mirror of
https://github.com/Ryujinx/Ryujinx.git
synced 2024-12-29 17:36:06 +00:00
Use ReadOnlySpan<byte> compiler optimization for static data (#3130)
This commit is contained in:
parent
7e9011673b
commit
9ca040c0ff
|
@ -1,13 +1,14 @@
|
||||||
// https://www.intel.com/content/dam/doc/white-paper/advanced-encryption-standard-new-instructions-set-paper.pdf
|
// https://www.intel.com/content/dam/doc/white-paper/advanced-encryption-standard-new-instructions-set-paper.pdf
|
||||||
|
|
||||||
using ARMeilleure.State;
|
using ARMeilleure.State;
|
||||||
|
using System;
|
||||||
|
|
||||||
namespace ARMeilleure.Instructions
|
namespace ARMeilleure.Instructions
|
||||||
{
|
{
|
||||||
static class CryptoHelper
|
static class CryptoHelper
|
||||||
{
|
{
|
||||||
#region "LookUp Tables"
|
#region "LookUp Tables"
|
||||||
private static readonly byte[] _sBox = new byte[]
|
private static ReadOnlySpan<byte> _sBox => new byte[]
|
||||||
{
|
{
|
||||||
0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
|
0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
|
||||||
0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
|
0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
|
||||||
|
@ -27,7 +28,7 @@ namespace ARMeilleure.Instructions
|
||||||
0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16
|
0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _invSBox = new byte[]
|
private static ReadOnlySpan<byte> _invSBox => new byte[]
|
||||||
{
|
{
|
||||||
0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38, 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
|
0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38, 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
|
||||||
0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87, 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
|
0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87, 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
|
||||||
|
@ -47,7 +48,7 @@ namespace ARMeilleure.Instructions
|
||||||
0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26, 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d
|
0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26, 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _gfMul02 = new byte[]
|
private static ReadOnlySpan<byte> _gfMul02 => new byte[]
|
||||||
{
|
{
|
||||||
0x00, 0x02, 0x04, 0x06, 0x08, 0x0a, 0x0c, 0x0e, 0x10, 0x12, 0x14, 0x16, 0x18, 0x1a, 0x1c, 0x1e,
