On 07/29/2014 10:27 AM, Gengyulei (Gengyl) wrote: > Thank you for your answer. I find the most time consuming process in > compiling a file is the optimization of the cgraph nodes (execute > all_passes), > This process is sequence, one node by one node. If we divide the cgraph nodes > into unrelated forest, we can parallel it, is this way feasible?
I suspect it is, but would require a lot of re-engineering. The GCC approach is to split source into compilation units of reasonable size and to optimize at link time. Andrew.