diff mbox series

[FFmpeg-devel,06/16] avutil: add nvtegra hwcontext

Message ID bb17171cf16ebcb4d15d9b998bc955754b133bdf.1717083800.git.averne381@gmail.com
State New
Headers show
Series NVidia Tegra hardware decoding backend | expand

Checks

Context Check Description
yinshiyou/commit_msg_loongarch64 warning Please wrap lines in the body of the commit message between 60 and 72 characters.
andriy/commit_msg_x86 warning Please wrap lines in the body of the commit message between 60 and 72 characters.
andriy/make_fate_x86 success Make fate finished
andriy/make_x86 warning New warnings during build

Commit Message

averne May 30, 2024, 7:43 p.m. UTC
This includes hwdevice and hwframes objects.
As the multimedia engines work with tiled surfaces (block linear in nvidia jargon), two frame transfer methods are implemented.
The first makes use of the VIC to perform the copy. Since some revisions of the VIC (such as the one found in the tegra X1) did not support 10+ bit formats, these go through two separate copy steps for the luma and chroma planes.
The second method copies on the CPU, and is used as a fallback if the VIC constraints are not satisfied.

Signed-off-by: averne <averne381@gmail.com>
---
 libavutil/Makefile             |   7 +-
 libavutil/hwcontext.c          |   4 +
 libavutil/hwcontext.h          |   1 +
 libavutil/hwcontext_internal.h |   1 +
 libavutil/hwcontext_nvtegra.c  | 880 +++++++++++++++++++++++++++++++++
 libavutil/hwcontext_nvtegra.h  |  85 ++++
 6 files changed, 976 insertions(+), 2 deletions(-)
 create mode 100644 libavutil/hwcontext_nvtegra.c
 create mode 100644 libavutil/hwcontext_nvtegra.h

Comments

Mark Thompson June 5, 2024, 8:47 p.m. UTC | #1
On 30/05/2024 20:43, averne wrote:
> This includes hwdevice and hwframes objects.
> As the multimedia engines work with tiled surfaces (block linear in nvidia jargon), two frame transfer methods are implemented.
> The first makes use of the VIC to perform the copy. Since some revisions of the VIC (such as the one found in the tegra X1) did not support 10+ bit formats, these go through two separate copy steps for the luma and chroma planes.
> The second method copies on the CPU, and is used as a fallback if the VIC constraints are not satisfied.
> 
> Signed-off-by: averne <averne381@gmail.com>
> ---
>  libavutil/Makefile             |   7 +-
>  libavutil/hwcontext.c          |   4 +
>  libavutil/hwcontext.h          |   1 +
>  libavutil/hwcontext_internal.h |   1 +
>  libavutil/hwcontext_nvtegra.c  | 880 +++++++++++++++++++++++++++++++++
>  libavutil/hwcontext_nvtegra.h  |  85 ++++
>  6 files changed, 976 insertions(+), 2 deletions(-)
>  create mode 100644 libavutil/hwcontext_nvtegra.c
>  create mode 100644 libavutil/hwcontext_nvtegra.h
> 
> ...> +
> +static int nvtegra_transfer_data(AVHWFramesContext *ctx, AVFrame *dst, const AVFrame *src) {
> +    const AVFrame *swframe;
> +    bool from;
> +    int num_planes, i;
> +
> +    from    = !dst->hw_frames_ctx;
> +    swframe = from ? dst : src;
> +
> +    if (swframe->hw_frames_ctx)
> +        return AVERROR(ENOSYS);
> +
> +    num_planes = av_pix_fmt_count_planes(swframe->format);
> +
> +    for (i = 0; i < num_planes; ++i) {
> +        if (((uintptr_t)swframe->data[i] & 0xff) || (swframe->linesize[i] & 0xff)) {
> +            av_log(ctx, AV_LOG_WARNING, "Frame address/pitch not aligned to 256, "
> +                                        "falling back to cpu transfer\n");
> +            return nvtegra_cpu_transfer_data(ctx, dst, src, num_planes, from);

Are you doing something somewhere to avoid this case?  It seems like it should be the normal one (given alignment is typically set signficantly lower than 256), so this warning would be very spammy.

> +        }
> +    }
> +
> +    return nvtegra_vic_transfer_data(ctx, dst, src, num_planes, from);
> +}
> +
> +const HWContextType ff_hwcontext_type_nvtegra = {
> +    .type                   = AV_HWDEVICE_TYPE_NVTEGRA,
> +    .name                   = "nvtegra",
> +
> +    .device_hwctx_size      = sizeof(NVTegraDevicePriv),
> +    .device_hwconfig_size   = 0,
> +    .frames_hwctx_size      = 0,
> +
> +    .device_create          = &nvtegra_device_create,
> +    .device_init            = &nvtegra_device_init,
> +    .device_uninit          = &nvtegra_device_uninit,
> +
> +    .frames_get_constraints = &nvtegra_frames_get_constraints,
> +    .frames_init            = &nvtegra_frames_init,
> +    .frames_uninit          = &nvtegra_frames_uninit,
> +    .frames_get_buffer      = &nvtegra_get_buffer,
> +
> +    .transfer_get_formats   = &nvtegra_transfer_get_formats,
> +    .transfer_data_to       = &nvtegra_transfer_data,
> +    .transfer_data_from     = &nvtegra_transfer_data,
> +
> +    .pix_fmts = (const enum AVPixelFormat[]) {
> +        AV_PIX_FMT_NVTEGRA,
> +        AV_PIX_FMT_NONE,
> +    },
> +};

What controls whether frames are linear or not?

It seems like the linear case could be exposed directly to the user rather than having to wrap it like this - the decoder could return read-only NV12 (or whatever) frames directly and they would work with other components.

> diff --git a/libavutil/hwcontext_nvtegra.h b/libavutil/hwcontext_nvtegra.h
> new file mode 100644
> index 0000000000..8a2383d304
> --- /dev/null
> +++ b/libavutil/hwcontext_nvtegra.h
> @@ -0,0 +1,85 @@
> +/*
> + * Copyright (c) 2024 averne <averne381@gmail.com>
> + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 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 General Public License for more details.
> + *
> + * You should have received a copy of the GNU 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.
> + */
> +
> +#ifndef AVUTIL_HWCONTEXT_NVTEGRA_H
> +#define AVUTIL_HWCONTEXT_NVTEGRA_H
> +
> +#include <stdint.h>
> +
> +#include "hwcontext.h"
> +#include "buffer.h"
> +#include "frame.h"
> +#include "pixfmt.h"
> +
> +#include "nvtegra.h"
> +
> +/*
> + * Encode a hardware revision into a version number
> + */
> +#define AV_NVTEGRA_ENCODE_REV(maj, min) (((maj & 0xff) << 8) | (min & 0xff))
> +
> +/*
> + * Decode a version number
> + */
> +static inline void av_nvtegra_decode_rev(int rev, int *maj, int *min) {
> +    *maj = (rev >> 8) & 0xff;
> +    *min = (rev >> 0) & 0xff;
> +}
> +
> +/**
> + * @file
> + * API-specific header for AV_HWDEVICE_TYPE_NVTEGRA.
> + *
> + * For user-allocated pools, AVHWFramesContext.pool must return AVBufferRefs
> + * with the data pointer set to an AVNVTegraMap.
> + */
> +
> +typedef struct AVNVTegraDeviceContext {
> +    /*
> +     * Hardware multimedia engines
> +     */
> +    AVNVTegraChannel nvdec_channel, nvenc_channel, nvjpg_channel, vic_channel;

Does a user need to supply all of these when making a device?

> +
> +    /*
> +     * Hardware revisions for associated engines, or 0 if invalid
> +     */
> +    int nvdec_version, nvenc_version, nvjpg_version, vic_version;

Why does a user setting up a device context need to supply the version numbers for each thing?

> +} AVNVTegraDeviceContext;
> +
> +typedef struct AVNVTegraFrame {
> +    /*
> +     * Reference to an AVNVTegraMap object
> +     */
> +    AVBufferRef *map_ref;
> +} AVNVTegraFrame;

What is the indirection doing here?  Can't it return the buffer inside this structure instead of making an intermediate structure?

> +
> +/*
> + * Helper to retrieve a map object from the corresponding frame
> + */
> +static inline AVNVTegraMap *av_nvtegra_frame_get_fbuf_map(const AVFrame *frame) {
> +    return (AVNVTegraMap *)((AVNVTegraFrame *)frame->buf[0]->data)->map_ref->data;
> +}
> +
> +/*
> + * Converts a pixel format to the equivalent code for the VIC engine
> + */
> +int av_nvtegra_pixfmt_to_vic(enum AVPixelFormat fmt);
> +
> +#endif /* AVUTIL_HWCONTEXT_NVTEGRA_H */

The mix of normal implementation code and weird specific detail for the particular platform is pretty nasty.  It does seem like exporting more of this into a separate library rather than embedding it in ffmpeg would be a good idea.

