Fix: compile host_build_graph orchestration with host g++#1330
Conversation
|
Warning Review limit reached
Next review available in: 42 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (1)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the orchestration compilation logic in simpler_setup/kernel_compiler.py to use the host toolchain when the runtime is 'host_build_graph'. Feedback on this change points out that the implementation silently falls back to the cross-compilation toolchain for any other runtime name, which violates the function's docstring stating that an unknown runtime name should raise a ValueError. It is recommended to explicitly validate the runtime name and raise an error if it is unsupported.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
ChaoWao
force-pushed
the
fix/hbg-orch-host-gxx
branch
from
35f3281 to
c22d5e8
Compare
host_build_graph dlopens the orchestration .so on the x86_64 host, but compile_orchestration routed a2a3/a5 to the aarch64 cross-compiler unconditionally — that routing predates host_build_graph. The AArch64 orch .so could not be dlopened by the x86_64 host runtime, surfacing as "dlopen failed: ... No such file or directory" and failing every host_build_graph scene test on a2a3 onboard. Route the orch toolchain by runtime instead of only by platform: host_build_graph uses HOST_GXX (the host-dlopen path, onboard and sim); tensormap_and_ringbuffer keeps AARCH64_GXX onboard (AICPU dlopen) and HOST_GXX for sim. Unknown runtime_name now raises ValueError, matching the docstring's documented contract. Verified on a2a3 onboard (910B2C): host_build_graph vector_example and paged_attention now pass; tensormap alternating_matmul_add unchanged (regression guard). a5 host_build_graph takes the same path but has no a5 silicon here to onboard-verify.
1 participant
Summary
host_build_graphdlopens the orchestration.soon the x86_64 host, butcompile_orchestrationrouted a2a3/a5 to the aarch64 cross-compiler unconditionally — that routing predateshost_build_graph. The AArch64 orch.socould not be dlopened by the x86_64 host runtime, surfacing asdlopen failed: ... No such file or directoryand failing everyhost_build_graphscene test on a2a3 onboard.host_build_graph→HOST_GXX(host-dlopen path, onboard + sim);tensormap_and_ringbuffer→AARCH64_GXXonboard (AICPU dlopen),HOST_GXXfor sim.elsebranch is byte-identical to before, sotensormap_and_ringbufferis unchanged.Root cause (evidence)
.so(a2a3host_build_graph) wasMachine: AArch64;libhost_runtime.soand the host areX86-64.lddreported "not a dynamic executable"; glibc's loader emits the misleadingNo such file or directoryon an ELF machine mismatch..soisX86-64,lddresolves all deps cleanly.Testing
Onboard a2a3 (910B2C), device 0 (run unlocked —
task-submitis not on this box; results are deterministic, not concurrency-related):host_build_graph/vector_examplehost_build_graph/paged_attention(the case #1185 claimed passed)tensormap_and_ringbuffer/alternating_matmul_add(regression guard)Note
a5
host_build_graphtakes the same code path (also host-dlopen) and is fixed by this change, but I have no a5 silicon here to onboard-verify — please run an a5host_build_graphcase on an a5 box / the a5 CI leg.