https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93512
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Segher said he thinks that ROTATE/ROTATERT on all targets will do the
truncation, so it is just a matter of proving that. Sure, one can add
explictly patterns like
(define_insn "rotl<mode>3_cntmask"
[(set (match_operand:GPR 0 "gpc_reg_operand" "=r")
(rotate:GPR (match_operand:GPR 1 "gpc_reg_operand" "r")
(and:SI (match_operand:SI 2 "reg_or_cint_operand" "rn")
(match_operand:SI 3 "const_int_operand" "n")))]
"UINTVAL (operands[3]) == GET_MODE_MASK (<MODE>mode)"
"rotl<wd>%I2 %0,%1,%<hH>2"
[(set_attr "type" "shift")
(set_attr "maybe_var_shift" "yes")])
and e.g. i386 target already does for a few patterns (but not all).
But:
$ find config -name \*.md | xargs grep '(rotate:' | wc -l
139
$ find config -name \*.md | xargs grep '(rotatert:' | wc -l
64
and that still doesn't include when rotates are macroized.