Thanks,

- Mark
averne June 29, 2024, 7:35 p.m. UTC | #2
Le 05/06/2024 à 22:47, Mark Thompson a écrit :> 
> On 30/05/2024 20:43, averne wrote:
>> This includes hwdevice and hwframes objects.
>> As the multimedia engines work with tiled surfaces (block linear in nvidia jargon), two frame transfer methods are implemented.
>> The first makes use of the VIC to perform the copy. Since some revisions of the VIC (such as the one found in the tegra X1) did not support 10+ bit formats, these go through two separate copy steps for the luma and chroma planes.
>> The second method copies on the CPU, and is used as a fallback if the VIC constraints are not satisfied.
>>
>> Signed-off-by: averne <averne381@gmail.com>
>> ---
>>  libavutil/Makefile             |   7 +-
>>  libavutil/hwcontext.c          |   4 +
>>  libavutil/hwcontext.h          |   1 +
>>  libavutil/hwcontext_internal.h |   1 +
>>  libavutil/hwcontext_nvtegra.c  | 880 +++++++++++++++++++++++++++++++++
>>  libavutil/hwcontext_nvtegra.h  |  85 ++++
>>  6 files changed, 976 insertions(+), 2 deletions(-)
>>  create mode 100644 libavutil/hwcontext_nvtegra.c
>>  create mode 100644 libavutil/hwcontext_nvtegra.h
>>
>> ...> +
>> +static int nvtegra_transfer_data(AVHWFramesContext *ctx, AVFrame *dst, const AVFrame *src) {
>> +    const AVFrame *swframe;
>> +    bool from;
>> +    int num_planes, i;
>> +
>> +    from    = !dst->hw_frames_ctx;
>> +    swframe = from ? dst : src;
>> +
>> +    if (swframe->hw_frames_ctx)
>> +        return AVERROR(ENOSYS);
>> +
>> +    num_planes = av_pix_fmt_count_planes(swframe->format);
>> +
>> +    for (i = 0; i < num_planes; ++i) {
>> +        if (((uintptr_t)swframe->data[i] & 0xff) || (swframe->linesize[i] & 0xff)) {
>> +            av_log(ctx, AV_LOG_WARNING, "Frame address/pitch not aligned to 256, "
>> +                                        "falling back to cpu transfer\n");
>> +            return nvtegra_cpu_transfer_data(ctx, dst, src, num_planes, from);
>
> Are you doing something somewhere to avoid this case?  It seems like it should be the normal one (given alignment is typically set signficantly lower than 256), so this warning would be very spammy.

Nothing is done on the FFmpeg side. I could print this warning 
just once per frames context.

>> +        }
>> +    }
>> +
>> +    return nvtegra_vic_transfer_data(ctx, dst, src, num_planes, from);
>> +}
>> +
>> +const HWContextType ff_hwcontext_type_nvtegra = {
>> +    .type                   = AV_HWDEVICE_TYPE_NVTEGRA,
>> +    .name                   = "nvtegra",
>> +
>> +    .device_hwctx_size      = sizeof(NVTegraDevicePriv),
>> +    .device_hwconfig_size   = 0,
>> +    .frames_hwctx_size      = 0,
>> +
>> +    .device_create          = &nvtegra_device_create,
>> +    .device_init            = &nvtegra_device_init,
>> +    .device_uninit          = &nvtegra_device_uninit,
>> +
>> +    .frames_get_constraints = &nvtegra_frames_get_constraints,
>> +    .frames_init            = &nvtegra_frames_init,
>> +    .frames_uninit          = &nvtegra_frames_uninit,
>> +    .frames_get_buffer      = &nvtegra_get_buffer,
>> +
>> +    .transfer_get_formats   = &nvtegra_transfer_get_formats,
>> +    .transfer_data_to       = &nvtegra_transfer_data,
>> +    .transfer_data_from     = &nvtegra_transfer_data,
>> +
>> +    .pix_fmts = (const enum AVPixelFormat[]) {
>> +        AV_PIX_FMT_NVTEGRA,
>> +        AV_PIX_FMT_NONE,
>> +    },
>> +};
>
> What controls whether frames are linear or not?
>
> It seems like the linear case could be exposed directly to the user rather than having to wrap it like this - the decoder could return read-only NV12 (or whatever) frames directly and they would work with other components.

NVDEC can only output in block linear format (tiled layout). I'm not 
very keen on exposing that data, considering cache management can be 
somewhat tricky. Typically, after a decode operations, the CPU cache
must be invalidated. 

>> diff --git a/libavutil/hwcontext_nvtegra.h b/libavutil/hwcontext_nvtegra.h
>> new file mode 100644
>> index 0000000000..8a2383d304
>> --- /dev/null
>> +++ b/libavutil/hwcontext_nvtegra.h
>> @@ -0,0 +1,85 @@
>> +/*
>> + * Copyright (c) 2024 averne <averne381@gmail.com>
>> + *
>> + * This file is part of FFmpeg.
>> + *
>> + * FFmpeg is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License as published by
>> + * the Free Software Foundation; either version 2 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 General Public License for more details.
>> + *
>> + * You should have received a copy of the GNU 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.
>> + */
>> +
>> +#ifndef AVUTIL_HWCONTEXT_NVTEGRA_H
>> +#define AVUTIL_HWCONTEXT_NVTEGRA_H
>> +
>> +#include <stdint.h>
>> +
>> +#include "hwcontext.h"
>> +#include "buffer.h"
>> +#include "frame.h"

