The benchmark producing these numbers were run on an Ubuntu 16.10 machine with 8 logical cores (4 physical, HT). The actual numbers are not to be interpreted as absolute performance figures, but rather as relative comparisons between the different allocators. For additional benchmark results, see the BENCHMARKS file.
Configuration of the thread and global caches can be important depending on your use pattern. See CACHE for a case study and some comments/guidelines.
Before calling any other function in the API, you MUST call the initialization function, either rpmalloc_initialize or pmalloc_initialize_config, or you will get undefined behaviour when calling other rpmalloc entry point.
Before terminating your use of the allocator, you SHOULD call rpmalloc_finalize in order to release caches and unmap virtual memory, as well as prepare the allocator for global scope cleanup at process exit or dynamic library unload depending on your use case.
The easiest way to use the library is simply adding rpmalloc.[h|c] to your project and compile them along with your sources. This contains only the rpmalloc specific entry points and does not provide internal hooks to process and/or thread creation at the moment. You are required to call these functions from your own code in order to initialize and finalize the allocator in your process and threads:
rpmalloc_initialize : Call at process start to initialize the allocator
rpmalloc_initialize_config : Optional entry point to call at process start to initialize the allocator with a custom memory mapping backend, memory page size and mapping granularity.
rpmalloc_finalize: Call at process exit to finalize the allocator
rpmalloc_thread_initialize: Call at each thread start to initialize the thread local data for the allocator
rpmalloc_thread_finalize: Call at each thread exit to finalize and release thread cache back to global cache
rpmalloc_config: Get the current runtime configuration of the allocator
Then simply use the rpmalloc/rpfree and the other malloc style replacement functions. Remember all allocations are 16-byte aligned, so no need to call the explicit rpmemalign/rpaligned_alloc/rpposix_memalign functions unless you need greater alignment, they are simply wrappers to make it easier to replace in existing code.
If you wish to override the standard library malloc family of functions and have automatic initialization/finalization of process and threads, define ENABLE_OVERRIDE to non-zero which will include the
malloc.c file in compilation of rpmalloc.c. The list of libc entry points replaced may not be complete, use libc replacement only as a convenience for testing the library on an existing code base, not a final solution.
For explicit first class heaps, see the rpmalloc_heap*_ API under first class heaps section, requiring RPMALLOC_FIRST_CLASS_HEAPS tp be defined to 1.
To compile as a static library run the configure python script which generates a Ninja build script, then build using ninja. The ninja build produces two static libraries, one named
rpmalloc and one named
rpmallocwrap, where the latter includes the libc entry point overrides.
The configure + ninja build also produces two shared object/dynamic libraries. The
rpmallocwrap shared library can be used with LD_PRELOAD/DYLD_INSERT_LIBRARIES to inject in a preexisting binary, replacing any malloc/free family of function calls. This is only implemented for Linux and macOS targets. The list of libc entry points replaced may not be complete, use preloading as a convenience for testing the library on an existing binary, not a final solution. The dynamic library also provides automatic init/fini of process and threads for all platforms.
The latest stable release is available in the master branch. For latest development code, use the develop branch.
Free memory pages are cached both per thread and in a global cache for all threads. The size of the thread caches is determined by an adaptive scheme where each cache is limited by a percentage of the maximum allocation count of the corresponding size class. The size of the global caches is determined by a multiple of the maximum of all thread caches. The factors controlling the cache sizes can be set by editing the individual defines in the
rpmalloc.c source file for fine tuned control.
ENABLE_UNLIMITED_CACHE: By default defined to 0, set to 1 to make all caches infinite, i.e never release spans to global cache unless thread finishes and never unmap memory pages back to the OS. Highest performance but largest memory overhead.
ENABLE_UNLIMITED_GLOBAL_CACHE: By default defined to 0, set to 1 to make global caches infinite, i.e never unmap memory pages back to the OS.
ENABLE_UNLIMITED_THREAD_CACHE: By default defined to 0, set to 1 to make thread caches infinite, i.e never release spans to global cache unless thread finishes.
