Changeset 58fe85a for libcfa/src/heap.cfa

Timestamp:

Jan 7, 2021, 3:27:00 PM (5 years ago)

Author:

Thierry Delisle <tdelisle@…>

Branches:

ADT, arm-eh, ast-experimental, enum, forall-pointer-decay, jacob/cs343-translation, master, new-ast-unique-expr, pthread-emulation, qualifiedEnum

Children:

2b4daf2, 64aeca0

Parents:

3c64c668 (diff), eef8dfb (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the (diff) links above to see all the changes relative to each parent.

Message:

Merge branch 'master' into park_unpark

File:

• libcfa/src/heap.cfa (modified) (59 diffs)

libcfa/src/heap.cfa

@@ -5 +5 @@
 // file "LICENCE" distributed with Cforall.
 //
-// heap.c --
+// heap.cfa --
 //
 // Author           : Peter A. Buhr
 // Created On       : Tue Dec 19 21:58:35 2017
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Tue Feb  4 10:04:51 2020
-// Update Count     : 648
+// Last Modified On : Wed Dec 16 12:28:25 2020
+// Update Count     : 1023
 //
 
 #include <unistd.h>                                                                             // sbrk, sysconf
+#include <stdlib.h>                                                                             // EXIT_FAILURE
 #include <stdbool.h>                                                                    // true, false
 #include <stdio.h>                                                                              // snprintf, fileno
 #include <errno.h>                                                                              // errno
 #include <string.h>                                                                             // memset, memcpy
-extern "C" {
+#include <limits.h>                                                                             // ULONG_MAX
+#include <malloc.h>                                                                             // memalign, malloc_usable_size
 #include <sys/mman.h>                                                                   // mmap, munmap
-} // extern "C"
-
-// #comment TD : Many of these should be merged into math I believe
-#include "bits/align.hfa"                                                               // libPow2
+
+#include "bits/align.hfa"                                                               // libAlign
 #include "bits/defs.hfa"                                                                // likely, unlikely
 #include "bits/locks.hfa"                                                               // __spinlock_t
 #include "startup.hfa"                                                                  // STARTUP_PRIORITY_MEMORY
-//#include "stdlib.hfa"                                                                 // bsearchl
-#include "malloc.h"
-
-#define MIN(x, y) (y > x ? x : y)
+#include "math.hfa"                                                                             // ceiling
+#include "bitmanip.hfa"                                                                 // is_pow2, ceiling2
 
 static bool traceHeap = false;
@@ -74 +72 @@
         // Define the default extension heap amount in units of bytes. When the uC++ supplied heap reaches the brk address,
         // the brk address is extended by the extension amount.
-        __CFA_DEFAULT_HEAP_EXPANSION__ = (1 * 1024 * 1024),
+        __CFA_DEFAULT_HEAP_EXPANSION__ = (10 * 1024 * 1024),
 
         // Define the mmap crossover point during allocation. Allocations less than this amount are allocated from buckets;
@@ -91 +89 @@
 
 #ifdef __CFA_DEBUG__
-static unsigned int allocFree;                                                  // running total of allocations minus frees
+static size_t allocUnfreed;                                                             // running total of allocations minus frees
 
 static void prtUnfreed() {
-        if ( allocFree != 0 ) {
+        if ( allocUnfreed != 0 ) {
                 // DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT.
                 char helpText[512];
-                int len = snprintf( helpText, sizeof(helpText), "CFA warning (UNIX pid:%ld) : program terminating with %u(0x%x) bytes of storage allocated but not freed.\n"
+                int len = snprintf( helpText, sizeof(helpText), "CFA warning (UNIX pid:%ld) : program terminating with %zu(0x%zx) bytes of storage allocated but not freed.\n"
                                                         "Possible cause is unfreed storage allocated by the program or system/library routines called from the program.\n",
-                                                        (long int)getpid(), allocFree, allocFree ); // always print the UNIX pid
+                                                        (long int)getpid(), allocUnfreed, allocUnfreed ); // always print the UNIX pid
                 __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
         } // if
@@ -106 +104 @@
 extern "C" {
         void heapAppStart() {                                                           // called by __cfaabi_appready_startup
-                allocFree = 0;
+                allocUnfreed = 0;
         } // heapAppStart
 
@@ -118 +116 @@
 
 // statically allocated variables => zero filled.
-static size_t pageSize;                                                                 // architecture pagesize
+size_t __page_size;                                                                             // architecture pagesize
+int __map_prot;                                                                                 // common mmap/mprotect protection
 static size_t heapExpand;                                                               // sbrk advance
 static size_t mmapStart;                                                                // cross over point for mmap
@@ -127 +126 @@
 #define LOCKFREE 1
 #define BUCKETLOCK SPINLOCK
-#if BUCKETLOCK == LOCKFREE
-#include <uStackLF.h>
+#if BUCKETLOCK == SPINLOCK
+#elif BUCKETLOCK == LOCKFREE
+#include <stackLockFree.hfa>
+#else
+        #error undefined lock type for bucket lock
 #endif // LOCKFREE
 
@@ -136 +138 @@
 
 struct HeapManager {
-//      struct FreeHeader;                                                                      // forward declaration
-
         struct Storage {
                 struct Header {                                                                 // header
@@ -145 +145 @@
                                                 struct {                                                // 4-byte word => 8-byte header, 8-byte word => 16-byte header
                                                         #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4
-                                                        uint32_t padding;                       // unused, force home/blocksize to overlay alignment in fake header
+                                                        uint64_t padding;                       // unused, force home/blocksize to overlay alignment in fake header
                                                         #endif // __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4
 
                                                         union {
-//                                                              FreeHeader * home;              // allocated block points back to home locations (must overlay alignment)
+                                                                // FreeHeader * home;           // allocated block points back to home locations (must overlay alignment)
+                                                                // 2nd low-order bit => zero filled
                                                                 void * home;                    // allocated block points back to home locations (must overlay alignment)
                                                                 size_t blockSize;               // size for munmap (must overlay alignment)
-                                                                #if BUCKLOCK == SPINLOCK
+                                                                #if BUCKETLOCK == SPINLOCK
                                                                 Storage * next;                 // freed block points next freed block of same size
                                                                 #endif // SPINLOCK
                                                         };
+                                                        size_t size;                            // allocation size in bytes
 
                                                         #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4
-                                                        uint32_t padding;                       // unused, force home/blocksize to overlay alignment in fake header
+                                                        uint64_t padding;                       // unused, force home/blocksize to overlay alignment in fake header
                                                         #endif // __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4
                                                 };
-                                                // future code
-                                                #if BUCKLOCK == LOCKFREE
-                                                Stack<Storage>::Link next;              // freed block points next freed block of same size (double-wide)
+                                                #if BUCKETLOCK == LOCKFREE
+                                                Link(Storage) next;                             // freed block points next freed block of same size (double-wide)
                                                 #endif // LOCKFREE
                                         };
                                 } real; // RealHeader
+
                                 struct FakeHeader {
                                         #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
-                                        uint32_t alignment;                                     // low-order bits of home/blockSize used for tricks
+                                        uint32_t alignment;                                     // 1st low-order bit => fake header & alignment
                                         #endif // __ORDER_LITTLE_ENDIAN__
 
@@ -187 +189 @@
 
         struct FreeHeader {
-                #if BUCKLOCK == SPINLOCK
+                #if BUCKETLOCK == SPINLOCK
                 __spinlock_t lock;                                                              // must be first field for alignment
                 Storage * freeList;
-                #elif BUCKLOCK == LOCKFREE
-                // future code
-                StackLF<Storage> freeList;
                 #else
-                        #error undefined lock type for bucket lock
-                #endif // SPINLOCK
+                StackLF(Storage) freeList;
+                #endif // BUCKETLOCK
                 size_t blockSize;                                                               // size of allocations on this list
         }; // FreeHeader
@@ -208 +207 @@
 }; // HeapManager
 
+#if BUCKETLOCK == LOCKFREE
+static inline {
+        Link(HeapManager.Storage) * ?`next( HeapManager.Storage * this ) { return &this->header.kind.real.next; }
+        void ?{}( HeapManager.FreeHeader & ) {}
+        void ^?{}( HeapManager.FreeHeader & ) {}
+} // distribution
+#endif // LOCKFREE
+
 static inline size_t getKey( const HeapManager.FreeHeader & freeheader ) { return freeheader.blockSize; }
 
