Re: [patch] combine.c if_then_else_cond() subreg modes

[Jeffrey A Law <law at redhat dot com>]

  In message <200101060215.VAA19089@greed.delorie.com>you write:
  > 
  > In emit-rtl.c, the comment for operand_subword says:
  > 
  > 	MODE is the mode of OP in case it is a CONST_INT.
Right.

  > But, in if_then_else_cond, when passed (subreg:SI (reg:DI 42) 1), we
  > pass SImode as the "mode".  However, shouldn't we pass the mode of the
  > inner register (DImode)?  Because, if_then_else_cond, if it returns a
  > const_int into true0 or false0, will return one sized for the reg's
  > mode, not the subreg's mode.
I don't think so.

What we're trying to do in this code is narrow the the true/false values,
so we want to use "mode" as-is instead of the mode of the inner part of
the subreg.  Why?  Because "mode" will always be smaller due to this
condition:

  /* If X is a normal SUBREG with both inner and outer modes integral,
     we can narrow both the true and false values of the inner expression,
     if there is a condition.  */
  else if (code == SUBREG && GET_MODE_CLASS (mode) == MODE_INT
           && GET_MODE_CLASS (GET_MODE (SUBREG_REG (x))) == MODE_INT
           && GET_MODE_SIZE (mode) <= GET_MODE_SIZE (GET_MODE (SUBREG_REG (x)))
           && 0 != (cond0 = if_then_else_cond (SUBREG_REG (x),
                                               &true0, &false0)))

Note the GET_MODE_CLASS ... test.

Presumably you're looking at this code for a reason :-)  Can you give more
details about what's going wrong?

jeff