vo_gpu_next: add new libplacebo-based renderer

As discussed in #8799, this will eventually replace vo_gpu. However, it
is not yet complete. Currently missing:

- OpenGL contexts
- hardware decoding
- blend-subtitles=video
- VOCTRL_SCREENSHOT

However, it's usable enough to cover most use cases, and as such is
enough to start getting in some crucial testing.

3 files changed, 96 insertions, 2 deletions

diff --git a/DOCS/interface-changes.rst b/DOCS/interface-changes.rst
index 84418c5ce3..6aa6fd907d 100644
--- a/DOCS/interface-changes.rst
+++ b/DOCS/interface-changes.rst
@@ -26,6 +26,11 @@ Interface changes
 
 ::
 
+ --- mpv 0.35.0 ---
+    - add the `--vo=gpu-next` video output driver, as well as the options
+      `--allow-delayed-peak-detect`, `--builtin-scalers`,
+      `--interpolation-preserve` `--lut`, `--lut-type`, `--image-lut`,
+      `--image-lut-type` and `--target-lut` along with it.
  --- mpv 0.34.0 ---
     - deprecate selecting by card number with `--drm-connector`, add
       `--drm-device` which can be used instead
diff --git a/DOCS/man/options.rst b/DOCS/man/options.rst
index 3ade7f764c..4b97915734 100644
--- a/DOCS/man/options.rst
+++ b/DOCS/man/options.rst
@@ -5013,8 +5013,9 @@ ALSA audio output options
 GPU renderer options
 -----------------------
 
-The following video options are currently all specific to ``--vo=gpu`` and
-``--vo=libmpv`` only, which are the only VOs that implement them.
+The following video options are currently all specific to ``--vo=gpu``,
+``--vo=libmpv`` and ``--vo=gpu-next``, which are the only VOs that implement
+them.
 
 ``--scale=<filter>``
     The filter function to use when upscaling video.
@@ -5191,6 +5192,12 @@ The following video options are currently all specific to ``--vo=gpu`` and
     will reproduce the source image perfectly if no scaling is performed.
     Enabled by default. Note that this option never affects ``--cscale``.
 
+``--builtin-scalers``
+    Allow using faster built-in replacements for common scalers such as
+    ``nearest``, ``bilinear`` or ``bicubic``. These have the disadvantage of
+    not being configurable, unlike normal scaler kernels. Defaults to
+    enabled. (This option only affects ``--vo=gpu-next``)
+
 ``--correct-downscaling``
     When using convolution based filters, extend the filter size when
     downscaling. Increases quality, but reduces performance while downscaling.
@@ -5267,6 +5274,15 @@ The following video options are currently all specific to ``--vo=gpu`` and
 
     Set this to ``-1`` to disable this logic.
 
+``--interpolation-preserve``
+    Preserve the previous frames' interpolated results even when renderer
+    parameters are changed - with the exception of options related to
+    cropping and video placement, which always invalidate the cache. Enabling
+    this option makes dynamic updates of renderer settings slightly smoother at
+    the cost of slightly higher latency in response to such changes. Defaults
+    to on. (Only affects ``--vo=gpu-next``, note that ``-vo=gpu`` always
+    invalidates interpolated frames)
+
 ``--opengl-pbo``
     Enable use of PBOs. On some drivers this can be faster, especially if the
     source video size is huge (e.g. so called "4K" video). On other drivers it
@@ -6139,6 +6155,29 @@ The following video options are currently all specific to ``--vo=gpu`` and
 
     NOTE: Only implemented on macOS.
 
+``--image-lut=<file>``
+    Specifies a custom LUT file (in Adobe .cube format) to apply to the colors
+    during image decoding. The exact interpretation of the LUT depends on
+    the value of ``--image-lut-type``. (Only for ``--vo=gpu-next``)
+
+``--image-lut-type=<value>``
+    Controls the interpretation of color values fed to and from the LUT
+    specified as ``--image-lut``. Valid values are:
+
+    auto
+        Chooses the interpretation of the LUT automatically from tagged
+        metadata, and otherwise falls back to ``native``. (Default)
+    native
+        Applied to the raw image contents in its native colorspace, before
+        decoding to RGB. For example, for a HDR10 image, this would be fed
+        PQ-encoded YCbCr values in the range 0.0 - 1.0.
+    normalized
+        Applied to the normalized RGB image contents, after decoding from
+        its native color encoding, but before linearization.
+    conversion
+        Fully replaces the color decoding. A LUT of this type should ingest the
+        image's native colorspace and output normalized non-linear RGB.
+
 ``--target-prim=<value>``
     Specifies the primaries of the display. Video colors will be adapted to
     this colorspace when ICC color management is not being used. Valid values
@@ -6254,6 +6293,12 @@ The following video options are currently all specific to ``--vo=gpu`` and
         In such a configuration, we highly recommend setting ``--tone-mapping``
         to ``mobius`` or even ``clip``.
 