@@ -214 +221 @@
 #define __STATISTICS__
 
-// Bucket size must be multiple of 16.
-// Powers of 2 are common allocation sizes, so make powers of 2 generate the minimum required size.
+// Size of array must harmonize with NoBucketSizes and individual bucket sizes must be multiple of 16.
+// Smaller multiples of 16 and powers of 2 are common allocation sizes, so make them generate the minimum required bucket size.
+// malloc(0) returns 0p, so no bucket is necessary for 0 bytes returning an address that can be freed.
 static const unsigned int bucketSizes[] @= {                    // different bucket sizes
-        16, 32, 48, 64 + sizeof(HeapManager.Storage), // 4
-        96, 112, 128 + sizeof(HeapManager.Storage), // 3
+        16 + sizeof(HeapManager.Storage), 32 + sizeof(HeapManager.Storage), 48 + sizeof(HeapManager.Storage), 64 + sizeof(HeapManager.Storage), // 4
+        96 + sizeof(HeapManager.Storage), 112 + sizeof(HeapManager.Storage), 128 + sizeof(HeapManager.Storage), // 3
         160, 192, 224, 256 + sizeof(HeapManager.Storage), // 4
         320, 384, 448, 512 + sizeof(HeapManager.Storage), // 4
@@ -236 +244 @@
 };
 
-static_assert( NoBucketSizes == sizeof(bucketSizes) / sizeof(bucketSizes[0]), "size of bucket array wrong" );
+static_assert( NoBucketSizes == sizeof(bucketSizes) / sizeof(bucketSizes[0] ), "size of bucket array wrong" );
 
 #ifdef FASTLOOKUP
@@ -243 +251 @@
 #endif // FASTLOOKUP
 
-static int mmapFd = -1;                                                                 // fake or actual fd for anonymous file
+static const off_t mmapFd = -1;                                                 // fake or actual fd for anonymous file
 #ifdef __CFA_DEBUG__
 static bool heapBoot = 0;                                                               // detect recursion during boot
 #endif // __CFA_DEBUG__
+
+// The constructor for heapManager is called explicitly in memory_startup.
 static HeapManager heapManager __attribute__(( aligned (128) )) @= {}; // size of cache line to prevent false sharing
 
@@ -252 +262 @@
 #ifdef __STATISTICS__
 // Heap statistics counters.
+static unsigned int malloc_calls;
+static unsigned long long int malloc_storage;
+static unsigned int aalloc_calls;
+static unsigned long long int aalloc_storage;
+static unsigned int calloc_calls;
+static unsigned long long int calloc_storage;
+static unsigned int memalign_calls;
+static unsigned long long int memalign_storage;
+static unsigned int amemalign_calls;
+static unsigned long long int amemalign_storage;
+static unsigned int cmemalign_calls;
+static unsigned long long int cmemalign_storage;
+static unsigned int resize_calls;
+static unsigned long long int resize_storage;
+static unsigned int realloc_calls;
+static unsigned long long int realloc_storage;
+static unsigned int free_calls;
+static unsigned long long int free_storage;
+static unsigned int mmap_calls;
 static unsigned long long int mmap_storage;
-static unsigned int mmap_calls;
+static unsigned int munmap_calls;
 static unsigned long long int munmap_storage;
-static unsigned int munmap_calls;
+static unsigned int sbrk_calls;
 static unsigned long long int sbrk_storage;
-static unsigned int sbrk_calls;
-static unsigned long long int malloc_storage;
-static unsigned int malloc_calls;
-static unsigned long long int free_storage;
-static unsigned int free_calls;
-static unsigned long long int calloc_storage;
-static unsigned int calloc_calls;
-static unsigned long long int memalign_storage;
-static unsigned int memalign_calls;
-static unsigned long long int cmemalign_storage;
-static unsigned int cmemalign_calls;
-static unsigned long long int realloc_storage;
-static unsigned int realloc_calls;
 // Statistics file descriptor (changed by malloc_stats_fd).
-static int statfd = STDERR_FILENO;                                              // default stderr
+static int stat_fd = STDERR_FILENO;                                             // default stderr
 
 // Use "write" because streams may be shutdown when calls are made.
 static void printStats() {
-        char helpText[512];
+        char helpText[1024];
         __cfaabi_bits_print_buffer( STDERR_FILENO, helpText, sizeof(helpText),
                                                                         "\nHeap statistics:\n"
                                                                         "  malloc: calls %u / storage %llu\n"
+                                                                        "  aalloc: calls %u / storage %llu\n"
                                                                         "  calloc: calls %u / storage %llu\n"
                                                                         "  memalign: calls %u / storage %llu\n"
+                                                                        "  amemalign: calls %u / storage %llu\n"
                                                                         "  cmemalign: calls %u / storage %llu\n"
+                                                                        "  resize: calls %u / storage %llu\n"
                                                                         "  realloc: calls %u / storage %llu\n"
                                                                         "  free: calls %u / storage %llu\n"
@@ -288 +307 @@
                                                                         "  sbrk: calls %u / storage %llu\n",
                                                                         malloc_calls, malloc_storage,
+                                                                        aalloc_calls, aalloc_storage,
                                                                         calloc_calls, calloc_storage,
                                                                         memalign_calls, memalign_storage,
+                                                                        amemalign_calls, amemalign_storage,
                                                                         cmemalign_calls, cmemalign_storage,
+                                                                        resize_calls, resize_storage,
                                                                         realloc_calls, realloc_storage,
                                                                         free_calls, free_storage,
@@ -300 +322 @@
 
 static int printStatsXML( FILE * stream ) {                             // see malloc_info
-        char helpText[512];
+        char helpText[1024];
         int len = snprintf( helpText, sizeof(helpText),
                                                 "<malloc version=\"1\">\n"
@@ -307 +329 @@
                                                 "</sizes>\n"
                                                 "<total type=\"malloc\" count=\"%u\" size=\"%llu\"/>\n"
+                                                "<total type=\"aalloc\" count=\"%u\" size=\"%llu\"/>\n"
                                                 "<total type=\"calloc\" count=\"%u\" size=\"%llu\"/>\n"
                                                 "<total type=\"memalign\" count=\"%u\" size=\"%llu\"/>\n"
+                                                "<total type=\"amemalign\" count=\"%u\" size=\"%llu\"/>\n"
                                                 "<total type=\"cmemalign\" count=\"%u\" size=\"%llu\"/>\n"
+                                                "<total type=\"resize\" count=\"%u\" size=\"%llu\"/>\n"
                                                 "<total type=\"realloc\" count=\"%u\" size=\"%llu\"/>\n"
                                                 "<total type=\"free\" count=\"%u\" size=\"%llu\"/>\n"
@@ -317 +342 @@
                                                 "</malloc>",
                                                 malloc_calls, malloc_storage,
+                                                aalloc_calls, aalloc_storage,
                                                 calloc_calls, calloc_storage,
                                                 memalign_calls, memalign_storage,
+                                                amemalign_calls, amemalign_storage,
                                                 cmemalign_calls, cmemalign_storage,
+                                                resize_calls, resize_storage,
                                                 realloc_calls, realloc_storage,
                                                 free_calls, free_storage,
@@ -332 +360 @@
 
 
-// static inline void noMemory() {
-//      abort( "Heap memory exhausted at %zu bytes.\n"
-//                 "Possible cause is very large memory allocation and/or large amount of unfreed storage allocated by the program or system/library routines.",
-//                 ((char *)(sbrk( 0 )) - (char *)(heapManager.heapBegin)) );
-// } // noMemory
-
-
-static inline void checkAlign( size_t alignment ) {
-        if ( alignment < libAlign() || ! libPow2( alignment ) ) {
-                abort( "Alignment %zu for memory allocation is less than %d and/or not a power of 2.", alignment, libAlign() );
-        } // if
-} // checkAlign
-
-
-static inline bool setHeapExpand( size_t value ) {
-  if ( heapExpand < pageSize ) return true;
-        heapExpand = value;
-        return false;
-} // setHeapExpand
-
-
 // thunk problem
 size_t Bsearchl( unsigned int key, const unsigned int * vals, size_t dim ) {
@@ -369 +376 @@
 
 static inline bool setMmapStart( size_t value ) {               // true => mmapped, false => sbrk
-  if ( value < pageSize || bucketSizes[NoBucketSizes - 1] < value ) return true;
+  if ( value < __page_size || bucketSizes[NoBucketSizes - 1] < value ) return false;
         mmapStart = value;                                                                      // set global
 
@@ -376 +383 @@
         assert( maxBucketsUsed < NoBucketSizes );                       // subscript failure ?
         assert( mmapStart <= bucketSizes[maxBucketsUsed] ); // search failure ?
-        return false;
+        return true;
 } // setMmapStart
+
+
+// <-------+----------------------------------------------------> bsize (bucket size)
+// |header |addr
+//==================================================================================
+//                   align/offset |
+// <-----------------<------------+-----------------------------> bsize (bucket size)
+//                   |fake-header | addr
+#define headerAddr( addr ) ((HeapManager.Storage.Header *)( (char *)addr - sizeof(HeapManager.Storage) ))
+#define realHeader( header ) ((HeapManager.Storage.Header *)((char *)header - header->kind.fake.offset))
+
+// <-------<<--------------------- dsize ---------------------->> bsize (bucket size)
+// |header |addr
+//==================================================================================
+//                   align/offset |
+// <------------------------------<<---------- dsize --------->>> bsize (bucket size)
+//                   |fake-header |addr
+#define dataStorage( bsize, addr, header ) (bsize - ( (char *)addr - (char *)header ))
+
+
+static inline void checkAlign( size_t alignment ) {
+        if ( alignment < libAlign() || ! is_pow2( alignment ) ) {
+                abort( "Alignment %zu for memory allocation is less than %d and/or not a power of 2.", alignment, libAlign() );
+        } // if
+} // checkAlign
 
