ROCm / rocmProfileData

MIT License
14 stars 8 forks source link

Graphkernel fix #38

Closed mwootton closed 1 year ago

mwootton commented 1 year ago

Have tracer add a placeholder entry in rocpd_kernelapi for graphLaunch calls. This allows graph kernels to appear as kernel subclasses and work with kernel analysis tools. The kernels are "dataless" as the params are only visible during the original capture phase. Add tracer logging of cudaLaunchKernelExC_v11060.