intptr_t)frame->data[1]; // index is a slice in texture array is - set to tell AMF which slice to use
> +            texture->lpVtbl->SetPrivateData(texture, &AMFTextureArrayIndexGUID, sizeof(index), &index);
> +
> +            res = ctx->context->pVtbl->CreateSurfaceFromDX11Native(ctx->context, texture, &surface, NULL); // wrap to AMF surface
> +            AMFAV_RETURN_IF_FALSE(avctx, res == AMF_OK, AVERROR(ENOMEM), "CreateSurfaceFromDX11Native() failed  with error %d\n", res);
> +            hw_surface = 1;
> +        }
> +        break;
> +#endif
> +#if CONFIG_DXVA2
> +    case AV_PIX_FMT_DXVA2_VLD:
> +        {
> +            IDirect3DSurface9 *texture = (IDirect3DSurface9 *)frame->data[3]; // actual texture
> +
> +            res = ctx->context->pVtbl->CreateSurfaceFromDX9Native(ctx->context, texture, &surface, NULL); // wrap to AMF surface
> +            AMFAV_RETURN_IF_FALSE(avctx, res == AMF_OK, AVERROR(ENOMEM), "CreateSurfaceFromDX9Native() failed  with error %d\n", res);
> +            hw_surface = 1;
> +        }
> +        break;
> +#endif
> +    default:
> +        {
> +            AMF_SURFACE_FORMAT amf_fmt = amf_av_to_amf_format(frame->format);
> +            res = ctx->context->pVtbl->AllocSurface(ctx->context, AMF_MEMORY_HOST, amf_fmt, frame->width, frame->height, &surface);
> +            AMFAV_RETURN_IF_FALSE(avctx, res == AMF_OK, AVERROR(ENOMEM), "AllocSurface() failed  with error %d\n", res);
> +            amf_copy_surface(avctx, frame, surface);
> +        }
> +        break;
> +    }
> +
> +    if (hw_surface) {
> +        // input HW surfaces can be vertically aligned by 16; tell AMF the real size
> +        surface->pVtbl->SetCrop(surface, 0, 0, frame->width, frame->height);
> +    }
> +
> +    surface->pVtbl->SetPts(surface, frame->pts);
> +    *ppSurface = surface;
> +    return 0;
> +}
> +
> +static int amf_scale_init(AVFilterContext *avctx)
> +{
> +    AMFScaleContext     *ctx = avctx->priv;
> +
> +    if (!strcmp(ctx->format_str, "same")) {
> +        ctx->format = AV_PIX_FMT_NONE;
> +    } else {
> +        ctx->format = av_get_pix_fmt(ctx->format_str);
> +        if (ctx->format == AV_PIX_FMT_NONE) {
> +            av_log(avctx, AV_LOG_ERROR, "Unrecognized pixel format: %s\n", ctx->format_str);
> +            return AVERROR(EINVAL);
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +static void amf_scale_uninit(AVFilterContext *avctx)
> +{
> +    AMFScaleContext *ctx = avctx->priv;
> +
> +    if (ctx->converter) {
> +        ctx->converter->pVtbl->Terminate(ctx->converter);
> +        ctx->converter->pVtbl->Release(ctx->converter);
> +        ctx->converter = NULL;
> +    }
> +
> +    av_buffer_unref(&ctx->amf_device_ref);
> +    av_buffer_unref(&ctx->hwdevice_ref);
> +    av_buffer_unref(&ctx->hwframes_in_ref);
> +    av_buffer_unref(&ctx->hwframes_out_ref);
> +}
> +
> +static int amf_scale_query_formats(AVFilterContext *avctx)
> +{
> +    const enum AVPixelFormat *output_pix_fmts;
> +    AVFilterFormats *input_formats;
> +    int err;
> +    int i;
> +    static const enum AVPixelFormat input_pix_fmts[] = {
> +        AV_PIX_FMT_NV12,
> +        AV_PIX_FMT_0RGB,
> +        AV_PIX_FMT_BGR0,
> +        AV_PIX_FMT_RGB0,
> +        AV_PIX_FMT_GRAY8,
> +        AV_PIX_FMT_YUV420P,
> +        AV_PIX_FMT_YUYV422,
> +        AV_PIX_FMT_NONE,
> +    };

Seems like it might be better to make this from the format map above to avoid duplication?  (The list does have to be identical or you can hit ugly failures below.)

