Re: [RFC][PATCH][mid-end] Optimize immediate choice in comparisons.

[Jeff Law <law@redhat.com>]

On 08/14/2018 11:01 AM, Vlad Lazar wrote:
> On 13/08/18 15:00, Richard Sandiford wrote:
>> Vlad Lazar <vlad.lazar@arm.com> writes:
>>> diff --git a/gcc/expmed.h b/gcc/expmed.h
>>> index
>>> 2890d9c9bbd034f01030dd551d544bf73e73b784..86a32a643fdd0fc9f396bd2c7904244bd484df16
>>> 100644
>>> --- a/gcc/expmed.h
>>> +++ b/gcc/expmed.h
>>> @@ -702,6 +702,10 @@ extern rtx emit_store_flag (rtx, enum rtx_code,
>>> rtx, rtx, machine_mode,
>>>    extern rtx emit_store_flag_force (rtx, enum rtx_code, rtx, rtx,
>>>                      machine_mode, int, int);
>>>    +extern void canonicalize_comparison (machine_mode, enum rtx_code
>>> *, rtx *);
>>> +
>>> +extern enum rtx_code canonicalized_cmp_code (enum rtx_code);
>>
>> It would probably be better to make the second function static (local
>> to expmed.c), since it's only used internally by canonicalize_comparison.
>> Switching the order of the two functions in expmed.c would avoid the need
>> for a forward declaration.
>>
>>> @@ -6153,6 +6156,97 @@ emit_store_flag_force (rtx target, enum
>>> rtx_code code, rtx op0, rtx op1,
>>>         return target;
>>>    }
>>> +
>>> +/* Choose the more appropiate immediate in comparisons.  The purpose
>>> of this
>>
>> Maybe worth saying "scalar integer comparisons", since the function
>> doesn't handle vector or floating-point comparisons.
>>
>>> +   is to end up with an immediate which can be loaded into a
>>> register in fewer
>>> +   moves, if possible.
>>> +
>>> +   For each integer comparison there exists an equivalent choice:
>>> +     i)   a >  b or a >= b + 1
>>> +     ii)  a <= b or a <  b + 1
>>> +     iii) a >= b or a >  b - 1
>>> +     iv)  a <  b or a <= b - 1
>>> +
>>> +   The function is called in the gimple expanders which handle
>>> GIMPLE_ASSIGN
>>> +   and GIMPLE_COND.  It assumes that the first operand of the
>>> comparison is a
>>> +   register and the second is a constant.
>>
>> This last paragraph doesn't seem accurate any more.  Probably best to
>> drop it, since there's no real need to say who the callers are if we
>> describe the interface.
>>
>>> +   mode is the mode of the first operand.
>>> +   code points to the comparison code.
>>> +   imm points to the rtx containing the immediate.  */
>>
>> GCC's convention is to put parameter names in caps in comments,
>> so "MODE is...", etc.
>>
>> On the IMM line, it might be worth adding "*IMM must satisfy
>> CONST_SCALAR_INT_P on entry and continues to satisfy CONST_SCALAR_INT_P
>> on exit."
>>
>>> +void canonicalize_comparison (machine_mode mode, enum rtx_code
>>> *code, rtx *imm)
>>> +{
>>> +  int to_add = 0;
>>> +
>>> +  if (!SCALAR_INT_MODE_P (mode))
>>> +    return;
>>> +
>>> +  /* Extract the immediate value from the rtx.  */
>>> +  wide_int imm_val = wi::shwi (INTVAL (*imm), mode);
>>
>> This should be:
>>
>>    rtx_mode_t imm_val (*imm, mode);
>>
>> so that it copes with all CONST_SCALAR_INT_P, not just CONST_INT.
>>
>>> +  if (*code == GT || *code == GTU || *code == LE || *code == LEU)
>>> +    to_add = 1;
>>> +
>>> +  if (*code == GE || *code == GEU || *code == LT || *code == LTU)
>>> +    to_add = -1;
>>
>> Might be better to have:
>>
>>    if (*code == GT || *code == GTU || *code == LE || *code == LEU)
>>      to_add = 1;
>>    else if (*code == GE || *code == GEU || *code == LT || *code == LTU)
>>      to_add = -1;
>>    else
>>      return;
>>
>> so that we exit early for other comparisons, rather than doing wasted
>> work.
>>
>>> +  /* Check for overflow/underflow in the case of signed values and
>>> +     wrapping around in the case of unsigned values.  If any occur
>>> +     cancel the optimization.  */
>>> +  wide_int max_val = wi::max_value (mode, SIGNED);
>>> +  wide_int min_val = wi::min_value (mode, SIGNED);
>>> +  if ((wi::cmps (imm_val, max_val) == 0 && to_add == 1)
>>> +      || (wi::cmps (imm_val, min_val) == 0 && to_add == -1))
>>> +    return;
>>
