Hi Julian, hi all, (I have not (yet) looked at the patch/testcase in depth - and I prefer that someone knowledgeable with C++ constructors comments. However:)
Regarding only nvptx - and the testcase on nvptx: On 12.05.23 14:02, Julian Brown wrote:
The approach taken is to generate a set of constructors to run on the target: this currently works for AMD GCN, but fails on NVPTX due to lack of constructor/destructor support there so far on mainline. (See the new test static-aggr-constructor-destructor-3.C for a reason why running constructors on the target is preferable to e.g. constructing on the host and then copying the resulting object to the target.)
(I think also the wording in OpenMP implies that it is constructed on the device and not copied over from the host, albeit the wording is not very clear.) For nvptx constructor/destructor support, see email thread "nvptx: Support global constructors/destructors via 'collect2' for offloading", e.g., https://gcc.gnu.org/pipermail/gcc-patches/2023-January/609711.html * * * If I run the testcase manually with nvptx offloading and -O0, I get: libgomp/testsuite/libgomp.c++/static-aggr-constructor-destructor-3.C:8:3: error: alias definitions not supported in this configuration and with -O1 and higher: libgomp/testsuite/libgomp.c++/static-aggr-constructor-destructor-3.C:36:1: sorry, unimplemented: global constructors not supported on this target As DEFAULT_CFLAGS is -O2 checking for the proper error works, i.e. something like (untested and ... to be filled in): /* { dg-message "sorry, ..." "..." target { offload_target_nvptx } } } /* { dg-excess-errors "Follow-up errors from mkoffload and lto-wrapper" { target { offload_target_nvptx } } */ Tobias ----------------- Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht München, HRB 106955
