Extremely high first call latency Julia 1.6 versus 1.5 with multiphysics PDE solver

Alexander-Barth · November 26, 2021, 12:40pm

I have an issue with compile time latency with a neural work in Julia 1.6 and 1.7-rc2 (compared to Julia 1.5) which might be related:

If one uses the option -O1 the compile times are vastly reduced. When inspecting the run with callgrind to CPUs time is mostly spend in LLVM even on the second call of the same function with the same arguments (using the default optimization level).

I tried to compile Julia from source changing the parameters max_methods, tupletype_depth, tuple_splat, inline_tupleret_bonus but without luck.

github.com

JuliaLang/julia/blob/master/base/compiler/types.jl#L69


      
          
          abstract type ForwardableArgtypes end
          
          struct AnalysisResults
              result
              next::AnalysisResults
              AnalysisResults(@nospecialize(result), next::AnalysisResults) = new(result, next)
              AnalysisResults(@nospecialize(result)) = new(result)
              # NullAnalysisResults() = new(nothing)
              # global const NULL_ANALYSIS_RESULTS = NullAnalysisResults()
          end
          const NULL_ANALYSIS_RESULTS = AnalysisResults(nothing)
          
          """
              InferenceResult(linfo::MethodInstance, [argtypes::ForwardableArgtypes, 𝕃::AbstractLattice])
          
          A type that represents the result of running type inference on a chunk of code.
          
          See also [`matching_cache_argtypes`](@ref).
          """
          mutable struct InferenceResult

When I interrupt Julia during the excessive compilation, it seemed to be busy in the DAGCombiner phase of LLVM.