v2.72.1

This release backports a number of bug fixes from master affecting the
v2.72.0 release, most notably extending the range of supported glslang
versions, as well as fixing the 3DLUT/ICC generation code.

Bug fixes:
- compatibility with glslang >= 8.13.3743
- compatibility with new glslang semantic versioning scheme
- fix broken shaders on some versions of GLES by defaulting to 32-bit
  precision for floating point math
- fix the vulkan API version passed to shaderc
- fix the extension check for glInvalidateTexImage
- fix pl_tex_create on older GLES versions
- fix OpenGL logging thread safety
- fix undefined memcmp() in pl_shader_av1_grain
- fix pl_render_target.repr being ignored by pl_render_image
- fix 3DLUT generation code (generated corrupt LUTs in all cases)
- fix symbol visibility on some versions of GCC
- fix potential overflow in BT.2390 shader
- properly restrict pl_shader_sample_polar to GLSL >= 130
- fix vulkan function loading of promoted core functions

Other changes:
- tiny performance gain in 3DLUT generation
- pl_opengl_create now logs GL_EXTENSIONS
- log some additional VkResult enum members
- improve several vulkan log messages by using friendly names of enums
- add a copy of the config.h variables to the libplacebo.pc file

v2.72.0

This is a major release with several key additions, most notably being
the support for custom, mpv-style "user shaders" (.hook), giving us
access to a large variety of pre-existing user shaders such as RAVU,
FSRCNNX, Anime4K, SSimSuperRes, KrigBilateral, NNEDI3, and more.

In addition to this, major additions include a completely refactored and
fixed AV1 grain generation shader, support for Vulkan versions higher
than 1.0, support for GPU-based timers, and improved interop APIs for
both Vulkan and OpenGL, and new and improved aspect ratio handling.

Finally, this release also brings with it a major change to the way HDR
and SDR content are mapped between each other, including a new
tone-mapping function based on the industry-standard ITU-R BT.2390 EETF.

Additions:
- add `pl_swapchain_hdr_metadata`, to set HDR metadata on supported
  swapchains (currently only vulkan with `VK_EXT_hdr_metadata`)
- add support for vulkan versions higher than 1.0, communicated via the
  new fields `api_version` and `max_api_version`
- add support for GPU-assisted validation and best practices layers, via
  the new field `pl_vk_inst_params.debug_extra`
- add helper functions for working with `pl_rect`s, including new aspect
  ratio handling helpers (`pl_rect2df_aspect_*`)
- add field `pl_vulkan_params.device_uuid` to allow choosing the vulkan
  device by its UUID
- add function `pl_vulkan_hold_raw`, to hold images without actually
  transitioning its layout and access mode
- add function `pl_vulkan_import`, to allow directly re-using an
  existing VkDevice rather than creating a new one; this requires
  communicating metadata about how the device was created
- add field `pl_vulkan_params.features` to allow loading optional extra
  device features at device creation time
- add support for mpv-style custom user shaders (.hook), using the set
  of functions in `<libplacebo/shaders/custom.h>`
- add `pl_render_high_quality_params`, enabling debanding and EWA
  scaling
- add `pl_timer` GPU resource type and associated API functions,
  allowing the GPU execution time of shaders and texture transfer
  operations to be measured directly
- add `PL_SHADER_SIG_SAMPLER`, allowing generated sampling shaders to
  directly accept the sampler to use as function parameters
- add `pl_image_set_chroma_location` to automatically apply the correct
  chroma location to any subsampled planes
- add `PL_TONE_MAPPING_BT_2390`, a tone mapping function based on the
  EETF from ITU-R Report BT.2390 (and make it the default)
- add `pl_peak_detect_params.overshoot_margin` to help combat clipping
  on certain types of rapid scene fade-ins
- add `pl_sampler_type` to allow encoding non-standard sampler types
  such as sampler2DRect, and also generalize samplers to allow e.g.
  usampler2D or isampler3D
- add `pl_opengl_wrap` and `pl_opengl_unwrap`, to allow directly mapping
  between OpenGL textures and the `pl_tex` abstraction

Changes:
- deprecate `pl_image.width/height`, which are now inferred
  automatically from the actual planes
- `pl_vulkan_wrap` now takes a `pl_vulkan_wrap_params` struct instead of
  directly accepting its parameters, including new fields `sample_mode`
  and `address_mode` to configure the created sampler
- change `pl_dispatch_compute` to allow optionally passing in a
  simulated framebuffer width/height, which will be used to translate
  vertex attributes (if any)
- undefine disabled `config.h` features, instead of defining them as 0
- remove debanding from `pl_render_default_params`
- refactor HDR<->SDR mapping; PL_COLOR_REF_WHITE has been removed and
  replaced by PL_COLOR_SDR_WHITE (203 cd/m^2) and PL_COLOR_SDR_WHITE_HLG
  (75% HLG), respectively
- completely refactor pl_shader_av1_grain`, which now samples directly
  from the passed texture rather than requiring the color be pre-sampled
- `pl_render_image` now infers the image primaries based on resolution,
  rather than always hard-coding `PL_COLOR_SPACE_UNKNOWN` as BT.709
- change `pl_render_target.dst_rect` from `pl_rect2d` to `pl_rect2df`,
  allowing more accurate aspect ratio handling, and correctly compensate
  for subpixel scaling ratios
- require `python3-mako` as a dependency of the `vulkan` feature
- `pl_chroma_location_offset` now treats `PL_CHROMA_UNKNOWN` as
  `PL_CHROMA_LEFT`, the de-facto standard chroma location
- the default value of `pl_color_map_params.tone_mapping_algo` is now
  `PL_TONE_MAPPING_BT_2390`

Fixes and performance improvements:
- fix shader generation when the GLSL version is explicitly overridden
- properly mark some shader failures (`pl_shader_is_failed`)
- fix texture invalidation on OpenGL
- correctly respect `pl_swapchain_frame.flipped` in
  `pl_render_target_from_swapchain`
- correctly validate descriptor uniqueness in `pl_pass_create`
- skip redundant matrix multiplication in `pl_shader_encode_color`
  wherever possible
- work around driver bugs w.r.t out-of-order buffer offsets by sorting
  all buffer variables by offset
- fix edge cases in vulkan swapchain usage flag checks
- fix excessive CPU usage in `pl_tex_download`
- reduce the number of unnecessary GPU flushes caused by `pl_buf_poll`
- fix issue where blending did not work on some drivers (e.g. nvidia)
- make the framebuffer discard check more aggressive
- fix computation of anti-aliased resizable orthogonal filters, e.g.
  when downscaling using `pl_filter_lanczos`
- fix external image memory barriers for exclusive mode images
- fix failure path of `pl_swapchain_submit_frame`
- fix various GLSL compatibility issues with av1 grain generation
- reduce maximum vulkan memory allocation slab size to conform to AMD
  recommendations
- fix build error when lcms is not available
- fix double-application of texture scale for e.g. 10-bit content when
  using separable scalers
- fix a multitude of bugs affecting av1 grain generation, especially for
  chroma planes
- fix segfault on vulkan device oom
- fix invalid shader generation on some platforms
- fix a multitude of bugs, edge cases and subtle off-by-ones related to
  chroma scaling and plane alignment
- add fallback code for edge case w.r.t chroma scaling and gpu resource
  exhaustion
- correctly load VK_KHR_swapchain in all circumstances that require
  accessing its functions
- minimize fbo usage inside `pl_renderer`, by re-using unused fbos
- tweak the work group size for polar scaling to perform better on
  modern GPUs (tested on RDNA)
- transparently upgrade fragment shaders to compute shaders on
  environments with async compute
- pick a more reasonable size for the dummy gpu's `max_group_threads`
- forbid 10-bit linear transfer functions from vulkan swapchains
- fix segfault when re-executing previously failed shaders
- fix swapchain creation errors on GLES 2
- explicitly mark all shader resources as non-aliased
- correctly specify shader storage buffers as coherent for shaders that
  require them
- fix various memory barrier synchronization issues for opengl

v2.72.0-rc2

This is a major release with several key additions, most notably being
the support for custom, mpv-style "user shaders" (.hook), giving us
access to a large variety of pre-existing user shaders such as RAVU,
FSRCNNX, Anime4K, SSimSuperRes, KrigBilateral, NNEDI3, and more.

In addition to this, major additions include a completely refactored and
fixed AV1 grain generation shader, support for Vulkan versions higher
than 1.0, support for GPU-based timers, and improved interop APIs for
both Vulkan and OpenGL, and new and improved aspect ratio handling.

Finally, this release also brings with it a major change to the way HDR
and SDR content are mapped between each other, including a new
tone-mapping function based on the industry-standard ITU-R BT.2390 EETF.

