Merge c9f0a78432 into fae61ed946

Unlike on Windows, it is perfectly acceptable to set symbol visibility
for static libraries, and it magically works. In fact, it even
beneficially does something useful.

This simplifies the code by avoiding special cases.

CMakeLists.txt

@@ -173,7 +173,8 @@ if(CMAKE_C_COMPILER_ID MATCHES "Intel")
     list(APPEND mi_cflags -Wall)
 endif()
 
-if(CMAKE_C_COMPILER_ID MATCHES "MSVC|Intel")
+# force C++ compilation with msvc or clang-cl to use modern C++ atomics
+if(CMAKE_C_COMPILER_ID MATCHES "MSVC|Intel" OR MI_CLANG_CL)
   set(MI_USE_CXX "ON")
 endif()
 
@@ -724,21 +725,19 @@ if (MI_BUILD_TESTS)
   if(MI_BUILD_SHARED AND NOT (MI_TRACK_ASAN OR MI_DEBUG_TSAN OR MI_DEBUG_UBSAN))
     add_executable(mimalloc-test-stress-dynamic test/test-stress.c)
     target_compile_definitions(mimalloc-test-stress-dynamic PRIVATE ${mi_defines} "USE_STD_MALLOC=1")
-    if(WIN32)
-      target_compile_definitions(mimalloc-test-stress-dynamic PRIVATE "MI_LINK_VERSION=1")
-    endif()
     target_compile_options(mimalloc-test-stress-dynamic PRIVATE ${mi_cflags})
     target_include_directories(mimalloc-test-stress-dynamic PRIVATE include)
-    target_link_libraries(mimalloc-test-stress-dynamic PRIVATE mimalloc ${mi_libraries})  # mi_version
     if(WIN32)
-      add_test(NAME test-stress-dynamic COMMAND ${CMAKE_COMMAND} -E env MIMALLOC_SHOW_STATS=1 $<TARGET_FILE:mimalloc-test-stress-dynamic>)
+      target_compile_definitions(mimalloc-test-stress-dynamic PRIVATE "MI_LINK_VERSION=1")  # link mi_version
+      target_link_libraries(mimalloc-test-stress-dynamic PRIVATE mimalloc ${mi_libraries})  # link mi_version
+      add_test(NAME test-stress-dynamic COMMAND ${CMAKE_COMMAND} -E env MIMALLOC_VERBOSE=1 $<TARGET_FILE:mimalloc-test-stress-dynamic>)
     else()
       if(APPLE)
         set(LD_PRELOAD "DYLD_INSERT_LIBRARIES")
       else()
         set(LD_PRELOAD "LD_PRELOAD")
       endif()
-      add_test(NAME test-stress-dynamic COMMAND ${CMAKE_COMMAND} -E env MIMALLOC_SHOW_STATS=1 ${LD_PRELOAD}=$<TARGET_FILE:mimalloc> $<TARGET_FILE:mimalloc-test-stress-dynamic>)
+      add_test(NAME test-stress-dynamic COMMAND ${CMAKE_COMMAND} -E env MIMALLOC_VERBOSE=1 ${LD_PRELOAD}=$<TARGET_FILE:mimalloc> $<TARGET_FILE:mimalloc-test-stress-dynamic>)
     endif()
   endif()
 endif()

include/mimalloc.h

@@ -60,11 +60,7 @@ terms of the MIT license. A copy of the license can be found in the file
   #define mi_attr_alloc_size2(s1,s2)
   #define mi_attr_alloc_align(p)
 #elif defined(__GNUC__)                 // includes clang and icc
-  #if defined(MI_SHARED_LIB) && defined(MI_SHARED_LIB_EXPORT)
-    #define mi_decl_export              __attribute__((visibility("default")))
-  #else
-    #define mi_decl_export
-  #endif
+  #define mi_decl_export              __attribute__((visibility("default")))
   #define mi_cdecl                      // leads to warnings... __attribute__((cdecl))
   #define mi_decl_restrict
   #define mi_attr_malloc                __attribute__((malloc))

