https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79295
--- Comment #2 from Michael Meissner <meissner at gcc dot gnu.org> ---
Yes, bcdadd requires all of its arguments to be altivec registers.
However, the pattern below is wrong:
(define_insn "bcd<bcd_add_sub>"
[(set (match_operand:V1TI 0 "register_operand" "")
(unspec:V1TI [(match_operand:V1TI 1 "register_operand" "")
(match_operand:V1TI 2 "register_operand" "")
(match_operand:QI 3 "const_0_to_1_operand" "")]
UNSPEC_BCD_ADD_SUB))
(clobber (reg:CCFP CR6_REGNO))]
"TARGET_P8_VECTOR"
"bcd<bcd_add_sub>. %0,%1,%2,%3"
[(set_attr "length" "4")
(set_attr "type" "vecsimple")])
It should have "v" constraints for each of the registers, and "n" for the
integer. It probably should use altivec_register_operand (or at least
gpc_reg_operand) instead of register_operand as well.
