marcel/Ryujinx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Amadeus: Add ARM SIMD fast path (#2069)

Browse Source 
Add fast paths in the audio renderer for AArch64 in all current fast paths.
This commit is contained in:
Mary
2021-03-02 23:50:46 +01:00
committed by GitHub
parent 3d04d7ef08
commit 31fca432a7
3 changed files with 97 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
Ryujinx.Audio/Renderer/Dsp/Command/MixCommand.cs
View File

@ -19,6 +19,7 @@ using System;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.Arm;
using System.Runtime.Intrinsics.X86;
namespace Ryujinx.Audio.Renderer.Dsp.Command
@ -89,6 +90,27 @@ namespace Ryujinx.Audio.Renderer.Dsp.Command
}
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private void ProcessMixAdvSimd(Span<float> outputMix, ReadOnlySpan<float> inputMix)
{
Vector128<float> volumeVec = Vector128.Create(Volume);
ReadOnlySpan<Vector128<float>> inputVec = MemoryMarshal.Cast<float, Vector128<float>>(inputMix);
Span<Vector128<float>> outputVec = MemoryMarshal.Cast<float, Vector128<float>>(outputMix);
int sisdStart = inputVec.Length * 4;
for (int i = 0; i < inputVec.Length; i++)
{
outputVec[i] = AdvSimd.Add(outputVec[i], AdvSimd.Ceiling(AdvSimd.Multiply(inputVec[i], volumeVec)));
}
for (int i = sisdStart; i < inputMix.Length; i++)
{
outputMix[i] += FloatingPointHelper.MultiplyRoundUp(inputMix[i], Volume);
}
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private void ProcessMixSlowPath(Span<float> outputMix, ReadOnlySpan<float> inputMix)
{
@ -108,6 +130,10 @@ namespace Ryujinx.Audio.Renderer.Dsp.Command
{
ProcessMixSse41(outputMix, inputMix);
}
else if (AdvSimd.IsSupported)
{
ProcessMixAdvSimd(outputMix, inputMix);
}
else
{
ProcessMixSlowPath(outputMix, inputMix);