https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #9 from Tobias Burnus <burnus at gcc dot gnu.org> --- (In reply to Jakub Jelinek from comment #8) > Lightly tested patch. Just quick manually testing "for-3.c" (I tried -O0 and -O3): * With nvptx offloading, it compiles + links – but at run time, I get on two systems: libgomp: cuLaunchKernel error: too many resources requested for launch and, on the third system, a SEGFAULT – which sounds as if it could be the same issue: #0 memcpy () at ../sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:145 #1 0x00007ffff63b2552 in ?? () from /usr/lib/x86_64-linux-gnu/libcuda.so.1 when executing libgomp/plugin/plugin-nvptx.c:2004 2004 r = CUDA_CALL_NOCHECK (cuLaunchKernel, function, teams, 1, 1, * For amdgcn, I get at startup: ... GCN debug: Released kernel dispatch: 0x7eb350 GCN debug: Copying 6000 bytes from host (0x7730c0) to device 0 (0x7ffeed8194d0) GCN warning: Could not find symbol for kernel in the code object Runtime message: HSA_STATUS_ERROR_INVALID_SYMBOL_NAME: There is no symbol with the given name. not found name: 'test_d_normal._omp_fn.0.kd' ... not found name: 'test_d_ds128_normal._omp_fn.0.kd' not found name: 'test_ds_normal._omp_fn.0.kd' ... [The .kd" comes from plugin/plugin-gcn.c's: sprintf (buf, "%s.kd", kernel->name); ] (I am now doing a full bootstrap now to ensure that that wasn't due to the incremental build.)
