First relevant commit:
windows.media.speech: Add a worker thread to the recognition session.
--
v2: windows.media.speech: Store recorded audio in a temporary ringbuffer.
windows.media.speech: Implement the audio capturing system.
https://gitlab.winehq.org/wine/wine/-/merge_requests/1948
--
v12: windows.media.speech: Implement the audio capturing system.
windows.media.speech: Allow the recognition session worker to be paused.
windows.media.speech: Add a worker thread to the recognition session.
https://gitlab.winehq.org/wine/wine/-/merge_requests/729
--
v4: vkd3d-shader/hlsl: Get rid of the "intrinsic" field of struct hlsl_ir_function.
vkd3d-shader/hlsl: Forbid returning void expressions from void functions.
vkd3d-shader/hlsl: Generate IR for user-defined function calls.
vkd3d-shader/hlsl: Avoid assuming that expressions have at least one argument.
tests: Test specifying a UAV address as an in/out parameter to a function.
https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/47
SPIR-V doesn't support parallelisation in patch constant functions.
Merging into one phase allows generation of a single patch constant
function and normalises to the shader model 6 pattern.
--
https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/60
This is not a regression, and probably not a visible bug fix either.
On the other hand, it is a clear bug fix, and a relatively safe one as well.
I don't know whether it makes more sense to defer this until after code freeze.
--
https://gitlab.winehq.org/wine/wine/-/merge_requests/1943