Test/fix FLoops.jl without reduce (#10)

JuliaFolds · Mar 2, 2021 · 86a13e5 · 86a13e5
1 parent 064aba9
commit 86a13e5
Show file tree

Hide file tree

Showing 5 changed files with 92 additions and 25 deletions.
diff --git a/src/transduce.jl b/src/transduce.jl
@@ -147,30 +147,34 @@ end
         acc = next(rf, acc, x)
     end
 
-    # combineblock
-    offsetb = (igl - 1) * groupsize()[1]
-    bound = max(0, nbasecases - offsetb)
-
-    # shared mem for a complete reduction
-    shared = @localmem(T, (2 * groupsize()[1]))
-    @inbounds shared[ill] = acc
-
-    m = ill - 1
-    t = ill
-    s = 1
-    c = nextpow(2, groupsize()[1]) >> 1
-    while c != 0
-        @synchronize
-        if t + s <= bound && iseven(m)
-            @inbounds shared[t] = _combine(rf, shared[t], shared[t+s])
-            m >>= 1
+    if Base.issingletontype(T)
+        # nothing to reduce (e.g., @floop w/o @reduce)
+    else
+        # combineblock
+        offsetb = (igl - 1) * groupsize()[1]
+        bound = max(0, nbasecases - offsetb)
+
+        # shared mem for a complete reduction
+        shared = @localmem(T, (2 * groupsize()[1]))
+        @inbounds shared[ill] = acc
+
+        m = ill - 1
+        t = ill
+        s = 1
+        c = nextpow(2, groupsize()[1]) >> 1
+        while c != 0
+            @synchronize
+            if t + s <= bound && iseven(m)
+                @inbounds shared[t] = _combine(rf, shared[t], shared[t+s])
+                m >>= 1
+            end
+            s <<= 1
+            c >>= 1
         end
-        s <<= 1
-        c >>= 1
-    end
 
-    if t == 1
-        @inbounds dest[igl] = shared[1]
+        if t == 1
+            @inbounds dest[igl] = shared[1]
+        end
     end
 end
 

diff --git a/test/device_tests.jl b/test/device_tests.jl
@@ -1,17 +1,30 @@
+using FLoops
 using Folds
 using FoldsKernelAbstractions
+using Referenceables
 using Test
 
+function inc!(xs, ex = nothing)
+    @floop ex for x in referenceable(xs)
+        x[] += 1
+    end
+    return xs
+end
+
 function test_size(
     device;
     groupsize,
     basesize = 2,
     extraitems = 0,
     n = groupsize * basesize + extraitems,
 )
-    @test Folds.sum(1:n, KAEx(device; groupsize, basesize)) == sum(1:n)
-    @test Folds.reduce(xor, 1:n, KAEx(device; groupsize, basesize); init = 0) ==
-          reduce(xor, 1:n; init = 0)
+    ex = KAEx(device; groupsize, basesize)
+
+    @test Folds.sum(1:n, ex) == sum(1:n)
+    @test Folds.reduce(xor, 1:n, ex; init = 0) == reduce(xor, 1:n; init = 0)
+
+    xs = FoldsKernelAbstractions.arrayfor(device)(1:n)
+    @test collect(inc!(xs, ex)) == 2:n+1
 end
 
 function test_sweep_sizes(device)

diff --git a/test/environments/main/Manifest.toml b/test/environments/main/Manifest.toml
@@ -104,6 +104,12 @@ git-tree-sha1 = "48920211c95a6da1914a06c44ec94be70e84ffff"
 uuid = "187b0558-2788-49d3-abe0-74a17ed4e7c9"
 version = "1.1.0"
 
+[[ContextVariablesX]]
+deps = ["Compat", "Logging", "UUIDs"]
+git-tree-sha1 = "fc81d9dd4ffdafb56680f01e6d9db464a6b3689d"
+uuid = "6add18c4-b38d-439d-96f6-d6bc489c04c5"
+version = "0.1.1"
+
 [[DataAPI]]
 git-tree-sha1 = "dfb3b7e89e395be1e25c2ad6d7690dc29cc53b1d"
 uuid = "9a962f9c-6df0-11e9-0e5d-c546b8b5ee8a"
@@ -142,6 +148,18 @@ git-tree-sha1 = "10407a39b87f29d47ebaca8edbc75d7c302ff93e"
 uuid = "e2ba6199-217a-4e67-a87a-7c52f15ade04"
 version = "0.1.3"
 
+[[FLoops]]
+deps = ["Compat", "FLoopsBase", "JuliaVariables", "MLStyle", "Setfield", "Transducers"]
+git-tree-sha1 = "e38ee2e0fdb95053d26593f92ffaa8353b9ff53f"
+uuid = "cc61a311-1640-44b5-9fba-1b764f453329"
+version = "0.1.6"
+
+[[FLoopsBase]]
+deps = ["ContextVariablesX"]
+git-tree-sha1 = "cf3d8b2527be12d204d06aba922b30339a9653dd"
+uuid = "b9860ae5-e623-471e-878b-f6a53c775ea6"
+version = "0.1.0"
+
 [[Folds]]
 deps = ["Accessors", "BangBang", "DefineSingletons", "Distributed", "InitialValues", "MicroCollections", "Test", "Transducers"]
 git-tree-sha1 = "1d508282913e9872b748088bea4a83726fcfdfed"
@@ -189,6 +207,12 @@ git-tree-sha1 = "a431f5f2ca3f4feef3bd7a5e94b8b8d4f2f647a0"
 uuid = "692b3bcd-3c85-4b1f-b108-f13ce0eb3210"
 version = "1.2.0"
 
