youmuliu/stgsharp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Better FMA

Browse Source
This commit is contained in:
Youmu Liu
2026-03-05 18:04:34 +08:00
parent dfa6f4ba92
commit 83997f64d0
21 changed files with 893 additions and 1030 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Mathematics/Numeric/MatrixCompute.double.cs
+13 -13
View File
@@ -58,7 +58,7 @@ namespace StgSharp.Mathematics.Numeric
MatrixKernel<double>* leftPtr = left.Buffer;
MatrixKernel<double>* rightPtr = right.Buffer;
MatrixKernel<double>* ansPtr = ans.Buffer;
MathematicsModule.GlobalContext.mat_sk[f64].buffer_add(leftPtr, rightPtr, ansPtr,
NumericalModule.GlobalContext.mat_sk[f64].buffer_add(leftPtr, rightPtr, ansPtr,
count);
return;
@@ -69,7 +69,7 @@ namespace StgSharp.Mathematics.Numeric
MatrixKernel<double>* leftPtr = left.Buffer;
MatrixKernel<double>* rightPtr = right.Buffer;
MatrixKernel<double>* ansPtr = ans.Buffer;
MathematicsModule.GlobalContext.mat_sk[f64].buffer_add(leftPtr, rightPtr, ansPtr,
NumericalModule.GlobalContext.mat_sk[f64].buffer_add(leftPtr, rightPtr, ansPtr,
count);
MatrixParallel.ReturnParallelResources(ref pool);
@@ -83,7 +83,7 @@ namespace StgSharp.Mathematics.Numeric
package->PrimCount = ans.KernelColumnLength;
package->SecCount = ans.KernelRowLength;
package->ComputeMode = new MatrixOpMode(MatrixIndexStyle.BUFFER_INDEX, MatrixOperationParam.LEFT_RIGHT_ANS_PARAM);
package->ComputeHandle = (IntPtr)MathematicsModule.GlobalContext.mat_sk[f64].buffer_add;
package->ComputeHandle = (IntPtr)NumericalModule.GlobalContext.mat_sk[f64].buffer_add;
MatrixParallelWrap taskGroup = MatrixParallel.ScheduleTask(package);
taskGroup.LaunchParallel(SleepMode.DeepSleep);
@@ -102,12 +102,12 @@ namespace StgSharp.Mathematics.Numeric
scalar->Data<double>(0) = value;
if (count <= 256)
{
MathematicsModule.GlobalContext.mat_sk[f64].buffer_fill(ans.Buffer, scalar, count);
NumericalModule.GlobalContext.mat_sk[f64].buffer_fill(ans.Buffer, scalar, count);
return;
} else if (count <= 1024)
{
MatrixParallelThread[] pool = MatrixParallel.LeadParallel(1);
MathematicsModule.GlobalContext.mat_sk[f64].buffer_fill(ans.Buffer, scalar, count);
NumericalModule.GlobalContext.mat_sk[f64].buffer_fill(ans.Buffer, scalar, count);
MatrixParallel.ReturnParallelResources(ref pool);
return;
}
@@ -117,7 +117,7 @@ namespace StgSharp.Mathematics.Numeric
package->PrimCount = ans.KernelColumnLength;
package->SecCount = ans.KernelRowLength;
package->ComputeMode = new MatrixOpMode(MatrixIndexStyle.BUFFER_INDEX, MatrixOperationParam.ANS_SCALAR_PARAM);
package->ComputeHandle = (IntPtr)MathematicsModule.GlobalContext.mat_sk[f64].buffer_fill;
package->ComputeHandle = (IntPtr)NumericalModule.GlobalContext.mat_sk[f64].buffer_fill;
package->Scalar = scalar;
MatrixParallelWrap taskGroup = MatrixParallel.ScheduleTask(package);
taskGroup.LaunchParallel(SleepMode.DeepSleep);