Le 05/06/2024 à 22:47, Mark Thompson a écrit :
> On 30/05/2024 20:43, averne wrote:
>> This includes hwdevice and hwframes objects.
>> As the multimedia engines work with tiled surfaces (block linear in nvidia jargon), two frame transfer methods are implemented.
>> The first makes use of the VIC to perform the copy. Since some revisions of the VIC (such as the one found in the tegra X1) did not support 10+ bit formats, these go through two separate copy steps for the luma and chroma planes.
>> The second method copies on the CPU, and is used as a fallback if the VIC constraints are not satisfied.
>>
>> Signed-off-by: averne <averne381@gmail.com>
>> ---
>>  libavutil/Makefile             |   7 +-
>>  libavutil/hwcontext.c          |   4 +
>>  libavutil/hwcontext.h          |   1 +
>>  libavutil/hwcontext_internal.h |   1 +
>>  libavutil/hwcontext_nvtegra.c  | 880 +++++++++++++++++++++++++++++++++
>>  libavutil/hwcontext_nvtegra.h  |  85 ++++
>>  6 files changed, 976 insertions(+), 2 deletions(-)
>>  create mode 100644 libavutil/hwcontext_nvtegra.c
>>  create mode 100644 libavutil/hwcontext_nvtegra.h
>>
>> ...> +
>> +static int nvtegra_transfer_data(AVHWFramesContext *ctx, AVFrame *dst, const AVFrame *src) {
>> +    const AVFrame *swframe;
>> +    bool from;
>> +    int num_planes, i;
>> +
>> +    from    = !dst->hw_frames_ctx;
>> +    swframe = from ? dst : src;
>> +
>> +    if (swframe->hw_frames_ctx)
>> +        return AVERROR(ENOSYS);
>> +
>> +    num_planes = av_pix_fmt_count_planes(swframe->format);
>> +
>> +    for (i = 0; i < num_planes; ++i) {
>> +        if (((uintptr_t)swframe->data[i] & 0xff) || (swframe->linesize[i] & 0xff)) {
>> +            av_log(ctx, AV_LOG_WARNING, "Frame address/pitch not aligned to 256, "
>> +                                        "falling back to cpu transfer\n");
>> +            return nvtegra_cpu_transfer_data(ctx, dst, src, num_planes, from);
> 
> Are you doing something somewhere to avoid this case?  It seems like it should be the normal one (given alignment is typically set signficantly lower than 256), so this warning would be very spammy.
> 
>> +        }
>> +    }
>> +
>> +    return nvtegra_vic_transfer_data(ctx, dst, src, num_planes, from);
>> +}
>> +
>> +const HWContextType ff_hwcontext_type_nvtegra = {
>> +    .type                   = AV_HWDEVICE_TYPE_NVTEGRA,
>> +    .name                   = "nvtegra",
>> +
>> +    .device_hwctx_size      = sizeof(NVTegraDevicePriv),
>> +    .device_hwconfig_size   = 0,
>> +    .frames_hwctx_size      = 0,
>> +
>> +    .device_create          = &nvtegra_device_create,
>> +    .device_init            = &nvtegra_device_init,
>> +    .device_uninit          = &nvtegra_device_uninit,
>> +
>> +    .frames_get_constraints = &nvtegra_frames_get_constraints,
>> +    .frames_init            = &nvtegra_frames_init,
>> +    .frames_uninit          = &nvtegra_frames_uninit,
>> +    .frames_get_buffer      = &nvtegra_get_buffer,
>> +
>> +    .transfer_get_formats   = &nvtegra_transfer_get_formats,
>> +    .transfer_data_to       = &nvtegra_transfer_data,
>> +    .transfer_data_from     = &nvtegra_transfer_data,
>> +
>> +    .pix_fmts = (const enum AVPixelFormat[]) {
>> +        AV_PIX_FMT_NVTEGRA,
>> +        AV_PIX_FMT_NONE,
>> +    },
>> +};
> 
> What controls whether frames are linear or not?
> 
> It seems like the linear case could be exposed directly to the user rather than having to wrap it like this - the decoder could return read-only NV12 (or whatever) frames directly and they would work with other components.
> 
>> diff --git a/libavutil/hwcontext_nvtegra.h b/libavutil/hwcontext_nvtegra.h
>> new file mode 100644
>> index 0000000000..8a2383d304
>> --- /dev/null
>> +++ b/libavutil/hwcontext_nvtegra.h
>> @@ -0,0 +1,85 @@
>> +/*
>> + * Copyright (c) 2024 averne <averne381@gmail.com>
>> + *
>> + * This file is part of FFmpeg.
>> + *
>> + * FFmpeg is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License as published by
>> + * the Free Software Foundation; either version 2 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 General Public License for more details.
>> + *
>> + * You should have received a copy of the GNU 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.
>> + */
>> +
>> +#ifndef AVUTIL_HWCONTEXT_NVTEGRA_H
>> +#define AVUTIL_HWCONTEXT_NVTEGRA_H
>> +
>> +#include <stdint.h>
>> +
>> +#include "hwcontext.h"
>> +#include "buffer.h"
>> +#include "frame.h"
>> +#include "pixfmt.h"
>> +
>> +#include "nvtegra.h"
>> +
>> +/*
>> + * Encode a hardware revision into a version number
>> + */
>> +#define AV_NVTEGRA_ENCODE_REV(maj, min) (((maj & 0xff) << 8) | (min & 0xff))
>> +
>> +/*
>> + * Decode a version number
>> + */
>> +static inline void av_nvtegra_decode_rev(int rev, int *maj, int *min) {
>> +    *maj = (rev >> 8) & 0xff;
>> +    *min = (rev >> 0) & 0xff;
>> +}
>> +
>> +/**
>> + * @file
>> + * API-specific header for AV_HWDEVICE_TYPE_NVTEGRA.
>> + *
>> + * For user-allocated pools, AVHWFramesContext.pool must return AVBufferRefs
>> + * with the data pointer set to an AVNVTegraMap.
>> + */
>> +
>> +typedef struct AVNVTegraDeviceContext {
>> +    /*
>> +     * Hardware multimedia engines
>> +     */
>> +    AVNVTegraChannel nvdec_channel, nvenc_channel, nvjpg_channel, vic_channel;
> 
> Does a user need to supply all of these when making a device?
> 
>> +
>> +    /*
>> +     * Hardware revisions for associated engines, or 0 if invalid
>> +     */
>> +    int nvdec_version, nvenc_version, nvjpg_version, vic_version;
> 
> Why does a user setting up a device context need to supply the version numbers for each thing?
> 
>> +} AVNVTegraDeviceContext;
>> +
>> +typedef struct AVNVTegraFrame {
>> +    /*
>> +     * Reference to an AVNVTegraMap object
>> +     */
>> +    AVBufferRef *map_ref;
>> +} AVNVTegraFrame;
> 
> What is the indirection doing here?  Can't it return the buffer inside this structure instead of making an intermediate structure?
> 
>> +
>> +/*
>> + * Helper to retrieve a map object from the corresponding frame
>> + */
>> +static inline AVNVTegraMap *av_nvtegra_frame_get_fbuf_map(const AVFrame *frame) {
>> +    return (AVNVTegraMap *)((AVNVTegraFrame *)frame->buf[0]->data)->map_ref->data;
>> +}
>> +
>> +/*
>> + * Converts a pixel format to the equivalent code for the VIC engine
>> + */
>> +int av_nvtegra_pixfmt_to_vic(enum AVPixelFormat fmt);
>> +
>> +#endif /* AVUTIL_HWCONTEXT_NVTEGRA_H */
> 
> The mix of normal implementation code and weird specific detail for the particular platform is pretty nasty.  It does seem like exporting more of this into a separate library rather than embedding it in ffmpeg would be a good idea.
> 
> Thanks,
> 
> - Mark
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
>> +#include "pixfmt.h"
>> +
>> +#include "nvtegra.h"
>> +
>> +/*
>> + * Encode a hardware revision into a version number
>> + */
>> +#define AV_NVTEGRA_ENCODE_REV(maj, min) (((maj & 0xff) << 8) | (min & 0xff))
>> +
>> +/*
>> + * Decode a version number
>> + */
>> +static inline void av_nvtegra_decode_rev(int rev, int *maj, int *min) {
>> +    *maj = (rev >> 8) & 0xff;
>> +    *min = (rev >> 0) & 0xff;
>> +}
>> +
>> +/**
>> + * @file
>> + * API-specific header for AV_HWDEVICE_TYPE_NVTEGRA.
>> + *
>> + * For user-allocated pools, AVHWFramesContext.pool must return AVBufferRefs
>> + * with the data pointer set to an AVNVTegraMap.
>> + */
>> +
>> +typedef struct AVNVTegraDeviceContext {
>> +    /*
>> +     * Hardware multimedia engines
>> +     */
>> +    AVNVTegraChannel nvdec_channel, nvenc_channel, nvjpg_channel, vic_channel;
>
> Does a user need to supply all of these when making a device?

These are filled out by the library when creating the hardware device. 

>> +
>> +    /*
>> +     * Hardware revisions for associated engines, or 0 if invalid
>> +     */
>> +    int nvdec_version, nvenc_version, nvjpg_version, vic_version;
>
> Why does a user setting up a device context need to supply the version numbers for each thing?

Same as above, the version is filled when creating the context.
Knowledge of the hardware revision is useful when determining hardware
capabilities.

>> +} AVNVTegraDeviceContext;
>> +
>> +typedef struct AVNVTegraFrame {
>> +    /*
>> +     * Reference to an AVNVTegraMap object
>> +     */
>> +    AVBufferRef *map_ref;
>> +} AVNVTegraFrame;
>
> What is the indirection doing here?  Can't it return the buffer inside this structure instead of making an intermediate structure?

That's an artifact of an optimization regarding my mpv graphics 
backend.
Since mapping/unmapping data in the GPU is sort of expensive (gmmu 
configuration), the mpv mapper would keep a reference to the frame 
object to avoid constantly doing mapping operations on the same 
memory ranges.
I'll remove that.

>> +
>> +/*
>> + * Helper to retrieve a map object from the corresponding frame
>> + */
>> +static inline AVNVTegraMap *av_nvtegra_frame_get_fbuf_map(const AVFrame *frame) {
>> +    return (AVNVTegraMap *)((AVNVTegraFrame *)frame->buf[0]->data)->map_ref->data;
>> +}
>> +
>> +/*
>> + * Converts a pixel format to the equivalent code for the VIC engine
>> + */
>> +int av_nvtegra_pixfmt_to_vic(enum AVPixelFormat fmt);
>> +
>> +#endif /* AVUTIL_HWCONTEXT_NVTEGRA_H */
>
> The mix of normal implementation code and weird specific detail for the particular platform is pretty nasty.  It does seem like exporting more of this into a separate library rather than embedding it in ffmpeg would be a good idea.
>
> Thanks,
>
> - Mark
diff mbox series

Patch

diff --git a/libavutil/Makefile b/libavutil/Makefile
index 733a23a8a3..44cd3f0dda 100644
--- a/libavutil/Makefile
+++ b/libavutil/Makefile
@@ -52,6 +52,7 @@  HEADERS = adler32.h                                                     \
           hwcontext_videotoolbox.h                                      \
           hwcontext_vdpau.h                                             \
           hwcontext_vulkan.h                                            \
+          hwcontext_nvtegra.h                                           \
           nvtegra.h                                                     \
           nvhost_ioctl.h                                                \
           nvmap_ioctl.h                                                 \
@@ -210,7 +211,7 @@  OBJS-$(CONFIG_VDPAU)                    += hwcontext_vdpau.o
 OBJS-$(CONFIG_VULKAN)                   += hwcontext_vulkan.o vulkan.o
 
 OBJS-$(!CONFIG_VULKAN)                  += hwcontext_stub.o
-OBJS-$(CONFIG_NVTEGRA)                  += nvtegra.o
+OBJS-$(CONFIG_NVTEGRA)                  += nvtegra.o hwcontext_nvtegra.o
 
 OBJS += $(COMPAT_OBJS:%=../compat/%)
 
@@ -233,7 +234,9 @@  SKIPHEADERS-$(CONFIG_VULKAN)           += hwcontext_vulkan.h vulkan.h   \
                                           vulkan_functions.h            \
                                           vulkan_loader.h
 SKIPHEADERS-$(CONFIG_NVTEGRA)          += nvtegra.h                     \
-                                          nvtegra_host1x.h
+                                          nvtegra_host1x.h              \
+                                          hwcontext_nvtegra.h
+
 
