search

ROCm / roctracer

ROCm Tracer Callback/Activity Library for Performance tracing AMD GPUs
https://rocm.docs.amd.com/projects/roctracer/en/latest/
Other
64 stars 30 forks source link

roctracer output for hipLaunchKernel does not include launch parameters #70

Closed lfmeadow closed 1 year ago

lfmeadow commented 2 years ago

e.g.:

rocprof --timestamp on --hip-trace -d profdata make run
...
in profdata/.../hip_api_trace.txt:

hipLaunchKernel(function_address=0x2011d8, numBlocks={}, dimBlocks={}, args=0x7ffde3d4d1d0, sharedMemBytes=0, stream=0)

The data for numBlocks etc is there for hipModuleLaunchKernel but not for hipLaunchKernel. It does show up in the rocprof counter output.

lmoriche commented 1 year ago

This should be fixed in ROCm-5.3:

hipLaunchKernel(function_address=0x200a50, numBlocks={z=1, y=1, x=2}, dimBlocks={z=1, y=1, x=128}, args=0x7ffc0ed19378, sharedMemBytes=1024, stream=0)