https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121296
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Last reconfirmed| |2025-07-29
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The reason is that vector lowering of .VEC_CONVERT does not yet support
multi-step conversion for all cases properly. There has been some
progress there for GCC 15 I think but it's not fully done.
Doing lowering before vectorization does at least allow us to re-vectorize
things, like cvt2 to
_73 = BIT_FIELD_REF <x_5(D), 256, 0>;
_74 = BIT_FIELD_REF <x_5(D), 256, 256>;
vect__10.24_75 = (vector(8) signed int) _73;
vect__10.24_76 = (vector(8) signed int) _74;
vect__10.23_77 = VEC_PACK_TRUNC_EXPR <vect__10.24_75, vect__10.24_76>;
_1 = {vect__10.23_77, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }};
_78 = BIT_FIELD_REF <y_6(D), 256, 0>;
_79 = BIT_FIELD_REF <y_6(D), 256, 256>;
vect__42.26_80 = (vector(8) signed int) _78;
vect__42.26_81 = (vector(8) signed int) _79;
vect__42.25_82 = VEC_PACK_TRUNC_EXPR <vect__42.26_80, vect__42.26_81>;
_3 = {vect__42.25_82, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }};
_7 = VEC_PERM_EXPR <_1, _3, { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13,
14, 15, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47 }>;
which shows some previously observed issues.
