Modulesfile environment file?

SYCL development
1

Hi,
did anyone hack together a modules file?

Brgds,
Tor

2

Hi Tor,
Welcome. When you say a modules file can you let us know what you are trying to do and what you mean by that?
Thanks,
Rod

3

Module environment as in Environment Modules — Modules 4.7.1+25-gdb74c4c documentation

Created one that seems to work. Any other variables that need to be included?

$ cat /cm/shared/modulefiles/computecpp/ce/2.5.0
#%Module

ComputeCPP Sycl compiler 2.5.0

set version 2.5.0

proc ModulesHelp { } {
puts stderr “\tSets up the environment for ComputeCpp Community Edition (CE) $version SYCL compiler \n”
puts stderr “\theterogeneous parallel programming platform that provides a conformant implementation of SYCL™ 1.2.1 Khronos specification. \n”
}

module-whatis "ComputeCPP is a SYCL compiler complient with Khronos 1.2.1 specifications. "
module-whatis "https://developer.codeplay.com/products/computecpp/ce/download "
module-whatis “Docs: $ firefox file:///cm/shared/apps/computecpp/ce/2.5.0/doc/api_pages/index.html &”

setenv ComputeCpp_DIR /cm/shared/apps/computecpp/ce/$version
setenv COMPUTECPP_ROOT /cm/shared/apps/computecpp/ce/$version
setenv COMPUTECPP_INCDIR /cm/shared/apps/computecpp/ce/$version/include
setenv COMPUTECPP_INCLUDEDIR /cm/shared/apps/computecpp/ce/$version/include
setenv COMPUTECPP_LIBDIR /cm/shared/apps/computecpp/ce/$version/lib
setenv COMPUTECPP_LIBRARYDIR /cm/shared/apps/computecpp/ce/$version/lib
prepend-path PATH /cm/shared/apps/computecpp/ce/$version/bin
prepend-path C_INCLUDE_PATH /cm/shared/apps/computecpp/ce/$version/include
prepend-path CPLUS_INCLUDE_PATH /cm/shared/apps/computecpp/ce/$version/include
prepend-path LIBRARY_PATH /cm/shared/apps/computecpp/ce/$version/lib
prepend-path LD_LIBRARY_PATH /cm/shared/apps/computecpp/ce/$version/lib
prepend-path PKG_CONFIG_PATH /cm/shared/apps/computecpp/ce/$version/lib/pkgconfig
prepend-path INFOPATH /cm/shared/apps/computecpp/ce/$version/doc/api_pages

set MSG “ComputeCPP CE $version (ComputeCpp-CE-2.5.0-aarch64-linux-gnu)”

4

Added COMPUTECPP_PACKAGE_ROOT_DIR for cmake.

5

Thank you. I’ve not seen that before but it looks like a useful set of information. I’ll ask the team if there is anything else that could be included there.

6

Module environment is very common in HPC systems to allow for clean install of many versions of compilers, tools and frameworks/SDKs. There are 2 variants of it. Environment modules and lmod. I use the former. Maybe you should include a section on it in your documentation? I would much appreciate feedback from you if I have left something out. Maybe OPENCL_VENDOR_PATH?

For instance, I notice that on a pure aarch64 node (Cavium Tx2 CN9980) without GPGPU I get

torel@n005:~$ module load computecpp/ce/2.5.0
torel@n005:~$ computecpp_info

ComputeCpp Info (CE 2.5.0 2021/04/23)

SYCL 1.2.1 revision 3

Device Info:

Cannot find any devices on the system. Please refer to your OpenCL vendor documentation.
Note that OPENCL_VENDOR_PATH is not defined. Some vendors may require this environment variable to be set.
torel@n005:~$

7

ARM computers don’t generally ship with a CPU OpenCL implementation (I don’t think ARM even provides such a thing). That seems to be the case here - so there’s nothing to run your code.

OpenCL with ARM NEON (without Mali GPU) available? - Stack Overflow this link seems to indicate that this is still true - though it is 5 years old, I haven’t heard anything to suggest the situation’s changed.

8

True. Will be adding some GPUs to the arm64 nodes now that Nvidia supports Cuda on aarch64 as well as nvhpc compiler. Must be ready for Grace. I suppose it will be exported. Will computecpp work with Nvidia offload engine?

torel@n005:~$ du -a /cm/shared/apps/computecpp/ce/2.5.0 | grep -i nvptx
22372 /cm/shared/apps/computecpp/ce/2.5.0/lib/ComputeCppTargets/nvptx64–nvidiacl.bc

9

ComputeCpp support for Nvidia devices is no longer officially supported. I would point you towards the work we are doing to add Nvidia device support to DPC++ however.

10

To add to it. Installed 2.11.0, and on aarch64 (Ubuntu 18.04.6LTS) with A40 GPU:

torel@n009:/cm/shared/apps/computecpp/ce/2.11.0$ uname -ar
Linux n009 5.4.0-148-generic #165~18.04.1-Ubuntu SMP Thu Apr 20 01:13:55 UTC 2023 aarch64 aarch64 aarch64 GNU/Linux
torel@n009:/cm/shared/apps/computecpp/ce/2.11.0$ cd
torel@n009:~$ module load computecpp/ce/2.11.0
torel@n009:~$ computecpp_info 
********************************************************************************

ComputeCpp Info (CE 2.11.0 2022/08/08)

SYCL 1.2.1 revision 3

********************************************************************************


Device Info:

Discovered 1 devices matching:
  platform    : <any>
  device type : <any>

--------------------------------------------------------------------------------
Device 0:

  Device is supported                     : UNTESTED - Untested OS
  Bitcode targets                         : spirv64 ptx64 
  CL_DEVICE_NAME                          : NVIDIA A40
  CL_DEVICE_VENDOR                        : NVIDIA Corporation
  CL_DRIVER_VERSION                       : 510.85.02
  CL_DEVICE_TYPE                          : CL_DEVICE_TYPE_GPU 

If you encounter problems when using any of these OpenCL devices, please consult
this website for known issues:
https://developer.codeplay.com/products/computecpp/ce/guides/platform-support?version=2.11.0

********************************************************************************

torel@n009:~$ compute
compute          compute++        compute-cl       computecpp_info  
torel@n009:~$ compute-cl --version
Codeplay ComputeCpp - CE 2.11.0 Device Compiler - clang version 8.0.0  (based on LLVM 8.0.0svn)
Target: aarch64-pc-windows-msvc
Thread model: posix
InstalledDir: /cm/shared/apps/computecpp/ce/2.11.0/bin
torel@n009:~$

Windows?