memcpy in target region

[zjin-lcf]

Is memcpy a supported in AMD clang++ ?

ld.lld: error: undefined symbol: memcpy

[zichguan-amd]

Hi @zjin-lcf, can you provide a sample snippet of what you are trying to achieve? Host memcpy requires string.h header.

[zjin-lcf]

It is memcpy in a target device region. I hope a sample snippet is not needed for the issue.

[zichguan-amd]

You mean omp_target_memcpy?

[zjin-lcf]

double f(...)
{
  ... 
  memcpy(y, ytemp, N * sizeof(double));

  return hnext;
}

[zichguan-amd]

Which version of aomp are you using? As of aomp 23.0-0 memcpy should be supported in target region.

[zjin-lcf]

If amdclang++ is different from aomp, please close the issue. Thanks.

amdclang++ --version
AMD clang version 22.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-7.2.0 26014 7b800a19466229b8479a78de19143dc33c3ab9b5)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /opt/rocm-7.2.0/lib/llvm/bin
Configuration file: /opt/rocm-7.2.0/lib/llvm/bin/clang++.cfg

[zichguan-amd]

aomp is different than the ROCm-packaged llvm. Can you please try the latest aomp release from https://github.com/ROCm/aomp/blob/aomp-dev/docs/MANYLINUX.md#install-from-release-binary-tarball? You can read more on https://github.com/ROCm/aomp/releases/tag/rel_23.0-0.

Use the amdclang++ from the aomp tarball to build your program. New OpenMP developments will happen in aomp first then trickle down to ROCm-packaged llvm.

[zjin-lcf]

Sure. Thank you for your answers very much.