Programming Languages (PL)¶

Overview¶

Optimizations for HorseIR: Static + Dynamic

Static Analysis¶

Intra-procedural optimizations

Constant propagation and foldering (CPF)
Common Sub-expression Elimination (CSE)
Peephole optimizations
Program slicing

Inter-procedural optimizations

Procedural inlining
Inter-procedural dead code elimination
Inter-procedural constant propagation
Procedure reordering

Others (before actual optimizations)

A framework for optimizing HorseIR code (In progress)
Alias analysis
Array access analysis

*red means not implemented yet.

Dynamic Analysis¶

The main goal is to provide a framework of efficient JIT-based optimizations to generate efficient code on-the-fly. For HorseIR, loop fusion based optimizations are particularlly critical in many cases. Many prior studies have shown that loop fusion based optimizations could be NP-hard to find an optimal solution. Therefore, we intend to explore a couple of heuristics as candidates for aggregating HorseIR primitive functions.

Further Opportunities¶

The list of collected compiler optimizations: read more