 
@@ -391 +423 @@
 static inline void fakeHeader( HeapManager.Storage.Header *& header, size_t & alignment ) {
         if ( unlikely( (header->kind.fake.alignment & 1) == 1 ) ) { // fake header ?
-                size_t offset = header->kind.fake.offset;
                 alignment = header->kind.fake.alignment & -2;   // remove flag from value
                 #ifdef __CFA_DEBUG__
                 checkAlign( alignment );                                                // check alignment
                 #endif // __CFA_DEBUG__
-                header = (HeapManager.Storage.Header *)((char *)header - offset);
+                header = realHeader( header );                                  // backup from fake to real header
+        } else {
+                alignment = libAlign();                                                 // => no fake header
         } // if
 } // fakeHeader
 
 
-// <-------+----------------------------------------------------> bsize (bucket size)
-// |header |addr
-//==================================================================================
-//                                | alignment
-// <-----------------<------------+-----------------------------> bsize (bucket size)
-//                   |fake-header | addr
-#define headerAddr( addr ) ((HeapManager.Storage.Header *)( (char *)addr - sizeof(HeapManager.Storage) ))
-
-// <-------<<--------------------- dsize ---------------------->> bsize (bucket size)
-// |header |addr
-//==================================================================================
-//                                | alignment
-// <------------------------------<<---------- dsize --------->>> bsize (bucket size)
-//                   |fake-header |addr
-#define dataStorage( bsize, addr, header ) (bsize - ( (char *)addr - (char *)header ))
-
-
-static inline bool headers( const char name[] __attribute__(( unused )), void * addr, HeapManager.Storage.Header *& header, HeapManager.FreeHeader *& freeElem, size_t & size, size_t & alignment ) with ( heapManager ) {
+static inline bool headers( const char name[] __attribute__(( unused )), void * addr, HeapManager.Storage.Header *& header, HeapManager.FreeHeader *& freeElem,
+                                                        size_t & size, size_t & alignment ) with( heapManager ) {
         header = headerAddr( addr );
 
-        if ( unlikely( heapEnd < addr ) ) {                                     // mmapped ?
+  if ( unlikely( addr < heapBegin || heapEnd < addr ) ) { // mmapped ?
                 fakeHeader( header, alignment );
                 size = header->kind.real.blockSize & -3;                // mmap size
@@ -428 +445 @@
 
         #ifdef __CFA_DEBUG__
-        checkHeader( addr < heapBegin || header < (HeapManager.Storage.Header *)heapBegin, name, addr ); // bad low address ?
+        checkHeader( header < (HeapManager.Storage.Header *)heapBegin, name, addr ); // bad low address ?
         #endif // __CFA_DEBUG__
 
@@ -449 +466 @@
 } // headers
 
-
-static inline void * extend( size_t size ) with ( heapManager ) {
+#ifdef __CFA_DEBUG__
+#if __SIZEOF_POINTER__ == 4
+#define MASK 0xdeadbeef
+#else
+#define MASK 0xdeadbeefdeadbeef
+#endif
+#define STRIDE size_t
+
+static void * Memset( void * addr, STRIDE size ) {              // debug only
+        if ( size % sizeof(STRIDE) != 0 ) abort( "Memset() : internal error, size %zd not multiple of %zd.", size, sizeof(STRIDE) );
+        if ( (STRIDE)addr % sizeof(STRIDE) != 0 ) abort( "Memset() : internal error, addr %p not multiple of %zd.", addr, sizeof(STRIDE) );
+
+        STRIDE * end = (STRIDE *)addr + size / sizeof(STRIDE);
+        for ( STRIDE * p = (STRIDE *)addr; p < end; p += 1 ) *p = MASK;
+        return addr;
+} // Memset
+#endif // __CFA_DEBUG__
+
+
+#define NO_MEMORY_MSG "insufficient heap memory available for allocating %zd new bytes."
+
+static inline void * extend( size_t size ) with( heapManager ) {
         lock( extlock __cfaabi_dbg_ctx2 );
         ptrdiff_t rem = heapRemaining - size;
@@ -456 +493 @@
                 // If the size requested is bigger than the current remaining storage, increase the size of the heap.
 
-                size_t increase = libCeiling( size > heapExpand ? size : heapExpand, libAlign() );
-                if ( sbrk( increase ) == (void *)-1 ) {
+                size_t increase = ceiling2( size > heapExpand ? size : heapExpand, __page_size );
+                // Do not call abort or strerror( errno ) as they may call malloc.
+                if ( sbrk( increase ) == (void *)-1 ) {                 // failed, no memory ?
                         unlock( extlock );
-                        errno = ENOMEM;
-                        return 0p;
+                        __cfaabi_bits_print_nolock( STDERR_FILENO, NO_MEMORY_MSG, size );
+                        _exit( EXIT_FAILURE );
+                } // if
+                if ( mprotect( (char *)heapEnd + heapRemaining, increase, __map_prot ) ) {
+                        unlock( extlock );
+                        __cfaabi_bits_print_nolock( STDERR_FILENO, "extend() : internal error, mprotect failure, heapEnd:%p size:%zd, errno:%d.\n", heapEnd, increase, errno );
+                        _exit( EXIT_FAILURE );
                 } // if
                 #ifdef __STATISTICS__
@@ -468 +511 @@
                 #ifdef __CFA_DEBUG__
                 // Set new memory to garbage so subsequent uninitialized usages might fail.
-                memset( (char *)heapEnd + heapRemaining, '\377', increase );
+                memset( (char *)heapEnd + heapRemaining, '\xde', increase );
+                //Memset( (char *)heapEnd + heapRemaining, increase );
                 #endif // __CFA_DEBUG__
                 rem = heapRemaining + increase - size;
@@ -481 +525 @@
 
 
-static inline void * doMalloc( size_t size ) with ( heapManager ) {
+static inline void * doMalloc( size_t size ) with( heapManager ) {
         HeapManager.Storage * block;                                            // pointer to new block of storage
 
@@ -487 +531 @@
         // along with the block and is a multiple of the alignment size.
 
-  if ( unlikely( size > ~0ul - sizeof(HeapManager.Storage) ) ) return 0p;
+  if ( unlikely( size > ULONG_MAX - sizeof(HeapManager.Storage) ) ) return 0p;
         size_t tsize = size + sizeof(HeapManager.Storage);
         if ( likely( tsize < mmapStart ) ) {                            // small size => sbrk
@@ -497 +541 @@
                         posn = Bsearchl( (unsigned int)tsize, bucketSizes, (size_t)maxBucketsUsed );
                 HeapManager.FreeHeader * freeElem = &freeLists[posn];
-                // #ifdef FASTLOOKUP
-                // if ( tsize < LookupSizes )
-                //      freeElem = &freeLists[lookup[tsize]];
-                // else
-                // #endif // FASTLOOKUP
-                //      freeElem = bsearchl( tsize, freeLists, (size_t)maxBucketsUsed ); // binary search
-                // HeapManager.FreeHeader * freeElem =
-                //      #ifdef FASTLOOKUP
-                //      tsize < LookupSizes ? &freeLists[lookup[tsize]] :
-                //      #endif // FASTLOOKUP
-                //      bsearchl( tsize, freeLists, (size_t)maxBucketsUsed ); // binary search
-                assert( freeElem <= &freeLists[maxBucketsUsed] ); // subscripting error ?
-                assert( tsize <= freeElem->blockSize );                 // search failure ?
+                verify( freeElem <= &freeLists[maxBucketsUsed] ); // subscripting error ?
+                verify( tsize <= freeElem->blockSize );                 // search failure ?
                 tsize = freeElem->blockSize;                                    // total space needed for request
 
                 // Spin until the lock is acquired for this particular size of block.
 
