Re: register renaming causes many ARM testsuite failures

[Bernd Schmidt <bernds at redhat dot com>]

On Tue, 2 Jan 2001, Philip Blundell wrote:

> >> The problem is that an insn like this matches the special `load multiple'
> >> pattern which generates an LDM instruction:
> >
> >Apparently the constraints on the ldmsi patterns aren't strict enough.
>
> Hmm, really?  I don't, offhand, see anything in regrename that would cause the
> constraints to have an effect here.  What were you thinking of in particular?

The function build_def_use uses constraints to determine which registers can
be used for renaming.  The relevant bit of code is this:

          /* Step 6: Begin new chains for writes inside operands.  */
          /* ??? Many targets have output constraints on the SET_DEST
             of a call insn, which is stupid, since these are certainly
             ABI defined hard registers.  Don't change calls at all.  */
          if (GET_CODE (insn) != CALL_INSN)
            for (i = 0; i < n_ops + recog_data.n_dups; i++)
              {
                int opn = i < n_ops ? i : recog_data.dup_num[i - n_ops];
                rtx *loc = (i < n_ops
                            ? recog_data.operand_loc[opn]
                            : recog_data.dup_loc[i - n_ops]);
                enum reg_class class = recog_op_alt[opn][alt].class;

                if (recog_data.operand_type[opn] == OP_OUT)
                  scan_rtx (insn, loc, class, mark_write, OP_OUT,
                            recog_op_alt[opn][alt].earlyclobber);
	     }


Bernd