Hi~
I try to access the defined private variable in cuda_module class after op execution in graph_runtime. How can I do it? Do I need to write packedfunc in cuda_module.cc?
Hi~
I try to access the defined private variable in cuda_module class after op execution in graph_runtime. How can I do it? Do I need to write packedfunc in cuda_module.cc?
What variable do you want to extract from cuda_module.cc
? For cuda kernel source, you can use runtime.imported_modules[0].get_source()", where
runtime is tvm.runtime.Module
.
Hi, thanks for your response. Actually I would like to extract(return) grid/block info as well as void**args to graph_runtime.cc, any ideas? I am trying to write packedfunc in cuda_module.cc
Hi, thanks for your response. Actually I would like to extract(return) grid/block info as well as void**args to graph_runtime.cc, any ideas? I am trying to write packedfunc in cuda_module.cc
That would be difficult, since graph runtime doesn’t know anything about CUDAWrappedFunc
or CUDAModuleNode
.
Is it possible to set some global variables to be accessed in graphruntime