Hi,
I currently have a problem of how to best preserve a register note across
an instruction split, e.g. I had to change a define_split like this (from
m68k.md):
(define_split
[(set (match_operand 0 "register_operand" "")
(zero_extend (match_operand 1 "nonimmediate_src_operand" "")))]
"!TARGET_CFV4 && reload_completed"
- [(set (match_dup 0)
- (const_int 0))
- (set (strict_low_part (match_dup 2))
- (match_dup 1))]
+ [(const_int 0)]
{
- operands[2] = gen_lowpart (GET_MODE (operands[1]), operands[0]);
+ rtx x, note;
+ emit_move_insn (operands[0], GEN_INT (0));
+ x = gen_lowpart (GET_MODE (operands[1]), operands[0]);
+ x = gen_rtx_STRICT_LOW_PART (VOIDmode, x);
+ x = emit_insn (gen_rtx_SET (VOIDmode, x, operands[1]));
+ note = find_reg_note (curr_insn, REG_INC, NULL);
+ if (note)
+ REG_NOTES (x) = gen_rtx_EXPR_LIST (REG_INC, XEXP (note, 0), NULL_RTX);
+ DONE;
})
My problem is now, fixing all splits like this, which may involve an auto
increment, is rather awkward and error prone, so I'm looking for a way to
make this simpler.
In the preparation statement one has no access to the instruction which
are generated by the pattern, so I hope it would be possible, to annotate
the pattern somehow, so genemit.c could generate the code to copy the note
automatically. The other possibility is to add a hook after the
instruction has been splitted.
Anyway, I appreciate any ideas on how to solve this, which would keep the
maintainance overhead as low as possible.
bye, Roman
