add startup numa sense

StgSharp.Common/HighPerformance/ProcessorAbstraction/LogicalCore.cs

@@ -0,0 +1,48 @@
+//-----------------------------------------------------------------------
+// -----------------------------------------------------------------------
+// file="LogicalCore"
+// Project: StgSharp
+// AuthorGroup: Nitload
+// Copyright (c) Nitload. All rights reserved.
+//     
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+//     
+// -----------------------------------------------------------------------
+// -----------------------------------------------------------------------
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Text;
+using System.Threading.Tasks;
+
+namespace StgSharp.Common.HighPerformance.ProcessorAbstraction
+{
+    public readonly struct LogicalCore
+    {
+
+        public readonly uint GlobalCpuId;
+        public readonly uint NumaGroupId;
+
+        /// <summary>
+        ///   Stays 0 on linux.
+        /// </summary>
+        public readonly uint NumaNodeId;
+
+    }
+}

StgSharp.Common/HighPerformance/ProcessorAbstraction/Numa.cs

@@ -0,0 +1,79 @@
+//-----------------------------------------------------------------------
+// -----------------------------------------------------------------------
+// file="Numa"
+// Project: StgSharp
+// AuthorGroup: Nitload
+// Copyright (c) Nitload. All rights reserved.
+//     
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+//     
+// -----------------------------------------------------------------------
+// -----------------------------------------------------------------------
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Runtime.InteropServices;
+using System.Text;
+using System.Threading.Tasks;
+
+namespace StgSharp.Common.HighPerformance.ProcessorAbstraction
+{
+    public static partial class Numa
+    {
+
+        private static NumaNode[] _numaNodes;
+
+        public static int NumaNodeCount { get; private set; }
+
+        public static ReadOnlySpan<NumaNode> AllNumaNode => _numaNodes;
+
+        [LibraryImport(Native.LibName, EntryPoint = "sn_get_numa_each")]
+        [UnmanagedCallConv(CallConvs = [typeof(CallConvCdecl)])]
+        [DefaultDllImportSearchPaths(DllImportSearchPath.AssemblyDirectory)]
+        public static partial int GetNumaEach(
+                                  int numaId,
+                                  Span<ulong> coreCountInNuma,
+                                  Span<ulong> globalCpuIdInNuma
+        );
+
+        [LibraryImport(Native.LibName, EntryPoint = "sn_get_numa_overall")]
+        [UnmanagedCallConv(CallConvs = [typeof(CallConvCdecl)])]
+        [DefaultDllImportSearchPaths(DllImportSearchPath.AssemblyDirectory)]
+        public static partial int GetNumaOverall(
+                                  ref int count,
+                                  Span<int> coreCountInNuma
+        );
+
+        internal static void Initialize()
+        {
+            Span<int> coreCountInNuma = stackalloc int[(Environment.ProcessorCount / 32) + 1];
+            int numaNodeCount = 0;
+            if (0 == GetNumaOverall(ref numaNodeCount, coreCountInNuma)) {
+                throw new InvalidOperationException("Failed to get NUMA overall information.");
+            }
+            _numaNodes = new NumaNode[numaNodeCount];
+            for (int i = 0; i < numaNodeCount; i++)
+            {
+                NumaNode node = new(coreCountInNuma[i]);
+                GetNumaEach(i, coreCountInNuma, node.GlobalCpuIds);
+            }
+        }
+
+    }
+}

StgSharp.Common/HighPerformance/ProcessorAbstraction/NumaNode.cs

@@ -0,0 +1,54 @@
+//-----------------------------------------------------------------------
+// -----------------------------------------------------------------------
+// file="NumaNode"
+// Project: StgSharp
+// AuthorGroup: Nitload
+// Copyright (c) Nitload. All rights reserved.
+//     
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+//     
+// -----------------------------------------------------------------------
+// -----------------------------------------------------------------------
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Runtime.InteropServices;
+using System.Text;
+using System.Threading.Tasks;
+
+namespace StgSharp.Common.HighPerformance.ProcessorAbstraction
+{
+    public class NumaNode
+    {
+
+        private LogicalCore[] _logicalCores;
+
+        public NumaNode(
+               int count
+        )
+        {
+            _logicalCores = new LogicalCore[count];
+        }
+
+        public int LogicalCoreCount => _logicalCores.Length;
+
+        public ReadOnlySpan<LogicalCore> LogicalCores => _logicalCores;
+
+    }
+}

