[Mlir-commits] [mlir] [mlir] [dataflow] further optimize dataflow compile time (PR #149804)

[donald chen]

cxy-1993 wrote:

I tested the solver's initialization and fixed-point convergence time, the actual compilation time data fluctuates significantly. After averaging multiple tests, i estimate that the analysis, which doesn't need this patch, would see an approximate 3-5% increase in compilation time. The introduced overhead is indeed quite substantial, so I think we should hold off on merging this patch for now.



> I think it's probably worthwhile to measure the baseline overhead this adds to the dataflow solver. The dataflow solver already has a bunch of hash map lookups as part of its inner loop so I do wonder how much slower it gets.



https://github.com/llvm/llvm-project/pull/149804