> +    static const enum AVPixelFormat output_pix_fmts_default[] = {
> +        AV_PIX_FMT_D3D11,
> +        AV_PIX_FMT_DXVA2_VLD,
> +        AV_PIX_FMT_NONE,
> +    };
> +    output_pix_fmts = output_pix_fmts_default;
> +
> +    //in case if hw_device_ctx is set to DXVA2 we change order of pixel formats to set DXVA2 be choosen by default
> +    //The order is ignored if hw_frames_ctx is not NULL on the config_output stage
> +    if (avctx->hw_device_ctx) {
> +        AVHWDeviceContext *device_ctx = (AVHWDeviceContext*)avctx->hw_device_ctx->data;
> +
> +        switch (device_ctx->type) {
> +    #if CONFIG_D3D11VA

#if at the start of the line.

> +        case AV_HWDEVICE_TYPE_D3D11VA:
> +            {
> +                static const enum AVPixelFormat output_pix_fmts_d3d11[] = {
> +                    AV_PIX_FMT_D3D11,
> +                    AV_PIX_FMT_NONE,
> +                };
> +                output_pix_fmts = output_pix_fmts_d3d11;
> +            }
> +            break;
> +    #endif
> +    #if CONFIG_DXVA2
> +        case AV_HWDEVICE_TYPE_DXVA2:
> +            {
> +                static const enum AVPixelFormat output_pix_fmts_dxva2[] = {
> +                    AV_PIX_FMT_DXVA2_VLD,
> +                    AV_PIX_FMT_NONE,
> +                };
> +                output_pix_fmts = output_pix_fmts_dxva2;
> +            }
> +            break;
> +    #endif
> +        default:
> +            {
> +                av_log(avctx, AV_LOG_ERROR, "Unsupported device : %s\n", av_hwdevice_get_type_name(device_ctx->type));

Just ignore it if you're given this rather than failing.  With the the graph-global device setup in the ffmpeg utility it's easy to end up with an unrelated device (OpenCL, say) attached to a filter which doesn't want it.

> +                return AVERROR(EINVAL);
> +            }
> +            break;
> +        }
> +    }
> +
> +    input_formats = ff_make_format_list(output_pix_fmts);
> +    if (!input_formats) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    for (i = 0; input_pix_fmts[i] != AV_PIX_FMT_NONE; i++) {
> +        err = ff_add_format(&input_formats, input_pix_fmts[i]);
> +        if (err < 0)
> +            return err;
> +    }
> +
> +    if ((err = ff_formats_ref(input_formats, &avctx->inputs[0]->out_formats)) < 0 ||
> +        (err = ff_formats_ref(ff_make_format_list(output_pix_fmts),
> +                              &avctx->outputs[0]->in_formats)) < 0)
> +        return err;
> +
> +    return 0;
> +}
> +
> +static int amf_scale_config_output(AVFilterLink *outlink)
> +{
> +    AVFilterContext *avctx = outlink->src;
> +    AVFilterLink   *inlink = avctx->inputs[0];
> +    AMFScaleContext  *ctx = avctx->priv;
> +    AVAMFDeviceContext *amf_ctx;
> +    AVHWFramesContext *hwframes_out;
> +    enum AVPixelFormat pix_fmt_in;
> +    AMFSize out_size;
> +    int err;
> +    AMF_RESULT res;
> +
> +    if ((err = ff_scale_eval_dimensions(avctx,
> +                                        ctx->w_expr, ctx->h_expr,
> +                                        inlink, outlink,
> +                                        &ctx->width, &ctx->height)) < 0)
> +        return err;
> +
> +    av_buffer_unref(&ctx->amf_device_ref);
> +    av_buffer_unref(&ctx->hwframes_in_ref);
> +    av_buffer_unref(&ctx->hwframes_out_ref);
> +
> +    if (inlink->hw_frames_ctx) {
> +        AVHWFramesContext *frames_ctx = (AVHWFramesContext*)inlink->hw_frames_ctx->data;
> +
> +        if (amf_av_to_amf_format(frames_ctx->sw_format) == AMF_SURFACE_UNKNOWN) {
> +            av_log(avctx, AV_LOG_ERROR, "Format of input frames context (%s) is not supported by AMF.\n",
> +                   av_get_pix_fmt_name(frames_ctx->sw_format));
> +            return AVERROR(EINVAL);
> +        }
> +
> +        err = av_hwdevice_ctx_create_derived(&ctx->amf_device_ref, AV_HWDEVICE_TYPE_AMF, frames_ctx->device_ref, 0);
> +        if (err < 0)
> +            return err;
> +
> +        ctx->hwframes_in_ref = av_buffer_ref(inlink->hw_frames_ctx);
> +        if (!ctx->hwframes_in_ref)
> +            return AVERROR(ENOMEM);
> +
> +        ctx->hwframes_out_ref = av_hwframe_ctx_alloc(frames_ctx->device_ref);
> +        if (!ctx->hwframes_out_ref)
> +            return AVERROR(ENOMEM);
> +
> +        hwframes_out = (AVHWFramesContext*)ctx->hwframes_out_ref->data;
> +        hwframes_out->format    = outlink->format;
> +        hwframes_out->sw_format = frames_ctx->sw_format;
> +        pix_fmt_in = frames_ctx->sw_format;
> +
> +    } else if (avctx->hw_device_ctx) {
> +        err = av_hwdevice_ctx_create_derived(&ctx->amf_device_ref, AV_HWDEVICE_TYPE_AMF, avctx->hw_device_ctx, 0);
> +        if (err < 0)
> +            return err;
> +
> +        ctx->hwdevice_ref = av_buffer_ref(avctx->hw_device_ctx);
> +        if (!ctx->hwdevice_ref)
> +            return AVERROR(ENOMEM);

