[batch] split cg compilation (hip)

ginkgo-project · Jul 5, 2024 · c8e5a61 · c8e5a61
1 parent 38b73e0
commit c8e5a61
Show file tree

Hide file tree

Showing 4 changed files with 324 additions and 224 deletions.
diff --git a/hip/CMakeLists.txt b/hip/CMakeLists.txt
@@ -3,6 +3,7 @@ include(${PROJECT_SOURCE_DIR}/cmake/template_instantiation.cmake)
 add_instantiation_files(. matrix/csr_kernels.instantiate.hip.cpp CSR_INSTANTIATE)
 add_instantiation_files(${PROJECT_SOURCE_DIR}/common/cuda_hip matrix/fbcsr_kernels.instantiate.cpp FBCSR_INSTANTIATE)
 add_instantiation_files(. solver/batch_bicgstab_kernels.instantiate.hip.cpp BATCH_BICGSTAB_INSTANTIATE)
+add_instantiation_files(. solver/batch_cg_kernels.instantiate.hip.cpp 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)
 set(GINKGO_HIP_SOURCES
@@ -36,7 +37,7 @@ set(GINKGO_HIP_SOURCES
     preconditioner/jacobi_generate_kernels.hip.cpp
     preconditioner/jacobi_simple_apply_kernels.hip.cpp
     ${BATCH_BICGSTAB_INSTANTIATE}
-    solver/batch_cg_kernels.hip.cpp
+    ${BATCH_CG_INSTANTIATE}
     solver/lower_trs_kernels.hip.cpp
     solver/upper_trs_kernels.hip.cpp
     ${GKO_UNIFIED_COMMON_SOURCES}

diff --git a/hip/solver/batch_cg_kernels.hip.cpp b/hip/solver/batch_cg_kernels.hip.cpp