CodeCosts.jl
This package provides a variant of @code_typed with estimated costs for the
inlining.
This helps find the factors which are preventing the SIMD vectorization.
julia> using CodeCosts
julia> f(x::T) where T = convert(T, max(x * 10.0, x / 3))
f (generic function with 1 method)
julia> @code_costs f(1.0f0)
CodeCostsInfo(
CodeInfo(
1 1 ─ %1 = Base.fpext(Base.Float64, x)::Float64
4 │ %2 = Base.mul_float(%1, 10.0)::Float64
20 │ %3 = Base.div_float(x, 3.0f0)::Float32
1 │ %4 = Base.fpext(Base.Float64, %3)::Float64
2 │ %5 = Base.lt_float(%2, %4)::Bool
1 │ %6 = Base.bitcast(Base.Int64, %4)::Int64
1 │ %7 = Base.slt_int(%6, 0)::Bool
1 │ %8 = Base.bitcast(Base.Int64, %2)::Int64
1 │ %9 = Base.slt_int(%8, 0)::Bool
0 │ %10 = Base.not_int(%7)::Bool
1 │ %11 = Base.and_int(%9, %10)::Bool
1 │ %12 = Base.or_int(%5, %11)::Bool
2 │ %13 = Base.ne_float(%2, %2)::Bool
1 │ %14 = Base.Math.ifelse(%13, %2, %4)::Float64
2 │ %15 = Base.ne_float(%4, %4)::Bool
1 │ %16 = Base.Math.ifelse(%15, %4, %2)::Float64
1 │ %17 = Base.Math.ifelse(%12, %14, %16)::Float64
1 │ %18 = Base.fptrunc(Base.Float32, %17)::Float32
0 └── return %18
)
, CodeCostsSummary(
zero: 2|
cheap: 12| 111111111111
middle: 10| 4===2=2=2=
expensive: 20| 20==================
total: 42| 100 (default threshold)
))