[OPENMP][NVPTX]Mark more functions as always_inline for better
authorAlexey Bataev <a.bataev@hotmail.com>
Tue, 21 May 2019 15:11:58 +0000 (15:11 +0000)
committerAlexey Bataev <a.bataev@hotmail.com>
Tue, 21 May 2019 15:11:58 +0000 (15:11 +0000)
commit1353fe814c52fd2eb0324cea1099ed79703896ff
tree13b6f4738d767d699d11a0e9a6ae68dbf70a27ca
parent6b544e2d0d0e55b57c95825ee03a9f7423174ada
[OPENMP][NVPTX]Mark more functions as always_inline for better
performance.

Internally generated functions must be marked as always_inlines in most
cases. Patch marks some extra reduction function + outlined parallel
functions as always_inline for better performance, but only if the
optimization is requested.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@361269 91177308-0d34-0410-b5e6-96231b3b80d8
23 files changed:
lib/CodeGen/CGOpenMPRuntime.cpp
lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
test/OpenMP/declare_target_codegen_globalization.cpp
test/OpenMP/for_reduction_codegen_UDR.cpp
test/OpenMP/nvptx_allocate_codegen.cpp
test/OpenMP/nvptx_data_sharing.cpp
test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
test/OpenMP/nvptx_lambda_capturing.cpp
test/OpenMP/nvptx_parallel_codegen.cpp
test/OpenMP/nvptx_parallel_for_codegen.cpp
test/OpenMP/nvptx_target_codegen.cpp
test/OpenMP/nvptx_target_parallel_codegen.cpp
test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp
test/OpenMP/nvptx_target_teams_codegen.cpp
test/OpenMP/nvptx_target_teams_distribute_codegen.cpp
test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp
test/OpenMP/nvptx_teams_codegen.cpp
test/OpenMP/nvptx_teams_reduction_codegen.cpp
test/OpenMP/taskloop_reduction_codegen.cpp
test/OpenMP/taskloop_simd_reduction_codegen.cpp