-                #if defined( SPINLOCK )
+                #if BUCKETLOCK == SPINLOCK
                 lock( freeElem->lock __cfaabi_dbg_ctx2 );
                 block = freeElem->freeList;                                             // remove node from stack
                 #else
-                block = freeElem->freeList.pop();
-                #endif // SPINLOCK
+                block = pop( freeElem->freeList );
+                #endif // BUCKETLOCK
                 if ( unlikely( block == 0p ) ) {                                // no free block ?
-                        #if defined( SPINLOCK )
+                        #if BUCKETLOCK == SPINLOCK
                         unlock( freeElem->lock );
-                        #endif // SPINLOCK
+                        #endif // BUCKETLOCK
 
                         // Freelist for that size was empty, so carve it out of the heap if there's enough left, or get some more
@@ -529 +562 @@
 
                         block = (HeapManager.Storage *)extend( tsize ); // mutual exclusion on call
-  if ( unlikely( block == 0p ) ) return 0p;
-                #if defined( SPINLOCK )
+                #if BUCKETLOCK == SPINLOCK
                 } else {
                         freeElem->freeList = block->header.kind.real.next;
                         unlock( freeElem->lock );
-                #endif // SPINLOCK
+                #endif // BUCKETLOCK
                 } // if
 
                 block->header.kind.real.home = freeElem;                // pointer back to free list of apropriate size
         } else {                                                                                        // large size => mmap
-  if ( unlikely( size > ~0ul - pageSize ) ) return 0p;
-                tsize = libCeiling( tsize, pageSize );                  // must be multiple of page size
+  if ( unlikely( size > ULONG_MAX - __page_size ) ) return 0p;
+                tsize = ceiling2( tsize, __page_size );                 // must be multiple of page size
                 #ifdef __STATISTICS__
                 __atomic_add_fetch( &mmap_calls, 1, __ATOMIC_SEQ_CST );
                 __atomic_add_fetch( &mmap_storage, tsize, __ATOMIC_SEQ_CST );
                 #endif // __STATISTICS__
-                block = (HeapManager.Storage *)mmap( 0, tsize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, mmapFd, 0 );
-                if ( block == (HeapManager.Storage *)MAP_FAILED ) {
+
+                block = (HeapManager.Storage *)mmap( 0, tsize, __map_prot, MAP_PRIVATE | MAP_ANONYMOUS, mmapFd, 0 );
+                if ( block == (HeapManager.Storage *)MAP_FAILED ) { // failed ?
+                        if ( errno == ENOMEM ) abort( NO_MEMORY_MSG, tsize ); // no memory
                         // Do not call strerror( errno ) as it may call malloc.
-                        abort( "(HeapManager &)0x%p.doMalloc() : internal error, mmap failure, size:%zu error:%d.", &heapManager, tsize, errno );
-                } // if
+                        abort( "(HeapManager &)0x%p.doMalloc() : internal error, mmap failure, size:%zu errno:%d.", &heapManager, tsize, errno );
+                } //if
                 #ifdef __CFA_DEBUG__
                 // Set new memory to garbage so subsequent uninitialized usages might fail.
-                memset( block, '\377', tsize );
+                memset( block, '\xde', tsize );
+                //Memset( block, tsize );
                 #endif // __CFA_DEBUG__
                 block->header.kind.real.blockSize = tsize;              // storage size for munmap
         } // if
 
+        block->header.kind.real.size = size;                            // store allocation size
         void * addr = &(block->data);                                           // adjust off header to user bytes
+        verify( ((uintptr_t)addr & (libAlign() - 1)) == 0 ); // minimum alignment ?
 
         #ifdef __CFA_DEBUG__
-        assert( ((uintptr_t)addr & (libAlign() - 1)) == 0 ); // minimum alignment ?
-        __atomic_add_fetch( &allocFree, tsize, __ATOMIC_SEQ_CST );
+        __atomic_add_fetch( &allocUnfreed, tsize, __ATOMIC_SEQ_CST );
         if ( traceHeap() ) {
                 enum { BufferSize = 64 };
                 char helpText[BufferSize];
                 int len = snprintf( helpText, BufferSize, "%p = Malloc( %zu ) (allocated %zu)\n", addr, size, tsize );
-                // int len = snprintf( helpText, BufferSize, "Malloc %p %zu\n", addr, size );
                 __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
         } // if
@@ -575 +610 @@
 
 
-static inline void doFree( void * addr ) with ( heapManager ) {
+static inline void doFree( void * addr ) with( heapManager ) {
         #ifdef __CFA_DEBUG__
         if ( unlikely( heapManager.heapBegin == 0p ) ) {
@@ -592 +627 @@
                 #endif // __STATISTICS__
                 if ( munmap( header, size ) == -1 ) {
-                        #ifdef __CFA_DEBUG__
                         abort( "Attempt to deallocate storage %p not allocated or with corrupt header.\n"
                                    "Possible cause is invalid pointer.",
                                    addr );
-                        #endif // __CFA_DEBUG__
                 } // if
         } else {
                 #ifdef __CFA_DEBUG__
                 // Set free memory to garbage so subsequent usages might fail.
-                memset( ((HeapManager.Storage *)header)->data, '\377', freeElem->blockSize - sizeof( HeapManager.Storage ) );
+                memset( ((HeapManager.Storage *)header)->data, '\xde', freeElem->blockSize - sizeof( HeapManager.Storage ) );
+                //Memset( ((HeapManager.Storage *)header)->data, freeElem->blockSize - sizeof( HeapManager.Storage ) );
                 #endif // __CFA_DEBUG__
 
@@ -607 +641 @@
                 free_storage += size;
                 #endif // __STATISTICS__
-                #if defined( SPINLOCK )
+                #if BUCKETLOCK == SPINLOCK
                 lock( freeElem->lock __cfaabi_dbg_ctx2 );               // acquire spin lock
                 header->kind.real.next = freeElem->freeList;    // push on stack
@@ -613 +647 @@
                 unlock( freeElem->lock );                                               // release spin lock
                 #else
-                freeElem->freeList.push( *(HeapManager.Storage *)header );
-                #endif // SPINLOCK
+                push( freeElem->freeList, *(HeapManager.Storage *)header );
+                #endif // BUCKETLOCK
         } // if
 
         #ifdef __CFA_DEBUG__
-        __atomic_add_fetch( &allocFree, -size, __ATOMIC_SEQ_CST );
+        __atomic_add_fetch( &allocUnfreed, -size, __ATOMIC_SEQ_CST );
         if ( traceHeap() ) {
-                enum { BufferSize = 64 };
-                char helpText[BufferSize];
+                char helpText[64];
                 int len = snprintf( helpText, sizeof(helpText), "Free( %p ) size:%zu\n", addr, size );
                 __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
@@ -629 +662 @@
 
 
-size_t prtFree( HeapManager & manager ) with ( manager ) {
+size_t prtFree( HeapManager & manager ) with( manager ) {
         size_t total = 0;
         #ifdef __STATISTICS__
@@ -641 +674 @@
                 #endif // __STATISTICS__
 
-                #if defined( SPINLOCK )
+                #if BUCKETLOCK == SPINLOCK
                 for ( HeapManager.Storage * p = freeLists[i].freeList; p != 0p; p = p->header.kind.real.next ) {
                 #else
-                for ( HeapManager.Storage * p = freeLists[i].freeList.top(); p != 0p; p = p->header.kind.real.next.top ) {
-                #endif // SPINLOCK
+                        for(;;) {
+//              for ( HeapManager.Storage * p = top( freeLists[i].freeList ); p != 0p; p = (p)`next->top ) {
+//              for ( HeapManager.Storage * p = top( freeLists[i].freeList ); p != 0p; /* p = getNext( p )->top */) {
+//                      HeapManager.Storage * temp = p->header.kind.real.next.top; // FIX ME: direct assignent fails, initialization works`
+//                      typeof(p) temp = (( p )`next)->top;                     // FIX ME: direct assignent fails, initialization works`
+//                      p = temp;
+                #endif // BUCKETLOCK
                         total += size;
                         #ifdef __STATISTICS__
@@ -665 +703 @@
 
 
-static void ?{}( HeapManager & manager ) with ( manager ) {
-        pageSize = sysconf( _SC_PAGESIZE );
+static void ?{}( HeapManager & manager ) with( manager ) {
+        __page_size = sysconf( _SC_PAGESIZE );
+        __map_prot = PROT_READ | PROT_WRITE | PROT_EXEC;
 
         for ( unsigned int i = 0; i < NoBucketSizes; i += 1 ) { // initialize the free lists
@@ -680 +719 @@
         #endif // FASTLOOKUP
 
-        if ( setMmapStart( default_mmap_start() ) ) {
+        if ( ! setMmapStart( default_mmap_start() ) ) {
                 abort( "HeapManager : internal error, mmap start initialization failure." );
         } // if
@@ -686 +725 @@
 
         char * end = (char *)sbrk( 0 );
-        sbrk( (char *)libCeiling( (long unsigned int)end, libAlign() ) - end ); // move start of heap to multiple of alignment
-        heapBegin = heapEnd = sbrk( 0 );                                        // get new start point
+        heapBegin = heapEnd = sbrk( (char *)ceiling2( (long unsigned int)end, __page_size ) - end ); // move start of heap to multiple of alignment
 } // HeapManager
 
@@ -695 +733 @@
         if ( traceHeapTerm() ) {
                 printStats();
-                // if ( prtfree() ) prtFree( heapManager, true );
+                // prtUnfreed() called in heapAppStop()
         } // if
         #endif // __STATISTICS__
@@ -704 +742 @@
 void memory_startup( void ) {
         #ifdef __CFA_DEBUG__
-        if ( unlikely( heapBoot ) ) {                                           // check for recursion during system boot
-                // DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT.
+        if ( heapBoot ) {                                                                       // check for recursion during system boot
                 abort( "boot() : internal error, recursively invoked during system boot." );
         } // if
@@ -711 +748 @@
         #endif // __CFA_DEBUG__
 
