https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114801
--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> --- https://developer.arm.com/documentation/101028/0012/14--M-profile-Vector-Extension--MVE--intrinsics suggests that it is a UB if all the bits for a single element aren't the same (i.e. false is all 0s, true is all 1s aka -1, anything else UB). So the testcase is invalid at runtime, right? It probably shouldn't error because it can be in dead code, and it certainly shouldn't ICE, it could emit __builtin_trap or it can just canonicalize it any way it likes it. Seems you are canonicalizing that to 1s, given the HW behavior I think it would be more correct to canonicalize to -1s. I think you could simply canonicalize on the CONST_INT, so else if (VALID_MVE_PRED_MODE (mode)) { if (CONST_INT_P (x) && (mode == V8BImode || mode == V4BImode)) { /* In V8BI or V4BI each element has 2 or 4 bits, if those bits aren't all the same, it is UB and gen_lowpart might ICE. Canonicalize all the 2 or 4 bits to all ones if any of them is non-zero. */ unsigned HOST_WIDE_INT xi = UINTVAL (xi); xi |= ((xi & 0x5555) << 1) | ((xi & 0xaaaa) >> 1); if (mode == V4BImode) xi |= ((xi & 0x3333) << 2) | ((xi & 0xcccc) >> 2); x = gen_int_mode (xi, HImode); } else if (SUBREG_P (x)) /* gen_lowpart on a SUBREG can ICE. */ x = force_reg (GET_MODE (x), x); x = gen_lowpart (mode, x); }