 TESTPROGS = adler32                                                     \
             aes                                                         \
diff --git a/libavutil/hwcontext.c b/libavutil/hwcontext.c
index fa99a0d8a4..8dd05147a4 100644
--- a/libavutil/hwcontext.c
+++ b/libavutil/hwcontext.c
@@ -65,6 +65,9 @@  static const HWContextType * const hw_table[] = {
 #endif
 #if CONFIG_VULKAN
     &ff_hwcontext_type_vulkan,
+#endif
+#if CONFIG_NVTEGRA
+    &ff_hwcontext_type_nvtegra,
 #endif
     NULL,
 };
@@ -82,6 +85,7 @@  static const char *const hw_type_names[] = {
     [AV_HWDEVICE_TYPE_VIDEOTOOLBOX] = "videotoolbox",
     [AV_HWDEVICE_TYPE_MEDIACODEC] = "mediacodec",
     [AV_HWDEVICE_TYPE_VULKAN] = "vulkan",
+    [AV_HWDEVICE_TYPE_NVTEGRA] = "nvtegra",
 };
 
 typedef struct FFHWDeviceContext {
diff --git a/libavutil/hwcontext.h b/libavutil/hwcontext.h
index bac30debae..d506281784 100644
--- a/libavutil/hwcontext.h
+++ b/libavutil/hwcontext.h
@@ -38,6 +38,7 @@  enum AVHWDeviceType {
     AV_HWDEVICE_TYPE_MEDIACODEC,
     AV_HWDEVICE_TYPE_VULKAN,
     AV_HWDEVICE_TYPE_D3D12VA,
+    AV_HWDEVICE_TYPE_NVTEGRA,
 };
 
 /**
diff --git a/libavutil/hwcontext_internal.h b/libavutil/hwcontext_internal.h
index e32b786238..478583abdd 100644
--- a/libavutil/hwcontext_internal.h
+++ b/libavutil/hwcontext_internal.h
@@ -163,5 +163,6 @@  extern const HWContextType ff_hwcontext_type_vdpau;
 extern const HWContextType ff_hwcontext_type_videotoolbox;
 extern const HWContextType ff_hwcontext_type_mediacodec;
 extern const HWContextType ff_hwcontext_type_vulkan;
+extern const HWContextType ff_hwcontext_type_nvtegra;
 