-        //assert( heapManager.heapBegin != 0 );
+        //verify( heapManager.heapBegin != 0 );
         //heapManager{};
-        if ( heapManager.heapBegin == 0p ) heapManager{};
+        if ( heapManager.heapBegin == 0p ) heapManager{};       // sanity check
 } // memory_startup
 
@@ -723 +760 @@
 
 static inline void * mallocNoStats( size_t size ) {             // necessary for malloc statistics
-        //assert( heapManager.heapBegin != 0 );
-        if ( unlikely( heapManager.heapBegin == 0p ) ) heapManager{}; // called before memory_startup ?
-        void * addr = doMalloc( size );
-        if ( unlikely( addr == 0p ) ) errno = ENOMEM;           // POSIX
-        return addr;
+        verify( heapManager.heapBegin != 0p );                          // called before memory_startup ?
+  if ( unlikely( size ) == 0 ) return 0p;                               // 0 BYTE ALLOCATION RETURNS NULL POINTER
+
+#if __SIZEOF_POINTER__ == 8
+        verify( size < ((typeof(size_t))1 << 48) );
+#endif // __SIZEOF_POINTER__ == 8
+        return doMalloc( size );
 } // mallocNoStats
 
 
-static inline void * callocNoStats( size_t noOfElems, size_t elemSize ) {
-        size_t size = noOfElems * elemSize;
+static inline void * callocNoStats( size_t dim, size_t elemSize ) {
+        size_t size = dim * elemSize;
+  if ( unlikely( size ) == 0 ) return 0p;                               // 0 BYTE ALLOCATION RETURNS NULL POINTER
         char * addr = (char *)mallocNoStats( size );
-  if ( unlikely( addr == 0p ) ) return 0p;
 
         HeapManager.Storage.Header * header;
         HeapManager.FreeHeader * freeElem;
         size_t bsize, alignment;
-        bool mapped __attribute__(( unused )) = headers( "calloc", addr, header, freeElem, bsize, alignment );
         #ifndef __CFA_DEBUG__
+        bool mapped =
+        #endif // __CFA_DEBUG__
+                headers( "calloc", addr, header, freeElem, bsize, alignment );
+        #ifndef __CFA_DEBUG__
+
         // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
         if ( ! mapped )
         #endif // __CFA_DEBUG__
-                // Zero entire data space even when > than size => realloc without a new allocation and zero fill works.
-                // <-------00000000000000000000000000000000000000000000000000000> bsize (bucket size)
+                // <-------0000000000000000000000000000UUUUUUUUUUUUUUUUUUUUUUUUU> bsize (bucket size) U => undefined
                 // `-header`-addr                      `-size
-                memset( addr, '\0', bsize - sizeof(HeapManager.Storage) ); // set to zeros
+                memset( addr, '\0', size );                                             // set to zeros
 
         header->kind.real.blockSize |= 2;                                       // mark as zero filled
@@ -754 +796 @@
 
 
-static inline void * memalignNoStats( size_t alignment, size_t size ) { // necessary for malloc statistics
+static inline void * memalignNoStats( size_t alignment, size_t size ) {
+  if ( unlikely( size ) == 0 ) return 0p;                               // 0 BYTE ALLOCATION RETURNS NULL POINTER
+
         #ifdef __CFA_DEBUG__
         checkAlign( alignment );                                                        // check alignment
@@ -772 +816 @@
         // add sizeof(Storage) for fake header
         char * addr = (char *)mallocNoStats( size + alignment - libAlign() + sizeof(HeapManager.Storage) );
-  if ( unlikely( addr == 0p ) ) return addr;
 
         // address in the block of the "next" alignment address
-        char * user = (char *)libCeiling( (uintptr_t)(addr + sizeof(HeapManager.Storage)), alignment );
+        char * user = (char *)ceiling2( (uintptr_t)(addr + sizeof(HeapManager.Storage)), alignment );
 
         // address of header from malloc
         HeapManager.Storage.Header * realHeader = headerAddr( addr );
+        realHeader->kind.real.size = size;                                      // correct size to eliminate above alignment offset
         // address of fake header * before* the alignment location
         HeapManager.Storage.Header * fakeHeader = headerAddr( user );
@@ -790 +834 @@
 
 
-static inline void * cmemalignNoStats( size_t alignment, size_t noOfElems, size_t elemSize ) {
-        size_t size = noOfElems * elemSize;
+static inline void * cmemalignNoStats( size_t alignment, size_t dim, size_t elemSize ) {
+        size_t size = dim * elemSize;
+  if ( unlikely( size ) == 0 ) return 0p;                               // 0 BYTE ALLOCATION RETURNS NULL POINTER
         char * addr = (char *)memalignNoStats( alignment, size );
-  if ( unlikely( addr == 0p ) ) return 0p;
+
         HeapManager.Storage.Header * header;
         HeapManager.FreeHeader * freeElem;
         size_t bsize;
-        bool mapped __attribute__(( unused )) = headers( "cmemalign", addr, header, freeElem, bsize, alignment );
         #ifndef __CFA_DEBUG__
+        bool mapped =
+        #endif // __CFA_DEBUG__
+                headers( "cmemalign", addr, header, freeElem, bsize, alignment );
+
         // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
+        #ifndef __CFA_DEBUG__
         if ( ! mapped )
         #endif // __CFA_DEBUG__
-                memset( addr, '\0', dataStorage( bsize, addr, header ) ); // set to zeros
-        header->kind.real.blockSize |= 2;                               // mark as zero filled
-
+                // <-------0000000000000000000000000000UUUUUUUUUUUUUUUUUUUUUUUUU> bsize (bucket size) U => undefined
+                // `-header`-addr                      `-size
+                memset( addr, '\0', size );                                             // set to zeros
+
+        header->kind.real.blockSize |= 2;                                       // mark as zero filled
         return addr;
 } // cmemalignNoStats
 
 
-// supported mallopt options
-#ifndef M_MMAP_THRESHOLD
-#define M_MMAP_THRESHOLD (-1)
-#endif // M_TOP_PAD
-#ifndef M_TOP_PAD
-#define M_TOP_PAD (-2)
-#endif // M_TOP_PAD
-
-
 extern "C" {
-        // The malloc() function allocates size bytes and returns a pointer to the allocated memory. The memory is not
-        // initialized. If size is 0, then malloc() returns either 0p, or a unique pointer value that can later be
-        // successfully passed to free().
+        // Allocates size bytes and returns a pointer to the allocated memory.  The contents are undefined. If size is 0,
+        // then malloc() returns a unique pointer value that can later be successfully passed to free().
         void * malloc( size_t size ) {
                 #ifdef __STATISTICS__
@@ -831 +872 @@
         } // malloc
 
-        // The calloc() function allocates memory for an array of nmemb elements of size bytes each and returns a pointer to
-        // the allocated memory. The memory is set to zero. If nmemb or size is 0, then calloc() returns either 0p, or a
-        // unique pointer value that can later be successfully passed to free().
-        void * calloc( size_t noOfElems, size_t elemSize ) {
+
+        // Same as malloc() except size bytes is an array of dim elements each of elemSize bytes.
+        void * aalloc( size_t dim, size_t elemSize ) {
+                size_t size = dim * elemSize;
+                #ifdef __STATISTICS__
+                __atomic_add_fetch( &aalloc_calls, 1, __ATOMIC_SEQ_CST );
+                __atomic_add_fetch( &aalloc_storage, size, __ATOMIC_SEQ_CST );
+                #endif // __STATISTICS__
+
+                return mallocNoStats( size );
+        } // aalloc
+
+
+        // Same as aalloc() with memory set to zero.
+        void * calloc( size_t dim, size_t elemSize ) {
                 #ifdef __STATISTICS__
                 __atomic_add_fetch( &calloc_calls, 1, __ATOMIC_SEQ_CST );
-                __atomic_add_fetch( &calloc_storage, noOfElems * elemSize, __ATOMIC_SEQ_CST );
-                #endif // __STATISTICS__
-
-                return callocNoStats( noOfElems, elemSize );
+                __atomic_add_fetch( &calloc_storage, dim * elemSize, __ATOMIC_SEQ_CST );
+                #endif // __STATISTICS__
+
+                return callocNoStats( dim, elemSize );
         } // calloc
 