Additions:
- add `pl_swapchain_hdr_metadata`, to set HDR metadata on supported
  swapchains (currently only vulkan with `VK_EXT_hdr_metadata`)
- add support for vulkan versions higher than 1.0, communicated via the
  new fields `api_version` and `max_api_version`
- add support for GPU-assisted validation and best practices layers, via
  the new field `pl_vk_inst_params.debug_extra`
- add helper functions for working with `pl_rect`s, including new aspect
  ratio handling helpers (`pl_rect2df_aspect_*`)
- add field `pl_vulkan_params.device_uuid` to allow choosing the vulkan
  device by its UUID
- add function `pl_vulkan_hold_raw`, to hold images without actually
  transitioning its layout and access mode
- add function `pl_vulkan_import`, to allow directly re-using an
  existing VkDevice rather than creating a new one; this requires
  communicating metadata about how the device was created
- add field `pl_vulkan_params.features` to allow loading optional extra
  device features at device creation time
- add support for mpv-style custom user shaders (.hook), using the set
  of functions in `<libplacebo/shaders/custom.h>`
- add `pl_render_high_quality_params`, enabling debanding and EWA
  scaling
- add `pl_timer` GPU resource type and associated API functions,
  allowing the GPU execution time of shaders and texture transfer
  operations to be measured directly
- add `PL_SHADER_SIG_SAMPLER`, allowing generated sampling shaders to
  directly accept the sampler to use as function parameters
- add `pl_image_set_chroma_location` to automatically apply the correct
  chroma location to any subsampled planes
- add `PL_TONE_MAPPING_BT_2390`, a tone mapping function based on the
  EETF from ITU-R Report BT.2390 (and make it the default)
- add `pl_peak_detect_params.overshoot_margin` to help combat clipping
  on certain types of rapid scene fade-ins
- add `pl_sampler_type` to allow encoding non-standard sampler types
  such as sampler2DRect, and also generalize samplers to allow e.g.
  usampler2D or isampler3D
- add `pl_opengl_wrap` and `pl_opengl_unwrap`, to allow directly mapping
  between OpenGL textures and the `pl_tex` abstraction

Changes:
- deprecate `pl_image.width/height`, which are now inferred
  automatically from the actual planes
- `pl_vulkan_wrap` now takes a `pl_vulkan_wrap_params` struct instead of
  directly accepting its parameters, including new fields `sample_mode`
  and `address_mode` to configure the created sampler
- change `pl_dispatch_compute` to allow optionally passing in a
  simulated framebuffer width/height, which will be used to translate
  vertex attributes (if any)
- undefine disabled `config.h` features, instead of defining them as 0
- remove debanding from `pl_render_default_params`
- refactor HDR<->SDR mapping; PL_COLOR_REF_WHITE has been removed and
  replaced by PL_COLOR_SDR_WHITE (203 cd/m^2) and PL_COLOR_SDR_WHITE_HLG
  (75% HLG), respectively
- completely refactor pl_shader_av1_grain`, which now samples directly
  from the passed texture rather than requiring the color be pre-sampled
- `pl_render_image` now infers the image primaries based on resolution,
  rather than always hard-coding `PL_COLOR_SPACE_UNKNOWN` as BT.709
- change `pl_render_target.dst_rect` from `pl_rect2d` to `pl_rect2df`,
  allowing more accurate aspect ratio handling, and correctly compensate
  for subpixel scaling ratios
- require `python3-mako` as a dependency of the `vulkan` feature
- `pl_chroma_location_offset` now treats `PL_CHROMA_UNKNOWN` as
  `PL_CHROMA_LEFT`, the de-facto standard chroma location
- the default value of `pl_color_map_params.tone_mapping_algo` is now
  `PL_TONE_MAPPING_BT_2390`

Fixes and performance improvements:
- fix shader generation when the GLSL version is explicitly overridden
- properly mark some shader failures (`pl_shader_is_failed`)
- fix texture invalidation on OpenGL
- correctly respect `pl_swapchain_frame.flipped` in
  `pl_render_target_from_swapchain`
- correctly validate descriptor uniqueness in `pl_pass_create`
- skip redundant matrix multiplication in `pl_shader_encode_color`
  wherever possible
- work around driver bugs w.r.t out-of-order buffer offsets by sorting
  all buffer variables by offset
- fix edge cases in vulkan swapchain usage flag checks
- fix excessive CPU usage in `pl_tex_download`
- reduce the number of unnecessary GPU flushes caused by `pl_buf_poll`
- fix issue where blending did not work on some drivers (e.g. nvidia)
- make the framebuffer discard check more aggressive
- fix computation of anti-aliased resizable orthogonal filters, e.g.
  when downscaling using `pl_filter_lanczos`
- fix external image memory barriers for exclusive mode images
- fix failure path of `pl_swapchain_submit_frame`
- fix various GLSL compatibility issues with av1 grain generation
- reduce maximum vulkan memory allocation slab size to conform to AMD
  recommendations
- fix build error when lcms is not available
- fix double-application of texture scale for e.g. 10-bit content when
  using separable scalers
- fix a multitude of bugs affecting av1 grain generation, especially for
  chroma planes
- fix segfault on vulkan device oom
- fix invalid shader generation on some platforms
- fix a multitude of bugs, edge cases and subtle off-by-ones related to
  chroma scaling and plane alignment
- add fallback code for edge case w.r.t chroma scaling and gpu resource
  exhaustion
- correctly load VK_KHR_swapchain in all circumstances that require
  accessing its functions
- minimize fbo usage inside `pl_renderer`, by re-using unused fbos
- tweak the work group size for polar scaling to perform better on
  modern GPUs (tested on RDNA)
- transparently upgrade fragment shaders to compute shaders on
  environments with async compute
- pick a more reasonable size for the dummy gpu's `max_group_threads`
- forbid 10-bit linear transfer functions from vulkan swapchains
- fix segfault when re-executing previously failed shaders
- fix swapchain creation errors on GLES 2
- explicitly mark all shader resources as non-aliased
- correctly specify shader storage buffers as coherent for shaders that
  require them
- fix various memory barrier synchronization issues for opengl

v2.72.0-rc1

This is a major release with several key additions, most notably being
the support for custom, mpv-style "user shaders" (.hook), giving us
access to a large variety of pre-existing user shaders such as RAVU,
FSRCNNX, Anime4K, SSimSuperRes, KrigBilateral, NNEDI3, and more.

In addition to this, major additions include a completely refactored and
fixed AV1 grain generation shader, support for Vulkan versions higher
than 1.0, support for GPU-based timers, and improved interop APIs for
both Vulkan and OpenGL, and new and improved aspect ratio handling.

Finally, this release also brings with it a major change to the way HDR
and SDR content are mapped between each other, including a new
tone-mapping function based on the industry-standard ITU-R BT.2390 EETF.

Additions:
- add `pl_swapchain_hdr_metadata`, to set HDR metadata on supported
  swapchains (currently only vulkan with `VK_EXT_hdr_metadata`)
- add support for vulkan versions higher than 1.0, communicated via the
  new fields `api_version` and `max_api_version`
- add support for GPU-assisted validation and best practices layers, via
  the new field `pl_vk_inst_params.debug_extra`
- add helper functions for working with `pl_rect`s, including new aspect
  ratio handling helpers (`pl_rect2df_aspect_*`)
- add field `pl_vulkan_params.device_uuid` to allow choosing the vulkan
  device by its UUID
- add function `pl_vulkan_hold_raw`, to hold images without actually
  transitioning its layout and access mode
- add function `pl_vulkan_import`, to allow directly re-using an
  existing VkDevice rather than creating a new one; this requires
  communicating metadata about how the device was created
- add field `pl_vulkan_params.features` to allow loading optional extra
  device features at device creation time
- add support for mpv-style custom user shaders (.hook), using the set
  of functions in `<libplacebo/shaders/custom.h>`
- add `pl_render_high_quality_params`, enabling debanding and EWA
  scaling
- add `pl_timer` GPU resource type and associated API functions,
  allowing the GPU execution time of shaders and texture transfer
  operations to be measured directly
- add `PL_SHADER_SIG_SAMPLER`, allowing generated sampling shaders to
  directly accept the sampler to use as function parameters
- add `pl_image_set_chroma_location` to automatically apply the correct
  chroma location to any subsampled planes
- add `PL_TONE_MAPPING_BT_2390`, a tone mapping function based on the
  EETF from ITU-R Report BT.2390 (and make it the default)
- add `pl_peak_detect_params.overshoot_margin` to help combat clipping
  on certain types of rapid scene fade-ins
- add `pl_sampler_type` to allow encoding non-standard sampler types
  such as sampler2DRect, and also generalize samplers to allow e.g.
  usampler2D or isampler3D
- add `pl_opengl_wrap` and `pl_opengl_unwrap`, to allow directly mapping
  between OpenGL textures and the `pl_tex` abstraction

Changes:
- deprecate `pl_image.width/height`, which are now inferred
  automatically from the actual planes
- `pl_vulkan_wrap` now takes a `pl_vulkan_wrap_params` struct instead of
  directly accepting its parameters, including new fields `sample_mode`
  and `address_mode` to configure the created sampler
- change `pl_dispatch_compute` to allow optionally passing in a
  simulated framebuffer width/height, which will be used to translate
  vertex attributes (if any)
- undefine disabled `config.h` features, instead of defining them as 0
- remove debanding from `pl_render_default_params`
- refactor HDR<->SDR mapping; PL_COLOR_REF_WHITE has been removed and
  replaced by PL_COLOR_SDR_WHITE (203 cd/m^2) and PL_COLOR_SDR_WHITE_HLG
  (75% HLG), respectively
- completely refactor pl_shader_av1_grain`, which now samples directly
  from the passed texture rather than requiring the color be pre-sampled
- `pl_render_image` now infers the image primaries based on resolution,
  rather than always hard-coding `PL_COLOR_SPACE_UNKNOWN` as BT.709
- change `pl_render_target.dst_rect` from `pl_rect2d` to `pl_rect2df`,
  allowing more accurate aspect ratio handling, and correctly compensate
  for subpixel scaling ratios
- require `python3-mako` as a dependency of the `vulkan` feature
- `pl_chroma_location_offset` now treats `PL_CHROMA_UNKNOWN` as
  `PL_CHROMA_LEFT`, the de-facto standard chroma location
- the default value of `pl_color_map_params.tone_mapping_algo` is now
  `PL_TONE_MAPPING_BT_2390`

