| Commit message (Collapse) | Author | Age |
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
sdl_cap.c wrote the same 32 in the GZF2 record format field for every
capture path, so gzf2_video.py's auto mode decoded ARGB8888/X-surface
frames (BGRA byte order on little-endian) as RGBA and swapped red and
blue. Make the field the documented pixel-order tag: 32 = RGBA8 from
the GL swap path (glReadPixels GL_RGBA), 33 = BGRA8 from
SDL_RenderReadPixels(ARGB8888) and X window surfaces; xgrab.c tags its
ZPixmap BGRA records 33 as well.
gzf2_video.py continues to accept both tags (and old all-32 captures
via --pix-fmt bgra), checks ffmpeg/ffprobe before indexing, and keeps
the final resampled endpoint so a 90-frame 60 fps capture encodes as
90 frames instead of 88.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
- cycle-r1780.sh: load vfio modules one per modprobe (the multi-module
form was silently treating the extra names as parameters and never
loading mdev); auto-detect the vgpu mdev type and prefer the 2 GiB
type used by the campaign.
- sdl_cap.c: gate capture-file creation to processes that can actually
read GL pixels (a child was truncating the capture), and log the
capture pid/pointers.
- gz_pbuf.c: LD_PRELOAD shim that gives an SDL2 GL client a GLX pbuffer
context on the guest's NVIDIA X server (the proven gl_frames path),
with SDL_GL_GetDrawableSize/SwapWindow interception.
|
|
|
video (Xvfb XGetImage capture, SDL present shim, window resize, XTEST input, QGA range fetch)
|