OpenGL rebuild project start!

源代码/StgSharp/Optimization/SSE.SqrtSS.cs

@@ -1,112 +1,53 @@
 using StgSharp.Math;
 using System;
 using System.Collections.Generic;
+using System.IO.Pipes;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 using System.Text;
-using static StgSharp.Optimization.CpuInfo;
 
 namespace StgSharp.Optimization
 {
     public static unsafe partial class SSE
     {
-        public static m128 sqrtssM128Cache = new m128();
-        public static m128* sqrtssM128CachePointer;
-
-        private static IntPtr sqrtssCodePointer;
-
-        private static sqrtssDelegate sqrtssDele;
-
-        private static byte[] sqrtssCodeBytes
-            = new byte[]
-        {
-            0x48,0xb8,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,  // mov rax, [pointer location]
-            0x48,0x8b,0x00,                                     // mov rax, [rax]
-            0xf3,0x0f,0x6f,0x00,                                // movqdu xmm0,xmmword ptr [rax]
-            0xf3,0x0f,0x51,0xc0,                                // sqrtss xmm0,xmm0
-            0xf3,0x0f,0x7f,0x00                                 // movdqu xmmword ptr [rax],xmm0
+        internal static readonly byte[] SqrtssCodeBytes = new byte[] {
+            0x48, 0x89, 0xc8,               //mov rax,rcx
+            0xf3, 0x0f, 0x6f, 0x00,         //movdqu xmm0,xmmword ptr [rax]
+            //0x66, 0x0f, 0x6f, 0x00,         //movdqa xmm0,xmmword ptr [rax]
+            0xf3,0x0f, 0x51, 0xc0,          //sqrtss xmm0,xmm0
+            0xf3, 0x0f, 0x7f, 0x00,         //movdqu xmmqord ptr [rax],xmm0
+            //0x66, 0x0f, 0x7f, 0x00,         //movdqa xmmword ptr [rax],xmm0
+            0xc3,                           //ret
         };
 
-        [DllImport("kernel32.dll", SetLastError = true)]
-        private static extern IntPtr VirtualAlloc(
-            IntPtr lpAddress,
-            UIntPtr dwSize,
-            AllocationType flAllocationType,
-            MemoryProtection flProtect
-        );
+        
 
-        [DllImport("kernel32")]
-        private static extern bool VirtualFree(
-                IntPtr lpAddress,
-                UInt32 dwSize,
-                UInt32 dwFreeType
-        );
+        static asmfuncDelegate sqrtssDelegate;
 
-        [Flags()]
-        protected enum AllocationType : uint
-        {
-            COMMIT = 0x1000,
-            RESERVE = 0x2000,
-            RESET = 0x80000,
-            LARGE_PAGES = 0x20000000,
-            PHYSICAL = 0x400000,
-            TOP_DOWN = 0x100000,
-            WRITE_WATCH = 0x200000
-        }
-
-        [Flags()]
-        protected enum MemoryProtection : uint
-        {
-            EXECUTE = 0x10,
-            EXECUTE_READ = 0x20,
-            EXECUTE_READWRITE = 0x40,
-            EXECUTE_WRITECOPY = 0x80,
-            NOACCESS = 0x01,
-            READONLY = 0x02,
-            READWRITE = 0x04,
-            WRITECOPY = 0x08,
-            GUARD_Modifierflag = 0x100,
-            NOCACHE_Modifierflag = 0x200,
-            WRITECOMBINE_Modifierflag = 0x400
-        }
-
-        private unsafe static void InitSqrtss()
-        {
-            byte[] sqrtssM128PointerBytes = BitConverter.GetBytes((ulong)sqrtssM128CachePointer);
-            for (int i = 0; i < sqrtssM128PointerBytes.Length; i++)
-            {
-                sqrtssCodeBytes[2+i] = sqrtssCodeBytes[7-i];
-            }
-
-            if (IntPtr.Size!=8)
-            {
-                throw new Exception("This SSE class support 64bits OS only presently!");
-            }
-
-            sqrtssCodePointer = VirtualAlloc(
+        private static IntPtr sqrtssHandle = VirtualAlloc(
                 IntPtr.Zero,
-                new UIntPtr((uint)sqrtssCodeBytes.Length),
-                AllocationType.COMMIT | AllocationType.RESERVE,
-                MemoryProtection.EXECUTE_READWRITE
-            );
+                SqrtssCodeBytes.Length,
+                MEM_COMMIT | MEM_RESERVE,
+                PAGE_EXECUTE_READWRITE);
 
-            Marshal.Copy(sqrtssCodeBytes, 0, sqrtssCodePointer, sqrtssCodeBytes.Length);
-            sqrtssDele = (sqrtssDelegate)Marshal.GetDelegateForFunctionPointer(sqrtssCodePointer,typeof(sqrtssDelegate));
+        private static void InitSqrtss()
+        {
+            Marshal.Copy(SqrtssCodeBytes,0,sqrtssHandle,SqrtssCodeBytes.Length);
+            sqrtssDelegate = Marshal.GetDelegateForFunctionPointer(sqrtssHandle,typeof(asmfuncDelegate)) as asmfuncDelegate;
         }
 
-        public static float Sqrtss(float a)
+        public unsafe static m128 Sqrtss(m128* m)
         {
-            fixed (m128* p = &sqrtssM128Cache)
-            {
-                sqrtssM128Cache.float0 = a;
-                sqrtssDele(sqrtssM128Cache);
-                return sqrtssM128Cache.float0;
-            }
+            sqrtssDelegate((ulong)m);
+            return *m;
         }
 
     }
 
-
-    [UnmanagedFunctionPointerAttribute(CallingConvention.ThisCall)]
-    internal delegate void sqrtssDelegate(m128 a);
+    /// <summary>
+    /// A delegate for asm sse functions
+    /// </summary>
+    /// <param name="m128Pointrter">The m128* pointer to the target m128 transformed in ulong form</param>
+    [UnmanagedFunctionPointerAttribute(CallingConvention.Cdecl)]
+    internal delegate void asmfuncDelegate(ulong m128Pointrter);
 }

源代码/StgSharp/Optimization/SSE.cs

@@ -1,18 +1,26 @@
 using System;
 using System.Collections.Generic;
+using System.Runtime.InteropServices;
 using System.Text;
 
 namespace StgSharp.Optimization
 {
     public static unsafe partial class SSE
     {
+        [DllImport("Kernel32.dll", EntryPoint = "VirtualAlloc")]
+        private static extern IntPtr VirtualAlloc(IntPtr address, int size, uint allocType, uint protect);
+        [DllImport("Kernel32.dll", EntryPoint = "VirtualFree")]
+        private static extern bool VirtualFree(IntPtr address, int size, uint freeType);
+
+        const uint MEM_COMMIT = 0x1000;
+        const uint MEM_RESERVE = 0x2000;
+        const uint PAGE_EXECUTE_READWRITE = 0x40;
+        const uint MEM_RELEASE = 0x8000;
+
         static SSE()
         {
             //编译模拟Sqrtss指令
-
-            var sqrtssCachePointer = stackalloc ulong[1];
             InitSqrtss();
-        
         }
 
     }

源代码/StgSharp/Optimization/m128.cs

@@ -6,7 +6,7 @@ using System.Text;
 
 namespace StgSharp.Optimization
 {
-    [StructLayout(LayoutKind.Explicit,Size =16)]
+    [StructLayout(LayoutKind.Explicit,Size =16,Pack =16)]
     public struct m128
     {
         [FieldOffset(0)] public float float0;