ENABLE_GLOBAL_CACHE: By default defined to 1, enables the global cache shared between all threads. Set to 0 to disable the global cache and directly unmap pages evicted from the thread cache.
ENABLE_THREAD_CACHE: By default defined to 1, enables the per-thread cache. Set to 0 to disable the thread cache and directly unmap pages no longer in use (also disables the global cache).
ENABLE_ADAPTIVE_THREAD_CACHE: Introduces a simple heuristics in the thread cache size, keeping 25% of the high water mark for each span count class.
Detailed statistics are available if ENABLE_STATISTICS is defined to 1 (default is 0, or disabled), either on compile command line or by setting the value in
rpmalloc.c. This will cause a slight overhead in runtime to collect statistics for each memory operation, and will also add 4 bytes overhead per allocation to track sizes.
Integer safety checks on all calls are enabled if ENABLE_VALIDATE_ARGS is defined to 1 (default is 0, or disabled), either on compile command line or by setting the value in
rpmalloc.c. If enabled, size arguments to the global entry points are verified not to cause integer overflows in calculations.
Asserts are enabled if ENABLE_ASSERTS is defined to 1 (default is 0, or disabled), either on compile command line or by setting the value in
To include malloc.c in compilation and provide overrides of standard library malloc entry points define ENABLE_OVERRIDE to 1. To enable automatic initialization of finalization of process and threads in order to preload the library into executables using standard library malloc, define ENABLE_PRELOAD to 1.
To enable the runtime configurable memory page and span sizes, define RPMALLOC_CONFIGURABLE to 1. By default, memory page size is determined by system APIs and memory span size is set to 64KiB.
To enable support for first class heaps, define RPMALLOC_FIRST_CLASS_HEAPS to 1. By default, the first class heap API is disabled.
The allocator has support for huge/large pages on Windows, Linux and MacOS. To enable it, pass a non-zero value in the config value
enable_huge_pages when initializing the allocator with
rpmalloc_initialize_config. If the system does not support huge pages it will be automatically disabled. You can query the status by looking at
enable_huge_pages in the config returned from a call to
rpmalloc_config after initialization is done.
The allocator is similar in spirit to tcmalloc from the Google Performance Toolkit. It uses separate heaps for each thread and partitions memory blocks according to a preconfigured set of size classes, up to 2MiB. Larger blocks are mapped and unmapped directly. Allocations for different size classes will be served from different set of memory pages, each “span” of pages is dedicated to one size class. Spans of pages can flow between threads when the thread cache overflows and are released to a global cache, or when the thread ends. Unlike tcmalloc, single blocks do not flow between threads, only entire spans of pages.
The allocator is based on a fixed but configurable page alignment (defaults to 64KiB) and 16 byte block alignment, where all runs of memory pages (spans) are mapped to this alignment boundary. On Windows this is automatically guaranteed up to 64KiB by the VirtualAlloc granularity, and on mmap systems it is achieved by oversizing the mapping and aligning the returned virtual memory address to the required boundaries. By aligning to a fixed size the free operation can locate the header of the memory span without having to do a table lookup (as tcmalloc does) by simply masking out the low bits of the address (for 64KiB this would be the low 16 bits).
Memory blocks are divided into three categories. For 64KiB span size/alignment the small blocks are [16, 1024] bytes, medium blocks (1024, 32256] bytes, and large blocks (32256, 2097120] bytes. The three categories are further divided in size classes. If the span size is changed, the small block classes remain but medium blocks go from (1024, span size] bytes.
Small blocks have a size class granularity of 16 bytes each in 64 buckets. Medium blocks have a granularity of 512 bytes, 61 buckets (default). Large blocks have the same granularity as the configured span size (default 64KiB). All allocations are fitted to these size class boundaries (an allocation of 36 bytes will allocate a block of 48 bytes). Each small and medium size class has an associated span (meaning a contiguous set of memory pages) configuration describing how many pages the size class will allocate each time the cache is empty and a new allocation is requested.
