devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity
Files
gpu-dense-solvers
eigen/ci/build.linux.gitlab-ci.yml
Rasmus Munk Larsen 58c44ef36d GPU: Add library dispatch module (DeviceMatrix, cuBLAS, cuSOLVER) 
Add Eigen/GPU module: A standalone GPU library dispatch layer where
DeviceMatrix<Scalar> operations map 1:1 to cuBLAS/cuSOLVER calls.
CPU and GPU solvers coexist in the same binary with compatible syntax.

Core infrastructure:
- DeviceMatrix<Scalar>: RAII dense column-major GPU memory wrapper with
  async host transfer (fromHost/toHost) and CUDA event-based cross-stream
  synchronization.
- GpuContext: Unified execution context owning a CUDA stream + cuBLAS
  handle + cuSOLVER handle. Thread-local default with explicit override
  via setThreadLocal(). Stream-borrowing constructor for integration.
- DeviceBuffer: Typed RAII device allocation with move semantics.

cuBLAS dispatch (expression syntax):
- GEMM: d_C = d_A.adjoint() * d_B (cublasXgemm)
- TRSM: d_X = d_A.triangularView<Lower>().solve(d_B) (cublasXtrsm)
- SYMM/HEMM: d_C = d_A.selfadjointView<Lower>() * d_B (cublasXsymm)
- SYRK/HERK: d_C = d_A * d_A.adjoint() (cublasXsyrk)

cuSOLVER dispatch:
- GpuLLT: Cached Cholesky factorization (cusolverDnXpotrf + Xpotrs)
- GpuLU: Cached LU factorization (cusolverDnXgetrf + Xgetrs)
- Solver chaining: auto x = d_A.llt().solve(d_B)
- Solver expressions with .device(ctx) for explicit stream control.

CI: Bump CUDA container to Ubuntu 22.04 (CMake 3.22), GCC 10->11,
Clang 12->14. Bump cmake_minimum_required to 3.17 for FindCUDAToolkit.

Tests: gpu_cublas.cpp, gpu_cusolver_llt.cpp, gpu_cusolver_lu.cpp,
gpu_device_matrix.cpp, gpu_library_example.cu
Benchmarks: bench_gpu_solvers.cpp, bench_gpu_chaining.cpp,
bench_gpu_batching.cpp
2026-04-09 19:05:25 -07:00

392 lines
14 KiB
YAML
Raw Permalink Blame History

# Base configuration for linux cross-compilation.
.build:linux:cross:
needs: []
extends: .common:linux:cross
stage: build
variables:
EIGEN_CI_BUILD_TARGET: buildtests
script:
- . ci/scripts/build.linux.script.sh
tags:
- saas-linux-2xlarge-amd64
rules:
- if: $CI_PIPELINE_SOURCE == "schedule" && $CI_PROJECT_NAMESPACE == "libeigen"
- if: $CI_PIPELINE_SOURCE == "web" && $CI_PROJECT_NAMESPACE == "libeigen"
- if: $CI_PIPELINE_SOURCE == "merge_request_event" && $CI_PROJECT_NAMESPACE == "libeigen" && $CI_MERGE_REQUEST_LABELS =~ "/all-tests/"
cache:
key: "$CI_JOB_NAME_SLUG-$CI_COMMIT_REF_SLUG-BUILD"
paths:
- ${EIGEN_CI_BUILDDIR}/
######## x86-64 ################################################################
.build:linux:cross:x86-64:
extends: .build:linux:cross
variables:
EIGEN_CI_TARGET_ARCH: x86_64
EIGEN_CI_CROSS_TARGET_TRIPLE: x86_64-linux-gnu
# GCC-10 (stable recent version)
build:linux:cross:x86-64:gcc-10:default:
extends: .build:linux:cross:x86-64
variables:
EIGEN_CI_C_COMPILER: gcc-10
EIGEN_CI_CXX_COMPILER: g++-10
EIGEN_CI_CROSS_INSTALL: g++-10-x86-64-linux-gnu
EIGEN_CI_CROSS_C_COMPILER: x86_64-linux-gnu-gcc-10
EIGEN_CI_CROSS_CXX_COMPILER: x86_64-linux-gnu-g++-10
# GCC-13 (strict support for _Float16)
build:linux:cross:x86-64:gcc-13:default:
extends: .build:linux:cross:x86-64
variables:
EIGEN_CI_C_COMPILER: gcc-13
EIGEN_CI_CXX_COMPILER: g++-13
EIGEN_CI_CROSS_INSTALL: g++-13-x86-64-linux-gnu