I don't think this reference to the input device does anything?

> +
> +        ctx->hwframes_out_ref = av_hwframe_ctx_alloc(ctx->hwdevice_ref);
> +        if (!ctx->hwframes_out_ref)
> +            return AVERROR(ENOMEM);
> +
> +        hwframes_out = (AVHWFramesContext*)ctx->hwframes_out_ref->data;
> +        hwframes_out->format    = outlink->format;
> +        hwframes_out->sw_format = inlink->format;
> +        pix_fmt_in = inlink->format;
> +
> +    } else {
> +        av_log(ctx, AV_LOG_ERROR, "A hardware device reference is required to init hwcontext_amf.\n");
> +        return AVERROR(EINVAL);
> +    }
> +
> +    if (ctx->format != AV_PIX_FMT_NONE) {
> +        hwframes_out->sw_format = ctx->format;
> +    }
> +
> +    outlink->w = ctx->width;
> +    outlink->h = ctx->height;
> +
> +    hwframes_out->width = outlink->w;
> +    hwframes_out->height = outlink->h;
> +
> +    err = av_hwframe_ctx_init(ctx->hwframes_out_ref);
> +    if (err < 0)
> +        return err;

Put some message here to explain the failure.  A user can easily hit this - e.g. by attempting to convert YUV -> RGB on a DXVA2 device, which only supports YUV surfaces.

> +
> +    outlink->hw_frames_ctx = av_buffer_ref(ctx->hwframes_out_ref);
> +    if (!outlink->hw_frames_ctx) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    amf_ctx = ((AVHWDeviceContext*)ctx->amf_device_ref->data)->hwctx;
> +    ctx->context = amf_ctx->context;
> +    ctx->factory = amf_ctx->factory;
> +
> +    res = ctx->factory->pVtbl->CreateComponent(ctx->factory, ctx->context, AMFVideoConverter, &ctx->converter);
> +    AMFAV_RETURN_IF_FALSE(ctx, res == AMF_OK, AVERROR_FILTER_NOT_FOUND, "CreateComponent(%ls) failed with error %d\n", AMFVideoConverter, res);

Not FILTER_NOT_FOUND.  I guess this means the component isn't available or you've run out of memory?  (Maybe there should be some sort of AMF error -> AVERROR mapping to avoid all these UNKNOWNs.  Getting all ENOMEM cases right would be nice, at least.)

> +
> +    AMF_ASSIGN_PROPERTY_INT64(res, ctx->converter, AMF_VIDEO_CONVERTER_OUTPUT_FORMAT, (amf_int32)amf_av_to_amf_format(hwframes_out->sw_format));
> +    AMFAV_RETURN_IF_FALSE(avctx, res == AMF_OK, AVERROR_UNKNOWN, "AMFConverter-SetProperty() failed with error %d\n", res);
> +
> +    out_size.width = outlink->w;
> +    out_size.height = outlink->h;
> +    AMF_ASSIGN_PROPERTY_SIZE(res, ctx->converter, AMF_VIDEO_CONVERTER_OUTPUT_SIZE, out_size);
> +    AMFAV_RETURN_IF_FALSE(avctx, res == AMF_OK, AVERROR_UNKNOWN, "AMFConverter-SetProperty() failed with error %d\n", res);
> +
> +    AMF_ASSIGN_PROPERTY_INT64(res, ctx->converter, AMF_VIDEO_CONVERTER_SCALE, (amf_int32)ctx->scale_type);
> +    AMFAV_RETURN_IF_FALSE(avctx, res == AMF_OK, AVERROR_UNKNOWN, "AMFConverter-SetProperty() failed with error %d\n", res);
> +
> +
> +    res = ctx->converter->pVtbl->Init(ctx->converter, amf_av_to_amf_format(pix_fmt_in), inlink->w, inlink->h);
> +    AMFAV_RETURN_IF_FALSE(avctx, res == AMF_OK, AVERROR_UNKNOWN, "AMFConverter-Init() failed with error %d\n", res);
> +
> +    return 0;