-        // The realloc() function changes the size of the memory block pointed to by ptr to size bytes. The contents will be
-        // unchanged in the range from the start of the region up to the minimum of the old and new sizes. If the new size
-        // is larger than the old size, the added memory will not be initialized.  If ptr is 0p, then the call is
-        // equivalent to malloc(size), for all values of size; if size is equal to zero, and ptr is not 0p, then the call
-        // is equivalent to free(ptr). Unless ptr is 0p, it must have been returned by an earlier call to malloc(),
-        // calloc() or realloc(). If the area pointed to was moved, a free(ptr) is done.
-        void * realloc( void * oaddr, size_t size ) {
-                #ifdef __STATISTICS__
-                __atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
+
+        // Change the size of the memory block pointed to by oaddr to size bytes. The contents are undefined.  If oaddr is
+        // 0p, then the call is equivalent to malloc(size), for all values of size; if size is equal to zero, and oaddr is
+        // not 0p, then the call is equivalent to free(oaddr). Unless oaddr is 0p, it must have been returned by an earlier
+        // call to malloc(), alloc(), calloc() or realloc(). If the area pointed to was moved, a free(oaddr) is done.
+        void * resize( void * oaddr, size_t size ) {
+                #ifdef __STATISTICS__
+                __atomic_add_fetch( &resize_calls, 1, __ATOMIC_SEQ_CST );
                 #endif // __STATISTICS__
 
                 // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
-          if ( unlikely( size == 0 ) ) { free( oaddr ); return mallocNoStats( size ); } // special cases
-          if ( unlikely( oaddr == 0p ) ) return mallocNoStats( size );
+          if ( unlikely( size == 0 ) ) { free( oaddr ); return 0p; } // special cases
+          if ( unlikely( oaddr == 0p ) ) {
+                        #ifdef __STATISTICS__
+                        __atomic_add_fetch( &resize_storage, size, __ATOMIC_SEQ_CST );
+                        #endif // __STATISTICS__
+                        return mallocNoStats( size );
+                } // if
 
                 HeapManager.Storage.Header * header;
                 HeapManager.FreeHeader * freeElem;
-                size_t bsize, oalign = 0;
+                size_t bsize, oalign;
+                headers( "resize", oaddr, header, freeElem, bsize, oalign );
+                size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket
+
+                // same size, DO NOT preserve STICKY PROPERTIES.
+                if ( oalign == libAlign() && size <= odsize && odsize <= size * 2 ) { // allow 50% wasted storage for smaller size
+                        header->kind.real.blockSize &= -2;                      // no alignment and turn off 0 fill
+                        header->kind.real.size = size;                          // reset allocation size
+                        return oaddr;
+                } // if
+
+                #ifdef __STATISTICS__
+                __atomic_add_fetch( &resize_storage, size, __ATOMIC_SEQ_CST );
+                #endif // __STATISTICS__
+
+                // change size, DO NOT preserve STICKY PROPERTIES.
+                free( oaddr );
+                return mallocNoStats( size );                                   // create new area
+        } // resize
+
+
+        // Same as resize() but the contents are unchanged in the range from the start of the region up to the minimum of
+        // the old and new sizes.
+        void * realloc( void * oaddr, size_t size ) {
+                #ifdef __STATISTICS__
+                __atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
+                #endif // __STATISTICS__
+
+                // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
+          if ( unlikely( size == 0 ) ) { free( oaddr ); return 0p; } // special cases
+          if ( unlikely( oaddr == 0p ) ) {
+                        #ifdef __STATISTICS__
+                        __atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST );
+                        #endif // __STATISTICS__
+                        return mallocNoStats( size );
+                } // if
+
+                HeapManager.Storage.Header * header;
+                HeapManager.FreeHeader * freeElem;
+                size_t bsize, oalign;
                 headers( "realloc", oaddr, header, freeElem, bsize, oalign );
 
                 size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket
-          if ( size <= odsize && odsize <= size * 2 ) { // allow up to 50% wasted storage in smaller size
-                        // Do not know size of original allocation => cannot do 0 fill for any additional space because do not know
-                        // where to start filling, i.e., do not overwrite existing values in space.
-                        //
-                        // This case does not result in a new profiler entry because the previous one still exists and it must match with
-                        // the free for this memory.  Hence, this realloc does not appear in the profiler output.
+                size_t osize = header->kind.real.size;                  // old allocation size
+                bool ozfill = (header->kind.real.blockSize & 2); // old allocation zero filled
+          if ( unlikely( size <= odsize ) && odsize <= size * 2 ) { // allow up to 50% wasted storage
+                        header->kind.real.size = size;                          // reset allocation size
+                        if ( unlikely( ozfill ) && size > osize ) {     // previous request zero fill and larger ?
+                                memset( (char *)oaddr + osize, '\0', size - osize ); // initialize added storage
+                        } // if
                         return oaddr;
                 } // if
 
                 #ifdef __STATISTICS__
-                __atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST );
+                __atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST );
                 #endif // __STATISTICS__
 
@@ -880 +976 @@
 
                 void * naddr;
-                if ( unlikely( oalign != 0 ) ) {                                // previous request memalign?
-                        if ( unlikely( header->kind.real.blockSize & 2 ) ) { // previous request zero fill
-                                naddr = cmemalignNoStats( oalign, 1, size ); // create new aligned area
-                        } else {
-                                naddr = memalignNoStats( oalign, size ); // create new aligned area
+                if ( likely( oalign == libAlign() ) ) {                 // previous request not aligned ?
+                        naddr = mallocNoStats( size );                          // create new area
+                } else {
+                        naddr = memalignNoStats( oalign, size );        // create new aligned area
+                } // if
+
+                headers( "realloc", naddr, header, freeElem, bsize, oalign );
+                memcpy( naddr, oaddr, min( osize, size ) );             // copy bytes
+                free( oaddr );
+
+                if ( unlikely( ozfill ) ) {                                             // previous request zero fill ?
+                        header->kind.real.blockSize |= 2;                       // mark new request as zero filled
+                        if ( size > osize ) {                                           // previous request larger ?
+                                memset( (char *)naddr + osize, '\0', size - osize ); // initialize added storage
                         } // if
-                } else {
-                        if ( unlikely( header->kind.real.blockSize & 2 ) ) { // previous request zero fill
-                                naddr = callocNoStats( 1, size );               // create new area
-                        } else {
-                                naddr = mallocNoStats( size );                  // create new area
-                        } // if
-                } // if
-          if ( unlikely( naddr == 0p ) ) return 0p;
-
-                headers( "realloc", naddr, header, freeElem, bsize, oalign );
-                size_t ndsize = dataStorage( bsize, naddr, header ); // data storage avilable in bucket
-                // To preserve prior fill, the entire bucket must be copied versus the size.
-                memcpy( naddr, oaddr, MIN( odsize, ndsize ) );  // copy bytes
-                free( oaddr );
+                } // if
                 return naddr;
         } // realloc
 
-        // The obsolete function memalign() allocates size bytes and returns a pointer to the allocated memory. The memory
-        // address will be a multiple of alignment, which must be a power of two.
+
+        // Same as malloc() except the memory address is a multiple of alignment, which must be a power of two. (obsolete)
         void * memalign( size_t alignment, size_t size ) {
                 #ifdef __STATISTICS__
@@ -915 +1007 @@
 
 
-        // The cmemalign() function is the same as calloc() with memory alignment.
-        void * cmemalign( size_t alignment, size_t noOfElems, size_t elemSize ) {
+        // Same as aalloc() with memory alignment.
+        void * amemalign( size_t alignment, size_t dim, size_t elemSize ) {
+                size_t size = dim * elemSize;
                 #ifdef __STATISTICS__
                 __atomic_add_fetch( &cmemalign_calls, 1, __ATOMIC_SEQ_CST );
-                __atomic_add_fetch( &cmemalign_storage, noOfElems * elemSize, __ATOMIC_SEQ_CST );
-                #endif // __STATISTICS__
-
-                return cmemalignNoStats( alignment, noOfElems, elemSize );
+                __atomic_add_fetch( &cmemalign_storage, size, __ATOMIC_SEQ_CST );
+                #endif // __STATISTICS__
+
+                return memalignNoStats( alignment, size );
+        } // amemalign
+
+
+        // Same as calloc() with memory alignment.
+        void * cmemalign( size_t alignment, size_t dim, size_t elemSize ) {
+                #ifdef __STATISTICS__
+                __atomic_add_fetch( &cmemalign_calls, 1, __ATOMIC_SEQ_CST );
+                __atomic_add_fetch( &cmemalign_storage, dim * elemSize, __ATOMIC_SEQ_CST );
+                #endif // __STATISTICS__
+
+                return cmemalignNoStats( alignment, dim, elemSize );
         } // cmemalign
 
-        // The function aligned_alloc() is the same as memalign(), except for the added restriction that size should be a
-        // multiple of alignment.
+
+        // Same as memalign(), but ISO/IEC 2011 C11 Section 7.22.2 states: the value of size shall be an integral multiple
+    // of alignment. This requirement is universally ignored.
         void * aligned_alloc( size_t alignment, size_t size ) {
                 return memalign( alignment, size );
@@ -932 +1037 @@
 
