wip: SpGeMM benchmark

Author: IgorErin

benchmarks/GraphBLAS-sharp.Benchmarks/Algorithms/BFS.fs

@@ -92,89 +92,87 @@ type Benchmarks<'elem when 'elem : struct>(
 
     abstract member Benchmark : unit -> unit
 
-module WithoutTransfer  =
-    type Benchmark<'elem when 'elem : struct>(
+type WithoutTransferBenchmark<'elem when 'elem : struct>(
+    buildFunToBenchmark,
+    converter: string -> 'elem,
+    boolConverter,
+    vertex) =
+
+    inherit Benchmarks<'elem>(
         buildFunToBenchmark,
-        converter: string -> 'elem,
+        converter,
         boolConverter,
-        vertex) =
+        vertex)
 
-        inherit Benchmarks<'elem>(
-            buildFunToBenchmark,
-            converter,
-            boolConverter,
-            vertex)
+    [<GlobalSetup>]
+    override this.GlobalSetup() =
+        this.ReadMatrix()
+        this.LoadMatrixToGPU()
 
-        [<GlobalSetup>]
-        override this.GlobalSetup() =
-            this.ReadMatrix()
-            this.LoadMatrixToGPU()
+    [<IterationCleanup>]
+    override this.IterationCleanup() =
+        this.ClearResult()
 
-        [<IterationCleanup>]
-        override this.IterationCleanup() =
-            this.ClearResult()
+    [<GlobalCleanup>]
+    override this.GlobalCleanup() =
+        this.ClearInputMatrix()
 
-        [<GlobalCleanup>]
-        override this.GlobalCleanup() =
-            this.ClearInputMatrix()
+    [<Benchmark>]
+    override this.Benchmark() =
+        this.BFS()
+        this.Processor.PostAndReply Msg.MsgNotifyMe
 
-        [<Benchmark>]
-        override this.Benchmark() =
-            this.BFS()
-            this.Processor.PostAndReply Msg.MsgNotifyMe
+type BFSWithoutTransferBenchmarkInt32() =
 
-    type Int() =
+    inherit WithoutTransferBenchmark<int>(
+        (singleSource ArithmeticOperations.intSumOption ArithmeticOperations.intMulOption),
+        int32,
+        (fun _ -> Utils.nextInt (System.Random())),
+        0)
 
-        inherit Benchmark<int>(
-            (singleSource ArithmeticOperations.intSumOption ArithmeticOperations.intMulOption),
-            int32,
-            (fun _ -> Utils.nextInt (System.Random())),
-            0)
+    static member InputMatrixProvider =
+        Benchmarks<_>.InputMatrixProviderBuilder "BFSBenchmarks.txt"
 
-        static member InputMatrixProvider =
-            Benchmark<_>.InputMatrixProviderBuilder "BFSBenchmarks.txt"
+type WithTransferBenchmark<'elem when 'elem : struct>(
+    buildFunToBenchmark,
+    converter: string -> 'elem,
+    boolConverter,
+    vertex) =
 
-module WithTransfer =
-    type Benchmark<'elem when 'elem : struct>(
+    inherit Benchmarks<'elem>(
         buildFunToBenchmark,
-        converter: string -> 'elem,
+        converter,
         boolConverter,
-        vertex) =
-
-        inherit Benchmarks<'elem>(
-            buildFunToBenchmark,
-            converter,
-            boolConverter,
-            vertex)
-
-        [<GlobalSetup>]
-        override this.GlobalSetup() =
-            this.ReadMatrix()
-
-        [<GlobalCleanup>]
-        override this.GlobalCleanup() =
-            this.ClearResult()
-
-        [<IterationCleanup>]
-        override this.IterationCleanup() =
-            this.ClearInputMatrix()
-            this.ClearResult()
-
-        [<Benchmark>]
-        override this.Benchmark() =
-            this.LoadMatrixToGPU()
-            this.BFS()
-            this.ResultLevels.ToHost this.Processor |> ignore
-            this.Processor.PostAndReply Msg.MsgNotifyMe
-
-    type Int() =
-
-        inherit Benchmark<int>(
-            (singleSource ArithmeticOperations.intSumOption ArithmeticOperations.intMulOption),
-            int32,
-            (fun _ -> Utils.nextInt (System.Random())),
-            0)
-
-        static member InputMatrixProvider =
-            Benchmark<_>.InputMatrixProviderBuilder "BFSBenchmarks.txt"
+        vertex)
+
+    [<GlobalSetup>]
+    override this.GlobalSetup() =
+        this.ReadMatrix()
+
+    [<GlobalCleanup>]
+    override this.GlobalCleanup() =
+        this.ClearResult()
+
+    [<IterationCleanup>]
+    override this.IterationCleanup() =
+        this.ClearInputMatrix()
+        this.ClearResult()
+
+    [<Benchmark>]
+    override this.Benchmark() =
+        this.LoadMatrixToGPU()
+        this.BFS()
+        this.ResultLevels.ToHost this.Processor |> ignore
+        this.Processor.PostAndReply Msg.MsgNotifyMe
+
+type BFSWithTransferBenchmarkInt32() =
+
+    inherit WithTransferBenchmark<int>(
+        (singleSource ArithmeticOperations.intSumOption ArithmeticOperations.intMulOption),
+        int32,
+        (fun _ -> Utils.nextInt (System.Random())),
+        0)
+
+    static member InputMatrixProvider =
+        Benchmarks<_>.InputMatrixProviderBuilder "BFSBenchmarks.txt"
 