A simple log message (at VERBOSE?) to say what this is now configured to do would be helpful - e.g. "Configured for rgba (sw) -> nv12 (dxva2)".  The automatic conversion filters can make it nonobvious what is actually happening, and a surprise software conversion somewhere can easily kill performance.

> +}
> +
> +static int amf_scale_filter_frame(AVFilterLink *inlink, AVFrame *in)
> +{
> +    AVFilterContext             *avctx = inlink->dst;
> +    AMFScaleContext             *ctx = avctx->priv;
> +    AVFilterLink                *outlink = avctx->outputs[0];
> +    AMF_RESULT  res;
> +    AMFSurface *surface_in;
> +    AMFSurface *surface_out;
> +    AMFData *data_out;
> +    enum AMF_VIDEO_CONVERTER_COLOR_PROFILE_ENUM amf_color_profile = AMF_VIDEO_CONVERTER_COLOR_PROFILE_UNKNOWN;
> +
> +    AVFrame *out = NULL;
> +    int ret = 0;
> +
> +    if (!ctx->converter)
> +        return AVERROR(EINVAL);

This can't happen, can it?  (Feel free to make it an assert in that case.)

> +
> +    ret = amf_avframe_to_amfsurface(avctx, in, &surface_in);
> +    if (ret < 0)
> +        goto fail;
> +
> +    res = ctx->converter->pVtbl->SubmitInput(ctx->converter, (AMFData*)surface_in);
> +    AMFAV_GOTO_FAIL_IF_FALSE(avctx, res == AMF_OK, AVERROR_UNKNOWN, "SubmitInput() failed with error %d\n", res);
> +
> +    res = ctx->converter->pVtbl->QueryOutput(ctx->converter, &data_out);
> +    AMFAV_GOTO_FAIL_IF_FALSE(avctx, res == AMF_OK, AVERROR_UNKNOWN, "QueryOutput() failed with error %d\n", res);
> +
> +    if (data_out) {
> +        AMFGuid guid = IID_AMFSurface();
> +        data_out->pVtbl->QueryInterface(data_out, &guid, (void**)&surface_out); // query for buffer interface
> +        data_out->pVtbl->Release(data_out);
> +    }

What is this if branch actually trying to do?  If it isn't taken then you will use surface_out uninitialised on the line following - should it be failing if there isn't data, or maybe calling QueryOutput again?

> +
> +    out = amf_amfsurface_to_avframe(avctx, surface_out);

Can fail.

> +
> +    ret = av_frame_copy_props(out, in);
> +    if (ret < 0)
> +        goto fail;
> +
> +    switch(in->colorspace) {
> +    case AVCOL_SPC_BT470BG:
> +    case AVCOL_SPC_SMPTE170M:
> +    case AVCOL_SPC_SMPTE240M:
> +        amf_color_profile = AMF_VIDEO_CONVERTER_COLOR_PROFILE_601;
> +        break;
> +    case AVCOL_SPC_BT709:
> +        amf_color_profile = AMF_VIDEO_CONVERTER_COLOR_PROFILE_709;
> +        break;
> +    case AVCOL_SPC_BT2020_NCL:
> +    case AVCOL_SPC_BT2020_CL:
> +        amf_color_profile = AMF_VIDEO_CONVERTER_COLOR_PROFILE_2020;
> +        break;
> +    case AVCOL_SPC_RGB:
> +        amf_color_profile = AMF_VIDEO_CONVERTER_COLOR_PROFILE_JPEG;
> +        break;
> +    default:
> +        amf_color_profile = AMF_VIDEO_CONVERTER_COLOR_PROFILE_UNKNOWN;
> +        break;
> +    }

Just looking at colorspace (aka matrix coefficients) doesn't feel sufficient for this.  Especially since you have JPEG as a type here, range feels like it needs to be taken into account.

