https://bugzilla.redhat.com/show_bug.cgi?id=2209759
--- Comment #8 from Jeremy Newton alexjnewt@fastmail.com ---
There also seem to be missing some perl requirements in your current spec:
Ah sorry, I forgot to include perl-generators to generate the perl requires. The update below is fixed.
A modernized patch of these issues would look sth like this
Thanks! I've updated with sed in %prep to make it a bit more flexible than a patch file, but thanks for the patch for illustration.
if ($DEVICE_LIB_PATH ne "$ROCM_PATH/amdgcn/bitcode") {
$HIPCXXFLAGS .= " --hip-device-lib-path=\"$DEVICE_LIB_PATH\"";
if ($DEVICE_LIB_PATH ne "$CLANG_RESOURCE_DIR/amdgcn/bitcode") {
$HIPCXXFLAGS .= " --rocm-device-lib-path=\"$DEVICE_LIB_PATH\"";
If I understand correctly, we always want to use --rocm-device-lib-path for redundancy, since I change the default device libs path in the fedora package.
I find it a bit odd that this square example pulls in the device libs
I think hipcc just always uses it. The perl hipcc is a bit hacky and limited, which is why I think they are rewriting it in c++ (see https://github.com/ROCm-Developer-Tools/HIPCC/blob/develop/README.md).
Here's the updated files: Spec URL: https://mystro256.fedorapeople.org/rocclr.spec SRPM URL: https://mystro256.fedorapeople.org/rocclr-5.5.1-2.fc39.src.rpm copr build: https://copr.fedorainfracloud.org/coprs/mystro256/rocm-hip/build/5988523/
I still need to look into those docs related rpmlint warnings, but I added a fix for aarch64, along with the fixes related to Philipp's comments. I'm going to see if I can get blender working, as that seems to be a good way to test if HIP is working.