Fixes and performance improvements:
- fix shader generation when the GLSL version is explicitly overridden
- properly mark some shader failures (`pl_shader_is_failed`)
- fix texture invalidation on OpenGL
- correctly respect `pl_swapchain_frame.flipped` in
  `pl_render_target_from_swapchain`
- correctly validate descriptor uniqueness in `pl_pass_create`
- skip redundant matrix multiplication in `pl_shader_encode_color`
  wherever possible
- work around driver bugs w.r.t out-of-order buffer offsets by sorting
  all buffer variables by offset
- fix edge cases in vulkan swapchain usage flag checks
- fix excessive CPU usage in `pl_tex_download`
- reduce the number of unnecessary GPU flushes caused by `pl_buf_poll`
- fix issue where blending did not work on some drivers (e.g. nvidia)
- make the framebuffer discard check more aggressive
- fix computation of anti-aliased resizable orthogonal filters, e.g.
  when downscaling using `pl_filter_lanczos`
- fix external image memory barriers for exclusive mode images
- fix failure path of `pl_swapchain_submit_frame`
- fix various GLSL compatibility issues with av1 grain generation
- reduce maximum vulkan memory allocation slab size to conform to AMD
  recommendations
- fix build error when lcms is not available
- fix double-application of texture scale for e.g. 10-bit content when
  using separable scalers
- fix a multitude of bugs affecting av1 grain generation, especially for
  chroma planes
- fix segfault on vulkan device oom
- fix invalid shader generation on some platforms
- fix a multitude of bugs, edge cases and subtle off-by-ones related to
  chroma scaling and plane alignment
- add fallback code for edge case w.r.t chroma scaling and gpu resource
  exhaustion
- correctly load VK_KHR_swapchain in all circumstances that require
  accessing its functions
- minimize fbo usage inside `pl_renderer`, by re-using unused fbos
- tweak the work group size for polar scaling to perform better on
  modern GPUs (tested on RDNA)
- transparently upgrade fragment shaders to compute shaders on
  environments with async compute
- pick a more reasonable size for the dummy gpu's `max_group_threads`
- forbid 10-bit linear transfer functions from vulkan swapchains
- fix segfault when re-executing previously failed shaders

v2.43.1

This release backports a number of bug fixes from master affecting the
v2.43.0 release, and also slightly extends the range of backwards
compatibility with respect to vulkan-headers versions. (This change is
not on master, so this can be seen as a compatibility release)

Bug fixes:
- fix GLSL version checks when overriding GLSL versions with
  `pl_shader_params.glsl`
- fix minor order-of-operations issue in AV1 grain synthesis
- fix glInvalidate* backwards compatibility issue
- fix compilation with older vulkan-headers (e.g. ubuntu 18.04)
- fix a bug where pl_shader_encode_color sometimes uploaded and applied
  an identity matrix unnecessarily
- work around upstream issues with MSL shader compilation by ensuring
  all buffer elements are sorted by offset
- fix hypothetical edge case with swapchain image usages forbidden by
  the swapchain image format
- fix an issue where blending overlays caused the rest of the image to
  be discarded

Other changes:
- log the vulkan surface selection choices
- log the overall library version on initialization
- add a few more pl_gpu validation checks
- significantly reduce the CPU consumpation of blocking texture downloads

v2.43.0

The major highlight of this release is the introduction of a new OpenGL
backend, documented in <libplacebo/opengl.h>, which is also the reason
for the major version bump.

Apart from the addition of OpenGL support, this release brings with it a
large number of fixes for backwards compatibility with older versions of
GLSL and GLES, allowing libplacebo to be used to generate legal shaders
for virtually any GLSL version in existence.

Supporting these changes, the test framework has been expanded
considerably to allow for automated testing of older OpenGL context
versions and older GLSL versions.

Additions:
- add `pl_gpu.pci`, containing the PCI address of the underlying GPU
- add `pl_image.av1_grain`, allowing `pl_renderer` to add AV1 grain
- add `pl_opengl` backend (see <libplacebo/opengl.h>)
- add `pl_color_levels_guess`, to infer color levels from pl_color_repr
- `pl_shader_encode_color` now handles non-linear color systems
- add `pl_render_params.force_dither`, to force use of the dither code
- add `pl_render_params.disable_fbos`, to forcibly disable FBOs
- add `pl_vk_inst_params.layers`, to allow loading extra vulkan layers
- add a way to restrict the maximum GLSL version provided by the various
  `pl_gpu` backends, for testing purposes

Changes:
- load all vulkan function pointers indirectly, using a user-provided
  `vkGetInstanceProcAddr` function
- vulkan can now be built without linking against libvulkan.so
- refactor av1 grain data struct, splitting it up into "grain
  metadata" and "shader params"
- `pl_swapchain_resize` now recreates suboptimal swapchains even if the
  size does not change
- `pl_tex_blit` now requires the new cap `PL_GPU_CAP_BLITTABLE_1D_3D`
  for 1D/3D textures
- `pl_tex_params.host_readable` now requires the new cap
  `PL_FMT_CAP_HOST_READABLE` on the chosen format
- `pl_vk_inst_create` will now automatically load any layers needed for
  extensions specified by the user

Fixes and performance improvements:
- make `pl_color_adjustment.gamma` actually do something
- fix av1 grain shader for separated planes
- fix memory corruption in `pl_vulkan_swapchain_suboptimal`
- fix access mask on API writes to read-writable vulkan buffers
- fix some instances of undefined behavior in av1 grain synthesis
- fix av1 grain scale for PL_COLOR_LEVELS_UNKNOWN
- fix use of local #include for config.h in installed headers
- fix various incorrect literals for compatibility with older GLSL
- fix potential type error in XYZ input handling
- fix numerous backwards compatibility issues for older GLSL and GLES
  versions, across the board
- quench some meson warnings, as well as some compiler warnings

v1.29.1

This is a minor bug fix follow-up, to address one major and two minor
issues with the v1.29.0 release.

Fixes:
- fix a segfault in the test framework on platforms without vulkan
- fix error message spam when rendering to minimized/hidden windows
- hackily fix compilation on certain glslang versions

v1.29.0

The key highlights of this release are the change of `pl_gpu` API
semantics to use explicit erroring instead of implicit undefined
behavior, and several important bug fixes related to swapchain resizing
(in particular, handling of suboptimal swapchains); as well as other
minor utility changes, bug fixes and performance improvements.

The new `pl_gpu` API allows clients to duplicate less code related to
checking parameters and capabilities, as long as they're fine with the
resulting error messages they might receive. In particular, libplacebo
no longer aborts on invalid parameters - which helped during initial
development, but is now suffocating clients more than it's helping.

Additions:
- add a new field `pl_fmt.internal_size`, which can differ from
  `texel_size` for emulated formats
- add 16f / 64f texture formats with host-emulated 32f representation
- expose a `pl_vulkan_choose_device` helper function
- add a mechanism for allowing the user to explicitly resize suboptimal
  swapchains rather than having libplacebo take care of it

Changes:
- make `pl_gpu` log and error gracefully on `illegal` parameters rather
  than doing undefined behavior (i.e. aborting)
- creating a `host_mapped` buffer now requires PL_GPU_CAP_MAPPED_BUFFERS
- drop `priv` fields from public structs

Fixes and performance improvements:
- make libplacebo slightly less log spammy in general
- skip the creation/use of semaphores for single-queue usage
- fix several bugs related to swapchain recreation and image acquisition
- fix creation of non-renderable but storable vulkan textures
- fix several bugs related to emulated texture downloads
- fix `pl_fmt` capabilities for some emulated formats
- fix stride/alignment of std430 vectors/matrices
- use push constants more aggressively, whenever possible
- slightly improve accurate of BT.2100 matrices
- fix overzealous rotation of vulkan queues
- fix dangling validation layer suppression in rare cases
- fix possible swapchain deadlock on VK_SUBOPTIMAL_KHR
- handle VK_SUBOPTIMAL_KHR more gracefully / efficiently

v1.21.0

This is a minor / house-keeping release, mostly to get some important
fixes into distros while also getting support for iGPUs requiring
non-coherent memory into a "stable" release.

Additions:
- `shaders.h` API users can now override the GLSL version
- add support for vulkan platforms with non-coherent memory (e.g. iGPUs)

Changes:
- default GLSL version is now 130 unless otherwise specified
- signature of `pl_shader_alloc` now uses a params struct
- PL_FIX_VER / PL_VERSION are no longer header constants
- the renderer will now automatically adjust deband grain scale for HDR

Fixes and performance improvements:
- on GLSL < 130, shaders now correctly use `texture2D` etc.
- fix a bug where async transfer was never actually used, due to
  incorrect alignment checks
- fixed some issues that could lead to symbol conflicts
- don't require large vertex buffers be host-visible (usually impossible)
- correctly detect swapchain image feature flags
- turned a few 1ms busy loops into infinite waits
- fix compilation on some platforms that require SPIRV-Tools
- fix vkCreateEvent error spam on MoltenVK
- bump queued command limit from 64 to 1024

v1.18.0

The major changes of this release include a new and completely overhauled
HDR tone mapping algorithm, support for importing backing memory for
textures and buffers (such as via dmabuf fds), as well as a swapchain
resizing API that, among other benefits, allows libplacebo to function
properly on Wayland.

