[PATCH] D29004: New OptimizationRemarkEmitter pass for MIR

Matthias Braun via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Jan 23 10:52:41 PST 2017


MatzeB added a comment.

Thinking some more about what is reported here: Reporting every single spill will produce a lot of data and most probably force optviewer to summarize/aggregate the information anyway to be any useful. My expectation would be that anything finer grained than per-loop is probably unnecessary, so:

Maybe a better approach would be to look whether reporting is enabled and if so walk the machine loops after greedy is finished. And then count the spill/reload instructions for each loop and only report that as a remark. A technique as used in AsmPrinter.cpp:emitComments() should be good enough to identify an instruction as performing a spill/reload.


https://reviews.llvm.org/D29004





More information about the llvm-commits mailing list