|
0x00, 0x02, 0x04, 0x06, 0x08, 0x0a, 0x0c, 0x0e, 0x10, 0x12, 0x14, 0x16, 0x18, 0x1a, 0x1c, 0x1e,
|
||||||
0x20, 0x22, 0x24, 0x26, 0x28, 0x2a, 0x2c, 0x2e, 0x30, 0x32, 0x34, 0x36, 0x38, 0x3a, 0x3c, 0x3e,
|
0x20, 0x22, 0x24, 0x26, 0x28, 0x2a, 0x2c, 0x2e, 0x30, 0x32, 0x34, 0x36, 0x38, 0x3a, 0x3c, 0x3e,
|
||||||
|
@ -67,7 +68,7 @@ namespace ARMeilleure.Instructions
|
||||||
0xfb, 0xf9, 0xff, 0xfd, 0xf3, 0xf1, 0xf7, 0xf5, 0xeb, 0xe9, 0xef, 0xed, 0xe3, 0xe1, 0xe7, 0xe5
|
0xfb, 0xf9, 0xff, 0xfd, 0xf3, 0xf1, 0xf7, 0xf5, 0xeb, 0xe9, 0xef, 0xed, 0xe3, 0xe1, 0xe7, 0xe5
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _gfMul03 = new byte[]
|
private static ReadOnlySpan<byte> _gfMul03 => new byte[]
|
||||||
{
|
{
|
||||||
0x00, 0x03, 0x06, 0x05, 0x0c, 0x0f, 0x0a, 0x09, 0x18, 0x1b, 0x1e, 0x1d, 0x14, 0x17, 0x12, 0x11,
|
0x00, 0x03, 0x06, 0x05, 0x0c, 0x0f, 0x0a, 0x09, 0x18, 0x1b, 0x1e, 0x1d, 0x14, 0x17, 0x12, 0x11,
|
||||||
0x30, 0x33, 0x36, 0x35, 0x3c, 0x3f, 0x3a, 0x39, 0x28, 0x2b, 0x2e, 0x2d, 0x24, 0x27, 0x22, 0x21,
|
0x30, 0x33, 0x36, 0x35, 0x3c, 0x3f, 0x3a, 0x39, 0x28, 0x2b, 0x2e, 0x2d, 0x24, 0x27, 0x22, 0x21,
|
||||||
|
@ -87,7 +88,7 @@ namespace ARMeilleure.Instructions
|
||||||
0x0b, 0x08, 0x0d, 0x0e, 0x07, 0x04, 0x01, 0x02, 0x13, 0x10, 0x15, 0x16, 0x1f, 0x1c, 0x19, 0x1a
|
0x0b, 0x08, 0x0d, 0x0e, 0x07, 0x04, 0x01, 0x02, 0x13, 0x10, 0x15, 0x16, 0x1f, 0x1c, 0x19, 0x1a
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _gfMul09 = new byte[]
|
private static ReadOnlySpan<byte> _gfMul09 => new byte[]
|
||||||
{
|
{
|
||||||
0x00, 0x09, 0x12, 0x1b, 0x24, 0x2d, 0x36, 0x3f, 0x48, 0x41, 0x5a, 0x53, 0x6c, 0x65, 0x7e, 0x77,
|
0x00, 0x09, 0x12, 0x1b, 0x24, 0x2d, 0x36, 0x3f, 0x48, 0x41, 0x5a, 0x53, 0x6c, 0x65, 0x7e, 0x77,
|
||||||
0x90, 0x99, 0x82, 0x8b, 0xb4, 0xbd, 0xa6, 0xaf, 0xd8, 0xd1, 0xca, 0xc3, 0xfc, 0xf5, 0xee, 0xe7,
|
0x90, 0x99, 0x82, 0x8b, 0xb4, 0xbd, 0xa6, 0xaf, 0xd8, 0xd1, 0xca, 0xc3, 0xfc, 0xf5, 0xee, 0xe7,
|
||||||
|
@ -107,7 +108,7 @@ namespace ARMeilleure.Instructions
|
||||||
0x31, 0x38, 0x23, 0x2a, 0x15, 0x1c, 0x07, 0x0e, 0x79, 0x70, 0x6b, 0x62, 0x5d, 0x54, 0x4f, 0x46
|
0x31, 0x38, 0x23, 0x2a, 0x15, 0x1c, 0x07, 0x0e, 0x79, 0x70, 0x6b, 0x62, 0x5d, 0x54, 0x4f, 0x46
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _gfMul0B = new byte[]
|
private static ReadOnlySpan<byte> _gfMul0B => new byte[]
|
||||||
{
|
{
|