There is also now support for generating CPU-backed "dummy resources"
for things like lookup tables or convolution kernels, for users who
prefer managing their own GPU resources. This allows `pl_shader`-level
API users to access the full host of libplacebo shaders without being
restricted the lack of a `pl_gpu`.

Additions:
- `pl_color_space` now has a `sig_scale` field, which allows modifying
   the signal range without affecting the transfer function's shape.
   (Effectively, by "stretching" it). This can be useful when dealing
   with HDR material encoded in a non-HDR curve such as linear light.
- Add support for importing handles (including `pl_sync` objects and
  `pl_shared_mem`). Currently, only memory can be imported.
- Add partial support for dmabuf fd interop. Currently does not support
  dma layout modifiers, so the usefulness of this feature is limited.
- Allow importing external memory when creating textures.
- Add a way to overexpose HDR material to bring its brightness up to
  more easily viewable levels.
- Add `pl_var_int` to go alongside the other `gpu.h` helpers.
- Add a way to create CPU-backed "dummy" instances of `pl_gpu`. These
  are not capable of compiling or executing any shaders, so they're only
  useful for users who want to generate and compile their own shaders.
  All GPU resources (textures, buffers) are backed by `malloc`.
- Add a new field to `pl_tex/buf_params` allowing users to attach
  arbitrary references (void*) that libplacebo will ignore.
- Add a function for changing the `pl_context_params` after creation.
- Add a new API function for dealing with swapchain resize events. This
  function can be used to both query and update the swapchain size, based
  on the usage. Not all platforms support all operations.
- Add new demo/skeleton based on GLFW.

Changes:
- Remove the `hdr_simulation` field from `pl_color_map_params`.
  Use `signal_scale` on the target color space instead.
- Subdivide handle capabilities based on object type (texture vs
   buffer).
- Completely redesign the HDR tone mapping algorithm. Most members of
  `pl_tone_mapping_params` have been updated, in particular the options
  related to desaturation and peak detection.
- Split up the HDR peak detection API into a "detection" and
  "application" stage, which can run independently. This avoids the
  one-frame delay inherent with the previous API, and thus allows
  tone mapping to be frame perfect.
- Remove superfluous field from `pl_shader_reset` left behind from a
  previous version of the API.

Fixes and performance improvements:
- Correctly apply the OOTF before performing linear colorspace
  operations on scene-referred content.
- Avoid clipping when tone-mapping out of gamut content by tone mapping
  before gamut mapping.
- Correctly reset the peak detection buffer state when running
  `pl_renderer_flush_cache`.
- Correctly use the image's original color space as the reference space
  when the content is prelinearized by the renderer. Fixes some cases
  where libplacebo could end up defaulting `PL_COLOR_TRC_AUTO` to a
  suboptimal curve.
- Correctly track the lifetime and ownership of individual vulkan buffer
  objects. Fixes some synchronization issues when attempting to use the
  same buffer from multiple different queue families.
- Fix integer overflow on implementations with very large texture size
  limits.
- Fix building against static libshaderc.

v1.7.0

This release marks the first major release of libplacebo, in tune with
the release of VLC 4, which will be the first major project using it.
Apart from API stability going forwards, this release brings with it a
new AV1 film grain shader, better interoperability between libplacebo
and external APIs like CUDA (via shared buffers and shared textures),
and ICtCp support.

While not strictly part of libplacebo, one of the highlights since the
previous release includes the existence of a new example file
`demos/video-filtering.c` which illustrates how one would use libplacebo
to do GPU-based image filtering in something like FFmpeg or mpv.

This release also marks our move from GitHub to VideoLAN's GitLab. May
they provide a home for us for a long time going forwards.

Additions:
- Add a new function `pl_gpu_finish` which blocks until all oustanding
  rendering on this `pl_gpu` is finished.
- Add new functions `pl_tex_recreate` and `pl_buf_recreate`, which work
  like `pl_tex/buf_create` but take a pointer to an existing tex/buf
  that will get destroyed + recreated only when necessary.
- Add a new function `pl_shader_is_failed` which will return true if a
  given shader is in a "failed" state. Shaders will be marked as failed
  on any internal/usage error, rather than them being silently ignored.
- Add a new enum `pl_channel` to clarify and encode friendly names for
  the often-referenced "canonical channel order".
- Add a new header `libplacebo/shaders/av1.h` which currently contains
  a function `pl_shader_av1_grain` for applying AV1 film grain on the
  GPU.
- Add a new concept of an "exportable" object (buffers and textures).
  Exportable objects can be exported using a handle and imported into
  other foreign APIs such as CUDA. The new functions `pl_buf_export` and
  `pl_tex_export` must be used to correctly synchronize access to the
  object. This also adds new fields `uuid` and `handle_caps` to
  `pl_gpu`.
- Supporting the previous feature, add a new field `memory_type` to
  `pl_buf_params` which can be used to influence what type of memory
  to allocate a buffer from. Currently only works for texture transfer
  buffers, since allocating uniform/storage buffers from non-VRAM makes
  little sense.
- Add a new synchronization primitive wrapper, `pl_sync`, which wraps a
  semaphore pair and must be used to synchronize access to textures with
  external, asynchronous APIs.
- Implement the ITU-R BT.2100 ICtCp color system. Since the libplacebo
  color systems are not strictly tied to any particular transfer
  function, we must explicitly mark which flavor of ICtCp is meant.
- Add a new field `instance_params` which can be used to influence the
  parameters used when `pl_vulkan_create` ends up creating an internal
  instance.
- Add a new function `pl_vulkan_unwrap` which allows users to unwrap a
  vulkan-baed `pl_tex` to expose the internal VkImage, allowing
  simultaneous use (via `pl_vulkan_hold/release`) similar to wrapped
  external images.
- Add new generic helper functions `pl_std430_layout` and
  `pl_std140_layout` which replace the old `pl_buf_uniform_layout`,
  `pl_buf_storage_layout` and `pl_push_constant_layout`.

Changes:
- Empty device names ("") can now be passed to `pl_vulkan_create`.
  They will be treated as if NULL was passed.
- The `out_plane` parameter of `pl_upload_plane` is now optional.
- Clarify/Relax the restrictions on `pl_buf` usage and polling. Users
  are technically free to use `pl_buf` for multiple simultaneous
  libplacebo operations. Buffer polling is only needed for accesses by the
  host.
- `pl_vulkan_hold` now returns a bool indicating success.
- `pl_buffer_var` has been moved from gpu.h's `pl_desc` to shaders.h's
  `pl_shader_desc`. Describing the individual variables of a descriptor
  binding had zero practical application.
- `pl_buf_uniform_layout`, `pl_buf_storage_layout` and
  `pl_push_constant_layout` are now macros for `pl_std140_layout` and
  `pl_std430_layout` (respectively). This changed the signature to drop
  the `pl_gpu` parameter.
- The `buf_offset` parameter to `pl_tex_transfer` no longer needs to be
  strictly aligned to a multiple of 4. The minimum alignment is now 1,
  however users are strongly recommended to stick to the multiple-of-4
  alignment (or ideally `align_tex_xfer_offset`) for performance
  reasons.
- The chromatic adaptation method in `pl_get_color_mapping_matrix` has
  been adjusted. We now use an LMS model derived from CIECAM97's revised
  linear Bradford matrix, rather than the non-linear matrix that was
  being used previously (incorrectly so, due to the lack of the required
  nonlinearity).
- The order of fields in `pl_rect3d` has been changed for consistency
  with the other rect types.

Fixes and performance improvements:
- Meson 0.47 is correctly marked as the minimum required version.
- Fix compilation on clang.
- Fix compilation on glslang git master.
- Fix compilation with older shaderc versions.
- Fix compilation on some platforms.
- Fix std140/std430 packing errors for vec3.
- Skip unnecessary flush in pl_buf_poll noop cases.
- Fix variable collision in sh_prng.
- Don't leak glslang internal symbols on supported platforms.
- Fix an issue where `pl_pass_run` was stricter than intended about
  compatibility with between `target` and `target_dummy`.
- Fix an issue where `pl_dispatch` could sometimes try dispatching
  shaders with an incompatible target.
- Fix an error in the heuristic for choosing the optimal image layout
  for vulkan render passes.
- Improved debugging messages in several places.
- Slightly speed up lookups from texture LUTs.
- Fix the addressing of shader LUTs in some hypothetical cases.
- Correctly flush the contents of host-readable buffers after
  modifications made by the GPU.
- Fix synchronization on `pl_buf_write` with non-mapped buffers.
- Fix undefined behavior when using push descriptors.
- Fix build issues on Android arm32.
- Slightly speed up some texture recreate operations by invalidating
  re-used textures.
- Fix an issue when trying to update large (>64k) VRAM-resident buffers.
- Fix two address calculation bugs in `pl_tex_blit`.
- Fix an over-read bug when the size of the vertex data changed for
  otherwise identical passes.
- Fix a misalignment that could theoretically happen with some
  combinations of (odd) texel sizes and device alignment requirements.
- Fix UB when creating "useless" images (without any usage flags).
- Fix a vulkan device memory leak when destroying large textures.
- Speed up compilation by skipping glslang when shaderc is available.
- Fix an alignment issue that could happen sometimes with
  `pl_buf_write` for odd write sizes.
- Fix an alignment bug when uploading partial textures when async
  transfer is enabled on some devices.
- Fix crash in `pl_color_primaries_is_wide_gamut` on DISPLAY_P3.
- Fix an error when re-using shader objects between polar and non-polar
  samplers. This is now safe to do.

v1.7.0-rc4

This release marks the first major release of libplacebo, in tune with
the release of VLC 4, which will be the first major project using it.
Apart from API stability going forwards, this release brings with it a
new AV1 film grain shader, better interoperability between libplacebo
and external APIs like CUDA (via shared buffers and shared textures),
and ICtCp support.

