Files
anematodeandJoost VandeVondele d5bbc6b67d [RfC] RISC-V port and universal binary
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
2026-07-03 20:29:50 +02:00

406 lines
13 KiB
YAML

name: Tests
on:
workflow_call:
jobs:
Test-Targets:
name: ${{ matrix.config.name }}
runs-on: ${{ matrix.config.os }}
env:
COMPCXX: ${{ matrix.config.compiler }}
COMP: ${{ matrix.config.comp }}
CXXFLAGS: "-Werror"
strategy:
fail-fast: false
matrix:
config:
- name: Ubuntu 22.04 GCC
os: ubuntu-22.04
compiler: g++
comp: gcc
run_32bit_tests: true
run_64bit_tests: true
shell: bash
- name: Ubuntu 22.04 Clang
os: ubuntu-22.04
compiler: clang++
comp: clang
run_32bit_tests: true
run_64bit_tests: true
shell: bash
- name: Android NDK aarch64
os: ubuntu-22.04
compiler: aarch64-linux-android29-clang++
comp: ndk
run_armv8_tests: true
shell: bash
- name: Android NDK arm
os: ubuntu-22.04
compiler: armv7a-linux-androideabi29-clang++
comp: ndk
run_armv7_tests: true
shell: bash
- name: Linux GCC ppc64
os: ubuntu-22.04
compiler: g++
comp: gcc
run_ppc64_tests: true
base_image: "ppc64le/alpine:latest"
platform: linux/ppc64le
shell: bash
- name: Linux GCC loongarch64
os: ubuntu-22.04
compiler: g++
comp: gcc
run_loongarch64_tests: true
base_image: "loongarch64/alpine:3.21"
platform: linux/loong64
shell: bash
- name: macOS 15 Apple Clang
os: macos-15-intel
compiler: clang++
comp: clang
run_64bit_tests: true
shell: bash
- name: macOS 15 Apple Clang M1
os: macos-15
compiler: clang++
comp: clang
run_64bit_tests: false
run_m1_tests: true
shell: bash
- name: macOS 15 GCC 11
os: macos-15-intel
compiler: g++-11
comp: gcc
run_64bit_tests: true
shell: bash
- name: Windows 2022 Mingw-w64 GCC x86_64
os: windows-2022
compiler: g++
comp: mingw
run_64bit_tests: true
msys_sys: mingw64
msys_env: x86_64-gcc
shell: msys2 {0}
- name: Windows 2022 Mingw-w64 GCC i686
os: windows-2022
compiler: g++
comp: mingw
run_32bit_tests: true
msys_sys: mingw32
msys_env: i686-gcc
shell: msys2 {0}
- name: Windows 2022 Mingw-w64 Clang x86_64
os: windows-2022
compiler: clang++
comp: clang
run_64bit_tests: true
msys_sys: clang64
msys_env: clang-x86_64-clang
shell: msys2 {0}
- name: Windows 11 Mingw-w64 Clang arm64
os: windows-11-arm
compiler: clang++
comp: clang
run_armv8_tests: true
msys_sys: clangarm64
msys_env: clang-aarch64-clang
shell: msys2 {0}
defaults:
run:
working-directory: src
shell: ${{ matrix.config.shell }}
steps:
- uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
with:
fetch-depth: 0
persist-credentials: false
- name: Download required linux packages
if: runner.os == 'Linux'
run: |
sudo apt update
sudo apt install expect valgrind g++-multilib qemu-user-static
- name: Install NDK
if: runner.os == 'Linux'
run: |
if [ $COMP == ndk ]; then
NDKV="27.2.12479018"
ANDROID_ROOT=/usr/local/lib/android
ANDROID_SDK_ROOT=$ANDROID_ROOT/sdk
SDKMANAGER=$ANDROID_SDK_ROOT/cmdline-tools/latest/bin/sdkmanager
echo "y" | $SDKMANAGER "ndk;$NDKV"
ANDROID_NDK_ROOT=$ANDROID_SDK_ROOT/ndk/$NDKV
ANDROID_NDK_BIN=$ANDROID_NDK_ROOT/toolchains/llvm/prebuilt/linux-x86_64/bin
echo "ANDROID_NDK_BIN=$ANDROID_NDK_BIN" >> $GITHUB_ENV
fi
- name: Set up QEMU
if: matrix.config.base_image
uses: docker/setup-qemu-action@ce360397dd3f832beb865e1373c09c0e9f86d70a # v4.0.0
- name: Set up Docker Buildx
if: matrix.config.base_image
uses: docker/setup-buildx-action@4d04d5d9486b7bd6fa91e7baf45bbb4f8b9deedd # v4.0.0
