Merge branch 'dev3' into dev3-binx

CMakeLists.txt

@@ -12,8 +12,7 @@ option(MI_XMALLOC           "Enable abort() call on memory allocation failure by
 option(MI_SHOW_ERRORS       "Show error and warning messages by default (only enabled by default in DEBUG mode)" OFF)
 option(MI_GUARDED           "Build with guard pages behind certain object allocations (implies MI_NO_PADDING=ON)" OFF)
 option(MI_USE_CXX           "Use the C++ compiler to compile the library (instead of the C compiler)" OFF)
-
+option(MI_OPT_ARCH          "Only for optimized builds: turn on architecture specific optimizations (for x64: '-march=haswell;-mavx2' (2013), for arm64: '-march=armv8.1-a' (2016))" OFF)
-option(MI_OPT_ARCH          "Only for optimized builds: turn on architecture specific optimizations (for x64: '-march=haswell;-mavx2' (2013), for arm64: '-march=armv8.1-a' (2016))" ON)
 option(MI_OPT_SIMD          "Use SIMD instructions (requires MI_OPT_ARCH to be enabled)" OFF)
 option(MI_SEE_ASM           "Generate assembly files" OFF)
 option(MI_OSX_INTERPOSE     "Use interpose to override standard malloc on macOS" ON)
@@ -121,9 +120,44 @@ if("${CMAKE_BINARY_DIR}" MATCHES ".*(S|s)ecure$")
   set(MI_SECURE "ON")
 endif()
 
+
+# Determine architecture
+set(MI_OPT_ARCH_FLAGS "")
+set(MI_ARCH "unknown")
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86|i[3456]86)$" OR CMAKE_GENERATOR_PLATFORM MATCHES "^(x86|Win32)$")
+  set(MI_ARCH "x86")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|x64|amd64|AMD64)$" OR CMAKE_GENERATOR_PLATFORM STREQUAL "x64" OR "x86_64" IN_LIST CMAKE_OSX_ARCHITECTURES) # must be before arm64
+  set(MI_ARCH "x64")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|armv[89].?|ARM64)$" OR CMAKE_GENERATOR_PLATFORM STREQUAL "ARM64" OR "arm64" IN_LIST CMAKE_OSX_ARCHITECTURES)
+  set(MI_ARCH "arm64")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm|armv[34567]|ARM)$")
+  set(MI_ARCH "arm32")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(riscv|riscv32|riscv64)$")
+  if(CMAKE_SIZEOF_VOID_P==4)
+    set(MI_ARCH "riscv32")
+  else()
+    set(MI_ARCH "riscv64")
+  endif()
+else()
+  set(MI_ARCH ${CMAKE_SYSTEM_PROCESSOR})
+endif()
+message(STATUS "Architecture: ${MI_ARCH}") # (${CMAKE_SYSTEM_PROCESSOR}, ${CMAKE_GENERATOR_PLATFORM}, ${CMAKE_GENERATOR})")
+
+# negative overrides (mainly to support vcpkg features)
+if(MI_NO_USE_CXX)
+  set(MI_USE_CXX "OFF")
+endif()
+if(MI_NO_OPT_ARCH)
+  set(MI_OPT_ARCH "OFF")
+elseif(MI_ARCH STREQUAL "arm64")
+  set(MI_OPT_ARCH "ON")  # enable armv8.1-a by default on arm64 unless MI_NO_OPT_ARCH is set
+endif()
+
+
 # -----------------------------------------------------------------------------
 # Process options
 # -----------------------------------------------------------------------------
+
 if(CMAKE_C_COMPILER_ID STREQUAL "Clang" AND CMAKE_CXX_COMPILER_FRONTEND_VARIANT STREQUAL "MSVC")
   set(MI_CLANG_CL "ON")
 endif()
@@ -143,27 +177,10 @@ if(CMAKE_C_COMPILER_ID MATCHES "Intel")
     list(APPEND mi_cflags -Wall)
 endif()
 
-# negative overrides (mainly to support vcpkg features)
-if(MI_NO_USE_CXX)
-  set(MI_USE_CXX "OFF")
-endif()
-if(MI_NO_OPT_ARCH)
-  set(MI_OPT_ARCH "OFF")
-endif()
-
-
 if(CMAKE_C_COMPILER_ID MATCHES "MSVC|Intel")
   set(MI_USE_CXX "ON")
 endif()
 
-if(CMAKE_BUILD_TYPE MATCHES "Release|RelWithDebInfo")
-  if (NOT MI_OPT_ARCH)
-    message(STATUS "Architecture specific optimizations are disabled (MI_OPT_ARCH=OFF)")
-  endif()
-#else()
-#  set(MI_OPT_ARCH OFF)
-endif()
-
 if(MI_OVERRIDE)
   message(STATUS "Override standard malloc (MI_OVERRIDE=ON)")
   if(APPLE)
@@ -370,28 +387,6 @@ if(MI_WIN_USE_FIXED_TLS)
   list(APPEND mi_defines MI_WIN_USE_FIXED_TLS=1)
 endif()
 