include/mimalloc/atomic.h

@@ -111,6 +111,7 @@ static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub);
 #define mi_atomic_cas_ptr_weak_release(tp,p,exp,des)    mi_atomic_cas_weak_release(p,exp,(tp*)des)
 #define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des)    mi_atomic_cas_weak_acq_rel(p,exp,(tp*)des)
 #define mi_atomic_cas_ptr_strong_release(tp,p,exp,des)  mi_atomic_cas_strong_release(p,exp,(tp*)des)
+#define mi_atomic_cas_ptr_strong_acq_rel(tp,p,exp,des)  mi_atomic_cas_strong_acq_rel(p,exp,(tp*)des)
 #define mi_atomic_exchange_ptr_relaxed(tp,p,x)          mi_atomic_exchange_relaxed(p,(tp*)x)
 #define mi_atomic_exchange_ptr_release(tp,p,x)          mi_atomic_exchange_release(p,(tp*)x)
 #define mi_atomic_exchange_ptr_acq_rel(tp,p,x)          mi_atomic_exchange_acq_rel(p,(tp*)x)
@@ -120,6 +121,7 @@ static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub);
 #define mi_atomic_cas_ptr_weak_release(tp,p,exp,des)    mi_atomic_cas_weak_release(p,exp,des)
 #define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des)    mi_atomic_cas_weak_acq_rel(p,exp,des)
 #define mi_atomic_cas_ptr_strong_release(tp,p,exp,des)  mi_atomic_cas_strong_release(p,exp,des)
+#define mi_atomic_cas_ptr_strong_acq_rel(tp,p,exp,des)  mi_atomic_cas_strong_acq_rel(p,exp,des)
 #define mi_atomic_exchange_ptr_relaxed(tp,p,x)          mi_atomic_exchange_relaxed(p,x)
 #define mi_atomic_exchange_ptr_release(tp,p,x)          mi_atomic_exchange_release(p,x)
 #define mi_atomic_exchange_ptr_acq_rel(tp,p,x)          mi_atomic_exchange_acq_rel(p,x)
@@ -303,6 +305,7 @@ static inline bool mi_atomic_casi64_strong_acq_rel(volatile _Atomic(int64_t*)p,
 #define mi_atomic_cas_ptr_weak_release(tp,p,exp,des)    mi_atomic_cas_weak_release((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des)
 #define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des)    mi_atomic_cas_weak_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des)
 #define mi_atomic_cas_ptr_strong_release(tp,p,exp,des)  mi_atomic_cas_strong_release((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des)
+#define mi_atomic_cas_ptr_strong_acq_rel(tp,p,exp,des)  mi_atomic_cas_strong_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des)
 #define mi_atomic_exchange_ptr_relaxed(tp,p,x)          (tp*)mi_atomic_exchange_relaxed((_Atomic(uintptr_t)*)(p),(uintptr_t)x)
 #define mi_atomic_exchange_ptr_release(tp,p,x)          (tp*)mi_atomic_exchange_release((_Atomic(uintptr_t)*)(p),(uintptr_t)x)
 #define mi_atomic_exchange_ptr_acq_rel(tp,p,x)          (tp*)mi_atomic_exchange_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t)x)

include/mimalloc/internal.h

@@ -140,9 +140,11 @@ void*       _mi_os_alloc_aligned_at_offset(size_t size, size_t alignment, size_t
 void*       _mi_os_get_aligned_hint(size_t try_alignment, size_t size);
 bool        _mi_os_use_large_page(size_t size, size_t alignment);
 size_t      _mi_os_large_page_size(void);
-
 void*       _mi_os_alloc_huge_os_pages(size_t pages, int numa_node, mi_msecs_t max_secs, size_t* pages_reserved, size_t* psize, mi_memid_t* memid);
 
+int         _mi_os_numa_node_count(void);
+int         _mi_os_numa_node(void);
+
 // arena.c
 mi_arena_id_t _mi_arena_id_none(void);
 void        _mi_arena_free(void* p, size_t size, size_t still_committed_size, mi_memid_t memid);
@@ -813,24 +815,6 @@ static inline uintptr_t _mi_random_shuffle(uintptr_t x) {
   return x;
 }
 
