set default settings to use minimal block sizes /4 (instead of /8), and no page reclaim on free

include/mimalloc/types.h

@@ -341,12 +341,12 @@ typedef struct mi_page_s {
 #endif
 
 // The max object size are checked to not waste more than 12.5% internally over the page sizes.
-#define MI_SMALL_MAX_OBJ_SIZE             ((MI_SMALL_PAGE_SIZE-MI_PAGE_INFO_SIZE)/8)   // < ~8 KiB
+#define MI_SMALL_MAX_OBJ_SIZE             ((MI_SMALL_PAGE_SIZE-MI_PAGE_INFO_SIZE)/4)   // < ~16 KiB
 #if MI_ENABLE_LARGE_PAGES
-#define MI_MEDIUM_MAX_OBJ_SIZE            ((MI_MEDIUM_PAGE_SIZE-MI_PAGE_INFO_SIZE)/8)  // < ~64 KiB
+#define MI_MEDIUM_MAX_OBJ_SIZE            ((MI_MEDIUM_PAGE_SIZE-MI_PAGE_INFO_SIZE)/4)  // < ~128 KiB
 #define MI_LARGE_MAX_OBJ_SIZE             (MI_LARGE_PAGE_SIZE/8)    // <= 512KiB // note: this must be a nice power of 2 or we get rounding issues with `_mi_bin`
 #else
-#define MI_MEDIUM_MAX_OBJ_SIZE            (MI_MEDIUM_PAGE_SIZE/8)   // <= 64 KiB
+#define MI_MEDIUM_MAX_OBJ_SIZE            (MI_MEDIUM_PAGE_SIZE/4)   // <= 128 KiB
 #define MI_LARGE_MAX_OBJ_SIZE             MI_MEDIUM_MAX_OBJ_SIZE    // note: this must be a nice power of 2 or we get rounding issues with `_mi_bin`
 #endif
 #define MI_LARGE_MAX_OBJ_WSIZE            (MI_LARGE_MAX_OBJ_SIZE/MI_SIZE_SIZE)

src/free.c

@@ -202,10 +202,11 @@ void mi_free(void* p) mi_attr_noexcept
 // Multi-threaded Free (`_mt`)
 // ------------------------------------------------------
 static bool mi_page_unown_from_free(mi_page_t* page, mi_block_t* mt_free);
-static inline bool mi_page_queue_len_is_atmost( mi_heap_t* heap, size_t block_size, size_t atmost) {
+static inline bool mi_page_queue_len_is_lower_as( mi_heap_t* heap, size_t block_size, long atmost) {
+  if (atmost <= 0) { return false; }
   mi_page_queue_t* const pq = mi_page_queue(heap,block_size);
   mi_assert_internal(pq!=NULL);
-  return (pq->count <= atmost);
+  return (pq->count < (size_t)atmost);
   /*
   for(mi_page_t* p = pq->first; p!=NULL; p = p->next, atmost--) {
     if (atmost == 0) { return false; }
@@ -256,12 +257,12 @@ static void mi_decl_noinline mi_free_try_collect_mt(mi_page_t* page, mi_block_t*
       // can we reclaim into this heap?
       if (heap != NULL && heap->allow_page_reclaim) {
         const long reclaim_max = _mi_option_get_fast(mi_option_page_reclaim_max);
-        if ((heap == page->heap && mi_page_queue_len_is_atmost(heap, page->block_size, reclaim_max)) ||  // only reclaim if we were the originating heap, and we have at most N pages already
+        if ((heap == page->heap && mi_page_queue_len_is_lower_as(heap, page->block_size, reclaim_max)) ||  // only reclaim if we were the originating heap, and we have at most N pages already
-          (reclaim_on_free == 1 &&               // OR if the reclaim across heaps is allowed
+            (reclaim_on_free == 1 &&               // OR if the reclaim across heaps is allowed
-            !mi_page_is_used_at_frac(page, 8) &&  //    and the page is not too full
+             !mi_page_is_used_at_frac(page, 8) &&  //    and the page is not too full
-            !heap->tld->is_in_threadpool &&       //    and not part of a threadpool
+             !heap->tld->is_in_threadpool &&       //    and not part of a threadpool
-            _mi_arena_memid_is_suitable(page->memid, heap->exclusive_arena))  // and the memory is suitable
+             _mi_arena_memid_is_suitable(page->memid, heap->exclusive_arena))  // and the memory is suitable
-          )
+           )
         {
           // first remove it from the abandoned pages in the arena -- this waits for any readers to finish
           _mi_arenas_page_unabandon(page);

src/options.c

@@ -169,14 +169,14 @@ static mi_option_desc_t options[_mi_option_last] =
          UNINIT, MI_OPTION(guarded_sample_rate)},       // 1 out of N allocations in the min/max range will be guarded (=4000)
   { 0,   UNINIT, MI_OPTION(guarded_sample_seed)},
   { 10000, UNINIT, MI_OPTION(generic_collect) },        // collect heaps every N (=10000) generic allocation calls
-  { 0,   UNINIT, MI_OPTION_LEGACY(page_reclaim_on_free, abandoned_reclaim_on_free) },// reclaim abandoned pages on a free: -1 = disable completely, 0 = only reclaim into the originating heap, 1 = reclaim on free across heaps
+  { -1,  UNINIT, MI_OPTION_LEGACY(page_reclaim_on_free, abandoned_reclaim_on_free) },// reclaim abandoned pages on a free: -1 = disable completely, 0 = only reclaim into the originating heap, 1 = reclaim on free across heaps
   { 2,   UNINIT, MI_OPTION(page_full_retain) },         // number of (small) pages to retain in the free page queues
   { 4,   UNINIT, MI_OPTION(page_max_candidates) },      // max search to find a best page candidate
   { 0,   UNINIT, MI_OPTION(max_vabits) },               // max virtual address space bits
   { MI_DEFAULT_PAGEMAP_COMMIT,
          UNINIT, MI_OPTION(pagemap_commit) },           // commit the full pagemap upfront?
-  { 1,   UNINIT, MI_OPTION(page_commit_on_demand) },    // commit pages on-demand (2 disables this only on overcommit systems (like Linux))
+  { 0,   UNINIT, MI_OPTION(page_commit_on_demand) },    // commit pages on-demand (2 disables this only on overcommit systems (like Linux))
-  { 16,  UNINIT, MI_OPTION(page_reclaim_max) },         // don't reclaim pages if we already own N pages (in that size class)
+  { 16,  UNINIT, MI_OPTION(page_reclaim_max) },         // don't reclaim pages if we already own < N pages (in that size class)
 };
 
 static void mi_option_init(mi_option_desc_t* desc);