Merge pull request #21 from ranocha/hr/precompile

precompile `__init__`

Project.toml

@@ -1,17 +1,19 @@
 name = "CPUSummary"
 uuid = "2a0fbf3d-bb9c-48f3-b0a9-814d99fd7ab9"
 authors = ["chriselrod <[email protected]> and contributors"]
-version = "0.2.2"
+version = "0.2.3"
 
 [deps]
 CpuId = "adafc99b-e345-5852-983c-f28acb93d879"
 IfElse = "615f187c-cbe4-4ef1-ba3b-2fcf58d6d173"
+PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
 Static = "aedffcd0-7271-4cad-89d0-dc628f76c6d3"
 
 [compat]
 CpuId = "0.3"
 IfElse = "0.1"
 Static = "0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8"
+PrecompileTools = "1.1"
 julia = "1.6"
 
 [extras]

src/CPUSummary.jl

@@ -66,6 +66,7 @@ function __init__()
     @eval sys_threads() = static($syst)
   end
   _extra_init()
+  return nothing
 end
 
 
@@ -94,4 +95,9 @@ function num_cache_levels()
   )
 end
 
+# explicit precompilation only on Julia v1.9 and newer
+if VERSION >= v"1.9"
+  include("precompile.jl")
+end
+
 end

src/precompile.jl

@@ -0,0 +1,10 @@
+using PrecompileTools: @compile_workload
+
+@compile_workload begin
+  __init__()
+  # `_extra_init()` is called by `__init__()`
+  # However, it does not seem to be recognized correctly since we can
+  # further reduce the time of `using CPUSummary` significantly by
+  # precompiling it here in addition ot `__init__()`.
+  _extra_init()
+end

src/x86.jl

@@ -53,6 +53,7 @@ function _extra_init()
   cs !== PrecompiledCacheSize && _eval_cache_size(cs)
   ci = CpuId.cacheinclusive()
   ci !== PrecompiledCacheInclusive && _eval_cache_inclusive(ci)
+  return nothing
 end