>> This needs to use the SIGNED/UNSIGNED choice appropriate for the
>> comparison (SIGNED for GT, UNSIGNED for GTU, etc.).  There doesn't
>> seem to be an existing function that says whether an rtx comparison
>> operation is signed or not (bit of a surprise), but there is
>> unsigned_condition, which returns the unsigned form of an rtx
>> comparison operator.  It might be worth adding something like:
>>
>>    /* Return true if integer comparison operator CODE interprets its
>> operands
>>       as unsigned.  */
>>
>>    inline bool
>>    unsigned_condition_p (rtx_code code)
>>    {
>>      return unsigned_condition (code) == code;
>>    }
>>
>> and using that to choose between SIGNED and UNSIGNED.
>>
>> Rather than using wi::cmp, a more direct way of checking for overflow
>> is to change this:
>>
>>> +  /* Generate a mov instruction for both cases and see whether the
>>> change
>>> +     was profitable.  */
>>> +  wide_int imm_modif = wi::add (imm_val, to_add);
>>
>> to use the overflow form of wi::add, i.e.:
>>
>>    wide_int imm_modif = wi::add (imm_val, to_add, sign, &overflow);
>>
>> and return if "overflow" is set.
>>
>>> +
>>> +  rtx reg = gen_reg_rtx (mode);
>>
>> gen_reg_rtx creates a new pseudo register that essentially stays
>> around until we've finished compiling the function.  It's better to use:
>>
>>    gen_rtx_REG (mode, LAST_VIRTUAL_REGISTER + 1);
>>
>> for cost calculations, so that we don't waste pseudo register numbers.
>>
>>> +  rtx new_imm = GEN_INT (imm_modif.to_shwi ());
>>
>> This should be:
>>
>>    rtx new_imm = immed_wide_int_const (imm_modif, mode);
>>
>> to cope with non-CONST_INT immediates, and to ensure that CONST_INT
>> immediates are properly sign-extended.
>>
>> (The rtx interfaces are a bit clunky, sorry.)
>>
>> Patch looks good to me with those changes, but someone else will
>> need to approve.
>>
>> Thanks,
>> Richard
>>
> 
> Thanks for taking the time to point everything out.
> 
> I've updated the patch according to Richard's comments.
> See the updated patch below. OK for trunk?
> 
> Thanks,
> Vlad
> 
> ---
> 
> This patch optimises the choice of immediates in integer comparisons.
> Integer
> comparisons allow for different choices (e.g. a > b is equivalent to a
>>= b+1)
> and there are cases where an incremented/decremented immediate can be
> loaded into a
> register in fewer instructions. The cases are as follows:
> 
> i)   a >  b or a >= b + 1
> ii)  a <= b or a <  b + 1
> iii) a >= b or a >  b - 1
> iv)  a <  b or a <= b - 1
> 
> For each comparison we check whether the equivalent can be performed in
> less instructions.
> This is done in the gimple expanders. Therefore, it requires a correct
> implementation of the
> TARGET_INSN_COST hook. The change is general and it applies to any
> integer comparison, regardless
> of types or location.
> 
> For example, on AArch64 for the following code:
> 
> int foo (int x)
> {
>   return x > 0xfefffffe;
> }
> 
> it generates:
> 
> mov    w1, -16777217
> cmp    w0, w1
> cset    w0, cs
> 
> instead of:
> 
> mov    w1, 65534
> movk    w1, 0xfeff, lsl 16
> cmp    w0, w1
> cset    w0, hi
> 
> Bootstrapped and regtested on aarch64-none-linux-gnu,
> x86_64-pc-linux-gnu and there are no regressions.
> 
> Thanks,
> Vlad
> 
> gcc/testsuite/
> Changelog for gcc/testsuite/Changelog
> 2018-08-14  Vlad Lazar  <vlad.lazar@arm.com>
> 
>     * gcc.target/aarch64/imm_choice_comparison.c: New.
> 
> gcc/
> Changelog for gcc/Changelog
> 2018-08-14  Vlad Lazar  <vlad.lazar@arm.com>
>     * expmed.h (canonicalize_comparison): New declaration.
>     * expmed.c (canonicalize_comparison, equivalent_cmp_code): New.
>     * expmed.c (emit_store_flag_1): Add call to canonicalize_comparison.
>     * optabs.c (prepare_cmp_insn): Likewise.
>     * rtl.h (unsigned_condition_p): New function which checks if a
> comparison operator is unsigned.
Thanks.  I fixed up the ChangeLog entry and installed this on the trunk.

Richard S -- thanks for working with Vlad to push this forward.

jeff