+[[JuliaVariables]]
+deps = ["MLStyle", "NameResolution"]
+git-tree-sha1 = "e0fcfa0a2f6122fbe13603764c5310dde00c5593"
+uuid = "b14d175d-62b4-44ba-8fb7-3064adc8c3ec"
+version = "0.2.3"
+
 [[KernelAbstractions]]
 deps = ["Adapt", "Cassette", "InteractiveUtils", "MacroTools", "SpecialFunctions", "StaticArrays", "UUIDs"]
 git-tree-sha1 = "39faf3047b6f03610d1a4a87071091be047e4681"
@@ -217,6 +241,11 @@ uuid = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 [[Logging]]
 uuid = "56ddb016-857b-54e1-b83d-db4d58db5568"
 
+[[MLStyle]]
+git-tree-sha1 = "937eda9ce36fcce082a42edd7181c8d23f4eb550"
+uuid = "d8e11817-5142-5d16-987a-aa16d5891078"
+version = "0.4.6"
+
 [[MacroTools]]
 deps = ["Markdown", "Random"]
 git-tree-sha1 = "6a8a2a625ab0dea913aba95c11370589e0239ff0"
@@ -242,6 +271,12 @@ git-tree-sha1 = "df42d0816edfc24f5b82a728f46381613c4dff79"
 uuid = "872c559c-99b0-510c-b3b7-b6c96a88d5cd"
 version = "0.7.14"
 
+[[NameResolution]]
+deps = ["PrettyPrint"]
+git-tree-sha1 = "1a0fa0e9613f46c9b8c11eee38ebb4f590013c5e"
+uuid = "71a1bf82-56d0-4bbc-8a3c-48b961074391"
+version = "0.1.5"
+
 [[OpenSpecFun_jll]]
 deps = ["Artifacts", "CompilerSupportLibraries_jll", "JLLWrappers", "Libdl", "Pkg"]
 git-tree-sha1 = "9db77584158d0ab52307f8c04f8e7c08ca76b5b3"
@@ -257,6 +292,11 @@ version = "1.4.0"
 deps = ["Dates", "LibGit2", "Libdl", "Logging", "Markdown", "Printf", "REPL", "Random", "SHA", "UUIDs"]
 uuid = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
 
+[[PrettyPrint]]
+git-tree-sha1 = "632eb4abab3449ab30c5e1afaa874f0b98b586e4"
+uuid = "8162dcfd-2161-5ef2-ae6c-7681170c5f98"
+version = "0.2.0"
+
 [[Printf]]
 deps = ["Unicode"]
 uuid = "de0858da-6303-5e67-8744-51eddeeeb8d7"
@@ -274,6 +314,12 @@ git-tree-sha1 = "57d8440b0c7d98fc4f889e478e80f268d534c9d5"
 uuid = "189a3867-3050-52da-a836-e630ba90ab69"
 version = "1.0.0"
 
+[[Referenceables]]
+deps = ["Adapt"]
+git-tree-sha1 = "8502ffd7a930b89a5bd612e748d65897955fafcc"
+uuid = "42d2dcc6-99eb-4e98-b66c-637b7d73030e"
+version = "0.1.1"
+
 [[Requires]]
 deps = ["UUIDs"]
 git-tree-sha1 = "cfbac6c1ed70c002ec6361e7fd334f02820d6419"

diff --git a/test/environments/main/Project.toml b/test/environments/main/Project.toml
@@ -3,7 +3,9 @@ Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
 Aqua = "4c88cf16-eb10-579e-8560-4a9242c79595"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 CUDAKernels = "72cfdca4-0801-4ab0-bf6a-d52aa10adc57"
+FLoops = "cc61a311-1640-44b5-9fba-1b764f453329"
 Folds = "41a02a25-b8f0-4f67-bc48-60067656b558"
 FoldsKernelAbstractions = "52461acb-21f3-48fe-a5ec-9a226775d6ac"
 KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
+Referenceables = "42d2dcc6-99eb-4e98-b66c-637b7d73030e"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
diff --git a/test/runtests.jl b/test/runtests.jl
@@ -5,6 +5,8 @@ import CUDA
 const TEST_CUDA = CUDA.has_cuda_gpu()
 const TEST_GPU = TEST_CUDA
 
+TEST_CUDA && CUDA.allowscalar(false)
+
 find_test(subdir = "") = sort([
     joinpath(subdir, file) for file in readdir(joinpath(@__DIR__, subdir)) if
     match(r"^test_.*\.jl$", file) !== nothing