Misaki/Misaki.HighPerformance
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add Vector type in SPMD to total of load vector size * lane width of number into memory for simd calculation.

Browse Source
This commit is contained in:
Misaki
2026-02-13 21:47:05 +09:00
parent 75d33d0763
commit 4f964b2d2a
22 changed files with 3682 additions and 447 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
Misaki.HighPerformance.Mathematics.SPMD/IJobSPMD.cs
View File

@@ -5,7 +5,7 @@ using System.Runtime.CompilerServices;
namespace Misaki.HighPerformance.Mathematics.SPMD;
public interface IJobSPMD<TNumber>
where TNumber : unmanaged, INumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
where TNumber : unmanaged, INumber<TNumber>, IBinaryNumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
{
void Execute<TLane>(int baseIndex, int threadIndex)
where TLane : ISPMD<TLane, TNumber>;
@@ -13,7 +13,7 @@ public interface IJobSPMD<TNumber>
internal struct SPMDJobWrapper<T, TNumber> : IJobParallelFor
where T : unmanaged, IJobSPMD<TNumber>
where TNumber : unmanaged, INumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
where TNumber : unmanaged, INumber<TNumber>, IBinaryNumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
{
public T innerJob;
public int totalCount;
@@ -40,7 +40,7 @@ internal struct SPMDJobWrapper<T, TNumber> : IJobParallelFor
internal struct SPMDScalerJobWrapper<T, TNumber> : IJobParallelFor
where T : unmanaged, IJobSPMD<TNumber>
where TNumber : unmanaged, INumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
where TNumber : unmanaged, INumber<TNumber>, IBinaryNumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
{
public T innerJob;
public int totalCount;
@@ -55,32 +55,42 @@ internal struct SPMDScalerJobWrapper<T, TNumber> : IJobParallelFor
public static class IJobParallelForSPMDExtensions
{
public static void Run<T, TNumber>(this ref T job, int totalCount, int threadIndex)
where T : unmanaged, IJobSPMD<TNumber>
where TNumber : unmanaged, INumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
where T : struct, IJobSPMD<TNumber>
where TNumber : unmanaged, INumber<TNumber>, IBinaryNumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
{
var iterations = (totalCount + WideLane<TNumber>.LaneWidth - 1) / WideLane<TNumber>.LaneWidth;
for (var loopIndex = 0; loopIndex < iterations; loopIndex++)
if (WideLane.IsSupported)
{
var baseIndex = loopIndex * WideLane<TNumber>.LaneWidth;
var remaining = totalCount - baseIndex;
var iterations = (totalCount + WideLane<TNumber>.LaneWidth - 1) / WideLane<TNumber>.LaneWidth;
for (var loopIndex = 0; loopIndex < iterations; loopIndex++)
{
var baseIndex = loopIndex * WideLane<TNumber>.LaneWidth;
var remaining = totalCount - baseIndex;
if (remaining >= WideLane<TNumber>.LaneWidth)
{
job.Execute<WideLane<TNumber>>(baseIndex, threadIndex);
}
else
{
for (var i = 0; i < remaining; i++)
if (remaining >= WideLane<TNumber>.LaneWidth)
{
job.Execute<ScalarLane<TNumber>>(baseIndex + i, threadIndex);
job.Execute<WideLane<TNumber>>(baseIndex, threadIndex);
}
else
{
for (var i = 0; i < remaining; i++)
{
job.Execute<ScalarLane<TNumber>>(baseIndex + i, threadIndex);
}
}
}
}
else
{
for (var loopIndex = 0; loopIndex < totalCount; loopIndex++)
{
job.Execute<ScalarLane<TNumber>>(loopIndex, threadIndex);
}
}
}
public static JobHandle ScheduleParallelSPDM<T, TNumber>(this JobScheduler jobScheduler, ref T job, int totalCount, int batchSize, int threadIndex, JobHandle dependency)
where T : unmanaged, IJobSPMD<TNumber>
where TNumber : unmanaged, INumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
where TNumber : unmanaged, INumber<TNumber>, IBinaryNumber<TNumber>, IMinMaxValue<TNumber>, IBitwiseOperators<TNumber, TNumber, TNumber>
{
if (WideLane.IsSupported)
{