Changes since version 1.0.1
# New features:
* Support for amdgpu native contexts
(Requires umerged changes to qemu, libdrm, and mesa,
for details see https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21658
# Fixes:
* many changes to the drm code to generalize the common code for
native context that is needed for AMDGPU and will be needed for Intel
## Venus:
- Enable support for new extensions:
* EXT_attachment_feedback_loop_layout
* VK_KHR_maintenance5
* VK_KHR_fragment_shading_rate support
* VK_EXT_external_memory_acquire_unmodified
* VK_KHR_format_feature_flags2
- Fixes and improvements:
* Fix a prior ring wait seqno validation
* handle ring fatal error more robustly
* hide priority request logging behind debug build
* lock device tracked object list
* log more on ring fatal
* only set ring thread prio if requested
* optimize to avoid locking each encoder write
* set forwarded nice priority for ring threads
## virgl/vrend:
* create transient dmabuf in resource_map
* fix "error: implicit declaration of function 'close'" for Windows
* fix "fatal error: poll.h: No such file or directory" for Windows
* Correct error value returned by virgl_fence_set_fd_locked()
* virgl_protocol: Use parenthesis in macros
* virgl_util: Add hash_table_search()
* virgl_util: hide slow trace behind debug build
* Require 4-byte alignment for all virglrenderer commands
* consider drm_initialized too
* Add XRGB2101010 support
* Check whether we actually got a valid OpenGL/GLES context
* Properly seperate the mirror_clamp feat reporting
* Support R16G16B16X16_FLOAT as GBM buffer
* Improve interaction between 24 bit formats used with GBM and OpenGL
* clean up error reporting in vrend_renderer_init
* conditionally build using gbm_bo_get_map_info()
* fix incorrect limit sent for max samplers
* reset dirty mask after binding samplers
* stop clobbering the texture bindings
* warn on missing sampler_view.texture
* workaround non-conformant sampler view pattern
* winsys_gbm: initialize pointer map_data before calling gbm_bo_map
* decode: print error if format is not correct
* shader: Never return out-of-range from varying_bit_from_semantic_and_index
* shader: Remove unnecessary condition
* util: Fix compiling set_dmabuf_name() using older kernel headers
* renderer: Verify that the iov has enough space for the texture data
* Add a fuzzer that sets up valid states to be able to test
## Improve compatibility with MS Windows:
* Fix resource_get_info_ext() failing for d3d resources
* Fixed logical operation handling so that the logic-op function is updated whenever needed.
* tgsi: fix "error: 'uint' undeclared" for Windows
## Build changes:
* Support Darwin hosts
* Disable strict aliasing