 #endif /* AVUTIL_HWCONTEXT_INTERNAL_H */
diff --git a/libavutil/hwcontext_nvtegra.c b/libavutil/hwcontext_nvtegra.c
new file mode 100644
index 0000000000..0f4d5a323b
--- /dev/null
+++ b/libavutil/hwcontext_nvtegra.c
@@ -0,0 +1,880 @@ 
+/*
+ * Copyright (c) 2024 averne <averne381@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU 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.
+ */
+
+#include <stdbool.h>
+
+#include "config.h"
+#include "pixdesc.h"
+#include "imgutils.h"
+#include "internal.h"
+#include "mem.h"
+#include "time.h"
+
+#include "hwcontext.h"
+#include "hwcontext_internal.h"
+
+#include "nvhost_ioctl.h"
+#include "nvmap_ioctl.h"
+#include "nvtegra_host1x.h"
+#include "clb0b6.h"
+#include "vic_drv.h"
+
+#include "hwcontext_nvtegra.h"
+
+typedef struct NVTegraDevicePriv {
+    /* The public AVNVTegraDeviceContext */
+    AVNVTegraDeviceContext p;
+
+    AVBufferRef *driver_state_ref;
+
+    AVNVTegraJobPool job_pool;
+    uint32_t vic_setup_off, vic_cmdbuf_off;
+} NVTegraDevicePriv;
+
+static const enum AVPixelFormat supported_sw_formats[] = {
+    AV_PIX_FMT_GRAY8,
+    AV_PIX_FMT_NV12,
+    AV_PIX_FMT_P010,
+    AV_PIX_FMT_YUV420P,
+};
+
+int av_nvtegra_pixfmt_to_vic(enum AVPixelFormat fmt) {
+    switch (fmt) {
+        case AV_PIX_FMT_GRAY8:
+            return NVB0B6_T_L8;
+        case AV_PIX_FMT_NV12:
+            return NVB0B6_T_Y8___U8V8_N420;
+        case AV_PIX_FMT_YUV420P:
+            return NVB0B6_T_Y8___U8___V8_N420;
+        case AV_PIX_FMT_RGB565:
+            return NVB0B6_T_R5G6B5;
+        case AV_PIX_FMT_RGB32:
+            return NVB0B6_T_A8R8G8B8;
+        case AV_PIX_FMT_BGR32:
+            return NVB0B6_T_A8B8G8R8;
+        case AV_PIX_FMT_RGB32_1:
+            return NVB0B6_T_R8G8B8A8;
+        case AV_PIX_FMT_BGR32_1:
+            return NVB0B6_T_B8G8R8A8;
+        case AV_PIX_FMT_0RGB32:
+            return NVB0B6_T_X8R8G8B8;
+        case AV_PIX_FMT_0BGR32:
+            return NVB0B6_T_X8B8G8R8;
+        default:
+            return -1;
+    }
+}
+
+static inline uint32_t nvtegra_surface_get_width_align(enum AVPixelFormat fmt, const AVComponentDescriptor *comp) {
+    int step = comp->step;
+
+    if (fmt != AV_PIX_FMT_NVTEGRA)
+        return 256 / step; /* Pitch linear surfaces must be aligned to 256B for VIC */
+
+    /*
+     * GOBs are 64B wide.
+     * In addition, we use a 32Bx8 cache width in VIC for block linear surfaces.
+     */
+    return 64 / step;
+}
+
+static inline uint32_t nvtegra_surface_get_height_align(enum AVPixelFormat fmt, const AVComponentDescriptor *comp) {
+    /* Height alignment is in terms of lines, not bytes, therefore we don't divide by the sample step */
+    if (fmt != AV_PIX_FMT_NVTEGRA)
+        return 4; /* We use 64Bx4 cache width in VIC for pitch linear surfaces */
+
+    /*
+     * GOBs are 8B high, and we use a GOB height of 2.
+     * In addition, we use a 32Bx8 cache width in VIC for block linear surfaces.
+     * We double this requirement to make sure it is respected for the subsampled chroma plane.
+     */
+    return 32;
+}
+
+static void nvtegra_device_uninit(AVHWDeviceContext *ctx) {
+    NVTegraDevicePriv       *priv = ctx->hwctx;
+    AVNVTegraDeviceContext *hwctx = &priv->p;
+
+    av_log(ctx, AV_LOG_DEBUG, "Deinitializing NVTEGRA device\n");
+
+    av_nvtegra_job_pool_uninit(&priv->job_pool);
+
+    if (hwctx->nvdec_version) {
+        av_nvtegra_channel_close(&hwctx->nvdec_channel);
+#ifdef __SWITCH__
+        mmuRequestFinalize(&hwctx->nvdec_channel.mmu_request);
+#endif
+    }
+
+    if (hwctx->nvjpg_version) {
+        av_nvtegra_channel_close(&hwctx->nvjpg_channel);
+#ifdef __SWITCH__
+        mmuRequestFinalize(&hwctx->nvjpg_channel.mmu_request);
+#endif
+    }
+
+    av_nvtegra_channel_close(&hwctx->vic_channel);
+
+    av_buffer_unref(&priv->driver_state_ref);
+}
+
+/*
+ * Hardware modules on the Tegra X1 (see t210.c in l4t kernel sources)
+ * - nvdec v2.0
+ * - nvenc v5.0
+ * - nvjpg v1.0
+ * - vic   v4.0
+ */
+
+static int nvtegra_device_init(AVHWDeviceContext *ctx) {
+    NVTegraDevicePriv       *priv = ctx->hwctx;
+    AVNVTegraDeviceContext *hwctx = &priv->p;
+
+    uint32_t vic_map_size;
+    int err;
+
+    av_log(ctx, AV_LOG_DEBUG, "Initializing NVTEGRA device\n");
+
+    err = av_nvtegra_channel_open(&hwctx->nvdec_channel, "/dev/nvhost-nvdec");
+    if (!err)
+        hwctx->nvdec_version = AV_NVTEGRA_ENCODE_REV(2,0);
+
+    err = av_nvtegra_channel_open(&hwctx->nvjpg_channel, "/dev/nvhost-nvjpg");
+    if (!err)
+        hwctx->nvjpg_version = AV_NVTEGRA_ENCODE_REV(1,0);
+
+    err = av_nvtegra_channel_open(&hwctx->vic_channel, "/dev/nvhost-vic");
+    if (err < 0)
+        goto fail;
+
+    hwctx->vic_version = AV_NVTEGRA_ENCODE_REV(4,0);
+
+    /* Note: Official code only sets this for the nvdec channel */
+    if (hwctx->nvdec_version) {
+        err = av_nvtegra_channel_set_submit_timeout(&hwctx->nvdec_channel, 1000);
+        if (err < 0)
+            goto fail;
+    }
+
+    if (hwctx->nvjpg_version) {
+        err = av_nvtegra_channel_set_submit_timeout(&hwctx->nvjpg_channel, 1000);
+        if (err < 0)
+            goto fail;
+    }
+
+    priv->vic_setup_off  = 0;
+    priv->vic_cmdbuf_off = FFALIGN(priv->vic_setup_off  + sizeof(VicConfigStruct),
+                                   AV_NVTEGRA_MAP_ALIGN);
+    vic_map_size         = FFALIGN(priv->vic_cmdbuf_off + AV_NVTEGRA_MAP_ALIGN,
+                                   0x1000);
+
+    err = av_nvtegra_job_pool_init(&priv->job_pool, &hwctx->vic_channel, vic_map_size,
+                                   priv->vic_cmdbuf_off, vic_map_size - priv->vic_cmdbuf_off);
+    if (err < 0)
+        goto fail;
+
+#ifndef __SWITCH__
+    hwctx->nvdec_channel.module_id = 0x75;
+    hwctx->nvjpg_channel.module_id = 0x76;
+#else
+    /*
+     * The NVHOST_IOCTL_CHANNEL_SET_CLK_RATE ioctl also exists on HOS but the clock rate
+     * will be reset when the console goes to sleep.
+     */
+    if (hwctx->nvdec_version) {
+        err = AVERROR(mmuRequestInitialize(&hwctx->nvdec_channel.mmu_request, (MmuModuleId)5, 8, false));
+        if (err < 0)
+            goto fail;
+    }
+
+    if (hwctx->nvjpg_version) {
+        err = AVERROR(mmuRequestInitialize(&hwctx->nvjpg_channel.mmu_request, MmuModuleId_Nvjpg, 8, false));
+        if (err < 0)
+            goto fail;
+    }
+#endif
+
+    return 0;
+
+fail:
+    nvtegra_device_uninit(ctx);
+    return err;
+}
+
+static int nvtegra_device_create(AVHWDeviceContext *ctx, const char *device,
+                                 AVDictionary *opts, int flags)
+{
+    NVTegraDevicePriv *priv = ctx->hwctx;
+
+    av_log(ctx, AV_LOG_DEBUG, "Creating NVTEGRA device\n");
+
+    priv->driver_state_ref = av_nvtegra_driver_init();
+    if (!priv->driver_state_ref) {
+        av_log(ctx, AV_LOG_ERROR, "Failed to create driver context, "
+                                  "make sure you are using a Tegra device\n");
+        return AVERROR(ENOSYS);
+    }
+
+    return 0;
+}
+
+static int nvtegra_frames_get_constraints(AVHWDeviceContext *ctx, const void *hwconfig,
+                                          AVHWFramesConstraints *constraints)
+{
+    av_log(ctx, AV_LOG_DEBUG, "Getting frame constraints for NVTEGRA device\n");
+
+    constraints->valid_sw_formats = av_malloc_array(FF_ARRAY_ELEMS(supported_sw_formats) + 1,
+                                                    sizeof(*constraints->valid_sw_formats));
+    if (!constraints->valid_sw_formats)
+        return AVERROR(ENOMEM);
+
+    for (int i = 0; i < FF_ARRAY_ELEMS(supported_sw_formats); ++i)
+        constraints->valid_sw_formats[i] = supported_sw_formats[i];
+    constraints->valid_sw_formats[FF_ARRAY_ELEMS(supported_sw_formats)] = AV_PIX_FMT_NONE;
+
+    constraints->valid_hw_formats = av_malloc_array(2, sizeof(*constraints->valid_hw_formats));
+    if (!constraints->valid_hw_formats)
+        return AVERROR(ENOMEM);
+
+    constraints->valid_hw_formats[0] = AV_PIX_FMT_NVTEGRA;
+    constraints->valid_hw_formats[1] = AV_PIX_FMT_NONE;
+
+    return 0;
+}
+
+static void nvtegra_map_free(void *opaque, uint8_t *data) {
+    AVNVTegraMap *map = (AVNVTegraMap *)data;
+
+    if (!map)
+        return;
+
+    av_nvtegra_map_destroy(map);
+
+    av_freep(&map);
+}
+
+static void nvtegra_frame_free(void *opaque, uint8_t *data) {
+    AVNVTegraFrame *frame = (AVNVTegraFrame *)data;
+
+    if (!frame)
+        return;
+
+    av_buffer_unref(&frame->map_ref);
+
+    av_freep(&frame);
+}
+
+static AVBufferRef *nvtegra_pool_alloc(void *opaque, size_t size) {
+    AVHWFramesContext        *ctx = opaque;
+    AVNVTegraDeviceContext *hwctx = &((NVTegraDevicePriv *)ctx->device_ctx->hwctx)->p;
+
+    AVBufferRef *buffer = NULL;
+    AVNVTegraFrame *frame = NULL;
+    AVNVTegraMap *map = NULL;
+    int err;
+
+    av_log(ctx, AV_LOG_DEBUG, "Creating surface from NVTEGRA device\n");
+
+    map = av_mallocz(sizeof(*map));
+    if (!map)
+        goto fail;
+
+    frame = av_mallocz(sizeof(*frame));
+    if (!map)
+        goto fail;
+
+    /*
+     * Framebuffers are allocated as CPU-cacheable, since they might get copied from
