RE: [PATCH] Disable loop2_invariant for -Os

["Zhenqiang Chen" <zhenqiang dot chen at arm dot com>]

>-----Original Message-----
>From: Richard Guenther [mailto:richard.guenther@gmail.com]
>Sent: 2012å6æ28æ 17:24
>To: Zhenqiang Chen
>Cc: gcc-patches@gcc.gnu.org
>Subject: Re: [PATCH] Disable loop2_invariant for -Os
>
>On Thu, Jun 28, 2012 at 10:33 AM, Zhenqiang Chen <zhenqiang.chen@arm.com>
>wrote:
>>>> diff --git a/gcc/loop-init.c b/gcc/loop-init.c index
>>>> 03f8f61..5d8cf73
>>>> 100644
>>>> --- a/gcc/loop-init.c
>>>> +++ b/gcc/loop-init.c
>>>> @@ -273,6 +273,12 @@ struct rtl_opt_pass pass_rtl_loop_done =
>>>>  static bool
>>>>  gate_rtl_move_loop_invariants (void)
>>>>  {
>>>> +  /* In general, invariant motion can not reduce code size. But it
>>>> + will
>>>> +     change the liverange of the invariant, which increases the
>>>> + register
>>>> +     pressure and might lead to more spilling.  */
>>>> +  if (optimize_function_for_size_p (cfun))
>>>> +    return false;
>>>> +
>>>
>>>Can you do this per loop instead?  Using optimize_loop_nest_for_size_p?
>>
>> Update it according to the comments.
>>
>> Thanks!
>> -Zhenqiang
>>
>> diff --git a/gcc/loop-invariant.c b/gcc/loop-invariant.c index
>> f8405dd..b0e84a7 100644
>> --- a/gcc/loop-invariant.c
>> +++ b/gcc/loop-invariant.c
>> @@ -1931,7 +1931,8 @@ move_loop_invariants (void)
>>       curr_loop = loop;
>>       /* move_single_loop_invariants for very large loops
>>         is time consuming and might need a lot of memory.  */
>> -      if (loop->num_nodes <= (unsigned)
>> LOOP_INVARIANT_MAX_BBS_IN_LOOP)
>> +      if (loop->num_nodes <= (unsigned)
>> + LOOP_INVARIANT_MAX_BBS_IN_LOOP
>> +         && ! optimize_loop_nest_for_size_p (loop))
>>        move_single_loop_invariants (loop);
>
>Wait - move_single_loop_invariants itself already uses
>optimize_loop_for_speed_p.
>And looking down it seems to have support for tracking spill cost (eventually only
>with -fira-loop-pressure) - please work out why this support is not working for
>you.

1) If -fira_loop_pressure is enabled, it reduces ~24% invariant motions in my tests. But it does not help on total code size. Seams there is issue to update the "regs_needed" after moving an invariant out of the loop (My benchmark logs show ~73% cases have more than one invariants moved).

During tracing, I found that move an integer constant out of the loop does not increase regs_needed. Function "get_pressure_class_and_nregs (rtx insn, int *nregs)" computes the "regs_needed".

   *nregs
      = ira_reg_class_max_nregs[pressure_class][GET_MODE (SET_SRC (set))];

In ARM, the insn to set an integer is like
     (set (reg:SI 183)
        (const_int 32 [0x20])) inv1.c:64 182 {*thumb1_movsi_insn}
     (nil))
GET_MODE (SET_SRC (set)) is VOIDMode and ira_reg_class_max_nregs[pressure_class][VOIDMode] is 0. In one of my test cases, it moves 4 integer constants out of the loop, which leads to spilling.

According to the algorithm in "calculate_loop_reg_pressure", moving an invariant out of the loop should impact on the register pressure. So I try to add the following code

  if (! (*nregs))
    *nregs = ira_reg_class_max_nregs[pressure_class][GET_MODE (reg)];

Logs show it reduces another 32% invariant motions. But the code size is still far from disabling the pass. Logs show -fira_loop_pressure impact other passes in addition to loop2_invariant (The result of "-fira_loop_pressure -fno-move-loop-invariants" is different from the result of "-fno-move-loop-invariants").

2) By default -fira_loop_pressure is not enabled for -Os, the logic to compute "regs_used" seams not sound. The following codes is from function "find_invariants_to_move"
    {
      unsigned int n_regs = DF_REG_SIZE (df);

      regs_used = 2;

      for (i = 0; i < n_regs; i++)
        {
          if (!DF_REGNO_FIRST_DEF (i) && DF_REGNO_LAST_USE (i))
            {
              /* This is a value that is used but not changed inside loop.  */
              regs_used++;
            }
        }
    }
* There is no loop related inform in the code.
* Benchmark logs show the condition (!DF_REGNO_FIRST_DEF (i) && DF_REGNO_LAST_USE (i)) is never true.

Thanks!
-Zhenqiang