Use lld on clang-aarch64-full-2stage #281
Conversation
We have another bot that uses lld, linaro-clang-aarch64-lld-2stage. However it does not build flang or mlir. I don't think we really need both, so I've added the lld settings to clang-aarch64-full-2stage as well. If this is stable then we can remove linaro-clang-aarch64-lld-2stage.
DavidSpickett
force-pushed
the
use-lld
branch
from
02a5785 to
7172116
Compare
| '--threads=32', '--build-threads=32'], | ||
| extra_cmake_args=[ | ||
| "-DCMAKE_C_FLAGS='-mcpu=cortex-a57'", | ||
| "-DCMAKE_CXX_FLAGS='-mcpu=cortex-a57'", | ||
| "-DLLVM_ENABLE_LLD=True", | ||
| "-DLLVM_LIT_ARGS='-v'", |
There was a problem hiding this comment.
I guess we should use -DCLANG_DEFAULT_LINKER=lld but I am little unsure about what -DLLVM_ENABLE_LLD=True does.
Also now that we are making changes to this builder, let change the cpu setting as well. I guess this will run on AWS which may use -mcpu=neoverse-v1 or something similar?
There was a problem hiding this comment.
Ampere Mountain Jade something but yes we should do that anyway I'll do that in a new PR.
|
True. Let me check what the SVE bots do right now. If those are all lld or close to it, maybe we keep this one using |
|
On second thought, I don't have the enthusiasm to figure out the risk of losing I did consider adding the MLIR testing to this builder also, so that it would be the same as the lld if we wanted to merge them in future. However, the lld bot also includes Flang, and adding that will probably break some memory limits, so I decided against doing that. In the future when SVE is standard, we can merge a lot of our non-AWS and AWS based builders, and this will be a much bigger saving. |
We have another bot that uses lld, linaro-clang-aarch64-lld-2stage. However it does not build flang or mlir.
I don't think we really need both, so I've added the lld settings to clang-aarch64-full-2stage as well.
If this is stable then we can remove linaro-clang-aarch64-lld-2stage.