||||||
0x00, 0x0b, 0x16, 0x1d, 0x2c, 0x27, 0x3a, 0x31, 0x58, 0x53, 0x4e, 0x45, 0x74, 0x7f, 0x62, 0x69,
|
0x00, 0x0b, 0x16, 0x1d, 0x2c, 0x27, 0x3a, 0x31, 0x58, 0x53, 0x4e, 0x45, 0x74, 0x7f, 0x62, 0x69,
|
||||||
0xb0, 0xbb, 0xa6, 0xad, 0x9c, 0x97, 0x8a, 0x81, 0xe8, 0xe3, 0xfe, 0xf5, 0xc4, 0xcf, 0xd2, 0xd9,
|
0xb0, 0xbb, 0xa6, 0xad, 0x9c, 0x97, 0x8a, 0x81, 0xe8, 0xe3, 0xfe, 0xf5, 0xc4, 0xcf, 0xd2, 0xd9,
|
||||||
|
@ -127,7 +128,7 @@ namespace ARMeilleure.Instructions
|
||||||
0xca, 0xc1, 0xdc, 0xd7, 0xe6, 0xed, 0xf0, 0xfb, 0x92, 0x99, 0x84, 0x8f, 0xbe, 0xb5, 0xa8, 0xa3
|
0xca, 0xc1, 0xdc, 0xd7, 0xe6, 0xed, 0xf0, 0xfb, 0x92, 0x99, 0x84, 0x8f, 0xbe, 0xb5, 0xa8, 0xa3
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _gfMul0D = new byte[]
|
private static ReadOnlySpan<byte> _gfMul0D => new byte[]
|
||||||
{
|
{
|
||||||
0x00, 0x0d, 0x1a, 0x17, 0x34, 0x39, 0x2e, 0x23, 0x68, 0x65, 0x72, 0x7f, 0x5c, 0x51, 0x46, 0x4b,
|
0x00, 0x0d, 0x1a, 0x17, 0x34, 0x39, 0x2e, 0x23, 0x68, 0x65, 0x72, 0x7f, 0x5c, 0x51, 0x46, 0x4b,
|
||||||
0xd0, 0xdd, 0xca, 0xc7, 0xe4, 0xe9, 0xfe, 0xf3, 0xb8, 0xb5, 0xa2, 0xaf, 0x8c, 0x81, 0x96, 0x9b,
|
0xd0, 0xdd, 0xca, 0xc7, 0xe4, 0xe9, 0xfe, 0xf3, 0xb8, 0xb5, 0xa2, 0xaf, 0x8c, 0x81, 0x96, 0x9b,
|
||||||
|
@ -147,7 +148,7 @@ namespace ARMeilleure.Instructions
|
||||||
0xdc, 0xd1, 0xc6, 0xcb, 0xe8, 0xe5, 0xf2, 0xff, 0xb4, 0xb9, 0xae, 0xa3, 0x80, 0x8d, 0x9a, 0x97
|
0xdc, 0xd1, 0xc6, 0xcb, 0xe8, 0xe5, 0xf2, 0xff, 0xb4, 0xb9, 0xae, 0xa3, 0x80, 0x8d, 0x9a, 0x97
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _gfMul0E = new byte[]
|
private static ReadOnlySpan<byte> _gfMul0E => new byte[]
|
||||||
{
|
{
|
||||||
0x00, 0x0e, 0x1c, 0x12, 0x38, 0x36, 0x24, 0x2a, 0x70, 0x7e, 0x6c, 0x62, 0x48, 0x46, 0x54, 0x5a,
|
0x00, 0x0e, 0x1c, 0x12, 0x38, 0x36, 0x24, 0x2a, 0x70, 0x7e, 0x6c, 0x62, 0x48, 0x46, 0x54, 0x5a,
|
||||||
0xe0, 0xee, 0xfc, 0xf2, 0xd8, 0xd6, 0xc4, 0xca, 0x90, 0x9e, 0x8c, 0x82, 0xa8, 0xa6, 0xb4, 0xba,
|
0xe0, 0xee, 0xfc, 0xf2, 0xd8, 0xd6, 0xc4, 0xca, 0x90, 0x9e, 0x8c, 0x82, 0xa8, 0xa6, 0xb4, 0xba,
|
||||||
|
@ -167,12 +168,12 @@ namespace ARMeilleure.Instructions
|
||||||
0xd7, 0xd9, 0xcb, 0xc5, 0xef, 0xe1, 0xf3, 0xfd, 0xa7, 0xa9, 0xbb, 0xb5, 0x9f, 0x91, 0x83, 0x8d
|
0xd7, 0xd9, 0xcb, 0xc5, 0xef, 0xe1, 0xf3, 0xfd, 0xa7, 0xa9, 0xbb, 0xb5, 0x9f, 0x91, 0x83, 0x8d
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _srPerm = new byte[]
|