Spans for small and medium blocks are cached in four levels to avoid calls to map/unmap memory pages. The first level is a per thread single active span for each size class. The second level is a per thread list of partially free spans for each size class. The third level is a per thread list of free spans. The fourth level is a global list of free spans.
Each span for a small and medium size class keeps track of how many blocks are allocated/free, as well as a list of which blocks that are free for allocation. To avoid locks, each span is completely owned by the allocating thread, and all cross-thread deallocations will be deferred to the owner thread through a separate free list per span.
Large blocks, or super spans, are cached in two levels. The first level is a per thread list of free super spans. The second level is a global list of free super spans.
By default the allocator uses OS APIs to map virtual memory pages as needed, either
VirtualAlloc on Windows or
mmap on POSIX systems. If you want to use your own custom memory mapping provider you can use rpmalloc_initialize_config and pass function pointers to map and unmap virtual memory. These function should reserve and free the requested number of bytes.
The returned memory address from the memory map function MUST be aligned to the memory page size and the memory span size (which ever is larger), both of which is configurable. Either provide the page and span sizes during initialization using rpmalloc_initialize_config, or use rpmalloc_config to find the required alignment which is equal to the maximum of page and span size. The span size MUST be a power of two in [4096, 262144] range, and be a multiple or divisor of the memory page size.
Memory mapping requests are always done in multiples of the memory page size. You can specify a custom page size when initializing rpmalloc with rpmalloc_initialize_config, or pass 0 to let rpmalloc determine the system memory page size using OS APIs. The page size MUST be a power of two.
To reduce system call overhead, memory spans are mapped in batches controlled by the
span_map_count configuration variable (which defaults to the
DEFAULT_SPAN_MAP_COUNT value if 0, which in turn is sized according to the cache configuration define, defaulting to 64). If the memory page size is larger than the span size, the number of spans to map in a single call will be adjusted to guarantee a multiple of the page size, and the spans will be kept mapped until the entire span range can be unmapped in one call (to avoid trying to unmap partial pages).
On macOS and iOS mmap requests are tagged with tag 240 for easy identification with the vmmap tool.
Super spans (spans a multiple > 1 of the span size) can be subdivided into smaller spans to fulfull a need to map a new span of memory. By default the allocator will greedily grab and break any larger span from the available caches before mapping new virtual memory. However, spans can currently not be glued together to form larger super spans again. Subspans can traverse the cache and be used by different threads individually.
A span that is a subspan of a larger super span can be individually decommitted to reduce physical memory pressure when the span is evicted from caches and scheduled to be unmapped. The entire original super span will keep track of the subspans it is broken up into, and when the entire range is decommitted tha super span will be unmapped. This allows platforms like Windows that require the entire virtual memory range that was mapped in a call to VirtualAlloc to be unmapped in one call to VirtualFree, while still decommitting individual pages in subspans (if the page size is smaller than the span size).
If you use a custom memory map/unmap function you need to take this into account by looking at the
release parameter given to the
memory_unmap function. It is set to 0 for decommitting individual pages and the total super span byte size for finally releasing the entire super span memory range.
There is no memory fragmentation by the allocator in the sense that it will not leave unallocated and unusable “holes” in the memory pages by calls to allocate and free blocks of different sizes. This is due to the fact that the memory pages allocated for each size class is split up in perfectly aligned blocks which are not reused for a request of a different size. The block freed by a call to
rpfree will always be immediately available for an allocation request within the same size class.
