Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

mixbench-hip does not compile #20

Open
oonkjbr opened this issue Aug 24, 2020 · 6 comments
Open

mixbench-hip does not compile #20

oonkjbr opened this issue Aug 24, 2020 · 6 comments

Comments

@oonkjbr
Copy link

oonkjbr commented Aug 24, 2020

I have installed the vanillad ROCm from https://github.com/ROCm-Developer-Tools/HIP/blob/master/INSTALL.md

I then set the HIP_PATH to this installation

cmake completes and gives the following build directory: CMakeCache.txt CMakeFiles cmake_install.cmake Makefile

However, running make then gives many errors:

$ make
[ 16%] Building HIPCC object CMakeFiles/mixbench-hip-ro.dir/mixbench-hip-ro_generated_mix_kernels_hip_ro.cpp.o
/opt/rocm/hip/include/hip/nvcc_detail/hip_runtime_api.h(1355): error: class "cudaPointerAttributes" has no member "memoryType"

/usr/local/cuda/include/cuda_fp16.h(1525): error: inline specifier allowed on function declarations only

/usr/local/cuda/include/cuda_fp16.h(1525): error: declaration is incompatible with overloaded function "__shfl_sync"
/usr/local/cuda/include/sm_30_intrinsics.h(133): here

...

Any suggestions on how to solve this?

@ekondis
Copy link
Owner

ekondis commented Aug 29, 2020

I believe that you have set the HIP_PATH env variable, e.g. as below:
export HIP_PATH=/opt/rocm/hip/

Are you building for CUDA target?

@oonkjbr
Copy link
Author

oonkjbr commented Sep 11, 2020

Thanks for your response. Indeed I have set the HIP_PATH as indicated by you and I am building for NVIDIA V100. Can the HIP part not be built for CUDA target, or is there something else that I need to change? I am quite interested in comparing the different software ecosystems, incl the benchmarks, for different type of GPU's.

@ekondis
Copy link
Owner

ekondis commented Sep 19, 2020

I have not used the CUDA target recently. Which is the version of the underlying CUDA installation?

I suspect that this is a problem with HIP expecting a member of the struct cudaPointerAttributes that no longer exists in recent CUDA versions (or just got renamed to "type"). See the relevant CUDA documentation reference.

@ekondis
Copy link
Owner

ekondis commented Sep 24, 2020

Since ROCm v3.8 is out, could you try again with this release?

@oonkjbr
Copy link
Author

oonkjbr commented Sep 25, 2020 via email

@oonkjbr
Copy link
Author

oonkjbr commented Sep 29, 2020 via email

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants