Re: [RFA] Avoid unnecessary load-immediate in coremark

[Richard Sandiford <richard.sandiford@arm.com>]

Jeff Law <jlaw@ventanamicro.com> writes:
> This is another minor improvement to coremark.   I suspect this only 
> improves code size as the load-immediate was likely issuing with the ret 
> statement on multi-issue machines.
>
>
> Basically we're failing to utilize conditional equivalences during the 
> post-reload CSE pass.  So if a particular block is only reached when a 
> certain condition holds (say for example a4 == 0) and the block has an 
> assignment like a4 = 0, we would fail to eliminate the unnecessary 
> assignment.

I wasn't sure (and was too lazy to try, sorry), but: is the reason
that we fail to catch this earlier because the two uses of r4 are
entirely separate (i.e. not from the same pseudo)?

> +	  /* Iterate over each incoming edge and see if they
> +	     all have the same implicit set.  */
> +	  FOR_EACH_EDGE (e, ei, bb->preds)
> +	    {
> +	      /* If the predecessor does not end in a conditional
> +		 jump, then it does not have an implicit set.  */
> +	      if (e->src != ENTRY_BLOCK_PTR_FOR_FN (cfun)
> +		  && !block_ends_with_condjump_p (e->src))
> +		{
> +		  found = false;
> +		  break;
> +		}
> +
> +	      /* We know the predecessor ends with a conditional
> +		 jump.  Now dig into the actal form of the jump
> +		 to potentially extract an implicit set.  */

Very minor, but it looked odd to fall through for the entry block.
How about:

	      if (e->src == ENTRY_BLOCK_PTR_FOR_FN (cfun)
		  || !block_ends_with_condjump_p (e->src))

?

> +	      rtx_insn *condjump = BB_END (e->src);
> +	      if (condjump
> +		  && any_condjump_p (condjump)
> +		  && onlyjump_p (condjump))
> +		{
> +		  /* Extract the condition.  */
> +		  rtx pat = PATTERN (condjump);
> +		  rtx i_t_e = SET_SRC (pat);
> +		  gcc_assert (GET_CODE (i_t_e) == IF_THEN_ELSE);
> +		  rtx cond = XEXP (i_t_e, 0);
> +		  if ((GET_CODE (cond) == EQ
> +		       && GET_CODE (XEXP (i_t_e, 1)) == LABEL_REF
> +		       && XEXP (XEXP (i_t_e, 1), 0) == BB_HEAD (bb))
> +		      || (GET_CODE (cond) == NE
> +			  && XEXP (i_t_e, 2) == pc_rtx
> +			  && e->src->next_bb == bb))
> +		    {
> +		      /* If this is the first time through record
> +			 the source and destination.  */
> +		      if (!dest)
> +			{
> +			  dest = XEXP (cond, 0);
> +			  src = XEXP (cond, 1);
> +			}
> +		      /* If this is not the first time through, then
> +			 verify the source and destination match.  */
> +		      else if (dest == XEXP (cond, 0) && src == XEXP (cond, 1))
> +			;

FWIW, agree with what Richi said about using rtx_equal_p here.  We don't
necessarily end up with shared hard regs, especially if they originated
from different pseudos.

Thanks,
Richard