[libc-commits] [libc] [libc] Use best-fit binary trie to make malloc logarithmic (PR #106259)

[via libc-commits]

================
@@ -9,199 +9,107 @@
 #ifndef LLVM_LIBC_SRC___SUPPORT_FREELIST_H
 #define LLVM_LIBC_SRC___SUPPORT_FREELIST_H
 
-#include "src/__support/CPP/array.h"
-#include "src/__support/CPP/cstddef.h"
-#include "src/__support/CPP/new.h"
-#include "src/__support/CPP/span.h"
-#include "src/__support/fixedvector.h"
-#include "src/__support/macros/config.h"
+#include "block.h"
 
 namespace LIBC_NAMESPACE_DECL {
 
-using cpp::span;
-
-/// Basic [freelist](https://en.wikipedia.org/wiki/Free_list) implementation
-/// for an allocator. This implementation buckets by chunk size, with a list
-/// of user-provided buckets. Each bucket is a linked list of storage chunks.
-/// Because this freelist uses the added chunks themselves as list nodes, there
-/// is a lower bound of `sizeof(FreeList.FreeListNode)` bytes for chunks which
-/// can be added to this freelist. There is also an implicit bucket for
-/// "everything else", for chunks which do not fit into a bucket.
-///
-/// Each added chunk will be added to the smallest bucket under which it fits.
-/// If it does not fit into any user-provided bucket, it will be added to the
-/// default bucket.
-///
-/// As an example, assume that the `FreeList` is configured with buckets of
-/// sizes {64, 128, 256, and 512} bytes. The internal state may look like the
-/// following:
-///
-/// @code{.unparsed}
-/// bucket[0] (64B) --> chunk[12B] --> chunk[42B] --> chunk[64B] --> NULL
-/// bucket[1] (128B) --> chunk[65B] --> chunk[72B] --> NULL
-/// bucket[2] (256B) --> NULL
-/// bucket[3] (512B) --> chunk[312B] --> chunk[512B] --> chunk[416B] --> NULL
-/// bucket[4] (implicit) --> chunk[1024B] --> chunk[513B] --> NULL
-/// @endcode
+/// A circularly-linked FIFO list storing free Blocks. All Blocks on a list
+/// are the same size.
 ///
-/// Note that added chunks should be aligned to a 4-byte boundary.
-template <size_t NUM_BUCKETS = 6> class FreeList {
+/// Allocating free blocks in FIFO order maximizes the amount of time before a
+/// free block is reused. This in turn maximizes the number of opportunities for
+/// it to be coalesced with an adjacent block, which tends to reduce heap
+/// fragmentation.
+class FreeList {
 public:
-  // Remove copy/move ctors
----------------
nopsledder wrote:

Recommend keeping these unless you have a reason not to.

https://github.com/llvm/llvm-project/pull/106259