 
-        // The function posix_memalign() allocates size bytes and places the address of the allocated memory in *memptr. The
-        // address of the allocated memory will be a multiple of alignment, which must be a power of two and a multiple of
-        // sizeof(void *). If size is 0, then posix_memalign() returns either 0p, or a unique pointer value that can later
-        // be successfully passed to free(3).
+        // Allocates size bytes and places the address of the allocated memory in *memptr. The address of the allocated
+        // memory shall be a multiple of alignment, which must be a power of two and a multiple of sizeof(void *). If size
+        // is 0, then posix_memalign() returns either 0p, or a unique pointer value that can later be successfully passed to
+        // free(3).
         int posix_memalign( void ** memptr, size_t alignment, size_t size ) {
-          if ( alignment < sizeof(void *) || ! libPow2( alignment ) ) return EINVAL; // check alignment
+          if ( alignment < libAlign() || ! is_pow2( alignment ) ) return EINVAL; // check alignment
                 * memptr = memalign( alignment, size );
-          if ( unlikely( * memptr == 0p ) ) return ENOMEM;
                 return 0;
         } // posix_memalign
 
-        // The obsolete function valloc() allocates size bytes and returns a pointer to the allocated memory. The memory
-        // address will be a multiple of the page size.  It is equivalent to memalign(sysconf(_SC_PAGESIZE),size).
+
+        // Allocates size bytes and returns a pointer to the allocated memory. The memory address shall be a multiple of the
+        // page size.  It is equivalent to memalign(sysconf(_SC_PAGESIZE),size).
         void * valloc( size_t size ) {
-                return memalign( pageSize, size );
+                return memalign( __page_size, size );
         } // valloc
 
 
-        // The free() function frees the memory space pointed to by ptr, which must have been returned by a previous call to
-        // malloc(), calloc() or realloc().  Otherwise, or if free(ptr) has already been called before, undefined behavior
-        // occurs. If ptr is 0p, no operation is performed.
+        // Same as valloc but rounds size to multiple of page size.
+        void * pvalloc( size_t size ) {
+                return memalign( __page_size, ceiling2( size, __page_size ) );
+        } // pvalloc
+
+
+        // Frees the memory space pointed to by ptr, which must have been returned by a previous call to malloc(), calloc()
+        // or realloc().  Otherwise, or if free(ptr) has already been called before, undefined behaviour occurs. If ptr is
+        // 0p, no operation is performed.
         void free( void * addr ) {
                 #ifdef __STATISTICS__
@@ -973 +1084 @@
 
 
-        // The malloc_alignment() function returns the alignment of the allocation.
+        // Returns the alignment of an allocation.
         size_t malloc_alignment( void * addr ) {
           if ( unlikely( addr == 0p ) ) return libAlign();      // minimum alignment
@@ -980 +1091 @@
                         return header->kind.fake.alignment & -2;        // remove flag from value
                 } else {
-                        return libAlign ();                                                     // minimum alignment
+                        return libAlign();                                                      // minimum alignment
                 } // if
         } // malloc_alignment
 
 
-        // The malloc_zero_fill() function returns true if the allocation is zero filled, i.e., initially allocated by calloc().
+        // Set the alignment for an the allocation and return previous alignment or 0 if no alignment.
+        size_t $malloc_alignment_set( void * addr, size_t alignment ) {
+          if ( unlikely( addr == 0p ) ) return libAlign();      // minimum alignment
+                size_t ret;
+                HeapManager.Storage.Header * header = headerAddr( addr );
+                if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
+                        ret = header->kind.fake.alignment & -2;         // remove flag from old value
+                        header->kind.fake.alignment = alignment | 1; // add flag to new value
+                } else {
+                        ret = 0;                                                                        // => no alignment to change
+                } // if
+                return ret;
+        } // $malloc_alignment_set
+
+
+        // Returns true if the allocation is zero filled, e.g., allocated by calloc().
         bool malloc_zero_fill( void * addr ) {
           if ( unlikely( addr == 0p ) ) return false;           // null allocation is not zero fill
                 HeapManager.Storage.Header * header = headerAddr( addr );
                 if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
-                        header = (HeapManager.Storage.Header *)((char *)header - header->kind.fake.offset);
-                } // if
-                return (header->kind.real.blockSize & 2) != 0;  // zero filled (calloc/cmemalign) ?
+                        header = realHeader( header );                          // backup from fake to real header
+                } // if
+                return (header->kind.real.blockSize & 2) != 0;  // zero filled ?
         } // malloc_zero_fill
 
-
-        // The malloc_usable_size() function returns the number of usable bytes in the block pointed to by ptr, a pointer to
-        // a block of memory allocated by malloc(3) or a related function.
+        // Set allocation is zero filled and return previous zero filled.
+        bool $malloc_zero_fill_set( void * addr ) {
+          if ( unlikely( addr == 0p ) ) return false;           // null allocation is not zero fill
+                HeapManager.Storage.Header * header = headerAddr( addr );
+                if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
+                        header = realHeader( header );                          // backup from fake to real header
+                } // if
+                bool ret = (header->kind.real.blockSize & 2) != 0; // zero filled ?
+                header->kind.real.blockSize |= 2;                               // mark as zero filled
+                return ret;
+        } // $malloc_zero_fill_set
+
+
+        // Returns original total allocation size (not bucket size) => array size is dimension * sizeif(T).
+        size_t malloc_size( void * addr ) {
+          if ( unlikely( addr == 0p ) ) return 0;                       // null allocation has zero size
+                HeapManager.Storage.Header * header = headerAddr( addr );
+                if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
+                        header = realHeader( header );                          // backup from fake to real header
+                } // if
+                return header->kind.real.size;
+        } // malloc_size
+
+        // Set allocation size and return previous size.
+        size_t $malloc_size_set( void * addr, size_t size ) {
+          if ( unlikely( addr == 0p ) ) return 0;                       // null allocation has 0 size
+                HeapManager.Storage.Header * header = headerAddr( addr );
+                if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
+                        header = realHeader( header );                          // backup from fake to real header
+                } // if
+                size_t ret = header->kind.real.size;
+                header->kind.real.size = size;
+                return ret;
+        } // $malloc_size_set
+
+
+        // Returns the number of usable bytes in the block pointed to by ptr, a pointer to a block of memory allocated by
+        // malloc or a related function.
         size_t malloc_usable_size( void * addr ) {
           if ( unlikely( addr == 0p ) ) return 0;                       // null allocation has 0 size
@@ -1005 +1166 @@
 
                 headers( "malloc_usable_size", addr, header, freeElem, bsize, alignment );
-                return dataStorage( bsize, addr, header );      // data storage in bucket
+                return dataStorage( bsize, addr, header );              // data storage in bucket
         } // malloc_usable_size
 
 
-        // The malloc_stats() function prints (on default standard error) statistics about memory allocated by malloc(3) and
-        // related functions.
+        // Prints (on default standard error) statistics about memory allocated by malloc and related functions.
         void malloc_stats( void ) {
                 #ifdef __STATISTICS__
@@ -1018 +1178 @@
         } // malloc_stats
 
-        // The malloc_stats_fd() function changes the file descripter where malloc_stats() writes the statistics.
+
+        // Changes the file descripter where malloc_stats() writes statistics.
         int malloc_stats_fd( int fd __attribute__(( unused )) ) {
                 #ifdef __STATISTICS__
-                int temp = statfd;
-                statfd = fd;
+                int temp = stat_fd;
+                stat_fd = fd;
                 return temp;
                 #else
@@ -1030 +1191 @@
 
 
-        // The mallopt() function adjusts parameters that control the behavior of the memory-allocation functions (see
-        // malloc(3)). The param argument specifies the parameter to be modified, and value specifies the new value for that
-        // parameter.
+        // Adjusts parameters that control the behaviour of the memory-allocation functions (see malloc). The param argument
+        // specifies the parameter to be modified, and value specifies the new value for that parameter.
         int mallopt( int option, int value ) {
                 choose( option ) {
                   case M_TOP_PAD:
-                        if ( setHeapExpand( value ) ) return 1;
+                        heapExpand = ceiling2( value, __page_size ); return 1;
                   case M_MMAP_THRESHOLD:
                         if ( setMmapStart( value ) ) return 1;
+                        break;
                 } // switch
                 return 0;                                                                               // error, unsupported
         } // mallopt
 
-        // The malloc_trim() function attempts to release free memory at the top of the heap (by calling sbrk(2) with a
-        // suitable argument).
+
+        // Attempt to release free memory at the top of the heap (by calling sbrk with a suitable argument).
         int malloc_trim( size_t ) {
                 return 0;                                                                               // => impossible to release memory
@@ -1050 +1211 @@
 