+     * during transfer operations. Cache management is done manually.
+     */
+    err = av_nvtegra_map_create(map, &hwctx->nvdec_channel, size, 0x100,
+                                NVMAP_HEAP_CARVEOUT_GENERIC, NVMAP_HANDLE_CACHEABLE);
+    if (err < 0)
+        goto fail;
+
+    /* Flush the CPU cache */
+    av_nvtegra_map_cache_op(map, NVMAP_CACHE_OP_WB, av_nvtegra_map_get_addr(map),
+                            av_nvtegra_map_get_size(map));
+
+    frame->map_ref = av_buffer_create((uint8_t *)map, sizeof(*map), nvtegra_map_free, ctx, 0);
+    if (!frame->map_ref)
+        goto fail;
+
+    buffer = av_buffer_create((uint8_t *)frame, sizeof(*frame), nvtegra_frame_free, ctx, 0);
+    if (!buffer)
+        goto fail;
+
+    return buffer;
+
+fail:
+    av_log(ctx, AV_LOG_ERROR, "Failed to create buffer\n");
+    nvtegra_frame_free(opaque, (uint8_t *)frame);
+    return NULL;
+}
+
+static int nvtegra_frames_init(AVHWFramesContext *ctx) {
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(ctx->sw_format);
+
+    uint32_t width_aligned, height_aligned, size;
+
+    av_log(ctx, AV_LOG_DEBUG, "Initializing frame pool for the NVTEGRA device\n");
+
+    if (!ctx->pool) {
+        width_aligned  = FFALIGN(ctx->width,  nvtegra_surface_get_width_align (ctx->format, &desc->comp[0]));
+        height_aligned = FFALIGN(ctx->height, nvtegra_surface_get_height_align(ctx->format, &desc->comp[0]));
+
+        size = av_image_get_buffer_size(ctx->sw_format, width_aligned, height_aligned,
+                                        nvtegra_surface_get_width_align(ctx->format, &desc->comp[0]));
+
+        ffhwframesctx(ctx)->pool_internal = av_buffer_pool_init2(size, ctx, nvtegra_pool_alloc, NULL);
+        if (!ffhwframesctx(ctx)->pool_internal)
+            return AVERROR(ENOMEM);
+    }
+
+    return 0;
+}
+
+static void nvtegra_frames_uninit(AVHWFramesContext *ctx) {
+    av_log(ctx, AV_LOG_DEBUG, "Deinitializing frame pool for the NVTEGRA device\n");
+}
+
+static int nvtegra_get_buffer(AVHWFramesContext *ctx, AVFrame *frame) {
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(ctx->sw_format);
+
+    AVNVTegraMap *map;
+    uint32_t width_aligned, height_aligned;
+    int err;
+
+    av_log(ctx, AV_LOG_DEBUG, "Getting frame buffer for NVTEGRA device\n");
+
+    frame->buf[0] = av_buffer_pool_get(ctx->pool);
+    if (!frame->buf[0])
+        return AVERROR(ENOMEM);
+
+    map = av_nvtegra_frame_get_fbuf_map(frame);
+
+    width_aligned  = FFALIGN(ctx->width,  nvtegra_surface_get_width_align (ctx->format, &desc->comp[0]));
+    height_aligned = FFALIGN(ctx->height, nvtegra_surface_get_height_align(ctx->format, &desc->comp[0]));
+
+    err = av_image_fill_arrays(frame->data, frame->linesize, av_nvtegra_map_get_addr(map),
+                               ctx->sw_format, width_aligned, height_aligned,
+                               nvtegra_surface_get_width_align(ctx->format, &desc->comp[0]));
+    if (err < 0)
+        return err;
+
+    frame->format = AV_PIX_FMT_NVTEGRA;
+    frame->width  = ctx->width;
+    frame->height = ctx->height;
+
+    return 0;
+}
+
+static int nvtegra_transfer_get_formats(AVHWFramesContext *ctx,
+                                        enum AVHWFrameTransferDirection dir,
+                                        enum AVPixelFormat **formats)
+{
+    enum AVPixelFormat *fmts;
+
+    av_log(ctx, AV_LOG_DEBUG, "Getting transfer formats for NVTEGRA device\n");
+
+    fmts = av_malloc_array(2, sizeof(**formats));
+    if (!fmts)
+        return AVERROR(ENOMEM);
+
+    fmts[0] = ctx->sw_format;
+    fmts[1] = AV_PIX_FMT_NONE;
+
+    *formats = fmts;
+    return 0;
+}
+
+static inline void nvtegra_cpu_copy_plane(void *dst, int dst_stride,
+                                          void *src, int src_stride, int h, bool from)
+{
+    /*
+     * Adapted from https://fgiesen.wordpress.com/2011/01/17/texture-tiling-and-swizzling/.
+     * We process 16x2 bytes at a time. Horizontally, this is the size of a linear atom
+     * in a 16Bx2 sector, conveniently also the size of a cache line and of a macroblock.
+     *
+     * NVDEC always uses a GOB height of 2 (block height of 16, in line with macroblock dimensions).
+     * The corresponding swizzling pattern is the following:
+     *    y3 y2 y1 y0 x5 x4 x3 x2 x1 x0
+     * x: ___x5_______x4____x3 x3 x1 x0
+     * y: y3____y2 y1____y0____________
+     *
+     * Addresses for the 4 lower bits can then be copied as-is (16 bytes).
+     * As a further optimization, the y0 bit is also handled within the same inner loop,
+     * which halves the total number of iterations.
+     *
+     * This function is declared inline with the expectation that the compiler will optimize
+     * the branches depending on the copy direction.
+     */
+
+    __uint128_t *src_ = src, *dst_ = dst, *src_line, *dst_line;
+    uint32_t ws = src_stride / sizeof(__uint128_t), wd = dst_stride / sizeof(__uint128_t),
+        w = FFMIN(ws, wd), offs_x = 0, offs_y = 0, offs_line;
+    uint32_t x_mask = -0x2e, y_mask = 0x2c;
+    int x, y;
+
+    for (y = 0; y < h; y += 2) {
+        dst_line = dst_ + (from ? y * wd : offs_y);
+        src_line = src_ + (from ? offs_y : y * ws);
+
+        offs_line = offs_x;
+        for (x = 0; x < w; ++x) {
+            dst_line[from ? x+0  : offs_line+0] = src_line[from ? offs_line+0 : x+0 ];
+            dst_line[from ? x+wd : offs_line+1] = src_line[from ? offs_line+1 : x+ws];
+            offs_line = (offs_line - x_mask) & x_mask;
+        }
+
+        offs_y = (offs_y - y_mask) & y_mask;
+
+        /* Wrap into next tile row */
+        if (!offs_y)
+            offs_x += from ? src_stride : dst_stride;
+    }
+}
+
+static int nvtegra_cpu_transfer_data(AVHWFramesContext *ctx, const AVFrame *dst, const AVFrame *src,
+                                     int num_planes, bool from)
+{
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(ctx->sw_format);
+    const AVFrame *hwframe, *swframe;
+    AVNVTegraMap *map;
+    int h, i;
+
+    hwframe = from ? src : dst, swframe = from ? dst : src;
+    map = av_nvtegra_frame_get_fbuf_map(hwframe);
+
+    if (swframe->format != ctx->sw_format) {
+        av_log(ctx, AV_LOG_ERROR, "Source and destination must have the same format for cpu transfers\n");
+        return AVERROR(EINVAL);
+    }
+
+    /* If we are transferring from a hardware frame, invalidate the CPU cache which might be stale */
+    if (from) {
+        av_nvtegra_map_cache_op(map, NVMAP_CACHE_OP_INV,
+                                av_nvtegra_map_get_addr(map), av_nvtegra_map_get_size(map));
+    }
+
+    /* Align the height to an even size */
+    h = FFALIGN(dst->height, 2);
+
+    for (i = 0; i < num_planes; ++i) {
+        if (map->is_linear) {
+            av_image_copy_plane(dst->data[i], dst->linesize[i], src->data[i], src->linesize[i],
+                                FFMIN(dst->linesize[i], src->linesize[i]),
+                                h >> (i ? desc->log2_chroma_h : 0));
+        } else {
+            /*
+             * Instanciate the same inlined function for both destinations,
+             * giving the compiler the opportunity to remove branching within the copy loops.
+             * (verified by decompilation at -O1 and higher for both gcc and clang)
+             */
+            if (from)
+                nvtegra_cpu_copy_plane(dst->data[i], dst->linesize[i], src->data[i], src->linesize[i],
+                                       h >> (i ? desc->log2_chroma_h : 0), true);
+            else
+                nvtegra_cpu_copy_plane(dst->data[i], dst->linesize[i], src->data[i], src->linesize[i],
+                                       h >> (i ? desc->log2_chroma_h : 0), false);
+        }
+    }
+
+    /* If we transferred to a hardware frame, flush the CPU cache to make the data visible to hardware engines */
+    if (!from) {
+        av_nvtegra_map_cache_op(map, NVMAP_CACHE_OP_WB,
+                                av_nvtegra_map_get_addr(map), av_nvtegra_map_get_size(map));
+    }
+
+    return 0;
+}
+
+static void nvtegra_vic_preprare_config(VicConfigStruct *config, const AVFrame *src, const AVFrame *dst,
+                                        enum AVPixelFormat fmt, bool is_16b_chroma)
+{
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(fmt);
+    bool input_linear = (src->format != AV_PIX_FMT_NVTEGRA) || av_nvtegra_frame_get_fbuf_map(src)->is_linear,
+        output_linear = (dst->format != AV_PIX_FMT_NVTEGRA) || av_nvtegra_frame_get_fbuf_map(dst)->is_linear;
+
+    /*
+     * The VIC engine has an undocumented limitation regarding height alignment,
+     * which should be padded to an even size.
+     */
+
+    /* Subsampled dimensions when emulating 16-bit chroma transfers, as input is always NV12 */
+    int divider   = !is_16b_chroma ? 1 : 2;
+    int src_width = src->width / divider, src_height = FFALIGN(src->height, 2) / divider,
+        dst_width = dst->width / divider, dst_height = FFALIGN(dst->height, 2) / divider;
+
+    *config = (VicConfigStruct){
+        .pipeConfig = {
+            .DownsampleHoriz            = 1 << 2, /* U9.2 */
+            .DownsampleVert             = 1 << 2, /* U9.2 */
