L4 map use no sync fast route when enouge predict ahead

src/Nitload.Common/HighPerformance/Memory/L4.Map.cs

@@ -35,11 +35,16 @@ namespace Nitload.Common.HighPerformance.Memory
                 {
                     try
                     {
-                        Predict();
-                        _ = Interlocked.Exchange(ref _swiss_swap_lock, 1);
-                        while (Volatile.Read(ref _swiss_reader_counter) != 0) { }
+                        int effectivePredictionCount = Predict();
+                        if (effectivePredictionCount > 0)
+                        {
+                            _ = Interlocked.Exchange(ref _swiss_swap_lock, 1);
+                            while (Volatile.Read(ref _swiss_reader_counter) != 0) { }
+
+                            _addressMap.SwapBuffers();
+                            Volatile.Write(ref _predict_count, effectivePredictionCount);
+                        }
 
-                        _addressMap.SwapBuffers();
                         if (_addressMap.TryGet(source, out handle)) {
                             throw new InvalidOperationException($"Unexpected map request: [address source: 0x{source.Address:x16}; layout profile: {source.Description:x16}].");
                         }
@@ -52,7 +57,7 @@ namespace Nitload.Common.HighPerformance.Memory
                 } else if (!TryGetAddress(source, out handle)) {
                     throw new InvalidOperationException($"Unexpected map request: [address source: 0x{source.Address:x16}; layout profile: {source.Description:x16}].");
                 }
-            } else if (!TryGetAddress(source, out handle)) {
+            } else if (!_addressMap.TryGet(source, out handle)) {
                 throw new InvalidOperationException($"Unexpected map request: [address source: 0x{source.Address:x16}; layout profile: {source.Description:x16}].");
             }
 

src/Nitload.Common/HighPerformance/Memory/L4.Predict.cs

@@ -16,7 +16,7 @@ namespace Nitload.Common.HighPerformance.Memory
     public unsafe partial class L4
     {
 
-        internal partial void Predict() { }
+        internal partial int Predict() => 0;
 
     }
 }

src/Nitload.Common/HighPerformance/Memory/L4.cs

@@ -81,7 +81,7 @@ namespace Nitload.Common.HighPerformance.Memory
             return _bufferAllocator.TryAlloc(size, out handle);
         }
 
-        internal partial void Predict();
+        internal partial int Predict();
 
     }
 }

src/Nitload.Native/nif-head/NitloadFrameWork.h

@@ -1,39 +0,0 @@
-#ifndef NIF_NATIVE_H
-#define NIF_NATIVE_H
-
-#ifdef __cplusplus
-#error "This header is not meant to be included in C++ code. Nitload.Native is a C library and should be used with C linkage."
-#endif
-
-#include "nif_internal.h"
-#include "nif_target.h"
-
-#include <immintrin.h>
-#include <assert.h>
-#include <glfw_function.h>
-
-#define byte char
-#define cs_char uint16_t
-#define nif_null_assert(ptr) assert(ptr != NULL)
-
-#ifndef NIF_
-#endif
-
-NIF_API void NIF_DECL nif_load_glfw_functions(GLFWFunctionTable *table);
-NIF_API void NIF_DECL nif_load_intrinsic_function(void *intrinsic_context, uint64_t id);
-NIF_API SIMDID NIF_DECL nif_get_simd_level_global();
-/* Per-core variant: always re-executes CPUID, never caches.
- * Must be called on the target core (after affinity is set). */
-NIF_API SIMDID NIF_DECL nif_get_simd_level_local();
-NIF_API SIMDID NIF_DECL nif_get_unite_simd(SIMDID left, SIMDID right);
-NIF_API int NIF_DECL nif_compare_simd(SIMDID left, SIMDID right);
-NIF_API void NIF_DECL nif_set_thread_affinity(int cpu_index, int processor_group_id);
-NIF_API void* NIF_DECL nif_numa_alloc(size_t size, int numa_id);
-NIF_API void NIF_DECL nif_numa_free(void *ptr, size_t size, int numa_id);
-NIF_API int NIF_DECL nif_get_numa_each(int numa_id,
-									  uint32_t *processor_group_ids,
-									  uint32_t *socket_ids,
-									  uint32_t *cpu_indices,
-									  uint32_t *numa_node_ids);
-NIF_API int NIF_DECL nif_get_numa_overall(int *out_numa_count, int *count_of_each_numa);
-#endif