Test validation of GPU-only function.

Manifest.toml

@@ -40,9 +40,9 @@ version = "0.9.37"
 
 [[Compat]]
 deps = ["Base64", "Dates", "DelimitedFiles", "Distributed", "InteractiveUtils", "LibGit2", "Libdl", "LinearAlgebra", "Markdown", "Mmap", "Pkg", "Printf", "REPL", "Random", "SHA", "Serialization", "SharedArrays", "Sockets", "SparseArrays", "Statistics", "Test", "UUIDs", "Unicode"]
-git-tree-sha1 = "4fecfd5485d3c5de4003e19f00c6898cccd40667"
+git-tree-sha1 = "ac4132ad78082518ec2037ae5770b6e796f7f956"
 uuid = "34da2185-b29b-5c13-b0c7-acf172513d20"
-version = "3.26.0"
+version = "3.27.0"
 
 [[CompilerSupportLibraries_jll]]
 deps = ["Artifacts", "Libdl"]
@@ -83,9 +83,9 @@ version = "6.2.2"
 
 [[GPUCompiler]]
 deps = ["DataStructures", "ExprTools", "InteractiveUtils", "LLVM", "Libdl", "Logging", "Scratch", "Serialization", "TimerOutputs", "UUIDs"]
-git-tree-sha1 = "a5a239b8c688f59872eb689edd75395e97cc6641"
+git-tree-sha1 = "7f13030524614d90c876332c980fd1edef2331e8"
 uuid = "61eb1bfa-7361-4325-ad38-22787b887f55"
-version = "0.11.2"
+version = "0.11.3"
 
 [[InteractiveUtils]]
 deps = ["Markdown"]

Project.toml

@@ -37,7 +37,7 @@ CEnum = "0.2, 0.3, 0.4"
 DataStructures = "0.17, 0.18"
 ExprTools = "0.1"
 GPUArrays = "6.1.0"
-GPUCompiler = "0.11.1"
+GPUCompiler = "0.11.3"
 LLVM = "3"
 MacroTools = "0.5"
 Memoize = "0.4"

test/device/intrinsics.jl

@@ -81,6 +81,21 @@ end
             @test testf(a->op.(a, T(2)), T[-1])
         end
     end
+
+    @testset "rsqrt" begin
+        # GPUCompiler.jl#173: a CUDA-only device function fails to validate
+        function kernel(a)
+            a[] = CUDA.rsqrt(a[])
+            return
+        end
+
+        # make sure this test uses an actual device function
+        @test_throws ErrorException kernel(ones(1))
+
+        a = CuArray{Float32}([4])
+        @cuda kernel(a)
+        @test Array(a) == [0.5]
+    end
 end