EIGEN_CI_CROSS_C_COMPILER: x86_64-linux-gnu-gcc-13
EIGEN_CI_CROSS_CXX_COMPILER: x86_64-linux-gnu-g++-13
build:linux:cross:x86-64:gcc-10:avx:
extends: build:linux:cross:x86-64:gcc-10:default
variables:
EIGEN_CI_ADDITIONAL_ARGS: "-DEIGEN_TEST_AVX=on"
build:linux:cross:x86-64:gcc-10:avx2:
extends: build:linux:cross:x86-64:gcc-10:default
variables:
EIGEN_CI_ADDITIONAL_ARGS: "-DEIGEN_TEST_AVX2=on"
build:linux:cross:x86-64:gcc-10:avx512dq:
extends: build:linux:cross:x86-64:gcc-10:default
variables:
EIGEN_CI_ADDITIONAL_ARGS: "-DEIGEN_TEST_AVX512DQ=on"
build:linux:cross:x86-64:gcc-13:avx512fp16:
extends: build:linux:cross:x86-64:gcc-13:default
variables:
EIGEN_CI_ADDITIONAL_ARGS: "-DEIGEN_TEST_AVX512FP16=on -DEIGEN_TEST_AVX512DQ=on -DEIGEN_TEST_F16C=on"
# Clang-14 (stable recent version)
build:linux:cross:x86-64:clang-14:default:
extends: .build:linux:cross:x86-64
variables:
EIGEN_CI_INSTALL: clang-14
EIGEN_CI_C_COMPILER: clang-14
EIGEN_CI_CXX_COMPILER: clang++-14
EIGEN_CI_CROSS_INSTALL: g++-10-x86-64-linux-gnu clang-14
build:linux:cross:x86-64:clang-14:avx2:
extends: build:linux:cross:x86-64:clang-14:default
variables:
EIGEN_CI_ADDITIONAL_ARGS: "-DEIGEN_TEST_AVX2=on"
build:linux:cross:x86-64:clang-14:avx512dq:
extends: build:linux:cross:x86-64:clang-14:default
variables:
EIGEN_CI_ADDITIONAL_ARGS: "-DEIGEN_TEST_AVX512DQ=on"
# Generic vector extension backend.
build:linux:cross:x86-64:clang-19:generic:avx512dq:
extends: build:linux:cross:x86-64:clang-14:default
variables:
EIGEN_CI_INSTALL: clang-19
EIGEN_CI_C_COMPILER: clang-19
EIGEN_CI_CXX_COMPILER: clang++-19
EIGEN_CI_CROSS_INSTALL: g++-14-x86-64-linux-gnu clang-19
EIGEN_CI_ADDITIONAL_ARGS: >
-DEIGEN_TEST_CUSTOM_CXX_FLAGS=-mfma;-mavx512dq;-DEIGEN_VECTORIZE_GENERIC=1
# Nightly full test suite in C++20 mode.
build:linux:cross:x86-64:clang-19:cxx20:nightly:
extends: .build:linux:cross:x86-64
variables:
EIGEN_CI_INSTALL: clang-19
EIGEN_CI_C_COMPILER: clang-19
EIGEN_CI_CXX_COMPILER: clang++-19
EIGEN_CI_CROSS_INSTALL: g++-14-x86-64-linux-gnu clang-19
EIGEN_CI_ADDITIONAL_ARGS: -DCMAKE_CXX_STANDARD=20
rules:
- if: $CI_PIPELINE_SOURCE == "schedule" && $CI_PROJECT_NAMESPACE == "libeigen" && $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
- if: $CI_PIPELINE_SOURCE == "web" && $CI_PROJECT_NAMESPACE == "libeigen"
- if: $CI_PIPELINE_SOURCE == "merge_request_event" && $CI_PROJECT_NAMESPACE == "libeigen" && $CI_MERGE_REQUEST_LABELS =~ "/all-tests/"
build:linux:docs:
extends: .build:linux:cross
variables:
EIGEN_CI_TARGET_ARCH: any
EIGEN_CI_BUILD_TARGET: doc
EIGEN_CI_INSTALL: ca-certificates clang flex python3 bison graphviz
EIGEN_CI_C_COMPILER: clang
EIGEN_CI_CXX_COMPILER: clang++
EIGEN_CI_BEFORE_SCRIPT: ". ci/scripts/build_and_install_doxygen.sh Release_1_13_2"
tags:
- saas-linux-medium-amd64
rules:
- if: $CI_PIPELINE_SOURCE == "schedule" && $CI_PROJECT_NAMESPACE == "libeigen"
- if: $CI_PIPELINE_SOURCE == "web" && $CI_PROJECT_NAMESPACE == "libeigen"
- if: $CI_PIPELINE_SOURCE == "push" && $CI_PROJECT_NAMESPACE == "libeigen" && $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
- if: $CI_PIPELINE_SOURCE == "merge_request_event" && $CI_PROJECT_NAMESPACE == "libeigen" && $CI_MERGE_REQUEST_LABELS =~ "/all-tests/"