While not strictly part of libplacebo, one of the highlights since the
previous release includes the existence of a new example file
`demos/video-filtering.c` which illustrates how one would use libplacebo
to do GPU-based image filtering in something like FFmpeg or mpv.

This release also marks our move from GitHub to VideoLAN's GitLab. May
they provide a home for us for a long time going forwards.

Additions:
- Add a new function `pl_gpu_finish` which blocks until all oustanding
  rendering on this `pl_gpu` is finished.
- Add new functions `pl_tex_recreate` and `pl_buf_recreate`, which work
  like `pl_tex/buf_create` but take a pointer to an existing tex/buf
  that will get destroyed + recreated only when necessary.
- Add a new function `pl_shader_is_failed` which will return true if a
  given shader is in a "failed" state. Shaders will be marked as failed
  on any internal/usage error, rather than them being silently ignored.
- Add a new enum `pl_channel` to clarify and encode friendly names for
  the often-referenced "canonical channel order".
- Add a new header `libplacebo/shaders/av1.h` which currently contains
  a function `pl_shader_av1_grain` for applying AV1 film grain on the
  GPU.
- Add a new concept of an "exportable" object (buffers and textures).
  Exportable objects can be exported using a handle and imported into
  other foreign APIs such as CUDA. The new functions `pl_buf_export` and
  `pl_tex_export` must be used to correctly synchronize access to the
  object. This also adds new fields `uuid` and `handle_caps` to
  `pl_gpu`.
- Supporting the previous feature, add a new field `memory_type` to
  `pl_buf_params` which can be used to influence what type of memory
  to allocate a buffer from. Currently only works for texture transfer
  buffers, since allocating uniform/storage buffers from non-VRAM makes
  little sense.
- Add a new synchronization primitive wrapper, `pl_sync`, which wraps a
  semaphore pair and must be used to synchronize access to textures with
  external, asynchronous APIs.
- Implement the ITU-R BT.2100 ICtCp color system. Since the libplacebo
  color systems are not strictly tied to any particular transfer
  function, we must explicitly mark which flavor of ICtCp is meant.
- Add a new field `instance_params` which can be used to influence the
  parameters used when `pl_vulkan_create` ends up creating an internal
  instance.
- Add a new function `pl_vulkan_unwrap` which allows users to unwrap a
  vulkan-baed `pl_tex` to expose the internal VkImage, allowing
  simultaneous use (via `pl_vulkan_hold/release`) similar to wrapped
  external images.
- Add new generic helper functions `pl_std430_layout` and
  `pl_std140_layout` which replace the old `pl_buf_uniform_layout`,
  `pl_buf_storage_layout` and `pl_push_constant_layout`.

Changes:
- Empty device names ("") can now be passed to `pl_vulkan_create`.
  They will be treated as if NULL was passed.
- The `out_plane` parameter of `pl_upload_plane` is now optional.
- Clarify/Relax the restrictions on `pl_buf` usage and polling. Users
  are technically free to use `pl_buf` for multiple simultaneous
  libplacebo operations. Buffer polling is only needed for accesses by the
  host.
- `pl_vulkan_hold` now returns a bool indicating success.
- `pl_buffer_var` has been moved from gpu.h's `pl_desc` to shaders.h's
  `pl_shader_desc`. Describing the individual variables of a descriptor
  binding had zero practical application.
- `pl_buf_uniform_layout`, `pl_buf_storage_layout` and
  `pl_push_constant_layout` are now macros for `pl_std140_layout` and
  `pl_std430_layout` (respectively). This changed the signature to drop
  the `pl_gpu` parameter.
- The `buf_offset` parameter to `pl_tex_transfer` no longer needs to be
  strictly aligned to a multiple of 4. The minimum alignment is now 1,
  however users are strongly recommended to stick to the multiple-of-4
  alignment (or ideally `align_tex_xfer_offset`) for performance
  reasons.
- The chromatic adaptation method in `pl_get_color_mapping_matrix` has
  been adjusted. We now use an LMS model derived from CIECAM97's revised
  linear Bradford matrix, rather than the non-linear matrix that was
  being used previously (incorrectly so, due to the lack of the required
  nonlinearity).
- The order of fields in `pl_rect3d` has been changed for consistency
  with the other rect types.

Fixes and performance improvements:
- Meson 0.47 is correctly marked as the minimum required version.
- Fix compilation on clang.
- Fix compilation on glslang git master.
- Fix compilation with older shaderc versions.
- Fix compilation on some platforms.
- Fix std140/std430 packing errors for vec3.
- Skip unnecessary flush in pl_buf_poll noop cases.
- Fix variable collision in sh_prng.
- Don't leak glslang internal symbols on supported platforms.
- Fix an issue where `pl_pass_run` was stricter than intended about
  compatibility with between `target` and `target_dummy`.
- Fix an issue where `pl_dispatch` could sometimes try dispatching
  shaders with an incompatible target.
- Fix an error in the heuristic for choosing the optimal image layout
  for vulkan render passes.
- Improved debugging messages in several places.
- Slightly speed up lookups from texture LUTs.
- Fix the addressing of shader LUTs in some hypothetical cases.
- Correctly flush the contents of host-readable buffers after
  modifications made by the GPU.
- Fix synchronization on `pl_buf_write` with non-mapped buffers.
- Fix undefined behavior when using push descriptors.
- Fix build issues on Android arm32.
- Slightly speed up some texture recreate operations by invalidating
  re-used textures.
- Fix an issue when trying to update large (>64k) VRAM-resident buffers.
- Fix two address calculation bugs in `pl_tex_blit`.
- Fix an over-read bug when the size of the vertex data changed for
  otherwise identical passes.
- Fix a misalignment that could theoretically happen with some
  combinations of (odd) texel sizes and device alignment requirements.
- Fix UB when creating "useless" images (without any usage flags).
- Fix a vulkan device memory leak when destroying large textures.
- Speed up compilation by skipping glslang when shaderc is available.
- Fix an alignment issue that could happen sometimes with
  `pl_buf_write` for odd write sizes.
- Fix an alignment bug when uploading partial textures when async
  transfer is enabled on some devices.
- Fix crash in `pl_color_primaries_is_wide_gamut` on DISPLAY_P3.
- Fix an error when re-using shader objects between polar and non-polar
  samplers. This is now safe to do.

v1.7.0-rc3

This release marks the first major release of libplacebo, in tune with
the release of VLC 4, which will be the first major project using it.
Apart from API stability going forwards, this release brings with it a
new AV1 film grain shader, better interoperability between libplacebo
and external APIs like CUDA (via shared buffers and shared textures),
and ICtCp support.

While not strictly part of libplacebo, one of the highlights since the
previous release includes the existence of a new example file
`demos/video-filtering.c` which illustrates how one would use libplacebo
to do GPU-based image filtering in something like FFmpeg or mpv.

This release also marks our move from GitHub to VideoLAN's GitLab. May
they provide a home for us for a long time going forwards.

Additions:
- Add a new function `pl_gpu_finish` which blocks until all oustanding
  rendering on this `pl_gpu` is finished.
- Add new functions `pl_tex_recreate` and `pl_buf_recreate`, which work
  like `pl_tex/buf_create` but take a pointer to an existing tex/buf
  that will get destroyed + recreated only when necessary.
- Add a new function `pl_shader_is_failed` which will return true if a
  given shader is in a "failed" state. Shaders will be marked as failed
  on any internal/usage error, rather than them being silently ignored.
- Add a new enum `pl_channel` to clarify and encode friendly names for
  the often-referenced "canonical channel order".
- Add a new header `libplacebo/shaders/av1.h` which currently contains
  a function `pl_shader_av1_grain` for applying AV1 film grain on the
  GPU.
- Add a new concept of an "exportable" object (buffers and textures).
  Exportable objects can be exported using a handle and imported into
  other foreign APIs such as CUDA. The new functions `pl_buf_export` and
  `pl_tex_export` must be used to correctly synchronize access to the
  object. This also adds new fields `uuid` and `handle_caps` to
  `pl_gpu`.
- Supporting the previous feature, add a new field `memory_type` to
  `pl_buf_params` which can be used to influence what type of memory
  to allocate a buffer from. Currently only works for texture transfer
  buffers, since allocating uniform/storage buffers from non-VRAM makes
  little sense.
- Add a new synchronization primitive wrapper, `pl_sync`, which wraps a
  semaphore pair and must be used to synchronize access to textures with
  external, asynchronous APIs.
- Implement the ITU-R BT.2100 ICtCp color system. Since the libplacebo
  color systems are not strictly tied to any particular transfer
  function, we must explicitly mark which flavor of ICtCp is meant.
- Add a new field `instance_params` which can be used to influence the
  parameters used when `pl_vulkan_create` ends up creating an internal
  instance.
- Add a new function `pl_vulkan_unwrap` which allows users to unwrap a
  vulkan-baed `pl_tex` to expose the internal VkImage, allowing
  simultaneous use (via `pl_vulkan_hold/release`) similar to wrapped
  external images.
- Add new generic helper functions `pl_std430_layout` and
  `pl_std140_layout` which replace the old `pl_buf_uniform_layout`,
  `pl_buf_storage_layout` and `pl_push_constant_layout`.

