diff --git a/llvm/projects/hpvm-tensor-rt/CMakeLists.txt b/llvm/projects/hpvm-tensor-rt/CMakeLists.txt index e0940eea58c3a14bda5b7d00d2033c7da8df68a7..d86c8fa3cb7cb70bbe23bda5cb21b57637ab9544 100644 --- a/llvm/projects/hpvm-tensor-rt/CMakeLists.txt +++ b/llvm/projects/hpvm-tensor-rt/CMakeLists.txt @@ -103,7 +103,10 @@ target_link_libraries(tensor_runtime ${LINK_LIBS}) cuda_add_library(tensor_runtime_install ${RUNTIME_SRCS}) cuda_add_cublas_to_target(tensor_runtime_install) target_link_libraries(tensor_runtime_install ${LINK_LIBS}) - +# tensor_runtime_install is built AFTER tensor_runtime because of a nvcc bug (bug?) +# that doesn't allow compiling the same file from multiple targets at once. +# Same for tensor_runtime_online. +add_dependencies(tensor_runtime_install tensor_runtime) # Online version remove_definitions(-DONLINE_PROFILING=false) @@ -112,7 +115,8 @@ remove_definitions(-DFP16_tuning=true) add_definitions(-DFP16_tuning=false) cuda_add_library(tensor_runtime_online ${RUNTIME_SRCS}) cuda_add_cublas_to_target(tensor_runtime_online) -target_link_libraries(tensor_runtime_online ${LINK_LIBS}) +target_link_libraries(tensor_runtime_online ${LINK_LIBS}) +add_dependencies(tensor_runtime_online tensor_runtime) # Adding new rule for building a cuDNN runtime library