https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70232
--- Comment #16 from Jeffrey A. Law <law at redhat dot com> --- So for thread paths noted in c#15 we have the following pieces of data 1. 69 statements to copy 2. 7 blocks to copy 3. Threads through latch, but does not create an irreducible loop 4. Eliminates a simple conditional branch, not a multiway branch tree-ssa-threadbackwards is used for 2 cases. Its first (and original) purpose was to catch FSM loops. Those are characterized by threading through the latch and eliminating a multi-way branch at the end of path. These are very profitable and as such they allow for a lot of copying (100 statements). The second case is as a replacement for the older forward-walking jump threading done by VRP/DOM that we're phasing out. These are less profitable and have much lower copying thresholds. #4 means that we've got the latter case. But the code is not identifying it as such. The code tests !threaded_through_latch when it should be testing !(threaded_through_latch && threaded_multiway_branch). Fixing that results in all those threads around the loop being rejected as needing too many statement copies. Only two small/short jump threads are realized. All the undesirable copying is gone and we have a stack size of 140 bytes.
