#pragma once #include "Internal/tlsf_allocator.inl.h" #include "heap_allocator.h" namespace baselib { BASELIB_CPP_INTERFACE { // tlsf_allocator (Two-Level Segregated Fit) // Lockless, dynamic-sized allocator capable of handling multiple concurrent allocations and deallocations (unless otherwise stated). // The cost (in processor instructions) allocating from the pool is O(1). // Allocating from the pool is lockless, except when capacity is required to increase, in which case the capacity is doubled for the a size range // of the particular allocation size request (see details below). // // Strict segregated fit allocation mechanism is applied, a requested size is rounded up to the next allocator provided size. // The granularity of provided sizes (size ranges) are defined by the `min_size`, `max_size` and `linear_subdivisions` parameters provided. // `optimal_size` can be called to obtain the actual/best fit size of an allocation for a certain requested size. // // A two-Level segregated fit allocator can be said to have two dimensions, or levels. // The first level provides size ranges of pow2 segments. // The second level provides size ranges of the first level pow2 segments size range divided by the `linear_subdivisions` parameter value. // Size range of a given size is be calculated as follows: // // int invSizeMask = ((1 << (int)log2(size)) / linear_subdivisions) - 1; // Inverse subdivision mask based on Pow2 of `size`, which is effectively range // int lowerBound = (size - 1 & ~invSizeMask) + 1; // int upperBound = lowerBound + invSizeMask; // // As an example, the (internal) size allocated for a requested size of 1500 with linear_subdivisions of 16 is 1536 range(1473-1536). // // Notes on performance/memory requirements: // // - This implementation is a segregated storage algorithm and does not, unlike a segregated fit algorithm (aka buddy allocator) split and coalesce // memory blocks. A segregated fit is well suited for a single threaded/lock-based implementation but would require multiple atomic operations to split // or coalesce blocks. // // - All allocators share a common base instance of the backing allocator `Allocator`, which is used for allocation when the capacity is required to // increase. Memory is only freed up when the tlsf allocator `deallocate_all` or destructor is invoked. // Furthermore, `deallocate_all` is optional to declare in the backing allocator `Allocator` and is if so invoked (once) instead of multiple `deallocate` // calls when `deallocate_all` (or destructor) is invoked on the tlsf allocator. // // - The allocator is constructed with only as many block allocators required for the selected min-max range with linear_subdivisions. // I.e. one allocator with (min_size, max_size, linear_subdivisions) 32,1024,8 has the same memory footprint as two 32,512,8 and 513,1024,8. // If either level allocator only requires a single allocator providing a range, code for calculating allocator indices is optimized away by template // construction. Additionally, if size is known at compile-time (const or sizeof) lookup can be optimized away by the compiler. // // - No overhead per allocation (no header information). // // - Internally, all memory block sizes must be rounded up to a multiple of alignment. I.e if alignment is 64, buckets containing 96 byte size allocations // will in internally use 128 byte blocks. Additionally, smallest size allocated will always be greater than or equal to `linear_subdivisions`. // // - The allocator relies on that the free memory pool must be persisted and read/write accessible as link information of free memory blocks are // read/written to by the allocator operations. // // Examples: // Range is within a single pow2 range, no subdivisions. No lookup code needed. // using BlockAllocator = tlsf_allocator<17, 32, 1>; // // Range is within a single pow2 range with 8 subdivisions, so in this case with linear increments (128/8=16) of bucket sizes. Second level lookup only. // using SegregatedFitAllocatorLinear = tlsf_allocator<129, 256, 8>; // // Range is several pow2 ranges, no subdivisions so pow2 size increments of bucket sizes. // using SegregatedFitAllocatorPow2 = tlsf_allocator<129, 2048, 1>; // // Range is several pow2 ranges, with 32 subdivisions each, so pow2 size increments where each pow2 contains an array of buckets with linear size // increments (pow2sz/32) of bucket sizes. // using TLSFAllocator = tlsf_allocator<129, 2048, 32>; // // // tlsf_allocator> // // min_size - valid minimum size of allocations. // max_size - valid maximum size of allocations. Must be less or equal to the size addressable by integral type `size_t` divided by two plus 1. // linear_subdivisions - number of linear subdivisions of second level allocators (defaults to 1). Must be a power of two and less or equal to `min_size` // Allocator - Backing memory allocator. Defaults to baselib heap_allocator. // template > class tlsf_allocator : protected detail::tlsf_allocator { using Impl = detail::tlsf_allocator; static_assert(min_size <= max_size, "min_size > max_size"); static_assert(min_size >= linear_subdivisions, "min_size < linear_subdivisions"); static_assert(max_size <= std::numeric_limits::max() / 2 + 