What do these AMF_VIDEO_CONVERTER_COLOR_PROFILEs actually mean in terms of the usual colour properties (primaries, transfer function, matrix coefficients, range)?

> +
> +    if (amf_color_profile != AMF_VIDEO_CONVERTER_COLOR_PROFILE_UNKNOWN) {
> +        AMF_ASSIGN_PROPERTY_INT64(res, ctx->converter, AMF_VIDEO_CONVERTER_COLOR_PROFILE, amf_color_profile);
> +    }
> +
> +    out->format = outlink->format;
> +    out->width  = outlink->w;
> +    out->height = outlink->h;
> +
> +    out->hw_frames_ctx = av_buffer_ref(ctx->hwframes_out_ref);
> +    if (!out->hw_frames_ctx) {
> +        ret = AVERROR(ENOMEM);
> +        goto fail;
> +    }
> +
> +    surface_in->pVtbl->Release(surface_in);
> +    surface_out->pVtbl->Release(surface_out);
> +
> +    if (inlink->sample_aspect_ratio.num) {
> +        outlink->sample_aspect_ratio = av_mul_q((AVRational){outlink->h * inlink->w, outlink->w * inlink->h}, inlink->sample_aspect_ratio);
> +    } else
> +        outlink->sample_aspect_ratio = inlink->sample_aspect_ratio;
> +
> +    av_frame_free(&in);
> +    return ff_filter_frame(outlink, out);
> +fail:
> +    av_frame_free(&in);
> +    av_frame_free(&out);
> +    return ret;
> +}
> +
> +#define OFFSET(x) offsetof(AMFScaleContext, x)
> +#define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
> +static const AVOption scale_amf_options[] = {
> +    { "w",      "Output video width",               OFFSET(w_expr),     AV_OPT_TYPE_STRING, { .str = "iw"   }, .flags = FLAGS },
> +    { "h",      "Output video height",              OFFSET(h_expr),     AV_OPT_TYPE_STRING, { .str = "ih"   }, .flags = FLAGS },
> +    { "format", "Output pixel format",              OFFSET(format_str), AV_OPT_TYPE_STRING, { .str = "same" }, .flags = FLAGS },
> +
> +    { "scale_type",    "Scale type",                OFFSET(scale_type),      AV_OPT_TYPE_INT,   { .i64 = AMF_VIDEO_CONVERTER_SCALE_BILINEAR },
> +        AMF_VIDEO_CONVERTER_SCALE_BILINEAR, AMF_VIDEO_CONVERTER_SCALE_BICUBIC, FLAGS, "scale_type" },

Do the scale types have significantly different performance?  If they are similar I think I would weakly suggest defaulting to the highest quality one.

> +    { "bilinear",      "Bilinear",      0,                       AV_OPT_TYPE_CONST, { .i64 = AMF_VIDEO_CONVERTER_SCALE_BILINEAR }, 0, 0, FLAGS, "scale_type" },
> +    { "bicubic",       "Bicubic",       0,                       AV_OPT_TYPE_CONST, { .i64 = AMF_VIDEO_CONVERTER_SCALE_BICUBIC },  0, 0, FLAGS, "scale_type" },
> +
> +    { NULL },
> +};
> +
> +
> +AVFILTER_DEFINE_CLASS(scale_amf);
> +
> +static const AVFilterPad amf_scale_inputs[] = {
> +    {
> +        .name         = "default",
> +        .type         = AVMEDIA_TYPE_VIDEO,
> +        .filter_frame = amf_scale_filter_frame,
> +    },
> +    { NULL }
> +};
> +
> +static const AVFilterPad amf_scale_outputs[] = {
> +    {
> +        .name         = "default",
> +        .type         = AVMEDIA_TYPE_VIDEO,
> +        .config_props = amf_scale_config_output,
> +    },
> +    { NULL }
> +};
> +
> +AVFilter ff_vf_scale_amf = {
> +    .name      = "scale_amf",
> +    .description = NULL_IF_CONFIG_SMALL("AMF video scaling and format conversion"),
> +
> +    .init          = amf_scale_init,
> +    .uninit        = amf_scale_uninit,
> +    .query_formats = amf_scale_query_formats,
> +
> +    .priv_size = sizeof(AMFScaleContext),
> +    .priv_class = &scale_amf_class,
> +
> +    .inputs    = amf_scale_inputs,
> +    .outputs   = amf_scale_outputs,
> +
> +    .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
> +};
> 

- Mark