+        },
+        .outputConfig = {
+            .AlphaFillMode              = !is_16b_chroma ? NVB0B6_DXVAHD_ALPHA_FILL_MODE_OPAQUE :
+                                                           NVB0B6_DXVAHD_ALPHA_FILL_MODE_SOURCE_STREAM,
+            .BackgroundAlpha            = 0,
+            .BackgroundR                = 0,
+            .BackgroundG                = 0,
+            .BackgroundB                = 0,
+            .TargetRectLeft             = 0,
+            .TargetRectRight            = dst_width  - 1,
+            .TargetRectTop              = 0,
+            .TargetRectBottom           = dst_height - 1,
+        },
+        .outputSurfaceConfig = {
+            .OutPixelFormat             = av_nvtegra_pixfmt_to_vic(fmt),
+            .OutSurfaceWidth            = dst_width  - 1,
+            .OutSurfaceHeight           = dst_height - 1,
+            .OutBlkKind                 = !output_linear ? NVB0B6_BLK_KIND_GENERIC_16Bx2 : NVB0B6_BLK_KIND_PITCH,
+            .OutBlkHeight               = !output_linear ? 1 : 0, /* GOB height 2 */
+            .OutLumaWidth               = (dst->linesize[0] / desc->comp[0].step) - 1,
+            .OutLumaHeight              = FFALIGN(dst_height, !output_linear ? 32 : 2) - 1,
+            .OutChromaWidth             = (desc->flags & AV_PIX_FMT_FLAG_RGB) ?
+                                          -1 : (dst->linesize[1] / desc->comp[1].step) - 1,
+            .OutChromaHeight            = (desc->flags & AV_PIX_FMT_FLAG_RGB) ? -1 :
+                                          (FFALIGN(dst_height, !output_linear ? 32 : 2) >> desc->log2_chroma_h) - 1,
+        },
+        .slotStruct = {
+            {
+                .slotConfig = {
+                    .SlotEnable         = 1,
+                    .CurrentFieldEnable = 1,
+                    .SoftClampLow       = 0,
+                    .SoftClampHigh      = 1023,
+                    .PlanarAlpha        = 1023,
+                    .ConstantAlpha      = 1,
+                    .SourceRectLeft     = 0,
+                    .SourceRectRight    = (src_width  - 1) << 16, /* U14.16 (for subpixel positioning) */
+                    .SourceRectTop      = 0,
+                    .SourceRectBottom   = (src_height - 1) << 16,
+                    .DestRectLeft       = 0,
+                    .DestRectRight      = src_width  - 1,
+                    .DestRectTop        = 0,
+                    .DestRectBottom     = src_height - 1,
+                },
+                .slotSurfaceConfig = {
+                    .SlotPixelFormat    = av_nvtegra_pixfmt_to_vic(fmt),
+                    .SlotChromaLocHoriz = ((desc->flags & AV_PIX_FMT_FLAG_RGB)          ||
+                                           src->chroma_location == AVCHROMA_LOC_TOPLEFT ||
+                                           src->chroma_location == AVCHROMA_LOC_LEFT    ||
+                                           src->chroma_location == AVCHROMA_LOC_BOTTOMLEFT) ? 0 : 1,
+                    .SlotChromaLocVert  = ((desc->flags & AV_PIX_FMT_FLAG_RGB)          ||
+                                           src->chroma_location == AVCHROMA_LOC_TOPLEFT ||
+                                           src->chroma_location == AVCHROMA_LOC_TOP) ? 0 :
+                                          (src->chroma_location == AVCHROMA_LOC_LEFT ||
+                                           src->chroma_location == AVCHROMA_LOC_CENTER) ? 1 : 2,
+                    .SlotBlkKind        = !input_linear ? NVB0B6_BLK_KIND_GENERIC_16Bx2 : NVB0B6_BLK_KIND_PITCH,
+                    .SlotBlkHeight      = !input_linear ? 1 : 0, /* GOB height 2 */
+                    .SlotCacheWidth     = !input_linear ? 1 : 3, /* 32Bx8 for block, 128Bx2 for pitch */
+                    .SlotSurfaceWidth   = src_width  - 1,
+                    .SlotSurfaceHeight  = src_height - 1,
+                    .SlotLumaWidth      = (src->linesize[0] / desc->comp[0].step) - 1,
+                    .SlotLumaHeight     = FFALIGN(src_height, !input_linear ? 32 : 2) - 1,
+                    .SlotChromaWidth    = (desc->flags & AV_PIX_FMT_FLAG_RGB) ?
+                                          -1 : (src->linesize[1] / desc->comp[1].step) - 1,
+                    .SlotChromaHeight   = (desc->flags & AV_PIX_FMT_FLAG_RGB) ? -1 :
+                                          (FFALIGN(src_height, !input_linear ? 32 : 2) >> desc->log2_chroma_h) - 1,
+                },
+            },
+        },
+    };
+}
+
+static int nvtegra_vic_prepare_cmdbuf(AVHWFramesContext *ctx, AVNVTegraJobPool *pool, AVNVTegraJob *job,
+                                      const AVFrame *src, const AVFrame *dst, enum AVPixelFormat fmt,
+                                      AVNVTegraMap **plane_maps, uint32_t *plane_offsets, int num_planes)
+{
+    NVTegraDevicePriv *priv = ctx->device_ctx->hwctx;
+    AVNVTegraCmdbuf *cmdbuf = &job->cmdbuf;
+
+    AVNVTegraMap *src_maps[4], *dst_maps[4];
+    uint32_t src_map_offsets[4], dst_map_offsets[4];
+    int src_reloc_type, dst_reloc_type, i, err;
+
+#define RELOC_VARS(frame) ({                                                             \
+    if (frame->format == AV_PIX_FMT_NVTEGRA) {                                           \
+        for (i = 0; i < FF_ARRAY_ELEMS(AV_JOIN(frame, _map_offsets)); ++i) {             \
+            AV_JOIN(frame, _maps       )[i] = av_nvtegra_frame_get_fbuf_map(frame);      \
+            AV_JOIN(frame, _map_offsets)[i] = frame->data[i] - frame->data[0];           \
+        }                                                                                \
+        AV_JOIN(frame, _reloc_type) = !av_nvtegra_frame_get_fbuf_map(frame)->is_linear ? \
+            NVHOST_RELOC_TYPE_BLOCK_LINEAR : NVHOST_RELOC_TYPE_PITCH_LINEAR;             \
+    } else {                                                                             \
+        for (i = 0; i < FF_ARRAY_ELEMS(AV_JOIN(frame, _map_offsets)); ++i) {             \
+            AV_JOIN(frame, _maps       )[i] = plane_maps   [i];                          \
+            AV_JOIN(frame, _map_offsets)[i] = plane_offsets[i];                          \
+        }                                                                                \
+        AV_JOIN(frame, _reloc_type) = NVHOST_RELOC_TYPE_PITCH_LINEAR;                    \
+    }                                                                                    \
+})
+
+    RELOC_VARS(src);
+    RELOC_VARS(dst);
+
+    err = av_nvtegra_cmdbuf_begin(cmdbuf, HOST1X_CLASS_VIC);
+    if (err < 0)
+        return err;
+
+    AV_NVTEGRA_PUSH_VALUE(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_SET_CONTROL_PARAMS,
+                          AV_NVTEGRA_VALUE(NVB0B6_VIDEO_COMPOSITOR_SET_CONTROL_PARAMS, CONFIG_STRUCT_SIZE, sizeof(VicConfigStruct) >> 4) |
+                          AV_NVTEGRA_VALUE(NVB0B6_VIDEO_COMPOSITOR_SET_CONTROL_PARAMS, GPTIMER_ON,         1)                            |
+                          AV_NVTEGRA_VALUE(NVB0B6_VIDEO_COMPOSITOR_SET_CONTROL_PARAMS, FALCON_CONTROL,     1));
+    AV_NVTEGRA_PUSH_RELOC(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_SET_CONFIG_STRUCT_OFFSET,
+                          &job->input_map, priv->vic_setup_off, NVHOST_RELOC_TYPE_DEFAULT);
+
+    switch (fmt) {
+        /* 16-bit transfer emulation */
+        case AV_PIX_FMT_RGB565:
+            /* Luma transfer */
+            AV_NVTEGRA_PUSH_RELOC(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_SET_SURFACE0_LUMA_OFFSET(0),
+                                  src_maps[0], src_map_offsets[0], src_reloc_type);
+            AV_NVTEGRA_PUSH_RELOC(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_SET_OUTPUT_SURFACE_LUMA_OFFSET,
+                                  dst_maps[0], dst_map_offsets[0], dst_reloc_type);
+            break;
+        case AV_PIX_FMT_RGB32:
+            /* Chroma transfer */
+            AV_NVTEGRA_PUSH_RELOC(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_SET_SURFACE0_LUMA_OFFSET(0),
+                                  src_maps[1], src_map_offsets[1], src_reloc_type);
+            AV_NVTEGRA_PUSH_RELOC(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_SET_OUTPUT_SURFACE_LUMA_OFFSET,
+                                  dst_maps[1], dst_map_offsets[1], dst_reloc_type);
+            break;
+
+        /* Normal transfers */
+        case AV_PIX_FMT_GRAY8:
+        case AV_PIX_FMT_NV12:
+        case AV_PIX_FMT_YUV420P:
+            for (i = 0; i < num_planes; ++i) {
+                AV_NVTEGRA_PUSH_RELOC(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_SET_SURFACE0_LUMA_OFFSET(0)    + i * sizeof(uint32_t),
+                                      src_maps[i], src_map_offsets[i], src_reloc_type);
+                AV_NVTEGRA_PUSH_RELOC(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_SET_OUTPUT_SURFACE_LUMA_OFFSET + i * sizeof(uint32_t),
+                                      dst_maps[i], dst_map_offsets[i], dst_reloc_type);
+            }
+            break;
+        default:
+            return AVERROR(EINVAL);
+    }
+
+    AV_NVTEGRA_PUSH_VALUE(cmdbuf, NVB0B6_VIDEO_COMPOSITOR_EXECUTE,
