[CI] Add gpu examples CI tests #490
Conversation
There was a problem hiding this comment.
Pull Request Overview
This PR refines shared memory open type checks in GPU-related BPF map implementations and introduces a new CI workflow for testing GPU examples. The main change standardizes how server-side vs. agent-side code paths are distinguished using more precise shared memory open type conditions.
- Updates conditional logic in GPU map implementations to use more accurate open type checks
- Disables existing CI workflows to conserve resources on the dev/gpu_example branch
- Adds a comprehensive new workflow for testing GPU examples with CUDA support
Reviewed Changes
Copilot reviewed 16 out of 16 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| runtime/src/bpf_map/gpu/nv_gpu_shared_array_map.cpp | Changed server/agent side detection from == SHM_CREATE_OR_OPEN to != SHM_OPEN_ONLY to properly handle both server open types (SHM_CREATE_OR_OPEN and SHM_REMOVE_AND_CREATE) |
| runtime/src/bpf_map/gpu/nv_gpu_ringbuf_map.cpp | Updated agent-side detection from != SHM_CREATE_OR_OPEN to != SHM_REMOVE_AND_CREATE |
| runtime/src/bpf_map/gpu/nv_gpu_array_map.cpp | Updated agent-side detection from != SHM_CREATE_OR_OPEN to != SHM_REMOVE_AND_CREATE |
| .github/workflows/test-gpu-examples.yml | Added new GPU examples testing workflow with CUDA 12.6 support and Python test orchestration |
| .github/workflows/*.yml (8 files) | Disabled existing workflows to save CI resources during development |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| { | ||
| if (shm_holder.global_shared_memory.get_open_type() != | ||
| shm_open_type::SHM_CREATE_OR_OPEN) { | ||
| shm_open_type::SHM_REMOVE_AND_CREATE) { |
There was a problem hiding this comment.
Inconsistent agent-side detection logic. This condition checks != SHM_REMOVE_AND_CREATE, which would be true for both SHM_OPEN_ONLY (agent) and SHM_CREATE_OR_OPEN (server). Based on the code at lines 41-43, the else branch should only execute on server side (returning server_gpu_shared_mem). The condition should be == SHM_OPEN_ONLY to match the pattern in nv_gpu_shared_array_map.cpp line 208-209.
| { | ||
| if (shm_holder.global_shared_memory.get_open_type() != | ||
| shm_open_type::SHM_CREATE_OR_OPEN) { | ||
| shm_open_type::SHM_REMOVE_AND_CREATE) { |
There was a problem hiding this comment.
Inconsistent agent-side detection logic. This condition checks != SHM_REMOVE_AND_CREATE, which would be true for both SHM_OPEN_ONLY (agent) and SHM_CREATE_OR_OPEN (server). Based on the else branch at line 42 returning local data buffer (which should be server-side behavior), the condition should be == SHM_OPEN_ONLY to match the pattern in nv_gpu_shared_array_map.cpp line 208-209.
Sy0307
force-pushed
the
dev/gpu_example
branch
2 times, most recently
from
5d9b91e to
d000b56
Compare
Sy0307
force-pushed
the
dev/gpu_example
branch
from
f20d25f to
44b4253
Compare
Sy0307
force-pushed
the
dev/gpu_example
branch
from
004a951 to
3022a5b
Compare
Sy0307
force-pushed
the
dev/gpu_example
branch
2 times, most recently
from
f2d799c to
39fe2a0
Compare
… default SM version assumed"
…uring correct tail management in the ring buffer.
…g line buffering and flushing output after printing counter values.
… map host memory and streamline shared memory registration.
…ers to ensure correct memory mapping and enhance shared memory management.
…use angle brackets for consistency with other includes.
…p pointers, replacing it with a no-op placeholder for clarity in the nv_attach_impl class.
Sy0307
force-pushed
the
dev/gpu_example
branch
2 times, most recently
from
aacd164 to
31f87ca
Compare
1 participant
To solve #487