Display: EGLINFO pipeline detection + GPU/CPU hint in kmscube & weston-wayland-test #248
Conversation
smuppand
requested review from
Vkenjam,
abbajaj806,
bhargav0610 and
maheswaqcom
|
@maheswaqcom Would you recommend failing the test if a CPU pipeline is detected instead of a GPU during detection? |
@smuppand first of all thanks for bringing this change. IMO, we should skip tests under Graphics (weston-simple-egl & KMScube) instead of failing, for better alignment to call out GPU HW acceleration not enabled |
smuppand
force-pushed
the
Display
branch
2 times, most recently
from
a838f6e to
0b36dc8
Compare
Thank you for clarifying the requirement. I've submitted the new patch to address the SKIP. |
abbajaj806
left a comment
abbajaj806
left a comment
There was a problem hiding this comment.
Please review the comments.
Introduce EGLINFO pipeline detection helpers in lib_display.sh to select a working platform (wayland/gbm/device/surfaceless) and print legacy-style details (EGL driver, GL_VENDOR, GL_RENDERER). Also emit an explicit hint indicating whether the selected path is GPU (msm/freedreno/etc.) or CPU (kms_swrast/llvmpipe/etc.) for CI clarity. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>
Call display_print_eglinfo_pipeline in kmscube (auto/ gbm) runner to record which EGL stack is active (GPU vs CPU) before executing the test workloads. This improves triage on boards where rendering silently falls back to CPU. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>
Call display_print_eglinfo_pipeline in weston-wayland-test (wayland) runner to record which EGL stack is active (GPU vs CPU) before executing the test workloads. This improves triage on boards where rendering silently falls back to CPU. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>
3 participants
What
[lib_display.sh](http://lib_display.sh/).EGL driver: msm,GL_VENDOR: freedreno, etc.)EGL driver: kms_swrast,GL_RENDERER: llvmpipe, etc.)kmscubeandweston-wayland-test[run.sh](http://run.sh/)to call the helper before running the test, so CI logs capture the active rendering path.Why
On some targets rendering can silently fall back to software (CPU) even though the test still runs. This change makes the pipeline explicit in logs, improving triage and reducing time to diagnose driver/stack issues.
Sample log - GPU
[INFO] 1970-01-01 06:03:38 - ---------------- EGLINFO pipeline detection (select one) ----------------
[INFO] 1970-01-01 06:03:38 - EGLINFO: Pipeline type: GPU (hardware)
[INFO] 1970-01-01 06:03:38 - EGLINFO: Pipeline=WAYLAND platform:
[INFO] 1970-01-01 06:03:38 - EGLINFO: EGL driver name: msm
[INFO] 1970-01-01 06:03:38 - EGLINFO: GL_VENDOR: freedreno
[INFO] 1970-01-01 06:03:38 - EGLINFO: GL_RENDERER: FD643
[INFO] 1970-01-01 06:03:38 - ---------------- End EGLINFO pipeline detection --------------------------
Sample log - CPU
[INFO] 2026-01-07 18:36:51 - ---------------- EGLINFO pipeline detection (select one) ----------------
[INFO] 2026-01-07 18:36:51 - EGLINFO: Pipeline type: CPU (software)
[INFO] 2026-01-07 18:36:51 - EGLINFO: Pipeline=GBM platform:
[INFO] 2026-01-07 18:36:51 - EGLINFO: EGL driver name: kms_swrast
[INFO] 2026-01-07 18:36:51 - EGLINFO: GL_VENDOR: Mesa
[INFO] 2026-01-07 18:36:51 - EGLINFO: GL_RENDERER: llvmpipe (LLVM 21.1.8, 128 bits)
[INFO] 2026-01-07 18:36:51 - ---------------- End EGLINFO pipeline detection --------------------------