-// -------------------------------------------------------------------
-// Optimize numa node access for the common case (= one node)
-// -------------------------------------------------------------------
-
-int    _mi_os_numa_node_get(void);
-size_t _mi_os_numa_node_count_get(void);
-
-extern mi_decl_hidden _Atomic(size_t) _mi_numa_node_count;
-static inline int _mi_os_numa_node(void) {
-  if mi_likely(mi_atomic_load_relaxed(&_mi_numa_node_count) == 1) { return 0; }
-  else return _mi_os_numa_node_get();
-}
-static inline size_t _mi_os_numa_node_count(void) {
-  const size_t count = mi_atomic_load_relaxed(&_mi_numa_node_count);
-  if mi_likely(count > 0) { return count; }
-  else return _mi_os_numa_node_count_get();
-}
-
 
 
 // -----------------------------------------------------------------------

readme.md

@@ -84,7 +84,7 @@ Enjoy!
 
 ### Releases
 
-* 2025-03-28, `v1.9.3`, `v2.2.3`, `v3.0.3` (beta): Various small bug and build fixes, including:
+* 2025-03-28, `v1.9.3`, `v2.2.3`, `v3.0.3` (beta) : Various small bug and build fixes, including:
   fix arm32 pre v7 builds, fix mingw build, get runtime statistics, improve statistic commit counts, 
   fix execution on non BMI1 x64 systems. 
 * 2025-03-06, `v1.9.2`, `v2.2.2`, `v3.0.2-beta`: Various small bug and build fixes. 
@@ -177,7 +177,7 @@ mimalloc is used in various large scale low-latency services and programs, for e
 
 Open `ide/vs2022/mimalloc.sln` in Visual Studio 2022 and build.
 The `mimalloc-lib` project builds a static library (in `out/msvc-x64`), while the
-`mimalloc-override-dll` project builds DLL for overriding malloc
+`mimalloc-override-dll` project builds a DLL for overriding malloc
 in the entire program.
 
 ## Linux, macOS, BSD, etc.

src/arena.c

@@ -1007,17 +1007,17 @@ int mi_reserve_huge_os_pages_interleave(size_t pages, size_t numa_nodes, size_t
   if (pages == 0) return 0;
 
   // pages per numa node
-  size_t numa_count = (numa_nodes > 0 ? numa_nodes : _mi_os_numa_node_count());
-  if (numa_count <= 0) numa_count = 1;
+  int numa_count = (numa_nodes > 0 && numa_nodes <= INT_MAX ? (int)numa_nodes : _mi_os_numa_node_count());
+  if (numa_count == 0) numa_count = 1;
   const size_t pages_per = pages / numa_count;
   const size_t pages_mod = pages % numa_count;
   const size_t timeout_per = (timeout_msecs==0 ? 0 : (timeout_msecs / numa_count) + 50);
 
   // reserve evenly among numa nodes
-  for (size_t numa_node = 0; numa_node < numa_count && pages > 0; numa_node++) {
+  for (int numa_node = 0; numa_node < numa_count && pages > 0; numa_node++) {
     size_t node_pages = pages_per;  // can be 0
-    if (numa_node < pages_mod) node_pages++;
-    int err = mi_reserve_huge_os_pages_at(node_pages, (int)numa_node, timeout_per);
+    if ((size_t)numa_node < pages_mod) node_pages++;
+    int err = mi_reserve_huge_os_pages_at(node_pages, numa_node, timeout_per);
     if (err) return err;
     if (pages < node_pages) {
       pages = 0;

src/free.c

@@ -340,7 +340,10 @@ mi_decl_nodiscard size_t mi_usable_size(const void* p) mi_attr_noexcept {
 
 void mi_free_size(void* p, size_t size) mi_attr_noexcept {
   MI_UNUSED_RELEASE(size);
-  mi_assert(p == NULL || size <= _mi_usable_size(p,"mi_free_size"));
+  #if MI_DEBUG
+  const size_t available = _mi_usable_size(p,"mi_free_size");
+  mi_assert(p == NULL || size <= available || available == 0 /* invalid pointer */ );
+  #endif
   mi_free(p);
 }
 

src/os.c

@@ -696,34 +696,47 @@ static void mi_os_free_huge_os_pages(void* p, size_t size) {
   }
 }
 
+
 /* ----------------------------------------------------------------------------
 Support NUMA aware allocation
 -----------------------------------------------------------------------------*/
 
