switch to SnoopPrecompile.jl (#164)

Project.toml

@@ -1,7 +1,7 @@
 name = "Octavian"
 uuid = "6fd5a793-0b7e-452c-907f-f8bfe9c57db4"
 authors = ["Chris Elrod", "Dilum Aluthge", "Mason Protter", "contributors"]
-version = "0.3.19"
+version = "0.3.20"
 
 [deps]
 ArrayInterface = "4fba245c-0d91-5ea0-9b3e-6abc04ee57a9"
@@ -11,6 +11,7 @@ LoopVectorization = "bdcacae8-1622-11e9-2a5c-532679323890"
 ManualMemory = "d125e4d3-2237-4719-b19c-fa641b8a4667"
 PolyesterWeave = "1d0040c9-8b98-4ee7-8388-3f51789ca0ad"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
+SnoopPrecompile = "66db9d55-30c0-4569-8b51-7e840670fc0c"
 Static = "aedffcd0-7271-4cad-89d0-dc628f76c6d3"
 ThreadingUtilities = "8290d209-cae3-49c0-8002-c8c24d57dab5"
 VectorizationBase = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
@@ -23,6 +24,7 @@ LoopVectorization = "0.12.86"
 ManualMemory = "0.1.1"
 PolyesterWeave = "0.1.1, 0.2"
 Requires = "1"
+SnoopPrecompile = "1"
 Static = "0.7.5, 0.8"
 ThreadingUtilities = "0.5"
 VectorizationBase = "0.21.15"

src/Octavian.jl

@@ -37,6 +37,8 @@ using ManualMemory: MemoryBuffer, load, store!
 
 using ThreadingUtilities: _atomic_add!, _atomic_load, _atomic_store!, launch, wait, SPIN
 
+using SnoopPrecompile: @precompile_setup, @precompile_all_calls
+
 if !(StaticInt <: Base.Integer)
   const Integer = Union{Base.Integer,StaticInt}
 end
@@ -63,18 +65,26 @@ include("complex_matmul.jl")
 include("init.jl") # `Octavian.__init__()` is defined in this file
 
 @static if VERSION >= v"1.8.0-beta1"
-  let
+  @precompile_setup begin
+    # Putting some things in `setup` can reduce the size of the
+    # precompile file and potentially make loading faster.
     __init__()
     A64 = rand(100, 100)
-    matmul(A64, A64)
-    matmul(A64', A64)
-    matmul(A64, A64')
-    matmul(A64', A64')
     A32 = rand(Float32, 100, 100)
-    matmul(A32, A32)
-    matmul(A32', A32)
-    matmul(A32, A32')
-    matmul(A32', A32')
+
+    @precompile_all_calls begin
+      # All calls in this block will be precompiled, regardless of whether
+      # they belong to Octavian.jl or not (on Julia 1.8 and higher).
+      matmul(A64, A64)
+      matmul(A64', A64)
+      matmul(A64, A64')
+      matmul(A64', A64')
+
+      matmul(A32, A32)
+      matmul(A32', A32)
+      matmul(A32, A32')
+      matmul(A32', A32')
+    end
   end
 end