> The middle-end pass NULL_TREE in previous, and then skip the adjust_stmt_cost > step which has count the cost of vr2gpr. > After Richard introduced more like slp_node with its vectype for recording > the cost, the adjust_stmt_cost will be hit and plus the cost of vr2gpr now. > And then fail to vectorize due to cost value changes of vr2gpr from 0 to 2. > > That means we need to reconsider how to pass the cost value from the command > line eventually. > If we need to add more options for cost, shall we consider to introduce a > config file for all possible costs values? > It may look like a json file which contains sorts of cost num, instead of add > it separately. (may have minor discuss with kito last summer I bet.)
For now I suggest we just go with a new param. As it's just for the testsuite it won't hurt. Another option would also be a CPU cost model (likewise just for testing) that has zero GPR <-> VR latency. Once a better option comes up we can easily get rid of both params again. -- Regards Robin
