Open Kerilk opened 3 years ago
CMake should be now more compatible with upstream HIP in master, although i have only tried it on a few samples. Please try it out and see if it works. You'll need to do a reinstall hipcl (there's no need to recompile clang+llvm).
Thanks, we'll try this out and let you know.
Thanks for this!
We tried it out on our system and it works to build more things than it used to! However, I was hoping that I could more or less use the exact same CMakeLists.txt from AMD hip to compile hipcl code.
I ran into one main issue: one of the codes I was trying to build with HIPCL uses the hip::device target from find_package(hip) to add compilation flags. hipcl doesn't define hip::device, but it defines just "hip::hipcl". Are there any plans to also define hip::device for hipcl so we can more easily use the same CMakeLists.txt as AMD hip?
I can respond with the CMakeLists.txt if that would help. Thanks!
Ping on this!
Sorry, i'm not aware of what is "hip::device" and what should it provide. Haven't seen it in the HIP CMake code. We could add an empty target but i'm not sure if that would work.
Hello,
Disclaimer: I am no expert in CMake...
We are currently trying to run hipcl on our test base: https://github.com/jz10/hip-test_suite Most simple examples are fine, and auto-tools build systems can also be configured without too much issues. Problems start arising when trying to build CMake based projects. The
FindHIP.cmake
of hipcl file seems to have diverged quite a bit from what HIP proposes nowadays: https://github.com/ROCm-Developer-Tools/HIP/blob/main/cmake/FindHIP.cmake In most cases this causes cmake to crash when a macro likehip_add_executable
is missing, or reject hipcl:As I already mentioned, I am no CMake expert so I don't feel confident adequately solving those issues. Nonetheless, it seems like a worthwhile investment, as hipcl is becoming increasingly stable, using it with bigger projects seems the likely next step.
Thanks