private static ReadOnlySpan<byte> _srPerm => new byte[]
|
||||||
{
|
{
|
||||||
0, 13, 10, 7, 4, 1, 14, 11, 8, 5, 2, 15, 12, 9, 6, 3
|
0, 13, 10, 7, 4, 1, 14, 11, 8, 5, 2, 15, 12, 9, 6, 3
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] _isrPerm = new byte[]
|
private static ReadOnlySpan<byte> _isrPerm => new byte[]
|
||||||
{
|
{
|
||||||
0, 5, 10, 15, 4, 9, 14, 3, 8, 13, 2, 7, 12, 1, 6, 11
|
0, 5, 10, 15, 4, 9, 14, 3, 8, 13, 2, 7, 12, 1, 6, 11
|
||||||
};
|
};
|
||||||
|
|
|
@ -824,7 +824,7 @@ namespace ARMeilleure.Instructions
|
||||||
return (ulong)(size - 1);
|
return (ulong)(size - 1);
|
||||||
}
|
}
|
||||||
|
|
||||||
private static readonly byte[] ClzNibbleTbl = { 4, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0 };
|
private static ReadOnlySpan<byte> ClzNibbleTbl => new byte[] { 4, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0 };
|
||||||
|
|
||||||
public static ulong CountLeadingZeros(ulong value, int size) // size is 8, 16, 32 or 64 (SIMD&FP or Base Inst.).
|
public static ulong CountLeadingZeros(ulong value, int size) // size is 8, 16, 32 or 64 (SIMD&FP or Base Inst.).
|
||||||
{
|
{
|
||||||
|
|
|
@ -1,8 +1,10 @@
|
||||||
|
using System;
|
||||||
|
|
||||||
namespace Ryujinx.Common
|
namespace Ryujinx.Common
|
||||||
{
|
{
|
||||||
public static class BitUtils
|
public static class BitUtils
|
||||||
{
|
{
|
||||||
private static readonly byte[] ClzNibbleTbl = { 4, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0 };
|
private static ReadOnlySpan<byte> ClzNibbleTbl => new byte[] { 4, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0 };
|
||||||
|
|
||||||
public static uint AlignUp(uint value, int size)
|
public static uint AlignUp(uint value, int size)
|
||||||
{
|
{
|
||||||
|
|
|
@ -43,7 +43,7 @@ namespace Ryujinx.Common
|
||||||
Prime32_1
|
Prime32_1
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] Xxh3KSecret = new byte[]
|
private static ReadOnlySpan<byte> Xxh3KSecret => new byte[]
|
||||||
{
|
{
|
||||||
0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
|
0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
|
||||||
0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
|
0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
|
||||||
|
|
|
@ -118,7 +118,7 @@ namespace Ryujinx.Graphics.Nvdec.FFmpeg.H264
|
||||||
}
|
}
|
||||||
|
|
||||||
// ZigZag LUTs from libavcodec.
|
// ZigZag LUTs from libavcodec.