+``--target-lut=<file>``
+    Specifies a custom LUT file (in Adobe .cube format) to apply to the colors
+    before display on-screen. This LUT is fed values in normalized RGB, after
+    encoding into the target colorspace, so after the application of
+    ``--target-trc``. (Only for ``--vo=gpu-next``)
+
 ``--tone-mapping=<value>``
     Specifies the algorithm used for tone-mapping images onto the target
     display. This is relevant for both HDR->SDR conversion as well as gamut
@@ -6335,6 +6380,14 @@ The following video options are currently all specific to ``--vo=gpu`` and
     The special value ``auto`` (default) will enable HDR peak computation
     automatically if compute shaders and SSBOs are supported.
 
+``--allow-delayed-peak-detect``
+    When using ``--hdr-compute-peak``, allow delaying the detected peak by a
+    frame when beneficial for performance. In particular, this is required to
+    avoid an unnecessary FBO indirection when no advanced rendering is required
+    otherwise. Has no effect if there already is an indirect pass, such as when
+    advanced scaling is enabled. Defaults to on. (Only affects
+    ``--vo=gpu-next``, note that ``--vo=gpu`` always delays the peak.)
+
 ``--hdr-peak-decay-rate=<1.0..1000.0>``
     The decay rate used for the HDR peak detection algorithm (default: 100.0).
     This is only relevant when ``--hdr-compute-peak`` is enabled. Higher values
@@ -6456,6 +6509,30 @@ The following video options are currently all specific to ``--vo=gpu`` and
     value ``inf`` causes the BT.1886 curve to be treated as a pure power gamma
     2.4 function.
 
+``--lut=<file>``
+    Specifies a custom LUT (in Adobe .cube format) to apply to the colors
+    as part of color conversion. The exact interpretation depends on the value
+    of ``--lut-type``. (Only for ``--vo=gpu-next``)
+
+``--lut-type=<value>``
+    Controls the interpretation of color values fed to and from the LUT
+    specified as ``--lut``. Valid values are:
+
+    auto
+        Chooses the interpretation of the LUT automatically from tagged
+        metadata, and otherwise falls back to ``native``. (Default)
+    native
+        Applied to raw image contents in its native RGB colorspace (non-linear
+        light), before conversion to the output color space.
+    normalized
+        Applied to the normalized RGB image contents, in linear light, before
+        conversion to the output color space.
+    conversion
+        Fully replaces the conversion from the image color space to the output
+        color space. If such a LUT is present, it has the highest priority, and
+        overrides any ICC profiles, as well as options related to tone mapping
+        and output colorimetry (``--target-prim``, ``--target-trc`` etc.).
+
 ``--blend-subtitles=<yes|video|no>``
     Blend subtitles directly onto upscaled video frames, before interpolation
     and/or color management (default: no). Enabling this causes subtitles to be
diff --git a/DOCS/man/vo.rst b/DOCS/man/vo.rst
index 7632f3c406..7897336266 100644
--- a/DOCS/man/vo.rst
+++ b/DOCS/man/vo.rst
@@ -257,6 +257,18 @@ Available video output drivers are:
     with ``rgb32f``. If you have problems, you can also try enabling the
     ``--gpu-dumb-mode=yes`` option.
 
+``gpu-next``
+    Experimental video renderer based on ``libplacebo``. This supports almost
+    the same set of features as ``--vo=gpu``. See `GPU renderer options`_ for a
+    list.
+
+    Currently, this only supports ``--gpu-api=vulkan``, and no hardware
+    decoding. Unlike ``--vo=gpu``, the FBO formats are not tunable, but you can
+    still set ``--gpu-dumb-mode=yes`` to forcibly disable their use.
+
+    Should generally be faster and higher quality, but some features may still
+    be missing or misbehave. Expect (and report!) bugs.
+
 ``sdl``
     SDL 2.0+ Render video output driver, depending on system with or without
     hardware acceleration. Should work on all platforms supported by SDL 2.0.