Re: [PING][PATCH] Move the check for any_condjump_p from sched-deps to target macros

[Jeff Law <law@redhat.com>]

On 04/26/2017 06:50 AM, Hurugalawadi, Naveen wrote:
> Hi Wilco,
> 
> Thanks for reviewing the patch.
> 
>>> The return false seems incorrect - it means a core can either have
>>> FUSE_CMP_BRANCH or FUSE_ALU_BRANCH but not both.
> Thanks for pointing out about the confusion.
> Modified the code as required.
> 
> Bootstrapped and Regression tested on AArch64 and X86_64.
> Please review the patch and let us know if its okay?
> 
> Thanks,
> Naveen
> 
> 
> fusion-anycond-jump-1.patch
> 
> 
> diff --git a/gcc/sched-deps.c b/gcc/sched-deps.c
> index b2393bf..b15a865 100644
> --- a/gcc/sched-deps.c
> +++ b/gcc/sched-deps.c
> @@ -2835,33 +2835,16 @@ sched_macro_fuse_insns (rtx_insn *insn)
>   {
>     rtx_insn *prev;
>   
> -  if (any_condjump_p (insn))
> -    {
> -      unsigned int condreg1, condreg2;
> -      rtx cc_reg_1;
> -      targetm.fixed_condition_code_regs (&condreg1, &condreg2);
> -      cc_reg_1 = gen_rtx_REG (CCmode, condreg1);
> -      prev = prev_nonnote_nondebug_insn (insn);
> -      if (!reg_referenced_p (cc_reg_1, PATTERN (insn))
> -          || !prev
> -          || !modified_in_p (cc_reg_1, prev))
> -        return;
> -    }
> -  else
> -    {
> -      rtx insn_set = single_set (insn);
> -
> -      prev = prev_nonnote_nondebug_insn (insn);
> -      if (!prev
> -          || !insn_set
> -          || !single_set (prev))
> -        return;
> +  rtx insn_set = single_set (insn);
>   
> -    }
> +  prev = prev_nonnote_nondebug_insn (insn);
> +  if (!prev
> +      || !insn_set
> +      || !single_set (prev))
> +    return;
>   
>     if (targetm.sched.macro_fusion_pair_p (prev, insn))
>       SCHED_GROUP_P (insn) = 1;
> -
>   }
Doesn't this avoid calling the target hook in cases where it used to 
call it before?

Consider a conditional jump inside a parallel that is not a single set.

Previously that could call the target hook and potentially set 
SCHED_GROUP_P.  But after your change I think it would return without 
ever querying the backend.

jeff