Failed STC (https://tests.stockfishchess.org/tests/view/6a4f36355529b8472df7ff7e):
LLR: -2.94 (-2.94,2.94) <0.00,2.00>
Total: 144768 W: 38003 L: 38051 D: 68714
Ptnml(0-2): 521, 17154, 37050, 17170, 489
Passed LTC (https://tests.stockfishchess.org/tests/view/6a501dad5529b8472df80103):
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 189888 W: 49613 L: 48984 D: 91291
Ptnml(0-2): 122, 20406, 53256, 21041, 119
Passed VLTC (https://tests.stockfishchess.org/tests/view/6a52892c5529b8472df80481):
LLR: 2.94 (-2.94,2.94) <0.00,2.00>
Total: 46124 W: 12156 L: 11865 D: 22103
Ptnml(0-2): 12, 4523, 13696, 4824, 7
Passed VVLTC (https://tests.stockfishchess.org/tests/view/6a566a285529b8472df80b84):
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 44698 W: 11805 L: 11507 D: 21386
Ptnml(0-2): 9, 3819, 14395, 4117, 9
This PR adds pawn pair features, a new set of features that is indexed by pairs of pawns occupying the same or adjacent files (hence "3-wide"). Because these are a superset of pawn–pawn interactions in threat inputs (including the recently added pawn-pusher inputs), those are removed. HalfKA features remain unchanged.
Pawn-pair features were invented by Jonathan for Pawnocchio, based on his observation that in a net trained on *all* pairs of pawns, the most important pawn pairs were those differing by at most one file.
sscg13 made the necessary changes to the trainer (https://github.com/official-stockfish/nnue-pytorch/pull/502) and initial changes to inference. The slowdown was originally far too large (some 10%); following major optimizations from Jonathan and a bit from me, the slowdown is small enough for a comfortable LTC pass; on my machine I get a 3.5% slowdown overall. I expect there are further "easy" speedups in the new code, so this gap should narrow a bit.
Nettest PR: https://github.com/vondele/nettest/pull/359
closes https://github.com/official-stockfish/Stockfish/pull/6982
Bench: 2718396
Co-Authored-By: anematode <timothy.herchen@gmail.com>
Co-Authored-By: Jonathan Hallström <lmj.hallstrom@gmail.com>
After a858defd33, the pthread implementation of `NativeThread` is always used (except MSVC), as:
1. `-DUSE_PTHREADS` is defined when `comp` is not `mingw`
2. the pthread implementation is used anyways because MinGW defines `__MINGW32__` or `__MINGW64__` macro
closes https://github.com/official-stockfish/Stockfish/pull/6985
No functional change
Fixes#6984, first reported by sqrmax on the Discord, causing incorrect detection of NUMA and L3 domains on Linux due to trailing whitespace when reading from sysfs.
This was introduced by #6844 which made `str_to_size_t` too strict, since `std::stoull` allows trailing characters (and leading whitespace, although this part is not relevant for us).
Before:
$ ./bins/stockfish.master speedtest
Stockfish dev-20260716-ebcea3ef by the Stockfish developers (see AUTHORS file)
info string Using 224 threads with NUMA node thread binding: 64/32:64/32:64/32:32/16
After:
$ ./stockfish speedtest
Stockfish dev-20260718-m-cb702902 by the Stockfish developers (see AUTHORS file)
info string Using 224 threads with NUMA node thread binding: 32/32:32/32:32/32:32/32:32/32:32/32:32/32
closes https://github.com/official-stockfish/Stockfish/pull/6989
No functional change
This PR:
- retires the unused distance() functions in bitboard.h including the 4K SquareDistance LUT,
- makes the 64K PopCnt16 LUT constexpr and only compiles it in if we don't use a popcount intrinsic (we can't fully remove the LUT because it's faster than the intrinsic in the absence of a popcount machine instruction),
- retires Bitboards::init().