- name: Build Docker container
if: matrix.config.base_image
run: |
docker buildx build --platform ${{ matrix.config.platform }} --load -t sf_builder - << EOF
FROM ${{ matrix.config.base_image }}
WORKDIR /app
RUN apk update && apk add make g++
CMD ["sh", "src/script.sh"]
EOF
- name: Download required macOS packages
if: runner.os == 'macOS'
run: brew install coreutils
- name: Install GCC on macOS
if: runner.os == 'macOS' && matrix.config.comp == 'gcc'
run: |
brew install gcc@11 || true
g++-11 --version
- name: Setup msys and install required packages
if: runner.os == 'Windows'
uses: msys2/setup-msys2@e9898307ac31d1a803454791be09ab9973336e1c # v2.31.1
with:
msystem: ${{ matrix.config.msys_sys }}
install: mingw-w64-${{ matrix.config.msys_env }} make git expect
- name: Download the used network from the fishtest framework
run: make net
- name: Extract the bench number from the commit history
run: |
for hash in $(git rev-list -100 HEAD); do
benchref=$(git show -s $hash | tac | grep -m 1 -o -x '[[:space:]]*\b[Bb]ench[ :]\+[1-9][0-9]\{5,7\}\b[[:space:]]*' | sed 's/[^0-9]//g') && break || true
done
[[ -n "$benchref" ]] && echo "benchref=$benchref" >> $GITHUB_ENV && echo "From commit: $hash" && echo "Reference bench: $benchref" || echo "No bench found"
- name: Check compiler
run: |
if [ -z "${{ matrix.config.base_image }}" ]; then
if [ $COMP == ndk ]; then
export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
fi
$COMPCXX -v
else
echo "$COMPCXX -v" > script.sh
docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}:/app sf_builder
fi
- name: Test help target
run: make help
- name: Check git
run: git --version
# x86-32 tests
- name: Test debug x86-32 build
if: matrix.config.run_32bit_tests
run: |
export CXXFLAGS="-Werror -D_GLIBCXX_DEBUG"
make clean
make -j4 ARCH=x86-32 optimize=no debug=yes build
../tests/signature.sh $benchref
- name: Test x86-32 build
if: matrix.config.run_32bit_tests
run: |
make clean
make -j4 ARCH=x86-32 build
../tests/signature.sh $benchref
- name: Test x86-32-sse41-popcnt build
if: matrix.config.run_32bit_tests
run: |
make clean
make -j4 ARCH=x86-32-sse41-popcnt build
../tests/signature.sh $benchref
- name: Test x86-32-sse2 build
if: matrix.config.run_32bit_tests
run: |
make clean
make -j4 ARCH=x86-32-sse2 build
../tests/signature.sh $benchref
- name: Test general-32 build
if: matrix.config.run_32bit_tests
run: |
make clean
make -j4 ARCH=general-32 build
../tests/signature.sh $benchref
# x86-64 tests
- name: Test debug x86-64-avx2 build
if: matrix.config.run_64bit_tests
run: |
export CXXFLAGS="-Werror -D_GLIBCXX_DEBUG"
make clean
make -j4 ARCH=x86-64-avx2 optimize=no debug=yes build
../tests/signature.sh $benchref
- name: Test x86-64-bmi2 build
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=x86-64-bmi2 build
../tests/signature.sh $benchref
- name: Test x86-64-avx2 build
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=x86-64-avx2 build
../tests/signature.sh $benchref
# Test a deprecated arch
- name: Test x86-64-modern build
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=x86-64-modern build
../tests/signature.sh $benchref
- name: Test x86-64-sse41-popcnt build
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=x86-64-sse41-popcnt build
../tests/signature.sh $benchref
- name: Test x86-64-ssse3 build
