Codex/issue 26 a5sim bgemm fix#402
Conversation
…m macro Move pto_cpu_sim_* simulation context APIs from device_runner.cpp into a dedicated cpu_sim_state.cpp/h. Add CPU_SIM_SET_TASK_COOKIE macro to inner_kernel.h (sim: real call, onboard: no-op), matching the existing STORE_RELEASE_FENCE platform polymorphism pattern. This ensures aicore_executor.cpp compiles cleanly on both sim and device targets.
![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 introduces a centralized CPU simulation state management system to better emulate hardware-specific execution contexts, such as block/subblock identities and task cookies. It adds thread-local storage for these contexts and a shared storage mechanism to support cross-core features like VEC_FIFO in simulation. The changes include updates to the kernel compiler for platform and core-specific flags, integration of state resets within the DeviceRunner, and the removal of simulation-specific code paths in test kernels to ensure they align with hardware implementations. All provided review comments were filtered out as they were purely evaluative or explanatory without offering actionable improvements, so I have no further feedback to provide.
1 participant
No description provided.