Passed STC:
https://tests.stockfishchess.org/tests/view/6a48e6d3f97ff95f78795a8f
LLR: 2.94 (-2.94,2.94) <-1.75,0.25>
Total: 162016 W: 42054 L: 41976 D: 77986
Ptnml(0-2): 383, 17375, 45454, 17373, 423
I didn't run LTC because the changes only really impact workers without popcount and there aren't any present on fishtest ATM. The correctness of the constexpr PopCnt16 LUT was verified locally with a x86-64 build; the build yields the same signature as master.
closes https://github.com/official-stockfish/Stockfish/pull/6952
No functional change
Add -fno-ipa-cp-clone for mingw just as we do for gcc. I get about a 1% speedup. My machine is pretty unreliable for bench so maybe someone can verify.
closes https://github.com/official-stockfish/Stockfish/pull/6973
No functional change
With the latest master I started getting:
"A breakpoint instruction (__debugbreak() statement or a similar call) was executed in Stockfish.exe." when debugging in MSVC.
I don't really understand the issue but Fable fixed it and I verified it works.
Here is its summary:
Bug: Calling _get_wpgmptr() in CommandLine::get_binary_directory() triggered a debug CRT assertion failure (__debugbreak()) on Windows. The CRT only initializes the wide program name _wpgmptr when the application uses a wide entry point (wmain). Since Stockfish uses the narrow main(int, char**) entry point, _wpgmptr remains nullptr, and the debug CRT's parameter validation asserts when _get_wpgmptr is called.
Fix: Replaced _get_wpgmptr() with GetModuleFileNameW(), which queries the executable path directly from the OS and does not depend on the CRT entry point type. The implementation uses a grow-and-retry loop, so paths longer than MAX_PATH are handled correctly. If the API fails, the code falls back to the original argv0-based behavior. The #ifdef _MSC_VER guard was also removed, since the fix no longer relies on the MSVC CRT — non-MSVC Windows builds (e.g., MinGW) now benefit from the accurate path as well.
closes https://github.com/official-stockfish/Stockfish/pull/6972
No functional change
The following have zero references repo-wide:
Move::to_sq_unchecked() (types.h): added in 542c30c2 for a
branchless correction-history update; its only caller was removed
in 5ae13d2b ("Remove Redundant Branchless Execution"), leaving the
method unused.
PipeDeleter (misc.h): the custom deleter for the std::unique_ptr<FILE,
PipeDeleter> that wrapped popen("lscpu"). That pipe was removed in
c8375c2f ("On linux use sysfs instead of lscpu"), so the struct has
been unused since.
Dropping PipeDeleter also removes the last user of <cstdio> in misc.h, so that include is dropped as well.
closes https://github.com/official-stockfish/Stockfish/pull/6948
No functional change
This reverts commit 9d4090e826.
causes very poor mate finding performance, ultimately causing a CI hang on a mate in 2
Testcase:
position fen k1B5/2p5/NbN5/P7/3p4/P2p4/P1prp3/2RbK3 b - - 0 1 setoption name Hash value 16
setoption name Threads value 4
go mate 2
ucinewgame
closes https://github.com/official-stockfish/Stockfish/pull/6966
Bench: 2466447
`Skill::pick_best()` assumes the root moves are sorted by score in descending order:
```cpp
Value topScore = rootMoves[0].score;
int delta = std::min(topScore - rootMoves[multiPV - 1].score, int(PawnValue));
```
That does not have to be so with Syzygy tablebases at the root, where the moves are ordered by `tbRank` instead. `rootMoves[0]` is then not the highest score and `delta` can go negative. Since `delta * (rng.rand<unsigned>() % int(weakness))` is evaluated as unsigned, a negative `delta` wraps to a large value and the `int(...)` cast overflows, so the `score + push >= maxScore` check never passes and `best` stays `Move::none()`. The caller
```cpp
std::swap(rootMoves[0],
*std::find(rootMoves.begin(), rootMoves.end(),
skill.best ? skill.best : skill.pick_best(rootMoves, multiPV)));
```
then dereferences `end()`. The result is a crash, or sometimes `bestmove (none)` / an illegal move, in tablebase endgames when `UCI_LimitStrength` is set (or `Skill Level` is below 20).