StgSharp.Common/StgSharp.Common.csproj

@@ -23,8 +23,4 @@
 		<None Remove="Graphics\**" />
 	</ItemGroup>
 
-	<ItemGroup>
-	  <Folder Include="HighPerformance\ProcessorAbstraction\" />
-	</ItemGroup>
-
 </Project>

StgSharp.Native/StgSharp.Native.vcxproj

@@ -185,7 +185,8 @@
     <ClCompile Include="sn-src\simdid.x86.c" />
     <ClCompile Include="sn-src\simdid_compare.x86.c" />
     <ClCompile Include="sn-src\sn_native.c" />
-    <ClCompile Include="sn-src\t.affinity.c" />
+    <ClCompile Include="sn-src\thread.linux.c" />
+    <ClCompile Include="sn-src\thread.windows.c" />
     <ClCompile Include="sn-src\vector.c" />
     <ClCompile Include="sn-src\vector_dot.c" />
   </ItemGroup>

StgSharp.Native/StgSharp.Native.vcxproj.filters

@@ -60,12 +60,15 @@
     <ClCompile Include="sn-src\mk.fma.x86.c">
       <Filter>源文件</Filter>
     </ClCompile>
-    <ClCompile Include="sn-src\t.affinity.c">
+    <ClCompile Include="sn-src\thread.windows.c">
       <Filter>源文件</Filter>
     </ClCompile>
     <ClCompile Include="sn-src\simdid_compare.x86.c">
       <Filter>源文件</Filter>
     </ClCompile>
+    <ClCompile Include="sn-src\thread.linux.c">
+      <Filter>源文件</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="sn-head\gl.h">

StgSharp.Native/sn-head/StgSharpNative.h

@@ -66,14 +66,18 @@ SN_API void SN_DECL unloadImageData(Image *out);
 SN_API char *SN_DECL readLog(void);
 SN_API void SN_DECL load_glfw_functions(GLFWFunctionTable *table);
 SN_API void SN_DECL load_intrinsic_function(void *intrinsic_context, uint64_t id);
-SN_API SIMDID SN_DECL sn_get_simd_level_global();
+
-/* Per-core variant: always re-executes CPUID, never caches.
+#pragma region numa
- * Must be called on the target core (after affinity is set). */
-SN_API SIMDID SN_DECL sn_get_simd_level_local();
 SN_API SIMDID SN_DECL sn_get_unite_simd(SIMDID left, SIMDID right);
 SN_API int SN_DECL sn_compare_simd(SIMDID left, SIMDID right);
 SN_API void SN_DECL sn_set_thread_affinity(int core_id);
+SN_API void SN_DECL sn_get_numa_overall(int *out_numa_count, int *count_of_each_numa);
+SN_API void SN_DECL sn_get_numa_each(int numa_id, uint64_t *in_numa_cpuid, uint64_t *global_cpuid);
+/* Per-core variant: always re-executes CPUID, never caches.
+ * Must be called on the target core (after affinity is set). */
+SN_API SIMDID SN_DECL sn_get_simd_level_local();
 
+#pragma endregion
 #ifdef __cplusplus
 }
 #endif

StgSharp.Native/sn-src/simdid.x86.c

@@ -340,16 +340,6 @@ SN_API SIMDID SN_DECL sn_get_simd_level_local()
         return simdid;
 }
 
-SN_API SIMDID SN_DECL sn_get_simd_level_global()
-{
-        if (global_simdid.mask) {
-                return global_simdid;
-        }
-
-        global_simdid = sn_get_simd_level_local();
-        return global_simdid;
-}
-
 #else
 
 SN_API SIMDID SN_DECL sn_get_simd_level_local()

StgSharp.Native/sn-src/thread.linux.c

@@ -0,0 +1,39 @@
+#include "sn_target.h"
+#include "StgSharpNative.h"
+
+#if SN_IS_OS(SN_OS_LINUX)
+#define _GNU_SOURCE
+#include <pthread.h>
+#include <sched.h>
+#include <unistd.h>
+
+/* Pin the calling thread to a single logical processor.
+ * core_id >= 0 : bind to that logical processor (global index, across groups).
+ * core_id <  0 : clear affinity constraint within the current group. */
+SN_API void SN_DECL sn_set_thread_affinity(int core_id)
+{
+        long ncpus = sysconf(_SC_NPROCESSORS_ONLN);
+
+        cpu_set_t *set = CPU_ALLOC(ncpus);
+        size_t size = CPU_ALLOC_SIZE(ncpus);
+        CPU_ZERO_S(size, set);
+
+        if (core_id < 0) {
+                for (long i = 0; i < ncpus; i++)
+                        CPU_SET_S((int)i, size, set);
+        } else {
+                CPU_SET_S(core_id, size, set);
+        }
+
+        pthread_setaffinity_np(pthread_self(), size, set);
+        CPU_FREE(set);
+}
+
+SN_API void SN_DECL sn_get_numa_overall(int *out_numa_count, int *count_of_each_numa)
+{
+}
+
+SN_API void SN_DECL sn_get_numa_each(int numa_id, int *in_numa_cpuid, int *global_cpuid)
+{
+}
+#endif