######## Sanitizers ############################################################
# ASAN/UBSAN builds are split into official/unsupported because the
# instrumented executables are too large to upload as a single artifact.
.build:linux:cross:x86-64:clang-19:default:asan-ubsan:
extends: .build:linux:cross:x86-64
# Disable cache for ASAN builds: instrumented .o files are so large that
# downloading the cache can take longer than a clean build from scratch.
cache: []
timeout: 2h
variables:
EIGEN_CI_INSTALL: clang-19 libclang-rt-19-dev
EIGEN_CI_C_COMPILER: clang-19
EIGEN_CI_CXX_COMPILER: clang++-19
EIGEN_CI_CROSS_INSTALL: g++-14-x86-64-linux-gnu clang-19 libclang-rt-19-dev
EIGEN_CI_ADDITIONAL_ARGS: >-
-DEIGEN_TEST_CUSTOM_CXX_FLAGS=-fsanitize=address,undefined;-fno-omit-frame-pointer;-fno-sanitize-recover=undefined
-DEIGEN_TEST_CUSTOM_LINKER_FLAGS=-fsanitize=address,undefined
build:linux:cross:x86-64:clang-19:default:asan-ubsan:official:
extends: .build:linux:cross:x86-64:clang-19:default:asan-ubsan
variables:
EIGEN_CI_BUILD_TARGET: BuildOfficial
build:linux:cross:x86-64:clang-19:default:asan-ubsan:unsupported:
extends: .build:linux:cross:x86-64:clang-19:default:asan-ubsan
variables:
EIGEN_CI_BUILD_TARGET: BuildUnsupported
######## NVHPC #################################################################
# NVHPC (nvc++) uses NVIDIA's HPC SDK container image with the compilers
# pre-installed. We override EIGEN_CI_INSTALL to avoid trying to apt-get
# install the compiler.
# Split into official/unsupported because nvc++ is so slow that the full
# build exceeds the 3-hour GitLab SaaS shared-runner timeout.
.build:linux:x86-64:nvhpc-26.1:
extends: .build:linux:cross:x86-64
image: nvcr.io/nvidia/nvhpc:26.1-devel-cuda13.1-ubuntu24.04
variables:
EIGEN_CI_C_COMPILER: nvc
EIGEN_CI_CXX_COMPILER: nvc++
EIGEN_CI_INSTALL: ""
EIGEN_CI_CROSS_INSTALL: ""
# NVHPC's compiler frontend is very memory-hungry with template-heavy code.
# The 2xlarge runner has 128 GB, so 16 jobs gives ~8 GB per process.
# The shuffled-batch build strategy spreads memory-hungry targets across
# batches, preventing simultaneous OOM-prone compilations.
EIGEN_CI_BUILD_JOBS: "16"
EIGEN_CI_FALLBACK_JOBS: "4"
build:linux:x86-64:nvhpc-26.1:default:official:
extends: .build:linux:x86-64:nvhpc-26.1
variables:
EIGEN_CI_BUILD_TARGET: BuildOfficial
build:linux:x86-64:nvhpc-26.1:default:unsupported:
extends: .build:linux:x86-64:nvhpc-26.1
variables:
EIGEN_CI_BUILD_TARGET: BuildUnsupported
######## CUDA ##################################################################
.build:linux:cuda:
extends: .build:linux:cross:x86-64
variables:
# Additional flags passed to the cuda compiler.
EIGEN_CI_CUDA_CXX_FLAGS: ""
# Compute architectures present in the GitLab CI runners.
EIGEN_CI_CUDA_COMPUTE_ARCH: "70;75"
EIGEN_CI_BUILD_TARGET: buildtests_gpu
EIGEN_CI_TEST_CUDA_CLANG: "off"
EIGEN_CI_TEST_CUDA_NVC: "off"
EIGEN_CI_ADDITIONAL_ARGS:
-DEIGEN_TEST_CUDA=on
-DEIGEN_CUDA_CXX_FLAGS=${EIGEN_CI_CUDA_CXX_FLAGS}
-DEIGEN_CUDA_COMPUTE_ARCH=${EIGEN_CI_CUDA_COMPUTE_ARCH}
-DEIGEN_TEST_CUDA_CLANG=${EIGEN_CI_TEST_CUDA_CLANG}
-DEIGEN_TEST_CUDA_NVC=${EIGEN_CI_TEST_CUDA_NVC}
tags:
# Build on regular linux to limit GPU cost.
- saas-linux-2xlarge-amd64
# GCC-11, CUDA-12.2
build:linux:cuda-12.2:gcc-11:
extends: .build:linux:cuda
image: nvidia/cuda:12.2.0-devel-ubuntu22.04
variables:
EIGEN_CI_C_COMPILER: gcc-11
EIGEN_CI_CXX_COMPILER: g++-11
# Clang-14, CUDA-12.2
build:linux:cuda-12.2:clang-14:
extends: build:linux:cuda-12.2:gcc-11
variables:
EIGEN_CI_C_COMPILER: clang-14
EIGEN_CI_CXX_COMPILER: clang++-14
EIGEN_CI_TEST_CUDA_CLANG: "on"
######### HIP ##################################################################
# Note: these are currently build-only, until we get an AMD-supported runner.
# ROCm HIP
build:linux:rocm-latest:gcc-10:
extends: .build:linux:cross
image: rocm/dev-ubuntu-24.04:6.3.1
variables:
EIGEN_CI_C_COMPILER: gcc-10
EIGEN_CI_CXX_COMPILER: g++-10
EIGEN_CI_BUILD_TARGET: buildtests_gpu
EIGEN_CI_ADDITIONAL_ARGS: -DEIGEN_TEST_HIP=on
cache: [] # Disable cache for ROCm, since it fails whenever the image updates.
######## Arm ###################################################################
.build:linux:cross:arm:
extends: .build:linux:cross
variables:
EIGEN_CI_TARGET_ARCH: arm
EIGEN_CI_CROSS_TARGET_TRIPLE: arm-linux-gnueabihf
EIGEN_CI_ADDITIONAL_ARGS: >
-DEIGEN_TEST_CUSTOM_CXX_FLAGS=-march=armv7-a;-mfpu=neon-vfpv4
-DCMAKE_SYSTEM_NAME=Linux
-DCMAKE_CROSSCOMPILING_EMULATOR=qemu-arm-static;-L;/usr/arm-linux-gnueabihf
build:linux:cross:arm:gcc-10:default:
extends: .build:linux:cross:arm
variables:
EIGEN_CI_CROSS_INSTALL: g++-10-arm-linux-gnueabihf qemu-user-static
EIGEN_CI_CROSS_C_COMPILER: arm-linux-gnueabihf-gcc-10
EIGEN_CI_CROSS_CXX_COMPILER: arm-linux-gnueabihf-g++-10
build:linux:cross:arm:clang-14:default:
extends: .build:linux:cross:arm
variables:
EIGEN_CI_INSTALL: clang-14
EIGEN_CI_C_COMPILER: clang-14
EIGEN_CI_CXX_COMPILER: clang++-14
EIGEN_CI_CROSS_INSTALL: g++-10-arm-linux-gnueabihf clang-14 qemu-user-static
######## aarch64 ###############################################################
.build:linux:cross:aarch64:
extends: .build:linux:cross
variables:
EIGEN_CI_TARGET_ARCH: aarch64
EIGEN_CI_CROSS_TARGET_TRIPLE: aarch64-linux-gnu
EIGEN_CI_ADDITIONAL_ARGS: -DEIGEN_TEST_CUSTOM_CXX_FLAGS=-march=armv8.2-a+fp16
tags:
- saas-linux-large-arm64
build:linux:cross:aarch64:gcc-10:default:
extends: .build:linux:cross:aarch64
variables:
EIGEN_CI_C_COMPILER: gcc-10
EIGEN_CI_CXX_COMPILER: g++-10
EIGEN_CI_CROSS_INSTALL: g++-10-aarch64-linux-gnu
EIGEN_CI_CROSS_C_COMPILER: aarch64-linux-gnu-gcc-10
EIGEN_CI_CROSS_CXX_COMPILER: aarch64-linux-gnu-g++-10
build:linux:cross:aarch64:clang-14:default:
extends: .build:linux:cross:aarch64
variables:
EIGEN_CI_INSTALL: clang-14
EIGEN_CI_C_COMPILER: clang-14
EIGEN_CI_CXX_COMPILER: clang++-14
EIGEN_CI_CROSS_INSTALL: g++-10-aarch64-linux-gnu clang-14
######## ppc64le ###############################################################
.build:linux:cross:ppc64le:
extends: .build:linux:cross
variables:
EIGEN_CI_TARGET_ARCH: ppc64le