Changes:
- Empty device names ("") can now be passed to `pl_vulkan_create`.
  They will be treated as if NULL was passed.
- The `out_plane` parameter of `pl_upload_plane` is now optional.
- Clarify/Relax the restrictions on `pl_buf` usage and polling. Users
  are technically free to use `pl_buf` for multiple simultaneous
  libplacebo operations. Buffer polling is only needed for accesses by the
  host.
- `pl_vulkan_hold` now returns a bool indicating success.
- `pl_buffer_var` has been moved from gpu.h's `pl_desc` to shaders.h's
  `pl_shader_desc`. Describing the individual variables of a descriptor
  binding had zero practical application.
- `pl_buf_uniform_layout`, `pl_buf_storage_layout` and
  `pl_push_constant_layout` are now macros for `pl_std140_layout` and
  `pl_std430_layout` (respectively). This changed the signature to drop
  the `pl_gpu` parameter.
- The `buf_offset` parameter to `pl_tex_transfer` no longer needs to be
  strictly aligned to a multiple of 4. The minimum alignment is now 1,
  however users are strongly recommended to stick to the multiple-of-4
  alignment (or ideally `align_tex_xfer_offset`) for performance
  reasons.
- The chromatic adaptation method in `pl_get_color_mapping_matrix` has
  been adjusted. We now use an LMS model derived from CIECAM97's revised
  linear Bradford matrix, rather than the non-linear matrix that was
  being used previously (incorrectly so, due to the lack of the required
  nonlinearity).
- The order of fields in `pl_rect3d` has been changed for consistency
  with the other rect types.

Fixes and performance improvements:
- Meson 0.47 is correctly marked as the minimum required version.
- Fix compilation on clang.
- Fix compilation on glslang git master.
- Fix compilation with older shaderc versions.
- Fix compilation on some platforms.
- Fix std140/std430 packing errors for vec3.
- Skip unnecessary flush in pl_buf_poll noop cases.
- Fix variable collision in sh_prng.
- Don't leak glslang internal symbols on supported platforms.
- Fix an issue where `pl_pass_run` was stricter than intended about
  compatibility with between `target` and `target_dummy`.
- Fix an issue where `pl_dispatch` could sometimes try dispatching
  shaders with an incompatible target.
- Fix an error in the heuristic for choosing the optimal image layout
  for vulkan render passes.
- Improved debugging messages in several places.
- Slightly speed up lookups from texture LUTs.
- Fix the addressing of shader LUTs in some hypothetical cases.
- Correctly flush the contents of host-readable buffers after
  modifications made by the GPU.
- Fix synchronization on `pl_buf_write` with non-mapped buffers.
- Fix undefined behavior when using push descriptors.
- Fix build issues on Android arm32.
- Slightly speed up some texture recreate operations by invalidating
  re-used textures.
- Fix an issue when trying to update large (>64k) VRAM-resident buffers.
- Fix two address calculation bugs in `pl_tex_blit`.
- Fix an over-read bug when the size of the vertex data changed for
  otherwise identical passes.
- Fix a misalignment that could theoretically happen with some
  combinations of (odd) texel sizes and device alignment requirements.
- Fix UB when creating "useless" images (without any usage flags).
- Fix a vulkan device memory leak when destroying large textures.
- Speed up compilation by skipping glslang when shaderc is available.
- Fix an alignment issue that could happen sometimes with
  `pl_buf_write` for odd write sizes.
- Fix an alignment bug when uploading partial textures when async
  transfer is enabled on some devices.
- Fix crash in `pl_color_primaries_is_wide_gamut` on DISPLAY_P3.
- Fix an error when re-using shader objects between polar and non-polar
  samplers. This is now safe to do.

v1.7.0-rc2

This release marks the first major release of libplacebo, in tune with
the release of VLC 4, which will be the first major project using it.
Apart from API stability going forwards, this release brings with it a
new AV1 film grain shader, better interoperability between libplacebo
and external APIs like CUDA (via shared buffers and shared textures),
and ICtCp support.

While not strictly part of libplacebo, one of the highlights since the
previous release includes the existence of a new example file
`demos/video-filtering.c` which illustrates how one would use libplacebo
to do GPU-based image filtering in something like FFmpeg or mpv.

Additions:
- Add a new function `pl_gpu_finish` which blocks until all oustanding
  rendering on this `pl_gpu` is finished.
- Add new functions `pl_tex_recreate` and `pl_buf_recreate`, which work
  like `pl_tex/buf_create` but take a pointer to an existing tex/buf
  that will get destroyed + recreated only when necessary.
- Add a new function `pl_shader_is_failed` which will return true if a
  given shader is in a "failed" state. Shaders will be marked as failed
  on any internal/usage error, rather than them being silently ignored.
- Add a new enum `pl_channel` to clarify and encode friendly names for
  the often-referenced "canonical channel order".
- Add a new header `libplacebo/shaders/av1.h` which currently contains
  a function `pl_shader_av1_grain` for applying AV1 film grain on the
  GPU.
- Add a new concept of an "exportable" object (buffers and textures).
  Exportable objects can be exported using a handle (currently only
  supporting fds on POSIX-style platforms) and imported into other
  foreign APIs such as CUDA. The new functions `pl_buf_export` and
  `pl_tex_export` must be used to correctly synchronize access to the
  object. This also adds new fields `uuid` and `handle_caps` to
  `pl_gpu`.
- Supporting the previous feature, add a new field `memory_type` to
  `pl_buf_params` which can be used to influence what type of memory
  to allocate a buffer from. Currently only works for texture transfer
  buffers, since allocating uniform/storage buffers from non-VRAM makes
  little sense.
- Add a new synchronization primitive wrapper, `pl_sync`, which wraps a
  semaphore pair and must be used to synchronize access to textures with
  external, asynchronous APIs.
- Implement the ITU-R BT.2100 ICtCp color system. Since the libplacebo
  color systems are not strictly tied to any particular transfer
  function, we must explicitly mark which flavor of ICtCp is meant.
- Add a new field `instance_params` which can be used to influence the
  parameters used when `pl_vulkan_create` ends up creating an internal
  instance.
- Add a new function `pl_vulkan_unwrap` which allows users to unwrap a
  vulkan-baed `pl_tex` to expose the internal VkImage, allowing
  simultaneous use (via `pl_vulkan_hold/release`) similar to wrapped
  external images.
- Add new generic helper functions `pl_std430_layout` and
  `pl_std140_layout` which replace the old `pl_buf_uniform_layout`,
  `pl_buf_storage_layout` and `pl_push_constant_layout`.

Changes:
- Empty device names ("") can now be passed to `pl_vulkan_create`.
  They will be treated as if NULL was passed.
- The `out_plane` parameter of `pl_upload_plane` is now optional.
- Clarify/Relax the restrictions on `pl_buf` usage and polling. Users
  are technically free to use `pl_buf` for multiple simultaneous
  libplacebo operations. Buffer polling is only needed for accesses by the
  host.
- `pl_vulkan_hold` now returns a bool indicating success.
- `pl_buffer_var` has been moved from gpu.h's `pl_desc` to shaders.h's
  `pl_shader_desc`. Describing the individual variables of a descriptor
  binding had zero practical application.
- `pl_buf_uniform_layout`, `pl_buf_storage_layout` and
  `pl_push_constant_layout` are now macros for `pl_std140_layout` and
  `pl_std430_layout` (respectively). This changed the signature to drop
  the `pl_gpu` parameter.

Fixes and performance improvements:
- Meson 0.47 is correctly marked as the minimum required version.
- Fix compilation on clang.
- Fix compilation on glslang git master.
- Fix compilation with older shaderc versions.
- Fix compilation on some platforms.
- Fix std140/std430 packing errors for vec3.
- Skip unnecessary flush in pl_buf_poll noop cases.
- Fix variable collision in sh_prng.
- Don't leak glslang internal symbols on supported platforms.
- Fix an issue where `pl_pass_run` was stricter than intended about
  compatibility with between `target` and `target_dummy`.
- Fix an issue where `pl_dispatch` could sometimes try dispatching
  shaders with an incompatible target.
- Fix an error in the heuristic for choosing the optimal image layout
  for vulkan render passes.
- Improved debugging messages in several places.
- Slightly speed up lookups from texture LUTs.
- Fix the addressing of shader LUTs in some hypothetical cases.
- Correctly flush the contents of host-readable buffers after
  modifications made by the GPU.
- Fix synchronization on `pl_buf_write` with non-mapped buffers.
- Fix undefined behavior when using push descriptors.
- Fix build issues on Android arm32.
- Slightly speed up some texture recreate operations by invalidating
  re-used textures.
- Fix an issue when trying to update large (>64k) VRAM-resident buffers.
- Fix two address calculation bugs in `pl_tex_blit`.
- Fix an over-read bug when the size of the vertex data changed for
  otherwise identical passes.
- Fix a misalignment that could theoretically happen with some
  combinations of (odd) texel sizes and device alignment requirements.
- Fix UB when creating "useless" images (without any usage flags).
- Fix a vulkan device memory leak when destroying large textures.
- Speed up compilation by skipping glslang when shaderc is available.

v1.7.0-rc1

This release marks the first major release of libplacebo, in tune with
the release of VLC 4, which will be the first major project using it.
Apart from API stability going forwards, this release brings with it a
new AV1 film grain shader, better interoperability between libplacebo
and external APIs like CUDA (via shared buffers and shared textures),
and ICtCp support.

