https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96912
Jakub Jelinek <jakub at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |jakub at gcc dot gnu.org --- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Note, in: typedef char V __attribute__((vector_size(16))); typedef long long W __attribute__((vector_size(16))); W foo (W x, W y, V m) { W t = (m < 0); return (~t & x) | (t & y); } V bar (V x, V y, V m) { V t = (m < 0); return (~t & x) | (t & y); } we actually optimize bar the way we should, seems it is forwprop1 that turns _1 = m_5(D) < { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; t_6 = VEC_COND_EXPR <_1, { -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 }, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }>; _2 = ~t_6; _3 = x_7(D) & _2; _4 = t_6 & y_8(D); _9 = _3 | _4; return _9; into: _1 = m_5(D) < { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; t_6 = VEC_COND_EXPR <_1, { -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 }, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }>; _2 = VEC_COND_EXPR <_1, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, { -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 }>; _3 = VEC_COND_EXPR <_1, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, x_7(D)>; _4 = VEC_COND_EXPR <_1, y_8(D), { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }>; _9 = VEC_COND_EXPR <_1, y_8(D), x_7(D)>; return _9; but the similar: _1 = m_6(D) < { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; _2 = VEC_COND_EXPR <_1, { -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 }, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }>; t_7 = VIEW_CONVERT_EXPR<W>(_2); _3 = ~t_7; _4 = x_8(D) & _3; _5 = t_7 & y_9(D); _10 = _4 | _5; return _10; in foo isn't optimized similarly. I'll look tomorrow at that, we should handle it likee bar with the VEC_COND_EXPR being done in the vector type corresponding to the comparison with VCEs to that and back.