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

GPU: Add BLAS-1 ops, DeviceScalar, device-resident SpMV, and CG interop (5/5)

Browse Source 
Add the operator interface needed for GPU iterative solvers:

- BLAS Level-1 on DeviceMatrix: dot(), norm(), squaredNorm(), setZero(),
  noalias(), operator+=/-=/\*= dispatching to cuBLAS axpy/scal/dot/nrm2.
- DeviceScalar<Scalar>: device-resident scalar returned by reductions.
  Defers host sync until value is read (implicit conversion). Device-side
  division via NPP for real types.
- GpuContext: stream-borrowing constructor, setThreadLocal(), cublasLtHandle(),
  cusparseHandle().
- GEMM upgraded from cublasGemmEx to cublasLtMatmul with heuristic algorithm
  selection and plan caching.
- GpuSparseContext: GpuContext& constructor for same-stream execution,
  deviceView() returning DeviceSparseView with operator* for device-resident
  SpMV (d_y = d_A * d_x).
- geam expressions: d_C = d_A + alpha * d_B via cublasXgeam.
- GpuSVD::matrixV() convenience wrapper.

These additions make DeviceMatrix usable as a VectorType in Eigen algorithm
templates. Conjugate gradient is the motivating example and is tested against
CPU ConjugateGradient for correctness.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Rasmus Munk Larsen
2026-04-09 19:54:13 -07:00
parent 43a95b62bb
commit 014f12f11a
22 changed files with 3363 additions and 151 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
test/CMakeLists.txt
View File

@@ -481,13 +481,13 @@ if(CUDA_FOUND AND EIGEN_TEST_CUDA)
ei_add_test(gpu_basic)
ei_add_test(gpu_library_example "" "CUDA::cusolver")
# DeviceMatrix tests: only CUDA runtime, no NVIDIA libraries.
# DeviceMatrix tests: CUDA runtime + cuBLAS + cuSOLVER (for BLAS-1 ops via GpuContext).
unset(EIGEN_ADD_TEST_FILENAME_EXTENSION)
add_executable(gpu_device_matrix gpu_device_matrix.cpp)
target_include_directories(gpu_device_matrix PRIVATE
"${CUDA_TOOLKIT_ROOT_DIR}/include"
"${CMAKE_CURRENT_BINARY_DIR}")
target_link_libraries(gpu_device_matrix Eigen3::Eigen CUDA::cudart)
target_link_libraries(gpu_device_matrix Eigen3::Eigen CUDA::cudart CUDA::cublas CUDA::cusolver CUDA::npps CUDA::nppc)
target_compile_definitions(gpu_device_matrix PRIVATE
EIGEN_TEST_MAX_SIZE=${EIGEN_TEST_MAX_SIZE}
EIGEN_TEST_PART_ALL=1)
@@ -575,7 +575,7 @@ if(CUDA_FOUND AND EIGEN_TEST_CUDA)
"${CUDA_TOOLKIT_ROOT_DIR}/include"
"${CMAKE_CURRENT_BINARY_DIR}")
target_link_libraries(gpu_cusparse_spmv
Eigen3::Eigen CUDA::cudart CUDA::cusparse)
Eigen3::Eigen CUDA::cudart CUDA::cusparse CUDA::cublas CUDA::cusolver)
target_compile_definitions(gpu_cusparse_spmv PRIVATE
EIGEN_TEST_MAX_SIZE=${EIGEN_TEST_MAX_SIZE}
EIGEN_TEST_PART_ALL=1)
@@ -584,6 +584,23 @@ if(CUDA_FOUND AND EIGEN_TEST_CUDA)
add_dependencies(buildtests_gpu gpu_cusparse_spmv)
set_property(TEST gpu_cusparse_spmv APPEND PROPERTY LABELS "Official;gpu")
set_property(TEST gpu_cusparse_spmv PROPERTY SKIP_RETURN_CODE 77)
# End-to-end GPU CG test: Eigen's ConjugateGradient with DeviceMatrix.
add_executable(gpu_cg gpu_cg.cpp)
target_include_directories(gpu_cg PRIVATE
"${CUDA_TOOLKIT_ROOT_DIR}/include"
"${CMAKE_CURRENT_BINARY_DIR}")
target_link_libraries(gpu_cg
Eigen3::Eigen CUDA::cudart CUDA::cusparse CUDA::cublas CUDA::cusolver CUDA::npps CUDA::nppc)
target_compile_definitions(gpu_cg PRIVATE
EIGEN_TEST_MAX_SIZE=${EIGEN_TEST_MAX_SIZE}
EIGEN_TEST_PART_ALL=1)
add_test(NAME gpu_cg COMMAND gpu_cg)
add_dependencies(buildtests gpu_cg)
add_dependencies(buildtests_gpu gpu_cg)
set_property(TEST gpu_cg APPEND PROPERTY LABELS "Official;gpu")
set_property(TEST gpu_cg PROPERTY SKIP_RETURN_CODE 77)
set(EIGEN_ADD_TEST_FILENAME_EXTENSION "cu")
endif()