Warning
This package is not maintained anymore. Please use e.g. JET.jl instead.
Traceur is essentially a codified version of the Julia performance tips. You run your code, it tells you about any obvious performance traps.
julia> using Traceur
julia> naive_relu(x) = x < 0 ? 0 : x
julia> @trace naive_relu(1.0)
naive_relu(::Float64) at none:1
returns Union{Float64, Int64}
1.0
julia> function naive_sum(xs)
s = 0
for x in xs
s += x
end
return s
end
julia> @trace naive_sum([1.])
Base.indexed_next(::Tuple{Int64,Bool}, ::Int64, ::Int64) at tuple.jl:54
returns Tuple{Union{Bool, Int64},Int64}
naive_sum(::Array{Float64,1}) at none:2
s is assigned as Int64 at line 2
s is assigned as Float64 at line 4
dynamic dispatch to s + x at line 4
returns Union{Float64, Int64}
1.0
julia> y = 1
julia> f(x) = x+y
julia> @trace f(1)
f(::Int64) at none:1
uses global variable Main.y
dynamic dispatch to x + Main.y at line 1
returns Any
2The heavily lifting is done by analyse, which takes a Call (essentially a (f, args...) tuple for each function called in the code). Most of the analysis steps work by retrieving the code_typed of the function, inspecting it for issues and emitting any warnings.
Suggestions for (or better, implementations of!) further analysis passes are welcome.