Using exclusive mode prefetch (#627)

* Using exclusive mode prefetch

The prefetching is always used to move the cache line to the current
core for writing.  This change makes it use exclusive mode prefetch
and enables it as a feature flag for x64.

* Debug platform for BSDs

* CI fixes

* More CI

* Update ARM prefetch

* Update x64 prefetch default

CMakeLists.txt

@@ -386,6 +386,11 @@ if(NOT SNMALLOC_HEADER_ONLY_LIBRARY)
       target_compile_definitions(${name} PRIVATE "SNMALLOC_EXPORT=__attribute__((visibility(\"default\")))")
       target_compile_options(${name} PRIVATE
         -fomit-frame-pointer -ffunction-sections)
+
+      check_cxx_compiler_flag("-Werror -Wextra -Wall -mprfchw" SUPPORT_PREFETCH_WRITE)
+      if (SUPPORT_PREFETCH_WRITE)
+        target_compile_options(${name} PRIVATE -mprfchw)
+      endif()
       # Static TLS model is unsupported on Haiku.
       if (NOT CMAKE_SYSTEM_NAME STREQUAL "Haiku")
         target_compile_options(${name} PRIVATE -ftls-model=initial-exec)

src/snmalloc/aal/aal.h

@@ -147,7 +147,7 @@ namespace snmalloc
     static inline void prefetch(void* ptr) noexcept
     {
 #if __has_builtin(__builtin_prefetch) && !defined(SNMALLOC_NO_AAL_BUILTINS)
-      __builtin_prefetch(ptr);
+      __builtin_prefetch(ptr, 1, 3);
 #else
       Arch::prefetch(ptr);
 #endif

src/snmalloc/aal/aal_arm.h

@@ -54,7 +54,7 @@ namespace snmalloc
 #elif __has_builtin(__builtin_prefetch) && !defined(SNMALLOC_NO_AAL_BUILTINS)
       __builtin_prefetch(ptr);
 #elif defined(SNMALLOC_VA_BITS_64)
-      __asm__ volatile("prfm pldl1keep, [%0]" : "=r"(ptr));
+      __asm__ volatile("prfm pstl1keep, [%0]" : "=r"(ptr));
 #else
       __asm__ volatile("pld\t[%0]" : "=r"(ptr));
 #endif

src/snmalloc/aal/aal_x86.h

@@ -78,7 +78,11 @@ namespace snmalloc
      */
     static inline void prefetch(void* ptr)
     {
-      _mm_prefetch(reinterpret_cast<const char*>(ptr), _MM_HINT_T0);
+#if defined(_MSC_VER)
+      _m_prefetchw(ptr);
+#else
+      _mm_prefetch(reinterpret_cast<const char*>(ptr), _MM_HINT_ET0);
+#endif
     }
 
     /**