jirayu_sri/stockfish
1
0
Fork 0
mirror of https://github.com/official-stockfish/Stockfish.git synced 2026-07-22 20:57:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add intrinsics for LSX and LASX

Browse Source 
This adds actual intrinsics of SIMD on loongarch64, with considerable performance improvements.

Benchmark on `Loongson-3A6000-HV Not Specified CPU @ 2.5GHz`

Baseline GCC 14 (debian trixie):
```
===========================
Total time (ms) : 17816
Nodes searched  : 2336177
Nodes/second    : 131128
```

With this PR:

GCC 16 (debian sid):
```
./stockfish compiler
Stockfish dev-20260516-52e8d9ef by the Stockfish developers (see AUTHORS file)

Compiled by                : g++ (GNUC) 16.1.0 on Linux
Compilation architecture   : loongarch64-lasx
Compilation settings       : 64bit LASX LSX
Compiler __VERSION__ macro : 16.1.0
===========================
Total time (ms) : 3518
Nodes searched  : 2336177
Nodes/second    : 664063
```
```
./stockfish compiler
Stockfish dev-20260516-52e8d9ef by the Stockfish developers (see AUTHORS file)

Compiled by                : g++ (GNUC) 16.1.0 on Linux
Compilation architecture   : loongarch64-lsx
Compilation settings       : 64bit LSX
Compiler __VERSION__ macro : 16.1.0
===========================
Total time (ms) : 4944
Nodes searched  : 2336177
Nodes/second    : 472527
```

clang 22 (debian sid):
```
./stockfish compiler
Stockfish dev-20260516-52e8d9ef by the Stockfish developers (see AUTHORS file)

Compiled by                : clang++ 22.1.5 on Linux
Compilation architecture   : loongarch64-lasx
Compilation settings       : 64bit LASX LSX
Compiler __VERSION__ macro : Debian Clang 22.1.5 (1)
===========================
Total time (ms) : 3401
Nodes searched  : 2336177
Nodes/second    : 686908
```
```
./stockfish compiler
Stockfish dev-20260516-52e8d9ef by the Stockfish developers (see AUTHORS file)

Compiled by                : clang++ 22.1.5 on Linux
Compilation architecture   : loongarch64-lsx
Compilation settings       : 64bit LSX
Compiler __VERSION__ macro : Debian Clang 22.1.5 (1)
===========================
Total time (ms) : 5501
Nodes searched  : 2336177
Nodes/second    : 424682
```

GCC 14 (debian trixie):
```
./stockfish compiler
Stockfish dev-20260516-52e8d9ef by the Stockfish developers (see AUTHORS file)

Compiled by                : g++ (GNUC) 14.2.0 on Linux
Compilation architecture   : loongarch64-lasx
Compilation settings       : 64bit LASX LSX
Compiler __VERSION__ macro : 14.2.0
===========================
Total time (ms) : 3559
Nodes searched  : 2336177
Nodes/second    : 656413
```
```
./stockfish compiler
Stockfish dev-20260516-52e8d9ef by the Stockfish developers (see AUTHORS file)

Compiled by                : g++ (GNUC) 14.2.0 on Linux
Compilation architecture   : loongarch64-lsx
Compilation settings       : 64bit LSX
Compiler __VERSION__ macro : 14.2.0
===========================
Total time (ms) : 5212
Nodes searched  : 2336177
Nodes/second    : 448230
```

clang 19 (debian trixie):
```
./stockfish compiler
Stockfish dev-20260516-52e8d9ef by the Stockfish developers (see AUTHORS file)

Compiled by                : clang++ 19.1.7 on Linux
Compilation architecture   : loongarch64-lasx
Compilation settings       : 64bit LASX LSX
Compiler __VERSION__ macro : Debian Clang 19.1.7 (3+b1)
===========================
Total time (ms) : 4830
Nodes searched  : 2336177
Nodes/second    : 483680
```
```
./stockfish compiler
Stockfish dev-20260516-52e8d9ef by the Stockfish developers (see AUTHORS file)

Compiled by                : clang++ 19.1.7 on Linux
Compilation architecture   : loongarch64-lsx
Compilation settings       : 64bit LSX
Compiler __VERSION__ macro : Debian Clang 19.1.7 (3+b1)
===========================
Total time (ms) : 5568
Nodes searched  : 2336177
Nodes/second    : 419572
```

closes https://github.com/official-stockfish/Stockfish/pull/6815

No functional change
This commit is contained in:
Wencey Wang
2026-05-17 20:42:36 +02:00
committed by Joost VandeVondele
parent 0366d092a8
commit ffffe1bdb5
11 changed files with 385 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/nnue/nnue_accumulator.cpp
+35
View File
@@ -516,6 +516,41 @@ Bitboard get_changed_pieces(const std::array<Piece, SQUARE_NB>& oldPieces,
sameBB |= static_cast<Bitboard>(equalMask) << i;
}
return ~sameBB;
#elif defined(USE_LASX)
static_assert(sizeof(Piece) == 1);
Bitboard changed = 0;
for (int i = 0; i < 64; i += 32)
{
const __m256i old_v = __lasx_xvld(reinterpret_cast<const void*>(&oldPieces[i]), 0);
const __m256i new_v = __lasx_xvld(reinterpret_cast<const void*>(&newPieces[i]), 0);
const __m256i diff = __lasx_xvxor_v(old_v, new_v);
const __m256i mask = __lasx_xvmsknz_b(diff);
const auto lo = static_cast<std::uint16_t>(__lasx_xvpickve2gr_d(mask, 0));
const auto hi = static_cast<std::uint16_t>(__lasx_xvpickve2gr_d(mask, 2));
changed |= (static_cast<Bitboard>(lo) | (static_cast<Bitboard>(hi) << 16)) << i;
}
return changed;
#elif defined(USE_LSX)
static_assert(sizeof(Piece) == 1);
Bitboard changed = 0;
for (int i = 0; i < 64; i += 16)
{
const __m128i old_v = __lsx_vld(reinterpret_cast<const void*>(&oldPieces[i]), 0);
const __m128i new_v = __lsx_vld(reinterpret_cast<const void*>(&newPieces[i]), 0);
const __m128i diff = __lsx_vxor_v(old_v, new_v);
const __m128i mask = __lsx_vmsknz_b(diff);
changed |= static_cast<Bitboard>(static_cast<std::uint16_t>(__lsx_vpickve2gr_d(mask, 0)))
<< i;
}
return changed;
#elif defined(USE_NEON)
uint8x16x4_t old_v = vld4q_u8(reinterpret_cast<const uint8_t*>(oldPieces.data()));
uint8x16x4_t new_v = vld4q_u8(reinterpret_cast<const uint8_t*>(newPieces.data()));