benchmarks/GraphBLAS-sharp.Benchmarks/Configs/Context.txt

@@ -1,3 +1,7 @@
+
+Gpu
+32
+
 NVIDIA*
 Gpu
 32

benchmarks/GraphBLAS-sharp.Benchmarks/Configs/SpGeMM.txt

@@ -0,0 +1 @@
+hollywood-2009.mtx

benchmarks/GraphBLAS-sharp.Benchmarks/GraphBLAS-sharp.Benchmarks.fsproj

@@ -26,6 +26,7 @@
     <Compile Include="Vector/Map2.fs" />
     <Compile Include="Algorithms/BFS.fs" />
     <Compile Include="Program.fs" />
+    <Folder Include="Datasets" />
   </ItemGroup>
   <Import Project="..\..\.paket\Paket.Restore.targets" />
 </Project>

benchmarks/GraphBLAS-sharp.Benchmarks/Helpers.fs

@@ -108,13 +108,16 @@ module Utils =
                     let queue = context.QueueProvider.CreateQueue()
 
                     { ClContext = context; Queue = queue })
+        let result =
+            seq {
+                for wgSize in workGroupSizes do
+                    for context in contexts do
+                        yield (context, wgSize)
+            }
 
-        seq {
-            for wgSize in workGroupSizes do
-                for context in contexts do
-                    yield (context, wgSize)
-        }
+        printfn "result length: %A" <| Seq.length result
 
+        result
     let nextSingle (random: System.Random) =
         let buffer = Array.zeroCreate<byte> 4
         random.NextBytes buffer

benchmarks/GraphBLAS-sharp.Benchmarks/Matrix/SpGeMM/Expand.fs

@@ -49,7 +49,7 @@ type Benchmarks<'elem when 'elem : struct>(
     static member AvailableContexts = Utils.availableContexts
 
     static member InputMatrixProviderBuilder pathToConfig =
-        let datasetFolder = "Mxm"
+        let datasetFolder = ""
         pathToConfig
         |> Utils.getMatricesFilenames
         |> Seq.map
@@ -59,7 +59,6 @@ type Benchmarks<'elem when 'elem : struct>(
                 match Path.GetExtension matrixFilename with
                 | ".mtx" ->
                     MtxReader(Utils.getFullPathToMatrix datasetFolder matrixFilename)
-                    , MtxReader(Utils.getFullPathToMatrix datasetFolder (matrixFilename))
                 | _ -> failwith "Unsupported matrix format")
 
     member this.FunToBenchmark =
@@ -145,4 +144,4 @@ module WithoutTransfer =
             )
 
         static member InputMatrixProvider =
-            Benchmarks<_>.InputMatrixProviderBuilder "MxmBenchmarks4Float32.txt"
+            Benchmarks<_>.InputMatrixProviderBuilder "SpGeMM.txt"

benchmarks/GraphBLAS-sharp.Benchmarks/Program.fs

@@ -4,7 +4,8 @@ open BenchmarkDotNet.Running
 [<EntryPoint>]
 let main argv =
     let benchmarks =
-        BenchmarkSwitcher [| typeof<Algorithms.BFS.WithoutTransfer.Int> |]
+        BenchmarkSwitcher [| typeof<Algorithms.BFS.BFSWithoutTransferBenchmarkInt32>
+                             typeof<Matrix.SpGeMM.Expand.WithoutTransfer.Float32> |]
 
     benchmarks.Run argv |> ignore
     0

src/GraphBLAS-sharp.Backend/Matrix/CSR/Matrix.fs

@@ -167,7 +167,7 @@ module Matrix =
         let pairwise = ClArray.pairwise clContext workGroupSize
 
         let subtract =
-            ClArray.map clContext workGroupSize <@ fun (fst, snd) -> snd - fst @>
+            ClArray.map <@ fun (fst, snd) -> snd - fst @> clContext workGroupSize
 
         fun (processor: MailboxProcessor<_>) allocationMode (matrix: ClMatrix.CSR<'b>) ->
             let pointerPairs =

src/GraphBLAS-sharp.Backend/Vector/Sparse/Vector.fs

@@ -65,6 +65,6 @@ module Vector =
     let reduce<'a when 'a: struct> (opAdd: Expr<'a -> 'a -> 'a>) (clContext: ClContext) workGroupSize =
 
         let reduce =
-            Reduce.reduce clContext workGroupSize opAdd
+            Reduce.reduce opAdd clContext workGroupSize
 
         fun (processor: MailboxProcessor<_>) (vector: ClVector.Sparse<'a>) -> reduce processor vector.Values

src/GraphBLAS-sharp.Backend/Vector/Vector.fs

@@ -39,7 +39,7 @@ module Vector =
             ClArray.zeroCreate clContext workGroupSize
 
         let map =
-            ClArray.map clContext workGroupSize <@ Some @>
+            ClArray.map <@ Some @> clContext workGroupSize
 
         fun (processor: MailboxProcessor<_>) allocationMode format size (elements: (int * 'a) list) ->
             match format with