dqily update on 20230423 by nitload youmu

Math/C/Project1/Project1/AsmSqrt.c

@@ -17,8 +17,16 @@ void main()
 
 	
 	float a = 400.0f;
+	a = sseSqrt(a);
+	printf("%f",a);
+	/**/
+}
+
+
+float sseSqrt(float x)
+{
 	float r[4];
-	float* pa = &a;
+	float* pa = &x;
 
 	__m128 b;
 	__m128 c;
@@ -27,21 +35,6 @@ void main()
 	c = _mm_sqrt_ss(b);
 	_mm_store_ss(r, c);
 
-	printf("%f", r[0]);
-	/**/
-}
-
-/*
-float sseSqrt(float x)
-{
-	float* a = &x;
-	float result[4];
-
-	__m128 b = _mm_load_ss(a);
-	__m128 c = _mm_sqrt_ss(b);
-
-	_mm_store_ss(result, c);
-
-	return result[0];
+	return r[0];
 }
 /**/

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

@@ -1,6 +1,7 @@
 using StgSharp.Math;
 using System;
 using System.Collections.Generic;
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 using System.Text;
 using static StgSharp.Optimization.CpuInfo;
@@ -10,6 +11,7 @@ namespace StgSharp.Optimization
     public static unsafe partial class SSE
     {
         public static m128 sqrtssM128Cache = new m128();
+        public static m128* sqrtssM128CachePointer;
 
         private static IntPtr sqrtssCodePointer;
 
@@ -18,7 +20,8 @@ namespace StgSharp.Optimization
         private static byte[] sqrtssCodeBytes
             = new byte[]
         {
-            0x48,0xb8,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,  // mov rax, [m128 location]
+            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
@@ -39,14 +42,37 @@ namespace StgSharp.Optimization
                 UInt32 dwFreeType
         );
 
+        [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()
         {
-            var marshalPtr = Marshal.AllocHGlobal(32);
-            var m128ptr = new IntPtr(16*((long)marshalPtr/16+1));
-            m128* sqrtssCachePointer = (m128*)m128ptr.ToPointer();
-
-            sqrtssM128Cache = *sqrtssCachePointer;
-            byte[] sqrtssM128PointerBytes = BitConverter.GetBytes((ulong)sqrtssCachePointer);
+            byte[] sqrtssM128PointerBytes = BitConverter.GetBytes((ulong)sqrtssM128CachePointer);
             for (int i = 0; i < sqrtssM128PointerBytes.Length; i++)
             {
                 sqrtssCodeBytes[2+i] = sqrtssCodeBytes[7-i];
@@ -70,14 +96,17 @@ namespace StgSharp.Optimization
 
         public static float Sqrtss(float a)
         {
-            sqrtssM128Cache.float0 = a;
-            sqrtssDele();
-            return sqrtssM128Cache.float0;
+            fixed (m128* p = &sqrtssM128Cache)
+            {
+                sqrtssM128Cache.float0 = a;
+                sqrtssDele(sqrtssM128Cache);
+                return sqrtssM128Cache.float0;
+            }
         }
 
     }
 
 
-    [UnmanagedFunctionPointerAttribute(CallingConvention.StdCall)]
-    internal delegate void sqrtssDelegate();
+    [UnmanagedFunctionPointerAttribute(CallingConvention.ThisCall)]
+    internal delegate void sqrtssDelegate(m128 a);
 }

源代码/StgSharp/Optimization/SSE.cs

@@ -9,7 +9,10 @@ namespace StgSharp.Optimization
         static SSE()
         {
             //编译模拟Sqrtss指令
+
+            var sqrtssCachePointer = stackalloc ulong[1];
             InitSqrtss();
+        
         }
 
     }

源代码/StgSharp/Optimization/m128.cs

@@ -62,4 +62,9 @@ namespace StgSharp.Optimization
             };
         }
     }
+
+    [StructLayout(LayoutKind.Explicit,Size = 32)]
+    internal struct m128cache
+    { 
+    }
 }

源代码/StgSharp/Optimization/x86.cs

@@ -1,11 +0,0 @@
-using System;
-using System.Collections.Generic;
-using System.Text;
-
-namespace StgSharp.Optimization
-{
-    public static unsafe class x86
-    {
-
-    }
-}

源代码/StgSharp/obj/Debug/netstandard2.1/StgSharp.csproj.CoreCompileInputs.cache

@@ -1 +1 @@
-508c7d643da3ffba0187af6df00e8fd1f0a8c855
+8cab6b2db52187a41fdaeb6d4e3db4542fe3cc35