650 Commits

Include renames

Author	SHA1	Message	Date
weaselbot	d040ed63a4	Move aarch64 Node16 SIMD index loads into assembly ... The aarch64 NEON paths in getNodeIndex/getNodeIndexExists, getChildGeq(Node16*), scan16, and checkMaxBetweenExclusiveImpl<Node16> loaded the full 16-element Node16::index array (and the Node48 reverseIndex array via scan16) with NEON intrinsics, then masked the result down to [0, numChildren). Only the in-use slots are initialized; the unused bytes are indeterminate, so the wide loads were undefined behavior in C++ (per [basic.indet]) even though the trailing lanes were discarded. MSan reports this on x86-64; on aarch64 it is the same UB but MSan's imprecise modeling doesn't flag it at -O0, so there is no red->green test. Mirror the existing x86-64 fix (commit 6fed133): implement the index operations in file-level assembly, where loading and operating on indeterminate values is well-defined. Add simd_aarch64.S with find_eq_16, find_ge_16, and mask_in_range_16. AArch64 lacks pmovmskb, so (like the prior NEON code) these return a 64-bit nibble mask rather than a 16-bit bitmask; the C++ call sites keep their existing nibble-mask arithmetic and only swap the inline NEON load/compare for the assembly call. The childMaxVersion compares stay in C++ NEON intrinsics, matching x86-64's compare16: those slots are always initialized to zero by the allocator, so the wide loads are defined. The assembly functions carry `bti c` landing pads and the same aeabi_feature_and_bits attributes the compiler emits for -mbranch-protection=standard, so the object stays BTI/PAC/GCS-compatible (and warning-free under -z force-bti). CMakeLists.txt builds simd_aarch64.S into the object library and the SIMD test/bench/fuzz targets on aarch64. Closes #68	2026-08-02 23:48:28 -04:00
andrew	6fed133212	Remove UB from indeterminate value handling ... Move SIMD operations on potentially-indeterminate Node16::index bytes into file-level assembly, where loading and operating on indeterminate values is well-defined (unlike C++). Restructure scalar fallback loops to iterate [0, numChildren) instead of [0, kMaxNodes). Fix TrivialSpan construction from indeterminate pointers in check::Job::init and insertPointWritesOrSorted to only construct when end.len > 0. Add MSan toolchain to the debug CI build to catch these issues going forward.	2026-08-02 21:06:53 -04:00
andrew	9d15af772e	Run msan in CI. Closes #66	2026-07-23 15:16:58 -04:00
weaselbot	63f9a139da	Fix undefined behavior on empty input in strinc() and prefixRange() ... strinc() in ConflictSet.cpp used std::string_view::size() (size_t) and subtracted 1 without first checking for an empty string. For the root node, getSearchPath() returns the empty string, so every debug correctness check underflowed size_t and relied on implementation-defined conversion to signed int. prefixRange() in Bench.cpp had the same loop shape. Although TrivialSpan::size() returns int, on an empty (or all-0xff) key the function then asserted and continued executing, allocating a zero-length buffer and writing before its start. Changes: - In strinc(), initialize index as signed int(str.size()) - 1 so the loop is skipped for empty input, and return ok=false cleanly. - In prefixRange(), initialize index the same way and call std::abort() after the assert so invalid input cannot fall through to an out-of-bounds write. - Replace C-style uint8_t casts with explicit static_casts. Closes #59	2026-06-29 13:52:02 -04:00
weaselbotandandrew	4dc5f7f75c	Fix self-move-assignment and leak in ConflictSet move-assignment ... The user-declared move-assignment operator overwrote `impl` without first destroying the existing implementation object, leaking all memory and resources owned by the left-hand side. Self-move-assignment also set `impl` to nullptr, leaving the object invalid and leaking the old state. Fix all three implementations (ConflictSet.cpp, SkipList.cpp, HashTable.cpp) to guard against self-assignment and to destroy/free the old `impl` before taking ownership of `other.impl`.	2026-06-22 14:08:59 -04:00
andrew	3a82d90914	Work around arm64 clang codegen issues breaking CI ... Disable preserve_none under ASan on aarch64: every clang tested (20, 21, trunk 22) miscompiles the continuation chains with that combination, crashing ~97% of fuzz corpus tests. The chains still work with the default calling convention. See #38. Mark the masked Node3 scan results defined for valgrind on aarch64: clang 21+ lowers the in-bounds tests through flags+csel, which memcheck models imprecisely, tainting bits the mask provably clears. See #39. Skip valgrind tests in the arm64 release job, since -DNVALGRIND compiles out the client requests the workaround relies on. Both issues track filing upstream bugs.	2026-06-12 16:35:57 -04:00
