crafterpika/MeloNX
0
0
Fork 0
mirror of https://git.743378673.xyz/MeloNX/MeloNX.git synced 2025-07-30 04:07:09 +02:00
Code Issues Projects Releases Packages Wiki Activity

Accelerate NVDEC VIC surface read/write and colorspace conversion with Arm64 HW intrinsics (#4351)

Browse source 
* Accelerate NVDEC VIC surface read/write and colorspace conversion with Arm64 HW intrinsics

* Improve ReadNv12 x86 SSE path
This commit is contained in:
gdkchan 2023-02-06 23:38:54 -03:00 • committed by GitHub
parent 3af1d284cb
commit 72a4aee19a
3 changed files with 420 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

121
Ryujinx.Graphics.Vic/Blender.cs
View file

@ -4,6 +4,7 @@ using System;
using System.Diagnostics;
using System.Runtime.CompilerServices;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.Arm;
using System.Runtime.Intrinsics.X86;
namespace Ryujinx.Graphics.Vic
@ -17,10 +18,18 @@ namespace Ryujinx.Graphics.Vic
int x2 = Math.Min(src.Width, x1 + targetRect.Width);
int y2 = Math.Min(src.Height, y1 + targetRect.Height);
if (Sse41.IsSupported && ((x1 | x2) & 3) == 0)
if (((x1 | x2) & 3) == 0)
{
BlendOneSse41(dst, src, ref slot, x1, y1, x2, y2);
return;
if (Sse41.IsSupported)
{
BlendOneSse41(dst, src, ref slot, x1, y1, x2, y2);
return;
}
else if (AdvSimd.IsSupported)
{
BlendOneAdvSimd(dst, src, ref slot, x1, y1, x2, y2);
return;
}
}
for (int y = y1; y < y2; y++)
@ -105,6 +114,84 @@ namespace Ryujinx.Graphics.Vic
}
}
private unsafe static void BlendOneAdvSimd(Surface dst, Surface src, ref SlotStruct slot, int x1, int y1, int x2, int y2)
{
Debug.Assert(((x1 | x2) & 3) == 0);
ref MatrixStruct mtx = ref slot.ColorMatrixStruct;
Vector128<int> col1 = Vector128.Create(mtx.MatrixCoeff00, mtx.MatrixCoeff10, mtx.MatrixCoeff20, 0);
Vector128<int> col2 = Vector128.Create(mtx.MatrixCoeff01, mtx.MatrixCoeff11, mtx.MatrixCoeff21, 0);
Vector128<int> col3 = Vector128.Create(mtx.MatrixCoeff02, mtx.MatrixCoeff12, mtx.MatrixCoeff22, 0);
Vector128<int> col4 = Vector128.Create(mtx.MatrixCoeff03, mtx.MatrixCoeff13, mtx.MatrixCoeff23, 0);
Vector128<int> rShift = Vector128.Create(-mtx.MatrixRShift);
Vector128<int> selMask = Vector128.Create(0, 0, 0, -1);
Vector128<ushort> clMin = Vector128.Create((ushort)slot.SlotConfig.SoftClampLow);
Vector128<ushort> clMax = Vector128.Create((ushort)slot.SlotConfig.SoftClampHigh);
fixed (Pixel* srcPtr = src.Data, dstPtr = dst.Data)
{
Pixel* ip = srcPtr;
Pixel* op = dstPtr;
if (mtx.MatrixEnable)
{
for (int y = y1; y < y2; y++, ip += src.Width, op += dst.Width)
{
for (int x = x1; x < x2; x += 4)
{
Vector128<ushort> pixel12 = AdvSimd.LoadVector128((ushort*)(ip + (uint)x));
Vector128<ushort> pixel34 = AdvSimd.LoadVector128((ushort*)(ip + (uint)x + 2));
Vector128<uint> pixel1 = AdvSimd.ZeroExtendWideningLower(pixel12.GetLower());
