back to matrix, but trouble in combining L4 predict with matrixparallelqueue

2026-03-19 19:53:22 +08:00
parent fd82829788
commit e521bed222
10 changed files with 291 additions and 142 deletions
@@ -0,0 +1,129 @@
+//-----------------------------------------------------------------------
+// -----------------------------------------------------------------------
+// file="MatrixCompute.Sequential"
+// Project: StgSharp
+// AuthorGroup: Nitload
+// Copyright (c) Nitload. All rights reserved.
+//     
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+//     
+// -----------------------------------------------------------------------
+// -----------------------------------------------------------------------
+using StgSharp.HighPerformance.Memory;
+using StgSharp.Mathematics.Numeric.Runtime;
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Text;
+using System.Threading.Tasks;
+
+using unsafe MkProc = delegate* unmanaged[Cdecl]<StgSharp.Mathematics.Internal.MatrixParallelTask*, void>;
+
+namespace StgSharp.Mathematics.Numeric.Linear
+{
+    public static unsafe partial class MatrixCompute
+    {
+
+        public static void Add(
+                           Matrix<float> left,
+                           Matrix<float> right,
+                           Matrix<float> ans
+        )
+        {
+            ArgumentNullException.ThrowIfNull(left, nameof(left));
+            ArgumentNullException.ThrowIfNull(right, nameof(right));
+            ArgumentNullException.ThrowIfNull(ans, nameof(ans));
+            if ((left.ColumnLength != right.ColumnLength) ||
+                (left.RowLength != right.RowLength) ||
+                (left.ColumnLength != ans.ColumnLength) ||
+                (left.RowLength != ans.RowLength)) {
+                throw new MatrixDimensionMismatchException<float>(
+                    left, right, ans
+                    );
+            }
+            MatrixElementType f32 = MatrixElementType.F32;
+            long count = ((long)left.KernelColumnLength) * left.KernelRowLength;
+
+            MatrixKernel<float>* leftPtr = left.Buffer;
+            MatrixKernel<float>* rightPtr = right.Buffer;
+            MatrixKernel<float>* ansPtr = ans.Buffer;
+            MkProc add = (MkProc)NumericalModule.GlobalContext.
+                mat_mk[f32.IntrinsicNode].
+                Get(MatrixIntrinsicHandle.Add);
+            MatrixParallelTask* t = stackalloc MatrixParallelTask[1];
+
+            t->Mat1 = (MatrixKernel*)leftPtr;
+            t->Mat2 = (MatrixKernel*)rightPtr;
+            t->Mat3 = (MatrixKernel*)ansPtr;
+            t->ElementType = MatrixElementType.F64;
+            t->Scalar.Data<long>(0) = count;
+            if (count <= 256)
+            {
+                add(t);
+
+                // .buffer_add(leftPtr, rightPtr, ansPtr, count);
+
+                return;
+            } else if (count <= 1024)
+            {
+                MatrixParallelThread[] pool = MatrixParallel.LeadParallel(1);
+                add(t);
+                MatrixParallel.ReturnParallelResources(ref pool);
+                return;
+            }
+            MatrixParallelWrap taskGroup = MatrixParallel.ScheduleTask<SequentialPredictor>((nuint)add, t);
+
+            taskGroup.LaunchParallel(SleepMode.DeepSleep);
+            taskGroup.Dispose();
+            return;
+        }
+
+        private sealed class SequentialPredictor : MatrixPredictor<SequentialPredictor>
+        {
+
+            public override bool Predict(
+                                 Span<CacheLinePrediction> node,
+                                 out int count
+            )
+            {
+                throw new NotImplementedException();
+            }
+
+            public override void Prefetch(
+                                 nuint origin,
+                                 ulong policy,
+                                 Span<byte> cache
+            )
+            {
+                throw new NotImplementedException();
+            }
+
+            public override void WriteBack(
+                                 nuint origin,
+                                 ulong policy,
+                                 ReadOnlySpan<byte> cache
+            )
+            {
+                throw new NotImplementedException();
+            }
+
+        }
+
+    }
+}
@@ -25,6 +25,7 @@
 //     
 // -----------------------------------------------------------------------
 // -----------------------------------------------------------------------
+using StgSharp.HighPerformance.Memory;
 using StgSharp.Internal;
 using StgSharp.Mathematics;
 using System;
@@ -37,7 +38,7 @@ using System.Threading.Tasks;

 namespace StgSharp.Mathematics.Numeric
 {
-    public partial class MatrixCompute
+    public static partial class MatrixCompute
    {

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
@@ -45,8 +46,8 @@ namespace StgSharp.Mathematics.Numeric
                                             MatrixKernel<T>* buffer,
                                             int colLength,
                                             int x,
-                                             int y)
-            where T: unmanaged, INumber<T>
+                                             int y
+        ) where T : unmanaged, INumber<T>
        {
            return (MatrixKernel*)(buffer + (colLength * x + y));
        }