Hi,
is there a way, in a CUDAgraph, to set the parameters to a kernel (block/grid size, shared memory) using the result of a previous node (computed in the same CUDAgraph, just a previous step) ?
Thanks in advance for your help/
cudaGraphKernelNodeSetGridDim allows you to set the grid dimentions of a kernel node from the device as long as you set that node to be updatable & have a reference to its device node handle.