UNSPEC_* handling in reg-stack.c

[Uros Bizjak <uros at kss-loka dot si>]

Hello!

I have a question regarding UNSPEC_* handling in subst_stack_regs_pat() 
function in reg-stack.c. As it is now, UNSPEC_SIN handles its input and 
ouptut as:

       src1 = get_true_reg (&XVECEXP (pat_src, 0, 0));
       emit_swap_insn (insn, regstack, *src1);

       src1_note = find_regno_note (insn, REG_DEAD, REGNO (*src1));
       ...
       if (src1_note)
         {
           replace_reg (&XEXP (src1_note, 0), FIRST_STACK_REG);
           regstack->top--;
           CLEAR_HARD_REG_BIT (regstack->reg_set, REGNO (*src1));
         }

I have made some test-cases that exercise this part of code, but input 
register was _never_ marked dead. This register is always replaced with 
its output, so reload pass always assings output and input to the same 
register. I would suggest changing this register handling to abort(), if 
input register is marked REG_DEAD. This would mean, that reload somehow 
failed:

       src1 = get_true_reg (&XVECEXP (pat_src, 0, 0));

       emit_swap_insn (insn, regstack, *src1);

       /* Input should never die, because it is
          replaced with output.  */
       src1_note = find_regno_note (insn, REG_DEAD, REGNO (*src1));
       if (src1_note)
         abort();

       if (STACK_REG_P (*dest))
         replace_reg (dest, FIRST_STACK_REG);

       replace_reg (src1, FIRST_STACK_REG);
       break;

I have introduced this change to UNSPEC_SIN & Co. input register 
handling, and also to input register of double-output UNSPEC_* 
instructions. The abort was never triggered, because reload pass always 
found appropriate input register (matched with the output register), 
that never dies. The situation is different in UNSPEC_FPATAN & Co., 
where second input register indeed dies, but it is unclear to me, why in 
this case we need:

       if (src1_note)  //redundant!!
         replace_reg (&XEXP (src1_note, 0), FIRST_STACK_REG);
       if (src2_note)
         replace_reg (&XEXP (src2_note, 0), FIRST_STACK_REG + 1);

I would like to introduce this change to handle double input, double 
output instructions (UNSPEC_FSCALE). UNSPEC_FSCALE is currently defined 
as "fscale;fstp %1" assembler, but with my prototype patch, "fstp %1" is 
generated automatically if one of UNSPEC_FSCALE outputs stays unused. 
(BTW: with TARGET_USE_FFREEP, when top fp register is marked REG_UNUSED, 
we could also emit "ffreep" instruction istead of "fstp").

Uros.