|
||||||
private static readonly byte[] ZigZagDirect = new byte[]
|
private static ReadOnlySpan<byte> ZigZagDirect => new byte[]
|
||||||
{
|
{
|
||||||
0, 1, 8, 16, 9, 2, 3, 10,
|
0, 1, 8, 16, 9, 2, 3, 10,
|
||||||
17, 24, 32, 25, 18, 11, 4, 5,
|
17, 24, 32, 25, 18, 11, 4, 5,
|
||||||
|
@ -130,7 +130,7 @@ namespace Ryujinx.Graphics.Nvdec.FFmpeg.H264
|
||||||
53, 60, 61, 54, 47, 55, 62, 63
|
53, 60, 61, 54, 47, 55, 62, 63
|
||||||
};
|
};
|
||||||
|
|
||||||
private static readonly byte[] ZigZagScan = new byte[]
|
private static ReadOnlySpan<byte> ZigZagScan => new byte[]
|
||||||
{
|
{
|
||||||
0 + 0 * 4, 1 + 0 * 4, 0 + 1 * 4, 0 + 2 * 4,
|
0 + 0 * 4, 1 + 0 * 4, 0 + 1 * 4, 0 + 2 * 4,
|
||||||
1 + 1 * 4, 2 + 0 * 4, 3 + 0 * 4, 2 + 1 * 4,
|
1 + 1 * 4, 2 + 0 * 4, 3 + 0 * 4, 2 + 1 * 4,
|
||||||
|
@ -140,7 +140,7 @@ namespace Ryujinx.Graphics.Nvdec.FFmpeg.H264
|
||||||
|
|
||||||
private static void WriteScalingList(ref H264BitStreamWriter writer, IArray<byte> list)
|
private static void WriteScalingList(ref H264BitStreamWriter writer, IArray<byte> list)
|
||||||
{
|
{
|
||||||
byte[] scan = list.Length == 16 ? ZigZagScan : ZigZagDirect;
|
ReadOnlySpan<byte> scan = list.Length == 16 ? ZigZagScan : ZigZagDirect;
|
||||||
|
|
||||||
int lastScale = 8;
|
int lastScale = 8;
|
||||||
|
|
||||||
|
|
|
@ -15,7 +15,7 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
|
||||||
|
|
||||||
public ISurface CreateSurface(int width, int height) => new Surface(width, height);
|
public ISurface CreateSurface(int width, int height) => new Surface(width, height);
|
||||||
|
|
||||||
private static readonly byte[] LiteralToFilter = new byte[]
|
private static ReadOnlySpan<byte> LiteralToFilter => new byte[]
|
||||||
{
|
{
|
||||||
Constants.EightTapSmooth,
|
Constants.EightTapSmooth,
|
||||||
Constants.EightTap,
|
Constants.EightTap,
|
||||||
|
|
|
@ -1,5 +1,6 @@
|
||||||
using Ryujinx.Graphics.Nvdec.Vp9.Common;
|
using Ryujinx.Graphics.Nvdec.Vp9.Common;
|
||||||
using Ryujinx.Graphics.Nvdec.Vp9.Types;
|
using Ryujinx.Graphics.Nvdec.Vp9.Types;
|
||||||
|
using System;
|
||||||
using static Ryujinx.Graphics.Nvdec.Vp9.Dsp.IntraPred;
|
using static Ryujinx.Graphics.Nvdec.Vp9.Dsp.IntraPred;
|
||||||
|
|
||||||
namespace Ryujinx.Graphics.Nvdec.Vp9
|
namespace Ryujinx.Graphics.Nvdec.Vp9
|
||||||
|
@ -24,7 +25,7 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
|
||||||
private const int NeedAbove = 1 << 2;
|
private const int NeedAbove = 1 << 2;
|
||||||
private const int NeedAboveRight = 1 << 3;
|
private const int NeedAboveRight = 1 << 3;
|
||||||
|
|
||||||
private static readonly byte[] ExtendModes = new byte[]
|
private static ReadOnlySpan<byte> ExtendModes => new byte[]
|
||||||
{
|
{
|
||||||
NeedAbove | NeedLeft, // DC
|
NeedAbove | NeedLeft, // DC
|
||||||
NeedAbove, // V
|
NeedAbove, // V
|
||||||
|
|
|
@ -10,7 +10,7 @@ namespace Ryujinx.Graphics.Nvdec.Vp9.Types
|
||||||
public short Row;
|
public short Row;
|
||||||
public short Col;
|
public short Col;
|
||||||
|
|
||||||
private static readonly byte[] LogInBase2 = new byte[]
|
private static ReadOnlySpan<byte> LogInBase2 => new byte[]
|
||||||
{
|
{
|
||||||
0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
|
0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
|
||||||
4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
|
4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
|
||||||
|
|
Loading…
Reference in a new issue