tt-lang-profile-optimize

External Resources

• Performance Tools Reference -- environment variables, perf summary, auto-profiling, signposts, Perfetto trace server
• TT-Lang Specification, Section 10 -- signpost and dprint language spec

Before You Start: Understand the Real Workload

Before optimizing, ask the user:

• What data sizes will this kernel run on in production? The test data may be much smaller than the real workload. Do not over-optimize for test data.
• Where will this run? Single chip? Multi-chip? This affects core budget.

Keep real-world constraints in mind throughout. For example, do not move everything to L1 just because the test data fits -- if the real data is larger, you need streaming. Optimizations must hold for the production workload, not just the test case.

Optimization Targets

Three goals, in priority order:

1. Maximize Core Utilization (target: 100%)