From 6a50807e01773951e727d897cdd3549e8e4249ce Mon Sep 17 00:00:00 2001
From: anematode <timothy.herchen@gmail.com>
Date: Sun, 10 May 2026 11:46:31 +0200
Subject: [PATCH] On Linux, allow shared-memory network weights to be backed by
 huge pages

This does require the user to have set `/sys/kernel/mm/transparent_hugepage/shmem_enabled` to something other than `never` (unfortunately the default, at least on Ubuntu).

```
Result of  50 runs
==================
base (...kfish.master) =    1814244  +/- 4396
test (./stockfish    ) =    1823756  +/- 3448
diff                   =      +9512  +/- 5496

speedup         = +0.0052
P(speedup > 0) =  0.9996
```

```
 Performance counter stats for './stockfish bench':

           237,561      dTLB-misses
     5,783,003,231      cycles

 Performance counter stats for './bins/stockfish.master bench':

         2,219,832      dTLB-misses
     5,932,779,255      cycles
```

closes https://github.com/official-stockfish/Stockfish/pull/6805

No functional change
---
 src/shm_linux.h | 41 +++++++++++++++++++++--------------------
 1 file changed, 21 insertions(+), 20 deletions(-)

diff --git a/src/shm_linux.h b/src/shm_linux.h
index 29c9e90f5..fd5a0d27a 100644
--- a/src/shm_linux.h
+++ b/src/shm_linux.h
@@ -141,23 +141,6 @@ class CleanupHooks {
 inline std::once_flag CleanupHooks::register_once_;
 
 
-inline int portable_fallocate(int fd, off_t offset, off_t length) {
-#ifdef __APPLE__
-    fstore_t store = {F_ALLOCATECONTIG, F_PEOFPOSMODE, offset, length, 0};
-    int      ret   = fcntl(fd, F_PREALLOCATE, &store);
-    if (ret == -1)
-    {
-        store.fst_flags = F_ALLOCATEALL;
-        ret             = fcntl(fd, F_PREALLOCATE, &store);
-    }
-    if (ret != -1)
-        ret = ftruncate(fd, offset + length);
-    return ret;
-#else
-    return posix_fallocate(fd, offset, length);
-#endif
-}
-
 }  // namespace detail
 
 template<typename T>
@@ -599,9 +582,6 @@ class SharedMemory: public detail::SharedMemoryBase {
         if (ftruncate(fd_, static_cast<off_t>(total_size_)) == -1)
             return false;
 
-        if (detail::portable_fallocate(fd_, 0, static_cast<off_t>(total_size_)) != 0)
-            return false;
-
         mapped_ptr_ = mmap(nullptr, total_size_, PROT_READ | PROT_WRITE, MAP_SHARED, fd_, 0);
         if (mapped_ptr_ == MAP_FAILED)
         {
@@ -609,6 +589,23 @@ class SharedMemory: public detail::SharedMemoryBase {
             return false;
         }
 
+#ifdef MADV_POPULATE_WRITE
+        // Pre-populate, attempting first with THP, to guarantee that the memory
+        // is allocated and avoid crashing on the first write.
+        madvise(mapped_ptr_, total_size_, MADV_HUGEPAGE);
+
+        const bool populated = madvise(mapped_ptr_, total_size_, MADV_POPULATE_WRITE) == 0;
+#else
+        const bool populated = false;
+#endif
+        // If the THP population failed, try with fallocate
+        if (!populated && posix_fallocate(fd_, 0, static_cast<off_t>(total_size_)) != 0)
+        {
+            munmap(mapped_ptr_, total_size_);
+            mapped_ptr_ = nullptr;
+            return false;
+        }
+
         data_ptr_ = static_cast<T*>(mapped_ptr_);
         header_ptr_ =
           reinterpret_cast<detail::ShmHeader*>(static_cast<char*>(mapped_ptr_) + sizeof(T));
@@ -642,6 +639,10 @@ class SharedMemory: public detail::SharedMemoryBase {
             return false;
         }
 
+#ifdef MADV_HUGEPAGE
+        madvise(mapped_ptr_, total_size_, MADV_HUGEPAGE);
+#endif
+
         data_ptr_   = static_cast<T*>(mapped_ptr_);
         header_ptr_ = std::launder(
           reinterpret_cast<detail::ShmHeader*>(static_cast<char*>(mapped_ptr_) + sizeof(T)));