https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125739
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> --- Confirmed with -O3 -fno-tree-dse -fno-tree-pre -fno-expensive-optimizations -fno-tree-sra -fwhole-program -fopt-info-vec -fno-tree-loop-vectorize, we vectorize t.c:27:7: optimized: basic block part vectorized using 8 byte vectors t.c:42:7: optimized: basic block part vectorized using 8 byte vectors we are vectorizing t.c:42:7: note: Vectorizing SLP tree: t.c:42:7: note: node 0x4c9bb40 (max_nunits=2, refcnt=1) vector(2) float t.c:42:7: note: op template: x.a = _9; t.c:42:7: note: stmt 0 x.a = _9; t.c:42:7: note: stmt 1 x.aa = _11; t.c:42:7: note: children 0x4c9bbe8 t.c:42:7: note: node 0x4c9bbe8 (max_nunits=2, refcnt=2) vector(2) float t.c:42:7: note: op template: _9 = x_a_lsm0.50_44 + 0.0; t.c:42:7: note: stmt 0 _9 = x_a_lsm0.50_44 + 0.0; t.c:42:7: note: stmt 1 _11 = x_aa_lsm0.49_57 + 0.0; t.c:42:7: note: children 0x4c9bc90 0x4c9bde0 t.c:42:7: note: node 0x4c9bc90 (max_nunits=2, refcnt=1) vector(2) float t.c:42:7: note: op template: x_a_lsm0.50_44 = PHI <_38(4), _9(10)> t.c:42:7: note: [l] stmt 0 x_a_lsm0.50_44 = PHI <_38(4), _9(10)> t.c:42:7: note: [l] stmt 1 x_aa_lsm0.49_57 = PHI <_2(4), _11(10)> t.c:42:7: note: children 0x4c9bd38 0x4c9bbe8 t.c:42:7: note: node 0x4c9bd38 (max_nunits=2, refcnt=1) vector(2) float t.c:42:7: note: op template: _38 = MEM[(float *)&x]; t.c:42:7: note: stmt 0 _38 = MEM[(float *)&x]; t.c:42:7: note: stmt 1 _2 = MEM[(float *)&x + 4B]; t.c:42:7: note: node (constant) 0x4c9bde0 (max_nunits=1, refcnt=1) vector(2) float but there's nothing obviously wrong happening.
