timyhac
commented
2 weeks ago @mitchrazga - this idea is worth exploring and I would be keen to see what improvements this would make possible.
I had a go at benchmarking this but was not able to see any allocation (other than the source buffer, which won't appear in the measurement) when using the current native method declaration nor when using a pointer-based one. This makes sense to me; that P/Invoke merely pins the buffer in place as opposed to making a copy.
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Jobs;
using BenchmarkDotNet.Running;
using System;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using System.Security;
public class Program
{
public static void Main(string[] args)
{
var summary = BenchmarkRunner.Run<Tests>();
}
}
[SimpleJob(RuntimeMoniker.Net472, baseline: true)]
[SimpleJob(RuntimeMoniker.Net80)]
[MemoryDiagnoser]
public class Tests
{
private byte[] data;
private int tag_id;
[Params(10, 30, 100, 300, 1000)]
public int N;
[GlobalSetup]
public void Setup()
{
tag_id = Native.plc_tag_create("protocol=ab-eip&gateway=%s&path=1,0&plc=controllogix&name=@raw", 1000);
var result = Native.plc_tag_set_size(tag_id, N);
data = new byte[N];
}
[Benchmark]
public int Original()
{
return plctag.Original(tag_id, 0, data, data.Length);
}
[Benchmark]
public int BytePtr()
{
return plctag.BytePtr(tag_id, 0, data);
}
}
public static class plctag
{
public static int Original(Int32 tag_id, int start_offset, byte[] buffer, int buffer_length)
{
return Native.Original(tag_id, start_offset, buffer, buffer_length);
}
public static int BytePtr(Int32 tag_id, int start_offset, Span<byte> buffer)
{
unsafe
{
fixed (byte* ptr = buffer)
{
return Native.BytePtr(tag_id, start_offset, ptr, buffer.Length);
}
}
}
}
[SuppressUnmanagedCodeSecurity]
static class Native
{
const string DLL_NAME = "plctag";
const string ENTRYPOINT = "plc_tag_get_raw_bytes";
[DllImport(DLL_NAME, EntryPoint = nameof(plc_tag_create), CallingConvention = CallingConvention.Cdecl, ExactSpelling = true, CharSet = CharSet.Ansi)]
public static extern Int32 plc_tag_create([MarshalAs(UnmanagedType.LPStr)] string lpString, int timeout);
[DllImport(DLL_NAME, EntryPoint = nameof(plc_tag_set_size), CallingConvention = CallingConvention.Cdecl, ExactSpelling = true)]
public static extern int plc_tag_set_size(Int32 tag, int new_size);
[DllImport(DLL_NAME, EntryPoint = ENTRYPOINT, CallingConvention = CallingConvention.Cdecl, ExactSpelling = true)]
public static extern int Original(Int32 tag_id, int start_offset, [Out] byte[] buffer, int buffer_length);
[DllImport(DLL_NAME, EntryPoint = ENTRYPOINT, CallingConvention = CallingConvention.Cdecl, ExactSpelling = true)]
public static extern unsafe int BytePtr(Int32 tag_id, int start_offset, byte* buffer, int buffer_length);
}The results of this are:
// * Summary *
BenchmarkDotNet v0.14.0, Windows 11 (10.0.22631.4249/23H2/2023Update/SunValley3)
11th Gen Intel Core i7-1195G7 2.90GHz, 1 CPU, 8 logical and 4 physical cores
.NET SDK 9.0.100-rc.1.24452.12
[Host] : .NET 8.0.10 (8.0.1024.46610), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
.NET 8.0 : .NET 8.0.10 (8.0.1024.46610), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
.NET Framework 4.7.2 : .NET Framework 4.8.1 (4.8.9277.0), X64 RyuJIT VectorSize=256
| Method | Job | Runtime | N | Mean | Error | StdDev | Ratio | RatioSD | Allocated | Alloc Ratio |
|--------- |--------------------- |--------------------- |----- |-----------:|---------:|---------:|------:|--------:|----------:|------------:|