While not strictly part of libplacebo, one of the highlights since the
previous release includes the existence of a new example file
`demos/video-filtering.c` which illustrates how one would use libplacebo
to do GPU-based image filtering in something like FFmpeg or mpv.

Additions:
- Add a new function `pl_gpu_finish` which blocks until all oustanding
  rendering on this `pl_gpu` is finished.
- Add new functions `pl_tex_recreate` and `pl_buf_recreate`, which work
  like `pl_tex/buf_create` but take a pointer to an existing tex/buf
  that will get destroyed + recreated only when necessary.
- Add a new function `pl_shader_is_failed` which will return true if a
  given shader is in a "failed" state. Shaders will be marked as failed
  on any internal/usage error, rather than them being silently ignored.
- Add a new enum `pl_channel` to clarify and encode friendly names for
  the often-referenced "canonical channel order".
- Add a new header `libplacebo/shaders/av1.h` which currently contains
  a function `pl_shader_av1_grain` for applying AV1 film grain on the
  GPU.
- Add a new concept of an "exportable" buffer. Exportable buffers can be
  exported using a handle (e.g. fd) and imported into other foreign APIs
  such as CUDA. The new function `pl_buf_export` must be used to
  correctly synchronize access to the buffer.
- Supporting the previous feature, add a new field `memory_type` to
  `pl_buf_params` which can be used to influence what type of memory
  to allocate a buffer from. Currently only works for texture transfer
  buffers, since allocating uniform/storage buffers from non-VRAM makes
  little sense.
- Implement the ITU-R BT.2100 ICtCp color system. Since the libplacebo
  color systems are not strictly tied to any particular transfer
  function, we must explicitly mark which flavor of ICtCp is meant.
- Add a new field `instance_params` which can be used to influence the
  parameters used when `pl_vulkan_create` ends up creating an internal
  instance.
- Add a new function `pl_vulkan_unwrap` which allows users to unwrap a
  vulkan-baed `pl_tex` to expose the internal VkImage, allowing
  simultaneous use (via `pl_vulkan_hold/release`) similar to wrapped
  external images.

Changes:
- Empty device names ("") can now be passed to `pl_vulkan_create`.
  They will be treated as if NULL was passed.
- The `out_plane` parameter of `pl_upload_plane` is now optional.
- Clarify/Relax the restrictions on `pl_buf` usage and polling. Users
  are technically free to use `pl_buf` for multiple simultaneous
  libplacebo operations. Buffer polling is only needed for accesses by the
  host.
- `pl_vulkan_hold` now returns a bool indicating success.

Fixes and performance improvements:
- Meson 0.47 is correctly marked as the minimum required version.
- Fix compilation on clang.
- Fix compilation on glslang git master.
- Fix std140/std430 packing errors for vec3.
- Skip unnecessary flush in pl_buf_poll noop cases.
- Fix variable collision in sh_prng.
- Don't leak glslang internal symbols on supported platforms.
- Fix an issue where `pl_pass_run` was stricter than intended about
  compatibility with between `target` and `target_dummy`.
- Fix an issue where `pl_dispatch` could sometimes try dispatching
  shaders with an incompatible target.
- Fix an error in the heuristic for choosing the optimal image layout
  for vulkan render passes.
- Improved debugging messages in several places.
- Slightly speed up lookups from texture LUTs.
- Fix the addressing of shader LUTs in some hypothetical cases.
- Correctly flush the contents of host-readable buffers after
  modifications made by the GPU.
- Fix synchronization on `pl_buf_write` with non-mapped buffers.
- Fix undefined behavior when using push descriptors.
- Fix build issues on Android arm32.
- Slightly speed up some texture recreate operations by invalidating
  re-used textures.
- Fix an issue when trying to update large (>64k) VRAM-resident buffers.
- Fix two address calculation bugs in `pl_tex_blit`.

v0.6.0

This release brings with it a number of minor fixes and changes, as well
as a new feature for simulating, and partially counteracting, various
forms of color blindness. For more information see the new
`pl_cone_params` structure and `pl_get_cone_matrix` function.

Additions:
- libplacebo can now be built against glslang directly instead of only
  via shaderc. Using the latter is still preferred as it allows us to
  get better diagnostics, but this should make it easier to port to
  platforms like android.
- VK_COMPOSITE_ALPHA_INHERIT_BIT_KHR is now supported, and gets treated
  as PL_ALPHA_UNKNOWN.
- Add a new feature for simulating various forms of color blindness
  (based on the LMS color model).

Changes:
- The build system now uses meson's built in tristate option syntax
  instead of doing it by hand. This requires using
  -Dfeature=enabled|disabled|auto instead of the old =true|false.
- Setting pl_render_params.upscaler/downscaler to NULL no longer allows
  automatically choosing bicubic, since this breaks various other
  optimizations that are only possible when using the built-in GPU
  sampling. This now only chooses bilinear/nearest depending on the
  texture/hardware capabilities.
- The public API of pl_shader_alloc was changed: the `uint8_t ident`
  field was removed. This makes no real sense to expose publicly, since
  users had no legitimate use for it other than setting it to `0`.

Fixes and performance improvements:
- The pl_dither_params.lut_size now gets correctly defaulted to 6.
- Fixed some use-after-scope bugs in the renderer.
- Fixed colorspace selection when using user-specified swapchain
  formats.
- Fixed a build issue on platforms without VK_EXT_swapchain_colorspace
- Peak detection is now correctly disabled if the buffer size is set to
  0. (Before, it would default back to 63 in that case)
- Fix the out-of-gamut warning feature. This previously stopped working
  for out-of-range tone mapped values.

v0.5.0

This release brings with it greatly improved support for HDR (including
HDR-in-SDR emulation, native HDR passthrough and improvements to the
HDR->SDR tone mapping algorithm), as well as support for ICC profiles
and external VkImage interop (needed for e.g. hardware decoding).

Additions:
- Add support for treating an SDR curve as an HDR curve by
  overriding/providing the true target peak. This essentially allows you
  to calibrate a true HDR device to a typical SDR curve and use it to
  display HDR media, at the cost of bit precision
- Add more values to enum pl_chroma_location
- Add a scale factor to struct pl_sample_src, which is useful for e.g.
  debanding on bit-shifted inputs
- Add a new function to infer/guess unknown pl_color_space fields
- Add support for 3DLUTs / ICC profiles (via LittleCMS), including
  support for assigning ICC profiles to the `pl_render_image` and
  `pl_render_target` structs
- Allow enabling optional extra vulkan device/instance extensions, as
  well as listing what extensions were actually loaded
- Add an API for VkImage interop with user code, e.g. to wrap mapped
  hardware decoding planes into `pl_tex`, plus all synchronization code
  and extra metadata necessary to make this interop safe
- Implement support for native HDR output on supported platforms

Changes:
- Separate the texture from the plane data struct in `pl_upload_plane`
- Swapchain FBOs are now guaranteed to be blittable (e.g. pl_tex_clear)
- Tone mapping is no longer done for desaturation steps. Users wanting
  high quality desaturation should use ICC profiles
- Change the types of various fields in struct pl_gpu_limits
- Slightly relax the sizing requirements of pl_tex_transfer
- Change the default value for pl_tex_transfer_params.stride_w/h.
  Instead of defaulting to the image size, they now default to the rect to
  be transferred
- Texture transfer buffer offsets must (now) be aligned to the texel
  size, in addition to being a multiple of 4
- Change the API of `pl_color_space_equal` to be consistent with the
  rest of the colorspace APIs
- Allow the `pl_upload_plane` helpers to upload from buffers as well
- Switch the type of mapped buffer pointers from char* to uint8_t*
- Rename DCI_P3 to Display_P3, and add DCI_P3 (using the actual DCI white
  point)

Fixes and performance improvements:
- Fix some compatibility issues with GLSL 120
- Prevent desaturation from exploding when tone mapping to HDR, or when
  tone mapping a very bright HDR scene to SDR
- Implement the fully parametrized BT.2100 (HLG) OOTF in both
  directions, based on the tagged/inferred signal peak
- Fix an issue preventing the out-of-gamut feature from working
  correctly on HDR displays
- Fix a segfault that occurs when instance creation fails
- Fix an issue where `peak_detect_frames` was sometimes too low
- Allow debanding to take care of implicit bilinear sampling in cases
  where no other scaler is requested, thus avoiding an unnecessary FBO
  indirection
- Only run the debanding algorithm on the parts of the image we will
  actually be displaying. This is relevant when cropping
- Fix an issue where PL_TONE_MAPPINg_MOBIUS could fail in certain
  (unrealistic) scenarios involving HDR-in-SDR emulation
- Fix an issue where vulkan could sometimes modify commands after
  submitting them, when recreating vertex buffers
- Make an effort to reuse the same vertex buffer when the contents don't
  change from the previous invocation
- Fix the glslang locale work-arounds on windows
- Fix an issue where trying to upload from a too-large texel buffer
  results in an assertion failure rather than a clean error
- Fix an issue where the usage of VkEvents prevented libplacebo from
  working on platforms where they were unsupported (e.g. MoltenVK)
- Support and use vulkan push descriptors where supported. This avoids
  having to maintain, update and synchronize descriptor sets
- Correctly disable signal linearization on low-bit depth FBOs
- Prevent a crash on the (pathological) case of a shader with no
  descriptors
- Fix a bug in the size limit calculations for 2D/3D LUTs
- Use optimization level "performance" instead of "size" for shader
  compilation
- Fix an abort() when trying to render to a non-storable FBO when using
  a final pass that requires compute shaders (e.g. HDR peak detection)

v0.2.1

This is a bugfix release, intended to backport a number of easy and/or
critical fixes onto v0.2.0, since this is effectively the current 'LTS'
due to its use in VLC 3.0. Despite the included fixes, it's still not
recommended to make use of `pl_renderer` in this release. v0.4.0 is the
first release where that feature could be considered officially
supported.