The fix computes the score range over the candidate moves directly, so `delta` stays non-negative and a valid move is always returned.
**To reproduce**:
```
setoption name UCI_LimitStrength value true
setoption name UCI_Elo value 2900
setoption name SyzygyPath value <syzygy tablebases path>
position fen 8/8/8/4k3/8/8/3BKN2/8 b - - 0 1
go wtime 250 btime 250 winc 100 binc 100
```
You need to call the go command a couple of times before crash, after which is ends with
```
bestmove a1e5 ponder (none)
```
and on the next `go wtime 250 btime 250 winc 100 binc 100` it crashes printing:
```
info string Available processors: 0-31
info string Using 1 thread
info string NNUE evaluation using nn-af1339a6dea3.nnue (106MiB, (83248, 1024, 31, 32, 1))
info string Network replica 1: Shared memory.
```
Bench is unchanged, since this code only runs under `UCI_LimitStrength` / `Skill Level`.
closes https://github.com/official-stockfish/Stockfish/pull/6957
No functional change
Reduce the number of *easy* ways to hit UB or an uncaught exception. This builds on Sopel's work on FEN validation.
Obviously SF is not and will never be a security boundary, but at least we can reduce the number of reports about uncaught `std::stoi` out of range, etc.
I also added a gentle fallback to `NumaConfig` specifically because that's something that (as a dev) I frequently set by hand in the CLI, but happy to revert if that's too annoying of a change.
closes https://github.com/official-stockfish/Stockfish/pull/6844
No functional change
In the spirit of previous such PRs, this PR proposes a collection of nonfunctional changes.
I am happy to incorporate changes from other devs, and revert some of the proposed changes if the maintainers ask me to.
Apart from trivial changes, the proposed changes so far include:
* A requested edit to `AUTHORS` and a resorting of all entries, following DIN 5007 for the treatment of any special characters.
* Exclude the two recent integer type renaming commits from git blame.
* Tightening of some static asserts in `history.h` to avoid overflows. (Note that rounding errors for floating point types could lead to the assert in `operator<<` triggering at run-time.)
* Re-instate the 0.5s maximal thinking time in case of a single legal move and reword the comment to make it clear that it should not be tuned.
* ~~A small refactoring of the network loading code thanks to @dubslow.~~
* A refactoring of the "dtz is dtm" code, also thanks to @dubslow.
closes https://github.com/official-stockfish/Stockfish/pull/6928
No functional change
GCC 12.2 added `__rbitll` intrinsic to `arm_acle.h`, so we can add a fallback for older versions
Clang 10 (our new minimum supported version) is unaffected by all this
closes https://github.com/official-stockfish/Stockfish/pull/6949
No functional change
Fixes problems with non-ASCII characters in paths. Fixes#3424.
Changes
- Bump the minimum supported Clang version to 11 to support std::filesystem
- Use std::optional for empty values instead of string sentinels like "None"
- Remove the FixedString class
- Move the EvalFile network path from the network struct to the engine
- Simplify parts of the command-line handling
- Remove the old dual-net logic from export_net
- And ofc we now support non-ascii paths for windows
Behavior Change
- Because the EvalFile is no longer part of the network, it won't be included in the hashing, thus the network sharing is no longer dependent on the EvalFile path
- As a result, if the binary version and network are the same, changing only the EvalFile location will now reuse the same shared network
This also fixes a master bug where:
```
export_net None
setoption name EvalFile value None
export_net None
Failed to export a net
```
would fail on the second export. With this patch, the second export succeeds.