| Original | .NET 8.0 | .NET 8.0 | 10 | 612.1 ns | 9.88 ns | 9.24 ns | 1.01 | 0.02 | - | NA |
| Original | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 10 | 608.7 ns | 9.68 ns | 8.08 ns | 1.00 | 0.02 | - | NA |
| | | | | | | | | | | |
| BytePtr | .NET 8.0 | .NET 8.0 | 10 | 605.0 ns | 11.80 ns | 11.59 ns | 0.94 | 0.02 | - | NA |
| BytePtr | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 10 | 643.1 ns | 11.88 ns | 10.53 ns | 1.00 | 0.02 | - | NA |
| | | | | | | | | | | |
| Original | .NET 8.0 | .NET 8.0 | 30 | 649.8 ns | 11.00 ns | 10.28 ns | 0.91 | 0.02 | - | NA |
| Original | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 30 | 713.3 ns | 13.79 ns | 14.76 ns | 1.00 | 0.03 | - | NA |
| | | | | | | | | | | |
| BytePtr | .NET 8.0 | .NET 8.0 | 30 | 682.8 ns | 10.50 ns | 9.31 ns | 0.93 | 0.02 | - | NA |
| BytePtr | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 30 | 737.1 ns | 14.55 ns | 13.61 ns | 1.00 | 0.03 | - | NA |
| | | | | | | | | | | |
| Original | .NET 8.0 | .NET 8.0 | 100 | 759.0 ns | 13.78 ns | 12.22 ns | 1.09 | 0.03 | - | NA |
| Original | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 100 | 694.8 ns | 11.25 ns | 13.40 ns | 1.00 | 0.03 | - | NA |
| | | | | | | | | | | |
| BytePtr | .NET 8.0 | .NET 8.0 | 100 | 731.4 ns | 14.61 ns | 15.01 ns | 1.08 | 0.03 | - | NA |
| BytePtr | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 100 | 679.6 ns | 13.56 ns | 15.61 ns | 1.00 | 0.03 | - | NA |
| | | | | | | | | | | |
| Original | .NET 8.0 | .NET 8.0 | 300 | 743.1 ns | 14.55 ns | 14.29 ns | 0.90 | 0.02 | - | NA |
| Original | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 300 | 830.3 ns | 14.75 ns | 12.31 ns | 1.00 | 0.02 | - | NA |
| | | | | | | | | | | |
| BytePtr | .NET 8.0 | .NET 8.0 | 300 | 764.9 ns | 14.71 ns | 12.28 ns | 0.90 | 0.04 | - | NA |
| BytePtr | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 300 | 850.1 ns | 17.03 ns | 33.61 ns | 1.00 | 0.05 | - | NA |
| | | | | | | | | | | |
| Original | .NET 8.0 | .NET 8.0 | 1000 | 1,045.4 ns | 19.85 ns | 18.56 ns | 1.04 | 0.03 | - | NA |
| Original | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 1000 | 1,006.0 ns | 18.89 ns | 17.67 ns | 1.00 | 0.02 | - | NA |
| | | | | | | | | | | |
| BytePtr | .NET 8.0 | .NET 8.0 | 1000 | 1,007.1 ns | 19.01 ns | 18.67 ns | 1.01 | 0.03 | - | NA |
| BytePtr | .NET Framework 4.7.2 | .NET Framework 4.7.2 | 1000 | 998.2 ns | 18.87 ns | 22.46 ns | 1.00 | 0.03 | - | NA |
MitchRazga
Would be beneficial if there were
(ReadOnly)Span<byte>overloads to reduce allocations and make using existing buffers easier. I'm happy to do a PR to implement if you are interested.The native method declarations in C# would need to change since they use
byte[]but won't be too much trouble since the native API uses a pointer anyway. https://github.com/libplctag/libplctag.NET/blob/e28ec36addeee8eae304607b49be52dccab1fcea/src/libplctag.NativeImport/NativeMethods.cs#L247-L251https://github.com/libplctag/libplctag/blob/c55bc5876d938dda1c609750cde5ae4812d7b8a8/src/lib/libplctag.h#L505-L506
There are a few ways to do this, one approach can be found in Memory\ and Span\ usage guidelines under:
Alternatively and ideally, would use P/Invoke source generation but this requires .NET 7+. Something like:
This would also make it possible to create
ReadAsyncandWriteAsyncwithMemory<byte>overloads too. Similar to Stream.ReadAsync-system-threading-cancellationtoken)) and Stream.WriteAsync-system-threading-cancellationtoken))Originally posted by @MitchRazga in https://github.com/libplctag/libplctag.NET/issues/394#issuecomment-2410778370