For a full list of backported changes, see the git log. But aside from
cosmetics, comment fixes, here are the most noteworthy changes:

Fixes:
- Fix some typos in shaders that lead to miscompilation
- Work around the lack of mix(vecN, vecN, bvecN) on GLSL 120
- Avoid the use of platform-dependent locale-invariant printf() calls
  for shader generation. Instead, use a custom implementation
- Work around libshaderc locale dependence (currently only relevant when
  using vulkan)
- Fix various errors with anamorphic content handling
- Fix various errors when using vulkan
- Add some more sanitization to RA calls, and fix some of the existing
  ones

v0.4.0

This release brings with it a major performance change to the renderer's
scaling pipeline, support for vulkan format emulation, improvements to
the HDR peak detection algorithm as well as greatly improved support for
compute shader-based rasterization.

In addition to the above features, this release renames the `RA`
abstraction to `pl_gpu`, and normalizes the `ra_` prefix to `pl_` in
order to match the rest of the API as well as prevent potential future
symbol conflicts with mpv's internal copy of RA.

Additions:
- Add `pl_shader_encode_color`, which is an inversion of
  `pl_shader_decode_color` (NOTE: Some features, such as inverse
  constant luminance or XYZ encoding, are not currently supported)
- Add API for overlay images. This can be used to directly overlay
  arbitrary planes (e.g. subtitles, or menu items) on top of the
  rendered image or the target FBO. Each plane is allowed to have its
  own color space/repr and format. See `struct pl_overlay` for more
  details
- Add various predefined `pl_color_repr` values
- Add scene change detection functionality to the HDR peak detection
  algorithm in `pl_shader_color_map`. The new algorithm is designed to
  allow using larger peak detection buffers without the associated risk
  of eye-adaptation-like effects
- Add support for texel buffers (uniform and storage) in the pl_buf API.
  These can be used to storage raw texels in buffer format, much like 1D
  textures, but with the usual buffer-related functionality
- Add support for vulkan format emulation, which allows users to
  directly upload and download e.g. rgb8 or rgb32f textures, even though
  these are not typically supported by GPUs. The format emulation uses
  compute shaders and texel storage buffers to fix the data
  representation on-GPU, and is significantly faster than doing the same
  with on-CPU software conversion routines
- Add a `pl_dispatch_compute` function, which behaves like
  `pl_dispatch_finish` but without the FBO output, blending and
  rasterization simulation logic. This is useful when dispatching
  pure-compute shaders. (Note: Not currently very useful for API users
  since there are no pure-compute shaders as part of the public API)

Changes:
- Rename the `ra_` series of abstractions to `pl_` for consistency. The
  old `struct ra` was renamed to `struct pl_gpu`
- The dispatch logic for compute shaders will now correctly simulate the
  dispatch rect, by manually translating coordinates in the shader.
- Disallow using compute shader scaling for flipped source rects. API
  users who relied on this behavior are encouraged to instead flip the
  coordinates of the resulting `pl_dispatch_finish` call
- Have the renderer infer `src_rect` and `dst_rect` when not given
- Have the renderer encode colors to the `target.repr` before outputting
  to screen. This includes, for example, conversion to YUV and dithering
  to the target representation's bit depth (where known)
- Add a `pl_blend_params` struct to `pl_dispatch_finish`, which can be
  used to enable blending (including compute shaders)
  including when using compute shaders.
- Subsampled planes must now be an integer multiple (or inverse thereof)
  of the reference size. If this is not the case, the renderer will
  assume the planes are cropped or expanded. This is needed to correctly
  render e.g. 4:2:0 subsampled JPEGs with odd sizes
- Require that all uses of `pl_image` with the same `signature` point to
  the same data, to allow safe cache purging
- Allow dispatching debanding and bicubic shaders at any size
- `pl_plane_find_fmt` may now be called with `out_map=NULL`
- Change the `pl_color_map_default_params` in response to the new scene
  detection feature. The default buffer size has been increased from 21
  to 64, which leads to a much more stable image that flickers less

Fixes and performance improvements:
- Fix typo that lead to incorrect anamorphic rendering
- Fix various rendering issues when using a flipped src_rect
- Fix a double-free that could happen as a result of rendering failing
  in an unexpected way.
- Fix handling of alpha values in `pl_shader_decode_color`
- Correctly set the queue family index for vulkan buffer barriers
- Work around a locale dependence in libshaderc by calling uselocale/XXX
  before calling into shaderc functions
- Remove POSIX dependence in the test suite
- Remove non-portable ffsll call
- Fix demos/sdl2.c building on windows
- Fix an issue where trying to polar sample multiple planes could result
  in a variable conflict
- Fix inverted plane shift offset calculation
- Respect the debanding settings in the renderer
- Sanitize `pl_pass_run_params.scissors` and silently drop no-ops
- Avoid dependence on exact float equality in the renderer's
  upscaling/downscaling checks
- Fix an off-by-one in the HDR peak computation code which lead to
  subtly too-dark output
- Correctly check for and enable required vulkan device features
- Switch to a custom implementation of locale-invariant printf, to avoid
  depending on broken platform-dependent versions
- Overhaul the renderer's scaling pipeline. The new code is
  significantly faster when rendering cropped images, and also fixes an
  issue where src_rect was not applied when FBOs are unavailable
- Minor performance tweaks here and there
- Cap the sizes of various buffer pools and command queues in order to
  avoid OOM scenarios in infinite loops that don't contain enough
  forward dependencies to force command flushing
- Correctly check for and catch buffer access overflow
- Don't segfault when `pl_buf_create` fails
- Don't segfault when running a compute shader without viewport/scissors
- Fix a 32-bit integer overflow that could occurr when using HDR peak
  detection on 4K content with large values of `peak_detect_frames`
- Fix an off-by-one in the generation of literal LUTs such as the type
  used by the dithering code when textures are unavailable, which led to
  these shaders miscompiling
- Various fixes for building on 32-bit systems
- Fix an uninitialized value in the renderer

v0.3.0

This update brings with it some major changes to the renderer, support
for swapchains, separated sampling, sigmoidization and more. It also
includes, for the first time, a graphical demo capable of rendering to
the screen - as well as a benchmark.

Additions:
- Add a new type (ra_swapchain) encapsulating the concept of a
  presentable surface (e.g. graphical window) which can be presented to
  the display, as well as an implementation for vulkan swapchains
- Add a simple benchmark suite
- Add a new helper utils/upload.h for simplifying the process of picking
  a compatible texture format and uploading raw texture data
- Add a helper function to setup a pl_render_target from a swapchain
- Add a graphical demo based on SDL2 (compiled with vulkan support)
- Add support for separated (orthogonal) scaling, and make the renderer
  use it
- Add support for using the optimized GPU-native sampling methods for
  nearest/bilinear/bicubic where possible
- Add support for sigmoidization/unsigmoidization shaders
- Make the renderer use linearization/sigmoidization where appropriate

Changes:
- Require vulkan >= v1.0.24, rather than the old >= 1.0.50
- Split up vulkan instance and device creation into separate helpers.
  This is needed to satisfy the typical VkInstance -> VkSurface -> VkDevice
  dependency dance required to create a renderable swapchain.
- Change the API of pl_dispatch_finish/abort to make it clear that they
  consume and take over ownership of the pl_shader
- Have the renderer also #include most of its API dependencies, to cut
  down the amount of #include spam needed to use it
- Rename pl_sample_polar_params and generalize it to include separated
  (orthogonal) scaling as well
- Various API changes to pl_render_params

Fixes:
- Allow running the build script without `git` installed
- Fix an invalid bounds check on ra_tex_blit causing it to fail for
  1D/2D textures
- Use global #include paths for all public headers, rather than relative
- Fix the src_rect calculation in the renderer, thus fixing an issue
  where anamorphic scaling produced the wrong result
- Fix a typo in the renderer, causing it to skip some planes
- Fix a memory leak caused by implicit deallocation of pl_shader
- Fix a bug in the renderer causing the image to stop updating after the
  first frame after a resize
- Fixes a bug where the color mapping / tone mapping shaders sometimes
  got (accidentally) triggered when the source colorspace is unknown
- Fix a bug where compute shaders were incorrectly dispatched after a
  resize of the dispatch target
- Fix a bug where the polar compute shader read from uninitialized
  memory when performing anamorphic scaling
- Various other bug fixes and performance tweaks