Quick Fishtest Check that nets load like normal.
https://tests.stockfishchess.org/tests/view/6a455593f97ff95f787954da
<img width="833" height="1137" alt="image" src="https://github.com/user-attachments/assets/fffee63b-c63a-4d5e-b13d-87d58bc28c88" />
closes https://github.com/official-stockfish/Stockfish/pull/6937
No functional change
Co-authored-by: anematode <timothy.herchen@gmail.com>
Performance on Spacemit K3, thanks @edolnx for testing
master:
Total time (ms) : 65200
Nodes searched : 3493826
Nodes/second : 53586
riscv-scalable-port:
Total time (ms) : 15834
Nodes searched : 3493826
Nodes/second : 220653
Also thanks to @camel-cdr for guidance on RVV programming, and https://cloud-v.co for supplying an RVV instance to test with
passed STC:
LLR: 2.81 (-2.94,2.94) <0.00,2.00>
Total: 1152 W: 527 L: 108 D: 517
Ptnml(0-2): 0, 17, 167, 348, 44
https://tests.stockfishchess.org/tests/view/6a39895b3036e45021aeb368
## Summary
We've had a `riscv64` target for a while, but haven't really optimized for it, in particular the vector extension (RVV).
RVV, like SVE, is based on a scalable vector system where the vector length ranges from 128 to 65536. In practice implementations are between 128 and 2048, and 256 bits is quite common (e.g. the Spacemit K3 system above). Unfortunately this doesn't fit well into the rest of our code which assumes a fixed vector length, so what I've done is bypass the `VECTOR` ifdef (which now basically means "FIXED_LENGTH_VECTOR") and just have RVV-specific paths.
The ability to explicitly control `vl` makes the code quite readable, in my opinion. We use LMUL>1 in most places to take advantage of multi-vector instructions. Generally the LMULs were chosen to best support a 256-bit vlen, which is very common, but by virtue of how the vlen control works, the code works with any vlen. In a couple places, i.e., `get_changed_pieces` and `AffineTransformSparseInput::propagate`, we have separate implementations depending on the vlen, because the optimal LMUL varies a lot between implementations.
One little wrinkle is that `load_as` is compiled to a sequence of byte loads, because although unaligned loads are legal in RVA23, the spec says that they *may* be extremely slow (even though they usually aren't, in actual hw), so compilers are conservative. Thus I aligned the relevant buffers and made the semantics of `load_as` that the operand is aligned, by adding a runtime assertion.
### Universal binary
Adding a universal binary is pretty easy and we can just cross-compile. There are two targets: baseline rv64gc and riscv64-rva23, which is actually a smaller subset of RVA23 that also works on some older processors that don't support the full thing. We use clang because GCC, until recently, has a nasty bug with LTO and RVV.
Like the universal ARM and x86 builds, we check all the builds in CI. In this case we run bench with multiple vlens, 128 through 1024.
In the meantime I deleted the existing broken and unused riscv64 tests.
### Follow-ups
- Optimizations
- zvdot4a8i path
closes https://github.com/official-stockfish/Stockfish/pull/6920
No functional change
Passed STC non-regression on avx512icl (https://tests.stockfishchess.org/tests/view/6a421b46f97ff95f78795061)
LLR: 2.96 (-2.94,2.94) <-1.75,0.25>
Total: 81632 W: 21240 L: 21075 D: 39317
Ptnml(0-2): 184, 8907, 22464, 9082, 179
Passed STC non-regression on bmi2 (https://tests.stockfishchess.org/tests/view/6a41f6f9f97ff95f7879503e)
LLR: 2.93 (-2.94,2.94) <-1.75,0.25>
Total: 78112 W: 20411 L: 20245 D: 37456
Ptnml(0-2): 188, 8449, 21632, 8583, 204
After #6845 , pext attacks are slower or equal to parallel-hyperbola attacks which works on all architectures that pext supports. SVE2 supports pext/pdep in vector registers, but it's much slower than the `rbit`-based HQ that we have on ARM, so I don't foresee the code being useful there.
Ofc I don't think we should remove the `bmi2` build or anything, because it's nice for people to be able to freely use pext when testing ideas, and I'm sure we'll some day find another use for it!
closes https://github.com/official-stockfish/Stockfish/pull/6939
No functional change
Co-authored-by: Dubslow <bunslow@gmail.com>
When compiling for Android (COMP=ndk) on macOS, the build fails because the
NDK's LLVM toolchain does not recognize the macOS-specific flags
-mdynamic-no-pic and -mmacosx-version-min.
