gdb, amdgpu: pass process id to amd_dbgapi_convert_address_space - #282
gdb, amdgpu: pass process id to amd_dbgapi_convert_address_space#282aktemur wants to merge 1 commit into
Conversation
Pass a process id to the dbgapi function `amd_dbgapi_convert_address_space`, which now requires this additional argument. Bug: AIROCGDB-644
| # greater or equal that version. It can be handy when testing with a newer | ||
| # version of the library. | ||
| PKG_CHECK_MODULES([AMD_DBGAPI], [amd-dbgapi >= 0.80.0], | ||
| PKG_CHECK_MODULES([AMD_DBGAPI], [amd-dbgapi >= 0.81.0], |
There was a problem hiding this comment.
I think we will need a transition phase for this… otherwise CI will never pass on ROCm-systems.
i.e. we need to first have ROCgdb which has a #if to select between the old and new dbgapi interface. Land this, and get it in TheRock. Once this is picked up, we can proceed and merge the dbgapi part (with passing CI), and later on we can do cleanup in GDB and remove the #if.
This is a real nuisance, but unfortunately, I do not think we have much of a way around this.
Ideally, in GDB's patch (at least the initial, this can be removed from the cleanup), GDB's configury would check the API and select based on this rather than a version number (using something like AC_TRY_COMPILE ). This way, we do not need to assume what order patches will land in dbgapi.
3 participants
This is the companion PR to ROCm/rocm-systems#10163.