1, "max_size > std::numeric_limits::max() / 2 + 1"); static_assert(baselib::Algorithm::IsPowerOfTwo(linear_subdivisions), "linear_subdivisions != pow2"); public: // non-copyable tlsf_allocator(const tlsf_allocator& other) = delete; tlsf_allocator& operator=(const tlsf_allocator& other) = delete; // non-movable (strictly speaking not needed but listed to signal intent) tlsf_allocator(tlsf_allocator&& other) = delete; tlsf_allocator& operator=(tlsf_allocator&& other) = delete; // Allocated memory is guaranteed to always be aligned to at least the value of `alignment`. static constexpr uint32_t alignment = Impl::alignment; // Creates a new instance tlsf_allocator() { atomic_thread_fence(memory_order_seq_cst); } // Destroy allocator, deallocates any memory allocated. // // If there are other threads currently accessing the allocator behavior is undefined. ~tlsf_allocator() {} // Allocates a memory block large enough to hold `size` number of bytes if allocation does not require increasing capacity. // // \returns Address to memory block of allocated memory or nullptr if failed or outside of size range. void* try_allocate(size_t size) { return owns(nullptr, size) ? Impl::try_allocate(size) : nullptr; } // Allocates a memory block large enough to hold `size` number of bytes. // // \returns Address to memory block of allocated memory or nullptr if failed or outside of size range void* allocate(size_t size) { return owns(nullptr, size) ? Impl::allocate(size) : nullptr; } // Reallocates previously allocated or reallocated memory pointed to by `ptr` from `old_size` to `new_size` number of bytes if reallocation does not // require increasing capacity. Passing `nullptr` in `ptr` yield the same result as calling `try_allocate`. // // \returns Address to memory block of reallocated memory or nullptr if failed or if `new_size` is outside of size range. void* try_reallocate(void* ptr, size_t old_size, size_t new_size) { return owns(nullptr, new_size) ? Impl::try_reallocate(ptr, old_size, new_size) : nullptr; } // Reallocates previously allocated or reallocated memory pointed to by `ptr` from `old_size` to `new_size` number of bytes. // Passing `nullptr` in `ptr` yield the same result as calling `allocate`. // // \returns Address to memory block of reallocated memory or nullptr if failed or if `new_size` is outside of size range void* reallocate(void* ptr, size_t old_size, size_t new_size) { return owns(nullptr, new_size) ? Impl::reallocate(ptr, old_size, new_size) : nullptr; } // Deallocates memory block previously allocated or reallocated with `size` pointed to by `ptr`. // Passing `nullptr` in `ptr` result in a no-op. // // \returns Always returns `true` bool deallocate(void* ptr, size_t size) { return Impl::deallocate(ptr, size); } // Free a linked list of allocations created using `batch_deallocate_link` with `size`. // `first` to `last` is first and last allocation of a `batch_deallocate_link` series of calls. // // \returns Always returns `true` bool batch_deallocate(void* ptr_first, void* ptr_last, size_t size) { return Impl::batch_deallocate(ptr_first, ptr_last, size); } // Link previously allocated memory of `size` to another. // // Use to create a linked list of allocations for use with `batch_deallocate(first, last, size)` // Size of linked allocations are required to be equal to `size`. // `nullptr` is a valid argument for `ptr_next`, but is not needed to terminate a linked list. // This is implicit transfer of ownership of the memory back to the allocator. // Memory of the allocation must not be accessed/modified once linked. void batch_deallocate_link(void* ptr, void* ptr_next, size_t size) { Impl::batch_deallocate_link(ptr, ptr_next); } // Release all resources and set capacity to zero // // Calling this function invalidates any currently allocated memory // If there are other threads currently accessing the allocator behavior is undefined. void deallocate_all() { Impl::deallocate_all(); } // Requests that the allocator capacity be at least enough to contain `capacity` for allocations of `size`. // // If `capacity` is less or equal to current capacity for allocations of `size`, the capacity is not affected. // Note that internally, `capacity` is rounded up to the nearest optimal allocation size based on `Allocator` attributes. // // \returns true if successful. bool reserve(size_t size, size_t capacity) { return owns(nullptr, size) ? Impl::reserve(size, capacity) : false; } // Get the current capacity of allocations with `size`. size_t capacity(size_t size) { return owns(nullptr, size) ? Impl::capacity(size) : 0; } // Calculate optimal allocation size given `size`. // // \returns Optimal size when allocating memory given `size` or zero if outside size range. static constexpr size_t optimal_size(const size_t size) { return owns(nullptr, size) ? Impl::optimal_size(size) : 0; } // Checks for the ownership allocation given `ptr` and `size` // It is implementation defined if either or both of `ptr` and `size` are considered to determine ownership. // This allocator does not consider `ptr`. // // \returns True if the allocator owns the allocation. static constexpr bool owns(const void *, size_t size) { return size - min_size <= max_size - min_size; } }; } }