[Bug] First compilation after fresh install always times out

[lyfne123]

Diagnosis

ptoas — After downloading/installing a new ptoas binary, the very first compilation task always times out. Subsequent compilations work normally. This suggests a cold-start overhead issue in ptoas (e.g. lazy initialization, cache warm-up, or dynamic library loading on first run).

Description

Every time a fresh ptoas binary is downloaded, the first compilation group encounters a 60-second timeout:

Failed to compile group 'qwen3_decode_layer_incore_1' [qwen3_decode_layer_incore_1_aic, qwen3_decode_layer_incore_1_aiv]: ptoas compilation timed out after 60s

Steps to reproduce:

1. Download a fresh ptoas binary from PTOAS GitHub Releases (e.g. ptoas-bin-macos-aarch64.tar.gz)
2. Extract and set PTOAS_ROOT to the extracted directory
3. Run any compilation that invokes ptoas, e.g. python3 examples/models/qwen3/qwen3_32b_decode_tilelet.py (from pypto-lib)
4. The first ptoas compilation group times out after 60s
5. Re-running the same script succeeds without timeout

Workaround: Run any trivial compilation first to "warm up" ptoas, then run the actual task.

Environment

Component	Version
ptoas	v0.17
pypto-lib	3acde04
pypto	50a84d7 (branch: main)

Host Platform

macOS (arm64)

Additional Context

This issue is reproducible on every fresh ptoas install. The ptoas --version check also exhibits similar behavior — it times out on the first invocation (observed with a 20s timeout). This strongly suggests a one-time cold-start cost in the ptoas binary.