However, there is memory fragmentation in the meaning that a request for x bytes followed by a request of y bytes where x and y are at least one size class different in size will return blocks that are at least one memory page apart in virtual address space. Only blocks of the same size will potentially be within the same memory page span.
rpmalloc keeps an “active span” and free list for each size class. This leads to back-to-back allocations will most likely be served from within the same span of memory pages (unless the span runs out of free blocks). The rpmalloc implementation will also use any “holes” in memory pages in semi-filled spans before using a completely free span.
rpmalloc provides a first class heap type with explicit heap control API. Heaps are maintained with calls to rpmalloc_heap_acquire and rpmalloc_heap_release and allocations/frees are done with rpmalloc_heap_alloc and rpmalloc_heap_free. See the
rpmalloc.h documentation for the full list of functions in the heap API. The main use case of explicit heap control is to scope allocations in a heap and release everything with a single call to rpmalloc_heap_free_all without having to maintain ownership of memory blocks. Note that the heap API is not thread-safe, the caller must make sure that each heap is only used in a single thread at any given time.
Compared to the some other allocators, rpmalloc does not suffer as much from a producer-consumer thread scenario where one thread allocates memory blocks and another thread frees the blocks. In some allocators the free blocks need to traverse both the thread cache of the thread doing the free operations as well as the global cache before being reused in the allocating thread. In rpmalloc the freed blocks will be reused as soon as the allocating thread needs to get new spans from the thread cache. This enables faster release of completely freed memory pages as blocks in a memory page will not be aliased between different owning threads.
Threads that keep ownership of allocated memory blocks within the thread and free the blocks from the same thread will have optimal performance.
Threads that have allocation patterns where the difference in memory usage high and low water marks fit within the thread cache thresholds in the allocator will never touch the global cache except during thread init/fini and have optimal performance. Tweaking the cache limits can be done on a per-size-class basis.
Since each thread cache maps spans of memory pages per size class, a thread that allocates just a few blocks of each size class (16, 32, ...) for many size classes will never fill each bucket, and thus map a lot of memory pages while only using a small fraction of the mapped memory. However, the wasted memory will always be less than 4KiB (or the configured memory page size) per size class as each span is initialized one memory page at a time. The cache for free spans will be reused by all size classes.
Threads that perform a lot of allocations and deallocations in a pattern that have a large difference in high and low water marks, and that difference is larger than the thread cache size, will put a lot of contention on the global cache. What will happen is the thread cache will overflow on each low water mark causing pages to be released to the global cache, then underflow on high water mark causing pages to be re-acquired from the global cache. This can be mitigated by changing the MAX_SPAN_CACHE_DIVISOR define in the source code (at the cost of higher average memory overhead).
VirtualAlloc has an internal granularity of 64KiB. However, mmap lacks this granularity control, and the implementation instead oversizes the memory mapping with configured span size to be able to always return a memory area with the required alignment. Since the extra memory pages are never touched this will not result in extra committed physical memory pages, but rather only increase virtual memory address space.
All entry points assume the passed values are valid, for example passing an invalid pointer to free would most likely result in a segmentation fault. The library does not try to guard against errors!.
To support global scope data doing dynamic allocation/deallocation such as C++ objects with custom constructors and destructors, the call to rpmalloc_finalize will not completely terminate the allocator but rather empty all caches and put the allocator in finalization mode. Once this call has been made, the allocator is no longer thread safe and expects all remaining calls to originate from global data destruction on main thread. Any spans or heaps becoming free during this phase will be immediately unmapped to allow correct teardown of the process or dynamic library without any leaks.
Johan Andersson at Embark has created a Rust wrapper available at rpmalloc-rs
Stas Denisov has created a C# wrapper available at Rpmalloc-CSharp
This is free and unencumbered software released into the public domain.
Anyone is free to copy, modify, publish, use, compile, sell, or distribute this software, either in source code form or as a compiled binary, for any purpose, commercial or non-commercial, and by any means.
In jurisdictions that recognize copyright laws, the author or authors of this software dedicate any and all copyright interest in the software to the public domain. We make this dedication for the benefit of the public at large and to the detriment of our heirs and successors. We intend this dedication to be an overt act of relinquishment in perpetuity of all present and future rights to this software under copyright law.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
For more information, please refer to http://unlicense.org
You can also use this software under the MIT license if public domain is not recognized in your country
The MIT License (MIT)
Copyright (c) 2017 Mattias Jansson
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.