-_Atomic(size_t)  _mi_numa_node_count; // = 0   // cache the node count
+static _Atomic(size_t) mi_numa_node_count; // = 0   // cache the node count
 
-size_t _mi_os_numa_node_count_get(void) {
-  size_t count = mi_atomic_load_acquire(&_mi_numa_node_count);
-  if (count <= 0) {
+int _mi_os_numa_node_count(void) {
+  size_t count = mi_atomic_load_acquire(&mi_numa_node_count);
+  if mi_unlikely(count == 0) {
     long ncount = mi_option_get(mi_option_use_numa_nodes); // given explicitly?
-    if (ncount > 0) {
+    if (ncount > 0 && ncount < INT_MAX) {
       count = (size_t)ncount;
     }
     else {
-      count = _mi_prim_numa_node_count(); // or detect dynamically
-      if (count == 0) count = 1;
+      const size_t n = _mi_prim_numa_node_count(); // or detect dynamically
+      if (n == 0 || n > INT_MAX) { count = 1; }
+                            else { count = n; }
     }
-    mi_atomic_store_release(&_mi_numa_node_count, count); // save it
+    mi_atomic_store_release(&mi_numa_node_count, count); // save it
     _mi_verbose_message("using %zd numa regions\n", count);
   }
-  return count;
+  mi_assert_internal(count > 0 && count <= INT_MAX);
+  return (int)count;
 }
 
-int _mi_os_numa_node_get(void) {
-  size_t numa_count = _mi_os_numa_node_count();
+static int mi_os_numa_node_get(void) {
+  int numa_count = _mi_os_numa_node_count();
   if (numa_count<=1) return 0; // optimize on single numa node systems: always node 0
   // never more than the node count and >= 0
-  size_t numa_node = _mi_prim_numa_node();
+  const size_t n = _mi_prim_numa_node();
+  int numa_node = (n < INT_MAX ? (int)n : 0);
   if (numa_node >= numa_count) { numa_node = numa_node % numa_count; }
-  return (int)numa_node;
+  return numa_node;
+}
+
+int _mi_os_numa_node(void) {
+  if mi_likely(mi_atomic_load_relaxed(&mi_numa_node_count) == 1) { 
+    return 0; 
+  }
+  else {
+    return mi_os_numa_node_get();
+  }
 }

src/prim/unix/prim.c

@@ -31,10 +31,10 @@ terms of the MIT license. A copy of the license can be found in the file
 
 #if defined(__linux__)
   #include <features.h>
-  #include <linux/prctl.h>  // PR_SET_VMA
-  //#if defined(MI_NO_THP)
-  #include <sys/prctl.h>    // THP disable
-  //#endif
+  #include <sys/prctl.h>    // THP disable, PR_SET_VMA
+  #if !defined(PR_SET_VMA)
+  #include <linux/prctl.h>
+  #endif
   #if defined(__GLIBC__)
   #include <linux/mman.h>   // linux mmap flags
   #else
@@ -208,7 +208,7 @@ static int unix_madvise(void* addr, size_t size, int advice) {
 
 static void* unix_mmap_prim(void* addr, size_t size, int protect_flags, int flags, int fd) {
   void* p = mmap(addr, size, protect_flags, flags, fd, 0 /* offset */);
-  #if (defined(__linux__) && defined(PR_SET_VMA))
+  #if defined(__linux__) && defined(PR_SET_VMA)
   if (p!=MAP_FAILED && p!=NULL) {
     prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, p, size, "mimalloc");
   }

src/stats.c

@@ -218,12 +218,14 @@ static void mi_stat_peak_print(const mi_stat_count_t* stat, const char* msg, int
   _mi_fprintf(out, arg, "\n");
 }
 
+#if MI_STAT>1
 static void mi_stat_total_print(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg) {
   _mi_fprintf(out, arg, "%10s:", msg);
   _mi_fprintf(out, arg, "%12s", " ");  // no peak
   mi_print_amount(stat->total, unit, out, arg);
   _mi_fprintf(out, arg, "\n");
 }
+#endif
 
 static void mi_stat_counter_print(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg ) {
   _mi_fprintf(out, arg, "%10s:", msg);