StgSharp.Native/sn-src/thread.windows.c

@@ -1,19 +1,13 @@
 #include "sn_target.h"
 #include "StgSharpNative.h"
 
-#if SN_IS_OS(SN_OS_LINUX)
+#if SN_IS_OS(SN_OS_WINDOWS)
-#define _GNU_SOURCE
-#include <pthread.h>
-#include <sched.h>
-#include <unistd.h>
-#endif
 
 /* Pin the calling thread to a single logical processor.
  * core_id >= 0 : bind to that logical processor (global index, across groups).
  * core_id <  0 : clear affinity constraint within the current group. */
 SN_API void SN_DECL sn_set_thread_affinity(int core_id)
 {
-#if SN_IS_OS(SN_OS_WINDOWS)
         if (core_id < 0) {
                 /* Query which group the thread currently belongs to */
                 GROUP_AFFINITY current;
@@ -23,33 +17,53 @@ SN_API void SN_DECL sn_set_thread_affinity(int core_id)
                 GROUP_AFFINITY ga;
                 memset(&ga, 0, sizeof(ga));
                 ga.Group = current.Group;
-                ga.Mask  = (count >= 64) ? ~(KAFFINITY)0
+                ga.Mask = (count >= 64) ? ~(KAFFINITY)0 : ((KAFFINITY)1 << count) - 1;
-                                         : ((KAFFINITY)1 << count) - 1;
                 SetThreadGroupAffinity(GetCurrentThread(), &ga, NULL);
                 return;
         }
         WORD group = (WORD)(core_id / 64);
-        int  index = core_id % 64;
+        int index = core_id % 64;
         DWORD count = GetActiveProcessorCount(group);
         if ((DWORD)index >= count)
                 return;
         GROUP_AFFINITY ga;
         memset(&ga, 0, sizeof(ga));
         ga.Group = group;
-        ga.Mask  = (KAFFINITY)1 << index;
+        ga.Mask = (KAFFINITY)1 << index;
         SetThreadGroupAffinity(GetCurrentThread(), &ga, NULL);
-
-#elif SN_IS_OS(SN_OS_LINUX)
-        cpu_set_t set;
-        if (core_id < 0) {
-                CPU_ZERO(&set);
-                long ncpus = sysconf(_SC_NPROCESSORS_ONLN);
-                for (long i = 0; i < ncpus; i++)
-                        CPU_SET((int)i, &set);
-        } else {
-                CPU_ZERO(&set);
-                CPU_SET(core_id, &set);
-        }
-        pthread_setaffinity_np(pthread_self(), sizeof(set), &set);
-#endif
 }
+
+SN_API int SN_DECL sn_get_numa_overall(int *out_numa_count, int *count_of_each_numa)
+{
+        ULONG highest;
+        if (!GetNumaHighestNodeNumber(&highest))
+                return 0;
+        *out_numa_count = (int)highest + 1;
+
+        for (UCHAR n = 0; n <= (UCHAR)highest; n++) {
+                ULONGLONG mask = 0;
+                GetNumaNodeProcessorMask(n, &mask);
+                count_of_each_numa[n] = (int)__popcnt64(mask);
+        }
+        return 1;
+}
+
+SN_API int SN_DECL sn_get_numa_each(int numa_id, uint64_t *internal, uint64_t *global)
+{
+        ULONGLONG mask = 0;
+        if (!GetNumaNodeProcessorMask((UCHAR)numa_id, &mask))
+                return 0;
+
+        int intra = 0;
+        while (mask) {
+                unsigned long idx;
+                _BitScanForward64(&idx, mask);
+                internal[intra] = intra;
+                global[intra] = (int)idx;
+                intra++;
+                mask &= mask - 1;
+        }
+        return 1;
+}
+
+#endif