-# Determine architecture
-set(MI_OPT_ARCH_FLAGS "")
-set(MI_ARCH "unknown")
-if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86|i[3456]86)$" OR CMAKE_GENERATOR_PLATFORM MATCHES "^(x86|Win32)$")
-  set(MI_ARCH "x86")
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|x64|amd64|AMD64)$" OR CMAKE_GENERATOR_PLATFORM STREQUAL "x64" OR "x86_64" IN_LIST CMAKE_OSX_ARCHITECTURES) # must be before arm64
-  set(MI_ARCH "x64")
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|armv[89].?|ARM64)$" OR CMAKE_GENERATOR_PLATFORM STREQUAL "ARM64" OR "arm64" IN_LIST CMAKE_OSX_ARCHITECTURES)
-  set(MI_ARCH "arm64")
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm|armv[34567]|ARM)$")
-  set(MI_ARCH "arm32")
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(riscv|riscv32|riscv64)$")
-  if(CMAKE_SIZEOF_VOID_P==4)
-    set(MI_ARCH "riscv32")
-  else()
-    set(MI_ARCH "riscv64")
-  endif()
-else()
-  set(MI_ARCH ${CMAKE_SYSTEM_PROCESSOR})
-endif()
-message(STATUS "Architecture: ${MI_ARCH}") # (${CMAKE_SYSTEM_PROCESSOR}, ${CMAKE_GENERATOR_PLATFORM}, ${CMAKE_GENERATOR})")
-
 # Check /proc/cpuinfo for an SV39 MMU and limit the virtual address bits.
 # (this will skip the aligned hinting in that case. Issue #939, #949)
 if (EXISTS /proc/cpuinfo)
@@ -448,7 +443,6 @@ endif()
 if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang|GNU|Intel" AND NOT CMAKE_SYSTEM_NAME MATCHES "Haiku")
   if(MI_OPT_ARCH)
     if(APPLE AND CMAKE_C_COMPILER_ID STREQUAL "AppleClang" AND CMAKE_OSX_ARCHITECTURES)   # to support multi-arch binaries (#999)
-      set(MI_OPT_ARCH_FLAGS "")
       if("arm64" IN_LIST CMAKE_OSX_ARCHITECTURES)
         list(APPEND MI_OPT_ARCH_FLAGS "-Xarch_arm64;-march=armv8.1-a")
       endif()

include/mimalloc/internal.h

@@ -160,7 +160,7 @@ bool          _mi_os_secure_guard_page_reset_at(void* addr);
 bool          _mi_os_secure_guard_page_reset_before(void* addr);
 
 int           _mi_os_numa_node(void);
-size_t        _mi_os_numa_node_count(void);
+int           _mi_os_numa_node_count(void);
 
 void*         _mi_os_alloc_aligned(size_t size, size_t alignment, bool commit, bool allow_large, mi_memid_t* memid);
 void*         _mi_os_alloc_aligned_at_offset(size_t size, size_t alignment, size_t align_offset, bool commit, bool allow_large, mi_memid_t* memid);

src/arena.c

@@ -1525,17 +1525,17 @@ 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());
+  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;
+  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++;
+    if ((size_t)numa_node < pages_mod) { node_pages++; }
-    int err = mi_reserve_huge_os_pages_at(node_pages, (int)numa_node, timeout_per);
+    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/init.c

@@ -663,7 +663,7 @@ static void mi_detect_cpu_features(void) {
   int32_t cpu_info[4];
   __cpuid(cpu_info, 7);
   _mi_cpu_has_fsrm = ((cpu_info[3] & (1 << 4)) != 0); // bit 4 of EDX : see <https://en.wikipedia.org/wiki/CPUID#EAX=7,_ECX=0:_Extended_Features>
-  _mi_cpu_has_erms = ((cpu_info[2] & (1 << 9)) != 0); // bit 9 of ECX : see <https://en.wikipedia.org/wiki/CPUID#EAX=7,_ECX=0:_Extended_Features>
+  _mi_cpu_has_erms = ((cpu_info[1] & (1 << 9)) != 0); // bit 9 of EBX : see <https://en.wikipedia.org/wiki/CPUID#EAX=7,_ECX=0:_Extended_Features>
 }
 #else
 static void mi_detect_cpu_features(void) {

src/os.c

@@ -694,18 +694,19 @@ static void mi_os_free_huge_os_pages(void* p, size_t size) {
 Support NUMA aware allocation
 -----------------------------------------------------------------------------*/
 
-static _Atomic(size_t)  _mi_numa_node_count; // = 0   // cache the node count
+static _Atomic(int)  _mi_numa_node_count; // = 0   // cache the node count
 
-size_t _mi_os_numa_node_count(void) {
+int _mi_os_numa_node_count(void) {
-  size_t count = mi_atomic_load_acquire(&_mi_numa_node_count);
+  int 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;
+      count = (int)ncount;
     }
     else {
-      count = _mi_prim_numa_node_count(); // or detect dynamically
+      const size_t n = _mi_prim_numa_node_count(); // or detect dynamically
-      if (count == 0) { count = 1; }
+      if (n == 0 || n > INT_MAX) { count = 1; }
+                            else { count = (int)n; }
     }
     mi_atomic_store_release(&_mi_numa_node_count, count); // save it
     _mi_verbose_message("using %zd numa regions\n", count);
@@ -715,12 +716,13 @@ size_t _mi_os_numa_node_count(void) {
 
 
 static int mi_os_numa_node_get(void) {
-  size_t numa_count = _mi_os_numa_node_count();
+  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) {