andrew	2642d453dc	Remove unnecessary branch for interleaved range writes	2024-12-11 21:53:44 -08:00
andrew	7166811387	Fix condition checking for erasing the root	2024-11-21 17:49:54 -08:00
andrew	81323972aa	Remove "memcpy common section to next node" ... Do it in a simpler, more robust/type safe way	2024-11-21 16:47:05 -08:00
andrew	8694ba8b6a	Remove unused field	2024-11-21 16:35:10 -08:00
andrew	0cea5565b5	Add artificial root parent and stop plumbing impl everywhere	2024-11-21 16:32:13 -08:00
andrew	972f16ed8f	Remove Node::endOfRange ... This should simplify storing leaves directly	2024-11-21 11:55:30 -08:00
andrew	2412684316	childMaxVersion memory needs to be defined ... Also remove some dead code, and only memset children to 0 for Node256. Other nodes detect absence of children other ways.	2024-11-21 10:35:05 -08:00
andrew	8251631087	Fix some unintentional generic usages of getChildAndMaxVersion	2024-11-20 21:50:48 -08:00
andrew	90fb2a9542	Remove some usages of generic getFirstChild	2024-11-20 21:43:47 -08:00
andrew	7c01f8ba0f	Remove some usages of maxVersion	2024-11-20 21:24:59 -08:00
andrew	0df2db7f8a	Remove some bad unlikely annotations found by -Wmisexpect ... Using a profile from server_bench	2024-11-20 19:06:50 -08:00
andrew	e125b599b5	Remove freeList, min/max capacity tracking ... The freelist doesn't seem to get a good hit rate. Policies other than capacity = minCapacity did not improve the rate we were resizing nodes, but did increase memory usage, so get rid of that too. Add a nodes_resized_total counter.	2024-11-20 14:45:56 -08:00
andrew	3f4d3b685a	More valgrind annotations	2024-11-20 13:36:30 -08:00
andrew	4198b8b090	Some prep for leafs-in-parents	2024-11-20 12:20:11 -08:00
andrew	8757d2387c	Call prefetch within TaggedNodePointer::getType ... Instead of at every call site	2024-11-20 12:07:32 -08:00
andrew	4a22b95d53	Remove state machine transitions "from" Node0 ... Those aren't used	2024-11-20 11:46:45 -08:00
andrew	03d6c7e471	Allocate maxCapacity instead of minCapacity	2024-11-19 16:13:57 -08:00
andrew	ceecc62a63	Disable freelist for macos ... It's faster on my mac m1 without the freelist	2024-11-19 14:28:15 -08:00
andrew	e5b9c03e77	Restore change that hurt codegen ... I don't understand, but this is necessary for good codegen somehow	2024-11-18 14:29:55 -08:00
andrew	ee5a84cd7b	Remove dead stores	2024-11-15 17:03:29 -08:00
andrew	77262ee2d3	Fix some grammar in a comment	2024-11-15 16:47:31 -08:00
andrew	2777e016ff	Be more consistent about TaggedNodePointer vs Node*	2024-11-15 16:30:05 -08:00
andrew	661ffcd843	Explain purpose for prefetches in getFirstChild	2024-11-15 16:29:42 -08:00
andrew	3a34d3cecb	Minor style improvements	2024-11-15 16:29:17 -08:00
andrew	189c73e3bd	Reduce Node3 size by 8	2024-11-15 15:53:18 -08:00
andrew	f5ec9f726a	Remove getFirstChildExists ... Once we know the type, for Node3 and higher we know the first child exists anyway	2024-11-15 13:05:11 -08:00
andrew	552fc11c5d	Prefetch second child to improve scan performance	2024-11-15 12:45:01 -08:00
andrew	08958d4109	Remove the reverse step for saving scan search path	2024-11-14 16:27:25 -08:00
andrew	b78e817e24	Only set HAS_AVX for x86_64	2024-11-12 18:16:30 -08:00
andrew	665a9313a4	Valgrind annotations for new free list	2024-11-12 18:11:09 -08:00
andrew	a92271a205	Build with -Wunused-variable	2024-11-12 17:50:27 -08:00
andrew	0dbfb4deae	Detect simd headers in c++ instead of cmake	2024-11-12 17:50:27 -08:00
andrew	b37feb58dd	Require musttail and preserve_none for interleaved	2024-11-12 17:50:27 -08:00
andrew	f85b92f8db	Improve next{Physical,Logical} codegen	2024-11-09 13:21:36 -08:00
andrew	3c44614311	Allocate from freelist with min/max capacity constraints	2024-11-08 21:35:13 -08:00
andrew	9c1ac3702e	Move index closer to start of Node{3,16} ... This should slightly improve cache hit rate	2024-11-08 20:54:00 -08:00
andrew	4494359ca2	Make insert_iterations for interleaved writes more closely match	2024-11-04 14:46:30 -08:00
andrew	4eaad39294	Maintain capacity invariant strictly	2024-11-04 13:43:02 -08:00
andrew	d6269c5b7c	Use interleaved even if preserve_none is missing	2024-11-01 15:12:59 -07:00
andrew	821179b8de	Remove dead code	2024-11-01 13:48:32 -07:00
andrew	681a961289	Dispatch on type pairs for end iter	2024-11-01 13:45:41 -07:00
andrew	c73a3da14c	Dispatch on type pairs for begin iter	2024-11-01 12:46:36 -07:00
andrew	5153d25cce	Dispatch on type pairs for common prefix iter	2024-11-01 12:06:48 -07:00
andrew	d2ec4e7fae	Fix use of wrong member in endIter	2024-11-01 11:17:26 -07:00