@@ -158,7 +158,7 @@ namespace StgSharp.Mathematics.Numeric
enumeration->Member<int>(0) = i;
enumeration->Member<int>(1) = j;
MathematicsModule.GlobalContext
NumericalModule.GlobalContext
.mat_sk[f64].kernel_tile_fma((MatrixKernel*)left.Buffer,
(MatrixKernel*)right.Buffer,
(MatrixKernel*)ans.Buffer,
@@ -209,7 +209,7 @@ namespace StgSharp.Mathematics.Numeric
{
MatrixKernel<double>* leftPtr = left.Buffer;
MatrixKernel<double>* ansPtr = ans.Buffer;
MathematicsModule.GlobalContext.mat_sk[f64].buffer_scalar_mul(leftPtr, ansPtr, scalar,
NumericalModule.GlobalContext.mat_sk[f64].buffer_scalar_mul(leftPtr, ansPtr, scalar,
count);
return;
@@ -219,7 +219,7 @@ namespace StgSharp.Mathematics.Numeric
MatrixKernel<double>* leftPtr = left.Buffer;
MatrixKernel<double>* ansPtr = ans.Buffer;
MathematicsModule.GlobalContext.mat_sk[f64].buffer_scalar_mul(leftPtr, ansPtr, scalar,
NumericalModule.GlobalContext.mat_sk[f64].buffer_scalar_mul(leftPtr, ansPtr, scalar,
count);
MatrixParallel.ReturnParallelResources(ref pool);
@@ -233,7 +233,7 @@ namespace StgSharp.Mathematics.Numeric
package->SecCount = left.KernelRowLength;
package->Scalar = scalar;
package->ComputeMode = new MatrixOpMode(MatrixIndexStyle.BUFFER_INDEX, MatrixOperationParam.RIGHT_ANS_SCALAR_PARAM);
package->ComputeHandle = (IntPtr)MathematicsModule.GlobalContext.mat_sk[f64].buffer_scalar_mul;
package->ComputeHandle = (IntPtr)NumericalModule.GlobalContext.mat_sk[f64].buffer_scalar_mul;
MatrixParallelWrap taskGroup = MatrixParallel.ScheduleTask(package);
taskGroup.LaunchParallel(SleepMode.DeepSleep);
@@ -266,7 +266,7 @@ namespace StgSharp.Mathematics.Numeric
MatrixKernel<double>* leftPtr = left.Buffer;
MatrixKernel<double>* rightPtr = right.Buffer;
MatrixKernel<double>* ansPtr = ans.Buffer;
MathematicsModule.GlobalContext.mat_sk[f64].buffer_sub(leftPtr, rightPtr, ansPtr,
NumericalModule.GlobalContext.mat_sk[f64].buffer_sub(leftPtr, rightPtr, ansPtr,
count);
return;
@@ -277,7 +277,7 @@ namespace StgSharp.Mathematics.Numeric
MatrixKernel<double>* leftPtr = left.Buffer;
MatrixKernel<double>* rightPtr = right.Buffer;
MatrixKernel<double>* ansPtr = ans.Buffer;
MathematicsModule.GlobalContext.mat_sk[f64].buffer_sub(leftPtr, rightPtr, ansPtr,
NumericalModule.GlobalContext.mat_sk[f64].buffer_sub(leftPtr, rightPtr, ansPtr,
count);
MatrixParallel.ReturnParallelResources(ref pool);
@@ -291,7 +291,7 @@ namespace StgSharp.Mathematics.Numeric
package->PrimCount = left.KernelColumnLength;
package->SecCount = left.KernelRowLength;
package->ComputeMode = new MatrixOpMode(MatrixIndexStyle.BUFFER_INDEX, MatrixOperationParam.LEFT_RIGHT_ANS_PARAM);
package->ComputeHandle = (IntPtr)MathematicsModule.GlobalContext.mat_sk[f64].buffer_sub;
package->ComputeHandle = (IntPtr)NumericalModule.GlobalContext.mat_sk[f64].buffer_sub;
MatrixParallelWrap taskGroup = MatrixParallel.ScheduleTask(package);
taskGroup.LaunchParallel(SleepMode.DeepSleep);