https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125124
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> --- This is now an issue with BB SLP partitioning. We have two integer reductions t.c:3:6: note: node 0x52f4f80 (max_nunits=4, refcnt=1) vector(4) int t.c:3:6: note: op template: patt_323 = (int) patt_322; t.c:3:6: note: [nv] stmt 0 patt_323 = (int) patt_322; t.c:3:6: note: [nv] stmt 1 patt_314 = (int) patt_313; t.c:3:6: note: [nv] stmt 2 patt_301 = (int) patt_300; t.c:3:6: note: [nv] stmt 3 patt_305 = (int) patt_304; t.c:3:6: note: children 0x52f4ed0 ... t.c:3:6: note: node 0x52f5500 (max_nunits=4, refcnt=1) vector(4) int t.c:3:6: note: op template: patt_354 = (int) patt_353; t.c:3:6: note: [nv] stmt 0 patt_354 = (int) patt_353; t.c:3:6: note: [nv] stmt 1 patt_345 = (int) patt_344; t.c:3:6: note: [nv] stmt 2 patt_332 = (int) patt_331; t.c:3:6: note: [nv] stmt 3 patt_336 = (int) patt_335; t.c:3:6: note: children 0x52f55b0 ... where vect_bb_partition_graph_r considers the recorded (pattern) SLP_TREE_SCALAR_STMTS only rather than the scalar coverage. The assert that triggers is exactly an assert that BB SLP partitioning worked, otherwise it should be redundant at this point.