 
-        // The malloc_info() function exports an XML string that describes the current state of the memory-allocation
-        // implementation in the caller.  The string is printed on the file stream stream.  The exported string includes
-        // information about all arenas (see malloc(3)).
+        // Exports an XML string that describes the current state of the memory-allocation implementation in the caller.
+        // The string is printed on the file stream stream.  The exported string includes information about all arenas (see
+        // malloc).
         int malloc_info( int options, FILE * stream ) {
-                if ( options != 0 ) { errno = EINVAL; return -1; }
+          if ( options != 0 ) { errno = EINVAL; return -1; }
+                #ifdef __STATISTICS__
                 return printStatsXML( stream );
+                #else
+                return 0;                                                                               // unsupported
+                #endif // __STATISTICS__
         } // malloc_info
 
 
-        // The malloc_get_state() function records the current state of all malloc(3) internal bookkeeping variables (but
-        // not the actual contents of the heap or the state of malloc_hook(3) functions pointers).  The state is recorded in
-        // a system-dependent opaque data structure dynamically allocated via malloc(3), and a pointer to that data
-        // structure is returned as the function result.  (It is the caller's responsibility to free(3) this memory.)
+        // Records the current state of all malloc internal bookkeeping variables (but not the actual contents of the heap
+        // or the state of malloc_hook functions pointers).  The state is recorded in a system-dependent opaque data
+        // structure dynamically allocated via malloc, and a pointer to that data structure is returned as the function
+        // result.  (The caller must free this memory.)
         void * malloc_get_state( void ) {
                 return 0p;                                                                              // unsupported
@@ -1068 +1233 @@
 
 
-        // The malloc_set_state() function restores the state of all malloc(3) internal bookkeeping variables to the values
-        // recorded in the opaque data structure pointed to by state.
-        int malloc_set_state( void * ptr ) {
+        // Restores the state of all malloc internal bookkeeping variables to the values recorded in the opaque data
+        // structure pointed to by state.
+        int malloc_set_state( void * ) {
                 return 0;                                                                               // unsupported
         } // malloc_set_state
@@ -1077 +1242 @@
 
 // Must have CFA linkage to overload with C linkage realloc.
-void * realloc( void * oaddr, size_t nalign, size_t size ) {
+void * resize( void * oaddr, size_t nalign, size_t size ) {
         #ifdef __STATISTICS__
-        __atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
+        __atomic_add_fetch( &resize_calls, 1, __ATOMIC_SEQ_CST );
         #endif // __STATISTICS__
 
-        // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
-  if ( unlikely( size == 0 ) ) { free( oaddr ); return mallocNoStats( size ); } // special cases
-  if ( unlikely( oaddr == 0p ) ) return mallocNoStats( size );
-
-        if ( unlikely( nalign == 0 ) ) nalign = libAlign();     // reset alignment to minimum
+        if ( unlikely( nalign < libAlign() ) ) nalign = libAlign(); // reset alignment to minimum
         #ifdef __CFA_DEBUG__
         else
@@ -1092 +1253 @@
         #endif // __CFA_DEBUG__
 
-        HeapManager.Storage.Header * header;
-        HeapManager.FreeHeader * freeElem;
-        size_t bsize, oalign = 0;
-        headers( "realloc", oaddr, header, freeElem, bsize, oalign );
-        size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket
-
-  if ( oalign != 0 && (uintptr_t)oaddr % nalign == 0 ) { // has alignment and just happens to work out
-                headerAddr( oaddr )->kind.fake.alignment = nalign | 1; // update alignment (could be the same)
-                return realloc( oaddr, size );
+        // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
+  if ( unlikely( size == 0 ) ) { free( oaddr ); return 0p; } // special cases
+  if ( unlikely( oaddr == 0p ) ) {
+                #ifdef __STATISTICS__
+                __atomic_add_fetch( &resize_storage, size, __ATOMIC_SEQ_CST );
+                #endif // __STATISTICS__
+                return memalignNoStats( nalign, size );
+        } // if
+
+        // Attempt to reuse existing alignment.
+        HeapManager.Storage.Header * header = headerAddr( oaddr );
+        bool isFakeHeader = header->kind.fake.alignment & 1; // old fake header ?
+        size_t oalign;
+        if ( isFakeHeader ) {
+                oalign = header->kind.fake.alignment & -2;              // old alignment
+                if ( (uintptr_t)oaddr % nalign == 0                             // lucky match ?
+                         && ( oalign <= nalign                                          // going down
+                                  || (oalign >= nalign && oalign <= 256) ) // little alignment storage wasted ?
+                        ) {
+                        headerAddr( oaddr )->kind.fake.alignment = nalign | 1; // update alignment (could be the same)
+                        HeapManager.FreeHeader * freeElem;
+                        size_t bsize, oalign;
+                        headers( "resize", oaddr, header, freeElem, bsize, oalign );
+                        size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket
+
+                        if ( size <= odsize && odsize <= size * 2 ) { // allow 50% wasted data storage
+                                headerAddr( oaddr )->kind.fake.alignment = nalign | 1; // update alignment (could be the same)
+
+                                header->kind.real.blockSize &= -2;              // turn off 0 fill
+                                header->kind.real.size = size;                  // reset allocation size
+                                return oaddr;
+                        } // if
+                } // if
+        } else if ( ! isFakeHeader                                                      // old real header (aligned on libAlign) ?
+                                && nalign == libAlign() ) {                             // new alignment also on libAlign => no fake header needed
+                return resize( oaddr, size );                                   // duplicate special case checks
         } // if
 
         #ifdef __STATISTICS__
+        __atomic_add_fetch( &resize_storage, size, __ATOMIC_SEQ_CST );
+        #endif // __STATISTICS__
+
+        // change size, DO NOT preserve STICKY PROPERTIES.
+        free( oaddr );
+        return memalignNoStats( nalign, size );                         // create new aligned area
+} // resize
+
+
+void * realloc( void * oaddr, size_t nalign, size_t size ) {
+        if ( unlikely( nalign < libAlign() ) ) nalign = libAlign(); // reset alignment to minimum
+        #ifdef __CFA_DEBUG__
+        else
+                checkAlign( nalign );                                                   // check alignment
+        #endif // __CFA_DEBUG__
+
+        // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
+  if ( unlikely( size == 0 ) ) { free( oaddr ); return 0p; } // special cases
+  if ( unlikely( oaddr == 0p ) ) {
+                #ifdef __STATISTICS__
+                __atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
+                __atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST );
+                #endif // __STATISTICS__
+                return memalignNoStats( nalign, size );
+        } // if
+
+        // Attempt to reuse existing alignment.
+        HeapManager.Storage.Header * header = headerAddr( oaddr );
+        bool isFakeHeader = header->kind.fake.alignment & 1; // old fake header ?
+        size_t oalign;
+        if ( isFakeHeader ) {
+                oalign = header->kind.fake.alignment & -2;              // old alignment
+                if ( (uintptr_t)oaddr % nalign == 0                             // lucky match ?
+                         && ( oalign <= nalign                                          // going down
+                                  || (oalign >= nalign && oalign <= 256) ) // little alignment storage wasted ?
+                        ) {
+                        headerAddr( oaddr )->kind.fake.alignment = nalign | 1; // update alignment (could be the same)
+                        return realloc( oaddr, size );                          // duplicate alignment and special case checks
+                } // if
+        } else if ( ! isFakeHeader                                                      // old real header (aligned on libAlign) ?
+                                && nalign == libAlign() )                               // new alignment also on libAlign => no fake header needed
+                return realloc( oaddr, size );                                  // duplicate alignment and special case checks
+
+        #ifdef __STATISTICS__
+        __atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
         __atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST );
         #endif // __STATISTICS__
 
+        HeapManager.FreeHeader * freeElem;
+        size_t bsize;
+        headers( "realloc", oaddr, header, freeElem, bsize, oalign );
+
         // change size and copy old content to new storage
 
-        void * naddr;
-        if ( unlikely( header->kind.real.blockSize & 2 ) ) { // previous request zero fill
-                naddr = cmemalignNoStats( nalign, 1, size );    // create new aligned area
-        } else {
-                naddr = memalignNoStats( nalign, size );                // create new aligned area
-        } // if
+        size_t osize = header->kind.real.size;                          // old allocation size
+        bool ozfill = (header->kind.real.blockSize & 2);        // old allocation zero filled
+
+        void * naddr = memalignNoStats( nalign, size );         // create new aligned area
 
         headers( "realloc", naddr, header, freeElem, bsize, oalign );
-        size_t ndsize = dataStorage( bsize, naddr, header ); // data storage avilable in bucket
-        // To preserve prior fill, the entire bucket must be copied versus the size.
-        memcpy( naddr, oaddr, MIN( odsize, ndsize ) );          // copy bytes
+        memcpy( naddr, oaddr, min( osize, size ) );                     // copy bytes
         free( oaddr );
+
+        if ( unlikely( ozfill ) ) {                                                     // previous request zero fill ?
+                header->kind.real.blockSize |= 2;                               // mark new request as zero filled
+                if ( size > osize ) {                                                   // previous request larger ?
+                        memset( (char *)naddr + osize, '\0', size - osize ); // initialize added storage
+                } // if
+        } // if
         return naddr;
 } // realloc