+                          AV_NVTEGRA_ENUM(NVB0B6_VIDEO_COMPOSITOR_EXECUTE, AWAKEN, ENABLE));
+
+    err = av_nvtegra_cmdbuf_add_syncpt_incr(cmdbuf, pool->channel->syncpt, 0);
+    if (err < 0)
+        return err;
+
+    err = av_nvtegra_cmdbuf_end(cmdbuf);
+    if (err < 0)
+        return err;
+
+    return 0;
+}
+
+static int nvtegra_vic_copy_plane(AVHWFramesContext *ctx, AVNVTegraJob *job,
+                                  const AVFrame *src, const AVFrame *dst,
+                                  enum AVPixelFormat fmt, AVNVTegraMap **plane_maps, uint32_t *plane_offsets,
+                                  int num_planes, bool is_chroma)
+{
+    NVTegraDevicePriv *priv = ctx->device_ctx->hwctx;
+
+    uint8_t *mem;
+    int err;
+
+    mem = av_nvtegra_map_get_addr(&job->input_map);
+
+    nvtegra_vic_preprare_config((VicConfigStruct *)(mem + priv->vic_setup_off),
+                                src, dst, fmt, is_chroma);
+
+    err = av_nvtegra_cmdbuf_clear(&job->cmdbuf);
+    if (err < 0)
+        return err;
+
+    err = nvtegra_vic_prepare_cmdbuf(ctx, &priv->job_pool, job, src, dst, fmt,
+                                     plane_maps, plane_offsets, num_planes);
+    if (err < 0)
+        goto fail;
+
+    err = av_nvtegra_job_submit(&priv->job_pool, job);
+    if (err < 0)
+        goto fail;
+
+    err = av_nvtegra_job_wait(&priv->job_pool, job, -1);
+    if (err < 0)
+        goto fail;
+
+fail:
+    return err;
+}
+
+static int nvtegra_vic_transfer_data(AVHWFramesContext *ctx, const AVFrame *dst, const AVFrame *src,
+                                     int num_planes, bool from)
+{
+    NVTegraDevicePriv       *priv = ctx->device_ctx->hwctx;
+    AVNVTegraDeviceContext *hwctx = &priv->p;
+
+    AVBufferRef *job_ref;
+    AVNVTegraJob *job;
+    const AVFrame *swframe;
+    uint8_t *map_bases[4];
+    AVNVTegraMap maps[4] = {0};
+    AVNVTegraMap *plane_maps[4];
+    uint32_t plane_offsets[4];
+    int num_maps, i, j, err;
+
+    swframe = from ? dst : src;
+
+    job_ref = av_nvtegra_job_pool_get(&priv->job_pool);
+    if (!job_ref) {
+        err = AVERROR(ENOMEM);
+        goto fail;
+    }
+
+    job = (AVNVTegraJob *)job_ref->data;
+
+    /* Create a map for each frame backing buffer */
+    for (i = 0; i < FF_ARRAY_ELEMS(maps); num_maps = ++i) {
+        if (!swframe->buf[i])
+            break;
+
+        /*
+         * In order to avoid a full-frame copy on the CPU, the provided memory
+         * is mapped into VIC and used directly during the transfer.
+         * The address and size are aligned to page boundaries.
+         * Cache management is performed manually to not affect data outside the buffer.
+         */
+        map_bases[i] = (uint8_t *)((uintptr_t)swframe->buf[i]->data & ~0xfff);
+        err = av_nvtegra_map_from_va(&maps[i], &hwctx->vic_channel, map_bases[i],
+                                     swframe->buf[i]->size + ((uintptr_t)swframe->buf[i]->data & 0xfff),
+                                     0x100, NVMAP_HANDLE_CACHEABLE);
+        if (err < 0)
+            goto fail;
+
+        err = av_nvtegra_map_map(&maps[i]);
+        if (err < 0)
+            goto fail;
+
+        /* Flush-invalidate the CPU cache prior to the transfer */
+        av_nvtegra_map_cache_op(&maps[i], NVMAP_CACHE_OP_WB_INV,
+                                ((uint8_t *)av_nvtegra_map_get_addr(&maps[i])) +
+                                    ((uintptr_t)swframe->buf[i]->data & 0xfff),
+                                swframe->buf[i]->size);
+    }
+
+    /* Find the corresponding map object and its offset for each plane  */
+    for (i = 0; i < num_planes; ++i) {
+        for (j = 0; j < FF_ARRAY_ELEMS(swframe->buf); ++j) {
+            if ((swframe->buf[j]->data <= swframe->data[i]) &&
+                    (swframe->data[i] < swframe->buf[j]->data + swframe->buf[j]->size))
+                break;
+        }
+
+        plane_maps   [i] = &maps[j];
+        plane_offsets[i] = swframe->data[i] - map_bases[j];
+    }
+
+    /* VIC expects planes in the reversed order */
+    if (swframe->format == AV_PIX_FMT_YUV420P) {
+        FFSWAP(AVNVTegraMap *, plane_maps   [1], plane_maps   [2]);
+        FFSWAP(uint32_t,       plane_offsets[1], plane_offsets[2]);
+    }
+
+    /*
+     * VIC2 does not support 16-bit YUV surfaces (eg. P010, P012, ...).
+     * Here we emulate them using two separates transfers for the luma and chroma planes
+     * (16-bit and 32-bit widths respectively).
+     */
+    if (swframe->format == AV_PIX_FMT_P010) {
+        err = nvtegra_vic_copy_plane(ctx, job, src, dst, AV_PIX_FMT_RGB565,
+                                     plane_maps, plane_offsets, 1, false);
+        if (err < 0)
+            goto fail;
+
+        err = nvtegra_vic_copy_plane(ctx, job, src, dst, AV_PIX_FMT_RGB32,
+                                     plane_maps, plane_offsets, 1, true);
+        if (err < 0)
+            goto fail;
+    } else {
+        err = nvtegra_vic_copy_plane(ctx, job, src, dst, swframe->format,
+                                     plane_maps, plane_offsets, num_planes, false);
+        if (err < 0)
+            goto fail;
+    }
+
+fail:
+    for (i = 0; i < num_maps; ++i) {
+        av_nvtegra_map_unmap(&maps[i]);
+        av_nvtegra_map_close(&maps[i]);
+    }
+
+    av_buffer_unref(&job_ref);
+
+    return err;
+}
+
+static int nvtegra_transfer_data(AVHWFramesContext *ctx, AVFrame *dst, const AVFrame *src) {
+    const AVFrame *swframe;
+    bool from;
+    int num_planes, i;
+
+    from    = !dst->hw_frames_ctx;
+    swframe = from ? dst : src;
+
+    if (swframe->hw_frames_ctx)
+        return AVERROR(ENOSYS);
+
+    num_planes = av_pix_fmt_count_planes(swframe->format);
+
+    for (i = 0; i < num_planes; ++i) {
+        if (((uintptr_t)swframe->data[i] & 0xff) || (swframe->linesize[i] & 0xff)) {
+            av_log(ctx, AV_LOG_WARNING, "Frame address/pitch not aligned to 256, "
+                                        "falling back to cpu transfer\n");
+            return nvtegra_cpu_transfer_data(ctx, dst, src, num_planes, from);
+        }
+    }
+
+    return nvtegra_vic_transfer_data(ctx, dst, src, num_planes, from);
+}
+
+const HWContextType ff_hwcontext_type_nvtegra = {
+    .type                   = AV_HWDEVICE_TYPE_NVTEGRA,
+    .name                   = "nvtegra",
+
+    .device_hwctx_size      = sizeof(NVTegraDevicePriv),
+    .device_hwconfig_size   = 0,
+    .frames_hwctx_size      = 0,
+
+    .device_create          = &nvtegra_device_create,
+    .device_init            = &nvtegra_device_init,
+    .device_uninit          = &nvtegra_device_uninit,
+
+    .frames_get_constraints = &nvtegra_frames_get_constraints,
+    .frames_init            = &nvtegra_frames_init,
+    .frames_uninit          = &nvtegra_frames_uninit,
+    .frames_get_buffer      = &nvtegra_get_buffer,
+
+    .transfer_get_formats   = &nvtegra_transfer_get_formats,
+    .transfer_data_to       = &nvtegra_transfer_data,
+    .transfer_data_from     = &nvtegra_transfer_data,
+
+    .pix_fmts = (const enum AVPixelFormat[]) {
+        AV_PIX_FMT_NVTEGRA,
+        AV_PIX_FMT_NONE,
+    },
+};
diff --git a/libavutil/hwcontext_nvtegra.h b/libavutil/hwcontext_nvtegra.h
new file mode 100644
index 0000000000..8a2383d304
--- /dev/null
+++ b/libavutil/hwcontext_nvtegra.h
@@ -0,0 +1,85 @@ 
+/*
+ * Copyright (c) 2024 averne <averne381@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU 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.
+ */
+
+#ifndef AVUTIL_HWCONTEXT_NVTEGRA_H
+#define AVUTIL_HWCONTEXT_NVTEGRA_H
+
+#include <stdint.h>
+
+#include "hwcontext.h"
+#include "buffer.h"
+#include "frame.h"
+#include "pixfmt.h"
+
+#include "nvtegra.h"
+
+/*
+ * Encode a hardware revision into a version number
+ */
+#define AV_NVTEGRA_ENCODE_REV(maj, min) (((maj & 0xff) << 8) | (min & 0xff))
+
+/*
+ * Decode a version number
+ */
+static inline void av_nvtegra_decode_rev(int rev, int *maj, int *min) {
+    *maj = (rev >> 8) & 0xff;
+    *min = (rev >> 0) & 0xff;
+}
+
+/**
+ * @file
+ * API-specific header for AV_HWDEVICE_TYPE_NVTEGRA.
+ *
+ * For user-allocated pools, AVHWFramesContext.pool must return AVBufferRefs
+ * with the data pointer set to an AVNVTegraMap.
+ */
+
+typedef struct AVNVTegraDeviceContext {
+    /*
+     * Hardware multimedia engines
+     */
+    AVNVTegraChannel nvdec_channel, nvenc_channel, nvjpg_channel, vic_channel;
+
+    /*
+     * Hardware revisions for associated engines, or 0 if invalid
+     */
+    int nvdec_version, nvenc_version, nvjpg_version, vic_version;
+} AVNVTegraDeviceContext;
+
+typedef struct AVNVTegraFrame {
+    /*
+     * Reference to an AVNVTegraMap object
+     */
+    AVBufferRef *map_ref;
+} AVNVTegraFrame;
+
+/*
+ * Helper to retrieve a map object from the corresponding frame
+ */
+static inline AVNVTegraMap *av_nvtegra_frame_get_fbuf_map(const AVFrame *frame) {
+    return (AVNVTegraMap *)((AVNVTegraFrame *)frame->buf[0]->data)->map_ref->data;
+}
+
+/*
+ * Converts a pixel format to the equivalent code for the VIC engine
+ */
+int av_nvtegra_pixfmt_to_vic(enum AVPixelFormat fmt);
+
+#endif /* AVUTIL_HWCONTEXT_NVTEGRA_H */