Vector128<uint> pixel2 = AdvSimd.ZeroExtendWideningUpper(pixel12);
Vector128<uint> pixel3 = AdvSimd.ZeroExtendWideningLower(pixel34.GetLower());
Vector128<uint> pixel4 = AdvSimd.ZeroExtendWideningUpper(pixel34);
Vector128<int> t1 = MatrixMultiplyAdvSimd(pixel1.AsInt32(), col1, col2, col3, col4, rShift, selMask);
Vector128<int> t2 = MatrixMultiplyAdvSimd(pixel2.AsInt32(), col1, col2, col3, col4, rShift, selMask);
Vector128<int> t3 = MatrixMultiplyAdvSimd(pixel3.AsInt32(), col1, col2, col3, col4, rShift, selMask);
Vector128<int> t4 = MatrixMultiplyAdvSimd(pixel4.AsInt32(), col1, col2, col3, col4, rShift, selMask);
Vector64<ushort> lower1 = AdvSimd.ExtractNarrowingSaturateUnsignedLower(t1);
Vector64<ushort> lower3 = AdvSimd.ExtractNarrowingSaturateUnsignedLower(t3);
pixel12 = AdvSimd.ExtractNarrowingSaturateUnsignedUpper(lower1, t2);
pixel34 = AdvSimd.ExtractNarrowingSaturateUnsignedUpper(lower3, t4);
pixel12 = AdvSimd.Min(pixel12, clMax);
pixel34 = AdvSimd.Min(pixel34, clMax);
pixel12 = AdvSimd.Max(pixel12, clMin);
pixel34 = AdvSimd.Max(pixel34, clMin);
AdvSimd.Store((ushort*)(op + (uint)x + 0), pixel12);
AdvSimd.Store((ushort*)(op + (uint)x + 2), pixel34);
}
}
}
else
{
for (int y = y1; y < y2; y++, ip += src.Width, op += dst.Width)
{
for (int x = x1; x < x2; x += 4)
{
Vector128<ushort> pixel12 = AdvSimd.LoadVector128((ushort*)(ip + (uint)x));
Vector128<ushort> pixel34 = AdvSimd.LoadVector128((ushort*)(ip + (uint)x + 2));
pixel12 = AdvSimd.Min(pixel12, clMax);
pixel34 = AdvSimd.Min(pixel34, clMax);
pixel12 = AdvSimd.Max(pixel12, clMin);
pixel34 = AdvSimd.Max(pixel34, clMin);
AdvSimd.Store((ushort*)(op + (uint)x + 0), pixel12);
AdvSimd.Store((ushort*)(op + (uint)x + 2), pixel34);
}
}
}
}
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static void MatrixMultiply(ref MatrixStruct mtx, int x, int y, int z, out int r, out int g, out int b)
{
@ -159,5 +246,33 @@ namespace Ryujinx.Graphics.Vic
return res;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static Vector128<int> MatrixMultiplyAdvSimd(
Vector128<int> pixel,
Vector128<int> col1,
Vector128<int> col2,
Vector128<int> col3,
Vector128<int> col4,
Vector128<int> rShift,
Vector128<int> selectMask)
{
Vector128<int> x = AdvSimd.DuplicateSelectedScalarToVector128(pixel, 0);
Vector128<int> y = AdvSimd.DuplicateSelectedScalarToVector128(pixel, 1);
Vector128<int> z = AdvSimd.DuplicateSelectedScalarToVector128(pixel, 2);
col1 = AdvSimd.Multiply(col1, x);
col2 = AdvSimd.Multiply(col2, y);
col3 = AdvSimd.Multiply(col3, z);
Vector128<int> res = AdvSimd.Add(col3, AdvSimd.Add(col1, col2));
res = AdvSimd.ShiftArithmetic(res, rShift);
res = AdvSimd.Add(res, col4);
res = AdvSimd.ShiftRightArithmetic(res, 8);
res = AdvSimd.BitwiseSelect(selectMask, pixel, res);
return res;
}
}
}