This PR wraps these flags with ifneq ($(COMP),ndk) condition to skip them
during Android cross-compilation.
Please squash and merge this PR.
closes https://github.com/official-stockfish/Stockfish/pull/6927
No functional change
This network is trained by adding the following binpacks, relabeled by @vondele and totaling 74B positions, to the distillation fine-tuning stage:
```
- vondele/from_kaggle_1_relabel/leela96-filt-v2.min.split_0.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_1_relabel/leela96-filt-v2.min.split_1.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_1_relabel/leela96-filt-v2.min.split_2.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_1_relabel/leela96-filt-v2.min.split_3.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_1_relabel/leela96-filt-v2.min.split_4.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_2_relabel/T60T70wIsRightFarseerT60T74T75T76.split_0.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_2_relabel/T60T70wIsRightFarseerT60T74T75T76.split_1.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_2_relabel/T60T70wIsRightFarseerT60T74T75T76.split_2.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_2_relabel/T60T70wIsRightFarseerT60T74T75T76.split_3.relabel-BT4-tf13tune.binpack
- vondele/from_kaggle_2_relabel/T60T70wIsRightFarseerT60T74T75T76.split_4.relabel-BT4-tf13tune.binpack
```
The relabeling effort has been completed during the testing of this patch, and a full training run is on the way. Thanks to @vondele, @anematode, @Disservin, @yl25946, and all who've contributed to the process.
Passed STC:
LLR: 2.95 (-2.94,2.94) <0.00,2.00>
Total: 24512 W: 6500 L: 6201 D: 11811
Ptnml(0-2): 69, 2772, 6300, 3021, 94
https://tests.stockfishchess.org/tests/view/6a40b7083036e45021aebbd6
Passed LTC:
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 15708 W: 4232 L: 3960 D: 7516
Ptnml(0-2): 8, 1572, 4420, 1848, 6
https://tests.stockfishchess.org/tests/view/6a413d293036e45021aebc84
nettest PR: https://github.com/vondele/nettest/pull/388
closes https://github.com/official-stockfish/Stockfish/pull/6932
Bench: 2102535
Co-authored-by: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Deduplicating Color-Specific Piece Validation.
The validation checks for the number of pawns and additional promoted pieces are duplicated for WHITE and BLACK. We can combine this logic into a single range-based for loop over both colors.
closes https://github.com/official-stockfish/Stockfish/pull/6922
No functional change
do_null_move copies the whole StateInfo from the previous state, which leaves capturedPiece holding the piece captured by the last real move, so inside the null-move subtree captured_piece() reports a stale capture. The priorCapture consumers in search are all guarded by prevSq != SQ_NONE or (ss-1)->currentMove.is_ok(), which are false at the null-move child, but the stalemate verification gate in qsearch reads captured_piece() unguarded and can be spuriously triggered by the stale value.
Clear the field, since a null move captures nothing.
Passed non-regression STC:
LLR: 2.93 (-2.94,2.94) <-1.75,0.25>
Total: 82784 W: 21172 L: 21011 D: 40601
Ptnml(0-2): 194, 8976, 22923, 9073, 226
https://tests.stockfishchess.org/tests/view/6a2b5b356b4aa63ddbf31518
Passed non-regression LTC:
LLR: 2.94 (-2.94,2.94) <-1.75,0.25>
Total: 118134 W: 30007 L: 29891 D: 58236
Ptnml(0-2): 66, 11856, 35108, 11970, 67
https://tests.stockfishchess.org/tests/view/6a2c4c9c0d5d4b19d0805301
closes https://github.com/official-stockfish/Stockfish/pull/6910
No functional change
The following have zero call sites repo-wide:
SearchManager:🆔 never read or written (also never initialized, nor was it ever used).
Search::Worker::elapsed_time(): never called. PV output uses tm.elapsed_time() (TimeManager) directly. (removed callers on 25361e5)
MovePicker::begin()/end(): unused private accessors. (removed callers on 8c2d21f)
closes https://github.com/official-stockfish/Stockfish/pull/6909
No functional change
