Amadeus: Add ARM SIMD fast path (#2069)

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
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 97 additions and 1 deletions

View file

@ -19,6 +19,7 @@ using System;
using System.Runtime.CompilerServices; using System.Runtime.CompilerServices;
using System.Runtime.InteropServices; using System.Runtime.InteropServices;
using System.Runtime.Intrinsics; using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.Arm;
using System.Runtime.Intrinsics.X86; using System.Runtime.Intrinsics.X86;
namespace Ryujinx.Audio.Renderer.Dsp.Command 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)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
private void ProcessMixSlowPath(Span<float> outputMix, ReadOnlySpan<float> inputMix) private void ProcessMixSlowPath(Span<float> outputMix, ReadOnlySpan<float> inputMix)
{ {
@ -108,6 +130,10 @@ namespace Ryujinx.Audio.Renderer.Dsp.Command
{ {
ProcessMixSse41(outputMix, inputMix); ProcessMixSse41(outputMix, inputMix);
} }
else if (AdvSimd.IsSupported)
{
ProcessMixAdvSimd(outputMix, inputMix);
}
else else
{ {
ProcessMixSlowPath(outputMix, inputMix); ProcessMixSlowPath(outputMix, inputMix);

View file

@ -19,6 +19,7 @@ using System;
using System.Runtime.CompilerServices; using System.Runtime.CompilerServices;
using System.Runtime.InteropServices; using System.Runtime.InteropServices;
using System.Runtime.Intrinsics; using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.Arm;
using System.Runtime.Intrinsics.X86; using System.Runtime.Intrinsics.X86;
namespace Ryujinx.Audio.Renderer.Dsp.Command namespace Ryujinx.Audio.Renderer.Dsp.Command
@ -89,6 +90,26 @@ namespace Ryujinx.Audio.Renderer.Dsp.Command
} }
} }
private void ProcessVolumeAdvSimd(Span<float> outputBuffer, ReadOnlySpan<float> inputBuffer)
{
Vector128<float> volumeVec = Vector128.Create(Volume);
ReadOnlySpan<Vector128<float>> inputVec = MemoryMarshal.Cast<float, Vector128<float>>(inputBuffer);
Span<Vector128<float>> outputVec = MemoryMarshal.Cast<float, Vector128<float>>(outputBuffer);
int sisdStart = inputVec.Length * 4;
for (int i = 0; i < inputVec.Length; i++)
{
outputVec[i] = AdvSimd.Ceiling(AdvSimd.Multiply(inputVec[i], volumeVec));
}
for (int i = sisdStart; i < inputBuffer.Length; i++)
{
outputBuffer[i] = FloatingPointHelper.MultiplyRoundUp(inputBuffer[i], Volume);
}
}
private void ProcessVolume(Span<float> outputBuffer, ReadOnlySpan<float> inputBuffer) private void ProcessVolume(Span<float> outputBuffer, ReadOnlySpan<float> inputBuffer)
{ {
if (Avx.IsSupported) if (Avx.IsSupported)
@ -99,6 +120,10 @@ namespace Ryujinx.Audio.Renderer.Dsp.Command
{ {
ProcessVolumeSse41(outputBuffer, inputBuffer); ProcessVolumeSse41(outputBuffer, inputBuffer);
} }
else if (AdvSimd.IsSupported)
{
ProcessVolumeAdvSimd(outputBuffer, inputBuffer);
}
else else
{ {
ProcessVolumeSlowPath(outputBuffer, inputBuffer); ProcessVolumeSlowPath(outputBuffer, inputBuffer);

View file

@ -26,6 +26,7 @@ using System.Diagnostics;
using System.Runtime.CompilerServices; using System.Runtime.CompilerServices;
using System.Runtime.InteropServices; using System.Runtime.InteropServices;
using System.Runtime.Intrinsics; using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.Arm;
using System.Runtime.Intrinsics.X86; using System.Runtime.Intrinsics.X86;
using static Ryujinx.Audio.Renderer.Parameter.VoiceInParameter; using static Ryujinx.Audio.Renderer.Parameter.VoiceInParameter;
@ -320,6 +321,24 @@ namespace Ryujinx.Audio.Renderer.Dsp
} }
} }
private static void ToFloatAdvSimd(Span<float> output, ReadOnlySpan<int> input, int sampleCount)
{
ReadOnlySpan<Vector128<int>> inputVec = MemoryMarshal.Cast<int, Vector128<int>>(input);
Span<Vector128<float>> outputVec = MemoryMarshal.Cast<float, Vector128<float>>(output);
int sisdStart = inputVec.Length * 4;
for (int i = 0; i < inputVec.Length; i++)
{
outputVec[i] = AdvSimd.ConvertToSingle(inputVec[i]);
}
for (int i = sisdStart; i < sampleCount; i++)
{
output[i] = input[i];
}
}
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
public static void ToFloatSlow(Span<float> output, ReadOnlySpan<int> input, int sampleCount) public static void ToFloatSlow(Span<float> output, ReadOnlySpan<int> input, int sampleCount)
{ {
@ -339,6 +358,10 @@ namespace Ryujinx.Audio.Renderer.Dsp
{ {
ToFloatSse2(output, input, sampleCount); ToFloatSse2(output, input, sampleCount);
} }
else if (AdvSimd.IsSupported)
{
ToFloatAdvSimd(output, input, sampleCount);
}
else else
{ {
ToFloatSlow(output, input, sampleCount); ToFloatSlow(output, input, sampleCount);
@ -372,7 +395,25 @@ namespace Ryujinx.Audio.Renderer.Dsp
for (int i = 0; i < inputVec.Length; i++) for (int i = 0; i < inputVec.Length; i++)
{ {
outputVec[i] = Avx.ConvertToVector128Int32(inputVec[i]); outputVec[i] = Sse2.ConvertToVector128Int32(inputVec[i]);
}
for (int i = sisdStart; i < sampleCount; i++)
{
output[i] = (int)input[i];
}
}
public static void ToIntAdvSimd(Span<int> output, ReadOnlySpan<float> input, int sampleCount)
{
ReadOnlySpan<Vector128<float>> inputVec = MemoryMarshal.Cast<float, Vector128<float>>(input);
Span<Vector128<int>> outputVec = MemoryMarshal.Cast<int, Vector128<int>>(output);
int sisdStart = inputVec.Length * 4;
for (int i = 0; i < inputVec.Length; i++)
{
outputVec[i] = AdvSimd.ConvertToInt32RoundToZero(inputVec[i]);
} }
for (int i = sisdStart; i < sampleCount; i++) for (int i = sisdStart; i < sampleCount; i++)
@ -400,6 +441,10 @@ namespace Ryujinx.Audio.Renderer.Dsp
{ {
ToIntSse2(output, input, sampleCount); ToIntSse2(output, input, sampleCount);
} }
else if (AdvSimd.IsSupported)
{
ToIntAdvSimd(output, input, sampleCount);
}
else else
{ {
ToIntSlow(output, input, sampleCount); ToIntSlow(output, input, sampleCount);