[batch] split cg compilation (cuda)

ginkgo-project · Jul 5, 2024 · 870ad69 · 870ad69
1 parent c8e5a61
commit 870ad69
Show file tree

Hide file tree

Showing 4 changed files with 380 additions and 240 deletions.
diff --git a/cuda/CMakeLists.txt b/cuda/CMakeLists.txt
@@ -4,6 +4,7 @@ include(${PROJECT_SOURCE_DIR}/cmake/template_instantiation.cmake)
 add_instantiation_files(. matrix/csr_kernels.instantiate.cu CSR_INSTANTIATE)
 add_instantiation_files(${PROJECT_SOURCE_DIR}/common/cuda_hip matrix/fbcsr_kernels.instantiate.cpp FBCSR_INSTANTIATE)
 add_instantiation_files(. solver/batch_bicgstab_kernels.instantiate.cu BATCH_BICGSTAB_INSTANTIATE)
+add_instantiation_files(. solver/batch_cg_kernels_launch.instantiate.cu BATCH_CG_INSTANTIATE)
 # we don't split up the dense kernels into distinct compilations
 list(APPEND GKO_UNIFIED_COMMON_SOURCES ${PROJECT_SOURCE_DIR}/common/unified/matrix/dense_kernels.instantiate.cpp)
 target_sources(ginkgo_cuda
@@ -39,7 +40,7 @@ target_sources(ginkgo_cuda
     preconditioner/jacobi_generate_kernels.cu
     preconditioner/jacobi_simple_apply_kernels.cu
     ${BATCH_BICGSTAB_INSTANTIATE}
-    solver/batch_cg_kernels.cu
+    ${BACTH_CG_INSTANTIATE}
     solver/lower_trs_kernels.cu
     solver/upper_trs_kernels.cu
     ${GKO_UNIFIED_COMMON_SOURCES}

diff --git a/cuda/solver/batch_cg_kernels.cu b/cuda/solver/batch_cg_kernels.cu