EIGEN_CI_CROSS_TARGET_TRIPLE: powerpc64le-linux-gnu
EIGEN_CI_ADDITIONAL_ARGS: >-
-DCMAKE_SYSTEM_NAME=Linux
-DCMAKE_CROSSCOMPILING_EMULATOR=qemu-ppc64le-static;-L;/usr/powerpc64le-linux-gnu
build:linux:cross:ppc64le:gcc-14:default:
extends: .build:linux:cross:ppc64le
variables:
EIGEN_CI_CROSS_INSTALL: g++-14-powerpc64le-linux-gnu qemu-user-static
EIGEN_CI_CROSS_C_COMPILER: powerpc64le-linux-gnu-gcc-14
EIGEN_CI_CROSS_CXX_COMPILER: powerpc64le-linux-gnu-g++-14
build:linux:cross:ppc64le:clang-16:default:
extends: .build:linux:cross:ppc64le
variables:
EIGEN_CI_C_COMPILER: clang-16
EIGEN_CI_CXX_COMPILER: clang++-16
EIGEN_CI_CROSS_INSTALL: g++-14-powerpc64le-linux-gnu clang-16 qemu-user-static
######## loongarch64 #################################################
.build:linux:cross:loongarch64:
extends: .build:linux:cross
variables:
EIGEN_CI_TARGET_ARCH: loongarch64
EIGEN_CI_CROSS_TARGET_TRIPLE: loongarch64-linux-gnu
# GCC-14 (minimum on Ubuntu 24)
build:linux:cross:loongarch64:gcc-14:default:
extends: .build:linux:cross:loongarch64
variables:
EIGEN_CI_CROSS_INSTALL: g++-14-loongarch64-linux-gnu qemu-user-static
EIGEN_CI_CROSS_C_COMPILER: loongarch64-linux-gnu-gcc-14
EIGEN_CI_CROSS_CXX_COMPILER: loongarch64-linux-gnu-g++-14
EIGEN_CI_ADDITIONAL_ARGS: >-
-DCMAKE_SYSTEM_NAME=Linux
-DCMAKE_CROSSCOMPILING_EMULATOR=qemu-loongarch64-static;-L;/usr/loongarch64-linux-gnu
-DEIGEN_TEST_LSX=on
######## MR Smoke Tests ########################################################
build:linux:cross:x86-64:gcc-10:default:smoketest:
extends: build:linux:cross:x86-64:gcc-10:default
variables:
EIGEN_CI_BUILD_TARGET: buildsmoketests
rules:
- if: $CI_PIPELINE_SOURCE == "merge_request_event"
tags:
- saas-linux-medium-amd64
build:linux:cross:x86-64:clang-14:default:smoketest:
extends: build:linux:cross:x86-64:clang-14:default
variables:
EIGEN_CI_BUILD_TARGET: buildsmoketests
rules:
- if: $CI_PIPELINE_SOURCE == "merge_request_event"
tags:
- saas-linux-medium-amd64
build:linux:aarch64:gcc-10:default:smoketest:
extends: build:linux:cross:aarch64:gcc-10:default
variables:
EIGEN_CI_BUILD_TARGET: buildsmoketests
rules:
- if: $CI_PIPELINE_SOURCE == "merge_request_event"
tags:
- saas-linux-medium-arm64
######## Sanitizer Smoke Tests #################################################
build:linux:cross:x86-64:clang-14:sanitizer:smoketest:
extends: build:linux:cross:x86-64:clang-14:default
variables:
EIGEN_CI_BUILD_TARGET: buildsmoketests
EIGEN_CI_INSTALL: clang-14 libclang-rt-14-dev
EIGEN_CI_CROSS_INSTALL: g++-10-x86-64-linux-gnu clang-14 libclang-rt-14-dev
EIGEN_CI_ADDITIONAL_ARGS: >-
-DEIGEN_TEST_CUSTOM_CXX_FLAGS=-fsanitize=address,undefined;-fno-sanitize-recover=address;-fno-omit-frame-pointer
-DEIGEN_TEST_CUSTOM_LINKER_FLAGS=-fsanitize=address,undefined
rules:
- if: $CI_PIPELINE_SOURCE == "merge_request_event"
tags:
- saas-linux-large-amd64
allow_failure: true
timeout: 30m
Reference in New Issue View Git Blame Copy Permalink