Alexey Akimov
Alexey Akimov
Fix issue https://github.com/ROCm-Developer-Tools/rocprofiler/issues/89
Added total amount of memory transferred through L2 cache for gfx908 target
Fixed L2CacheHit metric for gfx908 target
Sometimes it's very useful in debug process
Added explicit check for infinite recursion in EnableExecutableTracking function. It's impossible to guarantee any order of EnableExecutableTracking and InitHsaApiTable. Removing null check for hsa_api_.hsa_init will add dependency from number and...
way to reproduce: `cat input.xml` `` `sudo rm /tmp/*.txt` `cd /opt/rocm-4.5.2/hip/samples/2_Cookbook/0_MatrixTranspose` `sudo make` `sudo rocprof -i input.xml ./MatrixTranspose` `cd /opt/rocm-4.5.2/hip/samples/2_Cookbook/13_occupancy` `sudo make` `sudo rocprof -i input.xml ./occupancy` `cat input.csv` Index,KernelName,gpu-id,queue-id,queue-index,pid,tid,grd,wgr,lds,scr,vgpr,sgpr,fbar,sig,obj,SQ_WAVES...
Intercept Queue callbacks are checked for null. If the user forgets to initialize one of them, random code could be called. That leads to very tricky memory corruption bugs. I...
callback_data_ is not cleared in RemoveCallbacks fixed it
Hsa ops traces correlation id is number of hsa async-copy called. Hsa api does not log any correlation id, but can be calculated implicitly as number of current hsa_amd_memory_async_copy called....
Fixed --trace-start on/of not working issue. roctracer_stop() call in tool_load() function does nothing before hsa activities registered. That forces tracing to be enabled from start independently from option value. tool_load...