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=x86-64-ssse3 build
../tests/signature.sh $benchref
- name: Test x86-64-sse3-popcnt build
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=x86-64-sse3-popcnt build
../tests/signature.sh $benchref
- name: Test x86-64 build
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=x86-64 build
../tests/signature.sh $benchref
- name: Test general-64 build
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=general-64 build
../tests/signature.sh $benchref
- name: Test apple-silicon build
if: matrix.config.run_m1_tests
run: |
make clean
make -j4 ARCH=apple-silicon build
../tests/signature.sh $benchref
- name: Test apple-silicon build with system g++ (macOS g++ == clang++)
if: matrix.config.run_m1_tests
run: |
make clean
env -u COMPCXX make -j4 ARCH=apple-silicon build COMP=gcc
../tests/signature.sh $benchref
# armv8 tests
- name: Test armv8 build
if: matrix.config.run_armv8_tests
run: |
if [ $COMP == ndk ]; then
export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
export LDFLAGS="-static -Wno-unused-command-line-argument"
fi
make clean
make -j4 ARCH=armv8 build
../tests/signature.sh $benchref
- name: Test armv8-dotprod build
if: matrix.config.run_armv8_tests
run: |
if [ $COMP == ndk ]; then
export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
export LDFLAGS="-static -Wno-unused-command-line-argument"
fi
make clean
make -j4 ARCH=armv8-dotprod build
../tests/signature.sh $benchref
# armv7 tests
- name: Test armv7 build
if: matrix.config.run_armv7_tests
run: |
export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
export LDFLAGS="-static -Wno-unused-command-line-argument"
make clean
make -j4 ARCH=armv7 build
../tests/signature.sh $benchref
- name: Test armv7-neon build
if: matrix.config.run_armv7_tests
run: |
export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
export LDFLAGS="-static -Wno-unused-command-line-argument"
make clean
make -j4 ARCH=armv7-neon build
../tests/signature.sh $benchref
# ppc64 tests
- name: Test ppc64 build
if: matrix.config.run_ppc64_tests
run: |
echo "cd src && export LDFLAGS='-static' && make clean && make -j4 ARCH=ppc-64 build" > script.sh
docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}:/app sf_builder
../tests/signature.sh $benchref
# loongarch64 tests
- name: Test loongarch64 build
if: matrix.config.run_loongarch64_tests
run: |
echo "cd src && export LDFLAGS='-static' && make clean && make -j4 ARCH=loongarch64 build" > script.sh
docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}:/app sf_builder
../tests/signature.sh $benchref
- name: Test loongarch64-lsx build
if: matrix.config.run_loongarch64_tests
run: |
echo "cd src && export LDFLAGS='-static' && make clean && make -j4 ARCH=loongarch64-lsx build" > script.sh
docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}:/app sf_builder
../tests/signature.sh $benchref
- name: Test loongarch64-lasx build
if: matrix.config.run_loongarch64_tests
run: |
echo "cd src && export LDFLAGS='-static' && make clean && make -j4 ARCH=loongarch64-lasx build" > script.sh
docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}:/app sf_builder
../tests/signature.sh $benchref
# Other tests
- name: Check perft and search reproducibility
if: matrix.config.run_64bit_tests
run: |
make clean
make -j4 ARCH=x86-64-avx2 build
../tests/perft.sh
../tests/reprosearch.sh