[OpenMP][NVPTX] Disable OpenMPOpt when building deviceRTLs

We build `deviceRTLs` with `-O1` by default, which also triggers OpenMPOpt. When
the info cache is created, some attributes are removed. As a result, although we
mark a few functions `noinline`, they are still inlined when the bitcode library
is generated. This can cause an issue in middle end optimization.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D106710
This commit is contained in:
Shilei Tian 2021-07-25 10:38:15 -04:00
parent fa0910e6de
commit f1b8fa55d0

View File

@ -153,7 +153,7 @@ set(cuda_src_files
# Set flags for LLVM Bitcode compilation.
set(bc_flags -S -x c++ -O1 -std=c++14
-mllvm -openmp-opt-disable-internalization
-mllvm -openmp-opt-disable
-target nvptx64
-Xclang -emit-llvm-bc
-Xclang -aux-triple -Xclang ${aux_triple}