mlemanczyk/even-perfect-numbers-scanner
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Inline batch copies for GPU prime tester

Browse Source
This commit is contained in:
Marek Lemańczyk committed 2025-10-29 19:21:32 +01:00
1 parent 2b0fc5be13
commit c61df1c33b
1 file changed
+17 -13
Download Patch File Download Diff File Expand all files Collapse all files
PerfectNumbers.Core/PrimeTester.cs
+17 -13
View File
@@ -4,6 +4,7 @@ using System.Collections.Concurrent;
using System.Collections.Generic; using System.Collections.Generic;
using System.Numerics; using System.Numerics;
using System.Runtime.CompilerServices; using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using ILGPU; using ILGPU;
using ILGPU.Runtime; using ILGPU.Runtime;
using PerfectNumbers.Core.Gpu; using PerfectNumbers.Core.Gpu;
@@ -117,11 +118,13 @@ public sealed class PrimeTester
var input = gpu.Input; var input = gpu.Input;
var output = gpu.Output; var output = gpu.Output;
var inputView = input.View;
var outputView = output.View;
input.View.CopyFromCPU(ref value, 1); inputView.CopyFromCPU(ref value, 1);
kernel( kernel(
1, 1,
input.View, inputView,
gpu.DevicePrimesDefault.View, gpu.DevicePrimesDefault.View,
gpu.DevicePrimesLastOne.View, gpu.DevicePrimesLastOne.View,
gpu.DevicePrimesLastSeven.View, gpu.DevicePrimesLastSeven.View,
@@ -132,9 +135,9 @@ public sealed class PrimeTester
gpu.DevicePrimesPow2LastSeven.View, gpu.DevicePrimesPow2LastSeven.View,
gpu.DevicePrimesPow2LastThree.View, gpu.DevicePrimesPow2LastThree.View,
gpu.DevicePrimesPow2LastNine.View, gpu.DevicePrimesPow2LastNine.View,
output.View); outputView);
accelerator.Synchronize(); accelerator.Synchronize();
output.View.CopyToCPU(ref flag, 1); outputView.CopyToCPU(ref flag, 1);
gpu.Dispose(); gpu.Dispose();
limiter.Dispose(); limiter.Dispose();
@@ -163,8 +166,8 @@ public static bool IsPrimeCpu(ulong n, CancellationToken ct)
var input = gpu.Input; var input = gpu.Input;
var output = gpu.Output; var output = gpu.Output;
ArrayPool<ulong> pool = ThreadStaticPools.UlongPool; var inputView = input.View;
ulong[] temp = pool.Rent(batchSize); var outputView = output.View;
int pos = 0; int pos = 0;
while (pos < totalLength) while (pos < totalLength)
@@ -172,12 +175,13 @@ public static bool IsPrimeCpu(ulong n, CancellationToken ct)
int remaining = totalLength - pos; int remaining = totalLength - pos;
int count = remaining > batchSize ? batchSize : remaining; int count = remaining > batchSize ? batchSize : remaining;
values.Slice(pos, count).CopyTo(temp); var valueSlice = values.Slice(pos, count);
input.View.CopyFromCPU(ref temp[0], count); ref ulong valueRef = ref MemoryMarshal.GetReference(valueSlice);
inputView.CopyFromCPU(ref valueRef, count);
kernel( kernel(
count, count,
input.View, inputView,
gpu.DevicePrimesDefault.View, gpu.DevicePrimesDefault.View,
gpu.DevicePrimesLastOne.View, gpu.DevicePrimesLastOne.View,
gpu.DevicePrimesLastSeven.View, gpu.DevicePrimesLastSeven.View,
@@ -188,15 +192,15 @@ public static bool IsPrimeCpu(ulong n, CancellationToken ct)
gpu.DevicePrimesPow2LastSeven.View, gpu.DevicePrimesPow2LastSeven.View,
gpu.DevicePrimesPow2LastThree.View, gpu.DevicePrimesPow2LastThree.View,
gpu.DevicePrimesPow2LastNine.View, gpu.DevicePrimesPow2LastNine.View,
output.View); outputView);
accelerator.Synchronize(); accelerator.Synchronize();
output.View.CopyToCPU(ref results[pos], count); var resultSlice = results.Slice(pos, count);
ref byte resultRef = ref MemoryMarshal.GetReference(resultSlice);
outputView.CopyToCPU(ref resultRef, count);
pos += count; pos += count;
} }
pool.Return(temp, clearArray: false);
gpu.Dispose(); gpu.Dispose();
limiter.Dispose(); limiter.Dispose();
} }