Re: what can be in a group set?

[Richard Sandiford <richard.sandiford@linaro.org>]

Paolo Bonzini <bonzini@gnu.org> writes:
> df-scan.c has this code to deal with group sets:
>
>    /* It is legal to have a set destination be a parallel. */
>    if (GET_CODE (dst) == PARALLEL)
>      {
>        int i;
>
>        for (i = XVECLEN (dst, 0) - 1; i >= 0; i--)
>          {
>            rtx temp = XVECEXP (dst, 0, i);
>            if (GET_CODE (temp) == EXPR_LIST || GET_CODE (temp) == CLOBBER
>                || GET_CODE (temp) == SET)
>              df_def_record_1 (collection_rec,
>                               temp, bb, insn_info,
>                               GET_CODE (temp) == CLOBBER
>                               ? flags | DF_REF_MUST_CLOBBER : flags);
>          }
>        return;
>      }
>
> It seems to me that the case of (set (parallel [(set ...)])) and (set 
> (parallel [(clobber ...)])) is bogus.  I would like to simplify it to 
> the following:
>
>    /* It is legal to have a set destination be a parallel. */
>    if (GET_CODE (dst) == PARALLEL)
>      {
>        int i;
>
>        for (i = XVECLEN (dst, 0) - 1; i >= 0; i--)
>          {
>            rtx temp = XVECEXP (dst, 0, i);
>            assert (GET_CODE (temp) == EXPR_LIST);
>            df_def_record_1 (collection_rec, temp, bb, insn_info, flags);
>          }
>        return;
>      }
>
> Does this make sense?

Yeah, the docs seem pretty certain that expr_list is the only valid choice.

The docs also say that the first expr_list can be null:

  If @var{lval} is a @code{parallel}, it is used to represent the case of
  a function returning a structure in multiple registers.  Each element
  of the @code{parallel} is an @code{expr_list} whose first operand is a
  @code{reg} and whose second operand is a @code{const_int} representing the
  offset (in bytes) into the structure at which the data in that register
  corresponds.  The first element may be null to indicate that the structure
  is also passed partly in memory.

but I can't see any code to handle that.  Am I missing something,
or does the lack of a crash here mean that we can remove the last
sentence?

(It might have been added for symmetry with argument passing, where this
sort of thing is needed.  But if it isn't actually used or implemented for
returns, it might be less confusing to remove it.)

> See the attached patch for the overall thing I 
> was thinking of.
>
> Paolo
>
>         * df-scan.c (df_def_record_1): Assert a parallel must contain
>         an EXPR_LIST at this point.  Receive the LOC and move its
>         extraction...
>         (df_defs_record): ... here.  Remove superfluous braces.

Looks good.

Richard