diff --git a/Project.toml b/Project.toml
index e7229f2e..0addc92e 100644
--- a/Project.toml
+++ b/Project.toml
@@ -4,6 +4,7 @@ version = "1.6.2"
 
 [deps]
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 StaticArraysCore = "1e83bf80-4336-4d27-bf5d-d5a4f845583c"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
@@ -15,8 +16,9 @@ Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 StaticArraysStatisticsExt = "Statistics"
 
 [compat]
-julia = "1.6"
+PrecompileTools = "1"
 StaticArraysCore = "~1.4.0"
+julia = "1.6"
 
 [extras]
 BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
diff --git a/src/StaticArrays.jl b/src/StaticArrays.jl
index fad003ae..0e4b1afe 100644
--- a/src/StaticArrays.jl
+++ b/src/StaticArrays.jl
@@ -19,6 +19,8 @@ import LinearAlgebra: transpose, adjoint, dot, eigvals, eigen, lyap, tr,
                       normalize!, Eigen, det, logdet, logabsdet, cross, diff, qr, \
 using LinearAlgebra: checksquare
 
+using PrecompileTools
+
 # StaticArraysCore imports
 # there is intentionally no "using StaticArraysCore" to not take all symbols exported
 # from StaticArraysCore to make transitioning definitions to StaticArraysCore easier.
@@ -136,6 +138,5 @@ include("pinv.jl")
 end
 
 include("precompile.jl")
-_precompile_()
 
 end # module
diff --git a/src/precompile.jl b/src/precompile.jl
index 439b53ac..e0737110 100644
--- a/src/precompile.jl
+++ b/src/precompile.jl
@@ -1,35 +1,29 @@
-function _precompile_()
-    ccall(:jl_generating_output, Cint, ()) == 1 || return nothing
-    # These make a difference for Makie's TTFP and should help others too
-    # Total time for this to run is ~2s. To time it, disable:
-    #   - the `_precompile_()` call in StaticArrays.jl
-    #   - the jl_generating_output line above
-    # and then do
-    #     @time StaticArrays._precompile_()
-
-    for T in (Float32, Float64, Int)   # some common eltypes
-        for S = 2:4                    # some common sizes
-            L = S*S
-            @assert precompile(Tuple{typeof(*),SMatrix{S, S, T, L},SMatrix{S, S, T, L}})
-            @assert precompile(Tuple{typeof(*),SMatrix{S, S, T, L},SVector{S, T}})
-            @assert precompile(Tuple{typeof(inv),SMatrix{S, S, T, L}})
-            @assert precompile(Tuple{typeof(transpose),SMatrix{S, S, T, L}})
-            @assert precompile(Tuple{typeof(_adjoint),Size{(S, S)},SMatrix{S, S, T, L}})
-            @assert precompile(Tuple{Core.kwftype(typeof(minimum)),NamedTuple{(:dims,), Tuple{Int}},typeof(minimum),SMatrix{S, S, T, L}})
-            @assert precompile(Tuple{Core.kwftype(typeof(maximum)),NamedTuple{(:dims,), Tuple{Int}},typeof(maximum),SMatrix{S, S, T, L}})
-            @assert precompile(Tuple{typeof(getindex),SMatrix{S, S, T, L},SOneTo{S-1},SOneTo{S-1}})
+@setup_workload begin
+    @compile_workload begin
+        for T in (Float32, Float64, Int)   # some common eltypes
+            for S = 2:4                    # some common sizes
+                L = S*S
+                x = fill(0., SMatrix{S, S, T, L})
+                x * x
+                y = fill(0., SVector{S, T})
+                x * y
+                inv(x)
+                transpose(x)
+                _adjoint(Size(S, S), x)
+                minimum(x; dims=1)
+                minimum(x; dims=2)
+                maximum(x; dims=1)
+                maximum(x; dims=2)
+                getindex(x, SOneTo(S-1), SOneTo(S-1))
+                y .* x .* y'
+                zero(y)
+                zero(x)
+            end
         end
-    end
 
-    # TODO: These fail to precompile on v1.11 pre-release
-    if VERSION < v"1.11.0-0"
-        # Some expensive generators
-        @assert precompile(Tuple{typeof(which(__broadcast,(Any,Size,Tuple{Vararg{Size}},Vararg{Any},)).generator.gen),Any,Any,Any,Any,Any,Any})
-        @assert precompile(Tuple{typeof(which(_zeros,(Size,Type{<:StaticArray},)).generator.gen),Any,Any,Any,Type,Any})
-        @assert precompile(Tuple{typeof(which(_mapfoldl,(Any,Any,Colon,Any,Size,Vararg{StaticArray},)).generator.gen),Any,Any,Any,Any,Any,Any,Any,Any})
-    end
-    # broadcast_getindex
-    for m = 0:5, n = m:5
-        @assert precompile(Tuple{typeof(broadcast_getindex),NTuple{m,Int},Int,CartesianIndex{n}})
+        # broadcast_getindex
+        for m = 0:5, n = m:5
+            broadcast_getindex(Tuple(1:m), 1, CartesianIndex(n))
+        end
     end
 end