Mihai Preda

@ROCmSupport apparently you don't have the OpenCL environment correctly set up. Does "clinfo" work? (that is the probable reason of "device not found" -- i.e. there is no OpenCL device...

Can you please attach the output of "gpuowl -h", which if it detects any devices would print the list of devices at the end. And maybe the output of clinfo?...

BTW, were you able to run gpuowl in the past? what changed?

@ROCmSupport I don't know, quite possibly (I don't have ROCm 4.1 installed). In any case, a stack trace (with gdb \ \) can inform whether it's the same segfault.

@ROCmSupport probably you're using an old version of the gcc compiler (g++ --version). The easiest fix would be to upgrade the compiler. If you want to try with experimental/filesystem, you...

The segfault may be related to some degree to the inline assembly, because when compiling with inline assembly disabled (which can be done by adding "-use NO_ASM" to the gpuowl...

@ROCmSupport I don't understand, what information would you find helpful? When running gpuowl, try adding to the end of the command line "-use NO_ASM", e.g.: ``` ./gpuowl -prp 107000009 -time...

@b-sumner yes, gpuowl also supports dump with the command line option "-dump \", so obtaining those files is easy. I attach them here for convenience, based on the version at...

@b-sumner if it's any help, the ".i" below contains a single __asm() that triggers the segfault. [4M_0_gfx906_sramecc-_xnack-.i.txt](https://github.com/RadeonOpenCompute/ROCm/files/6643612/4M_0_gfx906_sramecc-_xnack-.i.txt)

I tried a fresh build of LLVM and comgr, and hit this assert: ``` gpuowl: /home/preda/llvm-project/llvm/lib/CodeGen/CalcSpillWeights.cpp:94: bool isRematerializable(const llvm::LiveInterval&, const llvm::LiveIntervals&, const llvm::VirtRegMap&, const llvm::TargetInstrInfo&): Assertion `MI && "Dead valno...