Closed
Description
We want to introduce DLTI support in GC with the following:
- Centralize the DLTI field definition used by GC. This can help front-end users know how many DLTI fields are used and need them to prepare.
- Add a verifier pass to verify the DLTI input from user. Raise error for unexpected input(such as a negative number of num_threads), and raise warn for missing fields, and provide a default value(such as 32K for L1_cache_size).
In the future, we may support arch=spr or arch=native which means GC needs to auto-detect many architecture details based on arch name. This issue doesn't cover this feature, just a mark.